linux/mm/memory.c

4230 lines
114 KiB
C
Raw Normal View History

/*
* linux/mm/memory.c
*
* Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
*/
/*
* demand-loading started 01.12.91 - seems it is high on the list of
* things wanted, and it should be easy to implement. - Linus
*/
/*
* Ok, demand-loading was easy, shared pages a little bit tricker. Shared
* pages started 02.12.91, seems to work. - Linus.
*
* Tested sharing by executing about 30 /bin/sh: under the old kernel it
* would have taken more than the 6M I have free, but it worked well as
* far as I could see.
*
* Also corrected some "invalidate()"s - I wasn't doing enough of them.
*/
/*
* Real VM (paging to/from disk) started 18.12.91. Much more work and
* thought has to go into this. Oh, well..
* 19.12.91 - works, somewhat. Sometimes I get faults, don't know why.
* Found it. Everything seems to work now.
* 20.12.91 - Ok, making the swap-device changeable like the root.
*/
/*
* 05.04.94 - Multi-page memory management added for v1.1.
* Idea by Alex Bligh (alex@cconcepts.co.uk)
*
* 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG
* (Gerhard.Wichert@pdb.siemens.de)
*
* Aug/Sep 2004 Changed to four level page tables (Andi Kleen)
*/
#include <linux/kernel_stat.h>
#include <linux/mm.h>
#include <linux/hugetlb.h>
#include <linux/mman.h>
#include <linux/swap.h>
#include <linux/highmem.h>
#include <linux/pagemap.h>
#include <linux/ksm.h>
#include <linux/rmap.h>
#include <linux/export.h>
#include <linux/delayacct.h>
#include <linux/init.h>
#include <linux/writeback.h>
Memory controller: memory accounting Add the accounting hooks. The accounting is carried out for RSS and Page Cache (unmapped) pages. There is now a common limit and accounting for both. The RSS accounting is accounted at page_add_*_rmap() and page_remove_rmap() time. Page cache is accounted at add_to_page_cache(), __delete_from_page_cache(). Swap cache is also accounted for. Each page's page_cgroup is protected with the last bit of the page_cgroup pointer, this makes handling of race conditions involving simultaneous mappings of a page easier. A reference count is kept in the page_cgroup to deal with cases where a page might be unmapped from the RSS of all tasks, but still lives in the page cache. Credits go to Vaidyanathan Srinivasan for helping with reference counting work of the page cgroup. Almost all of the page cache accounting code has help from Vaidyanathan Srinivasan. [hugh@veritas.com: fix swapoff breakage] [akpm@linux-foundation.org: fix locking] Signed-off-by: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com> Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com> Cc: Pavel Emelianov <xemul@openvz.org> Cc: Paul Menage <menage@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Kirill Korotaev <dev@sw.ru> Cc: Herbert Poetzl <herbert@13thfloor.at> Cc: David Rientjes <rientjes@google.com> Cc: <Valdis.Kletnieks@vt.edu> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-07 08:13:53 +00:00
#include <linux/memcontrol.h>
mmu-notifiers: core With KVM/GFP/XPMEM there isn't just the primary CPU MMU pointing to pages. There are secondary MMUs (with secondary sptes and secondary tlbs) too. sptes in the kvm case are shadow pagetables, but when I say spte in mmu-notifier context, I mean "secondary pte". In GRU case there's no actual secondary pte and there's only a secondary tlb because the GRU secondary MMU has no knowledge about sptes and every secondary tlb miss event in the MMU always generates a page fault that has to be resolved by the CPU (this is not the case of KVM where the a secondary tlb miss will walk sptes in hardware and it will refill the secondary tlb transparently to software if the corresponding spte is present). The same way zap_page_range has to invalidate the pte before freeing the page, the spte (and secondary tlb) must also be invalidated before any page is freed and reused. Currently we take a page_count pin on every page mapped by sptes, but that means the pages can't be swapped whenever they're mapped by any spte because they're part of the guest working set. Furthermore a spte unmap event can immediately lead to a page to be freed when the pin is released (so requiring the same complex and relatively slow tlb_gather smp safe logic we have in zap_page_range and that can be avoided completely if the spte unmap event doesn't require an unpin of the page previously mapped in the secondary MMU). The mmu notifiers allow kvm/GRU/XPMEM to attach to the tsk->mm and know when the VM is swapping or freeing or doing anything on the primary MMU so that the secondary MMU code can drop sptes before the pages are freed, avoiding all page pinning and allowing 100% reliable swapping of guest physical address space. Furthermore it avoids the code that teardown the mappings of the secondary MMU, to implement a logic like tlb_gather in zap_page_range that would require many IPI to flush other cpu tlbs, for each fixed number of spte unmapped. To make an example: if what happens on the primary MMU is a protection downgrade (from writeable to wrprotect) the secondary MMU mappings will be invalidated, and the next secondary-mmu-page-fault will call get_user_pages and trigger a do_wp_page through get_user_pages if it called get_user_pages with write=1, and it'll re-establishing an updated spte or secondary-tlb-mapping on the copied page. Or it will setup a readonly spte or readonly tlb mapping if it's a guest-read, if it calls get_user_pages with write=0. This is just an example. This allows to map any page pointed by any pte (and in turn visible in the primary CPU MMU), into a secondary MMU (be it a pure tlb like GRU, or an full MMU with both sptes and secondary-tlb like the shadow-pagetable layer with kvm), or a remote DMA in software like XPMEM (hence needing of schedule in XPMEM code to send the invalidate to the remote node, while no need to schedule in kvm/gru as it's an immediate event like invalidating primary-mmu pte). At least for KVM without this patch it's impossible to swap guests reliably. And having this feature and removing the page pin allows several other optimizations that simplify life considerably. Dependencies: 1) mm_take_all_locks() to register the mmu notifier when the whole VM isn't doing anything with "mm". This allows mmu notifier users to keep track if the VM is in the middle of the invalidate_range_begin/end critical section with an atomic counter incraese in range_begin and decreased in range_end. No secondary MMU page fault is allowed to map any spte or secondary tlb reference, while the VM is in the middle of range_begin/end as any page returned by get_user_pages in that critical section could later immediately be freed without any further ->invalidate_page notification (invalidate_range_begin/end works on ranges and ->invalidate_page isn't called immediately before freeing the page). To stop all page freeing and pagetable overwrites the mmap_sem must be taken in write mode and all other anon_vma/i_mmap locks must be taken too. 2) It'd be a waste to add branches in the VM if nobody could possibly run KVM/GRU/XPMEM on the kernel, so mmu notifiers will only enabled if CONFIG_KVM=m/y. In the current kernel kvm won't yet take advantage of mmu notifiers, but this already allows to compile a KVM external module against a kernel with mmu notifiers enabled and from the next pull from kvm.git we'll start using them. And GRU/XPMEM will also be able to continue the development by enabling KVM=m in their config, until they submit all GRU/XPMEM GPLv2 code to the mainline kernel. Then they can also enable MMU_NOTIFIERS in the same way KVM does it (even if KVM=n). This guarantees nobody selects MMU_NOTIFIER=y if KVM and GRU and XPMEM are all =n. The mmu_notifier_register call can fail because mm_take_all_locks may be interrupted by a signal and return -EINTR. Because mmu_notifier_reigster is used when a driver startup, a failure can be gracefully handled. Here an example of the change applied to kvm to register the mmu notifiers. Usually when a driver startups other allocations are required anyway and -ENOMEM failure paths exists already. struct kvm *kvm_arch_create_vm(void) { struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL); + int err; if (!kvm) return ERR_PTR(-ENOMEM); INIT_LIST_HEAD(&kvm->arch.active_mmu_pages); + kvm->arch.mmu_notifier.ops = &kvm_mmu_notifier_ops; + err = mmu_notifier_register(&kvm->arch.mmu_notifier, current->mm); + if (err) { + kfree(kvm); + return ERR_PTR(err); + } + return kvm; } mmu_notifier_unregister returns void and it's reliable. The patch also adds a few needed but missing includes that would prevent kernel to compile after these changes on non-x86 archs (x86 didn't need them by luck). [akpm@linux-foundation.org: coding-style fixes] [akpm@linux-foundation.org: fix mm/filemap_xip.c build] [akpm@linux-foundation.org: fix mm/mmu_notifier.c build] Signed-off-by: Andrea Arcangeli <andrea@qumranet.com> Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Christoph Lameter <cl@linux-foundation.org> Cc: Jack Steiner <steiner@sgi.com> Cc: Robin Holt <holt@sgi.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Kanoj Sarcar <kanojsarcar@yahoo.com> Cc: Roland Dreier <rdreier@cisco.com> Cc: Steve Wise <swise@opengridcomputing.com> Cc: Avi Kivity <avi@qumranet.com> Cc: Hugh Dickins <hugh@veritas.com> Cc: Rusty Russell <rusty@rustcorp.com.au> Cc: Anthony Liguori <aliguori@us.ibm.com> Cc: Chris Wright <chrisw@redhat.com> Cc: Marcelo Tosatti <marcelo@kvack.org> Cc: Eric Dumazet <dada1@cosmosbay.com> Cc: "Paul E. McKenney" <paulmck@us.ibm.com> Cc: Izik Eidus <izike@qumranet.com> Cc: Anthony Liguori <aliguori@us.ibm.com> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-28 22:46:29 +00:00
#include <linux/mmu_notifier.h>
badpage: replace page_remove_rmap Eeek and BUG Now that bad pages are kept out of circulation, there is no need for the infamous page_remove_rmap() BUG() - once that page is freed, its negative mapcount will issue a "Bad page state" message and the page won't be freed. Removing the BUG() allows more info, on subsequent pages, to be gathered. We do have more info about the page at this point than bad_page() can know - notably, what the pmd is, which might pinpoint something like low 64kB corruption - but page_remove_rmap() isn't given the address to find that. In practice, there is only one call to page_remove_rmap() which has ever reported anything, that from zap_pte_range() (usually on exit, sometimes on munmap). It has all the info, so remove page_remove_rmap()'s "Eeek" message and leave it all to zap_pte_range(). mm/memory.c already has a hardly used print_bad_pte() function, showing some of the appropriate info: extend it to show what we want for the rmap case: pte info, page info (when there is a page) and vma info to compare. zap_pte_range() already knows the pmd, but print_bad_pte() is easier to use if it works that out for itself. Some of this info is also shown in bad_page()'s "Bad page state" message. Keep them separate, but adjust them to match each other as far as possible. Say "Bad page map" in print_bad_pte(), and add a TAINT_BAD_PAGE there too. print_bad_pte() show current->comm unconditionally (though it should get repeated in the usually irrelevant stack trace): sorry, I misled Nick Piggin to make it conditional on vm_mm == current->mm, but current->mm is already NULL in the exit case. Usually current->comm is good, though exceptionally it may not be that of the mm (when "swapoff" for example). Signed-off-by: Hugh Dickins <hugh@veritas.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:40:08 +00:00
#include <linux/kallsyms.h>
#include <linux/swapops.h>
#include <linux/elf.h>
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h percpu.h is included by sched.h and module.h and thus ends up being included when building most .c files. percpu.h includes slab.h which in turn includes gfp.h making everything defined by the two files universally available and complicating inclusion dependencies. percpu.h -> slab.h dependency is about to be removed. Prepare for this change by updating users of gfp and slab facilities include those headers directly instead of assuming availability. As this conversion needs to touch large number of source files, the following script is used as the basis of conversion. http://userweb.kernel.org/~tj/misc/slabh-sweep.py The script does the followings. * Scan files for gfp and slab usages and update includes such that only the necessary includes are there. ie. if only gfp is used, gfp.h, if slab is used, slab.h. * When the script inserts a new include, it looks at the include blocks and try to put the new include such that its order conforms to its surrounding. It's put in the include block which contains core kernel includes, in the same order that the rest are ordered - alphabetical, Christmas tree, rev-Xmas-tree or at the end if there doesn't seem to be any matching order. * If the script can't find a place to put a new include (mostly because the file doesn't have fitting include block), it prints out an error message indicating which .h file needs to be added to the file. The conversion was done in the following steps. 1. The initial automatic conversion of all .c files updated slightly over 4000 files, deleting around 700 includes and adding ~480 gfp.h and ~3000 slab.h inclusions. The script emitted errors for ~400 files. 2. Each error was manually checked. Some didn't need the inclusion, some needed manual addition while adding it to implementation .h or embedding .c file was more appropriate for others. This step added inclusions to around 150 files. 3. The script was run again and the output was compared to the edits from #2 to make sure no file was left behind. 4. Several build tests were done and a couple of problems were fixed. e.g. lib/decompress_*.c used malloc/free() wrappers around slab APIs requiring slab.h to be added manually. 5. The script was run on all .h files but without automatically editing them as sprinkling gfp.h and slab.h inclusions around .h files could easily lead to inclusion dependency hell. Most gfp.h inclusion directives were ignored as stuff from gfp.h was usually wildly available and often used in preprocessor macros. Each slab.h inclusion directive was examined and added manually as necessary. 6. percpu.h was updated not to include slab.h. 7. Build test were done on the following configurations and failures were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my distributed build env didn't work with gcov compiles) and a few more options had to be turned off depending on archs to make things build (like ipr on powerpc/64 which failed due to missing writeq). * x86 and x86_64 UP and SMP allmodconfig and a custom test config. * powerpc and powerpc64 SMP allmodconfig * sparc and sparc64 SMP allmodconfig * ia64 SMP allmodconfig * s390 SMP allmodconfig * alpha SMP allmodconfig * um on x86_64 SMP allmodconfig 8. percpu.h modifications were reverted so that it could be applied as a separate patch and serve as bisection point. Given the fact that I had only a couple of failures from tests on step 6, I'm fairly confident about the coverage of this conversion patch. If there is a breakage, it's likely to be something in one of the arch headers which should be easily discoverable easily on most builds of the specific arch. Signed-off-by: Tejun Heo <tj@kernel.org> Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 08:04:11 +00:00
#include <linux/gfp.h>
#include <linux/migrate.h>
#include <linux/string.h>
#include <asm/io.h>
#include <asm/pgalloc.h>
#include <asm/uaccess.h>
#include <asm/tlb.h>
#include <asm/tlbflush.h>
#include <asm/pgtable.h>
#include "internal.h"
[PATCH] sparsemem memory model Sparsemem abstracts the use of discontiguous mem_maps[]. This kind of mem_map[] is needed by discontiguous memory machines (like in the old CONFIG_DISCONTIGMEM case) as well as memory hotplug systems. Sparsemem replaces DISCONTIGMEM when enabled, and it is hoped that it can eventually become a complete replacement. A significant advantage over DISCONTIGMEM is that it's completely separated from CONFIG_NUMA. When producing this patch, it became apparent in that NUMA and DISCONTIG are often confused. Another advantage is that sparse doesn't require each NUMA node's ranges to be contiguous. It can handle overlapping ranges between nodes with no problems, where DISCONTIGMEM currently throws away that memory. Sparsemem uses an array to provide different pfn_to_page() translations for each SECTION_SIZE area of physical memory. This is what allows the mem_map[] to be chopped up. In order to do quick pfn_to_page() operations, the section number of the page is encoded in page->flags. Part of the sparsemem infrastructure enables sharing of these bits more dynamically (at compile-time) between the page_zone() and sparsemem operations. However, on 32-bit architectures, the number of bits is quite limited, and may require growing the size of the page->flags type in certain conditions. Several things might force this to occur: a decrease in the SECTION_SIZE (if you want to hotplug smaller areas of memory), an increase in the physical address space, or an increase in the number of used page->flags. One thing to note is that, once sparsemem is present, the NUMA node information no longer needs to be stored in the page->flags. It might provide speed increases on certain platforms and will be stored there if there is room. But, if out of room, an alternate (theoretically slower) mechanism is used. This patch introduces CONFIG_FLATMEM. It is used in almost all cases where there used to be an #ifndef DISCONTIG, because SPARSEMEM and DISCONTIGMEM often have to compile out the same areas of code. Signed-off-by: Andy Whitcroft <apw@shadowen.org> Signed-off-by: Dave Hansen <haveblue@us.ibm.com> Signed-off-by: Martin Bligh <mbligh@aracnet.com> Signed-off-by: Adrian Bunk <bunk@stusta.de> Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com> Signed-off-by: Bob Picco <bob.picco@hp.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-06-23 07:07:54 +00:00
#ifndef CONFIG_NEED_MULTIPLE_NODES
/* use the per-pgdat data instead for discontigmem - mbligh */
unsigned long max_mapnr;
struct page *mem_map;
EXPORT_SYMBOL(max_mapnr);
EXPORT_SYMBOL(mem_map);
#endif
unsigned long num_physpages;
/*
* A number of key systems in x86 including ioremap() rely on the assumption
* that high_memory defines the upper bound on direct map memory, then end
* of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
* highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
* and ZONE_HIGHMEM.
*/
void * high_memory;
EXPORT_SYMBOL(num_physpages);
EXPORT_SYMBOL(high_memory);
/*
* Randomize the address space (stacks, mmaps, brk, etc.).
*
* ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
* as ancient (libc5 based) binaries can segfault. )
*/
int randomize_va_space __read_mostly =
#ifdef CONFIG_COMPAT_BRK
1;
#else
2;
#endif
static int __init disable_randmaps(char *s)
{
randomize_va_space = 0;
return 1;
}
__setup("norandmaps", disable_randmaps);
unsigned long zero_pfn __read_mostly;
unsigned long highest_memmap_pfn __read_mostly;
mm: reinstate ZERO_PAGE KAMEZAWA Hiroyuki has observed customers of earlier kernels taking advantage of the ZERO_PAGE: which we stopped do_anonymous_page() from using in 2.6.24. And there were a couple of regression reports on LKML. Following suggestions from Linus, reinstate do_anonymous_page() use of the ZERO_PAGE; but this time avoid dirtying its struct page cacheline with (map)count updates - let vm_normal_page() regard it as abnormal. Use it only on arches which __HAVE_ARCH_PTE_SPECIAL (x86, s390, sh32, most powerpc): that's not essential, but minimizes additional branches (keeping them in the unlikely pte_special case); and incidentally excludes mips (some models of which needed eight colours of ZERO_PAGE to avoid costly exceptions). Don't be fanatical about avoiding ZERO_PAGE updates: get_user_pages() callers won't want to make exceptions for it, so increment its count there. Changes to mlock and migration? happily seems not needed. In most places it's quicker to check pfn than struct page address: prepare a __read_mostly zero_pfn for that. Does get_dump_page() still need its ZERO_PAGE check? probably not, but keep it anyway. Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk> Acked-by: Rik van Riel <riel@redhat.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Minchan Kim <minchan.kim@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-09-22 00:03:30 +00:00
/*
* CONFIG_MMU architectures set up ZERO_PAGE in their paging_init()
*/
static int __init init_zero_pfn(void)
{
zero_pfn = page_to_pfn(ZERO_PAGE(0));
return 0;
}
core_initcall(init_zero_pfn);
#if defined(SPLIT_RSS_COUNTING)
void sync_mm_rss(struct mm_struct *mm)
{
int i;
for (i = 0; i < NR_MM_COUNTERS; i++) {
if (current->rss_stat.count[i]) {
add_mm_counter(mm, i, current->rss_stat.count[i]);
current->rss_stat.count[i] = 0;
}
}
current->rss_stat.events = 0;
}
static void add_mm_counter_fast(struct mm_struct *mm, int member, int val)
{
struct task_struct *task = current;
if (likely(task->mm == mm))
task->rss_stat.count[member] += val;
else
add_mm_counter(mm, member, val);
}
#define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1)
#define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1)
/* sync counter once per 64 page faults */
#define TASK_RSS_EVENTS_THRESH (64)
static void check_sync_rss_stat(struct task_struct *task)
{
if (unlikely(task != current))
return;
if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH))
sync_mm_rss(task->mm);
}
#else /* SPLIT_RSS_COUNTING */
#define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member)
#define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member)
static void check_sync_rss_stat(struct task_struct *task)
{
}
#endif /* SPLIT_RSS_COUNTING */
#ifdef HAVE_GENERIC_MMU_GATHER
static int tlb_next_batch(struct mmu_gather *tlb)
{
struct mmu_gather_batch *batch;
batch = tlb->active;
if (batch->next) {
tlb->active = batch->next;
return 1;
}
batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0);
if (!batch)
return 0;
batch->next = NULL;
batch->nr = 0;
batch->max = MAX_GATHER_BATCH;
tlb->active->next = batch;
tlb->active = batch;
return 1;
}
/* tlb_gather_mmu
* Called to initialize an (on-stack) mmu_gather structure for page-table
* tear-down from @mm. The @fullmm argument is used when @mm is without
* users and we're going to destroy the full address space (exit/execve).
*/
void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, bool fullmm)
{
tlb->mm = mm;
tlb->fullmm = fullmm;
tlb->start = -1UL;
tlb->end = 0;
tlb->need_flush = 0;
tlb->fast_mode = (num_possible_cpus() == 1);
tlb->local.next = NULL;
tlb->local.nr = 0;
tlb->local.max = ARRAY_SIZE(tlb->__pages);
tlb->active = &tlb->local;
#ifdef CONFIG_HAVE_RCU_TABLE_FREE
tlb->batch = NULL;
#endif
}
void tlb_flush_mmu(struct mmu_gather *tlb)
{
struct mmu_gather_batch *batch;
if (!tlb->need_flush)
return;
tlb->need_flush = 0;
tlb_flush(tlb);
#ifdef CONFIG_HAVE_RCU_TABLE_FREE
tlb_table_flush(tlb);
#endif
if (tlb_fast_mode(tlb))
return;
for (batch = &tlb->local; batch; batch = batch->next) {
free_pages_and_swap_cache(batch->pages, batch->nr);
batch->nr = 0;
}
tlb->active = &tlb->local;
}
/* tlb_finish_mmu
* Called at the end of the shootdown operation to free up any resources
* that were required.
*/
void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end)
{
struct mmu_gather_batch *batch, *next;
tlb->start = start;
tlb->end = end;
tlb_flush_mmu(tlb);
/* keep the page table cache within bounds */
check_pgt_cache();
for (batch = tlb->local.next; batch; batch = next) {
next = batch->next;
free_pages((unsigned long)batch, 0);
}
tlb->local.next = NULL;
}
/* __tlb_remove_page
* Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while
* handling the additional races in SMP caused by other CPUs caching valid
* mappings in their TLBs. Returns the number of free page slots left.
* When out of page slots we must call tlb_flush_mmu().
*/
int __tlb_remove_page(struct mmu_gather *tlb, struct page *page)
{
struct mmu_gather_batch *batch;
VM_BUG_ON(!tlb->need_flush);
if (tlb_fast_mode(tlb)) {
free_page_and_swap_cache(page);
return 1; /* avoid calling tlb_flush_mmu() */
}
batch = tlb->active;
batch->pages[batch->nr++] = page;
if (batch->nr == batch->max) {
if (!tlb_next_batch(tlb))
return 0;
batch = tlb->active;
}
VM_BUG_ON(batch->nr > batch->max);
return batch->max - batch->nr;
}
#endif /* HAVE_GENERIC_MMU_GATHER */
#ifdef CONFIG_HAVE_RCU_TABLE_FREE
/*
* See the comment near struct mmu_table_batch.
*/
static void tlb_remove_table_smp_sync(void *arg)
{
/* Simply deliver the interrupt */
}
static void tlb_remove_table_one(void *table)
{
/*
* This isn't an RCU grace period and hence the page-tables cannot be
* assumed to be actually RCU-freed.
*
* It is however sufficient for software page-table walkers that rely on
* IRQ disabling. See the comment near struct mmu_table_batch.
*/
smp_call_function(tlb_remove_table_smp_sync, NULL, 1);
__tlb_remove_table(table);
}
static void tlb_remove_table_rcu(struct rcu_head *head)
{
struct mmu_table_batch *batch;
int i;
batch = container_of(head, struct mmu_table_batch, rcu);
for (i = 0; i < batch->nr; i++)
__tlb_remove_table(batch->tables[i]);
free_page((unsigned long)batch);
}
void tlb_table_flush(struct mmu_gather *tlb)
{
struct mmu_table_batch **batch = &tlb->batch;
if (*batch) {
call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu);
*batch = NULL;
}
}
void tlb_remove_table(struct mmu_gather *tlb, void *table)
{
struct mmu_table_batch **batch = &tlb->batch;
tlb->need_flush = 1;
/*
* When there's less then two users of this mm there cannot be a
* concurrent page-table walk.
*/
if (atomic_read(&tlb->mm->mm_users) < 2) {
__tlb_remove_table(table);
return;
}
if (*batch == NULL) {
*batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN);
if (*batch == NULL) {
tlb_remove_table_one(table);
return;
}
(*batch)->nr = 0;
}
(*batch)->tables[(*batch)->nr++] = table;
if ((*batch)->nr == MAX_TABLE_BATCH)
tlb_table_flush(tlb);
}
#endif /* CONFIG_HAVE_RCU_TABLE_FREE */
/*
* If a p?d_bad entry is found while walking page tables, report
* the error, before resetting entry to p?d_none. Usually (but
* very seldom) called out from the p?d_none_or_clear_bad macros.
*/
void pgd_clear_bad(pgd_t *pgd)
{
pgd_ERROR(*pgd);
pgd_clear(pgd);
}
void pud_clear_bad(pud_t *pud)
{
pud_ERROR(*pud);
pud_clear(pud);
}
void pmd_clear_bad(pmd_t *pmd)
{
pmd_ERROR(*pmd);
pmd_clear(pmd);
}
/*
* Note: this doesn't free the actual pages themselves. That
* has been handled earlier when unmapping all the memory regions.
*/
static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
unsigned long addr)
{
CONFIG_HIGHPTE vs. sub-page page tables. Background: I've implemented 1K/2K page tables for s390. These sub-page page tables are required to properly support the s390 virtualization instruction with KVM. The SIE instruction requires that the page tables have 256 page table entries (pte) followed by 256 page status table entries (pgste). The pgstes are only required if the process is using the SIE instruction. The pgstes are updated by the hardware and by the hypervisor for a number of reasons, one of them is dirty and reference bit tracking. To avoid wasting memory the standard pte table allocation should return 1K/2K (31/64 bit) and 2K/4K if the process is using SIE. Problem: Page size on s390 is 4K, page table size is 1K or 2K. That means the s390 version for pte_alloc_one cannot return a pointer to a struct page. Trouble is that with the CONFIG_HIGHPTE feature on x86 pte_alloc_one cannot return a pointer to a pte either, since that would require more than 32 bit for the return value of pte_alloc_one (and the pte * would not be accessible since its not kmapped). Solution: The only solution I found to this dilemma is a new typedef: a pgtable_t. For s390 pgtable_t will be a (pte *) - to be introduced with a later patch. For everybody else it will be a (struct page *). The additional problem with the initialization of the ptl lock and the NR_PAGETABLE accounting is solved with a constructor pgtable_page_ctor and a destructor pgtable_page_dtor. The page table allocation and free functions need to call these two whenever a page table page is allocated or freed. pmd_populate will get a pgtable_t instead of a struct page pointer. To get the pgtable_t back from a pmd entry that has been installed with pmd_populate a new function pmd_pgtable is added. It replaces the pmd_page call in free_pte_range and apply_to_pte_range. Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:22:04 +00:00
pgtable_t token = pmd_pgtable(*pmd);
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
pmd_clear(pmd);
pte_free_tlb(tlb, token, addr);
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
tlb->mm->nr_ptes--;
}
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
unsigned long addr, unsigned long end,
unsigned long floor, unsigned long ceiling)
{
pmd_t *pmd;
unsigned long next;
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
unsigned long start;
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
start = addr;
pmd = pmd_offset(pud, addr);
do {
next = pmd_addr_end(addr, end);
if (pmd_none_or_clear_bad(pmd))
continue;
free_pte_range(tlb, pmd, addr);
} while (pmd++, addr = next, addr != end);
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
start &= PUD_MASK;
if (start < floor)
return;
if (ceiling) {
ceiling &= PUD_MASK;
if (!ceiling)
return;
}
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
if (end - 1 > ceiling - 1)
return;
pmd = pmd_offset(pud, start);
pud_clear(pud);
pmd_free_tlb(tlb, pmd, start);
}
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
unsigned long addr, unsigned long end,
unsigned long floor, unsigned long ceiling)
{
pud_t *pud;
unsigned long next;
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
unsigned long start;
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
start = addr;
pud = pud_offset(pgd, addr);
do {
next = pud_addr_end(addr, end);
if (pud_none_or_clear_bad(pud))
continue;
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
free_pmd_range(tlb, pud, addr, next, floor, ceiling);
} while (pud++, addr = next, addr != end);
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
start &= PGDIR_MASK;
if (start < floor)
return;
if (ceiling) {
ceiling &= PGDIR_MASK;
if (!ceiling)
return;
}
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
if (end - 1 > ceiling - 1)
return;
pud = pud_offset(pgd, start);
pgd_clear(pgd);
pud_free_tlb(tlb, pud, start);
}
/*
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
* This function frees user-level page tables of a process.
*
* Must be called with pagetable lock held.
*/
void free_pgd_range(struct mmu_gather *tlb,
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
unsigned long addr, unsigned long end,
unsigned long floor, unsigned long ceiling)
{
pgd_t *pgd;
unsigned long next;
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
/*
* The next few lines have given us lots of grief...
*
* Why are we testing PMD* at this top level? Because often
* there will be no work to do at all, and we'd prefer not to
* go all the way down to the bottom just to discover that.
*
* Why all these "- 1"s? Because 0 represents both the bottom
* of the address space and the top of it (using -1 for the
* top wouldn't help much: the masks would do the wrong thing).
* The rule is that addr 0 and floor 0 refer to the bottom of
* the address space, but end 0 and ceiling 0 refer to the top
* Comparisons need to use "end - 1" and "ceiling - 1" (though
* that end 0 case should be mythical).
*
* Wherever addr is brought up or ceiling brought down, we must
* be careful to reject "the opposite 0" before it confuses the
* subsequent tests. But what about where end is brought down
* by PMD_SIZE below? no, end can't go down to 0 there.
*
* Whereas we round start (addr) and ceiling down, by different
* masks at different levels, in order to test whether a table
* now has no other vmas using it, so can be freed, we don't
* bother to round floor or end up - the tests don't need that.
*/
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
addr &= PMD_MASK;
if (addr < floor) {
addr += PMD_SIZE;
if (!addr)
return;
}
if (ceiling) {
ceiling &= PMD_MASK;
if (!ceiling)
return;
}
if (end - 1 > ceiling - 1)
end -= PMD_SIZE;
if (addr > end - 1)
return;
pgd = pgd_offset(tlb->mm, addr);
do {
next = pgd_addr_end(addr, end);
if (pgd_none_or_clear_bad(pgd))
continue;
free_pud_range(tlb, pgd, addr, next, floor, ceiling);
} while (pgd++, addr = next, addr != end);
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
}
void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
unsigned long floor, unsigned long ceiling)
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
{
while (vma) {
struct vm_area_struct *next = vma->vm_next;
unsigned long addr = vma->vm_start;
[PATCH] mm: unlink vma before pagetables In most places the descent from pgd to pud to pmd to pte holds mmap_sem (exclusively or not), which ensures that free_pgtables cannot be freeing page tables from any level at the same time. But truncation and reverse mapping descend without mmap_sem. No problem: just make sure that a vma is unlinked from its prio_tree (or nonlinear list) and from its anon_vma list, after zapping the vma, but before freeing its page tables. Then neither vmtruncate nor rmap can reach that vma whose page tables are now volatile (nor do they need to reach it, since all its page entries have been zapped by this stage). The i_mmap_lock and anon_vma->lock already serialize this correctly; but the locking hierarchy is such that we cannot take them while holding page_table_lock. Well, we're trying to push that down anyway. So in this patch, move anon_vma_unlink and unlink_file_vma into free_pgtables, at the same time as moving page_table_lock around calls to unmap_vmas. tlb_gather_mmu and tlb_finish_mmu then fall outside the page_table_lock, but we made them preempt_disable and preempt_enable earlier; and a long source audit of all the architectures has shown no problem with removing page_table_lock from them. free_pgtables doesn't need page_table_lock for itself, nor for what it calls; tlb->mm->nr_ptes is usually protected by page_table_lock, but partly by non-exclusive mmap_sem - here it's decremented with exclusive mmap_sem, or mm_users 0. update_hiwater_rss and vm_unacct_memory don't need page_table_lock either. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:29 +00:00
/*
* Hide vma from rmap and truncate_pagecache before freeing
* pgtables
[PATCH] mm: unlink vma before pagetables In most places the descent from pgd to pud to pmd to pte holds mmap_sem (exclusively or not), which ensures that free_pgtables cannot be freeing page tables from any level at the same time. But truncation and reverse mapping descend without mmap_sem. No problem: just make sure that a vma is unlinked from its prio_tree (or nonlinear list) and from its anon_vma list, after zapping the vma, but before freeing its page tables. Then neither vmtruncate nor rmap can reach that vma whose page tables are now volatile (nor do they need to reach it, since all its page entries have been zapped by this stage). The i_mmap_lock and anon_vma->lock already serialize this correctly; but the locking hierarchy is such that we cannot take them while holding page_table_lock. Well, we're trying to push that down anyway. So in this patch, move anon_vma_unlink and unlink_file_vma into free_pgtables, at the same time as moving page_table_lock around calls to unmap_vmas. tlb_gather_mmu and tlb_finish_mmu then fall outside the page_table_lock, but we made them preempt_disable and preempt_enable earlier; and a long source audit of all the architectures has shown no problem with removing page_table_lock from them. free_pgtables doesn't need page_table_lock for itself, nor for what it calls; tlb->mm->nr_ptes is usually protected by page_table_lock, but partly by non-exclusive mmap_sem - here it's decremented with exclusive mmap_sem, or mm_users 0. update_hiwater_rss and vm_unacct_memory don't need page_table_lock either. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:29 +00:00
*/
mm: change anon_vma linking to fix multi-process server scalability issue The old anon_vma code can lead to scalability issues with heavily forking workloads. Specifically, each anon_vma will be shared between the parent process and all its child processes. In a workload with 1000 child processes and a VMA with 1000 anonymous pages per process that get COWed, this leads to a system with a million anonymous pages in the same anon_vma, each of which is mapped in just one of the 1000 processes. However, the current rmap code needs to walk them all, leading to O(N) scanning complexity for each page. This can result in systems where one CPU is walking the page tables of 1000 processes in page_referenced_one, while all other CPUs are stuck on the anon_vma lock. This leads to catastrophic failure for a benchmark like AIM7, where the total number of processes can reach in the tens of thousands. Real workloads are still a factor 10 less process intensive than AIM7, but they are catching up. This patch changes the way anon_vmas and VMAs are linked, which allows us to associate multiple anon_vmas with a VMA. At fork time, each child process gets its own anon_vmas, in which its COWed pages will be instantiated. The parents' anon_vma is also linked to the VMA, because non-COWed pages could be present in any of the children. This reduces rmap scanning complexity to O(1) for the pages of the 1000 child processes, with O(N) complexity for at most 1/N pages in the system. This reduces the average scanning cost in heavily forking workloads from O(N) to 2. The only real complexity in this patch stems from the fact that linking a VMA to anon_vmas now involves memory allocations. This means vma_adjust can fail, if it needs to attach a VMA to anon_vma structures. This in turn means error handling needs to be added to the calling functions. A second source of complexity is that, because there can be multiple anon_vmas, the anon_vma linking in vma_adjust can no longer be done under "the" anon_vma lock. To prevent the rmap code from walking up an incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h to make sure it is impossible to compile a kernel that needs both symbolic values for the same bitflag. Some test results: Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test box with 16GB RAM and not quite enough IO), the system ends up running >99% in system time, with every CPU on the same anon_vma lock in the pageout code. With these changes, AIM7 hits the cross-over point around 29.7k users. This happens with ~99% IO wait time, there never seems to be any spike in system time. The anon_vma lock contention appears to be resolved. [akpm@linux-foundation.org: cleanups] Signed-off-by: Rik van Riel <riel@redhat.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Larry Woodman <lwoodman@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com> Cc: Minchan Kim <minchan.kim@gmail.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 21:42:07 +00:00
unlink_anon_vmas(vma);
[PATCH] mm: unlink vma before pagetables In most places the descent from pgd to pud to pmd to pte holds mmap_sem (exclusively or not), which ensures that free_pgtables cannot be freeing page tables from any level at the same time. But truncation and reverse mapping descend without mmap_sem. No problem: just make sure that a vma is unlinked from its prio_tree (or nonlinear list) and from its anon_vma list, after zapping the vma, but before freeing its page tables. Then neither vmtruncate nor rmap can reach that vma whose page tables are now volatile (nor do they need to reach it, since all its page entries have been zapped by this stage). The i_mmap_lock and anon_vma->lock already serialize this correctly; but the locking hierarchy is such that we cannot take them while holding page_table_lock. Well, we're trying to push that down anyway. So in this patch, move anon_vma_unlink and unlink_file_vma into free_pgtables, at the same time as moving page_table_lock around calls to unmap_vmas. tlb_gather_mmu and tlb_finish_mmu then fall outside the page_table_lock, but we made them preempt_disable and preempt_enable earlier; and a long source audit of all the architectures has shown no problem with removing page_table_lock from them. free_pgtables doesn't need page_table_lock for itself, nor for what it calls; tlb->mm->nr_ptes is usually protected by page_table_lock, but partly by non-exclusive mmap_sem - here it's decremented with exclusive mmap_sem, or mm_users 0. update_hiwater_rss and vm_unacct_memory don't need page_table_lock either. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:29 +00:00
unlink_file_vma(vma);
[PATCH] hugepage: Fix hugepage logic in free_pgtables() free_pgtables() has special logic to call hugetlb_free_pgd_range() instead of the normal free_pgd_range() on hugepage VMAs. However, the test it uses to do so is incorrect: it calls is_hugepage_only_range on a hugepage sized range at the start of the vma. is_hugepage_only_range() will return true if the given range has any intersection with a hugepage address region, and in this case the given region need not be hugepage aligned. So, for example, this test can return true if called on, say, a 4k VMA immediately preceding a (nicely aligned) hugepage VMA. At present we get away with this because the powerpc version of hugetlb_free_pgd_range() is just a call to free_pgd_range(). On ia64 (the only other arch with a non-trivial is_hugepage_only_range()) we get away with it for a different reason; the hugepage area is not contiguous with the rest of the user address space, and VMAs are not permitted in between, so the test can't return a false positive there. Nonetheless this should be fixed. We do that in the patch below by replacing the is_hugepage_only_range() test with an explicit test of the VMA using is_vm_hugetlb_page(). This in turn changes behaviour for platforms where is_hugepage_only_range() returns false always (everything except powerpc and ia64). We address this by ensuring that hugetlb_free_pgd_range() is defined to be identical to free_pgd_range() (instead of a no-op) on everything except ia64. Even so, it will prevent some otherwise possible coalescing of calls down to free_pgd_range(). Since this only happens for hugepage VMAs, removing this small optimization seems unlikely to cause any trouble. This patch causes no regressions on the libhugetlbfs testsuite - ppc64 POWER5 (8-way), ppc64 G5 (2-way) and i386 Pentium M (UP). Signed-off-by: David Gibson <dwg@au1.ibm.com> Cc: William Lee Irwin III <wli@holomorphy.com> Acked-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-22 08:08:57 +00:00
if (is_vm_hugetlb_page(vma)) {
hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
floor, next? next->vm_start: ceiling);
} else {
/*
* Optimization: gather nearby vmas into one call down
*/
while (next && next->vm_start <= vma->vm_end + PMD_SIZE
&& !is_vm_hugetlb_page(next)) {
vma = next;
next = vma->vm_next;
mm: change anon_vma linking to fix multi-process server scalability issue The old anon_vma code can lead to scalability issues with heavily forking workloads. Specifically, each anon_vma will be shared between the parent process and all its child processes. In a workload with 1000 child processes and a VMA with 1000 anonymous pages per process that get COWed, this leads to a system with a million anonymous pages in the same anon_vma, each of which is mapped in just one of the 1000 processes. However, the current rmap code needs to walk them all, leading to O(N) scanning complexity for each page. This can result in systems where one CPU is walking the page tables of 1000 processes in page_referenced_one, while all other CPUs are stuck on the anon_vma lock. This leads to catastrophic failure for a benchmark like AIM7, where the total number of processes can reach in the tens of thousands. Real workloads are still a factor 10 less process intensive than AIM7, but they are catching up. This patch changes the way anon_vmas and VMAs are linked, which allows us to associate multiple anon_vmas with a VMA. At fork time, each child process gets its own anon_vmas, in which its COWed pages will be instantiated. The parents' anon_vma is also linked to the VMA, because non-COWed pages could be present in any of the children. This reduces rmap scanning complexity to O(1) for the pages of the 1000 child processes, with O(N) complexity for at most 1/N pages in the system. This reduces the average scanning cost in heavily forking workloads from O(N) to 2. The only real complexity in this patch stems from the fact that linking a VMA to anon_vmas now involves memory allocations. This means vma_adjust can fail, if it needs to attach a VMA to anon_vma structures. This in turn means error handling needs to be added to the calling functions. A second source of complexity is that, because there can be multiple anon_vmas, the anon_vma linking in vma_adjust can no longer be done under "the" anon_vma lock. To prevent the rmap code from walking up an incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h to make sure it is impossible to compile a kernel that needs both symbolic values for the same bitflag. Some test results: Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test box with 16GB RAM and not quite enough IO), the system ends up running >99% in system time, with every CPU on the same anon_vma lock in the pageout code. With these changes, AIM7 hits the cross-over point around 29.7k users. This happens with ~99% IO wait time, there never seems to be any spike in system time. The anon_vma lock contention appears to be resolved. [akpm@linux-foundation.org: cleanups] Signed-off-by: Rik van Riel <riel@redhat.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Larry Woodman <lwoodman@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com> Cc: Minchan Kim <minchan.kim@gmail.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 21:42:07 +00:00
unlink_anon_vmas(vma);
[PATCH] mm: unlink vma before pagetables In most places the descent from pgd to pud to pmd to pte holds mmap_sem (exclusively or not), which ensures that free_pgtables cannot be freeing page tables from any level at the same time. But truncation and reverse mapping descend without mmap_sem. No problem: just make sure that a vma is unlinked from its prio_tree (or nonlinear list) and from its anon_vma list, after zapping the vma, but before freeing its page tables. Then neither vmtruncate nor rmap can reach that vma whose page tables are now volatile (nor do they need to reach it, since all its page entries have been zapped by this stage). The i_mmap_lock and anon_vma->lock already serialize this correctly; but the locking hierarchy is such that we cannot take them while holding page_table_lock. Well, we're trying to push that down anyway. So in this patch, move anon_vma_unlink and unlink_file_vma into free_pgtables, at the same time as moving page_table_lock around calls to unmap_vmas. tlb_gather_mmu and tlb_finish_mmu then fall outside the page_table_lock, but we made them preempt_disable and preempt_enable earlier; and a long source audit of all the architectures has shown no problem with removing page_table_lock from them. free_pgtables doesn't need page_table_lock for itself, nor for what it calls; tlb->mm->nr_ptes is usually protected by page_table_lock, but partly by non-exclusive mmap_sem - here it's decremented with exclusive mmap_sem, or mm_users 0. update_hiwater_rss and vm_unacct_memory don't need page_table_lock either. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:29 +00:00
unlink_file_vma(vma);
}
free_pgd_range(tlb, addr, vma->vm_end,
floor, next? next->vm_start: ceiling);
}
[PATCH] freepgt: free_pgtables use vma list Recent woes with some arches needing their own pgd_addr_end macro; and 4-level clear_page_range regression since 2.6.10's clear_page_tables; and its long-standing well-known inefficiency in searching throughout the higher-level page tables for those few entries to clear and free: all can be blamed on ignoring the list of vmas when we free page tables. Replace exit_mmap's clear_page_range of the total user address space by free_pgtables operating on the mm's vma list; unmap_region use it in the same way, giving floor and ceiling beyond which it may not free tables. This brings lmbench fork/exec/sh numbers back to 2.6.10 (unless preempt is enabled, in which case latency fixes spoil unmap_vmas throughput). Beware: the do_mmap_pgoff driver failure case must now use unmap_region instead of zap_page_range, since a page table might have been allocated, and can only be freed while it is touched by some vma. Move free_pgtables from mmap.c to memory.c, where its lower levels are adapted from the clear_page_range levels. (Most of free_pgtables' old code was actually for a non-existent case, prev not properly set up, dating from before hch gave us split_vma.) Pass mmu_gather** in the public interfaces, since we might want to add latency lockdrops later; but no attempt to do so yet, going by vma should itself reduce latency. But what if is_hugepage_only_range? Those ia64 and ppc64 cases need careful examination: put that off until a later patch of the series. What of x86_64's 32bit vdso page __map_syscall32 maps outside any vma? And the range to sparc64's flush_tlb_pgtables? It's less clear to me now that we need to do more than is done here - every PMD_SIZE ever occupied will be flushed, do we really have to flush every PGDIR_SIZE ever partially occupied? A shame to complicate it unnecessarily. Special thanks to David Miller for time spent repairing my ceilings. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-04-19 20:29:15 +00:00
vma = next;
}
}
int __pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma,
pmd_t *pmd, unsigned long address)
{
CONFIG_HIGHPTE vs. sub-page page tables. Background: I've implemented 1K/2K page tables for s390. These sub-page page tables are required to properly support the s390 virtualization instruction with KVM. The SIE instruction requires that the page tables have 256 page table entries (pte) followed by 256 page status table entries (pgste). The pgstes are only required if the process is using the SIE instruction. The pgstes are updated by the hardware and by the hypervisor for a number of reasons, one of them is dirty and reference bit tracking. To avoid wasting memory the standard pte table allocation should return 1K/2K (31/64 bit) and 2K/4K if the process is using SIE. Problem: Page size on s390 is 4K, page table size is 1K or 2K. That means the s390 version for pte_alloc_one cannot return a pointer to a struct page. Trouble is that with the CONFIG_HIGHPTE feature on x86 pte_alloc_one cannot return a pointer to a pte either, since that would require more than 32 bit for the return value of pte_alloc_one (and the pte * would not be accessible since its not kmapped). Solution: The only solution I found to this dilemma is a new typedef: a pgtable_t. For s390 pgtable_t will be a (pte *) - to be introduced with a later patch. For everybody else it will be a (struct page *). The additional problem with the initialization of the ptl lock and the NR_PAGETABLE accounting is solved with a constructor pgtable_page_ctor and a destructor pgtable_page_dtor. The page table allocation and free functions need to call these two whenever a page table page is allocated or freed. pmd_populate will get a pgtable_t instead of a struct page pointer. To get the pgtable_t back from a pmd entry that has been installed with pmd_populate a new function pmd_pgtable is added. It replaces the pmd_page call in free_pte_range and apply_to_pte_range. Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:22:04 +00:00
pgtable_t new = pte_alloc_one(mm, address);
int wait_split_huge_page;
if (!new)
return -ENOMEM;
fix SMP data race in pagetable setup vs walking There is a possible data race in the page table walking code. After the split ptlock patches, it actually seems to have been introduced to the core code, but even before that I think it would have impacted some architectures (powerpc and sparc64, at least, walk the page tables without taking locks eg. see find_linux_pte()). The race is as follows: The pte page is allocated, zeroed, and its struct page gets its spinlock initialized. The mm-wide ptl is then taken, and then the pte page is inserted into the pagetables. At this point, the spinlock is not guaranteed to have ordered the previous stores to initialize the pte page with the subsequent store to put it in the page tables. So another Linux page table walker might be walking down (without any locks, because we have split-leaf-ptls), and find that new pte we've inserted. It might try to take the spinlock before the store from the other CPU initializes it. And subsequently it might read a pte_t out before stores from the other CPU have cleared the memory. There are also similar races in higher levels of the page tables. They obviously don't involve the spinlock, but could see uninitialized memory. Arch code and hardware pagetable walkers that walk the pagetables without locks could see similar uninitialized memory problems, regardless of whether split ptes are enabled or not. I prefer to put the barriers in core code, because that's where the higher level logic happens, but the page table accessors are per-arch, and open-coding them everywhere I don't think is an option. I'll put the read-side barriers in alpha arch code for now (other architectures perform data-dependent loads in order). Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-05-14 04:37:36 +00:00
/*
* Ensure all pte setup (eg. pte page lock and page clearing) are
* visible before the pte is made visible to other CPUs by being
* put into page tables.
*
* The other side of the story is the pointer chasing in the page
* table walking code (when walking the page table without locking;
* ie. most of the time). Fortunately, these data accesses consist
* of a chain of data-dependent loads, meaning most CPUs (alpha
* being the notable exception) will already guarantee loads are
* seen in-order. See the alpha page table accessors for the
* smp_read_barrier_depends() barriers in page table walking code.
*/
smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
spin_lock(&mm->page_table_lock);
wait_split_huge_page = 0;
if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
mm->nr_ptes++;
pmd_populate(mm, pmd, new);
CONFIG_HIGHPTE vs. sub-page page tables. Background: I've implemented 1K/2K page tables for s390. These sub-page page tables are required to properly support the s390 virtualization instruction with KVM. The SIE instruction requires that the page tables have 256 page table entries (pte) followed by 256 page status table entries (pgste). The pgstes are only required if the process is using the SIE instruction. The pgstes are updated by the hardware and by the hypervisor for a number of reasons, one of them is dirty and reference bit tracking. To avoid wasting memory the standard pte table allocation should return 1K/2K (31/64 bit) and 2K/4K if the process is using SIE. Problem: Page size on s390 is 4K, page table size is 1K or 2K. That means the s390 version for pte_alloc_one cannot return a pointer to a struct page. Trouble is that with the CONFIG_HIGHPTE feature on x86 pte_alloc_one cannot return a pointer to a pte either, since that would require more than 32 bit for the return value of pte_alloc_one (and the pte * would not be accessible since its not kmapped). Solution: The only solution I found to this dilemma is a new typedef: a pgtable_t. For s390 pgtable_t will be a (pte *) - to be introduced with a later patch. For everybody else it will be a (struct page *). The additional problem with the initialization of the ptl lock and the NR_PAGETABLE accounting is solved with a constructor pgtable_page_ctor and a destructor pgtable_page_dtor. The page table allocation and free functions need to call these two whenever a page table page is allocated or freed. pmd_populate will get a pgtable_t instead of a struct page pointer. To get the pgtable_t back from a pmd entry that has been installed with pmd_populate a new function pmd_pgtable is added. It replaces the pmd_page call in free_pte_range and apply_to_pte_range. Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:22:04 +00:00
new = NULL;
} else if (unlikely(pmd_trans_splitting(*pmd)))
wait_split_huge_page = 1;
spin_unlock(&mm->page_table_lock);
CONFIG_HIGHPTE vs. sub-page page tables. Background: I've implemented 1K/2K page tables for s390. These sub-page page tables are required to properly support the s390 virtualization instruction with KVM. The SIE instruction requires that the page tables have 256 page table entries (pte) followed by 256 page status table entries (pgste). The pgstes are only required if the process is using the SIE instruction. The pgstes are updated by the hardware and by the hypervisor for a number of reasons, one of them is dirty and reference bit tracking. To avoid wasting memory the standard pte table allocation should return 1K/2K (31/64 bit) and 2K/4K if the process is using SIE. Problem: Page size on s390 is 4K, page table size is 1K or 2K. That means the s390 version for pte_alloc_one cannot return a pointer to a struct page. Trouble is that with the CONFIG_HIGHPTE feature on x86 pte_alloc_one cannot return a pointer to a pte either, since that would require more than 32 bit for the return value of pte_alloc_one (and the pte * would not be accessible since its not kmapped). Solution: The only solution I found to this dilemma is a new typedef: a pgtable_t. For s390 pgtable_t will be a (pte *) - to be introduced with a later patch. For everybody else it will be a (struct page *). The additional problem with the initialization of the ptl lock and the NR_PAGETABLE accounting is solved with a constructor pgtable_page_ctor and a destructor pgtable_page_dtor. The page table allocation and free functions need to call these two whenever a page table page is allocated or freed. pmd_populate will get a pgtable_t instead of a struct page pointer. To get the pgtable_t back from a pmd entry that has been installed with pmd_populate a new function pmd_pgtable is added. It replaces the pmd_page call in free_pte_range and apply_to_pte_range. Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:22:04 +00:00
if (new)
pte_free(mm, new);
if (wait_split_huge_page)
wait_split_huge_page(vma->anon_vma, pmd);
return 0;
}
int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
{
pte_t *new = pte_alloc_one_kernel(&init_mm, address);
if (!new)
return -ENOMEM;
fix SMP data race in pagetable setup vs walking There is a possible data race in the page table walking code. After the split ptlock patches, it actually seems to have been introduced to the core code, but even before that I think it would have impacted some architectures (powerpc and sparc64, at least, walk the page tables without taking locks eg. see find_linux_pte()). The race is as follows: The pte page is allocated, zeroed, and its struct page gets its spinlock initialized. The mm-wide ptl is then taken, and then the pte page is inserted into the pagetables. At this point, the spinlock is not guaranteed to have ordered the previous stores to initialize the pte page with the subsequent store to put it in the page tables. So another Linux page table walker might be walking down (without any locks, because we have split-leaf-ptls), and find that new pte we've inserted. It might try to take the spinlock before the store from the other CPU initializes it. And subsequently it might read a pte_t out before stores from the other CPU have cleared the memory. There are also similar races in higher levels of the page tables. They obviously don't involve the spinlock, but could see uninitialized memory. Arch code and hardware pagetable walkers that walk the pagetables without locks could see similar uninitialized memory problems, regardless of whether split ptes are enabled or not. I prefer to put the barriers in core code, because that's where the higher level logic happens, but the page table accessors are per-arch, and open-coding them everywhere I don't think is an option. I'll put the read-side barriers in alpha arch code for now (other architectures perform data-dependent loads in order). Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-05-14 04:37:36 +00:00
smp_wmb(); /* See comment in __pte_alloc */
spin_lock(&init_mm.page_table_lock);
if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
pmd_populate_kernel(&init_mm, pmd, new);
CONFIG_HIGHPTE vs. sub-page page tables. Background: I've implemented 1K/2K page tables for s390. These sub-page page tables are required to properly support the s390 virtualization instruction with KVM. The SIE instruction requires that the page tables have 256 page table entries (pte) followed by 256 page status table entries (pgste). The pgstes are only required if the process is using the SIE instruction. The pgstes are updated by the hardware and by the hypervisor for a number of reasons, one of them is dirty and reference bit tracking. To avoid wasting memory the standard pte table allocation should return 1K/2K (31/64 bit) and 2K/4K if the process is using SIE. Problem: Page size on s390 is 4K, page table size is 1K or 2K. That means the s390 version for pte_alloc_one cannot return a pointer to a struct page. Trouble is that with the CONFIG_HIGHPTE feature on x86 pte_alloc_one cannot return a pointer to a pte either, since that would require more than 32 bit for the return value of pte_alloc_one (and the pte * would not be accessible since its not kmapped). Solution: The only solution I found to this dilemma is a new typedef: a pgtable_t. For s390 pgtable_t will be a (pte *) - to be introduced with a later patch. For everybody else it will be a (struct page *). The additional problem with the initialization of the ptl lock and the NR_PAGETABLE accounting is solved with a constructor pgtable_page_ctor and a destructor pgtable_page_dtor. The page table allocation and free functions need to call these two whenever a page table page is allocated or freed. pmd_populate will get a pgtable_t instead of a struct page pointer. To get the pgtable_t back from a pmd entry that has been installed with pmd_populate a new function pmd_pgtable is added. It replaces the pmd_page call in free_pte_range and apply_to_pte_range. Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:22:04 +00:00
new = NULL;
} else
VM_BUG_ON(pmd_trans_splitting(*pmd));
spin_unlock(&init_mm.page_table_lock);
CONFIG_HIGHPTE vs. sub-page page tables. Background: I've implemented 1K/2K page tables for s390. These sub-page page tables are required to properly support the s390 virtualization instruction with KVM. The SIE instruction requires that the page tables have 256 page table entries (pte) followed by 256 page status table entries (pgste). The pgstes are only required if the process is using the SIE instruction. The pgstes are updated by the hardware and by the hypervisor for a number of reasons, one of them is dirty and reference bit tracking. To avoid wasting memory the standard pte table allocation should return 1K/2K (31/64 bit) and 2K/4K if the process is using SIE. Problem: Page size on s390 is 4K, page table size is 1K or 2K. That means the s390 version for pte_alloc_one cannot return a pointer to a struct page. Trouble is that with the CONFIG_HIGHPTE feature on x86 pte_alloc_one cannot return a pointer to a pte either, since that would require more than 32 bit for the return value of pte_alloc_one (and the pte * would not be accessible since its not kmapped). Solution: The only solution I found to this dilemma is a new typedef: a pgtable_t. For s390 pgtable_t will be a (pte *) - to be introduced with a later patch. For everybody else it will be a (struct page *). The additional problem with the initialization of the ptl lock and the NR_PAGETABLE accounting is solved with a constructor pgtable_page_ctor and a destructor pgtable_page_dtor. The page table allocation and free functions need to call these two whenever a page table page is allocated or freed. pmd_populate will get a pgtable_t instead of a struct page pointer. To get the pgtable_t back from a pmd entry that has been installed with pmd_populate a new function pmd_pgtable is added. It replaces the pmd_page call in free_pte_range and apply_to_pte_range. Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:22:04 +00:00
if (new)
pte_free_kernel(&init_mm, new);
return 0;
}
static inline void init_rss_vec(int *rss)
{
memset(rss, 0, sizeof(int) * NR_MM_COUNTERS);
}
static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss)
{
int i;
if (current->mm == mm)
sync_mm_rss(mm);
for (i = 0; i < NR_MM_COUNTERS; i++)
if (rss[i])
add_mm_counter(mm, i, rss[i]);
}
2005-10-30 01:16:12 +00:00
/*
* This function is called to print an error when a bad pte
* is found. For example, we might have a PFN-mapped pte in
* a region that doesn't allow it.
2005-10-30 01:16:12 +00:00
*
* The calling function must still handle the error.
*/
badpage: replace page_remove_rmap Eeek and BUG Now that bad pages are kept out of circulation, there is no need for the infamous page_remove_rmap() BUG() - once that page is freed, its negative mapcount will issue a "Bad page state" message and the page won't be freed. Removing the BUG() allows more info, on subsequent pages, to be gathered. We do have more info about the page at this point than bad_page() can know - notably, what the pmd is, which might pinpoint something like low 64kB corruption - but page_remove_rmap() isn't given the address to find that. In practice, there is only one call to page_remove_rmap() which has ever reported anything, that from zap_pte_range() (usually on exit, sometimes on munmap). It has all the info, so remove page_remove_rmap()'s "Eeek" message and leave it all to zap_pte_range(). mm/memory.c already has a hardly used print_bad_pte() function, showing some of the appropriate info: extend it to show what we want for the rmap case: pte info, page info (when there is a page) and vma info to compare. zap_pte_range() already knows the pmd, but print_bad_pte() is easier to use if it works that out for itself. Some of this info is also shown in bad_page()'s "Bad page state" message. Keep them separate, but adjust them to match each other as far as possible. Say "Bad page map" in print_bad_pte(), and add a TAINT_BAD_PAGE there too. print_bad_pte() show current->comm unconditionally (though it should get repeated in the usually irrelevant stack trace): sorry, I misled Nick Piggin to make it conditional on vm_mm == current->mm, but current->mm is already NULL in the exit case. Usually current->comm is good, though exceptionally it may not be that of the mm (when "swapoff" for example). Signed-off-by: Hugh Dickins <hugh@veritas.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:40:08 +00:00
static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
pte_t pte, struct page *page)
2005-10-30 01:16:12 +00:00
{
badpage: replace page_remove_rmap Eeek and BUG Now that bad pages are kept out of circulation, there is no need for the infamous page_remove_rmap() BUG() - once that page is freed, its negative mapcount will issue a "Bad page state" message and the page won't be freed. Removing the BUG() allows more info, on subsequent pages, to be gathered. We do have more info about the page at this point than bad_page() can know - notably, what the pmd is, which might pinpoint something like low 64kB corruption - but page_remove_rmap() isn't given the address to find that. In practice, there is only one call to page_remove_rmap() which has ever reported anything, that from zap_pte_range() (usually on exit, sometimes on munmap). It has all the info, so remove page_remove_rmap()'s "Eeek" message and leave it all to zap_pte_range(). mm/memory.c already has a hardly used print_bad_pte() function, showing some of the appropriate info: extend it to show what we want for the rmap case: pte info, page info (when there is a page) and vma info to compare. zap_pte_range() already knows the pmd, but print_bad_pte() is easier to use if it works that out for itself. Some of this info is also shown in bad_page()'s "Bad page state" message. Keep them separate, but adjust them to match each other as far as possible. Say "Bad page map" in print_bad_pte(), and add a TAINT_BAD_PAGE there too. print_bad_pte() show current->comm unconditionally (though it should get repeated in the usually irrelevant stack trace): sorry, I misled Nick Piggin to make it conditional on vm_mm == current->mm, but current->mm is already NULL in the exit case. Usually current->comm is good, though exceptionally it may not be that of the mm (when "swapoff" for example). Signed-off-by: Hugh Dickins <hugh@veritas.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:40:08 +00:00
pgd_t *pgd = pgd_offset(vma->vm_mm, addr);
pud_t *pud = pud_offset(pgd, addr);
pmd_t *pmd = pmd_offset(pud, addr);
struct address_space *mapping;
pgoff_t index;
static unsigned long resume;
static unsigned long nr_shown;
static unsigned long nr_unshown;
/*
* Allow a burst of 60 reports, then keep quiet for that minute;
* or allow a steady drip of one report per second.
*/
if (nr_shown == 60) {
if (time_before(jiffies, resume)) {
nr_unshown++;
return;
}
if (nr_unshown) {
printk(KERN_ALERT
"BUG: Bad page map: %lu messages suppressed\n",
nr_unshown);
nr_unshown = 0;
}
nr_shown = 0;
}
if (nr_shown++ == 0)
resume = jiffies + 60 * HZ;
badpage: replace page_remove_rmap Eeek and BUG Now that bad pages are kept out of circulation, there is no need for the infamous page_remove_rmap() BUG() - once that page is freed, its negative mapcount will issue a "Bad page state" message and the page won't be freed. Removing the BUG() allows more info, on subsequent pages, to be gathered. We do have more info about the page at this point than bad_page() can know - notably, what the pmd is, which might pinpoint something like low 64kB corruption - but page_remove_rmap() isn't given the address to find that. In practice, there is only one call to page_remove_rmap() which has ever reported anything, that from zap_pte_range() (usually on exit, sometimes on munmap). It has all the info, so remove page_remove_rmap()'s "Eeek" message and leave it all to zap_pte_range(). mm/memory.c already has a hardly used print_bad_pte() function, showing some of the appropriate info: extend it to show what we want for the rmap case: pte info, page info (when there is a page) and vma info to compare. zap_pte_range() already knows the pmd, but print_bad_pte() is easier to use if it works that out for itself. Some of this info is also shown in bad_page()'s "Bad page state" message. Keep them separate, but adjust them to match each other as far as possible. Say "Bad page map" in print_bad_pte(), and add a TAINT_BAD_PAGE there too. print_bad_pte() show current->comm unconditionally (though it should get repeated in the usually irrelevant stack trace): sorry, I misled Nick Piggin to make it conditional on vm_mm == current->mm, but current->mm is already NULL in the exit case. Usually current->comm is good, though exceptionally it may not be that of the mm (when "swapoff" for example). Signed-off-by: Hugh Dickins <hugh@veritas.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:40:08 +00:00
mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;
index = linear_page_index(vma, addr);
printk(KERN_ALERT
"BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",
badpage: replace page_remove_rmap Eeek and BUG Now that bad pages are kept out of circulation, there is no need for the infamous page_remove_rmap() BUG() - once that page is freed, its negative mapcount will issue a "Bad page state" message and the page won't be freed. Removing the BUG() allows more info, on subsequent pages, to be gathered. We do have more info about the page at this point than bad_page() can know - notably, what the pmd is, which might pinpoint something like low 64kB corruption - but page_remove_rmap() isn't given the address to find that. In practice, there is only one call to page_remove_rmap() which has ever reported anything, that from zap_pte_range() (usually on exit, sometimes on munmap). It has all the info, so remove page_remove_rmap()'s "Eeek" message and leave it all to zap_pte_range(). mm/memory.c already has a hardly used print_bad_pte() function, showing some of the appropriate info: extend it to show what we want for the rmap case: pte info, page info (when there is a page) and vma info to compare. zap_pte_range() already knows the pmd, but print_bad_pte() is easier to use if it works that out for itself. Some of this info is also shown in bad_page()'s "Bad page state" message. Keep them separate, but adjust them to match each other as far as possible. Say "Bad page map" in print_bad_pte(), and add a TAINT_BAD_PAGE there too. print_bad_pte() show current->comm unconditionally (though it should get repeated in the usually irrelevant stack trace): sorry, I misled Nick Piggin to make it conditional on vm_mm == current->mm, but current->mm is already NULL in the exit case. Usually current->comm is good, though exceptionally it may not be that of the mm (when "swapoff" for example). Signed-off-by: Hugh Dickins <hugh@veritas.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:40:08 +00:00
current->comm,
(long long)pte_val(pte), (long long)pmd_val(*pmd));
if (page)
dump_page(page);
printk(KERN_ALERT
badpage: replace page_remove_rmap Eeek and BUG Now that bad pages are kept out of circulation, there is no need for the infamous page_remove_rmap() BUG() - once that page is freed, its negative mapcount will issue a "Bad page state" message and the page won't be freed. Removing the BUG() allows more info, on subsequent pages, to be gathered. We do have more info about the page at this point than bad_page() can know - notably, what the pmd is, which might pinpoint something like low 64kB corruption - but page_remove_rmap() isn't given the address to find that. In practice, there is only one call to page_remove_rmap() which has ever reported anything, that from zap_pte_range() (usually on exit, sometimes on munmap). It has all the info, so remove page_remove_rmap()'s "Eeek" message and leave it all to zap_pte_range(). mm/memory.c already has a hardly used print_bad_pte() function, showing some of the appropriate info: extend it to show what we want for the rmap case: pte info, page info (when there is a page) and vma info to compare. zap_pte_range() already knows the pmd, but print_bad_pte() is easier to use if it works that out for itself. Some of this info is also shown in bad_page()'s "Bad page state" message. Keep them separate, but adjust them to match each other as far as possible. Say "Bad page map" in print_bad_pte(), and add a TAINT_BAD_PAGE there too. print_bad_pte() show current->comm unconditionally (though it should get repeated in the usually irrelevant stack trace): sorry, I misled Nick Piggin to make it conditional on vm_mm == current->mm, but current->mm is already NULL in the exit case. Usually current->comm is good, though exceptionally it may not be that of the mm (when "swapoff" for example). Signed-off-by: Hugh Dickins <hugh@veritas.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:40:08 +00:00
"addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
(void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
/*
* Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y
*/
if (vma->vm_ops)
print_symbol(KERN_ALERT "vma->vm_ops->fault: %s\n",
badpage: replace page_remove_rmap Eeek and BUG Now that bad pages are kept out of circulation, there is no need for the infamous page_remove_rmap() BUG() - once that page is freed, its negative mapcount will issue a "Bad page state" message and the page won't be freed. Removing the BUG() allows more info, on subsequent pages, to be gathered. We do have more info about the page at this point than bad_page() can know - notably, what the pmd is, which might pinpoint something like low 64kB corruption - but page_remove_rmap() isn't given the address to find that. In practice, there is only one call to page_remove_rmap() which has ever reported anything, that from zap_pte_range() (usually on exit, sometimes on munmap). It has all the info, so remove page_remove_rmap()'s "Eeek" message and leave it all to zap_pte_range(). mm/memory.c already has a hardly used print_bad_pte() function, showing some of the appropriate info: extend it to show what we want for the rmap case: pte info, page info (when there is a page) and vma info to compare. zap_pte_range() already knows the pmd, but print_bad_pte() is easier to use if it works that out for itself. Some of this info is also shown in bad_page()'s "Bad page state" message. Keep them separate, but adjust them to match each other as far as possible. Say "Bad page map" in print_bad_pte(), and add a TAINT_BAD_PAGE there too. print_bad_pte() show current->comm unconditionally (though it should get repeated in the usually irrelevant stack trace): sorry, I misled Nick Piggin to make it conditional on vm_mm == current->mm, but current->mm is already NULL in the exit case. Usually current->comm is good, though exceptionally it may not be that of the mm (when "swapoff" for example). Signed-off-by: Hugh Dickins <hugh@veritas.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:40:08 +00:00
(unsigned long)vma->vm_ops->fault);
if (vma->vm_file && vma->vm_file->f_op)
print_symbol(KERN_ALERT "vma->vm_file->f_op->mmap: %s\n",
badpage: replace page_remove_rmap Eeek and BUG Now that bad pages are kept out of circulation, there is no need for the infamous page_remove_rmap() BUG() - once that page is freed, its negative mapcount will issue a "Bad page state" message and the page won't be freed. Removing the BUG() allows more info, on subsequent pages, to be gathered. We do have more info about the page at this point than bad_page() can know - notably, what the pmd is, which might pinpoint something like low 64kB corruption - but page_remove_rmap() isn't given the address to find that. In practice, there is only one call to page_remove_rmap() which has ever reported anything, that from zap_pte_range() (usually on exit, sometimes on munmap). It has all the info, so remove page_remove_rmap()'s "Eeek" message and leave it all to zap_pte_range(). mm/memory.c already has a hardly used print_bad_pte() function, showing some of the appropriate info: extend it to show what we want for the rmap case: pte info, page info (when there is a page) and vma info to compare. zap_pte_range() already knows the pmd, but print_bad_pte() is easier to use if it works that out for itself. Some of this info is also shown in bad_page()'s "Bad page state" message. Keep them separate, but adjust them to match each other as far as possible. Say "Bad page map" in print_bad_pte(), and add a TAINT_BAD_PAGE there too. print_bad_pte() show current->comm unconditionally (though it should get repeated in the usually irrelevant stack trace): sorry, I misled Nick Piggin to make it conditional on vm_mm == current->mm, but current->mm is already NULL in the exit case. Usually current->comm is good, though exceptionally it may not be that of the mm (when "swapoff" for example). Signed-off-by: Hugh Dickins <hugh@veritas.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:40:08 +00:00
(unsigned long)vma->vm_file->f_op->mmap);
2005-10-30 01:16:12 +00:00
dump_stack();
badpage: replace page_remove_rmap Eeek and BUG Now that bad pages are kept out of circulation, there is no need for the infamous page_remove_rmap() BUG() - once that page is freed, its negative mapcount will issue a "Bad page state" message and the page won't be freed. Removing the BUG() allows more info, on subsequent pages, to be gathered. We do have more info about the page at this point than bad_page() can know - notably, what the pmd is, which might pinpoint something like low 64kB corruption - but page_remove_rmap() isn't given the address to find that. In practice, there is only one call to page_remove_rmap() which has ever reported anything, that from zap_pte_range() (usually on exit, sometimes on munmap). It has all the info, so remove page_remove_rmap()'s "Eeek" message and leave it all to zap_pte_range(). mm/memory.c already has a hardly used print_bad_pte() function, showing some of the appropriate info: extend it to show what we want for the rmap case: pte info, page info (when there is a page) and vma info to compare. zap_pte_range() already knows the pmd, but print_bad_pte() is easier to use if it works that out for itself. Some of this info is also shown in bad_page()'s "Bad page state" message. Keep them separate, but adjust them to match each other as far as possible. Say "Bad page map" in print_bad_pte(), and add a TAINT_BAD_PAGE there too. print_bad_pte() show current->comm unconditionally (though it should get repeated in the usually irrelevant stack trace): sorry, I misled Nick Piggin to make it conditional on vm_mm == current->mm, but current->mm is already NULL in the exit case. Usually current->comm is good, though exceptionally it may not be that of the mm (when "swapoff" for example). Signed-off-by: Hugh Dickins <hugh@veritas.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:40:08 +00:00
add_taint(TAINT_BAD_PAGE);
2005-10-30 01:16:12 +00:00
}
mm: move all mmu notifier invocations to be done outside the PT lock In order to allow sleeping during mmu notifier calls, we need to avoid invoking them under the page table spinlock. This patch solves the problem by calling invalidate_page notification after releasing the lock (but before freeing the page itself), or by wrapping the page invalidation with calls to invalidate_range_begin and invalidate_range_end. To prevent accidental changes to the invalidate_range_end arguments after the call to invalidate_range_begin, the patch introduces a convention of saving the arguments in consistently named locals: unsigned long mmun_start; /* For mmu_notifiers */ unsigned long mmun_end; /* For mmu_notifiers */ ... mmun_start = ... mmun_end = ... mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); ... mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); The patch changes code to use this convention for all calls to mmu_notifier_invalidate_range_start/end, except those where the calls are close enough so that anyone who glances at the code can see the values aren't changing. This patchset is a preliminary step towards on-demand paging design to be added to the RDMA stack. Why do we want on-demand paging for Infiniband? Applications register memory with an RDMA adapter using system calls, and subsequently post IO operations that refer to the corresponding virtual addresses directly to HW. Until now, this was achieved by pinning the memory during the registration calls. The goal of on demand paging is to avoid pinning the pages of registered memory regions (MRs). This will allow users the same flexibility they get when swapping any other part of their processes address spaces. Instead of requiring the entire MR to fit in physical memory, we can allow the MR to be larger, and only fit the current working set in physical memory. Why should anyone care? What problems are users currently experiencing? This can make programming with RDMA much simpler. Today, developers that are working with more data than their RAM can hold need either to deregister and reregister memory regions throughout their process's life, or keep a single memory region and copy the data to it. On demand paging will allow these developers to register a single MR at the beginning of their process's life, and let the operating system manage which pages needs to be fetched at a given time. In the future, we might be able to provide a single memory access key for each process that would provide the entire process's address as one large memory region, and the developers wouldn't need to register memory regions at all. Is there any prospect that any other subsystems will utilise these infrastructural changes? If so, which and how, etc? As for other subsystems, I understand that XPMEM wanted to sleep in MMU notifiers, as Christoph Lameter wrote at http://lkml.indiana.edu/hypermail/linux/kernel/0802.1/0460.html and perhaps Andrea knows about other use cases. Scheduling in mmu notifications is required since we need to sync the hardware with the secondary page tables change. A TLB flush of an IO device is inherently slower than a CPU TLB flush, so our design works by sending the invalidation request to the device, and waiting for an interrupt before exiting the mmu notifier handler. Avi said: kvm may be a buyer. kvm::mmu_lock, which serializes guest page faults, also protects long operations such as destroying large ranges. It would be good to convert it into a spinlock, but as it is used inside mmu notifiers, this cannot be done. (there are alternatives, such as keeping the spinlock and using a generation counter to do the teardown in O(1), which is what the "may" is doing up there). [akpm@linux-foundation.orgpossible speed tweak in hugetlb_cow(), cleanups] Signed-off-by: Andrea Arcangeli <andrea@qumranet.com> Signed-off-by: Sagi Grimberg <sagig@mellanox.com> Signed-off-by: Haggai Eran <haggaie@mellanox.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com> Cc: Or Gerlitz <ogerlitz@mellanox.com> Cc: Haggai Eran <haggaie@mellanox.com> Cc: Shachar Raindel <raindel@mellanox.com> Cc: Liran Liss <liranl@mellanox.com> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Avi Kivity <avi@redhat.com> Cc: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-08 23:33:33 +00:00
static inline bool is_cow_mapping(vm_flags_t flags)
{
return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
}
/*
mm: introduce pte_special pte bit s390 for one, cannot implement VM_MIXEDMAP with pfn_valid, due to their memory model (which is more dynamic than most). Instead, they had proposed to implement it with an additional path through vm_normal_page(), using a bit in the pte to determine whether or not the page should be refcounted: vm_normal_page() { ... if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { if (vma->vm_flags & VM_MIXEDMAP) { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; #else if (!pfn_valid(pfn)) return NULL; #endif goto out; } ... } This is fine, however if we are allowed to use a bit in the pte to determine refcountedness, we can use that to _completely_ replace all the vma based schemes. So instead of adding more cases to the already complex vma-based scheme, we can have a clearly seperate and simple pte-based scheme (and get slightly better code generation in the process): vm_normal_page() { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; return pte_page(pte); #else ... #endif } And finally, we may rather make this concept usable by any architecture rather than making it s390 only, so implement a new type of pte state for this. Unfortunately the old vma based code must stay, because some architectures may not be able to spare pte bits. This makes vm_normal_page a little bit more ugly than we would like, but the 2 cases are clearly seperate. So introduce a pte_special pte state, and use it in mm/memory.c. It is currently a noop for all architectures, so this doesn't actually result in any compiled code changes to mm/memory.o. BTW: I haven't put vm_normal_page() into arch code as-per an earlier suggestion. The reason is that, regardless of where vm_normal_page is actually implemented, the *abstraction* is still exactly the same. Also, while it depends on whether the architecture has pte_special or not, that is the only two possible cases, and it really isn't an arch specific function -- the role of the arch code should be to provide primitive functions and accessors with which to build the core code; pte_special does that. We do not want architectures to know or care about vm_normal_page itself, and we definitely don't want them being able to invent something new there out of sight of mm/ code. If we made vm_normal_page an arch function, then we have to make vm_insert_mixed (next patch) an arch function too. So I don't think moving it to arch code fundamentally improves any abstractions, while it does practically make the code more difficult to follow, for both mm and arch developers, and easier to misuse. [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Acked-by: Carsten Otte <cotte@de.ibm.com> Cc: Jared Hulbert <jaredeh@gmail.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 09:13:00 +00:00
* vm_normal_page -- This function gets the "struct page" associated with a pte.
*
mm: introduce pte_special pte bit s390 for one, cannot implement VM_MIXEDMAP with pfn_valid, due to their memory model (which is more dynamic than most). Instead, they had proposed to implement it with an additional path through vm_normal_page(), using a bit in the pte to determine whether or not the page should be refcounted: vm_normal_page() { ... if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { if (vma->vm_flags & VM_MIXEDMAP) { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; #else if (!pfn_valid(pfn)) return NULL; #endif goto out; } ... } This is fine, however if we are allowed to use a bit in the pte to determine refcountedness, we can use that to _completely_ replace all the vma based schemes. So instead of adding more cases to the already complex vma-based scheme, we can have a clearly seperate and simple pte-based scheme (and get slightly better code generation in the process): vm_normal_page() { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; return pte_page(pte); #else ... #endif } And finally, we may rather make this concept usable by any architecture rather than making it s390 only, so implement a new type of pte state for this. Unfortunately the old vma based code must stay, because some architectures may not be able to spare pte bits. This makes vm_normal_page a little bit more ugly than we would like, but the 2 cases are clearly seperate. So introduce a pte_special pte state, and use it in mm/memory.c. It is currently a noop for all architectures, so this doesn't actually result in any compiled code changes to mm/memory.o. BTW: I haven't put vm_normal_page() into arch code as-per an earlier suggestion. The reason is that, regardless of where vm_normal_page is actually implemented, the *abstraction* is still exactly the same. Also, while it depends on whether the architecture has pte_special or not, that is the only two possible cases, and it really isn't an arch specific function -- the role of the arch code should be to provide primitive functions and accessors with which to build the core code; pte_special does that. We do not want architectures to know or care about vm_normal_page itself, and we definitely don't want them being able to invent something new there out of sight of mm/ code. If we made vm_normal_page an arch function, then we have to make vm_insert_mixed (next patch) an arch function too. So I don't think moving it to arch code fundamentally improves any abstractions, while it does practically make the code more difficult to follow, for both mm and arch developers, and easier to misuse. [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Acked-by: Carsten Otte <cotte@de.ibm.com> Cc: Jared Hulbert <jaredeh@gmail.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 09:13:00 +00:00
* "Special" mappings do not wish to be associated with a "struct page" (either
* it doesn't exist, or it exists but they don't want to touch it). In this
* case, NULL is returned here. "Normal" mappings do have a struct page.
mm: introduce VM_MIXEDMAP This series introduces some important infrastructure work. The overall result is that: 1. We now support XIP backed filesystems using memory that have no struct page allocated to them. And patches 6 and 7 actually implement this for s390. This is pretty important in a number of cases. As far as I understand, in the case of virtualisation (eg. s390), each guest may mount a readonly copy of the same filesystem (eg. the distro). Currently, guests need to allocate struct pages for this image. So if you have 100 guests, you already need to allocate more memory for the struct pages than the size of the image. I think. (Carsten?) For other (eg. embedded) systems, you may have a very large non- volatile filesystem. If you have to have struct pages for this, then your RAM consumption will go up proportionally to fs size. Even though it is just a small proportion, the RAM can be much more costly eg in terms of power, so every KB less that Linux uses makes it more attractive to a lot of these guys. 2. VM_MIXEDMAP allows us to support mappings where you actually do want to refcount _some_ pages in the mapping, but not others, and support COW on arbitrary (non-linear) mappings. Jared needs this for his NVRAM filesystem in progress. Future iterations of this filesystem will most likely want to migrate pages between pagecache and XIP backing, which is where the requirement for mixed (some refcounted, some not) comes from. 3. pte_special also has a peripheral usage that I need for my lockless get_user_pages patch. That was shown to speed up "oltp" on db2 by 10% on a 2 socket system, which is kind of significant because they scrounge for months to try to find 0.1% improvement on these workloads. I'm hoping we might finally be faster than AIX on pSeries with this :). My reference to lockless get_user_pages is not meant to justify this patchset (which doesn't include lockless gup), but just to show that pte_special is not some s390 specific thing that should be hidden in arch code or xip code: I definitely want to use it on at least x86 and powerpc as well. This patch: Introduce a new type of mapping, VM_MIXEDMAP. This is unlike VM_PFNMAP in that it can support COW mappings of arbitrary ranges including ranges without struct page *and* ranges with a struct page that we actually want to refcount (PFNMAP can only support COW in those cases where the un-COW-ed translations are mapped linearly in the virtual address, and can only support non refcounted ranges). VM_MIXEDMAP achieves this by refcounting all pfn_valid pages, and not refcounting !pfn_valid pages (which is not an option for VM_PFNMAP, because it needs to avoid refcounting pfn_valid pages eg. for /dev/mem mappings). Signed-off-by: Jared Hulbert <jaredeh@gmail.com> Signed-off-by: Nick Piggin <npiggin@suse.de> Acked-by: Carsten Otte <cotte@de.ibm.com> Cc: Jared Hulbert <jaredeh@gmail.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 09:12:58 +00:00
*
mm: introduce pte_special pte bit s390 for one, cannot implement VM_MIXEDMAP with pfn_valid, due to their memory model (which is more dynamic than most). Instead, they had proposed to implement it with an additional path through vm_normal_page(), using a bit in the pte to determine whether or not the page should be refcounted: vm_normal_page() { ... if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { if (vma->vm_flags & VM_MIXEDMAP) { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; #else if (!pfn_valid(pfn)) return NULL; #endif goto out; } ... } This is fine, however if we are allowed to use a bit in the pte to determine refcountedness, we can use that to _completely_ replace all the vma based schemes. So instead of adding more cases to the already complex vma-based scheme, we can have a clearly seperate and simple pte-based scheme (and get slightly better code generation in the process): vm_normal_page() { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; return pte_page(pte); #else ... #endif } And finally, we may rather make this concept usable by any architecture rather than making it s390 only, so implement a new type of pte state for this. Unfortunately the old vma based code must stay, because some architectures may not be able to spare pte bits. This makes vm_normal_page a little bit more ugly than we would like, but the 2 cases are clearly seperate. So introduce a pte_special pte state, and use it in mm/memory.c. It is currently a noop for all architectures, so this doesn't actually result in any compiled code changes to mm/memory.o. BTW: I haven't put vm_normal_page() into arch code as-per an earlier suggestion. The reason is that, regardless of where vm_normal_page is actually implemented, the *abstraction* is still exactly the same. Also, while it depends on whether the architecture has pte_special or not, that is the only two possible cases, and it really isn't an arch specific function -- the role of the arch code should be to provide primitive functions and accessors with which to build the core code; pte_special does that. We do not want architectures to know or care about vm_normal_page itself, and we definitely don't want them being able to invent something new there out of sight of mm/ code. If we made vm_normal_page an arch function, then we have to make vm_insert_mixed (next patch) an arch function too. So I don't think moving it to arch code fundamentally improves any abstractions, while it does practically make the code more difficult to follow, for both mm and arch developers, and easier to misuse. [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Acked-by: Carsten Otte <cotte@de.ibm.com> Cc: Jared Hulbert <jaredeh@gmail.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 09:13:00 +00:00
* There are 2 broad cases. Firstly, an architecture may define a pte_special()
* pte bit, in which case this function is trivial. Secondly, an architecture
* may not have a spare pte bit, which requires a more complicated scheme,
* described below.
*
* A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
* special mapping (even if there are underlying and valid "struct pages").
* COWed pages of a VM_PFNMAP are always normal.
*
mm: introduce VM_MIXEDMAP This series introduces some important infrastructure work. The overall result is that: 1. We now support XIP backed filesystems using memory that have no struct page allocated to them. And patches 6 and 7 actually implement this for s390. This is pretty important in a number of cases. As far as I understand, in the case of virtualisation (eg. s390), each guest may mount a readonly copy of the same filesystem (eg. the distro). Currently, guests need to allocate struct pages for this image. So if you have 100 guests, you already need to allocate more memory for the struct pages than the size of the image. I think. (Carsten?) For other (eg. embedded) systems, you may have a very large non- volatile filesystem. If you have to have struct pages for this, then your RAM consumption will go up proportionally to fs size. Even though it is just a small proportion, the RAM can be much more costly eg in terms of power, so every KB less that Linux uses makes it more attractive to a lot of these guys. 2. VM_MIXEDMAP allows us to support mappings where you actually do want to refcount _some_ pages in the mapping, but not others, and support COW on arbitrary (non-linear) mappings. Jared needs this for his NVRAM filesystem in progress. Future iterations of this filesystem will most likely want to migrate pages between pagecache and XIP backing, which is where the requirement for mixed (some refcounted, some not) comes from. 3. pte_special also has a peripheral usage that I need for my lockless get_user_pages patch. That was shown to speed up "oltp" on db2 by 10% on a 2 socket system, which is kind of significant because they scrounge for months to try to find 0.1% improvement on these workloads. I'm hoping we might finally be faster than AIX on pSeries with this :). My reference to lockless get_user_pages is not meant to justify this patchset (which doesn't include lockless gup), but just to show that pte_special is not some s390 specific thing that should be hidden in arch code or xip code: I definitely want to use it on at least x86 and powerpc as well. This patch: Introduce a new type of mapping, VM_MIXEDMAP. This is unlike VM_PFNMAP in that it can support COW mappings of arbitrary ranges including ranges without struct page *and* ranges with a struct page that we actually want to refcount (PFNMAP can only support COW in those cases where the un-COW-ed translations are mapped linearly in the virtual address, and can only support non refcounted ranges). VM_MIXEDMAP achieves this by refcounting all pfn_valid pages, and not refcounting !pfn_valid pages (which is not an option for VM_PFNMAP, because it needs to avoid refcounting pfn_valid pages eg. for /dev/mem mappings). Signed-off-by: Jared Hulbert <jaredeh@gmail.com> Signed-off-by: Nick Piggin <npiggin@suse.de> Acked-by: Carsten Otte <cotte@de.ibm.com> Cc: Jared Hulbert <jaredeh@gmail.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 09:12:58 +00:00
* The way we recognize COWed pages within VM_PFNMAP mappings is through the
* rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
mm: introduce pte_special pte bit s390 for one, cannot implement VM_MIXEDMAP with pfn_valid, due to their memory model (which is more dynamic than most). Instead, they had proposed to implement it with an additional path through vm_normal_page(), using a bit in the pte to determine whether or not the page should be refcounted: vm_normal_page() { ... if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { if (vma->vm_flags & VM_MIXEDMAP) { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; #else if (!pfn_valid(pfn)) return NULL; #endif goto out; } ... } This is fine, however if we are allowed to use a bit in the pte to determine refcountedness, we can use that to _completely_ replace all the vma based schemes. So instead of adding more cases to the already complex vma-based scheme, we can have a clearly seperate and simple pte-based scheme (and get slightly better code generation in the process): vm_normal_page() { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; return pte_page(pte); #else ... #endif } And finally, we may rather make this concept usable by any architecture rather than making it s390 only, so implement a new type of pte state for this. Unfortunately the old vma based code must stay, because some architectures may not be able to spare pte bits. This makes vm_normal_page a little bit more ugly than we would like, but the 2 cases are clearly seperate. So introduce a pte_special pte state, and use it in mm/memory.c. It is currently a noop for all architectures, so this doesn't actually result in any compiled code changes to mm/memory.o. BTW: I haven't put vm_normal_page() into arch code as-per an earlier suggestion. The reason is that, regardless of where vm_normal_page is actually implemented, the *abstraction* is still exactly the same. Also, while it depends on whether the architecture has pte_special or not, that is the only two possible cases, and it really isn't an arch specific function -- the role of the arch code should be to provide primitive functions and accessors with which to build the core code; pte_special does that. We do not want architectures to know or care about vm_normal_page itself, and we definitely don't want them being able to invent something new there out of sight of mm/ code. If we made vm_normal_page an arch function, then we have to make vm_insert_mixed (next patch) an arch function too. So I don't think moving it to arch code fundamentally improves any abstractions, while it does practically make the code more difficult to follow, for both mm and arch developers, and easier to misuse. [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Acked-by: Carsten Otte <cotte@de.ibm.com> Cc: Jared Hulbert <jaredeh@gmail.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 09:13:00 +00:00
* set, and the vm_pgoff will point to the first PFN mapped: thus every special
* mapping will always honor the rule
*
* pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
*
mm: introduce pte_special pte bit s390 for one, cannot implement VM_MIXEDMAP with pfn_valid, due to their memory model (which is more dynamic than most). Instead, they had proposed to implement it with an additional path through vm_normal_page(), using a bit in the pte to determine whether or not the page should be refcounted: vm_normal_page() { ... if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { if (vma->vm_flags & VM_MIXEDMAP) { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; #else if (!pfn_valid(pfn)) return NULL; #endif goto out; } ... } This is fine, however if we are allowed to use a bit in the pte to determine refcountedness, we can use that to _completely_ replace all the vma based schemes. So instead of adding more cases to the already complex vma-based scheme, we can have a clearly seperate and simple pte-based scheme (and get slightly better code generation in the process): vm_normal_page() { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; return pte_page(pte); #else ... #endif } And finally, we may rather make this concept usable by any architecture rather than making it s390 only, so implement a new type of pte state for this. Unfortunately the old vma based code must stay, because some architectures may not be able to spare pte bits. This makes vm_normal_page a little bit more ugly than we would like, but the 2 cases are clearly seperate. So introduce a pte_special pte state, and use it in mm/memory.c. It is currently a noop for all architectures, so this doesn't actually result in any compiled code changes to mm/memory.o. BTW: I haven't put vm_normal_page() into arch code as-per an earlier suggestion. The reason is that, regardless of where vm_normal_page is actually implemented, the *abstraction* is still exactly the same. Also, while it depends on whether the architecture has pte_special or not, that is the only two possible cases, and it really isn't an arch specific function -- the role of the arch code should be to provide primitive functions and accessors with which to build the core code; pte_special does that. We do not want architectures to know or care about vm_normal_page itself, and we definitely don't want them being able to invent something new there out of sight of mm/ code. If we made vm_normal_page an arch function, then we have to make vm_insert_mixed (next patch) an arch function too. So I don't think moving it to arch code fundamentally improves any abstractions, while it does practically make the code more difficult to follow, for both mm and arch developers, and easier to misuse. [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Acked-by: Carsten Otte <cotte@de.ibm.com> Cc: Jared Hulbert <jaredeh@gmail.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 09:13:00 +00:00
* And for normal mappings this is false.
*
* This restricts such mappings to be a linear translation from virtual address
* to pfn. To get around this restriction, we allow arbitrary mappings so long
* as the vma is not a COW mapping; in that case, we know that all ptes are
* special (because none can have been COWed).
mm: introduce VM_MIXEDMAP This series introduces some important infrastructure work. The overall result is that: 1. We now support XIP backed filesystems using memory that have no struct page allocated to them. And patches 6 and 7 actually implement this for s390. This is pretty important in a number of cases. As far as I understand, in the case of virtualisation (eg. s390), each guest may mount a readonly copy of the same filesystem (eg. the distro). Currently, guests need to allocate struct pages for this image. So if you have 100 guests, you already need to allocate more memory for the struct pages than the size of the image. I think. (Carsten?) For other (eg. embedded) systems, you may have a very large non- volatile filesystem. If you have to have struct pages for this, then your RAM consumption will go up proportionally to fs size. Even though it is just a small proportion, the RAM can be much more costly eg in terms of power, so every KB less that Linux uses makes it more attractive to a lot of these guys. 2. VM_MIXEDMAP allows us to support mappings where you actually do want to refcount _some_ pages in the mapping, but not others, and support COW on arbitrary (non-linear) mappings. Jared needs this for his NVRAM filesystem in progress. Future iterations of this filesystem will most likely want to migrate pages between pagecache and XIP backing, which is where the requirement for mixed (some refcounted, some not) comes from. 3. pte_special also has a peripheral usage that I need for my lockless get_user_pages patch. That was shown to speed up "oltp" on db2 by 10% on a 2 socket system, which is kind of significant because they scrounge for months to try to find 0.1% improvement on these workloads. I'm hoping we might finally be faster than AIX on pSeries with this :). My reference to lockless get_user_pages is not meant to justify this patchset (which doesn't include lockless gup), but just to show that pte_special is not some s390 specific thing that should be hidden in arch code or xip code: I definitely want to use it on at least x86 and powerpc as well. This patch: Introduce a new type of mapping, VM_MIXEDMAP. This is unlike VM_PFNMAP in that it can support COW mappings of arbitrary ranges including ranges without struct page *and* ranges with a struct page that we actually want to refcount (PFNMAP can only support COW in those cases where the un-COW-ed translations are mapped linearly in the virtual address, and can only support non refcounted ranges). VM_MIXEDMAP achieves this by refcounting all pfn_valid pages, and not refcounting !pfn_valid pages (which is not an option for VM_PFNMAP, because it needs to avoid refcounting pfn_valid pages eg. for /dev/mem mappings). Signed-off-by: Jared Hulbert <jaredeh@gmail.com> Signed-off-by: Nick Piggin <npiggin@suse.de> Acked-by: Carsten Otte <cotte@de.ibm.com> Cc: Jared Hulbert <jaredeh@gmail.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 09:12:58 +00:00
*
*
mm: introduce pte_special pte bit s390 for one, cannot implement VM_MIXEDMAP with pfn_valid, due to their memory model (which is more dynamic than most). Instead, they had proposed to implement it with an additional path through vm_normal_page(), using a bit in the pte to determine whether or not the page should be refcounted: vm_normal_page() { ... if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { if (vma->vm_flags & VM_MIXEDMAP) { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; #else if (!pfn_valid(pfn)) return NULL; #endif goto out; } ... } This is fine, however if we are allowed to use a bit in the pte to determine refcountedness, we can use that to _completely_ replace all the vma based schemes. So instead of adding more cases to the already complex vma-based scheme, we can have a clearly seperate and simple pte-based scheme (and get slightly better code generation in the process): vm_normal_page() { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; return pte_page(pte); #else ... #endif } And finally, we may rather make this concept usable by any architecture rather than making it s390 only, so implement a new type of pte state for this. Unfortunately the old vma based code must stay, because some architectures may not be able to spare pte bits. This makes vm_normal_page a little bit more ugly than we would like, but the 2 cases are clearly seperate. So introduce a pte_special pte state, and use it in mm/memory.c. It is currently a noop for all architectures, so this doesn't actually result in any compiled code changes to mm/memory.o. BTW: I haven't put vm_normal_page() into arch code as-per an earlier suggestion. The reason is that, regardless of where vm_normal_page is actually implemented, the *abstraction* is still exactly the same. Also, while it depends on whether the architecture has pte_special or not, that is the only two possible cases, and it really isn't an arch specific function -- the role of the arch code should be to provide primitive functions and accessors with which to build the core code; pte_special does that. We do not want architectures to know or care about vm_normal_page itself, and we definitely don't want them being able to invent something new there out of sight of mm/ code. If we made vm_normal_page an arch function, then we have to make vm_insert_mixed (next patch) an arch function too. So I don't think moving it to arch code fundamentally improves any abstractions, while it does practically make the code more difficult to follow, for both mm and arch developers, and easier to misuse. [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Acked-by: Carsten Otte <cotte@de.ibm.com> Cc: Jared Hulbert <jaredeh@gmail.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 09:13:00 +00:00
* In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
mm: introduce VM_MIXEDMAP This series introduces some important infrastructure work. The overall result is that: 1. We now support XIP backed filesystems using memory that have no struct page allocated to them. And patches 6 and 7 actually implement this for s390. This is pretty important in a number of cases. As far as I understand, in the case of virtualisation (eg. s390), each guest may mount a readonly copy of the same filesystem (eg. the distro). Currently, guests need to allocate struct pages for this image. So if you have 100 guests, you already need to allocate more memory for the struct pages than the size of the image. I think. (Carsten?) For other (eg. embedded) systems, you may have a very large non- volatile filesystem. If you have to have struct pages for this, then your RAM consumption will go up proportionally to fs size. Even though it is just a small proportion, the RAM can be much more costly eg in terms of power, so every KB less that Linux uses makes it more attractive to a lot of these guys. 2. VM_MIXEDMAP allows us to support mappings where you actually do want to refcount _some_ pages in the mapping, but not others, and support COW on arbitrary (non-linear) mappings. Jared needs this for his NVRAM filesystem in progress. Future iterations of this filesystem will most likely want to migrate pages between pagecache and XIP backing, which is where the requirement for mixed (some refcounted, some not) comes from. 3. pte_special also has a peripheral usage that I need for my lockless get_user_pages patch. That was shown to speed up "oltp" on db2 by 10% on a 2 socket system, which is kind of significant because they scrounge for months to try to find 0.1% improvement on these workloads. I'm hoping we might finally be faster than AIX on pSeries with this :). My reference to lockless get_user_pages is not meant to justify this patchset (which doesn't include lockless gup), but just to show that pte_special is not some s390 specific thing that should be hidden in arch code or xip code: I definitely want to use it on at least x86 and powerpc as well. This patch: Introduce a new type of mapping, VM_MIXEDMAP. This is unlike VM_PFNMAP in that it can support COW mappings of arbitrary ranges including ranges without struct page *and* ranges with a struct page that we actually want to refcount (PFNMAP can only support COW in those cases where the un-COW-ed translations are mapped linearly in the virtual address, and can only support non refcounted ranges). VM_MIXEDMAP achieves this by refcounting all pfn_valid pages, and not refcounting !pfn_valid pages (which is not an option for VM_PFNMAP, because it needs to avoid refcounting pfn_valid pages eg. for /dev/mem mappings). Signed-off-by: Jared Hulbert <jaredeh@gmail.com> Signed-off-by: Nick Piggin <npiggin@suse.de> Acked-by: Carsten Otte <cotte@de.ibm.com> Cc: Jared Hulbert <jaredeh@gmail.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 09:12:58 +00:00
*
* VM_MIXEDMAP mappings can likewise contain memory with or without "struct
* page" backing, however the difference is that _all_ pages with a struct
* page (that is, those where pfn_valid is true) are refcounted and considered
* normal pages by the VM. The disadvantage is that pages are refcounted
* (which can be slower and simply not an option for some PFNMAP users). The
* advantage is that we don't have to follow the strict linearity rule of
* PFNMAP mappings in order to support COWable mappings.
*
*/
mm: introduce pte_special pte bit s390 for one, cannot implement VM_MIXEDMAP with pfn_valid, due to their memory model (which is more dynamic than most). Instead, they had proposed to implement it with an additional path through vm_normal_page(), using a bit in the pte to determine whether or not the page should be refcounted: vm_normal_page() { ... if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { if (vma->vm_flags & VM_MIXEDMAP) { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; #else if (!pfn_valid(pfn)) return NULL; #endif goto out; } ... } This is fine, however if we are allowed to use a bit in the pte to determine refcountedness, we can use that to _completely_ replace all the vma based schemes. So instead of adding more cases to the already complex vma-based scheme, we can have a clearly seperate and simple pte-based scheme (and get slightly better code generation in the process): vm_normal_page() { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; return pte_page(pte); #else ... #endif } And finally, we may rather make this concept usable by any architecture rather than making it s390 only, so implement a new type of pte state for this. Unfortunately the old vma based code must stay, because some architectures may not be able to spare pte bits. This makes vm_normal_page a little bit more ugly than we would like, but the 2 cases are clearly seperate. So introduce a pte_special pte state, and use it in mm/memory.c. It is currently a noop for all architectures, so this doesn't actually result in any compiled code changes to mm/memory.o. BTW: I haven't put vm_normal_page() into arch code as-per an earlier suggestion. The reason is that, regardless of where vm_normal_page is actually implemented, the *abstraction* is still exactly the same. Also, while it depends on whether the architecture has pte_special or not, that is the only two possible cases, and it really isn't an arch specific function -- the role of the arch code should be to provide primitive functions and accessors with which to build the core code; pte_special does that. We do not want architectures to know or care about vm_normal_page itself, and we definitely don't want them being able to invent something new there out of sight of mm/ code. If we made vm_normal_page an arch function, then we have to make vm_insert_mixed (next patch) an arch function too. So I don't think moving it to arch code fundamentally improves any abstractions, while it does practically make the code more difficult to follow, for both mm and arch developers, and easier to misuse. [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Acked-by: Carsten Otte <cotte@de.ibm.com> Cc: Jared Hulbert <jaredeh@gmail.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 09:13:00 +00:00
#ifdef __HAVE_ARCH_PTE_SPECIAL
# define HAVE_PTE_SPECIAL 1
#else
# define HAVE_PTE_SPECIAL 0
#endif
struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
pte_t pte)
{
unsigned long pfn = pte_pfn(pte);
mm: introduce pte_special pte bit s390 for one, cannot implement VM_MIXEDMAP with pfn_valid, due to their memory model (which is more dynamic than most). Instead, they had proposed to implement it with an additional path through vm_normal_page(), using a bit in the pte to determine whether or not the page should be refcounted: vm_normal_page() { ... if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { if (vma->vm_flags & VM_MIXEDMAP) { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; #else if (!pfn_valid(pfn)) return NULL; #endif goto out; } ... } This is fine, however if we are allowed to use a bit in the pte to determine refcountedness, we can use that to _completely_ replace all the vma based schemes. So instead of adding more cases to the already complex vma-based scheme, we can have a clearly seperate and simple pte-based scheme (and get slightly better code generation in the process): vm_normal_page() { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; return pte_page(pte); #else ... #endif } And finally, we may rather make this concept usable by any architecture rather than making it s390 only, so implement a new type of pte state for this. Unfortunately the old vma based code must stay, because some architectures may not be able to spare pte bits. This makes vm_normal_page a little bit more ugly than we would like, but the 2 cases are clearly seperate. So introduce a pte_special pte state, and use it in mm/memory.c. It is currently a noop for all architectures, so this doesn't actually result in any compiled code changes to mm/memory.o. BTW: I haven't put vm_normal_page() into arch code as-per an earlier suggestion. The reason is that, regardless of where vm_normal_page is actually implemented, the *abstraction* is still exactly the same. Also, while it depends on whether the architecture has pte_special or not, that is the only two possible cases, and it really isn't an arch specific function -- the role of the arch code should be to provide primitive functions and accessors with which to build the core code; pte_special does that. We do not want architectures to know or care about vm_normal_page itself, and we definitely don't want them being able to invent something new there out of sight of mm/ code. If we made vm_normal_page an arch function, then we have to make vm_insert_mixed (next patch) an arch function too. So I don't think moving it to arch code fundamentally improves any abstractions, while it does practically make the code more difficult to follow, for both mm and arch developers, and easier to misuse. [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Acked-by: Carsten Otte <cotte@de.ibm.com> Cc: Jared Hulbert <jaredeh@gmail.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 09:13:00 +00:00
if (HAVE_PTE_SPECIAL) {
if (likely(!pte_special(pte)))
goto check_pfn;
mm: reinstate ZERO_PAGE KAMEZAWA Hiroyuki has observed customers of earlier kernels taking advantage of the ZERO_PAGE: which we stopped do_anonymous_page() from using in 2.6.24. And there were a couple of regression reports on LKML. Following suggestions from Linus, reinstate do_anonymous_page() use of the ZERO_PAGE; but this time avoid dirtying its struct page cacheline with (map)count updates - let vm_normal_page() regard it as abnormal. Use it only on arches which __HAVE_ARCH_PTE_SPECIAL (x86, s390, sh32, most powerpc): that's not essential, but minimizes additional branches (keeping them in the unlikely pte_special case); and incidentally excludes mips (some models of which needed eight colours of ZERO_PAGE to avoid costly exceptions). Don't be fanatical about avoiding ZERO_PAGE updates: get_user_pages() callers won't want to make exceptions for it, so increment its count there. Changes to mlock and migration? happily seems not needed. In most places it's quicker to check pfn than struct page address: prepare a __read_mostly zero_pfn for that. Does get_dump_page() still need its ZERO_PAGE check? probably not, but keep it anyway. Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk> Acked-by: Rik van Riel <riel@redhat.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Minchan Kim <minchan.kim@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-09-22 00:03:30 +00:00
if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))
return NULL;
if (!is_zero_pfn(pfn))
print_bad_pte(vma, addr, pte, NULL);
mm: introduce pte_special pte bit s390 for one, cannot implement VM_MIXEDMAP with pfn_valid, due to their memory model (which is more dynamic than most). Instead, they had proposed to implement it with an additional path through vm_normal_page(), using a bit in the pte to determine whether or not the page should be refcounted: vm_normal_page() { ... if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { if (vma->vm_flags & VM_MIXEDMAP) { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; #else if (!pfn_valid(pfn)) return NULL; #endif goto out; } ... } This is fine, however if we are allowed to use a bit in the pte to determine refcountedness, we can use that to _completely_ replace all the vma based schemes. So instead of adding more cases to the already complex vma-based scheme, we can have a clearly seperate and simple pte-based scheme (and get slightly better code generation in the process): vm_normal_page() { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; return pte_page(pte); #else ... #endif } And finally, we may rather make this concept usable by any architecture rather than making it s390 only, so implement a new type of pte state for this. Unfortunately the old vma based code must stay, because some architectures may not be able to spare pte bits. This makes vm_normal_page a little bit more ugly than we would like, but the 2 cases are clearly seperate. So introduce a pte_special pte state, and use it in mm/memory.c. It is currently a noop for all architectures, so this doesn't actually result in any compiled code changes to mm/memory.o. BTW: I haven't put vm_normal_page() into arch code as-per an earlier suggestion. The reason is that, regardless of where vm_normal_page is actually implemented, the *abstraction* is still exactly the same. Also, while it depends on whether the architecture has pte_special or not, that is the only two possible cases, and it really isn't an arch specific function -- the role of the arch code should be to provide primitive functions and accessors with which to build the core code; pte_special does that. We do not want architectures to know or care about vm_normal_page itself, and we definitely don't want them being able to invent something new there out of sight of mm/ code. If we made vm_normal_page an arch function, then we have to make vm_insert_mixed (next patch) an arch function too. So I don't think moving it to arch code fundamentally improves any abstractions, while it does practically make the code more difficult to follow, for both mm and arch developers, and easier to misuse. [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Acked-by: Carsten Otte <cotte@de.ibm.com> Cc: Jared Hulbert <jaredeh@gmail.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 09:13:00 +00:00
return NULL;
}
/* !HAVE_PTE_SPECIAL case follows: */
mm: introduce VM_MIXEDMAP This series introduces some important infrastructure work. The overall result is that: 1. We now support XIP backed filesystems using memory that have no struct page allocated to them. And patches 6 and 7 actually implement this for s390. This is pretty important in a number of cases. As far as I understand, in the case of virtualisation (eg. s390), each guest may mount a readonly copy of the same filesystem (eg. the distro). Currently, guests need to allocate struct pages for this image. So if you have 100 guests, you already need to allocate more memory for the struct pages than the size of the image. I think. (Carsten?) For other (eg. embedded) systems, you may have a very large non- volatile filesystem. If you have to have struct pages for this, then your RAM consumption will go up proportionally to fs size. Even though it is just a small proportion, the RAM can be much more costly eg in terms of power, so every KB less that Linux uses makes it more attractive to a lot of these guys. 2. VM_MIXEDMAP allows us to support mappings where you actually do want to refcount _some_ pages in the mapping, but not others, and support COW on arbitrary (non-linear) mappings. Jared needs this for his NVRAM filesystem in progress. Future iterations of this filesystem will most likely want to migrate pages between pagecache and XIP backing, which is where the requirement for mixed (some refcounted, some not) comes from. 3. pte_special also has a peripheral usage that I need for my lockless get_user_pages patch. That was shown to speed up "oltp" on db2 by 10% on a 2 socket system, which is kind of significant because they scrounge for months to try to find 0.1% improvement on these workloads. I'm hoping we might finally be faster than AIX on pSeries with this :). My reference to lockless get_user_pages is not meant to justify this patchset (which doesn't include lockless gup), but just to show that pte_special is not some s390 specific thing that should be hidden in arch code or xip code: I definitely want to use it on at least x86 and powerpc as well. This patch: Introduce a new type of mapping, VM_MIXEDMAP. This is unlike VM_PFNMAP in that it can support COW mappings of arbitrary ranges including ranges without struct page *and* ranges with a struct page that we actually want to refcount (PFNMAP can only support COW in those cases where the un-COW-ed translations are mapped linearly in the virtual address, and can only support non refcounted ranges). VM_MIXEDMAP achieves this by refcounting all pfn_valid pages, and not refcounting !pfn_valid pages (which is not an option for VM_PFNMAP, because it needs to avoid refcounting pfn_valid pages eg. for /dev/mem mappings). Signed-off-by: Jared Hulbert <jaredeh@gmail.com> Signed-off-by: Nick Piggin <npiggin@suse.de> Acked-by: Carsten Otte <cotte@de.ibm.com> Cc: Jared Hulbert <jaredeh@gmail.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 09:12:58 +00:00
if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
if (vma->vm_flags & VM_MIXEDMAP) {
if (!pfn_valid(pfn))
return NULL;
goto out;
} else {
mm: introduce pte_special pte bit s390 for one, cannot implement VM_MIXEDMAP with pfn_valid, due to their memory model (which is more dynamic than most). Instead, they had proposed to implement it with an additional path through vm_normal_page(), using a bit in the pte to determine whether or not the page should be refcounted: vm_normal_page() { ... if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { if (vma->vm_flags & VM_MIXEDMAP) { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; #else if (!pfn_valid(pfn)) return NULL; #endif goto out; } ... } This is fine, however if we are allowed to use a bit in the pte to determine refcountedness, we can use that to _completely_ replace all the vma based schemes. So instead of adding more cases to the already complex vma-based scheme, we can have a clearly seperate and simple pte-based scheme (and get slightly better code generation in the process): vm_normal_page() { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; return pte_page(pte); #else ... #endif } And finally, we may rather make this concept usable by any architecture rather than making it s390 only, so implement a new type of pte state for this. Unfortunately the old vma based code must stay, because some architectures may not be able to spare pte bits. This makes vm_normal_page a little bit more ugly than we would like, but the 2 cases are clearly seperate. So introduce a pte_special pte state, and use it in mm/memory.c. It is currently a noop for all architectures, so this doesn't actually result in any compiled code changes to mm/memory.o. BTW: I haven't put vm_normal_page() into arch code as-per an earlier suggestion. The reason is that, regardless of where vm_normal_page is actually implemented, the *abstraction* is still exactly the same. Also, while it depends on whether the architecture has pte_special or not, that is the only two possible cases, and it really isn't an arch specific function -- the role of the arch code should be to provide primitive functions and accessors with which to build the core code; pte_special does that. We do not want architectures to know or care about vm_normal_page itself, and we definitely don't want them being able to invent something new there out of sight of mm/ code. If we made vm_normal_page an arch function, then we have to make vm_insert_mixed (next patch) an arch function too. So I don't think moving it to arch code fundamentally improves any abstractions, while it does practically make the code more difficult to follow, for both mm and arch developers, and easier to misuse. [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Acked-by: Carsten Otte <cotte@de.ibm.com> Cc: Jared Hulbert <jaredeh@gmail.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 09:13:00 +00:00
unsigned long off;
off = (addr - vma->vm_start) >> PAGE_SHIFT;
mm: introduce VM_MIXEDMAP This series introduces some important infrastructure work. The overall result is that: 1. We now support XIP backed filesystems using memory that have no struct page allocated to them. And patches 6 and 7 actually implement this for s390. This is pretty important in a number of cases. As far as I understand, in the case of virtualisation (eg. s390), each guest may mount a readonly copy of the same filesystem (eg. the distro). Currently, guests need to allocate struct pages for this image. So if you have 100 guests, you already need to allocate more memory for the struct pages than the size of the image. I think. (Carsten?) For other (eg. embedded) systems, you may have a very large non- volatile filesystem. If you have to have struct pages for this, then your RAM consumption will go up proportionally to fs size. Even though it is just a small proportion, the RAM can be much more costly eg in terms of power, so every KB less that Linux uses makes it more attractive to a lot of these guys. 2. VM_MIXEDMAP allows us to support mappings where you actually do want to refcount _some_ pages in the mapping, but not others, and support COW on arbitrary (non-linear) mappings. Jared needs this for his NVRAM filesystem in progress. Future iterations of this filesystem will most likely want to migrate pages between pagecache and XIP backing, which is where the requirement for mixed (some refcounted, some not) comes from. 3. pte_special also has a peripheral usage that I need for my lockless get_user_pages patch. That was shown to speed up "oltp" on db2 by 10% on a 2 socket system, which is kind of significant because they scrounge for months to try to find 0.1% improvement on these workloads. I'm hoping we might finally be faster than AIX on pSeries with this :). My reference to lockless get_user_pages is not meant to justify this patchset (which doesn't include lockless gup), but just to show that pte_special is not some s390 specific thing that should be hidden in arch code or xip code: I definitely want to use it on at least x86 and powerpc as well. This patch: Introduce a new type of mapping, VM_MIXEDMAP. This is unlike VM_PFNMAP in that it can support COW mappings of arbitrary ranges including ranges without struct page *and* ranges with a struct page that we actually want to refcount (PFNMAP can only support COW in those cases where the un-COW-ed translations are mapped linearly in the virtual address, and can only support non refcounted ranges). VM_MIXEDMAP achieves this by refcounting all pfn_valid pages, and not refcounting !pfn_valid pages (which is not an option for VM_PFNMAP, because it needs to avoid refcounting pfn_valid pages eg. for /dev/mem mappings). Signed-off-by: Jared Hulbert <jaredeh@gmail.com> Signed-off-by: Nick Piggin <npiggin@suse.de> Acked-by: Carsten Otte <cotte@de.ibm.com> Cc: Jared Hulbert <jaredeh@gmail.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 09:12:58 +00:00
if (pfn == vma->vm_pgoff + off)
return NULL;
if (!is_cow_mapping(vma->vm_flags))
return NULL;
}
}
if (is_zero_pfn(pfn))
return NULL;
check_pfn:
if (unlikely(pfn > highest_memmap_pfn)) {
print_bad_pte(vma, addr, pte, NULL);
return NULL;
}
/*
mm: introduce pte_special pte bit s390 for one, cannot implement VM_MIXEDMAP with pfn_valid, due to their memory model (which is more dynamic than most). Instead, they had proposed to implement it with an additional path through vm_normal_page(), using a bit in the pte to determine whether or not the page should be refcounted: vm_normal_page() { ... if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { if (vma->vm_flags & VM_MIXEDMAP) { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; #else if (!pfn_valid(pfn)) return NULL; #endif goto out; } ... } This is fine, however if we are allowed to use a bit in the pte to determine refcountedness, we can use that to _completely_ replace all the vma based schemes. So instead of adding more cases to the already complex vma-based scheme, we can have a clearly seperate and simple pte-based scheme (and get slightly better code generation in the process): vm_normal_page() { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; return pte_page(pte); #else ... #endif } And finally, we may rather make this concept usable by any architecture rather than making it s390 only, so implement a new type of pte state for this. Unfortunately the old vma based code must stay, because some architectures may not be able to spare pte bits. This makes vm_normal_page a little bit more ugly than we would like, but the 2 cases are clearly seperate. So introduce a pte_special pte state, and use it in mm/memory.c. It is currently a noop for all architectures, so this doesn't actually result in any compiled code changes to mm/memory.o. BTW: I haven't put vm_normal_page() into arch code as-per an earlier suggestion. The reason is that, regardless of where vm_normal_page is actually implemented, the *abstraction* is still exactly the same. Also, while it depends on whether the architecture has pte_special or not, that is the only two possible cases, and it really isn't an arch specific function -- the role of the arch code should be to provide primitive functions and accessors with which to build the core code; pte_special does that. We do not want architectures to know or care about vm_normal_page itself, and we definitely don't want them being able to invent something new there out of sight of mm/ code. If we made vm_normal_page an arch function, then we have to make vm_insert_mixed (next patch) an arch function too. So I don't think moving it to arch code fundamentally improves any abstractions, while it does practically make the code more difficult to follow, for both mm and arch developers, and easier to misuse. [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Acked-by: Carsten Otte <cotte@de.ibm.com> Cc: Jared Hulbert <jaredeh@gmail.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 09:13:00 +00:00
* NOTE! We still have PageReserved() pages in the page tables.
* eg. VDSO mappings can cause them to exist.
*/
mm: introduce VM_MIXEDMAP This series introduces some important infrastructure work. The overall result is that: 1. We now support XIP backed filesystems using memory that have no struct page allocated to them. And patches 6 and 7 actually implement this for s390. This is pretty important in a number of cases. As far as I understand, in the case of virtualisation (eg. s390), each guest may mount a readonly copy of the same filesystem (eg. the distro). Currently, guests need to allocate struct pages for this image. So if you have 100 guests, you already need to allocate more memory for the struct pages than the size of the image. I think. (Carsten?) For other (eg. embedded) systems, you may have a very large non- volatile filesystem. If you have to have struct pages for this, then your RAM consumption will go up proportionally to fs size. Even though it is just a small proportion, the RAM can be much more costly eg in terms of power, so every KB less that Linux uses makes it more attractive to a lot of these guys. 2. VM_MIXEDMAP allows us to support mappings where you actually do want to refcount _some_ pages in the mapping, but not others, and support COW on arbitrary (non-linear) mappings. Jared needs this for his NVRAM filesystem in progress. Future iterations of this filesystem will most likely want to migrate pages between pagecache and XIP backing, which is where the requirement for mixed (some refcounted, some not) comes from. 3. pte_special also has a peripheral usage that I need for my lockless get_user_pages patch. That was shown to speed up "oltp" on db2 by 10% on a 2 socket system, which is kind of significant because they scrounge for months to try to find 0.1% improvement on these workloads. I'm hoping we might finally be faster than AIX on pSeries with this :). My reference to lockless get_user_pages is not meant to justify this patchset (which doesn't include lockless gup), but just to show that pte_special is not some s390 specific thing that should be hidden in arch code or xip code: I definitely want to use it on at least x86 and powerpc as well. This patch: Introduce a new type of mapping, VM_MIXEDMAP. This is unlike VM_PFNMAP in that it can support COW mappings of arbitrary ranges including ranges without struct page *and* ranges with a struct page that we actually want to refcount (PFNMAP can only support COW in those cases where the un-COW-ed translations are mapped linearly in the virtual address, and can only support non refcounted ranges). VM_MIXEDMAP achieves this by refcounting all pfn_valid pages, and not refcounting !pfn_valid pages (which is not an option for VM_PFNMAP, because it needs to avoid refcounting pfn_valid pages eg. for /dev/mem mappings). Signed-off-by: Jared Hulbert <jaredeh@gmail.com> Signed-off-by: Nick Piggin <npiggin@suse.de> Acked-by: Carsten Otte <cotte@de.ibm.com> Cc: Jared Hulbert <jaredeh@gmail.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 09:12:58 +00:00
out:
return pfn_to_page(pfn);
}
/*
* copy one vm_area from one task to the other. Assumes the page tables
* already present in the new task to be cleared in the whole range
* covered by this vma.
*/
swap_info: swap count continuations Swap is duplicated (reference count incremented by one) whenever the same swap page is inserted into another mm (when forking finds a swap entry in place of a pte, or when reclaim unmaps a pte to insert the swap entry). swap_info_struct's vmalloc'ed swap_map is the array of these reference counts: but what happens when the unsigned short (or unsigned char since the preceding patch) is full? (and its high bit is kept for a cache flag) We then lose track of it, never freeing, leaving it in use until swapoff: at which point we _hope_ that a single pass will have found all instances, assume there are no more, and will lose user data if we're wrong. Swapping of KSM pages has not yet been enabled; but it is implemented, and makes it very easy for a user to overflow the maximum swap count: possible with ordinary process pages, but unlikely, even when pid_max has been raised from PID_MAX_DEFAULT. This patch implements swap count continuations: when the count overflows, a continuation page is allocated and linked to the original vmalloc'ed map page, and this used to hold the continuation counts for that entry and its neighbours. These continuation pages are seldom referenced: the common paths all work on the original swap_map, only referring to a continuation page when the low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-15 01:58:46 +00:00
static inline unsigned long
copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
2005-10-30 01:16:12 +00:00
pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
unsigned long addr, int *rss)
{
2005-10-30 01:16:12 +00:00
unsigned long vm_flags = vma->vm_flags;
pte_t pte = *src_pte;
struct page *page;
/* pte contains position in swap or file, so copy. */
if (unlikely(!pte_present(pte))) {
if (!pte_file(pte)) {
[PATCH] Swapless page migration: add R/W migration entries Implement read/write migration ptes We take the upper two swapfiles for the two types of migration ptes and define a series of macros in swapops.h. The VM is modified to handle the migration entries. migration entries can only be encountered when the page they are pointing to is locked. This limits the number of places one has to fix. We also check in copy_pte_range and in mprotect_pte_range() for migration ptes. We check for migration ptes in do_swap_cache and call a function that will then wait on the page lock. This allows us to effectively stop all accesses to apge. Migration entries are created by try_to_unmap if called for migration and removed by local functions in migrate.c From: Hugh Dickins <hugh@veritas.com> Several times while testing swapless page migration (I've no NUMA, just hacking it up to migrate recklessly while running load), I've hit the BUG_ON(!PageLocked(p)) in migration_entry_to_page. This comes from an orphaned migration entry, unrelated to the current correctly locked migration, but hit by remove_anon_migration_ptes as it checks an address in each vma of the anon_vma list. Such an orphan may be left behind if an earlier migration raced with fork: copy_one_pte can duplicate a migration entry from parent to child, after remove_anon_migration_ptes has checked the child vma, but before it has removed it from the parent vma. (If the process were later to fault on this orphaned entry, it would hit the same BUG from migration_entry_wait.) This could be fixed by locking anon_vma in copy_one_pte, but we'd rather not. There's no such problem with file pages, because vma_prio_tree_add adds child vma after parent vma, and the page table locking at each end is enough to serialize. Follow that example with anon_vma: add new vmas to the tail instead of the head. (There's no corresponding problem when inserting migration entries, because a missed pte will leave the page count and mapcount high, which is allowed for. And there's no corresponding problem when migrating via swap, because a leftover swap entry will be correctly faulted. But the swapless method has no refcounting of its entries.) From: Ingo Molnar <mingo@elte.hu> pte_unmap_unlock() takes the pte pointer as an argument. From: Hugh Dickins <hugh@veritas.com> Several times while testing swapless page migration, gcc has tried to exec a pointer instead of a string: smells like COW mappings are not being properly write-protected on fork. The protection in copy_one_pte looks very convincing, until at last you realize that the second arg to make_migration_entry is a boolean "write", and SWP_MIGRATION_READ is 30. Anyway, it's better done like in change_pte_range, using is_write_migration_entry and make_migration_entry_read. From: Hugh Dickins <hugh@veritas.com> Remove unnecessary obfuscation from sys_swapon's range check on swap type, which blew up causing memory corruption once swapless migration made MAX_SWAPFILES no longer 2 ^ MAX_SWAPFILES_SHIFT. Signed-off-by: Hugh Dickins <hugh@veritas.com> Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Christoph Lameter <clameter@engr.sgi.com> Signed-off-by: Ingo Molnar <mingo@elte.hu> From: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 09:03:35 +00:00
swp_entry_t entry = pte_to_swp_entry(pte);
swap_info: swap count continuations Swap is duplicated (reference count incremented by one) whenever the same swap page is inserted into another mm (when forking finds a swap entry in place of a pte, or when reclaim unmaps a pte to insert the swap entry). swap_info_struct's vmalloc'ed swap_map is the array of these reference counts: but what happens when the unsigned short (or unsigned char since the preceding patch) is full? (and its high bit is kept for a cache flag) We then lose track of it, never freeing, leaving it in use until swapoff: at which point we _hope_ that a single pass will have found all instances, assume there are no more, and will lose user data if we're wrong. Swapping of KSM pages has not yet been enabled; but it is implemented, and makes it very easy for a user to overflow the maximum swap count: possible with ordinary process pages, but unlikely, even when pid_max has been raised from PID_MAX_DEFAULT. This patch implements swap count continuations: when the count overflows, a continuation page is allocated and linked to the original vmalloc'ed map page, and this used to hold the continuation counts for that entry and its neighbours. These continuation pages are seldom referenced: the common paths all work on the original swap_map, only referring to a continuation page when the low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-15 01:58:46 +00:00
if (swap_duplicate(entry) < 0)
return entry.val;
/* make sure dst_mm is on swapoff's mmlist. */
if (unlikely(list_empty(&dst_mm->mmlist))) {
spin_lock(&mmlist_lock);
if (list_empty(&dst_mm->mmlist))
list_add(&dst_mm->mmlist,
&src_mm->mmlist);
spin_unlock(&mmlist_lock);
}
if (likely(!non_swap_entry(entry)))
rss[MM_SWAPENTS]++;
else if (is_migration_entry(entry)) {
page = migration_entry_to_page(entry);
if (PageAnon(page))
rss[MM_ANONPAGES]++;
else
rss[MM_FILEPAGES]++;
if (is_write_migration_entry(entry) &&
is_cow_mapping(vm_flags)) {
/*
* COW mappings require pages in both
* parent and child to be set to read.
*/
make_migration_entry_read(&entry);
pte = swp_entry_to_pte(entry);
set_pte_at(src_mm, addr, src_pte, pte);
}
[PATCH] Swapless page migration: add R/W migration entries Implement read/write migration ptes We take the upper two swapfiles for the two types of migration ptes and define a series of macros in swapops.h. The VM is modified to handle the migration entries. migration entries can only be encountered when the page they are pointing to is locked. This limits the number of places one has to fix. We also check in copy_pte_range and in mprotect_pte_range() for migration ptes. We check for migration ptes in do_swap_cache and call a function that will then wait on the page lock. This allows us to effectively stop all accesses to apge. Migration entries are created by try_to_unmap if called for migration and removed by local functions in migrate.c From: Hugh Dickins <hugh@veritas.com> Several times while testing swapless page migration (I've no NUMA, just hacking it up to migrate recklessly while running load), I've hit the BUG_ON(!PageLocked(p)) in migration_entry_to_page. This comes from an orphaned migration entry, unrelated to the current correctly locked migration, but hit by remove_anon_migration_ptes as it checks an address in each vma of the anon_vma list. Such an orphan may be left behind if an earlier migration raced with fork: copy_one_pte can duplicate a migration entry from parent to child, after remove_anon_migration_ptes has checked the child vma, but before it has removed it from the parent vma. (If the process were later to fault on this orphaned entry, it would hit the same BUG from migration_entry_wait.) This could be fixed by locking anon_vma in copy_one_pte, but we'd rather not. There's no such problem with file pages, because vma_prio_tree_add adds child vma after parent vma, and the page table locking at each end is enough to serialize. Follow that example with anon_vma: add new vmas to the tail instead of the head. (There's no corresponding problem when inserting migration entries, because a missed pte will leave the page count and mapcount high, which is allowed for. And there's no corresponding problem when migrating via swap, because a leftover swap entry will be correctly faulted. But the swapless method has no refcounting of its entries.) From: Ingo Molnar <mingo@elte.hu> pte_unmap_unlock() takes the pte pointer as an argument. From: Hugh Dickins <hugh@veritas.com> Several times while testing swapless page migration, gcc has tried to exec a pointer instead of a string: smells like COW mappings are not being properly write-protected on fork. The protection in copy_one_pte looks very convincing, until at last you realize that the second arg to make_migration_entry is a boolean "write", and SWP_MIGRATION_READ is 30. Anyway, it's better done like in change_pte_range, using is_write_migration_entry and make_migration_entry_read. From: Hugh Dickins <hugh@veritas.com> Remove unnecessary obfuscation from sys_swapon's range check on swap type, which blew up causing memory corruption once swapless migration made MAX_SWAPFILES no longer 2 ^ MAX_SWAPFILES_SHIFT. Signed-off-by: Hugh Dickins <hugh@veritas.com> Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Christoph Lameter <clameter@engr.sgi.com> Signed-off-by: Ingo Molnar <mingo@elte.hu> From: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 09:03:35 +00:00
}
}
goto out_set_pte;
}
/*
* If it's a COW mapping, write protect it both
* in the parent and the child
*/
if (is_cow_mapping(vm_flags)) {
ptep_set_wrprotect(src_mm, addr, src_pte);
pte = pte_wrprotect(pte);
}
/*
* If it's a shared mapping, mark it clean in
* the child
*/
if (vm_flags & VM_SHARED)
pte = pte_mkclean(pte);
pte = pte_mkold(pte);
page = vm_normal_page(vma, addr, pte);
if (page) {
get_page(page);
page_dup_rmap(page);
if (PageAnon(page))
rss[MM_ANONPAGES]++;
else
rss[MM_FILEPAGES]++;
}
out_set_pte:
set_pte_at(dst_mm, addr, dst_pte, pte);
swap_info: swap count continuations Swap is duplicated (reference count incremented by one) whenever the same swap page is inserted into another mm (when forking finds a swap entry in place of a pte, or when reclaim unmaps a pte to insert the swap entry). swap_info_struct's vmalloc'ed swap_map is the array of these reference counts: but what happens when the unsigned short (or unsigned char since the preceding patch) is full? (and its high bit is kept for a cache flag) We then lose track of it, never freeing, leaving it in use until swapoff: at which point we _hope_ that a single pass will have found all instances, assume there are no more, and will lose user data if we're wrong. Swapping of KSM pages has not yet been enabled; but it is implemented, and makes it very easy for a user to overflow the maximum swap count: possible with ordinary process pages, but unlikely, even when pid_max has been raised from PID_MAX_DEFAULT. This patch implements swap count continuations: when the count overflows, a continuation page is allocated and linked to the original vmalloc'ed map page, and this used to hold the continuation counts for that entry and its neighbours. These continuation pages are seldom referenced: the common paths all work on the original swap_map, only referring to a continuation page when the low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-15 01:58:46 +00:00
return 0;
}
thp: transparent hugepage core Lately I've been working to make KVM use hugepages transparently without the usual restrictions of hugetlbfs. Some of the restrictions I'd like to see removed: 1) hugepages have to be swappable or the guest physical memory remains locked in RAM and can't be paged out to swap 2) if a hugepage allocation fails, regular pages should be allocated instead and mixed in the same vma without any failure and without userland noticing 3) if some task quits and more hugepages become available in the buddy, guest physical memory backed by regular pages should be relocated on hugepages automatically in regions under madvise(MADV_HUGEPAGE) (ideally event driven by waking up the kernel deamon if the order=HPAGE_PMD_SHIFT-PAGE_SHIFT list becomes not null) 4) avoidance of reservation and maximization of use of hugepages whenever possible. Reservation (needed to avoid runtime fatal faliures) may be ok for 1 machine with 1 database with 1 database cache with 1 database cache size known at boot time. It's definitely not feasible with a virtualization hypervisor usage like RHEV-H that runs an unknown number of virtual machines with an unknown size of each virtual machine with an unknown amount of pagecache that could be potentially useful in the host for guest not using O_DIRECT (aka cache=off). hugepages in the virtualization hypervisor (and also in the guest!) are much more important than in a regular host not using virtualization, becasue with NPT/EPT they decrease the tlb-miss cacheline accesses from 24 to 19 in case only the hypervisor uses transparent hugepages, and they decrease the tlb-miss cacheline accesses from 19 to 15 in case both the linux hypervisor and the linux guest both uses this patch (though the guest will limit the addition speedup to anonymous regions only for now...). Even more important is that the tlb miss handler is much slower on a NPT/EPT guest than for a regular shadow paging or no-virtualization scenario. So maximizing the amount of virtual memory cached by the TLB pays off significantly more with NPT/EPT than without (even if there would be no significant speedup in the tlb-miss runtime). The first (and more tedious) part of this work requires allowing the VM to handle anonymous hugepages mixed with regular pages transparently on regular anonymous vmas. This is what this patch tries to achieve in the least intrusive possible way. We want hugepages and hugetlb to be used in a way so that all applications can benefit without changes (as usual we leverage the KVM virtualization design: by improving the Linux VM at large, KVM gets the performance boost too). The most important design choice is: always fallback to 4k allocation if the hugepage allocation fails! This is the _very_ opposite of some large pagecache patches that failed with -EIO back then if a 64k (or similar) allocation failed... Second important decision (to reduce the impact of the feature on the existing pagetable handling code) is that at any time we can split an hugepage into 512 regular pages and it has to be done with an operation that can't fail. This way the reliability of the swapping isn't decreased (no need to allocate memory when we are short on memory to swap) and it's trivial to plug a split_huge_page* one-liner where needed without polluting the VM. Over time we can teach mprotect, mremap and friends to handle pmd_trans_huge natively without calling split_huge_page*. The fact it can't fail isn't just for swap: if split_huge_page would return -ENOMEM (instead of the current void) we'd need to rollback the mprotect from the middle of it (ideally including undoing the split_vma) which would be a big change and in the very wrong direction (it'd likely be simpler not to call split_huge_page at all and to teach mprotect and friends to handle hugepages instead of rolling them back from the middle). In short the very value of split_huge_page is that it can't fail. The collapsing and madvise(MADV_HUGEPAGE) part will remain separated and incremental and it'll just be an "harmless" addition later if this initial part is agreed upon. It also should be noted that locking-wise replacing regular pages with hugepages is going to be very easy if compared to what I'm doing below in split_huge_page, as it will only happen when page_count(page) matches page_mapcount(page) if we can take the PG_lock and mmap_sem in write mode. collapse_huge_page will be a "best effort" that (unlike split_huge_page) can fail at the minimal sign of trouble and we can try again later. collapse_huge_page will be similar to how KSM works and the madvise(MADV_HUGEPAGE) will work similar to madvise(MADV_MERGEABLE). The default I like is that transparent hugepages are used at page fault time. This can be changed with /sys/kernel/mm/transparent_hugepage/enabled. The control knob can be set to three values "always", "madvise", "never" which mean respectively that hugepages are always used, or only inside madvise(MADV_HUGEPAGE) regions, or never used. /sys/kernel/mm/transparent_hugepage/defrag instead controls if the hugepage allocation should defrag memory aggressively "always", only inside "madvise" regions, or "never". The pmd_trans_splitting/pmd_trans_huge locking is very solid. The put_page (from get_user_page users that can't use mmu notifier like O_DIRECT) that runs against a __split_huge_page_refcount instead was a pain to serialize in a way that would result always in a coherent page count for both tail and head. I think my locking solution with a compound_lock taken only after the page_first is valid and is still a PageHead should be safe but it surely needs review from SMP race point of view. In short there is no current existing way to serialize the O_DIRECT final put_page against split_huge_page_refcount so I had to invent a new one (O_DIRECT loses knowledge on the mapping status by the time gup_fast returns so...). And I didn't want to impact all gup/gup_fast users for now, maybe if we change the gup interface substantially we can avoid this locking, I admit I didn't think too much about it because changing the gup unpinning interface would be invasive. If we ignored O_DIRECT we could stick to the existing compound refcounting code, by simply adding a get_user_pages_fast_flags(foll_flags) where KVM (and any other mmu notifier user) would call it without FOLL_GET (and if FOLL_GET isn't set we'd just BUG_ON if nobody registered itself in the current task mmu notifier list yet). But O_DIRECT is fundamental for decent performance of virtualized I/O on fast storage so we can't avoid it to solve the race of put_page against split_huge_page_refcount to achieve a complete hugepage feature for KVM. Swap and oom works fine (well just like with regular pages ;). MMU notifier is handled transparently too, with the exception of the young bit on the pmd, that didn't have a range check but I think KVM will be fine because the whole point of hugepages is that EPT/NPT will also use a huge pmd when they notice gup returns pages with PageCompound set, so they won't care of a range and there's just the pmd young bit to check in that case. NOTE: in some cases if the L2 cache is small, this may slowdown and waste memory during COWs because 4M of memory are accessed in a single fault instead of 8k (the payoff is that after COW the program can run faster). So we might want to switch the copy_huge_page (and clear_huge_page too) to not temporal stores. I also extensively researched ways to avoid this cache trashing with a full prefault logic that would cow in 8k/16k/32k/64k up to 1M (I can send those patches that fully implemented prefault) but I concluded they're not worth it and they add an huge additional complexity and they remove all tlb benefits until the full hugepage has been faulted in, to save a little bit of memory and some cache during app startup, but they still don't improve substantially the cache-trashing during startup if the prefault happens in >4k chunks. One reason is that those 4k pte entries copied are still mapped on a perfectly cache-colored hugepage, so the trashing is the worst one can generate in those copies (cow of 4k page copies aren't so well colored so they trashes less, but again this results in software running faster after the page fault). Those prefault patches allowed things like a pte where post-cow pages were local 4k regular anon pages and the not-yet-cowed pte entries were pointing in the middle of some hugepage mapped read-only. If it doesn't payoff substantially with todays hardware it will payoff even less in the future with larger l2 caches, and the prefault logic would blot the VM a lot. If one is emebdded transparent_hugepage can be disabled during boot with sysfs or with the boot commandline parameter transparent_hugepage=0 (or transparent_hugepage=2 to restrict hugepages inside madvise regions) that will ensure not a single hugepage is allocated at boot time. It is simple enough to just disable transparent hugepage globally and let transparent hugepages be allocated selectively by applications in the MADV_HUGEPAGE region (both at page fault time, and if enabled with the collapse_huge_page too through the kernel daemon). This patch supports only hugepages mapped in the pmd, archs that have smaller hugepages will not fit in this patch alone. Also some archs like power have certain tlb limits that prevents mixing different page size in the same regions so they will not fit in this framework that requires "graceful fallback" to basic PAGE_SIZE in case of physical memory fragmentation. hugetlbfs remains a perfect fit for those because its software limits happen to match the hardware limits. hugetlbfs also remains a perfect fit for hugepage sizes like 1GByte that cannot be hoped to be found not fragmented after a certain system uptime and that would be very expensive to defragment with relocation, so requiring reservation. hugetlbfs is the "reservation way", the point of transparent hugepages is not to have any reservation at all and maximizing the use of cache and hugepages at all times automatically. Some performance result: vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largep ages3 memset page fault 1566023 memset tlb miss 453854 memset second tlb miss 453321 random access tlb miss 41635 random access second tlb miss 41658 vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largepages3 memset page fault 1566471 memset tlb miss 453375 memset second tlb miss 453320 random access tlb miss 41636 random access second tlb miss 41637 vmx andrea # ./largepages3 memset page fault 1566642 memset tlb miss 453417 memset second tlb miss 453313 random access tlb miss 41630 random access second tlb miss 41647 vmx andrea # ./largepages3 memset page fault 1566872 memset tlb miss 453418 memset second tlb miss 453315 random access tlb miss 41618 random access second tlb miss 41659 vmx andrea # echo 0 > /proc/sys/vm/transparent_hugepage vmx andrea # ./largepages3 memset page fault 2182476 memset tlb miss 460305 memset second tlb miss 460179 random access tlb miss 44483 random access second tlb miss 44186 vmx andrea # ./largepages3 memset page fault 2182791 memset tlb miss 460742 memset second tlb miss 459962 random access tlb miss 43981 random access second tlb miss 43988 ============ #include <stdio.h> #include <stdlib.h> #include <string.h> #include <sys/time.h> #define SIZE (3UL*1024*1024*1024) int main() { char *p = malloc(SIZE), *p2; struct timeval before, after; gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset page fault %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); return 0; } ============ Signed-off-by: Andrea Arcangeli <aarcange@redhat.com> Acked-by: Rik van Riel <riel@redhat.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-13 23:46:52 +00:00
int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
unsigned long addr, unsigned long end)
{
pte_t *orig_src_pte, *orig_dst_pte;
pte_t *src_pte, *dst_pte;
spinlock_t *src_ptl, *dst_ptl;
int progress = 0;
int rss[NR_MM_COUNTERS];
swap_info: swap count continuations Swap is duplicated (reference count incremented by one) whenever the same swap page is inserted into another mm (when forking finds a swap entry in place of a pte, or when reclaim unmaps a pte to insert the swap entry). swap_info_struct's vmalloc'ed swap_map is the array of these reference counts: but what happens when the unsigned short (or unsigned char since the preceding patch) is full? (and its high bit is kept for a cache flag) We then lose track of it, never freeing, leaving it in use until swapoff: at which point we _hope_ that a single pass will have found all instances, assume there are no more, and will lose user data if we're wrong. Swapping of KSM pages has not yet been enabled; but it is implemented, and makes it very easy for a user to overflow the maximum swap count: possible with ordinary process pages, but unlikely, even when pid_max has been raised from PID_MAX_DEFAULT. This patch implements swap count continuations: when the count overflows, a continuation page is allocated and linked to the original vmalloc'ed map page, and this used to hold the continuation counts for that entry and its neighbours. These continuation pages are seldom referenced: the common paths all work on the original swap_map, only referring to a continuation page when the low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-15 01:58:46 +00:00
swp_entry_t entry = (swp_entry_t){0};
again:
init_rss_vec(rss);
dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
if (!dst_pte)
return -ENOMEM;
src_pte = pte_offset_map(src_pmd, addr);
[PATCH] mm: split page table lock Christoph Lameter demonstrated very poor scalability on the SGI 512-way, with a many-threaded application which concurrently initializes different parts of a large anonymous area. This patch corrects that, by using a separate spinlock per page table page, to guard the page table entries in that page, instead of using the mm's single page_table_lock. (But even then, page_table_lock is still used to guard page table allocation, and anon_vma allocation.) In this implementation, the spinlock is tucked inside the struct page of the page table page: with a BUILD_BUG_ON in case it overflows - which it would in the case of 32-bit PA-RISC with spinlock debugging enabled. Splitting the lock is not quite for free: another cacheline access. Ideally, I suppose we would use split ptlock only for multi-threaded processes on multi-cpu machines; but deciding that dynamically would have its own costs. So for now enable it by config, at some number of cpus - since the Kconfig language doesn't support inequalities, let preprocessor compare that with NR_CPUS. But I don't think it's worth being user-configurable: for good testing of both split and unsplit configs, split now at 4 cpus, and perhaps change that to 8 later. There is a benefit even for singly threaded processes: kswapd can be attacking one part of the mm while another part is busy faulting. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:40 +00:00
src_ptl = pte_lockptr(src_mm, src_pmd);
spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
orig_src_pte = src_pte;
orig_dst_pte = dst_pte;
arch_enter_lazy_mmu_mode();
do {
/*
* We are holding two locks at this point - either of them
* could generate latencies in another task on another CPU.
*/
if (progress >= 32) {
progress = 0;
if (need_resched() ||
spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
break;
}
if (pte_none(*src_pte)) {
progress++;
continue;
}
swap_info: swap count continuations Swap is duplicated (reference count incremented by one) whenever the same swap page is inserted into another mm (when forking finds a swap entry in place of a pte, or when reclaim unmaps a pte to insert the swap entry). swap_info_struct's vmalloc'ed swap_map is the array of these reference counts: but what happens when the unsigned short (or unsigned char since the preceding patch) is full? (and its high bit is kept for a cache flag) We then lose track of it, never freeing, leaving it in use until swapoff: at which point we _hope_ that a single pass will have found all instances, assume there are no more, and will lose user data if we're wrong. Swapping of KSM pages has not yet been enabled; but it is implemented, and makes it very easy for a user to overflow the maximum swap count: possible with ordinary process pages, but unlikely, even when pid_max has been raised from PID_MAX_DEFAULT. This patch implements swap count continuations: when the count overflows, a continuation page is allocated and linked to the original vmalloc'ed map page, and this used to hold the continuation counts for that entry and its neighbours. These continuation pages are seldom referenced: the common paths all work on the original swap_map, only referring to a continuation page when the low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-15 01:58:46 +00:00
entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte,
vma, addr, rss);
if (entry.val)
break;
progress += 8;
} while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
arch_leave_lazy_mmu_mode();
spin_unlock(src_ptl);
pte_unmap(orig_src_pte);
add_mm_rss_vec(dst_mm, rss);
pte_unmap_unlock(orig_dst_pte, dst_ptl);
cond_resched();
swap_info: swap count continuations Swap is duplicated (reference count incremented by one) whenever the same swap page is inserted into another mm (when forking finds a swap entry in place of a pte, or when reclaim unmaps a pte to insert the swap entry). swap_info_struct's vmalloc'ed swap_map is the array of these reference counts: but what happens when the unsigned short (or unsigned char since the preceding patch) is full? (and its high bit is kept for a cache flag) We then lose track of it, never freeing, leaving it in use until swapoff: at which point we _hope_ that a single pass will have found all instances, assume there are no more, and will lose user data if we're wrong. Swapping of KSM pages has not yet been enabled; but it is implemented, and makes it very easy for a user to overflow the maximum swap count: possible with ordinary process pages, but unlikely, even when pid_max has been raised from PID_MAX_DEFAULT. This patch implements swap count continuations: when the count overflows, a continuation page is allocated and linked to the original vmalloc'ed map page, and this used to hold the continuation counts for that entry and its neighbours. These continuation pages are seldom referenced: the common paths all work on the original swap_map, only referring to a continuation page when the low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-15 01:58:46 +00:00
if (entry.val) {
if (add_swap_count_continuation(entry, GFP_KERNEL) < 0)
return -ENOMEM;
progress = 0;
}
if (addr != end)
goto again;
return 0;
}
static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
unsigned long addr, unsigned long end)
{
pmd_t *src_pmd, *dst_pmd;
unsigned long next;
dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
if (!dst_pmd)
return -ENOMEM;
src_pmd = pmd_offset(src_pud, addr);
do {
next = pmd_addr_end(addr, end);
thp: transparent hugepage core Lately I've been working to make KVM use hugepages transparently without the usual restrictions of hugetlbfs. Some of the restrictions I'd like to see removed: 1) hugepages have to be swappable or the guest physical memory remains locked in RAM and can't be paged out to swap 2) if a hugepage allocation fails, regular pages should be allocated instead and mixed in the same vma without any failure and without userland noticing 3) if some task quits and more hugepages become available in the buddy, guest physical memory backed by regular pages should be relocated on hugepages automatically in regions under madvise(MADV_HUGEPAGE) (ideally event driven by waking up the kernel deamon if the order=HPAGE_PMD_SHIFT-PAGE_SHIFT list becomes not null) 4) avoidance of reservation and maximization of use of hugepages whenever possible. Reservation (needed to avoid runtime fatal faliures) may be ok for 1 machine with 1 database with 1 database cache with 1 database cache size known at boot time. It's definitely not feasible with a virtualization hypervisor usage like RHEV-H that runs an unknown number of virtual machines with an unknown size of each virtual machine with an unknown amount of pagecache that could be potentially useful in the host for guest not using O_DIRECT (aka cache=off). hugepages in the virtualization hypervisor (and also in the guest!) are much more important than in a regular host not using virtualization, becasue with NPT/EPT they decrease the tlb-miss cacheline accesses from 24 to 19 in case only the hypervisor uses transparent hugepages, and they decrease the tlb-miss cacheline accesses from 19 to 15 in case both the linux hypervisor and the linux guest both uses this patch (though the guest will limit the addition speedup to anonymous regions only for now...). Even more important is that the tlb miss handler is much slower on a NPT/EPT guest than for a regular shadow paging or no-virtualization scenario. So maximizing the amount of virtual memory cached by the TLB pays off significantly more with NPT/EPT than without (even if there would be no significant speedup in the tlb-miss runtime). The first (and more tedious) part of this work requires allowing the VM to handle anonymous hugepages mixed with regular pages transparently on regular anonymous vmas. This is what this patch tries to achieve in the least intrusive possible way. We want hugepages and hugetlb to be used in a way so that all applications can benefit without changes (as usual we leverage the KVM virtualization design: by improving the Linux VM at large, KVM gets the performance boost too). The most important design choice is: always fallback to 4k allocation if the hugepage allocation fails! This is the _very_ opposite of some large pagecache patches that failed with -EIO back then if a 64k (or similar) allocation failed... Second important decision (to reduce the impact of the feature on the existing pagetable handling code) is that at any time we can split an hugepage into 512 regular pages and it has to be done with an operation that can't fail. This way the reliability of the swapping isn't decreased (no need to allocate memory when we are short on memory to swap) and it's trivial to plug a split_huge_page* one-liner where needed without polluting the VM. Over time we can teach mprotect, mremap and friends to handle pmd_trans_huge natively without calling split_huge_page*. The fact it can't fail isn't just for swap: if split_huge_page would return -ENOMEM (instead of the current void) we'd need to rollback the mprotect from the middle of it (ideally including undoing the split_vma) which would be a big change and in the very wrong direction (it'd likely be simpler not to call split_huge_page at all and to teach mprotect and friends to handle hugepages instead of rolling them back from the middle). In short the very value of split_huge_page is that it can't fail. The collapsing and madvise(MADV_HUGEPAGE) part will remain separated and incremental and it'll just be an "harmless" addition later if this initial part is agreed upon. It also should be noted that locking-wise replacing regular pages with hugepages is going to be very easy if compared to what I'm doing below in split_huge_page, as it will only happen when page_count(page) matches page_mapcount(page) if we can take the PG_lock and mmap_sem in write mode. collapse_huge_page will be a "best effort" that (unlike split_huge_page) can fail at the minimal sign of trouble and we can try again later. collapse_huge_page will be similar to how KSM works and the madvise(MADV_HUGEPAGE) will work similar to madvise(MADV_MERGEABLE). The default I like is that transparent hugepages are used at page fault time. This can be changed with /sys/kernel/mm/transparent_hugepage/enabled. The control knob can be set to three values "always", "madvise", "never" which mean respectively that hugepages are always used, or only inside madvise(MADV_HUGEPAGE) regions, or never used. /sys/kernel/mm/transparent_hugepage/defrag instead controls if the hugepage allocation should defrag memory aggressively "always", only inside "madvise" regions, or "never". The pmd_trans_splitting/pmd_trans_huge locking is very solid. The put_page (from get_user_page users that can't use mmu notifier like O_DIRECT) that runs against a __split_huge_page_refcount instead was a pain to serialize in a way that would result always in a coherent page count for both tail and head. I think my locking solution with a compound_lock taken only after the page_first is valid and is still a PageHead should be safe but it surely needs review from SMP race point of view. In short there is no current existing way to serialize the O_DIRECT final put_page against split_huge_page_refcount so I had to invent a new one (O_DIRECT loses knowledge on the mapping status by the time gup_fast returns so...). And I didn't want to impact all gup/gup_fast users for now, maybe if we change the gup interface substantially we can avoid this locking, I admit I didn't think too much about it because changing the gup unpinning interface would be invasive. If we ignored O_DIRECT we could stick to the existing compound refcounting code, by simply adding a get_user_pages_fast_flags(foll_flags) where KVM (and any other mmu notifier user) would call it without FOLL_GET (and if FOLL_GET isn't set we'd just BUG_ON if nobody registered itself in the current task mmu notifier list yet). But O_DIRECT is fundamental for decent performance of virtualized I/O on fast storage so we can't avoid it to solve the race of put_page against split_huge_page_refcount to achieve a complete hugepage feature for KVM. Swap and oom works fine (well just like with regular pages ;). MMU notifier is handled transparently too, with the exception of the young bit on the pmd, that didn't have a range check but I think KVM will be fine because the whole point of hugepages is that EPT/NPT will also use a huge pmd when they notice gup returns pages with PageCompound set, so they won't care of a range and there's just the pmd young bit to check in that case. NOTE: in some cases if the L2 cache is small, this may slowdown and waste memory during COWs because 4M of memory are accessed in a single fault instead of 8k (the payoff is that after COW the program can run faster). So we might want to switch the copy_huge_page (and clear_huge_page too) to not temporal stores. I also extensively researched ways to avoid this cache trashing with a full prefault logic that would cow in 8k/16k/32k/64k up to 1M (I can send those patches that fully implemented prefault) but I concluded they're not worth it and they add an huge additional complexity and they remove all tlb benefits until the full hugepage has been faulted in, to save a little bit of memory and some cache during app startup, but they still don't improve substantially the cache-trashing during startup if the prefault happens in >4k chunks. One reason is that those 4k pte entries copied are still mapped on a perfectly cache-colored hugepage, so the trashing is the worst one can generate in those copies (cow of 4k page copies aren't so well colored so they trashes less, but again this results in software running faster after the page fault). Those prefault patches allowed things like a pte where post-cow pages were local 4k regular anon pages and the not-yet-cowed pte entries were pointing in the middle of some hugepage mapped read-only. If it doesn't payoff substantially with todays hardware it will payoff even less in the future with larger l2 caches, and the prefault logic would blot the VM a lot. If one is emebdded transparent_hugepage can be disabled during boot with sysfs or with the boot commandline parameter transparent_hugepage=0 (or transparent_hugepage=2 to restrict hugepages inside madvise regions) that will ensure not a single hugepage is allocated at boot time. It is simple enough to just disable transparent hugepage globally and let transparent hugepages be allocated selectively by applications in the MADV_HUGEPAGE region (both at page fault time, and if enabled with the collapse_huge_page too through the kernel daemon). This patch supports only hugepages mapped in the pmd, archs that have smaller hugepages will not fit in this patch alone. Also some archs like power have certain tlb limits that prevents mixing different page size in the same regions so they will not fit in this framework that requires "graceful fallback" to basic PAGE_SIZE in case of physical memory fragmentation. hugetlbfs remains a perfect fit for those because its software limits happen to match the hardware limits. hugetlbfs also remains a perfect fit for hugepage sizes like 1GByte that cannot be hoped to be found not fragmented after a certain system uptime and that would be very expensive to defragment with relocation, so requiring reservation. hugetlbfs is the "reservation way", the point of transparent hugepages is not to have any reservation at all and maximizing the use of cache and hugepages at all times automatically. Some performance result: vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largep ages3 memset page fault 1566023 memset tlb miss 453854 memset second tlb miss 453321 random access tlb miss 41635 random access second tlb miss 41658 vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largepages3 memset page fault 1566471 memset tlb miss 453375 memset second tlb miss 453320 random access tlb miss 41636 random access second tlb miss 41637 vmx andrea # ./largepages3 memset page fault 1566642 memset tlb miss 453417 memset second tlb miss 453313 random access tlb miss 41630 random access second tlb miss 41647 vmx andrea # ./largepages3 memset page fault 1566872 memset tlb miss 453418 memset second tlb miss 453315 random access tlb miss 41618 random access second tlb miss 41659 vmx andrea # echo 0 > /proc/sys/vm/transparent_hugepage vmx andrea # ./largepages3 memset page fault 2182476 memset tlb miss 460305 memset second tlb miss 460179 random access tlb miss 44483 random access second tlb miss 44186 vmx andrea # ./largepages3 memset page fault 2182791 memset tlb miss 460742 memset second tlb miss 459962 random access tlb miss 43981 random access second tlb miss 43988 ============ #include <stdio.h> #include <stdlib.h> #include <string.h> #include <sys/time.h> #define SIZE (3UL*1024*1024*1024) int main() { char *p = malloc(SIZE), *p2; struct timeval before, after; gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset page fault %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); return 0; } ============ Signed-off-by: Andrea Arcangeli <aarcange@redhat.com> Acked-by: Rik van Riel <riel@redhat.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-13 23:46:52 +00:00
if (pmd_trans_huge(*src_pmd)) {
int err;
VM_BUG_ON(next-addr != HPAGE_PMD_SIZE);
thp: transparent hugepage core Lately I've been working to make KVM use hugepages transparently without the usual restrictions of hugetlbfs. Some of the restrictions I'd like to see removed: 1) hugepages have to be swappable or the guest physical memory remains locked in RAM and can't be paged out to swap 2) if a hugepage allocation fails, regular pages should be allocated instead and mixed in the same vma without any failure and without userland noticing 3) if some task quits and more hugepages become available in the buddy, guest physical memory backed by regular pages should be relocated on hugepages automatically in regions under madvise(MADV_HUGEPAGE) (ideally event driven by waking up the kernel deamon if the order=HPAGE_PMD_SHIFT-PAGE_SHIFT list becomes not null) 4) avoidance of reservation and maximization of use of hugepages whenever possible. Reservation (needed to avoid runtime fatal faliures) may be ok for 1 machine with 1 database with 1 database cache with 1 database cache size known at boot time. It's definitely not feasible with a virtualization hypervisor usage like RHEV-H that runs an unknown number of virtual machines with an unknown size of each virtual machine with an unknown amount of pagecache that could be potentially useful in the host for guest not using O_DIRECT (aka cache=off). hugepages in the virtualization hypervisor (and also in the guest!) are much more important than in a regular host not using virtualization, becasue with NPT/EPT they decrease the tlb-miss cacheline accesses from 24 to 19 in case only the hypervisor uses transparent hugepages, and they decrease the tlb-miss cacheline accesses from 19 to 15 in case both the linux hypervisor and the linux guest both uses this patch (though the guest will limit the addition speedup to anonymous regions only for now...). Even more important is that the tlb miss handler is much slower on a NPT/EPT guest than for a regular shadow paging or no-virtualization scenario. So maximizing the amount of virtual memory cached by the TLB pays off significantly more with NPT/EPT than without (even if there would be no significant speedup in the tlb-miss runtime). The first (and more tedious) part of this work requires allowing the VM to handle anonymous hugepages mixed with regular pages transparently on regular anonymous vmas. This is what this patch tries to achieve in the least intrusive possible way. We want hugepages and hugetlb to be used in a way so that all applications can benefit without changes (as usual we leverage the KVM virtualization design: by improving the Linux VM at large, KVM gets the performance boost too). The most important design choice is: always fallback to 4k allocation if the hugepage allocation fails! This is the _very_ opposite of some large pagecache patches that failed with -EIO back then if a 64k (or similar) allocation failed... Second important decision (to reduce the impact of the feature on the existing pagetable handling code) is that at any time we can split an hugepage into 512 regular pages and it has to be done with an operation that can't fail. This way the reliability of the swapping isn't decreased (no need to allocate memory when we are short on memory to swap) and it's trivial to plug a split_huge_page* one-liner where needed without polluting the VM. Over time we can teach mprotect, mremap and friends to handle pmd_trans_huge natively without calling split_huge_page*. The fact it can't fail isn't just for swap: if split_huge_page would return -ENOMEM (instead of the current void) we'd need to rollback the mprotect from the middle of it (ideally including undoing the split_vma) which would be a big change and in the very wrong direction (it'd likely be simpler not to call split_huge_page at all and to teach mprotect and friends to handle hugepages instead of rolling them back from the middle). In short the very value of split_huge_page is that it can't fail. The collapsing and madvise(MADV_HUGEPAGE) part will remain separated and incremental and it'll just be an "harmless" addition later if this initial part is agreed upon. It also should be noted that locking-wise replacing regular pages with hugepages is going to be very easy if compared to what I'm doing below in split_huge_page, as it will only happen when page_count(page) matches page_mapcount(page) if we can take the PG_lock and mmap_sem in write mode. collapse_huge_page will be a "best effort" that (unlike split_huge_page) can fail at the minimal sign of trouble and we can try again later. collapse_huge_page will be similar to how KSM works and the madvise(MADV_HUGEPAGE) will work similar to madvise(MADV_MERGEABLE). The default I like is that transparent hugepages are used at page fault time. This can be changed with /sys/kernel/mm/transparent_hugepage/enabled. The control knob can be set to three values "always", "madvise", "never" which mean respectively that hugepages are always used, or only inside madvise(MADV_HUGEPAGE) regions, or never used. /sys/kernel/mm/transparent_hugepage/defrag instead controls if the hugepage allocation should defrag memory aggressively "always", only inside "madvise" regions, or "never". The pmd_trans_splitting/pmd_trans_huge locking is very solid. The put_page (from get_user_page users that can't use mmu notifier like O_DIRECT) that runs against a __split_huge_page_refcount instead was a pain to serialize in a way that would result always in a coherent page count for both tail and head. I think my locking solution with a compound_lock taken only after the page_first is valid and is still a PageHead should be safe but it surely needs review from SMP race point of view. In short there is no current existing way to serialize the O_DIRECT final put_page against split_huge_page_refcount so I had to invent a new one (O_DIRECT loses knowledge on the mapping status by the time gup_fast returns so...). And I didn't want to impact all gup/gup_fast users for now, maybe if we change the gup interface substantially we can avoid this locking, I admit I didn't think too much about it because changing the gup unpinning interface would be invasive. If we ignored O_DIRECT we could stick to the existing compound refcounting code, by simply adding a get_user_pages_fast_flags(foll_flags) where KVM (and any other mmu notifier user) would call it without FOLL_GET (and if FOLL_GET isn't set we'd just BUG_ON if nobody registered itself in the current task mmu notifier list yet). But O_DIRECT is fundamental for decent performance of virtualized I/O on fast storage so we can't avoid it to solve the race of put_page against split_huge_page_refcount to achieve a complete hugepage feature for KVM. Swap and oom works fine (well just like with regular pages ;). MMU notifier is handled transparently too, with the exception of the young bit on the pmd, that didn't have a range check but I think KVM will be fine because the whole point of hugepages is that EPT/NPT will also use a huge pmd when they notice gup returns pages with PageCompound set, so they won't care of a range and there's just the pmd young bit to check in that case. NOTE: in some cases if the L2 cache is small, this may slowdown and waste memory during COWs because 4M of memory are accessed in a single fault instead of 8k (the payoff is that after COW the program can run faster). So we might want to switch the copy_huge_page (and clear_huge_page too) to not temporal stores. I also extensively researched ways to avoid this cache trashing with a full prefault logic that would cow in 8k/16k/32k/64k up to 1M (I can send those patches that fully implemented prefault) but I concluded they're not worth it and they add an huge additional complexity and they remove all tlb benefits until the full hugepage has been faulted in, to save a little bit of memory and some cache during app startup, but they still don't improve substantially the cache-trashing during startup if the prefault happens in >4k chunks. One reason is that those 4k pte entries copied are still mapped on a perfectly cache-colored hugepage, so the trashing is the worst one can generate in those copies (cow of 4k page copies aren't so well colored so they trashes less, but again this results in software running faster after the page fault). Those prefault patches allowed things like a pte where post-cow pages were local 4k regular anon pages and the not-yet-cowed pte entries were pointing in the middle of some hugepage mapped read-only. If it doesn't payoff substantially with todays hardware it will payoff even less in the future with larger l2 caches, and the prefault logic would blot the VM a lot. If one is emebdded transparent_hugepage can be disabled during boot with sysfs or with the boot commandline parameter transparent_hugepage=0 (or transparent_hugepage=2 to restrict hugepages inside madvise regions) that will ensure not a single hugepage is allocated at boot time. It is simple enough to just disable transparent hugepage globally and let transparent hugepages be allocated selectively by applications in the MADV_HUGEPAGE region (both at page fault time, and if enabled with the collapse_huge_page too through the kernel daemon). This patch supports only hugepages mapped in the pmd, archs that have smaller hugepages will not fit in this patch alone. Also some archs like power have certain tlb limits that prevents mixing different page size in the same regions so they will not fit in this framework that requires "graceful fallback" to basic PAGE_SIZE in case of physical memory fragmentation. hugetlbfs remains a perfect fit for those because its software limits happen to match the hardware limits. hugetlbfs also remains a perfect fit for hugepage sizes like 1GByte that cannot be hoped to be found not fragmented after a certain system uptime and that would be very expensive to defragment with relocation, so requiring reservation. hugetlbfs is the "reservation way", the point of transparent hugepages is not to have any reservation at all and maximizing the use of cache and hugepages at all times automatically. Some performance result: vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largep ages3 memset page fault 1566023 memset tlb miss 453854 memset second tlb miss 453321 random access tlb miss 41635 random access second tlb miss 41658 vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largepages3 memset page fault 1566471 memset tlb miss 453375 memset second tlb miss 453320 random access tlb miss 41636 random access second tlb miss 41637 vmx andrea # ./largepages3 memset page fault 1566642 memset tlb miss 453417 memset second tlb miss 453313 random access tlb miss 41630 random access second tlb miss 41647 vmx andrea # ./largepages3 memset page fault 1566872 memset tlb miss 453418 memset second tlb miss 453315 random access tlb miss 41618 random access second tlb miss 41659 vmx andrea # echo 0 > /proc/sys/vm/transparent_hugepage vmx andrea # ./largepages3 memset page fault 2182476 memset tlb miss 460305 memset second tlb miss 460179 random access tlb miss 44483 random access second tlb miss 44186 vmx andrea # ./largepages3 memset page fault 2182791 memset tlb miss 460742 memset second tlb miss 459962 random access tlb miss 43981 random access second tlb miss 43988 ============ #include <stdio.h> #include <stdlib.h> #include <string.h> #include <sys/time.h> #define SIZE (3UL*1024*1024*1024) int main() { char *p = malloc(SIZE), *p2; struct timeval before, after; gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset page fault %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); return 0; } ============ Signed-off-by: Andrea Arcangeli <aarcange@redhat.com> Acked-by: Rik van Riel <riel@redhat.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-13 23:46:52 +00:00
err = copy_huge_pmd(dst_mm, src_mm,
dst_pmd, src_pmd, addr, vma);
if (err == -ENOMEM)
return -ENOMEM;
if (!err)
continue;
/* fall through */
}
if (pmd_none_or_clear_bad(src_pmd))
continue;
if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
vma, addr, next))
return -ENOMEM;
} while (dst_pmd++, src_pmd++, addr = next, addr != end);
return 0;
}
static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
unsigned long addr, unsigned long end)
{
pud_t *src_pud, *dst_pud;
unsigned long next;
dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
if (!dst_pud)
return -ENOMEM;
src_pud = pud_offset(src_pgd, addr);
do {
next = pud_addr_end(addr, end);
if (pud_none_or_clear_bad(src_pud))
continue;
if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
vma, addr, next))
return -ENOMEM;
} while (dst_pud++, src_pud++, addr = next, addr != end);
return 0;
}
int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
struct vm_area_struct *vma)
{
pgd_t *src_pgd, *dst_pgd;
unsigned long next;
unsigned long addr = vma->vm_start;
unsigned long end = vma->vm_end;
mm: move all mmu notifier invocations to be done outside the PT lock In order to allow sleeping during mmu notifier calls, we need to avoid invoking them under the page table spinlock. This patch solves the problem by calling invalidate_page notification after releasing the lock (but before freeing the page itself), or by wrapping the page invalidation with calls to invalidate_range_begin and invalidate_range_end. To prevent accidental changes to the invalidate_range_end arguments after the call to invalidate_range_begin, the patch introduces a convention of saving the arguments in consistently named locals: unsigned long mmun_start; /* For mmu_notifiers */ unsigned long mmun_end; /* For mmu_notifiers */ ... mmun_start = ... mmun_end = ... mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); ... mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); The patch changes code to use this convention for all calls to mmu_notifier_invalidate_range_start/end, except those where the calls are close enough so that anyone who glances at the code can see the values aren't changing. This patchset is a preliminary step towards on-demand paging design to be added to the RDMA stack. Why do we want on-demand paging for Infiniband? Applications register memory with an RDMA adapter using system calls, and subsequently post IO operations that refer to the corresponding virtual addresses directly to HW. Until now, this was achieved by pinning the memory during the registration calls. The goal of on demand paging is to avoid pinning the pages of registered memory regions (MRs). This will allow users the same flexibility they get when swapping any other part of their processes address spaces. Instead of requiring the entire MR to fit in physical memory, we can allow the MR to be larger, and only fit the current working set in physical memory. Why should anyone care? What problems are users currently experiencing? This can make programming with RDMA much simpler. Today, developers that are working with more data than their RAM can hold need either to deregister and reregister memory regions throughout their process's life, or keep a single memory region and copy the data to it. On demand paging will allow these developers to register a single MR at the beginning of their process's life, and let the operating system manage which pages needs to be fetched at a given time. In the future, we might be able to provide a single memory access key for each process that would provide the entire process's address as one large memory region, and the developers wouldn't need to register memory regions at all. Is there any prospect that any other subsystems will utilise these infrastructural changes? If so, which and how, etc? As for other subsystems, I understand that XPMEM wanted to sleep in MMU notifiers, as Christoph Lameter wrote at http://lkml.indiana.edu/hypermail/linux/kernel/0802.1/0460.html and perhaps Andrea knows about other use cases. Scheduling in mmu notifications is required since we need to sync the hardware with the secondary page tables change. A TLB flush of an IO device is inherently slower than a CPU TLB flush, so our design works by sending the invalidation request to the device, and waiting for an interrupt before exiting the mmu notifier handler. Avi said: kvm may be a buyer. kvm::mmu_lock, which serializes guest page faults, also protects long operations such as destroying large ranges. It would be good to convert it into a spinlock, but as it is used inside mmu notifiers, this cannot be done. (there are alternatives, such as keeping the spinlock and using a generation counter to do the teardown in O(1), which is what the "may" is doing up there). [akpm@linux-foundation.orgpossible speed tweak in hugetlb_cow(), cleanups] Signed-off-by: Andrea Arcangeli <andrea@qumranet.com> Signed-off-by: Sagi Grimberg <sagig@mellanox.com> Signed-off-by: Haggai Eran <haggaie@mellanox.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com> Cc: Or Gerlitz <ogerlitz@mellanox.com> Cc: Haggai Eran <haggaie@mellanox.com> Cc: Shachar Raindel <raindel@mellanox.com> Cc: Liran Liss <liranl@mellanox.com> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Avi Kivity <avi@redhat.com> Cc: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-08 23:33:33 +00:00
unsigned long mmun_start; /* For mmu_notifiers */
unsigned long mmun_end; /* For mmu_notifiers */
bool is_cow;
mmu-notifiers: core With KVM/GFP/XPMEM there isn't just the primary CPU MMU pointing to pages. There are secondary MMUs (with secondary sptes and secondary tlbs) too. sptes in the kvm case are shadow pagetables, but when I say spte in mmu-notifier context, I mean "secondary pte". In GRU case there's no actual secondary pte and there's only a secondary tlb because the GRU secondary MMU has no knowledge about sptes and every secondary tlb miss event in the MMU always generates a page fault that has to be resolved by the CPU (this is not the case of KVM where the a secondary tlb miss will walk sptes in hardware and it will refill the secondary tlb transparently to software if the corresponding spte is present). The same way zap_page_range has to invalidate the pte before freeing the page, the spte (and secondary tlb) must also be invalidated before any page is freed and reused. Currently we take a page_count pin on every page mapped by sptes, but that means the pages can't be swapped whenever they're mapped by any spte because they're part of the guest working set. Furthermore a spte unmap event can immediately lead to a page to be freed when the pin is released (so requiring the same complex and relatively slow tlb_gather smp safe logic we have in zap_page_range and that can be avoided completely if the spte unmap event doesn't require an unpin of the page previously mapped in the secondary MMU). The mmu notifiers allow kvm/GRU/XPMEM to attach to the tsk->mm and know when the VM is swapping or freeing or doing anything on the primary MMU so that the secondary MMU code can drop sptes before the pages are freed, avoiding all page pinning and allowing 100% reliable swapping of guest physical address space. Furthermore it avoids the code that teardown the mappings of the secondary MMU, to implement a logic like tlb_gather in zap_page_range that would require many IPI to flush other cpu tlbs, for each fixed number of spte unmapped. To make an example: if what happens on the primary MMU is a protection downgrade (from writeable to wrprotect) the secondary MMU mappings will be invalidated, and the next secondary-mmu-page-fault will call get_user_pages and trigger a do_wp_page through get_user_pages if it called get_user_pages with write=1, and it'll re-establishing an updated spte or secondary-tlb-mapping on the copied page. Or it will setup a readonly spte or readonly tlb mapping if it's a guest-read, if it calls get_user_pages with write=0. This is just an example. This allows to map any page pointed by any pte (and in turn visible in the primary CPU MMU), into a secondary MMU (be it a pure tlb like GRU, or an full MMU with both sptes and secondary-tlb like the shadow-pagetable layer with kvm), or a remote DMA in software like XPMEM (hence needing of schedule in XPMEM code to send the invalidate to the remote node, while no need to schedule in kvm/gru as it's an immediate event like invalidating primary-mmu pte). At least for KVM without this patch it's impossible to swap guests reliably. And having this feature and removing the page pin allows several other optimizations that simplify life considerably. Dependencies: 1) mm_take_all_locks() to register the mmu notifier when the whole VM isn't doing anything with "mm". This allows mmu notifier users to keep track if the VM is in the middle of the invalidate_range_begin/end critical section with an atomic counter incraese in range_begin and decreased in range_end. No secondary MMU page fault is allowed to map any spte or secondary tlb reference, while the VM is in the middle of range_begin/end as any page returned by get_user_pages in that critical section could later immediately be freed without any further ->invalidate_page notification (invalidate_range_begin/end works on ranges and ->invalidate_page isn't called immediately before freeing the page). To stop all page freeing and pagetable overwrites the mmap_sem must be taken in write mode and all other anon_vma/i_mmap locks must be taken too. 2) It'd be a waste to add branches in the VM if nobody could possibly run KVM/GRU/XPMEM on the kernel, so mmu notifiers will only enabled if CONFIG_KVM=m/y. In the current kernel kvm won't yet take advantage of mmu notifiers, but this already allows to compile a KVM external module against a kernel with mmu notifiers enabled and from the next pull from kvm.git we'll start using them. And GRU/XPMEM will also be able to continue the development by enabling KVM=m in their config, until they submit all GRU/XPMEM GPLv2 code to the mainline kernel. Then they can also enable MMU_NOTIFIERS in the same way KVM does it (even if KVM=n). This guarantees nobody selects MMU_NOTIFIER=y if KVM and GRU and XPMEM are all =n. The mmu_notifier_register call can fail because mm_take_all_locks may be interrupted by a signal and return -EINTR. Because mmu_notifier_reigster is used when a driver startup, a failure can be gracefully handled. Here an example of the change applied to kvm to register the mmu notifiers. Usually when a driver startups other allocations are required anyway and -ENOMEM failure paths exists already. struct kvm *kvm_arch_create_vm(void) { struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL); + int err; if (!kvm) return ERR_PTR(-ENOMEM); INIT_LIST_HEAD(&kvm->arch.active_mmu_pages); + kvm->arch.mmu_notifier.ops = &kvm_mmu_notifier_ops; + err = mmu_notifier_register(&kvm->arch.mmu_notifier, current->mm); + if (err) { + kfree(kvm); + return ERR_PTR(err); + } + return kvm; } mmu_notifier_unregister returns void and it's reliable. The patch also adds a few needed but missing includes that would prevent kernel to compile after these changes on non-x86 archs (x86 didn't need them by luck). [akpm@linux-foundation.org: coding-style fixes] [akpm@linux-foundation.org: fix mm/filemap_xip.c build] [akpm@linux-foundation.org: fix mm/mmu_notifier.c build] Signed-off-by: Andrea Arcangeli <andrea@qumranet.com> Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Christoph Lameter <cl@linux-foundation.org> Cc: Jack Steiner <steiner@sgi.com> Cc: Robin Holt <holt@sgi.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Kanoj Sarcar <kanojsarcar@yahoo.com> Cc: Roland Dreier <rdreier@cisco.com> Cc: Steve Wise <swise@opengridcomputing.com> Cc: Avi Kivity <avi@qumranet.com> Cc: Hugh Dickins <hugh@veritas.com> Cc: Rusty Russell <rusty@rustcorp.com.au> Cc: Anthony Liguori <aliguori@us.ibm.com> Cc: Chris Wright <chrisw@redhat.com> Cc: Marcelo Tosatti <marcelo@kvack.org> Cc: Eric Dumazet <dada1@cosmosbay.com> Cc: "Paul E. McKenney" <paulmck@us.ibm.com> Cc: Izik Eidus <izike@qumranet.com> Cc: Anthony Liguori <aliguori@us.ibm.com> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-28 22:46:29 +00:00
int ret;
/*
* Don't copy ptes where a page fault will fill them correctly.
* Fork becomes much lighter when there are big shared or private
* readonly mappings. The tradeoff is that copy_page_range is more
* efficient than faulting.
*/
mm: kill vma flag VM_INSERTPAGE Merge VM_INSERTPAGE into VM_MIXEDMAP. VM_MIXEDMAP VMA can mix pure-pfn ptes, special ptes and normal ptes. Now copy_page_range() always copies VM_MIXEDMAP VMA on fork like VM_PFNMAP. If driver populates whole VMA at mmap() it probably not expects page-faults. This patch removes special check from vma_wants_writenotify() which disables pages write tracking for VMA populated via vm_instert_page(). BDI below mapped file should not use dirty-accounting, moreover do_wp_page() can handle this. vm_insert_page() still marks vma after first usage. Usually it is called from f_op->mmap() handler under mm->mmap_sem write-lock, so it able to change vma->vm_flags. Caller must set VM_MIXEDMAP at mmap time if it wants to call this function from other places, for example from page-fault handler. Signed-off-by: Konstantin Khlebnikov <khlebnikov@openvz.org> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Carsten Otte <cotte@de.ibm.com> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Eric Paris <eparis@redhat.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Hugh Dickins <hughd@google.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: James Morris <james.l.morris@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Kentaro Takeda <takedakn@nttdata.co.jp> Cc: Matt Helsley <matthltc@us.ibm.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> Cc: Suresh Siddha <suresh.b.siddha@intel.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Venkatesh Pallipadi <venki@google.com> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-08 23:28:40 +00:00
if (!(vma->vm_flags & (VM_HUGETLB | VM_NONLINEAR |
VM_PFNMAP | VM_MIXEDMAP))) {
if (!vma->anon_vma)
return 0;
}
if (is_vm_hugetlb_page(vma))
return copy_hugetlb_page_range(dst_mm, src_mm, vma);
mm, x86, pat: rework linear pfn-mmap tracking Replace the generic vma-flag VM_PFN_AT_MMAP with x86-only VM_PAT. We can toss mapping address from remap_pfn_range() into track_pfn_vma_new(), and collect all PAT-related logic together in arch/x86/. This patch also restores orignal frustration-free is_cow_mapping() check in remap_pfn_range(), as it was before commit v2.6.28-rc8-88-g3c8bb73 ("x86: PAT: store vm_pgoff for all linear_over_vma_region mappings - v3") is_linear_pfn_mapping() checks can be removed from mm/huge_memory.c, because it already handled by VM_PFNMAP in VM_NO_THP bit-mask. [suresh.b.siddha@intel.com: Reset the VM_PAT flag as part of untrack_pfn_vma()] Signed-off-by: Konstantin Khlebnikov <khlebnikov@openvz.org> Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com> Cc: Venkatesh Pallipadi <venki@google.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Ingo Molnar <mingo@redhat.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Carsten Otte <cotte@de.ibm.com> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Eric Paris <eparis@redhat.com> Cc: Hugh Dickins <hughd@google.com> Cc: James Morris <james.l.morris@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Kentaro Takeda <takedakn@nttdata.co.jp> Cc: Matt Helsley <matthltc@us.ibm.com> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Venkatesh Pallipadi <venki@google.com> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-08 23:28:34 +00:00
if (unlikely(vma->vm_flags & VM_PFNMAP)) {
/*
* We do not free on error cases below as remove_vma
* gets called on error from higher level routine
*/
x86, pat: separate the pfn attribute tracking for remap_pfn_range and vm_insert_pfn With PAT enabled, vm_insert_pfn() looks up the existing pfn memory attribute and uses it. Expectation is that the driver reserves the memory attributes for the pfn before calling vm_insert_pfn(). remap_pfn_range() (when called for the whole vma) will setup a new attribute (based on the prot argument) for the specified pfn range. This addresses the legacy usage which typically calls remap_pfn_range() with a desired memory attribute. For ranges smaller than the vma size (which is typically not the case), remap_pfn_range() will use the existing memory attribute for the pfn range. Expose two different API's for these different behaviors. track_pfn_insert() for tracking the pfn attribute set by vm_insert_pfn() and track_pfn_remap() for the remap_pfn_range(). This cleanup also prepares the ground for the track/untrack pfn vma routines to take over the ownership of setting PAT specific vm_flag in the 'vma'. [khlebnikov@openvz.org: Clear checks in track_pfn_remap()] [akpm@linux-foundation.org: tweak a few comments] Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com> Signed-off-by: Konstantin Khlebnikov <khlebnikov@openvz.org> Cc: Venkatesh Pallipadi <venki@google.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Ingo Molnar <mingo@redhat.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Carsten Otte <cotte@de.ibm.com> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Eric Paris <eparis@redhat.com> Cc: Hugh Dickins <hughd@google.com> Cc: James Morris <james.l.morris@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Kentaro Takeda <takedakn@nttdata.co.jp> Cc: Konstantin Khlebnikov <khlebnikov@openvz.org> Cc: Matt Helsley <matthltc@us.ibm.com> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> Cc: Suresh Siddha <suresh.b.siddha@intel.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-08 23:28:29 +00:00
ret = track_pfn_copy(vma);
if (ret)
return ret;
}
mmu-notifiers: core With KVM/GFP/XPMEM there isn't just the primary CPU MMU pointing to pages. There are secondary MMUs (with secondary sptes and secondary tlbs) too. sptes in the kvm case are shadow pagetables, but when I say spte in mmu-notifier context, I mean "secondary pte". In GRU case there's no actual secondary pte and there's only a secondary tlb because the GRU secondary MMU has no knowledge about sptes and every secondary tlb miss event in the MMU always generates a page fault that has to be resolved by the CPU (this is not the case of KVM where the a secondary tlb miss will walk sptes in hardware and it will refill the secondary tlb transparently to software if the corresponding spte is present). The same way zap_page_range has to invalidate the pte before freeing the page, the spte (and secondary tlb) must also be invalidated before any page is freed and reused. Currently we take a page_count pin on every page mapped by sptes, but that means the pages can't be swapped whenever they're mapped by any spte because they're part of the guest working set. Furthermore a spte unmap event can immediately lead to a page to be freed when the pin is released (so requiring the same complex and relatively slow tlb_gather smp safe logic we have in zap_page_range and that can be avoided completely if the spte unmap event doesn't require an unpin of the page previously mapped in the secondary MMU). The mmu notifiers allow kvm/GRU/XPMEM to attach to the tsk->mm and know when the VM is swapping or freeing or doing anything on the primary MMU so that the secondary MMU code can drop sptes before the pages are freed, avoiding all page pinning and allowing 100% reliable swapping of guest physical address space. Furthermore it avoids the code that teardown the mappings of the secondary MMU, to implement a logic like tlb_gather in zap_page_range that would require many IPI to flush other cpu tlbs, for each fixed number of spte unmapped. To make an example: if what happens on the primary MMU is a protection downgrade (from writeable to wrprotect) the secondary MMU mappings will be invalidated, and the next secondary-mmu-page-fault will call get_user_pages and trigger a do_wp_page through get_user_pages if it called get_user_pages with write=1, and it'll re-establishing an updated spte or secondary-tlb-mapping on the copied page. Or it will setup a readonly spte or readonly tlb mapping if it's a guest-read, if it calls get_user_pages with write=0. This is just an example. This allows to map any page pointed by any pte (and in turn visible in the primary CPU MMU), into a secondary MMU (be it a pure tlb like GRU, or an full MMU with both sptes and secondary-tlb like the shadow-pagetable layer with kvm), or a remote DMA in software like XPMEM (hence needing of schedule in XPMEM code to send the invalidate to the remote node, while no need to schedule in kvm/gru as it's an immediate event like invalidating primary-mmu pte). At least for KVM without this patch it's impossible to swap guests reliably. And having this feature and removing the page pin allows several other optimizations that simplify life considerably. Dependencies: 1) mm_take_all_locks() to register the mmu notifier when the whole VM isn't doing anything with "mm". This allows mmu notifier users to keep track if the VM is in the middle of the invalidate_range_begin/end critical section with an atomic counter incraese in range_begin and decreased in range_end. No secondary MMU page fault is allowed to map any spte or secondary tlb reference, while the VM is in the middle of range_begin/end as any page returned by get_user_pages in that critical section could later immediately be freed without any further ->invalidate_page notification (invalidate_range_begin/end works on ranges and ->invalidate_page isn't called immediately before freeing the page). To stop all page freeing and pagetable overwrites the mmap_sem must be taken in write mode and all other anon_vma/i_mmap locks must be taken too. 2) It'd be a waste to add branches in the VM if nobody could possibly run KVM/GRU/XPMEM on the kernel, so mmu notifiers will only enabled if CONFIG_KVM=m/y. In the current kernel kvm won't yet take advantage of mmu notifiers, but this already allows to compile a KVM external module against a kernel with mmu notifiers enabled and from the next pull from kvm.git we'll start using them. And GRU/XPMEM will also be able to continue the development by enabling KVM=m in their config, until they submit all GRU/XPMEM GPLv2 code to the mainline kernel. Then they can also enable MMU_NOTIFIERS in the same way KVM does it (even if KVM=n). This guarantees nobody selects MMU_NOTIFIER=y if KVM and GRU and XPMEM are all =n. The mmu_notifier_register call can fail because mm_take_all_locks may be interrupted by a signal and return -EINTR. Because mmu_notifier_reigster is used when a driver startup, a failure can be gracefully handled. Here an example of the change applied to kvm to register the mmu notifiers. Usually when a driver startups other allocations are required anyway and -ENOMEM failure paths exists already. struct kvm *kvm_arch_create_vm(void) { struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL); + int err; if (!kvm) return ERR_PTR(-ENOMEM); INIT_LIST_HEAD(&kvm->arch.active_mmu_pages); + kvm->arch.mmu_notifier.ops = &kvm_mmu_notifier_ops; + err = mmu_notifier_register(&kvm->arch.mmu_notifier, current->mm); + if (err) { + kfree(kvm); + return ERR_PTR(err); + } + return kvm; } mmu_notifier_unregister returns void and it's reliable. The patch also adds a few needed but missing includes that would prevent kernel to compile after these changes on non-x86 archs (x86 didn't need them by luck). [akpm@linux-foundation.org: coding-style fixes] [akpm@linux-foundation.org: fix mm/filemap_xip.c build] [akpm@linux-foundation.org: fix mm/mmu_notifier.c build] Signed-off-by: Andrea Arcangeli <andrea@qumranet.com> Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Christoph Lameter <cl@linux-foundation.org> Cc: Jack Steiner <steiner@sgi.com> Cc: Robin Holt <holt@sgi.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Kanoj Sarcar <kanojsarcar@yahoo.com> Cc: Roland Dreier <rdreier@cisco.com> Cc: Steve Wise <swise@opengridcomputing.com> Cc: Avi Kivity <avi@qumranet.com> Cc: Hugh Dickins <hugh@veritas.com> Cc: Rusty Russell <rusty@rustcorp.com.au> Cc: Anthony Liguori <aliguori@us.ibm.com> Cc: Chris Wright <chrisw@redhat.com> Cc: Marcelo Tosatti <marcelo@kvack.org> Cc: Eric Dumazet <dada1@cosmosbay.com> Cc: "Paul E. McKenney" <paulmck@us.ibm.com> Cc: Izik Eidus <izike@qumranet.com> Cc: Anthony Liguori <aliguori@us.ibm.com> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-28 22:46:29 +00:00
/*
* We need to invalidate the secondary MMU mappings only when
* there could be a permission downgrade on the ptes of the
* parent mm. And a permission downgrade will only happen if
* is_cow_mapping() returns true.
*/
mm: move all mmu notifier invocations to be done outside the PT lock In order to allow sleeping during mmu notifier calls, we need to avoid invoking them under the page table spinlock. This patch solves the problem by calling invalidate_page notification after releasing the lock (but before freeing the page itself), or by wrapping the page invalidation with calls to invalidate_range_begin and invalidate_range_end. To prevent accidental changes to the invalidate_range_end arguments after the call to invalidate_range_begin, the patch introduces a convention of saving the arguments in consistently named locals: unsigned long mmun_start; /* For mmu_notifiers */ unsigned long mmun_end; /* For mmu_notifiers */ ... mmun_start = ... mmun_end = ... mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); ... mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); The patch changes code to use this convention for all calls to mmu_notifier_invalidate_range_start/end, except those where the calls are close enough so that anyone who glances at the code can see the values aren't changing. This patchset is a preliminary step towards on-demand paging design to be added to the RDMA stack. Why do we want on-demand paging for Infiniband? Applications register memory with an RDMA adapter using system calls, and subsequently post IO operations that refer to the corresponding virtual addresses directly to HW. Until now, this was achieved by pinning the memory during the registration calls. The goal of on demand paging is to avoid pinning the pages of registered memory regions (MRs). This will allow users the same flexibility they get when swapping any other part of their processes address spaces. Instead of requiring the entire MR to fit in physical memory, we can allow the MR to be larger, and only fit the current working set in physical memory. Why should anyone care? What problems are users currently experiencing? This can make programming with RDMA much simpler. Today, developers that are working with more data than their RAM can hold need either to deregister and reregister memory regions throughout their process's life, or keep a single memory region and copy the data to it. On demand paging will allow these developers to register a single MR at the beginning of their process's life, and let the operating system manage which pages needs to be fetched at a given time. In the future, we might be able to provide a single memory access key for each process that would provide the entire process's address as one large memory region, and the developers wouldn't need to register memory regions at all. Is there any prospect that any other subsystems will utilise these infrastructural changes? If so, which and how, etc? As for other subsystems, I understand that XPMEM wanted to sleep in MMU notifiers, as Christoph Lameter wrote at http://lkml.indiana.edu/hypermail/linux/kernel/0802.1/0460.html and perhaps Andrea knows about other use cases. Scheduling in mmu notifications is required since we need to sync the hardware with the secondary page tables change. A TLB flush of an IO device is inherently slower than a CPU TLB flush, so our design works by sending the invalidation request to the device, and waiting for an interrupt before exiting the mmu notifier handler. Avi said: kvm may be a buyer. kvm::mmu_lock, which serializes guest page faults, also protects long operations such as destroying large ranges. It would be good to convert it into a spinlock, but as it is used inside mmu notifiers, this cannot be done. (there are alternatives, such as keeping the spinlock and using a generation counter to do the teardown in O(1), which is what the "may" is doing up there). [akpm@linux-foundation.orgpossible speed tweak in hugetlb_cow(), cleanups] Signed-off-by: Andrea Arcangeli <andrea@qumranet.com> Signed-off-by: Sagi Grimberg <sagig@mellanox.com> Signed-off-by: Haggai Eran <haggaie@mellanox.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com> Cc: Or Gerlitz <ogerlitz@mellanox.com> Cc: Haggai Eran <haggaie@mellanox.com> Cc: Shachar Raindel <raindel@mellanox.com> Cc: Liran Liss <liranl@mellanox.com> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Avi Kivity <avi@redhat.com> Cc: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-08 23:33:33 +00:00
is_cow = is_cow_mapping(vma->vm_flags);
mmun_start = addr;
mmun_end = end;
if (is_cow)
mmu_notifier_invalidate_range_start(src_mm, mmun_start,
mmun_end);
mmu-notifiers: core With KVM/GFP/XPMEM there isn't just the primary CPU MMU pointing to pages. There are secondary MMUs (with secondary sptes and secondary tlbs) too. sptes in the kvm case are shadow pagetables, but when I say spte in mmu-notifier context, I mean "secondary pte". In GRU case there's no actual secondary pte and there's only a secondary tlb because the GRU secondary MMU has no knowledge about sptes and every secondary tlb miss event in the MMU always generates a page fault that has to be resolved by the CPU (this is not the case of KVM where the a secondary tlb miss will walk sptes in hardware and it will refill the secondary tlb transparently to software if the corresponding spte is present). The same way zap_page_range has to invalidate the pte before freeing the page, the spte (and secondary tlb) must also be invalidated before any page is freed and reused. Currently we take a page_count pin on every page mapped by sptes, but that means the pages can't be swapped whenever they're mapped by any spte because they're part of the guest working set. Furthermore a spte unmap event can immediately lead to a page to be freed when the pin is released (so requiring the same complex and relatively slow tlb_gather smp safe logic we have in zap_page_range and that can be avoided completely if the spte unmap event doesn't require an unpin of the page previously mapped in the secondary MMU). The mmu notifiers allow kvm/GRU/XPMEM to attach to the tsk->mm and know when the VM is swapping or freeing or doing anything on the primary MMU so that the secondary MMU code can drop sptes before the pages are freed, avoiding all page pinning and allowing 100% reliable swapping of guest physical address space. Furthermore it avoids the code that teardown the mappings of the secondary MMU, to implement a logic like tlb_gather in zap_page_range that would require many IPI to flush other cpu tlbs, for each fixed number of spte unmapped. To make an example: if what happens on the primary MMU is a protection downgrade (from writeable to wrprotect) the secondary MMU mappings will be invalidated, and the next secondary-mmu-page-fault will call get_user_pages and trigger a do_wp_page through get_user_pages if it called get_user_pages with write=1, and it'll re-establishing an updated spte or secondary-tlb-mapping on the copied page. Or it will setup a readonly spte or readonly tlb mapping if it's a guest-read, if it calls get_user_pages with write=0. This is just an example. This allows to map any page pointed by any pte (and in turn visible in the primary CPU MMU), into a secondary MMU (be it a pure tlb like GRU, or an full MMU with both sptes and secondary-tlb like the shadow-pagetable layer with kvm), or a remote DMA in software like XPMEM (hence needing of schedule in XPMEM code to send the invalidate to the remote node, while no need to schedule in kvm/gru as it's an immediate event like invalidating primary-mmu pte). At least for KVM without this patch it's impossible to swap guests reliably. And having this feature and removing the page pin allows several other optimizations that simplify life considerably. Dependencies: 1) mm_take_all_locks() to register the mmu notifier when the whole VM isn't doing anything with "mm". This allows mmu notifier users to keep track if the VM is in the middle of the invalidate_range_begin/end critical section with an atomic counter incraese in range_begin and decreased in range_end. No secondary MMU page fault is allowed to map any spte or secondary tlb reference, while the VM is in the middle of range_begin/end as any page returned by get_user_pages in that critical section could later immediately be freed without any further ->invalidate_page notification (invalidate_range_begin/end works on ranges and ->invalidate_page isn't called immediately before freeing the page). To stop all page freeing and pagetable overwrites the mmap_sem must be taken in write mode and all other anon_vma/i_mmap locks must be taken too. 2) It'd be a waste to add branches in the VM if nobody could possibly run KVM/GRU/XPMEM on the kernel, so mmu notifiers will only enabled if CONFIG_KVM=m/y. In the current kernel kvm won't yet take advantage of mmu notifiers, but this already allows to compile a KVM external module against a kernel with mmu notifiers enabled and from the next pull from kvm.git we'll start using them. And GRU/XPMEM will also be able to continue the development by enabling KVM=m in their config, until they submit all GRU/XPMEM GPLv2 code to the mainline kernel. Then they can also enable MMU_NOTIFIERS in the same way KVM does it (even if KVM=n). This guarantees nobody selects MMU_NOTIFIER=y if KVM and GRU and XPMEM are all =n. The mmu_notifier_register call can fail because mm_take_all_locks may be interrupted by a signal and return -EINTR. Because mmu_notifier_reigster is used when a driver startup, a failure can be gracefully handled. Here an example of the change applied to kvm to register the mmu notifiers. Usually when a driver startups other allocations are required anyway and -ENOMEM failure paths exists already. struct kvm *kvm_arch_create_vm(void) { struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL); + int err; if (!kvm) return ERR_PTR(-ENOMEM); INIT_LIST_HEAD(&kvm->arch.active_mmu_pages); + kvm->arch.mmu_notifier.ops = &kvm_mmu_notifier_ops; + err = mmu_notifier_register(&kvm->arch.mmu_notifier, current->mm); + if (err) { + kfree(kvm); + return ERR_PTR(err); + } + return kvm; } mmu_notifier_unregister returns void and it's reliable. The patch also adds a few needed but missing includes that would prevent kernel to compile after these changes on non-x86 archs (x86 didn't need them by luck). [akpm@linux-foundation.org: coding-style fixes] [akpm@linux-foundation.org: fix mm/filemap_xip.c build] [akpm@linux-foundation.org: fix mm/mmu_notifier.c build] Signed-off-by: Andrea Arcangeli <andrea@qumranet.com> Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Christoph Lameter <cl@linux-foundation.org> Cc: Jack Steiner <steiner@sgi.com> Cc: Robin Holt <holt@sgi.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Kanoj Sarcar <kanojsarcar@yahoo.com> Cc: Roland Dreier <rdreier@cisco.com> Cc: Steve Wise <swise@opengridcomputing.com> Cc: Avi Kivity <avi@qumranet.com> Cc: Hugh Dickins <hugh@veritas.com> Cc: Rusty Russell <rusty@rustcorp.com.au> Cc: Anthony Liguori <aliguori@us.ibm.com> Cc: Chris Wright <chrisw@redhat.com> Cc: Marcelo Tosatti <marcelo@kvack.org> Cc: Eric Dumazet <dada1@cosmosbay.com> Cc: "Paul E. McKenney" <paulmck@us.ibm.com> Cc: Izik Eidus <izike@qumranet.com> Cc: Anthony Liguori <aliguori@us.ibm.com> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-28 22:46:29 +00:00
ret = 0;
dst_pgd = pgd_offset(dst_mm, addr);
src_pgd = pgd_offset(src_mm, addr);
do {
next = pgd_addr_end(addr, end);
if (pgd_none_or_clear_bad(src_pgd))
continue;
mmu-notifiers: core With KVM/GFP/XPMEM there isn't just the primary CPU MMU pointing to pages. There are secondary MMUs (with secondary sptes and secondary tlbs) too. sptes in the kvm case are shadow pagetables, but when I say spte in mmu-notifier context, I mean "secondary pte". In GRU case there's no actual secondary pte and there's only a secondary tlb because the GRU secondary MMU has no knowledge about sptes and every secondary tlb miss event in the MMU always generates a page fault that has to be resolved by the CPU (this is not the case of KVM where the a secondary tlb miss will walk sptes in hardware and it will refill the secondary tlb transparently to software if the corresponding spte is present). The same way zap_page_range has to invalidate the pte before freeing the page, the spte (and secondary tlb) must also be invalidated before any page is freed and reused. Currently we take a page_count pin on every page mapped by sptes, but that means the pages can't be swapped whenever they're mapped by any spte because they're part of the guest working set. Furthermore a spte unmap event can immediately lead to a page to be freed when the pin is released (so requiring the same complex and relatively slow tlb_gather smp safe logic we have in zap_page_range and that can be avoided completely if the spte unmap event doesn't require an unpin of the page previously mapped in the secondary MMU). The mmu notifiers allow kvm/GRU/XPMEM to attach to the tsk->mm and know when the VM is swapping or freeing or doing anything on the primary MMU so that the secondary MMU code can drop sptes before the pages are freed, avoiding all page pinning and allowing 100% reliable swapping of guest physical address space. Furthermore it avoids the code that teardown the mappings of the secondary MMU, to implement a logic like tlb_gather in zap_page_range that would require many IPI to flush other cpu tlbs, for each fixed number of spte unmapped. To make an example: if what happens on the primary MMU is a protection downgrade (from writeable to wrprotect) the secondary MMU mappings will be invalidated, and the next secondary-mmu-page-fault will call get_user_pages and trigger a do_wp_page through get_user_pages if it called get_user_pages with write=1, and it'll re-establishing an updated spte or secondary-tlb-mapping on the copied page. Or it will setup a readonly spte or readonly tlb mapping if it's a guest-read, if it calls get_user_pages with write=0. This is just an example. This allows to map any page pointed by any pte (and in turn visible in the primary CPU MMU), into a secondary MMU (be it a pure tlb like GRU, or an full MMU with both sptes and secondary-tlb like the shadow-pagetable layer with kvm), or a remote DMA in software like XPMEM (hence needing of schedule in XPMEM code to send the invalidate to the remote node, while no need to schedule in kvm/gru as it's an immediate event like invalidating primary-mmu pte). At least for KVM without this patch it's impossible to swap guests reliably. And having this feature and removing the page pin allows several other optimizations that simplify life considerably. Dependencies: 1) mm_take_all_locks() to register the mmu notifier when the whole VM isn't doing anything with "mm". This allows mmu notifier users to keep track if the VM is in the middle of the invalidate_range_begin/end critical section with an atomic counter incraese in range_begin and decreased in range_end. No secondary MMU page fault is allowed to map any spte or secondary tlb reference, while the VM is in the middle of range_begin/end as any page returned by get_user_pages in that critical section could later immediately be freed without any further ->invalidate_page notification (invalidate_range_begin/end works on ranges and ->invalidate_page isn't called immediately before freeing the page). To stop all page freeing and pagetable overwrites the mmap_sem must be taken in write mode and all other anon_vma/i_mmap locks must be taken too. 2) It'd be a waste to add branches in the VM if nobody could possibly run KVM/GRU/XPMEM on the kernel, so mmu notifiers will only enabled if CONFIG_KVM=m/y. In the current kernel kvm won't yet take advantage of mmu notifiers, but this already allows to compile a KVM external module against a kernel with mmu notifiers enabled and from the next pull from kvm.git we'll start using them. And GRU/XPMEM will also be able to continue the development by enabling KVM=m in their config, until they submit all GRU/XPMEM GPLv2 code to the mainline kernel. Then they can also enable MMU_NOTIFIERS in the same way KVM does it (even if KVM=n). This guarantees nobody selects MMU_NOTIFIER=y if KVM and GRU and XPMEM are all =n. The mmu_notifier_register call can fail because mm_take_all_locks may be interrupted by a signal and return -EINTR. Because mmu_notifier_reigster is used when a driver startup, a failure can be gracefully handled. Here an example of the change applied to kvm to register the mmu notifiers. Usually when a driver startups other allocations are required anyway and -ENOMEM failure paths exists already. struct kvm *kvm_arch_create_vm(void) { struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL); + int err; if (!kvm) return ERR_PTR(-ENOMEM); INIT_LIST_HEAD(&kvm->arch.active_mmu_pages); + kvm->arch.mmu_notifier.ops = &kvm_mmu_notifier_ops; + err = mmu_notifier_register(&kvm->arch.mmu_notifier, current->mm); + if (err) { + kfree(kvm); + return ERR_PTR(err); + } + return kvm; } mmu_notifier_unregister returns void and it's reliable. The patch also adds a few needed but missing includes that would prevent kernel to compile after these changes on non-x86 archs (x86 didn't need them by luck). [akpm@linux-foundation.org: coding-style fixes] [akpm@linux-foundation.org: fix mm/filemap_xip.c build] [akpm@linux-foundation.org: fix mm/mmu_notifier.c build] Signed-off-by: Andrea Arcangeli <andrea@qumranet.com> Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Christoph Lameter <cl@linux-foundation.org> Cc: Jack Steiner <steiner@sgi.com> Cc: Robin Holt <holt@sgi.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Kanoj Sarcar <kanojsarcar@yahoo.com> Cc: Roland Dreier <rdreier@cisco.com> Cc: Steve Wise <swise@opengridcomputing.com> Cc: Avi Kivity <avi@qumranet.com> Cc: Hugh Dickins <hugh@veritas.com> Cc: Rusty Russell <rusty@rustcorp.com.au> Cc: Anthony Liguori <aliguori@us.ibm.com> Cc: Chris Wright <chrisw@redhat.com> Cc: Marcelo Tosatti <marcelo@kvack.org> Cc: Eric Dumazet <dada1@cosmosbay.com> Cc: "Paul E. McKenney" <paulmck@us.ibm.com> Cc: Izik Eidus <izike@qumranet.com> Cc: Anthony Liguori <aliguori@us.ibm.com> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-28 22:46:29 +00:00
if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
vma, addr, next))) {
ret = -ENOMEM;
break;
}
} while (dst_pgd++, src_pgd++, addr = next, addr != end);
mmu-notifiers: core With KVM/GFP/XPMEM there isn't just the primary CPU MMU pointing to pages. There are secondary MMUs (with secondary sptes and secondary tlbs) too. sptes in the kvm case are shadow pagetables, but when I say spte in mmu-notifier context, I mean "secondary pte". In GRU case there's no actual secondary pte and there's only a secondary tlb because the GRU secondary MMU has no knowledge about sptes and every secondary tlb miss event in the MMU always generates a page fault that has to be resolved by the CPU (this is not the case of KVM where the a secondary tlb miss will walk sptes in hardware and it will refill the secondary tlb transparently to software if the corresponding spte is present). The same way zap_page_range has to invalidate the pte before freeing the page, the spte (and secondary tlb) must also be invalidated before any page is freed and reused. Currently we take a page_count pin on every page mapped by sptes, but that means the pages can't be swapped whenever they're mapped by any spte because they're part of the guest working set. Furthermore a spte unmap event can immediately lead to a page to be freed when the pin is released (so requiring the same complex and relatively slow tlb_gather smp safe logic we have in zap_page_range and that can be avoided completely if the spte unmap event doesn't require an unpin of the page previously mapped in the secondary MMU). The mmu notifiers allow kvm/GRU/XPMEM to attach to the tsk->mm and know when the VM is swapping or freeing or doing anything on the primary MMU so that the secondary MMU code can drop sptes before the pages are freed, avoiding all page pinning and allowing 100% reliable swapping of guest physical address space. Furthermore it avoids the code that teardown the mappings of the secondary MMU, to implement a logic like tlb_gather in zap_page_range that would require many IPI to flush other cpu tlbs, for each fixed number of spte unmapped. To make an example: if what happens on the primary MMU is a protection downgrade (from writeable to wrprotect) the secondary MMU mappings will be invalidated, and the next secondary-mmu-page-fault will call get_user_pages and trigger a do_wp_page through get_user_pages if it called get_user_pages with write=1, and it'll re-establishing an updated spte or secondary-tlb-mapping on the copied page. Or it will setup a readonly spte or readonly tlb mapping if it's a guest-read, if it calls get_user_pages with write=0. This is just an example. This allows to map any page pointed by any pte (and in turn visible in the primary CPU MMU), into a secondary MMU (be it a pure tlb like GRU, or an full MMU with both sptes and secondary-tlb like the shadow-pagetable layer with kvm), or a remote DMA in software like XPMEM (hence needing of schedule in XPMEM code to send the invalidate to the remote node, while no need to schedule in kvm/gru as it's an immediate event like invalidating primary-mmu pte). At least for KVM without this patch it's impossible to swap guests reliably. And having this feature and removing the page pin allows several other optimizations that simplify life considerably. Dependencies: 1) mm_take_all_locks() to register the mmu notifier when the whole VM isn't doing anything with "mm". This allows mmu notifier users to keep track if the VM is in the middle of the invalidate_range_begin/end critical section with an atomic counter incraese in range_begin and decreased in range_end. No secondary MMU page fault is allowed to map any spte or secondary tlb reference, while the VM is in the middle of range_begin/end as any page returned by get_user_pages in that critical section could later immediately be freed without any further ->invalidate_page notification (invalidate_range_begin/end works on ranges and ->invalidate_page isn't called immediately before freeing the page). To stop all page freeing and pagetable overwrites the mmap_sem must be taken in write mode and all other anon_vma/i_mmap locks must be taken too. 2) It'd be a waste to add branches in the VM if nobody could possibly run KVM/GRU/XPMEM on the kernel, so mmu notifiers will only enabled if CONFIG_KVM=m/y. In the current kernel kvm won't yet take advantage of mmu notifiers, but this already allows to compile a KVM external module against a kernel with mmu notifiers enabled and from the next pull from kvm.git we'll start using them. And GRU/XPMEM will also be able to continue the development by enabling KVM=m in their config, until they submit all GRU/XPMEM GPLv2 code to the mainline kernel. Then they can also enable MMU_NOTIFIERS in the same way KVM does it (even if KVM=n). This guarantees nobody selects MMU_NOTIFIER=y if KVM and GRU and XPMEM are all =n. The mmu_notifier_register call can fail because mm_take_all_locks may be interrupted by a signal and return -EINTR. Because mmu_notifier_reigster is used when a driver startup, a failure can be gracefully handled. Here an example of the change applied to kvm to register the mmu notifiers. Usually when a driver startups other allocations are required anyway and -ENOMEM failure paths exists already. struct kvm *kvm_arch_create_vm(void) { struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL); + int err; if (!kvm) return ERR_PTR(-ENOMEM); INIT_LIST_HEAD(&kvm->arch.active_mmu_pages); + kvm->arch.mmu_notifier.ops = &kvm_mmu_notifier_ops; + err = mmu_notifier_register(&kvm->arch.mmu_notifier, current->mm); + if (err) { + kfree(kvm); + return ERR_PTR(err); + } + return kvm; } mmu_notifier_unregister returns void and it's reliable. The patch also adds a few needed but missing includes that would prevent kernel to compile after these changes on non-x86 archs (x86 didn't need them by luck). [akpm@linux-foundation.org: coding-style fixes] [akpm@linux-foundation.org: fix mm/filemap_xip.c build] [akpm@linux-foundation.org: fix mm/mmu_notifier.c build] Signed-off-by: Andrea Arcangeli <andrea@qumranet.com> Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Christoph Lameter <cl@linux-foundation.org> Cc: Jack Steiner <steiner@sgi.com> Cc: Robin Holt <holt@sgi.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Kanoj Sarcar <kanojsarcar@yahoo.com> Cc: Roland Dreier <rdreier@cisco.com> Cc: Steve Wise <swise@opengridcomputing.com> Cc: Avi Kivity <avi@qumranet.com> Cc: Hugh Dickins <hugh@veritas.com> Cc: Rusty Russell <rusty@rustcorp.com.au> Cc: Anthony Liguori <aliguori@us.ibm.com> Cc: Chris Wright <chrisw@redhat.com> Cc: Marcelo Tosatti <marcelo@kvack.org> Cc: Eric Dumazet <dada1@cosmosbay.com> Cc: "Paul E. McKenney" <paulmck@us.ibm.com> Cc: Izik Eidus <izike@qumranet.com> Cc: Anthony Liguori <aliguori@us.ibm.com> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-28 22:46:29 +00:00
mm: move all mmu notifier invocations to be done outside the PT lock In order to allow sleeping during mmu notifier calls, we need to avoid invoking them under the page table spinlock. This patch solves the problem by calling invalidate_page notification after releasing the lock (but before freeing the page itself), or by wrapping the page invalidation with calls to invalidate_range_begin and invalidate_range_end. To prevent accidental changes to the invalidate_range_end arguments after the call to invalidate_range_begin, the patch introduces a convention of saving the arguments in consistently named locals: unsigned long mmun_start; /* For mmu_notifiers */ unsigned long mmun_end; /* For mmu_notifiers */ ... mmun_start = ... mmun_end = ... mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); ... mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); The patch changes code to use this convention for all calls to mmu_notifier_invalidate_range_start/end, except those where the calls are close enough so that anyone who glances at the code can see the values aren't changing. This patchset is a preliminary step towards on-demand paging design to be added to the RDMA stack. Why do we want on-demand paging for Infiniband? Applications register memory with an RDMA adapter using system calls, and subsequently post IO operations that refer to the corresponding virtual addresses directly to HW. Until now, this was achieved by pinning the memory during the registration calls. The goal of on demand paging is to avoid pinning the pages of registered memory regions (MRs). This will allow users the same flexibility they get when swapping any other part of their processes address spaces. Instead of requiring the entire MR to fit in physical memory, we can allow the MR to be larger, and only fit the current working set in physical memory. Why should anyone care? What problems are users currently experiencing? This can make programming with RDMA much simpler. Today, developers that are working with more data than their RAM can hold need either to deregister and reregister memory regions throughout their process's life, or keep a single memory region and copy the data to it. On demand paging will allow these developers to register a single MR at the beginning of their process's life, and let the operating system manage which pages needs to be fetched at a given time. In the future, we might be able to provide a single memory access key for each process that would provide the entire process's address as one large memory region, and the developers wouldn't need to register memory regions at all. Is there any prospect that any other subsystems will utilise these infrastructural changes? If so, which and how, etc? As for other subsystems, I understand that XPMEM wanted to sleep in MMU notifiers, as Christoph Lameter wrote at http://lkml.indiana.edu/hypermail/linux/kernel/0802.1/0460.html and perhaps Andrea knows about other use cases. Scheduling in mmu notifications is required since we need to sync the hardware with the secondary page tables change. A TLB flush of an IO device is inherently slower than a CPU TLB flush, so our design works by sending the invalidation request to the device, and waiting for an interrupt before exiting the mmu notifier handler. Avi said: kvm may be a buyer. kvm::mmu_lock, which serializes guest page faults, also protects long operations such as destroying large ranges. It would be good to convert it into a spinlock, but as it is used inside mmu notifiers, this cannot be done. (there are alternatives, such as keeping the spinlock and using a generation counter to do the teardown in O(1), which is what the "may" is doing up there). [akpm@linux-foundation.orgpossible speed tweak in hugetlb_cow(), cleanups] Signed-off-by: Andrea Arcangeli <andrea@qumranet.com> Signed-off-by: Sagi Grimberg <sagig@mellanox.com> Signed-off-by: Haggai Eran <haggaie@mellanox.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com> Cc: Or Gerlitz <ogerlitz@mellanox.com> Cc: Haggai Eran <haggaie@mellanox.com> Cc: Shachar Raindel <raindel@mellanox.com> Cc: Liran Liss <liranl@mellanox.com> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Avi Kivity <avi@redhat.com> Cc: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-08 23:33:33 +00:00
if (is_cow)
mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end);
mmu-notifiers: core With KVM/GFP/XPMEM there isn't just the primary CPU MMU pointing to pages. There are secondary MMUs (with secondary sptes and secondary tlbs) too. sptes in the kvm case are shadow pagetables, but when I say spte in mmu-notifier context, I mean "secondary pte". In GRU case there's no actual secondary pte and there's only a secondary tlb because the GRU secondary MMU has no knowledge about sptes and every secondary tlb miss event in the MMU always generates a page fault that has to be resolved by the CPU (this is not the case of KVM where the a secondary tlb miss will walk sptes in hardware and it will refill the secondary tlb transparently to software if the corresponding spte is present). The same way zap_page_range has to invalidate the pte before freeing the page, the spte (and secondary tlb) must also be invalidated before any page is freed and reused. Currently we take a page_count pin on every page mapped by sptes, but that means the pages can't be swapped whenever they're mapped by any spte because they're part of the guest working set. Furthermore a spte unmap event can immediately lead to a page to be freed when the pin is released (so requiring the same complex and relatively slow tlb_gather smp safe logic we have in zap_page_range and that can be avoided completely if the spte unmap event doesn't require an unpin of the page previously mapped in the secondary MMU). The mmu notifiers allow kvm/GRU/XPMEM to attach to the tsk->mm and know when the VM is swapping or freeing or doing anything on the primary MMU so that the secondary MMU code can drop sptes before the pages are freed, avoiding all page pinning and allowing 100% reliable swapping of guest physical address space. Furthermore it avoids the code that teardown the mappings of the secondary MMU, to implement a logic like tlb_gather in zap_page_range that would require many IPI to flush other cpu tlbs, for each fixed number of spte unmapped. To make an example: if what happens on the primary MMU is a protection downgrade (from writeable to wrprotect) the secondary MMU mappings will be invalidated, and the next secondary-mmu-page-fault will call get_user_pages and trigger a do_wp_page through get_user_pages if it called get_user_pages with write=1, and it'll re-establishing an updated spte or secondary-tlb-mapping on the copied page. Or it will setup a readonly spte or readonly tlb mapping if it's a guest-read, if it calls get_user_pages with write=0. This is just an example. This allows to map any page pointed by any pte (and in turn visible in the primary CPU MMU), into a secondary MMU (be it a pure tlb like GRU, or an full MMU with both sptes and secondary-tlb like the shadow-pagetable layer with kvm), or a remote DMA in software like XPMEM (hence needing of schedule in XPMEM code to send the invalidate to the remote node, while no need to schedule in kvm/gru as it's an immediate event like invalidating primary-mmu pte). At least for KVM without this patch it's impossible to swap guests reliably. And having this feature and removing the page pin allows several other optimizations that simplify life considerably. Dependencies: 1) mm_take_all_locks() to register the mmu notifier when the whole VM isn't doing anything with "mm". This allows mmu notifier users to keep track if the VM is in the middle of the invalidate_range_begin/end critical section with an atomic counter incraese in range_begin and decreased in range_end. No secondary MMU page fault is allowed to map any spte or secondary tlb reference, while the VM is in the middle of range_begin/end as any page returned by get_user_pages in that critical section could later immediately be freed without any further ->invalidate_page notification (invalidate_range_begin/end works on ranges and ->invalidate_page isn't called immediately before freeing the page). To stop all page freeing and pagetable overwrites the mmap_sem must be taken in write mode and all other anon_vma/i_mmap locks must be taken too. 2) It'd be a waste to add branches in the VM if nobody could possibly run KVM/GRU/XPMEM on the kernel, so mmu notifiers will only enabled if CONFIG_KVM=m/y. In the current kernel kvm won't yet take advantage of mmu notifiers, but this already allows to compile a KVM external module against a kernel with mmu notifiers enabled and from the next pull from kvm.git we'll start using them. And GRU/XPMEM will also be able to continue the development by enabling KVM=m in their config, until they submit all GRU/XPMEM GPLv2 code to the mainline kernel. Then they can also enable MMU_NOTIFIERS in the same way KVM does it (even if KVM=n). This guarantees nobody selects MMU_NOTIFIER=y if KVM and GRU and XPMEM are all =n. The mmu_notifier_register call can fail because mm_take_all_locks may be interrupted by a signal and return -EINTR. Because mmu_notifier_reigster is used when a driver startup, a failure can be gracefully handled. Here an example of the change applied to kvm to register the mmu notifiers. Usually when a driver startups other allocations are required anyway and -ENOMEM failure paths exists already. struct kvm *kvm_arch_create_vm(void) { struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL); + int err; if (!kvm) return ERR_PTR(-ENOMEM); INIT_LIST_HEAD(&kvm->arch.active_mmu_pages); + kvm->arch.mmu_notifier.ops = &kvm_mmu_notifier_ops; + err = mmu_notifier_register(&kvm->arch.mmu_notifier, current->mm); + if (err) { + kfree(kvm); + return ERR_PTR(err); + } + return kvm; } mmu_notifier_unregister returns void and it's reliable. The patch also adds a few needed but missing includes that would prevent kernel to compile after these changes on non-x86 archs (x86 didn't need them by luck). [akpm@linux-foundation.org: coding-style fixes] [akpm@linux-foundation.org: fix mm/filemap_xip.c build] [akpm@linux-foundation.org: fix mm/mmu_notifier.c build] Signed-off-by: Andrea Arcangeli <andrea@qumranet.com> Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Christoph Lameter <cl@linux-foundation.org> Cc: Jack Steiner <steiner@sgi.com> Cc: Robin Holt <holt@sgi.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Kanoj Sarcar <kanojsarcar@yahoo.com> Cc: Roland Dreier <rdreier@cisco.com> Cc: Steve Wise <swise@opengridcomputing.com> Cc: Avi Kivity <avi@qumranet.com> Cc: Hugh Dickins <hugh@veritas.com> Cc: Rusty Russell <rusty@rustcorp.com.au> Cc: Anthony Liguori <aliguori@us.ibm.com> Cc: Chris Wright <chrisw@redhat.com> Cc: Marcelo Tosatti <marcelo@kvack.org> Cc: Eric Dumazet <dada1@cosmosbay.com> Cc: "Paul E. McKenney" <paulmck@us.ibm.com> Cc: Izik Eidus <izike@qumranet.com> Cc: Anthony Liguori <aliguori@us.ibm.com> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-28 22:46:29 +00:00
return ret;
}
static unsigned long zap_pte_range(struct mmu_gather *tlb,
2005-10-30 01:16:12 +00:00
struct vm_area_struct *vma, pmd_t *pmd,
unsigned long addr, unsigned long end,
struct zap_details *details)
{
2005-10-30 01:16:12 +00:00
struct mm_struct *mm = tlb->mm;
mm: mmu_gather rework Rework the existing mmu_gather infrastructure. The direct purpose of these patches was to allow preemptible mmu_gather, but even without that I think these patches provide an improvement to the status quo. The first 9 patches rework the mmu_gather infrastructure. For review purpose I've split them into generic and per-arch patches with the last of those a generic cleanup. The next patch provides generic RCU page-table freeing, and the followup is a patch converting s390 to use this. I've also got 4 patches from DaveM lined up (not included in this series) that uses this to implement gup_fast() for sparc64. Then there is one patch that extends the generic mmu_gather batching. After that follow the mm preemptibility patches, these make part of the mm a lot more preemptible. It converts i_mmap_lock and anon_vma->lock to mutexes which together with the mmu_gather rework makes mmu_gather preemptible as well. Making i_mmap_lock a mutex also enables a clean-up of the truncate code. This also allows for preemptible mmu_notifiers, something that XPMEM I think wants. Furthermore, it removes the new and universially detested unmap_mutex. This patch: Remove the first obstacle towards a fully preemptible mmu_gather. The current scheme assumes mmu_gather is always done with preemption disabled and uses per-cpu storage for the page batches. Change this to try and allocate a page for batching and in case of failure, use a small on-stack array to make some progress. Preemptible mmu_gather is desired in general and usable once i_mmap_lock becomes a mutex. Doing it before the mutex conversion saves us from having to rework the code by moving the mmu_gather bits inside the pte_lock. Also avoid flushing the tlb batches from under the pte lock, this is useful even without the i_mmap_lock conversion as it significantly reduces pte lock hold times. [akpm@linux-foundation.org: fix comment tpyo] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Miller <davem@davemloft.net> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Jeff Dike <jdike@addtoit.com> Cc: Richard Weinberger <richard@nod.at> Cc: Tony Luck <tony.luck@intel.com> Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Acked-by: Hugh Dickins <hughd@google.com> Acked-by: Mel Gorman <mel@csn.ul.ie> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Namhyung Kim <namhyung@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-05-25 00:11:45 +00:00
int force_flush = 0;
int rss[NR_MM_COUNTERS];
spinlock_t *ptl;
mm: fix wrong kunmap_atomic() pointer Running a ktest.pl test, I hit the following bug on x86_32: ------------[ cut here ]------------ WARNING: at arch/x86/mm/highmem_32.c:81 __kunmap_atomic+0x64/0xc1() Hardware name: Modules linked in: Pid: 93, comm: sh Not tainted 2.6.39-test+ #1 Call Trace: [<c04450da>] warn_slowpath_common+0x7c/0x91 [<c042f5df>] ? __kunmap_atomic+0x64/0xc1 [<c042f5df>] ? __kunmap_atomic+0x64/0xc1^M [<c0445111>] warn_slowpath_null+0x22/0x24 [<c042f5df>] __kunmap_atomic+0x64/0xc1 [<c04d4a22>] unmap_vmas+0x43a/0x4e0 [<c04d9065>] exit_mmap+0x91/0xd2 [<c0443057>] mmput+0x43/0xad [<c0448358>] exit_mm+0x111/0x119 [<c044855f>] do_exit+0x1ff/0x5fa [<c0454ea2>] ? set_current_blocked+0x3c/0x40 [<c0454f24>] ? sigprocmask+0x7e/0x8e [<c0448b55>] do_group_exit+0x65/0x88 [<c0448b90>] sys_exit_group+0x18/0x1c [<c0c3915f>] sysenter_do_call+0x12/0x38 ---[ end trace 8055f74ea3c0eb62 ]--- Running a ktest.pl git bisect, found the culprit: commit e303297e6c3a ("mm: extended batches for generic mmu_gather") But although this was the commit triggering the bug, it was not the one originally responsible for the bug. That was commit d16dfc550f53 ("mm: mmu_gather rework"). The code in zap_pte_range() has something that looks like the following: pte = pte_offset_map_lock(mm, pmd, addr, &ptl); do { [...] } while (pte++, addr += PAGE_SIZE, addr != end); pte_unmap_unlock(pte - 1, ptl); The pte starts off pointing at the first element in the page table directory that was returned by the pte_offset_map_lock(). When it's done with the page, pte will be pointing to anything between the next entry and the first entry of the next page inclusive. By doing a pte - 1, this puts the pte back onto the original page, which is all that pte_unmap_unlock() needs. In most archs (64 bit), this is not an issue as the pte is ignored in the pte_unmap_unlock(). But on 32 bit archs, where things may be kmapped, it is essential that the pte passed to pte_unmap_unlock() resides on the same page that was given by pte_offest_map_lock(). The problem came in d16dfc55 ("mm: mmu_gather rework") where it introduced a "break;" from the while loop. This alone did not seem to easily trigger the bug. But the modifications made by e303297e6 caused that "break;" to be hit on the first iteration, before the pte++. The pte not being incremented will now cause pte_unmap_unlock(pte - 1) to be pointing to the previous page. This will cause the wrong page to be unmapped, and also trigger the warning above. The simple solution is to just save the pointer given by pte_offset_map_lock() and use it in the unlock. Signed-off-by: Steven Rostedt <rostedt@goodmis.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Acked-by: Hugh Dickins <hughd@google.com> Cc: Mel Gorman <mel@csn.ul.ie> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-06-15 22:08:23 +00:00
pte_t *start_pte;
pte_t *pte;
mm: mmu_gather rework Rework the existing mmu_gather infrastructure. The direct purpose of these patches was to allow preemptible mmu_gather, but even without that I think these patches provide an improvement to the status quo. The first 9 patches rework the mmu_gather infrastructure. For review purpose I've split them into generic and per-arch patches with the last of those a generic cleanup. The next patch provides generic RCU page-table freeing, and the followup is a patch converting s390 to use this. I've also got 4 patches from DaveM lined up (not included in this series) that uses this to implement gup_fast() for sparc64. Then there is one patch that extends the generic mmu_gather batching. After that follow the mm preemptibility patches, these make part of the mm a lot more preemptible. It converts i_mmap_lock and anon_vma->lock to mutexes which together with the mmu_gather rework makes mmu_gather preemptible as well. Making i_mmap_lock a mutex also enables a clean-up of the truncate code. This also allows for preemptible mmu_notifiers, something that XPMEM I think wants. Furthermore, it removes the new and universially detested unmap_mutex. This patch: Remove the first obstacle towards a fully preemptible mmu_gather. The current scheme assumes mmu_gather is always done with preemption disabled and uses per-cpu storage for the page batches. Change this to try and allocate a page for batching and in case of failure, use a small on-stack array to make some progress. Preemptible mmu_gather is desired in general and usable once i_mmap_lock becomes a mutex. Doing it before the mutex conversion saves us from having to rework the code by moving the mmu_gather bits inside the pte_lock. Also avoid flushing the tlb batches from under the pte lock, this is useful even without the i_mmap_lock conversion as it significantly reduces pte lock hold times. [akpm@linux-foundation.org: fix comment tpyo] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Miller <davem@davemloft.net> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Jeff Dike <jdike@addtoit.com> Cc: Richard Weinberger <richard@nod.at> Cc: Tony Luck <tony.luck@intel.com> Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Acked-by: Hugh Dickins <hughd@google.com> Acked-by: Mel Gorman <mel@csn.ul.ie> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Namhyung Kim <namhyung@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-05-25 00:11:45 +00:00
again:
init_rss_vec(rss);
mm: fix wrong kunmap_atomic() pointer Running a ktest.pl test, I hit the following bug on x86_32: ------------[ cut here ]------------ WARNING: at arch/x86/mm/highmem_32.c:81 __kunmap_atomic+0x64/0xc1() Hardware name: Modules linked in: Pid: 93, comm: sh Not tainted 2.6.39-test+ #1 Call Trace: [<c04450da>] warn_slowpath_common+0x7c/0x91 [<c042f5df>] ? __kunmap_atomic+0x64/0xc1 [<c042f5df>] ? __kunmap_atomic+0x64/0xc1^M [<c0445111>] warn_slowpath_null+0x22/0x24 [<c042f5df>] __kunmap_atomic+0x64/0xc1 [<c04d4a22>] unmap_vmas+0x43a/0x4e0 [<c04d9065>] exit_mmap+0x91/0xd2 [<c0443057>] mmput+0x43/0xad [<c0448358>] exit_mm+0x111/0x119 [<c044855f>] do_exit+0x1ff/0x5fa [<c0454ea2>] ? set_current_blocked+0x3c/0x40 [<c0454f24>] ? sigprocmask+0x7e/0x8e [<c0448b55>] do_group_exit+0x65/0x88 [<c0448b90>] sys_exit_group+0x18/0x1c [<c0c3915f>] sysenter_do_call+0x12/0x38 ---[ end trace 8055f74ea3c0eb62 ]--- Running a ktest.pl git bisect, found the culprit: commit e303297e6c3a ("mm: extended batches for generic mmu_gather") But although this was the commit triggering the bug, it was not the one originally responsible for the bug. That was commit d16dfc550f53 ("mm: mmu_gather rework"). The code in zap_pte_range() has something that looks like the following: pte = pte_offset_map_lock(mm, pmd, addr, &ptl); do { [...] } while (pte++, addr += PAGE_SIZE, addr != end); pte_unmap_unlock(pte - 1, ptl); The pte starts off pointing at the first element in the page table directory that was returned by the pte_offset_map_lock(). When it's done with the page, pte will be pointing to anything between the next entry and the first entry of the next page inclusive. By doing a pte - 1, this puts the pte back onto the original page, which is all that pte_unmap_unlock() needs. In most archs (64 bit), this is not an issue as the pte is ignored in the pte_unmap_unlock(). But on 32 bit archs, where things may be kmapped, it is essential that the pte passed to pte_unmap_unlock() resides on the same page that was given by pte_offest_map_lock(). The problem came in d16dfc55 ("mm: mmu_gather rework") where it introduced a "break;" from the while loop. This alone did not seem to easily trigger the bug. But the modifications made by e303297e6 caused that "break;" to be hit on the first iteration, before the pte++. The pte not being incremented will now cause pte_unmap_unlock(pte - 1) to be pointing to the previous page. This will cause the wrong page to be unmapped, and also trigger the warning above. The simple solution is to just save the pointer given by pte_offset_map_lock() and use it in the unlock. Signed-off-by: Steven Rostedt <rostedt@goodmis.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Acked-by: Hugh Dickins <hughd@google.com> Cc: Mel Gorman <mel@csn.ul.ie> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-06-15 22:08:23 +00:00
start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
pte = start_pte;
arch_enter_lazy_mmu_mode();
do {
pte_t ptent = *pte;
if (pte_none(ptent)) {
continue;
}
if (pte_present(ptent)) {
struct page *page;
page = vm_normal_page(vma, addr, ptent);
if (unlikely(details) && page) {
/*
* unmap_shared_mapping_pages() wants to
* invalidate cache without truncating:
* unmap shared but keep private pages.
*/
if (details->check_mapping &&
details->check_mapping != page->mapping)
continue;
/*
* Each page->index must be checked when
* invalidating or truncating nonlinear.
*/
if (details->nonlinear_vma &&
(page->index < details->first_index ||
page->index > details->last_index))
continue;
}
2005-10-30 01:16:12 +00:00
ptent = ptep_get_and_clear_full(mm, addr, pte,
2005-09-03 22:55:04 +00:00
tlb->fullmm);
tlb_remove_tlb_entry(tlb, pte, addr);
if (unlikely(!page))
continue;
if (unlikely(details) && details->nonlinear_vma
&& linear_page_index(details->nonlinear_vma,
addr) != page->index)
2005-10-30 01:16:12 +00:00
set_pte_at(mm, addr, pte,
pgoff_to_pte(page->index));
if (PageAnon(page))
rss[MM_ANONPAGES]--;
else {
if (pte_dirty(ptent))
set_page_dirty(page);
if (pte_young(ptent) &&
likely(!VM_SequentialReadHint(vma)))
mark_page_accessed(page);
rss[MM_FILEPAGES]--;
}
page_remove_rmap(page);
badpage: replace page_remove_rmap Eeek and BUG Now that bad pages are kept out of circulation, there is no need for the infamous page_remove_rmap() BUG() - once that page is freed, its negative mapcount will issue a "Bad page state" message and the page won't be freed. Removing the BUG() allows more info, on subsequent pages, to be gathered. We do have more info about the page at this point than bad_page() can know - notably, what the pmd is, which might pinpoint something like low 64kB corruption - but page_remove_rmap() isn't given the address to find that. In practice, there is only one call to page_remove_rmap() which has ever reported anything, that from zap_pte_range() (usually on exit, sometimes on munmap). It has all the info, so remove page_remove_rmap()'s "Eeek" message and leave it all to zap_pte_range(). mm/memory.c already has a hardly used print_bad_pte() function, showing some of the appropriate info: extend it to show what we want for the rmap case: pte info, page info (when there is a page) and vma info to compare. zap_pte_range() already knows the pmd, but print_bad_pte() is easier to use if it works that out for itself. Some of this info is also shown in bad_page()'s "Bad page state" message. Keep them separate, but adjust them to match each other as far as possible. Say "Bad page map" in print_bad_pte(), and add a TAINT_BAD_PAGE there too. print_bad_pte() show current->comm unconditionally (though it should get repeated in the usually irrelevant stack trace): sorry, I misled Nick Piggin to make it conditional on vm_mm == current->mm, but current->mm is already NULL in the exit case. Usually current->comm is good, though exceptionally it may not be that of the mm (when "swapoff" for example). Signed-off-by: Hugh Dickins <hugh@veritas.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:40:08 +00:00
if (unlikely(page_mapcount(page) < 0))
print_bad_pte(vma, addr, ptent, page);
mm: mmu_gather rework Rework the existing mmu_gather infrastructure. The direct purpose of these patches was to allow preemptible mmu_gather, but even without that I think these patches provide an improvement to the status quo. The first 9 patches rework the mmu_gather infrastructure. For review purpose I've split them into generic and per-arch patches with the last of those a generic cleanup. The next patch provides generic RCU page-table freeing, and the followup is a patch converting s390 to use this. I've also got 4 patches from DaveM lined up (not included in this series) that uses this to implement gup_fast() for sparc64. Then there is one patch that extends the generic mmu_gather batching. After that follow the mm preemptibility patches, these make part of the mm a lot more preemptible. It converts i_mmap_lock and anon_vma->lock to mutexes which together with the mmu_gather rework makes mmu_gather preemptible as well. Making i_mmap_lock a mutex also enables a clean-up of the truncate code. This also allows for preemptible mmu_notifiers, something that XPMEM I think wants. Furthermore, it removes the new and universially detested unmap_mutex. This patch: Remove the first obstacle towards a fully preemptible mmu_gather. The current scheme assumes mmu_gather is always done with preemption disabled and uses per-cpu storage for the page batches. Change this to try and allocate a page for batching and in case of failure, use a small on-stack array to make some progress. Preemptible mmu_gather is desired in general and usable once i_mmap_lock becomes a mutex. Doing it before the mutex conversion saves us from having to rework the code by moving the mmu_gather bits inside the pte_lock. Also avoid flushing the tlb batches from under the pte lock, this is useful even without the i_mmap_lock conversion as it significantly reduces pte lock hold times. [akpm@linux-foundation.org: fix comment tpyo] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Miller <davem@davemloft.net> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Jeff Dike <jdike@addtoit.com> Cc: Richard Weinberger <richard@nod.at> Cc: Tony Luck <tony.luck@intel.com> Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Acked-by: Hugh Dickins <hughd@google.com> Acked-by: Mel Gorman <mel@csn.ul.ie> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Namhyung Kim <namhyung@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-05-25 00:11:45 +00:00
force_flush = !__tlb_remove_page(tlb, page);
if (force_flush)
break;
continue;
}
/*
* If details->check_mapping, we leave swap entries;
* if details->nonlinear_vma, we leave file entries.
*/
if (unlikely(details))
continue;
if (pte_file(ptent)) {
if (unlikely(!(vma->vm_flags & VM_NONLINEAR)))
print_bad_pte(vma, addr, ptent, NULL);
} else {
swp_entry_t entry = pte_to_swp_entry(ptent);
if (!non_swap_entry(entry))
rss[MM_SWAPENTS]--;
else if (is_migration_entry(entry)) {
struct page *page;
page = migration_entry_to_page(entry);
if (PageAnon(page))
rss[MM_ANONPAGES]--;
else
rss[MM_FILEPAGES]--;
}
if (unlikely(!free_swap_and_cache(entry)))
print_bad_pte(vma, addr, ptent, NULL);
}
pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
} while (pte++, addr += PAGE_SIZE, addr != end);
add_mm_rss_vec(mm, rss);
arch_leave_lazy_mmu_mode();
mm: fix wrong kunmap_atomic() pointer Running a ktest.pl test, I hit the following bug on x86_32: ------------[ cut here ]------------ WARNING: at arch/x86/mm/highmem_32.c:81 __kunmap_atomic+0x64/0xc1() Hardware name: Modules linked in: Pid: 93, comm: sh Not tainted 2.6.39-test+ #1 Call Trace: [<c04450da>] warn_slowpath_common+0x7c/0x91 [<c042f5df>] ? __kunmap_atomic+0x64/0xc1 [<c042f5df>] ? __kunmap_atomic+0x64/0xc1^M [<c0445111>] warn_slowpath_null+0x22/0x24 [<c042f5df>] __kunmap_atomic+0x64/0xc1 [<c04d4a22>] unmap_vmas+0x43a/0x4e0 [<c04d9065>] exit_mmap+0x91/0xd2 [<c0443057>] mmput+0x43/0xad [<c0448358>] exit_mm+0x111/0x119 [<c044855f>] do_exit+0x1ff/0x5fa [<c0454ea2>] ? set_current_blocked+0x3c/0x40 [<c0454f24>] ? sigprocmask+0x7e/0x8e [<c0448b55>] do_group_exit+0x65/0x88 [<c0448b90>] sys_exit_group+0x18/0x1c [<c0c3915f>] sysenter_do_call+0x12/0x38 ---[ end trace 8055f74ea3c0eb62 ]--- Running a ktest.pl git bisect, found the culprit: commit e303297e6c3a ("mm: extended batches for generic mmu_gather") But although this was the commit triggering the bug, it was not the one originally responsible for the bug. That was commit d16dfc550f53 ("mm: mmu_gather rework"). The code in zap_pte_range() has something that looks like the following: pte = pte_offset_map_lock(mm, pmd, addr, &ptl); do { [...] } while (pte++, addr += PAGE_SIZE, addr != end); pte_unmap_unlock(pte - 1, ptl); The pte starts off pointing at the first element in the page table directory that was returned by the pte_offset_map_lock(). When it's done with the page, pte will be pointing to anything between the next entry and the first entry of the next page inclusive. By doing a pte - 1, this puts the pte back onto the original page, which is all that pte_unmap_unlock() needs. In most archs (64 bit), this is not an issue as the pte is ignored in the pte_unmap_unlock(). But on 32 bit archs, where things may be kmapped, it is essential that the pte passed to pte_unmap_unlock() resides on the same page that was given by pte_offest_map_lock(). The problem came in d16dfc55 ("mm: mmu_gather rework") where it introduced a "break;" from the while loop. This alone did not seem to easily trigger the bug. But the modifications made by e303297e6 caused that "break;" to be hit on the first iteration, before the pte++. The pte not being incremented will now cause pte_unmap_unlock(pte - 1) to be pointing to the previous page. This will cause the wrong page to be unmapped, and also trigger the warning above. The simple solution is to just save the pointer given by pte_offset_map_lock() and use it in the unlock. Signed-off-by: Steven Rostedt <rostedt@goodmis.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Acked-by: Hugh Dickins <hughd@google.com> Cc: Mel Gorman <mel@csn.ul.ie> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-06-15 22:08:23 +00:00
pte_unmap_unlock(start_pte, ptl);
mm: mmu_gather rework Rework the existing mmu_gather infrastructure. The direct purpose of these patches was to allow preemptible mmu_gather, but even without that I think these patches provide an improvement to the status quo. The first 9 patches rework the mmu_gather infrastructure. For review purpose I've split them into generic and per-arch patches with the last of those a generic cleanup. The next patch provides generic RCU page-table freeing, and the followup is a patch converting s390 to use this. I've also got 4 patches from DaveM lined up (not included in this series) that uses this to implement gup_fast() for sparc64. Then there is one patch that extends the generic mmu_gather batching. After that follow the mm preemptibility patches, these make part of the mm a lot more preemptible. It converts i_mmap_lock and anon_vma->lock to mutexes which together with the mmu_gather rework makes mmu_gather preemptible as well. Making i_mmap_lock a mutex also enables a clean-up of the truncate code. This also allows for preemptible mmu_notifiers, something that XPMEM I think wants. Furthermore, it removes the new and universially detested unmap_mutex. This patch: Remove the first obstacle towards a fully preemptible mmu_gather. The current scheme assumes mmu_gather is always done with preemption disabled and uses per-cpu storage for the page batches. Change this to try and allocate a page for batching and in case of failure, use a small on-stack array to make some progress. Preemptible mmu_gather is desired in general and usable once i_mmap_lock becomes a mutex. Doing it before the mutex conversion saves us from having to rework the code by moving the mmu_gather bits inside the pte_lock. Also avoid flushing the tlb batches from under the pte lock, this is useful even without the i_mmap_lock conversion as it significantly reduces pte lock hold times. [akpm@linux-foundation.org: fix comment tpyo] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Miller <davem@davemloft.net> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Jeff Dike <jdike@addtoit.com> Cc: Richard Weinberger <richard@nod.at> Cc: Tony Luck <tony.luck@intel.com> Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Acked-by: Hugh Dickins <hughd@google.com> Acked-by: Mel Gorman <mel@csn.ul.ie> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Namhyung Kim <namhyung@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-05-25 00:11:45 +00:00
/*
* mmu_gather ran out of room to batch pages, we break out of
* the PTE lock to avoid doing the potential expensive TLB invalidate
* and page-free while holding it.
*/
if (force_flush) {
force_flush = 0;
#ifdef HAVE_GENERIC_MMU_GATHER
tlb->start = addr;
tlb->end = end;
#endif
mm: mmu_gather rework Rework the existing mmu_gather infrastructure. The direct purpose of these patches was to allow preemptible mmu_gather, but even without that I think these patches provide an improvement to the status quo. The first 9 patches rework the mmu_gather infrastructure. For review purpose I've split them into generic and per-arch patches with the last of those a generic cleanup. The next patch provides generic RCU page-table freeing, and the followup is a patch converting s390 to use this. I've also got 4 patches from DaveM lined up (not included in this series) that uses this to implement gup_fast() for sparc64. Then there is one patch that extends the generic mmu_gather batching. After that follow the mm preemptibility patches, these make part of the mm a lot more preemptible. It converts i_mmap_lock and anon_vma->lock to mutexes which together with the mmu_gather rework makes mmu_gather preemptible as well. Making i_mmap_lock a mutex also enables a clean-up of the truncate code. This also allows for preemptible mmu_notifiers, something that XPMEM I think wants. Furthermore, it removes the new and universially detested unmap_mutex. This patch: Remove the first obstacle towards a fully preemptible mmu_gather. The current scheme assumes mmu_gather is always done with preemption disabled and uses per-cpu storage for the page batches. Change this to try and allocate a page for batching and in case of failure, use a small on-stack array to make some progress. Preemptible mmu_gather is desired in general and usable once i_mmap_lock becomes a mutex. Doing it before the mutex conversion saves us from having to rework the code by moving the mmu_gather bits inside the pte_lock. Also avoid flushing the tlb batches from under the pte lock, this is useful even without the i_mmap_lock conversion as it significantly reduces pte lock hold times. [akpm@linux-foundation.org: fix comment tpyo] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Miller <davem@davemloft.net> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Jeff Dike <jdike@addtoit.com> Cc: Richard Weinberger <richard@nod.at> Cc: Tony Luck <tony.luck@intel.com> Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Acked-by: Hugh Dickins <hughd@google.com> Acked-by: Mel Gorman <mel@csn.ul.ie> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Namhyung Kim <namhyung@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-05-25 00:11:45 +00:00
tlb_flush_mmu(tlb);
if (addr != end)
goto again;
}
return addr;
}
static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
2005-10-30 01:16:12 +00:00
struct vm_area_struct *vma, pud_t *pud,
unsigned long addr, unsigned long end,
struct zap_details *details)
{
pmd_t *pmd;
unsigned long next;
pmd = pmd_offset(pud, addr);
do {
next = pmd_addr_end(addr, end);
thp: transparent hugepage core Lately I've been working to make KVM use hugepages transparently without the usual restrictions of hugetlbfs. Some of the restrictions I'd like to see removed: 1) hugepages have to be swappable or the guest physical memory remains locked in RAM and can't be paged out to swap 2) if a hugepage allocation fails, regular pages should be allocated instead and mixed in the same vma without any failure and without userland noticing 3) if some task quits and more hugepages become available in the buddy, guest physical memory backed by regular pages should be relocated on hugepages automatically in regions under madvise(MADV_HUGEPAGE) (ideally event driven by waking up the kernel deamon if the order=HPAGE_PMD_SHIFT-PAGE_SHIFT list becomes not null) 4) avoidance of reservation and maximization of use of hugepages whenever possible. Reservation (needed to avoid runtime fatal faliures) may be ok for 1 machine with 1 database with 1 database cache with 1 database cache size known at boot time. It's definitely not feasible with a virtualization hypervisor usage like RHEV-H that runs an unknown number of virtual machines with an unknown size of each virtual machine with an unknown amount of pagecache that could be potentially useful in the host for guest not using O_DIRECT (aka cache=off). hugepages in the virtualization hypervisor (and also in the guest!) are much more important than in a regular host not using virtualization, becasue with NPT/EPT they decrease the tlb-miss cacheline accesses from 24 to 19 in case only the hypervisor uses transparent hugepages, and they decrease the tlb-miss cacheline accesses from 19 to 15 in case both the linux hypervisor and the linux guest both uses this patch (though the guest will limit the addition speedup to anonymous regions only for now...). Even more important is that the tlb miss handler is much slower on a NPT/EPT guest than for a regular shadow paging or no-virtualization scenario. So maximizing the amount of virtual memory cached by the TLB pays off significantly more with NPT/EPT than without (even if there would be no significant speedup in the tlb-miss runtime). The first (and more tedious) part of this work requires allowing the VM to handle anonymous hugepages mixed with regular pages transparently on regular anonymous vmas. This is what this patch tries to achieve in the least intrusive possible way. We want hugepages and hugetlb to be used in a way so that all applications can benefit without changes (as usual we leverage the KVM virtualization design: by improving the Linux VM at large, KVM gets the performance boost too). The most important design choice is: always fallback to 4k allocation if the hugepage allocation fails! This is the _very_ opposite of some large pagecache patches that failed with -EIO back then if a 64k (or similar) allocation failed... Second important decision (to reduce the impact of the feature on the existing pagetable handling code) is that at any time we can split an hugepage into 512 regular pages and it has to be done with an operation that can't fail. This way the reliability of the swapping isn't decreased (no need to allocate memory when we are short on memory to swap) and it's trivial to plug a split_huge_page* one-liner where needed without polluting the VM. Over time we can teach mprotect, mremap and friends to handle pmd_trans_huge natively without calling split_huge_page*. The fact it can't fail isn't just for swap: if split_huge_page would return -ENOMEM (instead of the current void) we'd need to rollback the mprotect from the middle of it (ideally including undoing the split_vma) which would be a big change and in the very wrong direction (it'd likely be simpler not to call split_huge_page at all and to teach mprotect and friends to handle hugepages instead of rolling them back from the middle). In short the very value of split_huge_page is that it can't fail. The collapsing and madvise(MADV_HUGEPAGE) part will remain separated and incremental and it'll just be an "harmless" addition later if this initial part is agreed upon. It also should be noted that locking-wise replacing regular pages with hugepages is going to be very easy if compared to what I'm doing below in split_huge_page, as it will only happen when page_count(page) matches page_mapcount(page) if we can take the PG_lock and mmap_sem in write mode. collapse_huge_page will be a "best effort" that (unlike split_huge_page) can fail at the minimal sign of trouble and we can try again later. collapse_huge_page will be similar to how KSM works and the madvise(MADV_HUGEPAGE) will work similar to madvise(MADV_MERGEABLE). The default I like is that transparent hugepages are used at page fault time. This can be changed with /sys/kernel/mm/transparent_hugepage/enabled. The control knob can be set to three values "always", "madvise", "never" which mean respectively that hugepages are always used, or only inside madvise(MADV_HUGEPAGE) regions, or never used. /sys/kernel/mm/transparent_hugepage/defrag instead controls if the hugepage allocation should defrag memory aggressively "always", only inside "madvise" regions, or "never". The pmd_trans_splitting/pmd_trans_huge locking is very solid. The put_page (from get_user_page users that can't use mmu notifier like O_DIRECT) that runs against a __split_huge_page_refcount instead was a pain to serialize in a way that would result always in a coherent page count for both tail and head. I think my locking solution with a compound_lock taken only after the page_first is valid and is still a PageHead should be safe but it surely needs review from SMP race point of view. In short there is no current existing way to serialize the O_DIRECT final put_page against split_huge_page_refcount so I had to invent a new one (O_DIRECT loses knowledge on the mapping status by the time gup_fast returns so...). And I didn't want to impact all gup/gup_fast users for now, maybe if we change the gup interface substantially we can avoid this locking, I admit I didn't think too much about it because changing the gup unpinning interface would be invasive. If we ignored O_DIRECT we could stick to the existing compound refcounting code, by simply adding a get_user_pages_fast_flags(foll_flags) where KVM (and any other mmu notifier user) would call it without FOLL_GET (and if FOLL_GET isn't set we'd just BUG_ON if nobody registered itself in the current task mmu notifier list yet). But O_DIRECT is fundamental for decent performance of virtualized I/O on fast storage so we can't avoid it to solve the race of put_page against split_huge_page_refcount to achieve a complete hugepage feature for KVM. Swap and oom works fine (well just like with regular pages ;). MMU notifier is handled transparently too, with the exception of the young bit on the pmd, that didn't have a range check but I think KVM will be fine because the whole point of hugepages is that EPT/NPT will also use a huge pmd when they notice gup returns pages with PageCompound set, so they won't care of a range and there's just the pmd young bit to check in that case. NOTE: in some cases if the L2 cache is small, this may slowdown and waste memory during COWs because 4M of memory are accessed in a single fault instead of 8k (the payoff is that after COW the program can run faster). So we might want to switch the copy_huge_page (and clear_huge_page too) to not temporal stores. I also extensively researched ways to avoid this cache trashing with a full prefault logic that would cow in 8k/16k/32k/64k up to 1M (I can send those patches that fully implemented prefault) but I concluded they're not worth it and they add an huge additional complexity and they remove all tlb benefits until the full hugepage has been faulted in, to save a little bit of memory and some cache during app startup, but they still don't improve substantially the cache-trashing during startup if the prefault happens in >4k chunks. One reason is that those 4k pte entries copied are still mapped on a perfectly cache-colored hugepage, so the trashing is the worst one can generate in those copies (cow of 4k page copies aren't so well colored so they trashes less, but again this results in software running faster after the page fault). Those prefault patches allowed things like a pte where post-cow pages were local 4k regular anon pages and the not-yet-cowed pte entries were pointing in the middle of some hugepage mapped read-only. If it doesn't payoff substantially with todays hardware it will payoff even less in the future with larger l2 caches, and the prefault logic would blot the VM a lot. If one is emebdded transparent_hugepage can be disabled during boot with sysfs or with the boot commandline parameter transparent_hugepage=0 (or transparent_hugepage=2 to restrict hugepages inside madvise regions) that will ensure not a single hugepage is allocated at boot time. It is simple enough to just disable transparent hugepage globally and let transparent hugepages be allocated selectively by applications in the MADV_HUGEPAGE region (both at page fault time, and if enabled with the collapse_huge_page too through the kernel daemon). This patch supports only hugepages mapped in the pmd, archs that have smaller hugepages will not fit in this patch alone. Also some archs like power have certain tlb limits that prevents mixing different page size in the same regions so they will not fit in this framework that requires "graceful fallback" to basic PAGE_SIZE in case of physical memory fragmentation. hugetlbfs remains a perfect fit for those because its software limits happen to match the hardware limits. hugetlbfs also remains a perfect fit for hugepage sizes like 1GByte that cannot be hoped to be found not fragmented after a certain system uptime and that would be very expensive to defragment with relocation, so requiring reservation. hugetlbfs is the "reservation way", the point of transparent hugepages is not to have any reservation at all and maximizing the use of cache and hugepages at all times automatically. Some performance result: vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largep ages3 memset page fault 1566023 memset tlb miss 453854 memset second tlb miss 453321 random access tlb miss 41635 random access second tlb miss 41658 vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largepages3 memset page fault 1566471 memset tlb miss 453375 memset second tlb miss 453320 random access tlb miss 41636 random access second tlb miss 41637 vmx andrea # ./largepages3 memset page fault 1566642 memset tlb miss 453417 memset second tlb miss 453313 random access tlb miss 41630 random access second tlb miss 41647 vmx andrea # ./largepages3 memset page fault 1566872 memset tlb miss 453418 memset second tlb miss 453315 random access tlb miss 41618 random access second tlb miss 41659 vmx andrea # echo 0 > /proc/sys/vm/transparent_hugepage vmx andrea # ./largepages3 memset page fault 2182476 memset tlb miss 460305 memset second tlb miss 460179 random access tlb miss 44483 random access second tlb miss 44186 vmx andrea # ./largepages3 memset page fault 2182791 memset tlb miss 460742 memset second tlb miss 459962 random access tlb miss 43981 random access second tlb miss 43988 ============ #include <stdio.h> #include <stdlib.h> #include <string.h> #include <sys/time.h> #define SIZE (3UL*1024*1024*1024) int main() { char *p = malloc(SIZE), *p2; struct timeval before, after; gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset page fault %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); return 0; } ============ Signed-off-by: Andrea Arcangeli <aarcange@redhat.com> Acked-by: Rik van Riel <riel@redhat.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-13 23:46:52 +00:00
if (pmd_trans_huge(*pmd)) {
mm: thp: fix pmd_bad() triggering in code paths holding mmap_sem read mode In some cases it may happen that pmd_none_or_clear_bad() is called with the mmap_sem hold in read mode. In those cases the huge page faults can allocate hugepmds under pmd_none_or_clear_bad() and that can trigger a false positive from pmd_bad() that will not like to see a pmd materializing as trans huge. It's not khugepaged causing the problem, khugepaged holds the mmap_sem in write mode (and all those sites must hold the mmap_sem in read mode to prevent pagetables to go away from under them, during code review it seems vm86 mode on 32bit kernels requires that too unless it's restricted to 1 thread per process or UP builds). The race is only with the huge pagefaults that can convert a pmd_none() into a pmd_trans_huge(). Effectively all these pmd_none_or_clear_bad() sites running with mmap_sem in read mode are somewhat speculative with the page faults, and the result is always undefined when they run simultaneously. This is probably why it wasn't common to run into this. For example if the madvise(MADV_DONTNEED) runs zap_page_range() shortly before the page fault, the hugepage will not be zapped, if the page fault runs first it will be zapped. Altering pmd_bad() not to error out if it finds hugepmds won't be enough to fix this, because zap_pmd_range would then proceed to call zap_pte_range (which would be incorrect if the pmd become a pmd_trans_huge()). The simplest way to fix this is to read the pmd in the local stack (regardless of what we read, no need of actual CPU barriers, only compiler barrier needed), and be sure it is not changing under the code that computes its value. Even if the real pmd is changing under the value we hold on the stack, we don't care. If we actually end up in zap_pte_range it means the pmd was not none already and it was not huge, and it can't become huge from under us (khugepaged locking explained above). All we need is to enforce that there is no way anymore that in a code path like below, pmd_trans_huge can be false, but pmd_none_or_clear_bad can run into a hugepmd. The overhead of a barrier() is just a compiler tweak and should not be measurable (I only added it for THP builds). I don't exclude different compiler versions may have prevented the race too by caching the value of *pmd on the stack (that hasn't been verified, but it wouldn't be impossible considering pmd_none_or_clear_bad, pmd_bad, pmd_trans_huge, pmd_none are all inlines and there's no external function called in between pmd_trans_huge and pmd_none_or_clear_bad). if (pmd_trans_huge(*pmd)) { if (next-addr != HPAGE_PMD_SIZE) { VM_BUG_ON(!rwsem_is_locked(&tlb->mm->mmap_sem)); split_huge_page_pmd(vma->vm_mm, pmd); } else if (zap_huge_pmd(tlb, vma, pmd, addr)) continue; /* fall through */ } if (pmd_none_or_clear_bad(pmd)) Because this race condition could be exercised without special privileges this was reported in CVE-2012-1179. The race was identified and fully explained by Ulrich who debugged it. I'm quoting his accurate explanation below, for reference. ====== start quote ======= mapcount 0 page_mapcount 1 kernel BUG at mm/huge_memory.c:1384! At some point prior to the panic, a "bad pmd ..." message similar to the following is logged on the console: mm/memory.c:145: bad pmd ffff8800376e1f98(80000000314000e7). The "bad pmd ..." message is logged by pmd_clear_bad() before it clears the page's PMD table entry. 143 void pmd_clear_bad(pmd_t *pmd) 144 { -> 145 pmd_ERROR(*pmd); 146 pmd_clear(pmd); 147 } After the PMD table entry has been cleared, there is an inconsistency between the actual number of PMD table entries that are mapping the page and the page's map count (_mapcount field in struct page). When the page is subsequently reclaimed, __split_huge_page() detects this inconsistency. 1381 if (mapcount != page_mapcount(page)) 1382 printk(KERN_ERR "mapcount %d page_mapcount %d\n", 1383 mapcount, page_mapcount(page)); -> 1384 BUG_ON(mapcount != page_mapcount(page)); The root cause of the problem is a race of two threads in a multithreaded process. Thread B incurs a page fault on a virtual address that has never been accessed (PMD entry is zero) while Thread A is executing an madvise() system call on a virtual address within the same 2 MB (huge page) range. virtual address space .---------------------. | | | | .-|---------------------| | | | | | |<-- B(fault) | | | 2 MB | |/////////////////////|-. huge < |/////////////////////| > A(range) page | |/////////////////////|-' | | | | | | '-|---------------------| | | | | '---------------------' - Thread A is executing an madvise(..., MADV_DONTNEED) system call on the virtual address range "A(range)" shown in the picture. sys_madvise // Acquire the semaphore in shared mode. down_read(&current->mm->mmap_sem) ... madvise_vma switch (behavior) case MADV_DONTNEED: madvise_dontneed zap_page_range unmap_vmas unmap_page_range zap_pud_range zap_pmd_range // // Assume that this huge page has never been accessed. // I.e. content of the PMD entry is zero (not mapped). // if (pmd_trans_huge(*pmd)) { // We don't get here due to the above assumption. } // // Assume that Thread B incurred a page fault and .---------> // sneaks in here as shown below. | // | if (pmd_none_or_clear_bad(pmd)) | { | if (unlikely(pmd_bad(*pmd))) | pmd_clear_bad | { | pmd_ERROR | // Log "bad pmd ..." message here. | pmd_clear | // Clear the page's PMD entry. | // Thread B incremented the map count | // in page_add_new_anon_rmap(), but | // now the page is no longer mapped | // by a PMD entry (-> inconsistency). | } | } | v - Thread B is handling a page fault on virtual address "B(fault)" shown in the picture. ... do_page_fault __do_page_fault // Acquire the semaphore in shared mode. down_read_trylock(&mm->mmap_sem) ... handle_mm_fault if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) // We get here due to the above assumption (PMD entry is zero). do_huge_pmd_anonymous_page alloc_hugepage_vma // Allocate a new transparent huge page here. ... __do_huge_pmd_anonymous_page ... spin_lock(&mm->page_table_lock) ... page_add_new_anon_rmap // Here we increment the page's map count (starts at -1). atomic_set(&page->_mapcount, 0) set_pmd_at // Here we set the page's PMD entry which will be cleared // when Thread A calls pmd_clear_bad(). ... spin_unlock(&mm->page_table_lock) The mmap_sem does not prevent the race because both threads are acquiring it in shared mode (down_read). Thread B holds the page_table_lock while the page's map count and PMD table entry are updated. However, Thread A does not synchronize on that lock. ====== end quote ======= [akpm@linux-foundation.org: checkpatch fixes] Reported-by: Ulrich Obergfell <uobergfe@redhat.com> Signed-off-by: Andrea Arcangeli <aarcange@redhat.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Cc: Mel Gorman <mgorman@suse.de> Cc: Hugh Dickins <hughd@google.com> Cc: Dave Jones <davej@redhat.com> Acked-by: Larry Woodman <lwoodman@redhat.com> Acked-by: Rik van Riel <riel@redhat.com> Cc: <stable@vger.kernel.org> [2.6.38+] Cc: Mark Salter <msalter@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-03-21 23:33:42 +00:00
if (next - addr != HPAGE_PMD_SIZE) {
#ifdef CONFIG_DEBUG_VM
if (!rwsem_is_locked(&tlb->mm->mmap_sem)) {
pr_err("%s: mmap_sem is unlocked! addr=0x%lx end=0x%lx vma->vm_start=0x%lx vma->vm_end=0x%lx\n",
__func__, addr, end,
vma->vm_start,
vma->vm_end);
BUG();
}
#endif
split_huge_page_pmd(vma, addr, pmd);
} else if (zap_huge_pmd(tlb, vma, pmd, addr))
mm: thp: fix pmd_bad() triggering in code paths holding mmap_sem read mode In some cases it may happen that pmd_none_or_clear_bad() is called with the mmap_sem hold in read mode. In those cases the huge page faults can allocate hugepmds under pmd_none_or_clear_bad() and that can trigger a false positive from pmd_bad() that will not like to see a pmd materializing as trans huge. It's not khugepaged causing the problem, khugepaged holds the mmap_sem in write mode (and all those sites must hold the mmap_sem in read mode to prevent pagetables to go away from under them, during code review it seems vm86 mode on 32bit kernels requires that too unless it's restricted to 1 thread per process or UP builds). The race is only with the huge pagefaults that can convert a pmd_none() into a pmd_trans_huge(). Effectively all these pmd_none_or_clear_bad() sites running with mmap_sem in read mode are somewhat speculative with the page faults, and the result is always undefined when they run simultaneously. This is probably why it wasn't common to run into this. For example if the madvise(MADV_DONTNEED) runs zap_page_range() shortly before the page fault, the hugepage will not be zapped, if the page fault runs first it will be zapped. Altering pmd_bad() not to error out if it finds hugepmds won't be enough to fix this, because zap_pmd_range would then proceed to call zap_pte_range (which would be incorrect if the pmd become a pmd_trans_huge()). The simplest way to fix this is to read the pmd in the local stack (regardless of what we read, no need of actual CPU barriers, only compiler barrier needed), and be sure it is not changing under the code that computes its value. Even if the real pmd is changing under the value we hold on the stack, we don't care. If we actually end up in zap_pte_range it means the pmd was not none already and it was not huge, and it can't become huge from under us (khugepaged locking explained above). All we need is to enforce that there is no way anymore that in a code path like below, pmd_trans_huge can be false, but pmd_none_or_clear_bad can run into a hugepmd. The overhead of a barrier() is just a compiler tweak and should not be measurable (I only added it for THP builds). I don't exclude different compiler versions may have prevented the race too by caching the value of *pmd on the stack (that hasn't been verified, but it wouldn't be impossible considering pmd_none_or_clear_bad, pmd_bad, pmd_trans_huge, pmd_none are all inlines and there's no external function called in between pmd_trans_huge and pmd_none_or_clear_bad). if (pmd_trans_huge(*pmd)) { if (next-addr != HPAGE_PMD_SIZE) { VM_BUG_ON(!rwsem_is_locked(&tlb->mm->mmap_sem)); split_huge_page_pmd(vma->vm_mm, pmd); } else if (zap_huge_pmd(tlb, vma, pmd, addr)) continue; /* fall through */ } if (pmd_none_or_clear_bad(pmd)) Because this race condition could be exercised without special privileges this was reported in CVE-2012-1179. The race was identified and fully explained by Ulrich who debugged it. I'm quoting his accurate explanation below, for reference. ====== start quote ======= mapcount 0 page_mapcount 1 kernel BUG at mm/huge_memory.c:1384! At some point prior to the panic, a "bad pmd ..." message similar to the following is logged on the console: mm/memory.c:145: bad pmd ffff8800376e1f98(80000000314000e7). The "bad pmd ..." message is logged by pmd_clear_bad() before it clears the page's PMD table entry. 143 void pmd_clear_bad(pmd_t *pmd) 144 { -> 145 pmd_ERROR(*pmd); 146 pmd_clear(pmd); 147 } After the PMD table entry has been cleared, there is an inconsistency between the actual number of PMD table entries that are mapping the page and the page's map count (_mapcount field in struct page). When the page is subsequently reclaimed, __split_huge_page() detects this inconsistency. 1381 if (mapcount != page_mapcount(page)) 1382 printk(KERN_ERR "mapcount %d page_mapcount %d\n", 1383 mapcount, page_mapcount(page)); -> 1384 BUG_ON(mapcount != page_mapcount(page)); The root cause of the problem is a race of two threads in a multithreaded process. Thread B incurs a page fault on a virtual address that has never been accessed (PMD entry is zero) while Thread A is executing an madvise() system call on a virtual address within the same 2 MB (huge page) range. virtual address space .---------------------. | | | | .-|---------------------| | | | | | |<-- B(fault) | | | 2 MB | |/////////////////////|-. huge < |/////////////////////| > A(range) page | |/////////////////////|-' | | | | | | '-|---------------------| | | | | '---------------------' - Thread A is executing an madvise(..., MADV_DONTNEED) system call on the virtual address range "A(range)" shown in the picture. sys_madvise // Acquire the semaphore in shared mode. down_read(&current->mm->mmap_sem) ... madvise_vma switch (behavior) case MADV_DONTNEED: madvise_dontneed zap_page_range unmap_vmas unmap_page_range zap_pud_range zap_pmd_range // // Assume that this huge page has never been accessed. // I.e. content of the PMD entry is zero (not mapped). // if (pmd_trans_huge(*pmd)) { // We don't get here due to the above assumption. } // // Assume that Thread B incurred a page fault and .---------> // sneaks in here as shown below. | // | if (pmd_none_or_clear_bad(pmd)) | { | if (unlikely(pmd_bad(*pmd))) | pmd_clear_bad | { | pmd_ERROR | // Log "bad pmd ..." message here. | pmd_clear | // Clear the page's PMD entry. | // Thread B incremented the map count | // in page_add_new_anon_rmap(), but | // now the page is no longer mapped | // by a PMD entry (-> inconsistency). | } | } | v - Thread B is handling a page fault on virtual address "B(fault)" shown in the picture. ... do_page_fault __do_page_fault // Acquire the semaphore in shared mode. down_read_trylock(&mm->mmap_sem) ... handle_mm_fault if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) // We get here due to the above assumption (PMD entry is zero). do_huge_pmd_anonymous_page alloc_hugepage_vma // Allocate a new transparent huge page here. ... __do_huge_pmd_anonymous_page ... spin_lock(&mm->page_table_lock) ... page_add_new_anon_rmap // Here we increment the page's map count (starts at -1). atomic_set(&page->_mapcount, 0) set_pmd_at // Here we set the page's PMD entry which will be cleared // when Thread A calls pmd_clear_bad(). ... spin_unlock(&mm->page_table_lock) The mmap_sem does not prevent the race because both threads are acquiring it in shared mode (down_read). Thread B holds the page_table_lock while the page's map count and PMD table entry are updated. However, Thread A does not synchronize on that lock. ====== end quote ======= [akpm@linux-foundation.org: checkpatch fixes] Reported-by: Ulrich Obergfell <uobergfe@redhat.com> Signed-off-by: Andrea Arcangeli <aarcange@redhat.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Cc: Mel Gorman <mgorman@suse.de> Cc: Hugh Dickins <hughd@google.com> Cc: Dave Jones <davej@redhat.com> Acked-by: Larry Woodman <lwoodman@redhat.com> Acked-by: Rik van Riel <riel@redhat.com> Cc: <stable@vger.kernel.org> [2.6.38+] Cc: Mark Salter <msalter@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-03-21 23:33:42 +00:00
goto next;
thp: transparent hugepage core Lately I've been working to make KVM use hugepages transparently without the usual restrictions of hugetlbfs. Some of the restrictions I'd like to see removed: 1) hugepages have to be swappable or the guest physical memory remains locked in RAM and can't be paged out to swap 2) if a hugepage allocation fails, regular pages should be allocated instead and mixed in the same vma without any failure and without userland noticing 3) if some task quits and more hugepages become available in the buddy, guest physical memory backed by regular pages should be relocated on hugepages automatically in regions under madvise(MADV_HUGEPAGE) (ideally event driven by waking up the kernel deamon if the order=HPAGE_PMD_SHIFT-PAGE_SHIFT list becomes not null) 4) avoidance of reservation and maximization of use of hugepages whenever possible. Reservation (needed to avoid runtime fatal faliures) may be ok for 1 machine with 1 database with 1 database cache with 1 database cache size known at boot time. It's definitely not feasible with a virtualization hypervisor usage like RHEV-H that runs an unknown number of virtual machines with an unknown size of each virtual machine with an unknown amount of pagecache that could be potentially useful in the host for guest not using O_DIRECT (aka cache=off). hugepages in the virtualization hypervisor (and also in the guest!) are much more important than in a regular host not using virtualization, becasue with NPT/EPT they decrease the tlb-miss cacheline accesses from 24 to 19 in case only the hypervisor uses transparent hugepages, and they decrease the tlb-miss cacheline accesses from 19 to 15 in case both the linux hypervisor and the linux guest both uses this patch (though the guest will limit the addition speedup to anonymous regions only for now...). Even more important is that the tlb miss handler is much slower on a NPT/EPT guest than for a regular shadow paging or no-virtualization scenario. So maximizing the amount of virtual memory cached by the TLB pays off significantly more with NPT/EPT than without (even if there would be no significant speedup in the tlb-miss runtime). The first (and more tedious) part of this work requires allowing the VM to handle anonymous hugepages mixed with regular pages transparently on regular anonymous vmas. This is what this patch tries to achieve in the least intrusive possible way. We want hugepages and hugetlb to be used in a way so that all applications can benefit without changes (as usual we leverage the KVM virtualization design: by improving the Linux VM at large, KVM gets the performance boost too). The most important design choice is: always fallback to 4k allocation if the hugepage allocation fails! This is the _very_ opposite of some large pagecache patches that failed with -EIO back then if a 64k (or similar) allocation failed... Second important decision (to reduce the impact of the feature on the existing pagetable handling code) is that at any time we can split an hugepage into 512 regular pages and it has to be done with an operation that can't fail. This way the reliability of the swapping isn't decreased (no need to allocate memory when we are short on memory to swap) and it's trivial to plug a split_huge_page* one-liner where needed without polluting the VM. Over time we can teach mprotect, mremap and friends to handle pmd_trans_huge natively without calling split_huge_page*. The fact it can't fail isn't just for swap: if split_huge_page would return -ENOMEM (instead of the current void) we'd need to rollback the mprotect from the middle of it (ideally including undoing the split_vma) which would be a big change and in the very wrong direction (it'd likely be simpler not to call split_huge_page at all and to teach mprotect and friends to handle hugepages instead of rolling them back from the middle). In short the very value of split_huge_page is that it can't fail. The collapsing and madvise(MADV_HUGEPAGE) part will remain separated and incremental and it'll just be an "harmless" addition later if this initial part is agreed upon. It also should be noted that locking-wise replacing regular pages with hugepages is going to be very easy if compared to what I'm doing below in split_huge_page, as it will only happen when page_count(page) matches page_mapcount(page) if we can take the PG_lock and mmap_sem in write mode. collapse_huge_page will be a "best effort" that (unlike split_huge_page) can fail at the minimal sign of trouble and we can try again later. collapse_huge_page will be similar to how KSM works and the madvise(MADV_HUGEPAGE) will work similar to madvise(MADV_MERGEABLE). The default I like is that transparent hugepages are used at page fault time. This can be changed with /sys/kernel/mm/transparent_hugepage/enabled. The control knob can be set to three values "always", "madvise", "never" which mean respectively that hugepages are always used, or only inside madvise(MADV_HUGEPAGE) regions, or never used. /sys/kernel/mm/transparent_hugepage/defrag instead controls if the hugepage allocation should defrag memory aggressively "always", only inside "madvise" regions, or "never". The pmd_trans_splitting/pmd_trans_huge locking is very solid. The put_page (from get_user_page users that can't use mmu notifier like O_DIRECT) that runs against a __split_huge_page_refcount instead was a pain to serialize in a way that would result always in a coherent page count for both tail and head. I think my locking solution with a compound_lock taken only after the page_first is valid and is still a PageHead should be safe but it surely needs review from SMP race point of view. In short there is no current existing way to serialize the O_DIRECT final put_page against split_huge_page_refcount so I had to invent a new one (O_DIRECT loses knowledge on the mapping status by the time gup_fast returns so...). And I didn't want to impact all gup/gup_fast users for now, maybe if we change the gup interface substantially we can avoid this locking, I admit I didn't think too much about it because changing the gup unpinning interface would be invasive. If we ignored O_DIRECT we could stick to the existing compound refcounting code, by simply adding a get_user_pages_fast_flags(foll_flags) where KVM (and any other mmu notifier user) would call it without FOLL_GET (and if FOLL_GET isn't set we'd just BUG_ON if nobody registered itself in the current task mmu notifier list yet). But O_DIRECT is fundamental for decent performance of virtualized I/O on fast storage so we can't avoid it to solve the race of put_page against split_huge_page_refcount to achieve a complete hugepage feature for KVM. Swap and oom works fine (well just like with regular pages ;). MMU notifier is handled transparently too, with the exception of the young bit on the pmd, that didn't have a range check but I think KVM will be fine because the whole point of hugepages is that EPT/NPT will also use a huge pmd when they notice gup returns pages with PageCompound set, so they won't care of a range and there's just the pmd young bit to check in that case. NOTE: in some cases if the L2 cache is small, this may slowdown and waste memory during COWs because 4M of memory are accessed in a single fault instead of 8k (the payoff is that after COW the program can run faster). So we might want to switch the copy_huge_page (and clear_huge_page too) to not temporal stores. I also extensively researched ways to avoid this cache trashing with a full prefault logic that would cow in 8k/16k/32k/64k up to 1M (I can send those patches that fully implemented prefault) but I concluded they're not worth it and they add an huge additional complexity and they remove all tlb benefits until the full hugepage has been faulted in, to save a little bit of memory and some cache during app startup, but they still don't improve substantially the cache-trashing during startup if the prefault happens in >4k chunks. One reason is that those 4k pte entries copied are still mapped on a perfectly cache-colored hugepage, so the trashing is the worst one can generate in those copies (cow of 4k page copies aren't so well colored so they trashes less, but again this results in software running faster after the page fault). Those prefault patches allowed things like a pte where post-cow pages were local 4k regular anon pages and the not-yet-cowed pte entries were pointing in the middle of some hugepage mapped read-only. If it doesn't payoff substantially with todays hardware it will payoff even less in the future with larger l2 caches, and the prefault logic would blot the VM a lot. If one is emebdded transparent_hugepage can be disabled during boot with sysfs or with the boot commandline parameter transparent_hugepage=0 (or transparent_hugepage=2 to restrict hugepages inside madvise regions) that will ensure not a single hugepage is allocated at boot time. It is simple enough to just disable transparent hugepage globally and let transparent hugepages be allocated selectively by applications in the MADV_HUGEPAGE region (both at page fault time, and if enabled with the collapse_huge_page too through the kernel daemon). This patch supports only hugepages mapped in the pmd, archs that have smaller hugepages will not fit in this patch alone. Also some archs like power have certain tlb limits that prevents mixing different page size in the same regions so they will not fit in this framework that requires "graceful fallback" to basic PAGE_SIZE in case of physical memory fragmentation. hugetlbfs remains a perfect fit for those because its software limits happen to match the hardware limits. hugetlbfs also remains a perfect fit for hugepage sizes like 1GByte that cannot be hoped to be found not fragmented after a certain system uptime and that would be very expensive to defragment with relocation, so requiring reservation. hugetlbfs is the "reservation way", the point of transparent hugepages is not to have any reservation at all and maximizing the use of cache and hugepages at all times automatically. Some performance result: vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largep ages3 memset page fault 1566023 memset tlb miss 453854 memset second tlb miss 453321 random access tlb miss 41635 random access second tlb miss 41658 vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largepages3 memset page fault 1566471 memset tlb miss 453375 memset second tlb miss 453320 random access tlb miss 41636 random access second tlb miss 41637 vmx andrea # ./largepages3 memset page fault 1566642 memset tlb miss 453417 memset second tlb miss 453313 random access tlb miss 41630 random access second tlb miss 41647 vmx andrea # ./largepages3 memset page fault 1566872 memset tlb miss 453418 memset second tlb miss 453315 random access tlb miss 41618 random access second tlb miss 41659 vmx andrea # echo 0 > /proc/sys/vm/transparent_hugepage vmx andrea # ./largepages3 memset page fault 2182476 memset tlb miss 460305 memset second tlb miss 460179 random access tlb miss 44483 random access second tlb miss 44186 vmx andrea # ./largepages3 memset page fault 2182791 memset tlb miss 460742 memset second tlb miss 459962 random access tlb miss 43981 random access second tlb miss 43988 ============ #include <stdio.h> #include <stdlib.h> #include <string.h> #include <sys/time.h> #define SIZE (3UL*1024*1024*1024) int main() { char *p = malloc(SIZE), *p2; struct timeval before, after; gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset page fault %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); return 0; } ============ Signed-off-by: Andrea Arcangeli <aarcange@redhat.com> Acked-by: Rik van Riel <riel@redhat.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-13 23:46:52 +00:00
/* fall through */
}
mm: thp: fix pmd_bad() triggering in code paths holding mmap_sem read mode In some cases it may happen that pmd_none_or_clear_bad() is called with the mmap_sem hold in read mode. In those cases the huge page faults can allocate hugepmds under pmd_none_or_clear_bad() and that can trigger a false positive from pmd_bad() that will not like to see a pmd materializing as trans huge. It's not khugepaged causing the problem, khugepaged holds the mmap_sem in write mode (and all those sites must hold the mmap_sem in read mode to prevent pagetables to go away from under them, during code review it seems vm86 mode on 32bit kernels requires that too unless it's restricted to 1 thread per process or UP builds). The race is only with the huge pagefaults that can convert a pmd_none() into a pmd_trans_huge(). Effectively all these pmd_none_or_clear_bad() sites running with mmap_sem in read mode are somewhat speculative with the page faults, and the result is always undefined when they run simultaneously. This is probably why it wasn't common to run into this. For example if the madvise(MADV_DONTNEED) runs zap_page_range() shortly before the page fault, the hugepage will not be zapped, if the page fault runs first it will be zapped. Altering pmd_bad() not to error out if it finds hugepmds won't be enough to fix this, because zap_pmd_range would then proceed to call zap_pte_range (which would be incorrect if the pmd become a pmd_trans_huge()). The simplest way to fix this is to read the pmd in the local stack (regardless of what we read, no need of actual CPU barriers, only compiler barrier needed), and be sure it is not changing under the code that computes its value. Even if the real pmd is changing under the value we hold on the stack, we don't care. If we actually end up in zap_pte_range it means the pmd was not none already and it was not huge, and it can't become huge from under us (khugepaged locking explained above). All we need is to enforce that there is no way anymore that in a code path like below, pmd_trans_huge can be false, but pmd_none_or_clear_bad can run into a hugepmd. The overhead of a barrier() is just a compiler tweak and should not be measurable (I only added it for THP builds). I don't exclude different compiler versions may have prevented the race too by caching the value of *pmd on the stack (that hasn't been verified, but it wouldn't be impossible considering pmd_none_or_clear_bad, pmd_bad, pmd_trans_huge, pmd_none are all inlines and there's no external function called in between pmd_trans_huge and pmd_none_or_clear_bad). if (pmd_trans_huge(*pmd)) { if (next-addr != HPAGE_PMD_SIZE) { VM_BUG_ON(!rwsem_is_locked(&tlb->mm->mmap_sem)); split_huge_page_pmd(vma->vm_mm, pmd); } else if (zap_huge_pmd(tlb, vma, pmd, addr)) continue; /* fall through */ } if (pmd_none_or_clear_bad(pmd)) Because this race condition could be exercised without special privileges this was reported in CVE-2012-1179. The race was identified and fully explained by Ulrich who debugged it. I'm quoting his accurate explanation below, for reference. ====== start quote ======= mapcount 0 page_mapcount 1 kernel BUG at mm/huge_memory.c:1384! At some point prior to the panic, a "bad pmd ..." message similar to the following is logged on the console: mm/memory.c:145: bad pmd ffff8800376e1f98(80000000314000e7). The "bad pmd ..." message is logged by pmd_clear_bad() before it clears the page's PMD table entry. 143 void pmd_clear_bad(pmd_t *pmd) 144 { -> 145 pmd_ERROR(*pmd); 146 pmd_clear(pmd); 147 } After the PMD table entry has been cleared, there is an inconsistency between the actual number of PMD table entries that are mapping the page and the page's map count (_mapcount field in struct page). When the page is subsequently reclaimed, __split_huge_page() detects this inconsistency. 1381 if (mapcount != page_mapcount(page)) 1382 printk(KERN_ERR "mapcount %d page_mapcount %d\n", 1383 mapcount, page_mapcount(page)); -> 1384 BUG_ON(mapcount != page_mapcount(page)); The root cause of the problem is a race of two threads in a multithreaded process. Thread B incurs a page fault on a virtual address that has never been accessed (PMD entry is zero) while Thread A is executing an madvise() system call on a virtual address within the same 2 MB (huge page) range. virtual address space .---------------------. | | | | .-|---------------------| | | | | | |<-- B(fault) | | | 2 MB | |/////////////////////|-. huge < |/////////////////////| > A(range) page | |/////////////////////|-' | | | | | | '-|---------------------| | | | | '---------------------' - Thread A is executing an madvise(..., MADV_DONTNEED) system call on the virtual address range "A(range)" shown in the picture. sys_madvise // Acquire the semaphore in shared mode. down_read(&current->mm->mmap_sem) ... madvise_vma switch (behavior) case MADV_DONTNEED: madvise_dontneed zap_page_range unmap_vmas unmap_page_range zap_pud_range zap_pmd_range // // Assume that this huge page has never been accessed. // I.e. content of the PMD entry is zero (not mapped). // if (pmd_trans_huge(*pmd)) { // We don't get here due to the above assumption. } // // Assume that Thread B incurred a page fault and .---------> // sneaks in here as shown below. | // | if (pmd_none_or_clear_bad(pmd)) | { | if (unlikely(pmd_bad(*pmd))) | pmd_clear_bad | { | pmd_ERROR | // Log "bad pmd ..." message here. | pmd_clear | // Clear the page's PMD entry. | // Thread B incremented the map count | // in page_add_new_anon_rmap(), but | // now the page is no longer mapped | // by a PMD entry (-> inconsistency). | } | } | v - Thread B is handling a page fault on virtual address "B(fault)" shown in the picture. ... do_page_fault __do_page_fault // Acquire the semaphore in shared mode. down_read_trylock(&mm->mmap_sem) ... handle_mm_fault if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) // We get here due to the above assumption (PMD entry is zero). do_huge_pmd_anonymous_page alloc_hugepage_vma // Allocate a new transparent huge page here. ... __do_huge_pmd_anonymous_page ... spin_lock(&mm->page_table_lock) ... page_add_new_anon_rmap // Here we increment the page's map count (starts at -1). atomic_set(&page->_mapcount, 0) set_pmd_at // Here we set the page's PMD entry which will be cleared // when Thread A calls pmd_clear_bad(). ... spin_unlock(&mm->page_table_lock) The mmap_sem does not prevent the race because both threads are acquiring it in shared mode (down_read). Thread B holds the page_table_lock while the page's map count and PMD table entry are updated. However, Thread A does not synchronize on that lock. ====== end quote ======= [akpm@linux-foundation.org: checkpatch fixes] Reported-by: Ulrich Obergfell <uobergfe@redhat.com> Signed-off-by: Andrea Arcangeli <aarcange@redhat.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Cc: Mel Gorman <mgorman@suse.de> Cc: Hugh Dickins <hughd@google.com> Cc: Dave Jones <davej@redhat.com> Acked-by: Larry Woodman <lwoodman@redhat.com> Acked-by: Rik van Riel <riel@redhat.com> Cc: <stable@vger.kernel.org> [2.6.38+] Cc: Mark Salter <msalter@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-03-21 23:33:42 +00:00
/*
* Here there can be other concurrent MADV_DONTNEED or
* trans huge page faults running, and if the pmd is
* none or trans huge it can change under us. This is
* because MADV_DONTNEED holds the mmap_sem in read
* mode.
*/
if (pmd_none_or_trans_huge_or_clear_bad(pmd))
goto next;
next = zap_pte_range(tlb, vma, pmd, addr, next, details);
mm: thp: fix pmd_bad() triggering in code paths holding mmap_sem read mode In some cases it may happen that pmd_none_or_clear_bad() is called with the mmap_sem hold in read mode. In those cases the huge page faults can allocate hugepmds under pmd_none_or_clear_bad() and that can trigger a false positive from pmd_bad() that will not like to see a pmd materializing as trans huge. It's not khugepaged causing the problem, khugepaged holds the mmap_sem in write mode (and all those sites must hold the mmap_sem in read mode to prevent pagetables to go away from under them, during code review it seems vm86 mode on 32bit kernels requires that too unless it's restricted to 1 thread per process or UP builds). The race is only with the huge pagefaults that can convert a pmd_none() into a pmd_trans_huge(). Effectively all these pmd_none_or_clear_bad() sites running with mmap_sem in read mode are somewhat speculative with the page faults, and the result is always undefined when they run simultaneously. This is probably why it wasn't common to run into this. For example if the madvise(MADV_DONTNEED) runs zap_page_range() shortly before the page fault, the hugepage will not be zapped, if the page fault runs first it will be zapped. Altering pmd_bad() not to error out if it finds hugepmds won't be enough to fix this, because zap_pmd_range would then proceed to call zap_pte_range (which would be incorrect if the pmd become a pmd_trans_huge()). The simplest way to fix this is to read the pmd in the local stack (regardless of what we read, no need of actual CPU barriers, only compiler barrier needed), and be sure it is not changing under the code that computes its value. Even if the real pmd is changing under the value we hold on the stack, we don't care. If we actually end up in zap_pte_range it means the pmd was not none already and it was not huge, and it can't become huge from under us (khugepaged locking explained above). All we need is to enforce that there is no way anymore that in a code path like below, pmd_trans_huge can be false, but pmd_none_or_clear_bad can run into a hugepmd. The overhead of a barrier() is just a compiler tweak and should not be measurable (I only added it for THP builds). I don't exclude different compiler versions may have prevented the race too by caching the value of *pmd on the stack (that hasn't been verified, but it wouldn't be impossible considering pmd_none_or_clear_bad, pmd_bad, pmd_trans_huge, pmd_none are all inlines and there's no external function called in between pmd_trans_huge and pmd_none_or_clear_bad). if (pmd_trans_huge(*pmd)) { if (next-addr != HPAGE_PMD_SIZE) { VM_BUG_ON(!rwsem_is_locked(&tlb->mm->mmap_sem)); split_huge_page_pmd(vma->vm_mm, pmd); } else if (zap_huge_pmd(tlb, vma, pmd, addr)) continue; /* fall through */ } if (pmd_none_or_clear_bad(pmd)) Because this race condition could be exercised without special privileges this was reported in CVE-2012-1179. The race was identified and fully explained by Ulrich who debugged it. I'm quoting his accurate explanation below, for reference. ====== start quote ======= mapcount 0 page_mapcount 1 kernel BUG at mm/huge_memory.c:1384! At some point prior to the panic, a "bad pmd ..." message similar to the following is logged on the console: mm/memory.c:145: bad pmd ffff8800376e1f98(80000000314000e7). The "bad pmd ..." message is logged by pmd_clear_bad() before it clears the page's PMD table entry. 143 void pmd_clear_bad(pmd_t *pmd) 144 { -> 145 pmd_ERROR(*pmd); 146 pmd_clear(pmd); 147 } After the PMD table entry has been cleared, there is an inconsistency between the actual number of PMD table entries that are mapping the page and the page's map count (_mapcount field in struct page). When the page is subsequently reclaimed, __split_huge_page() detects this inconsistency. 1381 if (mapcount != page_mapcount(page)) 1382 printk(KERN_ERR "mapcount %d page_mapcount %d\n", 1383 mapcount, page_mapcount(page)); -> 1384 BUG_ON(mapcount != page_mapcount(page)); The root cause of the problem is a race of two threads in a multithreaded process. Thread B incurs a page fault on a virtual address that has never been accessed (PMD entry is zero) while Thread A is executing an madvise() system call on a virtual address within the same 2 MB (huge page) range. virtual address space .---------------------. | | | | .-|---------------------| | | | | | |<-- B(fault) | | | 2 MB | |/////////////////////|-. huge < |/////////////////////| > A(range) page | |/////////////////////|-' | | | | | | '-|---------------------| | | | | '---------------------' - Thread A is executing an madvise(..., MADV_DONTNEED) system call on the virtual address range "A(range)" shown in the picture. sys_madvise // Acquire the semaphore in shared mode. down_read(&current->mm->mmap_sem) ... madvise_vma switch (behavior) case MADV_DONTNEED: madvise_dontneed zap_page_range unmap_vmas unmap_page_range zap_pud_range zap_pmd_range // // Assume that this huge page has never been accessed. // I.e. content of the PMD entry is zero (not mapped). // if (pmd_trans_huge(*pmd)) { // We don't get here due to the above assumption. } // // Assume that Thread B incurred a page fault and .---------> // sneaks in here as shown below. | // | if (pmd_none_or_clear_bad(pmd)) | { | if (unlikely(pmd_bad(*pmd))) | pmd_clear_bad | { | pmd_ERROR | // Log "bad pmd ..." message here. | pmd_clear | // Clear the page's PMD entry. | // Thread B incremented the map count | // in page_add_new_anon_rmap(), but | // now the page is no longer mapped | // by a PMD entry (-> inconsistency). | } | } | v - Thread B is handling a page fault on virtual address "B(fault)" shown in the picture. ... do_page_fault __do_page_fault // Acquire the semaphore in shared mode. down_read_trylock(&mm->mmap_sem) ... handle_mm_fault if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) // We get here due to the above assumption (PMD entry is zero). do_huge_pmd_anonymous_page alloc_hugepage_vma // Allocate a new transparent huge page here. ... __do_huge_pmd_anonymous_page ... spin_lock(&mm->page_table_lock) ... page_add_new_anon_rmap // Here we increment the page's map count (starts at -1). atomic_set(&page->_mapcount, 0) set_pmd_at // Here we set the page's PMD entry which will be cleared // when Thread A calls pmd_clear_bad(). ... spin_unlock(&mm->page_table_lock) The mmap_sem does not prevent the race because both threads are acquiring it in shared mode (down_read). Thread B holds the page_table_lock while the page's map count and PMD table entry are updated. However, Thread A does not synchronize on that lock. ====== end quote ======= [akpm@linux-foundation.org: checkpatch fixes] Reported-by: Ulrich Obergfell <uobergfe@redhat.com> Signed-off-by: Andrea Arcangeli <aarcange@redhat.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Cc: Mel Gorman <mgorman@suse.de> Cc: Hugh Dickins <hughd@google.com> Cc: Dave Jones <davej@redhat.com> Acked-by: Larry Woodman <lwoodman@redhat.com> Acked-by: Rik van Riel <riel@redhat.com> Cc: <stable@vger.kernel.org> [2.6.38+] Cc: Mark Salter <msalter@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-03-21 23:33:42 +00:00
next:
cond_resched();
} while (pmd++, addr = next, addr != end);
return addr;
}
static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
2005-10-30 01:16:12 +00:00
struct vm_area_struct *vma, pgd_t *pgd,
unsigned long addr, unsigned long end,
struct zap_details *details)
{
pud_t *pud;
unsigned long next;
pud = pud_offset(pgd, addr);
do {
next = pud_addr_end(addr, end);
if (pud_none_or_clear_bad(pud))
continue;
next = zap_pmd_range(tlb, vma, pud, addr, next, details);
} while (pud++, addr = next, addr != end);
return addr;
}
static void unmap_page_range(struct mmu_gather *tlb,
struct vm_area_struct *vma,
unsigned long addr, unsigned long end,
struct zap_details *details)
{
pgd_t *pgd;
unsigned long next;
if (details && !details->check_mapping && !details->nonlinear_vma)
details = NULL;
BUG_ON(addr >= end);
memcg: coalesce uncharge during unmap/truncate In massive parallel enviroment, res_counter can be a performance bottleneck. One strong techinque to reduce lock contention is reducing calls by coalescing some amount of calls into one. Considering charge/uncharge chatacteristic, - charge is done one by one via demand-paging. - uncharge is done by - in chunk at munmap, truncate, exit, execve... - one by one via vmscan/paging. It seems we have a chance to coalesce uncharges for improving scalability at unmap/truncation. This patch is a for coalescing uncharge. For avoiding scattering memcg's structure to functions under /mm, this patch adds memcg batch uncharge information to the task. A reason for per-task batching is for making use of caller's context information. We do batched uncharge (deleyed uncharge) when truncation/unmap occurs but do direct uncharge when uncharge is called by memory reclaim (vmscan.c). The degree of coalescing depends on callers - at invalidate/trucate... pagevec size - at unmap ....ZAP_BLOCK_SIZE (memory itself will be freed in this degree.) Then, we'll not coalescing too much. On x86-64 8cpu server, I tested overheads of memcg at page fault by running a program which does map/fault/unmap in a loop. Running a task per a cpu by taskset and see sum of the number of page faults in 60secs. [without memcg config] 40156968 page-faults # 0.085 M/sec ( +- 0.046% ) 27.67 cache-miss/faults [root cgroup] 36659599 page-faults # 0.077 M/sec ( +- 0.247% ) 31.58 miss/faults [in a child cgroup] 18444157 page-faults # 0.039 M/sec ( +- 0.133% ) 69.96 miss/faults [child with this patch] 27133719 page-faults # 0.057 M/sec ( +- 0.155% ) 47.16 miss/faults We can see some amounts of improvement. (root cgroup doesn't affected by this patch) Another patch for "charge" will follow this and above will be improved more. Changelog(since 2009/10/02): - renamed filed of memcg_batch (as pages to bytes, memsw to memsw_bytes) - some clean up and commentary/description updates. - added initialize code to copy_process(). (possible bug fix) Changelog(old): - fixed !CONFIG_MEM_CGROUP case. - rebased onto the latest mmotm + softlimit fix patches. - unified patch for callers - added commetns. - make ->do_batch as bool. - removed css_get() at el. We don't need it. Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Balbir Singh <balbir@in.ibm.com> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-16 00:47:03 +00:00
mem_cgroup_uncharge_start();
tlb_start_vma(tlb, vma);
pgd = pgd_offset(vma->vm_mm, addr);
do {
next = pgd_addr_end(addr, end);
if (pgd_none_or_clear_bad(pgd))
continue;
next = zap_pud_range(tlb, vma, pgd, addr, next, details);
} while (pgd++, addr = next, addr != end);
tlb_end_vma(tlb, vma);
memcg: coalesce uncharge during unmap/truncate In massive parallel enviroment, res_counter can be a performance bottleneck. One strong techinque to reduce lock contention is reducing calls by coalescing some amount of calls into one. Considering charge/uncharge chatacteristic, - charge is done one by one via demand-paging. - uncharge is done by - in chunk at munmap, truncate, exit, execve... - one by one via vmscan/paging. It seems we have a chance to coalesce uncharges for improving scalability at unmap/truncation. This patch is a for coalescing uncharge. For avoiding scattering memcg's structure to functions under /mm, this patch adds memcg batch uncharge information to the task. A reason for per-task batching is for making use of caller's context information. We do batched uncharge (deleyed uncharge) when truncation/unmap occurs but do direct uncharge when uncharge is called by memory reclaim (vmscan.c). The degree of coalescing depends on callers - at invalidate/trucate... pagevec size - at unmap ....ZAP_BLOCK_SIZE (memory itself will be freed in this degree.) Then, we'll not coalescing too much. On x86-64 8cpu server, I tested overheads of memcg at page fault by running a program which does map/fault/unmap in a loop. Running a task per a cpu by taskset and see sum of the number of page faults in 60secs. [without memcg config] 40156968 page-faults # 0.085 M/sec ( +- 0.046% ) 27.67 cache-miss/faults [root cgroup] 36659599 page-faults # 0.077 M/sec ( +- 0.247% ) 31.58 miss/faults [in a child cgroup] 18444157 page-faults # 0.039 M/sec ( +- 0.133% ) 69.96 miss/faults [child with this patch] 27133719 page-faults # 0.057 M/sec ( +- 0.155% ) 47.16 miss/faults We can see some amounts of improvement. (root cgroup doesn't affected by this patch) Another patch for "charge" will follow this and above will be improved more. Changelog(since 2009/10/02): - renamed filed of memcg_batch (as pages to bytes, memsw to memsw_bytes) - some clean up and commentary/description updates. - added initialize code to copy_process(). (possible bug fix) Changelog(old): - fixed !CONFIG_MEM_CGROUP case. - rebased onto the latest mmotm + softlimit fix patches. - unified patch for callers - added commetns. - make ->do_batch as bool. - removed css_get() at el. We don't need it. Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Balbir Singh <balbir@in.ibm.com> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-16 00:47:03 +00:00
mem_cgroup_uncharge_end();
}
static void unmap_single_vma(struct mmu_gather *tlb,
struct vm_area_struct *vma, unsigned long start_addr,
unsigned long end_addr,
struct zap_details *details)
{
unsigned long start = max(vma->vm_start, start_addr);
unsigned long end;
if (start >= vma->vm_end)
return;
end = min(vma->vm_end, end_addr);
if (end <= vma->vm_start)
return;
uprobes/core: Decrement uprobe count before the pages are unmapped Uprobes has a callback (uprobe_munmap()) in the unmap path to maintain the uprobes count. In the exit path this callback gets called in unlink_file_vma(). However by the time unlink_file_vma() is called, the pages would have been unmapped (in unmap_vmas()) and the task->rss_stat counts accounted (in zap_pte_range()). If the exiting process has probepoints, uprobe_munmap() checks if the breakpoint instruction was around before decrementing the probe count. This results in a file backed page being reread by uprobe_munmap() and hence it does not find the breakpoint. This patch fixes this problem by moving the callback to unmap_single_vma(). Since unmap_single_vma() may not unmap the complete vma, add start and end parameters to uprobe_munmap(). This bug became apparent courtesy of commit c3f0327f8e9d ("mm: add rss counters consistency check"). Signed-off-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com> Cc: Jim Keniston <jkenisto@linux.vnet.ibm.com> Cc: Linux-mm <linux-mm@kvack.org> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Andi Kleen <andi@firstfloor.org> Cc: Christoph Hellwig <hch@infradead.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Arnaldo Carvalho de Melo <acme@infradead.org> Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Cc: Anton Arapov <anton@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Link: http://lkml.kernel.org/r/20120411103527.23245.9835.sendpatchset@srdronam.in.ibm.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2012-04-11 10:35:27 +00:00
if (vma->vm_file)
uprobe_munmap(vma, start, end);
mm, x86, pat: rework linear pfn-mmap tracking Replace the generic vma-flag VM_PFN_AT_MMAP with x86-only VM_PAT. We can toss mapping address from remap_pfn_range() into track_pfn_vma_new(), and collect all PAT-related logic together in arch/x86/. This patch also restores orignal frustration-free is_cow_mapping() check in remap_pfn_range(), as it was before commit v2.6.28-rc8-88-g3c8bb73 ("x86: PAT: store vm_pgoff for all linear_over_vma_region mappings - v3") is_linear_pfn_mapping() checks can be removed from mm/huge_memory.c, because it already handled by VM_PFNMAP in VM_NO_THP bit-mask. [suresh.b.siddha@intel.com: Reset the VM_PAT flag as part of untrack_pfn_vma()] Signed-off-by: Konstantin Khlebnikov <khlebnikov@openvz.org> Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com> Cc: Venkatesh Pallipadi <venki@google.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Ingo Molnar <mingo@redhat.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Carsten Otte <cotte@de.ibm.com> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Eric Paris <eparis@redhat.com> Cc: Hugh Dickins <hughd@google.com> Cc: James Morris <james.l.morris@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Kentaro Takeda <takedakn@nttdata.co.jp> Cc: Matt Helsley <matthltc@us.ibm.com> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Venkatesh Pallipadi <venki@google.com> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-08 23:28:34 +00:00
if (unlikely(vma->vm_flags & VM_PFNMAP))
x86, pat: separate the pfn attribute tracking for remap_pfn_range and vm_insert_pfn With PAT enabled, vm_insert_pfn() looks up the existing pfn memory attribute and uses it. Expectation is that the driver reserves the memory attributes for the pfn before calling vm_insert_pfn(). remap_pfn_range() (when called for the whole vma) will setup a new attribute (based on the prot argument) for the specified pfn range. This addresses the legacy usage which typically calls remap_pfn_range() with a desired memory attribute. For ranges smaller than the vma size (which is typically not the case), remap_pfn_range() will use the existing memory attribute for the pfn range. Expose two different API's for these different behaviors. track_pfn_insert() for tracking the pfn attribute set by vm_insert_pfn() and track_pfn_remap() for the remap_pfn_range(). This cleanup also prepares the ground for the track/untrack pfn vma routines to take over the ownership of setting PAT specific vm_flag in the 'vma'. [khlebnikov@openvz.org: Clear checks in track_pfn_remap()] [akpm@linux-foundation.org: tweak a few comments] Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com> Signed-off-by: Konstantin Khlebnikov <khlebnikov@openvz.org> Cc: Venkatesh Pallipadi <venki@google.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Ingo Molnar <mingo@redhat.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Carsten Otte <cotte@de.ibm.com> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Eric Paris <eparis@redhat.com> Cc: Hugh Dickins <hughd@google.com> Cc: James Morris <james.l.morris@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Kentaro Takeda <takedakn@nttdata.co.jp> Cc: Konstantin Khlebnikov <khlebnikov@openvz.org> Cc: Matt Helsley <matthltc@us.ibm.com> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> Cc: Suresh Siddha <suresh.b.siddha@intel.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-08 23:28:29 +00:00
untrack_pfn(vma, 0, 0);
if (start != end) {
if (unlikely(is_vm_hugetlb_page(vma))) {
/*
* It is undesirable to test vma->vm_file as it
* should be non-null for valid hugetlb area.
* However, vm_file will be NULL in the error
* cleanup path of do_mmap_pgoff. When
* hugetlbfs ->mmap method fails,
* do_mmap_pgoff() nullifies vma->vm_file
* before calling this function to clean up.
* Since no pte has actually been setup, it is
* safe to do nothing in this case.
*/
if (vma->vm_file) {
mutex_lock(&vma->vm_file->f_mapping->i_mmap_mutex);
mm: hugetlbfs: close race during teardown of hugetlbfs shared page tables If a process creates a large hugetlbfs mapping that is eligible for page table sharing and forks heavily with children some of whom fault and others which destroy the mapping then it is possible for page tables to get corrupted. Some teardowns of the mapping encounter a "bad pmd" and output a message to the kernel log. The final teardown will trigger a BUG_ON in mm/filemap.c. This was reproduced in 3.4 but is known to have existed for a long time and goes back at least as far as 2.6.37. It was probably was introduced in 2.6.20 by [39dde65c: shared page table for hugetlb page]. The messages look like this; [ ..........] Lots of bad pmd messages followed by this [ 127.164256] mm/memory.c:391: bad pmd ffff880412e04fe8(80000003de4000e7). [ 127.164257] mm/memory.c:391: bad pmd ffff880412e04ff0(80000003de6000e7). [ 127.164258] mm/memory.c:391: bad pmd ffff880412e04ff8(80000003de0000e7). [ 127.186778] ------------[ cut here ]------------ [ 127.186781] kernel BUG at mm/filemap.c:134! [ 127.186782] invalid opcode: 0000 [#1] SMP [ 127.186783] CPU 7 [ 127.186784] Modules linked in: af_packet cpufreq_conservative cpufreq_userspace cpufreq_powersave acpi_cpufreq mperf ext3 jbd dm_mod coretemp crc32c_intel usb_storage ghash_clmulni_intel aesni_intel i2c_i801 r8169 mii uas sr_mod cdrom sg iTCO_wdt iTCO_vendor_support shpchp serio_raw cryptd aes_x86_64 e1000e pci_hotplug dcdbas aes_generic container microcode ext4 mbcache jbd2 crc16 sd_mod crc_t10dif i915 drm_kms_helper drm i2c_algo_bit ehci_hcd ahci libahci usbcore rtc_cmos usb_common button i2c_core intel_agp video intel_gtt fan processor thermal thermal_sys hwmon ata_generic pata_atiixp libata scsi_mod [ 127.186801] [ 127.186802] Pid: 9017, comm: hugetlbfs-test Not tainted 3.4.0-autobuild #53 Dell Inc. OptiPlex 990/06D7TR [ 127.186804] RIP: 0010:[<ffffffff810ed6ce>] [<ffffffff810ed6ce>] __delete_from_page_cache+0x15e/0x160 [ 127.186809] RSP: 0000:ffff8804144b5c08 EFLAGS: 00010002 [ 127.186810] RAX: 0000000000000001 RBX: ffffea000a5c9000 RCX: 00000000ffffffc0 [ 127.186811] RDX: 0000000000000000 RSI: 0000000000000009 RDI: ffff88042dfdad00 [ 127.186812] RBP: ffff8804144b5c18 R08: 0000000000000009 R09: 0000000000000003 [ 127.186813] R10: 0000000000000000 R11: 000000000000002d R12: ffff880412ff83d8 [ 127.186814] R13: ffff880412ff83d8 R14: 0000000000000000 R15: ffff880412ff83d8 [ 127.186815] FS: 00007fe18ed2c700(0000) GS:ffff88042dce0000(0000) knlGS:0000000000000000 [ 127.186816] CS: 0010 DS: 0000 ES: 0000 CR0: 000000008005003b [ 127.186817] CR2: 00007fe340000503 CR3: 0000000417a14000 CR4: 00000000000407e0 [ 127.186818] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 [ 127.186819] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400 [ 127.186820] Process hugetlbfs-test (pid: 9017, threadinfo ffff8804144b4000, task ffff880417f803c0) [ 127.186821] Stack: [ 127.186822] ffffea000a5c9000 0000000000000000 ffff8804144b5c48 ffffffff810ed83b [ 127.186824] ffff8804144b5c48 000000000000138a 0000000000001387 ffff8804144b5c98 [ 127.186825] ffff8804144b5d48 ffffffff811bc925 ffff8804144b5cb8 0000000000000000 [ 127.186827] Call Trace: [ 127.186829] [<ffffffff810ed83b>] delete_from_page_cache+0x3b/0x80 [ 127.186832] [<ffffffff811bc925>] truncate_hugepages+0x115/0x220 [ 127.186834] [<ffffffff811bca43>] hugetlbfs_evict_inode+0x13/0x30 [ 127.186837] [<ffffffff811655c7>] evict+0xa7/0x1b0 [ 127.186839] [<ffffffff811657a3>] iput_final+0xd3/0x1f0 [ 127.186840] [<ffffffff811658f9>] iput+0x39/0x50 [ 127.186842] [<ffffffff81162708>] d_kill+0xf8/0x130 [ 127.186843] [<ffffffff81162812>] dput+0xd2/0x1a0 [ 127.186845] [<ffffffff8114e2d0>] __fput+0x170/0x230 [ 127.186848] [<ffffffff81236e0e>] ? rb_erase+0xce/0x150 [ 127.186849] [<ffffffff8114e3ad>] fput+0x1d/0x30 [ 127.186851] [<ffffffff81117db7>] remove_vma+0x37/0x80 [ 127.186853] [<ffffffff81119182>] do_munmap+0x2d2/0x360 [ 127.186855] [<ffffffff811cc639>] sys_shmdt+0xc9/0x170 [ 127.186857] [<ffffffff81410a39>] system_call_fastpath+0x16/0x1b [ 127.186858] Code: 0f 1f 44 00 00 48 8b 43 08 48 8b 00 48 8b 40 28 8b b0 40 03 00 00 85 f6 0f 88 df fe ff ff 48 89 df e8 e7 cb 05 00 e9 d2 fe ff ff <0f> 0b 55 83 e2 fd 48 89 e5 48 83 ec 30 48 89 5d d8 4c 89 65 e0 [ 127.186868] RIP [<ffffffff810ed6ce>] __delete_from_page_cache+0x15e/0x160 [ 127.186870] RSP <ffff8804144b5c08> [ 127.186871] ---[ end trace 7cbac5d1db69f426 ]--- The bug is a race and not always easy to reproduce. To reproduce it I was doing the following on a single socket I7-based machine with 16G of RAM. $ hugeadm --pool-pages-max DEFAULT:13G $ echo $((18*1048576*1024)) > /proc/sys/kernel/shmmax $ echo $((18*1048576*1024)) > /proc/sys/kernel/shmall $ for i in `seq 1 9000`; do ./hugetlbfs-test; done On my particular machine, it usually triggers within 10 minutes but enabling debug options can change the timing such that it never hits. Once the bug is triggered, the machine is in trouble and needs to be rebooted. The machine will respond but processes accessing proc like "ps aux" will hang due to the BUG_ON. shutdown will also hang and needs a hard reset or a sysrq-b. The basic problem is a race between page table sharing and teardown. For the most part page table sharing depends on i_mmap_mutex. In some cases, it is also taking the mm->page_table_lock for the PTE updates but with shared page tables, it is the i_mmap_mutex that is more important. Unfortunately it appears to be also insufficient. Consider the following situation Process A Process B --------- --------- hugetlb_fault shmdt LockWrite(mmap_sem) do_munmap unmap_region unmap_vmas unmap_single_vma unmap_hugepage_range Lock(i_mmap_mutex) Lock(mm->page_table_lock) huge_pmd_unshare/unmap tables <--- (1) Unlock(mm->page_table_lock) Unlock(i_mmap_mutex) huge_pte_alloc ... Lock(i_mmap_mutex) ... vma_prio_walk, find svma, spte ... Lock(mm->page_table_lock) ... share spte ... Unlock(mm->page_table_lock) ... Unlock(i_mmap_mutex) ... hugetlb_no_page <--- (2) free_pgtables unlink_file_vma hugetlb_free_pgd_range remove_vma_list In this scenario, it is possible for Process A to share page tables with Process B that is trying to tear them down. The i_mmap_mutex on its own does not prevent Process A walking Process B's page tables. At (1) above, the page tables are not shared yet so it unmaps the PMDs. Process A sets up page table sharing and at (2) faults a new entry. Process B then trips up on it in free_pgtables. This patch fixes the problem by adding a new function __unmap_hugepage_range_final that is only called when the VMA is about to be destroyed. This function clears VM_MAYSHARE during unmap_hugepage_range() under the i_mmap_mutex. This makes the VMA ineligible for sharing and avoids the race. Superficially this looks like it would then be vunerable to truncate and madvise issues but hugetlbfs has its own truncate handlers so does not use unmap_mapping_range() and does not support madvise(DONTNEED). This should be treated as a -stable candidate if it is merged. Test program is as follows. The test case was mostly written by Michal Hocko with a few minor changes to reproduce this bug. ==== CUT HERE ==== static size_t huge_page_size = (2UL << 20); static size_t nr_huge_page_A = 512; static size_t nr_huge_page_B = 5632; unsigned int get_random(unsigned int max) { struct timeval tv; gettimeofday(&tv, NULL); srandom(tv.tv_usec); return random() % max; } static void play(void *addr, size_t size) { unsigned char *start = addr, *end = start + size, *a; start += get_random(size/2); /* we could itterate on huge pages but let's give it more time. */ for (a = start; a < end; a += 4096) *a = 0; } int main(int argc, char **argv) { key_t key = IPC_PRIVATE; size_t sizeA = nr_huge_page_A * huge_page_size; size_t sizeB = nr_huge_page_B * huge_page_size; int shmidA, shmidB; void *addrA = NULL, *addrB = NULL; int nr_children = 300, n = 0; if ((shmidA = shmget(key, sizeA, IPC_CREAT|SHM_HUGETLB|0660)) == -1) { perror("shmget:"); return 1; } if ((addrA = shmat(shmidA, addrA, SHM_R|SHM_W)) == (void *)-1UL) { perror("shmat"); return 1; } if ((shmidB = shmget(key, sizeB, IPC_CREAT|SHM_HUGETLB|0660)) == -1) { perror("shmget:"); return 1; } if ((addrB = shmat(shmidB, addrB, SHM_R|SHM_W)) == (void *)-1UL) { perror("shmat"); return 1; } fork_child: switch(fork()) { case 0: switch (n%3) { case 0: play(addrA, sizeA); break; case 1: play(addrB, sizeB); break; case 2: break; } break; case -1: perror("fork:"); break; default: if (++n < nr_children) goto fork_child; play(addrA, sizeA); break; } shmdt(addrA); shmdt(addrB); do { wait(NULL); } while (--n > 0); shmctl(shmidA, IPC_RMID, NULL); shmctl(shmidB, IPC_RMID, NULL); return 0; } [akpm@linux-foundation.org: name the declaration's args, fix CONFIG_HUGETLBFS=n build] Signed-off-by: Hugh Dickins <hughd@google.com> Reviewed-by: Michal Hocko <mhocko@suse.cz> Signed-off-by: Mel Gorman <mgorman@suse.de> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-07-31 23:46:20 +00:00
__unmap_hugepage_range_final(tlb, vma, start, end, NULL);
mutex_unlock(&vma->vm_file->f_mapping->i_mmap_mutex);
}
} else
unmap_page_range(tlb, vma, start, end, details);
}
}
/**
* unmap_vmas - unmap a range of memory covered by a list of vma's
* @tlb: address of the caller's struct mmu_gather
* @vma: the starting vma
* @start_addr: virtual address at which to start unmapping
* @end_addr: virtual address at which to end unmapping
*
* Unmap all pages in the vma list.
*
* Only addresses between `start' and `end' will be unmapped.
*
* The VMA list must be sorted in ascending virtual address order.
*
* unmap_vmas() assumes that the caller will flush the whole unmapped address
* range after unmap_vmas() returns. So the only responsibility here is to
* ensure that any thus-far unmapped pages are flushed before unmap_vmas()
* drops the lock and schedules.
*/
void unmap_vmas(struct mmu_gather *tlb,
struct vm_area_struct *vma, unsigned long start_addr,
unsigned long end_addr)
{
mmu-notifiers: core With KVM/GFP/XPMEM there isn't just the primary CPU MMU pointing to pages. There are secondary MMUs (with secondary sptes and secondary tlbs) too. sptes in the kvm case are shadow pagetables, but when I say spte in mmu-notifier context, I mean "secondary pte". In GRU case there's no actual secondary pte and there's only a secondary tlb because the GRU secondary MMU has no knowledge about sptes and every secondary tlb miss event in the MMU always generates a page fault that has to be resolved by the CPU (this is not the case of KVM where the a secondary tlb miss will walk sptes in hardware and it will refill the secondary tlb transparently to software if the corresponding spte is present). The same way zap_page_range has to invalidate the pte before freeing the page, the spte (and secondary tlb) must also be invalidated before any page is freed and reused. Currently we take a page_count pin on every page mapped by sptes, but that means the pages can't be swapped whenever they're mapped by any spte because they're part of the guest working set. Furthermore a spte unmap event can immediately lead to a page to be freed when the pin is released (so requiring the same complex and relatively slow tlb_gather smp safe logic we have in zap_page_range and that can be avoided completely if the spte unmap event doesn't require an unpin of the page previously mapped in the secondary MMU). The mmu notifiers allow kvm/GRU/XPMEM to attach to the tsk->mm and know when the VM is swapping or freeing or doing anything on the primary MMU so that the secondary MMU code can drop sptes before the pages are freed, avoiding all page pinning and allowing 100% reliable swapping of guest physical address space. Furthermore it avoids the code that teardown the mappings of the secondary MMU, to implement a logic like tlb_gather in zap_page_range that would require many IPI to flush other cpu tlbs, for each fixed number of spte unmapped. To make an example: if what happens on the primary MMU is a protection downgrade (from writeable to wrprotect) the secondary MMU mappings will be invalidated, and the next secondary-mmu-page-fault will call get_user_pages and trigger a do_wp_page through get_user_pages if it called get_user_pages with write=1, and it'll re-establishing an updated spte or secondary-tlb-mapping on the copied page. Or it will setup a readonly spte or readonly tlb mapping if it's a guest-read, if it calls get_user_pages with write=0. This is just an example. This allows to map any page pointed by any pte (and in turn visible in the primary CPU MMU), into a secondary MMU (be it a pure tlb like GRU, or an full MMU with both sptes and secondary-tlb like the shadow-pagetable layer with kvm), or a remote DMA in software like XPMEM (hence needing of schedule in XPMEM code to send the invalidate to the remote node, while no need to schedule in kvm/gru as it's an immediate event like invalidating primary-mmu pte). At least for KVM without this patch it's impossible to swap guests reliably. And having this feature and removing the page pin allows several other optimizations that simplify life considerably. Dependencies: 1) mm_take_all_locks() to register the mmu notifier when the whole VM isn't doing anything with "mm". This allows mmu notifier users to keep track if the VM is in the middle of the invalidate_range_begin/end critical section with an atomic counter incraese in range_begin and decreased in range_end. No secondary MMU page fault is allowed to map any spte or secondary tlb reference, while the VM is in the middle of range_begin/end as any page returned by get_user_pages in that critical section could later immediately be freed without any further ->invalidate_page notification (invalidate_range_begin/end works on ranges and ->invalidate_page isn't called immediately before freeing the page). To stop all page freeing and pagetable overwrites the mmap_sem must be taken in write mode and all other anon_vma/i_mmap locks must be taken too. 2) It'd be a waste to add branches in the VM if nobody could possibly run KVM/GRU/XPMEM on the kernel, so mmu notifiers will only enabled if CONFIG_KVM=m/y. In the current kernel kvm won't yet take advantage of mmu notifiers, but this already allows to compile a KVM external module against a kernel with mmu notifiers enabled and from the next pull from kvm.git we'll start using them. And GRU/XPMEM will also be able to continue the development by enabling KVM=m in their config, until they submit all GRU/XPMEM GPLv2 code to the mainline kernel. Then they can also enable MMU_NOTIFIERS in the same way KVM does it (even if KVM=n). This guarantees nobody selects MMU_NOTIFIER=y if KVM and GRU and XPMEM are all =n. The mmu_notifier_register call can fail because mm_take_all_locks may be interrupted by a signal and return -EINTR. Because mmu_notifier_reigster is used when a driver startup, a failure can be gracefully handled. Here an example of the change applied to kvm to register the mmu notifiers. Usually when a driver startups other allocations are required anyway and -ENOMEM failure paths exists already. struct kvm *kvm_arch_create_vm(void) { struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL); + int err; if (!kvm) return ERR_PTR(-ENOMEM); INIT_LIST_HEAD(&kvm->arch.active_mmu_pages); + kvm->arch.mmu_notifier.ops = &kvm_mmu_notifier_ops; + err = mmu_notifier_register(&kvm->arch.mmu_notifier, current->mm); + if (err) { + kfree(kvm); + return ERR_PTR(err); + } + return kvm; } mmu_notifier_unregister returns void and it's reliable. The patch also adds a few needed but missing includes that would prevent kernel to compile after these changes on non-x86 archs (x86 didn't need them by luck). [akpm@linux-foundation.org: coding-style fixes] [akpm@linux-foundation.org: fix mm/filemap_xip.c build] [akpm@linux-foundation.org: fix mm/mmu_notifier.c build] Signed-off-by: Andrea Arcangeli <andrea@qumranet.com> Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Christoph Lameter <cl@linux-foundation.org> Cc: Jack Steiner <steiner@sgi.com> Cc: Robin Holt <holt@sgi.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Kanoj Sarcar <kanojsarcar@yahoo.com> Cc: Roland Dreier <rdreier@cisco.com> Cc: Steve Wise <swise@opengridcomputing.com> Cc: Avi Kivity <avi@qumranet.com> Cc: Hugh Dickins <hugh@veritas.com> Cc: Rusty Russell <rusty@rustcorp.com.au> Cc: Anthony Liguori <aliguori@us.ibm.com> Cc: Chris Wright <chrisw@redhat.com> Cc: Marcelo Tosatti <marcelo@kvack.org> Cc: Eric Dumazet <dada1@cosmosbay.com> Cc: "Paul E. McKenney" <paulmck@us.ibm.com> Cc: Izik Eidus <izike@qumranet.com> Cc: Anthony Liguori <aliguori@us.ibm.com> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-28 22:46:29 +00:00
struct mm_struct *mm = vma->vm_mm;
mmu-notifiers: core With KVM/GFP/XPMEM there isn't just the primary CPU MMU pointing to pages. There are secondary MMUs (with secondary sptes and secondary tlbs) too. sptes in the kvm case are shadow pagetables, but when I say spte in mmu-notifier context, I mean "secondary pte". In GRU case there's no actual secondary pte and there's only a secondary tlb because the GRU secondary MMU has no knowledge about sptes and every secondary tlb miss event in the MMU always generates a page fault that has to be resolved by the CPU (this is not the case of KVM where the a secondary tlb miss will walk sptes in hardware and it will refill the secondary tlb transparently to software if the corresponding spte is present). The same way zap_page_range has to invalidate the pte before freeing the page, the spte (and secondary tlb) must also be invalidated before any page is freed and reused. Currently we take a page_count pin on every page mapped by sptes, but that means the pages can't be swapped whenever they're mapped by any spte because they're part of the guest working set. Furthermore a spte unmap event can immediately lead to a page to be freed when the pin is released (so requiring the same complex and relatively slow tlb_gather smp safe logic we have in zap_page_range and that can be avoided completely if the spte unmap event doesn't require an unpin of the page previously mapped in the secondary MMU). The mmu notifiers allow kvm/GRU/XPMEM to attach to the tsk->mm and know when the VM is swapping or freeing or doing anything on the primary MMU so that the secondary MMU code can drop sptes before the pages are freed, avoiding all page pinning and allowing 100% reliable swapping of guest physical address space. Furthermore it avoids the code that teardown the mappings of the secondary MMU, to implement a logic like tlb_gather in zap_page_range that would require many IPI to flush other cpu tlbs, for each fixed number of spte unmapped. To make an example: if what happens on the primary MMU is a protection downgrade (from writeable to wrprotect) the secondary MMU mappings will be invalidated, and the next secondary-mmu-page-fault will call get_user_pages and trigger a do_wp_page through get_user_pages if it called get_user_pages with write=1, and it'll re-establishing an updated spte or secondary-tlb-mapping on the copied page. Or it will setup a readonly spte or readonly tlb mapping if it's a guest-read, if it calls get_user_pages with write=0. This is just an example. This allows to map any page pointed by any pte (and in turn visible in the primary CPU MMU), into a secondary MMU (be it a pure tlb like GRU, or an full MMU with both sptes and secondary-tlb like the shadow-pagetable layer with kvm), or a remote DMA in software like XPMEM (hence needing of schedule in XPMEM code to send the invalidate to the remote node, while no need to schedule in kvm/gru as it's an immediate event like invalidating primary-mmu pte). At least for KVM without this patch it's impossible to swap guests reliably. And having this feature and removing the page pin allows several other optimizations that simplify life considerably. Dependencies: 1) mm_take_all_locks() to register the mmu notifier when the whole VM isn't doing anything with "mm". This allows mmu notifier users to keep track if the VM is in the middle of the invalidate_range_begin/end critical section with an atomic counter incraese in range_begin and decreased in range_end. No secondary MMU page fault is allowed to map any spte or secondary tlb reference, while the VM is in the middle of range_begin/end as any page returned by get_user_pages in that critical section could later immediately be freed without any further ->invalidate_page notification (invalidate_range_begin/end works on ranges and ->invalidate_page isn't called immediately before freeing the page). To stop all page freeing and pagetable overwrites the mmap_sem must be taken in write mode and all other anon_vma/i_mmap locks must be taken too. 2) It'd be a waste to add branches in the VM if nobody could possibly run KVM/GRU/XPMEM on the kernel, so mmu notifiers will only enabled if CONFIG_KVM=m/y. In the current kernel kvm won't yet take advantage of mmu notifiers, but this already allows to compile a KVM external module against a kernel with mmu notifiers enabled and from the next pull from kvm.git we'll start using them. And GRU/XPMEM will also be able to continue the development by enabling KVM=m in their config, until they submit all GRU/XPMEM GPLv2 code to the mainline kernel. Then they can also enable MMU_NOTIFIERS in the same way KVM does it (even if KVM=n). This guarantees nobody selects MMU_NOTIFIER=y if KVM and GRU and XPMEM are all =n. The mmu_notifier_register call can fail because mm_take_all_locks may be interrupted by a signal and return -EINTR. Because mmu_notifier_reigster is used when a driver startup, a failure can be gracefully handled. Here an example of the change applied to kvm to register the mmu notifiers. Usually when a driver startups other allocations are required anyway and -ENOMEM failure paths exists already. struct kvm *kvm_arch_create_vm(void) { struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL); + int err; if (!kvm) return ERR_PTR(-ENOMEM); INIT_LIST_HEAD(&kvm->arch.active_mmu_pages); + kvm->arch.mmu_notifier.ops = &kvm_mmu_notifier_ops; + err = mmu_notifier_register(&kvm->arch.mmu_notifier, current->mm); + if (err) { + kfree(kvm); + return ERR_PTR(err); + } + return kvm; } mmu_notifier_unregister returns void and it's reliable. The patch also adds a few needed but missing includes that would prevent kernel to compile after these changes on non-x86 archs (x86 didn't need them by luck). [akpm@linux-foundation.org: coding-style fixes] [akpm@linux-foundation.org: fix mm/filemap_xip.c build] [akpm@linux-foundation.org: fix mm/mmu_notifier.c build] Signed-off-by: Andrea Arcangeli <andrea@qumranet.com> Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Christoph Lameter <cl@linux-foundation.org> Cc: Jack Steiner <steiner@sgi.com> Cc: Robin Holt <holt@sgi.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Kanoj Sarcar <kanojsarcar@yahoo.com> Cc: Roland Dreier <rdreier@cisco.com> Cc: Steve Wise <swise@opengridcomputing.com> Cc: Avi Kivity <avi@qumranet.com> Cc: Hugh Dickins <hugh@veritas.com> Cc: Rusty Russell <rusty@rustcorp.com.au> Cc: Anthony Liguori <aliguori@us.ibm.com> Cc: Chris Wright <chrisw@redhat.com> Cc: Marcelo Tosatti <marcelo@kvack.org> Cc: Eric Dumazet <dada1@cosmosbay.com> Cc: "Paul E. McKenney" <paulmck@us.ibm.com> Cc: Izik Eidus <izike@qumranet.com> Cc: Anthony Liguori <aliguori@us.ibm.com> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-28 22:46:29 +00:00
mmu_notifier_invalidate_range_start(mm, start_addr, end_addr);
for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next)
unmap_single_vma(tlb, vma, start_addr, end_addr, NULL);
mmu-notifiers: core With KVM/GFP/XPMEM there isn't just the primary CPU MMU pointing to pages. There are secondary MMUs (with secondary sptes and secondary tlbs) too. sptes in the kvm case are shadow pagetables, but when I say spte in mmu-notifier context, I mean "secondary pte". In GRU case there's no actual secondary pte and there's only a secondary tlb because the GRU secondary MMU has no knowledge about sptes and every secondary tlb miss event in the MMU always generates a page fault that has to be resolved by the CPU (this is not the case of KVM where the a secondary tlb miss will walk sptes in hardware and it will refill the secondary tlb transparently to software if the corresponding spte is present). The same way zap_page_range has to invalidate the pte before freeing the page, the spte (and secondary tlb) must also be invalidated before any page is freed and reused. Currently we take a page_count pin on every page mapped by sptes, but that means the pages can't be swapped whenever they're mapped by any spte because they're part of the guest working set. Furthermore a spte unmap event can immediately lead to a page to be freed when the pin is released (so requiring the same complex and relatively slow tlb_gather smp safe logic we have in zap_page_range and that can be avoided completely if the spte unmap event doesn't require an unpin of the page previously mapped in the secondary MMU). The mmu notifiers allow kvm/GRU/XPMEM to attach to the tsk->mm and know when the VM is swapping or freeing or doing anything on the primary MMU so that the secondary MMU code can drop sptes before the pages are freed, avoiding all page pinning and allowing 100% reliable swapping of guest physical address space. Furthermore it avoids the code that teardown the mappings of the secondary MMU, to implement a logic like tlb_gather in zap_page_range that would require many IPI to flush other cpu tlbs, for each fixed number of spte unmapped. To make an example: if what happens on the primary MMU is a protection downgrade (from writeable to wrprotect) the secondary MMU mappings will be invalidated, and the next secondary-mmu-page-fault will call get_user_pages and trigger a do_wp_page through get_user_pages if it called get_user_pages with write=1, and it'll re-establishing an updated spte or secondary-tlb-mapping on the copied page. Or it will setup a readonly spte or readonly tlb mapping if it's a guest-read, if it calls get_user_pages with write=0. This is just an example. This allows to map any page pointed by any pte (and in turn visible in the primary CPU MMU), into a secondary MMU (be it a pure tlb like GRU, or an full MMU with both sptes and secondary-tlb like the shadow-pagetable layer with kvm), or a remote DMA in software like XPMEM (hence needing of schedule in XPMEM code to send the invalidate to the remote node, while no need to schedule in kvm/gru as it's an immediate event like invalidating primary-mmu pte). At least for KVM without this patch it's impossible to swap guests reliably. And having this feature and removing the page pin allows several other optimizations that simplify life considerably. Dependencies: 1) mm_take_all_locks() to register the mmu notifier when the whole VM isn't doing anything with "mm". This allows mmu notifier users to keep track if the VM is in the middle of the invalidate_range_begin/end critical section with an atomic counter incraese in range_begin and decreased in range_end. No secondary MMU page fault is allowed to map any spte or secondary tlb reference, while the VM is in the middle of range_begin/end as any page returned by get_user_pages in that critical section could later immediately be freed without any further ->invalidate_page notification (invalidate_range_begin/end works on ranges and ->invalidate_page isn't called immediately before freeing the page). To stop all page freeing and pagetable overwrites the mmap_sem must be taken in write mode and all other anon_vma/i_mmap locks must be taken too. 2) It'd be a waste to add branches in the VM if nobody could possibly run KVM/GRU/XPMEM on the kernel, so mmu notifiers will only enabled if CONFIG_KVM=m/y. In the current kernel kvm won't yet take advantage of mmu notifiers, but this already allows to compile a KVM external module against a kernel with mmu notifiers enabled and from the next pull from kvm.git we'll start using them. And GRU/XPMEM will also be able to continue the development by enabling KVM=m in their config, until they submit all GRU/XPMEM GPLv2 code to the mainline kernel. Then they can also enable MMU_NOTIFIERS in the same way KVM does it (even if KVM=n). This guarantees nobody selects MMU_NOTIFIER=y if KVM and GRU and XPMEM are all =n. The mmu_notifier_register call can fail because mm_take_all_locks may be interrupted by a signal and return -EINTR. Because mmu_notifier_reigster is used when a driver startup, a failure can be gracefully handled. Here an example of the change applied to kvm to register the mmu notifiers. Usually when a driver startups other allocations are required anyway and -ENOMEM failure paths exists already. struct kvm *kvm_arch_create_vm(void) { struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL); + int err; if (!kvm) return ERR_PTR(-ENOMEM); INIT_LIST_HEAD(&kvm->arch.active_mmu_pages); + kvm->arch.mmu_notifier.ops = &kvm_mmu_notifier_ops; + err = mmu_notifier_register(&kvm->arch.mmu_notifier, current->mm); + if (err) { + kfree(kvm); + return ERR_PTR(err); + } + return kvm; } mmu_notifier_unregister returns void and it's reliable. The patch also adds a few needed but missing includes that would prevent kernel to compile after these changes on non-x86 archs (x86 didn't need them by luck). [akpm@linux-foundation.org: coding-style fixes] [akpm@linux-foundation.org: fix mm/filemap_xip.c build] [akpm@linux-foundation.org: fix mm/mmu_notifier.c build] Signed-off-by: Andrea Arcangeli <andrea@qumranet.com> Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Christoph Lameter <cl@linux-foundation.org> Cc: Jack Steiner <steiner@sgi.com> Cc: Robin Holt <holt@sgi.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Kanoj Sarcar <kanojsarcar@yahoo.com> Cc: Roland Dreier <rdreier@cisco.com> Cc: Steve Wise <swise@opengridcomputing.com> Cc: Avi Kivity <avi@qumranet.com> Cc: Hugh Dickins <hugh@veritas.com> Cc: Rusty Russell <rusty@rustcorp.com.au> Cc: Anthony Liguori <aliguori@us.ibm.com> Cc: Chris Wright <chrisw@redhat.com> Cc: Marcelo Tosatti <marcelo@kvack.org> Cc: Eric Dumazet <dada1@cosmosbay.com> Cc: "Paul E. McKenney" <paulmck@us.ibm.com> Cc: Izik Eidus <izike@qumranet.com> Cc: Anthony Liguori <aliguori@us.ibm.com> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-28 22:46:29 +00:00
mmu_notifier_invalidate_range_end(mm, start_addr, end_addr);
}
/**
* zap_page_range - remove user pages in a given range
* @vma: vm_area_struct holding the applicable pages
* @start: starting address of pages to zap
* @size: number of bytes to zap
* @details: details of nonlinear truncation or shared cache invalidation
*
* Caller must protect the VMA list
*/
void zap_page_range(struct vm_area_struct *vma, unsigned long start,
unsigned long size, struct zap_details *details)
{
struct mm_struct *mm = vma->vm_mm;
mm: mmu_gather rework Rework the existing mmu_gather infrastructure. The direct purpose of these patches was to allow preemptible mmu_gather, but even without that I think these patches provide an improvement to the status quo. The first 9 patches rework the mmu_gather infrastructure. For review purpose I've split them into generic and per-arch patches with the last of those a generic cleanup. The next patch provides generic RCU page-table freeing, and the followup is a patch converting s390 to use this. I've also got 4 patches from DaveM lined up (not included in this series) that uses this to implement gup_fast() for sparc64. Then there is one patch that extends the generic mmu_gather batching. After that follow the mm preemptibility patches, these make part of the mm a lot more preemptible. It converts i_mmap_lock and anon_vma->lock to mutexes which together with the mmu_gather rework makes mmu_gather preemptible as well. Making i_mmap_lock a mutex also enables a clean-up of the truncate code. This also allows for preemptible mmu_notifiers, something that XPMEM I think wants. Furthermore, it removes the new and universially detested unmap_mutex. This patch: Remove the first obstacle towards a fully preemptible mmu_gather. The current scheme assumes mmu_gather is always done with preemption disabled and uses per-cpu storage for the page batches. Change this to try and allocate a page for batching and in case of failure, use a small on-stack array to make some progress. Preemptible mmu_gather is desired in general and usable once i_mmap_lock becomes a mutex. Doing it before the mutex conversion saves us from having to rework the code by moving the mmu_gather bits inside the pte_lock. Also avoid flushing the tlb batches from under the pte lock, this is useful even without the i_mmap_lock conversion as it significantly reduces pte lock hold times. [akpm@linux-foundation.org: fix comment tpyo] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Miller <davem@davemloft.net> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Jeff Dike <jdike@addtoit.com> Cc: Richard Weinberger <richard@nod.at> Cc: Tony Luck <tony.luck@intel.com> Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Acked-by: Hugh Dickins <hughd@google.com> Acked-by: Mel Gorman <mel@csn.ul.ie> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Namhyung Kim <namhyung@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-05-25 00:11:45 +00:00
struct mmu_gather tlb;
unsigned long end = start + size;
lru_add_drain();
mm: mmu_gather rework Rework the existing mmu_gather infrastructure. The direct purpose of these patches was to allow preemptible mmu_gather, but even without that I think these patches provide an improvement to the status quo. The first 9 patches rework the mmu_gather infrastructure. For review purpose I've split them into generic and per-arch patches with the last of those a generic cleanup. The next patch provides generic RCU page-table freeing, and the followup is a patch converting s390 to use this. I've also got 4 patches from DaveM lined up (not included in this series) that uses this to implement gup_fast() for sparc64. Then there is one patch that extends the generic mmu_gather batching. After that follow the mm preemptibility patches, these make part of the mm a lot more preemptible. It converts i_mmap_lock and anon_vma->lock to mutexes which together with the mmu_gather rework makes mmu_gather preemptible as well. Making i_mmap_lock a mutex also enables a clean-up of the truncate code. This also allows for preemptible mmu_notifiers, something that XPMEM I think wants. Furthermore, it removes the new and universially detested unmap_mutex. This patch: Remove the first obstacle towards a fully preemptible mmu_gather. The current scheme assumes mmu_gather is always done with preemption disabled and uses per-cpu storage for the page batches. Change this to try and allocate a page for batching and in case of failure, use a small on-stack array to make some progress. Preemptible mmu_gather is desired in general and usable once i_mmap_lock becomes a mutex. Doing it before the mutex conversion saves us from having to rework the code by moving the mmu_gather bits inside the pte_lock. Also avoid flushing the tlb batches from under the pte lock, this is useful even without the i_mmap_lock conversion as it significantly reduces pte lock hold times. [akpm@linux-foundation.org: fix comment tpyo] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Miller <davem@davemloft.net> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Jeff Dike <jdike@addtoit.com> Cc: Richard Weinberger <richard@nod.at> Cc: Tony Luck <tony.luck@intel.com> Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Acked-by: Hugh Dickins <hughd@google.com> Acked-by: Mel Gorman <mel@csn.ul.ie> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Namhyung Kim <namhyung@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-05-25 00:11:45 +00:00
tlb_gather_mmu(&tlb, mm, 0);
[PATCH] mm: update_hiwaters just in time update_mem_hiwater has attracted various criticisms, in particular from those concerned with mm scalability. Originally it was called whenever rss or total_vm got raised. Then many of those callsites were replaced by a timer tick call from account_system_time. Now Frank van Maarseveen reports that to be found inadequate. How about this? Works for Frank. Replace update_mem_hiwater, a poor combination of two unrelated ops, by macros update_hiwater_rss and update_hiwater_vm. Don't attempt to keep mm->hiwater_rss up to date at timer tick, nor every time we raise rss (usually by 1): those are hot paths. Do the opposite, update only when about to lower rss (usually by many), or just before final accounting in do_exit. Handle mm->hiwater_vm in the same way, though it's much less of an issue. Demand that whoever collects these hiwater statistics do the work of taking the maximum with rss or total_vm. And there has been no collector of these hiwater statistics in the tree. The new convention needs an example, so match Frank's usage by adding a VmPeak line above VmSize to /proc/<pid>/status, and also a VmHWM line above VmRSS (High-Water-Mark or High-Water-Memory). There was a particular anomaly during mremap move, that hiwater_vm might be captured too high. A fleeting such anomaly remains, but it's quickly corrected now, whereas before it would stick. What locking? None: if the app is racy then these statistics will be racy, it's not worth any overhead to make them exact. But whenever it suits, hiwater_vm is updated under exclusive mmap_sem, and hiwater_rss under page_table_lock (for now) or with preemption disabled (later on): without going to any trouble, minimize the time between reading current values and updating, to minimize those occasions when a racing thread bumps a count up and back down in between. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:18 +00:00
update_hiwater_rss(mm);
mmu_notifier_invalidate_range_start(mm, start, end);
for ( ; vma && vma->vm_start < end; vma = vma->vm_next)
unmap_single_vma(&tlb, vma, start, end, details);
mmu_notifier_invalidate_range_end(mm, start, end);
tlb_finish_mmu(&tlb, start, end);
}
/**
* zap_page_range_single - remove user pages in a given range
* @vma: vm_area_struct holding the applicable pages
* @address: starting address of pages to zap
* @size: number of bytes to zap
* @details: details of nonlinear truncation or shared cache invalidation
*
* The range must fit into one VMA.
*/
static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address,
unsigned long size, struct zap_details *details)
{
struct mm_struct *mm = vma->vm_mm;
mm: mmu_gather rework Rework the existing mmu_gather infrastructure. The direct purpose of these patches was to allow preemptible mmu_gather, but even without that I think these patches provide an improvement to the status quo. The first 9 patches rework the mmu_gather infrastructure. For review purpose I've split them into generic and per-arch patches with the last of those a generic cleanup. The next patch provides generic RCU page-table freeing, and the followup is a patch converting s390 to use this. I've also got 4 patches from DaveM lined up (not included in this series) that uses this to implement gup_fast() for sparc64. Then there is one patch that extends the generic mmu_gather batching. After that follow the mm preemptibility patches, these make part of the mm a lot more preemptible. It converts i_mmap_lock and anon_vma->lock to mutexes which together with the mmu_gather rework makes mmu_gather preemptible as well. Making i_mmap_lock a mutex also enables a clean-up of the truncate code. This also allows for preemptible mmu_notifiers, something that XPMEM I think wants. Furthermore, it removes the new and universially detested unmap_mutex. This patch: Remove the first obstacle towards a fully preemptible mmu_gather. The current scheme assumes mmu_gather is always done with preemption disabled and uses per-cpu storage for the page batches. Change this to try and allocate a page for batching and in case of failure, use a small on-stack array to make some progress. Preemptible mmu_gather is desired in general and usable once i_mmap_lock becomes a mutex. Doing it before the mutex conversion saves us from having to rework the code by moving the mmu_gather bits inside the pte_lock. Also avoid flushing the tlb batches from under the pte lock, this is useful even without the i_mmap_lock conversion as it significantly reduces pte lock hold times. [akpm@linux-foundation.org: fix comment tpyo] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Miller <davem@davemloft.net> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Jeff Dike <jdike@addtoit.com> Cc: Richard Weinberger <richard@nod.at> Cc: Tony Luck <tony.luck@intel.com> Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Acked-by: Hugh Dickins <hughd@google.com> Acked-by: Mel Gorman <mel@csn.ul.ie> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Namhyung Kim <namhyung@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-05-25 00:11:45 +00:00
struct mmu_gather tlb;
unsigned long end = address + size;
lru_add_drain();
mm: mmu_gather rework Rework the existing mmu_gather infrastructure. The direct purpose of these patches was to allow preemptible mmu_gather, but even without that I think these patches provide an improvement to the status quo. The first 9 patches rework the mmu_gather infrastructure. For review purpose I've split them into generic and per-arch patches with the last of those a generic cleanup. The next patch provides generic RCU page-table freeing, and the followup is a patch converting s390 to use this. I've also got 4 patches from DaveM lined up (not included in this series) that uses this to implement gup_fast() for sparc64. Then there is one patch that extends the generic mmu_gather batching. After that follow the mm preemptibility patches, these make part of the mm a lot more preemptible. It converts i_mmap_lock and anon_vma->lock to mutexes which together with the mmu_gather rework makes mmu_gather preemptible as well. Making i_mmap_lock a mutex also enables a clean-up of the truncate code. This also allows for preemptible mmu_notifiers, something that XPMEM I think wants. Furthermore, it removes the new and universially detested unmap_mutex. This patch: Remove the first obstacle towards a fully preemptible mmu_gather. The current scheme assumes mmu_gather is always done with preemption disabled and uses per-cpu storage for the page batches. Change this to try and allocate a page for batching and in case of failure, use a small on-stack array to make some progress. Preemptible mmu_gather is desired in general and usable once i_mmap_lock becomes a mutex. Doing it before the mutex conversion saves us from having to rework the code by moving the mmu_gather bits inside the pte_lock. Also avoid flushing the tlb batches from under the pte lock, this is useful even without the i_mmap_lock conversion as it significantly reduces pte lock hold times. [akpm@linux-foundation.org: fix comment tpyo] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Miller <davem@davemloft.net> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Jeff Dike <jdike@addtoit.com> Cc: Richard Weinberger <richard@nod.at> Cc: Tony Luck <tony.luck@intel.com> Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Acked-by: Hugh Dickins <hughd@google.com> Acked-by: Mel Gorman <mel@csn.ul.ie> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Namhyung Kim <namhyung@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-05-25 00:11:45 +00:00
tlb_gather_mmu(&tlb, mm, 0);
[PATCH] mm: update_hiwaters just in time update_mem_hiwater has attracted various criticisms, in particular from those concerned with mm scalability. Originally it was called whenever rss or total_vm got raised. Then many of those callsites were replaced by a timer tick call from account_system_time. Now Frank van Maarseveen reports that to be found inadequate. How about this? Works for Frank. Replace update_mem_hiwater, a poor combination of two unrelated ops, by macros update_hiwater_rss and update_hiwater_vm. Don't attempt to keep mm->hiwater_rss up to date at timer tick, nor every time we raise rss (usually by 1): those are hot paths. Do the opposite, update only when about to lower rss (usually by many), or just before final accounting in do_exit. Handle mm->hiwater_vm in the same way, though it's much less of an issue. Demand that whoever collects these hiwater statistics do the work of taking the maximum with rss or total_vm. And there has been no collector of these hiwater statistics in the tree. The new convention needs an example, so match Frank's usage by adding a VmPeak line above VmSize to /proc/<pid>/status, and also a VmHWM line above VmRSS (High-Water-Mark or High-Water-Memory). There was a particular anomaly during mremap move, that hiwater_vm might be captured too high. A fleeting such anomaly remains, but it's quickly corrected now, whereas before it would stick. What locking? None: if the app is racy then these statistics will be racy, it's not worth any overhead to make them exact. But whenever it suits, hiwater_vm is updated under exclusive mmap_sem, and hiwater_rss under page_table_lock (for now) or with preemption disabled (later on): without going to any trouble, minimize the time between reading current values and updating, to minimize those occasions when a racing thread bumps a count up and back down in between. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:18 +00:00
update_hiwater_rss(mm);
mmu_notifier_invalidate_range_start(mm, address, end);
unmap_single_vma(&tlb, vma, address, end, details);
mmu_notifier_invalidate_range_end(mm, address, end);
mm: mmu_gather rework Rework the existing mmu_gather infrastructure. The direct purpose of these patches was to allow preemptible mmu_gather, but even without that I think these patches provide an improvement to the status quo. The first 9 patches rework the mmu_gather infrastructure. For review purpose I've split them into generic and per-arch patches with the last of those a generic cleanup. The next patch provides generic RCU page-table freeing, and the followup is a patch converting s390 to use this. I've also got 4 patches from DaveM lined up (not included in this series) that uses this to implement gup_fast() for sparc64. Then there is one patch that extends the generic mmu_gather batching. After that follow the mm preemptibility patches, these make part of the mm a lot more preemptible. It converts i_mmap_lock and anon_vma->lock to mutexes which together with the mmu_gather rework makes mmu_gather preemptible as well. Making i_mmap_lock a mutex also enables a clean-up of the truncate code. This also allows for preemptible mmu_notifiers, something that XPMEM I think wants. Furthermore, it removes the new and universially detested unmap_mutex. This patch: Remove the first obstacle towards a fully preemptible mmu_gather. The current scheme assumes mmu_gather is always done with preemption disabled and uses per-cpu storage for the page batches. Change this to try and allocate a page for batching and in case of failure, use a small on-stack array to make some progress. Preemptible mmu_gather is desired in general and usable once i_mmap_lock becomes a mutex. Doing it before the mutex conversion saves us from having to rework the code by moving the mmu_gather bits inside the pte_lock. Also avoid flushing the tlb batches from under the pte lock, this is useful even without the i_mmap_lock conversion as it significantly reduces pte lock hold times. [akpm@linux-foundation.org: fix comment tpyo] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Miller <davem@davemloft.net> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Jeff Dike <jdike@addtoit.com> Cc: Richard Weinberger <richard@nod.at> Cc: Tony Luck <tony.luck@intel.com> Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Acked-by: Hugh Dickins <hughd@google.com> Acked-by: Mel Gorman <mel@csn.ul.ie> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Namhyung Kim <namhyung@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-05-25 00:11:45 +00:00
tlb_finish_mmu(&tlb, address, end);
}
/**
* zap_vma_ptes - remove ptes mapping the vma
* @vma: vm_area_struct holding ptes to be zapped
* @address: starting address of pages to zap
* @size: number of bytes to zap
*
* This function only unmaps ptes assigned to VM_PFNMAP vmas.
*
* The entire address range must be fully contained within the vma.
*
* Returns 0 if successful.
*/
int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
unsigned long size)
{
if (address < vma->vm_start || address + size > vma->vm_end ||
!(vma->vm_flags & VM_PFNMAP))
return -1;
zap_page_range_single(vma, address, size, NULL);
return 0;
}
EXPORT_SYMBOL_GPL(zap_vma_ptes);
/**
* follow_page - look up a page descriptor from a user-virtual address
* @vma: vm_area_struct mapping @address
* @address: virtual address to look up
* @flags: flags modifying lookup behaviour
*
* @flags can have FOLL_ flags set, defined in <linux/mm.h>
*
* Returns the mapped (struct page *), %NULL if no mapping exists, or
* an error pointer if there is a mapping to something not represented
* by a page descriptor (see also vm_normal_page()).
*/
struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
unsigned int flags)
{
pgd_t *pgd;
pud_t *pud;
pmd_t *pmd;
pte_t *ptep, pte;
spinlock_t *ptl;
struct page *page;
struct mm_struct *mm = vma->vm_mm;
page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
if (!IS_ERR(page)) {
BUG_ON(flags & FOLL_GET);
goto out;
}
page = NULL;
pgd = pgd_offset(mm, address);
if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
goto no_page_table;
pud = pud_offset(pgd, address);
if (pud_none(*pud))
goto no_page_table;
if (pud_huge(*pud) && vma->vm_flags & VM_HUGETLB) {
BUG_ON(flags & FOLL_GET);
page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE);
goto out;
}
if (unlikely(pud_bad(*pud)))
goto no_page_table;
pmd = pmd_offset(pud, address);
x86: fix PAE pmd_bad bootup warning Fix warning from pmd_bad() at bootup on a HIGHMEM64G HIGHPTE x86_32. That came from 9fc34113f6880b215cbea4e7017fc818700384c2 x86: debug pmd_bad(); but we understand now that the typecasting was wrong for PAE in the previous version: pagetable pages above 4GB looked bad and stopped Arjan from booting. And revert that cded932b75ab0a5f9181ee3da34a0a488d1a14fd x86: fix pmd_bad and pud_bad to support huge pages. It was the wrong way round: we shouldn't weaken every pmd_bad and pud_bad check to let huge pages slip through - in part they check that we _don't_ have a huge page where it's not expected. Put the x86 pmd_bad() and pud_bad() definitions back to what they have long been: they can be improved (x86_32 should use PTE_MASK, to stop PAE thinking junk in the upper word is good; and x86_64 should follow x86_32's stricter comparison, to stop thinking any subset of required bits is good); but that should be a later patch. Fix Hans' good observation that follow_page() will never find pmd_huge() because that would have already failed the pmd_bad test: test pmd_huge in between the pmd_none and pmd_bad tests. Tighten x86's pmd_huge() check? No, once it's a hugepage entry, it can get quite far from a good pmd: for example, PROT_NONE leaves it with only ACCESSED of the KERN_PGTABLE bits. However... though follow_page() contains this and another test for huge pages, so it's nice to keep it working on them, where does it actually get called on a huge page? get_user_pages() checks is_vm_hugetlb_page(vma) to to call alternative hugetlb processing, as does unmap_vmas() and others. Signed-off-by: Hugh Dickins <hugh@veritas.com> Earlier-version-tested-by: Ingo Molnar <mingo@elte.hu> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Jeff Chua <jeff.chua.linux@gmail.com> Cc: Hans Rosenfeld <hans.rosenfeld@amd.com> Cc: Arjan van de Ven <arjan@linux.intel.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-05-06 19:49:23 +00:00
if (pmd_none(*pmd))
goto no_page_table;
thp: transparent hugepage core Lately I've been working to make KVM use hugepages transparently without the usual restrictions of hugetlbfs. Some of the restrictions I'd like to see removed: 1) hugepages have to be swappable or the guest physical memory remains locked in RAM and can't be paged out to swap 2) if a hugepage allocation fails, regular pages should be allocated instead and mixed in the same vma without any failure and without userland noticing 3) if some task quits and more hugepages become available in the buddy, guest physical memory backed by regular pages should be relocated on hugepages automatically in regions under madvise(MADV_HUGEPAGE) (ideally event driven by waking up the kernel deamon if the order=HPAGE_PMD_SHIFT-PAGE_SHIFT list becomes not null) 4) avoidance of reservation and maximization of use of hugepages whenever possible. Reservation (needed to avoid runtime fatal faliures) may be ok for 1 machine with 1 database with 1 database cache with 1 database cache size known at boot time. It's definitely not feasible with a virtualization hypervisor usage like RHEV-H that runs an unknown number of virtual machines with an unknown size of each virtual machine with an unknown amount of pagecache that could be potentially useful in the host for guest not using O_DIRECT (aka cache=off). hugepages in the virtualization hypervisor (and also in the guest!) are much more important than in a regular host not using virtualization, becasue with NPT/EPT they decrease the tlb-miss cacheline accesses from 24 to 19 in case only the hypervisor uses transparent hugepages, and they decrease the tlb-miss cacheline accesses from 19 to 15 in case both the linux hypervisor and the linux guest both uses this patch (though the guest will limit the addition speedup to anonymous regions only for now...). Even more important is that the tlb miss handler is much slower on a NPT/EPT guest than for a regular shadow paging or no-virtualization scenario. So maximizing the amount of virtual memory cached by the TLB pays off significantly more with NPT/EPT than without (even if there would be no significant speedup in the tlb-miss runtime). The first (and more tedious) part of this work requires allowing the VM to handle anonymous hugepages mixed with regular pages transparently on regular anonymous vmas. This is what this patch tries to achieve in the least intrusive possible way. We want hugepages and hugetlb to be used in a way so that all applications can benefit without changes (as usual we leverage the KVM virtualization design: by improving the Linux VM at large, KVM gets the performance boost too). The most important design choice is: always fallback to 4k allocation if the hugepage allocation fails! This is the _very_ opposite of some large pagecache patches that failed with -EIO back then if a 64k (or similar) allocation failed... Second important decision (to reduce the impact of the feature on the existing pagetable handling code) is that at any time we can split an hugepage into 512 regular pages and it has to be done with an operation that can't fail. This way the reliability of the swapping isn't decreased (no need to allocate memory when we are short on memory to swap) and it's trivial to plug a split_huge_page* one-liner where needed without polluting the VM. Over time we can teach mprotect, mremap and friends to handle pmd_trans_huge natively without calling split_huge_page*. The fact it can't fail isn't just for swap: if split_huge_page would return -ENOMEM (instead of the current void) we'd need to rollback the mprotect from the middle of it (ideally including undoing the split_vma) which would be a big change and in the very wrong direction (it'd likely be simpler not to call split_huge_page at all and to teach mprotect and friends to handle hugepages instead of rolling them back from the middle). In short the very value of split_huge_page is that it can't fail. The collapsing and madvise(MADV_HUGEPAGE) part will remain separated and incremental and it'll just be an "harmless" addition later if this initial part is agreed upon. It also should be noted that locking-wise replacing regular pages with hugepages is going to be very easy if compared to what I'm doing below in split_huge_page, as it will only happen when page_count(page) matches page_mapcount(page) if we can take the PG_lock and mmap_sem in write mode. collapse_huge_page will be a "best effort" that (unlike split_huge_page) can fail at the minimal sign of trouble and we can try again later. collapse_huge_page will be similar to how KSM works and the madvise(MADV_HUGEPAGE) will work similar to madvise(MADV_MERGEABLE). The default I like is that transparent hugepages are used at page fault time. This can be changed with /sys/kernel/mm/transparent_hugepage/enabled. The control knob can be set to three values "always", "madvise", "never" which mean respectively that hugepages are always used, or only inside madvise(MADV_HUGEPAGE) regions, or never used. /sys/kernel/mm/transparent_hugepage/defrag instead controls if the hugepage allocation should defrag memory aggressively "always", only inside "madvise" regions, or "never". The pmd_trans_splitting/pmd_trans_huge locking is very solid. The put_page (from get_user_page users that can't use mmu notifier like O_DIRECT) that runs against a __split_huge_page_refcount instead was a pain to serialize in a way that would result always in a coherent page count for both tail and head. I think my locking solution with a compound_lock taken only after the page_first is valid and is still a PageHead should be safe but it surely needs review from SMP race point of view. In short there is no current existing way to serialize the O_DIRECT final put_page against split_huge_page_refcount so I had to invent a new one (O_DIRECT loses knowledge on the mapping status by the time gup_fast returns so...). And I didn't want to impact all gup/gup_fast users for now, maybe if we change the gup interface substantially we can avoid this locking, I admit I didn't think too much about it because changing the gup unpinning interface would be invasive. If we ignored O_DIRECT we could stick to the existing compound refcounting code, by simply adding a get_user_pages_fast_flags(foll_flags) where KVM (and any other mmu notifier user) would call it without FOLL_GET (and if FOLL_GET isn't set we'd just BUG_ON if nobody registered itself in the current task mmu notifier list yet). But O_DIRECT is fundamental for decent performance of virtualized I/O on fast storage so we can't avoid it to solve the race of put_page against split_huge_page_refcount to achieve a complete hugepage feature for KVM. Swap and oom works fine (well just like with regular pages ;). MMU notifier is handled transparently too, with the exception of the young bit on the pmd, that didn't have a range check but I think KVM will be fine because the whole point of hugepages is that EPT/NPT will also use a huge pmd when they notice gup returns pages with PageCompound set, so they won't care of a range and there's just the pmd young bit to check in that case. NOTE: in some cases if the L2 cache is small, this may slowdown and waste memory during COWs because 4M of memory are accessed in a single fault instead of 8k (the payoff is that after COW the program can run faster). So we might want to switch the copy_huge_page (and clear_huge_page too) to not temporal stores. I also extensively researched ways to avoid this cache trashing with a full prefault logic that would cow in 8k/16k/32k/64k up to 1M (I can send those patches that fully implemented prefault) but I concluded they're not worth it and they add an huge additional complexity and they remove all tlb benefits until the full hugepage has been faulted in, to save a little bit of memory and some cache during app startup, but they still don't improve substantially the cache-trashing during startup if the prefault happens in >4k chunks. One reason is that those 4k pte entries copied are still mapped on a perfectly cache-colored hugepage, so the trashing is the worst one can generate in those copies (cow of 4k page copies aren't so well colored so they trashes less, but again this results in software running faster after the page fault). Those prefault patches allowed things like a pte where post-cow pages were local 4k regular anon pages and the not-yet-cowed pte entries were pointing in the middle of some hugepage mapped read-only. If it doesn't payoff substantially with todays hardware it will payoff even less in the future with larger l2 caches, and the prefault logic would blot the VM a lot. If one is emebdded transparent_hugepage can be disabled during boot with sysfs or with the boot commandline parameter transparent_hugepage=0 (or transparent_hugepage=2 to restrict hugepages inside madvise regions) that will ensure not a single hugepage is allocated at boot time. It is simple enough to just disable transparent hugepage globally and let transparent hugepages be allocated selectively by applications in the MADV_HUGEPAGE region (both at page fault time, and if enabled with the collapse_huge_page too through the kernel daemon). This patch supports only hugepages mapped in the pmd, archs that have smaller hugepages will not fit in this patch alone. Also some archs like power have certain tlb limits that prevents mixing different page size in the same regions so they will not fit in this framework that requires "graceful fallback" to basic PAGE_SIZE in case of physical memory fragmentation. hugetlbfs remains a perfect fit for those because its software limits happen to match the hardware limits. hugetlbfs also remains a perfect fit for hugepage sizes like 1GByte that cannot be hoped to be found not fragmented after a certain system uptime and that would be very expensive to defragment with relocation, so requiring reservation. hugetlbfs is the "reservation way", the point of transparent hugepages is not to have any reservation at all and maximizing the use of cache and hugepages at all times automatically. Some performance result: vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largep ages3 memset page fault 1566023 memset tlb miss 453854 memset second tlb miss 453321 random access tlb miss 41635 random access second tlb miss 41658 vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largepages3 memset page fault 1566471 memset tlb miss 453375 memset second tlb miss 453320 random access tlb miss 41636 random access second tlb miss 41637 vmx andrea # ./largepages3 memset page fault 1566642 memset tlb miss 453417 memset second tlb miss 453313 random access tlb miss 41630 random access second tlb miss 41647 vmx andrea # ./largepages3 memset page fault 1566872 memset tlb miss 453418 memset second tlb miss 453315 random access tlb miss 41618 random access second tlb miss 41659 vmx andrea # echo 0 > /proc/sys/vm/transparent_hugepage vmx andrea # ./largepages3 memset page fault 2182476 memset tlb miss 460305 memset second tlb miss 460179 random access tlb miss 44483 random access second tlb miss 44186 vmx andrea # ./largepages3 memset page fault 2182791 memset tlb miss 460742 memset second tlb miss 459962 random access tlb miss 43981 random access second tlb miss 43988 ============ #include <stdio.h> #include <stdlib.h> #include <string.h> #include <sys/time.h> #define SIZE (3UL*1024*1024*1024) int main() { char *p = malloc(SIZE), *p2; struct timeval before, after; gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset page fault %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); return 0; } ============ Signed-off-by: Andrea Arcangeli <aarcange@redhat.com> Acked-by: Rik van Riel <riel@redhat.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-13 23:46:52 +00:00
if (pmd_huge(*pmd) && vma->vm_flags & VM_HUGETLB) {
BUG_ON(flags & FOLL_GET);
page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
goto out;
}
if ((flags & FOLL_NUMA) && pmd_numa(*pmd))
goto no_page_table;
thp: transparent hugepage core Lately I've been working to make KVM use hugepages transparently without the usual restrictions of hugetlbfs. Some of the restrictions I'd like to see removed: 1) hugepages have to be swappable or the guest physical memory remains locked in RAM and can't be paged out to swap 2) if a hugepage allocation fails, regular pages should be allocated instead and mixed in the same vma without any failure and without userland noticing 3) if some task quits and more hugepages become available in the buddy, guest physical memory backed by regular pages should be relocated on hugepages automatically in regions under madvise(MADV_HUGEPAGE) (ideally event driven by waking up the kernel deamon if the order=HPAGE_PMD_SHIFT-PAGE_SHIFT list becomes not null) 4) avoidance of reservation and maximization of use of hugepages whenever possible. Reservation (needed to avoid runtime fatal faliures) may be ok for 1 machine with 1 database with 1 database cache with 1 database cache size known at boot time. It's definitely not feasible with a virtualization hypervisor usage like RHEV-H that runs an unknown number of virtual machines with an unknown size of each virtual machine with an unknown amount of pagecache that could be potentially useful in the host for guest not using O_DIRECT (aka cache=off). hugepages in the virtualization hypervisor (and also in the guest!) are much more important than in a regular host not using virtualization, becasue with NPT/EPT they decrease the tlb-miss cacheline accesses from 24 to 19 in case only the hypervisor uses transparent hugepages, and they decrease the tlb-miss cacheline accesses from 19 to 15 in case both the linux hypervisor and the linux guest both uses this patch (though the guest will limit the addition speedup to anonymous regions only for now...). Even more important is that the tlb miss handler is much slower on a NPT/EPT guest than for a regular shadow paging or no-virtualization scenario. So maximizing the amount of virtual memory cached by the TLB pays off significantly more with NPT/EPT than without (even if there would be no significant speedup in the tlb-miss runtime). The first (and more tedious) part of this work requires allowing the VM to handle anonymous hugepages mixed with regular pages transparently on regular anonymous vmas. This is what this patch tries to achieve in the least intrusive possible way. We want hugepages and hugetlb to be used in a way so that all applications can benefit without changes (as usual we leverage the KVM virtualization design: by improving the Linux VM at large, KVM gets the performance boost too). The most important design choice is: always fallback to 4k allocation if the hugepage allocation fails! This is the _very_ opposite of some large pagecache patches that failed with -EIO back then if a 64k (or similar) allocation failed... Second important decision (to reduce the impact of the feature on the existing pagetable handling code) is that at any time we can split an hugepage into 512 regular pages and it has to be done with an operation that can't fail. This way the reliability of the swapping isn't decreased (no need to allocate memory when we are short on memory to swap) and it's trivial to plug a split_huge_page* one-liner where needed without polluting the VM. Over time we can teach mprotect, mremap and friends to handle pmd_trans_huge natively without calling split_huge_page*. The fact it can't fail isn't just for swap: if split_huge_page would return -ENOMEM (instead of the current void) we'd need to rollback the mprotect from the middle of it (ideally including undoing the split_vma) which would be a big change and in the very wrong direction (it'd likely be simpler not to call split_huge_page at all and to teach mprotect and friends to handle hugepages instead of rolling them back from the middle). In short the very value of split_huge_page is that it can't fail. The collapsing and madvise(MADV_HUGEPAGE) part will remain separated and incremental and it'll just be an "harmless" addition later if this initial part is agreed upon. It also should be noted that locking-wise replacing regular pages with hugepages is going to be very easy if compared to what I'm doing below in split_huge_page, as it will only happen when page_count(page) matches page_mapcount(page) if we can take the PG_lock and mmap_sem in write mode. collapse_huge_page will be a "best effort" that (unlike split_huge_page) can fail at the minimal sign of trouble and we can try again later. collapse_huge_page will be similar to how KSM works and the madvise(MADV_HUGEPAGE) will work similar to madvise(MADV_MERGEABLE). The default I like is that transparent hugepages are used at page fault time. This can be changed with /sys/kernel/mm/transparent_hugepage/enabled. The control knob can be set to three values "always", "madvise", "never" which mean respectively that hugepages are always used, or only inside madvise(MADV_HUGEPAGE) regions, or never used. /sys/kernel/mm/transparent_hugepage/defrag instead controls if the hugepage allocation should defrag memory aggressively "always", only inside "madvise" regions, or "never". The pmd_trans_splitting/pmd_trans_huge locking is very solid. The put_page (from get_user_page users that can't use mmu notifier like O_DIRECT) that runs against a __split_huge_page_refcount instead was a pain to serialize in a way that would result always in a coherent page count for both tail and head. I think my locking solution with a compound_lock taken only after the page_first is valid and is still a PageHead should be safe but it surely needs review from SMP race point of view. In short there is no current existing way to serialize the O_DIRECT final put_page against split_huge_page_refcount so I had to invent a new one (O_DIRECT loses knowledge on the mapping status by the time gup_fast returns so...). And I didn't want to impact all gup/gup_fast users for now, maybe if we change the gup interface substantially we can avoid this locking, I admit I didn't think too much about it because changing the gup unpinning interface would be invasive. If we ignored O_DIRECT we could stick to the existing compound refcounting code, by simply adding a get_user_pages_fast_flags(foll_flags) where KVM (and any other mmu notifier user) would call it without FOLL_GET (and if FOLL_GET isn't set we'd just BUG_ON if nobody registered itself in the current task mmu notifier list yet). But O_DIRECT is fundamental for decent performance of virtualized I/O on fast storage so we can't avoid it to solve the race of put_page against split_huge_page_refcount to achieve a complete hugepage feature for KVM. Swap and oom works fine (well just like with regular pages ;). MMU notifier is handled transparently too, with the exception of the young bit on the pmd, that didn't have a range check but I think KVM will be fine because the whole point of hugepages is that EPT/NPT will also use a huge pmd when they notice gup returns pages with PageCompound set, so they won't care of a range and there's just the pmd young bit to check in that case. NOTE: in some cases if the L2 cache is small, this may slowdown and waste memory during COWs because 4M of memory are accessed in a single fault instead of 8k (the payoff is that after COW the program can run faster). So we might want to switch the copy_huge_page (and clear_huge_page too) to not temporal stores. I also extensively researched ways to avoid this cache trashing with a full prefault logic that would cow in 8k/16k/32k/64k up to 1M (I can send those patches that fully implemented prefault) but I concluded they're not worth it and they add an huge additional complexity and they remove all tlb benefits until the full hugepage has been faulted in, to save a little bit of memory and some cache during app startup, but they still don't improve substantially the cache-trashing during startup if the prefault happens in >4k chunks. One reason is that those 4k pte entries copied are still mapped on a perfectly cache-colored hugepage, so the trashing is the worst one can generate in those copies (cow of 4k page copies aren't so well colored so they trashes less, but again this results in software running faster after the page fault). Those prefault patches allowed things like a pte where post-cow pages were local 4k regular anon pages and the not-yet-cowed pte entries were pointing in the middle of some hugepage mapped read-only. If it doesn't payoff substantially with todays hardware it will payoff even less in the future with larger l2 caches, and the prefault logic would blot the VM a lot. If one is emebdded transparent_hugepage can be disabled during boot with sysfs or with the boot commandline parameter transparent_hugepage=0 (or transparent_hugepage=2 to restrict hugepages inside madvise regions) that will ensure not a single hugepage is allocated at boot time. It is simple enough to just disable transparent hugepage globally and let transparent hugepages be allocated selectively by applications in the MADV_HUGEPAGE region (both at page fault time, and if enabled with the collapse_huge_page too through the kernel daemon). This patch supports only hugepages mapped in the pmd, archs that have smaller hugepages will not fit in this patch alone. Also some archs like power have certain tlb limits that prevents mixing different page size in the same regions so they will not fit in this framework that requires "graceful fallback" to basic PAGE_SIZE in case of physical memory fragmentation. hugetlbfs remains a perfect fit for those because its software limits happen to match the hardware limits. hugetlbfs also remains a perfect fit for hugepage sizes like 1GByte that cannot be hoped to be found not fragmented after a certain system uptime and that would be very expensive to defragment with relocation, so requiring reservation. hugetlbfs is the "reservation way", the point of transparent hugepages is not to have any reservation at all and maximizing the use of cache and hugepages at all times automatically. Some performance result: vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largep ages3 memset page fault 1566023 memset tlb miss 453854 memset second tlb miss 453321 random access tlb miss 41635 random access second tlb miss 41658 vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largepages3 memset page fault 1566471 memset tlb miss 453375 memset second tlb miss 453320 random access tlb miss 41636 random access second tlb miss 41637 vmx andrea # ./largepages3 memset page fault 1566642 memset tlb miss 453417 memset second tlb miss 453313 random access tlb miss 41630 random access second tlb miss 41647 vmx andrea # ./largepages3 memset page fault 1566872 memset tlb miss 453418 memset second tlb miss 453315 random access tlb miss 41618 random access second tlb miss 41659 vmx andrea # echo 0 > /proc/sys/vm/transparent_hugepage vmx andrea # ./largepages3 memset page fault 2182476 memset tlb miss 460305 memset second tlb miss 460179 random access tlb miss 44483 random access second tlb miss 44186 vmx andrea # ./largepages3 memset page fault 2182791 memset tlb miss 460742 memset second tlb miss 459962 random access tlb miss 43981 random access second tlb miss 43988 ============ #include <stdio.h> #include <stdlib.h> #include <string.h> #include <sys/time.h> #define SIZE (3UL*1024*1024*1024) int main() { char *p = malloc(SIZE), *p2; struct timeval before, after; gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset page fault %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); return 0; } ============ Signed-off-by: Andrea Arcangeli <aarcange@redhat.com> Acked-by: Rik van Riel <riel@redhat.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-13 23:46:52 +00:00
if (pmd_trans_huge(*pmd)) {
if (flags & FOLL_SPLIT) {
split_huge_page_pmd(vma, address, pmd);
goto split_fallthrough;
}
thp: transparent hugepage core Lately I've been working to make KVM use hugepages transparently without the usual restrictions of hugetlbfs. Some of the restrictions I'd like to see removed: 1) hugepages have to be swappable or the guest physical memory remains locked in RAM and can't be paged out to swap 2) if a hugepage allocation fails, regular pages should be allocated instead and mixed in the same vma without any failure and without userland noticing 3) if some task quits and more hugepages become available in the buddy, guest physical memory backed by regular pages should be relocated on hugepages automatically in regions under madvise(MADV_HUGEPAGE) (ideally event driven by waking up the kernel deamon if the order=HPAGE_PMD_SHIFT-PAGE_SHIFT list becomes not null) 4) avoidance of reservation and maximization of use of hugepages whenever possible. Reservation (needed to avoid runtime fatal faliures) may be ok for 1 machine with 1 database with 1 database cache with 1 database cache size known at boot time. It's definitely not feasible with a virtualization hypervisor usage like RHEV-H that runs an unknown number of virtual machines with an unknown size of each virtual machine with an unknown amount of pagecache that could be potentially useful in the host for guest not using O_DIRECT (aka cache=off). hugepages in the virtualization hypervisor (and also in the guest!) are much more important than in a regular host not using virtualization, becasue with NPT/EPT they decrease the tlb-miss cacheline accesses from 24 to 19 in case only the hypervisor uses transparent hugepages, and they decrease the tlb-miss cacheline accesses from 19 to 15 in case both the linux hypervisor and the linux guest both uses this patch (though the guest will limit the addition speedup to anonymous regions only for now...). Even more important is that the tlb miss handler is much slower on a NPT/EPT guest than for a regular shadow paging or no-virtualization scenario. So maximizing the amount of virtual memory cached by the TLB pays off significantly more with NPT/EPT than without (even if there would be no significant speedup in the tlb-miss runtime). The first (and more tedious) part of this work requires allowing the VM to handle anonymous hugepages mixed with regular pages transparently on regular anonymous vmas. This is what this patch tries to achieve in the least intrusive possible way. We want hugepages and hugetlb to be used in a way so that all applications can benefit without changes (as usual we leverage the KVM virtualization design: by improving the Linux VM at large, KVM gets the performance boost too). The most important design choice is: always fallback to 4k allocation if the hugepage allocation fails! This is the _very_ opposite of some large pagecache patches that failed with -EIO back then if a 64k (or similar) allocation failed... Second important decision (to reduce the impact of the feature on the existing pagetable handling code) is that at any time we can split an hugepage into 512 regular pages and it has to be done with an operation that can't fail. This way the reliability of the swapping isn't decreased (no need to allocate memory when we are short on memory to swap) and it's trivial to plug a split_huge_page* one-liner where needed without polluting the VM. Over time we can teach mprotect, mremap and friends to handle pmd_trans_huge natively without calling split_huge_page*. The fact it can't fail isn't just for swap: if split_huge_page would return -ENOMEM (instead of the current void) we'd need to rollback the mprotect from the middle of it (ideally including undoing the split_vma) which would be a big change and in the very wrong direction (it'd likely be simpler not to call split_huge_page at all and to teach mprotect and friends to handle hugepages instead of rolling them back from the middle). In short the very value of split_huge_page is that it can't fail. The collapsing and madvise(MADV_HUGEPAGE) part will remain separated and incremental and it'll just be an "harmless" addition later if this initial part is agreed upon. It also should be noted that locking-wise replacing regular pages with hugepages is going to be very easy if compared to what I'm doing below in split_huge_page, as it will only happen when page_count(page) matches page_mapcount(page) if we can take the PG_lock and mmap_sem in write mode. collapse_huge_page will be a "best effort" that (unlike split_huge_page) can fail at the minimal sign of trouble and we can try again later. collapse_huge_page will be similar to how KSM works and the madvise(MADV_HUGEPAGE) will work similar to madvise(MADV_MERGEABLE). The default I like is that transparent hugepages are used at page fault time. This can be changed with /sys/kernel/mm/transparent_hugepage/enabled. The control knob can be set to three values "always", "madvise", "never" which mean respectively that hugepages are always used, or only inside madvise(MADV_HUGEPAGE) regions, or never used. /sys/kernel/mm/transparent_hugepage/defrag instead controls if the hugepage allocation should defrag memory aggressively "always", only inside "madvise" regions, or "never". The pmd_trans_splitting/pmd_trans_huge locking is very solid. The put_page (from get_user_page users that can't use mmu notifier like O_DIRECT) that runs against a __split_huge_page_refcount instead was a pain to serialize in a way that would result always in a coherent page count for both tail and head. I think my locking solution with a compound_lock taken only after the page_first is valid and is still a PageHead should be safe but it surely needs review from SMP race point of view. In short there is no current existing way to serialize the O_DIRECT final put_page against split_huge_page_refcount so I had to invent a new one (O_DIRECT loses knowledge on the mapping status by the time gup_fast returns so...). And I didn't want to impact all gup/gup_fast users for now, maybe if we change the gup interface substantially we can avoid this locking, I admit I didn't think too much about it because changing the gup unpinning interface would be invasive. If we ignored O_DIRECT we could stick to the existing compound refcounting code, by simply adding a get_user_pages_fast_flags(foll_flags) where KVM (and any other mmu notifier user) would call it without FOLL_GET (and if FOLL_GET isn't set we'd just BUG_ON if nobody registered itself in the current task mmu notifier list yet). But O_DIRECT is fundamental for decent performance of virtualized I/O on fast storage so we can't avoid it to solve the race of put_page against split_huge_page_refcount to achieve a complete hugepage feature for KVM. Swap and oom works fine (well just like with regular pages ;). MMU notifier is handled transparently too, with the exception of the young bit on the pmd, that didn't have a range check but I think KVM will be fine because the whole point of hugepages is that EPT/NPT will also use a huge pmd when they notice gup returns pages with PageCompound set, so they won't care of a range and there's just the pmd young bit to check in that case. NOTE: in some cases if the L2 cache is small, this may slowdown and waste memory during COWs because 4M of memory are accessed in a single fault instead of 8k (the payoff is that after COW the program can run faster). So we might want to switch the copy_huge_page (and clear_huge_page too) to not temporal stores. I also extensively researched ways to avoid this cache trashing with a full prefault logic that would cow in 8k/16k/32k/64k up to 1M (I can send those patches that fully implemented prefault) but I concluded they're not worth it and they add an huge additional complexity and they remove all tlb benefits until the full hugepage has been faulted in, to save a little bit of memory and some cache during app startup, but they still don't improve substantially the cache-trashing during startup if the prefault happens in >4k chunks. One reason is that those 4k pte entries copied are still mapped on a perfectly cache-colored hugepage, so the trashing is the worst one can generate in those copies (cow of 4k page copies aren't so well colored so they trashes less, but again this results in software running faster after the page fault). Those prefault patches allowed things like a pte where post-cow pages were local 4k regular anon pages and the not-yet-cowed pte entries were pointing in the middle of some hugepage mapped read-only. If it doesn't payoff substantially with todays hardware it will payoff even less in the future with larger l2 caches, and the prefault logic would blot the VM a lot. If one is emebdded transparent_hugepage can be disabled during boot with sysfs or with the boot commandline parameter transparent_hugepage=0 (or transparent_hugepage=2 to restrict hugepages inside madvise regions) that will ensure not a single hugepage is allocated at boot time. It is simple enough to just disable transparent hugepage globally and let transparent hugepages be allocated selectively by applications in the MADV_HUGEPAGE region (both at page fault time, and if enabled with the collapse_huge_page too through the kernel daemon). This patch supports only hugepages mapped in the pmd, archs that have smaller hugepages will not fit in this patch alone. Also some archs like power have certain tlb limits that prevents mixing different page size in the same regions so they will not fit in this framework that requires "graceful fallback" to basic PAGE_SIZE in case of physical memory fragmentation. hugetlbfs remains a perfect fit for those because its software limits happen to match the hardware limits. hugetlbfs also remains a perfect fit for hugepage sizes like 1GByte that cannot be hoped to be found not fragmented after a certain system uptime and that would be very expensive to defragment with relocation, so requiring reservation. hugetlbfs is the "reservation way", the point of transparent hugepages is not to have any reservation at all and maximizing the use of cache and hugepages at all times automatically. Some performance result: vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largep ages3 memset page fault 1566023 memset tlb miss 453854 memset second tlb miss 453321 random access tlb miss 41635 random access second tlb miss 41658 vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largepages3 memset page fault 1566471 memset tlb miss 453375 memset second tlb miss 453320 random access tlb miss 41636 random access second tlb miss 41637 vmx andrea # ./largepages3 memset page fault 1566642 memset tlb miss 453417 memset second tlb miss 453313 random access tlb miss 41630 random access second tlb miss 41647 vmx andrea # ./largepages3 memset page fault 1566872 memset tlb miss 453418 memset second tlb miss 453315 random access tlb miss 41618 random access second tlb miss 41659 vmx andrea # echo 0 > /proc/sys/vm/transparent_hugepage vmx andrea # ./largepages3 memset page fault 2182476 memset tlb miss 460305 memset second tlb miss 460179 random access tlb miss 44483 random access second tlb miss 44186 vmx andrea # ./largepages3 memset page fault 2182791 memset tlb miss 460742 memset second tlb miss 459962 random access tlb miss 43981 random access second tlb miss 43988 ============ #include <stdio.h> #include <stdlib.h> #include <string.h> #include <sys/time.h> #define SIZE (3UL*1024*1024*1024) int main() { char *p = malloc(SIZE), *p2; struct timeval before, after; gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset page fault %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); return 0; } ============ Signed-off-by: Andrea Arcangeli <aarcange@redhat.com> Acked-by: Rik van Riel <riel@redhat.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-13 23:46:52 +00:00
spin_lock(&mm->page_table_lock);
if (likely(pmd_trans_huge(*pmd))) {
if (unlikely(pmd_trans_splitting(*pmd))) {
spin_unlock(&mm->page_table_lock);
wait_split_huge_page(vma->anon_vma, pmd);
} else {
mm, thp: fix mapped pages avoiding unevictable list on mlock When a transparent hugepage is mapped and it is included in an mlock() range, follow_page() incorrectly avoids setting the page's mlock bit and moving it to the unevictable lru. This is evident if you try to mlock(), munlock(), and then mlock() a range again. Currently: #define MAP_SIZE (4 << 30) /* 4GB */ void *ptr = mmap(NULL, MAP_SIZE, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0); mlock(ptr, MAP_SIZE); $ grep -E "Unevictable|Inactive\(anon" /proc/meminfo Inactive(anon): 6304 kB Unevictable: 4213924 kB munlock(ptr, MAP_SIZE); Inactive(anon): 4186252 kB Unevictable: 19652 kB mlock(ptr, MAP_SIZE); Inactive(anon): 4198556 kB Unevictable: 21684 kB Notice that less than 2MB was added to the unevictable list; this is because these pages in the range are not transparent hugepages since the 4GB range was allocated with mmap() and has no specific alignment. If posix_memalign() were used instead, unevictable would not have grown at all on the second mlock(). The fix is to call mlock_vma_page() so that the mlock bit is set and the page is added to the unevictable list. With this patch: mlock(ptr, MAP_SIZE); Inactive(anon): 4056 kB Unevictable: 4213940 kB munlock(ptr, MAP_SIZE); Inactive(anon): 4198268 kB Unevictable: 19636 kB mlock(ptr, MAP_SIZE); Inactive(anon): 4008 kB Unevictable: 4213940 kB Signed-off-by: David Rientjes <rientjes@google.com> Acked-by: Hugh Dickins <hughd@google.com> Reviewed-by: Andrea Arcangeli <aarcange@redhat.com> Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michel Lespinasse <walken@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-08 23:34:03 +00:00
page = follow_trans_huge_pmd(vma, address,
thp: transparent hugepage core Lately I've been working to make KVM use hugepages transparently without the usual restrictions of hugetlbfs. Some of the restrictions I'd like to see removed: 1) hugepages have to be swappable or the guest physical memory remains locked in RAM and can't be paged out to swap 2) if a hugepage allocation fails, regular pages should be allocated instead and mixed in the same vma without any failure and without userland noticing 3) if some task quits and more hugepages become available in the buddy, guest physical memory backed by regular pages should be relocated on hugepages automatically in regions under madvise(MADV_HUGEPAGE) (ideally event driven by waking up the kernel deamon if the order=HPAGE_PMD_SHIFT-PAGE_SHIFT list becomes not null) 4) avoidance of reservation and maximization of use of hugepages whenever possible. Reservation (needed to avoid runtime fatal faliures) may be ok for 1 machine with 1 database with 1 database cache with 1 database cache size known at boot time. It's definitely not feasible with a virtualization hypervisor usage like RHEV-H that runs an unknown number of virtual machines with an unknown size of each virtual machine with an unknown amount of pagecache that could be potentially useful in the host for guest not using O_DIRECT (aka cache=off). hugepages in the virtualization hypervisor (and also in the guest!) are much more important than in a regular host not using virtualization, becasue with NPT/EPT they decrease the tlb-miss cacheline accesses from 24 to 19 in case only the hypervisor uses transparent hugepages, and they decrease the tlb-miss cacheline accesses from 19 to 15 in case both the linux hypervisor and the linux guest both uses this patch (though the guest will limit the addition speedup to anonymous regions only for now...). Even more important is that the tlb miss handler is much slower on a NPT/EPT guest than for a regular shadow paging or no-virtualization scenario. So maximizing the amount of virtual memory cached by the TLB pays off significantly more with NPT/EPT than without (even if there would be no significant speedup in the tlb-miss runtime). The first (and more tedious) part of this work requires allowing the VM to handle anonymous hugepages mixed with regular pages transparently on regular anonymous vmas. This is what this patch tries to achieve in the least intrusive possible way. We want hugepages and hugetlb to be used in a way so that all applications can benefit without changes (as usual we leverage the KVM virtualization design: by improving the Linux VM at large, KVM gets the performance boost too). The most important design choice is: always fallback to 4k allocation if the hugepage allocation fails! This is the _very_ opposite of some large pagecache patches that failed with -EIO back then if a 64k (or similar) allocation failed... Second important decision (to reduce the impact of the feature on the existing pagetable handling code) is that at any time we can split an hugepage into 512 regular pages and it has to be done with an operation that can't fail. This way the reliability of the swapping isn't decreased (no need to allocate memory when we are short on memory to swap) and it's trivial to plug a split_huge_page* one-liner where needed without polluting the VM. Over time we can teach mprotect, mremap and friends to handle pmd_trans_huge natively without calling split_huge_page*. The fact it can't fail isn't just for swap: if split_huge_page would return -ENOMEM (instead of the current void) we'd need to rollback the mprotect from the middle of it (ideally including undoing the split_vma) which would be a big change and in the very wrong direction (it'd likely be simpler not to call split_huge_page at all and to teach mprotect and friends to handle hugepages instead of rolling them back from the middle). In short the very value of split_huge_page is that it can't fail. The collapsing and madvise(MADV_HUGEPAGE) part will remain separated and incremental and it'll just be an "harmless" addition later if this initial part is agreed upon. It also should be noted that locking-wise replacing regular pages with hugepages is going to be very easy if compared to what I'm doing below in split_huge_page, as it will only happen when page_count(page) matches page_mapcount(page) if we can take the PG_lock and mmap_sem in write mode. collapse_huge_page will be a "best effort" that (unlike split_huge_page) can fail at the minimal sign of trouble and we can try again later. collapse_huge_page will be similar to how KSM works and the madvise(MADV_HUGEPAGE) will work similar to madvise(MADV_MERGEABLE). The default I like is that transparent hugepages are used at page fault time. This can be changed with /sys/kernel/mm/transparent_hugepage/enabled. The control knob can be set to three values "always", "madvise", "never" which mean respectively that hugepages are always used, or only inside madvise(MADV_HUGEPAGE) regions, or never used. /sys/kernel/mm/transparent_hugepage/defrag instead controls if the hugepage allocation should defrag memory aggressively "always", only inside "madvise" regions, or "never". The pmd_trans_splitting/pmd_trans_huge locking is very solid. The put_page (from get_user_page users that can't use mmu notifier like O_DIRECT) that runs against a __split_huge_page_refcount instead was a pain to serialize in a way that would result always in a coherent page count for both tail and head. I think my locking solution with a compound_lock taken only after the page_first is valid and is still a PageHead should be safe but it surely needs review from SMP race point of view. In short there is no current existing way to serialize the O_DIRECT final put_page against split_huge_page_refcount so I had to invent a new one (O_DIRECT loses knowledge on the mapping status by the time gup_fast returns so...). And I didn't want to impact all gup/gup_fast users for now, maybe if we change the gup interface substantially we can avoid this locking, I admit I didn't think too much about it because changing the gup unpinning interface would be invasive. If we ignored O_DIRECT we could stick to the existing compound refcounting code, by simply adding a get_user_pages_fast_flags(foll_flags) where KVM (and any other mmu notifier user) would call it without FOLL_GET (and if FOLL_GET isn't set we'd just BUG_ON if nobody registered itself in the current task mmu notifier list yet). But O_DIRECT is fundamental for decent performance of virtualized I/O on fast storage so we can't avoid it to solve the race of put_page against split_huge_page_refcount to achieve a complete hugepage feature for KVM. Swap and oom works fine (well just like with regular pages ;). MMU notifier is handled transparently too, with the exception of the young bit on the pmd, that didn't have a range check but I think KVM will be fine because the whole point of hugepages is that EPT/NPT will also use a huge pmd when they notice gup returns pages with PageCompound set, so they won't care of a range and there's just the pmd young bit to check in that case. NOTE: in some cases if the L2 cache is small, this may slowdown and waste memory during COWs because 4M of memory are accessed in a single fault instead of 8k (the payoff is that after COW the program can run faster). So we might want to switch the copy_huge_page (and clear_huge_page too) to not temporal stores. I also extensively researched ways to avoid this cache trashing with a full prefault logic that would cow in 8k/16k/32k/64k up to 1M (I can send those patches that fully implemented prefault) but I concluded they're not worth it and they add an huge additional complexity and they remove all tlb benefits until the full hugepage has been faulted in, to save a little bit of memory and some cache during app startup, but they still don't improve substantially the cache-trashing during startup if the prefault happens in >4k chunks. One reason is that those 4k pte entries copied are still mapped on a perfectly cache-colored hugepage, so the trashing is the worst one can generate in those copies (cow of 4k page copies aren't so well colored so they trashes less, but again this results in software running faster after the page fault). Those prefault patches allowed things like a pte where post-cow pages were local 4k regular anon pages and the not-yet-cowed pte entries were pointing in the middle of some hugepage mapped read-only. If it doesn't payoff substantially with todays hardware it will payoff even less in the future with larger l2 caches, and the prefault logic would blot the VM a lot. If one is emebdded transparent_hugepage can be disabled during boot with sysfs or with the boot commandline parameter transparent_hugepage=0 (or transparent_hugepage=2 to restrict hugepages inside madvise regions) that will ensure not a single hugepage is allocated at boot time. It is simple enough to just disable transparent hugepage globally and let transparent hugepages be allocated selectively by applications in the MADV_HUGEPAGE region (both at page fault time, and if enabled with the collapse_huge_page too through the kernel daemon). This patch supports only hugepages mapped in the pmd, archs that have smaller hugepages will not fit in this patch alone. Also some archs like power have certain tlb limits that prevents mixing different page size in the same regions so they will not fit in this framework that requires "graceful fallback" to basic PAGE_SIZE in case of physical memory fragmentation. hugetlbfs remains a perfect fit for those because its software limits happen to match the hardware limits. hugetlbfs also remains a perfect fit for hugepage sizes like 1GByte that cannot be hoped to be found not fragmented after a certain system uptime and that would be very expensive to defragment with relocation, so requiring reservation. hugetlbfs is the "reservation way", the point of transparent hugepages is not to have any reservation at all and maximizing the use of cache and hugepages at all times automatically. Some performance result: vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largep ages3 memset page fault 1566023 memset tlb miss 453854 memset second tlb miss 453321 random access tlb miss 41635 random access second tlb miss 41658 vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largepages3 memset page fault 1566471 memset tlb miss 453375 memset second tlb miss 453320 random access tlb miss 41636 random access second tlb miss 41637 vmx andrea # ./largepages3 memset page fault 1566642 memset tlb miss 453417 memset second tlb miss 453313 random access tlb miss 41630 random access second tlb miss 41647 vmx andrea # ./largepages3 memset page fault 1566872 memset tlb miss 453418 memset second tlb miss 453315 random access tlb miss 41618 random access second tlb miss 41659 vmx andrea # echo 0 > /proc/sys/vm/transparent_hugepage vmx andrea # ./largepages3 memset page fault 2182476 memset tlb miss 460305 memset second tlb miss 460179 random access tlb miss 44483 random access second tlb miss 44186 vmx andrea # ./largepages3 memset page fault 2182791 memset tlb miss 460742 memset second tlb miss 459962 random access tlb miss 43981 random access second tlb miss 43988 ============ #include <stdio.h> #include <stdlib.h> #include <string.h> #include <sys/time.h> #define SIZE (3UL*1024*1024*1024) int main() { char *p = malloc(SIZE), *p2; struct timeval before, after; gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset page fault %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); return 0; } ============ Signed-off-by: Andrea Arcangeli <aarcange@redhat.com> Acked-by: Rik van Riel <riel@redhat.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-13 23:46:52 +00:00
pmd, flags);
spin_unlock(&mm->page_table_lock);
goto out;
}
} else
spin_unlock(&mm->page_table_lock);
/* fall through */
}
split_fallthrough:
x86: fix PAE pmd_bad bootup warning Fix warning from pmd_bad() at bootup on a HIGHMEM64G HIGHPTE x86_32. That came from 9fc34113f6880b215cbea4e7017fc818700384c2 x86: debug pmd_bad(); but we understand now that the typecasting was wrong for PAE in the previous version: pagetable pages above 4GB looked bad and stopped Arjan from booting. And revert that cded932b75ab0a5f9181ee3da34a0a488d1a14fd x86: fix pmd_bad and pud_bad to support huge pages. It was the wrong way round: we shouldn't weaken every pmd_bad and pud_bad check to let huge pages slip through - in part they check that we _don't_ have a huge page where it's not expected. Put the x86 pmd_bad() and pud_bad() definitions back to what they have long been: they can be improved (x86_32 should use PTE_MASK, to stop PAE thinking junk in the upper word is good; and x86_64 should follow x86_32's stricter comparison, to stop thinking any subset of required bits is good); but that should be a later patch. Fix Hans' good observation that follow_page() will never find pmd_huge() because that would have already failed the pmd_bad test: test pmd_huge in between the pmd_none and pmd_bad tests. Tighten x86's pmd_huge() check? No, once it's a hugepage entry, it can get quite far from a good pmd: for example, PROT_NONE leaves it with only ACCESSED of the KERN_PGTABLE bits. However... though follow_page() contains this and another test for huge pages, so it's nice to keep it working on them, where does it actually get called on a huge page? get_user_pages() checks is_vm_hugetlb_page(vma) to to call alternative hugetlb processing, as does unmap_vmas() and others. Signed-off-by: Hugh Dickins <hugh@veritas.com> Earlier-version-tested-by: Ingo Molnar <mingo@elte.hu> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Jeff Chua <jeff.chua.linux@gmail.com> Cc: Hans Rosenfeld <hans.rosenfeld@amd.com> Cc: Arjan van de Ven <arjan@linux.intel.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-05-06 19:49:23 +00:00
if (unlikely(pmd_bad(*pmd)))
goto no_page_table;
ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
pte = *ptep;
if (!pte_present(pte))
Reinstate ZERO_PAGE optimization in 'get_user_pages()' and fix XIP KAMEZAWA Hiroyuki and Oleg Nesterov point out that since the commit 557ed1fa2620dc119adb86b34c614e152a629a80 ("remove ZERO_PAGE") removed the ZERO_PAGE from the VM mappings, any users of get_user_pages() will generally now populate the VM with real empty pages needlessly. We used to get the ZERO_PAGE when we did the "handle_mm_fault()", but since fault handling no longer uses ZERO_PAGE for new anonymous pages, we now need to handle that special case in follow_page() instead. In particular, the removal of ZERO_PAGE effectively removed the core file writing optimization where we would skip writing pages that had not been populated at all, and increased memory pressure a lot by allocating all those useless newly zeroed pages. This reinstates the optimization by making the unmapped PTE case the same as for a non-existent page table, which already did this correctly. While at it, this also fixes the XIP case for follow_page(), where the caller could not differentiate between the case of a page that simply could not be used (because it had no "struct page" associated with it) and a page that just wasn't mapped. We do that by simply returning an error pointer for pages that could not be turned into a "struct page *". The error is arbitrarily picked to be EFAULT, since that was what get_user_pages() already used for the equivalent IO-mapped page case. [ Also removed an impossible test for pte_offset_map_lock() failing: that's not how that function works ] Acked-by: Oleg Nesterov <oleg@tv-sign.ru> Acked-by: Nick Piggin <npiggin@suse.de> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Hugh Dickins <hugh@veritas.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland McGrath <roland@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-06-20 18:18:25 +00:00
goto no_page;
if ((flags & FOLL_NUMA) && pte_numa(pte))
goto no_page;
if ((flags & FOLL_WRITE) && !pte_write(pte))
goto unlock;
mm: reinstate ZERO_PAGE KAMEZAWA Hiroyuki has observed customers of earlier kernels taking advantage of the ZERO_PAGE: which we stopped do_anonymous_page() from using in 2.6.24. And there were a couple of regression reports on LKML. Following suggestions from Linus, reinstate do_anonymous_page() use of the ZERO_PAGE; but this time avoid dirtying its struct page cacheline with (map)count updates - let vm_normal_page() regard it as abnormal. Use it only on arches which __HAVE_ARCH_PTE_SPECIAL (x86, s390, sh32, most powerpc): that's not essential, but minimizes additional branches (keeping them in the unlikely pte_special case); and incidentally excludes mips (some models of which needed eight colours of ZERO_PAGE to avoid costly exceptions). Don't be fanatical about avoiding ZERO_PAGE updates: get_user_pages() callers won't want to make exceptions for it, so increment its count there. Changes to mlock and migration? happily seems not needed. In most places it's quicker to check pfn than struct page address: prepare a __read_mostly zero_pfn for that. Does get_dump_page() still need its ZERO_PAGE check? probably not, but keep it anyway. Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk> Acked-by: Rik van Riel <riel@redhat.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Minchan Kim <minchan.kim@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-09-22 00:03:30 +00:00
page = vm_normal_page(vma, address, pte);
mm: reinstate ZERO_PAGE KAMEZAWA Hiroyuki has observed customers of earlier kernels taking advantage of the ZERO_PAGE: which we stopped do_anonymous_page() from using in 2.6.24. And there were a couple of regression reports on LKML. Following suggestions from Linus, reinstate do_anonymous_page() use of the ZERO_PAGE; but this time avoid dirtying its struct page cacheline with (map)count updates - let vm_normal_page() regard it as abnormal. Use it only on arches which __HAVE_ARCH_PTE_SPECIAL (x86, s390, sh32, most powerpc): that's not essential, but minimizes additional branches (keeping them in the unlikely pte_special case); and incidentally excludes mips (some models of which needed eight colours of ZERO_PAGE to avoid costly exceptions). Don't be fanatical about avoiding ZERO_PAGE updates: get_user_pages() callers won't want to make exceptions for it, so increment its count there. Changes to mlock and migration? happily seems not needed. In most places it's quicker to check pfn than struct page address: prepare a __read_mostly zero_pfn for that. Does get_dump_page() still need its ZERO_PAGE check? probably not, but keep it anyway. Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk> Acked-by: Rik van Riel <riel@redhat.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Minchan Kim <minchan.kim@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-09-22 00:03:30 +00:00
if (unlikely(!page)) {
if ((flags & FOLL_DUMP) ||
!is_zero_pfn(pte_pfn(pte)))
mm: reinstate ZERO_PAGE KAMEZAWA Hiroyuki has observed customers of earlier kernels taking advantage of the ZERO_PAGE: which we stopped do_anonymous_page() from using in 2.6.24. And there were a couple of regression reports on LKML. Following suggestions from Linus, reinstate do_anonymous_page() use of the ZERO_PAGE; but this time avoid dirtying its struct page cacheline with (map)count updates - let vm_normal_page() regard it as abnormal. Use it only on arches which __HAVE_ARCH_PTE_SPECIAL (x86, s390, sh32, most powerpc): that's not essential, but minimizes additional branches (keeping them in the unlikely pte_special case); and incidentally excludes mips (some models of which needed eight colours of ZERO_PAGE to avoid costly exceptions). Don't be fanatical about avoiding ZERO_PAGE updates: get_user_pages() callers won't want to make exceptions for it, so increment its count there. Changes to mlock and migration? happily seems not needed. In most places it's quicker to check pfn than struct page address: prepare a __read_mostly zero_pfn for that. Does get_dump_page() still need its ZERO_PAGE check? probably not, but keep it anyway. Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk> Acked-by: Rik van Riel <riel@redhat.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Minchan Kim <minchan.kim@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-09-22 00:03:30 +00:00
goto bad_page;
page = pte_page(pte);
}
if (flags & FOLL_GET)
mm: thp: tail page refcounting fix Michel while working on the working set estimation code, noticed that calling get_page_unless_zero() on a random pfn_to_page(random_pfn) wasn't safe, if the pfn ended up being a tail page of a transparent hugepage under splitting by __split_huge_page_refcount(). He then found the problem could also theoretically materialize with page_cache_get_speculative() during the speculative radix tree lookups that uses get_page_unless_zero() in SMP if the radix tree page is freed and reallocated and get_user_pages is called on it before page_cache_get_speculative has a chance to call get_page_unless_zero(). So the best way to fix the problem is to keep page_tail->_count zero at all times. This will guarantee that get_page_unless_zero() can never succeed on any tail page. page_tail->_mapcount is guaranteed zero and is unused for all tail pages of a compound page, so we can simply account the tail page references there and transfer them to tail_page->_count in __split_huge_page_refcount() (in addition to the head_page->_mapcount). While debugging this s/_count/_mapcount/ change I also noticed get_page is called by direct-io.c on pages returned by get_user_pages. That wasn't entirely safe because the two atomic_inc in get_page weren't atomic. As opposed to other get_user_page users like secondary-MMU page fault to establish the shadow pagetables would never call any superflous get_page after get_user_page returns. It's safer to make get_page universally safe for tail pages and to use get_page_foll() within follow_page (inside get_user_pages()). get_page_foll() is safe to do the refcounting for tail pages without taking any locks because it is run within PT lock protected critical sections (PT lock for pte and page_table_lock for pmd_trans_huge). The standard get_page() as invoked by direct-io instead will now take the compound_lock but still only for tail pages. The direct-io paths are usually I/O bound and the compound_lock is per THP so very finegrined, so there's no risk of scalability issues with it. A simple direct-io benchmarks with all lockdep prove locking and spinlock debugging infrastructure enabled shows identical performance and no overhead. So it's worth it. Ideally direct-io should stop calling get_page() on pages returned by get_user_pages(). The spinlock in get_page() is already optimized away for no-THP builds but doing get_page() on tail pages returned by GUP is generally a rare operation and usually only run in I/O paths. This new refcounting on page_tail->_mapcount in addition to avoiding new RCU critical sections will also allow the working set estimation code to work without any further complexity associated to the tail page refcounting with THP. Signed-off-by: Andrea Arcangeli <aarcange@redhat.com> Reported-by: Michel Lespinasse <walken@google.com> Reviewed-by: Michel Lespinasse <walken@google.com> Reviewed-by: Minchan Kim <minchan.kim@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Hugh Dickins <hughd@google.com> Cc: Johannes Weiner <jweiner@redhat.com> Cc: Rik van Riel <riel@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Gibson <david@gibson.dropbear.id.au> Cc: <stable@kernel.org> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-11-02 20:36:59 +00:00
get_page_foll(page);
if (flags & FOLL_TOUCH) {
if ((flags & FOLL_WRITE) &&
!pte_dirty(pte) && !PageDirty(page))
set_page_dirty(page);
/*
* pte_mkyoung() would be more correct here, but atomic care
* is needed to avoid losing the dirty bit: it is easier to use
* mark_page_accessed().
*/
mark_page_accessed(page);
}
if ((flags & FOLL_MLOCK) && (vma->vm_flags & VM_LOCKED)) {
/*
* The preliminary mapping check is mainly to avoid the
* pointless overhead of lock_page on the ZERO_PAGE
* which might bounce very badly if there is contention.
*
* If the page is already locked, we don't need to
* handle it now - vmscan will handle it later if and
* when it attempts to reclaim the page.
*/
if (page->mapping && trylock_page(page)) {
lru_add_drain(); /* push cached pages to LRU */
/*
mm: use clear_page_mlock() in page_remove_rmap() We had thought that pages could no longer get freed while still marked as mlocked; but Johannes Weiner posted this program to demonstrate that truncating an mlocked private file mapping containing COWed pages is still mishandled: #include <sys/types.h> #include <sys/mman.h> #include <sys/stat.h> #include <stdlib.h> #include <unistd.h> #include <fcntl.h> #include <stdio.h> int main(void) { char *map; int fd; system("grep mlockfreed /proc/vmstat"); fd = open("chigurh", O_CREAT|O_EXCL|O_RDWR); unlink("chigurh"); ftruncate(fd, 4096); map = mmap(NULL, 4096, PROT_WRITE, MAP_PRIVATE, fd, 0); map[0] = 11; mlock(map, sizeof(fd)); ftruncate(fd, 0); close(fd); munlock(map, sizeof(fd)); munmap(map, 4096); system("grep mlockfreed /proc/vmstat"); return 0; } The anon COWed pages are not caught by truncation's clear_page_mlock() of the pagecache pages; but unmap_mapping_range() unmaps them, so we ought to look out for them there in page_remove_rmap(). Indeed, why should truncation or invalidation be doing the clear_page_mlock() when removing from pagecache? mlock is a property of mapping in userspace, not a property of pagecache: an mlocked unmapped page is nonsensical. Reported-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Hugh Dickins <hughd@google.com> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Rik van Riel <riel@redhat.com> Cc: Michel Lespinasse <walken@google.com> Cc: Ying Han <yinghan@google.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-08 23:33:19 +00:00
* Because we lock page here, and migration is
* blocked by the pte's page reference, and we
* know the page is still mapped, we don't even
* need to check for file-cache page truncation.
*/
mm: use clear_page_mlock() in page_remove_rmap() We had thought that pages could no longer get freed while still marked as mlocked; but Johannes Weiner posted this program to demonstrate that truncating an mlocked private file mapping containing COWed pages is still mishandled: #include <sys/types.h> #include <sys/mman.h> #include <sys/stat.h> #include <stdlib.h> #include <unistd.h> #include <fcntl.h> #include <stdio.h> int main(void) { char *map; int fd; system("grep mlockfreed /proc/vmstat"); fd = open("chigurh", O_CREAT|O_EXCL|O_RDWR); unlink("chigurh"); ftruncate(fd, 4096); map = mmap(NULL, 4096, PROT_WRITE, MAP_PRIVATE, fd, 0); map[0] = 11; mlock(map, sizeof(fd)); ftruncate(fd, 0); close(fd); munlock(map, sizeof(fd)); munmap(map, 4096); system("grep mlockfreed /proc/vmstat"); return 0; } The anon COWed pages are not caught by truncation's clear_page_mlock() of the pagecache pages; but unmap_mapping_range() unmaps them, so we ought to look out for them there in page_remove_rmap(). Indeed, why should truncation or invalidation be doing the clear_page_mlock() when removing from pagecache? mlock is a property of mapping in userspace, not a property of pagecache: an mlocked unmapped page is nonsensical. Reported-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Hugh Dickins <hughd@google.com> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Rik van Riel <riel@redhat.com> Cc: Michel Lespinasse <walken@google.com> Cc: Ying Han <yinghan@google.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-08 23:33:19 +00:00
mlock_vma_page(page);
unlock_page(page);
}
}
unlock:
pte_unmap_unlock(ptep, ptl);
out:
return page;
Reinstate ZERO_PAGE optimization in 'get_user_pages()' and fix XIP KAMEZAWA Hiroyuki and Oleg Nesterov point out that since the commit 557ed1fa2620dc119adb86b34c614e152a629a80 ("remove ZERO_PAGE") removed the ZERO_PAGE from the VM mappings, any users of get_user_pages() will generally now populate the VM with real empty pages needlessly. We used to get the ZERO_PAGE when we did the "handle_mm_fault()", but since fault handling no longer uses ZERO_PAGE for new anonymous pages, we now need to handle that special case in follow_page() instead. In particular, the removal of ZERO_PAGE effectively removed the core file writing optimization where we would skip writing pages that had not been populated at all, and increased memory pressure a lot by allocating all those useless newly zeroed pages. This reinstates the optimization by making the unmapped PTE case the same as for a non-existent page table, which already did this correctly. While at it, this also fixes the XIP case for follow_page(), where the caller could not differentiate between the case of a page that simply could not be used (because it had no "struct page" associated with it) and a page that just wasn't mapped. We do that by simply returning an error pointer for pages that could not be turned into a "struct page *". The error is arbitrarily picked to be EFAULT, since that was what get_user_pages() already used for the equivalent IO-mapped page case. [ Also removed an impossible test for pte_offset_map_lock() failing: that's not how that function works ] Acked-by: Oleg Nesterov <oleg@tv-sign.ru> Acked-by: Nick Piggin <npiggin@suse.de> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Hugh Dickins <hugh@veritas.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland McGrath <roland@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-06-20 18:18:25 +00:00
bad_page:
pte_unmap_unlock(ptep, ptl);
return ERR_PTR(-EFAULT);
no_page:
pte_unmap_unlock(ptep, ptl);
if (!pte_none(pte))
return page;
no_page_table:
/*
* When core dumping an enormous anonymous area that nobody
* has touched so far, we don't want to allocate unnecessary pages or
* page tables. Return error instead of NULL to skip handle_mm_fault,
* then get_dump_page() will return NULL to leave a hole in the dump.
* But we can only make this optimization where a hole would surely
* be zero-filled if handle_mm_fault() actually did handle it.
*/
if ((flags & FOLL_DUMP) &&
(!vma->vm_ops || !vma->vm_ops->fault))
return ERR_PTR(-EFAULT);
return page;
}
static inline int stack_guard_page(struct vm_area_struct *vma, unsigned long addr)
{
return stack_guard_page_start(vma, addr) ||
stack_guard_page_end(vma, addr+PAGE_SIZE);
}
/**
* __get_user_pages() - pin user pages in memory
* @tsk: task_struct of target task
* @mm: mm_struct of target mm
* @start: starting user address
* @nr_pages: number of pages from start to pin
* @gup_flags: flags modifying pin behaviour
* @pages: array that receives pointers to the pages pinned.
* Should be at least nr_pages long. Or NULL, if caller
* only intends to ensure the pages are faulted in.
* @vmas: array of pointers to vmas corresponding to each page.
* Or NULL if the caller does not require them.
* @nonblocking: whether waiting for disk IO or mmap_sem contention
*
* Returns number of pages pinned. This may be fewer than the number
* requested. If nr_pages is 0 or negative, returns 0. If no pages
* were pinned, returns -errno. Each page returned must be released
* with a put_page() call when it is finished with. vmas will only
* remain valid while mmap_sem is held.
*
* Must be called with mmap_sem held for read or write.
*
* __get_user_pages walks a process's page tables and takes a reference to
* each struct page that each user address corresponds to at a given
* instant. That is, it takes the page that would be accessed if a user
* thread accesses the given user virtual address at that instant.
*
* This does not guarantee that the page exists in the user mappings when
* __get_user_pages returns, and there may even be a completely different
* page there in some cases (eg. if mmapped pagecache has been invalidated
* and subsequently re faulted). However it does guarantee that the page
* won't be freed completely. And mostly callers simply care that the page
* contains data that was valid *at some point in time*. Typically, an IO
* or similar operation cannot guarantee anything stronger anyway because
* locks can't be held over the syscall boundary.
*
* If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If
* the page is written to, set_page_dirty (or set_page_dirty_lock, as
* appropriate) must be called after the page is finished with, and
* before put_page is called.
*
* If @nonblocking != NULL, __get_user_pages will not wait for disk IO
* or mmap_sem contention, and if waiting is needed to pin all pages,
* *@nonblocking will be set to 0.
*
* In most cases, get_user_pages or get_user_pages_fast should be used
* instead of __get_user_pages. __get_user_pages should be used only if
* you need some special @gup_flags.
*/
mlock: mlocked pages are unevictable Make sure that mlocked pages also live on the unevictable LRU, so kswapd will not scan them over and over again. This is achieved through various strategies: 1) add yet another page flag--PG_mlocked--to indicate that the page is locked for efficient testing in vmscan and, optionally, fault path. This allows early culling of unevictable pages, preventing them from getting to page_referenced()/try_to_unmap(). Also allows separate accounting of mlock'd pages, as Nick's original patch did. Note: Nick's original mlock patch used a PG_mlocked flag. I had removed this in favor of the PG_unevictable flag + an mlock_count [new page struct member]. I restored the PG_mlocked flag to eliminate the new count field. 2) add the mlock/unevictable infrastructure to mm/mlock.c, with internal APIs in mm/internal.h. This is a rework of Nick's original patch to these files, taking into account that mlocked pages are now kept on unevictable LRU list. 3) update vmscan.c:page_evictable() to check PageMlocked() and, if vma passed in, the vm_flags. Note that the vma will only be passed in for new pages in the fault path; and then only if the "cull unevictable pages in fault path" patch is included. 4) add try_to_unlock() to rmap.c to walk a page's rmap and ClearPageMlocked() if no other vmas have it mlocked. Reuses as much of try_to_unmap() as possible. This effectively replaces the use of one of the lru list links as an mlock count. If this mechanism let's pages in mlocked vmas leak through w/o PG_mlocked set [I don't know that it does], we should catch them later in try_to_unmap(). One hopes this will be rare, as it will be relatively expensive. Original mm/internal.h, mm/rmap.c and mm/mlock.c changes: Signed-off-by: Nick Piggin <npiggin@suse.de> splitlru: introduce __get_user_pages(): New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS. because current get_user_pages() can't grab PROT_NONE pages theresore it cause PROT_NONE pages can't munlock. [akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch] [akpm@linux-foundation.org: untangle patch interdependencies] [akpm@linux-foundation.org: fix things after out-of-order merging] [hugh@veritas.com: fix page-flags mess] [lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm'] [kosaki.motohiro@jp.fujitsu.com: build fix] [kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments] [kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()] Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Signed-off-by: Rik van Riel <riel@redhat.com> Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Dave Hansen <dave@linux.vnet.ibm.com> Cc: Matt Mackall <mpm@selenic.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 03:26:44 +00:00
int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
unsigned long start, int nr_pages, unsigned int gup_flags,
struct page **pages, struct vm_area_struct **vmas,
int *nonblocking)
{
int i;
unsigned long vm_flags;
if (nr_pages <= 0)
return 0;
VM_BUG_ON(!!pages != !!(gup_flags & FOLL_GET));
/*
* Require read or write permissions.
* If FOLL_FORCE is set, we only require the "MAY" flags.
*/
vm_flags = (gup_flags & FOLL_WRITE) ?
(VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
vm_flags &= (gup_flags & FOLL_FORCE) ?
(VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
/*
* If FOLL_FORCE and FOLL_NUMA are both set, handle_mm_fault
* would be called on PROT_NONE ranges. We must never invoke
* handle_mm_fault on PROT_NONE ranges or the NUMA hinting
* page faults would unprotect the PROT_NONE ranges if
* _PAGE_NUMA and _PAGE_PROTNONE are sharing the same pte/pmd
* bitflag. So to avoid that, don't set FOLL_NUMA if
* FOLL_FORCE is set.
*/
if (!(gup_flags & FOLL_FORCE))
gup_flags |= FOLL_NUMA;
i = 0;
do {
struct vm_area_struct *vma;
vma = find_extend_vma(mm, start);
if (!vma && in_gate_area(mm, start)) {
unsigned long pg = start & PAGE_MASK;
pgd_t *pgd;
pud_t *pud;
pmd_t *pmd;
pte_t *pte;
mlock: mlocked pages are unevictable Make sure that mlocked pages also live on the unevictable LRU, so kswapd will not scan them over and over again. This is achieved through various strategies: 1) add yet another page flag--PG_mlocked--to indicate that the page is locked for efficient testing in vmscan and, optionally, fault path. This allows early culling of unevictable pages, preventing them from getting to page_referenced()/try_to_unmap(). Also allows separate accounting of mlock'd pages, as Nick's original patch did. Note: Nick's original mlock patch used a PG_mlocked flag. I had removed this in favor of the PG_unevictable flag + an mlock_count [new page struct member]. I restored the PG_mlocked flag to eliminate the new count field. 2) add the mlock/unevictable infrastructure to mm/mlock.c, with internal APIs in mm/internal.h. This is a rework of Nick's original patch to these files, taking into account that mlocked pages are now kept on unevictable LRU list. 3) update vmscan.c:page_evictable() to check PageMlocked() and, if vma passed in, the vm_flags. Note that the vma will only be passed in for new pages in the fault path; and then only if the "cull unevictable pages in fault path" patch is included. 4) add try_to_unlock() to rmap.c to walk a page's rmap and ClearPageMlocked() if no other vmas have it mlocked. Reuses as much of try_to_unmap() as possible. This effectively replaces the use of one of the lru list links as an mlock count. If this mechanism let's pages in mlocked vmas leak through w/o PG_mlocked set [I don't know that it does], we should catch them later in try_to_unmap(). One hopes this will be rare, as it will be relatively expensive. Original mm/internal.h, mm/rmap.c and mm/mlock.c changes: Signed-off-by: Nick Piggin <npiggin@suse.de> splitlru: introduce __get_user_pages(): New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS. because current get_user_pages() can't grab PROT_NONE pages theresore it cause PROT_NONE pages can't munlock. [akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch] [akpm@linux-foundation.org: untangle patch interdependencies] [akpm@linux-foundation.org: fix things after out-of-order merging] [hugh@veritas.com: fix page-flags mess] [lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm'] [kosaki.motohiro@jp.fujitsu.com: build fix] [kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments] [kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()] Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Signed-off-by: Rik van Riel <riel@redhat.com> Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Dave Hansen <dave@linux.vnet.ibm.com> Cc: Matt Mackall <mpm@selenic.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 03:26:44 +00:00
/* user gate pages are read-only */
if (gup_flags & FOLL_WRITE)
return i ? : -EFAULT;
if (pg > TASK_SIZE)
pgd = pgd_offset_k(pg);
else
pgd = pgd_offset_gate(mm, pg);
BUG_ON(pgd_none(*pgd));
pud = pud_offset(pgd, pg);
BUG_ON(pud_none(*pud));
pmd = pmd_offset(pud, pg);
if (pmd_none(*pmd))
return i ? : -EFAULT;
VM_BUG_ON(pmd_trans_huge(*pmd));
pte = pte_offset_map(pmd, pg);
if (pte_none(*pte)) {
pte_unmap(pte);
return i ? : -EFAULT;
}
vma = get_gate_vma(mm);
if (pages) {
struct page *page;
page = vm_normal_page(vma, start, *pte);
if (!page) {
if (!(gup_flags & FOLL_DUMP) &&
is_zero_pfn(pte_pfn(*pte)))
page = pte_page(*pte);
else {
pte_unmap(pte);
return i ? : -EFAULT;
}
}
pages[i] = page;
get_page(page);
}
pte_unmap(pte);
goto next_page;
}
mlock: mlocked pages are unevictable Make sure that mlocked pages also live on the unevictable LRU, so kswapd will not scan them over and over again. This is achieved through various strategies: 1) add yet another page flag--PG_mlocked--to indicate that the page is locked for efficient testing in vmscan and, optionally, fault path. This allows early culling of unevictable pages, preventing them from getting to page_referenced()/try_to_unmap(). Also allows separate accounting of mlock'd pages, as Nick's original patch did. Note: Nick's original mlock patch used a PG_mlocked flag. I had removed this in favor of the PG_unevictable flag + an mlock_count [new page struct member]. I restored the PG_mlocked flag to eliminate the new count field. 2) add the mlock/unevictable infrastructure to mm/mlock.c, with internal APIs in mm/internal.h. This is a rework of Nick's original patch to these files, taking into account that mlocked pages are now kept on unevictable LRU list. 3) update vmscan.c:page_evictable() to check PageMlocked() and, if vma passed in, the vm_flags. Note that the vma will only be passed in for new pages in the fault path; and then only if the "cull unevictable pages in fault path" patch is included. 4) add try_to_unlock() to rmap.c to walk a page's rmap and ClearPageMlocked() if no other vmas have it mlocked. Reuses as much of try_to_unmap() as possible. This effectively replaces the use of one of the lru list links as an mlock count. If this mechanism let's pages in mlocked vmas leak through w/o PG_mlocked set [I don't know that it does], we should catch them later in try_to_unmap(). One hopes this will be rare, as it will be relatively expensive. Original mm/internal.h, mm/rmap.c and mm/mlock.c changes: Signed-off-by: Nick Piggin <npiggin@suse.de> splitlru: introduce __get_user_pages(): New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS. because current get_user_pages() can't grab PROT_NONE pages theresore it cause PROT_NONE pages can't munlock. [akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch] [akpm@linux-foundation.org: untangle patch interdependencies] [akpm@linux-foundation.org: fix things after out-of-order merging] [hugh@veritas.com: fix page-flags mess] [lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm'] [kosaki.motohiro@jp.fujitsu.com: build fix] [kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments] [kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()] Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Signed-off-by: Rik van Riel <riel@redhat.com> Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Dave Hansen <dave@linux.vnet.ibm.com> Cc: Matt Mackall <mpm@selenic.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 03:26:44 +00:00
if (!vma ||
(vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
!(vm_flags & vma->vm_flags))
return i ? : -EFAULT;
if (is_vm_hugetlb_page(vma)) {
i = follow_hugetlb_page(mm, vma, pages, vmas,
&start, &nr_pages, i, gup_flags);
continue;
}
do {
struct page *page;
unsigned int foll_flags = gup_flags;
/*
mm: make get_user_pages() interruptible The initial implementation of checking TIF_MEMDIE covers the cases of OOM killing. If the process has been OOM killed, the TIF_MEMDIE is set and it return immediately. This patch includes: 1. add the case that the SIGKILL is sent by user processes. The process can try to get_user_pages() unlimited memory even if a user process has sent a SIGKILL to it(maybe a monitor find the process exceed its memory limit and try to kill it). In the old implementation, the SIGKILL won't be handled until the get_user_pages() returns. 2. change the return value to be ERESTARTSYS. It makes no sense to return ENOMEM if the get_user_pages returned by getting a SIGKILL signal. Considering the general convention for a system call interrupted by a signal is ERESTARTNOSYS, so the current return value is consistant to that. Lee: An unfortunate side effect of "make-get_user_pages-interruptible" is that it prevents a SIGKILL'd task from munlock-ing pages that it had mlocked, resulting in freeing of mlocked pages. Freeing of mlocked pages, in itself, is not so bad. We just count them now--altho' I had hoped to remove this stat and add PG_MLOCKED to the free pages flags check. However, consider pages in shared libraries mapped by more than one task that a task mlocked--e.g., via mlockall(). If the task that mlocked the pages exits via SIGKILL, these pages would be left mlocked and unevictable. Proposed fix: Add another GUP flag to ignore sigkill when calling get_user_pages from munlock()--similar to Kosaki Motohiro's 'IGNORE_VMA_PERMISSIONS flag for the same purpose. We are not actually allocating memory in this case, which "make-get_user_pages-interruptible" intends to avoid. We're just munlocking pages that are already resident and mapped, and we're reusing get_user_pages() to access those pages. ?? Maybe we should combine 'IGNORE_VMA_PERMISSIONS and '_IGNORE_SIGKILL into a single flag: GUP_FLAGS_MUNLOCK ??? [Lee.Schermerhorn@hp.com: ignore sigkill in get_user_pages during munlock] Signed-off-by: Paul Menage <menage@google.com> Signed-off-by: Ying Han <yinghan@google.com> Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Reviewed-by: Pekka Enberg <penberg@cs.helsinki.fi> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Hugh Dickins <hugh@veritas.com> Cc: Oleg Nesterov <oleg@tv-sign.ru> Cc: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Rohit Seth <rohitseth@google.com> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:40:18 +00:00
* If we have a pending SIGKILL, don't keep faulting
* pages and potentially allocating memory.
*/
if (unlikely(fatal_signal_pending(current)))
mm: make get_user_pages() interruptible The initial implementation of checking TIF_MEMDIE covers the cases of OOM killing. If the process has been OOM killed, the TIF_MEMDIE is set and it return immediately. This patch includes: 1. add the case that the SIGKILL is sent by user processes. The process can try to get_user_pages() unlimited memory even if a user process has sent a SIGKILL to it(maybe a monitor find the process exceed its memory limit and try to kill it). In the old implementation, the SIGKILL won't be handled until the get_user_pages() returns. 2. change the return value to be ERESTARTSYS. It makes no sense to return ENOMEM if the get_user_pages returned by getting a SIGKILL signal. Considering the general convention for a system call interrupted by a signal is ERESTARTNOSYS, so the current return value is consistant to that. Lee: An unfortunate side effect of "make-get_user_pages-interruptible" is that it prevents a SIGKILL'd task from munlock-ing pages that it had mlocked, resulting in freeing of mlocked pages. Freeing of mlocked pages, in itself, is not so bad. We just count them now--altho' I had hoped to remove this stat and add PG_MLOCKED to the free pages flags check. However, consider pages in shared libraries mapped by more than one task that a task mlocked--e.g., via mlockall(). If the task that mlocked the pages exits via SIGKILL, these pages would be left mlocked and unevictable. Proposed fix: Add another GUP flag to ignore sigkill when calling get_user_pages from munlock()--similar to Kosaki Motohiro's 'IGNORE_VMA_PERMISSIONS flag for the same purpose. We are not actually allocating memory in this case, which "make-get_user_pages-interruptible" intends to avoid. We're just munlocking pages that are already resident and mapped, and we're reusing get_user_pages() to access those pages. ?? Maybe we should combine 'IGNORE_VMA_PERMISSIONS and '_IGNORE_SIGKILL into a single flag: GUP_FLAGS_MUNLOCK ??? [Lee.Schermerhorn@hp.com: ignore sigkill in get_user_pages during munlock] Signed-off-by: Paul Menage <menage@google.com> Signed-off-by: Ying Han <yinghan@google.com> Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Reviewed-by: Pekka Enberg <penberg@cs.helsinki.fi> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Hugh Dickins <hugh@veritas.com> Cc: Oleg Nesterov <oleg@tv-sign.ru> Cc: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Rohit Seth <rohitseth@google.com> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:40:18 +00:00
return i ? i : -ERESTARTSYS;
cond_resched();
while (!(page = follow_page(vma, start, foll_flags))) {
int ret;
unsigned int fault_flags = 0;
/* For mlock, just skip the stack guard page. */
if (foll_flags & FOLL_MLOCK) {
if (stack_guard_page(vma, start))
goto next_page;
}
if (foll_flags & FOLL_WRITE)
fault_flags |= FAULT_FLAG_WRITE;
if (nonblocking)
fault_flags |= FAULT_FLAG_ALLOW_RETRY;
if (foll_flags & FOLL_NOWAIT)
fault_flags |= (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT);
ret = handle_mm_fault(mm, vma, start,
fault_flags);
mm: fault feedback #2 This patch completes Linus's wish that the fault return codes be made into bit flags, which I agree makes everything nicer. This requires requires all handle_mm_fault callers to be modified (possibly the modifications should go further and do things like fault accounting in handle_mm_fault -- however that would be for another patch). [akpm@linux-foundation.org: fix alpha build] [akpm@linux-foundation.org: fix s390 build] [akpm@linux-foundation.org: fix sparc build] [akpm@linux-foundation.org: fix sparc64 build] [akpm@linux-foundation.org: fix ia64 build] Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Richard Henderson <rth@twiddle.net> Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Ian Molton <spyro@f2s.com> Cc: Bryan Wu <bryan.wu@analog.com> Cc: Mikael Starvik <starvik@axis.com> Cc: David Howells <dhowells@redhat.com> Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Hirokazu Takata <takata@linux-m32r.org> Cc: Geert Uytterhoeven <geert@linux-m68k.org> Cc: Roman Zippel <zippel@linux-m68k.org> Cc: Greg Ungerer <gerg@uclinux.org> Cc: Matthew Wilcox <willy@debian.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Kazumoto Kojima <kkojima@rr.iij4u.or.jp> Cc: Richard Curnow <rc@rc0.org.uk> Cc: William Lee Irwin III <wli@holomorphy.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Jeff Dike <jdike@addtoit.com> Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it> Cc: Miles Bader <uclinux-v850@lsi.nec.co.jp> Cc: Chris Zankel <chris@zankel.net> Acked-by: Kyle McMartin <kyle@mcmartin.ca> Acked-by: Haavard Skinnemoen <hskinnemoen@atmel.com> Acked-by: Ralf Baechle <ralf@linux-mips.org> Acked-by: Andi Kleen <ak@muc.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> [ Still apparently needs some ARM and PPC loving - Linus ] Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:47:05 +00:00
if (ret & VM_FAULT_ERROR) {
if (ret & VM_FAULT_OOM)
return i ? i : -ENOMEM;
if (ret & (VM_FAULT_HWPOISON |
VM_FAULT_HWPOISON_LARGE)) {
if (i)
return i;
else if (gup_flags & FOLL_HWPOISON)
return -EHWPOISON;
else
return -EFAULT;
}
if (ret & VM_FAULT_SIGBUS)
mm: fault feedback #2 This patch completes Linus's wish that the fault return codes be made into bit flags, which I agree makes everything nicer. This requires requires all handle_mm_fault callers to be modified (possibly the modifications should go further and do things like fault accounting in handle_mm_fault -- however that would be for another patch). [akpm@linux-foundation.org: fix alpha build] [akpm@linux-foundation.org: fix s390 build] [akpm@linux-foundation.org: fix sparc build] [akpm@linux-foundation.org: fix sparc64 build] [akpm@linux-foundation.org: fix ia64 build] Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Richard Henderson <rth@twiddle.net> Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Ian Molton <spyro@f2s.com> Cc: Bryan Wu <bryan.wu@analog.com> Cc: Mikael Starvik <starvik@axis.com> Cc: David Howells <dhowells@redhat.com> Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Hirokazu Takata <takata@linux-m32r.org> Cc: Geert Uytterhoeven <geert@linux-m68k.org> Cc: Roman Zippel <zippel@linux-m68k.org> Cc: Greg Ungerer <gerg@uclinux.org> Cc: Matthew Wilcox <willy@debian.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Kazumoto Kojima <kkojima@rr.iij4u.or.jp> Cc: Richard Curnow <rc@rc0.org.uk> Cc: William Lee Irwin III <wli@holomorphy.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Jeff Dike <jdike@addtoit.com> Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it> Cc: Miles Bader <uclinux-v850@lsi.nec.co.jp> Cc: Chris Zankel <chris@zankel.net> Acked-by: Kyle McMartin <kyle@mcmartin.ca> Acked-by: Haavard Skinnemoen <hskinnemoen@atmel.com> Acked-by: Ralf Baechle <ralf@linux-mips.org> Acked-by: Andi Kleen <ak@muc.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> [ Still apparently needs some ARM and PPC loving - Linus ] Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:47:05 +00:00
return i ? i : -EFAULT;
BUG();
}
if (tsk) {
if (ret & VM_FAULT_MAJOR)
tsk->maj_flt++;
else
tsk->min_flt++;
}
mm: fault feedback #2 This patch completes Linus's wish that the fault return codes be made into bit flags, which I agree makes everything nicer. This requires requires all handle_mm_fault callers to be modified (possibly the modifications should go further and do things like fault accounting in handle_mm_fault -- however that would be for another patch). [akpm@linux-foundation.org: fix alpha build] [akpm@linux-foundation.org: fix s390 build] [akpm@linux-foundation.org: fix sparc build] [akpm@linux-foundation.org: fix sparc64 build] [akpm@linux-foundation.org: fix ia64 build] Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Richard Henderson <rth@twiddle.net> Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Ian Molton <spyro@f2s.com> Cc: Bryan Wu <bryan.wu@analog.com> Cc: Mikael Starvik <starvik@axis.com> Cc: David Howells <dhowells@redhat.com> Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Hirokazu Takata <takata@linux-m32r.org> Cc: Geert Uytterhoeven <geert@linux-m68k.org> Cc: Roman Zippel <zippel@linux-m68k.org> Cc: Greg Ungerer <gerg@uclinux.org> Cc: Matthew Wilcox <willy@debian.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Kazumoto Kojima <kkojima@rr.iij4u.or.jp> Cc: Richard Curnow <rc@rc0.org.uk> Cc: William Lee Irwin III <wli@holomorphy.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Jeff Dike <jdike@addtoit.com> Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it> Cc: Miles Bader <uclinux-v850@lsi.nec.co.jp> Cc: Chris Zankel <chris@zankel.net> Acked-by: Kyle McMartin <kyle@mcmartin.ca> Acked-by: Haavard Skinnemoen <hskinnemoen@atmel.com> Acked-by: Ralf Baechle <ralf@linux-mips.org> Acked-by: Andi Kleen <ak@muc.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> [ Still apparently needs some ARM and PPC loving - Linus ] Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:47:05 +00:00
if (ret & VM_FAULT_RETRY) {
if (nonblocking)
*nonblocking = 0;
return i;
}
/*
mm: fault feedback #2 This patch completes Linus's wish that the fault return codes be made into bit flags, which I agree makes everything nicer. This requires requires all handle_mm_fault callers to be modified (possibly the modifications should go further and do things like fault accounting in handle_mm_fault -- however that would be for another patch). [akpm@linux-foundation.org: fix alpha build] [akpm@linux-foundation.org: fix s390 build] [akpm@linux-foundation.org: fix sparc build] [akpm@linux-foundation.org: fix sparc64 build] [akpm@linux-foundation.org: fix ia64 build] Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Richard Henderson <rth@twiddle.net> Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Ian Molton <spyro@f2s.com> Cc: Bryan Wu <bryan.wu@analog.com> Cc: Mikael Starvik <starvik@axis.com> Cc: David Howells <dhowells@redhat.com> Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Hirokazu Takata <takata@linux-m32r.org> Cc: Geert Uytterhoeven <geert@linux-m68k.org> Cc: Roman Zippel <zippel@linux-m68k.org> Cc: Greg Ungerer <gerg@uclinux.org> Cc: Matthew Wilcox <willy@debian.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Kazumoto Kojima <kkojima@rr.iij4u.or.jp> Cc: Richard Curnow <rc@rc0.org.uk> Cc: William Lee Irwin III <wli@holomorphy.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Jeff Dike <jdike@addtoit.com> Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it> Cc: Miles Bader <uclinux-v850@lsi.nec.co.jp> Cc: Chris Zankel <chris@zankel.net> Acked-by: Kyle McMartin <kyle@mcmartin.ca> Acked-by: Haavard Skinnemoen <hskinnemoen@atmel.com> Acked-by: Ralf Baechle <ralf@linux-mips.org> Acked-by: Andi Kleen <ak@muc.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> [ Still apparently needs some ARM and PPC loving - Linus ] Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:47:05 +00:00
* The VM_FAULT_WRITE bit tells us that
* do_wp_page has broken COW when necessary,
* even if maybe_mkwrite decided not to set
* pte_write. We can thus safely do subsequent
* page lookups as if they were reads. But only
* do so when looping for pte_write is futile:
* in some cases userspace may also be wanting
* to write to the gotten user page, which a
* read fault here might prevent (a readonly
* page might get reCOWed by userspace write).
*/
if ((ret & VM_FAULT_WRITE) &&
!(vma->vm_flags & VM_WRITE))
foll_flags &= ~FOLL_WRITE;
mm: fault feedback #2 This patch completes Linus's wish that the fault return codes be made into bit flags, which I agree makes everything nicer. This requires requires all handle_mm_fault callers to be modified (possibly the modifications should go further and do things like fault accounting in handle_mm_fault -- however that would be for another patch). [akpm@linux-foundation.org: fix alpha build] [akpm@linux-foundation.org: fix s390 build] [akpm@linux-foundation.org: fix sparc build] [akpm@linux-foundation.org: fix sparc64 build] [akpm@linux-foundation.org: fix ia64 build] Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Richard Henderson <rth@twiddle.net> Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Ian Molton <spyro@f2s.com> Cc: Bryan Wu <bryan.wu@analog.com> Cc: Mikael Starvik <starvik@axis.com> Cc: David Howells <dhowells@redhat.com> Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Hirokazu Takata <takata@linux-m32r.org> Cc: Geert Uytterhoeven <geert@linux-m68k.org> Cc: Roman Zippel <zippel@linux-m68k.org> Cc: Greg Ungerer <gerg@uclinux.org> Cc: Matthew Wilcox <willy@debian.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Kazumoto Kojima <kkojima@rr.iij4u.or.jp> Cc: Richard Curnow <rc@rc0.org.uk> Cc: William Lee Irwin III <wli@holomorphy.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Jeff Dike <jdike@addtoit.com> Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it> Cc: Miles Bader <uclinux-v850@lsi.nec.co.jp> Cc: Chris Zankel <chris@zankel.net> Acked-by: Kyle McMartin <kyle@mcmartin.ca> Acked-by: Haavard Skinnemoen <hskinnemoen@atmel.com> Acked-by: Ralf Baechle <ralf@linux-mips.org> Acked-by: Andi Kleen <ak@muc.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> [ Still apparently needs some ARM and PPC loving - Linus ] Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:47:05 +00:00
[PATCH] page fault retry with NOPAGE_REFAULT Add a way for a no_page() handler to request a retry of the faulting instruction. It goes back to userland on page faults and just tries again in get_user_pages(). I added a cond_resched() in the loop in that later case. The problem I have with signal and spufs is an actual bug affecting apps and I don't see other ways of fixing it. In addition, we are having issues with infiniband and 64k pages (related to the way the hypervisor deals with some HV cards) that will require us to muck around with the MMU from within the IB driver's no_page() (it's a pSeries specific driver) and return to the caller the same way using NOPAGE_REFAULT. And to add to this, the graphics folks have been following a new approach of memory management that involves transparently swapping objects between video ram and main meory. To do that, they need installing PTEs from a no_page() handler as well and that also requires returning with NOPAGE_REFAULT. (For the later, they are currently using io_remap_pfn_range to install one PTE from no_page() which is a bit racy, we need to add a check for the PTE having already been installed afer taking the lock, but that's ok, they are only at the proof-of-concept stage. I'll send a patch adding a "clean" function to do that, we can use that from spufs too and get rid of the sparsemem hacks we do to create struct page for SPEs. Basically, that provides a generic solution for being able to have no_page() map hardware devices, which is something that I think sound driver folks have been asking for some time too). All of these things depend on having the NOPAGE_REFAULT exit path from no_page() handlers. Signed-off-by: Benjamin Herrenchmidt <benh@kernel.crashing.org> Cc: Hugh Dickins <hugh@veritas.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-10-06 07:43:53 +00:00
cond_resched();
}
Reinstate ZERO_PAGE optimization in 'get_user_pages()' and fix XIP KAMEZAWA Hiroyuki and Oleg Nesterov point out that since the commit 557ed1fa2620dc119adb86b34c614e152a629a80 ("remove ZERO_PAGE") removed the ZERO_PAGE from the VM mappings, any users of get_user_pages() will generally now populate the VM with real empty pages needlessly. We used to get the ZERO_PAGE when we did the "handle_mm_fault()", but since fault handling no longer uses ZERO_PAGE for new anonymous pages, we now need to handle that special case in follow_page() instead. In particular, the removal of ZERO_PAGE effectively removed the core file writing optimization where we would skip writing pages that had not been populated at all, and increased memory pressure a lot by allocating all those useless newly zeroed pages. This reinstates the optimization by making the unmapped PTE case the same as for a non-existent page table, which already did this correctly. While at it, this also fixes the XIP case for follow_page(), where the caller could not differentiate between the case of a page that simply could not be used (because it had no "struct page" associated with it) and a page that just wasn't mapped. We do that by simply returning an error pointer for pages that could not be turned into a "struct page *". The error is arbitrarily picked to be EFAULT, since that was what get_user_pages() already used for the equivalent IO-mapped page case. [ Also removed an impossible test for pte_offset_map_lock() failing: that's not how that function works ] Acked-by: Oleg Nesterov <oleg@tv-sign.ru> Acked-by: Nick Piggin <npiggin@suse.de> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Hugh Dickins <hugh@veritas.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland McGrath <roland@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-06-20 18:18:25 +00:00
if (IS_ERR(page))
return i ? i : PTR_ERR(page);
if (pages) {
pages[i] = page;
flush_anon_page(vma, page, start);
flush_dcache_page(page);
}
next_page:
if (vmas)
vmas[i] = vma;
i++;
start += PAGE_SIZE;
nr_pages--;
} while (nr_pages && start < vma->vm_end);
} while (nr_pages);
return i;
}
EXPORT_SYMBOL(__get_user_pages);
mlock: mlocked pages are unevictable Make sure that mlocked pages also live on the unevictable LRU, so kswapd will not scan them over and over again. This is achieved through various strategies: 1) add yet another page flag--PG_mlocked--to indicate that the page is locked for efficient testing in vmscan and, optionally, fault path. This allows early culling of unevictable pages, preventing them from getting to page_referenced()/try_to_unmap(). Also allows separate accounting of mlock'd pages, as Nick's original patch did. Note: Nick's original mlock patch used a PG_mlocked flag. I had removed this in favor of the PG_unevictable flag + an mlock_count [new page struct member]. I restored the PG_mlocked flag to eliminate the new count field. 2) add the mlock/unevictable infrastructure to mm/mlock.c, with internal APIs in mm/internal.h. This is a rework of Nick's original patch to these files, taking into account that mlocked pages are now kept on unevictable LRU list. 3) update vmscan.c:page_evictable() to check PageMlocked() and, if vma passed in, the vm_flags. Note that the vma will only be passed in for new pages in the fault path; and then only if the "cull unevictable pages in fault path" patch is included. 4) add try_to_unlock() to rmap.c to walk a page's rmap and ClearPageMlocked() if no other vmas have it mlocked. Reuses as much of try_to_unmap() as possible. This effectively replaces the use of one of the lru list links as an mlock count. If this mechanism let's pages in mlocked vmas leak through w/o PG_mlocked set [I don't know that it does], we should catch them later in try_to_unmap(). One hopes this will be rare, as it will be relatively expensive. Original mm/internal.h, mm/rmap.c and mm/mlock.c changes: Signed-off-by: Nick Piggin <npiggin@suse.de> splitlru: introduce __get_user_pages(): New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS. because current get_user_pages() can't grab PROT_NONE pages theresore it cause PROT_NONE pages can't munlock. [akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch] [akpm@linux-foundation.org: untangle patch interdependencies] [akpm@linux-foundation.org: fix things after out-of-order merging] [hugh@veritas.com: fix page-flags mess] [lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm'] [kosaki.motohiro@jp.fujitsu.com: build fix] [kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments] [kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()] Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Signed-off-by: Rik van Riel <riel@redhat.com> Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Dave Hansen <dave@linux.vnet.ibm.com> Cc: Matt Mackall <mpm@selenic.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 03:26:44 +00:00
mm/futex: fix futex writes on archs with SW tracking of dirty & young I haven't reproduced it myself but the fail scenario is that on such machines (notably ARM and some embedded powerpc), if you manage to hit that futex path on a writable page whose dirty bit has gone from the PTE, you'll livelock inside the kernel from what I can tell. It will go in a loop of trying the atomic access, failing, trying gup to "fix it up", getting succcess from gup, go back to the atomic access, failing again because dirty wasn't fixed etc... So I think you essentially hang in the kernel. The scenario is probably rare'ish because affected architecture are embedded and tend to not swap much (if at all) so we probably rarely hit the case where dirty is missing or young is missing, but I think Shan has a piece of SW that can reliably reproduce it using a shared writable mapping & fork or something like that. On archs who use SW tracking of dirty & young, a page without dirty is effectively mapped read-only and a page without young unaccessible in the PTE. Additionally, some architectures might lazily flush the TLB when relaxing write protection (by doing only a local flush), and expect a fault to invalidate the stale entry if it's still present on another processor. The futex code assumes that if the "in_atomic()" access -EFAULT's, it can "fix it up" by causing get_user_pages() which would then be equivalent to taking the fault. However that isn't the case. get_user_pages() will not call handle_mm_fault() in the case where the PTE seems to have the right permissions, regardless of the dirty and young state. It will eventually update those bits ... in the struct page, but not in the PTE. Additionally, it will not handle the lazy TLB flushing that can be required by some architectures in the fault case. Basically, gup is the wrong interface for the job. The patch provides a more appropriate one which boils down to just calling handle_mm_fault() since what we are trying to do is simulate a real page fault. The futex code currently attempts to write to user memory within a pagefault disabled section, and if that fails, tries to fix it up using get_user_pages(). This doesn't work on archs where the dirty and young bits are maintained by software, since they will gate access permission in the TLB, and will not be updated by gup(). In addition, there's an expectation on some archs that a spurious write fault triggers a local TLB flush, and that is missing from the picture as well. I decided that adding those "features" to gup() would be too much for this already too complex function, and instead added a new simpler fixup_user_fault() which is essentially a wrapper around handle_mm_fault() which the futex code can call. [akpm@linux-foundation.org: coding-style fixes] [akpm@linux-foundation.org: fix some nits Darren saw, fiddle comment layout] Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Reported-by: Shan Hai <haishan.bai@gmail.com> Tested-by: Shan Hai <haishan.bai@gmail.com> Cc: David Laight <David.Laight@ACULAB.COM> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Darren Hart <darren.hart@intel.com> Cc: <stable@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-07-26 00:12:32 +00:00
/*
* fixup_user_fault() - manually resolve a user page fault
* @tsk: the task_struct to use for page fault accounting, or
* NULL if faults are not to be recorded.
* @mm: mm_struct of target mm
* @address: user address
* @fault_flags:flags to pass down to handle_mm_fault()
*
* This is meant to be called in the specific scenario where for locking reasons
* we try to access user memory in atomic context (within a pagefault_disable()
* section), this returns -EFAULT, and we want to resolve the user fault before
* trying again.
*
* Typically this is meant to be used by the futex code.
*
* The main difference with get_user_pages() is that this function will
* unconditionally call handle_mm_fault() which will in turn perform all the
* necessary SW fixup of the dirty and young bits in the PTE, while
* handle_mm_fault() only guarantees to update these in the struct page.
*
* This is important for some architectures where those bits also gate the
* access permission to the page because they are maintained in software. On
* such architectures, gup() will not be enough to make a subsequent access
* succeed.
*
* This should be called with the mm_sem held for read.
*/
int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm,
unsigned long address, unsigned int fault_flags)
{
struct vm_area_struct *vma;
int ret;
vma = find_extend_vma(mm, address);
if (!vma || address < vma->vm_start)
return -EFAULT;
ret = handle_mm_fault(mm, vma, address, fault_flags);
if (ret & VM_FAULT_ERROR) {
if (ret & VM_FAULT_OOM)
return -ENOMEM;
if (ret & (VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_LARGE))
return -EHWPOISON;
if (ret & VM_FAULT_SIGBUS)
return -EFAULT;
BUG();
}
if (tsk) {
if (ret & VM_FAULT_MAJOR)
tsk->maj_flt++;
else
tsk->min_flt++;
}
return 0;
}
/*
* get_user_pages() - pin user pages in memory
* @tsk: the task_struct to use for page fault accounting, or
* NULL if faults are not to be recorded.
* @mm: mm_struct of target mm
* @start: starting user address
* @nr_pages: number of pages from start to pin
* @write: whether pages will be written to by the caller
* @force: whether to force write access even if user mapping is
* readonly. This will result in the page being COWed even
* in MAP_SHARED mappings. You do not want this.
* @pages: array that receives pointers to the pages pinned.
* Should be at least nr_pages long. Or NULL, if caller
* only intends to ensure the pages are faulted in.
* @vmas: array of pointers to vmas corresponding to each page.
* Or NULL if the caller does not require them.
*
* Returns number of pages pinned. This may be fewer than the number
* requested. If nr_pages is 0 or negative, returns 0. If no pages
* were pinned, returns -errno. Each page returned must be released
* with a put_page() call when it is finished with. vmas will only
* remain valid while mmap_sem is held.
*
* Must be called with mmap_sem held for read or write.
*
* get_user_pages walks a process's page tables and takes a reference to
* each struct page that each user address corresponds to at a given
* instant. That is, it takes the page that would be accessed if a user
* thread accesses the given user virtual address at that instant.
*
* This does not guarantee that the page exists in the user mappings when
* get_user_pages returns, and there may even be a completely different
* page there in some cases (eg. if mmapped pagecache has been invalidated
* and subsequently re faulted). However it does guarantee that the page
* won't be freed completely. And mostly callers simply care that the page
* contains data that was valid *at some point in time*. Typically, an IO
* or similar operation cannot guarantee anything stronger anyway because
* locks can't be held over the syscall boundary.
*
* If write=0, the page must not be written to. If the page is written to,
* set_page_dirty (or set_page_dirty_lock, as appropriate) must be called
* after the page is finished with, and before put_page is called.
*
* get_user_pages is typically used for fewer-copy IO operations, to get a
* handle on the memory by some means other than accesses via the user virtual
* addresses. The pages may be submitted for DMA to devices or accessed via
* their kernel linear mapping (via the kmap APIs). Care should be taken to
* use the correct cache flushing APIs.
*
* See also get_user_pages_fast, for performance critical applications.
*/
mlock: mlocked pages are unevictable Make sure that mlocked pages also live on the unevictable LRU, so kswapd will not scan them over and over again. This is achieved through various strategies: 1) add yet another page flag--PG_mlocked--to indicate that the page is locked for efficient testing in vmscan and, optionally, fault path. This allows early culling of unevictable pages, preventing them from getting to page_referenced()/try_to_unmap(). Also allows separate accounting of mlock'd pages, as Nick's original patch did. Note: Nick's original mlock patch used a PG_mlocked flag. I had removed this in favor of the PG_unevictable flag + an mlock_count [new page struct member]. I restored the PG_mlocked flag to eliminate the new count field. 2) add the mlock/unevictable infrastructure to mm/mlock.c, with internal APIs in mm/internal.h. This is a rework of Nick's original patch to these files, taking into account that mlocked pages are now kept on unevictable LRU list. 3) update vmscan.c:page_evictable() to check PageMlocked() and, if vma passed in, the vm_flags. Note that the vma will only be passed in for new pages in the fault path; and then only if the "cull unevictable pages in fault path" patch is included. 4) add try_to_unlock() to rmap.c to walk a page's rmap and ClearPageMlocked() if no other vmas have it mlocked. Reuses as much of try_to_unmap() as possible. This effectively replaces the use of one of the lru list links as an mlock count. If this mechanism let's pages in mlocked vmas leak through w/o PG_mlocked set [I don't know that it does], we should catch them later in try_to_unmap(). One hopes this will be rare, as it will be relatively expensive. Original mm/internal.h, mm/rmap.c and mm/mlock.c changes: Signed-off-by: Nick Piggin <npiggin@suse.de> splitlru: introduce __get_user_pages(): New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS. because current get_user_pages() can't grab PROT_NONE pages theresore it cause PROT_NONE pages can't munlock. [akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch] [akpm@linux-foundation.org: untangle patch interdependencies] [akpm@linux-foundation.org: fix things after out-of-order merging] [hugh@veritas.com: fix page-flags mess] [lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm'] [kosaki.motohiro@jp.fujitsu.com: build fix] [kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments] [kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()] Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Signed-off-by: Rik van Riel <riel@redhat.com> Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Dave Hansen <dave@linux.vnet.ibm.com> Cc: Matt Mackall <mpm@selenic.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 03:26:44 +00:00
int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
unsigned long start, int nr_pages, int write, int force,
mlock: mlocked pages are unevictable Make sure that mlocked pages also live on the unevictable LRU, so kswapd will not scan them over and over again. This is achieved through various strategies: 1) add yet another page flag--PG_mlocked--to indicate that the page is locked for efficient testing in vmscan and, optionally, fault path. This allows early culling of unevictable pages, preventing them from getting to page_referenced()/try_to_unmap(). Also allows separate accounting of mlock'd pages, as Nick's original patch did. Note: Nick's original mlock patch used a PG_mlocked flag. I had removed this in favor of the PG_unevictable flag + an mlock_count [new page struct member]. I restored the PG_mlocked flag to eliminate the new count field. 2) add the mlock/unevictable infrastructure to mm/mlock.c, with internal APIs in mm/internal.h. This is a rework of Nick's original patch to these files, taking into account that mlocked pages are now kept on unevictable LRU list. 3) update vmscan.c:page_evictable() to check PageMlocked() and, if vma passed in, the vm_flags. Note that the vma will only be passed in for new pages in the fault path; and then only if the "cull unevictable pages in fault path" patch is included. 4) add try_to_unlock() to rmap.c to walk a page's rmap and ClearPageMlocked() if no other vmas have it mlocked. Reuses as much of try_to_unmap() as possible. This effectively replaces the use of one of the lru list links as an mlock count. If this mechanism let's pages in mlocked vmas leak through w/o PG_mlocked set [I don't know that it does], we should catch them later in try_to_unmap(). One hopes this will be rare, as it will be relatively expensive. Original mm/internal.h, mm/rmap.c and mm/mlock.c changes: Signed-off-by: Nick Piggin <npiggin@suse.de> splitlru: introduce __get_user_pages(): New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS. because current get_user_pages() can't grab PROT_NONE pages theresore it cause PROT_NONE pages can't munlock. [akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch] [akpm@linux-foundation.org: untangle patch interdependencies] [akpm@linux-foundation.org: fix things after out-of-order merging] [hugh@veritas.com: fix page-flags mess] [lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm'] [kosaki.motohiro@jp.fujitsu.com: build fix] [kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments] [kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()] Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Signed-off-by: Rik van Riel <riel@redhat.com> Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Dave Hansen <dave@linux.vnet.ibm.com> Cc: Matt Mackall <mpm@selenic.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 03:26:44 +00:00
struct page **pages, struct vm_area_struct **vmas)
{
int flags = FOLL_TOUCH;
mlock: mlocked pages are unevictable Make sure that mlocked pages also live on the unevictable LRU, so kswapd will not scan them over and over again. This is achieved through various strategies: 1) add yet another page flag--PG_mlocked--to indicate that the page is locked for efficient testing in vmscan and, optionally, fault path. This allows early culling of unevictable pages, preventing them from getting to page_referenced()/try_to_unmap(). Also allows separate accounting of mlock'd pages, as Nick's original patch did. Note: Nick's original mlock patch used a PG_mlocked flag. I had removed this in favor of the PG_unevictable flag + an mlock_count [new page struct member]. I restored the PG_mlocked flag to eliminate the new count field. 2) add the mlock/unevictable infrastructure to mm/mlock.c, with internal APIs in mm/internal.h. This is a rework of Nick's original patch to these files, taking into account that mlocked pages are now kept on unevictable LRU list. 3) update vmscan.c:page_evictable() to check PageMlocked() and, if vma passed in, the vm_flags. Note that the vma will only be passed in for new pages in the fault path; and then only if the "cull unevictable pages in fault path" patch is included. 4) add try_to_unlock() to rmap.c to walk a page's rmap and ClearPageMlocked() if no other vmas have it mlocked. Reuses as much of try_to_unmap() as possible. This effectively replaces the use of one of the lru list links as an mlock count. If this mechanism let's pages in mlocked vmas leak through w/o PG_mlocked set [I don't know that it does], we should catch them later in try_to_unmap(). One hopes this will be rare, as it will be relatively expensive. Original mm/internal.h, mm/rmap.c and mm/mlock.c changes: Signed-off-by: Nick Piggin <npiggin@suse.de> splitlru: introduce __get_user_pages(): New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS. because current get_user_pages() can't grab PROT_NONE pages theresore it cause PROT_NONE pages can't munlock. [akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch] [akpm@linux-foundation.org: untangle patch interdependencies] [akpm@linux-foundation.org: fix things after out-of-order merging] [hugh@veritas.com: fix page-flags mess] [lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm'] [kosaki.motohiro@jp.fujitsu.com: build fix] [kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments] [kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()] Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Signed-off-by: Rik van Riel <riel@redhat.com> Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Dave Hansen <dave@linux.vnet.ibm.com> Cc: Matt Mackall <mpm@selenic.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 03:26:44 +00:00
if (pages)
flags |= FOLL_GET;
mlock: mlocked pages are unevictable Make sure that mlocked pages also live on the unevictable LRU, so kswapd will not scan them over and over again. This is achieved through various strategies: 1) add yet another page flag--PG_mlocked--to indicate that the page is locked for efficient testing in vmscan and, optionally, fault path. This allows early culling of unevictable pages, preventing them from getting to page_referenced()/try_to_unmap(). Also allows separate accounting of mlock'd pages, as Nick's original patch did. Note: Nick's original mlock patch used a PG_mlocked flag. I had removed this in favor of the PG_unevictable flag + an mlock_count [new page struct member]. I restored the PG_mlocked flag to eliminate the new count field. 2) add the mlock/unevictable infrastructure to mm/mlock.c, with internal APIs in mm/internal.h. This is a rework of Nick's original patch to these files, taking into account that mlocked pages are now kept on unevictable LRU list. 3) update vmscan.c:page_evictable() to check PageMlocked() and, if vma passed in, the vm_flags. Note that the vma will only be passed in for new pages in the fault path; and then only if the "cull unevictable pages in fault path" patch is included. 4) add try_to_unlock() to rmap.c to walk a page's rmap and ClearPageMlocked() if no other vmas have it mlocked. Reuses as much of try_to_unmap() as possible. This effectively replaces the use of one of the lru list links as an mlock count. If this mechanism let's pages in mlocked vmas leak through w/o PG_mlocked set [I don't know that it does], we should catch them later in try_to_unmap(). One hopes this will be rare, as it will be relatively expensive. Original mm/internal.h, mm/rmap.c and mm/mlock.c changes: Signed-off-by: Nick Piggin <npiggin@suse.de> splitlru: introduce __get_user_pages(): New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS. because current get_user_pages() can't grab PROT_NONE pages theresore it cause PROT_NONE pages can't munlock. [akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch] [akpm@linux-foundation.org: untangle patch interdependencies] [akpm@linux-foundation.org: fix things after out-of-order merging] [hugh@veritas.com: fix page-flags mess] [lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm'] [kosaki.motohiro@jp.fujitsu.com: build fix] [kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments] [kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()] Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Signed-off-by: Rik van Riel <riel@redhat.com> Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Dave Hansen <dave@linux.vnet.ibm.com> Cc: Matt Mackall <mpm@selenic.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 03:26:44 +00:00
if (write)
flags |= FOLL_WRITE;
mlock: mlocked pages are unevictable Make sure that mlocked pages also live on the unevictable LRU, so kswapd will not scan them over and over again. This is achieved through various strategies: 1) add yet another page flag--PG_mlocked--to indicate that the page is locked for efficient testing in vmscan and, optionally, fault path. This allows early culling of unevictable pages, preventing them from getting to page_referenced()/try_to_unmap(). Also allows separate accounting of mlock'd pages, as Nick's original patch did. Note: Nick's original mlock patch used a PG_mlocked flag. I had removed this in favor of the PG_unevictable flag + an mlock_count [new page struct member]. I restored the PG_mlocked flag to eliminate the new count field. 2) add the mlock/unevictable infrastructure to mm/mlock.c, with internal APIs in mm/internal.h. This is a rework of Nick's original patch to these files, taking into account that mlocked pages are now kept on unevictable LRU list. 3) update vmscan.c:page_evictable() to check PageMlocked() and, if vma passed in, the vm_flags. Note that the vma will only be passed in for new pages in the fault path; and then only if the "cull unevictable pages in fault path" patch is included. 4) add try_to_unlock() to rmap.c to walk a page's rmap and ClearPageMlocked() if no other vmas have it mlocked. Reuses as much of try_to_unmap() as possible. This effectively replaces the use of one of the lru list links as an mlock count. If this mechanism let's pages in mlocked vmas leak through w/o PG_mlocked set [I don't know that it does], we should catch them later in try_to_unmap(). One hopes this will be rare, as it will be relatively expensive. Original mm/internal.h, mm/rmap.c and mm/mlock.c changes: Signed-off-by: Nick Piggin <npiggin@suse.de> splitlru: introduce __get_user_pages(): New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS. because current get_user_pages() can't grab PROT_NONE pages theresore it cause PROT_NONE pages can't munlock. [akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch] [akpm@linux-foundation.org: untangle patch interdependencies] [akpm@linux-foundation.org: fix things after out-of-order merging] [hugh@veritas.com: fix page-flags mess] [lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm'] [kosaki.motohiro@jp.fujitsu.com: build fix] [kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments] [kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()] Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Signed-off-by: Rik van Riel <riel@redhat.com> Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Dave Hansen <dave@linux.vnet.ibm.com> Cc: Matt Mackall <mpm@selenic.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 03:26:44 +00:00
if (force)
flags |= FOLL_FORCE;
mlock: mlocked pages are unevictable Make sure that mlocked pages also live on the unevictable LRU, so kswapd will not scan them over and over again. This is achieved through various strategies: 1) add yet another page flag--PG_mlocked--to indicate that the page is locked for efficient testing in vmscan and, optionally, fault path. This allows early culling of unevictable pages, preventing them from getting to page_referenced()/try_to_unmap(). Also allows separate accounting of mlock'd pages, as Nick's original patch did. Note: Nick's original mlock patch used a PG_mlocked flag. I had removed this in favor of the PG_unevictable flag + an mlock_count [new page struct member]. I restored the PG_mlocked flag to eliminate the new count field. 2) add the mlock/unevictable infrastructure to mm/mlock.c, with internal APIs in mm/internal.h. This is a rework of Nick's original patch to these files, taking into account that mlocked pages are now kept on unevictable LRU list. 3) update vmscan.c:page_evictable() to check PageMlocked() and, if vma passed in, the vm_flags. Note that the vma will only be passed in for new pages in the fault path; and then only if the "cull unevictable pages in fault path" patch is included. 4) add try_to_unlock() to rmap.c to walk a page's rmap and ClearPageMlocked() if no other vmas have it mlocked. Reuses as much of try_to_unmap() as possible. This effectively replaces the use of one of the lru list links as an mlock count. If this mechanism let's pages in mlocked vmas leak through w/o PG_mlocked set [I don't know that it does], we should catch them later in try_to_unmap(). One hopes this will be rare, as it will be relatively expensive. Original mm/internal.h, mm/rmap.c and mm/mlock.c changes: Signed-off-by: Nick Piggin <npiggin@suse.de> splitlru: introduce __get_user_pages(): New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS. because current get_user_pages() can't grab PROT_NONE pages theresore it cause PROT_NONE pages can't munlock. [akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch] [akpm@linux-foundation.org: untangle patch interdependencies] [akpm@linux-foundation.org: fix things after out-of-order merging] [hugh@veritas.com: fix page-flags mess] [lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm'] [kosaki.motohiro@jp.fujitsu.com: build fix] [kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments] [kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()] Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Signed-off-by: Rik van Riel <riel@redhat.com> Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Dave Hansen <dave@linux.vnet.ibm.com> Cc: Matt Mackall <mpm@selenic.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 03:26:44 +00:00
return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
NULL);
mlock: mlocked pages are unevictable Make sure that mlocked pages also live on the unevictable LRU, so kswapd will not scan them over and over again. This is achieved through various strategies: 1) add yet another page flag--PG_mlocked--to indicate that the page is locked for efficient testing in vmscan and, optionally, fault path. This allows early culling of unevictable pages, preventing them from getting to page_referenced()/try_to_unmap(). Also allows separate accounting of mlock'd pages, as Nick's original patch did. Note: Nick's original mlock patch used a PG_mlocked flag. I had removed this in favor of the PG_unevictable flag + an mlock_count [new page struct member]. I restored the PG_mlocked flag to eliminate the new count field. 2) add the mlock/unevictable infrastructure to mm/mlock.c, with internal APIs in mm/internal.h. This is a rework of Nick's original patch to these files, taking into account that mlocked pages are now kept on unevictable LRU list. 3) update vmscan.c:page_evictable() to check PageMlocked() and, if vma passed in, the vm_flags. Note that the vma will only be passed in for new pages in the fault path; and then only if the "cull unevictable pages in fault path" patch is included. 4) add try_to_unlock() to rmap.c to walk a page's rmap and ClearPageMlocked() if no other vmas have it mlocked. Reuses as much of try_to_unmap() as possible. This effectively replaces the use of one of the lru list links as an mlock count. If this mechanism let's pages in mlocked vmas leak through w/o PG_mlocked set [I don't know that it does], we should catch them later in try_to_unmap(). One hopes this will be rare, as it will be relatively expensive. Original mm/internal.h, mm/rmap.c and mm/mlock.c changes: Signed-off-by: Nick Piggin <npiggin@suse.de> splitlru: introduce __get_user_pages(): New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS. because current get_user_pages() can't grab PROT_NONE pages theresore it cause PROT_NONE pages can't munlock. [akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch] [akpm@linux-foundation.org: untangle patch interdependencies] [akpm@linux-foundation.org: fix things after out-of-order merging] [hugh@veritas.com: fix page-flags mess] [lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm'] [kosaki.motohiro@jp.fujitsu.com: build fix] [kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments] [kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()] Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Signed-off-by: Rik van Riel <riel@redhat.com> Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Dave Hansen <dave@linux.vnet.ibm.com> Cc: Matt Mackall <mpm@selenic.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 03:26:44 +00:00
}
EXPORT_SYMBOL(get_user_pages);
/**
* get_dump_page() - pin user page in memory while writing it to core dump
* @addr: user address
*
* Returns struct page pointer of user page pinned for dump,
* to be freed afterwards by page_cache_release() or put_page().
*
* Returns NULL on any kind of failure - a hole must then be inserted into
* the corefile, to preserve alignment with its headers; and also returns
* NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found -
* allowing a hole to be left in the corefile to save diskspace.
*
* Called without mmap_sem, but after all other threads have been killed.
*/
#ifdef CONFIG_ELF_CORE
struct page *get_dump_page(unsigned long addr)
{
struct vm_area_struct *vma;
struct page *page;
if (__get_user_pages(current, current->mm, addr, 1,
FOLL_FORCE | FOLL_DUMP | FOLL_GET, &page, &vma,
NULL) < 1)
return NULL;
flush_cache_page(vma, addr, page_to_pfn(page));
return page;
}
#endif /* CONFIG_ELF_CORE */
pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr,
spinlock_t **ptl)
{
pgd_t * pgd = pgd_offset(mm, addr);
pud_t * pud = pud_alloc(mm, pgd, addr);
if (pud) {
pmd_t * pmd = pmd_alloc(mm, pud, addr);
if (pmd) {
VM_BUG_ON(pmd_trans_huge(*pmd));
return pte_alloc_map_lock(mm, pmd, addr, ptl);
}
}
return NULL;
}
/*
* This is the old fallback for page remapping.
*
* For historical reasons, it only allows reserved pages. Only
* old drivers should use this, and they needed to mark their
* pages reserved for the old functions anyway.
*/
static int insert_page(struct vm_area_struct *vma, unsigned long addr,
struct page *page, pgprot_t prot)
{
struct mm_struct *mm = vma->vm_mm;
int retval;
pte_t *pte;
Memory controller: memory accounting Add the accounting hooks. The accounting is carried out for RSS and Page Cache (unmapped) pages. There is now a common limit and accounting for both. The RSS accounting is accounted at page_add_*_rmap() and page_remove_rmap() time. Page cache is accounted at add_to_page_cache(), __delete_from_page_cache(). Swap cache is also accounted for. Each page's page_cgroup is protected with the last bit of the page_cgroup pointer, this makes handling of race conditions involving simultaneous mappings of a page easier. A reference count is kept in the page_cgroup to deal with cases where a page might be unmapped from the RSS of all tasks, but still lives in the page cache. Credits go to Vaidyanathan Srinivasan for helping with reference counting work of the page cgroup. Almost all of the page cache accounting code has help from Vaidyanathan Srinivasan. [hugh@veritas.com: fix swapoff breakage] [akpm@linux-foundation.org: fix locking] Signed-off-by: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com> Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com> Cc: Pavel Emelianov <xemul@openvz.org> Cc: Paul Menage <menage@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Kirill Korotaev <dev@sw.ru> Cc: Herbert Poetzl <herbert@13thfloor.at> Cc: David Rientjes <rientjes@google.com> Cc: <Valdis.Kletnieks@vt.edu> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-07 08:13:53 +00:00
spinlock_t *ptl;
retval = -EINVAL;
if (PageAnon(page))
goto out;
retval = -ENOMEM;
flush_dcache_page(page);
pte = get_locked_pte(mm, addr, &ptl);
if (!pte)
goto out;
retval = -EBUSY;
if (!pte_none(*pte))
goto out_unlock;
/* Ok, finally just insert the thing.. */
get_page(page);
inc_mm_counter_fast(mm, MM_FILEPAGES);
page_add_file_rmap(page);
set_pte_at(mm, addr, pte, mk_pte(page, prot));
retval = 0;
Memory controller: memory accounting Add the accounting hooks. The accounting is carried out for RSS and Page Cache (unmapped) pages. There is now a common limit and accounting for both. The RSS accounting is accounted at page_add_*_rmap() and page_remove_rmap() time. Page cache is accounted at add_to_page_cache(), __delete_from_page_cache(). Swap cache is also accounted for. Each page's page_cgroup is protected with the last bit of the page_cgroup pointer, this makes handling of race conditions involving simultaneous mappings of a page easier. A reference count is kept in the page_cgroup to deal with cases where a page might be unmapped from the RSS of all tasks, but still lives in the page cache. Credits go to Vaidyanathan Srinivasan for helping with reference counting work of the page cgroup. Almost all of the page cache accounting code has help from Vaidyanathan Srinivasan. [hugh@veritas.com: fix swapoff breakage] [akpm@linux-foundation.org: fix locking] Signed-off-by: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com> Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com> Cc: Pavel Emelianov <xemul@openvz.org> Cc: Paul Menage <menage@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Kirill Korotaev <dev@sw.ru> Cc: Herbert Poetzl <herbert@13thfloor.at> Cc: David Rientjes <rientjes@google.com> Cc: <Valdis.Kletnieks@vt.edu> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-07 08:13:53 +00:00
pte_unmap_unlock(pte, ptl);
return retval;
out_unlock:
pte_unmap_unlock(pte, ptl);
out:
return retval;
}
/**
* vm_insert_page - insert single page into user vma
* @vma: user vma to map to
* @addr: target user address of this page
* @page: source kernel page
*
* This allows drivers to insert individual pages they've allocated
* into a user vma.
*
* The page has to be a nice clean _individual_ kernel allocation.
* If you allocate a compound page, you need to have marked it as
* such (__GFP_COMP), or manually just split the page up yourself
* (see split_page()).
*
* NOTE! Traditionally this was done with "remap_pfn_range()" which
* took an arbitrary page protection parameter. This doesn't allow
* that. Your vma protection will have to be set up correctly, which
* means that if you want a shared writable mapping, you'd better
* ask for a shared writable mapping!
*
* The page does not need to be reserved.
mm: kill vma flag VM_INSERTPAGE Merge VM_INSERTPAGE into VM_MIXEDMAP. VM_MIXEDMAP VMA can mix pure-pfn ptes, special ptes and normal ptes. Now copy_page_range() always copies VM_MIXEDMAP VMA on fork like VM_PFNMAP. If driver populates whole VMA at mmap() it probably not expects page-faults. This patch removes special check from vma_wants_writenotify() which disables pages write tracking for VMA populated via vm_instert_page(). BDI below mapped file should not use dirty-accounting, moreover do_wp_page() can handle this. vm_insert_page() still marks vma after first usage. Usually it is called from f_op->mmap() handler under mm->mmap_sem write-lock, so it able to change vma->vm_flags. Caller must set VM_MIXEDMAP at mmap time if it wants to call this function from other places, for example from page-fault handler. Signed-off-by: Konstantin Khlebnikov <khlebnikov@openvz.org> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Carsten Otte <cotte@de.ibm.com> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Eric Paris <eparis@redhat.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Hugh Dickins <hughd@google.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: James Morris <james.l.morris@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Kentaro Takeda <takedakn@nttdata.co.jp> Cc: Matt Helsley <matthltc@us.ibm.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> Cc: Suresh Siddha <suresh.b.siddha@intel.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Venkatesh Pallipadi <venki@google.com> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-08 23:28:40 +00:00
*
* Usually this function is called from f_op->mmap() handler
* under mm->mmap_sem write-lock, so it can change vma->vm_flags.
* Caller must set VM_MIXEDMAP on vma if it wants to call this
* function from other places, for example from page-fault handler.
*/
int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
struct page *page)
{
if (addr < vma->vm_start || addr >= vma->vm_end)
return -EFAULT;
if (!page_count(page))
return -EINVAL;
mm: kill vma flag VM_INSERTPAGE Merge VM_INSERTPAGE into VM_MIXEDMAP. VM_MIXEDMAP VMA can mix pure-pfn ptes, special ptes and normal ptes. Now copy_page_range() always copies VM_MIXEDMAP VMA on fork like VM_PFNMAP. If driver populates whole VMA at mmap() it probably not expects page-faults. This patch removes special check from vma_wants_writenotify() which disables pages write tracking for VMA populated via vm_instert_page(). BDI below mapped file should not use dirty-accounting, moreover do_wp_page() can handle this. vm_insert_page() still marks vma after first usage. Usually it is called from f_op->mmap() handler under mm->mmap_sem write-lock, so it able to change vma->vm_flags. Caller must set VM_MIXEDMAP at mmap time if it wants to call this function from other places, for example from page-fault handler. Signed-off-by: Konstantin Khlebnikov <khlebnikov@openvz.org> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Carsten Otte <cotte@de.ibm.com> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Eric Paris <eparis@redhat.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Hugh Dickins <hughd@google.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: James Morris <james.l.morris@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Kentaro Takeda <takedakn@nttdata.co.jp> Cc: Matt Helsley <matthltc@us.ibm.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> Cc: Suresh Siddha <suresh.b.siddha@intel.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Venkatesh Pallipadi <venki@google.com> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-08 23:28:40 +00:00
if (!(vma->vm_flags & VM_MIXEDMAP)) {
BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem));
BUG_ON(vma->vm_flags & VM_PFNMAP);
vma->vm_flags |= VM_MIXEDMAP;
}
return insert_page(vma, addr, page, vma->vm_page_prot);
}
EXPORT_SYMBOL(vm_insert_page);
static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
unsigned long pfn, pgprot_t prot)
{
struct mm_struct *mm = vma->vm_mm;
int retval;
pte_t *pte, entry;
spinlock_t *ptl;
retval = -ENOMEM;
pte = get_locked_pte(mm, addr, &ptl);
if (!pte)
goto out;
retval = -EBUSY;
if (!pte_none(*pte))
goto out_unlock;
/* Ok, finally just insert the thing.. */
entry = pte_mkspecial(pfn_pte(pfn, prot));
set_pte_at(mm, addr, pte, entry);
update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */
retval = 0;
out_unlock:
pte_unmap_unlock(pte, ptl);
out:
return retval;
}
/**
* vm_insert_pfn - insert single pfn into user vma
* @vma: user vma to map to
* @addr: target user address of this page
* @pfn: source kernel pfn
*
* Similar to vm_insert_page, this allows drivers to insert individual pages
* they've allocated into a user vma. Same comments apply.
*
* This function should only be called from a vm_ops->fault handler, and
* in that case the handler should return NULL.
*
* vma cannot be a COW mapping.
*
* As this is called only for pages that do not currently exist, we
* do not need to flush old virtual caches or the TLB.
*/
int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
unsigned long pfn)
{
int ret;
pgprot_t pgprot = vma->vm_page_prot;
mm: introduce pte_special pte bit s390 for one, cannot implement VM_MIXEDMAP with pfn_valid, due to their memory model (which is more dynamic than most). Instead, they had proposed to implement it with an additional path through vm_normal_page(), using a bit in the pte to determine whether or not the page should be refcounted: vm_normal_page() { ... if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { if (vma->vm_flags & VM_MIXEDMAP) { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; #else if (!pfn_valid(pfn)) return NULL; #endif goto out; } ... } This is fine, however if we are allowed to use a bit in the pte to determine refcountedness, we can use that to _completely_ replace all the vma based schemes. So instead of adding more cases to the already complex vma-based scheme, we can have a clearly seperate and simple pte-based scheme (and get slightly better code generation in the process): vm_normal_page() { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; return pte_page(pte); #else ... #endif } And finally, we may rather make this concept usable by any architecture rather than making it s390 only, so implement a new type of pte state for this. Unfortunately the old vma based code must stay, because some architectures may not be able to spare pte bits. This makes vm_normal_page a little bit more ugly than we would like, but the 2 cases are clearly seperate. So introduce a pte_special pte state, and use it in mm/memory.c. It is currently a noop for all architectures, so this doesn't actually result in any compiled code changes to mm/memory.o. BTW: I haven't put vm_normal_page() into arch code as-per an earlier suggestion. The reason is that, regardless of where vm_normal_page is actually implemented, the *abstraction* is still exactly the same. Also, while it depends on whether the architecture has pte_special or not, that is the only two possible cases, and it really isn't an arch specific function -- the role of the arch code should be to provide primitive functions and accessors with which to build the core code; pte_special does that. We do not want architectures to know or care about vm_normal_page itself, and we definitely don't want them being able to invent something new there out of sight of mm/ code. If we made vm_normal_page an arch function, then we have to make vm_insert_mixed (next patch) an arch function too. So I don't think moving it to arch code fundamentally improves any abstractions, while it does practically make the code more difficult to follow, for both mm and arch developers, and easier to misuse. [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Acked-by: Carsten Otte <cotte@de.ibm.com> Cc: Jared Hulbert <jaredeh@gmail.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 09:13:00 +00:00
/*
* Technically, architectures with pte_special can avoid all these
* restrictions (same for remap_pfn_range). However we would like
* consistency in testing and feature parity among all, so we should
* try to keep these invariants in place for everybody.
*/
mm: introduce VM_MIXEDMAP This series introduces some important infrastructure work. The overall result is that: 1. We now support XIP backed filesystems using memory that have no struct page allocated to them. And patches 6 and 7 actually implement this for s390. This is pretty important in a number of cases. As far as I understand, in the case of virtualisation (eg. s390), each guest may mount a readonly copy of the same filesystem (eg. the distro). Currently, guests need to allocate struct pages for this image. So if you have 100 guests, you already need to allocate more memory for the struct pages than the size of the image. I think. (Carsten?) For other (eg. embedded) systems, you may have a very large non- volatile filesystem. If you have to have struct pages for this, then your RAM consumption will go up proportionally to fs size. Even though it is just a small proportion, the RAM can be much more costly eg in terms of power, so every KB less that Linux uses makes it more attractive to a lot of these guys. 2. VM_MIXEDMAP allows us to support mappings where you actually do want to refcount _some_ pages in the mapping, but not others, and support COW on arbitrary (non-linear) mappings. Jared needs this for his NVRAM filesystem in progress. Future iterations of this filesystem will most likely want to migrate pages between pagecache and XIP backing, which is where the requirement for mixed (some refcounted, some not) comes from. 3. pte_special also has a peripheral usage that I need for my lockless get_user_pages patch. That was shown to speed up "oltp" on db2 by 10% on a 2 socket system, which is kind of significant because they scrounge for months to try to find 0.1% improvement on these workloads. I'm hoping we might finally be faster than AIX on pSeries with this :). My reference to lockless get_user_pages is not meant to justify this patchset (which doesn't include lockless gup), but just to show that pte_special is not some s390 specific thing that should be hidden in arch code or xip code: I definitely want to use it on at least x86 and powerpc as well. This patch: Introduce a new type of mapping, VM_MIXEDMAP. This is unlike VM_PFNMAP in that it can support COW mappings of arbitrary ranges including ranges without struct page *and* ranges with a struct page that we actually want to refcount (PFNMAP can only support COW in those cases where the un-COW-ed translations are mapped linearly in the virtual address, and can only support non refcounted ranges). VM_MIXEDMAP achieves this by refcounting all pfn_valid pages, and not refcounting !pfn_valid pages (which is not an option for VM_PFNMAP, because it needs to avoid refcounting pfn_valid pages eg. for /dev/mem mappings). Signed-off-by: Jared Hulbert <jaredeh@gmail.com> Signed-off-by: Nick Piggin <npiggin@suse.de> Acked-by: Carsten Otte <cotte@de.ibm.com> Cc: Jared Hulbert <jaredeh@gmail.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 09:12:58 +00:00
BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
(VM_PFNMAP|VM_MIXEDMAP));
BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
if (addr < vma->vm_start || addr >= vma->vm_end)
return -EFAULT;
x86, pat: separate the pfn attribute tracking for remap_pfn_range and vm_insert_pfn With PAT enabled, vm_insert_pfn() looks up the existing pfn memory attribute and uses it. Expectation is that the driver reserves the memory attributes for the pfn before calling vm_insert_pfn(). remap_pfn_range() (when called for the whole vma) will setup a new attribute (based on the prot argument) for the specified pfn range. This addresses the legacy usage which typically calls remap_pfn_range() with a desired memory attribute. For ranges smaller than the vma size (which is typically not the case), remap_pfn_range() will use the existing memory attribute for the pfn range. Expose two different API's for these different behaviors. track_pfn_insert() for tracking the pfn attribute set by vm_insert_pfn() and track_pfn_remap() for the remap_pfn_range(). This cleanup also prepares the ground for the track/untrack pfn vma routines to take over the ownership of setting PAT specific vm_flag in the 'vma'. [khlebnikov@openvz.org: Clear checks in track_pfn_remap()] [akpm@linux-foundation.org: tweak a few comments] Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com> Signed-off-by: Konstantin Khlebnikov <khlebnikov@openvz.org> Cc: Venkatesh Pallipadi <venki@google.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Ingo Molnar <mingo@redhat.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Carsten Otte <cotte@de.ibm.com> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Eric Paris <eparis@redhat.com> Cc: Hugh Dickins <hughd@google.com> Cc: James Morris <james.l.morris@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Kentaro Takeda <takedakn@nttdata.co.jp> Cc: Konstantin Khlebnikov <khlebnikov@openvz.org> Cc: Matt Helsley <matthltc@us.ibm.com> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> Cc: Suresh Siddha <suresh.b.siddha@intel.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-08 23:28:29 +00:00
if (track_pfn_insert(vma, &pgprot, pfn))
return -EINVAL;
ret = insert_pfn(vma, addr, pfn, pgprot);
return ret;
}
EXPORT_SYMBOL(vm_insert_pfn);
int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
unsigned long pfn)
{
BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));
if (addr < vma->vm_start || addr >= vma->vm_end)
return -EFAULT;
/*
* If we don't have pte special, then we have to use the pfn_valid()
* based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
* refcount the page if pfn_valid is true (hence insert_page rather
* than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP
* without pte special, it would there be refcounted as a normal page.
*/
if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) {
struct page *page;
page = pfn_to_page(pfn);
return insert_page(vma, addr, page, vma->vm_page_prot);
}
return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
}
EXPORT_SYMBOL(vm_insert_mixed);
/*
* maps a range of physical memory into the requested pages. the old
* mappings are removed. any references to nonexistent pages results
* in null mappings (currently treated as "copy-on-access")
*/
static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
unsigned long addr, unsigned long end,
unsigned long pfn, pgprot_t prot)
{
pte_t *pte;
spinlock_t *ptl;
pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
if (!pte)
return -ENOMEM;
arch_enter_lazy_mmu_mode();
do {
BUG_ON(!pte_none(*pte));
mm: introduce pte_special pte bit s390 for one, cannot implement VM_MIXEDMAP with pfn_valid, due to their memory model (which is more dynamic than most). Instead, they had proposed to implement it with an additional path through vm_normal_page(), using a bit in the pte to determine whether or not the page should be refcounted: vm_normal_page() { ... if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { if (vma->vm_flags & VM_MIXEDMAP) { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; #else if (!pfn_valid(pfn)) return NULL; #endif goto out; } ... } This is fine, however if we are allowed to use a bit in the pte to determine refcountedness, we can use that to _completely_ replace all the vma based schemes. So instead of adding more cases to the already complex vma-based scheme, we can have a clearly seperate and simple pte-based scheme (and get slightly better code generation in the process): vm_normal_page() { #ifdef s390 if (!mixedmap_refcount_pte(pte)) return NULL; return pte_page(pte); #else ... #endif } And finally, we may rather make this concept usable by any architecture rather than making it s390 only, so implement a new type of pte state for this. Unfortunately the old vma based code must stay, because some architectures may not be able to spare pte bits. This makes vm_normal_page a little bit more ugly than we would like, but the 2 cases are clearly seperate. So introduce a pte_special pte state, and use it in mm/memory.c. It is currently a noop for all architectures, so this doesn't actually result in any compiled code changes to mm/memory.o. BTW: I haven't put vm_normal_page() into arch code as-per an earlier suggestion. The reason is that, regardless of where vm_normal_page is actually implemented, the *abstraction* is still exactly the same. Also, while it depends on whether the architecture has pte_special or not, that is the only two possible cases, and it really isn't an arch specific function -- the role of the arch code should be to provide primitive functions and accessors with which to build the core code; pte_special does that. We do not want architectures to know or care about vm_normal_page itself, and we definitely don't want them being able to invent something new there out of sight of mm/ code. If we made vm_normal_page an arch function, then we have to make vm_insert_mixed (next patch) an arch function too. So I don't think moving it to arch code fundamentally improves any abstractions, while it does practically make the code more difficult to follow, for both mm and arch developers, and easier to misuse. [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Acked-by: Carsten Otte <cotte@de.ibm.com> Cc: Jared Hulbert <jaredeh@gmail.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 09:13:00 +00:00
set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
pfn++;
} while (pte++, addr += PAGE_SIZE, addr != end);
arch_leave_lazy_mmu_mode();
pte_unmap_unlock(pte - 1, ptl);
return 0;
}
static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
unsigned long addr, unsigned long end,
unsigned long pfn, pgprot_t prot)
{
pmd_t *pmd;
unsigned long next;
pfn -= addr >> PAGE_SHIFT;
pmd = pmd_alloc(mm, pud, addr);
if (!pmd)
return -ENOMEM;
VM_BUG_ON(pmd_trans_huge(*pmd));
do {
next = pmd_addr_end(addr, end);
if (remap_pte_range(mm, pmd, addr, next,
pfn + (addr >> PAGE_SHIFT), prot))
return -ENOMEM;
} while (pmd++, addr = next, addr != end);
return 0;
}
static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
unsigned long addr, unsigned long end,
unsigned long pfn, pgprot_t prot)
{
pud_t *pud;
unsigned long next;
pfn -= addr >> PAGE_SHIFT;
pud = pud_alloc(mm, pgd, addr);
if (!pud)
return -ENOMEM;
do {
next = pud_addr_end(addr, end);
if (remap_pmd_range(mm, pud, addr, next,
pfn + (addr >> PAGE_SHIFT), prot))
return -ENOMEM;
} while (pud++, addr = next, addr != end);
return 0;
}
/**
* remap_pfn_range - remap kernel memory to userspace
* @vma: user vma to map to
* @addr: target user address to start at
* @pfn: physical address of kernel memory
* @size: size of map area
* @prot: page protection flags for this mapping
*
* Note: this is only safe if the mm semaphore is held when called.
*/
int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
unsigned long pfn, unsigned long size, pgprot_t prot)
{
pgd_t *pgd;
unsigned long next;
unsigned long end = addr + PAGE_ALIGN(size);
struct mm_struct *mm = vma->vm_mm;
int err;
/*
* Physically remapped pages are special. Tell the
* rest of the world about it:
* VM_IO tells people not to look at these pages
* (accesses can have side effects).
* VM_PFNMAP tells the core MM that the base pages are just
* raw PFN mappings, and do not have a "struct page" associated
* with them.
mm: kill vma flag VM_RESERVED and mm->reserved_vm counter A long time ago, in v2.4, VM_RESERVED kept swapout process off VMA, currently it lost original meaning but still has some effects: | effect | alternative flags -+------------------------+--------------------------------------------- 1| account as reserved_vm | VM_IO 2| skip in core dump | VM_IO, VM_DONTDUMP 3| do not merge or expand | VM_IO, VM_DONTEXPAND, VM_HUGETLB, VM_PFNMAP 4| do not mlock | VM_IO, VM_DONTEXPAND, VM_HUGETLB, VM_PFNMAP This patch removes reserved_vm counter from mm_struct. Seems like nobody cares about it, it does not exported into userspace directly, it only reduces total_vm showed in proc. Thus VM_RESERVED can be replaced with VM_IO or pair VM_DONTEXPAND | VM_DONTDUMP. remap_pfn_range() and io_remap_pfn_range() set VM_IO|VM_DONTEXPAND|VM_DONTDUMP. remap_vmalloc_range() set VM_DONTEXPAND | VM_DONTDUMP. [akpm@linux-foundation.org: drivers/vfio/pci/vfio_pci.c fixup] Signed-off-by: Konstantin Khlebnikov <khlebnikov@openvz.org> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Carsten Otte <cotte@de.ibm.com> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Eric Paris <eparis@redhat.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Hugh Dickins <hughd@google.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: James Morris <james.l.morris@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Kentaro Takeda <takedakn@nttdata.co.jp> Cc: Matt Helsley <matthltc@us.ibm.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> Cc: Suresh Siddha <suresh.b.siddha@intel.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Venkatesh Pallipadi <venki@google.com> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-08 23:29:02 +00:00
* VM_DONTEXPAND
* Disable vma merging and expanding with mremap().
* VM_DONTDUMP
* Omit vma from core dump, even when VM_IO turned off.
*
* There's a horrible special case to handle copy-on-write
* behaviour that some programs depend on. We mark the "original"
* un-COW'ed pages by matching them up with "vma->vm_pgoff".
mm, x86, pat: rework linear pfn-mmap tracking Replace the generic vma-flag VM_PFN_AT_MMAP with x86-only VM_PAT. We can toss mapping address from remap_pfn_range() into track_pfn_vma_new(), and collect all PAT-related logic together in arch/x86/. This patch also restores orignal frustration-free is_cow_mapping() check in remap_pfn_range(), as it was before commit v2.6.28-rc8-88-g3c8bb73 ("x86: PAT: store vm_pgoff for all linear_over_vma_region mappings - v3") is_linear_pfn_mapping() checks can be removed from mm/huge_memory.c, because it already handled by VM_PFNMAP in VM_NO_THP bit-mask. [suresh.b.siddha@intel.com: Reset the VM_PAT flag as part of untrack_pfn_vma()] Signed-off-by: Konstantin Khlebnikov <khlebnikov@openvz.org> Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com> Cc: Venkatesh Pallipadi <venki@google.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Ingo Molnar <mingo@redhat.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Carsten Otte <cotte@de.ibm.com> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Eric Paris <eparis@redhat.com> Cc: Hugh Dickins <hughd@google.com> Cc: James Morris <james.l.morris@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Kentaro Takeda <takedakn@nttdata.co.jp> Cc: Matt Helsley <matthltc@us.ibm.com> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Venkatesh Pallipadi <venki@google.com> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-08 23:28:34 +00:00
* See vm_normal_page() for details.
*/
mm, x86, pat: rework linear pfn-mmap tracking Replace the generic vma-flag VM_PFN_AT_MMAP with x86-only VM_PAT. We can toss mapping address from remap_pfn_range() into track_pfn_vma_new(), and collect all PAT-related logic together in arch/x86/. This patch also restores orignal frustration-free is_cow_mapping() check in remap_pfn_range(), as it was before commit v2.6.28-rc8-88-g3c8bb73 ("x86: PAT: store vm_pgoff for all linear_over_vma_region mappings - v3") is_linear_pfn_mapping() checks can be removed from mm/huge_memory.c, because it already handled by VM_PFNMAP in VM_NO_THP bit-mask. [suresh.b.siddha@intel.com: Reset the VM_PAT flag as part of untrack_pfn_vma()] Signed-off-by: Konstantin Khlebnikov <khlebnikov@openvz.org> Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com> Cc: Venkatesh Pallipadi <venki@google.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Ingo Molnar <mingo@redhat.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Carsten Otte <cotte@de.ibm.com> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Eric Paris <eparis@redhat.com> Cc: Hugh Dickins <hughd@google.com> Cc: James Morris <james.l.morris@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Kentaro Takeda <takedakn@nttdata.co.jp> Cc: Matt Helsley <matthltc@us.ibm.com> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Venkatesh Pallipadi <venki@google.com> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-08 23:28:34 +00:00
if (is_cow_mapping(vma->vm_flags)) {
if (addr != vma->vm_start || end != vma->vm_end)
return -EINVAL;
vma->vm_pgoff = pfn;
mm, x86, pat: rework linear pfn-mmap tracking Replace the generic vma-flag VM_PFN_AT_MMAP with x86-only VM_PAT. We can toss mapping address from remap_pfn_range() into track_pfn_vma_new(), and collect all PAT-related logic together in arch/x86/. This patch also restores orignal frustration-free is_cow_mapping() check in remap_pfn_range(), as it was before commit v2.6.28-rc8-88-g3c8bb73 ("x86: PAT: store vm_pgoff for all linear_over_vma_region mappings - v3") is_linear_pfn_mapping() checks can be removed from mm/huge_memory.c, because it already handled by VM_PFNMAP in VM_NO_THP bit-mask. [suresh.b.siddha@intel.com: Reset the VM_PAT flag as part of untrack_pfn_vma()] Signed-off-by: Konstantin Khlebnikov <khlebnikov@openvz.org> Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com> Cc: Venkatesh Pallipadi <venki@google.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Ingo Molnar <mingo@redhat.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Carsten Otte <cotte@de.ibm.com> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Eric Paris <eparis@redhat.com> Cc: Hugh Dickins <hughd@google.com> Cc: James Morris <james.l.morris@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Kentaro Takeda <takedakn@nttdata.co.jp> Cc: Matt Helsley <matthltc@us.ibm.com> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Venkatesh Pallipadi <venki@google.com> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-08 23:28:34 +00:00
}
err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size));
if (err)
x86: PAT: store vm_pgoff for all linear_over_vma_region mappings - v3 Impact: Code transformation, new functions added should have no effect. Drivers use mmap followed by pgprot_* and remap_pfn_range or vm_insert_pfn, in order to export reserved memory to userspace. Currently, such mappings are not tracked and hence not kept consistent with other mappings (/dev/mem, pci resource, ioremap) for the sme memory, that may exist in the system. The following patchset adds x86 PAT attribute tracking and untracking for pfnmap related APIs. First three patches in the patchset are changing the generic mm code to fit in this tracking. Last four patches are x86 specific to make things work with x86 PAT code. The patchset aso introduces pgprot_writecombine interface, which gives writecombine mapping when enabled, falling back to pgprot_noncached otherwise. This patch: While working on x86 PAT, we faced some hurdles with trackking remap_pfn_range() regions, as we do not have any information to say whether that PFNMAP mapping is linear for the entire vma range or it is smaller granularity regions within the vma. A simple solution to this is to use vm_pgoff as an indicator for linear mapping over the vma region. Currently, remap_pfn_range only sets vm_pgoff for COW mappings. Below patch changes the logic and sets the vm_pgoff irrespective of COW. This will still not be enough for the case where pfn is zero (vma region mapped to physical address zero). But, for all the other cases, we can look at pfnmap VMAs and say whether the mappng is for the entire vma region or not. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com> Signed-off-by: H. Peter Anvin <hpa@zytor.com>
2008-12-18 19:41:27 +00:00
return -EINVAL;
mm: kill vma flag VM_RESERVED and mm->reserved_vm counter A long time ago, in v2.4, VM_RESERVED kept swapout process off VMA, currently it lost original meaning but still has some effects: | effect | alternative flags -+------------------------+--------------------------------------------- 1| account as reserved_vm | VM_IO 2| skip in core dump | VM_IO, VM_DONTDUMP 3| do not merge or expand | VM_IO, VM_DONTEXPAND, VM_HUGETLB, VM_PFNMAP 4| do not mlock | VM_IO, VM_DONTEXPAND, VM_HUGETLB, VM_PFNMAP This patch removes reserved_vm counter from mm_struct. Seems like nobody cares about it, it does not exported into userspace directly, it only reduces total_vm showed in proc. Thus VM_RESERVED can be replaced with VM_IO or pair VM_DONTEXPAND | VM_DONTDUMP. remap_pfn_range() and io_remap_pfn_range() set VM_IO|VM_DONTEXPAND|VM_DONTDUMP. remap_vmalloc_range() set VM_DONTEXPAND | VM_DONTDUMP. [akpm@linux-foundation.org: drivers/vfio/pci/vfio_pci.c fixup] Signed-off-by: Konstantin Khlebnikov <khlebnikov@openvz.org> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Carsten Otte <cotte@de.ibm.com> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Eric Paris <eparis@redhat.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Hugh Dickins <hughd@google.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: James Morris <james.l.morris@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Kentaro Takeda <takedakn@nttdata.co.jp> Cc: Matt Helsley <matthltc@us.ibm.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> Cc: Suresh Siddha <suresh.b.siddha@intel.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Venkatesh Pallipadi <venki@google.com> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-08 23:29:02 +00:00
vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
BUG_ON(addr >= end);
pfn -= addr >> PAGE_SHIFT;
pgd = pgd_offset(mm, addr);
flush_cache_range(vma, addr, end);
do {
next = pgd_addr_end(addr, end);
err = remap_pud_range(mm, pgd, addr, next,
pfn + (addr >> PAGE_SHIFT), prot);
if (err)
break;
} while (pgd++, addr = next, addr != end);
if (err)
x86, pat: separate the pfn attribute tracking for remap_pfn_range and vm_insert_pfn With PAT enabled, vm_insert_pfn() looks up the existing pfn memory attribute and uses it. Expectation is that the driver reserves the memory attributes for the pfn before calling vm_insert_pfn(). remap_pfn_range() (when called for the whole vma) will setup a new attribute (based on the prot argument) for the specified pfn range. This addresses the legacy usage which typically calls remap_pfn_range() with a desired memory attribute. For ranges smaller than the vma size (which is typically not the case), remap_pfn_range() will use the existing memory attribute for the pfn range. Expose two different API's for these different behaviors. track_pfn_insert() for tracking the pfn attribute set by vm_insert_pfn() and track_pfn_remap() for the remap_pfn_range(). This cleanup also prepares the ground for the track/untrack pfn vma routines to take over the ownership of setting PAT specific vm_flag in the 'vma'. [khlebnikov@openvz.org: Clear checks in track_pfn_remap()] [akpm@linux-foundation.org: tweak a few comments] Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com> Signed-off-by: Konstantin Khlebnikov <khlebnikov@openvz.org> Cc: Venkatesh Pallipadi <venki@google.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Ingo Molnar <mingo@redhat.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Carsten Otte <cotte@de.ibm.com> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Eric Paris <eparis@redhat.com> Cc: Hugh Dickins <hughd@google.com> Cc: James Morris <james.l.morris@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Kentaro Takeda <takedakn@nttdata.co.jp> Cc: Konstantin Khlebnikov <khlebnikov@openvz.org> Cc: Matt Helsley <matthltc@us.ibm.com> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> Cc: Suresh Siddha <suresh.b.siddha@intel.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-08 23:28:29 +00:00
untrack_pfn(vma, pfn, PAGE_ALIGN(size));
return err;
}
EXPORT_SYMBOL(remap_pfn_range);
static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
unsigned long addr, unsigned long end,
pte_fn_t fn, void *data)
{
pte_t *pte;
int err;
CONFIG_HIGHPTE vs. sub-page page tables. Background: I've implemented 1K/2K page tables for s390. These sub-page page tables are required to properly support the s390 virtualization instruction with KVM. The SIE instruction requires that the page tables have 256 page table entries (pte) followed by 256 page status table entries (pgste). The pgstes are only required if the process is using the SIE instruction. The pgstes are updated by the hardware and by the hypervisor for a number of reasons, one of them is dirty and reference bit tracking. To avoid wasting memory the standard pte table allocation should return 1K/2K (31/64 bit) and 2K/4K if the process is using SIE. Problem: Page size on s390 is 4K, page table size is 1K or 2K. That means the s390 version for pte_alloc_one cannot return a pointer to a struct page. Trouble is that with the CONFIG_HIGHPTE feature on x86 pte_alloc_one cannot return a pointer to a pte either, since that would require more than 32 bit for the return value of pte_alloc_one (and the pte * would not be accessible since its not kmapped). Solution: The only solution I found to this dilemma is a new typedef: a pgtable_t. For s390 pgtable_t will be a (pte *) - to be introduced with a later patch. For everybody else it will be a (struct page *). The additional problem with the initialization of the ptl lock and the NR_PAGETABLE accounting is solved with a constructor pgtable_page_ctor and a destructor pgtable_page_dtor. The page table allocation and free functions need to call these two whenever a page table page is allocated or freed. pmd_populate will get a pgtable_t instead of a struct page pointer. To get the pgtable_t back from a pmd entry that has been installed with pmd_populate a new function pmd_pgtable is added. It replaces the pmd_page call in free_pte_range and apply_to_pte_range. Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:22:04 +00:00
pgtable_t token;
spinlock_t *uninitialized_var(ptl);
pte = (mm == &init_mm) ?
pte_alloc_kernel(pmd, addr) :
pte_alloc_map_lock(mm, pmd, addr, &ptl);
if (!pte)
return -ENOMEM;
BUG_ON(pmd_huge(*pmd));
arch_enter_lazy_mmu_mode();
CONFIG_HIGHPTE vs. sub-page page tables. Background: I've implemented 1K/2K page tables for s390. These sub-page page tables are required to properly support the s390 virtualization instruction with KVM. The SIE instruction requires that the page tables have 256 page table entries (pte) followed by 256 page status table entries (pgste). The pgstes are only required if the process is using the SIE instruction. The pgstes are updated by the hardware and by the hypervisor for a number of reasons, one of them is dirty and reference bit tracking. To avoid wasting memory the standard pte table allocation should return 1K/2K (31/64 bit) and 2K/4K if the process is using SIE. Problem: Page size on s390 is 4K, page table size is 1K or 2K. That means the s390 version for pte_alloc_one cannot return a pointer to a struct page. Trouble is that with the CONFIG_HIGHPTE feature on x86 pte_alloc_one cannot return a pointer to a pte either, since that would require more than 32 bit for the return value of pte_alloc_one (and the pte * would not be accessible since its not kmapped). Solution: The only solution I found to this dilemma is a new typedef: a pgtable_t. For s390 pgtable_t will be a (pte *) - to be introduced with a later patch. For everybody else it will be a (struct page *). The additional problem with the initialization of the ptl lock and the NR_PAGETABLE accounting is solved with a constructor pgtable_page_ctor and a destructor pgtable_page_dtor. The page table allocation and free functions need to call these two whenever a page table page is allocated or freed. pmd_populate will get a pgtable_t instead of a struct page pointer. To get the pgtable_t back from a pmd entry that has been installed with pmd_populate a new function pmd_pgtable is added. It replaces the pmd_page call in free_pte_range and apply_to_pte_range. Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:22:04 +00:00
token = pmd_pgtable(*pmd);
do {
err = fn(pte++, token, addr, data);
if (err)
break;
} while (addr += PAGE_SIZE, addr != end);
arch_leave_lazy_mmu_mode();
if (mm != &init_mm)
pte_unmap_unlock(pte-1, ptl);
return err;
}
static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
unsigned long addr, unsigned long end,
pte_fn_t fn, void *data)
{
pmd_t *pmd;
unsigned long next;
int err;
BUG_ON(pud_huge(*pud));
pmd = pmd_alloc(mm, pud, addr);
if (!pmd)
return -ENOMEM;
do {
next = pmd_addr_end(addr, end);
err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
if (err)
break;
} while (pmd++, addr = next, addr != end);
return err;
}
static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
unsigned long addr, unsigned long end,
pte_fn_t fn, void *data)
{
pud_t *pud;
unsigned long next;
int err;
pud = pud_alloc(mm, pgd, addr);
if (!pud)
return -ENOMEM;
do {
next = pud_addr_end(addr, end);
err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
if (err)
break;
} while (pud++, addr = next, addr != end);
return err;
}
/*
* Scan a region of virtual memory, filling in page tables as necessary
* and calling a provided function on each leaf page table.
*/
int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
unsigned long size, pte_fn_t fn, void *data)
{
pgd_t *pgd;
unsigned long next;
unsigned long end = addr + size;
int err;
BUG_ON(addr >= end);
pgd = pgd_offset(mm, addr);
do {
next = pgd_addr_end(addr, end);
err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
if (err)
break;
} while (pgd++, addr = next, addr != end);
return err;
}
EXPORT_SYMBOL_GPL(apply_to_page_range);
/*
* handle_pte_fault chooses page fault handler according to an entry
* which was read non-atomically. Before making any commitment, on
* those architectures or configurations (e.g. i386 with PAE) which
* might give a mix of unmatched parts, do_swap_page and do_nonlinear_fault
* must check under lock before unmapping the pte and proceeding
* (but do_wp_page is only called after already making such a check;
* and do_anonymous_page can safely check later on).
*/
[PATCH] mm: split page table lock Christoph Lameter demonstrated very poor scalability on the SGI 512-way, with a many-threaded application which concurrently initializes different parts of a large anonymous area. This patch corrects that, by using a separate spinlock per page table page, to guard the page table entries in that page, instead of using the mm's single page_table_lock. (But even then, page_table_lock is still used to guard page table allocation, and anon_vma allocation.) In this implementation, the spinlock is tucked inside the struct page of the page table page: with a BUILD_BUG_ON in case it overflows - which it would in the case of 32-bit PA-RISC with spinlock debugging enabled. Splitting the lock is not quite for free: another cacheline access. Ideally, I suppose we would use split ptlock only for multi-threaded processes on multi-cpu machines; but deciding that dynamically would have its own costs. So for now enable it by config, at some number of cpus - since the Kconfig language doesn't support inequalities, let preprocessor compare that with NR_CPUS. But I don't think it's worth being user-configurable: for good testing of both split and unsplit configs, split now at 4 cpus, and perhaps change that to 8 later. There is a benefit even for singly threaded processes: kswapd can be attacking one part of the mm while another part is busy faulting. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:40 +00:00
static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
pte_t *page_table, pte_t orig_pte)
{
int same = 1;
#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
if (sizeof(pte_t) > sizeof(unsigned long)) {
[PATCH] mm: split page table lock Christoph Lameter demonstrated very poor scalability on the SGI 512-way, with a many-threaded application which concurrently initializes different parts of a large anonymous area. This patch corrects that, by using a separate spinlock per page table page, to guard the page table entries in that page, instead of using the mm's single page_table_lock. (But even then, page_table_lock is still used to guard page table allocation, and anon_vma allocation.) In this implementation, the spinlock is tucked inside the struct page of the page table page: with a BUILD_BUG_ON in case it overflows - which it would in the case of 32-bit PA-RISC with spinlock debugging enabled. Splitting the lock is not quite for free: another cacheline access. Ideally, I suppose we would use split ptlock only for multi-threaded processes on multi-cpu machines; but deciding that dynamically would have its own costs. So for now enable it by config, at some number of cpus - since the Kconfig language doesn't support inequalities, let preprocessor compare that with NR_CPUS. But I don't think it's worth being user-configurable: for good testing of both split and unsplit configs, split now at 4 cpus, and perhaps change that to 8 later. There is a benefit even for singly threaded processes: kswapd can be attacking one part of the mm while another part is busy faulting. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:40 +00:00
spinlock_t *ptl = pte_lockptr(mm, pmd);
spin_lock(ptl);
same = pte_same(*page_table, orig_pte);
[PATCH] mm: split page table lock Christoph Lameter demonstrated very poor scalability on the SGI 512-way, with a many-threaded application which concurrently initializes different parts of a large anonymous area. This patch corrects that, by using a separate spinlock per page table page, to guard the page table entries in that page, instead of using the mm's single page_table_lock. (But even then, page_table_lock is still used to guard page table allocation, and anon_vma allocation.) In this implementation, the spinlock is tucked inside the struct page of the page table page: with a BUILD_BUG_ON in case it overflows - which it would in the case of 32-bit PA-RISC with spinlock debugging enabled. Splitting the lock is not quite for free: another cacheline access. Ideally, I suppose we would use split ptlock only for multi-threaded processes on multi-cpu machines; but deciding that dynamically would have its own costs. So for now enable it by config, at some number of cpus - since the Kconfig language doesn't support inequalities, let preprocessor compare that with NR_CPUS. But I don't think it's worth being user-configurable: for good testing of both split and unsplit configs, split now at 4 cpus, and perhaps change that to 8 later. There is a benefit even for singly threaded processes: kswapd can be attacking one part of the mm while another part is busy faulting. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:40 +00:00
spin_unlock(ptl);
}
#endif
pte_unmap(page_table);
return same;
}
static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma)
{
/*
* If the source page was a PFN mapping, we don't have
* a "struct page" for it. We do a best-effort copy by
* just copying from the original user address. If that
* fails, we just zero-fill it. Live with it.
*/
if (unlikely(!src)) {
void *kaddr = kmap_atomic(dst);
void __user *uaddr = (void __user *)(va & PAGE_MASK);
/*
* This really shouldn't fail, because the page is there
* in the page tables. But it might just be unreadable,
* in which case we just give up and fill the result with
* zeroes.
*/
if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
clear_page(kaddr);
kunmap_atomic(kaddr);
flush_dcache_page(dst);
mm: fix PageUptodate data race After running SetPageUptodate, preceeding stores to the page contents to actually bring it uptodate may not be ordered with the store to set the page uptodate. Therefore, another CPU which checks PageUptodate is true, then reads the page contents can get stale data. Fix this by having an smp_wmb before SetPageUptodate, and smp_rmb after PageUptodate. Many places that test PageUptodate, do so with the page locked, and this would be enough to ensure memory ordering in those places if SetPageUptodate were only called while the page is locked. Unfortunately that is not always the case for some filesystems, but it could be an idea for the future. Also bring the handling of anonymous page uptodateness in line with that of file backed page management, by marking anon pages as uptodate when they _are_ uptodate, rather than when our implementation requires that they be marked as such. Doing allows us to get rid of the smp_wmb's in the page copying functions, which were especially added for anonymous pages for an analogous memory ordering problem. Both file and anonymous pages are handled with the same barriers. FAQ: Q. Why not do this in flush_dcache_page? A. Firstly, flush_dcache_page handles only one side (the smb side) of the ordering protocol; we'd still need smp_rmb somewhere. Secondly, hiding away memory barriers in a completely unrelated function is nasty; at least in the PageUptodate macros, they are located together with (half) the operations involved in the ordering. Thirdly, the smp_wmb is only required when first bringing the page uptodate, wheras flush_dcache_page should be called each time it is written to through the kernel mapping. It is logically the wrong place to put it. Q. Why does this increase my text size / reduce my performance / etc. A. Because it is adding the necessary instructions to eliminate the data-race. Q. Can it be improved? A. Yes, eg. if you were to create a rule that all SetPageUptodate operations run under the page lock, we could avoid the smp_rmb places where PageUptodate is queried under the page lock. Requires audit of all filesystems and at least some would need reworking. That's great you're interested, I'm eagerly awaiting your patches. Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 06:29:34 +00:00
} else
copy_user_highpage(dst, src, va, vma);
}
/*
* This routine handles present pages, when users try to write
* to a shared page. It is done by copying the page to a new address
* and decrementing the shared-page counter for the old page.
*
* Note that this routine assumes that the protection checks have been
* done by the caller (the low-level page fault routine in most cases).
* Thus we can safely just mark it writable once we've done any necessary
* COW.
*
* We also mark the page dirty at this point even though the page will
* change only once the write actually happens. This avoids a few races,
* and potentially makes it more efficient.
*
* We enter with non-exclusive mmap_sem (to exclude vma changes,
* but allow concurrent faults), with pte both mapped and locked.
* We return with mmap_sem still held, but pte unmapped and unlocked.
*/
static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
unsigned long address, pte_t *page_table, pmd_t *pmd,
spinlock_t *ptl, pte_t orig_pte)
__releases(ptl)
{
mm: move all mmu notifier invocations to be done outside the PT lock In order to allow sleeping during mmu notifier calls, we need to avoid invoking them under the page table spinlock. This patch solves the problem by calling invalidate_page notification after releasing the lock (but before freeing the page itself), or by wrapping the page invalidation with calls to invalidate_range_begin and invalidate_range_end. To prevent accidental changes to the invalidate_range_end arguments after the call to invalidate_range_begin, the patch introduces a convention of saving the arguments in consistently named locals: unsigned long mmun_start; /* For mmu_notifiers */ unsigned long mmun_end; /* For mmu_notifiers */ ... mmun_start = ... mmun_end = ... mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); ... mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); The patch changes code to use this convention for all calls to mmu_notifier_invalidate_range_start/end, except those where the calls are close enough so that anyone who glances at the code can see the values aren't changing. This patchset is a preliminary step towards on-demand paging design to be added to the RDMA stack. Why do we want on-demand paging for Infiniband? Applications register memory with an RDMA adapter using system calls, and subsequently post IO operations that refer to the corresponding virtual addresses directly to HW. Until now, this was achieved by pinning the memory during the registration calls. The goal of on demand paging is to avoid pinning the pages of registered memory regions (MRs). This will allow users the same flexibility they get when swapping any other part of their processes address spaces. Instead of requiring the entire MR to fit in physical memory, we can allow the MR to be larger, and only fit the current working set in physical memory. Why should anyone care? What problems are users currently experiencing? This can make programming with RDMA much simpler. Today, developers that are working with more data than their RAM can hold need either to deregister and reregister memory regions throughout their process's life, or keep a single memory region and copy the data to it. On demand paging will allow these developers to register a single MR at the beginning of their process's life, and let the operating system manage which pages needs to be fetched at a given time. In the future, we might be able to provide a single memory access key for each process that would provide the entire process's address as one large memory region, and the developers wouldn't need to register memory regions at all. Is there any prospect that any other subsystems will utilise these infrastructural changes? If so, which and how, etc? As for other subsystems, I understand that XPMEM wanted to sleep in MMU notifiers, as Christoph Lameter wrote at http://lkml.indiana.edu/hypermail/linux/kernel/0802.1/0460.html and perhaps Andrea knows about other use cases. Scheduling in mmu notifications is required since we need to sync the hardware with the secondary page tables change. A TLB flush of an IO device is inherently slower than a CPU TLB flush, so our design works by sending the invalidation request to the device, and waiting for an interrupt before exiting the mmu notifier handler. Avi said: kvm may be a buyer. kvm::mmu_lock, which serializes guest page faults, also protects long operations such as destroying large ranges. It would be good to convert it into a spinlock, but as it is used inside mmu notifiers, this cannot be done. (there are alternatives, such as keeping the spinlock and using a generation counter to do the teardown in O(1), which is what the "may" is doing up there). [akpm@linux-foundation.orgpossible speed tweak in hugetlb_cow(), cleanups] Signed-off-by: Andrea Arcangeli <andrea@qumranet.com> Signed-off-by: Sagi Grimberg <sagig@mellanox.com> Signed-off-by: Haggai Eran <haggaie@mellanox.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com> Cc: Or Gerlitz <ogerlitz@mellanox.com> Cc: Haggai Eran <haggaie@mellanox.com> Cc: Shachar Raindel <raindel@mellanox.com> Cc: Liran Liss <liranl@mellanox.com> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Avi Kivity <avi@redhat.com> Cc: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-08 23:33:33 +00:00
struct page *old_page, *new_page = NULL;
pte_t entry;
do_wp_page: remove the 'reuse' flag mlocking a shared, writable vma currently causes the corresponding pages to be marked as dirty and queued for writeback. This seems rather unnecessary given that the pages are not being actually modified during mlock. It is understood that for non-shared mappings (file or anon) we want to use a write fault in order to break COW, but there is just no such need for shared mappings. The first two patches in this series do not introduce any behavior change. The intent there is to make it obvious that dirtying file pages is only done in the (writable, shared) case. I think this clarifies the code, but I wouldn't mind dropping these two patches if there is no consensus about them. The last patch is where we actually avoid dirtying shared mappings during mlock. Note that as a side effect of this, we won't call page_mkwrite() for the mappings that define it, and won't be pre-allocating data blocks at the FS level if the mapped file was sparsely allocated. My understanding is that mlock does not need to provide such guarantee, as evidenced by the fact that it never did for the filesystems that don't define page_mkwrite() - including some common ones like ext3. However, I would like to gather feedback on this from filesystem people as a precaution. If this turns out to be a showstopper, maybe block preallocation can be added back on using a different interface. Large shared mlocks are getting significantly (>2x) faster in my tests, as the disk can be fully used for reading the file instead of having to share between this and writeback. This patch: Reorganize the code to remove the 'reuse' flag. No behavior changes. Signed-off-by: Michel Lespinasse <walken@google.com> Cc: Hugh Dickins <hughd@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Kosaki Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Theodore Tso <tytso@google.com> Cc: Michael Rubin <mrubin@google.com> Cc: Suleiman Souhlal <suleiman@google.com> Cc: Dave Chinner <david@fromorbit.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-13 23:46:07 +00:00
int ret = 0;
int page_mkwrite = 0;
[PATCH] mm: tracking shared dirty pages Tracking of dirty pages in shared writeable mmap()s. The idea is simple: write protect clean shared writeable pages, catch the write-fault, make writeable and set dirty. On page write-back clean all the PTE dirty bits and write protect them once again. The implementation is a tad harder, mainly because the default backing_dev_info capabilities were too loosely maintained. Hence it is not enough to test the backing_dev_info for cap_account_dirty. The current heuristic is as follows, a VMA is eligible when: - its shared writeable (vm_flags & (VM_WRITE|VM_SHARED)) == (VM_WRITE|VM_SHARED) - it is not a 'special' mapping (vm_flags & (VM_PFNMAP|VM_INSERTPAGE)) == 0 - the backing_dev_info is cap_account_dirty mapping_cap_account_dirty(vma->vm_file->f_mapping) - f_op->mmap() didn't change the default page protection Page from remap_pfn_range() are explicitly excluded because their COW semantics are already horrid enough (see vm_normal_page() in do_wp_page()) and because they don't have a backing store anyway. mprotect() is taught about the new behaviour as well. However it overrides the last condition. Cleaning the pages on write-back is done with page_mkclean() a new rmap call. It can be called on any page, but is currently only implemented for mapped pages, if the page is found the be of a VMA that accounts dirty pages it will also wrprotect the PTE. Finally, in fs/buffers.c:try_to_free_buffers(); remove clear_page_dirty() from under ->private_lock. This seems to be safe, since ->private_lock is used to serialize access to the buffers, not the page itself. This is needed because clear_page_dirty() will call into page_mkclean() and would thereby violate locking order. [dhowells@redhat.com: Provide a page_mkclean() implementation for NOMMU] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-26 06:30:57 +00:00
struct page *dirty_page = NULL;
unsigned long mmun_start = 0; /* For mmu_notifiers */
unsigned long mmun_end = 0; /* For mmu_notifiers */
old_page = vm_normal_page(vma, address, orig_pte);
mm: dirty page accounting vs VM_MIXEDMAP Dirty page accounting accurately measures the amound of dirty pages in writable shared mappings by mapping the pages RO (as indicated by vma_wants_writenotify). We then trap on first write and call set_page_dirty() on the page, after which we map the page RW and continue execution. When we launder dirty pages, we call clear_page_dirty_for_io() which clears both the dirty flag, and maps the page RO again before we start writeout so that the story can repeat itself. vma_wants_writenotify() excludes VM_PFNMAP on the basis that we cannot do the regular dirty page stuff on raw PFNs and the memory isn't going anywhere anyway. The recently introduced VM_MIXEDMAP mixes both !pfn_valid() and pfn_valid() pages in a single mapping. We can't do dirty page accounting on !pfn_valid() pages as stated above, and mapping them RO causes them to be COW'ed on write, which breaks VM_SHARED semantics. Excluding VM_MIXEDMAP in vma_wants_writenotify() would mean we don't do the regular dirty page accounting for the pfn_valid() pages, which would bring back all the head-aches from inaccurate dirty page accounting. So instead, we let the !pfn_valid() pages get mapped RO, but fix them up unconditionally in the fault path. Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Acked-by: Hugh Dickins <hugh@veritas.com> Cc: "Jared Hulbert" <jaredeh@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-04 16:59:24 +00:00
if (!old_page) {
/*
* VM_MIXEDMAP !pfn_valid() case
*
* We should not cow pages in a shared writeable mapping.
* Just mark the pages writable as we can't do any dirty
* accounting on raw pfn maps.
*/
if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
(VM_WRITE|VM_SHARED))
goto reuse;
goto gotten;
mm: dirty page accounting vs VM_MIXEDMAP Dirty page accounting accurately measures the amound of dirty pages in writable shared mappings by mapping the pages RO (as indicated by vma_wants_writenotify). We then trap on first write and call set_page_dirty() on the page, after which we map the page RW and continue execution. When we launder dirty pages, we call clear_page_dirty_for_io() which clears both the dirty flag, and maps the page RO again before we start writeout so that the story can repeat itself. vma_wants_writenotify() excludes VM_PFNMAP on the basis that we cannot do the regular dirty page stuff on raw PFNs and the memory isn't going anywhere anyway. The recently introduced VM_MIXEDMAP mixes both !pfn_valid() and pfn_valid() pages in a single mapping. We can't do dirty page accounting on !pfn_valid() pages as stated above, and mapping them RO causes them to be COW'ed on write, which breaks VM_SHARED semantics. Excluding VM_MIXEDMAP in vma_wants_writenotify() would mean we don't do the regular dirty page accounting for the pfn_valid() pages, which would bring back all the head-aches from inaccurate dirty page accounting. So instead, we let the !pfn_valid() pages get mapped RO, but fix them up unconditionally in the fault path. Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Acked-by: Hugh Dickins <hugh@veritas.com> Cc: "Jared Hulbert" <jaredeh@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-04 16:59:24 +00:00
}
[PATCH] mm: tracking shared dirty pages Tracking of dirty pages in shared writeable mmap()s. The idea is simple: write protect clean shared writeable pages, catch the write-fault, make writeable and set dirty. On page write-back clean all the PTE dirty bits and write protect them once again. The implementation is a tad harder, mainly because the default backing_dev_info capabilities were too loosely maintained. Hence it is not enough to test the backing_dev_info for cap_account_dirty. The current heuristic is as follows, a VMA is eligible when: - its shared writeable (vm_flags & (VM_WRITE|VM_SHARED)) == (VM_WRITE|VM_SHARED) - it is not a 'special' mapping (vm_flags & (VM_PFNMAP|VM_INSERTPAGE)) == 0 - the backing_dev_info is cap_account_dirty mapping_cap_account_dirty(vma->vm_file->f_mapping) - f_op->mmap() didn't change the default page protection Page from remap_pfn_range() are explicitly excluded because their COW semantics are already horrid enough (see vm_normal_page() in do_wp_page()) and because they don't have a backing store anyway. mprotect() is taught about the new behaviour as well. However it overrides the last condition. Cleaning the pages on write-back is done with page_mkclean() a new rmap call. It can be called on any page, but is currently only implemented for mapped pages, if the page is found the be of a VMA that accounts dirty pages it will also wrprotect the PTE. Finally, in fs/buffers.c:try_to_free_buffers(); remove clear_page_dirty() from under ->private_lock. This seems to be safe, since ->private_lock is used to serialize access to the buffers, not the page itself. This is needed because clear_page_dirty() will call into page_mkclean() and would thereby violate locking order. [dhowells@redhat.com: Provide a page_mkclean() implementation for NOMMU] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-26 06:30:57 +00:00
/*
* Take out anonymous pages first, anonymous shared vmas are
* not dirty accountable.
[PATCH] mm: tracking shared dirty pages Tracking of dirty pages in shared writeable mmap()s. The idea is simple: write protect clean shared writeable pages, catch the write-fault, make writeable and set dirty. On page write-back clean all the PTE dirty bits and write protect them once again. The implementation is a tad harder, mainly because the default backing_dev_info capabilities were too loosely maintained. Hence it is not enough to test the backing_dev_info for cap_account_dirty. The current heuristic is as follows, a VMA is eligible when: - its shared writeable (vm_flags & (VM_WRITE|VM_SHARED)) == (VM_WRITE|VM_SHARED) - it is not a 'special' mapping (vm_flags & (VM_PFNMAP|VM_INSERTPAGE)) == 0 - the backing_dev_info is cap_account_dirty mapping_cap_account_dirty(vma->vm_file->f_mapping) - f_op->mmap() didn't change the default page protection Page from remap_pfn_range() are explicitly excluded because their COW semantics are already horrid enough (see vm_normal_page() in do_wp_page()) and because they don't have a backing store anyway. mprotect() is taught about the new behaviour as well. However it overrides the last condition. Cleaning the pages on write-back is done with page_mkclean() a new rmap call. It can be called on any page, but is currently only implemented for mapped pages, if the page is found the be of a VMA that accounts dirty pages it will also wrprotect the PTE. Finally, in fs/buffers.c:try_to_free_buffers(); remove clear_page_dirty() from under ->private_lock. This seems to be safe, since ->private_lock is used to serialize access to the buffers, not the page itself. This is needed because clear_page_dirty() will call into page_mkclean() and would thereby violate locking order. [dhowells@redhat.com: Provide a page_mkclean() implementation for NOMMU] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-26 06:30:57 +00:00
*/
if (PageAnon(old_page) && !PageKsm(old_page)) {
if (!trylock_page(old_page)) {
page_cache_get(old_page);
pte_unmap_unlock(page_table, ptl);
lock_page(old_page);
page_table = pte_offset_map_lock(mm, pmd, address,
&ptl);
if (!pte_same(*page_table, orig_pte)) {
unlock_page(old_page);
goto unlock;
}
page_cache_release(old_page);
}
do_wp_page: remove the 'reuse' flag mlocking a shared, writable vma currently causes the corresponding pages to be marked as dirty and queued for writeback. This seems rather unnecessary given that the pages are not being actually modified during mlock. It is understood that for non-shared mappings (file or anon) we want to use a write fault in order to break COW, but there is just no such need for shared mappings. The first two patches in this series do not introduce any behavior change. The intent there is to make it obvious that dirtying file pages is only done in the (writable, shared) case. I think this clarifies the code, but I wouldn't mind dropping these two patches if there is no consensus about them. The last patch is where we actually avoid dirtying shared mappings during mlock. Note that as a side effect of this, we won't call page_mkwrite() for the mappings that define it, and won't be pre-allocating data blocks at the FS level if the mapped file was sparsely allocated. My understanding is that mlock does not need to provide such guarantee, as evidenced by the fact that it never did for the filesystems that don't define page_mkwrite() - including some common ones like ext3. However, I would like to gather feedback on this from filesystem people as a precaution. If this turns out to be a showstopper, maybe block preallocation can be added back on using a different interface. Large shared mlocks are getting significantly (>2x) faster in my tests, as the disk can be fully used for reading the file instead of having to share between this and writeback. This patch: Reorganize the code to remove the 'reuse' flag. No behavior changes. Signed-off-by: Michel Lespinasse <walken@google.com> Cc: Hugh Dickins <hughd@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Kosaki Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Theodore Tso <tytso@google.com> Cc: Michael Rubin <mrubin@google.com> Cc: Suleiman Souhlal <suleiman@google.com> Cc: Dave Chinner <david@fromorbit.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-13 23:46:07 +00:00
if (reuse_swap_page(old_page)) {
/*
* The page is all ours. Move it to our anon_vma so
* the rmap code will not search our parent or siblings.
* Protected against the rmap code by the page lock.
*/
page_move_anon_rmap(old_page, vma, address);
do_wp_page: remove the 'reuse' flag mlocking a shared, writable vma currently causes the corresponding pages to be marked as dirty and queued for writeback. This seems rather unnecessary given that the pages are not being actually modified during mlock. It is understood that for non-shared mappings (file or anon) we want to use a write fault in order to break COW, but there is just no such need for shared mappings. The first two patches in this series do not introduce any behavior change. The intent there is to make it obvious that dirtying file pages is only done in the (writable, shared) case. I think this clarifies the code, but I wouldn't mind dropping these two patches if there is no consensus about them. The last patch is where we actually avoid dirtying shared mappings during mlock. Note that as a side effect of this, we won't call page_mkwrite() for the mappings that define it, and won't be pre-allocating data blocks at the FS level if the mapped file was sparsely allocated. My understanding is that mlock does not need to provide such guarantee, as evidenced by the fact that it never did for the filesystems that don't define page_mkwrite() - including some common ones like ext3. However, I would like to gather feedback on this from filesystem people as a precaution. If this turns out to be a showstopper, maybe block preallocation can be added back on using a different interface. Large shared mlocks are getting significantly (>2x) faster in my tests, as the disk can be fully used for reading the file instead of having to share between this and writeback. This patch: Reorganize the code to remove the 'reuse' flag. No behavior changes. Signed-off-by: Michel Lespinasse <walken@google.com> Cc: Hugh Dickins <hughd@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Kosaki Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Theodore Tso <tytso@google.com> Cc: Michael Rubin <mrubin@google.com> Cc: Suleiman Souhlal <suleiman@google.com> Cc: Dave Chinner <david@fromorbit.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-13 23:46:07 +00:00
unlock_page(old_page);
goto reuse;
}
unlock_page(old_page);
} else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
[PATCH] mm: tracking shared dirty pages Tracking of dirty pages in shared writeable mmap()s. The idea is simple: write protect clean shared writeable pages, catch the write-fault, make writeable and set dirty. On page write-back clean all the PTE dirty bits and write protect them once again. The implementation is a tad harder, mainly because the default backing_dev_info capabilities were too loosely maintained. Hence it is not enough to test the backing_dev_info for cap_account_dirty. The current heuristic is as follows, a VMA is eligible when: - its shared writeable (vm_flags & (VM_WRITE|VM_SHARED)) == (VM_WRITE|VM_SHARED) - it is not a 'special' mapping (vm_flags & (VM_PFNMAP|VM_INSERTPAGE)) == 0 - the backing_dev_info is cap_account_dirty mapping_cap_account_dirty(vma->vm_file->f_mapping) - f_op->mmap() didn't change the default page protection Page from remap_pfn_range() are explicitly excluded because their COW semantics are already horrid enough (see vm_normal_page() in do_wp_page()) and because they don't have a backing store anyway. mprotect() is taught about the new behaviour as well. However it overrides the last condition. Cleaning the pages on write-back is done with page_mkclean() a new rmap call. It can be called on any page, but is currently only implemented for mapped pages, if the page is found the be of a VMA that accounts dirty pages it will also wrprotect the PTE. Finally, in fs/buffers.c:try_to_free_buffers(); remove clear_page_dirty() from under ->private_lock. This seems to be safe, since ->private_lock is used to serialize access to the buffers, not the page itself. This is needed because clear_page_dirty() will call into page_mkclean() and would thereby violate locking order. [dhowells@redhat.com: Provide a page_mkclean() implementation for NOMMU] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-26 06:30:57 +00:00
(VM_WRITE|VM_SHARED))) {
/*
* Only catch write-faults on shared writable pages,
* read-only shared pages can get COWed by
* get_user_pages(.write=1, .force=1).
*/
[PATCH] add page_mkwrite() vm_operations method Add a new VMA operation to notify a filesystem or other driver about the MMU generating a fault because userspace attempted to write to a page mapped through a read-only PTE. This facility permits the filesystem or driver to: (*) Implement storage allocation/reservation on attempted write, and so to deal with problems such as ENOSPC more gracefully (perhaps by generating SIGBUS). (*) Delay making the page writable until the contents have been written to a backing cache. This is useful for NFS/AFS when using FS-Cache/CacheFS. It permits the filesystem to have some guarantee about the state of the cache. (*) Account and limit number of dirty pages. This is one piece of the puzzle needed to make shared writable mapping work safely in FUSE. Needed by cachefs (Or is it cachefiles? Or fscache? <head spins>). At least four other groups have stated an interest in it or a desire to use the functionality it provides: FUSE, OCFS2, NTFS and JFFS2. Also, things like EXT3 really ought to use it to deal with the case of shared-writable mmap encountering ENOSPC before we permit the page to be dirtied. From: Peter Zijlstra <a.p.zijlstra@chello.nl> get_user_pages(.write=1, .force=1) can generate COW hits on read-only shared mappings, this patch traps those as mkpage_write candidates and fails to handle them the old way. Signed-off-by: David Howells <dhowells@redhat.com> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Joel Becker <Joel.Becker@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Cc: Anton Altaparmakov <aia21@cantab.net> Cc: David Woodhouse <dwmw2@infradead.org> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 09:03:43 +00:00
if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
struct vm_fault vmf;
int tmp;
vmf.virtual_address = (void __user *)(address &
PAGE_MASK);
vmf.pgoff = old_page->index;
vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
vmf.page = old_page;
[PATCH] add page_mkwrite() vm_operations method Add a new VMA operation to notify a filesystem or other driver about the MMU generating a fault because userspace attempted to write to a page mapped through a read-only PTE. This facility permits the filesystem or driver to: (*) Implement storage allocation/reservation on attempted write, and so to deal with problems such as ENOSPC more gracefully (perhaps by generating SIGBUS). (*) Delay making the page writable until the contents have been written to a backing cache. This is useful for NFS/AFS when using FS-Cache/CacheFS. It permits the filesystem to have some guarantee about the state of the cache. (*) Account and limit number of dirty pages. This is one piece of the puzzle needed to make shared writable mapping work safely in FUSE. Needed by cachefs (Or is it cachefiles? Or fscache? <head spins>). At least four other groups have stated an interest in it or a desire to use the functionality it provides: FUSE, OCFS2, NTFS and JFFS2. Also, things like EXT3 really ought to use it to deal with the case of shared-writable mmap encountering ENOSPC before we permit the page to be dirtied. From: Peter Zijlstra <a.p.zijlstra@chello.nl> get_user_pages(.write=1, .force=1) can generate COW hits on read-only shared mappings, this patch traps those as mkpage_write candidates and fails to handle them the old way. Signed-off-by: David Howells <dhowells@redhat.com> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Joel Becker <Joel.Becker@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Cc: Anton Altaparmakov <aia21@cantab.net> Cc: David Woodhouse <dwmw2@infradead.org> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 09:03:43 +00:00
/*
* Notify the address space that the page is about to
* become writable so that it can prohibit this or wait
* for the page to get into an appropriate state.
*
* We do this without the lock held, so that it can
* sleep if it needs to.
*/
page_cache_get(old_page);
pte_unmap_unlock(page_table, ptl);
tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
if (unlikely(tmp &
(VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
ret = tmp;
[PATCH] add page_mkwrite() vm_operations method Add a new VMA operation to notify a filesystem or other driver about the MMU generating a fault because userspace attempted to write to a page mapped through a read-only PTE. This facility permits the filesystem or driver to: (*) Implement storage allocation/reservation on attempted write, and so to deal with problems such as ENOSPC more gracefully (perhaps by generating SIGBUS). (*) Delay making the page writable until the contents have been written to a backing cache. This is useful for NFS/AFS when using FS-Cache/CacheFS. It permits the filesystem to have some guarantee about the state of the cache. (*) Account and limit number of dirty pages. This is one piece of the puzzle needed to make shared writable mapping work safely in FUSE. Needed by cachefs (Or is it cachefiles? Or fscache? <head spins>). At least four other groups have stated an interest in it or a desire to use the functionality it provides: FUSE, OCFS2, NTFS and JFFS2. Also, things like EXT3 really ought to use it to deal with the case of shared-writable mmap encountering ENOSPC before we permit the page to be dirtied. From: Peter Zijlstra <a.p.zijlstra@chello.nl> get_user_pages(.write=1, .force=1) can generate COW hits on read-only shared mappings, this patch traps those as mkpage_write candidates and fails to handle them the old way. Signed-off-by: David Howells <dhowells@redhat.com> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Joel Becker <Joel.Becker@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Cc: Anton Altaparmakov <aia21@cantab.net> Cc: David Woodhouse <dwmw2@infradead.org> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 09:03:43 +00:00
goto unwritable_page;
}
mm: close page_mkwrite races Change page_mkwrite to allow implementations to return with the page locked, and also change it's callers (in page fault paths) to hold the lock until the page is marked dirty. This allows the filesystem to have full control of page dirtying events coming from the VM. Rather than simply hold the page locked over the page_mkwrite call, we call page_mkwrite with the page unlocked and allow callers to return with it locked, so filesystems can avoid LOR conditions with page lock. The problem with the current scheme is this: a filesystem that wants to associate some metadata with a page as long as the page is dirty, will perform this manipulation in its ->page_mkwrite. It currently then must return with the page unlocked and may not hold any other locks (according to existing page_mkwrite convention). In this window, the VM could write out the page, clearing page-dirty. The filesystem has no good way to detect that a dirty pte is about to be attached, so it will happily write out the page, at which point, the filesystem may manipulate the metadata to reflect that the page is no longer dirty. It is not always possible to perform the required metadata manipulation in ->set_page_dirty, because that function cannot block or fail. The filesystem may need to allocate some data structure, for example. And the VM cannot mark the pte dirty before page_mkwrite, because page_mkwrite is allowed to fail, so we must not allow any window where the page could be written to if page_mkwrite does fail. This solution of holding the page locked over the 3 critical operations (page_mkwrite, setting the pte dirty, and finally setting the page dirty) closes out races nicely, preventing page cleaning for writeout being initiated in that window. This provides the filesystem with a strong synchronisation against the VM here. - Sage needs this race closed for ceph filesystem. - Trond for NFS (http://bugzilla.kernel.org/show_bug.cgi?id=12913). - I need it for fsblock. - I suspect other filesystems may need it too (eg. btrfs). - I have converted buffer.c to the new locking. Even simple block allocation under dirty pages might be susceptible to i_size changing under partial page at the end of file (we also have a buffer.c-side problem here, but it cannot be fixed properly without this patch). - Other filesystems (eg. NFS, maybe btrfs) will need to change their page_mkwrite functions themselves. [ This also moves page_mkwrite another step closer to fault, which should eventually allow page_mkwrite to be moved into ->fault, and thus avoiding a filesystem calldown and page lock/unlock cycle in __do_fault. ] [akpm@linux-foundation.org: fix derefs of NULL ->mapping] Cc: Sage Weil <sage@newdream.net> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Valdis Kletnieks <Valdis.Kletnieks@vt.edu> Cc: <stable@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-30 22:08:16 +00:00
if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
lock_page(old_page);
if (!old_page->mapping) {
ret = 0; /* retry the fault */
unlock_page(old_page);
goto unwritable_page;
}
} else
VM_BUG_ON(!PageLocked(old_page));
[PATCH] add page_mkwrite() vm_operations method Add a new VMA operation to notify a filesystem or other driver about the MMU generating a fault because userspace attempted to write to a page mapped through a read-only PTE. This facility permits the filesystem or driver to: (*) Implement storage allocation/reservation on attempted write, and so to deal with problems such as ENOSPC more gracefully (perhaps by generating SIGBUS). (*) Delay making the page writable until the contents have been written to a backing cache. This is useful for NFS/AFS when using FS-Cache/CacheFS. It permits the filesystem to have some guarantee about the state of the cache. (*) Account and limit number of dirty pages. This is one piece of the puzzle needed to make shared writable mapping work safely in FUSE. Needed by cachefs (Or is it cachefiles? Or fscache? <head spins>). At least four other groups have stated an interest in it or a desire to use the functionality it provides: FUSE, OCFS2, NTFS and JFFS2. Also, things like EXT3 really ought to use it to deal with the case of shared-writable mmap encountering ENOSPC before we permit the page to be dirtied. From: Peter Zijlstra <a.p.zijlstra@chello.nl> get_user_pages(.write=1, .force=1) can generate COW hits on read-only shared mappings, this patch traps those as mkpage_write candidates and fails to handle them the old way. Signed-off-by: David Howells <dhowells@redhat.com> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Joel Becker <Joel.Becker@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Cc: Anton Altaparmakov <aia21@cantab.net> Cc: David Woodhouse <dwmw2@infradead.org> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 09:03:43 +00:00
/*
* Since we dropped the lock we need to revalidate
* the PTE as someone else may have changed it. If
* they did, we just return, as we can count on the
* MMU to tell us if they didn't also make it writable.
*/
page_table = pte_offset_map_lock(mm, pmd, address,
&ptl);
mm: close page_mkwrite races Change page_mkwrite to allow implementations to return with the page locked, and also change it's callers (in page fault paths) to hold the lock until the page is marked dirty. This allows the filesystem to have full control of page dirtying events coming from the VM. Rather than simply hold the page locked over the page_mkwrite call, we call page_mkwrite with the page unlocked and allow callers to return with it locked, so filesystems can avoid LOR conditions with page lock. The problem with the current scheme is this: a filesystem that wants to associate some metadata with a page as long as the page is dirty, will perform this manipulation in its ->page_mkwrite. It currently then must return with the page unlocked and may not hold any other locks (according to existing page_mkwrite convention). In this window, the VM could write out the page, clearing page-dirty. The filesystem has no good way to detect that a dirty pte is about to be attached, so it will happily write out the page, at which point, the filesystem may manipulate the metadata to reflect that the page is no longer dirty. It is not always possible to perform the required metadata manipulation in ->set_page_dirty, because that function cannot block or fail. The filesystem may need to allocate some data structure, for example. And the VM cannot mark the pte dirty before page_mkwrite, because page_mkwrite is allowed to fail, so we must not allow any window where the page could be written to if page_mkwrite does fail. This solution of holding the page locked over the 3 critical operations (page_mkwrite, setting the pte dirty, and finally setting the page dirty) closes out races nicely, preventing page cleaning for writeout being initiated in that window. This provides the filesystem with a strong synchronisation against the VM here. - Sage needs this race closed for ceph filesystem. - Trond for NFS (http://bugzilla.kernel.org/show_bug.cgi?id=12913). - I need it for fsblock. - I suspect other filesystems may need it too (eg. btrfs). - I have converted buffer.c to the new locking. Even simple block allocation under dirty pages might be susceptible to i_size changing under partial page at the end of file (we also have a buffer.c-side problem here, but it cannot be fixed properly without this patch). - Other filesystems (eg. NFS, maybe btrfs) will need to change their page_mkwrite functions themselves. [ This also moves page_mkwrite another step closer to fault, which should eventually allow page_mkwrite to be moved into ->fault, and thus avoiding a filesystem calldown and page lock/unlock cycle in __do_fault. ] [akpm@linux-foundation.org: fix derefs of NULL ->mapping] Cc: Sage Weil <sage@newdream.net> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Valdis Kletnieks <Valdis.Kletnieks@vt.edu> Cc: <stable@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-30 22:08:16 +00:00
if (!pte_same(*page_table, orig_pte)) {
unlock_page(old_page);
[PATCH] add page_mkwrite() vm_operations method Add a new VMA operation to notify a filesystem or other driver about the MMU generating a fault because userspace attempted to write to a page mapped through a read-only PTE. This facility permits the filesystem or driver to: (*) Implement storage allocation/reservation on attempted write, and so to deal with problems such as ENOSPC more gracefully (perhaps by generating SIGBUS). (*) Delay making the page writable until the contents have been written to a backing cache. This is useful for NFS/AFS when using FS-Cache/CacheFS. It permits the filesystem to have some guarantee about the state of the cache. (*) Account and limit number of dirty pages. This is one piece of the puzzle needed to make shared writable mapping work safely in FUSE. Needed by cachefs (Or is it cachefiles? Or fscache? <head spins>). At least four other groups have stated an interest in it or a desire to use the functionality it provides: FUSE, OCFS2, NTFS and JFFS2. Also, things like EXT3 really ought to use it to deal with the case of shared-writable mmap encountering ENOSPC before we permit the page to be dirtied. From: Peter Zijlstra <a.p.zijlstra@chello.nl> get_user_pages(.write=1, .force=1) can generate COW hits on read-only shared mappings, this patch traps those as mkpage_write candidates and fails to handle them the old way. Signed-off-by: David Howells <dhowells@redhat.com> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Joel Becker <Joel.Becker@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Cc: Anton Altaparmakov <aia21@cantab.net> Cc: David Woodhouse <dwmw2@infradead.org> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 09:03:43 +00:00
goto unlock;
mm: close page_mkwrite races Change page_mkwrite to allow implementations to return with the page locked, and also change it's callers (in page fault paths) to hold the lock until the page is marked dirty. This allows the filesystem to have full control of page dirtying events coming from the VM. Rather than simply hold the page locked over the page_mkwrite call, we call page_mkwrite with the page unlocked and allow callers to return with it locked, so filesystems can avoid LOR conditions with page lock. The problem with the current scheme is this: a filesystem that wants to associate some metadata with a page as long as the page is dirty, will perform this manipulation in its ->page_mkwrite. It currently then must return with the page unlocked and may not hold any other locks (according to existing page_mkwrite convention). In this window, the VM could write out the page, clearing page-dirty. The filesystem has no good way to detect that a dirty pte is about to be attached, so it will happily write out the page, at which point, the filesystem may manipulate the metadata to reflect that the page is no longer dirty. It is not always possible to perform the required metadata manipulation in ->set_page_dirty, because that function cannot block or fail. The filesystem may need to allocate some data structure, for example. And the VM cannot mark the pte dirty before page_mkwrite, because page_mkwrite is allowed to fail, so we must not allow any window where the page could be written to if page_mkwrite does fail. This solution of holding the page locked over the 3 critical operations (page_mkwrite, setting the pte dirty, and finally setting the page dirty) closes out races nicely, preventing page cleaning for writeout being initiated in that window. This provides the filesystem with a strong synchronisation against the VM here. - Sage needs this race closed for ceph filesystem. - Trond for NFS (http://bugzilla.kernel.org/show_bug.cgi?id=12913). - I need it for fsblock. - I suspect other filesystems may need it too (eg. btrfs). - I have converted buffer.c to the new locking. Even simple block allocation under dirty pages might be susceptible to i_size changing under partial page at the end of file (we also have a buffer.c-side problem here, but it cannot be fixed properly without this patch). - Other filesystems (eg. NFS, maybe btrfs) will need to change their page_mkwrite functions themselves. [ This also moves page_mkwrite another step closer to fault, which should eventually allow page_mkwrite to be moved into ->fault, and thus avoiding a filesystem calldown and page lock/unlock cycle in __do_fault. ] [akpm@linux-foundation.org: fix derefs of NULL ->mapping] Cc: Sage Weil <sage@newdream.net> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Valdis Kletnieks <Valdis.Kletnieks@vt.edu> Cc: <stable@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-30 22:08:16 +00:00
}
page_mkwrite = 1;
}
[PATCH] mm: tracking shared dirty pages Tracking of dirty pages in shared writeable mmap()s. The idea is simple: write protect clean shared writeable pages, catch the write-fault, make writeable and set dirty. On page write-back clean all the PTE dirty bits and write protect them once again. The implementation is a tad harder, mainly because the default backing_dev_info capabilities were too loosely maintained. Hence it is not enough to test the backing_dev_info for cap_account_dirty. The current heuristic is as follows, a VMA is eligible when: - its shared writeable (vm_flags & (VM_WRITE|VM_SHARED)) == (VM_WRITE|VM_SHARED) - it is not a 'special' mapping (vm_flags & (VM_PFNMAP|VM_INSERTPAGE)) == 0 - the backing_dev_info is cap_account_dirty mapping_cap_account_dirty(vma->vm_file->f_mapping) - f_op->mmap() didn't change the default page protection Page from remap_pfn_range() are explicitly excluded because their COW semantics are already horrid enough (see vm_normal_page() in do_wp_page()) and because they don't have a backing store anyway. mprotect() is taught about the new behaviour as well. However it overrides the last condition. Cleaning the pages on write-back is done with page_mkclean() a new rmap call. It can be called on any page, but is currently only implemented for mapped pages, if the page is found the be of a VMA that accounts dirty pages it will also wrprotect the PTE. Finally, in fs/buffers.c:try_to_free_buffers(); remove clear_page_dirty() from under ->private_lock. This seems to be safe, since ->private_lock is used to serialize access to the buffers, not the page itself. This is needed because clear_page_dirty() will call into page_mkclean() and would thereby violate locking order. [dhowells@redhat.com: Provide a page_mkclean() implementation for NOMMU] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-26 06:30:57 +00:00
dirty_page = old_page;
get_page(dirty_page);
[PATCH] add page_mkwrite() vm_operations method Add a new VMA operation to notify a filesystem or other driver about the MMU generating a fault because userspace attempted to write to a page mapped through a read-only PTE. This facility permits the filesystem or driver to: (*) Implement storage allocation/reservation on attempted write, and so to deal with problems such as ENOSPC more gracefully (perhaps by generating SIGBUS). (*) Delay making the page writable until the contents have been written to a backing cache. This is useful for NFS/AFS when using FS-Cache/CacheFS. It permits the filesystem to have some guarantee about the state of the cache. (*) Account and limit number of dirty pages. This is one piece of the puzzle needed to make shared writable mapping work safely in FUSE. Needed by cachefs (Or is it cachefiles? Or fscache? <head spins>). At least four other groups have stated an interest in it or a desire to use the functionality it provides: FUSE, OCFS2, NTFS and JFFS2. Also, things like EXT3 really ought to use it to deal with the case of shared-writable mmap encountering ENOSPC before we permit the page to be dirtied. From: Peter Zijlstra <a.p.zijlstra@chello.nl> get_user_pages(.write=1, .force=1) can generate COW hits on read-only shared mappings, this patch traps those as mkpage_write candidates and fails to handle them the old way. Signed-off-by: David Howells <dhowells@redhat.com> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Joel Becker <Joel.Becker@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Cc: Anton Altaparmakov <aia21@cantab.net> Cc: David Woodhouse <dwmw2@infradead.org> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 09:03:43 +00:00
mm: dirty page accounting vs VM_MIXEDMAP Dirty page accounting accurately measures the amound of dirty pages in writable shared mappings by mapping the pages RO (as indicated by vma_wants_writenotify). We then trap on first write and call set_page_dirty() on the page, after which we map the page RW and continue execution. When we launder dirty pages, we call clear_page_dirty_for_io() which clears both the dirty flag, and maps the page RO again before we start writeout so that the story can repeat itself. vma_wants_writenotify() excludes VM_PFNMAP on the basis that we cannot do the regular dirty page stuff on raw PFNs and the memory isn't going anywhere anyway. The recently introduced VM_MIXEDMAP mixes both !pfn_valid() and pfn_valid() pages in a single mapping. We can't do dirty page accounting on !pfn_valid() pages as stated above, and mapping them RO causes them to be COW'ed on write, which breaks VM_SHARED semantics. Excluding VM_MIXEDMAP in vma_wants_writenotify() would mean we don't do the regular dirty page accounting for the pfn_valid() pages, which would bring back all the head-aches from inaccurate dirty page accounting. So instead, we let the !pfn_valid() pages get mapped RO, but fix them up unconditionally in the fault path. Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Acked-by: Hugh Dickins <hugh@veritas.com> Cc: "Jared Hulbert" <jaredeh@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-04 16:59:24 +00:00
reuse:
[PATCH] add page_mkwrite() vm_operations method Add a new VMA operation to notify a filesystem or other driver about the MMU generating a fault because userspace attempted to write to a page mapped through a read-only PTE. This facility permits the filesystem or driver to: (*) Implement storage allocation/reservation on attempted write, and so to deal with problems such as ENOSPC more gracefully (perhaps by generating SIGBUS). (*) Delay making the page writable until the contents have been written to a backing cache. This is useful for NFS/AFS when using FS-Cache/CacheFS. It permits the filesystem to have some guarantee about the state of the cache. (*) Account and limit number of dirty pages. This is one piece of the puzzle needed to make shared writable mapping work safely in FUSE. Needed by cachefs (Or is it cachefiles? Or fscache? <head spins>). At least four other groups have stated an interest in it or a desire to use the functionality it provides: FUSE, OCFS2, NTFS and JFFS2. Also, things like EXT3 really ought to use it to deal with the case of shared-writable mmap encountering ENOSPC before we permit the page to be dirtied. From: Peter Zijlstra <a.p.zijlstra@chello.nl> get_user_pages(.write=1, .force=1) can generate COW hits on read-only shared mappings, this patch traps those as mkpage_write candidates and fails to handle them the old way. Signed-off-by: David Howells <dhowells@redhat.com> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Joel Becker <Joel.Becker@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Cc: Anton Altaparmakov <aia21@cantab.net> Cc: David Woodhouse <dwmw2@infradead.org> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 09:03:43 +00:00
flush_cache_page(vma, address, pte_pfn(orig_pte));
entry = pte_mkyoung(orig_pte);
entry = maybe_mkwrite(pte_mkdirty(entry), vma);
if (ptep_set_access_flags(vma, address, page_table, entry,1))
update_mmu_cache(vma, address, page_table);
pte_unmap_unlock(page_table, ptl);
[PATCH] add page_mkwrite() vm_operations method Add a new VMA operation to notify a filesystem or other driver about the MMU generating a fault because userspace attempted to write to a page mapped through a read-only PTE. This facility permits the filesystem or driver to: (*) Implement storage allocation/reservation on attempted write, and so to deal with problems such as ENOSPC more gracefully (perhaps by generating SIGBUS). (*) Delay making the page writable until the contents have been written to a backing cache. This is useful for NFS/AFS when using FS-Cache/CacheFS. It permits the filesystem to have some guarantee about the state of the cache. (*) Account and limit number of dirty pages. This is one piece of the puzzle needed to make shared writable mapping work safely in FUSE. Needed by cachefs (Or is it cachefiles? Or fscache? <head spins>). At least four other groups have stated an interest in it or a desire to use the functionality it provides: FUSE, OCFS2, NTFS and JFFS2. Also, things like EXT3 really ought to use it to deal with the case of shared-writable mmap encountering ENOSPC before we permit the page to be dirtied. From: Peter Zijlstra <a.p.zijlstra@chello.nl> get_user_pages(.write=1, .force=1) can generate COW hits on read-only shared mappings, this patch traps those as mkpage_write candidates and fails to handle them the old way. Signed-off-by: David Howells <dhowells@redhat.com> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Joel Becker <Joel.Becker@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Cc: Anton Altaparmakov <aia21@cantab.net> Cc: David Woodhouse <dwmw2@infradead.org> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 09:03:43 +00:00
ret |= VM_FAULT_WRITE;
if (!dirty_page)
return ret;
/*
* Yes, Virginia, this is actually required to prevent a race
* with clear_page_dirty_for_io() from clearing the page dirty
* bit after it clear all dirty ptes, but before a racing
* do_wp_page installs a dirty pte.
*
* __do_fault is protected similarly.
*/
if (!page_mkwrite) {
wait_on_page_locked(dirty_page);
set_page_dirty_balance(dirty_page, page_mkwrite);
/* file_update_time outside page_lock */
if (vma->vm_file)
file_update_time(vma->vm_file);
}
put_page(dirty_page);
if (page_mkwrite) {
struct address_space *mapping = dirty_page->mapping;
set_page_dirty(dirty_page);
unlock_page(dirty_page);
page_cache_release(dirty_page);
if (mapping) {
/*
* Some device drivers do not set page.mapping
* but still dirty their pages
*/
balance_dirty_pages_ratelimited(mapping);
}
}
return ret;
}
/*
* Ok, we need to copy. Oh, well..
*/
2005-10-30 01:16:12 +00:00
page_cache_get(old_page);
gotten:
pte_unmap_unlock(page_table, ptl);
if (unlikely(anon_vma_prepare(vma)))
goto oom;
mm: reinstate ZERO_PAGE KAMEZAWA Hiroyuki has observed customers of earlier kernels taking advantage of the ZERO_PAGE: which we stopped do_anonymous_page() from using in 2.6.24. And there were a couple of regression reports on LKML. Following suggestions from Linus, reinstate do_anonymous_page() use of the ZERO_PAGE; but this time avoid dirtying its struct page cacheline with (map)count updates - let vm_normal_page() regard it as abnormal. Use it only on arches which __HAVE_ARCH_PTE_SPECIAL (x86, s390, sh32, most powerpc): that's not essential, but minimizes additional branches (keeping them in the unlikely pte_special case); and incidentally excludes mips (some models of which needed eight colours of ZERO_PAGE to avoid costly exceptions). Don't be fanatical about avoiding ZERO_PAGE updates: get_user_pages() callers won't want to make exceptions for it, so increment its count there. Changes to mlock and migration? happily seems not needed. In most places it's quicker to check pfn than struct page address: prepare a __read_mostly zero_pfn for that. Does get_dump_page() still need its ZERO_PAGE check? probably not, but keep it anyway. Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk> Acked-by: Rik van Riel <riel@redhat.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Minchan Kim <minchan.kim@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-09-22 00:03:30 +00:00
if (is_zero_pfn(pte_pfn(orig_pte))) {
mm: reinstate ZERO_PAGE KAMEZAWA Hiroyuki has observed customers of earlier kernels taking advantage of the ZERO_PAGE: which we stopped do_anonymous_page() from using in 2.6.24. And there were a couple of regression reports on LKML. Following suggestions from Linus, reinstate do_anonymous_page() use of the ZERO_PAGE; but this time avoid dirtying its struct page cacheline with (map)count updates - let vm_normal_page() regard it as abnormal. Use it only on arches which __HAVE_ARCH_PTE_SPECIAL (x86, s390, sh32, most powerpc): that's not essential, but minimizes additional branches (keeping them in the unlikely pte_special case); and incidentally excludes mips (some models of which needed eight colours of ZERO_PAGE to avoid costly exceptions). Don't be fanatical about avoiding ZERO_PAGE updates: get_user_pages() callers won't want to make exceptions for it, so increment its count there. Changes to mlock and migration? happily seems not needed. In most places it's quicker to check pfn than struct page address: prepare a __read_mostly zero_pfn for that. Does get_dump_page() still need its ZERO_PAGE check? probably not, but keep it anyway. Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk> Acked-by: Rik van Riel <riel@redhat.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Minchan Kim <minchan.kim@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-09-22 00:03:30 +00:00
new_page = alloc_zeroed_user_highpage_movable(vma, address);
if (!new_page)
goto oom;
} else {
new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
if (!new_page)
goto oom;
cow_user_page(new_page, old_page, address, vma);
}
__SetPageUptodate(new_page);
if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL))
Memory controller: memory accounting Add the accounting hooks. The accounting is carried out for RSS and Page Cache (unmapped) pages. There is now a common limit and accounting for both. The RSS accounting is accounted at page_add_*_rmap() and page_remove_rmap() time. Page cache is accounted at add_to_page_cache(), __delete_from_page_cache(). Swap cache is also accounted for. Each page's page_cgroup is protected with the last bit of the page_cgroup pointer, this makes handling of race conditions involving simultaneous mappings of a page easier. A reference count is kept in the page_cgroup to deal with cases where a page might be unmapped from the RSS of all tasks, but still lives in the page cache. Credits go to Vaidyanathan Srinivasan for helping with reference counting work of the page cgroup. Almost all of the page cache accounting code has help from Vaidyanathan Srinivasan. [hugh@veritas.com: fix swapoff breakage] [akpm@linux-foundation.org: fix locking] Signed-off-by: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com> Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com> Cc: Pavel Emelianov <xemul@openvz.org> Cc: Paul Menage <menage@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Kirill Korotaev <dev@sw.ru> Cc: Herbert Poetzl <herbert@13thfloor.at> Cc: David Rientjes <rientjes@google.com> Cc: <Valdis.Kletnieks@vt.edu> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-07 08:13:53 +00:00
goto oom_free_new;
mmun_start = address & PAGE_MASK;
mmun_end = mmun_start + PAGE_SIZE;
mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
/*
* Re-check the pte - we dropped the lock
*/
page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
if (likely(pte_same(*page_table, orig_pte))) {
if (old_page) {
if (!PageAnon(old_page)) {
dec_mm_counter_fast(mm, MM_FILEPAGES);
inc_mm_counter_fast(mm, MM_ANONPAGES);
}
} else
inc_mm_counter_fast(mm, MM_ANONPAGES);
flush_cache_page(vma, address, pte_pfn(orig_pte));
entry = mk_pte(new_page, vma->vm_page_prot);
entry = maybe_mkwrite(pte_mkdirty(entry), vma);
[PATCH] mm: fix a race condition under SMC + COW Failing context is a multi threaded process context and the failing sequence is as follows. One thread T0 doing self modifying code on page X on processor P0 and another thread T1 doing COW (breaking the COW setup as part of just happened fork() in another thread T2) on the same page X on processor P1. T0 doing SMC can endup modifying the new page Y (allocated by the T1 doing COW on P1) but because of different I/D TLB's, P0 ITLB will not see the new mapping till the flush TLB IPI from P1 is received. During this interval, if T0 executes the code created by SMC it can result in an app error (as ITLB still points to old page X and endup executing the content in page X rather than using the content in page Y). Fix this issue by first clearing the PTE and flushing it, before updating it with new entry. Hugh sayeth: I was a bit sceptical, in the habit of thinking that Self Modifying Code must look such issues itself: but I guess there's nothing it can do to avoid this one. Fair enough, what you're changing it to is pretty much what powerpc and s390 were already doing, and is a more robust way of proceeding, consistent with how ptes are set everywhere else. The ptep_clear_flush is a bit heavy-handed (it's anxious to return the pte that was atomically cleared), but we'd have to wander through lots of arches to get the right minimal behaviour. It'd also be nice to eliminate ptep_establish completely, now only used to define other macros/inlines: it always seemed obfuscation to me, what you've got there now is clearer. Let's put those cleanups on a TODO list. Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com> Acked-by: "David S. Miller" <davem@davemloft.net> Acked-by: Hugh Dickins <hugh@veritas.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-29 08:58:42 +00:00
/*
* Clear the pte entry and flush it first, before updating the
* pte with the new entry. This will avoid a race condition
* seen in the presence of one thread doing SMC and another
* thread doing COW.
*/
ksm: add mmu_notifier set_pte_at_notify() KSM is a linux driver that allows dynamicly sharing identical memory pages between one or more processes. Unlike tradtional page sharing that is made at the allocation of the memory, ksm do it dynamicly after the memory was created. Memory is periodically scanned; identical pages are identified and merged. The sharing is made in a transparent way to the processes that use it. Ksm is highly important for hypervisors (kvm), where in production enviorments there might be many copys of the same data data among the host memory. This kind of data can be: similar kernels, librarys, cache, and so on. Even that ksm was wrote for kvm, any userspace application that want to use it to share its data can try it. Ksm may be useful for any application that might have similar (page aligment) data strctures among the memory, ksm will find this data merge it to one copy, and even if it will be changed and thereforew copy on writed, ksm will merge it again as soon as it will be identical again. Another reason to consider using ksm is the fact that it might simplify alot the userspace code of application that want to use shared private data, instead that the application will mange shared area, ksm will do this for the application, and even write to this data will be allowed without any synchinization acts from the application. Ksm was designed to be a loadable module that doesn't change the VM code of linux. This patch: The set_pte_at_notify() macro allows setting a pte in the shadow page table directly, instead of flushing the shadow page table entry and then getting vmexit to set it. It uses a new change_pte() callback to do so. set_pte_at_notify() is an optimization for kvm, and other users of mmu_notifiers, for COW pages. It is useful for kvm when ksm is used, because it allows kvm not to have to receive vmexit and only then map the ksm page into the shadow page table, but instead map it directly at the same time as Linux maps the page into the host page table. Users of mmu_notifiers who don't implement new mmu_notifier_change_pte() callback will just receive the mmu_notifier_invalidate_page() callback. Signed-off-by: Izik Eidus <ieidus@redhat.com> Signed-off-by: Chris Wright <chrisw@redhat.com> Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Rik van Riel <riel@redhat.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Balbir Singh <balbir@in.ibm.com> Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Avi Kivity <avi@redhat.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-09-22 00:01:51 +00:00
ptep_clear_flush(vma, address, page_table);
page_add_new_anon_rmap(new_page, vma, address);
ksm: add mmu_notifier set_pte_at_notify() KSM is a linux driver that allows dynamicly sharing identical memory pages between one or more processes. Unlike tradtional page sharing that is made at the allocation of the memory, ksm do it dynamicly after the memory was created. Memory is periodically scanned; identical pages are identified and merged. The sharing is made in a transparent way to the processes that use it. Ksm is highly important for hypervisors (kvm), where in production enviorments there might be many copys of the same data data among the host memory. This kind of data can be: similar kernels, librarys, cache, and so on. Even that ksm was wrote for kvm, any userspace application that want to use it to share its data can try it. Ksm may be useful for any application that might have similar (page aligment) data strctures among the memory, ksm will find this data merge it to one copy, and even if it will be changed and thereforew copy on writed, ksm will merge it again as soon as it will be identical again. Another reason to consider using ksm is the fact that it might simplify alot the userspace code of application that want to use shared private data, instead that the application will mange shared area, ksm will do this for the application, and even write to this data will be allowed without any synchinization acts from the application. Ksm was designed to be a loadable module that doesn't change the VM code of linux. This patch: The set_pte_at_notify() macro allows setting a pte in the shadow page table directly, instead of flushing the shadow page table entry and then getting vmexit to set it. It uses a new change_pte() callback to do so. set_pte_at_notify() is an optimization for kvm, and other users of mmu_notifiers, for COW pages. It is useful for kvm when ksm is used, because it allows kvm not to have to receive vmexit and only then map the ksm page into the shadow page table, but instead map it directly at the same time as Linux maps the page into the host page table. Users of mmu_notifiers who don't implement new mmu_notifier_change_pte() callback will just receive the mmu_notifier_invalidate_page() callback. Signed-off-by: Izik Eidus <ieidus@redhat.com> Signed-off-by: Chris Wright <chrisw@redhat.com> Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Rik van Riel <riel@redhat.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Balbir Singh <balbir@in.ibm.com> Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Avi Kivity <avi@redhat.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-09-22 00:01:51 +00:00
/*
* We call the notify macro here because, when using secondary
* mmu page tables (such as kvm shadow page tables), we want the
* new page to be mapped directly into the secondary page table.
*/
set_pte_at_notify(mm, address, page_table, entry);
update_mmu_cache(vma, address, page_table);
if (old_page) {
/*
* Only after switching the pte to the new page may
* we remove the mapcount here. Otherwise another
* process may come and find the rmap count decremented
* before the pte is switched to the new page, and
* "reuse" the old page writing into it while our pte
* here still points into it and can be read by other
* threads.
*
* The critical issue is to order this
* page_remove_rmap with the ptp_clear_flush above.
* Those stores are ordered by (if nothing else,)
* the barrier present in the atomic_add_negative
* in page_remove_rmap.
*
* Then the TLB flush in ptep_clear_flush ensures that
* no process can access the old page before the
* decremented mapcount is visible. And the old page
* cannot be reused until after the decremented
* mapcount is visible. So transitively, TLBs to
* old page will be flushed before it can be reused.
*/
page_remove_rmap(old_page);
}
/* Free the old page.. */
new_page = old_page;
ret |= VM_FAULT_WRITE;
Memory controller: memory accounting Add the accounting hooks. The accounting is carried out for RSS and Page Cache (unmapped) pages. There is now a common limit and accounting for both. The RSS accounting is accounted at page_add_*_rmap() and page_remove_rmap() time. Page cache is accounted at add_to_page_cache(), __delete_from_page_cache(). Swap cache is also accounted for. Each page's page_cgroup is protected with the last bit of the page_cgroup pointer, this makes handling of race conditions involving simultaneous mappings of a page easier. A reference count is kept in the page_cgroup to deal with cases where a page might be unmapped from the RSS of all tasks, but still lives in the page cache. Credits go to Vaidyanathan Srinivasan for helping with reference counting work of the page cgroup. Almost all of the page cache accounting code has help from Vaidyanathan Srinivasan. [hugh@veritas.com: fix swapoff breakage] [akpm@linux-foundation.org: fix locking] Signed-off-by: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com> Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com> Cc: Pavel Emelianov <xemul@openvz.org> Cc: Paul Menage <menage@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Kirill Korotaev <dev@sw.ru> Cc: Herbert Poetzl <herbert@13thfloor.at> Cc: David Rientjes <rientjes@google.com> Cc: <Valdis.Kletnieks@vt.edu> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-07 08:13:53 +00:00
} else
mem_cgroup_uncharge_page(new_page);
if (new_page)
page_cache_release(new_page);
unlock:
pte_unmap_unlock(page_table, ptl);
if (mmun_end > mmun_start)
mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
if (old_page) {
/*
* Don't let another task, with possibly unlocked vma,
* keep the mlocked page.
*/
if ((ret & VM_FAULT_WRITE) && (vma->vm_flags & VM_LOCKED)) {
lock_page(old_page); /* LRU manipulation */
munlock_vma_page(old_page);
unlock_page(old_page);
}
page_cache_release(old_page);
}
return ret;
Memory controller: memory accounting Add the accounting hooks. The accounting is carried out for RSS and Page Cache (unmapped) pages. There is now a common limit and accounting for both. The RSS accounting is accounted at page_add_*_rmap() and page_remove_rmap() time. Page cache is accounted at add_to_page_cache(), __delete_from_page_cache(). Swap cache is also accounted for. Each page's page_cgroup is protected with the last bit of the page_cgroup pointer, this makes handling of race conditions involving simultaneous mappings of a page easier. A reference count is kept in the page_cgroup to deal with cases where a page might be unmapped from the RSS of all tasks, but still lives in the page cache. Credits go to Vaidyanathan Srinivasan for helping with reference counting work of the page cgroup. Almost all of the page cache accounting code has help from Vaidyanathan Srinivasan. [hugh@veritas.com: fix swapoff breakage] [akpm@linux-foundation.org: fix locking] Signed-off-by: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com> Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com> Cc: Pavel Emelianov <xemul@openvz.org> Cc: Paul Menage <menage@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Kirill Korotaev <dev@sw.ru> Cc: Herbert Poetzl <herbert@13thfloor.at> Cc: David Rientjes <rientjes@google.com> Cc: <Valdis.Kletnieks@vt.edu> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-07 08:13:53 +00:00
oom_free_new:
page_cache_release(new_page);
oom:
if (old_page)
page_cache_release(old_page);
return VM_FAULT_OOM;
[PATCH] add page_mkwrite() vm_operations method Add a new VMA operation to notify a filesystem or other driver about the MMU generating a fault because userspace attempted to write to a page mapped through a read-only PTE. This facility permits the filesystem or driver to: (*) Implement storage allocation/reservation on attempted write, and so to deal with problems such as ENOSPC more gracefully (perhaps by generating SIGBUS). (*) Delay making the page writable until the contents have been written to a backing cache. This is useful for NFS/AFS when using FS-Cache/CacheFS. It permits the filesystem to have some guarantee about the state of the cache. (*) Account and limit number of dirty pages. This is one piece of the puzzle needed to make shared writable mapping work safely in FUSE. Needed by cachefs (Or is it cachefiles? Or fscache? <head spins>). At least four other groups have stated an interest in it or a desire to use the functionality it provides: FUSE, OCFS2, NTFS and JFFS2. Also, things like EXT3 really ought to use it to deal with the case of shared-writable mmap encountering ENOSPC before we permit the page to be dirtied. From: Peter Zijlstra <a.p.zijlstra@chello.nl> get_user_pages(.write=1, .force=1) can generate COW hits on read-only shared mappings, this patch traps those as mkpage_write candidates and fails to handle them the old way. Signed-off-by: David Howells <dhowells@redhat.com> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Joel Becker <Joel.Becker@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Cc: Anton Altaparmakov <aia21@cantab.net> Cc: David Woodhouse <dwmw2@infradead.org> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 09:03:43 +00:00
unwritable_page:
page_cache_release(old_page);
return ret;
}
static void unmap_mapping_range_vma(struct vm_area_struct *vma,
unsigned long start_addr, unsigned long end_addr,
struct zap_details *details)
{
zap_page_range_single(vma, start_addr, end_addr - start_addr, details);
}
static inline void unmap_mapping_range_tree(struct rb_root *root,
struct zap_details *details)
{
struct vm_area_struct *vma;
pgoff_t vba, vea, zba, zea;
vma_interval_tree_foreach(vma, root,
details->first_index, details->last_index) {
vba = vma->vm_pgoff;
vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
/* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
zba = details->first_index;
if (zba < vba)
zba = vba;
zea = details->last_index;
if (zea > vea)
zea = vea;
unmap_mapping_range_vma(vma,
((zba - vba) << PAGE_SHIFT) + vma->vm_start,
((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
details);
}
}
static inline void unmap_mapping_range_list(struct list_head *head,
struct zap_details *details)
{
struct vm_area_struct *vma;
/*
* In nonlinear VMAs there is no correspondence between virtual address
* offset and file offset. So we must perform an exhaustive search
* across *all* the pages in each nonlinear VMA, not just the pages
* whose virtual address lies outside the file truncation point.
*/
list_for_each_entry(vma, head, shared.nonlinear) {
details->nonlinear_vma = vma;
unmap_mapping_range_vma(vma, vma->vm_start, vma->vm_end, details);
}
}
/**
* unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file.
* @mapping: the address space containing mmaps to be unmapped.
* @holebegin: byte in first page to unmap, relative to the start of
* the underlying file. This will be rounded down to a PAGE_SIZE
* boundary. Note that this is different from truncate_pagecache(), which
* must keep the partial page. In contrast, we must get rid of
* partial pages.
* @holelen: size of prospective hole in bytes. This will be rounded
* up to a PAGE_SIZE boundary. A holelen of zero truncates to the
* end of the file.
* @even_cows: 1 when truncating a file, unmap even private COWed pages;
* but 0 when invalidating pagecache, don't throw away private data.
*/
void unmap_mapping_range(struct address_space *mapping,
loff_t const holebegin, loff_t const holelen, int even_cows)
{
struct zap_details details;
pgoff_t hba = holebegin >> PAGE_SHIFT;
pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
/* Check for overflow. */
if (sizeof(holelen) > sizeof(hlen)) {
long long holeend =
(holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
if (holeend & ~(long long)ULONG_MAX)
hlen = ULONG_MAX - hba + 1;
}
details.check_mapping = even_cows? NULL: mapping;
details.nonlinear_vma = NULL;
details.first_index = hba;
details.last_index = hba + hlen - 1;
if (details.last_index < details.first_index)
details.last_index = ULONG_MAX;
mutex_lock(&mapping->i_mmap_mutex);
if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap)))
unmap_mapping_range_tree(&mapping->i_mmap, &details);
if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
mutex_unlock(&mapping->i_mmap_mutex);
}
EXPORT_SYMBOL(unmap_mapping_range);
/*
* We enter with non-exclusive mmap_sem (to exclude vma changes,
* but allow concurrent faults), and pte mapped but not yet locked.
* We return with mmap_sem still held, but pte unmapped and unlocked.
*/
static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
unsigned long address, pte_t *page_table, pmd_t *pmd,
unsigned int flags, pte_t orig_pte)
{
spinlock_t *ptl;
struct page *page, *swapcache = NULL;
swp_entry_t entry;
pte_t pte;
mm: retry page fault when blocking on disk transfer This change reduces mmap_sem hold times that are caused by waiting for disk transfers when accessing file mapped VMAs. It introduces the VM_FAULT_ALLOW_RETRY flag, which indicates that the call site wants mmap_sem to be released if blocking on a pending disk transfer. In that case, filemap_fault() returns the VM_FAULT_RETRY status bit and do_page_fault() will then re-acquire mmap_sem and retry the page fault. It is expected that the retry will hit the same page which will now be cached, and thus it will complete with a low mmap_sem hold time. Tests: - microbenchmark: thread A mmaps a large file and does random read accesses to the mmaped area - achieves about 55 iterations/s. Thread B does mmap/munmap in a loop at a separate location - achieves 55 iterations/s before, 15000 iterations/s after. - We are seeing related effects in some applications in house, which show significant performance regressions when running without this change. [akpm@linux-foundation.org: fix warning & crash] Signed-off-by: Michel Lespinasse <walken@google.com> Acked-by: Rik van Riel <riel@redhat.com> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Reviewed-by: Wu Fengguang <fengguang.wu@intel.com> Cc: Ying Han <yinghan@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Ingo Molnar <mingo@elte.hu> Cc: Thomas Gleixner <tglx@linutronix.de> Acked-by: "H. Peter Anvin" <hpa@zytor.com> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-26 21:21:57 +00:00
int locked;
struct mem_cgroup *ptr;
int exclusive = 0;
mm: fault feedback #2 This patch completes Linus's wish that the fault return codes be made into bit flags, which I agree makes everything nicer. This requires requires all handle_mm_fault callers to be modified (possibly the modifications should go further and do things like fault accounting in handle_mm_fault -- however that would be for another patch). [akpm@linux-foundation.org: fix alpha build] [akpm@linux-foundation.org: fix s390 build] [akpm@linux-foundation.org: fix sparc build] [akpm@linux-foundation.org: fix sparc64 build] [akpm@linux-foundation.org: fix ia64 build] Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Richard Henderson <rth@twiddle.net> Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Ian Molton <spyro@f2s.com> Cc: Bryan Wu <bryan.wu@analog.com> Cc: Mikael Starvik <starvik@axis.com> Cc: David Howells <dhowells@redhat.com> Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Hirokazu Takata <takata@linux-m32r.org> Cc: Geert Uytterhoeven <geert@linux-m68k.org> Cc: Roman Zippel <zippel@linux-m68k.org> Cc: Greg Ungerer <gerg@uclinux.org> Cc: Matthew Wilcox <willy@debian.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Kazumoto Kojima <kkojima@rr.iij4u.or.jp> Cc: Richard Curnow <rc@rc0.org.uk> Cc: William Lee Irwin III <wli@holomorphy.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Jeff Dike <jdike@addtoit.com> Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it> Cc: Miles Bader <uclinux-v850@lsi.nec.co.jp> Cc: Chris Zankel <chris@zankel.net> Acked-by: Kyle McMartin <kyle@mcmartin.ca> Acked-by: Haavard Skinnemoen <hskinnemoen@atmel.com> Acked-by: Ralf Baechle <ralf@linux-mips.org> Acked-by: Andi Kleen <ak@muc.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> [ Still apparently needs some ARM and PPC loving - Linus ] Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:47:05 +00:00
int ret = 0;
[PATCH] mm: split page table lock Christoph Lameter demonstrated very poor scalability on the SGI 512-way, with a many-threaded application which concurrently initializes different parts of a large anonymous area. This patch corrects that, by using a separate spinlock per page table page, to guard the page table entries in that page, instead of using the mm's single page_table_lock. (But even then, page_table_lock is still used to guard page table allocation, and anon_vma allocation.) In this implementation, the spinlock is tucked inside the struct page of the page table page: with a BUILD_BUG_ON in case it overflows - which it would in the case of 32-bit PA-RISC with spinlock debugging enabled. Splitting the lock is not quite for free: another cacheline access. Ideally, I suppose we would use split ptlock only for multi-threaded processes on multi-cpu machines; but deciding that dynamically would have its own costs. So for now enable it by config, at some number of cpus - since the Kconfig language doesn't support inequalities, let preprocessor compare that with NR_CPUS. But I don't think it's worth being user-configurable: for good testing of both split and unsplit configs, split now at 4 cpus, and perhaps change that to 8 later. There is a benefit even for singly threaded processes: kswapd can be attacking one part of the mm while another part is busy faulting. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:40 +00:00
if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
goto out;
entry = pte_to_swp_entry(orig_pte);
if (unlikely(non_swap_entry(entry))) {
if (is_migration_entry(entry)) {
migration_entry_wait(mm, pmd, address);
} else if (is_hwpoison_entry(entry)) {
ret = VM_FAULT_HWPOISON;
} else {
print_bad_pte(vma, address, orig_pte, NULL);
ret = VM_FAULT_SIGBUS;
}
[PATCH] Swapless page migration: add R/W migration entries Implement read/write migration ptes We take the upper two swapfiles for the two types of migration ptes and define a series of macros in swapops.h. The VM is modified to handle the migration entries. migration entries can only be encountered when the page they are pointing to is locked. This limits the number of places one has to fix. We also check in copy_pte_range and in mprotect_pte_range() for migration ptes. We check for migration ptes in do_swap_cache and call a function that will then wait on the page lock. This allows us to effectively stop all accesses to apge. Migration entries are created by try_to_unmap if called for migration and removed by local functions in migrate.c From: Hugh Dickins <hugh@veritas.com> Several times while testing swapless page migration (I've no NUMA, just hacking it up to migrate recklessly while running load), I've hit the BUG_ON(!PageLocked(p)) in migration_entry_to_page. This comes from an orphaned migration entry, unrelated to the current correctly locked migration, but hit by remove_anon_migration_ptes as it checks an address in each vma of the anon_vma list. Such an orphan may be left behind if an earlier migration raced with fork: copy_one_pte can duplicate a migration entry from parent to child, after remove_anon_migration_ptes has checked the child vma, but before it has removed it from the parent vma. (If the process were later to fault on this orphaned entry, it would hit the same BUG from migration_entry_wait.) This could be fixed by locking anon_vma in copy_one_pte, but we'd rather not. There's no such problem with file pages, because vma_prio_tree_add adds child vma after parent vma, and the page table locking at each end is enough to serialize. Follow that example with anon_vma: add new vmas to the tail instead of the head. (There's no corresponding problem when inserting migration entries, because a missed pte will leave the page count and mapcount high, which is allowed for. And there's no corresponding problem when migrating via swap, because a leftover swap entry will be correctly faulted. But the swapless method has no refcounting of its entries.) From: Ingo Molnar <mingo@elte.hu> pte_unmap_unlock() takes the pte pointer as an argument. From: Hugh Dickins <hugh@veritas.com> Several times while testing swapless page migration, gcc has tried to exec a pointer instead of a string: smells like COW mappings are not being properly write-protected on fork. The protection in copy_one_pte looks very convincing, until at last you realize that the second arg to make_migration_entry is a boolean "write", and SWP_MIGRATION_READ is 30. Anyway, it's better done like in change_pte_range, using is_write_migration_entry and make_migration_entry_read. From: Hugh Dickins <hugh@veritas.com> Remove unnecessary obfuscation from sys_swapon's range check on swap type, which blew up causing memory corruption once swapless migration made MAX_SWAPFILES no longer 2 ^ MAX_SWAPFILES_SHIFT. Signed-off-by: Hugh Dickins <hugh@veritas.com> Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Christoph Lameter <clameter@engr.sgi.com> Signed-off-by: Ingo Molnar <mingo@elte.hu> From: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 09:03:35 +00:00
goto out;
}
delayacct_set_flag(DELAYACCT_PF_SWAPIN);
page = lookup_swap_cache(entry);
if (!page) {
page = swapin_readahead(entry,
GFP_HIGHUSER_MOVABLE, vma, address);
if (!page) {
/*
* Back out if somebody else faulted in this pte
* while we released the pte lock.
*/
page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
if (likely(pte_same(*page_table, orig_pte)))
ret = VM_FAULT_OOM;
delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
goto unlock;
}
/* Had to read the page from swap area: Major fault */
ret = VM_FAULT_MAJOR;
[PATCH] Light weight event counters The remaining counters in page_state after the zoned VM counter patches have been applied are all just for show in /proc/vmstat. They have no essential function for the VM. We use a simple increment of per cpu variables. In order to avoid the most severe races we disable preempt. Preempt does not prevent the race between an increment and an interrupt handler incrementing the same statistics counter. However, that race is exceedingly rare, we may only loose one increment or so and there is no requirement (at least not in kernel) that the vm event counters have to be accurate. In the non preempt case this results in a simple increment for each counter. For many architectures this will be reduced by the compiler to a single instruction. This single instruction is atomic for i386 and x86_64. And therefore even the rare race condition in an interrupt is avoided for both architectures in most cases. The patchset also adds an off switch for embedded systems that allows a building of linux kernels without these counters. The implementation of these counters is through inline code that hopefully results in only a single instruction increment instruction being emitted (i386, x86_64) or in the increment being hidden though instruction concurrency (EPIC architectures such as ia64 can get that done). Benefits: - VM event counter operations usually reduce to a single inline instruction on i386 and x86_64. - No interrupt disable, only preempt disable for the preempt case. Preempt disable can also be avoided by moving the counter into a spinlock. - Handling is similar to zoned VM counters. - Simple and easily extendable. - Can be omitted to reduce memory use for embedded use. References: RFC http://marc.theaimsgroup.com/?l=linux-kernel&m=113512330605497&w=2 RFC http://marc.theaimsgroup.com/?l=linux-kernel&m=114988082814934&w=2 local_t http://marc.theaimsgroup.com/?l=linux-kernel&m=114991748606690&w=2 V2 http://marc.theaimsgroup.com/?t=115014808400007&r=1&w=2 V3 http://marc.theaimsgroup.com/?l=linux-kernel&m=115024767022346&w=2 V4 http://marc.theaimsgroup.com/?l=linux-kernel&m=115047968808926&w=2 Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-30 08:55:45 +00:00
count_vm_event(PGMAJFAULT);
memcg: add the pagefault count into memcg stats Two new stats in per-memcg memory.stat which tracks the number of page faults and number of major page faults. "pgfault" "pgmajfault" They are different from "pgpgin"/"pgpgout" stat which count number of pages charged/discharged to the cgroup and have no meaning of reading/ writing page to disk. It is valuable to track the two stats for both measuring application's performance as well as the efficiency of the kernel page reclaim path. Counting pagefaults per process is useful, but we also need the aggregated value since processes are monitored and controlled in cgroup basis in memcg. Functional test: check the total number of pgfault/pgmajfault of all memcgs and compare with global vmstat value: $ cat /proc/vmstat | grep fault pgfault 1070751 pgmajfault 553 $ cat /dev/cgroup/memory.stat | grep fault pgfault 1071138 pgmajfault 553 total_pgfault 1071142 total_pgmajfault 553 $ cat /dev/cgroup/A/memory.stat | grep fault pgfault 199 pgmajfault 0 total_pgfault 199 total_pgmajfault 0 Performance test: run page fault test(pft) wit 16 thread on faulting in 15G anon pages in 16G container. There is no regression noticed on the "flt/cpu/s" Sample output from pft: TAG pft:anon-sys-default: Gb Thr CLine User System Wall flt/cpu/s fault/wsec 15 16 1 0.67s 233.41s 14.76s 16798.546 266356.260 +-------------------------------------------------------------------------+ N Min Max Median Avg Stddev x 10 16682.962 17344.027 16913.524 16928.812 166.5362 + 10 16695.568 16923.896 16820.604 16824.652 84.816568 No difference proven at 95.0% confidence [akpm@linux-foundation.org: fix build] [hughd@google.com: shmem fix] Signed-off-by: Ying Han <yinghan@google.com> Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Reviewed-by: Minchan Kim <minchan.kim@gmail.com> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com> Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-05-26 23:25:38 +00:00
mem_cgroup_count_vm_event(mm, PGMAJFAULT);
} else if (PageHWPoison(page)) {
/*
* hwpoisoned dirty swapcache pages are kept for killing
* owner processes (which may be unknown at hwpoison time)
*/
ret = VM_FAULT_HWPOISON;
delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
goto out_release;
}
mm: retry page fault when blocking on disk transfer This change reduces mmap_sem hold times that are caused by waiting for disk transfers when accessing file mapped VMAs. It introduces the VM_FAULT_ALLOW_RETRY flag, which indicates that the call site wants mmap_sem to be released if blocking on a pending disk transfer. In that case, filemap_fault() returns the VM_FAULT_RETRY status bit and do_page_fault() will then re-acquire mmap_sem and retry the page fault. It is expected that the retry will hit the same page which will now be cached, and thus it will complete with a low mmap_sem hold time. Tests: - microbenchmark: thread A mmaps a large file and does random read accesses to the mmaped area - achieves about 55 iterations/s. Thread B does mmap/munmap in a loop at a separate location - achieves 55 iterations/s before, 15000 iterations/s after. - We are seeing related effects in some applications in house, which show significant performance regressions when running without this change. [akpm@linux-foundation.org: fix warning & crash] Signed-off-by: Michel Lespinasse <walken@google.com> Acked-by: Rik van Riel <riel@redhat.com> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Reviewed-by: Wu Fengguang <fengguang.wu@intel.com> Cc: Ying Han <yinghan@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Ingo Molnar <mingo@elte.hu> Cc: Thomas Gleixner <tglx@linutronix.de> Acked-by: "H. Peter Anvin" <hpa@zytor.com> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-26 21:21:57 +00:00
locked = lock_page_or_retry(page, mm, flags);
delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
mm: retry page fault when blocking on disk transfer This change reduces mmap_sem hold times that are caused by waiting for disk transfers when accessing file mapped VMAs. It introduces the VM_FAULT_ALLOW_RETRY flag, which indicates that the call site wants mmap_sem to be released if blocking on a pending disk transfer. In that case, filemap_fault() returns the VM_FAULT_RETRY status bit and do_page_fault() will then re-acquire mmap_sem and retry the page fault. It is expected that the retry will hit the same page which will now be cached, and thus it will complete with a low mmap_sem hold time. Tests: - microbenchmark: thread A mmaps a large file and does random read accesses to the mmaped area - achieves about 55 iterations/s. Thread B does mmap/munmap in a loop at a separate location - achieves 55 iterations/s before, 15000 iterations/s after. - We are seeing related effects in some applications in house, which show significant performance regressions when running without this change. [akpm@linux-foundation.org: fix warning & crash] Signed-off-by: Michel Lespinasse <walken@google.com> Acked-by: Rik van Riel <riel@redhat.com> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Reviewed-by: Wu Fengguang <fengguang.wu@intel.com> Cc: Ying Han <yinghan@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Ingo Molnar <mingo@elte.hu> Cc: Thomas Gleixner <tglx@linutronix.de> Acked-by: "H. Peter Anvin" <hpa@zytor.com> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-26 21:21:57 +00:00
if (!locked) {
ret |= VM_FAULT_RETRY;
goto out_release;
}
/*
mm: further fix swapin race condition Commit 4969c1192d15 ("mm: fix swapin race condition") is now agreed to be incomplete. There's a race, not very much less likely than the original race envisaged, in which it is further necessary to check that the swapcache page's swap has not changed. Here's the reasoning: cast in terms of reuse_swap_page(), but probably could be reformulated to rely on try_to_free_swap() instead, or on swapoff+swapon. A, faults into do_swap_page(): does page1 = lookup_swap_cache(swap1) and comes through the lock_page(page1). B, a racing thread of the same process, faults on the same address: does page1 = lookup_swap_cache(swap1) and now waits in lock_page(page1), but for whatever reason is unlucky not to get the lock any time soon. A carries on through do_swap_page(), a write fault, but cannot reuse the swap page1 (another reference to swap1). Unlocks the page1 (but B doesn't get it yet), does COW in do_wp_page(), page2 now in that pte. C, perhaps the parent of A+B, comes in and write faults the same swap page1 into its mm, reuse_swap_page() succeeds this time, swap1 is freed. kswapd comes in after some time (B still unlucky) and swaps out some pages from A+B and C: it allocates the original swap1 to page2 in A+B, and some other swap2 to the original page1 now in C. But does not immediately free page1 (actually it couldn't: B holds a reference), leaving it in swap cache for now. B at last gets the lock on page1, hooray! Is PageSwapCache(page1)? Yes. Is pte_same(*page_table, orig_pte)? Yes, because page2 has now been given the swap1 which page1 used to have. So B proceeds to insert page1 into A+B's page_table, though its content now belongs to C, quite different from what A wrote there. B ought to have checked that page1's swap was still swap1. Signed-off-by: Hugh Dickins <hughd@google.com> Reviewed-by: Rik van Riel <riel@redhat.com> Cc: stable@kernel.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-09-20 02:40:22 +00:00
* Make sure try_to_free_swap or reuse_swap_page or swapoff did not
* release the swapcache from under us. The page pin, and pte_same
* test below, are not enough to exclude that. Even if it is still
* swapcache, we need to check that the page's swap has not changed.
*/
mm: further fix swapin race condition Commit 4969c1192d15 ("mm: fix swapin race condition") is now agreed to be incomplete. There's a race, not very much less likely than the original race envisaged, in which it is further necessary to check that the swapcache page's swap has not changed. Here's the reasoning: cast in terms of reuse_swap_page(), but probably could be reformulated to rely on try_to_free_swap() instead, or on swapoff+swapon. A, faults into do_swap_page(): does page1 = lookup_swap_cache(swap1) and comes through the lock_page(page1). B, a racing thread of the same process, faults on the same address: does page1 = lookup_swap_cache(swap1) and now waits in lock_page(page1), but for whatever reason is unlucky not to get the lock any time soon. A carries on through do_swap_page(), a write fault, but cannot reuse the swap page1 (another reference to swap1). Unlocks the page1 (but B doesn't get it yet), does COW in do_wp_page(), page2 now in that pte. C, perhaps the parent of A+B, comes in and write faults the same swap page1 into its mm, reuse_swap_page() succeeds this time, swap1 is freed. kswapd comes in after some time (B still unlucky) and swaps out some pages from A+B and C: it allocates the original swap1 to page2 in A+B, and some other swap2 to the original page1 now in C. But does not immediately free page1 (actually it couldn't: B holds a reference), leaving it in swap cache for now. B at last gets the lock on page1, hooray! Is PageSwapCache(page1)? Yes. Is pte_same(*page_table, orig_pte)? Yes, because page2 has now been given the swap1 which page1 used to have. So B proceeds to insert page1 into A+B's page_table, though its content now belongs to C, quite different from what A wrote there. B ought to have checked that page1's swap was still swap1. Signed-off-by: Hugh Dickins <hughd@google.com> Reviewed-by: Rik van Riel <riel@redhat.com> Cc: stable@kernel.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-09-20 02:40:22 +00:00
if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val))
goto out_page;
if (ksm_might_need_to_copy(page, vma, address)) {
swapcache = page;
page = ksm_does_need_to_copy(page, vma, address);
if (unlikely(!page)) {
ret = VM_FAULT_OOM;
page = swapcache;
swapcache = NULL;
goto out_page;
}
ksm: let shared pages be swappable Initial implementation for swapping out KSM's shared pages: add page_referenced_ksm() and try_to_unmap_ksm(), which rmap.c calls when faced with a PageKsm page. Most of what's needed can be got from the rmap_items listed from the stable_node of the ksm page, without discovering the actual vma: so in this patch just fake up a struct vma for page_referenced_one() or try_to_unmap_one(), then refine that in the next patch. Add VM_NONLINEAR to ksm_madvise()'s list of exclusions: it has always been implicit there (being only set with VM_SHARED, already excluded), but let's make it explicit, to help justify the lack of nonlinear unmap. Rely on the page lock to protect against concurrent modifications to that page's node of the stable tree. The awkward part is not swapout but swapin: do_swap_page() and page_add_anon_rmap() now have to allow for new possibilities - perhaps a ksm page still in swapcache, perhaps a swapcache page associated with one location in one anon_vma now needed for another location or anon_vma. (And the vma might even be no longer VM_MERGEABLE when that happens.) ksm_might_need_to_copy() checks for that case, and supplies a duplicate page when necessary, simply leaving it to a subsequent pass of ksmd to rediscover the identity and merge them back into one ksm page. Disappointingly primitive: but the alternative would have to accumulate unswappable info about the swapped out ksm pages, limiting swappability. Remove page_add_ksm_rmap(): page_add_anon_rmap() now has to allow for the particular case it was handling, so just use it instead. Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk> Cc: Izik Eidus <ieidus@redhat.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Chris Wright <chrisw@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-15 01:59:24 +00:00
}
if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) {
Memory controller: memory accounting Add the accounting hooks. The accounting is carried out for RSS and Page Cache (unmapped) pages. There is now a common limit and accounting for both. The RSS accounting is accounted at page_add_*_rmap() and page_remove_rmap() time. Page cache is accounted at add_to_page_cache(), __delete_from_page_cache(). Swap cache is also accounted for. Each page's page_cgroup is protected with the last bit of the page_cgroup pointer, this makes handling of race conditions involving simultaneous mappings of a page easier. A reference count is kept in the page_cgroup to deal with cases where a page might be unmapped from the RSS of all tasks, but still lives in the page cache. Credits go to Vaidyanathan Srinivasan for helping with reference counting work of the page cgroup. Almost all of the page cache accounting code has help from Vaidyanathan Srinivasan. [hugh@veritas.com: fix swapoff breakage] [akpm@linux-foundation.org: fix locking] Signed-off-by: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com> Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com> Cc: Pavel Emelianov <xemul@openvz.org> Cc: Paul Menage <menage@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Kirill Korotaev <dev@sw.ru> Cc: Herbert Poetzl <herbert@13thfloor.at> Cc: David Rientjes <rientjes@google.com> Cc: <Valdis.Kletnieks@vt.edu> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-07 08:13:53 +00:00
ret = VM_FAULT_OOM;
goto out_page;
Memory controller: memory accounting Add the accounting hooks. The accounting is carried out for RSS and Page Cache (unmapped) pages. There is now a common limit and accounting for both. The RSS accounting is accounted at page_add_*_rmap() and page_remove_rmap() time. Page cache is accounted at add_to_page_cache(), __delete_from_page_cache(). Swap cache is also accounted for. Each page's page_cgroup is protected with the last bit of the page_cgroup pointer, this makes handling of race conditions involving simultaneous mappings of a page easier. A reference count is kept in the page_cgroup to deal with cases where a page might be unmapped from the RSS of all tasks, but still lives in the page cache. Credits go to Vaidyanathan Srinivasan for helping with reference counting work of the page cgroup. Almost all of the page cache accounting code has help from Vaidyanathan Srinivasan. [hugh@veritas.com: fix swapoff breakage] [akpm@linux-foundation.org: fix locking] Signed-off-by: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com> Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com> Cc: Pavel Emelianov <xemul@openvz.org> Cc: Paul Menage <menage@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Kirill Korotaev <dev@sw.ru> Cc: Herbert Poetzl <herbert@13thfloor.at> Cc: David Rientjes <rientjes@google.com> Cc: <Valdis.Kletnieks@vt.edu> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-07 08:13:53 +00:00
}
/*
* Back out if somebody else already faulted in this pte.
*/
page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
if (unlikely(!pte_same(*page_table, orig_pte)))
goto out_nomap;
if (unlikely(!PageUptodate(page))) {
ret = VM_FAULT_SIGBUS;
goto out_nomap;
}
memcg: mem+swap controller core This patch implements per cgroup limit for usage of memory+swap. However there are SwapCache, double counting of swap-cache and swap-entry is avoided. Mem+Swap controller works as following. - memory usage is limited by memory.limit_in_bytes. - memory + swap usage is limited by memory.memsw_limit_in_bytes. This has following benefits. - A user can limit total resource usage of mem+swap. Without this, because memory resource controller doesn't take care of usage of swap, a process can exhaust all the swap (by memory leak.) We can avoid this case. And Swap is shared resource but it cannot be reclaimed (goes back to memory) until it's used. This characteristic can be trouble when the memory is divided into some parts by cpuset or memcg. Assume group A and group B. After some application executes, the system can be.. Group A -- very large free memory space but occupy 99% of swap. Group B -- under memory shortage but cannot use swap...it's nearly full. Ability to set appropriate swap limit for each group is required. Maybe someone wonder "why not swap but mem+swap ?" - The global LRU(kswapd) can swap out arbitrary pages. Swap-out means to move account from memory to swap...there is no change in usage of mem+swap. In other words, when we want to limit the usage of swap without affecting global LRU, mem+swap limit is better than just limiting swap. Accounting target information is stored in swap_cgroup which is per swap entry record. Charge is done as following. map - charge page and memsw. unmap - uncharge page/memsw if not SwapCache. swap-out (__delete_from_swap_cache) - uncharge page - record mem_cgroup information to swap_cgroup. swap-in (do_swap_page) - charged as page and memsw. record in swap_cgroup is cleared. memsw accounting is decremented. swap-free (swap_free()) - if swap entry is freed, memsw is uncharged by PAGE_SIZE. There are people work under never-swap environments and consider swap as something bad. For such people, this mem+swap controller extension is just an overhead. This overhead is avoided by config or boot option. (see Kconfig. detail is not in this patch.) TODO: - maybe more optimization can be don in swap-in path. (but not very safe.) But we just do simple accounting at this stage. [nishimura@mxp.nes.nec.co.jp: make resize limit hold mutex] [hugh@veritas.com: memswap controller core swapcache fixes] Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Li Zefan <lizf@cn.fujitsu.com> Cc: Balbir Singh <balbir@in.ibm.com> Cc: Pavel Emelyanov <xemul@openvz.org> Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:00 +00:00
/*
* The page isn't present yet, go ahead with the fault.
*
* Be careful about the sequence of operations here.
* To get its accounting right, reuse_swap_page() must be called
* while the page is counted on swap but not yet in mapcount i.e.
* before page_add_anon_rmap() and swap_free(); try_to_free_swap()
* must be called after the swap_free(), or it will never succeed.
memcg: fix swap accounting leak Fix swapin charge operation of memcg. Now, memcg has hooks to swap-out operation and checks SwapCache is really unused or not. That check depends on contents of struct page. I.e. If PageAnon(page) && page_mapped(page), the page is recoginized as still-in-use. Now, reuse_swap_page() calles delete_from_swap_cache() before establishment of any rmap. Then, in followinig sequence (Page fault with WRITE) try_charge() (charge += PAGESIZE) commit_charge() (Check page_cgroup is used or not..) reuse_swap_page() -> delete_from_swapcache() -> mem_cgroup_uncharge_swapcache() (charge -= PAGESIZE) ...... New charge is uncharged soon.... To avoid this, move commit_charge() after page_mapcount() goes up to 1. By this, try_charge() (usage += PAGESIZE) reuse_swap_page() (may usage -= PAGESIZE if PCG_USED is set) commit_charge() (If page_cgroup is not marked as PCG_USED, add new charge.) Accounting will be correct. Changelog (v2) -> (v3) - fixed invalid charge to swp_entry==0. - updated documentation. Changelog (v1) -> (v2) - fixed comment. [nishimura@mxp.nes.nec.co.jp: swap accounting leak doc fix] Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com> Tested-by: Balbir Singh <balbir@linux.vnet.ibm.com> Cc: Hugh Dickins <hugh@veritas.com> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:31 +00:00
* Because delete_from_swap_page() may be called by reuse_swap_page(),
* mem_cgroup_commit_charge_swapin() may not be able to find swp_entry
* in page->private. In this case, a record in swap_cgroup is silently
* discarded at swap_free().
memcg: mem+swap controller core This patch implements per cgroup limit for usage of memory+swap. However there are SwapCache, double counting of swap-cache and swap-entry is avoided. Mem+Swap controller works as following. - memory usage is limited by memory.limit_in_bytes. - memory + swap usage is limited by memory.memsw_limit_in_bytes. This has following benefits. - A user can limit total resource usage of mem+swap. Without this, because memory resource controller doesn't take care of usage of swap, a process can exhaust all the swap (by memory leak.) We can avoid this case. And Swap is shared resource but it cannot be reclaimed (goes back to memory) until it's used. This characteristic can be trouble when the memory is divided into some parts by cpuset or memcg. Assume group A and group B. After some application executes, the system can be.. Group A -- very large free memory space but occupy 99% of swap. Group B -- under memory shortage but cannot use swap...it's nearly full. Ability to set appropriate swap limit for each group is required. Maybe someone wonder "why not swap but mem+swap ?" - The global LRU(kswapd) can swap out arbitrary pages. Swap-out means to move account from memory to swap...there is no change in usage of mem+swap. In other words, when we want to limit the usage of swap without affecting global LRU, mem+swap limit is better than just limiting swap. Accounting target information is stored in swap_cgroup which is per swap entry record. Charge is done as following. map - charge page and memsw. unmap - uncharge page/memsw if not SwapCache. swap-out (__delete_from_swap_cache) - uncharge page - record mem_cgroup information to swap_cgroup. swap-in (do_swap_page) - charged as page and memsw. record in swap_cgroup is cleared. memsw accounting is decremented. swap-free (swap_free()) - if swap entry is freed, memsw is uncharged by PAGE_SIZE. There are people work under never-swap environments and consider swap as something bad. For such people, this mem+swap controller extension is just an overhead. This overhead is avoided by config or boot option. (see Kconfig. detail is not in this patch.) TODO: - maybe more optimization can be don in swap-in path. (but not very safe.) But we just do simple accounting at this stage. [nishimura@mxp.nes.nec.co.jp: make resize limit hold mutex] [hugh@veritas.com: memswap controller core swapcache fixes] Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Li Zefan <lizf@cn.fujitsu.com> Cc: Balbir Singh <balbir@in.ibm.com> Cc: Pavel Emelyanov <xemul@openvz.org> Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:00 +00:00
*/
inc_mm_counter_fast(mm, MM_ANONPAGES);
dec_mm_counter_fast(mm, MM_SWAPENTS);
pte = mk_pte(page, vma->vm_page_prot);
if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) {
pte = maybe_mkwrite(pte_mkdirty(pte), vma);
flags &= ~FAULT_FLAG_WRITE;
ret |= VM_FAULT_WRITE;
exclusive = 1;
}
flush_icache_page(vma, page);
set_pte_at(mm, address, page_table, pte);
do_page_add_anon_rmap(page, vma, address, exclusive);
memcg: fix swap accounting leak Fix swapin charge operation of memcg. Now, memcg has hooks to swap-out operation and checks SwapCache is really unused or not. That check depends on contents of struct page. I.e. If PageAnon(page) && page_mapped(page), the page is recoginized as still-in-use. Now, reuse_swap_page() calles delete_from_swap_cache() before establishment of any rmap. Then, in followinig sequence (Page fault with WRITE) try_charge() (charge += PAGESIZE) commit_charge() (Check page_cgroup is used or not..) reuse_swap_page() -> delete_from_swapcache() -> mem_cgroup_uncharge_swapcache() (charge -= PAGESIZE) ...... New charge is uncharged soon.... To avoid this, move commit_charge() after page_mapcount() goes up to 1. By this, try_charge() (usage += PAGESIZE) reuse_swap_page() (may usage -= PAGESIZE if PCG_USED is set) commit_charge() (If page_cgroup is not marked as PCG_USED, add new charge.) Accounting will be correct. Changelog (v2) -> (v3) - fixed invalid charge to swp_entry==0. - updated documentation. Changelog (v1) -> (v2) - fixed comment. [nishimura@mxp.nes.nec.co.jp: swap accounting leak doc fix] Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com> Tested-by: Balbir Singh <balbir@linux.vnet.ibm.com> Cc: Hugh Dickins <hugh@veritas.com> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:31 +00:00
/* It's better to call commit-charge after rmap is established */
mem_cgroup_commit_charge_swapin(page, ptr);
swap_free(entry);
mlock: mlocked pages are unevictable Make sure that mlocked pages also live on the unevictable LRU, so kswapd will not scan them over and over again. This is achieved through various strategies: 1) add yet another page flag--PG_mlocked--to indicate that the page is locked for efficient testing in vmscan and, optionally, fault path. This allows early culling of unevictable pages, preventing them from getting to page_referenced()/try_to_unmap(). Also allows separate accounting of mlock'd pages, as Nick's original patch did. Note: Nick's original mlock patch used a PG_mlocked flag. I had removed this in favor of the PG_unevictable flag + an mlock_count [new page struct member]. I restored the PG_mlocked flag to eliminate the new count field. 2) add the mlock/unevictable infrastructure to mm/mlock.c, with internal APIs in mm/internal.h. This is a rework of Nick's original patch to these files, taking into account that mlocked pages are now kept on unevictable LRU list. 3) update vmscan.c:page_evictable() to check PageMlocked() and, if vma passed in, the vm_flags. Note that the vma will only be passed in for new pages in the fault path; and then only if the "cull unevictable pages in fault path" patch is included. 4) add try_to_unlock() to rmap.c to walk a page's rmap and ClearPageMlocked() if no other vmas have it mlocked. Reuses as much of try_to_unmap() as possible. This effectively replaces the use of one of the lru list links as an mlock count. If this mechanism let's pages in mlocked vmas leak through w/o PG_mlocked set [I don't know that it does], we should catch them later in try_to_unmap(). One hopes this will be rare, as it will be relatively expensive. Original mm/internal.h, mm/rmap.c and mm/mlock.c changes: Signed-off-by: Nick Piggin <npiggin@suse.de> splitlru: introduce __get_user_pages(): New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS. because current get_user_pages() can't grab PROT_NONE pages theresore it cause PROT_NONE pages can't munlock. [akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch] [akpm@linux-foundation.org: untangle patch interdependencies] [akpm@linux-foundation.org: fix things after out-of-order merging] [hugh@veritas.com: fix page-flags mess] [lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm'] [kosaki.motohiro@jp.fujitsu.com: build fix] [kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments] [kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()] Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Signed-off-by: Rik van Riel <riel@redhat.com> Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Dave Hansen <dave@linux.vnet.ibm.com> Cc: Matt Mackall <mpm@selenic.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 03:26:44 +00:00
if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
mm: try_to_free_swap replaces remove_exclusive_swap_page remove_exclusive_swap_page(): its problem is in living up to its name. It doesn't matter if someone else has a reference to the page (raised page_count); it doesn't matter if the page is mapped into userspace (raised page_mapcount - though that hints it may be worth keeping the swap): all that matters is that there be no more references to the swap (and no writeback in progress). swapoff (try_to_unuse) has been removing pages from swapcache for years, with no concern for page count or page mapcount, and we used to have a comment in lookup_swap_cache() recognizing that: if you go for a page of swapcache, you'll get the right page, but it could have been removed from swapcache by the time you get page lock. So, give up asking for exclusivity: get rid of remove_exclusive_swap_page(), and remove_exclusive_swap_page_ref() and remove_exclusive_swap_page_count() which were spawned for the recent LRU work: replace them by the simpler try_to_free_swap() which just checks page_swapcount(). Similarly, remove the page_count limitation from free_swap_and_count(), but assume that it's worth holding on to the swap if page is mapped and swap nowhere near full. Add a vm_swap_full() test in free_swap_cache()? It would be consistent, but I think we probably have enough for now. Signed-off-by: Hugh Dickins <hugh@veritas.com> Cc: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Rik van Riel <riel@redhat.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Robin Holt <holt@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:39:36 +00:00
try_to_free_swap(page);
unlock_page(page);
if (swapcache) {
/*
* Hold the lock to avoid the swap entry to be reused
* until we take the PT lock for the pte_same() check
* (to avoid false positives from pte_same). For
* further safety release the lock after the swap_free
* so that the swap count won't change under a
* parallel locked swapcache.
*/
unlock_page(swapcache);
page_cache_release(swapcache);
}
if (flags & FAULT_FLAG_WRITE) {
ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
if (ret & VM_FAULT_ERROR)
ret &= VM_FAULT_ERROR;
goto out;
}
/* No need to invalidate - it was non-present before */
update_mmu_cache(vma, address, page_table);
unlock:
pte_unmap_unlock(page_table, ptl);
out:
return ret;
out_nomap:
memcg: introduce charge-commit-cancel style of functions There is a small race in do_swap_page(). When the page swapped-in is charged, the mapcount can be greater than 0. But, at the same time some process (shares it ) call unmap and make mapcount 1->0 and the page is uncharged. CPUA CPUB mapcount == 1. (1) charge if mapcount==0 zap_pte_range() (2) mapcount 1 => 0. (3) uncharge(). (success) (4) set page's rmap() mapcount 0=>1 Then, this swap page's account is leaked. For fixing this, I added a new interface. - charge account to res_counter by PAGE_SIZE and try to free pages if necessary. - commit register page_cgroup and add to LRU if necessary. - cancel uncharge PAGE_SIZE because of do_swap_page failure. CPUA (1) charge (always) (2) set page's rmap (mapcount > 0) (3) commit charge was necessary or not after set_pte(). This protocol uses PCG_USED bit on page_cgroup for avoiding over accounting. Usual mem_cgroup_charge_common() does charge -> commit at a time. And this patch also adds following function to clarify all charges. - mem_cgroup_newpage_charge() ....replacement for mem_cgroup_charge() called against newly allocated anon pages. - mem_cgroup_charge_migrate_fixup() called only from remove_migration_ptes(). we'll have to rewrite this later.(this patch just keeps old behavior) This function will be removed by additional patch to make migration clearer. Good for clarifying "what we do" Then, we have 4 following charge points. - newpage - swap-in - add-to-cache. - migration. [akpm@linux-foundation.org: add missing inline directives to stubs] Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Cc: Balbir Singh <balbir@in.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:48 +00:00
mem_cgroup_cancel_charge_swapin(ptr);
pte_unmap_unlock(page_table, ptl);
out_page:
unlock_page(page);
out_release:
page_cache_release(page);
if (swapcache) {
unlock_page(swapcache);
page_cache_release(swapcache);
}
return ret;
}
/*
* This is like a special single-page "expand_{down|up}wards()",
* except we must first make sure that 'address{-|+}PAGE_SIZE'
* doesn't hit another vma.
*/
static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address)
{
address &= PAGE_MASK;
if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) {
struct vm_area_struct *prev = vma->vm_prev;
/*
* Is there a mapping abutting this one below?
*
* That's only ok if it's the same stack mapping
* that has gotten split..
*/
if (prev && prev->vm_end == address)
return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM;
expand_downwards(vma, address - PAGE_SIZE);
}
if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) {
struct vm_area_struct *next = vma->vm_next;
/* As VM_GROWSDOWN but s/below/above/ */
if (next && next->vm_start == address + PAGE_SIZE)
return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM;
expand_upwards(vma, address + PAGE_SIZE);
}
return 0;
}
/*
* We enter with non-exclusive mmap_sem (to exclude vma changes,
* but allow concurrent faults), and pte mapped but not yet locked.
* We return with mmap_sem still held, but pte unmapped and unlocked.
*/
static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
unsigned long address, pte_t *page_table, pmd_t *pmd,
unsigned int flags)
{
struct page *page;
spinlock_t *ptl;
pte_t entry;
pte_unmap(page_table);
/* Check if we need to add a guard page to the stack */
if (check_stack_guard_page(vma, address) < 0)
return VM_FAULT_SIGBUS;
/* Use the zero-page for reads */
if (!(flags & FAULT_FLAG_WRITE)) {
entry = pte_mkspecial(pfn_pte(my_zero_pfn(address),
vma->vm_page_prot));
page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
mm: reinstate ZERO_PAGE KAMEZAWA Hiroyuki has observed customers of earlier kernels taking advantage of the ZERO_PAGE: which we stopped do_anonymous_page() from using in 2.6.24. And there were a couple of regression reports on LKML. Following suggestions from Linus, reinstate do_anonymous_page() use of the ZERO_PAGE; but this time avoid dirtying its struct page cacheline with (map)count updates - let vm_normal_page() regard it as abnormal. Use it only on arches which __HAVE_ARCH_PTE_SPECIAL (x86, s390, sh32, most powerpc): that's not essential, but minimizes additional branches (keeping them in the unlikely pte_special case); and incidentally excludes mips (some models of which needed eight colours of ZERO_PAGE to avoid costly exceptions). Don't be fanatical about avoiding ZERO_PAGE updates: get_user_pages() callers won't want to make exceptions for it, so increment its count there. Changes to mlock and migration? happily seems not needed. In most places it's quicker to check pfn than struct page address: prepare a __read_mostly zero_pfn for that. Does get_dump_page() still need its ZERO_PAGE check? probably not, but keep it anyway. Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk> Acked-by: Rik van Riel <riel@redhat.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Minchan Kim <minchan.kim@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-09-22 00:03:30 +00:00
if (!pte_none(*page_table))
goto unlock;
goto setpte;
}
remove ZERO_PAGE The commit b5810039a54e5babf428e9a1e89fc1940fabff11 contains the note A last caveat: the ZERO_PAGE is now refcounted and managed with rmap (and thus mapcounted and count towards shared rss). These writes to the struct page could cause excessive cacheline bouncing on big systems. There are a number of ways this could be addressed if it is an issue. And indeed this cacheline bouncing has shown up on large SGI systems. There was a situation where an Altix system was essentially livelocked tearing down ZERO_PAGE pagetables when an HPC app aborted during startup. This situation can be avoided in userspace, but it does highlight the potential scalability problem with refcounting ZERO_PAGE, and corner cases where it can really hurt (we don't want the system to livelock!). There are several broad ways to fix this problem: 1. add back some special casing to avoid refcounting ZERO_PAGE 2. per-node or per-cpu ZERO_PAGES 3. remove the ZERO_PAGE completely I will argue for 3. The others should also fix the problem, but they result in more complex code than does 3, with little or no real benefit that I can see. Why? Inserting a ZERO_PAGE for anonymous read faults appears to be a false optimisation: if an application is performance critical, it would not be doing many read faults of new memory, or at least it could be expected to write to that memory soon afterwards. If cache or memory use is critical, it should not be working with a significant number of ZERO_PAGEs anyway (a more compact representation of zeroes should be used). As a sanity check -- mesuring on my desktop system, there are never many mappings to the ZERO_PAGE (eg. 2 or 3), thus memory usage here should not increase much without it. When running a make -j4 kernel compile on my dual core system, there are about 1,000 mappings to the ZERO_PAGE created per second, but about 1,000 ZERO_PAGE COW faults per second (less than 1 ZERO_PAGE mapping per second is torn down without being COWed). So removing ZERO_PAGE will save 1,000 page faults per second when running kbuild, while keeping it only saves less than 1 page clearing operation per second. 1 page clear is cheaper than a thousand faults, presumably, so there isn't an obvious loss. Neither the logical argument nor these basic tests give a guarantee of no regressions. However, this is a reasonable opportunity to try to remove the ZERO_PAGE from the pagefault path. If it is found to cause regressions, we can reintroduce it and just avoid refcounting it. The /dev/zero ZERO_PAGE usage and TLB tricks also get nuked. I don't see much use to them except on benchmarks. All other users of ZERO_PAGE are converted just to use ZERO_PAGE(0) for simplicity. We can look at replacing them all and maybe ripping out ZERO_PAGE completely when we are more satisfied with this solution. Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus "snif" Torvalds <torvalds@linux-foundation.org>
2007-10-16 08:24:40 +00:00
/* Allocate our own private page. */
if (unlikely(anon_vma_prepare(vma)))
goto oom;
page = alloc_zeroed_user_highpage_movable(vma, address);
if (!page)
goto oom;
mm: fix PageUptodate data race After running SetPageUptodate, preceeding stores to the page contents to actually bring it uptodate may not be ordered with the store to set the page uptodate. Therefore, another CPU which checks PageUptodate is true, then reads the page contents can get stale data. Fix this by having an smp_wmb before SetPageUptodate, and smp_rmb after PageUptodate. Many places that test PageUptodate, do so with the page locked, and this would be enough to ensure memory ordering in those places if SetPageUptodate were only called while the page is locked. Unfortunately that is not always the case for some filesystems, but it could be an idea for the future. Also bring the handling of anonymous page uptodateness in line with that of file backed page management, by marking anon pages as uptodate when they _are_ uptodate, rather than when our implementation requires that they be marked as such. Doing allows us to get rid of the smp_wmb's in the page copying functions, which were especially added for anonymous pages for an analogous memory ordering problem. Both file and anonymous pages are handled with the same barriers. FAQ: Q. Why not do this in flush_dcache_page? A. Firstly, flush_dcache_page handles only one side (the smb side) of the ordering protocol; we'd still need smp_rmb somewhere. Secondly, hiding away memory barriers in a completely unrelated function is nasty; at least in the PageUptodate macros, they are located together with (half) the operations involved in the ordering. Thirdly, the smp_wmb is only required when first bringing the page uptodate, wheras flush_dcache_page should be called each time it is written to through the kernel mapping. It is logically the wrong place to put it. Q. Why does this increase my text size / reduce my performance / etc. A. Because it is adding the necessary instructions to eliminate the data-race. Q. Can it be improved? A. Yes, eg. if you were to create a rule that all SetPageUptodate operations run under the page lock, we could avoid the smp_rmb places where PageUptodate is queried under the page lock. Requires audit of all filesystems and at least some would need reworking. That's great you're interested, I'm eagerly awaiting your patches. Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 06:29:34 +00:00
__SetPageUptodate(page);
if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL))
Memory controller: memory accounting Add the accounting hooks. The accounting is carried out for RSS and Page Cache (unmapped) pages. There is now a common limit and accounting for both. The RSS accounting is accounted at page_add_*_rmap() and page_remove_rmap() time. Page cache is accounted at add_to_page_cache(), __delete_from_page_cache(). Swap cache is also accounted for. Each page's page_cgroup is protected with the last bit of the page_cgroup pointer, this makes handling of race conditions involving simultaneous mappings of a page easier. A reference count is kept in the page_cgroup to deal with cases where a page might be unmapped from the RSS of all tasks, but still lives in the page cache. Credits go to Vaidyanathan Srinivasan for helping with reference counting work of the page cgroup. Almost all of the page cache accounting code has help from Vaidyanathan Srinivasan. [hugh@veritas.com: fix swapoff breakage] [akpm@linux-foundation.org: fix locking] Signed-off-by: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com> Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com> Cc: Pavel Emelianov <xemul@openvz.org> Cc: Paul Menage <menage@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Kirill Korotaev <dev@sw.ru> Cc: Herbert Poetzl <herbert@13thfloor.at> Cc: David Rientjes <rientjes@google.com> Cc: <Valdis.Kletnieks@vt.edu> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-07 08:13:53 +00:00
goto oom_free_page;
remove ZERO_PAGE The commit b5810039a54e5babf428e9a1e89fc1940fabff11 contains the note A last caveat: the ZERO_PAGE is now refcounted and managed with rmap (and thus mapcounted and count towards shared rss). These writes to the struct page could cause excessive cacheline bouncing on big systems. There are a number of ways this could be addressed if it is an issue. And indeed this cacheline bouncing has shown up on large SGI systems. There was a situation where an Altix system was essentially livelocked tearing down ZERO_PAGE pagetables when an HPC app aborted during startup. This situation can be avoided in userspace, but it does highlight the potential scalability problem with refcounting ZERO_PAGE, and corner cases where it can really hurt (we don't want the system to livelock!). There are several broad ways to fix this problem: 1. add back some special casing to avoid refcounting ZERO_PAGE 2. per-node or per-cpu ZERO_PAGES 3. remove the ZERO_PAGE completely I will argue for 3. The others should also fix the problem, but they result in more complex code than does 3, with little or no real benefit that I can see. Why? Inserting a ZERO_PAGE for anonymous read faults appears to be a false optimisation: if an application is performance critical, it would not be doing many read faults of new memory, or at least it could be expected to write to that memory soon afterwards. If cache or memory use is critical, it should not be working with a significant number of ZERO_PAGEs anyway (a more compact representation of zeroes should be used). As a sanity check -- mesuring on my desktop system, there are never many mappings to the ZERO_PAGE (eg. 2 or 3), thus memory usage here should not increase much without it. When running a make -j4 kernel compile on my dual core system, there are about 1,000 mappings to the ZERO_PAGE created per second, but about 1,000 ZERO_PAGE COW faults per second (less than 1 ZERO_PAGE mapping per second is torn down without being COWed). So removing ZERO_PAGE will save 1,000 page faults per second when running kbuild, while keeping it only saves less than 1 page clearing operation per second. 1 page clear is cheaper than a thousand faults, presumably, so there isn't an obvious loss. Neither the logical argument nor these basic tests give a guarantee of no regressions. However, this is a reasonable opportunity to try to remove the ZERO_PAGE from the pagefault path. If it is found to cause regressions, we can reintroduce it and just avoid refcounting it. The /dev/zero ZERO_PAGE usage and TLB tricks also get nuked. I don't see much use to them except on benchmarks. All other users of ZERO_PAGE are converted just to use ZERO_PAGE(0) for simplicity. We can look at replacing them all and maybe ripping out ZERO_PAGE completely when we are more satisfied with this solution. Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus "snif" Torvalds <torvalds@linux-foundation.org>
2007-10-16 08:24:40 +00:00
entry = mk_pte(page, vma->vm_page_prot);
if (vma->vm_flags & VM_WRITE)
entry = pte_mkwrite(pte_mkdirty(entry));
remove ZERO_PAGE The commit b5810039a54e5babf428e9a1e89fc1940fabff11 contains the note A last caveat: the ZERO_PAGE is now refcounted and managed with rmap (and thus mapcounted and count towards shared rss). These writes to the struct page could cause excessive cacheline bouncing on big systems. There are a number of ways this could be addressed if it is an issue. And indeed this cacheline bouncing has shown up on large SGI systems. There was a situation where an Altix system was essentially livelocked tearing down ZERO_PAGE pagetables when an HPC app aborted during startup. This situation can be avoided in userspace, but it does highlight the potential scalability problem with refcounting ZERO_PAGE, and corner cases where it can really hurt (we don't want the system to livelock!). There are several broad ways to fix this problem: 1. add back some special casing to avoid refcounting ZERO_PAGE 2. per-node or per-cpu ZERO_PAGES 3. remove the ZERO_PAGE completely I will argue for 3. The others should also fix the problem, but they result in more complex code than does 3, with little or no real benefit that I can see. Why? Inserting a ZERO_PAGE for anonymous read faults appears to be a false optimisation: if an application is performance critical, it would not be doing many read faults of new memory, or at least it could be expected to write to that memory soon afterwards. If cache or memory use is critical, it should not be working with a significant number of ZERO_PAGEs anyway (a more compact representation of zeroes should be used). As a sanity check -- mesuring on my desktop system, there are never many mappings to the ZERO_PAGE (eg. 2 or 3), thus memory usage here should not increase much without it. When running a make -j4 kernel compile on my dual core system, there are about 1,000 mappings to the ZERO_PAGE created per second, but about 1,000 ZERO_PAGE COW faults per second (less than 1 ZERO_PAGE mapping per second is torn down without being COWed). So removing ZERO_PAGE will save 1,000 page faults per second when running kbuild, while keeping it only saves less than 1 page clearing operation per second. 1 page clear is cheaper than a thousand faults, presumably, so there isn't an obvious loss. Neither the logical argument nor these basic tests give a guarantee of no regressions. However, this is a reasonable opportunity to try to remove the ZERO_PAGE from the pagefault path. If it is found to cause regressions, we can reintroduce it and just avoid refcounting it. The /dev/zero ZERO_PAGE usage and TLB tricks also get nuked. I don't see much use to them except on benchmarks. All other users of ZERO_PAGE are converted just to use ZERO_PAGE(0) for simplicity. We can look at replacing them all and maybe ripping out ZERO_PAGE completely when we are more satisfied with this solution. Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus "snif" Torvalds <torvalds@linux-foundation.org>
2007-10-16 08:24:40 +00:00
page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
ksm: fix deadlock with munlock in exit_mmap Rawhide users have reported hang at startup when cryptsetup is run: the same problem can be simply reproduced by running a program int main() { mlockall(MCL_CURRENT | MCL_FUTURE); return 0; } The problem is that exit_mmap() applies munlock_vma_pages_all() to clean up VM_LOCKED areas, and its current implementation (stupidly) tries to fault in absent pages, for example where PROT_NONE prevented them being faulted in when mlocking. Whereas the "ksm: fix oom deadlock" patch, knowing there's a race by which KSM might try to fault in pages after exit_mmap() had finally zapped the range, backs out of such faults doing nothing when its ksm_test_exit() notices mm_users 0. So revert that part of "ksm: fix oom deadlock" which moved the ksm_exit() call from before exit_mmap() to the middle of exit_mmap(); and remove those ksm_test_exit() checks from the page fault paths, so allowing the munlocking to proceed without interference. ksm_exit, if there are rmap_items still chained on this mm slot, takes mmap_sem write side: so preventing KSM from working on an mm while exit_mmap runs. And KSM will bail out as soon as it notices that mm_users is already zero, thanks to its internal ksm_test_exit checks. So that when a task is killed by OOM killer or the user, KSM will not indefinitely prevent it from running exit_mmap to release its memory. This does break a part of what "ksm: fix oom deadlock" was trying to achieve. When unmerging KSM (echo 2 >/sys/kernel/mm/ksm), and even when ksmd itself has to cancel a KSM page, it is possible that the first OOM-kill victim would be the KSM process being faulted: then its memory won't be freed until a second victim has been selected (freeing memory for the unmerging fault to complete). But the OOM killer is already liable to kill a second victim once the intended victim's p->mm goes to NULL: so there's not much point in rejecting this KSM patch before fixing that OOM behaviour. It is very much more important to allow KSM users to boot up, than to haggle over an unlikely and poorly supported OOM case. We also intend to fix munlocking to not fault pages: at which point this patch _could_ be reverted; though that would be controversial, so we hope to find a better solution. Signed-off-by: Andrea Arcangeli <aarcange@redhat.com> Acked-by: Justin M. Forbes <jforbes@redhat.com> Acked-for-now-by: Hugh Dickins <hugh.dickins@tiscali.co.uk> Cc: Izik Eidus <ieidus@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-09-22 00:02:22 +00:00
if (!pte_none(*page_table))
remove ZERO_PAGE The commit b5810039a54e5babf428e9a1e89fc1940fabff11 contains the note A last caveat: the ZERO_PAGE is now refcounted and managed with rmap (and thus mapcounted and count towards shared rss). These writes to the struct page could cause excessive cacheline bouncing on big systems. There are a number of ways this could be addressed if it is an issue. And indeed this cacheline bouncing has shown up on large SGI systems. There was a situation where an Altix system was essentially livelocked tearing down ZERO_PAGE pagetables when an HPC app aborted during startup. This situation can be avoided in userspace, but it does highlight the potential scalability problem with refcounting ZERO_PAGE, and corner cases where it can really hurt (we don't want the system to livelock!). There are several broad ways to fix this problem: 1. add back some special casing to avoid refcounting ZERO_PAGE 2. per-node or per-cpu ZERO_PAGES 3. remove the ZERO_PAGE completely I will argue for 3. The others should also fix the problem, but they result in more complex code than does 3, with little or no real benefit that I can see. Why? Inserting a ZERO_PAGE for anonymous read faults appears to be a false optimisation: if an application is performance critical, it would not be doing many read faults of new memory, or at least it could be expected to write to that memory soon afterwards. If cache or memory use is critical, it should not be working with a significant number of ZERO_PAGEs anyway (a more compact representation of zeroes should be used). As a sanity check -- mesuring on my desktop system, there are never many mappings to the ZERO_PAGE (eg. 2 or 3), thus memory usage here should not increase much without it. When running a make -j4 kernel compile on my dual core system, there are about 1,000 mappings to the ZERO_PAGE created per second, but about 1,000 ZERO_PAGE COW faults per second (less than 1 ZERO_PAGE mapping per second is torn down without being COWed). So removing ZERO_PAGE will save 1,000 page faults per second when running kbuild, while keeping it only saves less than 1 page clearing operation per second. 1 page clear is cheaper than a thousand faults, presumably, so there isn't an obvious loss. Neither the logical argument nor these basic tests give a guarantee of no regressions. However, this is a reasonable opportunity to try to remove the ZERO_PAGE from the pagefault path. If it is found to cause regressions, we can reintroduce it and just avoid refcounting it. The /dev/zero ZERO_PAGE usage and TLB tricks also get nuked. I don't see much use to them except on benchmarks. All other users of ZERO_PAGE are converted just to use ZERO_PAGE(0) for simplicity. We can look at replacing them all and maybe ripping out ZERO_PAGE completely when we are more satisfied with this solution. Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus "snif" Torvalds <torvalds@linux-foundation.org>
2007-10-16 08:24:40 +00:00
goto release;
ksm: fix oom deadlock There's a now-obvious deadlock in KSM's out-of-memory handling: imagine ksmd or KSM_RUN_UNMERGE handling, holding ksm_thread_mutex, trying to allocate a page to break KSM in an mm which becomes the OOM victim (quite likely in the unmerge case): it's killed and goes to exit, and hangs there waiting to acquire ksm_thread_mutex. Clearly we must not require ksm_thread_mutex in __ksm_exit, simple though that made everything else: perhaps use mmap_sem somehow? And part of the answer lies in the comments on unmerge_ksm_pages: __ksm_exit should also leave all the rmap_item removal to ksmd. But there's a fundamental problem, that KSM relies upon mmap_sem to guarantee the consistency of the mm it's dealing with, yet exit_mmap tears down an mm without taking mmap_sem. And bumping mm_users won't help at all, that just ensures that the pages the OOM killer assumes are on their way to being freed will not be freed. The best answer seems to be, to move the ksm_exit callout from just before exit_mmap, to the middle of exit_mmap: after the mm's pages have been freed (if the mmu_gather is flushed), but before its page tables and vma structures have been freed; and down_write,up_write mmap_sem there to serialize with KSM's own reliance on mmap_sem. But KSM then needs to be careful, whenever it downs mmap_sem, to check that the mm is not already exiting: there's a danger of using find_vma on a layout that's being torn apart, or writing into page tables which have been freed for reuse; and even do_anonymous_page and __do_fault need to check they're not being called by break_ksm to reinstate a pte after zap_pte_range has zapped that page table. Though it might be clearer to add an exiting flag, set while holding mmap_sem in __ksm_exit, that wouldn't cover the issue of reinstating a zapped pte. All we need is to check whether mm_users is 0 - but must remember that ksmd may detect that before __ksm_exit is reached. So, ksm_test_exit(mm) added to comment such checks on mm->mm_users. __ksm_exit now has to leave clearing up the rmap_items to ksmd, that needs ksm_thread_mutex; but shift the exiting mm just after the ksm_scan cursor so that it will soon be dealt with. __ksm_enter raise mm_count to hold the mm_struct, ksmd's exit processing (exactly like its processing when it finds all VM_MERGEABLEs unmapped) mmdrop it, similar procedure for KSM_RUN_UNMERGE (which has stopped ksmd). But also give __ksm_exit a fast path: when there's no complication (no rmap_items attached to mm and it's not at the ksm_scan cursor), it can safely do all the exiting work itself. This is not just an optimization: when ksmd is not running, the raised mm_count would otherwise leak mm_structs. Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk> Acked-by: Izik Eidus <ieidus@redhat.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-09-22 00:02:20 +00:00
inc_mm_counter_fast(mm, MM_ANONPAGES);
remove ZERO_PAGE The commit b5810039a54e5babf428e9a1e89fc1940fabff11 contains the note A last caveat: the ZERO_PAGE is now refcounted and managed with rmap (and thus mapcounted and count towards shared rss). These writes to the struct page could cause excessive cacheline bouncing on big systems. There are a number of ways this could be addressed if it is an issue. And indeed this cacheline bouncing has shown up on large SGI systems. There was a situation where an Altix system was essentially livelocked tearing down ZERO_PAGE pagetables when an HPC app aborted during startup. This situation can be avoided in userspace, but it does highlight the potential scalability problem with refcounting ZERO_PAGE, and corner cases where it can really hurt (we don't want the system to livelock!). There are several broad ways to fix this problem: 1. add back some special casing to avoid refcounting ZERO_PAGE 2. per-node or per-cpu ZERO_PAGES 3. remove the ZERO_PAGE completely I will argue for 3. The others should also fix the problem, but they result in more complex code than does 3, with little or no real benefit that I can see. Why? Inserting a ZERO_PAGE for anonymous read faults appears to be a false optimisation: if an application is performance critical, it would not be doing many read faults of new memory, or at least it could be expected to write to that memory soon afterwards. If cache or memory use is critical, it should not be working with a significant number of ZERO_PAGEs anyway (a more compact representation of zeroes should be used). As a sanity check -- mesuring on my desktop system, there are never many mappings to the ZERO_PAGE (eg. 2 or 3), thus memory usage here should not increase much without it. When running a make -j4 kernel compile on my dual core system, there are about 1,000 mappings to the ZERO_PAGE created per second, but about 1,000 ZERO_PAGE COW faults per second (less than 1 ZERO_PAGE mapping per second is torn down without being COWed). So removing ZERO_PAGE will save 1,000 page faults per second when running kbuild, while keeping it only saves less than 1 page clearing operation per second. 1 page clear is cheaper than a thousand faults, presumably, so there isn't an obvious loss. Neither the logical argument nor these basic tests give a guarantee of no regressions. However, this is a reasonable opportunity to try to remove the ZERO_PAGE from the pagefault path. If it is found to cause regressions, we can reintroduce it and just avoid refcounting it. The /dev/zero ZERO_PAGE usage and TLB tricks also get nuked. I don't see much use to them except on benchmarks. All other users of ZERO_PAGE are converted just to use ZERO_PAGE(0) for simplicity. We can look at replacing them all and maybe ripping out ZERO_PAGE completely when we are more satisfied with this solution. Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus "snif" Torvalds <torvalds@linux-foundation.org>
2007-10-16 08:24:40 +00:00
page_add_new_anon_rmap(page, vma, address);
mm: reinstate ZERO_PAGE KAMEZAWA Hiroyuki has observed customers of earlier kernels taking advantage of the ZERO_PAGE: which we stopped do_anonymous_page() from using in 2.6.24. And there were a couple of regression reports on LKML. Following suggestions from Linus, reinstate do_anonymous_page() use of the ZERO_PAGE; but this time avoid dirtying its struct page cacheline with (map)count updates - let vm_normal_page() regard it as abnormal. Use it only on arches which __HAVE_ARCH_PTE_SPECIAL (x86, s390, sh32, most powerpc): that's not essential, but minimizes additional branches (keeping them in the unlikely pte_special case); and incidentally excludes mips (some models of which needed eight colours of ZERO_PAGE to avoid costly exceptions). Don't be fanatical about avoiding ZERO_PAGE updates: get_user_pages() callers won't want to make exceptions for it, so increment its count there. Changes to mlock and migration? happily seems not needed. In most places it's quicker to check pfn than struct page address: prepare a __read_mostly zero_pfn for that. Does get_dump_page() still need its ZERO_PAGE check? probably not, but keep it anyway. Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk> Acked-by: Rik van Riel <riel@redhat.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Minchan Kim <minchan.kim@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-09-22 00:03:30 +00:00
setpte:
set_pte_at(mm, address, page_table, entry);
/* No need to invalidate - it was non-present before */
update_mmu_cache(vma, address, page_table);
unlock:
pte_unmap_unlock(page_table, ptl);
mm: fault feedback #2 This patch completes Linus's wish that the fault return codes be made into bit flags, which I agree makes everything nicer. This requires requires all handle_mm_fault callers to be modified (possibly the modifications should go further and do things like fault accounting in handle_mm_fault -- however that would be for another patch). [akpm@linux-foundation.org: fix alpha build] [akpm@linux-foundation.org: fix s390 build] [akpm@linux-foundation.org: fix sparc build] [akpm@linux-foundation.org: fix sparc64 build] [akpm@linux-foundation.org: fix ia64 build] Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Richard Henderson <rth@twiddle.net> Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Ian Molton <spyro@f2s.com> Cc: Bryan Wu <bryan.wu@analog.com> Cc: Mikael Starvik <starvik@axis.com> Cc: David Howells <dhowells@redhat.com> Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Hirokazu Takata <takata@linux-m32r.org> Cc: Geert Uytterhoeven <geert@linux-m68k.org> Cc: Roman Zippel <zippel@linux-m68k.org> Cc: Greg Ungerer <gerg@uclinux.org> Cc: Matthew Wilcox <willy@debian.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Kazumoto Kojima <kkojima@rr.iij4u.or.jp> Cc: Richard Curnow <rc@rc0.org.uk> Cc: William Lee Irwin III <wli@holomorphy.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Jeff Dike <jdike@addtoit.com> Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it> Cc: Miles Bader <uclinux-v850@lsi.nec.co.jp> Cc: Chris Zankel <chris@zankel.net> Acked-by: Kyle McMartin <kyle@mcmartin.ca> Acked-by: Haavard Skinnemoen <hskinnemoen@atmel.com> Acked-by: Ralf Baechle <ralf@linux-mips.org> Acked-by: Andi Kleen <ak@muc.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> [ Still apparently needs some ARM and PPC loving - Linus ] Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:47:05 +00:00
return 0;
release:
Memory controller: memory accounting Add the accounting hooks. The accounting is carried out for RSS and Page Cache (unmapped) pages. There is now a common limit and accounting for both. The RSS accounting is accounted at page_add_*_rmap() and page_remove_rmap() time. Page cache is accounted at add_to_page_cache(), __delete_from_page_cache(). Swap cache is also accounted for. Each page's page_cgroup is protected with the last bit of the page_cgroup pointer, this makes handling of race conditions involving simultaneous mappings of a page easier. A reference count is kept in the page_cgroup to deal with cases where a page might be unmapped from the RSS of all tasks, but still lives in the page cache. Credits go to Vaidyanathan Srinivasan for helping with reference counting work of the page cgroup. Almost all of the page cache accounting code has help from Vaidyanathan Srinivasan. [hugh@veritas.com: fix swapoff breakage] [akpm@linux-foundation.org: fix locking] Signed-off-by: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com> Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com> Cc: Pavel Emelianov <xemul@openvz.org> Cc: Paul Menage <menage@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Kirill Korotaev <dev@sw.ru> Cc: Herbert Poetzl <herbert@13thfloor.at> Cc: David Rientjes <rientjes@google.com> Cc: <Valdis.Kletnieks@vt.edu> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-07 08:13:53 +00:00
mem_cgroup_uncharge_page(page);
page_cache_release(page);
goto unlock;
Memory controller: memory accounting Add the accounting hooks. The accounting is carried out for RSS and Page Cache (unmapped) pages. There is now a common limit and accounting for both. The RSS accounting is accounted at page_add_*_rmap() and page_remove_rmap() time. Page cache is accounted at add_to_page_cache(), __delete_from_page_cache(). Swap cache is also accounted for. Each page's page_cgroup is protected with the last bit of the page_cgroup pointer, this makes handling of race conditions involving simultaneous mappings of a page easier. A reference count is kept in the page_cgroup to deal with cases where a page might be unmapped from the RSS of all tasks, but still lives in the page cache. Credits go to Vaidyanathan Srinivasan for helping with reference counting work of the page cgroup. Almost all of the page cache accounting code has help from Vaidyanathan Srinivasan. [hugh@veritas.com: fix swapoff breakage] [akpm@linux-foundation.org: fix locking] Signed-off-by: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com> Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com> Cc: Pavel Emelianov <xemul@openvz.org> Cc: Paul Menage <menage@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Kirill Korotaev <dev@sw.ru> Cc: Herbert Poetzl <herbert@13thfloor.at> Cc: David Rientjes <rientjes@google.com> Cc: <Valdis.Kletnieks@vt.edu> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-07 08:13:53 +00:00
oom_free_page:
page_cache_release(page);
oom:
return VM_FAULT_OOM;
}
/*
mm: merge populate and nopage into fault (fixes nonlinear) Nonlinear mappings are (AFAIKS) simply a virtual memory concept that encodes the virtual address -> file offset differently from linear mappings. ->populate is a layering violation because the filesystem/pagecache code should need to know anything about the virtual memory mapping. The hitch here is that the ->nopage handler didn't pass down enough information (ie. pgoff). But it is more logical to pass pgoff rather than have the ->nopage function calculate it itself anyway (because that's a similar layering violation). Having the populate handler install the pte itself is likewise a nasty thing to be doing. This patch introduces a new fault handler that replaces ->nopage and ->populate and (later) ->nopfn. Most of the old mechanism is still in place so there is a lot of duplication and nice cleanups that can be removed if everyone switches over. The rationale for doing this in the first place is that nonlinear mappings are subject to the pagefault vs invalidate/truncate race too, and it seemed stupid to duplicate the synchronisation logic rather than just consolidate the two. After this patch, MAP_NONBLOCK no longer sets up ptes for pages present in pagecache. Seems like a fringe functionality anyway. NOPAGE_REFAULT is removed. This should be implemented with ->fault, and no users have hit mainline yet. [akpm@linux-foundation.org: cleanup] [randy.dunlap@oracle.com: doc. fixes for readahead] [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:59 +00:00
* __do_fault() tries to create a new page mapping. It aggressively
* tries to share with existing pages, but makes a separate copy if
mm: merge populate and nopage into fault (fixes nonlinear) Nonlinear mappings are (AFAIKS) simply a virtual memory concept that encodes the virtual address -> file offset differently from linear mappings. ->populate is a layering violation because the filesystem/pagecache code should need to know anything about the virtual memory mapping. The hitch here is that the ->nopage handler didn't pass down enough information (ie. pgoff). But it is more logical to pass pgoff rather than have the ->nopage function calculate it itself anyway (because that's a similar layering violation). Having the populate handler install the pte itself is likewise a nasty thing to be doing. This patch introduces a new fault handler that replaces ->nopage and ->populate and (later) ->nopfn. Most of the old mechanism is still in place so there is a lot of duplication and nice cleanups that can be removed if everyone switches over. The rationale for doing this in the first place is that nonlinear mappings are subject to the pagefault vs invalidate/truncate race too, and it seemed stupid to duplicate the synchronisation logic rather than just consolidate the two. After this patch, MAP_NONBLOCK no longer sets up ptes for pages present in pagecache. Seems like a fringe functionality anyway. NOPAGE_REFAULT is removed. This should be implemented with ->fault, and no users have hit mainline yet. [akpm@linux-foundation.org: cleanup] [randy.dunlap@oracle.com: doc. fixes for readahead] [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:59 +00:00
* the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
* the next page fault.
*
* As this is called only for pages that do not currently exist, we
* do not need to flush old virtual caches or the TLB.
*
* We enter with non-exclusive mmap_sem (to exclude vma changes,
* but allow concurrent faults), and pte neither mapped nor locked.
* We return with mmap_sem still held, but pte unmapped and unlocked.
*/
mm: merge populate and nopage into fault (fixes nonlinear) Nonlinear mappings are (AFAIKS) simply a virtual memory concept that encodes the virtual address -> file offset differently from linear mappings. ->populate is a layering violation because the filesystem/pagecache code should need to know anything about the virtual memory mapping. The hitch here is that the ->nopage handler didn't pass down enough information (ie. pgoff). But it is more logical to pass pgoff rather than have the ->nopage function calculate it itself anyway (because that's a similar layering violation). Having the populate handler install the pte itself is likewise a nasty thing to be doing. This patch introduces a new fault handler that replaces ->nopage and ->populate and (later) ->nopfn. Most of the old mechanism is still in place so there is a lot of duplication and nice cleanups that can be removed if everyone switches over. The rationale for doing this in the first place is that nonlinear mappings are subject to the pagefault vs invalidate/truncate race too, and it seemed stupid to duplicate the synchronisation logic rather than just consolidate the two. After this patch, MAP_NONBLOCK no longer sets up ptes for pages present in pagecache. Seems like a fringe functionality anyway. NOPAGE_REFAULT is removed. This should be implemented with ->fault, and no users have hit mainline yet. [akpm@linux-foundation.org: cleanup] [randy.dunlap@oracle.com: doc. fixes for readahead] [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:59 +00:00
static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
unsigned long address, pmd_t *pmd,
mm: merge populate and nopage into fault (fixes nonlinear) Nonlinear mappings are (AFAIKS) simply a virtual memory concept that encodes the virtual address -> file offset differently from linear mappings. ->populate is a layering violation because the filesystem/pagecache code should need to know anything about the virtual memory mapping. The hitch here is that the ->nopage handler didn't pass down enough information (ie. pgoff). But it is more logical to pass pgoff rather than have the ->nopage function calculate it itself anyway (because that's a similar layering violation). Having the populate handler install the pte itself is likewise a nasty thing to be doing. This patch introduces a new fault handler that replaces ->nopage and ->populate and (later) ->nopfn. Most of the old mechanism is still in place so there is a lot of duplication and nice cleanups that can be removed if everyone switches over. The rationale for doing this in the first place is that nonlinear mappings are subject to the pagefault vs invalidate/truncate race too, and it seemed stupid to duplicate the synchronisation logic rather than just consolidate the two. After this patch, MAP_NONBLOCK no longer sets up ptes for pages present in pagecache. Seems like a fringe functionality anyway. NOPAGE_REFAULT is removed. This should be implemented with ->fault, and no users have hit mainline yet. [akpm@linux-foundation.org: cleanup] [randy.dunlap@oracle.com: doc. fixes for readahead] [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:59 +00:00
pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
{
pte_t *page_table;
spinlock_t *ptl;
struct page *page;
mm: preallocate page before lock_page() at filemap COW Currently we are keeping faulted page locked throughout whole __do_fault call (except for page_mkwrite code path) after calling file system's fault code. If we do early COW, we allocate a new page which has to be charged for a memcg (mem_cgroup_newpage_charge). This function, however, might block for unbounded amount of time if memcg oom killer is disabled or fork-bomb is running because the only way out of the OOM situation is either an external event or OOM-situation fix. In the end we are keeping the faulted page locked and blocking other processes from faulting it in which is not good at all because we are basically punishing potentially an unrelated process for OOM condition in a different group (I have seen stuck system because of ld-2.11.1.so being locked). We can do test easily. % cgcreate -g memory:A % cgset -r memory.limit_in_bytes=64M A % cgset -r memory.memsw.limit_in_bytes=64M A % cd kernel_dir; cgexec -g memory:A make -j Then, the whole system will live-locked until you kill 'make -j' by hands (or push reboot...) This is because some important page in a a shared library are locked. Considering again, the new page is not necessary to be allocated with lock_page() held. And usual page allocation may dive into long memory reclaim loop with holding lock_page() and can cause very long latency. There are 3 ways. 1. do allocation/charge before lock_page() Pros. - simple and can handle page allocation in the same manner. This will reduce holding time of lock_page() in general. Cons. - we do page allocation even if ->fault() returns error. 2. do charge after unlock_page(). Even if charge fails, it's just OOM. Pros. - no impact to non-memcg path. Cons. - implemenation requires special cares of LRU and we need to modify page_add_new_anon_rmap()... 3. do unlock->charge->lock again method. Pros. - no impact to non-memcg path. Cons. - This may kill LOCK_PAGE_RETRY optimization. We need to release lock and get it again... This patch moves "charge" and memory allocation for COW page before lock_page(). Then, we can avoid scanning LRU with holding a lock on a page and latency under lock_page() will be reduced. Then, above livelock disappears. [akpm@linux-foundation.org: fix code layout] Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Reported-by: Lutz Vieweg <lvml@5t9.de> Original-idea-by: Michal Hocko <mhocko@suse.cz> Cc: Michal Hocko <mhocko@suse.cz> Cc: Ying Han <yinghan@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-07-26 00:12:27 +00:00
struct page *cow_page;
pte_t entry;
int anon = 0;
[PATCH] mm: tracking shared dirty pages Tracking of dirty pages in shared writeable mmap()s. The idea is simple: write protect clean shared writeable pages, catch the write-fault, make writeable and set dirty. On page write-back clean all the PTE dirty bits and write protect them once again. The implementation is a tad harder, mainly because the default backing_dev_info capabilities were too loosely maintained. Hence it is not enough to test the backing_dev_info for cap_account_dirty. The current heuristic is as follows, a VMA is eligible when: - its shared writeable (vm_flags & (VM_WRITE|VM_SHARED)) == (VM_WRITE|VM_SHARED) - it is not a 'special' mapping (vm_flags & (VM_PFNMAP|VM_INSERTPAGE)) == 0 - the backing_dev_info is cap_account_dirty mapping_cap_account_dirty(vma->vm_file->f_mapping) - f_op->mmap() didn't change the default page protection Page from remap_pfn_range() are explicitly excluded because their COW semantics are already horrid enough (see vm_normal_page() in do_wp_page()) and because they don't have a backing store anyway. mprotect() is taught about the new behaviour as well. However it overrides the last condition. Cleaning the pages on write-back is done with page_mkclean() a new rmap call. It can be called on any page, but is currently only implemented for mapped pages, if the page is found the be of a VMA that accounts dirty pages it will also wrprotect the PTE. Finally, in fs/buffers.c:try_to_free_buffers(); remove clear_page_dirty() from under ->private_lock. This seems to be safe, since ->private_lock is used to serialize access to the buffers, not the page itself. This is needed because clear_page_dirty() will call into page_mkclean() and would thereby violate locking order. [dhowells@redhat.com: Provide a page_mkclean() implementation for NOMMU] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-26 06:30:57 +00:00
struct page *dirty_page = NULL;
struct vm_fault vmf;
int ret;
int page_mkwrite = 0;
mm: merge populate and nopage into fault (fixes nonlinear) Nonlinear mappings are (AFAIKS) simply a virtual memory concept that encodes the virtual address -> file offset differently from linear mappings. ->populate is a layering violation because the filesystem/pagecache code should need to know anything about the virtual memory mapping. The hitch here is that the ->nopage handler didn't pass down enough information (ie. pgoff). But it is more logical to pass pgoff rather than have the ->nopage function calculate it itself anyway (because that's a similar layering violation). Having the populate handler install the pte itself is likewise a nasty thing to be doing. This patch introduces a new fault handler that replaces ->nopage and ->populate and (later) ->nopfn. Most of the old mechanism is still in place so there is a lot of duplication and nice cleanups that can be removed if everyone switches over. The rationale for doing this in the first place is that nonlinear mappings are subject to the pagefault vs invalidate/truncate race too, and it seemed stupid to duplicate the synchronisation logic rather than just consolidate the two. After this patch, MAP_NONBLOCK no longer sets up ptes for pages present in pagecache. Seems like a fringe functionality anyway. NOPAGE_REFAULT is removed. This should be implemented with ->fault, and no users have hit mainline yet. [akpm@linux-foundation.org: cleanup] [randy.dunlap@oracle.com: doc. fixes for readahead] [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:59 +00:00
mm: preallocate page before lock_page() at filemap COW Currently we are keeping faulted page locked throughout whole __do_fault call (except for page_mkwrite code path) after calling file system's fault code. If we do early COW, we allocate a new page which has to be charged for a memcg (mem_cgroup_newpage_charge). This function, however, might block for unbounded amount of time if memcg oom killer is disabled or fork-bomb is running because the only way out of the OOM situation is either an external event or OOM-situation fix. In the end we are keeping the faulted page locked and blocking other processes from faulting it in which is not good at all because we are basically punishing potentially an unrelated process for OOM condition in a different group (I have seen stuck system because of ld-2.11.1.so being locked). We can do test easily. % cgcreate -g memory:A % cgset -r memory.limit_in_bytes=64M A % cgset -r memory.memsw.limit_in_bytes=64M A % cd kernel_dir; cgexec -g memory:A make -j Then, the whole system will live-locked until you kill 'make -j' by hands (or push reboot...) This is because some important page in a a shared library are locked. Considering again, the new page is not necessary to be allocated with lock_page() held. And usual page allocation may dive into long memory reclaim loop with holding lock_page() and can cause very long latency. There are 3 ways. 1. do allocation/charge before lock_page() Pros. - simple and can handle page allocation in the same manner. This will reduce holding time of lock_page() in general. Cons. - we do page allocation even if ->fault() returns error. 2. do charge after unlock_page(). Even if charge fails, it's just OOM. Pros. - no impact to non-memcg path. Cons. - implemenation requires special cares of LRU and we need to modify page_add_new_anon_rmap()... 3. do unlock->charge->lock again method. Pros. - no impact to non-memcg path. Cons. - This may kill LOCK_PAGE_RETRY optimization. We need to release lock and get it again... This patch moves "charge" and memory allocation for COW page before lock_page(). Then, we can avoid scanning LRU with holding a lock on a page and latency under lock_page() will be reduced. Then, above livelock disappears. [akpm@linux-foundation.org: fix code layout] Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Reported-by: Lutz Vieweg <lvml@5t9.de> Original-idea-by: Michal Hocko <mhocko@suse.cz> Cc: Michal Hocko <mhocko@suse.cz> Cc: Ying Han <yinghan@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-07-26 00:12:27 +00:00
/*
* If we do COW later, allocate page befor taking lock_page()
* on the file cache page. This will reduce lock holding time.
*/
if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) {
if (unlikely(anon_vma_prepare(vma)))
return VM_FAULT_OOM;
cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
if (!cow_page)
return VM_FAULT_OOM;
if (mem_cgroup_newpage_charge(cow_page, mm, GFP_KERNEL)) {
page_cache_release(cow_page);
return VM_FAULT_OOM;
}
} else
cow_page = NULL;
vmf.virtual_address = (void __user *)(address & PAGE_MASK);
vmf.pgoff = pgoff;
vmf.flags = flags;
vmf.page = NULL;
ret = vma->vm_ops->fault(vma, &vmf);
mm: retry page fault when blocking on disk transfer This change reduces mmap_sem hold times that are caused by waiting for disk transfers when accessing file mapped VMAs. It introduces the VM_FAULT_ALLOW_RETRY flag, which indicates that the call site wants mmap_sem to be released if blocking on a pending disk transfer. In that case, filemap_fault() returns the VM_FAULT_RETRY status bit and do_page_fault() will then re-acquire mmap_sem and retry the page fault. It is expected that the retry will hit the same page which will now be cached, and thus it will complete with a low mmap_sem hold time. Tests: - microbenchmark: thread A mmaps a large file and does random read accesses to the mmaped area - achieves about 55 iterations/s. Thread B does mmap/munmap in a loop at a separate location - achieves 55 iterations/s before, 15000 iterations/s after. - We are seeing related effects in some applications in house, which show significant performance regressions when running without this change. [akpm@linux-foundation.org: fix warning & crash] Signed-off-by: Michel Lespinasse <walken@google.com> Acked-by: Rik van Riel <riel@redhat.com> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Reviewed-by: Wu Fengguang <fengguang.wu@intel.com> Cc: Ying Han <yinghan@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Ingo Molnar <mingo@elte.hu> Cc: Thomas Gleixner <tglx@linutronix.de> Acked-by: "H. Peter Anvin" <hpa@zytor.com> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-26 21:21:57 +00:00
if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE |
VM_FAULT_RETRY)))
mm: preallocate page before lock_page() at filemap COW Currently we are keeping faulted page locked throughout whole __do_fault call (except for page_mkwrite code path) after calling file system's fault code. If we do early COW, we allocate a new page which has to be charged for a memcg (mem_cgroup_newpage_charge). This function, however, might block for unbounded amount of time if memcg oom killer is disabled or fork-bomb is running because the only way out of the OOM situation is either an external event or OOM-situation fix. In the end we are keeping the faulted page locked and blocking other processes from faulting it in which is not good at all because we are basically punishing potentially an unrelated process for OOM condition in a different group (I have seen stuck system because of ld-2.11.1.so being locked). We can do test easily. % cgcreate -g memory:A % cgset -r memory.limit_in_bytes=64M A % cgset -r memory.memsw.limit_in_bytes=64M A % cd kernel_dir; cgexec -g memory:A make -j Then, the whole system will live-locked until you kill 'make -j' by hands (or push reboot...) This is because some important page in a a shared library are locked. Considering again, the new page is not necessary to be allocated with lock_page() held. And usual page allocation may dive into long memory reclaim loop with holding lock_page() and can cause very long latency. There are 3 ways. 1. do allocation/charge before lock_page() Pros. - simple and can handle page allocation in the same manner. This will reduce holding time of lock_page() in general. Cons. - we do page allocation even if ->fault() returns error. 2. do charge after unlock_page(). Even if charge fails, it's just OOM. Pros. - no impact to non-memcg path. Cons. - implemenation requires special cares of LRU and we need to modify page_add_new_anon_rmap()... 3. do unlock->charge->lock again method. Pros. - no impact to non-memcg path. Cons. - This may kill LOCK_PAGE_RETRY optimization. We need to release lock and get it again... This patch moves "charge" and memory allocation for COW page before lock_page(). Then, we can avoid scanning LRU with holding a lock on a page and latency under lock_page() will be reduced. Then, above livelock disappears. [akpm@linux-foundation.org: fix code layout] Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Reported-by: Lutz Vieweg <lvml@5t9.de> Original-idea-by: Michal Hocko <mhocko@suse.cz> Cc: Michal Hocko <mhocko@suse.cz> Cc: Ying Han <yinghan@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-07-26 00:12:27 +00:00
goto uncharge_out;
if (unlikely(PageHWPoison(vmf.page))) {
if (ret & VM_FAULT_LOCKED)
unlock_page(vmf.page);
mm: preallocate page before lock_page() at filemap COW Currently we are keeping faulted page locked throughout whole __do_fault call (except for page_mkwrite code path) after calling file system's fault code. If we do early COW, we allocate a new page which has to be charged for a memcg (mem_cgroup_newpage_charge). This function, however, might block for unbounded amount of time if memcg oom killer is disabled or fork-bomb is running because the only way out of the OOM situation is either an external event or OOM-situation fix. In the end we are keeping the faulted page locked and blocking other processes from faulting it in which is not good at all because we are basically punishing potentially an unrelated process for OOM condition in a different group (I have seen stuck system because of ld-2.11.1.so being locked). We can do test easily. % cgcreate -g memory:A % cgset -r memory.limit_in_bytes=64M A % cgset -r memory.memsw.limit_in_bytes=64M A % cd kernel_dir; cgexec -g memory:A make -j Then, the whole system will live-locked until you kill 'make -j' by hands (or push reboot...) This is because some important page in a a shared library are locked. Considering again, the new page is not necessary to be allocated with lock_page() held. And usual page allocation may dive into long memory reclaim loop with holding lock_page() and can cause very long latency. There are 3 ways. 1. do allocation/charge before lock_page() Pros. - simple and can handle page allocation in the same manner. This will reduce holding time of lock_page() in general. Cons. - we do page allocation even if ->fault() returns error. 2. do charge after unlock_page(). Even if charge fails, it's just OOM. Pros. - no impact to non-memcg path. Cons. - implemenation requires special cares of LRU and we need to modify page_add_new_anon_rmap()... 3. do unlock->charge->lock again method. Pros. - no impact to non-memcg path. Cons. - This may kill LOCK_PAGE_RETRY optimization. We need to release lock and get it again... This patch moves "charge" and memory allocation for COW page before lock_page(). Then, we can avoid scanning LRU with holding a lock on a page and latency under lock_page() will be reduced. Then, above livelock disappears. [akpm@linux-foundation.org: fix code layout] Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Reported-by: Lutz Vieweg <lvml@5t9.de> Original-idea-by: Michal Hocko <mhocko@suse.cz> Cc: Michal Hocko <mhocko@suse.cz> Cc: Ying Han <yinghan@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-07-26 00:12:27 +00:00
ret = VM_FAULT_HWPOISON;
goto uncharge_out;
}
mm: fix fault vs invalidate race for linear mappings Fix the race between invalidate_inode_pages and do_no_page. Andrea Arcangeli identified a subtle race between invalidation of pages from pagecache with userspace mappings, and do_no_page. The issue is that invalidation has to shoot down all mappings to the page, before it can be discarded from the pagecache. Between shooting down ptes to a particular page, and actually dropping the struct page from the pagecache, do_no_page from any process might fault on that page and establish a new mapping to the page just before it gets discarded from the pagecache. The most common case where such invalidation is used is in file truncation. This case was catered for by doing a sort of open-coded seqlock between the file's i_size, and its truncate_count. Truncation will decrease i_size, then increment truncate_count before unmapping userspace pages; do_no_page will read truncate_count, then find the page if it is within i_size, and then check truncate_count under the page table lock and back out and retry if it had subsequently been changed (ptl will serialise against unmapping, and ensure a potentially updated truncate_count is actually visible). Complexity and documentation issues aside, the locking protocol fails in the case where we would like to invalidate pagecache inside i_size. do_no_page can come in anytime and filemap_nopage is not aware of the invalidation in progress (as it is when it is outside i_size). The end result is that dangling (->mapping == NULL) pages that appear to be from a particular file may be mapped into userspace with nonsense data. Valid mappings to the same place will see a different page. Andrea implemented two working fixes, one using a real seqlock, another using a page->flags bit. He also proposed using the page lock in do_no_page, but that was initially considered too heavyweight. However, it is not a global or per-file lock, and the page cacheline is modified in do_no_page to increment _count and _mapcount anyway, so a further modification should not be a large performance hit. Scalability is not an issue. This patch implements this latter approach. ->nopage implementations return with the page locked if it is possible for their underlying file to be invalidated (in that case, they must set a special vm_flags bit to indicate so). do_no_page only unlocks the page after setting up the mapping completely. invalidation is excluded because it holds the page lock during invalidation of each page (and ensures that the page is not mapped while holding the lock). This also allows significant simplifications in do_no_page, because we have the page locked in the right place in the pagecache from the start. Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:57 +00:00
/*
* For consistency in subsequent calls, make the faulted page always
mm: fix fault vs invalidate race for linear mappings Fix the race between invalidate_inode_pages and do_no_page. Andrea Arcangeli identified a subtle race between invalidation of pages from pagecache with userspace mappings, and do_no_page. The issue is that invalidation has to shoot down all mappings to the page, before it can be discarded from the pagecache. Between shooting down ptes to a particular page, and actually dropping the struct page from the pagecache, do_no_page from any process might fault on that page and establish a new mapping to the page just before it gets discarded from the pagecache. The most common case where such invalidation is used is in file truncation. This case was catered for by doing a sort of open-coded seqlock between the file's i_size, and its truncate_count. Truncation will decrease i_size, then increment truncate_count before unmapping userspace pages; do_no_page will read truncate_count, then find the page if it is within i_size, and then check truncate_count under the page table lock and back out and retry if it had subsequently been changed (ptl will serialise against unmapping, and ensure a potentially updated truncate_count is actually visible). Complexity and documentation issues aside, the locking protocol fails in the case where we would like to invalidate pagecache inside i_size. do_no_page can come in anytime and filemap_nopage is not aware of the invalidation in progress (as it is when it is outside i_size). The end result is that dangling (->mapping == NULL) pages that appear to be from a particular file may be mapped into userspace with nonsense data. Valid mappings to the same place will see a different page. Andrea implemented two working fixes, one using a real seqlock, another using a page->flags bit. He also proposed using the page lock in do_no_page, but that was initially considered too heavyweight. However, it is not a global or per-file lock, and the page cacheline is modified in do_no_page to increment _count and _mapcount anyway, so a further modification should not be a large performance hit. Scalability is not an issue. This patch implements this latter approach. ->nopage implementations return with the page locked if it is possible for their underlying file to be invalidated (in that case, they must set a special vm_flags bit to indicate so). do_no_page only unlocks the page after setting up the mapping completely. invalidation is excluded because it holds the page lock during invalidation of each page (and ensures that the page is not mapped while holding the lock). This also allows significant simplifications in do_no_page, because we have the page locked in the right place in the pagecache from the start. Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:57 +00:00
* locked.
*/
mm: fault feedback #2 This patch completes Linus's wish that the fault return codes be made into bit flags, which I agree makes everything nicer. This requires requires all handle_mm_fault callers to be modified (possibly the modifications should go further and do things like fault accounting in handle_mm_fault -- however that would be for another patch). [akpm@linux-foundation.org: fix alpha build] [akpm@linux-foundation.org: fix s390 build] [akpm@linux-foundation.org: fix sparc build] [akpm@linux-foundation.org: fix sparc64 build] [akpm@linux-foundation.org: fix ia64 build] Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Richard Henderson <rth@twiddle.net> Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Ian Molton <spyro@f2s.com> Cc: Bryan Wu <bryan.wu@analog.com> Cc: Mikael Starvik <starvik@axis.com> Cc: David Howells <dhowells@redhat.com> Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Hirokazu Takata <takata@linux-m32r.org> Cc: Geert Uytterhoeven <geert@linux-m68k.org> Cc: Roman Zippel <zippel@linux-m68k.org> Cc: Greg Ungerer <gerg@uclinux.org> Cc: Matthew Wilcox <willy@debian.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Kazumoto Kojima <kkojima@rr.iij4u.or.jp> Cc: Richard Curnow <rc@rc0.org.uk> Cc: William Lee Irwin III <wli@holomorphy.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Jeff Dike <jdike@addtoit.com> Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it> Cc: Miles Bader <uclinux-v850@lsi.nec.co.jp> Cc: Chris Zankel <chris@zankel.net> Acked-by: Kyle McMartin <kyle@mcmartin.ca> Acked-by: Haavard Skinnemoen <hskinnemoen@atmel.com> Acked-by: Ralf Baechle <ralf@linux-mips.org> Acked-by: Andi Kleen <ak@muc.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> [ Still apparently needs some ARM and PPC loving - Linus ] Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:47:05 +00:00
if (unlikely(!(ret & VM_FAULT_LOCKED)))
lock_page(vmf.page);
mm: merge populate and nopage into fault (fixes nonlinear) Nonlinear mappings are (AFAIKS) simply a virtual memory concept that encodes the virtual address -> file offset differently from linear mappings. ->populate is a layering violation because the filesystem/pagecache code should need to know anything about the virtual memory mapping. The hitch here is that the ->nopage handler didn't pass down enough information (ie. pgoff). But it is more logical to pass pgoff rather than have the ->nopage function calculate it itself anyway (because that's a similar layering violation). Having the populate handler install the pte itself is likewise a nasty thing to be doing. This patch introduces a new fault handler that replaces ->nopage and ->populate and (later) ->nopfn. Most of the old mechanism is still in place so there is a lot of duplication and nice cleanups that can be removed if everyone switches over. The rationale for doing this in the first place is that nonlinear mappings are subject to the pagefault vs invalidate/truncate race too, and it seemed stupid to duplicate the synchronisation logic rather than just consolidate the two. After this patch, MAP_NONBLOCK no longer sets up ptes for pages present in pagecache. Seems like a fringe functionality anyway. NOPAGE_REFAULT is removed. This should be implemented with ->fault, and no users have hit mainline yet. [akpm@linux-foundation.org: cleanup] [randy.dunlap@oracle.com: doc. fixes for readahead] [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:59 +00:00
else
VM_BUG_ON(!PageLocked(vmf.page));
mm: fix fault vs invalidate race for linear mappings Fix the race between invalidate_inode_pages and do_no_page. Andrea Arcangeli identified a subtle race between invalidation of pages from pagecache with userspace mappings, and do_no_page. The issue is that invalidation has to shoot down all mappings to the page, before it can be discarded from the pagecache. Between shooting down ptes to a particular page, and actually dropping the struct page from the pagecache, do_no_page from any process might fault on that page and establish a new mapping to the page just before it gets discarded from the pagecache. The most common case where such invalidation is used is in file truncation. This case was catered for by doing a sort of open-coded seqlock between the file's i_size, and its truncate_count. Truncation will decrease i_size, then increment truncate_count before unmapping userspace pages; do_no_page will read truncate_count, then find the page if it is within i_size, and then check truncate_count under the page table lock and back out and retry if it had subsequently been changed (ptl will serialise against unmapping, and ensure a potentially updated truncate_count is actually visible). Complexity and documentation issues aside, the locking protocol fails in the case where we would like to invalidate pagecache inside i_size. do_no_page can come in anytime and filemap_nopage is not aware of the invalidation in progress (as it is when it is outside i_size). The end result is that dangling (->mapping == NULL) pages that appear to be from a particular file may be mapped into userspace with nonsense data. Valid mappings to the same place will see a different page. Andrea implemented two working fixes, one using a real seqlock, another using a page->flags bit. He also proposed using the page lock in do_no_page, but that was initially considered too heavyweight. However, it is not a global or per-file lock, and the page cacheline is modified in do_no_page to increment _count and _mapcount anyway, so a further modification should not be a large performance hit. Scalability is not an issue. This patch implements this latter approach. ->nopage implementations return with the page locked if it is possible for their underlying file to be invalidated (in that case, they must set a special vm_flags bit to indicate so). do_no_page only unlocks the page after setting up the mapping completely. invalidation is excluded because it holds the page lock during invalidation of each page (and ensures that the page is not mapped while holding the lock). This also allows significant simplifications in do_no_page, because we have the page locked in the right place in the pagecache from the start. Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:57 +00:00
/*
* Should we do an early C-O-W break?
*/
page = vmf.page;
mm: merge populate and nopage into fault (fixes nonlinear) Nonlinear mappings are (AFAIKS) simply a virtual memory concept that encodes the virtual address -> file offset differently from linear mappings. ->populate is a layering violation because the filesystem/pagecache code should need to know anything about the virtual memory mapping. The hitch here is that the ->nopage handler didn't pass down enough information (ie. pgoff). But it is more logical to pass pgoff rather than have the ->nopage function calculate it itself anyway (because that's a similar layering violation). Having the populate handler install the pte itself is likewise a nasty thing to be doing. This patch introduces a new fault handler that replaces ->nopage and ->populate and (later) ->nopfn. Most of the old mechanism is still in place so there is a lot of duplication and nice cleanups that can be removed if everyone switches over. The rationale for doing this in the first place is that nonlinear mappings are subject to the pagefault vs invalidate/truncate race too, and it seemed stupid to duplicate the synchronisation logic rather than just consolidate the two. After this patch, MAP_NONBLOCK no longer sets up ptes for pages present in pagecache. Seems like a fringe functionality anyway. NOPAGE_REFAULT is removed. This should be implemented with ->fault, and no users have hit mainline yet. [akpm@linux-foundation.org: cleanup] [randy.dunlap@oracle.com: doc. fixes for readahead] [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:59 +00:00
if (flags & FAULT_FLAG_WRITE) {
[PATCH] add page_mkwrite() vm_operations method Add a new VMA operation to notify a filesystem or other driver about the MMU generating a fault because userspace attempted to write to a page mapped through a read-only PTE. This facility permits the filesystem or driver to: (*) Implement storage allocation/reservation on attempted write, and so to deal with problems such as ENOSPC more gracefully (perhaps by generating SIGBUS). (*) Delay making the page writable until the contents have been written to a backing cache. This is useful for NFS/AFS when using FS-Cache/CacheFS. It permits the filesystem to have some guarantee about the state of the cache. (*) Account and limit number of dirty pages. This is one piece of the puzzle needed to make shared writable mapping work safely in FUSE. Needed by cachefs (Or is it cachefiles? Or fscache? <head spins>). At least four other groups have stated an interest in it or a desire to use the functionality it provides: FUSE, OCFS2, NTFS and JFFS2. Also, things like EXT3 really ought to use it to deal with the case of shared-writable mmap encountering ENOSPC before we permit the page to be dirtied. From: Peter Zijlstra <a.p.zijlstra@chello.nl> get_user_pages(.write=1, .force=1) can generate COW hits on read-only shared mappings, this patch traps those as mkpage_write candidates and fails to handle them the old way. Signed-off-by: David Howells <dhowells@redhat.com> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Joel Becker <Joel.Becker@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Cc: Anton Altaparmakov <aia21@cantab.net> Cc: David Woodhouse <dwmw2@infradead.org> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 09:03:43 +00:00
if (!(vma->vm_flags & VM_SHARED)) {
mm: preallocate page before lock_page() at filemap COW Currently we are keeping faulted page locked throughout whole __do_fault call (except for page_mkwrite code path) after calling file system's fault code. If we do early COW, we allocate a new page which has to be charged for a memcg (mem_cgroup_newpage_charge). This function, however, might block for unbounded amount of time if memcg oom killer is disabled or fork-bomb is running because the only way out of the OOM situation is either an external event or OOM-situation fix. In the end we are keeping the faulted page locked and blocking other processes from faulting it in which is not good at all because we are basically punishing potentially an unrelated process for OOM condition in a different group (I have seen stuck system because of ld-2.11.1.so being locked). We can do test easily. % cgcreate -g memory:A % cgset -r memory.limit_in_bytes=64M A % cgset -r memory.memsw.limit_in_bytes=64M A % cd kernel_dir; cgexec -g memory:A make -j Then, the whole system will live-locked until you kill 'make -j' by hands (or push reboot...) This is because some important page in a a shared library are locked. Considering again, the new page is not necessary to be allocated with lock_page() held. And usual page allocation may dive into long memory reclaim loop with holding lock_page() and can cause very long latency. There are 3 ways. 1. do allocation/charge before lock_page() Pros. - simple and can handle page allocation in the same manner. This will reduce holding time of lock_page() in general. Cons. - we do page allocation even if ->fault() returns error. 2. do charge after unlock_page(). Even if charge fails, it's just OOM. Pros. - no impact to non-memcg path. Cons. - implemenation requires special cares of LRU and we need to modify page_add_new_anon_rmap()... 3. do unlock->charge->lock again method. Pros. - no impact to non-memcg path. Cons. - This may kill LOCK_PAGE_RETRY optimization. We need to release lock and get it again... This patch moves "charge" and memory allocation for COW page before lock_page(). Then, we can avoid scanning LRU with holding a lock on a page and latency under lock_page() will be reduced. Then, above livelock disappears. [akpm@linux-foundation.org: fix code layout] Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Reported-by: Lutz Vieweg <lvml@5t9.de> Original-idea-by: Michal Hocko <mhocko@suse.cz> Cc: Michal Hocko <mhocko@suse.cz> Cc: Ying Han <yinghan@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-07-26 00:12:27 +00:00
page = cow_page;
mm: merge populate and nopage into fault (fixes nonlinear) Nonlinear mappings are (AFAIKS) simply a virtual memory concept that encodes the virtual address -> file offset differently from linear mappings. ->populate is a layering violation because the filesystem/pagecache code should need to know anything about the virtual memory mapping. The hitch here is that the ->nopage handler didn't pass down enough information (ie. pgoff). But it is more logical to pass pgoff rather than have the ->nopage function calculate it itself anyway (because that's a similar layering violation). Having the populate handler install the pte itself is likewise a nasty thing to be doing. This patch introduces a new fault handler that replaces ->nopage and ->populate and (later) ->nopfn. Most of the old mechanism is still in place so there is a lot of duplication and nice cleanups that can be removed if everyone switches over. The rationale for doing this in the first place is that nonlinear mappings are subject to the pagefault vs invalidate/truncate race too, and it seemed stupid to duplicate the synchronisation logic rather than just consolidate the two. After this patch, MAP_NONBLOCK no longer sets up ptes for pages present in pagecache. Seems like a fringe functionality anyway. NOPAGE_REFAULT is removed. This should be implemented with ->fault, and no users have hit mainline yet. [akpm@linux-foundation.org: cleanup] [randy.dunlap@oracle.com: doc. fixes for readahead] [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:59 +00:00
anon = 1;
copy_user_highpage(page, vmf.page, address, vma);
mm: fix PageUptodate data race After running SetPageUptodate, preceeding stores to the page contents to actually bring it uptodate may not be ordered with the store to set the page uptodate. Therefore, another CPU which checks PageUptodate is true, then reads the page contents can get stale data. Fix this by having an smp_wmb before SetPageUptodate, and smp_rmb after PageUptodate. Many places that test PageUptodate, do so with the page locked, and this would be enough to ensure memory ordering in those places if SetPageUptodate were only called while the page is locked. Unfortunately that is not always the case for some filesystems, but it could be an idea for the future. Also bring the handling of anonymous page uptodateness in line with that of file backed page management, by marking anon pages as uptodate when they _are_ uptodate, rather than when our implementation requires that they be marked as such. Doing allows us to get rid of the smp_wmb's in the page copying functions, which were especially added for anonymous pages for an analogous memory ordering problem. Both file and anonymous pages are handled with the same barriers. FAQ: Q. Why not do this in flush_dcache_page? A. Firstly, flush_dcache_page handles only one side (the smb side) of the ordering protocol; we'd still need smp_rmb somewhere. Secondly, hiding away memory barriers in a completely unrelated function is nasty; at least in the PageUptodate macros, they are located together with (half) the operations involved in the ordering. Thirdly, the smp_wmb is only required when first bringing the page uptodate, wheras flush_dcache_page should be called each time it is written to through the kernel mapping. It is logically the wrong place to put it. Q. Why does this increase my text size / reduce my performance / etc. A. Because it is adding the necessary instructions to eliminate the data-race. Q. Can it be improved? A. Yes, eg. if you were to create a rule that all SetPageUptodate operations run under the page lock, we could avoid the smp_rmb places where PageUptodate is queried under the page lock. Requires audit of all filesystems and at least some would need reworking. That's great you're interested, I'm eagerly awaiting your patches. Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 06:29:34 +00:00
__SetPageUptodate(page);
[PATCH] add page_mkwrite() vm_operations method Add a new VMA operation to notify a filesystem or other driver about the MMU generating a fault because userspace attempted to write to a page mapped through a read-only PTE. This facility permits the filesystem or driver to: (*) Implement storage allocation/reservation on attempted write, and so to deal with problems such as ENOSPC more gracefully (perhaps by generating SIGBUS). (*) Delay making the page writable until the contents have been written to a backing cache. This is useful for NFS/AFS when using FS-Cache/CacheFS. It permits the filesystem to have some guarantee about the state of the cache. (*) Account and limit number of dirty pages. This is one piece of the puzzle needed to make shared writable mapping work safely in FUSE. Needed by cachefs (Or is it cachefiles? Or fscache? <head spins>). At least four other groups have stated an interest in it or a desire to use the functionality it provides: FUSE, OCFS2, NTFS and JFFS2. Also, things like EXT3 really ought to use it to deal with the case of shared-writable mmap encountering ENOSPC before we permit the page to be dirtied. From: Peter Zijlstra <a.p.zijlstra@chello.nl> get_user_pages(.write=1, .force=1) can generate COW hits on read-only shared mappings, this patch traps those as mkpage_write candidates and fails to handle them the old way. Signed-off-by: David Howells <dhowells@redhat.com> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Joel Becker <Joel.Becker@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Cc: Anton Altaparmakov <aia21@cantab.net> Cc: David Woodhouse <dwmw2@infradead.org> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 09:03:43 +00:00
} else {
mm: merge populate and nopage into fault (fixes nonlinear) Nonlinear mappings are (AFAIKS) simply a virtual memory concept that encodes the virtual address -> file offset differently from linear mappings. ->populate is a layering violation because the filesystem/pagecache code should need to know anything about the virtual memory mapping. The hitch here is that the ->nopage handler didn't pass down enough information (ie. pgoff). But it is more logical to pass pgoff rather than have the ->nopage function calculate it itself anyway (because that's a similar layering violation). Having the populate handler install the pte itself is likewise a nasty thing to be doing. This patch introduces a new fault handler that replaces ->nopage and ->populate and (later) ->nopfn. Most of the old mechanism is still in place so there is a lot of duplication and nice cleanups that can be removed if everyone switches over. The rationale for doing this in the first place is that nonlinear mappings are subject to the pagefault vs invalidate/truncate race too, and it seemed stupid to duplicate the synchronisation logic rather than just consolidate the two. After this patch, MAP_NONBLOCK no longer sets up ptes for pages present in pagecache. Seems like a fringe functionality anyway. NOPAGE_REFAULT is removed. This should be implemented with ->fault, and no users have hit mainline yet. [akpm@linux-foundation.org: cleanup] [randy.dunlap@oracle.com: doc. fixes for readahead] [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:59 +00:00
/*
* If the page will be shareable, see if the backing
[PATCH] add page_mkwrite() vm_operations method Add a new VMA operation to notify a filesystem or other driver about the MMU generating a fault because userspace attempted to write to a page mapped through a read-only PTE. This facility permits the filesystem or driver to: (*) Implement storage allocation/reservation on attempted write, and so to deal with problems such as ENOSPC more gracefully (perhaps by generating SIGBUS). (*) Delay making the page writable until the contents have been written to a backing cache. This is useful for NFS/AFS when using FS-Cache/CacheFS. It permits the filesystem to have some guarantee about the state of the cache. (*) Account and limit number of dirty pages. This is one piece of the puzzle needed to make shared writable mapping work safely in FUSE. Needed by cachefs (Or is it cachefiles? Or fscache? <head spins>). At least four other groups have stated an interest in it or a desire to use the functionality it provides: FUSE, OCFS2, NTFS and JFFS2. Also, things like EXT3 really ought to use it to deal with the case of shared-writable mmap encountering ENOSPC before we permit the page to be dirtied. From: Peter Zijlstra <a.p.zijlstra@chello.nl> get_user_pages(.write=1, .force=1) can generate COW hits on read-only shared mappings, this patch traps those as mkpage_write candidates and fails to handle them the old way. Signed-off-by: David Howells <dhowells@redhat.com> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Joel Becker <Joel.Becker@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Cc: Anton Altaparmakov <aia21@cantab.net> Cc: David Woodhouse <dwmw2@infradead.org> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 09:03:43 +00:00
* address space wants to know that the page is about
mm: merge populate and nopage into fault (fixes nonlinear) Nonlinear mappings are (AFAIKS) simply a virtual memory concept that encodes the virtual address -> file offset differently from linear mappings. ->populate is a layering violation because the filesystem/pagecache code should need to know anything about the virtual memory mapping. The hitch here is that the ->nopage handler didn't pass down enough information (ie. pgoff). But it is more logical to pass pgoff rather than have the ->nopage function calculate it itself anyway (because that's a similar layering violation). Having the populate handler install the pte itself is likewise a nasty thing to be doing. This patch introduces a new fault handler that replaces ->nopage and ->populate and (later) ->nopfn. Most of the old mechanism is still in place so there is a lot of duplication and nice cleanups that can be removed if everyone switches over. The rationale for doing this in the first place is that nonlinear mappings are subject to the pagefault vs invalidate/truncate race too, and it seemed stupid to duplicate the synchronisation logic rather than just consolidate the two. After this patch, MAP_NONBLOCK no longer sets up ptes for pages present in pagecache. Seems like a fringe functionality anyway. NOPAGE_REFAULT is removed. This should be implemented with ->fault, and no users have hit mainline yet. [akpm@linux-foundation.org: cleanup] [randy.dunlap@oracle.com: doc. fixes for readahead] [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:59 +00:00
* to become writable
*/
if (vma->vm_ops->page_mkwrite) {
int tmp;
unlock_page(page);
mm: close page_mkwrite races Change page_mkwrite to allow implementations to return with the page locked, and also change it's callers (in page fault paths) to hold the lock until the page is marked dirty. This allows the filesystem to have full control of page dirtying events coming from the VM. Rather than simply hold the page locked over the page_mkwrite call, we call page_mkwrite with the page unlocked and allow callers to return with it locked, so filesystems can avoid LOR conditions with page lock. The problem with the current scheme is this: a filesystem that wants to associate some metadata with a page as long as the page is dirty, will perform this manipulation in its ->page_mkwrite. It currently then must return with the page unlocked and may not hold any other locks (according to existing page_mkwrite convention). In this window, the VM could write out the page, clearing page-dirty. The filesystem has no good way to detect that a dirty pte is about to be attached, so it will happily write out the page, at which point, the filesystem may manipulate the metadata to reflect that the page is no longer dirty. It is not always possible to perform the required metadata manipulation in ->set_page_dirty, because that function cannot block or fail. The filesystem may need to allocate some data structure, for example. And the VM cannot mark the pte dirty before page_mkwrite, because page_mkwrite is allowed to fail, so we must not allow any window where the page could be written to if page_mkwrite does fail. This solution of holding the page locked over the 3 critical operations (page_mkwrite, setting the pte dirty, and finally setting the page dirty) closes out races nicely, preventing page cleaning for writeout being initiated in that window. This provides the filesystem with a strong synchronisation against the VM here. - Sage needs this race closed for ceph filesystem. - Trond for NFS (http://bugzilla.kernel.org/show_bug.cgi?id=12913). - I need it for fsblock. - I suspect other filesystems may need it too (eg. btrfs). - I have converted buffer.c to the new locking. Even simple block allocation under dirty pages might be susceptible to i_size changing under partial page at the end of file (we also have a buffer.c-side problem here, but it cannot be fixed properly without this patch). - Other filesystems (eg. NFS, maybe btrfs) will need to change their page_mkwrite functions themselves. [ This also moves page_mkwrite another step closer to fault, which should eventually allow page_mkwrite to be moved into ->fault, and thus avoiding a filesystem calldown and page lock/unlock cycle in __do_fault. ] [akpm@linux-foundation.org: fix derefs of NULL ->mapping] Cc: Sage Weil <sage@newdream.net> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Valdis Kletnieks <Valdis.Kletnieks@vt.edu> Cc: <stable@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-30 22:08:16 +00:00
vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
if (unlikely(tmp &
(VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
ret = tmp;
mm: close page_mkwrite races Change page_mkwrite to allow implementations to return with the page locked, and also change it's callers (in page fault paths) to hold the lock until the page is marked dirty. This allows the filesystem to have full control of page dirtying events coming from the VM. Rather than simply hold the page locked over the page_mkwrite call, we call page_mkwrite with the page unlocked and allow callers to return with it locked, so filesystems can avoid LOR conditions with page lock. The problem with the current scheme is this: a filesystem that wants to associate some metadata with a page as long as the page is dirty, will perform this manipulation in its ->page_mkwrite. It currently then must return with the page unlocked and may not hold any other locks (according to existing page_mkwrite convention). In this window, the VM could write out the page, clearing page-dirty. The filesystem has no good way to detect that a dirty pte is about to be attached, so it will happily write out the page, at which point, the filesystem may manipulate the metadata to reflect that the page is no longer dirty. It is not always possible to perform the required metadata manipulation in ->set_page_dirty, because that function cannot block or fail. The filesystem may need to allocate some data structure, for example. And the VM cannot mark the pte dirty before page_mkwrite, because page_mkwrite is allowed to fail, so we must not allow any window where the page could be written to if page_mkwrite does fail. This solution of holding the page locked over the 3 critical operations (page_mkwrite, setting the pte dirty, and finally setting the page dirty) closes out races nicely, preventing page cleaning for writeout being initiated in that window. This provides the filesystem with a strong synchronisation against the VM here. - Sage needs this race closed for ceph filesystem. - Trond for NFS (http://bugzilla.kernel.org/show_bug.cgi?id=12913). - I need it for fsblock. - I suspect other filesystems may need it too (eg. btrfs). - I have converted buffer.c to the new locking. Even simple block allocation under dirty pages might be susceptible to i_size changing under partial page at the end of file (we also have a buffer.c-side problem here, but it cannot be fixed properly without this patch). - Other filesystems (eg. NFS, maybe btrfs) will need to change their page_mkwrite functions themselves. [ This also moves page_mkwrite another step closer to fault, which should eventually allow page_mkwrite to be moved into ->fault, and thus avoiding a filesystem calldown and page lock/unlock cycle in __do_fault. ] [akpm@linux-foundation.org: fix derefs of NULL ->mapping] Cc: Sage Weil <sage@newdream.net> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Valdis Kletnieks <Valdis.Kletnieks@vt.edu> Cc: <stable@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-30 22:08:16 +00:00
goto unwritable_page;
}
mm: close page_mkwrite races Change page_mkwrite to allow implementations to return with the page locked, and also change it's callers (in page fault paths) to hold the lock until the page is marked dirty. This allows the filesystem to have full control of page dirtying events coming from the VM. Rather than simply hold the page locked over the page_mkwrite call, we call page_mkwrite with the page unlocked and allow callers to return with it locked, so filesystems can avoid LOR conditions with page lock. The problem with the current scheme is this: a filesystem that wants to associate some metadata with a page as long as the page is dirty, will perform this manipulation in its ->page_mkwrite. It currently then must return with the page unlocked and may not hold any other locks (according to existing page_mkwrite convention). In this window, the VM could write out the page, clearing page-dirty. The filesystem has no good way to detect that a dirty pte is about to be attached, so it will happily write out the page, at which point, the filesystem may manipulate the metadata to reflect that the page is no longer dirty. It is not always possible to perform the required metadata manipulation in ->set_page_dirty, because that function cannot block or fail. The filesystem may need to allocate some data structure, for example. And the VM cannot mark the pte dirty before page_mkwrite, because page_mkwrite is allowed to fail, so we must not allow any window where the page could be written to if page_mkwrite does fail. This solution of holding the page locked over the 3 critical operations (page_mkwrite, setting the pte dirty, and finally setting the page dirty) closes out races nicely, preventing page cleaning for writeout being initiated in that window. This provides the filesystem with a strong synchronisation against the VM here. - Sage needs this race closed for ceph filesystem. - Trond for NFS (http://bugzilla.kernel.org/show_bug.cgi?id=12913). - I need it for fsblock. - I suspect other filesystems may need it too (eg. btrfs). - I have converted buffer.c to the new locking. Even simple block allocation under dirty pages might be susceptible to i_size changing under partial page at the end of file (we also have a buffer.c-side problem here, but it cannot be fixed properly without this patch). - Other filesystems (eg. NFS, maybe btrfs) will need to change their page_mkwrite functions themselves. [ This also moves page_mkwrite another step closer to fault, which should eventually allow page_mkwrite to be moved into ->fault, and thus avoiding a filesystem calldown and page lock/unlock cycle in __do_fault. ] [akpm@linux-foundation.org: fix derefs of NULL ->mapping] Cc: Sage Weil <sage@newdream.net> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Valdis Kletnieks <Valdis.Kletnieks@vt.edu> Cc: <stable@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-30 22:08:16 +00:00
if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
lock_page(page);
if (!page->mapping) {
ret = 0; /* retry the fault */
unlock_page(page);
goto unwritable_page;
}
} else
VM_BUG_ON(!PageLocked(page));
page_mkwrite = 1;
[PATCH] add page_mkwrite() vm_operations method Add a new VMA operation to notify a filesystem or other driver about the MMU generating a fault because userspace attempted to write to a page mapped through a read-only PTE. This facility permits the filesystem or driver to: (*) Implement storage allocation/reservation on attempted write, and so to deal with problems such as ENOSPC more gracefully (perhaps by generating SIGBUS). (*) Delay making the page writable until the contents have been written to a backing cache. This is useful for NFS/AFS when using FS-Cache/CacheFS. It permits the filesystem to have some guarantee about the state of the cache. (*) Account and limit number of dirty pages. This is one piece of the puzzle needed to make shared writable mapping work safely in FUSE. Needed by cachefs (Or is it cachefiles? Or fscache? <head spins>). At least four other groups have stated an interest in it or a desire to use the functionality it provides: FUSE, OCFS2, NTFS and JFFS2. Also, things like EXT3 really ought to use it to deal with the case of shared-writable mmap encountering ENOSPC before we permit the page to be dirtied. From: Peter Zijlstra <a.p.zijlstra@chello.nl> get_user_pages(.write=1, .force=1) can generate COW hits on read-only shared mappings, this patch traps those as mkpage_write candidates and fails to handle them the old way. Signed-off-by: David Howells <dhowells@redhat.com> Cc: Miklos Szeredi <miklos@szeredi.hu> Cc: Joel Becker <Joel.Becker@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Cc: Anton Altaparmakov <aia21@cantab.net> Cc: David Woodhouse <dwmw2@infradead.org> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 09:03:43 +00:00
}
}
mm: merge populate and nopage into fault (fixes nonlinear) Nonlinear mappings are (AFAIKS) simply a virtual memory concept that encodes the virtual address -> file offset differently from linear mappings. ->populate is a layering violation because the filesystem/pagecache code should need to know anything about the virtual memory mapping. The hitch here is that the ->nopage handler didn't pass down enough information (ie. pgoff). But it is more logical to pass pgoff rather than have the ->nopage function calculate it itself anyway (because that's a similar layering violation). Having the populate handler install the pte itself is likewise a nasty thing to be doing. This patch introduces a new fault handler that replaces ->nopage and ->populate and (later) ->nopfn. Most of the old mechanism is still in place so there is a lot of duplication and nice cleanups that can be removed if everyone switches over. The rationale for doing this in the first place is that nonlinear mappings are subject to the pagefault vs invalidate/truncate race too, and it seemed stupid to duplicate the synchronisation logic rather than just consolidate the two. After this patch, MAP_NONBLOCK no longer sets up ptes for pages present in pagecache. Seems like a fringe functionality anyway. NOPAGE_REFAULT is removed. This should be implemented with ->fault, and no users have hit mainline yet. [akpm@linux-foundation.org: cleanup] [randy.dunlap@oracle.com: doc. fixes for readahead] [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:59 +00:00
}
page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
/*
* This silly early PAGE_DIRTY setting removes a race
* due to the bad i386 page protection. But it's valid
* for other architectures too.
*
* Note that if FAULT_FLAG_WRITE is set, we either now have
* an exclusive copy of the page, or this is a shared mapping,
* so we can make it writable and dirty to avoid having to
* handle that later.
*/
/* Only go through if we didn't race with anybody else... */
ksm: fix deadlock with munlock in exit_mmap Rawhide users have reported hang at startup when cryptsetup is run: the same problem can be simply reproduced by running a program int main() { mlockall(MCL_CURRENT | MCL_FUTURE); return 0; } The problem is that exit_mmap() applies munlock_vma_pages_all() to clean up VM_LOCKED areas, and its current implementation (stupidly) tries to fault in absent pages, for example where PROT_NONE prevented them being faulted in when mlocking. Whereas the "ksm: fix oom deadlock" patch, knowing there's a race by which KSM might try to fault in pages after exit_mmap() had finally zapped the range, backs out of such faults doing nothing when its ksm_test_exit() notices mm_users 0. So revert that part of "ksm: fix oom deadlock" which moved the ksm_exit() call from before exit_mmap() to the middle of exit_mmap(); and remove those ksm_test_exit() checks from the page fault paths, so allowing the munlocking to proceed without interference. ksm_exit, if there are rmap_items still chained on this mm slot, takes mmap_sem write side: so preventing KSM from working on an mm while exit_mmap runs. And KSM will bail out as soon as it notices that mm_users is already zero, thanks to its internal ksm_test_exit checks. So that when a task is killed by OOM killer or the user, KSM will not indefinitely prevent it from running exit_mmap to release its memory. This does break a part of what "ksm: fix oom deadlock" was trying to achieve. When unmerging KSM (echo 2 >/sys/kernel/mm/ksm), and even when ksmd itself has to cancel a KSM page, it is possible that the first OOM-kill victim would be the KSM process being faulted: then its memory won't be freed until a second victim has been selected (freeing memory for the unmerging fault to complete). But the OOM killer is already liable to kill a second victim once the intended victim's p->mm goes to NULL: so there's not much point in rejecting this KSM patch before fixing that OOM behaviour. It is very much more important to allow KSM users to boot up, than to haggle over an unlikely and poorly supported OOM case. We also intend to fix munlocking to not fault pages: at which point this patch _could_ be reverted; though that would be controversial, so we hope to find a better solution. Signed-off-by: Andrea Arcangeli <aarcange@redhat.com> Acked-by: Justin M. Forbes <jforbes@redhat.com> Acked-for-now-by: Hugh Dickins <hugh.dickins@tiscali.co.uk> Cc: Izik Eidus <ieidus@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-09-22 00:02:22 +00:00
if (likely(pte_same(*page_table, orig_pte))) {
mm: fix fault vs invalidate race for linear mappings Fix the race between invalidate_inode_pages and do_no_page. Andrea Arcangeli identified a subtle race between invalidation of pages from pagecache with userspace mappings, and do_no_page. The issue is that invalidation has to shoot down all mappings to the page, before it can be discarded from the pagecache. Between shooting down ptes to a particular page, and actually dropping the struct page from the pagecache, do_no_page from any process might fault on that page and establish a new mapping to the page just before it gets discarded from the pagecache. The most common case where such invalidation is used is in file truncation. This case was catered for by doing a sort of open-coded seqlock between the file's i_size, and its truncate_count. Truncation will decrease i_size, then increment truncate_count before unmapping userspace pages; do_no_page will read truncate_count, then find the page if it is within i_size, and then check truncate_count under the page table lock and back out and retry if it had subsequently been changed (ptl will serialise against unmapping, and ensure a potentially updated truncate_count is actually visible). Complexity and documentation issues aside, the locking protocol fails in the case where we would like to invalidate pagecache inside i_size. do_no_page can come in anytime and filemap_nopage is not aware of the invalidation in progress (as it is when it is outside i_size). The end result is that dangling (->mapping == NULL) pages that appear to be from a particular file may be mapped into userspace with nonsense data. Valid mappings to the same place will see a different page. Andrea implemented two working fixes, one using a real seqlock, another using a page->flags bit. He also proposed using the page lock in do_no_page, but that was initially considered too heavyweight. However, it is not a global or per-file lock, and the page cacheline is modified in do_no_page to increment _count and _mapcount anyway, so a further modification should not be a large performance hit. Scalability is not an issue. This patch implements this latter approach. ->nopage implementations return with the page locked if it is possible for their underlying file to be invalidated (in that case, they must set a special vm_flags bit to indicate so). do_no_page only unlocks the page after setting up the mapping completely. invalidation is excluded because it holds the page lock during invalidation of each page (and ensures that the page is not mapped while holding the lock). This also allows significant simplifications in do_no_page, because we have the page locked in the right place in the pagecache from the start. Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:57 +00:00
flush_icache_page(vma, page);
entry = mk_pte(page, vma->vm_page_prot);
mm: merge populate and nopage into fault (fixes nonlinear) Nonlinear mappings are (AFAIKS) simply a virtual memory concept that encodes the virtual address -> file offset differently from linear mappings. ->populate is a layering violation because the filesystem/pagecache code should need to know anything about the virtual memory mapping. The hitch here is that the ->nopage handler didn't pass down enough information (ie. pgoff). But it is more logical to pass pgoff rather than have the ->nopage function calculate it itself anyway (because that's a similar layering violation). Having the populate handler install the pte itself is likewise a nasty thing to be doing. This patch introduces a new fault handler that replaces ->nopage and ->populate and (later) ->nopfn. Most of the old mechanism is still in place so there is a lot of duplication and nice cleanups that can be removed if everyone switches over. The rationale for doing this in the first place is that nonlinear mappings are subject to the pagefault vs invalidate/truncate race too, and it seemed stupid to duplicate the synchronisation logic rather than just consolidate the two. After this patch, MAP_NONBLOCK no longer sets up ptes for pages present in pagecache. Seems like a fringe functionality anyway. NOPAGE_REFAULT is removed. This should be implemented with ->fault, and no users have hit mainline yet. [akpm@linux-foundation.org: cleanup] [randy.dunlap@oracle.com: doc. fixes for readahead] [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:59 +00:00
if (flags & FAULT_FLAG_WRITE)
entry = maybe_mkwrite(pte_mkdirty(entry), vma);
if (anon) {
inc_mm_counter_fast(mm, MM_ANONPAGES);
swap: cull unevictable pages in fault path In the fault paths that install new anonymous pages, check whether the page is evictable or not using lru_cache_add_active_or_unevictable(). If the page is evictable, just add it to the active lru list [via the pagevec cache], else add it to the unevictable list. This "proactive" culling in the fault path mimics the handling of mlocked pages in Nick Piggin's series to keep mlocked pages off the lru lists. Notes: 1) This patch is optional--e.g., if one is concerned about the additional test in the fault path. We can defer the moving of nonreclaimable pages until when vmscan [shrink_*_list()] encounters them. Vmscan will only need to handle such pages once, but if there are a lot of them it could impact system performance. 2) The 'vma' argument to page_evictable() is require to notice that we're faulting a page into an mlock()ed vma w/o having to scan the page's rmap in the fault path. Culling mlock()ed anon pages is currently the only reason for this patch. 3) We can't cull swap pages in read_swap_cache_async() because the vma argument doesn't necessarily correspond to the swap cache offset passed in by swapin_readahead(). This could [did!] result in mlocking pages in non-VM_LOCKED vmas if [when] we tried to cull in this path. 4) Move set_pte_at() to after where we add page to lru to keep it hidden from other tasks that might walk the page table. We already do it in this order in do_anonymous() page. And, these are COW'd anon pages. Is this safe? [riel@redhat.com: undo an overzealous code cleanup] Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Signed-off-by: Rik van Riel <riel@redhat.com> Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 03:26:52 +00:00
page_add_new_anon_rmap(page, vma, address);
} else {
inc_mm_counter_fast(mm, MM_FILEPAGES);
mm: fix fault vs invalidate race for linear mappings Fix the race between invalidate_inode_pages and do_no_page. Andrea Arcangeli identified a subtle race between invalidation of pages from pagecache with userspace mappings, and do_no_page. The issue is that invalidation has to shoot down all mappings to the page, before it can be discarded from the pagecache. Between shooting down ptes to a particular page, and actually dropping the struct page from the pagecache, do_no_page from any process might fault on that page and establish a new mapping to the page just before it gets discarded from the pagecache. The most common case where such invalidation is used is in file truncation. This case was catered for by doing a sort of open-coded seqlock between the file's i_size, and its truncate_count. Truncation will decrease i_size, then increment truncate_count before unmapping userspace pages; do_no_page will read truncate_count, then find the page if it is within i_size, and then check truncate_count under the page table lock and back out and retry if it had subsequently been changed (ptl will serialise against unmapping, and ensure a potentially updated truncate_count is actually visible). Complexity and documentation issues aside, the locking protocol fails in the case where we would like to invalidate pagecache inside i_size. do_no_page can come in anytime and filemap_nopage is not aware of the invalidation in progress (as it is when it is outside i_size). The end result is that dangling (->mapping == NULL) pages that appear to be from a particular file may be mapped into userspace with nonsense data. Valid mappings to the same place will see a different page. Andrea implemented two working fixes, one using a real seqlock, another using a page->flags bit. He also proposed using the page lock in do_no_page, but that was initially considered too heavyweight. However, it is not a global or per-file lock, and the page cacheline is modified in do_no_page to increment _count and _mapcount anyway, so a further modification should not be a large performance hit. Scalability is not an issue. This patch implements this latter approach. ->nopage implementations return with the page locked if it is possible for their underlying file to be invalidated (in that case, they must set a special vm_flags bit to indicate so). do_no_page only unlocks the page after setting up the mapping completely. invalidation is excluded because it holds the page lock during invalidation of each page (and ensures that the page is not mapped while holding the lock). This also allows significant simplifications in do_no_page, because we have the page locked in the right place in the pagecache from the start. Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:57 +00:00
page_add_file_rmap(page);
mm: merge populate and nopage into fault (fixes nonlinear) Nonlinear mappings are (AFAIKS) simply a virtual memory concept that encodes the virtual address -> file offset differently from linear mappings. ->populate is a layering violation because the filesystem/pagecache code should need to know anything about the virtual memory mapping. The hitch here is that the ->nopage handler didn't pass down enough information (ie. pgoff). But it is more logical to pass pgoff rather than have the ->nopage function calculate it itself anyway (because that's a similar layering violation). Having the populate handler install the pte itself is likewise a nasty thing to be doing. This patch introduces a new fault handler that replaces ->nopage and ->populate and (later) ->nopfn. Most of the old mechanism is still in place so there is a lot of duplication and nice cleanups that can be removed if everyone switches over. The rationale for doing this in the first place is that nonlinear mappings are subject to the pagefault vs invalidate/truncate race too, and it seemed stupid to duplicate the synchronisation logic rather than just consolidate the two. After this patch, MAP_NONBLOCK no longer sets up ptes for pages present in pagecache. Seems like a fringe functionality anyway. NOPAGE_REFAULT is removed. This should be implemented with ->fault, and no users have hit mainline yet. [akpm@linux-foundation.org: cleanup] [randy.dunlap@oracle.com: doc. fixes for readahead] [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:59 +00:00
if (flags & FAULT_FLAG_WRITE) {
mm: fix fault vs invalidate race for linear mappings Fix the race between invalidate_inode_pages and do_no_page. Andrea Arcangeli identified a subtle race between invalidation of pages from pagecache with userspace mappings, and do_no_page. The issue is that invalidation has to shoot down all mappings to the page, before it can be discarded from the pagecache. Between shooting down ptes to a particular page, and actually dropping the struct page from the pagecache, do_no_page from any process might fault on that page and establish a new mapping to the page just before it gets discarded from the pagecache. The most common case where such invalidation is used is in file truncation. This case was catered for by doing a sort of open-coded seqlock between the file's i_size, and its truncate_count. Truncation will decrease i_size, then increment truncate_count before unmapping userspace pages; do_no_page will read truncate_count, then find the page if it is within i_size, and then check truncate_count under the page table lock and back out and retry if it had subsequently been changed (ptl will serialise against unmapping, and ensure a potentially updated truncate_count is actually visible). Complexity and documentation issues aside, the locking protocol fails in the case where we would like to invalidate pagecache inside i_size. do_no_page can come in anytime and filemap_nopage is not aware of the invalidation in progress (as it is when it is outside i_size). The end result is that dangling (->mapping == NULL) pages that appear to be from a particular file may be mapped into userspace with nonsense data. Valid mappings to the same place will see a different page. Andrea implemented two working fixes, one using a real seqlock, another using a page->flags bit. He also proposed using the page lock in do_no_page, but that was initially considered too heavyweight. However, it is not a global or per-file lock, and the page cacheline is modified in do_no_page to increment _count and _mapcount anyway, so a further modification should not be a large performance hit. Scalability is not an issue. This patch implements this latter approach. ->nopage implementations return with the page locked if it is possible for their underlying file to be invalidated (in that case, they must set a special vm_flags bit to indicate so). do_no_page only unlocks the page after setting up the mapping completely. invalidation is excluded because it holds the page lock during invalidation of each page (and ensures that the page is not mapped while holding the lock). This also allows significant simplifications in do_no_page, because we have the page locked in the right place in the pagecache from the start. Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:57 +00:00
dirty_page = page;
[PATCH] mm: tracking shared dirty pages Tracking of dirty pages in shared writeable mmap()s. The idea is simple: write protect clean shared writeable pages, catch the write-fault, make writeable and set dirty. On page write-back clean all the PTE dirty bits and write protect them once again. The implementation is a tad harder, mainly because the default backing_dev_info capabilities were too loosely maintained. Hence it is not enough to test the backing_dev_info for cap_account_dirty. The current heuristic is as follows, a VMA is eligible when: - its shared writeable (vm_flags & (VM_WRITE|VM_SHARED)) == (VM_WRITE|VM_SHARED) - it is not a 'special' mapping (vm_flags & (VM_PFNMAP|VM_INSERTPAGE)) == 0 - the backing_dev_info is cap_account_dirty mapping_cap_account_dirty(vma->vm_file->f_mapping) - f_op->mmap() didn't change the default page protection Page from remap_pfn_range() are explicitly excluded because their COW semantics are already horrid enough (see vm_normal_page() in do_wp_page()) and because they don't have a backing store anyway. mprotect() is taught about the new behaviour as well. However it overrides the last condition. Cleaning the pages on write-back is done with page_mkclean() a new rmap call. It can be called on any page, but is currently only implemented for mapped pages, if the page is found the be of a VMA that accounts dirty pages it will also wrprotect the PTE. Finally, in fs/buffers.c:try_to_free_buffers(); remove clear_page_dirty() from under ->private_lock. This seems to be safe, since ->private_lock is used to serialize access to the buffers, not the page itself. This is needed because clear_page_dirty() will call into page_mkclean() and would thereby violate locking order. [dhowells@redhat.com: Provide a page_mkclean() implementation for NOMMU] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-26 06:30:57 +00:00
get_page(dirty_page);
}
}
swap: cull unevictable pages in fault path In the fault paths that install new anonymous pages, check whether the page is evictable or not using lru_cache_add_active_or_unevictable(). If the page is evictable, just add it to the active lru list [via the pagevec cache], else add it to the unevictable list. This "proactive" culling in the fault path mimics the handling of mlocked pages in Nick Piggin's series to keep mlocked pages off the lru lists. Notes: 1) This patch is optional--e.g., if one is concerned about the additional test in the fault path. We can defer the moving of nonreclaimable pages until when vmscan [shrink_*_list()] encounters them. Vmscan will only need to handle such pages once, but if there are a lot of them it could impact system performance. 2) The 'vma' argument to page_evictable() is require to notice that we're faulting a page into an mlock()ed vma w/o having to scan the page's rmap in the fault path. Culling mlock()ed anon pages is currently the only reason for this patch. 3) We can't cull swap pages in read_swap_cache_async() because the vma argument doesn't necessarily correspond to the swap cache offset passed in by swapin_readahead(). This could [did!] result in mlocking pages in non-VM_LOCKED vmas if [when] we tried to cull in this path. 4) Move set_pte_at() to after where we add page to lru to keep it hidden from other tasks that might walk the page table. We already do it in this order in do_anonymous() page. And, these are COW'd anon pages. Is this safe? [riel@redhat.com: undo an overzealous code cleanup] Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Signed-off-by: Rik van Riel <riel@redhat.com> Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 03:26:52 +00:00
set_pte_at(mm, address, page_table, entry);
mm: fix fault vs invalidate race for linear mappings Fix the race between invalidate_inode_pages and do_no_page. Andrea Arcangeli identified a subtle race between invalidation of pages from pagecache with userspace mappings, and do_no_page. The issue is that invalidation has to shoot down all mappings to the page, before it can be discarded from the pagecache. Between shooting down ptes to a particular page, and actually dropping the struct page from the pagecache, do_no_page from any process might fault on that page and establish a new mapping to the page just before it gets discarded from the pagecache. The most common case where such invalidation is used is in file truncation. This case was catered for by doing a sort of open-coded seqlock between the file's i_size, and its truncate_count. Truncation will decrease i_size, then increment truncate_count before unmapping userspace pages; do_no_page will read truncate_count, then find the page if it is within i_size, and then check truncate_count under the page table lock and back out and retry if it had subsequently been changed (ptl will serialise against unmapping, and ensure a potentially updated truncate_count is actually visible). Complexity and documentation issues aside, the locking protocol fails in the case where we would like to invalidate pagecache inside i_size. do_no_page can come in anytime and filemap_nopage is not aware of the invalidation in progress (as it is when it is outside i_size). The end result is that dangling (->mapping == NULL) pages that appear to be from a particular file may be mapped into userspace with nonsense data. Valid mappings to the same place will see a different page. Andrea implemented two working fixes, one using a real seqlock, another using a page->flags bit. He also proposed using the page lock in do_no_page, but that was initially considered too heavyweight. However, it is not a global or per-file lock, and the page cacheline is modified in do_no_page to increment _count and _mapcount anyway, so a further modification should not be a large performance hit. Scalability is not an issue. This patch implements this latter approach. ->nopage implementations return with the page locked if it is possible for their underlying file to be invalidated (in that case, they must set a special vm_flags bit to indicate so). do_no_page only unlocks the page after setting up the mapping completely. invalidation is excluded because it holds the page lock during invalidation of each page (and ensures that the page is not mapped while holding the lock). This also allows significant simplifications in do_no_page, because we have the page locked in the right place in the pagecache from the start. Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:57 +00:00
/* no need to invalidate: a not-present page won't be cached */
update_mmu_cache(vma, address, page_table);
} else {
mm: preallocate page before lock_page() at filemap COW Currently we are keeping faulted page locked throughout whole __do_fault call (except for page_mkwrite code path) after calling file system's fault code. If we do early COW, we allocate a new page which has to be charged for a memcg (mem_cgroup_newpage_charge). This function, however, might block for unbounded amount of time if memcg oom killer is disabled or fork-bomb is running because the only way out of the OOM situation is either an external event or OOM-situation fix. In the end we are keeping the faulted page locked and blocking other processes from faulting it in which is not good at all because we are basically punishing potentially an unrelated process for OOM condition in a different group (I have seen stuck system because of ld-2.11.1.so being locked). We can do test easily. % cgcreate -g memory:A % cgset -r memory.limit_in_bytes=64M A % cgset -r memory.memsw.limit_in_bytes=64M A % cd kernel_dir; cgexec -g memory:A make -j Then, the whole system will live-locked until you kill 'make -j' by hands (or push reboot...) This is because some important page in a a shared library are locked. Considering again, the new page is not necessary to be allocated with lock_page() held. And usual page allocation may dive into long memory reclaim loop with holding lock_page() and can cause very long latency. There are 3 ways. 1. do allocation/charge before lock_page() Pros. - simple and can handle page allocation in the same manner. This will reduce holding time of lock_page() in general. Cons. - we do page allocation even if ->fault() returns error. 2. do charge after unlock_page(). Even if charge fails, it's just OOM. Pros. - no impact to non-memcg path. Cons. - implemenation requires special cares of LRU and we need to modify page_add_new_anon_rmap()... 3. do unlock->charge->lock again method. Pros. - no impact to non-memcg path. Cons. - This may kill LOCK_PAGE_RETRY optimization. We need to release lock and get it again... This patch moves "charge" and memory allocation for COW page before lock_page(). Then, we can avoid scanning LRU with holding a lock on a page and latency under lock_page() will be reduced. Then, above livelock disappears. [akpm@linux-foundation.org: fix code layout] Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Reported-by: Lutz Vieweg <lvml@5t9.de> Original-idea-by: Michal Hocko <mhocko@suse.cz> Cc: Michal Hocko <mhocko@suse.cz> Cc: Ying Han <yinghan@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-07-26 00:12:27 +00:00
if (cow_page)
mem_cgroup_uncharge_page(cow_page);
mm: fix fault vs invalidate race for linear mappings Fix the race between invalidate_inode_pages and do_no_page. Andrea Arcangeli identified a subtle race between invalidation of pages from pagecache with userspace mappings, and do_no_page. The issue is that invalidation has to shoot down all mappings to the page, before it can be discarded from the pagecache. Between shooting down ptes to a particular page, and actually dropping the struct page from the pagecache, do_no_page from any process might fault on that page and establish a new mapping to the page just before it gets discarded from the pagecache. The most common case where such invalidation is used is in file truncation. This case was catered for by doing a sort of open-coded seqlock between the file's i_size, and its truncate_count. Truncation will decrease i_size, then increment truncate_count before unmapping userspace pages; do_no_page will read truncate_count, then find the page if it is within i_size, and then check truncate_count under the page table lock and back out and retry if it had subsequently been changed (ptl will serialise against unmapping, and ensure a potentially updated truncate_count is actually visible). Complexity and documentation issues aside, the locking protocol fails in the case where we would like to invalidate pagecache inside i_size. do_no_page can come in anytime and filemap_nopage is not aware of the invalidation in progress (as it is when it is outside i_size). The end result is that dangling (->mapping == NULL) pages that appear to be from a particular file may be mapped into userspace with nonsense data. Valid mappings to the same place will see a different page. Andrea implemented two working fixes, one using a real seqlock, another using a page->flags bit. He also proposed using the page lock in do_no_page, but that was initially considered too heavyweight. However, it is not a global or per-file lock, and the page cacheline is modified in do_no_page to increment _count and _mapcount anyway, so a further modification should not be a large performance hit. Scalability is not an issue. This patch implements this latter approach. ->nopage implementations return with the page locked if it is possible for their underlying file to be invalidated (in that case, they must set a special vm_flags bit to indicate so). do_no_page only unlocks the page after setting up the mapping completely. invalidation is excluded because it holds the page lock during invalidation of each page (and ensures that the page is not mapped while holding the lock). This also allows significant simplifications in do_no_page, because we have the page locked in the right place in the pagecache from the start. Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:57 +00:00
if (anon)
page_cache_release(page);
else
mm: merge populate and nopage into fault (fixes nonlinear) Nonlinear mappings are (AFAIKS) simply a virtual memory concept that encodes the virtual address -> file offset differently from linear mappings. ->populate is a layering violation because the filesystem/pagecache code should need to know anything about the virtual memory mapping. The hitch here is that the ->nopage handler didn't pass down enough information (ie. pgoff). But it is more logical to pass pgoff rather than have the ->nopage function calculate it itself anyway (because that's a similar layering violation). Having the populate handler install the pte itself is likewise a nasty thing to be doing. This patch introduces a new fault handler that replaces ->nopage and ->populate and (later) ->nopfn. Most of the old mechanism is still in place so there is a lot of duplication and nice cleanups that can be removed if everyone switches over. The rationale for doing this in the first place is that nonlinear mappings are subject to the pagefault vs invalidate/truncate race too, and it seemed stupid to duplicate the synchronisation logic rather than just consolidate the two. After this patch, MAP_NONBLOCK no longer sets up ptes for pages present in pagecache. Seems like a fringe functionality anyway. NOPAGE_REFAULT is removed. This should be implemented with ->fault, and no users have hit mainline yet. [akpm@linux-foundation.org: cleanup] [randy.dunlap@oracle.com: doc. fixes for readahead] [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:59 +00:00
anon = 1; /* no anon but release faulted_page */
}
pte_unmap_unlock(page_table, ptl);
mm: fix fault vs invalidate race for linear mappings Fix the race between invalidate_inode_pages and do_no_page. Andrea Arcangeli identified a subtle race between invalidation of pages from pagecache with userspace mappings, and do_no_page. The issue is that invalidation has to shoot down all mappings to the page, before it can be discarded from the pagecache. Between shooting down ptes to a particular page, and actually dropping the struct page from the pagecache, do_no_page from any process might fault on that page and establish a new mapping to the page just before it gets discarded from the pagecache. The most common case where such invalidation is used is in file truncation. This case was catered for by doing a sort of open-coded seqlock between the file's i_size, and its truncate_count. Truncation will decrease i_size, then increment truncate_count before unmapping userspace pages; do_no_page will read truncate_count, then find the page if it is within i_size, and then check truncate_count under the page table lock and back out and retry if it had subsequently been changed (ptl will serialise against unmapping, and ensure a potentially updated truncate_count is actually visible). Complexity and documentation issues aside, the locking protocol fails in the case where we would like to invalidate pagecache inside i_size. do_no_page can come in anytime and filemap_nopage is not aware of the invalidation in progress (as it is when it is outside i_size). The end result is that dangling (->mapping == NULL) pages that appear to be from a particular file may be mapped into userspace with nonsense data. Valid mappings to the same place will see a different page. Andrea implemented two working fixes, one using a real seqlock, another using a page->flags bit. He also proposed using the page lock in do_no_page, but that was initially considered too heavyweight. However, it is not a global or per-file lock, and the page cacheline is modified in do_no_page to increment _count and _mapcount anyway, so a further modification should not be a large performance hit. Scalability is not an issue. This patch implements this latter approach. ->nopage implementations return with the page locked if it is possible for their underlying file to be invalidated (in that case, they must set a special vm_flags bit to indicate so). do_no_page only unlocks the page after setting up the mapping completely. invalidation is excluded because it holds the page lock during invalidation of each page (and ensures that the page is not mapped while holding the lock). This also allows significant simplifications in do_no_page, because we have the page locked in the right place in the pagecache from the start. Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:57 +00:00
mm: close page_mkwrite races Change page_mkwrite to allow implementations to return with the page locked, and also change it's callers (in page fault paths) to hold the lock until the page is marked dirty. This allows the filesystem to have full control of page dirtying events coming from the VM. Rather than simply hold the page locked over the page_mkwrite call, we call page_mkwrite with the page unlocked and allow callers to return with it locked, so filesystems can avoid LOR conditions with page lock. The problem with the current scheme is this: a filesystem that wants to associate some metadata with a page as long as the page is dirty, will perform this manipulation in its ->page_mkwrite. It currently then must return with the page unlocked and may not hold any other locks (according to existing page_mkwrite convention). In this window, the VM could write out the page, clearing page-dirty. The filesystem has no good way to detect that a dirty pte is about to be attached, so it will happily write out the page, at which point, the filesystem may manipulate the metadata to reflect that the page is no longer dirty. It is not always possible to perform the required metadata manipulation in ->set_page_dirty, because that function cannot block or fail. The filesystem may need to allocate some data structure, for example. And the VM cannot mark the pte dirty before page_mkwrite, because page_mkwrite is allowed to fail, so we must not allow any window where the page could be written to if page_mkwrite does fail. This solution of holding the page locked over the 3 critical operations (page_mkwrite, setting the pte dirty, and finally setting the page dirty) closes out races nicely, preventing page cleaning for writeout being initiated in that window. This provides the filesystem with a strong synchronisation against the VM here. - Sage needs this race closed for ceph filesystem. - Trond for NFS (http://bugzilla.kernel.org/show_bug.cgi?id=12913). - I need it for fsblock. - I suspect other filesystems may need it too (eg. btrfs). - I have converted buffer.c to the new locking. Even simple block allocation under dirty pages might be susceptible to i_size changing under partial page at the end of file (we also have a buffer.c-side problem here, but it cannot be fixed properly without this patch). - Other filesystems (eg. NFS, maybe btrfs) will need to change their page_mkwrite functions themselves. [ This also moves page_mkwrite another step closer to fault, which should eventually allow page_mkwrite to be moved into ->fault, and thus avoiding a filesystem calldown and page lock/unlock cycle in __do_fault. ] [akpm@linux-foundation.org: fix derefs of NULL ->mapping] Cc: Sage Weil <sage@newdream.net> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Valdis Kletnieks <Valdis.Kletnieks@vt.edu> Cc: <stable@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-30 22:08:16 +00:00
if (dirty_page) {
struct address_space *mapping = page->mapping;
int dirtied = 0;
mm: close page_mkwrite races Change page_mkwrite to allow implementations to return with the page locked, and also change it's callers (in page fault paths) to hold the lock until the page is marked dirty. This allows the filesystem to have full control of page dirtying events coming from the VM. Rather than simply hold the page locked over the page_mkwrite call, we call page_mkwrite with the page unlocked and allow callers to return with it locked, so filesystems can avoid LOR conditions with page lock. The problem with the current scheme is this: a filesystem that wants to associate some metadata with a page as long as the page is dirty, will perform this manipulation in its ->page_mkwrite. It currently then must return with the page unlocked and may not hold any other locks (according to existing page_mkwrite convention). In this window, the VM could write out the page, clearing page-dirty. The filesystem has no good way to detect that a dirty pte is about to be attached, so it will happily write out the page, at which point, the filesystem may manipulate the metadata to reflect that the page is no longer dirty. It is not always possible to perform the required metadata manipulation in ->set_page_dirty, because that function cannot block or fail. The filesystem may need to allocate some data structure, for example. And the VM cannot mark the pte dirty before page_mkwrite, because page_mkwrite is allowed to fail, so we must not allow any window where the page could be written to if page_mkwrite does fail. This solution of holding the page locked over the 3 critical operations (page_mkwrite, setting the pte dirty, and finally setting the page dirty) closes out races nicely, preventing page cleaning for writeout being initiated in that window. This provides the filesystem with a strong synchronisation against the VM here. - Sage needs this race closed for ceph filesystem. - Trond for NFS (http://bugzilla.kernel.org/show_bug.cgi?id=12913). - I need it for fsblock. - I suspect other filesystems may need it too (eg. btrfs). - I have converted buffer.c to the new locking. Even simple block allocation under dirty pages might be susceptible to i_size changing under partial page at the end of file (we also have a buffer.c-side problem here, but it cannot be fixed properly without this patch). - Other filesystems (eg. NFS, maybe btrfs) will need to change their page_mkwrite functions themselves. [ This also moves page_mkwrite another step closer to fault, which should eventually allow page_mkwrite to be moved into ->fault, and thus avoiding a filesystem calldown and page lock/unlock cycle in __do_fault. ] [akpm@linux-foundation.org: fix derefs of NULL ->mapping] Cc: Sage Weil <sage@newdream.net> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Valdis Kletnieks <Valdis.Kletnieks@vt.edu> Cc: <stable@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-30 22:08:16 +00:00
if (set_page_dirty(dirty_page))
dirtied = 1;
mm: close page_mkwrite races Change page_mkwrite to allow implementations to return with the page locked, and also change it's callers (in page fault paths) to hold the lock until the page is marked dirty. This allows the filesystem to have full control of page dirtying events coming from the VM. Rather than simply hold the page locked over the page_mkwrite call, we call page_mkwrite with the page unlocked and allow callers to return with it locked, so filesystems can avoid LOR conditions with page lock. The problem with the current scheme is this: a filesystem that wants to associate some metadata with a page as long as the page is dirty, will perform this manipulation in its ->page_mkwrite. It currently then must return with the page unlocked and may not hold any other locks (according to existing page_mkwrite convention). In this window, the VM could write out the page, clearing page-dirty. The filesystem has no good way to detect that a dirty pte is about to be attached, so it will happily write out the page, at which point, the filesystem may manipulate the metadata to reflect that the page is no longer dirty. It is not always possible to perform the required metadata manipulation in ->set_page_dirty, because that function cannot block or fail. The filesystem may need to allocate some data structure, for example. And the VM cannot mark the pte dirty before page_mkwrite, because page_mkwrite is allowed to fail, so we must not allow any window where the page could be written to if page_mkwrite does fail. This solution of holding the page locked over the 3 critical operations (page_mkwrite, setting the pte dirty, and finally setting the page dirty) closes out races nicely, preventing page cleaning for writeout being initiated in that window. This provides the filesystem with a strong synchronisation against the VM here. - Sage needs this race closed for ceph filesystem. - Trond for NFS (http://bugzilla.kernel.org/show_bug.cgi?id=12913). - I need it for fsblock. - I suspect other filesystems may need it too (eg. btrfs). - I have converted buffer.c to the new locking. Even simple block allocation under dirty pages might be susceptible to i_size changing under partial page at the end of file (we also have a buffer.c-side problem here, but it cannot be fixed properly without this patch). - Other filesystems (eg. NFS, maybe btrfs) will need to change their page_mkwrite functions themselves. [ This also moves page_mkwrite another step closer to fault, which should eventually allow page_mkwrite to be moved into ->fault, and thus avoiding a filesystem calldown and page lock/unlock cycle in __do_fault. ] [akpm@linux-foundation.org: fix derefs of NULL ->mapping] Cc: Sage Weil <sage@newdream.net> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Valdis Kletnieks <Valdis.Kletnieks@vt.edu> Cc: <stable@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-30 22:08:16 +00:00
unlock_page(dirty_page);
[PATCH] mm: tracking shared dirty pages Tracking of dirty pages in shared writeable mmap()s. The idea is simple: write protect clean shared writeable pages, catch the write-fault, make writeable and set dirty. On page write-back clean all the PTE dirty bits and write protect them once again. The implementation is a tad harder, mainly because the default backing_dev_info capabilities were too loosely maintained. Hence it is not enough to test the backing_dev_info for cap_account_dirty. The current heuristic is as follows, a VMA is eligible when: - its shared writeable (vm_flags & (VM_WRITE|VM_SHARED)) == (VM_WRITE|VM_SHARED) - it is not a 'special' mapping (vm_flags & (VM_PFNMAP|VM_INSERTPAGE)) == 0 - the backing_dev_info is cap_account_dirty mapping_cap_account_dirty(vma->vm_file->f_mapping) - f_op->mmap() didn't change the default page protection Page from remap_pfn_range() are explicitly excluded because their COW semantics are already horrid enough (see vm_normal_page() in do_wp_page()) and because they don't have a backing store anyway. mprotect() is taught about the new behaviour as well. However it overrides the last condition. Cleaning the pages on write-back is done with page_mkclean() a new rmap call. It can be called on any page, but is currently only implemented for mapped pages, if the page is found the be of a VMA that accounts dirty pages it will also wrprotect the PTE. Finally, in fs/buffers.c:try_to_free_buffers(); remove clear_page_dirty() from under ->private_lock. This seems to be safe, since ->private_lock is used to serialize access to the buffers, not the page itself. This is needed because clear_page_dirty() will call into page_mkclean() and would thereby violate locking order. [dhowells@redhat.com: Provide a page_mkclean() implementation for NOMMU] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-26 06:30:57 +00:00
put_page(dirty_page);
if ((dirtied || page_mkwrite) && mapping) {
mm: close page_mkwrite races Change page_mkwrite to allow implementations to return with the page locked, and also change it's callers (in page fault paths) to hold the lock until the page is marked dirty. This allows the filesystem to have full control of page dirtying events coming from the VM. Rather than simply hold the page locked over the page_mkwrite call, we call page_mkwrite with the page unlocked and allow callers to return with it locked, so filesystems can avoid LOR conditions with page lock. The problem with the current scheme is this: a filesystem that wants to associate some metadata with a page as long as the page is dirty, will perform this manipulation in its ->page_mkwrite. It currently then must return with the page unlocked and may not hold any other locks (according to existing page_mkwrite convention). In this window, the VM could write out the page, clearing page-dirty. The filesystem has no good way to detect that a dirty pte is about to be attached, so it will happily write out the page, at which point, the filesystem may manipulate the metadata to reflect that the page is no longer dirty. It is not always possible to perform the required metadata manipulation in ->set_page_dirty, because that function cannot block or fail. The filesystem may need to allocate some data structure, for example. And the VM cannot mark the pte dirty before page_mkwrite, because page_mkwrite is allowed to fail, so we must not allow any window where the page could be written to if page_mkwrite does fail. This solution of holding the page locked over the 3 critical operations (page_mkwrite, setting the pte dirty, and finally setting the page dirty) closes out races nicely, preventing page cleaning for writeout being initiated in that window. This provides the filesystem with a strong synchronisation against the VM here. - Sage needs this race closed for ceph filesystem. - Trond for NFS (http://bugzilla.kernel.org/show_bug.cgi?id=12913). - I need it for fsblock. - I suspect other filesystems may need it too (eg. btrfs). - I have converted buffer.c to the new locking. Even simple block allocation under dirty pages might be susceptible to i_size changing under partial page at the end of file (we also have a buffer.c-side problem here, but it cannot be fixed properly without this patch). - Other filesystems (eg. NFS, maybe btrfs) will need to change their page_mkwrite functions themselves. [ This also moves page_mkwrite another step closer to fault, which should eventually allow page_mkwrite to be moved into ->fault, and thus avoiding a filesystem calldown and page lock/unlock cycle in __do_fault. ] [akpm@linux-foundation.org: fix derefs of NULL ->mapping] Cc: Sage Weil <sage@newdream.net> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Valdis Kletnieks <Valdis.Kletnieks@vt.edu> Cc: <stable@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-30 22:08:16 +00:00
/*
* Some device drivers do not set page.mapping but still
* dirty their pages
*/
balance_dirty_pages_ratelimited(mapping);
}
/* file_update_time outside page_lock */
if (vma->vm_file && !page_mkwrite)
mm: close page_mkwrite races Change page_mkwrite to allow implementations to return with the page locked, and also change it's callers (in page fault paths) to hold the lock until the page is marked dirty. This allows the filesystem to have full control of page dirtying events coming from the VM. Rather than simply hold the page locked over the page_mkwrite call, we call page_mkwrite with the page unlocked and allow callers to return with it locked, so filesystems can avoid LOR conditions with page lock. The problem with the current scheme is this: a filesystem that wants to associate some metadata with a page as long as the page is dirty, will perform this manipulation in its ->page_mkwrite. It currently then must return with the page unlocked and may not hold any other locks (according to existing page_mkwrite convention). In this window, the VM could write out the page, clearing page-dirty. The filesystem has no good way to detect that a dirty pte is about to be attached, so it will happily write out the page, at which point, the filesystem may manipulate the metadata to reflect that the page is no longer dirty. It is not always possible to perform the required metadata manipulation in ->set_page_dirty, because that function cannot block or fail. The filesystem may need to allocate some data structure, for example. And the VM cannot mark the pte dirty before page_mkwrite, because page_mkwrite is allowed to fail, so we must not allow any window where the page could be written to if page_mkwrite does fail. This solution of holding the page locked over the 3 critical operations (page_mkwrite, setting the pte dirty, and finally setting the page dirty) closes out races nicely, preventing page cleaning for writeout being initiated in that window. This provides the filesystem with a strong synchronisation against the VM here. - Sage needs this race closed for ceph filesystem. - Trond for NFS (http://bugzilla.kernel.org/show_bug.cgi?id=12913). - I need it for fsblock. - I suspect other filesystems may need it too (eg. btrfs). - I have converted buffer.c to the new locking. Even simple block allocation under dirty pages might be susceptible to i_size changing under partial page at the end of file (we also have a buffer.c-side problem here, but it cannot be fixed properly without this patch). - Other filesystems (eg. NFS, maybe btrfs) will need to change their page_mkwrite functions themselves. [ This also moves page_mkwrite another step closer to fault, which should eventually allow page_mkwrite to be moved into ->fault, and thus avoiding a filesystem calldown and page lock/unlock cycle in __do_fault. ] [akpm@linux-foundation.org: fix derefs of NULL ->mapping] Cc: Sage Weil <sage@newdream.net> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Valdis Kletnieks <Valdis.Kletnieks@vt.edu> Cc: <stable@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-30 22:08:16 +00:00
file_update_time(vma->vm_file);
} else {
unlock_page(vmf.page);
if (anon)
page_cache_release(vmf.page);
[PATCH] mm: tracking shared dirty pages Tracking of dirty pages in shared writeable mmap()s. The idea is simple: write protect clean shared writeable pages, catch the write-fault, make writeable and set dirty. On page write-back clean all the PTE dirty bits and write protect them once again. The implementation is a tad harder, mainly because the default backing_dev_info capabilities were too loosely maintained. Hence it is not enough to test the backing_dev_info for cap_account_dirty. The current heuristic is as follows, a VMA is eligible when: - its shared writeable (vm_flags & (VM_WRITE|VM_SHARED)) == (VM_WRITE|VM_SHARED) - it is not a 'special' mapping (vm_flags & (VM_PFNMAP|VM_INSERTPAGE)) == 0 - the backing_dev_info is cap_account_dirty mapping_cap_account_dirty(vma->vm_file->f_mapping) - f_op->mmap() didn't change the default page protection Page from remap_pfn_range() are explicitly excluded because their COW semantics are already horrid enough (see vm_normal_page() in do_wp_page()) and because they don't have a backing store anyway. mprotect() is taught about the new behaviour as well. However it overrides the last condition. Cleaning the pages on write-back is done with page_mkclean() a new rmap call. It can be called on any page, but is currently only implemented for mapped pages, if the page is found the be of a VMA that accounts dirty pages it will also wrprotect the PTE. Finally, in fs/buffers.c:try_to_free_buffers(); remove clear_page_dirty() from under ->private_lock. This seems to be safe, since ->private_lock is used to serialize access to the buffers, not the page itself. This is needed because clear_page_dirty() will call into page_mkclean() and would thereby violate locking order. [dhowells@redhat.com: Provide a page_mkclean() implementation for NOMMU] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Hugh Dickins <hugh@veritas.com> Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-26 06:30:57 +00:00
}
mm: fix fault vs invalidate race for linear mappings Fix the race between invalidate_inode_pages and do_no_page. Andrea Arcangeli identified a subtle race between invalidation of pages from pagecache with userspace mappings, and do_no_page. The issue is that invalidation has to shoot down all mappings to the page, before it can be discarded from the pagecache. Between shooting down ptes to a particular page, and actually dropping the struct page from the pagecache, do_no_page from any process might fault on that page and establish a new mapping to the page just before it gets discarded from the pagecache. The most common case where such invalidation is used is in file truncation. This case was catered for by doing a sort of open-coded seqlock between the file's i_size, and its truncate_count. Truncation will decrease i_size, then increment truncate_count before unmapping userspace pages; do_no_page will read truncate_count, then find the page if it is within i_size, and then check truncate_count under the page table lock and back out and retry if it had subsequently been changed (ptl will serialise against unmapping, and ensure a potentially updated truncate_count is actually visible). Complexity and documentation issues aside, the locking protocol fails in the case where we would like to invalidate pagecache inside i_size. do_no_page can come in anytime and filemap_nopage is not aware of the invalidation in progress (as it is when it is outside i_size). The end result is that dangling (->mapping == NULL) pages that appear to be from a particular file may be mapped into userspace with nonsense data. Valid mappings to the same place will see a different page. Andrea implemented two working fixes, one using a real seqlock, another using a page->flags bit. He also proposed using the page lock in do_no_page, but that was initially considered too heavyweight. However, it is not a global or per-file lock, and the page cacheline is modified in do_no_page to increment _count and _mapcount anyway, so a further modification should not be a large performance hit. Scalability is not an issue. This patch implements this latter approach. ->nopage implementations return with the page locked if it is possible for their underlying file to be invalidated (in that case, they must set a special vm_flags bit to indicate so). do_no_page only unlocks the page after setting up the mapping completely. invalidation is excluded because it holds the page lock during invalidation of each page (and ensures that the page is not mapped while holding the lock). This also allows significant simplifications in do_no_page, because we have the page locked in the right place in the pagecache from the start. Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:57 +00:00
mm: fault feedback #2 This patch completes Linus's wish that the fault return codes be made into bit flags, which I agree makes everything nicer. This requires requires all handle_mm_fault callers to be modified (possibly the modifications should go further and do things like fault accounting in handle_mm_fault -- however that would be for another patch). [akpm@linux-foundation.org: fix alpha build] [akpm@linux-foundation.org: fix s390 build] [akpm@linux-foundation.org: fix sparc build] [akpm@linux-foundation.org: fix sparc64 build] [akpm@linux-foundation.org: fix ia64 build] Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Richard Henderson <rth@twiddle.net> Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Ian Molton <spyro@f2s.com> Cc: Bryan Wu <bryan.wu@analog.com> Cc: Mikael Starvik <starvik@axis.com> Cc: David Howells <dhowells@redhat.com> Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Hirokazu Takata <takata@linux-m32r.org> Cc: Geert Uytterhoeven <geert@linux-m68k.org> Cc: Roman Zippel <zippel@linux-m68k.org> Cc: Greg Ungerer <gerg@uclinux.org> Cc: Matthew Wilcox <willy@debian.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Kazumoto Kojima <kkojima@rr.iij4u.or.jp> Cc: Richard Curnow <rc@rc0.org.uk> Cc: William Lee Irwin III <wli@holomorphy.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Jeff Dike <jdike@addtoit.com> Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it> Cc: Miles Bader <uclinux-v850@lsi.nec.co.jp> Cc: Chris Zankel <chris@zankel.net> Acked-by: Kyle McMartin <kyle@mcmartin.ca> Acked-by: Haavard Skinnemoen <hskinnemoen@atmel.com> Acked-by: Ralf Baechle <ralf@linux-mips.org> Acked-by: Andi Kleen <ak@muc.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> [ Still apparently needs some ARM and PPC loving - Linus ] Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:47:05 +00:00
return ret;
mm: close page_mkwrite races Change page_mkwrite to allow implementations to return with the page locked, and also change it's callers (in page fault paths) to hold the lock until the page is marked dirty. This allows the filesystem to have full control of page dirtying events coming from the VM. Rather than simply hold the page locked over the page_mkwrite call, we call page_mkwrite with the page unlocked and allow callers to return with it locked, so filesystems can avoid LOR conditions with page lock. The problem with the current scheme is this: a filesystem that wants to associate some metadata with a page as long as the page is dirty, will perform this manipulation in its ->page_mkwrite. It currently then must return with the page unlocked and may not hold any other locks (according to existing page_mkwrite convention). In this window, the VM could write out the page, clearing page-dirty. The filesystem has no good way to detect that a dirty pte is about to be attached, so it will happily write out the page, at which point, the filesystem may manipulate the metadata to reflect that the page is no longer dirty. It is not always possible to perform the required metadata manipulation in ->set_page_dirty, because that function cannot block or fail. The filesystem may need to allocate some data structure, for example. And the VM cannot mark the pte dirty before page_mkwrite, because page_mkwrite is allowed to fail, so we must not allow any window where the page could be written to if page_mkwrite does fail. This solution of holding the page locked over the 3 critical operations (page_mkwrite, setting the pte dirty, and finally setting the page dirty) closes out races nicely, preventing page cleaning for writeout being initiated in that window. This provides the filesystem with a strong synchronisation against the VM here. - Sage needs this race closed for ceph filesystem. - Trond for NFS (http://bugzilla.kernel.org/show_bug.cgi?id=12913). - I need it for fsblock. - I suspect other filesystems may need it too (eg. btrfs). - I have converted buffer.c to the new locking. Even simple block allocation under dirty pages might be susceptible to i_size changing under partial page at the end of file (we also have a buffer.c-side problem here, but it cannot be fixed properly without this patch). - Other filesystems (eg. NFS, maybe btrfs) will need to change their page_mkwrite functions themselves. [ This also moves page_mkwrite another step closer to fault, which should eventually allow page_mkwrite to be moved into ->fault, and thus avoiding a filesystem calldown and page lock/unlock cycle in __do_fault. ] [akpm@linux-foundation.org: fix derefs of NULL ->mapping] Cc: Sage Weil <sage@newdream.net> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Valdis Kletnieks <Valdis.Kletnieks@vt.edu> Cc: <stable@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-30 22:08:16 +00:00
unwritable_page:
page_cache_release(page);
return ret;
mm: preallocate page before lock_page() at filemap COW Currently we are keeping faulted page locked throughout whole __do_fault call (except for page_mkwrite code path) after calling file system's fault code. If we do early COW, we allocate a new page which has to be charged for a memcg (mem_cgroup_newpage_charge). This function, however, might block for unbounded amount of time if memcg oom killer is disabled or fork-bomb is running because the only way out of the OOM situation is either an external event or OOM-situation fix. In the end we are keeping the faulted page locked and blocking other processes from faulting it in which is not good at all because we are basically punishing potentially an unrelated process for OOM condition in a different group (I have seen stuck system because of ld-2.11.1.so being locked). We can do test easily. % cgcreate -g memory:A % cgset -r memory.limit_in_bytes=64M A % cgset -r memory.memsw.limit_in_bytes=64M A % cd kernel_dir; cgexec -g memory:A make -j Then, the whole system will live-locked until you kill 'make -j' by hands (or push reboot...) This is because some important page in a a shared library are locked. Considering again, the new page is not necessary to be allocated with lock_page() held. And usual page allocation may dive into long memory reclaim loop with holding lock_page() and can cause very long latency. There are 3 ways. 1. do allocation/charge before lock_page() Pros. - simple and can handle page allocation in the same manner. This will reduce holding time of lock_page() in general. Cons. - we do page allocation even if ->fault() returns error. 2. do charge after unlock_page(). Even if charge fails, it's just OOM. Pros. - no impact to non-memcg path. Cons. - implemenation requires special cares of LRU and we need to modify page_add_new_anon_rmap()... 3. do unlock->charge->lock again method. Pros. - no impact to non-memcg path. Cons. - This may kill LOCK_PAGE_RETRY optimization. We need to release lock and get it again... This patch moves "charge" and memory allocation for COW page before lock_page(). Then, we can avoid scanning LRU with holding a lock on a page and latency under lock_page() will be reduced. Then, above livelock disappears. [akpm@linux-foundation.org: fix code layout] Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Reported-by: Lutz Vieweg <lvml@5t9.de> Original-idea-by: Michal Hocko <mhocko@suse.cz> Cc: Michal Hocko <mhocko@suse.cz> Cc: Ying Han <yinghan@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-07-26 00:12:27 +00:00
uncharge_out:
/* fs's fault handler get error */
if (cow_page) {
mem_cgroup_uncharge_page(cow_page);
page_cache_release(cow_page);
}
return ret;
mm: merge populate and nopage into fault (fixes nonlinear) Nonlinear mappings are (AFAIKS) simply a virtual memory concept that encodes the virtual address -> file offset differently from linear mappings. ->populate is a layering violation because the filesystem/pagecache code should need to know anything about the virtual memory mapping. The hitch here is that the ->nopage handler didn't pass down enough information (ie. pgoff). But it is more logical to pass pgoff rather than have the ->nopage function calculate it itself anyway (because that's a similar layering violation). Having the populate handler install the pte itself is likewise a nasty thing to be doing. This patch introduces a new fault handler that replaces ->nopage and ->populate and (later) ->nopfn. Most of the old mechanism is still in place so there is a lot of duplication and nice cleanups that can be removed if everyone switches over. The rationale for doing this in the first place is that nonlinear mappings are subject to the pagefault vs invalidate/truncate race too, and it seemed stupid to duplicate the synchronisation logic rather than just consolidate the two. After this patch, MAP_NONBLOCK no longer sets up ptes for pages present in pagecache. Seems like a fringe functionality anyway. NOPAGE_REFAULT is removed. This should be implemented with ->fault, and no users have hit mainline yet. [akpm@linux-foundation.org: cleanup] [randy.dunlap@oracle.com: doc. fixes for readahead] [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:59 +00:00
}
mm: fix fault vs invalidate race for linear mappings Fix the race between invalidate_inode_pages and do_no_page. Andrea Arcangeli identified a subtle race between invalidation of pages from pagecache with userspace mappings, and do_no_page. The issue is that invalidation has to shoot down all mappings to the page, before it can be discarded from the pagecache. Between shooting down ptes to a particular page, and actually dropping the struct page from the pagecache, do_no_page from any process might fault on that page and establish a new mapping to the page just before it gets discarded from the pagecache. The most common case where such invalidation is used is in file truncation. This case was catered for by doing a sort of open-coded seqlock between the file's i_size, and its truncate_count. Truncation will decrease i_size, then increment truncate_count before unmapping userspace pages; do_no_page will read truncate_count, then find the page if it is within i_size, and then check truncate_count under the page table lock and back out and retry if it had subsequently been changed (ptl will serialise against unmapping, and ensure a potentially updated truncate_count is actually visible). Complexity and documentation issues aside, the locking protocol fails in the case where we would like to invalidate pagecache inside i_size. do_no_page can come in anytime and filemap_nopage is not aware of the invalidation in progress (as it is when it is outside i_size). The end result is that dangling (->mapping == NULL) pages that appear to be from a particular file may be mapped into userspace with nonsense data. Valid mappings to the same place will see a different page. Andrea implemented two working fixes, one using a real seqlock, another using a page->flags bit. He also proposed using the page lock in do_no_page, but that was initially considered too heavyweight. However, it is not a global or per-file lock, and the page cacheline is modified in do_no_page to increment _count and _mapcount anyway, so a further modification should not be a large performance hit. Scalability is not an issue. This patch implements this latter approach. ->nopage implementations return with the page locked if it is possible for their underlying file to be invalidated (in that case, they must set a special vm_flags bit to indicate so). do_no_page only unlocks the page after setting up the mapping completely. invalidation is excluded because it holds the page lock during invalidation of each page (and ensures that the page is not mapped while holding the lock). This also allows significant simplifications in do_no_page, because we have the page locked in the right place in the pagecache from the start. Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:57 +00:00
mm: merge populate and nopage into fault (fixes nonlinear) Nonlinear mappings are (AFAIKS) simply a virtual memory concept that encodes the virtual address -> file offset differently from linear mappings. ->populate is a layering violation because the filesystem/pagecache code should need to know anything about the virtual memory mapping. The hitch here is that the ->nopage handler didn't pass down enough information (ie. pgoff). But it is more logical to pass pgoff rather than have the ->nopage function calculate it itself anyway (because that's a similar layering violation). Having the populate handler install the pte itself is likewise a nasty thing to be doing. This patch introduces a new fault handler that replaces ->nopage and ->populate and (later) ->nopfn. Most of the old mechanism is still in place so there is a lot of duplication and nice cleanups that can be removed if everyone switches over. The rationale for doing this in the first place is that nonlinear mappings are subject to the pagefault vs invalidate/truncate race too, and it seemed stupid to duplicate the synchronisation logic rather than just consolidate the two. After this patch, MAP_NONBLOCK no longer sets up ptes for pages present in pagecache. Seems like a fringe functionality anyway. NOPAGE_REFAULT is removed. This should be implemented with ->fault, and no users have hit mainline yet. [akpm@linux-foundation.org: cleanup] [randy.dunlap@oracle.com: doc. fixes for readahead] [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:59 +00:00
static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
unsigned long address, pte_t *page_table, pmd_t *pmd,
unsigned int flags, pte_t orig_pte)
mm: merge populate and nopage into fault (fixes nonlinear) Nonlinear mappings are (AFAIKS) simply a virtual memory concept that encodes the virtual address -> file offset differently from linear mappings. ->populate is a layering violation because the filesystem/pagecache code should need to know anything about the virtual memory mapping. The hitch here is that the ->nopage handler didn't pass down enough information (ie. pgoff). But it is more logical to pass pgoff rather than have the ->nopage function calculate it itself anyway (because that's a similar layering violation). Having the populate handler install the pte itself is likewise a nasty thing to be doing. This patch introduces a new fault handler that replaces ->nopage and ->populate and (later) ->nopfn. Most of the old mechanism is still in place so there is a lot of duplication and nice cleanups that can be removed if everyone switches over. The rationale for doing this in the first place is that nonlinear mappings are subject to the pagefault vs invalidate/truncate race too, and it seemed stupid to duplicate the synchronisation logic rather than just consolidate the two. After this patch, MAP_NONBLOCK no longer sets up ptes for pages present in pagecache. Seems like a fringe functionality anyway. NOPAGE_REFAULT is removed. This should be implemented with ->fault, and no users have hit mainline yet. [akpm@linux-foundation.org: cleanup] [randy.dunlap@oracle.com: doc. fixes for readahead] [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:59 +00:00
{
pgoff_t pgoff = (((address & PAGE_MASK)
- vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
mm: merge populate and nopage into fault (fixes nonlinear) Nonlinear mappings are (AFAIKS) simply a virtual memory concept that encodes the virtual address -> file offset differently from linear mappings. ->populate is a layering violation because the filesystem/pagecache code should need to know anything about the virtual memory mapping. The hitch here is that the ->nopage handler didn't pass down enough information (ie. pgoff). But it is more logical to pass pgoff rather than have the ->nopage function calculate it itself anyway (because that's a similar layering violation). Having the populate handler install the pte itself is likewise a nasty thing to be doing. This patch introduces a new fault handler that replaces ->nopage and ->populate and (later) ->nopfn. Most of the old mechanism is still in place so there is a lot of duplication and nice cleanups that can be removed if everyone switches over. The rationale for doing this in the first place is that nonlinear mappings are subject to the pagefault vs invalidate/truncate race too, and it seemed stupid to duplicate the synchronisation logic rather than just consolidate the two. After this patch, MAP_NONBLOCK no longer sets up ptes for pages present in pagecache. Seems like a fringe functionality anyway. NOPAGE_REFAULT is removed. This should be implemented with ->fault, and no users have hit mainline yet. [akpm@linux-foundation.org: cleanup] [randy.dunlap@oracle.com: doc. fixes for readahead] [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:59 +00:00
pte_unmap(page_table);
return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
mm: merge populate and nopage into fault (fixes nonlinear) Nonlinear mappings are (AFAIKS) simply a virtual memory concept that encodes the virtual address -> file offset differently from linear mappings. ->populate is a layering violation because the filesystem/pagecache code should need to know anything about the virtual memory mapping. The hitch here is that the ->nopage handler didn't pass down enough information (ie. pgoff). But it is more logical to pass pgoff rather than have the ->nopage function calculate it itself anyway (because that's a similar layering violation). Having the populate handler install the pte itself is likewise a nasty thing to be doing. This patch introduces a new fault handler that replaces ->nopage and ->populate and (later) ->nopfn. Most of the old mechanism is still in place so there is a lot of duplication and nice cleanups that can be removed if everyone switches over. The rationale for doing this in the first place is that nonlinear mappings are subject to the pagefault vs invalidate/truncate race too, and it seemed stupid to duplicate the synchronisation logic rather than just consolidate the two. After this patch, MAP_NONBLOCK no longer sets up ptes for pages present in pagecache. Seems like a fringe functionality anyway. NOPAGE_REFAULT is removed. This should be implemented with ->fault, and no users have hit mainline yet. [akpm@linux-foundation.org: cleanup] [randy.dunlap@oracle.com: doc. fixes for readahead] [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:59 +00:00
}
/*
* Fault of a previously existing named mapping. Repopulate the pte
* from the encoded file_pte if possible. This enables swappable
* nonlinear vmas.
*
* We enter with non-exclusive mmap_sem (to exclude vma changes,
* but allow concurrent faults), and pte mapped but not yet locked.
* We return with mmap_sem still held, but pte unmapped and unlocked.
*/
static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
unsigned long address, pte_t *page_table, pmd_t *pmd,
unsigned int flags, pte_t orig_pte)
{
pgoff_t pgoff;
flags |= FAULT_FLAG_NONLINEAR;
[PATCH] mm: split page table lock Christoph Lameter demonstrated very poor scalability on the SGI 512-way, with a many-threaded application which concurrently initializes different parts of a large anonymous area. This patch corrects that, by using a separate spinlock per page table page, to guard the page table entries in that page, instead of using the mm's single page_table_lock. (But even then, page_table_lock is still used to guard page table allocation, and anon_vma allocation.) In this implementation, the spinlock is tucked inside the struct page of the page table page: with a BUILD_BUG_ON in case it overflows - which it would in the case of 32-bit PA-RISC with spinlock debugging enabled. Splitting the lock is not quite for free: another cacheline access. Ideally, I suppose we would use split ptlock only for multi-threaded processes on multi-cpu machines; but deciding that dynamically would have its own costs. So for now enable it by config, at some number of cpus - since the Kconfig language doesn't support inequalities, let preprocessor compare that with NR_CPUS. But I don't think it's worth being user-configurable: for good testing of both split and unsplit configs, split now at 4 cpus, and perhaps change that to 8 later. There is a benefit even for singly threaded processes: kswapd can be attacking one part of the mm while another part is busy faulting. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:40 +00:00
if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
mm: fault feedback #2 This patch completes Linus's wish that the fault return codes be made into bit flags, which I agree makes everything nicer. This requires requires all handle_mm_fault callers to be modified (possibly the modifications should go further and do things like fault accounting in handle_mm_fault -- however that would be for another patch). [akpm@linux-foundation.org: fix alpha build] [akpm@linux-foundation.org: fix s390 build] [akpm@linux-foundation.org: fix sparc build] [akpm@linux-foundation.org: fix sparc64 build] [akpm@linux-foundation.org: fix ia64 build] Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Richard Henderson <rth@twiddle.net> Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Ian Molton <spyro@f2s.com> Cc: Bryan Wu <bryan.wu@analog.com> Cc: Mikael Starvik <starvik@axis.com> Cc: David Howells <dhowells@redhat.com> Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Hirokazu Takata <takata@linux-m32r.org> Cc: Geert Uytterhoeven <geert@linux-m68k.org> Cc: Roman Zippel <zippel@linux-m68k.org> Cc: Greg Ungerer <gerg@uclinux.org> Cc: Matthew Wilcox <willy@debian.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Kazumoto Kojima <kkojima@rr.iij4u.or.jp> Cc: Richard Curnow <rc@rc0.org.uk> Cc: William Lee Irwin III <wli@holomorphy.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Jeff Dike <jdike@addtoit.com> Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it> Cc: Miles Bader <uclinux-v850@lsi.nec.co.jp> Cc: Chris Zankel <chris@zankel.net> Acked-by: Kyle McMartin <kyle@mcmartin.ca> Acked-by: Haavard Skinnemoen <hskinnemoen@atmel.com> Acked-by: Ralf Baechle <ralf@linux-mips.org> Acked-by: Andi Kleen <ak@muc.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> [ Still apparently needs some ARM and PPC loving - Linus ] Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:47:05 +00:00
return 0;
if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
/*
* Page table corrupted: show pte and kill process.
*/
badpage: replace page_remove_rmap Eeek and BUG Now that bad pages are kept out of circulation, there is no need for the infamous page_remove_rmap() BUG() - once that page is freed, its negative mapcount will issue a "Bad page state" message and the page won't be freed. Removing the BUG() allows more info, on subsequent pages, to be gathered. We do have more info about the page at this point than bad_page() can know - notably, what the pmd is, which might pinpoint something like low 64kB corruption - but page_remove_rmap() isn't given the address to find that. In practice, there is only one call to page_remove_rmap() which has ever reported anything, that from zap_pte_range() (usually on exit, sometimes on munmap). It has all the info, so remove page_remove_rmap()'s "Eeek" message and leave it all to zap_pte_range(). mm/memory.c already has a hardly used print_bad_pte() function, showing some of the appropriate info: extend it to show what we want for the rmap case: pte info, page info (when there is a page) and vma info to compare. zap_pte_range() already knows the pmd, but print_bad_pte() is easier to use if it works that out for itself. Some of this info is also shown in bad_page()'s "Bad page state" message. Keep them separate, but adjust them to match each other as far as possible. Say "Bad page map" in print_bad_pte(), and add a TAINT_BAD_PAGE there too. print_bad_pte() show current->comm unconditionally (though it should get repeated in the usually irrelevant stack trace): sorry, I misled Nick Piggin to make it conditional on vm_mm == current->mm, but current->mm is already NULL in the exit case. Usually current->comm is good, though exceptionally it may not be that of the mm (when "swapoff" for example). Signed-off-by: Hugh Dickins <hugh@veritas.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-06 22:40:08 +00:00
print_bad_pte(vma, address, orig_pte, NULL);
return VM_FAULT_SIGBUS;
}
pgoff = pte_to_pgoff(orig_pte);
return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
}
int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
unsigned long addr, int current_nid)
{
get_page(page);
count_vm_numa_event(NUMA_HINT_FAULTS);
if (current_nid == numa_node_id())
count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL);
return mpol_misplaced(page, vma, addr);
}
int do_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,
unsigned long addr, pte_t pte, pte_t *ptep, pmd_t *pmd)
{
struct page *page = NULL;
spinlock_t *ptl;
int current_nid = -1;
int target_nid;
bool migrated = false;
/*
* The "pte" at this point cannot be used safely without
* validation through pte_unmap_same(). It's of NUMA type but
* the pfn may be screwed if the read is non atomic.
*
* ptep_modify_prot_start is not called as this is clearing
* the _PAGE_NUMA bit and it is not really expected that there
* would be concurrent hardware modifications to the PTE.
*/
ptl = pte_lockptr(mm, pmd);
spin_lock(ptl);
if (unlikely(!pte_same(*ptep, pte))) {
pte_unmap_unlock(ptep, ptl);
goto out;
}
pte = pte_mknonnuma(pte);
set_pte_at(mm, addr, ptep, pte);
update_mmu_cache(vma, addr, ptep);
page = vm_normal_page(vma, addr, pte);
if (!page) {
pte_unmap_unlock(ptep, ptl);
return 0;
}
current_nid = page_to_nid(page);
target_nid = numa_migrate_prep(page, vma, addr, current_nid);
pte_unmap_unlock(ptep, ptl);
if (target_nid == -1) {
/*
* Account for the fault against the current node if it not
* being replaced regardless of where the page is located.
*/
current_nid = numa_node_id();
put_page(page);
goto out;
}
/* Migrate to the requested node */
migrated = migrate_misplaced_page(page, target_nid);
if (migrated)
current_nid = target_nid;
out:
if (current_nid != -1)
task_numa_fault(current_nid, 1, migrated);
return 0;
}
/* NUMA hinting page fault entry point for regular pmds */
#ifdef CONFIG_NUMA_BALANCING
static int do_pmd_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,
unsigned long addr, pmd_t *pmdp)
{
pmd_t pmd;
pte_t *pte, *orig_pte;
unsigned long _addr = addr & PMD_MASK;
unsigned long offset;
spinlock_t *ptl;
bool numa = false;
mm: numa: Add pte updates, hinting and migration stats It is tricky to quantify the basic cost of automatic NUMA placement in a meaningful manner. This patch adds some vmstats that can be used as part of a basic costing model. u = basic unit = sizeof(void *) Ca = cost of struct page access = sizeof(struct page) / u Cpte = Cost PTE access = Ca Cupdate = Cost PTE update = (2 * Cpte) + (2 * Wlock) where Cpte is incurred twice for a read and a write and Wlock is a constant representing the cost of taking or releasing a lock Cnumahint = Cost of a minor page fault = some high constant e.g. 1000 Cpagerw = Cost to read or write a full page = Ca + PAGE_SIZE/u Ci = Cost of page isolation = Ca + Wi where Wi is a constant that should reflect the approximate cost of the locking operation Cpagecopy = Cpagerw + (Cpagerw * Wnuma) + Ci + (Ci * Wnuma) where Wnuma is the approximate NUMA factor. 1 is local. 1.2 would imply that remote accesses are 20% more expensive Balancing cost = Cpte * numa_pte_updates + Cnumahint * numa_hint_faults + Ci * numa_pages_migrated + Cpagecopy * numa_pages_migrated Note that numa_pages_migrated is used as a measure of how many pages were isolated even though it would miss pages that failed to migrate. A vmstat counter could have been added for it but the isolation cost is pretty marginal in comparison to the overall cost so it seemed overkill. The ideal way to measure automatic placement benefit would be to count the number of remote accesses versus local accesses and do something like benefit = (remote_accesses_before - remove_access_after) * Wnuma but the information is not readily available. As a workload converges, the expection would be that the number of remote numa hints would reduce to 0. convergence = numa_hint_faults_local / numa_hint_faults where this is measured for the last N number of numa hints recorded. When the workload is fully converged the value is 1. This can measure if the placement policy is converging and how fast it is doing it. Signed-off-by: Mel Gorman <mgorman@suse.de> Acked-by: Rik van Riel <riel@redhat.com>
2012-11-02 14:52:48 +00:00
int local_nid = numa_node_id();
spin_lock(&mm->page_table_lock);
pmd = *pmdp;
if (pmd_numa(pmd)) {
set_pmd_at(mm, _addr, pmdp, pmd_mknonnuma(pmd));
numa = true;
}
spin_unlock(&mm->page_table_lock);
if (!numa)
return 0;
/* we're in a page fault so some vma must be in the range */
BUG_ON(!vma);
BUG_ON(vma->vm_start >= _addr + PMD_SIZE);
offset = max(_addr, vma->vm_start) & ~PMD_MASK;
VM_BUG_ON(offset >= PMD_SIZE);
orig_pte = pte = pte_offset_map_lock(mm, pmdp, _addr, &ptl);
pte += offset >> PAGE_SHIFT;
for (addr = _addr + offset; addr < _addr + PMD_SIZE; pte++, addr += PAGE_SIZE) {
pte_t pteval = *pte;
struct page *page;
int curr_nid = local_nid;
int target_nid;
bool migrated;
if (!pte_present(pteval))
continue;
if (!pte_numa(pteval))
continue;
if (addr >= vma->vm_end) {
vma = find_vma(mm, addr);
/* there's a pte present so there must be a vma */
BUG_ON(!vma);
BUG_ON(addr < vma->vm_start);
}
if (pte_numa(pteval)) {
pteval = pte_mknonnuma(pteval);
set_pte_at(mm, addr, pte, pteval);
}
page = vm_normal_page(vma, addr, pteval);
if (unlikely(!page))
continue;
/* only check non-shared pages */
if (unlikely(page_mapcount(page) != 1))
continue;
/*
* Note that the NUMA fault is later accounted to either
* the node that is currently running or where the page is
* migrated to.
*/
curr_nid = local_nid;
target_nid = numa_migrate_prep(page, vma, addr,
page_to_nid(page));
if (target_nid == -1) {
put_page(page);
continue;
}
/* Migrate to the requested node */
pte_unmap_unlock(pte, ptl);
migrated = migrate_misplaced_page(page, target_nid);
if (migrated)
curr_nid = target_nid;
task_numa_fault(curr_nid, 1, migrated);
mm: numa: Add pte updates, hinting and migration stats It is tricky to quantify the basic cost of automatic NUMA placement in a meaningful manner. This patch adds some vmstats that can be used as part of a basic costing model. u = basic unit = sizeof(void *) Ca = cost of struct page access = sizeof(struct page) / u Cpte = Cost PTE access = Ca Cupdate = Cost PTE update = (2 * Cpte) + (2 * Wlock) where Cpte is incurred twice for a read and a write and Wlock is a constant representing the cost of taking or releasing a lock Cnumahint = Cost of a minor page fault = some high constant e.g. 1000 Cpagerw = Cost to read or write a full page = Ca + PAGE_SIZE/u Ci = Cost of page isolation = Ca + Wi where Wi is a constant that should reflect the approximate cost of the locking operation Cpagecopy = Cpagerw + (Cpagerw * Wnuma) + Ci + (Ci * Wnuma) where Wnuma is the approximate NUMA factor. 1 is local. 1.2 would imply that remote accesses are 20% more expensive Balancing cost = Cpte * numa_pte_updates + Cnumahint * numa_hint_faults + Ci * numa_pages_migrated + Cpagecopy * numa_pages_migrated Note that numa_pages_migrated is used as a measure of how many pages were isolated even though it would miss pages that failed to migrate. A vmstat counter could have been added for it but the isolation cost is pretty marginal in comparison to the overall cost so it seemed overkill. The ideal way to measure automatic placement benefit would be to count the number of remote accesses versus local accesses and do something like benefit = (remote_accesses_before - remove_access_after) * Wnuma but the information is not readily available. As a workload converges, the expection would be that the number of remote numa hints would reduce to 0. convergence = numa_hint_faults_local / numa_hint_faults where this is measured for the last N number of numa hints recorded. When the workload is fully converged the value is 1. This can measure if the placement policy is converging and how fast it is doing it. Signed-off-by: Mel Gorman <mgorman@suse.de> Acked-by: Rik van Riel <riel@redhat.com>
2012-11-02 14:52:48 +00:00
pte = pte_offset_map_lock(mm, pmdp, addr, &ptl);
}
pte_unmap_unlock(orig_pte, ptl);
return 0;
}
#else
static int do_pmd_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,
unsigned long addr, pmd_t *pmdp)
{
BUG();
return 0;
}
#endif /* CONFIG_NUMA_BALANCING */
/*
* These routines also need to handle stuff like marking pages dirty
* and/or accessed for architectures that don't do it in hardware (most
* RISC architectures). The early dirtying is also good on the i386.
*
* There is also a hook called "update_mmu_cache()" that architectures
* with external mmu caches can use to update those (ie the Sparc or
* PowerPC hashed page tables that act as extended TLBs).
*
* We enter with non-exclusive mmap_sem (to exclude vma changes,
* but allow concurrent faults), and pte mapped but not yet locked.
* We return with mmap_sem still held, but pte unmapped and unlocked.
*/
thp: transparent hugepage core Lately I've been working to make KVM use hugepages transparently without the usual restrictions of hugetlbfs. Some of the restrictions I'd like to see removed: 1) hugepages have to be swappable or the guest physical memory remains locked in RAM and can't be paged out to swap 2) if a hugepage allocation fails, regular pages should be allocated instead and mixed in the same vma without any failure and without userland noticing 3) if some task quits and more hugepages become available in the buddy, guest physical memory backed by regular pages should be relocated on hugepages automatically in regions under madvise(MADV_HUGEPAGE) (ideally event driven by waking up the kernel deamon if the order=HPAGE_PMD_SHIFT-PAGE_SHIFT list becomes not null) 4) avoidance of reservation and maximization of use of hugepages whenever possible. Reservation (needed to avoid runtime fatal faliures) may be ok for 1 machine with 1 database with 1 database cache with 1 database cache size known at boot time. It's definitely not feasible with a virtualization hypervisor usage like RHEV-H that runs an unknown number of virtual machines with an unknown size of each virtual machine with an unknown amount of pagecache that could be potentially useful in the host for guest not using O_DIRECT (aka cache=off). hugepages in the virtualization hypervisor (and also in the guest!) are much more important than in a regular host not using virtualization, becasue with NPT/EPT they decrease the tlb-miss cacheline accesses from 24 to 19 in case only the hypervisor uses transparent hugepages, and they decrease the tlb-miss cacheline accesses from 19 to 15 in case both the linux hypervisor and the linux guest both uses this patch (though the guest will limit the addition speedup to anonymous regions only for now...). Even more important is that the tlb miss handler is much slower on a NPT/EPT guest than for a regular shadow paging or no-virtualization scenario. So maximizing the amount of virtual memory cached by the TLB pays off significantly more with NPT/EPT than without (even if there would be no significant speedup in the tlb-miss runtime). The first (and more tedious) part of this work requires allowing the VM to handle anonymous hugepages mixed with regular pages transparently on regular anonymous vmas. This is what this patch tries to achieve in the least intrusive possible way. We want hugepages and hugetlb to be used in a way so that all applications can benefit without changes (as usual we leverage the KVM virtualization design: by improving the Linux VM at large, KVM gets the performance boost too). The most important design choice is: always fallback to 4k allocation if the hugepage allocation fails! This is the _very_ opposite of some large pagecache patches that failed with -EIO back then if a 64k (or similar) allocation failed... Second important decision (to reduce the impact of the feature on the existing pagetable handling code) is that at any time we can split an hugepage into 512 regular pages and it has to be done with an operation that can't fail. This way the reliability of the swapping isn't decreased (no need to allocate memory when we are short on memory to swap) and it's trivial to plug a split_huge_page* one-liner where needed without polluting the VM. Over time we can teach mprotect, mremap and friends to handle pmd_trans_huge natively without calling split_huge_page*. The fact it can't fail isn't just for swap: if split_huge_page would return -ENOMEM (instead of the current void) we'd need to rollback the mprotect from the middle of it (ideally including undoing the split_vma) which would be a big change and in the very wrong direction (it'd likely be simpler not to call split_huge_page at all and to teach mprotect and friends to handle hugepages instead of rolling them back from the middle). In short the very value of split_huge_page is that it can't fail. The collapsing and madvise(MADV_HUGEPAGE) part will remain separated and incremental and it'll just be an "harmless" addition later if this initial part is agreed upon. It also should be noted that locking-wise replacing regular pages with hugepages is going to be very easy if compared to what I'm doing below in split_huge_page, as it will only happen when page_count(page) matches page_mapcount(page) if we can take the PG_lock and mmap_sem in write mode. collapse_huge_page will be a "best effort" that (unlike split_huge_page) can fail at the minimal sign of trouble and we can try again later. collapse_huge_page will be similar to how KSM works and the madvise(MADV_HUGEPAGE) will work similar to madvise(MADV_MERGEABLE). The default I like is that transparent hugepages are used at page fault time. This can be changed with /sys/kernel/mm/transparent_hugepage/enabled. The control knob can be set to three values "always", "madvise", "never" which mean respectively that hugepages are always used, or only inside madvise(MADV_HUGEPAGE) regions, or never used. /sys/kernel/mm/transparent_hugepage/defrag instead controls if the hugepage allocation should defrag memory aggressively "always", only inside "madvise" regions, or "never". The pmd_trans_splitting/pmd_trans_huge locking is very solid. The put_page (from get_user_page users that can't use mmu notifier like O_DIRECT) that runs against a __split_huge_page_refcount instead was a pain to serialize in a way that would result always in a coherent page count for both tail and head. I think my locking solution with a compound_lock taken only after the page_first is valid and is still a PageHead should be safe but it surely needs review from SMP race point of view. In short there is no current existing way to serialize the O_DIRECT final put_page against split_huge_page_refcount so I had to invent a new one (O_DIRECT loses knowledge on the mapping status by the time gup_fast returns so...). And I didn't want to impact all gup/gup_fast users for now, maybe if we change the gup interface substantially we can avoid this locking, I admit I didn't think too much about it because changing the gup unpinning interface would be invasive. If we ignored O_DIRECT we could stick to the existing compound refcounting code, by simply adding a get_user_pages_fast_flags(foll_flags) where KVM (and any other mmu notifier user) would call it without FOLL_GET (and if FOLL_GET isn't set we'd just BUG_ON if nobody registered itself in the current task mmu notifier list yet). But O_DIRECT is fundamental for decent performance of virtualized I/O on fast storage so we can't avoid it to solve the race of put_page against split_huge_page_refcount to achieve a complete hugepage feature for KVM. Swap and oom works fine (well just like with regular pages ;). MMU notifier is handled transparently too, with the exception of the young bit on the pmd, that didn't have a range check but I think KVM will be fine because the whole point of hugepages is that EPT/NPT will also use a huge pmd when they notice gup returns pages with PageCompound set, so they won't care of a range and there's just the pmd young bit to check in that case. NOTE: in some cases if the L2 cache is small, this may slowdown and waste memory during COWs because 4M of memory are accessed in a single fault instead of 8k (the payoff is that after COW the program can run faster). So we might want to switch the copy_huge_page (and clear_huge_page too) to not temporal stores. I also extensively researched ways to avoid this cache trashing with a full prefault logic that would cow in 8k/16k/32k/64k up to 1M (I can send those patches that fully implemented prefault) but I concluded they're not worth it and they add an huge additional complexity and they remove all tlb benefits until the full hugepage has been faulted in, to save a little bit of memory and some cache during app startup, but they still don't improve substantially the cache-trashing during startup if the prefault happens in >4k chunks. One reason is that those 4k pte entries copied are still mapped on a perfectly cache-colored hugepage, so the trashing is the worst one can generate in those copies (cow of 4k page copies aren't so well colored so they trashes less, but again this results in software running faster after the page fault). Those prefault patches allowed things like a pte where post-cow pages were local 4k regular anon pages and the not-yet-cowed pte entries were pointing in the middle of some hugepage mapped read-only. If it doesn't payoff substantially with todays hardware it will payoff even less in the future with larger l2 caches, and the prefault logic would blot the VM a lot. If one is emebdded transparent_hugepage can be disabled during boot with sysfs or with the boot commandline parameter transparent_hugepage=0 (or transparent_hugepage=2 to restrict hugepages inside madvise regions) that will ensure not a single hugepage is allocated at boot time. It is simple enough to just disable transparent hugepage globally and let transparent hugepages be allocated selectively by applications in the MADV_HUGEPAGE region (both at page fault time, and if enabled with the collapse_huge_page too through the kernel daemon). This patch supports only hugepages mapped in the pmd, archs that have smaller hugepages will not fit in this patch alone. Also some archs like power have certain tlb limits that prevents mixing different page size in the same regions so they will not fit in this framework that requires "graceful fallback" to basic PAGE_SIZE in case of physical memory fragmentation. hugetlbfs remains a perfect fit for those because its software limits happen to match the hardware limits. hugetlbfs also remains a perfect fit for hugepage sizes like 1GByte that cannot be hoped to be found not fragmented after a certain system uptime and that would be very expensive to defragment with relocation, so requiring reservation. hugetlbfs is the "reservation way", the point of transparent hugepages is not to have any reservation at all and maximizing the use of cache and hugepages at all times automatically. Some performance result: vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largep ages3 memset page fault 1566023 memset tlb miss 453854 memset second tlb miss 453321 random access tlb miss 41635 random access second tlb miss 41658 vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largepages3 memset page fault 1566471 memset tlb miss 453375 memset second tlb miss 453320 random access tlb miss 41636 random access second tlb miss 41637 vmx andrea # ./largepages3 memset page fault 1566642 memset tlb miss 453417 memset second tlb miss 453313 random access tlb miss 41630 random access second tlb miss 41647 vmx andrea # ./largepages3 memset page fault 1566872 memset tlb miss 453418 memset second tlb miss 453315 random access tlb miss 41618 random access second tlb miss 41659 vmx andrea # echo 0 > /proc/sys/vm/transparent_hugepage vmx andrea # ./largepages3 memset page fault 2182476 memset tlb miss 460305 memset second tlb miss 460179 random access tlb miss 44483 random access second tlb miss 44186 vmx andrea # ./largepages3 memset page fault 2182791 memset tlb miss 460742 memset second tlb miss 459962 random access tlb miss 43981 random access second tlb miss 43988 ============ #include <stdio.h> #include <stdlib.h> #include <string.h> #include <sys/time.h> #define SIZE (3UL*1024*1024*1024) int main() { char *p = malloc(SIZE), *p2; struct timeval before, after; gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset page fault %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); return 0; } ============ Signed-off-by: Andrea Arcangeli <aarcange@redhat.com> Acked-by: Rik van Riel <riel@redhat.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-13 23:46:52 +00:00
int handle_pte_fault(struct mm_struct *mm,
struct vm_area_struct *vma, unsigned long address,
pte_t *pte, pmd_t *pmd, unsigned int flags)
{
pte_t entry;
spinlock_t *ptl;
entry = *pte;
if (!pte_present(entry)) {
if (pte_none(entry)) {
if (vma->vm_ops) {
if (likely(vma->vm_ops->fault))
mm: merge populate and nopage into fault (fixes nonlinear) Nonlinear mappings are (AFAIKS) simply a virtual memory concept that encodes the virtual address -> file offset differently from linear mappings. ->populate is a layering violation because the filesystem/pagecache code should need to know anything about the virtual memory mapping. The hitch here is that the ->nopage handler didn't pass down enough information (ie. pgoff). But it is more logical to pass pgoff rather than have the ->nopage function calculate it itself anyway (because that's a similar layering violation). Having the populate handler install the pte itself is likewise a nasty thing to be doing. This patch introduces a new fault handler that replaces ->nopage and ->populate and (later) ->nopfn. Most of the old mechanism is still in place so there is a lot of duplication and nice cleanups that can be removed if everyone switches over. The rationale for doing this in the first place is that nonlinear mappings are subject to the pagefault vs invalidate/truncate race too, and it seemed stupid to duplicate the synchronisation logic rather than just consolidate the two. After this patch, MAP_NONBLOCK no longer sets up ptes for pages present in pagecache. Seems like a fringe functionality anyway. NOPAGE_REFAULT is removed. This should be implemented with ->fault, and no users have hit mainline yet. [akpm@linux-foundation.org: cleanup] [randy.dunlap@oracle.com: doc. fixes for readahead] [akpm@linux-foundation.org: build fix] Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com> Cc: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:46:59 +00:00
return do_linear_fault(mm, vma, address,
pte, pmd, flags, entry);
}
return do_anonymous_page(mm, vma, address,
pte, pmd, flags);
}
if (pte_file(entry))
return do_nonlinear_fault(mm, vma, address,
pte, pmd, flags, entry);
return do_swap_page(mm, vma, address,
pte, pmd, flags, entry);
}
if (pte_numa(entry))
return do_numa_page(mm, vma, address, entry, pte, pmd);
[PATCH] mm: split page table lock Christoph Lameter demonstrated very poor scalability on the SGI 512-way, with a many-threaded application which concurrently initializes different parts of a large anonymous area. This patch corrects that, by using a separate spinlock per page table page, to guard the page table entries in that page, instead of using the mm's single page_table_lock. (But even then, page_table_lock is still used to guard page table allocation, and anon_vma allocation.) In this implementation, the spinlock is tucked inside the struct page of the page table page: with a BUILD_BUG_ON in case it overflows - which it would in the case of 32-bit PA-RISC with spinlock debugging enabled. Splitting the lock is not quite for free: another cacheline access. Ideally, I suppose we would use split ptlock only for multi-threaded processes on multi-cpu machines; but deciding that dynamically would have its own costs. So for now enable it by config, at some number of cpus - since the Kconfig language doesn't support inequalities, let preprocessor compare that with NR_CPUS. But I don't think it's worth being user-configurable: for good testing of both split and unsplit configs, split now at 4 cpus, and perhaps change that to 8 later. There is a benefit even for singly threaded processes: kswapd can be attacking one part of the mm while another part is busy faulting. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:40 +00:00
ptl = pte_lockptr(mm, pmd);
spin_lock(ptl);
if (unlikely(!pte_same(*pte, entry)))
goto unlock;
if (flags & FAULT_FLAG_WRITE) {
if (!pte_write(entry))
return do_wp_page(mm, vma, address,
pte, pmd, ptl, entry);
entry = pte_mkdirty(entry);
}
entry = pte_mkyoung(entry);
if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) {
update_mmu_cache(vma, address, pte);
} else {
/*
* This is needed only for protection faults but the arch code
* is not yet telling us if this is a protection fault or not.
* This still avoids useless tlb flushes for .text page faults
* with threads.
*/
if (flags & FAULT_FLAG_WRITE)
flush_tlb_fix_spurious_fault(vma, address);
}
unlock:
pte_unmap_unlock(pte, ptl);
mm: fault feedback #2 This patch completes Linus's wish that the fault return codes be made into bit flags, which I agree makes everything nicer. This requires requires all handle_mm_fault callers to be modified (possibly the modifications should go further and do things like fault accounting in handle_mm_fault -- however that would be for another patch). [akpm@linux-foundation.org: fix alpha build] [akpm@linux-foundation.org: fix s390 build] [akpm@linux-foundation.org: fix sparc build] [akpm@linux-foundation.org: fix sparc64 build] [akpm@linux-foundation.org: fix ia64 build] Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Richard Henderson <rth@twiddle.net> Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Ian Molton <spyro@f2s.com> Cc: Bryan Wu <bryan.wu@analog.com> Cc: Mikael Starvik <starvik@axis.com> Cc: David Howells <dhowells@redhat.com> Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Hirokazu Takata <takata@linux-m32r.org> Cc: Geert Uytterhoeven <geert@linux-m68k.org> Cc: Roman Zippel <zippel@linux-m68k.org> Cc: Greg Ungerer <gerg@uclinux.org> Cc: Matthew Wilcox <willy@debian.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Kazumoto Kojima <kkojima@rr.iij4u.or.jp> Cc: Richard Curnow <rc@rc0.org.uk> Cc: William Lee Irwin III <wli@holomorphy.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Jeff Dike <jdike@addtoit.com> Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it> Cc: Miles Bader <uclinux-v850@lsi.nec.co.jp> Cc: Chris Zankel <chris@zankel.net> Acked-by: Kyle McMartin <kyle@mcmartin.ca> Acked-by: Haavard Skinnemoen <hskinnemoen@atmel.com> Acked-by: Ralf Baechle <ralf@linux-mips.org> Acked-by: Andi Kleen <ak@muc.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> [ Still apparently needs some ARM and PPC loving - Linus ] Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:47:05 +00:00
return 0;
}
/*
* By the time we get here, we already hold the mm semaphore
*/
mm: fault feedback #2 This patch completes Linus's wish that the fault return codes be made into bit flags, which I agree makes everything nicer. This requires requires all handle_mm_fault callers to be modified (possibly the modifications should go further and do things like fault accounting in handle_mm_fault -- however that would be for another patch). [akpm@linux-foundation.org: fix alpha build] [akpm@linux-foundation.org: fix s390 build] [akpm@linux-foundation.org: fix sparc build] [akpm@linux-foundation.org: fix sparc64 build] [akpm@linux-foundation.org: fix ia64 build] Signed-off-by: Nick Piggin <npiggin@suse.de> Cc: Richard Henderson <rth@twiddle.net> Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Ian Molton <spyro@f2s.com> Cc: Bryan Wu <bryan.wu@analog.com> Cc: Mikael Starvik <starvik@axis.com> Cc: David Howells <dhowells@redhat.com> Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Hirokazu Takata <takata@linux-m32r.org> Cc: Geert Uytterhoeven <geert@linux-m68k.org> Cc: Roman Zippel <zippel@linux-m68k.org> Cc: Greg Ungerer <gerg@uclinux.org> Cc: Matthew Wilcox <willy@debian.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Kazumoto Kojima <kkojima@rr.iij4u.or.jp> Cc: Richard Curnow <rc@rc0.org.uk> Cc: William Lee Irwin III <wli@holomorphy.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Jeff Dike <jdike@addtoit.com> Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it> Cc: Miles Bader <uclinux-v850@lsi.nec.co.jp> Cc: Chris Zankel <chris@zankel.net> Acked-by: Kyle McMartin <kyle@mcmartin.ca> Acked-by: Haavard Skinnemoen <hskinnemoen@atmel.com> Acked-by: Ralf Baechle <ralf@linux-mips.org> Acked-by: Andi Kleen <ak@muc.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> [ Still apparently needs some ARM and PPC loving - Linus ] Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-19 08:47:05 +00:00
int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
unsigned long address, unsigned int flags)
{
pgd_t *pgd;
pud_t *pud;
pmd_t *pmd;
pte_t *pte;
__set_current_state(TASK_RUNNING);
[PATCH] Light weight event counters The remaining counters in page_state after the zoned VM counter patches have been applied are all just for show in /proc/vmstat. They have no essential function for the VM. We use a simple increment of per cpu variables. In order to avoid the most severe races we disable preempt. Preempt does not prevent the race between an increment and an interrupt handler incrementing the same statistics counter. However, that race is exceedingly rare, we may only loose one increment or so and there is no requirement (at least not in kernel) that the vm event counters have to be accurate. In the non preempt case this results in a simple increment for each counter. For many architectures this will be reduced by the compiler to a single instruction. This single instruction is atomic for i386 and x86_64. And therefore even the rare race condition in an interrupt is avoided for both architectures in most cases. The patchset also adds an off switch for embedded systems that allows a building of linux kernels without these counters. The implementation of these counters is through inline code that hopefully results in only a single instruction increment instruction being emitted (i386, x86_64) or in the increment being hidden though instruction concurrency (EPIC architectures such as ia64 can get that done). Benefits: - VM event counter operations usually reduce to a single inline instruction on i386 and x86_64. - No interrupt disable, only preempt disable for the preempt case. Preempt disable can also be avoided by moving the counter into a spinlock. - Handling is similar to zoned VM counters. - Simple and easily extendable. - Can be omitted to reduce memory use for embedded use. References: RFC http://marc.theaimsgroup.com/?l=linux-kernel&m=113512330605497&w=2 RFC http://marc.theaimsgroup.com/?l=linux-kernel&m=114988082814934&w=2 local_t http://marc.theaimsgroup.com/?l=linux-kernel&m=114991748606690&w=2 V2 http://marc.theaimsgroup.com/?t=115014808400007&r=1&w=2 V3 http://marc.theaimsgroup.com/?l=linux-kernel&m=115024767022346&w=2 V4 http://marc.theaimsgroup.com/?l=linux-kernel&m=115047968808926&w=2 Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-30 08:55:45 +00:00
count_vm_event(PGFAULT);
memcg: add the pagefault count into memcg stats Two new stats in per-memcg memory.stat which tracks the number of page faults and number of major page faults. "pgfault" "pgmajfault" They are different from "pgpgin"/"pgpgout" stat which count number of pages charged/discharged to the cgroup and have no meaning of reading/ writing page to disk. It is valuable to track the two stats for both measuring application's performance as well as the efficiency of the kernel page reclaim path. Counting pagefaults per process is useful, but we also need the aggregated value since processes are monitored and controlled in cgroup basis in memcg. Functional test: check the total number of pgfault/pgmajfault of all memcgs and compare with global vmstat value: $ cat /proc/vmstat | grep fault pgfault 1070751 pgmajfault 553 $ cat /dev/cgroup/memory.stat | grep fault pgfault 1071138 pgmajfault 553 total_pgfault 1071142 total_pgmajfault 553 $ cat /dev/cgroup/A/memory.stat | grep fault pgfault 199 pgmajfault 0 total_pgfault 199 total_pgmajfault 0 Performance test: run page fault test(pft) wit 16 thread on faulting in 15G anon pages in 16G container. There is no regression noticed on the "flt/cpu/s" Sample output from pft: TAG pft:anon-sys-default: Gb Thr CLine User System Wall flt/cpu/s fault/wsec 15 16 1 0.67s 233.41s 14.76s 16798.546 266356.260 +-------------------------------------------------------------------------+ N Min Max Median Avg Stddev x 10 16682.962 17344.027 16913.524 16928.812 166.5362 + 10 16695.568 16923.896 16820.604 16824.652 84.816568 No difference proven at 95.0% confidence [akpm@linux-foundation.org: fix build] [hughd@google.com: shmem fix] Signed-off-by: Ying Han <yinghan@google.com> Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Reviewed-by: Minchan Kim <minchan.kim@gmail.com> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com> Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-05-26 23:25:38 +00:00
mem_cgroup_count_vm_event(mm, PGFAULT);
/* do counter updates before entering really critical section. */
check_sync_rss_stat(current);
if (unlikely(is_vm_hugetlb_page(vma)))
return hugetlb_fault(mm, vma, address, flags);
retry:
pgd = pgd_offset(mm, address);
pud = pud_alloc(mm, pgd, address);
if (!pud)
return VM_FAULT_OOM;
pmd = pmd_alloc(mm, pud, address);
if (!pmd)
return VM_FAULT_OOM;
thp: transparent hugepage core Lately I've been working to make KVM use hugepages transparently without the usual restrictions of hugetlbfs. Some of the restrictions I'd like to see removed: 1) hugepages have to be swappable or the guest physical memory remains locked in RAM and can't be paged out to swap 2) if a hugepage allocation fails, regular pages should be allocated instead and mixed in the same vma without any failure and without userland noticing 3) if some task quits and more hugepages become available in the buddy, guest physical memory backed by regular pages should be relocated on hugepages automatically in regions under madvise(MADV_HUGEPAGE) (ideally event driven by waking up the kernel deamon if the order=HPAGE_PMD_SHIFT-PAGE_SHIFT list becomes not null) 4) avoidance of reservation and maximization of use of hugepages whenever possible. Reservation (needed to avoid runtime fatal faliures) may be ok for 1 machine with 1 database with 1 database cache with 1 database cache size known at boot time. It's definitely not feasible with a virtualization hypervisor usage like RHEV-H that runs an unknown number of virtual machines with an unknown size of each virtual machine with an unknown amount of pagecache that could be potentially useful in the host for guest not using O_DIRECT (aka cache=off). hugepages in the virtualization hypervisor (and also in the guest!) are much more important than in a regular host not using virtualization, becasue with NPT/EPT they decrease the tlb-miss cacheline accesses from 24 to 19 in case only the hypervisor uses transparent hugepages, and they decrease the tlb-miss cacheline accesses from 19 to 15 in case both the linux hypervisor and the linux guest both uses this patch (though the guest will limit the addition speedup to anonymous regions only for now...). Even more important is that the tlb miss handler is much slower on a NPT/EPT guest than for a regular shadow paging or no-virtualization scenario. So maximizing the amount of virtual memory cached by the TLB pays off significantly more with NPT/EPT than without (even if there would be no significant speedup in the tlb-miss runtime). The first (and more tedious) part of this work requires allowing the VM to handle anonymous hugepages mixed with regular pages transparently on regular anonymous vmas. This is what this patch tries to achieve in the least intrusive possible way. We want hugepages and hugetlb to be used in a way so that all applications can benefit without changes (as usual we leverage the KVM virtualization design: by improving the Linux VM at large, KVM gets the performance boost too). The most important design choice is: always fallback to 4k allocation if the hugepage allocation fails! This is the _very_ opposite of some large pagecache patches that failed with -EIO back then if a 64k (or similar) allocation failed... Second important decision (to reduce the impact of the feature on the existing pagetable handling code) is that at any time we can split an hugepage into 512 regular pages and it has to be done with an operation that can't fail. This way the reliability of the swapping isn't decreased (no need to allocate memory when we are short on memory to swap) and it's trivial to plug a split_huge_page* one-liner where needed without polluting the VM. Over time we can teach mprotect, mremap and friends to handle pmd_trans_huge natively without calling split_huge_page*. The fact it can't fail isn't just for swap: if split_huge_page would return -ENOMEM (instead of the current void) we'd need to rollback the mprotect from the middle of it (ideally including undoing the split_vma) which would be a big change and in the very wrong direction (it'd likely be simpler not to call split_huge_page at all and to teach mprotect and friends to handle hugepages instead of rolling them back from the middle). In short the very value of split_huge_page is that it can't fail. The collapsing and madvise(MADV_HUGEPAGE) part will remain separated and incremental and it'll just be an "harmless" addition later if this initial part is agreed upon. It also should be noted that locking-wise replacing regular pages with hugepages is going to be very easy if compared to what I'm doing below in split_huge_page, as it will only happen when page_count(page) matches page_mapcount(page) if we can take the PG_lock and mmap_sem in write mode. collapse_huge_page will be a "best effort" that (unlike split_huge_page) can fail at the minimal sign of trouble and we can try again later. collapse_huge_page will be similar to how KSM works and the madvise(MADV_HUGEPAGE) will work similar to madvise(MADV_MERGEABLE). The default I like is that transparent hugepages are used at page fault time. This can be changed with /sys/kernel/mm/transparent_hugepage/enabled. The control knob can be set to three values "always", "madvise", "never" which mean respectively that hugepages are always used, or only inside madvise(MADV_HUGEPAGE) regions, or never used. /sys/kernel/mm/transparent_hugepage/defrag instead controls if the hugepage allocation should defrag memory aggressively "always", only inside "madvise" regions, or "never". The pmd_trans_splitting/pmd_trans_huge locking is very solid. The put_page (from get_user_page users that can't use mmu notifier like O_DIRECT) that runs against a __split_huge_page_refcount instead was a pain to serialize in a way that would result always in a coherent page count for both tail and head. I think my locking solution with a compound_lock taken only after the page_first is valid and is still a PageHead should be safe but it surely needs review from SMP race point of view. In short there is no current existing way to serialize the O_DIRECT final put_page against split_huge_page_refcount so I had to invent a new one (O_DIRECT loses knowledge on the mapping status by the time gup_fast returns so...). And I didn't want to impact all gup/gup_fast users for now, maybe if we change the gup interface substantially we can avoid this locking, I admit I didn't think too much about it because changing the gup unpinning interface would be invasive. If we ignored O_DIRECT we could stick to the existing compound refcounting code, by simply adding a get_user_pages_fast_flags(foll_flags) where KVM (and any other mmu notifier user) would call it without FOLL_GET (and if FOLL_GET isn't set we'd just BUG_ON if nobody registered itself in the current task mmu notifier list yet). But O_DIRECT is fundamental for decent performance of virtualized I/O on fast storage so we can't avoid it to solve the race of put_page against split_huge_page_refcount to achieve a complete hugepage feature for KVM. Swap and oom works fine (well just like with regular pages ;). MMU notifier is handled transparently too, with the exception of the young bit on the pmd, that didn't have a range check but I think KVM will be fine because the whole point of hugepages is that EPT/NPT will also use a huge pmd when they notice gup returns pages with PageCompound set, so they won't care of a range and there's just the pmd young bit to check in that case. NOTE: in some cases if the L2 cache is small, this may slowdown and waste memory during COWs because 4M of memory are accessed in a single fault instead of 8k (the payoff is that after COW the program can run faster). So we might want to switch the copy_huge_page (and clear_huge_page too) to not temporal stores. I also extensively researched ways to avoid this cache trashing with a full prefault logic that would cow in 8k/16k/32k/64k up to 1M (I can send those patches that fully implemented prefault) but I concluded they're not worth it and they add an huge additional complexity and they remove all tlb benefits until the full hugepage has been faulted in, to save a little bit of memory and some cache during app startup, but they still don't improve substantially the cache-trashing during startup if the prefault happens in >4k chunks. One reason is that those 4k pte entries copied are still mapped on a perfectly cache-colored hugepage, so the trashing is the worst one can generate in those copies (cow of 4k page copies aren't so well colored so they trashes less, but again this results in software running faster after the page fault). Those prefault patches allowed things like a pte where post-cow pages were local 4k regular anon pages and the not-yet-cowed pte entries were pointing in the middle of some hugepage mapped read-only. If it doesn't payoff substantially with todays hardware it will payoff even less in the future with larger l2 caches, and the prefault logic would blot the VM a lot. If one is emebdded transparent_hugepage can be disabled during boot with sysfs or with the boot commandline parameter transparent_hugepage=0 (or transparent_hugepage=2 to restrict hugepages inside madvise regions) that will ensure not a single hugepage is allocated at boot time. It is simple enough to just disable transparent hugepage globally and let transparent hugepages be allocated selectively by applications in the MADV_HUGEPAGE region (both at page fault time, and if enabled with the collapse_huge_page too through the kernel daemon). This patch supports only hugepages mapped in the pmd, archs that have smaller hugepages will not fit in this patch alone. Also some archs like power have certain tlb limits that prevents mixing different page size in the same regions so they will not fit in this framework that requires "graceful fallback" to basic PAGE_SIZE in case of physical memory fragmentation. hugetlbfs remains a perfect fit for those because its software limits happen to match the hardware limits. hugetlbfs also remains a perfect fit for hugepage sizes like 1GByte that cannot be hoped to be found not fragmented after a certain system uptime and that would be very expensive to defragment with relocation, so requiring reservation. hugetlbfs is the "reservation way", the point of transparent hugepages is not to have any reservation at all and maximizing the use of cache and hugepages at all times automatically. Some performance result: vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largep ages3 memset page fault 1566023 memset tlb miss 453854 memset second tlb miss 453321 random access tlb miss 41635 random access second tlb miss 41658 vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largepages3 memset page fault 1566471 memset tlb miss 453375 memset second tlb miss 453320 random access tlb miss 41636 random access second tlb miss 41637 vmx andrea # ./largepages3 memset page fault 1566642 memset tlb miss 453417 memset second tlb miss 453313 random access tlb miss 41630 random access second tlb miss 41647 vmx andrea # ./largepages3 memset page fault 1566872 memset tlb miss 453418 memset second tlb miss 453315 random access tlb miss 41618 random access second tlb miss 41659 vmx andrea # echo 0 > /proc/sys/vm/transparent_hugepage vmx andrea # ./largepages3 memset page fault 2182476 memset tlb miss 460305 memset second tlb miss 460179 random access tlb miss 44483 random access second tlb miss 44186 vmx andrea # ./largepages3 memset page fault 2182791 memset tlb miss 460742 memset second tlb miss 459962 random access tlb miss 43981 random access second tlb miss 43988 ============ #include <stdio.h> #include <stdlib.h> #include <string.h> #include <sys/time.h> #define SIZE (3UL*1024*1024*1024) int main() { char *p = malloc(SIZE), *p2; struct timeval before, after; gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset page fault %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); return 0; } ============ Signed-off-by: Andrea Arcangeli <aarcange@redhat.com> Acked-by: Rik van Riel <riel@redhat.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-13 23:46:52 +00:00
if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) {
if (!vma->vm_ops)
return do_huge_pmd_anonymous_page(mm, vma, address,
pmd, flags);
} else {
pmd_t orig_pmd = *pmd;
int ret;
thp: transparent hugepage core Lately I've been working to make KVM use hugepages transparently without the usual restrictions of hugetlbfs. Some of the restrictions I'd like to see removed: 1) hugepages have to be swappable or the guest physical memory remains locked in RAM and can't be paged out to swap 2) if a hugepage allocation fails, regular pages should be allocated instead and mixed in the same vma without any failure and without userland noticing 3) if some task quits and more hugepages become available in the buddy, guest physical memory backed by regular pages should be relocated on hugepages automatically in regions under madvise(MADV_HUGEPAGE) (ideally event driven by waking up the kernel deamon if the order=HPAGE_PMD_SHIFT-PAGE_SHIFT list becomes not null) 4) avoidance of reservation and maximization of use of hugepages whenever possible. Reservation (needed to avoid runtime fatal faliures) may be ok for 1 machine with 1 database with 1 database cache with 1 database cache size known at boot time. It's definitely not feasible with a virtualization hypervisor usage like RHEV-H that runs an unknown number of virtual machines with an unknown size of each virtual machine with an unknown amount of pagecache that could be potentially useful in the host for guest not using O_DIRECT (aka cache=off). hugepages in the virtualization hypervisor (and also in the guest!) are much more important than in a regular host not using virtualization, becasue with NPT/EPT they decrease the tlb-miss cacheline accesses from 24 to 19 in case only the hypervisor uses transparent hugepages, and they decrease the tlb-miss cacheline accesses from 19 to 15 in case both the linux hypervisor and the linux guest both uses this patch (though the guest will limit the addition speedup to anonymous regions only for now...). Even more important is that the tlb miss handler is much slower on a NPT/EPT guest than for a regular shadow paging or no-virtualization scenario. So maximizing the amount of virtual memory cached by the TLB pays off significantly more with NPT/EPT than without (even if there would be no significant speedup in the tlb-miss runtime). The first (and more tedious) part of this work requires allowing the VM to handle anonymous hugepages mixed with regular pages transparently on regular anonymous vmas. This is what this patch tries to achieve in the least intrusive possible way. We want hugepages and hugetlb to be used in a way so that all applications can benefit without changes (as usual we leverage the KVM virtualization design: by improving the Linux VM at large, KVM gets the performance boost too). The most important design choice is: always fallback to 4k allocation if the hugepage allocation fails! This is the _very_ opposite of some large pagecache patches that failed with -EIO back then if a 64k (or similar) allocation failed... Second important decision (to reduce the impact of the feature on the existing pagetable handling code) is that at any time we can split an hugepage into 512 regular pages and it has to be done with an operation that can't fail. This way the reliability of the swapping isn't decreased (no need to allocate memory when we are short on memory to swap) and it's trivial to plug a split_huge_page* one-liner where needed without polluting the VM. Over time we can teach mprotect, mremap and friends to handle pmd_trans_huge natively without calling split_huge_page*. The fact it can't fail isn't just for swap: if split_huge_page would return -ENOMEM (instead of the current void) we'd need to rollback the mprotect from the middle of it (ideally including undoing the split_vma) which would be a big change and in the very wrong direction (it'd likely be simpler not to call split_huge_page at all and to teach mprotect and friends to handle hugepages instead of rolling them back from the middle). In short the very value of split_huge_page is that it can't fail. The collapsing and madvise(MADV_HUGEPAGE) part will remain separated and incremental and it'll just be an "harmless" addition later if this initial part is agreed upon. It also should be noted that locking-wise replacing regular pages with hugepages is going to be very easy if compared to what I'm doing below in split_huge_page, as it will only happen when page_count(page) matches page_mapcount(page) if we can take the PG_lock and mmap_sem in write mode. collapse_huge_page will be a "best effort" that (unlike split_huge_page) can fail at the minimal sign of trouble and we can try again later. collapse_huge_page will be similar to how KSM works and the madvise(MADV_HUGEPAGE) will work similar to madvise(MADV_MERGEABLE). The default I like is that transparent hugepages are used at page fault time. This can be changed with /sys/kernel/mm/transparent_hugepage/enabled. The control knob can be set to three values "always", "madvise", "never" which mean respectively that hugepages are always used, or only inside madvise(MADV_HUGEPAGE) regions, or never used. /sys/kernel/mm/transparent_hugepage/defrag instead controls if the hugepage allocation should defrag memory aggressively "always", only inside "madvise" regions, or "never". The pmd_trans_splitting/pmd_trans_huge locking is very solid. The put_page (from get_user_page users that can't use mmu notifier like O_DIRECT) that runs against a __split_huge_page_refcount instead was a pain to serialize in a way that would result always in a coherent page count for both tail and head. I think my locking solution with a compound_lock taken only after the page_first is valid and is still a PageHead should be safe but it surely needs review from SMP race point of view. In short there is no current existing way to serialize the O_DIRECT final put_page against split_huge_page_refcount so I had to invent a new one (O_DIRECT loses knowledge on the mapping status by the time gup_fast returns so...). And I didn't want to impact all gup/gup_fast users for now, maybe if we change the gup interface substantially we can avoid this locking, I admit I didn't think too much about it because changing the gup unpinning interface would be invasive. If we ignored O_DIRECT we could stick to the existing compound refcounting code, by simply adding a get_user_pages_fast_flags(foll_flags) where KVM (and any other mmu notifier user) would call it without FOLL_GET (and if FOLL_GET isn't set we'd just BUG_ON if nobody registered itself in the current task mmu notifier list yet). But O_DIRECT is fundamental for decent performance of virtualized I/O on fast storage so we can't avoid it to solve the race of put_page against split_huge_page_refcount to achieve a complete hugepage feature for KVM. Swap and oom works fine (well just like with regular pages ;). MMU notifier is handled transparently too, with the exception of the young bit on the pmd, that didn't have a range check but I think KVM will be fine because the whole point of hugepages is that EPT/NPT will also use a huge pmd when they notice gup returns pages with PageCompound set, so they won't care of a range and there's just the pmd young bit to check in that case. NOTE: in some cases if the L2 cache is small, this may slowdown and waste memory during COWs because 4M of memory are accessed in a single fault instead of 8k (the payoff is that after COW the program can run faster). So we might want to switch the copy_huge_page (and clear_huge_page too) to not temporal stores. I also extensively researched ways to avoid this cache trashing with a full prefault logic that would cow in 8k/16k/32k/64k up to 1M (I can send those patches that fully implemented prefault) but I concluded they're not worth it and they add an huge additional complexity and they remove all tlb benefits until the full hugepage has been faulted in, to save a little bit of memory and some cache during app startup, but they still don't improve substantially the cache-trashing during startup if the prefault happens in >4k chunks. One reason is that those 4k pte entries copied are still mapped on a perfectly cache-colored hugepage, so the trashing is the worst one can generate in those copies (cow of 4k page copies aren't so well colored so they trashes less, but again this results in software running faster after the page fault). Those prefault patches allowed things like a pte where post-cow pages were local 4k regular anon pages and the not-yet-cowed pte entries were pointing in the middle of some hugepage mapped read-only. If it doesn't payoff substantially with todays hardware it will payoff even less in the future with larger l2 caches, and the prefault logic would blot the VM a lot. If one is emebdded transparent_hugepage can be disabled during boot with sysfs or with the boot commandline parameter transparent_hugepage=0 (or transparent_hugepage=2 to restrict hugepages inside madvise regions) that will ensure not a single hugepage is allocated at boot time. It is simple enough to just disable transparent hugepage globally and let transparent hugepages be allocated selectively by applications in the MADV_HUGEPAGE region (both at page fault time, and if enabled with the collapse_huge_page too through the kernel daemon). This patch supports only hugepages mapped in the pmd, archs that have smaller hugepages will not fit in this patch alone. Also some archs like power have certain tlb limits that prevents mixing different page size in the same regions so they will not fit in this framework that requires "graceful fallback" to basic PAGE_SIZE in case of physical memory fragmentation. hugetlbfs remains a perfect fit for those because its software limits happen to match the hardware limits. hugetlbfs also remains a perfect fit for hugepage sizes like 1GByte that cannot be hoped to be found not fragmented after a certain system uptime and that would be very expensive to defragment with relocation, so requiring reservation. hugetlbfs is the "reservation way", the point of transparent hugepages is not to have any reservation at all and maximizing the use of cache and hugepages at all times automatically. Some performance result: vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largep ages3 memset page fault 1566023 memset tlb miss 453854 memset second tlb miss 453321 random access tlb miss 41635 random access second tlb miss 41658 vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largepages3 memset page fault 1566471 memset tlb miss 453375 memset second tlb miss 453320 random access tlb miss 41636 random access second tlb miss 41637 vmx andrea # ./largepages3 memset page fault 1566642 memset tlb miss 453417 memset second tlb miss 453313 random access tlb miss 41630 random access second tlb miss 41647 vmx andrea # ./largepages3 memset page fault 1566872 memset tlb miss 453418 memset second tlb miss 453315 random access tlb miss 41618 random access second tlb miss 41659 vmx andrea # echo 0 > /proc/sys/vm/transparent_hugepage vmx andrea # ./largepages3 memset page fault 2182476 memset tlb miss 460305 memset second tlb miss 460179 random access tlb miss 44483 random access second tlb miss 44186 vmx andrea # ./largepages3 memset page fault 2182791 memset tlb miss 460742 memset second tlb miss 459962 random access tlb miss 43981 random access second tlb miss 43988 ============ #include <stdio.h> #include <stdlib.h> #include <string.h> #include <sys/time.h> #define SIZE (3UL*1024*1024*1024) int main() { char *p = malloc(SIZE), *p2; struct timeval before, after; gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset page fault %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); return 0; } ============ Signed-off-by: Andrea Arcangeli <aarcange@redhat.com> Acked-by: Rik van Riel <riel@redhat.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-13 23:46:52 +00:00
barrier();
if (pmd_trans_huge(orig_pmd)) {
unsigned int dirty = flags & FAULT_FLAG_WRITE;
Automatic NUMA Balancing V11 -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.18 (GNU/Linux) iQIcBAABAgAGBQJQx0kQAAoJEHzG/DNEskfi4fQP/R5PRovayroZALBMLnVJDaLD Ttr9p40VNXbiJ+MfRgatJjSSJZ4Jl+fC3NEqBhcwVZhckZZb9R2s0WtrSQo5+ZbB vdRfiuKoCaKM4cSZ08C12uTvsF6xjhjd27CTUlMkyOcDoKxMEFKelv0hocSxe4Wo xqlv3eF+VsY7kE1BNbgBP06SX4tDpIHRxXfqJPMHaSKQmre+cU0xG2GcEu3QGbHT DEDTI788YSaWLmBfMC+kWoaQl1+bV/FYvavIAS8/o4K9IKvgR42VzrXmaFaqrbgb 72ksa6xfAi57yTmZHqyGmts06qYeBbPpKI+yIhCMInxA9CY3lPbvHppRf0RQOyzj YOi4hovGEMJKE+BCILukhJcZ9jCTtS3zut6v1rdvR88f4y7uhR9RfmRfsxuW7PNj 3Rmh191+n0lVWDmhOs2psXuCLJr3LEiA0dFffN1z8REUTtTAZMsj8Rz+SvBNAZDR hsJhERVeXB6X5uQ5rkLDzbn1Zic60LjVw7LIp6SF2OYf/YKaF8vhyWOA8dyCEu8W CGo7AoG0BO8tIIr8+LvFe8CweypysZImx4AjCfIs4u9pu/v11zmBvO9NO5yfuObF BreEERYgTes/UITxn1qdIW4/q+Nr0iKO3CTqsmu6L1GfCz3/XzPGs3U26fUhllqi Ka0JKgnWvsa6ez6FSzKI =ivQa -----END PGP SIGNATURE----- Merge tag 'balancenuma-v11' of git://git.kernel.org/pub/scm/linux/kernel/git/mel/linux-balancenuma Pull Automatic NUMA Balancing bare-bones from Mel Gorman: "There are three implementations for NUMA balancing, this tree (balancenuma), numacore which has been developed in tip/master and autonuma which is in aa.git. In almost all respects balancenuma is the dumbest of the three because its main impact is on the VM side with no attempt to be smart about scheduling. In the interest of getting the ball rolling, it would be desirable to see this much merged for 3.8 with the view to building scheduler smarts on top and adapting the VM where required for 3.9. The most recent set of comparisons available from different people are mel: https://lkml.org/lkml/2012/12/9/108 mingo: https://lkml.org/lkml/2012/12/7/331 tglx: https://lkml.org/lkml/2012/12/10/437 srikar: https://lkml.org/lkml/2012/12/10/397 The results are a mixed bag. In my own tests, balancenuma does reasonably well. It's dumb as rocks and does not regress against mainline. On the other hand, Ingo's tests shows that balancenuma is incapable of converging for this workloads driven by perf which is bad but is potentially explained by the lack of scheduler smarts. Thomas' results show balancenuma improves on mainline but falls far short of numacore or autonuma. Srikar's results indicate we all suffer on a large machine with imbalanced node sizes. My own testing showed that recent numacore results have improved dramatically, particularly in the last week but not universally. We've butted heads heavily on system CPU usage and high levels of migration even when it shows that overall performance is better. There are also cases where it regresses. Of interest is that for specjbb in some configurations it will regress for lower numbers of warehouses and show gains for higher numbers which is not reported by the tool by default and sometimes missed in treports. Recently I reported for numacore that the JVM was crashing with NullPointerExceptions but currently it's unclear what the source of this problem is. Initially I thought it was in how numacore batch handles PTEs but I'm no longer think this is the case. It's possible numacore is just able to trigger it due to higher rates of migration. These reports were quite late in the cycle so I/we would like to start with this tree as it contains much of the code we can agree on and has not changed significantly over the last 2-3 weeks." * tag 'balancenuma-v11' of git://git.kernel.org/pub/scm/linux/kernel/git/mel/linux-balancenuma: (50 commits) mm/rmap, migration: Make rmap_walk_anon() and try_to_unmap_anon() more scalable mm/rmap: Convert the struct anon_vma::mutex to an rwsem mm: migrate: Account a transhuge page properly when rate limiting mm: numa: Account for failed allocations and isolations as migration failures mm: numa: Add THP migration for the NUMA working set scanning fault case build fix mm: numa: Add THP migration for the NUMA working set scanning fault case. mm: sched: numa: Delay PTE scanning until a task is scheduled on a new node mm: sched: numa: Control enabling and disabling of NUMA balancing if !SCHED_DEBUG mm: sched: numa: Control enabling and disabling of NUMA balancing mm: sched: Adapt the scanning rate if a NUMA hinting fault does not migrate mm: numa: Use a two-stage filter to restrict pages being migrated for unlikely task<->node relationships mm: numa: migrate: Set last_nid on newly allocated page mm: numa: split_huge_page: Transfer last_nid on tail page mm: numa: Introduce last_nid to the page frame sched: numa: Slowly increase the scanning period as NUMA faults are handled mm: numa: Rate limit setting of pte_numa if node is saturated mm: numa: Rate limit the amount of memory that is migrated between nodes mm: numa: Structures for Migrate On Fault per NUMA migration rate limiting mm: numa: Migrate pages handled during a pmd_numa hinting fault mm: numa: Migrate on reference policy ...
2012-12-16 22:33:25 +00:00
if (pmd_numa(orig_pmd))
return do_huge_pmd_numa_page(mm, vma, address,
orig_pmd, pmd);
Automatic NUMA Balancing V11 -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.18 (GNU/Linux) iQIcBAABAgAGBQJQx0kQAAoJEHzG/DNEskfi4fQP/R5PRovayroZALBMLnVJDaLD Ttr9p40VNXbiJ+MfRgatJjSSJZ4Jl+fC3NEqBhcwVZhckZZb9R2s0WtrSQo5+ZbB vdRfiuKoCaKM4cSZ08C12uTvsF6xjhjd27CTUlMkyOcDoKxMEFKelv0hocSxe4Wo xqlv3eF+VsY7kE1BNbgBP06SX4tDpIHRxXfqJPMHaSKQmre+cU0xG2GcEu3QGbHT DEDTI788YSaWLmBfMC+kWoaQl1+bV/FYvavIAS8/o4K9IKvgR42VzrXmaFaqrbgb 72ksa6xfAi57yTmZHqyGmts06qYeBbPpKI+yIhCMInxA9CY3lPbvHppRf0RQOyzj YOi4hovGEMJKE+BCILukhJcZ9jCTtS3zut6v1rdvR88f4y7uhR9RfmRfsxuW7PNj 3Rmh191+n0lVWDmhOs2psXuCLJr3LEiA0dFffN1z8REUTtTAZMsj8Rz+SvBNAZDR hsJhERVeXB6X5uQ5rkLDzbn1Zic60LjVw7LIp6SF2OYf/YKaF8vhyWOA8dyCEu8W CGo7AoG0BO8tIIr8+LvFe8CweypysZImx4AjCfIs4u9pu/v11zmBvO9NO5yfuObF BreEERYgTes/UITxn1qdIW4/q+Nr0iKO3CTqsmu6L1GfCz3/XzPGs3U26fUhllqi Ka0JKgnWvsa6ez6FSzKI =ivQa -----END PGP SIGNATURE----- Merge tag 'balancenuma-v11' of git://git.kernel.org/pub/scm/linux/kernel/git/mel/linux-balancenuma Pull Automatic NUMA Balancing bare-bones from Mel Gorman: "There are three implementations for NUMA balancing, this tree (balancenuma), numacore which has been developed in tip/master and autonuma which is in aa.git. In almost all respects balancenuma is the dumbest of the three because its main impact is on the VM side with no attempt to be smart about scheduling. In the interest of getting the ball rolling, it would be desirable to see this much merged for 3.8 with the view to building scheduler smarts on top and adapting the VM where required for 3.9. The most recent set of comparisons available from different people are mel: https://lkml.org/lkml/2012/12/9/108 mingo: https://lkml.org/lkml/2012/12/7/331 tglx: https://lkml.org/lkml/2012/12/10/437 srikar: https://lkml.org/lkml/2012/12/10/397 The results are a mixed bag. In my own tests, balancenuma does reasonably well. It's dumb as rocks and does not regress against mainline. On the other hand, Ingo's tests shows that balancenuma is incapable of converging for this workloads driven by perf which is bad but is potentially explained by the lack of scheduler smarts. Thomas' results show balancenuma improves on mainline but falls far short of numacore or autonuma. Srikar's results indicate we all suffer on a large machine with imbalanced node sizes. My own testing showed that recent numacore results have improved dramatically, particularly in the last week but not universally. We've butted heads heavily on system CPU usage and high levels of migration even when it shows that overall performance is better. There are also cases where it regresses. Of interest is that for specjbb in some configurations it will regress for lower numbers of warehouses and show gains for higher numbers which is not reported by the tool by default and sometimes missed in treports. Recently I reported for numacore that the JVM was crashing with NullPointerExceptions but currently it's unclear what the source of this problem is. Initially I thought it was in how numacore batch handles PTEs but I'm no longer think this is the case. It's possible numacore is just able to trigger it due to higher rates of migration. These reports were quite late in the cycle so I/we would like to start with this tree as it contains much of the code we can agree on and has not changed significantly over the last 2-3 weeks." * tag 'balancenuma-v11' of git://git.kernel.org/pub/scm/linux/kernel/git/mel/linux-balancenuma: (50 commits) mm/rmap, migration: Make rmap_walk_anon() and try_to_unmap_anon() more scalable mm/rmap: Convert the struct anon_vma::mutex to an rwsem mm: migrate: Account a transhuge page properly when rate limiting mm: numa: Account for failed allocations and isolations as migration failures mm: numa: Add THP migration for the NUMA working set scanning fault case build fix mm: numa: Add THP migration for the NUMA working set scanning fault case. mm: sched: numa: Delay PTE scanning until a task is scheduled on a new node mm: sched: numa: Control enabling and disabling of NUMA balancing if !SCHED_DEBUG mm: sched: numa: Control enabling and disabling of NUMA balancing mm: sched: Adapt the scanning rate if a NUMA hinting fault does not migrate mm: numa: Use a two-stage filter to restrict pages being migrated for unlikely task<->node relationships mm: numa: migrate: Set last_nid on newly allocated page mm: numa: split_huge_page: Transfer last_nid on tail page mm: numa: Introduce last_nid to the page frame sched: numa: Slowly increase the scanning period as NUMA faults are handled mm: numa: Rate limit setting of pte_numa if node is saturated mm: numa: Rate limit the amount of memory that is migrated between nodes mm: numa: Structures for Migrate On Fault per NUMA migration rate limiting mm: numa: Migrate pages handled during a pmd_numa hinting fault mm: numa: Migrate on reference policy ...
2012-12-16 22:33:25 +00:00
if (dirty && !pmd_write(orig_pmd)) {
ret = do_huge_pmd_wp_page(mm, vma, address, pmd,
orig_pmd);
/*
* If COW results in an oom, the huge pmd will
* have been split, so retry the fault on the
* pte for a smaller charge.
*/
if (unlikely(ret & VM_FAULT_OOM))
goto retry;
return ret;
} else {
huge_pmd_set_accessed(mm, vma, address, pmd,
orig_pmd, dirty);
}
thp: transparent hugepage core Lately I've been working to make KVM use hugepages transparently without the usual restrictions of hugetlbfs. Some of the restrictions I'd like to see removed: 1) hugepages have to be swappable or the guest physical memory remains locked in RAM and can't be paged out to swap 2) if a hugepage allocation fails, regular pages should be allocated instead and mixed in the same vma without any failure and without userland noticing 3) if some task quits and more hugepages become available in the buddy, guest physical memory backed by regular pages should be relocated on hugepages automatically in regions under madvise(MADV_HUGEPAGE) (ideally event driven by waking up the kernel deamon if the order=HPAGE_PMD_SHIFT-PAGE_SHIFT list becomes not null) 4) avoidance of reservation and maximization of use of hugepages whenever possible. Reservation (needed to avoid runtime fatal faliures) may be ok for 1 machine with 1 database with 1 database cache with 1 database cache size known at boot time. It's definitely not feasible with a virtualization hypervisor usage like RHEV-H that runs an unknown number of virtual machines with an unknown size of each virtual machine with an unknown amount of pagecache that could be potentially useful in the host for guest not using O_DIRECT (aka cache=off). hugepages in the virtualization hypervisor (and also in the guest!) are much more important than in a regular host not using virtualization, becasue with NPT/EPT they decrease the tlb-miss cacheline accesses from 24 to 19 in case only the hypervisor uses transparent hugepages, and they decrease the tlb-miss cacheline accesses from 19 to 15 in case both the linux hypervisor and the linux guest both uses this patch (though the guest will limit the addition speedup to anonymous regions only for now...). Even more important is that the tlb miss handler is much slower on a NPT/EPT guest than for a regular shadow paging or no-virtualization scenario. So maximizing the amount of virtual memory cached by the TLB pays off significantly more with NPT/EPT than without (even if there would be no significant speedup in the tlb-miss runtime). The first (and more tedious) part of this work requires allowing the VM to handle anonymous hugepages mixed with regular pages transparently on regular anonymous vmas. This is what this patch tries to achieve in the least intrusive possible way. We want hugepages and hugetlb to be used in a way so that all applications can benefit without changes (as usual we leverage the KVM virtualization design: by improving the Linux VM at large, KVM gets the performance boost too). The most important design choice is: always fallback to 4k allocation if the hugepage allocation fails! This is the _very_ opposite of some large pagecache patches that failed with -EIO back then if a 64k (or similar) allocation failed... Second important decision (to reduce the impact of the feature on the existing pagetable handling code) is that at any time we can split an hugepage into 512 regular pages and it has to be done with an operation that can't fail. This way the reliability of the swapping isn't decreased (no need to allocate memory when we are short on memory to swap) and it's trivial to plug a split_huge_page* one-liner where needed without polluting the VM. Over time we can teach mprotect, mremap and friends to handle pmd_trans_huge natively without calling split_huge_page*. The fact it can't fail isn't just for swap: if split_huge_page would return -ENOMEM (instead of the current void) we'd need to rollback the mprotect from the middle of it (ideally including undoing the split_vma) which would be a big change and in the very wrong direction (it'd likely be simpler not to call split_huge_page at all and to teach mprotect and friends to handle hugepages instead of rolling them back from the middle). In short the very value of split_huge_page is that it can't fail. The collapsing and madvise(MADV_HUGEPAGE) part will remain separated and incremental and it'll just be an "harmless" addition later if this initial part is agreed upon. It also should be noted that locking-wise replacing regular pages with hugepages is going to be very easy if compared to what I'm doing below in split_huge_page, as it will only happen when page_count(page) matches page_mapcount(page) if we can take the PG_lock and mmap_sem in write mode. collapse_huge_page will be a "best effort" that (unlike split_huge_page) can fail at the minimal sign of trouble and we can try again later. collapse_huge_page will be similar to how KSM works and the madvise(MADV_HUGEPAGE) will work similar to madvise(MADV_MERGEABLE). The default I like is that transparent hugepages are used at page fault time. This can be changed with /sys/kernel/mm/transparent_hugepage/enabled. The control knob can be set to three values "always", "madvise", "never" which mean respectively that hugepages are always used, or only inside madvise(MADV_HUGEPAGE) regions, or never used. /sys/kernel/mm/transparent_hugepage/defrag instead controls if the hugepage allocation should defrag memory aggressively "always", only inside "madvise" regions, or "never". The pmd_trans_splitting/pmd_trans_huge locking is very solid. The put_page (from get_user_page users that can't use mmu notifier like O_DIRECT) that runs against a __split_huge_page_refcount instead was a pain to serialize in a way that would result always in a coherent page count for both tail and head. I think my locking solution with a compound_lock taken only after the page_first is valid and is still a PageHead should be safe but it surely needs review from SMP race point of view. In short there is no current existing way to serialize the O_DIRECT final put_page against split_huge_page_refcount so I had to invent a new one (O_DIRECT loses knowledge on the mapping status by the time gup_fast returns so...). And I didn't want to impact all gup/gup_fast users for now, maybe if we change the gup interface substantially we can avoid this locking, I admit I didn't think too much about it because changing the gup unpinning interface would be invasive. If we ignored O_DIRECT we could stick to the existing compound refcounting code, by simply adding a get_user_pages_fast_flags(foll_flags) where KVM (and any other mmu notifier user) would call it without FOLL_GET (and if FOLL_GET isn't set we'd just BUG_ON if nobody registered itself in the current task mmu notifier list yet). But O_DIRECT is fundamental for decent performance of virtualized I/O on fast storage so we can't avoid it to solve the race of put_page against split_huge_page_refcount to achieve a complete hugepage feature for KVM. Swap and oom works fine (well just like with regular pages ;). MMU notifier is handled transparently too, with the exception of the young bit on the pmd, that didn't have a range check but I think KVM will be fine because the whole point of hugepages is that EPT/NPT will also use a huge pmd when they notice gup returns pages with PageCompound set, so they won't care of a range and there's just the pmd young bit to check in that case. NOTE: in some cases if the L2 cache is small, this may slowdown and waste memory during COWs because 4M of memory are accessed in a single fault instead of 8k (the payoff is that after COW the program can run faster). So we might want to switch the copy_huge_page (and clear_huge_page too) to not temporal stores. I also extensively researched ways to avoid this cache trashing with a full prefault logic that would cow in 8k/16k/32k/64k up to 1M (I can send those patches that fully implemented prefault) but I concluded they're not worth it and they add an huge additional complexity and they remove all tlb benefits until the full hugepage has been faulted in, to save a little bit of memory and some cache during app startup, but they still don't improve substantially the cache-trashing during startup if the prefault happens in >4k chunks. One reason is that those 4k pte entries copied are still mapped on a perfectly cache-colored hugepage, so the trashing is the worst one can generate in those copies (cow of 4k page copies aren't so well colored so they trashes less, but again this results in software running faster after the page fault). Those prefault patches allowed things like a pte where post-cow pages were local 4k regular anon pages and the not-yet-cowed pte entries were pointing in the middle of some hugepage mapped read-only. If it doesn't payoff substantially with todays hardware it will payoff even less in the future with larger l2 caches, and the prefault logic would blot the VM a lot. If one is emebdded transparent_hugepage can be disabled during boot with sysfs or with the boot commandline parameter transparent_hugepage=0 (or transparent_hugepage=2 to restrict hugepages inside madvise regions) that will ensure not a single hugepage is allocated at boot time. It is simple enough to just disable transparent hugepage globally and let transparent hugepages be allocated selectively by applications in the MADV_HUGEPAGE region (both at page fault time, and if enabled with the collapse_huge_page too through the kernel daemon). This patch supports only hugepages mapped in the pmd, archs that have smaller hugepages will not fit in this patch alone. Also some archs like power have certain tlb limits that prevents mixing different page size in the same regions so they will not fit in this framework that requires "graceful fallback" to basic PAGE_SIZE in case of physical memory fragmentation. hugetlbfs remains a perfect fit for those because its software limits happen to match the hardware limits. hugetlbfs also remains a perfect fit for hugepage sizes like 1GByte that cannot be hoped to be found not fragmented after a certain system uptime and that would be very expensive to defragment with relocation, so requiring reservation. hugetlbfs is the "reservation way", the point of transparent hugepages is not to have any reservation at all and maximizing the use of cache and hugepages at all times automatically. Some performance result: vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largep ages3 memset page fault 1566023 memset tlb miss 453854 memset second tlb miss 453321 random access tlb miss 41635 random access second tlb miss 41658 vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largepages3 memset page fault 1566471 memset tlb miss 453375 memset second tlb miss 453320 random access tlb miss 41636 random access second tlb miss 41637 vmx andrea # ./largepages3 memset page fault 1566642 memset tlb miss 453417 memset second tlb miss 453313 random access tlb miss 41630 random access second tlb miss 41647 vmx andrea # ./largepages3 memset page fault 1566872 memset tlb miss 453418 memset second tlb miss 453315 random access tlb miss 41618 random access second tlb miss 41659 vmx andrea # echo 0 > /proc/sys/vm/transparent_hugepage vmx andrea # ./largepages3 memset page fault 2182476 memset tlb miss 460305 memset second tlb miss 460179 random access tlb miss 44483 random access second tlb miss 44186 vmx andrea # ./largepages3 memset page fault 2182791 memset tlb miss 460742 memset second tlb miss 459962 random access tlb miss 43981 random access second tlb miss 43988 ============ #include <stdio.h> #include <stdlib.h> #include <string.h> #include <sys/time.h> #define SIZE (3UL*1024*1024*1024) int main() { char *p = malloc(SIZE), *p2; struct timeval before, after; gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset page fault %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); return 0; } ============ Signed-off-by: Andrea Arcangeli <aarcange@redhat.com> Acked-by: Rik van Riel <riel@redhat.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-13 23:46:52 +00:00
return 0;
}
}
if (pmd_numa(*pmd))
return do_pmd_numa_page(mm, vma, address, pmd);
thp: transparent hugepage core Lately I've been working to make KVM use hugepages transparently without the usual restrictions of hugetlbfs. Some of the restrictions I'd like to see removed: 1) hugepages have to be swappable or the guest physical memory remains locked in RAM and can't be paged out to swap 2) if a hugepage allocation fails, regular pages should be allocated instead and mixed in the same vma without any failure and without userland noticing 3) if some task quits and more hugepages become available in the buddy, guest physical memory backed by regular pages should be relocated on hugepages automatically in regions under madvise(MADV_HUGEPAGE) (ideally event driven by waking up the kernel deamon if the order=HPAGE_PMD_SHIFT-PAGE_SHIFT list becomes not null) 4) avoidance of reservation and maximization of use of hugepages whenever possible. Reservation (needed to avoid runtime fatal faliures) may be ok for 1 machine with 1 database with 1 database cache with 1 database cache size known at boot time. It's definitely not feasible with a virtualization hypervisor usage like RHEV-H that runs an unknown number of virtual machines with an unknown size of each virtual machine with an unknown amount of pagecache that could be potentially useful in the host for guest not using O_DIRECT (aka cache=off). hugepages in the virtualization hypervisor (and also in the guest!) are much more important than in a regular host not using virtualization, becasue with NPT/EPT they decrease the tlb-miss cacheline accesses from 24 to 19 in case only the hypervisor uses transparent hugepages, and they decrease the tlb-miss cacheline accesses from 19 to 15 in case both the linux hypervisor and the linux guest both uses this patch (though the guest will limit the addition speedup to anonymous regions only for now...). Even more important is that the tlb miss handler is much slower on a NPT/EPT guest than for a regular shadow paging or no-virtualization scenario. So maximizing the amount of virtual memory cached by the TLB pays off significantly more with NPT/EPT than without (even if there would be no significant speedup in the tlb-miss runtime). The first (and more tedious) part of this work requires allowing the VM to handle anonymous hugepages mixed with regular pages transparently on regular anonymous vmas. This is what this patch tries to achieve in the least intrusive possible way. We want hugepages and hugetlb to be used in a way so that all applications can benefit without changes (as usual we leverage the KVM virtualization design: by improving the Linux VM at large, KVM gets the performance boost too). The most important design choice is: always fallback to 4k allocation if the hugepage allocation fails! This is the _very_ opposite of some large pagecache patches that failed with -EIO back then if a 64k (or similar) allocation failed... Second important decision (to reduce the impact of the feature on the existing pagetable handling code) is that at any time we can split an hugepage into 512 regular pages and it has to be done with an operation that can't fail. This way the reliability of the swapping isn't decreased (no need to allocate memory when we are short on memory to swap) and it's trivial to plug a split_huge_page* one-liner where needed without polluting the VM. Over time we can teach mprotect, mremap and friends to handle pmd_trans_huge natively without calling split_huge_page*. The fact it can't fail isn't just for swap: if split_huge_page would return -ENOMEM (instead of the current void) we'd need to rollback the mprotect from the middle of it (ideally including undoing the split_vma) which would be a big change and in the very wrong direction (it'd likely be simpler not to call split_huge_page at all and to teach mprotect and friends to handle hugepages instead of rolling them back from the middle). In short the very value of split_huge_page is that it can't fail. The collapsing and madvise(MADV_HUGEPAGE) part will remain separated and incremental and it'll just be an "harmless" addition later if this initial part is agreed upon. It also should be noted that locking-wise replacing regular pages with hugepages is going to be very easy if compared to what I'm doing below in split_huge_page, as it will only happen when page_count(page) matches page_mapcount(page) if we can take the PG_lock and mmap_sem in write mode. collapse_huge_page will be a "best effort" that (unlike split_huge_page) can fail at the minimal sign of trouble and we can try again later. collapse_huge_page will be similar to how KSM works and the madvise(MADV_HUGEPAGE) will work similar to madvise(MADV_MERGEABLE). The default I like is that transparent hugepages are used at page fault time. This can be changed with /sys/kernel/mm/transparent_hugepage/enabled. The control knob can be set to three values "always", "madvise", "never" which mean respectively that hugepages are always used, or only inside madvise(MADV_HUGEPAGE) regions, or never used. /sys/kernel/mm/transparent_hugepage/defrag instead controls if the hugepage allocation should defrag memory aggressively "always", only inside "madvise" regions, or "never". The pmd_trans_splitting/pmd_trans_huge locking is very solid. The put_page (from get_user_page users that can't use mmu notifier like O_DIRECT) that runs against a __split_huge_page_refcount instead was a pain to serialize in a way that would result always in a coherent page count for both tail and head. I think my locking solution with a compound_lock taken only after the page_first is valid and is still a PageHead should be safe but it surely needs review from SMP race point of view. In short there is no current existing way to serialize the O_DIRECT final put_page against split_huge_page_refcount so I had to invent a new one (O_DIRECT loses knowledge on the mapping status by the time gup_fast returns so...). And I didn't want to impact all gup/gup_fast users for now, maybe if we change the gup interface substantially we can avoid this locking, I admit I didn't think too much about it because changing the gup unpinning interface would be invasive. If we ignored O_DIRECT we could stick to the existing compound refcounting code, by simply adding a get_user_pages_fast_flags(foll_flags) where KVM (and any other mmu notifier user) would call it without FOLL_GET (and if FOLL_GET isn't set we'd just BUG_ON if nobody registered itself in the current task mmu notifier list yet). But O_DIRECT is fundamental for decent performance of virtualized I/O on fast storage so we can't avoid it to solve the race of put_page against split_huge_page_refcount to achieve a complete hugepage feature for KVM. Swap and oom works fine (well just like with regular pages ;). MMU notifier is handled transparently too, with the exception of the young bit on the pmd, that didn't have a range check but I think KVM will be fine because the whole point of hugepages is that EPT/NPT will also use a huge pmd when they notice gup returns pages with PageCompound set, so they won't care of a range and there's just the pmd young bit to check in that case. NOTE: in some cases if the L2 cache is small, this may slowdown and waste memory during COWs because 4M of memory are accessed in a single fault instead of 8k (the payoff is that after COW the program can run faster). So we might want to switch the copy_huge_page (and clear_huge_page too) to not temporal stores. I also extensively researched ways to avoid this cache trashing with a full prefault logic that would cow in 8k/16k/32k/64k up to 1M (I can send those patches that fully implemented prefault) but I concluded they're not worth it and they add an huge additional complexity and they remove all tlb benefits until the full hugepage has been faulted in, to save a little bit of memory and some cache during app startup, but they still don't improve substantially the cache-trashing during startup if the prefault happens in >4k chunks. One reason is that those 4k pte entries copied are still mapped on a perfectly cache-colored hugepage, so the trashing is the worst one can generate in those copies (cow of 4k page copies aren't so well colored so they trashes less, but again this results in software running faster after the page fault). Those prefault patches allowed things like a pte where post-cow pages were local 4k regular anon pages and the not-yet-cowed pte entries were pointing in the middle of some hugepage mapped read-only. If it doesn't payoff substantially with todays hardware it will payoff even less in the future with larger l2 caches, and the prefault logic would blot the VM a lot. If one is emebdded transparent_hugepage can be disabled during boot with sysfs or with the boot commandline parameter transparent_hugepage=0 (or transparent_hugepage=2 to restrict hugepages inside madvise regions) that will ensure not a single hugepage is allocated at boot time. It is simple enough to just disable transparent hugepage globally and let transparent hugepages be allocated selectively by applications in the MADV_HUGEPAGE region (both at page fault time, and if enabled with the collapse_huge_page too through the kernel daemon). This patch supports only hugepages mapped in the pmd, archs that have smaller hugepages will not fit in this patch alone. Also some archs like power have certain tlb limits that prevents mixing different page size in the same regions so they will not fit in this framework that requires "graceful fallback" to basic PAGE_SIZE in case of physical memory fragmentation. hugetlbfs remains a perfect fit for those because its software limits happen to match the hardware limits. hugetlbfs also remains a perfect fit for hugepage sizes like 1GByte that cannot be hoped to be found not fragmented after a certain system uptime and that would be very expensive to defragment with relocation, so requiring reservation. hugetlbfs is the "reservation way", the point of transparent hugepages is not to have any reservation at all and maximizing the use of cache and hugepages at all times automatically. Some performance result: vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largep ages3 memset page fault 1566023 memset tlb miss 453854 memset second tlb miss 453321 random access tlb miss 41635 random access second tlb miss 41658 vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largepages3 memset page fault 1566471 memset tlb miss 453375 memset second tlb miss 453320 random access tlb miss 41636 random access second tlb miss 41637 vmx andrea # ./largepages3 memset page fault 1566642 memset tlb miss 453417 memset second tlb miss 453313 random access tlb miss 41630 random access second tlb miss 41647 vmx andrea # ./largepages3 memset page fault 1566872 memset tlb miss 453418 memset second tlb miss 453315 random access tlb miss 41618 random access second tlb miss 41659 vmx andrea # echo 0 > /proc/sys/vm/transparent_hugepage vmx andrea # ./largepages3 memset page fault 2182476 memset tlb miss 460305 memset second tlb miss 460179 random access tlb miss 44483 random access second tlb miss 44186 vmx andrea # ./largepages3 memset page fault 2182791 memset tlb miss 460742 memset second tlb miss 459962 random access tlb miss 43981 random access second tlb miss 43988 ============ #include <stdio.h> #include <stdlib.h> #include <string.h> #include <sys/time.h> #define SIZE (3UL*1024*1024*1024) int main() { char *p = malloc(SIZE), *p2; struct timeval before, after; gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset page fault %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); return 0; } ============ Signed-off-by: Andrea Arcangeli <aarcange@redhat.com> Acked-by: Rik van Riel <riel@redhat.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-13 23:46:52 +00:00
/*
* Use __pte_alloc instead of pte_alloc_map, because we can't
* run pte_offset_map on the pmd, if an huge pmd could
* materialize from under us from a different thread.
*/
mm: Check if PTE is already allocated during page fault With transparent hugepage support, handle_mm_fault() has to be careful that a normal PMD has been established before handling a PTE fault. To achieve this, it used __pte_alloc() directly instead of pte_alloc_map as pte_alloc_map is unsafe to run against a huge PMD. pte_offset_map() is called once it is known the PMD is safe. pte_alloc_map() is smart enough to check if a PTE is already present before calling __pte_alloc but this check was lost. As a consequence, PTEs may be allocated unnecessarily and the page table lock taken. Thi useless PTE does get cleaned up but it's a performance hit which is visible in page_test from aim9. This patch simply re-adds the check normally done by pte_alloc_map to check if the PTE needs to be allocated before taking the page table lock. The effect is noticable in page_test from aim9. AIM9 2.6.38-vanilla 2.6.38-checkptenone creat-clo 446.10 ( 0.00%) 424.47 (-5.10%) page_test 38.10 ( 0.00%) 42.04 ( 9.37%) brk_test 52.45 ( 0.00%) 51.57 (-1.71%) exec_test 382.00 ( 0.00%) 456.90 (16.39%) fork_test 60.11 ( 0.00%) 67.79 (11.34%) MMTests Statistics: duration Total Elapsed Time (seconds) 611.90 612.22 (While this affects 2.6.38, it is a performance rather than a functional bug and normally outside the rules -stable. While the big performance differences are to a microbench, the difference in fork and exec performance may be significant enough that -stable wants to consider the patch) Reported-by: Raz Ben Yehuda <raziebe@gmail.com> Signed-off-by: Mel Gorman <mgorman@suse.de> Signed-off-by: Andrea Arcangeli <aarcange@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Rik van Riel <riel@redhat.com> [ Picked this up from the AutoNUMA tree to help it upstream and to allow apples-to-apples performance comparisons. ] Signed-off-by: Ingo Molnar <mingo@kernel.org>
2011-10-12 19:06:51 +00:00
if (unlikely(pmd_none(*pmd)) &&
unlikely(__pte_alloc(mm, vma, pmd, address)))
return VM_FAULT_OOM;
thp: transparent hugepage core Lately I've been working to make KVM use hugepages transparently without the usual restrictions of hugetlbfs. Some of the restrictions I'd like to see removed: 1) hugepages have to be swappable or the guest physical memory remains locked in RAM and can't be paged out to swap 2) if a hugepage allocation fails, regular pages should be allocated instead and mixed in the same vma without any failure and without userland noticing 3) if some task quits and more hugepages become available in the buddy, guest physical memory backed by regular pages should be relocated on hugepages automatically in regions under madvise(MADV_HUGEPAGE) (ideally event driven by waking up the kernel deamon if the order=HPAGE_PMD_SHIFT-PAGE_SHIFT list becomes not null) 4) avoidance of reservation and maximization of use of hugepages whenever possible. Reservation (needed to avoid runtime fatal faliures) may be ok for 1 machine with 1 database with 1 database cache with 1 database cache size known at boot time. It's definitely not feasible with a virtualization hypervisor usage like RHEV-H that runs an unknown number of virtual machines with an unknown size of each virtual machine with an unknown amount of pagecache that could be potentially useful in the host for guest not using O_DIRECT (aka cache=off). hugepages in the virtualization hypervisor (and also in the guest!) are much more important than in a regular host not using virtualization, becasue with NPT/EPT they decrease the tlb-miss cacheline accesses from 24 to 19 in case only the hypervisor uses transparent hugepages, and they decrease the tlb-miss cacheline accesses from 19 to 15 in case both the linux hypervisor and the linux guest both uses this patch (though the guest will limit the addition speedup to anonymous regions only for now...). Even more important is that the tlb miss handler is much slower on a NPT/EPT guest than for a regular shadow paging or no-virtualization scenario. So maximizing the amount of virtual memory cached by the TLB pays off significantly more with NPT/EPT than without (even if there would be no significant speedup in the tlb-miss runtime). The first (and more tedious) part of this work requires allowing the VM to handle anonymous hugepages mixed with regular pages transparently on regular anonymous vmas. This is what this patch tries to achieve in the least intrusive possible way. We want hugepages and hugetlb to be used in a way so that all applications can benefit without changes (as usual we leverage the KVM virtualization design: by improving the Linux VM at large, KVM gets the performance boost too). The most important design choice is: always fallback to 4k allocation if the hugepage allocation fails! This is the _very_ opposite of some large pagecache patches that failed with -EIO back then if a 64k (or similar) allocation failed... Second important decision (to reduce the impact of the feature on the existing pagetable handling code) is that at any time we can split an hugepage into 512 regular pages and it has to be done with an operation that can't fail. This way the reliability of the swapping isn't decreased (no need to allocate memory when we are short on memory to swap) and it's trivial to plug a split_huge_page* one-liner where needed without polluting the VM. Over time we can teach mprotect, mremap and friends to handle pmd_trans_huge natively without calling split_huge_page*. The fact it can't fail isn't just for swap: if split_huge_page would return -ENOMEM (instead of the current void) we'd need to rollback the mprotect from the middle of it (ideally including undoing the split_vma) which would be a big change and in the very wrong direction (it'd likely be simpler not to call split_huge_page at all and to teach mprotect and friends to handle hugepages instead of rolling them back from the middle). In short the very value of split_huge_page is that it can't fail. The collapsing and madvise(MADV_HUGEPAGE) part will remain separated and incremental and it'll just be an "harmless" addition later if this initial part is agreed upon. It also should be noted that locking-wise replacing regular pages with hugepages is going to be very easy if compared to what I'm doing below in split_huge_page, as it will only happen when page_count(page) matches page_mapcount(page) if we can take the PG_lock and mmap_sem in write mode. collapse_huge_page will be a "best effort" that (unlike split_huge_page) can fail at the minimal sign of trouble and we can try again later. collapse_huge_page will be similar to how KSM works and the madvise(MADV_HUGEPAGE) will work similar to madvise(MADV_MERGEABLE). The default I like is that transparent hugepages are used at page fault time. This can be changed with /sys/kernel/mm/transparent_hugepage/enabled. The control knob can be set to three values "always", "madvise", "never" which mean respectively that hugepages are always used, or only inside madvise(MADV_HUGEPAGE) regions, or never used. /sys/kernel/mm/transparent_hugepage/defrag instead controls if the hugepage allocation should defrag memory aggressively "always", only inside "madvise" regions, or "never". The pmd_trans_splitting/pmd_trans_huge locking is very solid. The put_page (from get_user_page users that can't use mmu notifier like O_DIRECT) that runs against a __split_huge_page_refcount instead was a pain to serialize in a way that would result always in a coherent page count for both tail and head. I think my locking solution with a compound_lock taken only after the page_first is valid and is still a PageHead should be safe but it surely needs review from SMP race point of view. In short there is no current existing way to serialize the O_DIRECT final put_page against split_huge_page_refcount so I had to invent a new one (O_DIRECT loses knowledge on the mapping status by the time gup_fast returns so...). And I didn't want to impact all gup/gup_fast users for now, maybe if we change the gup interface substantially we can avoid this locking, I admit I didn't think too much about it because changing the gup unpinning interface would be invasive. If we ignored O_DIRECT we could stick to the existing compound refcounting code, by simply adding a get_user_pages_fast_flags(foll_flags) where KVM (and any other mmu notifier user) would call it without FOLL_GET (and if FOLL_GET isn't set we'd just BUG_ON if nobody registered itself in the current task mmu notifier list yet). But O_DIRECT is fundamental for decent performance of virtualized I/O on fast storage so we can't avoid it to solve the race of put_page against split_huge_page_refcount to achieve a complete hugepage feature for KVM. Swap and oom works fine (well just like with regular pages ;). MMU notifier is handled transparently too, with the exception of the young bit on the pmd, that didn't have a range check but I think KVM will be fine because the whole point of hugepages is that EPT/NPT will also use a huge pmd when they notice gup returns pages with PageCompound set, so they won't care of a range and there's just the pmd young bit to check in that case. NOTE: in some cases if the L2 cache is small, this may slowdown and waste memory during COWs because 4M of memory are accessed in a single fault instead of 8k (the payoff is that after COW the program can run faster). So we might want to switch the copy_huge_page (and clear_huge_page too) to not temporal stores. I also extensively researched ways to avoid this cache trashing with a full prefault logic that would cow in 8k/16k/32k/64k up to 1M (I can send those patches that fully implemented prefault) but I concluded they're not worth it and they add an huge additional complexity and they remove all tlb benefits until the full hugepage has been faulted in, to save a little bit of memory and some cache during app startup, but they still don't improve substantially the cache-trashing during startup if the prefault happens in >4k chunks. One reason is that those 4k pte entries copied are still mapped on a perfectly cache-colored hugepage, so the trashing is the worst one can generate in those copies (cow of 4k page copies aren't so well colored so they trashes less, but again this results in software running faster after the page fault). Those prefault patches allowed things like a pte where post-cow pages were local 4k regular anon pages and the not-yet-cowed pte entries were pointing in the middle of some hugepage mapped read-only. If it doesn't payoff substantially with todays hardware it will payoff even less in the future with larger l2 caches, and the prefault logic would blot the VM a lot. If one is emebdded transparent_hugepage can be disabled during boot with sysfs or with the boot commandline parameter transparent_hugepage=0 (or transparent_hugepage=2 to restrict hugepages inside madvise regions) that will ensure not a single hugepage is allocated at boot time. It is simple enough to just disable transparent hugepage globally and let transparent hugepages be allocated selectively by applications in the MADV_HUGEPAGE region (both at page fault time, and if enabled with the collapse_huge_page too through the kernel daemon). This patch supports only hugepages mapped in the pmd, archs that have smaller hugepages will not fit in this patch alone. Also some archs like power have certain tlb limits that prevents mixing different page size in the same regions so they will not fit in this framework that requires "graceful fallback" to basic PAGE_SIZE in case of physical memory fragmentation. hugetlbfs remains a perfect fit for those because its software limits happen to match the hardware limits. hugetlbfs also remains a perfect fit for hugepage sizes like 1GByte that cannot be hoped to be found not fragmented after a certain system uptime and that would be very expensive to defragment with relocation, so requiring reservation. hugetlbfs is the "reservation way", the point of transparent hugepages is not to have any reservation at all and maximizing the use of cache and hugepages at all times automatically. Some performance result: vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largep ages3 memset page fault 1566023 memset tlb miss 453854 memset second tlb miss 453321 random access tlb miss 41635 random access second tlb miss 41658 vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largepages3 memset page fault 1566471 memset tlb miss 453375 memset second tlb miss 453320 random access tlb miss 41636 random access second tlb miss 41637 vmx andrea # ./largepages3 memset page fault 1566642 memset tlb miss 453417 memset second tlb miss 453313 random access tlb miss 41630 random access second tlb miss 41647 vmx andrea # ./largepages3 memset page fault 1566872 memset tlb miss 453418 memset second tlb miss 453315 random access tlb miss 41618 random access second tlb miss 41659 vmx andrea # echo 0 > /proc/sys/vm/transparent_hugepage vmx andrea # ./largepages3 memset page fault 2182476 memset tlb miss 460305 memset second tlb miss 460179 random access tlb miss 44483 random access second tlb miss 44186 vmx andrea # ./largepages3 memset page fault 2182791 memset tlb miss 460742 memset second tlb miss 459962 random access tlb miss 43981 random access second tlb miss 43988 ============ #include <stdio.h> #include <stdlib.h> #include <string.h> #include <sys/time.h> #define SIZE (3UL*1024*1024*1024) int main() { char *p = malloc(SIZE), *p2; struct timeval before, after; gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset page fault %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); memset(p, 0, SIZE); gettimeofday(&after, NULL); printf("memset second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); gettimeofday(&before, NULL); for (p2 = p; p2 < p+SIZE; p2 += 4096) *p2 = 0; gettimeofday(&after, NULL); printf("random access second tlb miss %Lu\n", (after.tv_sec-before.tv_sec)*1000000UL + after.tv_usec-before.tv_usec); return 0; } ============ Signed-off-by: Andrea Arcangeli <aarcange@redhat.com> Acked-by: Rik van Riel <riel@redhat.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-13 23:46:52 +00:00
/* if an huge pmd materialized from under us just retry later */
if (unlikely(pmd_trans_huge(*pmd)))
return 0;
/*
* A regular pmd is established and it can't morph into a huge pmd
* from under us anymore at this point because we hold the mmap_sem
* read mode and khugepaged takes it in write mode. So now it's
* safe to run pte_offset_map().
*/
pte = pte_offset_map(pmd, address);
return handle_pte_fault(mm, vma, address, pte, pmd, flags);
}
#ifndef __PAGETABLE_PUD_FOLDED
/*
* Allocate page upper directory.
[PATCH] mm: init_mm without ptlock First step in pushing down the page_table_lock. init_mm.page_table_lock has been used throughout the architectures (usually for ioremap): not to serialize kernel address space allocation (that's usually vmlist_lock), but because pud_alloc,pmd_alloc,pte_alloc_kernel expect caller holds it. Reverse that: don't lock or unlock init_mm.page_table_lock in any of the architectures; instead rely on pud_alloc,pmd_alloc,pte_alloc_kernel to take and drop it when allocating a new one, to check lest a racing task already did. Similarly no page_table_lock in vmalloc's map_vm_area. Some temporary ugliness in __pud_alloc and __pmd_alloc: since they also handle user mms, which are converted only by a later patch, for now they have to lock differently according to whether or not it's init_mm. If sources get muddled, there's a danger that an arch source taking init_mm.page_table_lock will be mixed with common source also taking it (or neither take it). So break the rules and make another change, which should break the build for such a mismatch: remove the redundant mm arg from pte_alloc_kernel (ppc64 scrapped its distinct ioremap_mm in 2.6.13). Exceptions: arm26 used pte_alloc_kernel on user mm, now pte_alloc_map; ia64 used pte_alloc_map on init_mm, now pte_alloc_kernel; parisc had bad args to pmd_alloc and pte_alloc_kernel in unused USE_HPPA_IOREMAP code; ppc64 map_io_page forgot to unlock on failure; ppc mmu_mapin_ram and ppc64 im_free took page_table_lock for no good reason. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:21 +00:00
* We've already handled the fast-path in-line.
*/
int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
{
pud_t *new = pud_alloc_one(mm, address);
if (!new)
return -ENOMEM;
fix SMP data race in pagetable setup vs walking There is a possible data race in the page table walking code. After the split ptlock patches, it actually seems to have been introduced to the core code, but even before that I think it would have impacted some architectures (powerpc and sparc64, at least, walk the page tables without taking locks eg. see find_linux_pte()). The race is as follows: The pte page is allocated, zeroed, and its struct page gets its spinlock initialized. The mm-wide ptl is then taken, and then the pte page is inserted into the pagetables. At this point, the spinlock is not guaranteed to have ordered the previous stores to initialize the pte page with the subsequent store to put it in the page tables. So another Linux page table walker might be walking down (without any locks, because we have split-leaf-ptls), and find that new pte we've inserted. It might try to take the spinlock before the store from the other CPU initializes it. And subsequently it might read a pte_t out before stores from the other CPU have cleared the memory. There are also similar races in higher levels of the page tables. They obviously don't involve the spinlock, but could see uninitialized memory. Arch code and hardware pagetable walkers that walk the pagetables without locks could see similar uninitialized memory problems, regardless of whether split ptes are enabled or not. I prefer to put the barriers in core code, because that's where the higher level logic happens, but the page table accessors are per-arch, and open-coding them everywhere I don't think is an option. I'll put the read-side barriers in alpha arch code for now (other architectures perform data-dependent loads in order). Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-05-14 04:37:36 +00:00
smp_wmb(); /* See comment in __pte_alloc */
[PATCH] mm: init_mm without ptlock First step in pushing down the page_table_lock. init_mm.page_table_lock has been used throughout the architectures (usually for ioremap): not to serialize kernel address space allocation (that's usually vmlist_lock), but because pud_alloc,pmd_alloc,pte_alloc_kernel expect caller holds it. Reverse that: don't lock or unlock init_mm.page_table_lock in any of the architectures; instead rely on pud_alloc,pmd_alloc,pte_alloc_kernel to take and drop it when allocating a new one, to check lest a racing task already did. Similarly no page_table_lock in vmalloc's map_vm_area. Some temporary ugliness in __pud_alloc and __pmd_alloc: since they also handle user mms, which are converted only by a later patch, for now they have to lock differently according to whether or not it's init_mm. If sources get muddled, there's a danger that an arch source taking init_mm.page_table_lock will be mixed with common source also taking it (or neither take it). So break the rules and make another change, which should break the build for such a mismatch: remove the redundant mm arg from pte_alloc_kernel (ppc64 scrapped its distinct ioremap_mm in 2.6.13). Exceptions: arm26 used pte_alloc_kernel on user mm, now pte_alloc_map; ia64 used pte_alloc_map on init_mm, now pte_alloc_kernel; parisc had bad args to pmd_alloc and pte_alloc_kernel in unused USE_HPPA_IOREMAP code; ppc64 map_io_page forgot to unlock on failure; ppc mmu_mapin_ram and ppc64 im_free took page_table_lock for no good reason. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:21 +00:00
spin_lock(&mm->page_table_lock);
if (pgd_present(*pgd)) /* Another has populated it */
pud_free(mm, new);
else
pgd_populate(mm, pgd, new);
spin_unlock(&mm->page_table_lock);
return 0;
}
#endif /* __PAGETABLE_PUD_FOLDED */
#ifndef __PAGETABLE_PMD_FOLDED
/*
* Allocate page middle directory.
[PATCH] mm: init_mm without ptlock First step in pushing down the page_table_lock. init_mm.page_table_lock has been used throughout the architectures (usually for ioremap): not to serialize kernel address space allocation (that's usually vmlist_lock), but because pud_alloc,pmd_alloc,pte_alloc_kernel expect caller holds it. Reverse that: don't lock or unlock init_mm.page_table_lock in any of the architectures; instead rely on pud_alloc,pmd_alloc,pte_alloc_kernel to take and drop it when allocating a new one, to check lest a racing task already did. Similarly no page_table_lock in vmalloc's map_vm_area. Some temporary ugliness in __pud_alloc and __pmd_alloc: since they also handle user mms, which are converted only by a later patch, for now they have to lock differently according to whether or not it's init_mm. If sources get muddled, there's a danger that an arch source taking init_mm.page_table_lock will be mixed with common source also taking it (or neither take it). So break the rules and make another change, which should break the build for such a mismatch: remove the redundant mm arg from pte_alloc_kernel (ppc64 scrapped its distinct ioremap_mm in 2.6.13). Exceptions: arm26 used pte_alloc_kernel on user mm, now pte_alloc_map; ia64 used pte_alloc_map on init_mm, now pte_alloc_kernel; parisc had bad args to pmd_alloc and pte_alloc_kernel in unused USE_HPPA_IOREMAP code; ppc64 map_io_page forgot to unlock on failure; ppc mmu_mapin_ram and ppc64 im_free took page_table_lock for no good reason. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:21 +00:00
* We've already handled the fast-path in-line.
*/
int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
{
pmd_t *new = pmd_alloc_one(mm, address);
if (!new)
return -ENOMEM;
fix SMP data race in pagetable setup vs walking There is a possible data race in the page table walking code. After the split ptlock patches, it actually seems to have been introduced to the core code, but even before that I think it would have impacted some architectures (powerpc and sparc64, at least, walk the page tables without taking locks eg. see find_linux_pte()). The race is as follows: The pte page is allocated, zeroed, and its struct page gets its spinlock initialized. The mm-wide ptl is then taken, and then the pte page is inserted into the pagetables. At this point, the spinlock is not guaranteed to have ordered the previous stores to initialize the pte page with the subsequent store to put it in the page tables. So another Linux page table walker might be walking down (without any locks, because we have split-leaf-ptls), and find that new pte we've inserted. It might try to take the spinlock before the store from the other CPU initializes it. And subsequently it might read a pte_t out before stores from the other CPU have cleared the memory. There are also similar races in higher levels of the page tables. They obviously don't involve the spinlock, but could see uninitialized memory. Arch code and hardware pagetable walkers that walk the pagetables without locks could see similar uninitialized memory problems, regardless of whether split ptes are enabled or not. I prefer to put the barriers in core code, because that's where the higher level logic happens, but the page table accessors are per-arch, and open-coding them everywhere I don't think is an option. I'll put the read-side barriers in alpha arch code for now (other architectures perform data-dependent loads in order). Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-05-14 04:37:36 +00:00
smp_wmb(); /* See comment in __pte_alloc */
[PATCH] mm: init_mm without ptlock First step in pushing down the page_table_lock. init_mm.page_table_lock has been used throughout the architectures (usually for ioremap): not to serialize kernel address space allocation (that's usually vmlist_lock), but because pud_alloc,pmd_alloc,pte_alloc_kernel expect caller holds it. Reverse that: don't lock or unlock init_mm.page_table_lock in any of the architectures; instead rely on pud_alloc,pmd_alloc,pte_alloc_kernel to take and drop it when allocating a new one, to check lest a racing task already did. Similarly no page_table_lock in vmalloc's map_vm_area. Some temporary ugliness in __pud_alloc and __pmd_alloc: since they also handle user mms, which are converted only by a later patch, for now they have to lock differently according to whether or not it's init_mm. If sources get muddled, there's a danger that an arch source taking init_mm.page_table_lock will be mixed with common source also taking it (or neither take it). So break the rules and make another change, which should break the build for such a mismatch: remove the redundant mm arg from pte_alloc_kernel (ppc64 scrapped its distinct ioremap_mm in 2.6.13). Exceptions: arm26 used pte_alloc_kernel on user mm, now pte_alloc_map; ia64 used pte_alloc_map on init_mm, now pte_alloc_kernel; parisc had bad args to pmd_alloc and pte_alloc_kernel in unused USE_HPPA_IOREMAP code; ppc64 map_io_page forgot to unlock on failure; ppc mmu_mapin_ram and ppc64 im_free took page_table_lock for no good reason. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 01:16:21 +00:00
spin_lock(&mm->page_table_lock);
#ifndef __ARCH_HAS_4LEVEL_HACK
if (pud_present(*pud)) /* Another has populated it */
pmd_free(mm, new);
else
pud_populate(mm, pud, new);
#else
if (pgd_present(*pud)) /* Another has populated it */
pmd_free(mm, new);
else
pgd_populate(mm, pud, new);
#endif /* __ARCH_HAS_4LEVEL_HACK */
spin_unlock(&mm->page_table_lock);
return 0;
[PATCH] Workaround for gcc 2.96 (undefined references) LD .tmp_vmlinux1 mm/built-in.o(.text+0x100d6): In function `copy_page_range': : undefined reference to `__pud_alloc' mm/built-in.o(.text+0x1010b): In function `copy_page_range': : undefined reference to `__pmd_alloc' mm/built-in.o(.text+0x11ef4): In function `__handle_mm_fault': : undefined reference to `__pud_alloc' fs/built-in.o(.text+0xc930): In function `install_arg_page': : undefined reference to `__pud_alloc' make: *** [.tmp_vmlinux1] Error 1 Those missing references in mm/memory.c arise from this code in include/linux/mm.h, combined with the fact that __PGTABLE_PMD_FOLDED and __PGTABLE_PUD_FOLDED are both set and __ARCH_HAS_4LEVEL_HACK is not: /* * The following ifdef needed to get the 4level-fixup.h header to work. * Remove it when 4level-fixup.h has been removed. */ #if defined(CONFIG_MMU) && !defined(__ARCH_HAS_4LEVEL_HACK) static inline pud_t *pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) { return (unlikely(pgd_none(*pgd)) && __pud_alloc(mm, pgd, address))? NULL: pud_offset(pgd, address); } static inline pmd_t *pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) { return (unlikely(pud_none(*pud)) && __pmd_alloc(mm, pud, address))? NULL: pmd_offset(pud, address); } #endif /* CONFIG_MMU && !__ARCH_HAS_4LEVEL_HACK */ With my configuration the pgd_none and pud_none routines are inlines returning a constant 0. Apparently the old compiler avoids generating calls to __pud_alloc and __pmd_alloc but still lists them as undefined references in the module's symbol table. I don't know which change caused this problem. I think it was added somewhere between 2.6.14 and 2.6.15-rc1, because I remember building several 2.6.14-rc kernels without difficulty. However I can't point to an individual culprit. Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-11-28 21:43:44 +00:00
}
#endif /* __PAGETABLE_PMD_FOLDED */
int make_pages_present(unsigned long addr, unsigned long end)
{
int ret, len, write;
struct vm_area_struct * vma;
vma = find_vma(current->mm, addr);
if (!vma)
mlock() fix return values Halesh says: Please find the below testcase provide to test mlock. Test Case : =========================== #include <sys/resource.h> #include <stdio.h> #include <sys/stat.h> #include <sys/types.h> #include <unistd.h> #include <sys/mman.h> #include <fcntl.h> #include <errno.h> #include <stdlib.h> int main(void) { int fd,ret, i = 0; char *addr, *addr1 = NULL; unsigned int page_size; struct rlimit rlim; if (0 != geteuid()) { printf("Execute this pgm as root\n"); exit(1); } /* create a file */ if ((fd = open("mmap_test.c",O_RDWR|O_CREAT,0755)) == -1) { printf("cant create test file\n"); exit(1); } page_size = sysconf(_SC_PAGE_SIZE); /* set the MEMLOCK limit */ rlim.rlim_cur = 2000; rlim.rlim_max = 2000; if ((ret = setrlimit(RLIMIT_MEMLOCK,&rlim)) != 0) { printf("Cant change limit values\n"); exit(1); } addr = 0; while (1) { /* map a page into memory each time*/ if ((addr = (char *) mmap(addr,page_size, PROT_READ | PROT_WRITE,MAP_SHARED,fd,0)) == MAP_FAILED) { printf("cant do mmap on file\n"); exit(1); } if (0 == i) addr1 = addr; i++; errno = 0; /* lock the mapped memory pagewise*/ if ((ret = mlock((char *)addr, 1500)) == -1) { printf("errno value is %d\n", errno); printf("cant lock maped region\n"); exit(1); } addr = addr + page_size; } } ====================================================== This testcase results in an mlock() failure with errno 14 that is EFAULT, but it has nowhere been specified that mlock() will return EFAULT. When I tested the same on older kernels like 2.6.18, I got the correct result i.e errno 12 (ENOMEM). I think in source code mlock(2), setting errno ENOMEM has been missed in do_mlock() , on mlock_fixup() failure. SUSv3 requires the following behavior frmo mlock(2). [ENOMEM] Some or all of the address range specified by the addr and len arguments does not correspond to valid mapped pages in the address space of the process. [EAGAIN] Some or all of the memory identified by the operation could not be locked when the call was made. This rule isn't so nice and slighly strange. but many people think POSIX/SUS compliance is important. Reported-by: Halesh Sadashiv <halesh.sadashiv@ap.sony.com> Tested-by: Halesh Sadashiv <halesh.sadashiv@ap.sony.com> Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: <stable@kernel.org> [2.6.25.x, 2.6.26.x] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-08-04 20:41:14 +00:00
return -ENOMEM;
/*
* We want to touch writable mappings with a write fault in order
* to break COW, except for shared mappings because these don't COW
* and we would not want to dirty them for nothing.
*/
write = (vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE;
BUG_ON(addr >= end);
BUG_ON(end > vma->vm_end);
len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
ret = get_user_pages(current, current->mm, addr,
len, write, 0, NULL, NULL);
if (ret < 0)
return ret;
return ret == len ? 0 : -EFAULT;
}
#if !defined(__HAVE_ARCH_GATE_AREA)
#if defined(AT_SYSINFO_EHDR)
static struct vm_area_struct gate_vma;
static int __init gate_vma_init(void)
{
gate_vma.vm_mm = NULL;
gate_vma.vm_start = FIXADDR_USER_START;
gate_vma.vm_end = FIXADDR_USER_END;
gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
gate_vma.vm_page_prot = __P101;
coredump: remove VM_ALWAYSDUMP flag The motivation for this patchset was that I was looking at a way for a qemu-kvm process, to exclude the guest memory from its core dump, which can be quite large. There are already a number of filter flags in /proc/<pid>/coredump_filter, however, these allow one to specify 'types' of kernel memory, not specific address ranges (which is needed in this case). Since there are no more vma flags available, the first patch eliminates the need for the 'VM_ALWAYSDUMP' flag. The flag is used internally by the kernel to mark vdso and vsyscall pages. However, it is simple enough to check if a vma covers a vdso or vsyscall page without the need for this flag. The second patch then replaces the 'VM_ALWAYSDUMP' flag with a new 'VM_NODUMP' flag, which can be set by userspace using new madvise flags: 'MADV_DONTDUMP', and unset via 'MADV_DODUMP'. The core dump filters continue to work the same as before unless 'MADV_DONTDUMP' is set on the region. The qemu code which implements this features is at: http://people.redhat.com/~jbaron/qemu-dump/qemu-dump.patch In my testing the qemu core dump shrunk from 383MB -> 13MB with this patch. I also believe that the 'MADV_DONTDUMP' flag might be useful for security sensitive apps, which might want to select which areas are dumped. This patch: The VM_ALWAYSDUMP flag is currently used by the coredump code to indicate that a vma is part of a vsyscall or vdso section. However, we can determine if a vma is in one these sections by checking it against the gate_vma and checking for a non-NULL return value from arch_vma_name(). Thus, freeing a valuable vma bit. Signed-off-by: Jason Baron <jbaron@redhat.com> Acked-by: Roland McGrath <roland@hack.frob.com> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Avi Kivity <avi@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-03-23 22:02:51 +00:00
return 0;
}
__initcall(gate_vma_init);
#endif
struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
{
#ifdef AT_SYSINFO_EHDR
return &gate_vma;
#else
return NULL;
#endif
}
int in_gate_area_no_mm(unsigned long addr)
{
#ifdef AT_SYSINFO_EHDR
if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
return 1;
#endif
return 0;
}
#endif /* __HAVE_ARCH_GATE_AREA */
static int __follow_pte(struct mm_struct *mm, unsigned long address,
pte_t **ptepp, spinlock_t **ptlp)
{
pgd_t *pgd;
pud_t *pud;
pmd_t *pmd;
pte_t *ptep;
pgd = pgd_offset(mm, address);
if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
goto out;
pud = pud_offset(pgd, address);
if (pud_none(*pud) || unlikely(pud_bad(*pud)))
goto out;
pmd = pmd_offset(pud, address);
VM_BUG_ON(pmd_trans_huge(*pmd));
if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
goto out;
/* We cannot handle huge page PFN maps. Luckily they don't exist. */
if (pmd_huge(*pmd))
goto out;
ptep = pte_offset_map_lock(mm, pmd, address, ptlp);
if (!ptep)
goto out;
if (!pte_present(*ptep))
goto unlock;
*ptepp = ptep;
return 0;
unlock:
pte_unmap_unlock(ptep, *ptlp);
out:
return -EINVAL;
}
static inline int follow_pte(struct mm_struct *mm, unsigned long address,
pte_t **ptepp, spinlock_t **ptlp)
{
int res;
/* (void) is needed to make gcc happy */
(void) __cond_lock(*ptlp,
!(res = __follow_pte(mm, address, ptepp, ptlp)));
return res;
}
/**
* follow_pfn - look up PFN at a user virtual address
* @vma: memory mapping
* @address: user virtual address
* @pfn: location to store found PFN
*
* Only IO mappings and raw PFN mappings are allowed.
*
* Returns zero and the pfn at @pfn on success, -ve otherwise.
*/
int follow_pfn(struct vm_area_struct *vma, unsigned long address,
unsigned long *pfn)
{
int ret = -EINVAL;
spinlock_t *ptl;
pte_t *ptep;
if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
return ret;
ret = follow_pte(vma->vm_mm, address, &ptep, &ptl);
if (ret)
return ret;
*pfn = pte_pfn(*ptep);
pte_unmap_unlock(ptep, ptl);
return 0;
}
EXPORT_SYMBOL(follow_pfn);
#ifdef CONFIG_HAVE_IOREMAP_PROT
int follow_phys(struct vm_area_struct *vma,
unsigned long address, unsigned int flags,
unsigned long *prot, resource_size_t *phys)
{
int ret = -EINVAL;
pte_t *ptep, pte;
spinlock_t *ptl;
if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
goto out;
if (follow_pte(vma->vm_mm, address, &ptep, &ptl))
goto out;
pte = *ptep;
if ((flags & FOLL_WRITE) && !pte_write(pte))
goto unlock;
*prot = pgprot_val(pte_pgprot(pte));
*phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
ret = 0;
unlock:
pte_unmap_unlock(ptep, ptl);
out:
return ret;
}
int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
void *buf, int len, int write)
{
resource_size_t phys_addr;
unsigned long prot = 0;
void __iomem *maddr;
int offset = addr & (PAGE_SIZE-1);
if (follow_phys(vma, addr, write, &prot, &phys_addr))
return -EINVAL;
maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot);
if (write)
memcpy_toio(maddr + offset, buf, len);
else
memcpy_fromio(buf, maddr + offset, len);
iounmap(maddr);
return len;
}
#endif
/*
* Access another process' address space as given in mm. If non-NULL, use the
* given task for page fault accounting.
*/
static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
unsigned long addr, void *buf, int len, int write)
{
struct vm_area_struct *vma;
void *old_buf = buf;
down_read(&mm->mmap_sem);
/* ignore errors, just check how much was successfully transferred */
while (len) {
int bytes, ret, offset;
void *maddr;
struct page *page = NULL;
ret = get_user_pages(tsk, mm, addr, 1,
write, 1, &page, &vma);
if (ret <= 0) {
/*
* Check if this is a VM_IO | VM_PFNMAP VMA, which
* we can access using slightly different code.
*/
#ifdef CONFIG_HAVE_IOREMAP_PROT
vma = find_vma(mm, addr);
mm: check that we have the right vma in __access_remote_vm() In __access_remote_vm() we need to check that we have found the right vma, not the following vma before we try to access it. Otherwise we might call the vma's access routine with an address which does not fall inside the vma. It was discovered on a current kernel but with an unreleased driver, from memory it was strace leading to a kernel bad access, but it obviously depends on what the access implementation does. Looking at other access implementations I only see: $ git grep -A 5 vm_operations|grep access arch/powerpc/platforms/cell/spufs/file.c- .access = spufs_mem_mmap_access, arch/x86/pci/i386.c- .access = generic_access_phys, drivers/char/mem.c- .access = generic_access_phys fs/sysfs/bin.c- .access = bin_access, The spufs one looks like it might behave badly given the wrong vma, it assumes vma->vm_file->private_data is a spu_context, and looks like it would probably blow up pretty quickly if it wasn't. generic_access_phys() only uses the vma to check vm_flags and get the mm, and then walks page tables using the address. So it should bail on the vm_flags check, or at worst let you access some other VM_IO mapping. And bin_access() just proxies to another access implementation. Signed-off-by: Michael Ellerman <michael@ellerman.id.au> Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-04-14 22:22:10 +00:00
if (!vma || vma->vm_start > addr)
break;
if (vma->vm_ops && vma->vm_ops->access)
ret = vma->vm_ops->access(vma, addr, buf,
len, write);
if (ret <= 0)
#endif
break;
bytes = ret;
} else {
bytes = len;
offset = addr & (PAGE_SIZE-1);
if (bytes > PAGE_SIZE-offset)
bytes = PAGE_SIZE-offset;
maddr = kmap(page);
if (write) {
copy_to_user_page(vma, page, addr,
maddr + offset, buf, bytes);
set_page_dirty_lock(page);
} else {
copy_from_user_page(vma, page, addr,
buf, maddr + offset, bytes);
}
kunmap(page);
page_cache_release(page);
}
len -= bytes;
buf += bytes;
addr += bytes;
}
up_read(&mm->mmap_sem);
return buf - old_buf;
}
/**
* access_remote_vm - access another process' address space
* @mm: the mm_struct of the target address space
* @addr: start address to access
* @buf: source or destination buffer
* @len: number of bytes to transfer
* @write: whether the access is a write
*
* The caller must hold a reference on @mm.
*/
int access_remote_vm(struct mm_struct *mm, unsigned long addr,
void *buf, int len, int write)
{
return __access_remote_vm(NULL, mm, addr, buf, len, write);
}
/*
* Access another process' address space.
* Source/target buffer must be kernel space,
* Do not walk the page table directly, use get_user_pages
*/
int access_process_vm(struct task_struct *tsk, unsigned long addr,
void *buf, int len, int write)
{
struct mm_struct *mm;
int ret;
mm = get_task_mm(tsk);
if (!mm)
return 0;
ret = __access_remote_vm(tsk, mm, addr, buf, len, write);
mmput(mm);
return ret;
}
/*
* Print the name of a VMA.
*/
void print_vma_addr(char *prefix, unsigned long ip)
{
struct mm_struct *mm = current->mm;
struct vm_area_struct *vma;
x86: fix "BUG: sleeping function called from invalid context" in print_vma_addr() Jiri Kosina reported the following deadlock scenario with show_unhandled_signals enabled: [ 68.379022] gnome-settings-[2941] trap int3 ip:3d2c840f34 sp:7fff36f5d100 error:0<3>BUG: sleeping function called from invalid context at kernel/rwsem.c:21 [ 68.379039] in_atomic():1, irqs_disabled():0 [ 68.379044] no locks held by gnome-settings-/2941. [ 68.379050] Pid: 2941, comm: gnome-settings- Not tainted 2.6.25-rc1 #30 [ 68.379054] [ 68.379056] Call Trace: [ 68.379061] <#DB> [<ffffffff81064883>] ? __debug_show_held_locks+0x13/0x30 [ 68.379109] [<ffffffff81036765>] __might_sleep+0xe5/0x110 [ 68.379123] [<ffffffff812f2240>] down_read+0x20/0x70 [ 68.379137] [<ffffffff8109cdca>] print_vma_addr+0x3a/0x110 [ 68.379152] [<ffffffff8100f435>] do_trap+0xf5/0x170 [ 68.379168] [<ffffffff8100f52b>] do_int3+0x7b/0xe0 [ 68.379180] [<ffffffff812f4a6f>] int3+0x9f/0xd0 [ 68.379203] <<EOE>> [ 68.379229] in libglib-2.0.so.0.1505.0[3d2c800000+dc000] and tracked it down to: commit 03252919b79891063cf99145612360efbdf9500b Author: Andi Kleen <ak@suse.de> Date: Wed Jan 30 13:33:18 2008 +0100 x86: print which shared library/executable faulted in segfault etc. messages the problem is that we call down_read() from an atomic context. Solve this by returning from print_vma_addr() if the preempt count is elevated. Update preempt_conditional_sti / preempt_conditional_cli to unconditionally lift the preempt count even on !CONFIG_PREEMPT. Reported-by: Jiri Kosina <jkosina@suse.cz> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-02-13 19:21:06 +00:00
/*
* Do not print if we are in atomic
* contexts (in exception stacks, etc.):
*/
if (preempt_count())
return;
down_read(&mm->mmap_sem);
vma = find_vma(mm, ip);
if (vma && vma->vm_file) {
struct file *f = vma->vm_file;
char *buf = (char *)__get_free_page(GFP_KERNEL);
if (buf) {
char *p;
p = d_path(&f->f_path, buf, PAGE_SIZE);
if (IS_ERR(p))
p = "?";
printk("%s%s[%lx+%lx]", prefix, kbasename(p),
vma->vm_start,
vma->vm_end - vma->vm_start);
free_page((unsigned long)buf);
}
}
up_read(&mm->mmap_sem);
}
#ifdef CONFIG_PROVE_LOCKING
void might_fault(void)
{
/*
* Some code (nfs/sunrpc) uses socket ops on kernel memory while
* holding the mmap_sem, this is safe because kernel memory doesn't
* get paged out, therefore we'll never actually fault, and the
* below annotations will generate false positives.
*/
if (segment_eq(get_fs(), KERNEL_DS))
return;
might_sleep();
/*
* it would be nicer only to annotate paths which are not under
* pagefault_disable, however that requires a larger audit and
* providing helpers like get_user_atomic.
*/
if (!in_atomic() && current->mm)
might_lock_read(&current->mm->mmap_sem);
}
EXPORT_SYMBOL(might_fault);
#endif
#if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS)
static void clear_gigantic_page(struct page *page,
unsigned long addr,
unsigned int pages_per_huge_page)
{
int i;
struct page *p = page;
might_sleep();
for (i = 0; i < pages_per_huge_page;
i++, p = mem_map_next(p, page, i)) {
cond_resched();
clear_user_highpage(p, addr + i * PAGE_SIZE);
}
}
void clear_huge_page(struct page *page,
unsigned long addr, unsigned int pages_per_huge_page)
{
int i;
if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
clear_gigantic_page(page, addr, pages_per_huge_page);
return;
}
might_sleep();
for (i = 0; i < pages_per_huge_page; i++) {
cond_resched();
clear_user_highpage(page + i, addr + i * PAGE_SIZE);
}
}
static void copy_user_gigantic_page(struct page *dst, struct page *src,
unsigned long addr,
struct vm_area_struct *vma,
unsigned int pages_per_huge_page)
{
int i;
struct page *dst_base = dst;
struct page *src_base = src;
for (i = 0; i < pages_per_huge_page; ) {
cond_resched();
copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma);
i++;
dst = mem_map_next(dst, dst_base, i);
src = mem_map_next(src, src_base, i);
}
}
void copy_user_huge_page(struct page *dst, struct page *src,
unsigned long addr, struct vm_area_struct *vma,
unsigned int pages_per_huge_page)
{
int i;
if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
copy_user_gigantic_page(dst, src, addr, vma,
pages_per_huge_page);
return;
}
might_sleep();
for (i = 0; i < pages_per_huge_page; i++) {
cond_resched();
copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma);
}
}
#endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */