linux/mm/swap_state.c

965 lines
26 KiB
C
Raw Normal View History

License cleanup: add SPDX GPL-2.0 license identifier to files with no license Many source files in the tree are missing licensing information, which makes it harder for compliance tools to determine the correct license. By default all files without license information are under the default license of the kernel, which is GPL version 2. Update the files which contain no license information with the 'GPL-2.0' SPDX license identifier. The SPDX identifier is a legally binding shorthand, which can be used instead of the full boiler plate text. This patch is based on work done by Thomas Gleixner and Kate Stewart and Philippe Ombredanne. How this work was done: Patches were generated and checked against linux-4.14-rc6 for a subset of the use cases: - file had no licensing information it it. - file was a */uapi/* one with no licensing information in it, - file was a */uapi/* one with existing licensing information, Further patches will be generated in subsequent months to fix up cases where non-standard license headers were used, and references to license had to be inferred by heuristics based on keywords. The analysis to determine which SPDX License Identifier to be applied to a file was done in a spreadsheet of side by side results from of the output of two independent scanners (ScanCode & Windriver) producing SPDX tag:value files created by Philippe Ombredanne. Philippe prepared the base worksheet, and did an initial spot review of a few 1000 files. The 4.13 kernel was the starting point of the analysis with 60,537 files assessed. Kate Stewart did a file by file comparison of the scanner results in the spreadsheet to determine which SPDX license identifier(s) to be applied to the file. She confirmed any determination that was not immediately clear with lawyers working with the Linux Foundation. Criteria used to select files for SPDX license identifier tagging was: - Files considered eligible had to be source code files. - Make and config files were included as candidates if they contained >5 lines of source - File already had some variant of a license header in it (even if <5 lines). All documentation files were explicitly excluded. The following heuristics were used to determine which SPDX license identifiers to apply. - when both scanners couldn't find any license traces, file was considered to have no license information in it, and the top level COPYING file license applied. For non */uapi/* files that summary was: SPDX license identifier # files ---------------------------------------------------|------- GPL-2.0 11139 and resulted in the first patch in this series. If that file was a */uapi/* path one, it was "GPL-2.0 WITH Linux-syscall-note" otherwise it was "GPL-2.0". Results of that was: SPDX license identifier # files ---------------------------------------------------|------- GPL-2.0 WITH Linux-syscall-note 930 and resulted in the second patch in this series. - if a file had some form of licensing information in it, and was one of the */uapi/* ones, it was denoted with the Linux-syscall-note if any GPL family license was found in the file or had no licensing in it (per prior point). Results summary: SPDX license identifier # files ---------------------------------------------------|------ GPL-2.0 WITH Linux-syscall-note 270 GPL-2.0+ WITH Linux-syscall-note 169 ((GPL-2.0 WITH Linux-syscall-note) OR BSD-2-Clause) 21 ((GPL-2.0 WITH Linux-syscall-note) OR BSD-3-Clause) 17 LGPL-2.1+ WITH Linux-syscall-note 15 GPL-1.0+ WITH Linux-syscall-note 14 ((GPL-2.0+ WITH Linux-syscall-note) OR BSD-3-Clause) 5 LGPL-2.0+ WITH Linux-syscall-note 4 LGPL-2.1 WITH Linux-syscall-note 3 ((GPL-2.0 WITH Linux-syscall-note) OR MIT) 3 ((GPL-2.0 WITH Linux-syscall-note) AND MIT) 1 and that resulted in the third patch in this series. - when the two scanners agreed on the detected license(s), that became the concluded license(s). - when there was disagreement between the two scanners (one detected a license but the other didn't, or they both detected different licenses) a manual inspection of the file occurred. - In most cases a manual inspection of the information in the file resulted in a clear resolution of the license that should apply (and which scanner probably needed to revisit its heuristics). - When it was not immediately clear, the license identifier was confirmed with lawyers working with the Linux Foundation. - If there was any question as to the appropriate license identifier, the file was flagged for further research and to be revisited later in time. In total, over 70 hours of logged manual review was done on the spreadsheet to determine the SPDX license identifiers to apply to the source files by Kate, Philippe, Thomas and, in some cases, confirmation by lawyers working with the Linux Foundation. Kate also obtained a third independent scan of the 4.13 code base from FOSSology, and compared selected files where the other two scanners disagreed against that SPDX file, to see if there was new insights. The Windriver scanner is based on an older version of FOSSology in part, so they are related. Thomas did random spot checks in about 500 files from the spreadsheets for the uapi headers and agreed with SPDX license identifier in the files he inspected. For the non-uapi files Thomas did random spot checks in about 15000 files. In initial set of patches against 4.14-rc6, 3 files were found to have copy/paste license identifier errors, and have been fixed to reflect the correct identifier. Additionally Philippe spent 10 hours this week doing a detailed manual inspection and review of the 12,461 patched files from the initial patch version early this week with: - a full scancode scan run, collecting the matched texts, detected license ids and scores - reviewing anything where there was a license detected (about 500+ files) to ensure that the applied SPDX license was correct - reviewing anything where there was no detection but the patch license was not GPL-2.0 WITH Linux-syscall-note to ensure that the applied SPDX license was correct This produced a worksheet with 20 files needing minor correction. This worksheet was then exported into 3 different .csv files for the different types of files to be modified. These .csv files were then reviewed by Greg. Thomas wrote a script to parse the csv files and add the proper SPDX tag to the file, in the format that the file expected. This script was further refined by Greg based on the output to detect more types of files automatically and to distinguish between header and source .c files (which need different comment types.) Finally Greg ran the script using the .csv files to generate the patches. Reviewed-by: Kate Stewart <kstewart@linuxfoundation.org> Reviewed-by: Philippe Ombredanne <pombredanne@nexb.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-11-01 14:07:57 +00:00
// SPDX-License-Identifier: GPL-2.0
/*
* linux/mm/swap_state.c
*
* Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
* Swap reorganised 29.12.95, Stephen Tweedie
*
* Rewritten to use page cache, (C) 1998 Stephen Tweedie
*/
#include <linux/mm.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/kernel_stat.h>
mempolicy: alloc_pages_mpol() for NUMA policy without vma Shrink shmem's stack usage by eliminating the pseudo-vma from its folio allocation. alloc_pages_mpol(gfp, order, pol, ilx, nid) becomes the principal actor for passing mempolicy choice down to __alloc_pages(), rather than vma_alloc_folio(gfp, order, vma, addr, hugepage). vma_alloc_folio() and alloc_pages() remain, but as wrappers around alloc_pages_mpol(). alloc_pages_bulk_*() untouched, except to provide the additional args to policy_nodemask(), which subsumes policy_node(). Cleanup throughout, cutting out some unhelpful "helpers". It would all be much simpler without MPOL_INTERLEAVE, but that adds a dynamic to the constant mpol: complicated by v3.6 commit 09c231cb8bfd ("tmpfs: distribute interleave better across nodes"), which added ino bias to the interleave, hidden from mm/mempolicy.c until this commit. Hence "ilx" throughout, the "interleave index". Originally I thought it could be done just with nid, but that's wrong: the nodemask may come from the shared policy layer below a shmem vma, or it may come from the task layer above a shmem vma; and without the final nodemask then nodeid cannot be decided. And how ilx is applied depends also on page order. The interleave index is almost always irrelevant unless MPOL_INTERLEAVE: with one exception in alloc_pages_mpol(), where the NO_INTERLEAVE_INDEX passed down from vma-less alloc_pages() is also used as hint not to use THP-style hugepage allocation - to avoid the overhead of a hugepage arg (though I don't understand why we never just added a GFP bit for THP - if it actually needs a different allocation strategy from other pages of the same order). vma_alloc_folio() still carries its hugepage arg here, but it is not used, and should be removed when agreed. get_vma_policy() no longer allows a NULL vma: over time I believe we've eradicated all the places which used to need it e.g. swapoff and madvise used to pass NULL vma to read_swap_cache_async(), but now know the vma. [hughd@google.com: handle NULL mpol being passed to __read_swap_cache_async()] Link: https://lkml.kernel.org/r/ea419956-4751-0102-21f7-9c93cb957892@google.com Link: https://lkml.kernel.org/r/74e34633-6060-f5e3-aee-7040d43f2e93@google.com Link: https://lkml.kernel.org/r/1738368e-bac0-fd11-ed7f-b87142a939fe@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Hildenbrand <david@redhat.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Huang Ying <ying.huang@intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Nhat Pham <nphamcs@gmail.com> Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Tejun heo <tj@kernel.org> Cc: Vishal Moola (Oracle) <vishal.moola@gmail.com> Cc: Yang Shi <shy828301@gmail.com> Cc: Yosry Ahmed <yosryahmed@google.com> Cc: Domenico Cerasuolo <mimmocerasuolo@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-10-19 20:39:08 +00:00
#include <linux/mempolicy.h>
#include <linux/swap.h>
#include <linux/swapops.h>
#include <linux/init.h>
#include <linux/pagemap.h>
#include <linux/pagevec.h>
#include <linux/backing-dev.h>
swap: allow swap readahead to be merged Swap readahead works fine, but the I/O to disk is almost always done in page size requests, despite the fact that readahead submits 1<<page-cluster pages at a time. On older kernels the old per device plugging behavior might have captured this and merged the requests, but currently all comes down to much more I/Os than required. On a single device this might not be an issue, but as soon as a server runs on shared san resources savin I/Os not only improves swapin throughput but also provides a lower resource utilization. With a load running KVM in a lot of memory overcommitment (the hot memory is 1.5 times the host memory) swapping throughput improves significantly and the lead feels more responsive as well as achieves more throughput. In a test setup with 16 swap disks running blocktrace on one of those disks shows the improved merging: Prior: Reads Queued: 560,888, 2,243MiB Writes Queued: 226,242, 904,968KiB Read Dispatches: 544,701, 2,243MiB Write Dispatches: 159,318, 904,968KiB Reads Requeued: 0 Writes Requeued: 0 Reads Completed: 544,716, 2,243MiB Writes Completed: 159,321, 904,980KiB Read Merges: 16,187, 64,748KiB Write Merges: 61,744, 246,976KiB IO unplugs: 149,614 Timer unplugs: 2,940 With the patch: Reads Queued: 734,315, 2,937MiB Writes Queued: 300,188, 1,200MiB Read Dispatches: 214,972, 2,937MiB Write Dispatches: 215,176, 1,200MiB Reads Requeued: 0 Writes Requeued: 0 Reads Completed: 214,971, 2,937MiB Writes Completed: 215,177, 1,200MiB Read Merges: 519,343, 2,077MiB Write Merges: 73,325, 293,300KiB IO unplugs: 337,130 Timer unplugs: 11,184 I got ~10% to ~40% more throughput in my cases and at the same time much lower cpu consumption when broken down per transferred kilobyte (the majority of that due to saved interrupts and better cache handling). In a shared SAN others might get an additional benefit as well, because this now causes less protocol overhead. Signed-off-by: Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com> Acked-by: Rik van Riel <riel@redhat.com> Acked-by: Jens Axboe <axboe@kernel.dk> Reviewed-by: Minchan Kim <minchan@kernel.org> Cc: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-07-31 23:41:44 +00:00
#include <linux/blkdev.h>
#include <linux/migrate.h>
mm/swap: split swap cache into 64MB trunks The patch is to improve the scalability of the swap out/in via using fine grained locks for the swap cache. In current kernel, one address space will be used for each swap device. And in the common configuration, the number of the swap device is very small (one is typical). This causes the heavy lock contention on the radix tree of the address space if multiple tasks swap out/in concurrently. But in fact, there is no dependency between pages in the swap cache. So that, we can split the one shared address space for each swap device into several address spaces to reduce the lock contention. In the patch, the shared address space is split into 64MB trunks. 64MB is chosen to balance the memory space usage and effect of lock contention reduction. The size of struct address_space on x86_64 architecture is 408B, so with the patch, 6528B more memory will be used for every 1GB swap space on x86_64 architecture. One address space is still shared for the swap entries in the same 64M trunks. To avoid lock contention for the first round of swap space allocation, the order of the swap clusters in the initial free clusters list is changed. The swap space distance between the consecutive swap clusters in the free cluster list is at least 64M. After the first round of allocation, the swap clusters are expected to be freed randomly, so the lock contention should be reduced effectively. Link: http://lkml.kernel.org/r/735bab895e64c930581ffb0a05b661e01da82bc5.1484082593.git.tim.c.chen@linux.intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com> Cc: Aaron Lu <aaron.lu@intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Christian Borntraeger <borntraeger@de.ibm.com> Cc: Dave Hansen <dave.hansen@intel.com> Cc: Hillf Danton <hillf.zj@alibaba-inc.com> Cc: Huang Ying <ying.huang@intel.com> Cc: Hugh Dickins <hughd@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Jonathan Corbet <corbet@lwn.net> escreveu: Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Vladimir Davydov <vdavydov.dev@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-02-22 23:45:26 +00:00
#include <linux/vmalloc.h>
mm/swap: add cache for swap slots allocation We add per cpu caches for swap slots that can be allocated and freed quickly without the need to touch the swap info lock. Two separate caches are maintained for swap slots allocated and swap slots returned. This is to allow the swap slots to be returned to the global pool in a batch so they will have a chance to be coaelesced with other slots in a cluster. We do not reuse the slots that are returned right away, as it may increase fragmentation of the slots. The swap allocation cache is protected by a mutex as we may sleep when searching for empty slots in cache. The swap free cache is protected by a spin lock as we cannot sleep in the free path. We refill the swap slots cache when we run out of slots, and we disable the swap slots cache and drain the slots if the global number of slots fall below a low watermark threshold. We re-enable the cache agian when the slots available are above a high watermark. [ying.huang@intel.com: use raw_cpu_ptr over this_cpu_ptr for swap slots access] [tim.c.chen@linux.intel.com: add comments on locks in swap_slots.h] Link: http://lkml.kernel.org/r/20170118180327.GA24225@linux.intel.com Link: http://lkml.kernel.org/r/35de301a4eaa8daa2977de6e987f2c154385eb66.1484082593.git.tim.c.chen@linux.intel.com Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com> Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Reviewed-by: Michal Hocko <mhocko@suse.com> Cc: Aaron Lu <aaron.lu@intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Christian Borntraeger <borntraeger@de.ibm.com> Cc: Dave Hansen <dave.hansen@intel.com> Cc: Hillf Danton <hillf.zj@alibaba-inc.com> Cc: Huang Ying <ying.huang@intel.com> Cc: Hugh Dickins <hughd@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Jonathan Corbet <corbet@lwn.net> escreveu: Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Vladimir Davydov <vdavydov.dev@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-02-22 23:45:39 +00:00
#include <linux/swap_slots.h>
mm, THP, swap: delay splitting THP during swap out Patch series "THP swap: Delay splitting THP during swapping out", v11. This patchset is to optimize the performance of Transparent Huge Page (THP) swap. Recently, the performance of the storage devices improved so fast that we cannot saturate the disk bandwidth with single logical CPU when do page swap out even on a high-end server machine. Because the performance of the storage device improved faster than that of single logical CPU. And it seems that the trend will not change in the near future. On the other hand, the THP becomes more and more popular because of increased memory size. So it becomes necessary to optimize THP swap performance. The advantages of the THP swap support include: - Batch the swap operations for the THP to reduce lock acquiring/releasing, including allocating/freeing the swap space, adding/deleting to/from the swap cache, and writing/reading the swap space, etc. This will help improve the performance of the THP swap. - The THP swap space read/write will be 2M sequential IO. It is particularly helpful for the swap read, which are usually 4k random IO. This will improve the performance of the THP swap too. - It will help the memory fragmentation, especially when the THP is heavily used by the applications. The 2M continuous pages will be free up after THP swapping out. - It will improve the THP utilization on the system with the swap turned on. Because the speed for khugepaged to collapse the normal pages into the THP is quite slow. After the THP is split during the swapping out, it will take quite long time for the normal pages to collapse back into the THP after being swapped in. The high THP utilization helps the efficiency of the page based memory management too. There are some concerns regarding THP swap in, mainly because possible enlarged read/write IO size (for swap in/out) may put more overhead on the storage device. To deal with that, the THP swap in should be turned on only when necessary. For example, it can be selected via "always/never/madvise" logic, to be turned on globally, turned off globally, or turned on only for VMA with MADV_HUGEPAGE, etc. This patchset is the first step for the THP swap support. The plan is to delay splitting THP step by step, finally avoid splitting THP during the THP swapping out and swap out/in the THP as a whole. As the first step, in this patchset, the splitting huge page is delayed from almost the first step of swapping out to after allocating the swap space for the THP and adding the THP into the swap cache. This will reduce lock acquiring/releasing for the locks used for the swap cache management. With the patchset, the swap out throughput improves 15.5% (from about 3.73GB/s to about 4.31GB/s) in the vm-scalability swap-w-seq test case with 8 processes. The test is done on a Xeon E5 v3 system. The swap device used is a RAM simulated PMEM (persistent memory) device. To test the sequential swapping out, the test case creates 8 processes, which sequentially allocate and write to the anonymous pages until the RAM and part of the swap device is used up. This patch (of 5): In this patch, splitting huge page is delayed from almost the first step of swapping out to after allocating the swap space for the THP (Transparent Huge Page) and adding the THP into the swap cache. This will batch the corresponding operation, thus improve THP swap out throughput. This is the first step for the THP swap optimization. The plan is to delay splitting the THP step by step and avoid splitting the THP finally. In this patch, one swap cluster is used to hold the contents of each THP swapped out. So, the size of the swap cluster is changed to that of the THP (Transparent Huge Page) on x86_64 architecture (512). For other architectures which want such THP swap optimization, ARCH_USES_THP_SWAP_CLUSTER needs to be selected in the Kconfig file for the architecture. In effect, this will enlarge swap cluster size by 2 times on x86_64. Which may make it harder to find a free cluster when the swap space becomes fragmented. So that, this may reduce the continuous swap space allocation and sequential write in theory. The performance test in 0day shows no regressions caused by this. In the future of THP swap optimization, some information of the swapped out THP (such as compound map count) will be recorded in the swap_cluster_info data structure. The mem cgroup swap accounting functions are enhanced to support charge or uncharge a swap cluster backing a THP as a whole. The swap cluster allocate/free functions are added to allocate/free a swap cluster for a THP. A fair simple algorithm is used for swap cluster allocation, that is, only the first swap device in priority list will be tried to allocate the swap cluster. The function will fail if the trying is not successful, and the caller will fallback to allocate a single swap slot instead. This works good enough for normal cases. If the difference of the number of the free swap clusters among multiple swap devices is significant, it is possible that some THPs are split earlier than necessary. For example, this could be caused by big size difference among multiple swap devices. The swap cache functions is enhanced to support add/delete THP to/from the swap cache as a set of (HPAGE_PMD_NR) sub-pages. This may be enhanced in the future with multi-order radix tree. But because we will split the THP soon during swapping out, that optimization doesn't make much sense for this first step. The THP splitting functions are enhanced to support to split THP in swap cache during swapping out. The page lock will be held during allocating the swap cluster, adding the THP into the swap cache and splitting the THP. So in the code path other than swapping out, if the THP need to be split, the PageSwapCache(THP) will be always false. The swap cluster is only available for SSD, so the THP swap optimization in this patchset has no effect for HDD. [ying.huang@intel.com: fix two issues in THP optimize patch] Link: http://lkml.kernel.org/r/87k25ed8zo.fsf@yhuang-dev.intel.com [hannes@cmpxchg.org: extensive cleanups and simplifications, reduce code size] Link: http://lkml.kernel.org/r/20170515112522.32457-2-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Suggested-by: Andrew Morton <akpm@linux-foundation.org> [for config option] Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> [for changes in huge_memory.c and huge_mm.h] Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Ebru Akagunduz <ebru.akagunduz@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Tejun Heo <tj@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Shaohua Li <shli@kernel.org> Cc: Minchan Kim <minchan@kernel.org> 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>
2017-07-06 22:37:18 +00:00
#include <linux/huge_mm.h>
#include <linux/shmem_fs.h>
mm: fix swap cache node allocation mask Chris Murphy reports that a slightly overcommitted load, testing swap and zram along with i915, splats and keeps on splatting, when it had better fail less noisily: gnome-shell: page allocation failure: order:0, mode:0x400d0(__GFP_IO|__GFP_FS|__GFP_COMP|__GFP_RECLAIMABLE), nodemask=(null),cpuset=/,mems_allowed=0 CPU: 2 PID: 1155 Comm: gnome-shell Not tainted 5.7.0-1.fc33.x86_64 #1 Call Trace: dump_stack+0x64/0x88 warn_alloc.cold+0x75/0xd9 __alloc_pages_slowpath.constprop.0+0xcfa/0xd30 __alloc_pages_nodemask+0x2df/0x320 alloc_slab_page+0x195/0x310 allocate_slab+0x3c5/0x440 ___slab_alloc+0x40c/0x5f0 __slab_alloc+0x1c/0x30 kmem_cache_alloc+0x20e/0x220 xas_nomem+0x28/0x70 add_to_swap_cache+0x321/0x400 __read_swap_cache_async+0x105/0x240 swap_cluster_readahead+0x22c/0x2e0 shmem_swapin+0x8e/0xc0 shmem_swapin_page+0x196/0x740 shmem_getpage_gfp+0x3a2/0xa60 shmem_read_mapping_page_gfp+0x32/0x60 shmem_get_pages+0x155/0x5e0 [i915] __i915_gem_object_get_pages+0x68/0xa0 [i915] i915_vma_pin+0x3fe/0x6c0 [i915] eb_add_vma+0x10b/0x2c0 [i915] i915_gem_do_execbuffer+0x704/0x3430 [i915] i915_gem_execbuffer2_ioctl+0x1ea/0x3e0 [i915] drm_ioctl_kernel+0x86/0xd0 [drm] drm_ioctl+0x206/0x390 [drm] ksys_ioctl+0x82/0xc0 __x64_sys_ioctl+0x16/0x20 do_syscall_64+0x5b/0xf0 entry_SYSCALL_64_after_hwframe+0x44/0xa9 Reported on 5.7, but it goes back really to 3.1: when shmem_read_mapping_page_gfp() was implemented for use by i915, and allowed for __GFP_NORETRY and __GFP_NOWARN flags in most places, but missed swapin's "& GFP_KERNEL" mask for page tree node allocation in __read_swap_cache_async() - that was to mask off HIGHUSER_MOVABLE bits from what page cache uses, but GFP_RECLAIM_MASK is now what's needed. Link: https://bugzilla.kernel.org/show_bug.cgi?id=208085 Link: http://lkml.kernel.org/r/alpine.LSU.2.11.2006151330070.11064@eggly.anvils Fixes: 68da9f055755 ("tmpfs: pass gfp to shmem_getpage_gfp") Signed-off-by: Hugh Dickins <hughd@google.com> Reviewed-by: Vlastimil Babka <vbabka@suse.cz> Reviewed-by: Matthew Wilcox (Oracle) <willy@infradead.org> Reported-by: Chris Murphy <lists@colorremedies.com> Analyzed-by: Vlastimil Babka <vbabka@suse.cz> Analyzed-by: Matthew Wilcox <willy@infradead.org> Tested-by: Chris Murphy <lists@colorremedies.com> Cc: <stable@vger.kernel.org> [3.1+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-26 03:29:59 +00:00
#include "internal.h"
mm: create new mm/swap.h header file Patch series "MM changes to improve swap-over-NFS support". Assorted improvements for swap-via-filesystem. This is a resend of these patches, rebased on current HEAD. The only substantial changes is that swap_dirty_folio has replaced swap_set_page_dirty. Currently swap-via-fs (SWP_FS_OPS) doesn't work for any filesystem. It has previously worked for NFS but that broke a few releases back. This series changes to use a new ->swap_rw rather than ->readpage and ->direct_IO. It also makes other improvements. There is a companion series already in linux-next which fixes various issues with NFS. Once both series land, a final patch is needed which changes NFS over to use ->swap_rw. This patch (of 10): Many functions declared in include/linux/swap.h are only used within mm/ Create a new "mm/swap.h" and move some of these declarations there. Remove the redundant 'extern' from the function declarations. [akpm@linux-foundation.org: mm/memory-failure.c needs mm/swap.h] Link: https://lkml.kernel.org/r/164859751830.29473.5309689752169286816.stgit@noble.brown Link: https://lkml.kernel.org/r/164859778120.29473.11725907882296224053.stgit@noble.brown Signed-off-by: NeilBrown <neilb@suse.de> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: David Howells <dhowells@redhat.com> Tested-by: Geert Uytterhoeven <geert+renesas@glider.be> Cc: Trond Myklebust <trond.myklebust@hammerspace.com> Cc: Hugh Dickins <hughd@google.com> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Miaohe Lin <linmiaohe@huawei.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-10 01:20:47 +00:00
#include "swap.h"
/*
* swapper_space is a fiction, retained to simplify the path through
* vmscan's shrink_page_list.
*/
static const struct address_space_operations swap_aops = {
.writepage = swap_writepage,
.dirty_folio = noop_dirty_folio,
#ifdef CONFIG_MIGRATION
.migrate_folio = migrate_folio,
#endif
};
struct address_space *swapper_spaces[MAX_SWAPFILES] __read_mostly;
static unsigned int nr_swapper_spaces[MAX_SWAPFILES] __read_mostly;
static bool enable_vma_readahead __read_mostly = true;
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
#define SWAP_RA_WIN_SHIFT (PAGE_SHIFT / 2)
#define SWAP_RA_HITS_MASK ((1UL << SWAP_RA_WIN_SHIFT) - 1)
#define SWAP_RA_HITS_MAX SWAP_RA_HITS_MASK
#define SWAP_RA_WIN_MASK (~PAGE_MASK & ~SWAP_RA_HITS_MASK)
#define SWAP_RA_HITS(v) ((v) & SWAP_RA_HITS_MASK)
#define SWAP_RA_WIN(v) (((v) & SWAP_RA_WIN_MASK) >> SWAP_RA_WIN_SHIFT)
#define SWAP_RA_ADDR(v) ((v) & PAGE_MASK)
#define SWAP_RA_VAL(addr, win, hits) \
(((addr) & PAGE_MASK) | \
(((win) << SWAP_RA_WIN_SHIFT) & SWAP_RA_WIN_MASK) | \
((hits) & SWAP_RA_HITS_MASK))
/* Initial readahead hits is 4 to start up with a small window */
#define GET_SWAP_RA_VAL(vma) \
(atomic_long_read(&(vma)->swap_readahead_info) ? : 4)
swap: add a simple detector for inappropriate swapin readahead This is a patch to improve swap readahead algorithm. It's from Hugh and I slightly changed it. Hugh's original changelog: swapin readahead does a blind readahead, whether or not the swapin is sequential. This may be ok on harddisk, because large reads have relatively small costs, and if the readahead pages are unneeded they can be reclaimed easily - though, what if their allocation forced reclaim of useful pages? But on SSD devices large reads are more expensive than small ones: if the readahead pages are unneeded, reading them in caused significant overhead. This patch adds very simplistic random read detection. Stealing the PageReadahead technique from Konstantin Khlebnikov's patch, avoiding the vma/anon_vma sophistications of Shaohua Li's patch, swapin_nr_pages() simply looks at readahead's current success rate, and narrows or widens its readahead window accordingly. There is little science to its heuristic: it's about as stupid as can be whilst remaining effective. The table below shows elapsed times (in centiseconds) when running a single repetitive swapping load across a 1000MB mapping in 900MB ram with 1GB swap (the harddisk tests had taken painfully too long when I used mem=500M, but SSD shows similar results for that). Vanilla is the 3.6-rc7 kernel on which I started; Shaohua denotes his Sep 3 patch in mmotm and linux-next; HughOld denotes my Oct 1 patch which Shaohua showed to be defective; HughNew this Nov 14 patch, with page_cluster as usual at default of 3 (8-page reads); HughPC4 this same patch with page_cluster 4 (16-page reads); HughPC0 with page_cluster 0 (1-page reads: no readahead). HDD for swapping to harddisk, SSD for swapping to VertexII SSD. Seq for sequential access to the mapping, cycling five times around; Rand for the same number of random touches. Anon for a MAP_PRIVATE anon mapping; Shmem for a MAP_SHARED anon mapping, equivalent to tmpfs. One weakness of Shaohua's vma/anon_vma approach was that it did not optimize Shmem: seen below. Konstantin's approach was perhaps mistuned, 50% slower on Seq: did not compete and is not shown below. HDD Vanilla Shaohua HughOld HughNew HughPC4 HughPC0 Seq Anon 73921 76210 75611 76904 78191 121542 Seq Shmem 73601 73176 73855 72947 74543 118322 Rand Anon 895392 831243 871569 845197 846496 841680 Rand Shmem 1058375 1053486 827935 764955 764376 756489 SSD Vanilla Shaohua HughOld HughNew HughPC4 HughPC0 Seq Anon 24634 24198 24673 25107 21614 70018 Seq Shmem 24959 24932 25052 25703 22030 69678 Rand Anon 43014 26146 28075 25989 26935 25901 Rand Shmem 45349 45215 28249 24268 24138 24332 These tests are, of course, two extremes of a very simple case: under heavier mixed loads I've not yet observed any consistent improvement or degradation, and wider testing would be welcome. Shaohua Li: Test shows Vanilla is slightly better in sequential workload than Hugh's patch. I observed with Hugh's patch sometimes the readahead size is shrinked too fast (from 8 to 1 immediately) in sequential workload if there is no hit. And in such case, continuing doing readahead is good actually. I don't prepare a sophisticated algorithm for the sequential workload because so far we can't guarantee sequential accessed pages are swap out sequentially. So I slightly change Hugh's heuristic - don't shrink readahead size too fast. Here is my test result (unit second, 3 runs average): Vanilla Hugh New Seq 356 370 360 Random 4525 2447 2444 Attached graph is the swapin/swapout throughput I collected with 'vmstat 2'. The first part is running a random workload (till around 1200 of the x-axis) and the second part is running a sequential workload. swapin and swapout throughput are almost identical in steady state in both workloads. These are expected behavior. while in Vanilla, swapin is much bigger than swapout especially in random workload (because wrong readahead). Original patches by: Shaohua Li and Konstantin Khlebnikov. [fengguang.wu@intel.com: swapin_nr_pages() can be static] Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: Fengguang Wu <fengguang.wu@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Konstantin Khlebnikov <khlebnikov@openvz.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-02-06 20:04:21 +00:00
static atomic_t swapin_readahead_hits = ATOMIC_INIT(4);
void show_swap_cache_info(void)
{
printk("%lu pages in swap cache\n", total_swapcache_pages());
printk("Free swap = %ldkB\n", K(get_nr_swap_pages()));
printk("Total swap = %lukB\n", K(total_swap_pages));
}
void *get_shadow_from_swap_cache(swp_entry_t entry)
{
struct address_space *address_space = swap_address_space(entry);
pgoff_t idx = swp_offset(entry);
void *shadow;
shadow = xa_load(&address_space->i_pages, idx);
if (xa_is_value(shadow))
return shadow;
return NULL;
}
/*
* add_to_swap_cache resembles filemap_add_folio on swapper_space,
* but sets SwapCache flag and private instead of mapping and index.
*/
int add_to_swap_cache(struct folio *folio, swp_entry_t entry,
gfp_t gfp, void **shadowp)
{
struct address_space *address_space = swap_address_space(entry);
mm, THP, swap: delay splitting THP during swap out Patch series "THP swap: Delay splitting THP during swapping out", v11. This patchset is to optimize the performance of Transparent Huge Page (THP) swap. Recently, the performance of the storage devices improved so fast that we cannot saturate the disk bandwidth with single logical CPU when do page swap out even on a high-end server machine. Because the performance of the storage device improved faster than that of single logical CPU. And it seems that the trend will not change in the near future. On the other hand, the THP becomes more and more popular because of increased memory size. So it becomes necessary to optimize THP swap performance. The advantages of the THP swap support include: - Batch the swap operations for the THP to reduce lock acquiring/releasing, including allocating/freeing the swap space, adding/deleting to/from the swap cache, and writing/reading the swap space, etc. This will help improve the performance of the THP swap. - The THP swap space read/write will be 2M sequential IO. It is particularly helpful for the swap read, which are usually 4k random IO. This will improve the performance of the THP swap too. - It will help the memory fragmentation, especially when the THP is heavily used by the applications. The 2M continuous pages will be free up after THP swapping out. - It will improve the THP utilization on the system with the swap turned on. Because the speed for khugepaged to collapse the normal pages into the THP is quite slow. After the THP is split during the swapping out, it will take quite long time for the normal pages to collapse back into the THP after being swapped in. The high THP utilization helps the efficiency of the page based memory management too. There are some concerns regarding THP swap in, mainly because possible enlarged read/write IO size (for swap in/out) may put more overhead on the storage device. To deal with that, the THP swap in should be turned on only when necessary. For example, it can be selected via "always/never/madvise" logic, to be turned on globally, turned off globally, or turned on only for VMA with MADV_HUGEPAGE, etc. This patchset is the first step for the THP swap support. The plan is to delay splitting THP step by step, finally avoid splitting THP during the THP swapping out and swap out/in the THP as a whole. As the first step, in this patchset, the splitting huge page is delayed from almost the first step of swapping out to after allocating the swap space for the THP and adding the THP into the swap cache. This will reduce lock acquiring/releasing for the locks used for the swap cache management. With the patchset, the swap out throughput improves 15.5% (from about 3.73GB/s to about 4.31GB/s) in the vm-scalability swap-w-seq test case with 8 processes. The test is done on a Xeon E5 v3 system. The swap device used is a RAM simulated PMEM (persistent memory) device. To test the sequential swapping out, the test case creates 8 processes, which sequentially allocate and write to the anonymous pages until the RAM and part of the swap device is used up. This patch (of 5): In this patch, splitting huge page is delayed from almost the first step of swapping out to after allocating the swap space for the THP (Transparent Huge Page) and adding the THP into the swap cache. This will batch the corresponding operation, thus improve THP swap out throughput. This is the first step for the THP swap optimization. The plan is to delay splitting the THP step by step and avoid splitting the THP finally. In this patch, one swap cluster is used to hold the contents of each THP swapped out. So, the size of the swap cluster is changed to that of the THP (Transparent Huge Page) on x86_64 architecture (512). For other architectures which want such THP swap optimization, ARCH_USES_THP_SWAP_CLUSTER needs to be selected in the Kconfig file for the architecture. In effect, this will enlarge swap cluster size by 2 times on x86_64. Which may make it harder to find a free cluster when the swap space becomes fragmented. So that, this may reduce the continuous swap space allocation and sequential write in theory. The performance test in 0day shows no regressions caused by this. In the future of THP swap optimization, some information of the swapped out THP (such as compound map count) will be recorded in the swap_cluster_info data structure. The mem cgroup swap accounting functions are enhanced to support charge or uncharge a swap cluster backing a THP as a whole. The swap cluster allocate/free functions are added to allocate/free a swap cluster for a THP. A fair simple algorithm is used for swap cluster allocation, that is, only the first swap device in priority list will be tried to allocate the swap cluster. The function will fail if the trying is not successful, and the caller will fallback to allocate a single swap slot instead. This works good enough for normal cases. If the difference of the number of the free swap clusters among multiple swap devices is significant, it is possible that some THPs are split earlier than necessary. For example, this could be caused by big size difference among multiple swap devices. The swap cache functions is enhanced to support add/delete THP to/from the swap cache as a set of (HPAGE_PMD_NR) sub-pages. This may be enhanced in the future with multi-order radix tree. But because we will split the THP soon during swapping out, that optimization doesn't make much sense for this first step. The THP splitting functions are enhanced to support to split THP in swap cache during swapping out. The page lock will be held during allocating the swap cluster, adding the THP into the swap cache and splitting the THP. So in the code path other than swapping out, if the THP need to be split, the PageSwapCache(THP) will be always false. The swap cluster is only available for SSD, so the THP swap optimization in this patchset has no effect for HDD. [ying.huang@intel.com: fix two issues in THP optimize patch] Link: http://lkml.kernel.org/r/87k25ed8zo.fsf@yhuang-dev.intel.com [hannes@cmpxchg.org: extensive cleanups and simplifications, reduce code size] Link: http://lkml.kernel.org/r/20170515112522.32457-2-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Suggested-by: Andrew Morton <akpm@linux-foundation.org> [for config option] Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> [for changes in huge_memory.c and huge_mm.h] Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Ebru Akagunduz <ebru.akagunduz@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Tejun Heo <tj@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Shaohua Li <shli@kernel.org> Cc: Minchan Kim <minchan@kernel.org> 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>
2017-07-06 22:37:18 +00:00
pgoff_t idx = swp_offset(entry);
XA_STATE_ORDER(xas, &address_space->i_pages, idx, folio_order(folio));
unsigned long i, nr = folio_nr_pages(folio);
void *old;
xas_set_update(&xas, workingset_update_node);
VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
VM_BUG_ON_FOLIO(folio_test_swapcache(folio), folio);
VM_BUG_ON_FOLIO(!folio_test_swapbacked(folio), folio);
folio_ref_add(folio, nr);
folio_set_swapcache(folio);
folio->swap = entry;
do {
xas_lock_irq(&xas);
xas_create_range(&xas);
if (xas_error(&xas))
goto unlock;
for (i = 0; i < nr; i++) {
VM_BUG_ON_FOLIO(xas.xa_index != idx + i, folio);
if (shadowp) {
old = xas_load(&xas);
if (xa_is_value(old))
*shadowp = old;
}
xas_store(&xas, folio);
xas_next(&xas);
}
mm, THP, swap: delay splitting THP during swap out Patch series "THP swap: Delay splitting THP during swapping out", v11. This patchset is to optimize the performance of Transparent Huge Page (THP) swap. Recently, the performance of the storage devices improved so fast that we cannot saturate the disk bandwidth with single logical CPU when do page swap out even on a high-end server machine. Because the performance of the storage device improved faster than that of single logical CPU. And it seems that the trend will not change in the near future. On the other hand, the THP becomes more and more popular because of increased memory size. So it becomes necessary to optimize THP swap performance. The advantages of the THP swap support include: - Batch the swap operations for the THP to reduce lock acquiring/releasing, including allocating/freeing the swap space, adding/deleting to/from the swap cache, and writing/reading the swap space, etc. This will help improve the performance of the THP swap. - The THP swap space read/write will be 2M sequential IO. It is particularly helpful for the swap read, which are usually 4k random IO. This will improve the performance of the THP swap too. - It will help the memory fragmentation, especially when the THP is heavily used by the applications. The 2M continuous pages will be free up after THP swapping out. - It will improve the THP utilization on the system with the swap turned on. Because the speed for khugepaged to collapse the normal pages into the THP is quite slow. After the THP is split during the swapping out, it will take quite long time for the normal pages to collapse back into the THP after being swapped in. The high THP utilization helps the efficiency of the page based memory management too. There are some concerns regarding THP swap in, mainly because possible enlarged read/write IO size (for swap in/out) may put more overhead on the storage device. To deal with that, the THP swap in should be turned on only when necessary. For example, it can be selected via "always/never/madvise" logic, to be turned on globally, turned off globally, or turned on only for VMA with MADV_HUGEPAGE, etc. This patchset is the first step for the THP swap support. The plan is to delay splitting THP step by step, finally avoid splitting THP during the THP swapping out and swap out/in the THP as a whole. As the first step, in this patchset, the splitting huge page is delayed from almost the first step of swapping out to after allocating the swap space for the THP and adding the THP into the swap cache. This will reduce lock acquiring/releasing for the locks used for the swap cache management. With the patchset, the swap out throughput improves 15.5% (from about 3.73GB/s to about 4.31GB/s) in the vm-scalability swap-w-seq test case with 8 processes. The test is done on a Xeon E5 v3 system. The swap device used is a RAM simulated PMEM (persistent memory) device. To test the sequential swapping out, the test case creates 8 processes, which sequentially allocate and write to the anonymous pages until the RAM and part of the swap device is used up. This patch (of 5): In this patch, splitting huge page is delayed from almost the first step of swapping out to after allocating the swap space for the THP (Transparent Huge Page) and adding the THP into the swap cache. This will batch the corresponding operation, thus improve THP swap out throughput. This is the first step for the THP swap optimization. The plan is to delay splitting the THP step by step and avoid splitting the THP finally. In this patch, one swap cluster is used to hold the contents of each THP swapped out. So, the size of the swap cluster is changed to that of the THP (Transparent Huge Page) on x86_64 architecture (512). For other architectures which want such THP swap optimization, ARCH_USES_THP_SWAP_CLUSTER needs to be selected in the Kconfig file for the architecture. In effect, this will enlarge swap cluster size by 2 times on x86_64. Which may make it harder to find a free cluster when the swap space becomes fragmented. So that, this may reduce the continuous swap space allocation and sequential write in theory. The performance test in 0day shows no regressions caused by this. In the future of THP swap optimization, some information of the swapped out THP (such as compound map count) will be recorded in the swap_cluster_info data structure. The mem cgroup swap accounting functions are enhanced to support charge or uncharge a swap cluster backing a THP as a whole. The swap cluster allocate/free functions are added to allocate/free a swap cluster for a THP. A fair simple algorithm is used for swap cluster allocation, that is, only the first swap device in priority list will be tried to allocate the swap cluster. The function will fail if the trying is not successful, and the caller will fallback to allocate a single swap slot instead. This works good enough for normal cases. If the difference of the number of the free swap clusters among multiple swap devices is significant, it is possible that some THPs are split earlier than necessary. For example, this could be caused by big size difference among multiple swap devices. The swap cache functions is enhanced to support add/delete THP to/from the swap cache as a set of (HPAGE_PMD_NR) sub-pages. This may be enhanced in the future with multi-order radix tree. But because we will split the THP soon during swapping out, that optimization doesn't make much sense for this first step. The THP splitting functions are enhanced to support to split THP in swap cache during swapping out. The page lock will be held during allocating the swap cluster, adding the THP into the swap cache and splitting the THP. So in the code path other than swapping out, if the THP need to be split, the PageSwapCache(THP) will be always false. The swap cluster is only available for SSD, so the THP swap optimization in this patchset has no effect for HDD. [ying.huang@intel.com: fix two issues in THP optimize patch] Link: http://lkml.kernel.org/r/87k25ed8zo.fsf@yhuang-dev.intel.com [hannes@cmpxchg.org: extensive cleanups and simplifications, reduce code size] Link: http://lkml.kernel.org/r/20170515112522.32457-2-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Suggested-by: Andrew Morton <akpm@linux-foundation.org> [for config option] Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> [for changes in huge_memory.c and huge_mm.h] Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Ebru Akagunduz <ebru.akagunduz@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Tejun Heo <tj@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Shaohua Li <shli@kernel.org> Cc: Minchan Kim <minchan@kernel.org> 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>
2017-07-06 22:37:18 +00:00
address_space->nrpages += nr;
__node_stat_mod_folio(folio, NR_FILE_PAGES, nr);
__lruvec_stat_mod_folio(folio, NR_SWAPCACHE, nr);
unlock:
xas_unlock_irq(&xas);
} while (xas_nomem(&xas, gfp));
if (!xas_error(&xas))
return 0;
folio_clear_swapcache(folio);
folio_ref_sub(folio, nr);
return xas_error(&xas);
}
/*
* This must be called only on folios that have
* been verified to be in the swap cache.
*/
void __delete_from_swap_cache(struct folio *folio,
swp_entry_t entry, void *shadow)
{
struct address_space *address_space = swap_address_space(entry);
int i;
long nr = folio_nr_pages(folio);
pgoff_t idx = swp_offset(entry);
XA_STATE(xas, &address_space->i_pages, idx);
xas_set_update(&xas, workingset_update_node);
VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
VM_BUG_ON_FOLIO(!folio_test_swapcache(folio), folio);
VM_BUG_ON_FOLIO(folio_test_writeback(folio), folio);
mm, THP, swap: delay splitting THP during swap out Patch series "THP swap: Delay splitting THP during swapping out", v11. This patchset is to optimize the performance of Transparent Huge Page (THP) swap. Recently, the performance of the storage devices improved so fast that we cannot saturate the disk bandwidth with single logical CPU when do page swap out even on a high-end server machine. Because the performance of the storage device improved faster than that of single logical CPU. And it seems that the trend will not change in the near future. On the other hand, the THP becomes more and more popular because of increased memory size. So it becomes necessary to optimize THP swap performance. The advantages of the THP swap support include: - Batch the swap operations for the THP to reduce lock acquiring/releasing, including allocating/freeing the swap space, adding/deleting to/from the swap cache, and writing/reading the swap space, etc. This will help improve the performance of the THP swap. - The THP swap space read/write will be 2M sequential IO. It is particularly helpful for the swap read, which are usually 4k random IO. This will improve the performance of the THP swap too. - It will help the memory fragmentation, especially when the THP is heavily used by the applications. The 2M continuous pages will be free up after THP swapping out. - It will improve the THP utilization on the system with the swap turned on. Because the speed for khugepaged to collapse the normal pages into the THP is quite slow. After the THP is split during the swapping out, it will take quite long time for the normal pages to collapse back into the THP after being swapped in. The high THP utilization helps the efficiency of the page based memory management too. There are some concerns regarding THP swap in, mainly because possible enlarged read/write IO size (for swap in/out) may put more overhead on the storage device. To deal with that, the THP swap in should be turned on only when necessary. For example, it can be selected via "always/never/madvise" logic, to be turned on globally, turned off globally, or turned on only for VMA with MADV_HUGEPAGE, etc. This patchset is the first step for the THP swap support. The plan is to delay splitting THP step by step, finally avoid splitting THP during the THP swapping out and swap out/in the THP as a whole. As the first step, in this patchset, the splitting huge page is delayed from almost the first step of swapping out to after allocating the swap space for the THP and adding the THP into the swap cache. This will reduce lock acquiring/releasing for the locks used for the swap cache management. With the patchset, the swap out throughput improves 15.5% (from about 3.73GB/s to about 4.31GB/s) in the vm-scalability swap-w-seq test case with 8 processes. The test is done on a Xeon E5 v3 system. The swap device used is a RAM simulated PMEM (persistent memory) device. To test the sequential swapping out, the test case creates 8 processes, which sequentially allocate and write to the anonymous pages until the RAM and part of the swap device is used up. This patch (of 5): In this patch, splitting huge page is delayed from almost the first step of swapping out to after allocating the swap space for the THP (Transparent Huge Page) and adding the THP into the swap cache. This will batch the corresponding operation, thus improve THP swap out throughput. This is the first step for the THP swap optimization. The plan is to delay splitting the THP step by step and avoid splitting the THP finally. In this patch, one swap cluster is used to hold the contents of each THP swapped out. So, the size of the swap cluster is changed to that of the THP (Transparent Huge Page) on x86_64 architecture (512). For other architectures which want such THP swap optimization, ARCH_USES_THP_SWAP_CLUSTER needs to be selected in the Kconfig file for the architecture. In effect, this will enlarge swap cluster size by 2 times on x86_64. Which may make it harder to find a free cluster when the swap space becomes fragmented. So that, this may reduce the continuous swap space allocation and sequential write in theory. The performance test in 0day shows no regressions caused by this. In the future of THP swap optimization, some information of the swapped out THP (such as compound map count) will be recorded in the swap_cluster_info data structure. The mem cgroup swap accounting functions are enhanced to support charge or uncharge a swap cluster backing a THP as a whole. The swap cluster allocate/free functions are added to allocate/free a swap cluster for a THP. A fair simple algorithm is used for swap cluster allocation, that is, only the first swap device in priority list will be tried to allocate the swap cluster. The function will fail if the trying is not successful, and the caller will fallback to allocate a single swap slot instead. This works good enough for normal cases. If the difference of the number of the free swap clusters among multiple swap devices is significant, it is possible that some THPs are split earlier than necessary. For example, this could be caused by big size difference among multiple swap devices. The swap cache functions is enhanced to support add/delete THP to/from the swap cache as a set of (HPAGE_PMD_NR) sub-pages. This may be enhanced in the future with multi-order radix tree. But because we will split the THP soon during swapping out, that optimization doesn't make much sense for this first step. The THP splitting functions are enhanced to support to split THP in swap cache during swapping out. The page lock will be held during allocating the swap cluster, adding the THP into the swap cache and splitting the THP. So in the code path other than swapping out, if the THP need to be split, the PageSwapCache(THP) will be always false. The swap cluster is only available for SSD, so the THP swap optimization in this patchset has no effect for HDD. [ying.huang@intel.com: fix two issues in THP optimize patch] Link: http://lkml.kernel.org/r/87k25ed8zo.fsf@yhuang-dev.intel.com [hannes@cmpxchg.org: extensive cleanups and simplifications, reduce code size] Link: http://lkml.kernel.org/r/20170515112522.32457-2-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Suggested-by: Andrew Morton <akpm@linux-foundation.org> [for config option] Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> [for changes in huge_memory.c and huge_mm.h] Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Ebru Akagunduz <ebru.akagunduz@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Tejun Heo <tj@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Shaohua Li <shli@kernel.org> Cc: Minchan Kim <minchan@kernel.org> 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>
2017-07-06 22:37:18 +00:00
for (i = 0; i < nr; i++) {
void *entry = xas_store(&xas, shadow);
VM_BUG_ON_PAGE(entry != folio, entry);
xas_next(&xas);
mm, THP, swap: delay splitting THP during swap out Patch series "THP swap: Delay splitting THP during swapping out", v11. This patchset is to optimize the performance of Transparent Huge Page (THP) swap. Recently, the performance of the storage devices improved so fast that we cannot saturate the disk bandwidth with single logical CPU when do page swap out even on a high-end server machine. Because the performance of the storage device improved faster than that of single logical CPU. And it seems that the trend will not change in the near future. On the other hand, the THP becomes more and more popular because of increased memory size. So it becomes necessary to optimize THP swap performance. The advantages of the THP swap support include: - Batch the swap operations for the THP to reduce lock acquiring/releasing, including allocating/freeing the swap space, adding/deleting to/from the swap cache, and writing/reading the swap space, etc. This will help improve the performance of the THP swap. - The THP swap space read/write will be 2M sequential IO. It is particularly helpful for the swap read, which are usually 4k random IO. This will improve the performance of the THP swap too. - It will help the memory fragmentation, especially when the THP is heavily used by the applications. The 2M continuous pages will be free up after THP swapping out. - It will improve the THP utilization on the system with the swap turned on. Because the speed for khugepaged to collapse the normal pages into the THP is quite slow. After the THP is split during the swapping out, it will take quite long time for the normal pages to collapse back into the THP after being swapped in. The high THP utilization helps the efficiency of the page based memory management too. There are some concerns regarding THP swap in, mainly because possible enlarged read/write IO size (for swap in/out) may put more overhead on the storage device. To deal with that, the THP swap in should be turned on only when necessary. For example, it can be selected via "always/never/madvise" logic, to be turned on globally, turned off globally, or turned on only for VMA with MADV_HUGEPAGE, etc. This patchset is the first step for the THP swap support. The plan is to delay splitting THP step by step, finally avoid splitting THP during the THP swapping out and swap out/in the THP as a whole. As the first step, in this patchset, the splitting huge page is delayed from almost the first step of swapping out to after allocating the swap space for the THP and adding the THP into the swap cache. This will reduce lock acquiring/releasing for the locks used for the swap cache management. With the patchset, the swap out throughput improves 15.5% (from about 3.73GB/s to about 4.31GB/s) in the vm-scalability swap-w-seq test case with 8 processes. The test is done on a Xeon E5 v3 system. The swap device used is a RAM simulated PMEM (persistent memory) device. To test the sequential swapping out, the test case creates 8 processes, which sequentially allocate and write to the anonymous pages until the RAM and part of the swap device is used up. This patch (of 5): In this patch, splitting huge page is delayed from almost the first step of swapping out to after allocating the swap space for the THP (Transparent Huge Page) and adding the THP into the swap cache. This will batch the corresponding operation, thus improve THP swap out throughput. This is the first step for the THP swap optimization. The plan is to delay splitting the THP step by step and avoid splitting the THP finally. In this patch, one swap cluster is used to hold the contents of each THP swapped out. So, the size of the swap cluster is changed to that of the THP (Transparent Huge Page) on x86_64 architecture (512). For other architectures which want such THP swap optimization, ARCH_USES_THP_SWAP_CLUSTER needs to be selected in the Kconfig file for the architecture. In effect, this will enlarge swap cluster size by 2 times on x86_64. Which may make it harder to find a free cluster when the swap space becomes fragmented. So that, this may reduce the continuous swap space allocation and sequential write in theory. The performance test in 0day shows no regressions caused by this. In the future of THP swap optimization, some information of the swapped out THP (such as compound map count) will be recorded in the swap_cluster_info data structure. The mem cgroup swap accounting functions are enhanced to support charge or uncharge a swap cluster backing a THP as a whole. The swap cluster allocate/free functions are added to allocate/free a swap cluster for a THP. A fair simple algorithm is used for swap cluster allocation, that is, only the first swap device in priority list will be tried to allocate the swap cluster. The function will fail if the trying is not successful, and the caller will fallback to allocate a single swap slot instead. This works good enough for normal cases. If the difference of the number of the free swap clusters among multiple swap devices is significant, it is possible that some THPs are split earlier than necessary. For example, this could be caused by big size difference among multiple swap devices. The swap cache functions is enhanced to support add/delete THP to/from the swap cache as a set of (HPAGE_PMD_NR) sub-pages. This may be enhanced in the future with multi-order radix tree. But because we will split the THP soon during swapping out, that optimization doesn't make much sense for this first step. The THP splitting functions are enhanced to support to split THP in swap cache during swapping out. The page lock will be held during allocating the swap cluster, adding the THP into the swap cache and splitting the THP. So in the code path other than swapping out, if the THP need to be split, the PageSwapCache(THP) will be always false. The swap cluster is only available for SSD, so the THP swap optimization in this patchset has no effect for HDD. [ying.huang@intel.com: fix two issues in THP optimize patch] Link: http://lkml.kernel.org/r/87k25ed8zo.fsf@yhuang-dev.intel.com [hannes@cmpxchg.org: extensive cleanups and simplifications, reduce code size] Link: http://lkml.kernel.org/r/20170515112522.32457-2-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Suggested-by: Andrew Morton <akpm@linux-foundation.org> [for config option] Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> [for changes in huge_memory.c and huge_mm.h] Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Ebru Akagunduz <ebru.akagunduz@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Tejun Heo <tj@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Shaohua Li <shli@kernel.org> Cc: Minchan Kim <minchan@kernel.org> 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>
2017-07-06 22:37:18 +00:00
}
folio->swap.val = 0;
folio_clear_swapcache(folio);
mm, THP, swap: delay splitting THP during swap out Patch series "THP swap: Delay splitting THP during swapping out", v11. This patchset is to optimize the performance of Transparent Huge Page (THP) swap. Recently, the performance of the storage devices improved so fast that we cannot saturate the disk bandwidth with single logical CPU when do page swap out even on a high-end server machine. Because the performance of the storage device improved faster than that of single logical CPU. And it seems that the trend will not change in the near future. On the other hand, the THP becomes more and more popular because of increased memory size. So it becomes necessary to optimize THP swap performance. The advantages of the THP swap support include: - Batch the swap operations for the THP to reduce lock acquiring/releasing, including allocating/freeing the swap space, adding/deleting to/from the swap cache, and writing/reading the swap space, etc. This will help improve the performance of the THP swap. - The THP swap space read/write will be 2M sequential IO. It is particularly helpful for the swap read, which are usually 4k random IO. This will improve the performance of the THP swap too. - It will help the memory fragmentation, especially when the THP is heavily used by the applications. The 2M continuous pages will be free up after THP swapping out. - It will improve the THP utilization on the system with the swap turned on. Because the speed for khugepaged to collapse the normal pages into the THP is quite slow. After the THP is split during the swapping out, it will take quite long time for the normal pages to collapse back into the THP after being swapped in. The high THP utilization helps the efficiency of the page based memory management too. There are some concerns regarding THP swap in, mainly because possible enlarged read/write IO size (for swap in/out) may put more overhead on the storage device. To deal with that, the THP swap in should be turned on only when necessary. For example, it can be selected via "always/never/madvise" logic, to be turned on globally, turned off globally, or turned on only for VMA with MADV_HUGEPAGE, etc. This patchset is the first step for the THP swap support. The plan is to delay splitting THP step by step, finally avoid splitting THP during the THP swapping out and swap out/in the THP as a whole. As the first step, in this patchset, the splitting huge page is delayed from almost the first step of swapping out to after allocating the swap space for the THP and adding the THP into the swap cache. This will reduce lock acquiring/releasing for the locks used for the swap cache management. With the patchset, the swap out throughput improves 15.5% (from about 3.73GB/s to about 4.31GB/s) in the vm-scalability swap-w-seq test case with 8 processes. The test is done on a Xeon E5 v3 system. The swap device used is a RAM simulated PMEM (persistent memory) device. To test the sequential swapping out, the test case creates 8 processes, which sequentially allocate and write to the anonymous pages until the RAM and part of the swap device is used up. This patch (of 5): In this patch, splitting huge page is delayed from almost the first step of swapping out to after allocating the swap space for the THP (Transparent Huge Page) and adding the THP into the swap cache. This will batch the corresponding operation, thus improve THP swap out throughput. This is the first step for the THP swap optimization. The plan is to delay splitting the THP step by step and avoid splitting the THP finally. In this patch, one swap cluster is used to hold the contents of each THP swapped out. So, the size of the swap cluster is changed to that of the THP (Transparent Huge Page) on x86_64 architecture (512). For other architectures which want such THP swap optimization, ARCH_USES_THP_SWAP_CLUSTER needs to be selected in the Kconfig file for the architecture. In effect, this will enlarge swap cluster size by 2 times on x86_64. Which may make it harder to find a free cluster when the swap space becomes fragmented. So that, this may reduce the continuous swap space allocation and sequential write in theory. The performance test in 0day shows no regressions caused by this. In the future of THP swap optimization, some information of the swapped out THP (such as compound map count) will be recorded in the swap_cluster_info data structure. The mem cgroup swap accounting functions are enhanced to support charge or uncharge a swap cluster backing a THP as a whole. The swap cluster allocate/free functions are added to allocate/free a swap cluster for a THP. A fair simple algorithm is used for swap cluster allocation, that is, only the first swap device in priority list will be tried to allocate the swap cluster. The function will fail if the trying is not successful, and the caller will fallback to allocate a single swap slot instead. This works good enough for normal cases. If the difference of the number of the free swap clusters among multiple swap devices is significant, it is possible that some THPs are split earlier than necessary. For example, this could be caused by big size difference among multiple swap devices. The swap cache functions is enhanced to support add/delete THP to/from the swap cache as a set of (HPAGE_PMD_NR) sub-pages. This may be enhanced in the future with multi-order radix tree. But because we will split the THP soon during swapping out, that optimization doesn't make much sense for this first step. The THP splitting functions are enhanced to support to split THP in swap cache during swapping out. The page lock will be held during allocating the swap cluster, adding the THP into the swap cache and splitting the THP. So in the code path other than swapping out, if the THP need to be split, the PageSwapCache(THP) will be always false. The swap cluster is only available for SSD, so the THP swap optimization in this patchset has no effect for HDD. [ying.huang@intel.com: fix two issues in THP optimize patch] Link: http://lkml.kernel.org/r/87k25ed8zo.fsf@yhuang-dev.intel.com [hannes@cmpxchg.org: extensive cleanups and simplifications, reduce code size] Link: http://lkml.kernel.org/r/20170515112522.32457-2-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Suggested-by: Andrew Morton <akpm@linux-foundation.org> [for config option] Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> [for changes in huge_memory.c and huge_mm.h] Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Ebru Akagunduz <ebru.akagunduz@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Tejun Heo <tj@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Shaohua Li <shli@kernel.org> Cc: Minchan Kim <minchan@kernel.org> 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>
2017-07-06 22:37:18 +00:00
address_space->nrpages -= nr;
__node_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
__lruvec_stat_mod_folio(folio, NR_SWAPCACHE, -nr);
}
/**
* add_to_swap - allocate swap space for a folio
* @folio: folio we want to move to swap
*
* Allocate swap space for the folio and add the folio to the
* swap cache.
*
* Context: Caller needs to hold the folio lock.
* Return: Whether the folio was added to the swap cache.
*/
bool add_to_swap(struct folio *folio)
{
swp_entry_t entry;
int err;
VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
VM_BUG_ON_FOLIO(!folio_test_uptodate(folio), folio);
entry = folio_alloc_swap(folio);
if (!entry.val)
return false;
/*
* XArray node allocations from PF_MEMALLOC contexts could
* completely exhaust the page allocator. __GFP_NOMEMALLOC
* stops emergency reserves from being allocated.
*
* TODO: this could cause a theoretical memory reclaim
* deadlock in the swap out path.
*/
/*
mm: support madvise(MADV_FREE) Linux doesn't have an ability to free pages lazy while other OS already have been supported that named by madvise(MADV_FREE). The gain is clear that kernel can discard freed pages rather than swapping out or OOM if memory pressure happens. Without memory pressure, freed pages would be reused by userspace without another additional overhead(ex, page fault + allocation + zeroing). Jason Evans said: : Facebook has been using MAP_UNINITIALIZED : (https://lkml.org/lkml/2012/1/18/308) in some of its applications for : several years, but there are operational costs to maintaining this : out-of-tree in our kernel and in jemalloc, and we are anxious to retire it : in favor of MADV_FREE. When we first enabled MAP_UNINITIALIZED it : increased throughput for much of our workload by ~5%, and although the : benefit has decreased using newer hardware and kernels, there is still : enough benefit that we cannot reasonably retire it without a replacement. : : Aside from Facebook operations, there are numerous broadly used : applications that would benefit from MADV_FREE. The ones that immediately : come to mind are redis, varnish, and MariaDB. I don't have much insight : into Android internals and development process, but I would hope to see : MADV_FREE support eventually end up there as well to benefit applications : linked with the integrated jemalloc. : : jemalloc will use MADV_FREE once it becomes available in the Linux kernel. : In fact, jemalloc already uses MADV_FREE or equivalent everywhere it's : available: *BSD, OS X, Windows, and Solaris -- every platform except Linux : (and AIX, but I'm not sure it even compiles on AIX). The lack of : MADV_FREE on Linux forced me down a long series of increasingly : sophisticated heuristics for madvise() volume reduction, and even so this : remains a common performance issue for people using jemalloc on Linux. : Please integrate MADV_FREE; many people will benefit substantially. How it works: When madvise syscall is called, VM clears dirty bit of ptes of the range. If memory pressure happens, VM checks dirty bit of page table and if it found still "clean", it means it's a "lazyfree pages" so VM could discard the page instead of swapping out. Once there was store operation for the page before VM peek a page to reclaim, dirty bit is set so VM can swap out the page instead of discarding. One thing we should notice is that basically, MADV_FREE relies on dirty bit in page table entry to decide whether VM allows to discard the page or not. IOW, if page table entry includes marked dirty bit, VM shouldn't discard the page. However, as a example, if swap-in by read fault happens, page table entry doesn't have dirty bit so MADV_FREE could discard the page wrongly. For avoiding the problem, MADV_FREE did more checks with PageDirty and PageSwapCache. It worked out because swapped-in page lives on swap cache and since it is evicted from the swap cache, the page has PG_dirty flag. So both page flags check effectively prevent wrong discarding by MADV_FREE. However, a problem in above logic is that swapped-in page has PG_dirty still after they are removed from swap cache so VM cannot consider the page as freeable any more even if madvise_free is called in future. Look at below example for detail. ptr = malloc(); memset(ptr); .. .. .. heavy memory pressure so all of pages are swapped out .. .. var = *ptr; -> a page swapped-in and could be removed from swapcache. Then, page table doesn't mark dirty bit and page descriptor includes PG_dirty .. .. madvise_free(ptr); -> It doesn't clear PG_dirty of the page. .. .. .. .. heavy memory pressure again. .. In this time, VM cannot discard the page because the page .. has *PG_dirty* To solve the problem, this patch clears PG_dirty if only the page is owned exclusively by current process when madvise is called because PG_dirty represents ptes's dirtiness in several processes so we could clear it only if we own it exclusively. Firstly, heavy users would be general allocators(ex, jemalloc, tcmalloc and hope glibc supports it) and jemalloc/tcmalloc already have supported the feature for other OS(ex, FreeBSD) barrios@blaptop:~/benchmark/ebizzy$ lscpu Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Byte Order: Little Endian CPU(s): 12 On-line CPU(s) list: 0-11 Thread(s) per core: 1 Core(s) per socket: 1 Socket(s): 12 NUMA node(s): 1 Vendor ID: GenuineIntel CPU family: 6 Model: 2 Stepping: 3 CPU MHz: 3200.185 BogoMIPS: 6400.53 Virtualization: VT-x Hypervisor vendor: KVM Virtualization type: full L1d cache: 32K L1i cache: 32K L2 cache: 4096K NUMA node0 CPU(s): 0-11 ebizzy benchmark(./ebizzy -S 10 -n 512) Higher avg is better. vanilla-jemalloc MADV_free-jemalloc 1 thread records: 10 records: 10 avg: 2961.90 avg: 12069.70 std: 71.96(2.43%) std: 186.68(1.55%) max: 3070.00 max: 12385.00 min: 2796.00 min: 11746.00 2 thread records: 10 records: 10 avg: 5020.00 avg: 17827.00 std: 264.87(5.28%) std: 358.52(2.01%) max: 5244.00 max: 18760.00 min: 4251.00 min: 17382.00 4 thread records: 10 records: 10 avg: 8988.80 avg: 27930.80 std: 1175.33(13.08%) std: 3317.33(11.88%) max: 9508.00 max: 30879.00 min: 5477.00 min: 21024.00 8 thread records: 10 records: 10 avg: 13036.50 avg: 33739.40 std: 170.67(1.31%) std: 5146.22(15.25%) max: 13371.00 max: 40572.00 min: 12785.00 min: 24088.00 16 thread records: 10 records: 10 avg: 11092.40 avg: 31424.20 std: 710.60(6.41%) std: 3763.89(11.98%) max: 12446.00 max: 36635.00 min: 9949.00 min: 25669.00 32 thread records: 10 records: 10 avg: 11067.00 avg: 34495.80 std: 971.06(8.77%) std: 2721.36(7.89%) max: 12010.00 max: 38598.00 min: 9002.00 min: 30636.00 In summary, MADV_FREE is about much faster than MADV_DONTNEED. This patch (of 12): Add core MADV_FREE implementation. [akpm@linux-foundation.org: small cleanups] Signed-off-by: Minchan Kim <minchan@kernel.org> Acked-by: Michal Hocko <mhocko@suse.com> Acked-by: Hugh Dickins <hughd@google.com> Cc: Mika Penttil <mika.penttila@nextfour.com> Cc: Michael Kerrisk <mtk.manpages@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Rik van Riel <riel@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Jason Evans <je@fb.com> Cc: Daniel Micay <danielmicay@gmail.com> Cc: "Kirill A. Shutemov" <kirill@shutemov.name> Cc: Shaohua Li <shli@kernel.org> Cc: <yalin.wang2010@gmail.com> Cc: Andy Lutomirski <luto@amacapital.net> Cc: "James E.J. Bottomley" <jejb@parisc-linux.org> Cc: "Kirill A. Shutemov" <kirill@shutemov.name> Cc: "Shaohua Li" <shli@kernel.org> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Chen Gang <gang.chen.5i5j@gmail.com> Cc: Chris Zankel <chris@zankel.net> Cc: Darrick J. Wong <darrick.wong@oracle.com> Cc: David S. Miller <davem@davemloft.net> Cc: Helge Deller <deller@gmx.de> Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Cc: Matt Turner <mattst88@gmail.com> Cc: Max Filippov <jcmvbkbc@gmail.com> Cc: Ralf Baechle <ralf@linux-mips.org> Cc: Richard Henderson <rth@twiddle.net> Cc: Roland Dreier <roland@kernel.org> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Shaohua Li <shli@kernel.org> Cc: Will Deacon <will.deacon@arm.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-01-16 00:54:53 +00:00
* Add it to the swap cache.
*/
err = add_to_swap_cache(folio, entry,
__GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN, NULL);
mm, THP, swap: delay splitting THP during swap out Patch series "THP swap: Delay splitting THP during swapping out", v11. This patchset is to optimize the performance of Transparent Huge Page (THP) swap. Recently, the performance of the storage devices improved so fast that we cannot saturate the disk bandwidth with single logical CPU when do page swap out even on a high-end server machine. Because the performance of the storage device improved faster than that of single logical CPU. And it seems that the trend will not change in the near future. On the other hand, the THP becomes more and more popular because of increased memory size. So it becomes necessary to optimize THP swap performance. The advantages of the THP swap support include: - Batch the swap operations for the THP to reduce lock acquiring/releasing, including allocating/freeing the swap space, adding/deleting to/from the swap cache, and writing/reading the swap space, etc. This will help improve the performance of the THP swap. - The THP swap space read/write will be 2M sequential IO. It is particularly helpful for the swap read, which are usually 4k random IO. This will improve the performance of the THP swap too. - It will help the memory fragmentation, especially when the THP is heavily used by the applications. The 2M continuous pages will be free up after THP swapping out. - It will improve the THP utilization on the system with the swap turned on. Because the speed for khugepaged to collapse the normal pages into the THP is quite slow. After the THP is split during the swapping out, it will take quite long time for the normal pages to collapse back into the THP after being swapped in. The high THP utilization helps the efficiency of the page based memory management too. There are some concerns regarding THP swap in, mainly because possible enlarged read/write IO size (for swap in/out) may put more overhead on the storage device. To deal with that, the THP swap in should be turned on only when necessary. For example, it can be selected via "always/never/madvise" logic, to be turned on globally, turned off globally, or turned on only for VMA with MADV_HUGEPAGE, etc. This patchset is the first step for the THP swap support. The plan is to delay splitting THP step by step, finally avoid splitting THP during the THP swapping out and swap out/in the THP as a whole. As the first step, in this patchset, the splitting huge page is delayed from almost the first step of swapping out to after allocating the swap space for the THP and adding the THP into the swap cache. This will reduce lock acquiring/releasing for the locks used for the swap cache management. With the patchset, the swap out throughput improves 15.5% (from about 3.73GB/s to about 4.31GB/s) in the vm-scalability swap-w-seq test case with 8 processes. The test is done on a Xeon E5 v3 system. The swap device used is a RAM simulated PMEM (persistent memory) device. To test the sequential swapping out, the test case creates 8 processes, which sequentially allocate and write to the anonymous pages until the RAM and part of the swap device is used up. This patch (of 5): In this patch, splitting huge page is delayed from almost the first step of swapping out to after allocating the swap space for the THP (Transparent Huge Page) and adding the THP into the swap cache. This will batch the corresponding operation, thus improve THP swap out throughput. This is the first step for the THP swap optimization. The plan is to delay splitting the THP step by step and avoid splitting the THP finally. In this patch, one swap cluster is used to hold the contents of each THP swapped out. So, the size of the swap cluster is changed to that of the THP (Transparent Huge Page) on x86_64 architecture (512). For other architectures which want such THP swap optimization, ARCH_USES_THP_SWAP_CLUSTER needs to be selected in the Kconfig file for the architecture. In effect, this will enlarge swap cluster size by 2 times on x86_64. Which may make it harder to find a free cluster when the swap space becomes fragmented. So that, this may reduce the continuous swap space allocation and sequential write in theory. The performance test in 0day shows no regressions caused by this. In the future of THP swap optimization, some information of the swapped out THP (such as compound map count) will be recorded in the swap_cluster_info data structure. The mem cgroup swap accounting functions are enhanced to support charge or uncharge a swap cluster backing a THP as a whole. The swap cluster allocate/free functions are added to allocate/free a swap cluster for a THP. A fair simple algorithm is used for swap cluster allocation, that is, only the first swap device in priority list will be tried to allocate the swap cluster. The function will fail if the trying is not successful, and the caller will fallback to allocate a single swap slot instead. This works good enough for normal cases. If the difference of the number of the free swap clusters among multiple swap devices is significant, it is possible that some THPs are split earlier than necessary. For example, this could be caused by big size difference among multiple swap devices. The swap cache functions is enhanced to support add/delete THP to/from the swap cache as a set of (HPAGE_PMD_NR) sub-pages. This may be enhanced in the future with multi-order radix tree. But because we will split the THP soon during swapping out, that optimization doesn't make much sense for this first step. The THP splitting functions are enhanced to support to split THP in swap cache during swapping out. The page lock will be held during allocating the swap cluster, adding the THP into the swap cache and splitting the THP. So in the code path other than swapping out, if the THP need to be split, the PageSwapCache(THP) will be always false. The swap cluster is only available for SSD, so the THP swap optimization in this patchset has no effect for HDD. [ying.huang@intel.com: fix two issues in THP optimize patch] Link: http://lkml.kernel.org/r/87k25ed8zo.fsf@yhuang-dev.intel.com [hannes@cmpxchg.org: extensive cleanups and simplifications, reduce code size] Link: http://lkml.kernel.org/r/20170515112522.32457-2-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Suggested-by: Andrew Morton <akpm@linux-foundation.org> [for config option] Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> [for changes in huge_memory.c and huge_mm.h] Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Ebru Akagunduz <ebru.akagunduz@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Tejun Heo <tj@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Shaohua Li <shli@kernel.org> Cc: Minchan Kim <minchan@kernel.org> 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>
2017-07-06 22:37:18 +00:00
if (err)
/*
* add_to_swap_cache() doesn't return -EEXIST, so we can safely
* clear SWAP_HAS_CACHE flag.
*/
goto fail;
mm: fix data corruption caused by lazyfree page MADV_FREE clears pte dirty bit and then marks the page lazyfree (clear SwapBacked). There is no lock to prevent the page is added to swap cache between these two steps by page reclaim. If page reclaim finds such page, it will simply add the page to swap cache without pageout the page to swap because the page is marked as clean. Next time, page fault will read data from the swap slot which doesn't have the original data, so we have a data corruption. To fix issue, we mark the page dirty and pageout the page. However, we shouldn't dirty all pages which is clean and in swap cache. swapin page is swap cache and clean too. So we only dirty page which is added into swap cache in page reclaim, which shouldn't be swapin page. As Minchan suggested, simply dirty the page in add_to_swap can do the job. Fixes: 802a3a92ad7a ("mm: reclaim MADV_FREE pages") Link: http://lkml.kernel.org/r/08c84256b007bf3f63c91d94383bd9eb6fee2daa.1506446061.git.shli@fb.com Signed-off-by: Shaohua Li <shli@fb.com> Reported-by: Artem Savkov <asavkov@redhat.com> Acked-by: Michal Hocko <mhocko@suse.com> Acked-by: Minchan Kim <minchan@kernel.org> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Hillf Danton <hdanton@sina.com> Cc: Hugh Dickins <hughd@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: <stable@vger.kernel.org> [4.12+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-10-03 23:15:32 +00:00
/*
* Normally the folio will be dirtied in unmap because its
* pte should be dirty. A special case is MADV_FREE page. The
* page's pte could have dirty bit cleared but the folio's
* SwapBacked flag is still set because clearing the dirty bit
* and SwapBacked flag has no lock protected. For such folio,
* unmap will not set dirty bit for it, so folio reclaim will
* not write the folio out. This can cause data corruption when
* the folio is swapped in later. Always setting the dirty flag
* for the folio solves the problem.
mm: fix data corruption caused by lazyfree page MADV_FREE clears pte dirty bit and then marks the page lazyfree (clear SwapBacked). There is no lock to prevent the page is added to swap cache between these two steps by page reclaim. If page reclaim finds such page, it will simply add the page to swap cache without pageout the page to swap because the page is marked as clean. Next time, page fault will read data from the swap slot which doesn't have the original data, so we have a data corruption. To fix issue, we mark the page dirty and pageout the page. However, we shouldn't dirty all pages which is clean and in swap cache. swapin page is swap cache and clean too. So we only dirty page which is added into swap cache in page reclaim, which shouldn't be swapin page. As Minchan suggested, simply dirty the page in add_to_swap can do the job. Fixes: 802a3a92ad7a ("mm: reclaim MADV_FREE pages") Link: http://lkml.kernel.org/r/08c84256b007bf3f63c91d94383bd9eb6fee2daa.1506446061.git.shli@fb.com Signed-off-by: Shaohua Li <shli@fb.com> Reported-by: Artem Savkov <asavkov@redhat.com> Acked-by: Michal Hocko <mhocko@suse.com> Acked-by: Minchan Kim <minchan@kernel.org> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Hillf Danton <hdanton@sina.com> Cc: Hugh Dickins <hughd@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: <stable@vger.kernel.org> [4.12+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-10-03 23:15:32 +00:00
*/
folio_mark_dirty(folio);
mm, THP, swap: delay splitting THP during swap out Patch series "THP swap: Delay splitting THP during swapping out", v11. This patchset is to optimize the performance of Transparent Huge Page (THP) swap. Recently, the performance of the storage devices improved so fast that we cannot saturate the disk bandwidth with single logical CPU when do page swap out even on a high-end server machine. Because the performance of the storage device improved faster than that of single logical CPU. And it seems that the trend will not change in the near future. On the other hand, the THP becomes more and more popular because of increased memory size. So it becomes necessary to optimize THP swap performance. The advantages of the THP swap support include: - Batch the swap operations for the THP to reduce lock acquiring/releasing, including allocating/freeing the swap space, adding/deleting to/from the swap cache, and writing/reading the swap space, etc. This will help improve the performance of the THP swap. - The THP swap space read/write will be 2M sequential IO. It is particularly helpful for the swap read, which are usually 4k random IO. This will improve the performance of the THP swap too. - It will help the memory fragmentation, especially when the THP is heavily used by the applications. The 2M continuous pages will be free up after THP swapping out. - It will improve the THP utilization on the system with the swap turned on. Because the speed for khugepaged to collapse the normal pages into the THP is quite slow. After the THP is split during the swapping out, it will take quite long time for the normal pages to collapse back into the THP after being swapped in. The high THP utilization helps the efficiency of the page based memory management too. There are some concerns regarding THP swap in, mainly because possible enlarged read/write IO size (for swap in/out) may put more overhead on the storage device. To deal with that, the THP swap in should be turned on only when necessary. For example, it can be selected via "always/never/madvise" logic, to be turned on globally, turned off globally, or turned on only for VMA with MADV_HUGEPAGE, etc. This patchset is the first step for the THP swap support. The plan is to delay splitting THP step by step, finally avoid splitting THP during the THP swapping out and swap out/in the THP as a whole. As the first step, in this patchset, the splitting huge page is delayed from almost the first step of swapping out to after allocating the swap space for the THP and adding the THP into the swap cache. This will reduce lock acquiring/releasing for the locks used for the swap cache management. With the patchset, the swap out throughput improves 15.5% (from about 3.73GB/s to about 4.31GB/s) in the vm-scalability swap-w-seq test case with 8 processes. The test is done on a Xeon E5 v3 system. The swap device used is a RAM simulated PMEM (persistent memory) device. To test the sequential swapping out, the test case creates 8 processes, which sequentially allocate and write to the anonymous pages until the RAM and part of the swap device is used up. This patch (of 5): In this patch, splitting huge page is delayed from almost the first step of swapping out to after allocating the swap space for the THP (Transparent Huge Page) and adding the THP into the swap cache. This will batch the corresponding operation, thus improve THP swap out throughput. This is the first step for the THP swap optimization. The plan is to delay splitting the THP step by step and avoid splitting the THP finally. In this patch, one swap cluster is used to hold the contents of each THP swapped out. So, the size of the swap cluster is changed to that of the THP (Transparent Huge Page) on x86_64 architecture (512). For other architectures which want such THP swap optimization, ARCH_USES_THP_SWAP_CLUSTER needs to be selected in the Kconfig file for the architecture. In effect, this will enlarge swap cluster size by 2 times on x86_64. Which may make it harder to find a free cluster when the swap space becomes fragmented. So that, this may reduce the continuous swap space allocation and sequential write in theory. The performance test in 0day shows no regressions caused by this. In the future of THP swap optimization, some information of the swapped out THP (such as compound map count) will be recorded in the swap_cluster_info data structure. The mem cgroup swap accounting functions are enhanced to support charge or uncharge a swap cluster backing a THP as a whole. The swap cluster allocate/free functions are added to allocate/free a swap cluster for a THP. A fair simple algorithm is used for swap cluster allocation, that is, only the first swap device in priority list will be tried to allocate the swap cluster. The function will fail if the trying is not successful, and the caller will fallback to allocate a single swap slot instead. This works good enough for normal cases. If the difference of the number of the free swap clusters among multiple swap devices is significant, it is possible that some THPs are split earlier than necessary. For example, this could be caused by big size difference among multiple swap devices. The swap cache functions is enhanced to support add/delete THP to/from the swap cache as a set of (HPAGE_PMD_NR) sub-pages. This may be enhanced in the future with multi-order radix tree. But because we will split the THP soon during swapping out, that optimization doesn't make much sense for this first step. The THP splitting functions are enhanced to support to split THP in swap cache during swapping out. The page lock will be held during allocating the swap cluster, adding the THP into the swap cache and splitting the THP. So in the code path other than swapping out, if the THP need to be split, the PageSwapCache(THP) will be always false. The swap cluster is only available for SSD, so the THP swap optimization in this patchset has no effect for HDD. [ying.huang@intel.com: fix two issues in THP optimize patch] Link: http://lkml.kernel.org/r/87k25ed8zo.fsf@yhuang-dev.intel.com [hannes@cmpxchg.org: extensive cleanups and simplifications, reduce code size] Link: http://lkml.kernel.org/r/20170515112522.32457-2-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Suggested-by: Andrew Morton <akpm@linux-foundation.org> [for config option] Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> [for changes in huge_memory.c and huge_mm.h] Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Ebru Akagunduz <ebru.akagunduz@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Tejun Heo <tj@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Shaohua Li <shli@kernel.org> Cc: Minchan Kim <minchan@kernel.org> 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>
2017-07-06 22:37:18 +00:00
return true;
mm, THP, swap: delay splitting THP during swap out Patch series "THP swap: Delay splitting THP during swapping out", v11. This patchset is to optimize the performance of Transparent Huge Page (THP) swap. Recently, the performance of the storage devices improved so fast that we cannot saturate the disk bandwidth with single logical CPU when do page swap out even on a high-end server machine. Because the performance of the storage device improved faster than that of single logical CPU. And it seems that the trend will not change in the near future. On the other hand, the THP becomes more and more popular because of increased memory size. So it becomes necessary to optimize THP swap performance. The advantages of the THP swap support include: - Batch the swap operations for the THP to reduce lock acquiring/releasing, including allocating/freeing the swap space, adding/deleting to/from the swap cache, and writing/reading the swap space, etc. This will help improve the performance of the THP swap. - The THP swap space read/write will be 2M sequential IO. It is particularly helpful for the swap read, which are usually 4k random IO. This will improve the performance of the THP swap too. - It will help the memory fragmentation, especially when the THP is heavily used by the applications. The 2M continuous pages will be free up after THP swapping out. - It will improve the THP utilization on the system with the swap turned on. Because the speed for khugepaged to collapse the normal pages into the THP is quite slow. After the THP is split during the swapping out, it will take quite long time for the normal pages to collapse back into the THP after being swapped in. The high THP utilization helps the efficiency of the page based memory management too. There are some concerns regarding THP swap in, mainly because possible enlarged read/write IO size (for swap in/out) may put more overhead on the storage device. To deal with that, the THP swap in should be turned on only when necessary. For example, it can be selected via "always/never/madvise" logic, to be turned on globally, turned off globally, or turned on only for VMA with MADV_HUGEPAGE, etc. This patchset is the first step for the THP swap support. The plan is to delay splitting THP step by step, finally avoid splitting THP during the THP swapping out and swap out/in the THP as a whole. As the first step, in this patchset, the splitting huge page is delayed from almost the first step of swapping out to after allocating the swap space for the THP and adding the THP into the swap cache. This will reduce lock acquiring/releasing for the locks used for the swap cache management. With the patchset, the swap out throughput improves 15.5% (from about 3.73GB/s to about 4.31GB/s) in the vm-scalability swap-w-seq test case with 8 processes. The test is done on a Xeon E5 v3 system. The swap device used is a RAM simulated PMEM (persistent memory) device. To test the sequential swapping out, the test case creates 8 processes, which sequentially allocate and write to the anonymous pages until the RAM and part of the swap device is used up. This patch (of 5): In this patch, splitting huge page is delayed from almost the first step of swapping out to after allocating the swap space for the THP (Transparent Huge Page) and adding the THP into the swap cache. This will batch the corresponding operation, thus improve THP swap out throughput. This is the first step for the THP swap optimization. The plan is to delay splitting the THP step by step and avoid splitting the THP finally. In this patch, one swap cluster is used to hold the contents of each THP swapped out. So, the size of the swap cluster is changed to that of the THP (Transparent Huge Page) on x86_64 architecture (512). For other architectures which want such THP swap optimization, ARCH_USES_THP_SWAP_CLUSTER needs to be selected in the Kconfig file for the architecture. In effect, this will enlarge swap cluster size by 2 times on x86_64. Which may make it harder to find a free cluster when the swap space becomes fragmented. So that, this may reduce the continuous swap space allocation and sequential write in theory. The performance test in 0day shows no regressions caused by this. In the future of THP swap optimization, some information of the swapped out THP (such as compound map count) will be recorded in the swap_cluster_info data structure. The mem cgroup swap accounting functions are enhanced to support charge or uncharge a swap cluster backing a THP as a whole. The swap cluster allocate/free functions are added to allocate/free a swap cluster for a THP. A fair simple algorithm is used for swap cluster allocation, that is, only the first swap device in priority list will be tried to allocate the swap cluster. The function will fail if the trying is not successful, and the caller will fallback to allocate a single swap slot instead. This works good enough for normal cases. If the difference of the number of the free swap clusters among multiple swap devices is significant, it is possible that some THPs are split earlier than necessary. For example, this could be caused by big size difference among multiple swap devices. The swap cache functions is enhanced to support add/delete THP to/from the swap cache as a set of (HPAGE_PMD_NR) sub-pages. This may be enhanced in the future with multi-order radix tree. But because we will split the THP soon during swapping out, that optimization doesn't make much sense for this first step. The THP splitting functions are enhanced to support to split THP in swap cache during swapping out. The page lock will be held during allocating the swap cluster, adding the THP into the swap cache and splitting the THP. So in the code path other than swapping out, if the THP need to be split, the PageSwapCache(THP) will be always false. The swap cluster is only available for SSD, so the THP swap optimization in this patchset has no effect for HDD. [ying.huang@intel.com: fix two issues in THP optimize patch] Link: http://lkml.kernel.org/r/87k25ed8zo.fsf@yhuang-dev.intel.com [hannes@cmpxchg.org: extensive cleanups and simplifications, reduce code size] Link: http://lkml.kernel.org/r/20170515112522.32457-2-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Suggested-by: Andrew Morton <akpm@linux-foundation.org> [for config option] Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> [for changes in huge_memory.c and huge_mm.h] Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Ebru Akagunduz <ebru.akagunduz@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Tejun Heo <tj@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Shaohua Li <shli@kernel.org> Cc: Minchan Kim <minchan@kernel.org> 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>
2017-07-06 22:37:18 +00:00
fail:
put_swap_folio(folio, entry);
return false;
}
/*
* This must be called only on folios that have
* been verified to be in the swap cache and locked.
* It will never put the folio into the free list,
* the caller has a reference on the folio.
*/
void delete_from_swap_cache(struct folio *folio)
{
swp_entry_t entry = folio->swap;
struct address_space *address_space = swap_address_space(entry);
xa_lock_irq(&address_space->i_pages);
__delete_from_swap_cache(folio, entry, NULL);
xa_unlock_irq(&address_space->i_pages);
put_swap_folio(folio, entry);
folio_ref_sub(folio, folio_nr_pages(folio));
}
void clear_shadow_from_swap_cache(int type, unsigned long begin,
unsigned long end)
{
unsigned long curr = begin;
void *old;
for (;;) {
swp_entry_t entry = swp_entry(type, curr);
struct address_space *address_space = swap_address_space(entry);
XA_STATE(xas, &address_space->i_pages, curr);
xas_set_update(&xas, workingset_update_node);
xa_lock_irq(&address_space->i_pages);
xas_for_each(&xas, old, end) {
if (!xa_is_value(old))
continue;
xas_store(&xas, NULL);
}
xa_unlock_irq(&address_space->i_pages);
/* search the next swapcache until we meet end */
curr >>= SWAP_ADDRESS_SPACE_SHIFT;
curr++;
curr <<= SWAP_ADDRESS_SPACE_SHIFT;
if (curr > end)
break;
}
}
mm: ptep_get() conversion Convert all instances of direct pte_t* dereferencing to instead use ptep_get() helper. This means that by default, the accesses change from a C dereference to a READ_ONCE(). This is technically the correct thing to do since where pgtables are modified by HW (for access/dirty) they are volatile and therefore we should always ensure READ_ONCE() semantics. But more importantly, by always using the helper, it can be overridden by the architecture to fully encapsulate the contents of the pte. Arch code is deliberately not converted, as the arch code knows best. It is intended that arch code (arm64) will override the default with its own implementation that can (e.g.) hide certain bits from the core code, or determine young/dirty status by mixing in state from another source. Conversion was done using Coccinelle: ---- // $ make coccicheck \ // COCCI=ptepget.cocci \ // SPFLAGS="--include-headers" \ // MODE=patch virtual patch @ depends on patch @ pte_t *v; @@ - *v + ptep_get(v) ---- Then reviewed and hand-edited to avoid multiple unnecessary calls to ptep_get(), instead opting to store the result of a single call in a variable, where it is correct to do so. This aims to negate any cost of READ_ONCE() and will benefit arch-overrides that may be more complex. Included is a fix for an issue in an earlier version of this patch that was pointed out by kernel test robot. The issue arose because config MMU=n elides definition of the ptep helper functions, including ptep_get(). HUGETLB_PAGE=n configs still define a simple huge_ptep_clear_flush() for linking purposes, which dereferences the ptep. So when both configs are disabled, this caused a build error because ptep_get() is not defined. Fix by continuing to do a direct dereference when MMU=n. This is safe because for this config the arch code cannot be trying to virtualize the ptes because none of the ptep helpers are defined. Link: https://lkml.kernel.org/r/20230612151545.3317766-4-ryan.roberts@arm.com Reported-by: kernel test robot <lkp@intel.com> Link: https://lore.kernel.org/oe-kbuild-all/202305120142.yXsNEo6H-lkp@intel.com/ Signed-off-by: Ryan Roberts <ryan.roberts@arm.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Potapenko <glider@google.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Andrey Konovalov <andreyknvl@gmail.com> Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com> Cc: Christian Brauner <brauner@kernel.org> Cc: Christoph Hellwig <hch@infradead.org> Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Dave Airlie <airlied@gmail.com> Cc: Dimitri Sivanich <dimitri.sivanich@hpe.com> Cc: Dmitry Vyukov <dvyukov@google.com> Cc: Ian Rogers <irogers@google.com> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jérôme Glisse <jglisse@redhat.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Lorenzo Stoakes <lstoakes@gmail.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Matthew Wilcox <willy@infradead.org> Cc: Miaohe Lin <linmiaohe@huawei.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Mike Rapoport (IBM) <rppt@kernel.org> Cc: Muchun Song <muchun.song@linux.dev> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Naoya Horiguchi <naoya.horiguchi@nec.com> Cc: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Roman Gushchin <roman.gushchin@linux.dev> Cc: SeongJae Park <sj@kernel.org> Cc: Shakeel Butt <shakeelb@google.com> Cc: Uladzislau Rezki (Sony) <urezki@gmail.com> Cc: Vincenzo Frascino <vincenzo.frascino@arm.com> Cc: Yu Zhao <yuzhao@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-06-12 15:15:45 +00:00
/*
* If we are the only user, then try to free up the swap cache.
*
* Its ok to check the swapcache flag without the folio lock
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
* here because we are going to recheck again inside
* folio_free_swap() _with_ the lock.
* - Marcelo
*/
void free_swap_cache(struct folio *folio)
{
if (folio_test_swapcache(folio) && !folio_mapped(folio) &&
folio_trylock(folio)) {
folio_free_swap(folio);
folio_unlock(folio);
}
}
mm: ptep_get() conversion Convert all instances of direct pte_t* dereferencing to instead use ptep_get() helper. This means that by default, the accesses change from a C dereference to a READ_ONCE(). This is technically the correct thing to do since where pgtables are modified by HW (for access/dirty) they are volatile and therefore we should always ensure READ_ONCE() semantics. But more importantly, by always using the helper, it can be overridden by the architecture to fully encapsulate the contents of the pte. Arch code is deliberately not converted, as the arch code knows best. It is intended that arch code (arm64) will override the default with its own implementation that can (e.g.) hide certain bits from the core code, or determine young/dirty status by mixing in state from another source. Conversion was done using Coccinelle: ---- // $ make coccicheck \ // COCCI=ptepget.cocci \ // SPFLAGS="--include-headers" \ // MODE=patch virtual patch @ depends on patch @ pte_t *v; @@ - *v + ptep_get(v) ---- Then reviewed and hand-edited to avoid multiple unnecessary calls to ptep_get(), instead opting to store the result of a single call in a variable, where it is correct to do so. This aims to negate any cost of READ_ONCE() and will benefit arch-overrides that may be more complex. Included is a fix for an issue in an earlier version of this patch that was pointed out by kernel test robot. The issue arose because config MMU=n elides definition of the ptep helper functions, including ptep_get(). HUGETLB_PAGE=n configs still define a simple huge_ptep_clear_flush() for linking purposes, which dereferences the ptep. So when both configs are disabled, this caused a build error because ptep_get() is not defined. Fix by continuing to do a direct dereference when MMU=n. This is safe because for this config the arch code cannot be trying to virtualize the ptes because none of the ptep helpers are defined. Link: https://lkml.kernel.org/r/20230612151545.3317766-4-ryan.roberts@arm.com Reported-by: kernel test robot <lkp@intel.com> Link: https://lore.kernel.org/oe-kbuild-all/202305120142.yXsNEo6H-lkp@intel.com/ Signed-off-by: Ryan Roberts <ryan.roberts@arm.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Potapenko <glider@google.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Alex Williamson <alex.williamson@redhat.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Andrey Konovalov <andreyknvl@gmail.com> Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com> Cc: Christian Brauner <brauner@kernel.org> Cc: Christoph Hellwig <hch@infradead.org> Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Dave Airlie <airlied@gmail.com> Cc: Dimitri Sivanich <dimitri.sivanich@hpe.com> Cc: Dmitry Vyukov <dvyukov@google.com> Cc: Ian Rogers <irogers@google.com> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jérôme Glisse <jglisse@redhat.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Lorenzo Stoakes <lstoakes@gmail.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Matthew Wilcox <willy@infradead.org> Cc: Miaohe Lin <linmiaohe@huawei.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Mike Rapoport (IBM) <rppt@kernel.org> Cc: Muchun Song <muchun.song@linux.dev> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Naoya Horiguchi <naoya.horiguchi@nec.com> Cc: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Roman Gushchin <roman.gushchin@linux.dev> Cc: SeongJae Park <sj@kernel.org> Cc: Shakeel Butt <shakeelb@google.com> Cc: Uladzislau Rezki (Sony) <urezki@gmail.com> Cc: Vincenzo Frascino <vincenzo.frascino@arm.com> Cc: Yu Zhao <yuzhao@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-06-12 15:15:45 +00:00
/*
* Perform a free_page(), also freeing any swap cache associated with
* this page if it is the last user of the page.
*/
void free_page_and_swap_cache(struct page *page)
{
struct folio *folio = page_folio(page);
free_swap_cache(folio);
if (!is_huge_zero_folio(folio))
folio_put(folio);
}
/*
* Passed an array of pages, drop them all from swapcache and then release
* them. They are removed from the LRU and freed if this is their last use.
*/
void free_pages_and_swap_cache(struct encoded_page **pages, int nr)
{
struct folio_batch folios;
unsigned int refs[PAGEVEC_SIZE];
lru_add_drain();
folio_batch_init(&folios);
mm/mmu_gather: add __tlb_remove_folio_pages() Add __tlb_remove_folio_pages(), which will remove multiple consecutive pages that belong to the same large folio, instead of only a single page. We'll be using this function when optimizing unmapping/zapping of large folios that are mapped by PTEs. We're using the remaining spare bit in an encoded_page to indicate that the next enoced page in an array contains actually shifted "nr_pages". Teach swap/freeing code about putting multiple folio references, and delayed rmap handling to remove page ranges of a folio. This extension allows for still gathering almost as many small folios as we used to (-1, because we have to prepare for a possibly bigger next entry), but still allows for gathering consecutive pages that belong to the same large folio. Note that we don't pass the folio pointer, because it is not required for now. Further, we don't support page_size != PAGE_SIZE, it won't be required for simple PTE batching. We have to provide a separate s390 implementation, but it's fairly straight forward. Another, more invasive and likely more expensive, approach would be to use folio+range or a PFN range instead of page+nr_pages. But, we should do that consistently for the whole mmu_gather. For now, let's keep it simple and add "nr_pages" only. Note that it is now possible to gather significantly more pages: In the past, we were able to gather ~10000 pages, now we can also gather ~5000 folio fragments that span multiple pages. A folio fragment on x86-64 can span up to 512 pages (2 MiB THP) and on arm64 with 64k in theory 8192 pages (512 MiB THP). Gathering more memory is not considered something we should worry about, especially because these are already corner cases. While we can gather more total memory, we won't free more folio fragments. As long as page freeing time primarily only depends on the number of involved folios, there is no effective change for !preempt configurations. However, we'll adjust tlb_batch_pages_flush() separately to handle corner cases where page freeing time grows proportionally with the actual memory size. Link: https://lkml.kernel.org/r/20240214204435.167852-9-david@redhat.com Signed-off-by: David Hildenbrand <david@redhat.com> Reviewed-by: Ryan Roberts <ryan.roberts@arm.com> Cc: Alexander Gordeev <agordeev@linux.ibm.com> Cc: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Christian Borntraeger <borntraeger@linux.ibm.com> Cc: Christophe Leroy <christophe.leroy@csgroup.eu> Cc: Heiko Carstens <hca@linux.ibm.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Michael Ellerman <mpe@ellerman.id.au> Cc: Michal Hocko <mhocko@suse.com> Cc: "Naveen N. Rao" <naveen.n.rao@linux.ibm.com> Cc: Nicholas Piggin <npiggin@gmail.com> Cc: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Sven Schnelle <svens@linux.ibm.com> Cc: Vasily Gorbik <gor@linux.ibm.com> Cc: Will Deacon <will@kernel.org> Cc: Yin Fengwei <fengwei.yin@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-02-14 20:44:33 +00:00
for (int i = 0; i < nr; i++) {
struct folio *folio = page_folio(encoded_page_ptr(pages[i]));
mm/mmu_gather: add __tlb_remove_folio_pages() Add __tlb_remove_folio_pages(), which will remove multiple consecutive pages that belong to the same large folio, instead of only a single page. We'll be using this function when optimizing unmapping/zapping of large folios that are mapped by PTEs. We're using the remaining spare bit in an encoded_page to indicate that the next enoced page in an array contains actually shifted "nr_pages". Teach swap/freeing code about putting multiple folio references, and delayed rmap handling to remove page ranges of a folio. This extension allows for still gathering almost as many small folios as we used to (-1, because we have to prepare for a possibly bigger next entry), but still allows for gathering consecutive pages that belong to the same large folio. Note that we don't pass the folio pointer, because it is not required for now. Further, we don't support page_size != PAGE_SIZE, it won't be required for simple PTE batching. We have to provide a separate s390 implementation, but it's fairly straight forward. Another, more invasive and likely more expensive, approach would be to use folio+range or a PFN range instead of page+nr_pages. But, we should do that consistently for the whole mmu_gather. For now, let's keep it simple and add "nr_pages" only. Note that it is now possible to gather significantly more pages: In the past, we were able to gather ~10000 pages, now we can also gather ~5000 folio fragments that span multiple pages. A folio fragment on x86-64 can span up to 512 pages (2 MiB THP) and on arm64 with 64k in theory 8192 pages (512 MiB THP). Gathering more memory is not considered something we should worry about, especially because these are already corner cases. While we can gather more total memory, we won't free more folio fragments. As long as page freeing time primarily only depends on the number of involved folios, there is no effective change for !preempt configurations. However, we'll adjust tlb_batch_pages_flush() separately to handle corner cases where page freeing time grows proportionally with the actual memory size. Link: https://lkml.kernel.org/r/20240214204435.167852-9-david@redhat.com Signed-off-by: David Hildenbrand <david@redhat.com> Reviewed-by: Ryan Roberts <ryan.roberts@arm.com> Cc: Alexander Gordeev <agordeev@linux.ibm.com> Cc: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Christian Borntraeger <borntraeger@linux.ibm.com> Cc: Christophe Leroy <christophe.leroy@csgroup.eu> Cc: Heiko Carstens <hca@linux.ibm.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Michael Ellerman <mpe@ellerman.id.au> Cc: Michal Hocko <mhocko@suse.com> Cc: "Naveen N. Rao" <naveen.n.rao@linux.ibm.com> Cc: Nicholas Piggin <npiggin@gmail.com> Cc: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Sven Schnelle <svens@linux.ibm.com> Cc: Vasily Gorbik <gor@linux.ibm.com> Cc: Will Deacon <will@kernel.org> Cc: Yin Fengwei <fengwei.yin@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-02-14 20:44:33 +00:00
free_swap_cache(folio);
refs[folios.nr] = 1;
mm/mmu_gather: add __tlb_remove_folio_pages() Add __tlb_remove_folio_pages(), which will remove multiple consecutive pages that belong to the same large folio, instead of only a single page. We'll be using this function when optimizing unmapping/zapping of large folios that are mapped by PTEs. We're using the remaining spare bit in an encoded_page to indicate that the next enoced page in an array contains actually shifted "nr_pages". Teach swap/freeing code about putting multiple folio references, and delayed rmap handling to remove page ranges of a folio. This extension allows for still gathering almost as many small folios as we used to (-1, because we have to prepare for a possibly bigger next entry), but still allows for gathering consecutive pages that belong to the same large folio. Note that we don't pass the folio pointer, because it is not required for now. Further, we don't support page_size != PAGE_SIZE, it won't be required for simple PTE batching. We have to provide a separate s390 implementation, but it's fairly straight forward. Another, more invasive and likely more expensive, approach would be to use folio+range or a PFN range instead of page+nr_pages. But, we should do that consistently for the whole mmu_gather. For now, let's keep it simple and add "nr_pages" only. Note that it is now possible to gather significantly more pages: In the past, we were able to gather ~10000 pages, now we can also gather ~5000 folio fragments that span multiple pages. A folio fragment on x86-64 can span up to 512 pages (2 MiB THP) and on arm64 with 64k in theory 8192 pages (512 MiB THP). Gathering more memory is not considered something we should worry about, especially because these are already corner cases. While we can gather more total memory, we won't free more folio fragments. As long as page freeing time primarily only depends on the number of involved folios, there is no effective change for !preempt configurations. However, we'll adjust tlb_batch_pages_flush() separately to handle corner cases where page freeing time grows proportionally with the actual memory size. Link: https://lkml.kernel.org/r/20240214204435.167852-9-david@redhat.com Signed-off-by: David Hildenbrand <david@redhat.com> Reviewed-by: Ryan Roberts <ryan.roberts@arm.com> Cc: Alexander Gordeev <agordeev@linux.ibm.com> Cc: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Christian Borntraeger <borntraeger@linux.ibm.com> Cc: Christophe Leroy <christophe.leroy@csgroup.eu> Cc: Heiko Carstens <hca@linux.ibm.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Michael Ellerman <mpe@ellerman.id.au> Cc: Michal Hocko <mhocko@suse.com> Cc: "Naveen N. Rao" <naveen.n.rao@linux.ibm.com> Cc: Nicholas Piggin <npiggin@gmail.com> Cc: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Sven Schnelle <svens@linux.ibm.com> Cc: Vasily Gorbik <gor@linux.ibm.com> Cc: Will Deacon <will@kernel.org> Cc: Yin Fengwei <fengwei.yin@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-02-14 20:44:33 +00:00
if (unlikely(encoded_page_flags(pages[i]) &
ENCODED_PAGE_BIT_NR_PAGES_NEXT))
refs[folios.nr] = encoded_nr_pages(pages[++i]);
mm/mmu_gather: add __tlb_remove_folio_pages() Add __tlb_remove_folio_pages(), which will remove multiple consecutive pages that belong to the same large folio, instead of only a single page. We'll be using this function when optimizing unmapping/zapping of large folios that are mapped by PTEs. We're using the remaining spare bit in an encoded_page to indicate that the next enoced page in an array contains actually shifted "nr_pages". Teach swap/freeing code about putting multiple folio references, and delayed rmap handling to remove page ranges of a folio. This extension allows for still gathering almost as many small folios as we used to (-1, because we have to prepare for a possibly bigger next entry), but still allows for gathering consecutive pages that belong to the same large folio. Note that we don't pass the folio pointer, because it is not required for now. Further, we don't support page_size != PAGE_SIZE, it won't be required for simple PTE batching. We have to provide a separate s390 implementation, but it's fairly straight forward. Another, more invasive and likely more expensive, approach would be to use folio+range or a PFN range instead of page+nr_pages. But, we should do that consistently for the whole mmu_gather. For now, let's keep it simple and add "nr_pages" only. Note that it is now possible to gather significantly more pages: In the past, we were able to gather ~10000 pages, now we can also gather ~5000 folio fragments that span multiple pages. A folio fragment on x86-64 can span up to 512 pages (2 MiB THP) and on arm64 with 64k in theory 8192 pages (512 MiB THP). Gathering more memory is not considered something we should worry about, especially because these are already corner cases. While we can gather more total memory, we won't free more folio fragments. As long as page freeing time primarily only depends on the number of involved folios, there is no effective change for !preempt configurations. However, we'll adjust tlb_batch_pages_flush() separately to handle corner cases where page freeing time grows proportionally with the actual memory size. Link: https://lkml.kernel.org/r/20240214204435.167852-9-david@redhat.com Signed-off-by: David Hildenbrand <david@redhat.com> Reviewed-by: Ryan Roberts <ryan.roberts@arm.com> Cc: Alexander Gordeev <agordeev@linux.ibm.com> Cc: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Christian Borntraeger <borntraeger@linux.ibm.com> Cc: Christophe Leroy <christophe.leroy@csgroup.eu> Cc: Heiko Carstens <hca@linux.ibm.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Michael Ellerman <mpe@ellerman.id.au> Cc: Michal Hocko <mhocko@suse.com> Cc: "Naveen N. Rao" <naveen.n.rao@linux.ibm.com> Cc: Nicholas Piggin <npiggin@gmail.com> Cc: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Sven Schnelle <svens@linux.ibm.com> Cc: Vasily Gorbik <gor@linux.ibm.com> Cc: Will Deacon <will@kernel.org> Cc: Yin Fengwei <fengwei.yin@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-02-14 20:44:33 +00:00
if (folio_batch_add(&folios, folio) == 0)
folios_put_refs(&folios, refs);
mm/mmu_gather: add __tlb_remove_folio_pages() Add __tlb_remove_folio_pages(), which will remove multiple consecutive pages that belong to the same large folio, instead of only a single page. We'll be using this function when optimizing unmapping/zapping of large folios that are mapped by PTEs. We're using the remaining spare bit in an encoded_page to indicate that the next enoced page in an array contains actually shifted "nr_pages". Teach swap/freeing code about putting multiple folio references, and delayed rmap handling to remove page ranges of a folio. This extension allows for still gathering almost as many small folios as we used to (-1, because we have to prepare for a possibly bigger next entry), but still allows for gathering consecutive pages that belong to the same large folio. Note that we don't pass the folio pointer, because it is not required for now. Further, we don't support page_size != PAGE_SIZE, it won't be required for simple PTE batching. We have to provide a separate s390 implementation, but it's fairly straight forward. Another, more invasive and likely more expensive, approach would be to use folio+range or a PFN range instead of page+nr_pages. But, we should do that consistently for the whole mmu_gather. For now, let's keep it simple and add "nr_pages" only. Note that it is now possible to gather significantly more pages: In the past, we were able to gather ~10000 pages, now we can also gather ~5000 folio fragments that span multiple pages. A folio fragment on x86-64 can span up to 512 pages (2 MiB THP) and on arm64 with 64k in theory 8192 pages (512 MiB THP). Gathering more memory is not considered something we should worry about, especially because these are already corner cases. While we can gather more total memory, we won't free more folio fragments. As long as page freeing time primarily only depends on the number of involved folios, there is no effective change for !preempt configurations. However, we'll adjust tlb_batch_pages_flush() separately to handle corner cases where page freeing time grows proportionally with the actual memory size. Link: https://lkml.kernel.org/r/20240214204435.167852-9-david@redhat.com Signed-off-by: David Hildenbrand <david@redhat.com> Reviewed-by: Ryan Roberts <ryan.roberts@arm.com> Cc: Alexander Gordeev <agordeev@linux.ibm.com> Cc: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Christian Borntraeger <borntraeger@linux.ibm.com> Cc: Christophe Leroy <christophe.leroy@csgroup.eu> Cc: Heiko Carstens <hca@linux.ibm.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Michael Ellerman <mpe@ellerman.id.au> Cc: Michal Hocko <mhocko@suse.com> Cc: "Naveen N. Rao" <naveen.n.rao@linux.ibm.com> Cc: Nicholas Piggin <npiggin@gmail.com> Cc: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Sven Schnelle <svens@linux.ibm.com> Cc: Vasily Gorbik <gor@linux.ibm.com> Cc: Will Deacon <will@kernel.org> Cc: Yin Fengwei <fengwei.yin@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-02-14 20:44:33 +00:00
}
if (folios.nr)
folios_put_refs(&folios, refs);
}
static inline bool swap_use_vma_readahead(void)
{
return READ_ONCE(enable_vma_readahead) && !atomic_read(&nr_rotate_swap);
}
/*
* Lookup a swap entry in the swap cache. A found folio will be returned
* unlocked and with its refcount incremented - we rely on the kernel
* lock getting page table operations atomic even if we drop the folio
* lock before returning.
*
* Caller must lock the swap device or hold a reference to keep it valid.
*/
struct folio *swap_cache_get_folio(swp_entry_t entry,
struct vm_area_struct *vma, unsigned long addr)
{
struct folio *folio;
folio = filemap_get_folio(swap_address_space(entry), swp_offset(entry));
if (!IS_ERR(folio)) {
bool vma_ra = swap_use_vma_readahead();
bool readahead;
/*
* At the moment, we don't support PG_readahead for anon THP
* so let's bail out rather than confusing the readahead stat.
*/
if (unlikely(folio_test_large(folio)))
return folio;
readahead = folio_test_clear_readahead(folio);
if (vma && vma_ra) {
unsigned long ra_val;
int win, hits;
ra_val = GET_SWAP_RA_VAL(vma);
win = SWAP_RA_WIN(ra_val);
hits = SWAP_RA_HITS(ra_val);
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
if (readahead)
hits = min_t(int, hits + 1, SWAP_RA_HITS_MAX);
atomic_long_set(&vma->swap_readahead_info,
SWAP_RA_VAL(addr, win, hits));
}
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
if (readahead) {
mm, swap: add swap readahead hit statistics Patch series "mm, swap: VMA based swap readahead", v4. The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory space. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patchset, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. In addition to the swap readahead changes, some new sysfs interface is added to show the efficiency of the readahead algorithm and some other swap statistics. This new implementation will incur more small random read, on SSD, the improved correctness of estimation and readahead target should beat the potential increased overhead, this is also illustrated in the test results below. But on HDD, the overhead may beat the benefit, so the original implementation will be used by default. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. This patch (of 5): The statistics for total readahead pages and total readahead hits are recorded and exported via the following sysfs interface. /sys/kernel/mm/swap/ra_hits /sys/kernel/mm/swap/ra_total With them, the efficiency of the swap readahead could be measured, so that the swap readahead algorithm and parameters could be tuned accordingly. [akpm@linux-foundation.org: don't display swap stats if CONFIG_SWAP=n] Link: http://lkml.kernel.org/r/20170807054038.1843-2-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:29 +00:00
count_vm_event(SWAP_RA_HIT);
if (!vma || !vma_ra)
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
atomic_inc(&swapin_readahead_hits);
mm, swap: add swap readahead hit statistics Patch series "mm, swap: VMA based swap readahead", v4. The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory space. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patchset, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. In addition to the swap readahead changes, some new sysfs interface is added to show the efficiency of the readahead algorithm and some other swap statistics. This new implementation will incur more small random read, on SSD, the improved correctness of estimation and readahead target should beat the potential increased overhead, this is also illustrated in the test results below. But on HDD, the overhead may beat the benefit, so the original implementation will be used by default. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. This patch (of 5): The statistics for total readahead pages and total readahead hits are recorded and exported via the following sysfs interface. /sys/kernel/mm/swap/ra_hits /sys/kernel/mm/swap/ra_total With them, the efficiency of the swap readahead could be measured, so that the swap readahead algorithm and parameters could be tuned accordingly. [akpm@linux-foundation.org: don't display swap stats if CONFIG_SWAP=n] Link: http://lkml.kernel.org/r/20170807054038.1843-2-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:29 +00:00
}
} else {
folio = NULL;
swap: add a simple detector for inappropriate swapin readahead This is a patch to improve swap readahead algorithm. It's from Hugh and I slightly changed it. Hugh's original changelog: swapin readahead does a blind readahead, whether or not the swapin is sequential. This may be ok on harddisk, because large reads have relatively small costs, and if the readahead pages are unneeded they can be reclaimed easily - though, what if their allocation forced reclaim of useful pages? But on SSD devices large reads are more expensive than small ones: if the readahead pages are unneeded, reading them in caused significant overhead. This patch adds very simplistic random read detection. Stealing the PageReadahead technique from Konstantin Khlebnikov's patch, avoiding the vma/anon_vma sophistications of Shaohua Li's patch, swapin_nr_pages() simply looks at readahead's current success rate, and narrows or widens its readahead window accordingly. There is little science to its heuristic: it's about as stupid as can be whilst remaining effective. The table below shows elapsed times (in centiseconds) when running a single repetitive swapping load across a 1000MB mapping in 900MB ram with 1GB swap (the harddisk tests had taken painfully too long when I used mem=500M, but SSD shows similar results for that). Vanilla is the 3.6-rc7 kernel on which I started; Shaohua denotes his Sep 3 patch in mmotm and linux-next; HughOld denotes my Oct 1 patch which Shaohua showed to be defective; HughNew this Nov 14 patch, with page_cluster as usual at default of 3 (8-page reads); HughPC4 this same patch with page_cluster 4 (16-page reads); HughPC0 with page_cluster 0 (1-page reads: no readahead). HDD for swapping to harddisk, SSD for swapping to VertexII SSD. Seq for sequential access to the mapping, cycling five times around; Rand for the same number of random touches. Anon for a MAP_PRIVATE anon mapping; Shmem for a MAP_SHARED anon mapping, equivalent to tmpfs. One weakness of Shaohua's vma/anon_vma approach was that it did not optimize Shmem: seen below. Konstantin's approach was perhaps mistuned, 50% slower on Seq: did not compete and is not shown below. HDD Vanilla Shaohua HughOld HughNew HughPC4 HughPC0 Seq Anon 73921 76210 75611 76904 78191 121542 Seq Shmem 73601 73176 73855 72947 74543 118322 Rand Anon 895392 831243 871569 845197 846496 841680 Rand Shmem 1058375 1053486 827935 764955 764376 756489 SSD Vanilla Shaohua HughOld HughNew HughPC4 HughPC0 Seq Anon 24634 24198 24673 25107 21614 70018 Seq Shmem 24959 24932 25052 25703 22030 69678 Rand Anon 43014 26146 28075 25989 26935 25901 Rand Shmem 45349 45215 28249 24268 24138 24332 These tests are, of course, two extremes of a very simple case: under heavier mixed loads I've not yet observed any consistent improvement or degradation, and wider testing would be welcome. Shaohua Li: Test shows Vanilla is slightly better in sequential workload than Hugh's patch. I observed with Hugh's patch sometimes the readahead size is shrinked too fast (from 8 to 1 immediately) in sequential workload if there is no hit. And in such case, continuing doing readahead is good actually. I don't prepare a sophisticated algorithm for the sequential workload because so far we can't guarantee sequential accessed pages are swap out sequentially. So I slightly change Hugh's heuristic - don't shrink readahead size too fast. Here is my test result (unit second, 3 runs average): Vanilla Hugh New Seq 356 370 360 Random 4525 2447 2444 Attached graph is the swapin/swapout throughput I collected with 'vmstat 2'. The first part is running a random workload (till around 1200 of the x-axis) and the second part is running a sequential workload. swapin and swapout throughput are almost identical in steady state in both workloads. These are expected behavior. while in Vanilla, swapin is much bigger than swapout especially in random workload (because wrong readahead). Original patches by: Shaohua Li and Konstantin Khlebnikov. [fengguang.wu@intel.com: swapin_nr_pages() can be static] Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: Fengguang Wu <fengguang.wu@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Konstantin Khlebnikov <khlebnikov@openvz.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-02-06 20:04:21 +00:00
}
return folio;
}
/**
* filemap_get_incore_folio - Find and get a folio from the page or swap caches.
* @mapping: The address_space to search.
* @index: The page cache index.
*
* This differs from filemap_get_folio() in that it will also look for the
* folio in the swap cache.
*
* Return: The found folio or %NULL.
*/
struct folio *filemap_get_incore_folio(struct address_space *mapping,
pgoff_t index)
{
swp_entry_t swp;
struct swap_info_struct *si;
struct folio *folio = filemap_get_entry(mapping, index);
if (!folio)
return ERR_PTR(-ENOENT);
if (!xa_is_value(folio))
return folio;
if (!shmem_mapping(mapping))
return ERR_PTR(-ENOENT);
swp = radix_to_swp_entry(folio);
/* There might be swapin error entries in shmem mapping. */
if (non_swap_entry(swp))
return ERR_PTR(-ENOENT);
/* Prevent swapoff from happening to us */
si = get_swap_device(swp);
if (!si)
return ERR_PTR(-ENOENT);
index = swp_offset(swp);
folio = filemap_get_folio(swap_address_space(swp), index);
put_swap_device(si);
return folio;
}
struct folio *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
struct mempolicy *mpol, pgoff_t ilx, bool *new_page_allocated,
bool skip_if_exists)
{
mm, swap: fix race between swapoff and some swap operations When swapin is performed, after getting the swap entry information from the page table, system will swap in the swap entry, without any lock held to prevent the swap device from being swapoff. This may cause the race like below, CPU 1 CPU 2 ----- ----- do_swap_page swapin_readahead __read_swap_cache_async swapoff swapcache_prepare p->swap_map = NULL __swap_duplicate p->swap_map[?] /* !!! NULL pointer access */ Because swapoff is usually done when system shutdown only, the race may not hit many people in practice. But it is still a race need to be fixed. To fix the race, get_swap_device() is added to check whether the specified swap entry is valid in its swap device. If so, it will keep the swap entry valid via preventing the swap device from being swapoff, until put_swap_device() is called. Because swapoff() is very rare code path, to make the normal path runs as fast as possible, rcu_read_lock/unlock() and synchronize_rcu() instead of reference count is used to implement get/put_swap_device(). >From get_swap_device() to put_swap_device(), RCU reader side is locked, so synchronize_rcu() in swapoff() will wait until put_swap_device() is called. In addition to swap_map, cluster_info, etc. data structure in the struct swap_info_struct, the swap cache radix tree will be freed after swapoff, so this patch fixes the race between swap cache looking up and swapoff too. Races between some other swap cache usages and swapoff are fixed too via calling synchronize_rcu() between clearing PageSwapCache() and freeing swap cache data structure. Another possible method to fix this is to use preempt_off() + stop_machine() to prevent the swap device from being swapoff when its data structure is being accessed. The overhead in hot-path of both methods is similar. The advantages of RCU based method are, 1. stop_machine() may disturb the normal execution code path on other CPUs. 2. File cache uses RCU to protect its radix tree. If the similar mechanism is used for swap cache too, it is easier to share code between them. 3. RCU is used to protect swap cache in total_swapcache_pages() and exit_swap_address_space() already. The two mechanisms can be merged to simplify the logic. Link: http://lkml.kernel.org/r/20190522015423.14418-1-ying.huang@intel.com Fixes: 235b62176712 ("mm/swap: add cluster lock") Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Reviewed-by: Andrea Parri <andrea.parri@amarulasolutions.com> Not-nacked-by: Hugh Dickins <hughd@google.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Michal Hocko <mhocko@suse.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Tim Chen <tim.c.chen@linux.intel.com> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Jérôme Glisse <jglisse@redhat.com> Cc: Yang Shi <yang.shi@linux.alibaba.com> Cc: David Rientjes <rientjes@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Jan Kara <jack@suse.cz> Cc: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-07-12 03:55:33 +00:00
struct swap_info_struct *si;
struct folio *folio;
void *shadow = NULL;
mm: memcontrol: charge swapin pages on instantiation Right now, users that are otherwise memory controlled can easily escape their containment and allocate significant amounts of memory that they're not being charged for. That's because swap readahead pages are not being charged until somebody actually faults them into their page table. This can be exploited with MADV_WILLNEED, which triggers arbitrary readahead allocations without charging the pages. There are additional problems with the delayed charging of swap pages: 1. To implement refault/workingset detection for anonymous pages, we need to have a target LRU available at swapin time, but the LRU is not determinable until the page has been charged. 2. To implement per-cgroup LRU locking, we need page->mem_cgroup to be stable when the page is isolated from the LRU; otherwise, the locks change under us. But swapcache gets charged after it's already on the LRU, and even if we cannot isolate it ourselves (since charging is not exactly optional). The previous patch ensured we always maintain cgroup ownership records for swap pages. This patch moves the swapcache charging point from the fault handler to swapin time to fix all of the above problems. v2: simplify swapin error checking (Joonsoo) [hughd@google.com: fix livelock in __read_swap_cache_async()] Link: http://lkml.kernel.org/r/alpine.LSU.2.11.2005212246080.8458@eggly.anvils Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com> Cc: Hugh Dickins <hughd@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: "Kirill A. Shutemov" <kirill@shutemov.name> Cc: Michal Hocko <mhocko@suse.com> Cc: Roman Gushchin <guro@fb.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Balbir Singh <bsingharora@gmail.com> Cc: Rafael Aquini <aquini@redhat.com> Cc: Alex Shi <alex.shi@linux.alibaba.com> Link: http://lkml.kernel.org/r/20200508183105.225460-17-hannes@cmpxchg.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 23:02:17 +00:00
*new_page_allocated = false;
si = get_swap_device(entry);
if (!si)
return NULL;
mm: memcontrol: charge swapin pages on instantiation Right now, users that are otherwise memory controlled can easily escape their containment and allocate significant amounts of memory that they're not being charged for. That's because swap readahead pages are not being charged until somebody actually faults them into their page table. This can be exploited with MADV_WILLNEED, which triggers arbitrary readahead allocations without charging the pages. There are additional problems with the delayed charging of swap pages: 1. To implement refault/workingset detection for anonymous pages, we need to have a target LRU available at swapin time, but the LRU is not determinable until the page has been charged. 2. To implement per-cgroup LRU locking, we need page->mem_cgroup to be stable when the page is isolated from the LRU; otherwise, the locks change under us. But swapcache gets charged after it's already on the LRU, and even if we cannot isolate it ourselves (since charging is not exactly optional). The previous patch ensured we always maintain cgroup ownership records for swap pages. This patch moves the swapcache charging point from the fault handler to swapin time to fix all of the above problems. v2: simplify swapin error checking (Joonsoo) [hughd@google.com: fix livelock in __read_swap_cache_async()] Link: http://lkml.kernel.org/r/alpine.LSU.2.11.2005212246080.8458@eggly.anvils Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com> Cc: Hugh Dickins <hughd@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: "Kirill A. Shutemov" <kirill@shutemov.name> Cc: Michal Hocko <mhocko@suse.com> Cc: Roman Gushchin <guro@fb.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Balbir Singh <bsingharora@gmail.com> Cc: Rafael Aquini <aquini@redhat.com> Cc: Alex Shi <alex.shi@linux.alibaba.com> Link: http://lkml.kernel.org/r/20200508183105.225460-17-hannes@cmpxchg.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 23:02:17 +00:00
for (;;) {
int err;
/*
* First check the swap cache. Since this is normally
* called after swap_cache_get_folio() failed, re-calling
* that would confuse statistics.
*/
folio = filemap_get_folio(swap_address_space(entry),
swp_offset(entry));
if (!IS_ERR(folio))
goto got_folio;
mm/swap: skip readahead only when swap slot cache is enabled Because during swap off, a swap entry may have swap_map[] == SWAP_HAS_CACHE (for example, just allocated). If we return NULL in __read_swap_cache_async(), the swap off will abort. So when swap slot cache is disabled, (for swap off), we will wait for page to be put into swap cache in such race condition. This should not be a problem for swap slot cache, because swap slot cache should be drained after clearing swap_slot_cache_enabled. [ying.huang@intel.com: fix memory leak in __read_swap_cache_async()] Link: http://lkml.kernel.org/r/874lzt6znd.fsf@yhuang-dev.intel.com Link: http://lkml.kernel.org/r/5e2c5f6abe8e6eb0797408897b1bba80938e9b9d.1484082593.git.tim.c.chen@linux.intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com> Cc: Aaron Lu <aaron.lu@intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Christian Borntraeger <borntraeger@de.ibm.com> Cc: Dave Hansen <dave.hansen@intel.com> Cc: Hillf Danton <hillf.zj@alibaba-inc.com> Cc: Huang Ying <ying.huang@intel.com> Cc: Hugh Dickins <hughd@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Jonathan Corbet <corbet@lwn.net> escreveu: Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Vladimir Davydov <vdavydov.dev@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-02-22 23:45:46 +00:00
/*
* Just skip read ahead for unused swap slot.
* During swap_off when swap_slot_cache is disabled,
* we have to handle the race between putting
* swap entry in swap cache and marking swap slot
* as SWAP_HAS_CACHE. That's done in later part of code or
* else swap_off will be aborted if we return NULL.
*/
if (!swap_swapcount(si, entry) && swap_slot_cache_enabled)
goto fail_put_swap;
/*
* Get a new folio to read into from swap. Allocate it now,
mm: memcontrol: charge swapin pages on instantiation Right now, users that are otherwise memory controlled can easily escape their containment and allocate significant amounts of memory that they're not being charged for. That's because swap readahead pages are not being charged until somebody actually faults them into their page table. This can be exploited with MADV_WILLNEED, which triggers arbitrary readahead allocations without charging the pages. There are additional problems with the delayed charging of swap pages: 1. To implement refault/workingset detection for anonymous pages, we need to have a target LRU available at swapin time, but the LRU is not determinable until the page has been charged. 2. To implement per-cgroup LRU locking, we need page->mem_cgroup to be stable when the page is isolated from the LRU; otherwise, the locks change under us. But swapcache gets charged after it's already on the LRU, and even if we cannot isolate it ourselves (since charging is not exactly optional). The previous patch ensured we always maintain cgroup ownership records for swap pages. This patch moves the swapcache charging point from the fault handler to swapin time to fix all of the above problems. v2: simplify swapin error checking (Joonsoo) [hughd@google.com: fix livelock in __read_swap_cache_async()] Link: http://lkml.kernel.org/r/alpine.LSU.2.11.2005212246080.8458@eggly.anvils Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com> Cc: Hugh Dickins <hughd@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: "Kirill A. Shutemov" <kirill@shutemov.name> Cc: Michal Hocko <mhocko@suse.com> Cc: Roman Gushchin <guro@fb.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Balbir Singh <bsingharora@gmail.com> Cc: Rafael Aquini <aquini@redhat.com> Cc: Alex Shi <alex.shi@linux.alibaba.com> Link: http://lkml.kernel.org/r/20200508183105.225460-17-hannes@cmpxchg.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 23:02:17 +00:00
* before marking swap_map SWAP_HAS_CACHE, when -EEXIST will
* cause any racers to loop around until we add it to cache.
*/
mempolicy: alloc_pages_mpol() for NUMA policy without vma Shrink shmem's stack usage by eliminating the pseudo-vma from its folio allocation. alloc_pages_mpol(gfp, order, pol, ilx, nid) becomes the principal actor for passing mempolicy choice down to __alloc_pages(), rather than vma_alloc_folio(gfp, order, vma, addr, hugepage). vma_alloc_folio() and alloc_pages() remain, but as wrappers around alloc_pages_mpol(). alloc_pages_bulk_*() untouched, except to provide the additional args to policy_nodemask(), which subsumes policy_node(). Cleanup throughout, cutting out some unhelpful "helpers". It would all be much simpler without MPOL_INTERLEAVE, but that adds a dynamic to the constant mpol: complicated by v3.6 commit 09c231cb8bfd ("tmpfs: distribute interleave better across nodes"), which added ino bias to the interleave, hidden from mm/mempolicy.c until this commit. Hence "ilx" throughout, the "interleave index". Originally I thought it could be done just with nid, but that's wrong: the nodemask may come from the shared policy layer below a shmem vma, or it may come from the task layer above a shmem vma; and without the final nodemask then nodeid cannot be decided. And how ilx is applied depends also on page order. The interleave index is almost always irrelevant unless MPOL_INTERLEAVE: with one exception in alloc_pages_mpol(), where the NO_INTERLEAVE_INDEX passed down from vma-less alloc_pages() is also used as hint not to use THP-style hugepage allocation - to avoid the overhead of a hugepage arg (though I don't understand why we never just added a GFP bit for THP - if it actually needs a different allocation strategy from other pages of the same order). vma_alloc_folio() still carries its hugepage arg here, but it is not used, and should be removed when agreed. get_vma_policy() no longer allows a NULL vma: over time I believe we've eradicated all the places which used to need it e.g. swapoff and madvise used to pass NULL vma to read_swap_cache_async(), but now know the vma. [hughd@google.com: handle NULL mpol being passed to __read_swap_cache_async()] Link: https://lkml.kernel.org/r/ea419956-4751-0102-21f7-9c93cb957892@google.com Link: https://lkml.kernel.org/r/74e34633-6060-f5e3-aee-7040d43f2e93@google.com Link: https://lkml.kernel.org/r/1738368e-bac0-fd11-ed7f-b87142a939fe@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Hildenbrand <david@redhat.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Huang Ying <ying.huang@intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Nhat Pham <nphamcs@gmail.com> Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Tejun heo <tj@kernel.org> Cc: Vishal Moola (Oracle) <vishal.moola@gmail.com> Cc: Yang Shi <shy828301@gmail.com> Cc: Yosry Ahmed <yosryahmed@google.com> Cc: Domenico Cerasuolo <mimmocerasuolo@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-10-19 20:39:08 +00:00
folio = (struct folio *)alloc_pages_mpol(gfp_mask, 0,
mpol, ilx, numa_node_id());
if (!folio)
goto fail_put_swap;
/*
* Swap entry may have been freed since our caller observed it.
*/
err = swapcache_prepare(entry);
mm: memcontrol: charge swapin pages on instantiation Right now, users that are otherwise memory controlled can easily escape their containment and allocate significant amounts of memory that they're not being charged for. That's because swap readahead pages are not being charged until somebody actually faults them into their page table. This can be exploited with MADV_WILLNEED, which triggers arbitrary readahead allocations without charging the pages. There are additional problems with the delayed charging of swap pages: 1. To implement refault/workingset detection for anonymous pages, we need to have a target LRU available at swapin time, but the LRU is not determinable until the page has been charged. 2. To implement per-cgroup LRU locking, we need page->mem_cgroup to be stable when the page is isolated from the LRU; otherwise, the locks change under us. But swapcache gets charged after it's already on the LRU, and even if we cannot isolate it ourselves (since charging is not exactly optional). The previous patch ensured we always maintain cgroup ownership records for swap pages. This patch moves the swapcache charging point from the fault handler to swapin time to fix all of the above problems. v2: simplify swapin error checking (Joonsoo) [hughd@google.com: fix livelock in __read_swap_cache_async()] Link: http://lkml.kernel.org/r/alpine.LSU.2.11.2005212246080.8458@eggly.anvils Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com> Cc: Hugh Dickins <hughd@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: "Kirill A. Shutemov" <kirill@shutemov.name> Cc: Michal Hocko <mhocko@suse.com> Cc: Roman Gushchin <guro@fb.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Balbir Singh <bsingharora@gmail.com> Cc: Rafael Aquini <aquini@redhat.com> Cc: Alex Shi <alex.shi@linux.alibaba.com> Link: http://lkml.kernel.org/r/20200508183105.225460-17-hannes@cmpxchg.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 23:02:17 +00:00
if (!err)
break;
folio_put(folio);
mm: memcontrol: charge swapin pages on instantiation Right now, users that are otherwise memory controlled can easily escape their containment and allocate significant amounts of memory that they're not being charged for. That's because swap readahead pages are not being charged until somebody actually faults them into their page table. This can be exploited with MADV_WILLNEED, which triggers arbitrary readahead allocations without charging the pages. There are additional problems with the delayed charging of swap pages: 1. To implement refault/workingset detection for anonymous pages, we need to have a target LRU available at swapin time, but the LRU is not determinable until the page has been charged. 2. To implement per-cgroup LRU locking, we need page->mem_cgroup to be stable when the page is isolated from the LRU; otherwise, the locks change under us. But swapcache gets charged after it's already on the LRU, and even if we cannot isolate it ourselves (since charging is not exactly optional). The previous patch ensured we always maintain cgroup ownership records for swap pages. This patch moves the swapcache charging point from the fault handler to swapin time to fix all of the above problems. v2: simplify swapin error checking (Joonsoo) [hughd@google.com: fix livelock in __read_swap_cache_async()] Link: http://lkml.kernel.org/r/alpine.LSU.2.11.2005212246080.8458@eggly.anvils Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com> Cc: Hugh Dickins <hughd@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: "Kirill A. Shutemov" <kirill@shutemov.name> Cc: Michal Hocko <mhocko@suse.com> Cc: Roman Gushchin <guro@fb.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Balbir Singh <bsingharora@gmail.com> Cc: Rafael Aquini <aquini@redhat.com> Cc: Alex Shi <alex.shi@linux.alibaba.com> Link: http://lkml.kernel.org/r/20200508183105.225460-17-hannes@cmpxchg.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 23:02:17 +00:00
if (err != -EEXIST)
goto fail_put_swap;
mm: memcontrol: charge swapin pages on instantiation Right now, users that are otherwise memory controlled can easily escape their containment and allocate significant amounts of memory that they're not being charged for. That's because swap readahead pages are not being charged until somebody actually faults them into their page table. This can be exploited with MADV_WILLNEED, which triggers arbitrary readahead allocations without charging the pages. There are additional problems with the delayed charging of swap pages: 1. To implement refault/workingset detection for anonymous pages, we need to have a target LRU available at swapin time, but the LRU is not determinable until the page has been charged. 2. To implement per-cgroup LRU locking, we need page->mem_cgroup to be stable when the page is isolated from the LRU; otherwise, the locks change under us. But swapcache gets charged after it's already on the LRU, and even if we cannot isolate it ourselves (since charging is not exactly optional). The previous patch ensured we always maintain cgroup ownership records for swap pages. This patch moves the swapcache charging point from the fault handler to swapin time to fix all of the above problems. v2: simplify swapin error checking (Joonsoo) [hughd@google.com: fix livelock in __read_swap_cache_async()] Link: http://lkml.kernel.org/r/alpine.LSU.2.11.2005212246080.8458@eggly.anvils Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com> Cc: Hugh Dickins <hughd@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: "Kirill A. Shutemov" <kirill@shutemov.name> Cc: Michal Hocko <mhocko@suse.com> Cc: Roman Gushchin <guro@fb.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Balbir Singh <bsingharora@gmail.com> Cc: Rafael Aquini <aquini@redhat.com> Cc: Alex Shi <alex.shi@linux.alibaba.com> Link: http://lkml.kernel.org/r/20200508183105.225460-17-hannes@cmpxchg.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 23:02:17 +00:00
zswap: make shrinking memcg-aware Currently, we only have a single global LRU for zswap. This makes it impossible to perform worload-specific shrinking - an memcg cannot determine which pages in the pool it owns, and often ends up writing pages from other memcgs. This issue has been previously observed in practice and mitigated by simply disabling memcg-initiated shrinking: https://lore.kernel.org/all/20230530232435.3097106-1-nphamcs@gmail.com/T/#u This patch fully resolves the issue by replacing the global zswap LRU with memcg- and NUMA-specific LRUs, and modify the reclaim logic: a) When a store attempt hits an memcg limit, it now triggers a synchronous reclaim attempt that, if successful, allows the new hotter page to be accepted by zswap. b) If the store attempt instead hits the global zswap limit, it will trigger an asynchronous reclaim attempt, in which an memcg is selected for reclaim in a round-robin-like fashion. [nphamcs@gmail.com: use correct function for the onlineness check, use mem_cgroup_iter_break()] Link: https://lkml.kernel.org/r/20231205195419.2563217-1-nphamcs@gmail.com [nphamcs@gmail.com: drop the pool's reference at the end of the writeback step] Link: https://lkml.kernel.org/r/20231206030627.4155634-1-nphamcs@gmail.com Link: https://lkml.kernel.org/r/20231130194023.4102148-4-nphamcs@gmail.com Signed-off-by: Domenico Cerasuolo <cerasuolodomenico@gmail.com> Co-developed-by: Nhat Pham <nphamcs@gmail.com> Signed-off-by: Nhat Pham <nphamcs@gmail.com> Tested-by: Bagas Sanjaya <bagasdotme@gmail.com> Cc: Chris Li <chrisl@kernel.org> Cc: Dan Streetman <ddstreet@ieee.org> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Muchun Song <muchun.song@linux.dev> Cc: Roman Gushchin <roman.gushchin@linux.dev> Cc: Seth Jennings <sjenning@redhat.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Shuah Khan <shuah@kernel.org> Cc: Vitaly Wool <vitaly.wool@konsulko.com> Cc: Yosry Ahmed <yosryahmed@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-11-30 19:40:20 +00:00
/*
* Protect against a recursive call to __read_swap_cache_async()
* on the same entry waiting forever here because SWAP_HAS_CACHE
* is set but the folio is not the swap cache yet. This can
* happen today if mem_cgroup_swapin_charge_folio() below
* triggers reclaim through zswap, which may call
* __read_swap_cache_async() in the writeback path.
*/
if (skip_if_exists)
goto fail_put_swap;
/*
mm: memcontrol: charge swapin pages on instantiation Right now, users that are otherwise memory controlled can easily escape their containment and allocate significant amounts of memory that they're not being charged for. That's because swap readahead pages are not being charged until somebody actually faults them into their page table. This can be exploited with MADV_WILLNEED, which triggers arbitrary readahead allocations without charging the pages. There are additional problems with the delayed charging of swap pages: 1. To implement refault/workingset detection for anonymous pages, we need to have a target LRU available at swapin time, but the LRU is not determinable until the page has been charged. 2. To implement per-cgroup LRU locking, we need page->mem_cgroup to be stable when the page is isolated from the LRU; otherwise, the locks change under us. But swapcache gets charged after it's already on the LRU, and even if we cannot isolate it ourselves (since charging is not exactly optional). The previous patch ensured we always maintain cgroup ownership records for swap pages. This patch moves the swapcache charging point from the fault handler to swapin time to fix all of the above problems. v2: simplify swapin error checking (Joonsoo) [hughd@google.com: fix livelock in __read_swap_cache_async()] Link: http://lkml.kernel.org/r/alpine.LSU.2.11.2005212246080.8458@eggly.anvils Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com> Cc: Hugh Dickins <hughd@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: "Kirill A. Shutemov" <kirill@shutemov.name> Cc: Michal Hocko <mhocko@suse.com> Cc: Roman Gushchin <guro@fb.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Balbir Singh <bsingharora@gmail.com> Cc: Rafael Aquini <aquini@redhat.com> Cc: Alex Shi <alex.shi@linux.alibaba.com> Link: http://lkml.kernel.org/r/20200508183105.225460-17-hannes@cmpxchg.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 23:02:17 +00:00
* We might race against __delete_from_swap_cache(), and
* stumble across a swap_map entry whose SWAP_HAS_CACHE
* has not yet been cleared. Or race against another
* __read_swap_cache_async(), which has set SWAP_HAS_CACHE
* in swap_map, but not yet added its folio to swap cache.
*/
mm: swap: get rid of livelock in swapin readahead In our testing, a livelock task was found. Through sysrq printing, same stack was found every time, as follows: __swap_duplicate+0x58/0x1a0 swapcache_prepare+0x24/0x30 __read_swap_cache_async+0xac/0x220 read_swap_cache_async+0x58/0xa0 swapin_readahead+0x24c/0x628 do_swap_page+0x374/0x8a0 __handle_mm_fault+0x598/0xd60 handle_mm_fault+0x114/0x200 do_page_fault+0x148/0x4d0 do_translation_fault+0xb0/0xd4 do_mem_abort+0x50/0xb0 The reason for the livelock is that swapcache_prepare() always returns EEXIST, indicating that SWAP_HAS_CACHE has not been cleared, so that it cannot jump out of the loop. We suspect that the task that clears the SWAP_HAS_CACHE flag never gets a chance to run. We try to lower the priority of the task stuck in a livelock so that the task that clears the SWAP_HAS_CACHE flag will run. The results show that the system returns to normal after the priority is lowered. In our testing, multiple real-time tasks are bound to the same core, and the task in the livelock is the highest priority task of the core, so the livelocked task cannot be preempted. Although cond_resched() is used by __read_swap_cache_async, it is an empty function in the preemptive system and cannot achieve the purpose of releasing the CPU. A high-priority task cannot release the CPU unless preempted by a higher-priority task. But when this task is already the highest priority task on this core, other tasks will not be able to be scheduled. So we think we should replace cond_resched() with schedule_timeout_uninterruptible(1), schedule_timeout_interruptible will call set_current_state first to set the task state, so the task will be removed from the running queue, so as to achieve the purpose of giving up the CPU and prevent it from running in kernel mode for too long. (akpm: ugly hack becomes uglier. But it fixes the issue in a backportable-to-stable fashion while we hopefully work on something better) Link: https://lkml.kernel.org/r/20220221111749.1928222-1-cgel.zte@gmail.com Signed-off-by: Guo Ziliang <guo.ziliang@zte.com.cn> Reported-by: Zeal Robot <zealci@zte.com.cn> Reviewed-by: Ran Xiaokai <ran.xiaokai@zte.com.cn> Reviewed-by: Jiang Xuexin <jiang.xuexin@zte.com.cn> Reviewed-by: Yang Yang <yang.yang29@zte.com.cn> Acked-by: Hugh Dickins <hughd@google.com> Cc: Naoya Horiguchi <naoya.horiguchi@nec.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Roger Quadros <rogerq@kernel.org> Cc: Ziliang Guo <guo.ziliang@zte.com.cn> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2022-03-16 23:15:03 +00:00
schedule_timeout_uninterruptible(1);
mm: memcontrol: charge swapin pages on instantiation Right now, users that are otherwise memory controlled can easily escape their containment and allocate significant amounts of memory that they're not being charged for. That's because swap readahead pages are not being charged until somebody actually faults them into their page table. This can be exploited with MADV_WILLNEED, which triggers arbitrary readahead allocations without charging the pages. There are additional problems with the delayed charging of swap pages: 1. To implement refault/workingset detection for anonymous pages, we need to have a target LRU available at swapin time, but the LRU is not determinable until the page has been charged. 2. To implement per-cgroup LRU locking, we need page->mem_cgroup to be stable when the page is isolated from the LRU; otherwise, the locks change under us. But swapcache gets charged after it's already on the LRU, and even if we cannot isolate it ourselves (since charging is not exactly optional). The previous patch ensured we always maintain cgroup ownership records for swap pages. This patch moves the swapcache charging point from the fault handler to swapin time to fix all of the above problems. v2: simplify swapin error checking (Joonsoo) [hughd@google.com: fix livelock in __read_swap_cache_async()] Link: http://lkml.kernel.org/r/alpine.LSU.2.11.2005212246080.8458@eggly.anvils Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com> Cc: Hugh Dickins <hughd@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: "Kirill A. Shutemov" <kirill@shutemov.name> Cc: Michal Hocko <mhocko@suse.com> Cc: Roman Gushchin <guro@fb.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Balbir Singh <bsingharora@gmail.com> Cc: Rafael Aquini <aquini@redhat.com> Cc: Alex Shi <alex.shi@linux.alibaba.com> Link: http://lkml.kernel.org/r/20200508183105.225460-17-hannes@cmpxchg.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 23:02:17 +00:00
}
/*
* The swap entry is ours to swap in. Prepare the new folio.
mm: memcontrol: charge swapin pages on instantiation Right now, users that are otherwise memory controlled can easily escape their containment and allocate significant amounts of memory that they're not being charged for. That's because swap readahead pages are not being charged until somebody actually faults them into their page table. This can be exploited with MADV_WILLNEED, which triggers arbitrary readahead allocations without charging the pages. There are additional problems with the delayed charging of swap pages: 1. To implement refault/workingset detection for anonymous pages, we need to have a target LRU available at swapin time, but the LRU is not determinable until the page has been charged. 2. To implement per-cgroup LRU locking, we need page->mem_cgroup to be stable when the page is isolated from the LRU; otherwise, the locks change under us. But swapcache gets charged after it's already on the LRU, and even if we cannot isolate it ourselves (since charging is not exactly optional). The previous patch ensured we always maintain cgroup ownership records for swap pages. This patch moves the swapcache charging point from the fault handler to swapin time to fix all of the above problems. v2: simplify swapin error checking (Joonsoo) [hughd@google.com: fix livelock in __read_swap_cache_async()] Link: http://lkml.kernel.org/r/alpine.LSU.2.11.2005212246080.8458@eggly.anvils Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com> Cc: Hugh Dickins <hughd@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: "Kirill A. Shutemov" <kirill@shutemov.name> Cc: Michal Hocko <mhocko@suse.com> Cc: Roman Gushchin <guro@fb.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Balbir Singh <bsingharora@gmail.com> Cc: Rafael Aquini <aquini@redhat.com> Cc: Alex Shi <alex.shi@linux.alibaba.com> Link: http://lkml.kernel.org/r/20200508183105.225460-17-hannes@cmpxchg.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 23:02:17 +00:00
*/
__folio_set_locked(folio);
__folio_set_swapbacked(folio);
mm: memcontrol: charge swapin pages on instantiation Right now, users that are otherwise memory controlled can easily escape their containment and allocate significant amounts of memory that they're not being charged for. That's because swap readahead pages are not being charged until somebody actually faults them into their page table. This can be exploited with MADV_WILLNEED, which triggers arbitrary readahead allocations without charging the pages. There are additional problems with the delayed charging of swap pages: 1. To implement refault/workingset detection for anonymous pages, we need to have a target LRU available at swapin time, but the LRU is not determinable until the page has been charged. 2. To implement per-cgroup LRU locking, we need page->mem_cgroup to be stable when the page is isolated from the LRU; otherwise, the locks change under us. But swapcache gets charged after it's already on the LRU, and even if we cannot isolate it ourselves (since charging is not exactly optional). The previous patch ensured we always maintain cgroup ownership records for swap pages. This patch moves the swapcache charging point from the fault handler to swapin time to fix all of the above problems. v2: simplify swapin error checking (Joonsoo) [hughd@google.com: fix livelock in __read_swap_cache_async()] Link: http://lkml.kernel.org/r/alpine.LSU.2.11.2005212246080.8458@eggly.anvils Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com> Cc: Hugh Dickins <hughd@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: "Kirill A. Shutemov" <kirill@shutemov.name> Cc: Michal Hocko <mhocko@suse.com> Cc: Roman Gushchin <guro@fb.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Balbir Singh <bsingharora@gmail.com> Cc: Rafael Aquini <aquini@redhat.com> Cc: Alex Shi <alex.shi@linux.alibaba.com> Link: http://lkml.kernel.org/r/20200508183105.225460-17-hannes@cmpxchg.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 23:02:17 +00:00
if (mem_cgroup_swapin_charge_folio(folio, NULL, gfp_mask, entry))
mm: memcontrol: charge swapin pages on instantiation Right now, users that are otherwise memory controlled can easily escape their containment and allocate significant amounts of memory that they're not being charged for. That's because swap readahead pages are not being charged until somebody actually faults them into their page table. This can be exploited with MADV_WILLNEED, which triggers arbitrary readahead allocations without charging the pages. There are additional problems with the delayed charging of swap pages: 1. To implement refault/workingset detection for anonymous pages, we need to have a target LRU available at swapin time, but the LRU is not determinable until the page has been charged. 2. To implement per-cgroup LRU locking, we need page->mem_cgroup to be stable when the page is isolated from the LRU; otherwise, the locks change under us. But swapcache gets charged after it's already on the LRU, and even if we cannot isolate it ourselves (since charging is not exactly optional). The previous patch ensured we always maintain cgroup ownership records for swap pages. This patch moves the swapcache charging point from the fault handler to swapin time to fix all of the above problems. v2: simplify swapin error checking (Joonsoo) [hughd@google.com: fix livelock in __read_swap_cache_async()] Link: http://lkml.kernel.org/r/alpine.LSU.2.11.2005212246080.8458@eggly.anvils Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com> Cc: Hugh Dickins <hughd@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: "Kirill A. Shutemov" <kirill@shutemov.name> Cc: Michal Hocko <mhocko@suse.com> Cc: Roman Gushchin <guro@fb.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Balbir Singh <bsingharora@gmail.com> Cc: Rafael Aquini <aquini@redhat.com> Cc: Alex Shi <alex.shi@linux.alibaba.com> Link: http://lkml.kernel.org/r/20200508183105.225460-17-hannes@cmpxchg.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 23:02:17 +00:00
goto fail_unlock;
memcg: charge before adding to swapcache on swapin Currently the kernel adds the page, allocated for swapin, to the swapcache before charging the page. This is fine but now we want a per-memcg swapcache stat which is essential for folks who wants to transparently migrate from cgroup v1's memsw to cgroup v2's memory and swap counters. In addition charging a page before exposing it to other parts of the kernel is a step in the right direction. To correctly maintain the per-memcg swapcache stat, this patch has adopted to charge the page before adding it to swapcache. One challenge in this option is the failure case of add_to_swap_cache() on which we need to undo the mem_cgroup_charge(). Specifically undoing mem_cgroup_uncharge_swap() is not simple. To resolve the issue, this patch decouples the charging for swapin pages from mem_cgroup_charge(). Two new functions are introduced, mem_cgroup_swapin_charge_page() for just charging the swapin page and mem_cgroup_swapin_uncharge_swap() for uncharging the swap slot once the page has been successfully added to the swapcache. [shakeelb@google.com: set page->private before calling swap_readpage] Link: https://lkml.kernel.org/r/20210318015959.2986837-1-shakeelb@google.com Link: https://lkml.kernel.org/r/20210305212639.775498-1-shakeelb@google.com Signed-off-by: Shakeel Butt <shakeelb@google.com> Acked-by: Roman Gushchin <guro@fb.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by: Hugh Dickins <hughd@google.com> Tested-by: Heiko Carstens <hca@linux.ibm.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Stephen Rothwell <sfr@canb.auug.org.au> Cc: Christian Borntraeger <borntraeger@de.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-04-30 05:56:36 +00:00
/* May fail (-ENOMEM) if XArray node allocation failed. */
if (add_to_swap_cache(folio, entry, gfp_mask & GFP_RECLAIM_MASK, &shadow))
mm: memcontrol: charge swapin pages on instantiation Right now, users that are otherwise memory controlled can easily escape their containment and allocate significant amounts of memory that they're not being charged for. That's because swap readahead pages are not being charged until somebody actually faults them into their page table. This can be exploited with MADV_WILLNEED, which triggers arbitrary readahead allocations without charging the pages. There are additional problems with the delayed charging of swap pages: 1. To implement refault/workingset detection for anonymous pages, we need to have a target LRU available at swapin time, but the LRU is not determinable until the page has been charged. 2. To implement per-cgroup LRU locking, we need page->mem_cgroup to be stable when the page is isolated from the LRU; otherwise, the locks change under us. But swapcache gets charged after it's already on the LRU, and even if we cannot isolate it ourselves (since charging is not exactly optional). The previous patch ensured we always maintain cgroup ownership records for swap pages. This patch moves the swapcache charging point from the fault handler to swapin time to fix all of the above problems. v2: simplify swapin error checking (Joonsoo) [hughd@google.com: fix livelock in __read_swap_cache_async()] Link: http://lkml.kernel.org/r/alpine.LSU.2.11.2005212246080.8458@eggly.anvils Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com> Cc: Hugh Dickins <hughd@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: "Kirill A. Shutemov" <kirill@shutemov.name> Cc: Michal Hocko <mhocko@suse.com> Cc: Roman Gushchin <guro@fb.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Balbir Singh <bsingharora@gmail.com> Cc: Rafael Aquini <aquini@redhat.com> Cc: Alex Shi <alex.shi@linux.alibaba.com> Link: http://lkml.kernel.org/r/20200508183105.225460-17-hannes@cmpxchg.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 23:02:17 +00:00
goto fail_unlock;
memcg: charge before adding to swapcache on swapin Currently the kernel adds the page, allocated for swapin, to the swapcache before charging the page. This is fine but now we want a per-memcg swapcache stat which is essential for folks who wants to transparently migrate from cgroup v1's memsw to cgroup v2's memory and swap counters. In addition charging a page before exposing it to other parts of the kernel is a step in the right direction. To correctly maintain the per-memcg swapcache stat, this patch has adopted to charge the page before adding it to swapcache. One challenge in this option is the failure case of add_to_swap_cache() on which we need to undo the mem_cgroup_charge(). Specifically undoing mem_cgroup_uncharge_swap() is not simple. To resolve the issue, this patch decouples the charging for swapin pages from mem_cgroup_charge(). Two new functions are introduced, mem_cgroup_swapin_charge_page() for just charging the swapin page and mem_cgroup_swapin_uncharge_swap() for uncharging the swap slot once the page has been successfully added to the swapcache. [shakeelb@google.com: set page->private before calling swap_readpage] Link: https://lkml.kernel.org/r/20210318015959.2986837-1-shakeelb@google.com Link: https://lkml.kernel.org/r/20210305212639.775498-1-shakeelb@google.com Signed-off-by: Shakeel Butt <shakeelb@google.com> Acked-by: Roman Gushchin <guro@fb.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by: Hugh Dickins <hughd@google.com> Tested-by: Heiko Carstens <hca@linux.ibm.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Stephen Rothwell <sfr@canb.auug.org.au> Cc: Christian Borntraeger <borntraeger@de.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-04-30 05:56:36 +00:00
mem_cgroup_swapin_uncharge_swap(entry);
mm: memcontrol: charge swapin pages on instantiation Right now, users that are otherwise memory controlled can easily escape their containment and allocate significant amounts of memory that they're not being charged for. That's because swap readahead pages are not being charged until somebody actually faults them into their page table. This can be exploited with MADV_WILLNEED, which triggers arbitrary readahead allocations without charging the pages. There are additional problems with the delayed charging of swap pages: 1. To implement refault/workingset detection for anonymous pages, we need to have a target LRU available at swapin time, but the LRU is not determinable until the page has been charged. 2. To implement per-cgroup LRU locking, we need page->mem_cgroup to be stable when the page is isolated from the LRU; otherwise, the locks change under us. But swapcache gets charged after it's already on the LRU, and even if we cannot isolate it ourselves (since charging is not exactly optional). The previous patch ensured we always maintain cgroup ownership records for swap pages. This patch moves the swapcache charging point from the fault handler to swapin time to fix all of the above problems. v2: simplify swapin error checking (Joonsoo) [hughd@google.com: fix livelock in __read_swap_cache_async()] Link: http://lkml.kernel.org/r/alpine.LSU.2.11.2005212246080.8458@eggly.anvils Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com> Cc: Hugh Dickins <hughd@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: "Kirill A. Shutemov" <kirill@shutemov.name> Cc: Michal Hocko <mhocko@suse.com> Cc: Roman Gushchin <guro@fb.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Balbir Singh <bsingharora@gmail.com> Cc: Rafael Aquini <aquini@redhat.com> Cc: Alex Shi <alex.shi@linux.alibaba.com> Link: http://lkml.kernel.org/r/20200508183105.225460-17-hannes@cmpxchg.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 23:02:17 +00:00
if (shadow)
workingset_refault(folio, shadow);
mm: balance LRU lists based on relative thrashing Since the LRUs were split into anon and file lists, the VM has been balancing between page cache and anonymous pages based on per-list ratios of scanned vs. rotated pages. In most cases that tips page reclaim towards the list that is easier to reclaim and has the fewest actively used pages, but there are a few problems with it: 1. Refaults and LRU rotations are weighted the same way, even though one costs IO and the other costs a bit of CPU. 2. The less we scan an LRU list based on already observed rotations, the more we increase the sampling interval for new references, and rotations become even more likely on that list. This can enter a death spiral in which we stop looking at one list completely until the other one is all but annihilated by page reclaim. Since commit a528910e12ec ("mm: thrash detection-based file cache sizing") we have refault detection for the page cache. Along with swapin events, they are good indicators of when the file or anon list, respectively, is too small for its workingset and needs to grow. For example, if the page cache is thrashing, the cache pages need more time in memory, while there may be colder pages on the anonymous list. Likewise, if swapped pages are faulting back in, it indicates that we reclaim anonymous pages too aggressively and should back off. Replace LRU rotations with refaults and swapins as the basis for relative reclaim cost of the two LRUs. This will have the VM target list balances that incur the least amount of IO on aggregate. Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Michal Hocko <mhocko@suse.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@surriel.com> Link: http://lkml.kernel.org/r/20200520232525.798933-12-hannes@cmpxchg.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 23:03:03 +00:00
/* Caller will initiate read into locked folio */
folio_add_lru(folio);
mm: memcontrol: charge swapin pages on instantiation Right now, users that are otherwise memory controlled can easily escape their containment and allocate significant amounts of memory that they're not being charged for. That's because swap readahead pages are not being charged until somebody actually faults them into their page table. This can be exploited with MADV_WILLNEED, which triggers arbitrary readahead allocations without charging the pages. There are additional problems with the delayed charging of swap pages: 1. To implement refault/workingset detection for anonymous pages, we need to have a target LRU available at swapin time, but the LRU is not determinable until the page has been charged. 2. To implement per-cgroup LRU locking, we need page->mem_cgroup to be stable when the page is isolated from the LRU; otherwise, the locks change under us. But swapcache gets charged after it's already on the LRU, and even if we cannot isolate it ourselves (since charging is not exactly optional). The previous patch ensured we always maintain cgroup ownership records for swap pages. This patch moves the swapcache charging point from the fault handler to swapin time to fix all of the above problems. v2: simplify swapin error checking (Joonsoo) [hughd@google.com: fix livelock in __read_swap_cache_async()] Link: http://lkml.kernel.org/r/alpine.LSU.2.11.2005212246080.8458@eggly.anvils Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com> Cc: Hugh Dickins <hughd@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: "Kirill A. Shutemov" <kirill@shutemov.name> Cc: Michal Hocko <mhocko@suse.com> Cc: Roman Gushchin <guro@fb.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Balbir Singh <bsingharora@gmail.com> Cc: Rafael Aquini <aquini@redhat.com> Cc: Alex Shi <alex.shi@linux.alibaba.com> Link: http://lkml.kernel.org/r/20200508183105.225460-17-hannes@cmpxchg.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 23:02:17 +00:00
*new_page_allocated = true;
got_folio:
put_swap_device(si);
return folio;
mm: memcontrol: charge swapin pages on instantiation Right now, users that are otherwise memory controlled can easily escape their containment and allocate significant amounts of memory that they're not being charged for. That's because swap readahead pages are not being charged until somebody actually faults them into their page table. This can be exploited with MADV_WILLNEED, which triggers arbitrary readahead allocations without charging the pages. There are additional problems with the delayed charging of swap pages: 1. To implement refault/workingset detection for anonymous pages, we need to have a target LRU available at swapin time, but the LRU is not determinable until the page has been charged. 2. To implement per-cgroup LRU locking, we need page->mem_cgroup to be stable when the page is isolated from the LRU; otherwise, the locks change under us. But swapcache gets charged after it's already on the LRU, and even if we cannot isolate it ourselves (since charging is not exactly optional). The previous patch ensured we always maintain cgroup ownership records for swap pages. This patch moves the swapcache charging point from the fault handler to swapin time to fix all of the above problems. v2: simplify swapin error checking (Joonsoo) [hughd@google.com: fix livelock in __read_swap_cache_async()] Link: http://lkml.kernel.org/r/alpine.LSU.2.11.2005212246080.8458@eggly.anvils Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com> Cc: Hugh Dickins <hughd@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: "Kirill A. Shutemov" <kirill@shutemov.name> Cc: Michal Hocko <mhocko@suse.com> Cc: Roman Gushchin <guro@fb.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Balbir Singh <bsingharora@gmail.com> Cc: Rafael Aquini <aquini@redhat.com> Cc: Alex Shi <alex.shi@linux.alibaba.com> Link: http://lkml.kernel.org/r/20200508183105.225460-17-hannes@cmpxchg.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 23:02:17 +00:00
fail_unlock:
put_swap_folio(folio, entry);
folio_unlock(folio);
folio_put(folio);
fail_put_swap:
put_swap_device(si);
mm: memcontrol: charge swapin pages on instantiation Right now, users that are otherwise memory controlled can easily escape their containment and allocate significant amounts of memory that they're not being charged for. That's because swap readahead pages are not being charged until somebody actually faults them into their page table. This can be exploited with MADV_WILLNEED, which triggers arbitrary readahead allocations without charging the pages. There are additional problems with the delayed charging of swap pages: 1. To implement refault/workingset detection for anonymous pages, we need to have a target LRU available at swapin time, but the LRU is not determinable until the page has been charged. 2. To implement per-cgroup LRU locking, we need page->mem_cgroup to be stable when the page is isolated from the LRU; otherwise, the locks change under us. But swapcache gets charged after it's already on the LRU, and even if we cannot isolate it ourselves (since charging is not exactly optional). The previous patch ensured we always maintain cgroup ownership records for swap pages. This patch moves the swapcache charging point from the fault handler to swapin time to fix all of the above problems. v2: simplify swapin error checking (Joonsoo) [hughd@google.com: fix livelock in __read_swap_cache_async()] Link: http://lkml.kernel.org/r/alpine.LSU.2.11.2005212246080.8458@eggly.anvils Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Reviewed-by: Alex Shi <alex.shi@linux.alibaba.com> Cc: Hugh Dickins <hughd@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: "Kirill A. Shutemov" <kirill@shutemov.name> Cc: Michal Hocko <mhocko@suse.com> Cc: Roman Gushchin <guro@fb.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Balbir Singh <bsingharora@gmail.com> Cc: Rafael Aquini <aquini@redhat.com> Cc: Alex Shi <alex.shi@linux.alibaba.com> Link: http://lkml.kernel.org/r/20200508183105.225460-17-hannes@cmpxchg.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-03 23:02:17 +00:00
return NULL;
}
/*
* Locate a page of swap in physical memory, reserving swap cache space
* and reading the disk if it is not already cached.
* A failure return means that either the page allocation failed or that
* the swap entry is no longer in use.
*
* get/put_swap_device() aren't needed to call this function, because
* __read_swap_cache_async() call them and swap_read_folio() holds the
* swap cache folio lock.
*/
struct folio *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
struct vm_area_struct *vma, unsigned long addr,
struct swap_iocb **plug)
{
mempolicy: alloc_pages_mpol() for NUMA policy without vma Shrink shmem's stack usage by eliminating the pseudo-vma from its folio allocation. alloc_pages_mpol(gfp, order, pol, ilx, nid) becomes the principal actor for passing mempolicy choice down to __alloc_pages(), rather than vma_alloc_folio(gfp, order, vma, addr, hugepage). vma_alloc_folio() and alloc_pages() remain, but as wrappers around alloc_pages_mpol(). alloc_pages_bulk_*() untouched, except to provide the additional args to policy_nodemask(), which subsumes policy_node(). Cleanup throughout, cutting out some unhelpful "helpers". It would all be much simpler without MPOL_INTERLEAVE, but that adds a dynamic to the constant mpol: complicated by v3.6 commit 09c231cb8bfd ("tmpfs: distribute interleave better across nodes"), which added ino bias to the interleave, hidden from mm/mempolicy.c until this commit. Hence "ilx" throughout, the "interleave index". Originally I thought it could be done just with nid, but that's wrong: the nodemask may come from the shared policy layer below a shmem vma, or it may come from the task layer above a shmem vma; and without the final nodemask then nodeid cannot be decided. And how ilx is applied depends also on page order. The interleave index is almost always irrelevant unless MPOL_INTERLEAVE: with one exception in alloc_pages_mpol(), where the NO_INTERLEAVE_INDEX passed down from vma-less alloc_pages() is also used as hint not to use THP-style hugepage allocation - to avoid the overhead of a hugepage arg (though I don't understand why we never just added a GFP bit for THP - if it actually needs a different allocation strategy from other pages of the same order). vma_alloc_folio() still carries its hugepage arg here, but it is not used, and should be removed when agreed. get_vma_policy() no longer allows a NULL vma: over time I believe we've eradicated all the places which used to need it e.g. swapoff and madvise used to pass NULL vma to read_swap_cache_async(), but now know the vma. [hughd@google.com: handle NULL mpol being passed to __read_swap_cache_async()] Link: https://lkml.kernel.org/r/ea419956-4751-0102-21f7-9c93cb957892@google.com Link: https://lkml.kernel.org/r/74e34633-6060-f5e3-aee-7040d43f2e93@google.com Link: https://lkml.kernel.org/r/1738368e-bac0-fd11-ed7f-b87142a939fe@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Hildenbrand <david@redhat.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Huang Ying <ying.huang@intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Nhat Pham <nphamcs@gmail.com> Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Tejun heo <tj@kernel.org> Cc: Vishal Moola (Oracle) <vishal.moola@gmail.com> Cc: Yang Shi <shy828301@gmail.com> Cc: Yosry Ahmed <yosryahmed@google.com> Cc: Domenico Cerasuolo <mimmocerasuolo@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-10-19 20:39:08 +00:00
bool page_allocated;
struct mempolicy *mpol;
pgoff_t ilx;
struct folio *folio;
mempolicy: alloc_pages_mpol() for NUMA policy without vma Shrink shmem's stack usage by eliminating the pseudo-vma from its folio allocation. alloc_pages_mpol(gfp, order, pol, ilx, nid) becomes the principal actor for passing mempolicy choice down to __alloc_pages(), rather than vma_alloc_folio(gfp, order, vma, addr, hugepage). vma_alloc_folio() and alloc_pages() remain, but as wrappers around alloc_pages_mpol(). alloc_pages_bulk_*() untouched, except to provide the additional args to policy_nodemask(), which subsumes policy_node(). Cleanup throughout, cutting out some unhelpful "helpers". It would all be much simpler without MPOL_INTERLEAVE, but that adds a dynamic to the constant mpol: complicated by v3.6 commit 09c231cb8bfd ("tmpfs: distribute interleave better across nodes"), which added ino bias to the interleave, hidden from mm/mempolicy.c until this commit. Hence "ilx" throughout, the "interleave index". Originally I thought it could be done just with nid, but that's wrong: the nodemask may come from the shared policy layer below a shmem vma, or it may come from the task layer above a shmem vma; and without the final nodemask then nodeid cannot be decided. And how ilx is applied depends also on page order. The interleave index is almost always irrelevant unless MPOL_INTERLEAVE: with one exception in alloc_pages_mpol(), where the NO_INTERLEAVE_INDEX passed down from vma-less alloc_pages() is also used as hint not to use THP-style hugepage allocation - to avoid the overhead of a hugepage arg (though I don't understand why we never just added a GFP bit for THP - if it actually needs a different allocation strategy from other pages of the same order). vma_alloc_folio() still carries its hugepage arg here, but it is not used, and should be removed when agreed. get_vma_policy() no longer allows a NULL vma: over time I believe we've eradicated all the places which used to need it e.g. swapoff and madvise used to pass NULL vma to read_swap_cache_async(), but now know the vma. [hughd@google.com: handle NULL mpol being passed to __read_swap_cache_async()] Link: https://lkml.kernel.org/r/ea419956-4751-0102-21f7-9c93cb957892@google.com Link: https://lkml.kernel.org/r/74e34633-6060-f5e3-aee-7040d43f2e93@google.com Link: https://lkml.kernel.org/r/1738368e-bac0-fd11-ed7f-b87142a939fe@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Hildenbrand <david@redhat.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Huang Ying <ying.huang@intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Nhat Pham <nphamcs@gmail.com> Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Tejun heo <tj@kernel.org> Cc: Vishal Moola (Oracle) <vishal.moola@gmail.com> Cc: Yang Shi <shy828301@gmail.com> Cc: Yosry Ahmed <yosryahmed@google.com> Cc: Domenico Cerasuolo <mimmocerasuolo@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-10-19 20:39:08 +00:00
mpol = get_vma_policy(vma, addr, 0, &ilx);
folio = __read_swap_cache_async(entry, gfp_mask, mpol, ilx,
zswap: make shrinking memcg-aware Currently, we only have a single global LRU for zswap. This makes it impossible to perform worload-specific shrinking - an memcg cannot determine which pages in the pool it owns, and often ends up writing pages from other memcgs. This issue has been previously observed in practice and mitigated by simply disabling memcg-initiated shrinking: https://lore.kernel.org/all/20230530232435.3097106-1-nphamcs@gmail.com/T/#u This patch fully resolves the issue by replacing the global zswap LRU with memcg- and NUMA-specific LRUs, and modify the reclaim logic: a) When a store attempt hits an memcg limit, it now triggers a synchronous reclaim attempt that, if successful, allows the new hotter page to be accepted by zswap. b) If the store attempt instead hits the global zswap limit, it will trigger an asynchronous reclaim attempt, in which an memcg is selected for reclaim in a round-robin-like fashion. [nphamcs@gmail.com: use correct function for the onlineness check, use mem_cgroup_iter_break()] Link: https://lkml.kernel.org/r/20231205195419.2563217-1-nphamcs@gmail.com [nphamcs@gmail.com: drop the pool's reference at the end of the writeback step] Link: https://lkml.kernel.org/r/20231206030627.4155634-1-nphamcs@gmail.com Link: https://lkml.kernel.org/r/20231130194023.4102148-4-nphamcs@gmail.com Signed-off-by: Domenico Cerasuolo <cerasuolodomenico@gmail.com> Co-developed-by: Nhat Pham <nphamcs@gmail.com> Signed-off-by: Nhat Pham <nphamcs@gmail.com> Tested-by: Bagas Sanjaya <bagasdotme@gmail.com> Cc: Chris Li <chrisl@kernel.org> Cc: Dan Streetman <ddstreet@ieee.org> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Muchun Song <muchun.song@linux.dev> Cc: Roman Gushchin <roman.gushchin@linux.dev> Cc: Seth Jennings <sjenning@redhat.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Shuah Khan <shuah@kernel.org> Cc: Vitaly Wool <vitaly.wool@konsulko.com> Cc: Yosry Ahmed <yosryahmed@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-11-30 19:40:20 +00:00
&page_allocated, false);
mempolicy: alloc_pages_mpol() for NUMA policy without vma Shrink shmem's stack usage by eliminating the pseudo-vma from its folio allocation. alloc_pages_mpol(gfp, order, pol, ilx, nid) becomes the principal actor for passing mempolicy choice down to __alloc_pages(), rather than vma_alloc_folio(gfp, order, vma, addr, hugepage). vma_alloc_folio() and alloc_pages() remain, but as wrappers around alloc_pages_mpol(). alloc_pages_bulk_*() untouched, except to provide the additional args to policy_nodemask(), which subsumes policy_node(). Cleanup throughout, cutting out some unhelpful "helpers". It would all be much simpler without MPOL_INTERLEAVE, but that adds a dynamic to the constant mpol: complicated by v3.6 commit 09c231cb8bfd ("tmpfs: distribute interleave better across nodes"), which added ino bias to the interleave, hidden from mm/mempolicy.c until this commit. Hence "ilx" throughout, the "interleave index". Originally I thought it could be done just with nid, but that's wrong: the nodemask may come from the shared policy layer below a shmem vma, or it may come from the task layer above a shmem vma; and without the final nodemask then nodeid cannot be decided. And how ilx is applied depends also on page order. The interleave index is almost always irrelevant unless MPOL_INTERLEAVE: with one exception in alloc_pages_mpol(), where the NO_INTERLEAVE_INDEX passed down from vma-less alloc_pages() is also used as hint not to use THP-style hugepage allocation - to avoid the overhead of a hugepage arg (though I don't understand why we never just added a GFP bit for THP - if it actually needs a different allocation strategy from other pages of the same order). vma_alloc_folio() still carries its hugepage arg here, but it is not used, and should be removed when agreed. get_vma_policy() no longer allows a NULL vma: over time I believe we've eradicated all the places which used to need it e.g. swapoff and madvise used to pass NULL vma to read_swap_cache_async(), but now know the vma. [hughd@google.com: handle NULL mpol being passed to __read_swap_cache_async()] Link: https://lkml.kernel.org/r/ea419956-4751-0102-21f7-9c93cb957892@google.com Link: https://lkml.kernel.org/r/74e34633-6060-f5e3-aee-7040d43f2e93@google.com Link: https://lkml.kernel.org/r/1738368e-bac0-fd11-ed7f-b87142a939fe@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Hildenbrand <david@redhat.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Huang Ying <ying.huang@intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Nhat Pham <nphamcs@gmail.com> Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Tejun heo <tj@kernel.org> Cc: Vishal Moola (Oracle) <vishal.moola@gmail.com> Cc: Yang Shi <shy828301@gmail.com> Cc: Yosry Ahmed <yosryahmed@google.com> Cc: Domenico Cerasuolo <mimmocerasuolo@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-10-19 20:39:08 +00:00
mpol_cond_put(mpol);
mempolicy: alloc_pages_mpol() for NUMA policy without vma Shrink shmem's stack usage by eliminating the pseudo-vma from its folio allocation. alloc_pages_mpol(gfp, order, pol, ilx, nid) becomes the principal actor for passing mempolicy choice down to __alloc_pages(), rather than vma_alloc_folio(gfp, order, vma, addr, hugepage). vma_alloc_folio() and alloc_pages() remain, but as wrappers around alloc_pages_mpol(). alloc_pages_bulk_*() untouched, except to provide the additional args to policy_nodemask(), which subsumes policy_node(). Cleanup throughout, cutting out some unhelpful "helpers". It would all be much simpler without MPOL_INTERLEAVE, but that adds a dynamic to the constant mpol: complicated by v3.6 commit 09c231cb8bfd ("tmpfs: distribute interleave better across nodes"), which added ino bias to the interleave, hidden from mm/mempolicy.c until this commit. Hence "ilx" throughout, the "interleave index". Originally I thought it could be done just with nid, but that's wrong: the nodemask may come from the shared policy layer below a shmem vma, or it may come from the task layer above a shmem vma; and without the final nodemask then nodeid cannot be decided. And how ilx is applied depends also on page order. The interleave index is almost always irrelevant unless MPOL_INTERLEAVE: with one exception in alloc_pages_mpol(), where the NO_INTERLEAVE_INDEX passed down from vma-less alloc_pages() is also used as hint not to use THP-style hugepage allocation - to avoid the overhead of a hugepage arg (though I don't understand why we never just added a GFP bit for THP - if it actually needs a different allocation strategy from other pages of the same order). vma_alloc_folio() still carries its hugepage arg here, but it is not used, and should be removed when agreed. get_vma_policy() no longer allows a NULL vma: over time I believe we've eradicated all the places which used to need it e.g. swapoff and madvise used to pass NULL vma to read_swap_cache_async(), but now know the vma. [hughd@google.com: handle NULL mpol being passed to __read_swap_cache_async()] Link: https://lkml.kernel.org/r/ea419956-4751-0102-21f7-9c93cb957892@google.com Link: https://lkml.kernel.org/r/74e34633-6060-f5e3-aee-7040d43f2e93@google.com Link: https://lkml.kernel.org/r/1738368e-bac0-fd11-ed7f-b87142a939fe@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Hildenbrand <david@redhat.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Huang Ying <ying.huang@intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Nhat Pham <nphamcs@gmail.com> Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Tejun heo <tj@kernel.org> Cc: Vishal Moola (Oracle) <vishal.moola@gmail.com> Cc: Yang Shi <shy828301@gmail.com> Cc: Yosry Ahmed <yosryahmed@google.com> Cc: Domenico Cerasuolo <mimmocerasuolo@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-10-19 20:39:08 +00:00
if (page_allocated)
swap_read_folio(folio, false, plug);
return folio;
}
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
static unsigned int __swapin_nr_pages(unsigned long prev_offset,
unsigned long offset,
int hits,
int max_pages,
int prev_win)
swap: add a simple detector for inappropriate swapin readahead This is a patch to improve swap readahead algorithm. It's from Hugh and I slightly changed it. Hugh's original changelog: swapin readahead does a blind readahead, whether or not the swapin is sequential. This may be ok on harddisk, because large reads have relatively small costs, and if the readahead pages are unneeded they can be reclaimed easily - though, what if their allocation forced reclaim of useful pages? But on SSD devices large reads are more expensive than small ones: if the readahead pages are unneeded, reading them in caused significant overhead. This patch adds very simplistic random read detection. Stealing the PageReadahead technique from Konstantin Khlebnikov's patch, avoiding the vma/anon_vma sophistications of Shaohua Li's patch, swapin_nr_pages() simply looks at readahead's current success rate, and narrows or widens its readahead window accordingly. There is little science to its heuristic: it's about as stupid as can be whilst remaining effective. The table below shows elapsed times (in centiseconds) when running a single repetitive swapping load across a 1000MB mapping in 900MB ram with 1GB swap (the harddisk tests had taken painfully too long when I used mem=500M, but SSD shows similar results for that). Vanilla is the 3.6-rc7 kernel on which I started; Shaohua denotes his Sep 3 patch in mmotm and linux-next; HughOld denotes my Oct 1 patch which Shaohua showed to be defective; HughNew this Nov 14 patch, with page_cluster as usual at default of 3 (8-page reads); HughPC4 this same patch with page_cluster 4 (16-page reads); HughPC0 with page_cluster 0 (1-page reads: no readahead). HDD for swapping to harddisk, SSD for swapping to VertexII SSD. Seq for sequential access to the mapping, cycling five times around; Rand for the same number of random touches. Anon for a MAP_PRIVATE anon mapping; Shmem for a MAP_SHARED anon mapping, equivalent to tmpfs. One weakness of Shaohua's vma/anon_vma approach was that it did not optimize Shmem: seen below. Konstantin's approach was perhaps mistuned, 50% slower on Seq: did not compete and is not shown below. HDD Vanilla Shaohua HughOld HughNew HughPC4 HughPC0 Seq Anon 73921 76210 75611 76904 78191 121542 Seq Shmem 73601 73176 73855 72947 74543 118322 Rand Anon 895392 831243 871569 845197 846496 841680 Rand Shmem 1058375 1053486 827935 764955 764376 756489 SSD Vanilla Shaohua HughOld HughNew HughPC4 HughPC0 Seq Anon 24634 24198 24673 25107 21614 70018 Seq Shmem 24959 24932 25052 25703 22030 69678 Rand Anon 43014 26146 28075 25989 26935 25901 Rand Shmem 45349 45215 28249 24268 24138 24332 These tests are, of course, two extremes of a very simple case: under heavier mixed loads I've not yet observed any consistent improvement or degradation, and wider testing would be welcome. Shaohua Li: Test shows Vanilla is slightly better in sequential workload than Hugh's patch. I observed with Hugh's patch sometimes the readahead size is shrinked too fast (from 8 to 1 immediately) in sequential workload if there is no hit. And in such case, continuing doing readahead is good actually. I don't prepare a sophisticated algorithm for the sequential workload because so far we can't guarantee sequential accessed pages are swap out sequentially. So I slightly change Hugh's heuristic - don't shrink readahead size too fast. Here is my test result (unit second, 3 runs average): Vanilla Hugh New Seq 356 370 360 Random 4525 2447 2444 Attached graph is the swapin/swapout throughput I collected with 'vmstat 2'. The first part is running a random workload (till around 1200 of the x-axis) and the second part is running a sequential workload. swapin and swapout throughput are almost identical in steady state in both workloads. These are expected behavior. while in Vanilla, swapin is much bigger than swapout especially in random workload (because wrong readahead). Original patches by: Shaohua Li and Konstantin Khlebnikov. [fengguang.wu@intel.com: swapin_nr_pages() can be static] Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: Fengguang Wu <fengguang.wu@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Konstantin Khlebnikov <khlebnikov@openvz.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-02-06 20:04:21 +00:00
{
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
unsigned int pages, last_ra;
swap: add a simple detector for inappropriate swapin readahead This is a patch to improve swap readahead algorithm. It's from Hugh and I slightly changed it. Hugh's original changelog: swapin readahead does a blind readahead, whether or not the swapin is sequential. This may be ok on harddisk, because large reads have relatively small costs, and if the readahead pages are unneeded they can be reclaimed easily - though, what if their allocation forced reclaim of useful pages? But on SSD devices large reads are more expensive than small ones: if the readahead pages are unneeded, reading them in caused significant overhead. This patch adds very simplistic random read detection. Stealing the PageReadahead technique from Konstantin Khlebnikov's patch, avoiding the vma/anon_vma sophistications of Shaohua Li's patch, swapin_nr_pages() simply looks at readahead's current success rate, and narrows or widens its readahead window accordingly. There is little science to its heuristic: it's about as stupid as can be whilst remaining effective. The table below shows elapsed times (in centiseconds) when running a single repetitive swapping load across a 1000MB mapping in 900MB ram with 1GB swap (the harddisk tests had taken painfully too long when I used mem=500M, but SSD shows similar results for that). Vanilla is the 3.6-rc7 kernel on which I started; Shaohua denotes his Sep 3 patch in mmotm and linux-next; HughOld denotes my Oct 1 patch which Shaohua showed to be defective; HughNew this Nov 14 patch, with page_cluster as usual at default of 3 (8-page reads); HughPC4 this same patch with page_cluster 4 (16-page reads); HughPC0 with page_cluster 0 (1-page reads: no readahead). HDD for swapping to harddisk, SSD for swapping to VertexII SSD. Seq for sequential access to the mapping, cycling five times around; Rand for the same number of random touches. Anon for a MAP_PRIVATE anon mapping; Shmem for a MAP_SHARED anon mapping, equivalent to tmpfs. One weakness of Shaohua's vma/anon_vma approach was that it did not optimize Shmem: seen below. Konstantin's approach was perhaps mistuned, 50% slower on Seq: did not compete and is not shown below. HDD Vanilla Shaohua HughOld HughNew HughPC4 HughPC0 Seq Anon 73921 76210 75611 76904 78191 121542 Seq Shmem 73601 73176 73855 72947 74543 118322 Rand Anon 895392 831243 871569 845197 846496 841680 Rand Shmem 1058375 1053486 827935 764955 764376 756489 SSD Vanilla Shaohua HughOld HughNew HughPC4 HughPC0 Seq Anon 24634 24198 24673 25107 21614 70018 Seq Shmem 24959 24932 25052 25703 22030 69678 Rand Anon 43014 26146 28075 25989 26935 25901 Rand Shmem 45349 45215 28249 24268 24138 24332 These tests are, of course, two extremes of a very simple case: under heavier mixed loads I've not yet observed any consistent improvement or degradation, and wider testing would be welcome. Shaohua Li: Test shows Vanilla is slightly better in sequential workload than Hugh's patch. I observed with Hugh's patch sometimes the readahead size is shrinked too fast (from 8 to 1 immediately) in sequential workload if there is no hit. And in such case, continuing doing readahead is good actually. I don't prepare a sophisticated algorithm for the sequential workload because so far we can't guarantee sequential accessed pages are swap out sequentially. So I slightly change Hugh's heuristic - don't shrink readahead size too fast. Here is my test result (unit second, 3 runs average): Vanilla Hugh New Seq 356 370 360 Random 4525 2447 2444 Attached graph is the swapin/swapout throughput I collected with 'vmstat 2'. The first part is running a random workload (till around 1200 of the x-axis) and the second part is running a sequential workload. swapin and swapout throughput are almost identical in steady state in both workloads. These are expected behavior. while in Vanilla, swapin is much bigger than swapout especially in random workload (because wrong readahead). Original patches by: Shaohua Li and Konstantin Khlebnikov. [fengguang.wu@intel.com: swapin_nr_pages() can be static] Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: Fengguang Wu <fengguang.wu@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Konstantin Khlebnikov <khlebnikov@openvz.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-02-06 20:04:21 +00:00
/*
* This heuristic has been found to work well on both sequential and
* random loads, swapping to hard disk or to SSD: please don't ask
* what the "+ 2" means, it just happens to work well, that's all.
*/
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
pages = hits + 2;
swap: add a simple detector for inappropriate swapin readahead This is a patch to improve swap readahead algorithm. It's from Hugh and I slightly changed it. Hugh's original changelog: swapin readahead does a blind readahead, whether or not the swapin is sequential. This may be ok on harddisk, because large reads have relatively small costs, and if the readahead pages are unneeded they can be reclaimed easily - though, what if their allocation forced reclaim of useful pages? But on SSD devices large reads are more expensive than small ones: if the readahead pages are unneeded, reading them in caused significant overhead. This patch adds very simplistic random read detection. Stealing the PageReadahead technique from Konstantin Khlebnikov's patch, avoiding the vma/anon_vma sophistications of Shaohua Li's patch, swapin_nr_pages() simply looks at readahead's current success rate, and narrows or widens its readahead window accordingly. There is little science to its heuristic: it's about as stupid as can be whilst remaining effective. The table below shows elapsed times (in centiseconds) when running a single repetitive swapping load across a 1000MB mapping in 900MB ram with 1GB swap (the harddisk tests had taken painfully too long when I used mem=500M, but SSD shows similar results for that). Vanilla is the 3.6-rc7 kernel on which I started; Shaohua denotes his Sep 3 patch in mmotm and linux-next; HughOld denotes my Oct 1 patch which Shaohua showed to be defective; HughNew this Nov 14 patch, with page_cluster as usual at default of 3 (8-page reads); HughPC4 this same patch with page_cluster 4 (16-page reads); HughPC0 with page_cluster 0 (1-page reads: no readahead). HDD for swapping to harddisk, SSD for swapping to VertexII SSD. Seq for sequential access to the mapping, cycling five times around; Rand for the same number of random touches. Anon for a MAP_PRIVATE anon mapping; Shmem for a MAP_SHARED anon mapping, equivalent to tmpfs. One weakness of Shaohua's vma/anon_vma approach was that it did not optimize Shmem: seen below. Konstantin's approach was perhaps mistuned, 50% slower on Seq: did not compete and is not shown below. HDD Vanilla Shaohua HughOld HughNew HughPC4 HughPC0 Seq Anon 73921 76210 75611 76904 78191 121542 Seq Shmem 73601 73176 73855 72947 74543 118322 Rand Anon 895392 831243 871569 845197 846496 841680 Rand Shmem 1058375 1053486 827935 764955 764376 756489 SSD Vanilla Shaohua HughOld HughNew HughPC4 HughPC0 Seq Anon 24634 24198 24673 25107 21614 70018 Seq Shmem 24959 24932 25052 25703 22030 69678 Rand Anon 43014 26146 28075 25989 26935 25901 Rand Shmem 45349 45215 28249 24268 24138 24332 These tests are, of course, two extremes of a very simple case: under heavier mixed loads I've not yet observed any consistent improvement or degradation, and wider testing would be welcome. Shaohua Li: Test shows Vanilla is slightly better in sequential workload than Hugh's patch. I observed with Hugh's patch sometimes the readahead size is shrinked too fast (from 8 to 1 immediately) in sequential workload if there is no hit. And in such case, continuing doing readahead is good actually. I don't prepare a sophisticated algorithm for the sequential workload because so far we can't guarantee sequential accessed pages are swap out sequentially. So I slightly change Hugh's heuristic - don't shrink readahead size too fast. Here is my test result (unit second, 3 runs average): Vanilla Hugh New Seq 356 370 360 Random 4525 2447 2444 Attached graph is the swapin/swapout throughput I collected with 'vmstat 2'. The first part is running a random workload (till around 1200 of the x-axis) and the second part is running a sequential workload. swapin and swapout throughput are almost identical in steady state in both workloads. These are expected behavior. while in Vanilla, swapin is much bigger than swapout especially in random workload (because wrong readahead). Original patches by: Shaohua Li and Konstantin Khlebnikov. [fengguang.wu@intel.com: swapin_nr_pages() can be static] Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: Fengguang Wu <fengguang.wu@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Konstantin Khlebnikov <khlebnikov@openvz.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-02-06 20:04:21 +00:00
if (pages == 2) {
/*
* We can have no readahead hits to judge by: but must not get
* stuck here forever, so check for an adjacent offset instead
* (and don't even bother to check whether swap type is same).
*/
if (offset != prev_offset + 1 && offset != prev_offset - 1)
pages = 1;
} else {
unsigned int roundup = 4;
while (roundup < pages)
roundup <<= 1;
pages = roundup;
}
if (pages > max_pages)
pages = max_pages;
/* Don't shrink readahead too fast */
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
last_ra = prev_win / 2;
swap: add a simple detector for inappropriate swapin readahead This is a patch to improve swap readahead algorithm. It's from Hugh and I slightly changed it. Hugh's original changelog: swapin readahead does a blind readahead, whether or not the swapin is sequential. This may be ok on harddisk, because large reads have relatively small costs, and if the readahead pages are unneeded they can be reclaimed easily - though, what if their allocation forced reclaim of useful pages? But on SSD devices large reads are more expensive than small ones: if the readahead pages are unneeded, reading them in caused significant overhead. This patch adds very simplistic random read detection. Stealing the PageReadahead technique from Konstantin Khlebnikov's patch, avoiding the vma/anon_vma sophistications of Shaohua Li's patch, swapin_nr_pages() simply looks at readahead's current success rate, and narrows or widens its readahead window accordingly. There is little science to its heuristic: it's about as stupid as can be whilst remaining effective. The table below shows elapsed times (in centiseconds) when running a single repetitive swapping load across a 1000MB mapping in 900MB ram with 1GB swap (the harddisk tests had taken painfully too long when I used mem=500M, but SSD shows similar results for that). Vanilla is the 3.6-rc7 kernel on which I started; Shaohua denotes his Sep 3 patch in mmotm and linux-next; HughOld denotes my Oct 1 patch which Shaohua showed to be defective; HughNew this Nov 14 patch, with page_cluster as usual at default of 3 (8-page reads); HughPC4 this same patch with page_cluster 4 (16-page reads); HughPC0 with page_cluster 0 (1-page reads: no readahead). HDD for swapping to harddisk, SSD for swapping to VertexII SSD. Seq for sequential access to the mapping, cycling five times around; Rand for the same number of random touches. Anon for a MAP_PRIVATE anon mapping; Shmem for a MAP_SHARED anon mapping, equivalent to tmpfs. One weakness of Shaohua's vma/anon_vma approach was that it did not optimize Shmem: seen below. Konstantin's approach was perhaps mistuned, 50% slower on Seq: did not compete and is not shown below. HDD Vanilla Shaohua HughOld HughNew HughPC4 HughPC0 Seq Anon 73921 76210 75611 76904 78191 121542 Seq Shmem 73601 73176 73855 72947 74543 118322 Rand Anon 895392 831243 871569 845197 846496 841680 Rand Shmem 1058375 1053486 827935 764955 764376 756489 SSD Vanilla Shaohua HughOld HughNew HughPC4 HughPC0 Seq Anon 24634 24198 24673 25107 21614 70018 Seq Shmem 24959 24932 25052 25703 22030 69678 Rand Anon 43014 26146 28075 25989 26935 25901 Rand Shmem 45349 45215 28249 24268 24138 24332 These tests are, of course, two extremes of a very simple case: under heavier mixed loads I've not yet observed any consistent improvement or degradation, and wider testing would be welcome. Shaohua Li: Test shows Vanilla is slightly better in sequential workload than Hugh's patch. I observed with Hugh's patch sometimes the readahead size is shrinked too fast (from 8 to 1 immediately) in sequential workload if there is no hit. And in such case, continuing doing readahead is good actually. I don't prepare a sophisticated algorithm for the sequential workload because so far we can't guarantee sequential accessed pages are swap out sequentially. So I slightly change Hugh's heuristic - don't shrink readahead size too fast. Here is my test result (unit second, 3 runs average): Vanilla Hugh New Seq 356 370 360 Random 4525 2447 2444 Attached graph is the swapin/swapout throughput I collected with 'vmstat 2'. The first part is running a random workload (till around 1200 of the x-axis) and the second part is running a sequential workload. swapin and swapout throughput are almost identical in steady state in both workloads. These are expected behavior. while in Vanilla, swapin is much bigger than swapout especially in random workload (because wrong readahead). Original patches by: Shaohua Li and Konstantin Khlebnikov. [fengguang.wu@intel.com: swapin_nr_pages() can be static] Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: Fengguang Wu <fengguang.wu@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Konstantin Khlebnikov <khlebnikov@openvz.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-02-06 20:04:21 +00:00
if (pages < last_ra)
pages = last_ra;
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
return pages;
}
static unsigned long swapin_nr_pages(unsigned long offset)
{
static unsigned long prev_offset;
unsigned int hits, pages, max_pages;
static atomic_t last_readahead_pages;
max_pages = 1 << READ_ONCE(page_cluster);
if (max_pages <= 1)
return 1;
hits = atomic_xchg(&swapin_readahead_hits, 0);
mm/swap_state: fix a data race in swapin_nr_pages "prev_offset" is a static variable in swapin_nr_pages() that can be accessed concurrently with only mmap_sem held in read mode as noticed by KCSAN, BUG: KCSAN: data-race in swap_cluster_readahead / swap_cluster_readahead write to 0xffffffff92763830 of 8 bytes by task 14795 on cpu 17: swap_cluster_readahead+0x2a6/0x5e0 swapin_readahead+0x92/0x8dc do_swap_page+0x49b/0xf20 __handle_mm_fault+0xcfb/0xd70 handle_mm_fault+0xfc/0x2f0 do_page_fault+0x263/0x715 page_fault+0x34/0x40 1 lock held by (dnf)/14795: #0: ffff897bd2e98858 (&mm->mmap_sem#2){++++}-{3:3}, at: do_page_fault+0x143/0x715 do_user_addr_fault at arch/x86/mm/fault.c:1405 (inlined by) do_page_fault at arch/x86/mm/fault.c:1535 irq event stamp: 83493 count_memcg_event_mm+0x1a6/0x270 count_memcg_event_mm+0x119/0x270 __do_softirq+0x365/0x589 irq_exit+0xa2/0xc0 read to 0xffffffff92763830 of 8 bytes by task 1 on cpu 22: swap_cluster_readahead+0xfd/0x5e0 swapin_readahead+0x92/0x8dc do_swap_page+0x49b/0xf20 __handle_mm_fault+0xcfb/0xd70 handle_mm_fault+0xfc/0x2f0 do_page_fault+0x263/0x715 page_fault+0x34/0x40 1 lock held by systemd/1: #0: ffff897c38f14858 (&mm->mmap_sem#2){++++}-{3:3}, at: do_page_fault+0x143/0x715 irq event stamp: 43530289 count_memcg_event_mm+0x1a6/0x270 count_memcg_event_mm+0x119/0x270 __do_softirq+0x365/0x589 irq_exit+0xa2/0xc0 Signed-off-by: Qian Cai <cai@lca.pw> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Cc: Marco Elver <elver@google.com> Cc: Hugh Dickins <hughd@google.com> Link: http://lkml.kernel.org/r/20200402213748.2237-1-cai@lca.pw Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-02 04:48:40 +00:00
pages = __swapin_nr_pages(READ_ONCE(prev_offset), offset, hits,
max_pages,
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
atomic_read(&last_readahead_pages));
if (!hits)
mm/swap_state: fix a data race in swapin_nr_pages "prev_offset" is a static variable in swapin_nr_pages() that can be accessed concurrently with only mmap_sem held in read mode as noticed by KCSAN, BUG: KCSAN: data-race in swap_cluster_readahead / swap_cluster_readahead write to 0xffffffff92763830 of 8 bytes by task 14795 on cpu 17: swap_cluster_readahead+0x2a6/0x5e0 swapin_readahead+0x92/0x8dc do_swap_page+0x49b/0xf20 __handle_mm_fault+0xcfb/0xd70 handle_mm_fault+0xfc/0x2f0 do_page_fault+0x263/0x715 page_fault+0x34/0x40 1 lock held by (dnf)/14795: #0: ffff897bd2e98858 (&mm->mmap_sem#2){++++}-{3:3}, at: do_page_fault+0x143/0x715 do_user_addr_fault at arch/x86/mm/fault.c:1405 (inlined by) do_page_fault at arch/x86/mm/fault.c:1535 irq event stamp: 83493 count_memcg_event_mm+0x1a6/0x270 count_memcg_event_mm+0x119/0x270 __do_softirq+0x365/0x589 irq_exit+0xa2/0xc0 read to 0xffffffff92763830 of 8 bytes by task 1 on cpu 22: swap_cluster_readahead+0xfd/0x5e0 swapin_readahead+0x92/0x8dc do_swap_page+0x49b/0xf20 __handle_mm_fault+0xcfb/0xd70 handle_mm_fault+0xfc/0x2f0 do_page_fault+0x263/0x715 page_fault+0x34/0x40 1 lock held by systemd/1: #0: ffff897c38f14858 (&mm->mmap_sem#2){++++}-{3:3}, at: do_page_fault+0x143/0x715 irq event stamp: 43530289 count_memcg_event_mm+0x1a6/0x270 count_memcg_event_mm+0x119/0x270 __do_softirq+0x365/0x589 irq_exit+0xa2/0xc0 Signed-off-by: Qian Cai <cai@lca.pw> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Cc: Marco Elver <elver@google.com> Cc: Hugh Dickins <hughd@google.com> Link: http://lkml.kernel.org/r/20200402213748.2237-1-cai@lca.pw Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-02 04:48:40 +00:00
WRITE_ONCE(prev_offset, offset);
swap: add a simple detector for inappropriate swapin readahead This is a patch to improve swap readahead algorithm. It's from Hugh and I slightly changed it. Hugh's original changelog: swapin readahead does a blind readahead, whether or not the swapin is sequential. This may be ok on harddisk, because large reads have relatively small costs, and if the readahead pages are unneeded they can be reclaimed easily - though, what if their allocation forced reclaim of useful pages? But on SSD devices large reads are more expensive than small ones: if the readahead pages are unneeded, reading them in caused significant overhead. This patch adds very simplistic random read detection. Stealing the PageReadahead technique from Konstantin Khlebnikov's patch, avoiding the vma/anon_vma sophistications of Shaohua Li's patch, swapin_nr_pages() simply looks at readahead's current success rate, and narrows or widens its readahead window accordingly. There is little science to its heuristic: it's about as stupid as can be whilst remaining effective. The table below shows elapsed times (in centiseconds) when running a single repetitive swapping load across a 1000MB mapping in 900MB ram with 1GB swap (the harddisk tests had taken painfully too long when I used mem=500M, but SSD shows similar results for that). Vanilla is the 3.6-rc7 kernel on which I started; Shaohua denotes his Sep 3 patch in mmotm and linux-next; HughOld denotes my Oct 1 patch which Shaohua showed to be defective; HughNew this Nov 14 patch, with page_cluster as usual at default of 3 (8-page reads); HughPC4 this same patch with page_cluster 4 (16-page reads); HughPC0 with page_cluster 0 (1-page reads: no readahead). HDD for swapping to harddisk, SSD for swapping to VertexII SSD. Seq for sequential access to the mapping, cycling five times around; Rand for the same number of random touches. Anon for a MAP_PRIVATE anon mapping; Shmem for a MAP_SHARED anon mapping, equivalent to tmpfs. One weakness of Shaohua's vma/anon_vma approach was that it did not optimize Shmem: seen below. Konstantin's approach was perhaps mistuned, 50% slower on Seq: did not compete and is not shown below. HDD Vanilla Shaohua HughOld HughNew HughPC4 HughPC0 Seq Anon 73921 76210 75611 76904 78191 121542 Seq Shmem 73601 73176 73855 72947 74543 118322 Rand Anon 895392 831243 871569 845197 846496 841680 Rand Shmem 1058375 1053486 827935 764955 764376 756489 SSD Vanilla Shaohua HughOld HughNew HughPC4 HughPC0 Seq Anon 24634 24198 24673 25107 21614 70018 Seq Shmem 24959 24932 25052 25703 22030 69678 Rand Anon 43014 26146 28075 25989 26935 25901 Rand Shmem 45349 45215 28249 24268 24138 24332 These tests are, of course, two extremes of a very simple case: under heavier mixed loads I've not yet observed any consistent improvement or degradation, and wider testing would be welcome. Shaohua Li: Test shows Vanilla is slightly better in sequential workload than Hugh's patch. I observed with Hugh's patch sometimes the readahead size is shrinked too fast (from 8 to 1 immediately) in sequential workload if there is no hit. And in such case, continuing doing readahead is good actually. I don't prepare a sophisticated algorithm for the sequential workload because so far we can't guarantee sequential accessed pages are swap out sequentially. So I slightly change Hugh's heuristic - don't shrink readahead size too fast. Here is my test result (unit second, 3 runs average): Vanilla Hugh New Seq 356 370 360 Random 4525 2447 2444 Attached graph is the swapin/swapout throughput I collected with 'vmstat 2'. The first part is running a random workload (till around 1200 of the x-axis) and the second part is running a sequential workload. swapin and swapout throughput are almost identical in steady state in both workloads. These are expected behavior. while in Vanilla, swapin is much bigger than swapout especially in random workload (because wrong readahead). Original patches by: Shaohua Li and Konstantin Khlebnikov. [fengguang.wu@intel.com: swapin_nr_pages() can be static] Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: Fengguang Wu <fengguang.wu@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Konstantin Khlebnikov <khlebnikov@openvz.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-02-06 20:04:21 +00:00
atomic_set(&last_readahead_pages, pages);
return pages;
}
/**
* swap_cluster_readahead - swap in pages in hope we need them soon
* @entry: swap entry of this memory
* @gfp_mask: memory allocation flags
mempolicy: alloc_pages_mpol() for NUMA policy without vma Shrink shmem's stack usage by eliminating the pseudo-vma from its folio allocation. alloc_pages_mpol(gfp, order, pol, ilx, nid) becomes the principal actor for passing mempolicy choice down to __alloc_pages(), rather than vma_alloc_folio(gfp, order, vma, addr, hugepage). vma_alloc_folio() and alloc_pages() remain, but as wrappers around alloc_pages_mpol(). alloc_pages_bulk_*() untouched, except to provide the additional args to policy_nodemask(), which subsumes policy_node(). Cleanup throughout, cutting out some unhelpful "helpers". It would all be much simpler without MPOL_INTERLEAVE, but that adds a dynamic to the constant mpol: complicated by v3.6 commit 09c231cb8bfd ("tmpfs: distribute interleave better across nodes"), which added ino bias to the interleave, hidden from mm/mempolicy.c until this commit. Hence "ilx" throughout, the "interleave index". Originally I thought it could be done just with nid, but that's wrong: the nodemask may come from the shared policy layer below a shmem vma, or it may come from the task layer above a shmem vma; and without the final nodemask then nodeid cannot be decided. And how ilx is applied depends also on page order. The interleave index is almost always irrelevant unless MPOL_INTERLEAVE: with one exception in alloc_pages_mpol(), where the NO_INTERLEAVE_INDEX passed down from vma-less alloc_pages() is also used as hint not to use THP-style hugepage allocation - to avoid the overhead of a hugepage arg (though I don't understand why we never just added a GFP bit for THP - if it actually needs a different allocation strategy from other pages of the same order). vma_alloc_folio() still carries its hugepage arg here, but it is not used, and should be removed when agreed. get_vma_policy() no longer allows a NULL vma: over time I believe we've eradicated all the places which used to need it e.g. swapoff and madvise used to pass NULL vma to read_swap_cache_async(), but now know the vma. [hughd@google.com: handle NULL mpol being passed to __read_swap_cache_async()] Link: https://lkml.kernel.org/r/ea419956-4751-0102-21f7-9c93cb957892@google.com Link: https://lkml.kernel.org/r/74e34633-6060-f5e3-aee-7040d43f2e93@google.com Link: https://lkml.kernel.org/r/1738368e-bac0-fd11-ed7f-b87142a939fe@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Hildenbrand <david@redhat.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Huang Ying <ying.huang@intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Nhat Pham <nphamcs@gmail.com> Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Tejun heo <tj@kernel.org> Cc: Vishal Moola (Oracle) <vishal.moola@gmail.com> Cc: Yang Shi <shy828301@gmail.com> Cc: Yosry Ahmed <yosryahmed@google.com> Cc: Domenico Cerasuolo <mimmocerasuolo@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-10-19 20:39:08 +00:00
* @mpol: NUMA memory allocation policy to be applied
* @ilx: NUMA interleave index, for use only when MPOL_INTERLEAVE
*
* Returns the struct folio for entry and addr, after queueing swapin.
*
* Primitive swap readahead code. We simply read an aligned block of
* (1 << page_cluster) entries in the swap area. This method is chosen
* because it doesn't cost us any seek time. We also make sure to queue
* the 'original' request together with the readahead ones...
*
mempolicy: alloc_pages_mpol() for NUMA policy without vma Shrink shmem's stack usage by eliminating the pseudo-vma from its folio allocation. alloc_pages_mpol(gfp, order, pol, ilx, nid) becomes the principal actor for passing mempolicy choice down to __alloc_pages(), rather than vma_alloc_folio(gfp, order, vma, addr, hugepage). vma_alloc_folio() and alloc_pages() remain, but as wrappers around alloc_pages_mpol(). alloc_pages_bulk_*() untouched, except to provide the additional args to policy_nodemask(), which subsumes policy_node(). Cleanup throughout, cutting out some unhelpful "helpers". It would all be much simpler without MPOL_INTERLEAVE, but that adds a dynamic to the constant mpol: complicated by v3.6 commit 09c231cb8bfd ("tmpfs: distribute interleave better across nodes"), which added ino bias to the interleave, hidden from mm/mempolicy.c until this commit. Hence "ilx" throughout, the "interleave index". Originally I thought it could be done just with nid, but that's wrong: the nodemask may come from the shared policy layer below a shmem vma, or it may come from the task layer above a shmem vma; and without the final nodemask then nodeid cannot be decided. And how ilx is applied depends also on page order. The interleave index is almost always irrelevant unless MPOL_INTERLEAVE: with one exception in alloc_pages_mpol(), where the NO_INTERLEAVE_INDEX passed down from vma-less alloc_pages() is also used as hint not to use THP-style hugepage allocation - to avoid the overhead of a hugepage arg (though I don't understand why we never just added a GFP bit for THP - if it actually needs a different allocation strategy from other pages of the same order). vma_alloc_folio() still carries its hugepage arg here, but it is not used, and should be removed when agreed. get_vma_policy() no longer allows a NULL vma: over time I believe we've eradicated all the places which used to need it e.g. swapoff and madvise used to pass NULL vma to read_swap_cache_async(), but now know the vma. [hughd@google.com: handle NULL mpol being passed to __read_swap_cache_async()] Link: https://lkml.kernel.org/r/ea419956-4751-0102-21f7-9c93cb957892@google.com Link: https://lkml.kernel.org/r/74e34633-6060-f5e3-aee-7040d43f2e93@google.com Link: https://lkml.kernel.org/r/1738368e-bac0-fd11-ed7f-b87142a939fe@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Hildenbrand <david@redhat.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Huang Ying <ying.huang@intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Nhat Pham <nphamcs@gmail.com> Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Tejun heo <tj@kernel.org> Cc: Vishal Moola (Oracle) <vishal.moola@gmail.com> Cc: Yang Shi <shy828301@gmail.com> Cc: Yosry Ahmed <yosryahmed@google.com> Cc: Domenico Cerasuolo <mimmocerasuolo@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-10-19 20:39:08 +00:00
* Note: it is intentional that the same NUMA policy and interleave index
* are used for every page of the readahead: neighbouring pages on swap
* are fairly likely to have been swapped out from the same node.
*/
struct folio *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask,
mempolicy: alloc_pages_mpol() for NUMA policy without vma Shrink shmem's stack usage by eliminating the pseudo-vma from its folio allocation. alloc_pages_mpol(gfp, order, pol, ilx, nid) becomes the principal actor for passing mempolicy choice down to __alloc_pages(), rather than vma_alloc_folio(gfp, order, vma, addr, hugepage). vma_alloc_folio() and alloc_pages() remain, but as wrappers around alloc_pages_mpol(). alloc_pages_bulk_*() untouched, except to provide the additional args to policy_nodemask(), which subsumes policy_node(). Cleanup throughout, cutting out some unhelpful "helpers". It would all be much simpler without MPOL_INTERLEAVE, but that adds a dynamic to the constant mpol: complicated by v3.6 commit 09c231cb8bfd ("tmpfs: distribute interleave better across nodes"), which added ino bias to the interleave, hidden from mm/mempolicy.c until this commit. Hence "ilx" throughout, the "interleave index". Originally I thought it could be done just with nid, but that's wrong: the nodemask may come from the shared policy layer below a shmem vma, or it may come from the task layer above a shmem vma; and without the final nodemask then nodeid cannot be decided. And how ilx is applied depends also on page order. The interleave index is almost always irrelevant unless MPOL_INTERLEAVE: with one exception in alloc_pages_mpol(), where the NO_INTERLEAVE_INDEX passed down from vma-less alloc_pages() is also used as hint not to use THP-style hugepage allocation - to avoid the overhead of a hugepage arg (though I don't understand why we never just added a GFP bit for THP - if it actually needs a different allocation strategy from other pages of the same order). vma_alloc_folio() still carries its hugepage arg here, but it is not used, and should be removed when agreed. get_vma_policy() no longer allows a NULL vma: over time I believe we've eradicated all the places which used to need it e.g. swapoff and madvise used to pass NULL vma to read_swap_cache_async(), but now know the vma. [hughd@google.com: handle NULL mpol being passed to __read_swap_cache_async()] Link: https://lkml.kernel.org/r/ea419956-4751-0102-21f7-9c93cb957892@google.com Link: https://lkml.kernel.org/r/74e34633-6060-f5e3-aee-7040d43f2e93@google.com Link: https://lkml.kernel.org/r/1738368e-bac0-fd11-ed7f-b87142a939fe@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Hildenbrand <david@redhat.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Huang Ying <ying.huang@intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Nhat Pham <nphamcs@gmail.com> Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Tejun heo <tj@kernel.org> Cc: Vishal Moola (Oracle) <vishal.moola@gmail.com> Cc: Yang Shi <shy828301@gmail.com> Cc: Yosry Ahmed <yosryahmed@google.com> Cc: Domenico Cerasuolo <mimmocerasuolo@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-10-19 20:39:08 +00:00
struct mempolicy *mpol, pgoff_t ilx)
{
struct folio *folio;
swap: add a simple detector for inappropriate swapin readahead This is a patch to improve swap readahead algorithm. It's from Hugh and I slightly changed it. Hugh's original changelog: swapin readahead does a blind readahead, whether or not the swapin is sequential. This may be ok on harddisk, because large reads have relatively small costs, and if the readahead pages are unneeded they can be reclaimed easily - though, what if their allocation forced reclaim of useful pages? But on SSD devices large reads are more expensive than small ones: if the readahead pages are unneeded, reading them in caused significant overhead. This patch adds very simplistic random read detection. Stealing the PageReadahead technique from Konstantin Khlebnikov's patch, avoiding the vma/anon_vma sophistications of Shaohua Li's patch, swapin_nr_pages() simply looks at readahead's current success rate, and narrows or widens its readahead window accordingly. There is little science to its heuristic: it's about as stupid as can be whilst remaining effective. The table below shows elapsed times (in centiseconds) when running a single repetitive swapping load across a 1000MB mapping in 900MB ram with 1GB swap (the harddisk tests had taken painfully too long when I used mem=500M, but SSD shows similar results for that). Vanilla is the 3.6-rc7 kernel on which I started; Shaohua denotes his Sep 3 patch in mmotm and linux-next; HughOld denotes my Oct 1 patch which Shaohua showed to be defective; HughNew this Nov 14 patch, with page_cluster as usual at default of 3 (8-page reads); HughPC4 this same patch with page_cluster 4 (16-page reads); HughPC0 with page_cluster 0 (1-page reads: no readahead). HDD for swapping to harddisk, SSD for swapping to VertexII SSD. Seq for sequential access to the mapping, cycling five times around; Rand for the same number of random touches. Anon for a MAP_PRIVATE anon mapping; Shmem for a MAP_SHARED anon mapping, equivalent to tmpfs. One weakness of Shaohua's vma/anon_vma approach was that it did not optimize Shmem: seen below. Konstantin's approach was perhaps mistuned, 50% slower on Seq: did not compete and is not shown below. HDD Vanilla Shaohua HughOld HughNew HughPC4 HughPC0 Seq Anon 73921 76210 75611 76904 78191 121542 Seq Shmem 73601 73176 73855 72947 74543 118322 Rand Anon 895392 831243 871569 845197 846496 841680 Rand Shmem 1058375 1053486 827935 764955 764376 756489 SSD Vanilla Shaohua HughOld HughNew HughPC4 HughPC0 Seq Anon 24634 24198 24673 25107 21614 70018 Seq Shmem 24959 24932 25052 25703 22030 69678 Rand Anon 43014 26146 28075 25989 26935 25901 Rand Shmem 45349 45215 28249 24268 24138 24332 These tests are, of course, two extremes of a very simple case: under heavier mixed loads I've not yet observed any consistent improvement or degradation, and wider testing would be welcome. Shaohua Li: Test shows Vanilla is slightly better in sequential workload than Hugh's patch. I observed with Hugh's patch sometimes the readahead size is shrinked too fast (from 8 to 1 immediately) in sequential workload if there is no hit. And in such case, continuing doing readahead is good actually. I don't prepare a sophisticated algorithm for the sequential workload because so far we can't guarantee sequential accessed pages are swap out sequentially. So I slightly change Hugh's heuristic - don't shrink readahead size too fast. Here is my test result (unit second, 3 runs average): Vanilla Hugh New Seq 356 370 360 Random 4525 2447 2444 Attached graph is the swapin/swapout throughput I collected with 'vmstat 2'. The first part is running a random workload (till around 1200 of the x-axis) and the second part is running a sequential workload. swapin and swapout throughput are almost identical in steady state in both workloads. These are expected behavior. while in Vanilla, swapin is much bigger than swapout especially in random workload (because wrong readahead). Original patches by: Shaohua Li and Konstantin Khlebnikov. [fengguang.wu@intel.com: swapin_nr_pages() can be static] Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: Fengguang Wu <fengguang.wu@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Konstantin Khlebnikov <khlebnikov@openvz.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-02-06 20:04:21 +00:00
unsigned long entry_offset = swp_offset(entry);
unsigned long offset = entry_offset;
unsigned long start_offset, end_offset;
swap: add a simple detector for inappropriate swapin readahead This is a patch to improve swap readahead algorithm. It's from Hugh and I slightly changed it. Hugh's original changelog: swapin readahead does a blind readahead, whether or not the swapin is sequential. This may be ok on harddisk, because large reads have relatively small costs, and if the readahead pages are unneeded they can be reclaimed easily - though, what if their allocation forced reclaim of useful pages? But on SSD devices large reads are more expensive than small ones: if the readahead pages are unneeded, reading them in caused significant overhead. This patch adds very simplistic random read detection. Stealing the PageReadahead technique from Konstantin Khlebnikov's patch, avoiding the vma/anon_vma sophistications of Shaohua Li's patch, swapin_nr_pages() simply looks at readahead's current success rate, and narrows or widens its readahead window accordingly. There is little science to its heuristic: it's about as stupid as can be whilst remaining effective. The table below shows elapsed times (in centiseconds) when running a single repetitive swapping load across a 1000MB mapping in 900MB ram with 1GB swap (the harddisk tests had taken painfully too long when I used mem=500M, but SSD shows similar results for that). Vanilla is the 3.6-rc7 kernel on which I started; Shaohua denotes his Sep 3 patch in mmotm and linux-next; HughOld denotes my Oct 1 patch which Shaohua showed to be defective; HughNew this Nov 14 patch, with page_cluster as usual at default of 3 (8-page reads); HughPC4 this same patch with page_cluster 4 (16-page reads); HughPC0 with page_cluster 0 (1-page reads: no readahead). HDD for swapping to harddisk, SSD for swapping to VertexII SSD. Seq for sequential access to the mapping, cycling five times around; Rand for the same number of random touches. Anon for a MAP_PRIVATE anon mapping; Shmem for a MAP_SHARED anon mapping, equivalent to tmpfs. One weakness of Shaohua's vma/anon_vma approach was that it did not optimize Shmem: seen below. Konstantin's approach was perhaps mistuned, 50% slower on Seq: did not compete and is not shown below. HDD Vanilla Shaohua HughOld HughNew HughPC4 HughPC0 Seq Anon 73921 76210 75611 76904 78191 121542 Seq Shmem 73601 73176 73855 72947 74543 118322 Rand Anon 895392 831243 871569 845197 846496 841680 Rand Shmem 1058375 1053486 827935 764955 764376 756489 SSD Vanilla Shaohua HughOld HughNew HughPC4 HughPC0 Seq Anon 24634 24198 24673 25107 21614 70018 Seq Shmem 24959 24932 25052 25703 22030 69678 Rand Anon 43014 26146 28075 25989 26935 25901 Rand Shmem 45349 45215 28249 24268 24138 24332 These tests are, of course, two extremes of a very simple case: under heavier mixed loads I've not yet observed any consistent improvement or degradation, and wider testing would be welcome. Shaohua Li: Test shows Vanilla is slightly better in sequential workload than Hugh's patch. I observed with Hugh's patch sometimes the readahead size is shrinked too fast (from 8 to 1 immediately) in sequential workload if there is no hit. And in such case, continuing doing readahead is good actually. I don't prepare a sophisticated algorithm for the sequential workload because so far we can't guarantee sequential accessed pages are swap out sequentially. So I slightly change Hugh's heuristic - don't shrink readahead size too fast. Here is my test result (unit second, 3 runs average): Vanilla Hugh New Seq 356 370 360 Random 4525 2447 2444 Attached graph is the swapin/swapout throughput I collected with 'vmstat 2'. The first part is running a random workload (till around 1200 of the x-axis) and the second part is running a sequential workload. swapin and swapout throughput are almost identical in steady state in both workloads. These are expected behavior. while in Vanilla, swapin is much bigger than swapout especially in random workload (because wrong readahead). Original patches by: Shaohua Li and Konstantin Khlebnikov. [fengguang.wu@intel.com: swapin_nr_pages() can be static] Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: Fengguang Wu <fengguang.wu@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Konstantin Khlebnikov <khlebnikov@openvz.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-02-06 20:04:21 +00:00
unsigned long mask;
mm, swap: fix false error message in __swp_swapcount() When a page fault occurs for a swap entry, the physical swap readahead (not the VMA base swap readahead) may readahead several swap entries after the fault swap entry. The readahead algorithm calculates some of the swap entries to readahead via increasing the offset of the fault swap entry without checking whether they are beyond the end of the swap device and it relys on the __swp_swapcount() and swapcache_prepare() to check it. Although __swp_swapcount() checks for the swap entry passed in, it will complain with the error message as follow for the expected invalid swap entry. This may make the end users confused. swap_info_get: Bad swap offset entry 0200f8a7 To fix the false error message, the swap entry checking is added in swapin_readahead() to avoid to pass the out-of-bound swap entries and the swap entry reserved for the swap header to __swp_swapcount() and swapcache_prepare(). Link: http://lkml.kernel.org/r/20171102054225.22897-1-ying.huang@intel.com Fixes: e8c26ab60598 ("mm/swap: skip readahead for unreferenced swap slots") Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Reported-by: Christian Kujau <lists@nerdbynature.de> Acked-by: Minchan Kim <minchan@kernel.org> Suggested-by: Minchan Kim <minchan@kernel.org> Cc: Tim Chen <tim.c.chen@linux.intel.com> Cc: Michal Hocko <mhocko@suse.com> Cc: Hugh Dickins <hughd@google.com> Cc: <stable@vger.kernel.org> [4.11+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-11-16 01:33:15 +00:00
struct swap_info_struct *si = swp_swap_info(entry);
swap: allow swap readahead to be merged Swap readahead works fine, but the I/O to disk is almost always done in page size requests, despite the fact that readahead submits 1<<page-cluster pages at a time. On older kernels the old per device plugging behavior might have captured this and merged the requests, but currently all comes down to much more I/Os than required. On a single device this might not be an issue, but as soon as a server runs on shared san resources savin I/Os not only improves swapin throughput but also provides a lower resource utilization. With a load running KVM in a lot of memory overcommitment (the hot memory is 1.5 times the host memory) swapping throughput improves significantly and the lead feels more responsive as well as achieves more throughput. In a test setup with 16 swap disks running blocktrace on one of those disks shows the improved merging: Prior: Reads Queued: 560,888, 2,243MiB Writes Queued: 226,242, 904,968KiB Read Dispatches: 544,701, 2,243MiB Write Dispatches: 159,318, 904,968KiB Reads Requeued: 0 Writes Requeued: 0 Reads Completed: 544,716, 2,243MiB Writes Completed: 159,321, 904,980KiB Read Merges: 16,187, 64,748KiB Write Merges: 61,744, 246,976KiB IO unplugs: 149,614 Timer unplugs: 2,940 With the patch: Reads Queued: 734,315, 2,937MiB Writes Queued: 300,188, 1,200MiB Read Dispatches: 214,972, 2,937MiB Write Dispatches: 215,176, 1,200MiB Reads Requeued: 0 Writes Requeued: 0 Reads Completed: 214,971, 2,937MiB Writes Completed: 215,177, 1,200MiB Read Merges: 519,343, 2,077MiB Write Merges: 73,325, 293,300KiB IO unplugs: 337,130 Timer unplugs: 11,184 I got ~10% to ~40% more throughput in my cases and at the same time much lower cpu consumption when broken down per transferred kilobyte (the majority of that due to saved interrupts and better cache handling). In a shared SAN others might get an additional benefit as well, because this now causes less protocol overhead. Signed-off-by: Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com> Acked-by: Rik van Riel <riel@redhat.com> Acked-by: Jens Axboe <axboe@kernel.dk> Reviewed-by: Minchan Kim <minchan@kernel.org> Cc: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-07-31 23:41:44 +00:00
struct blk_plug plug;
struct swap_iocb *splug = NULL;
swap: remove remnants of polling from read_swap_cache_async Patch series "Per-VMA lock support for swap and userfaults", v7. When per-VMA locks were introduced in [1] several types of page faults would still fall back to mmap_lock to keep the patchset simple. Among them are swap and userfault pages. The main reason for skipping those cases was the fact that mmap_lock could be dropped while handling these faults and that required additional logic to be implemented. Implement the mechanism to allow per-VMA locks to be dropped for these cases. First, change handle_mm_fault to drop per-VMA locks when returning VM_FAULT_RETRY or VM_FAULT_COMPLETED to be consistent with the way mmap_lock is handled. Then change folio_lock_or_retry to accept vm_fault and return vm_fault_t which simplifies later patches. Finally allow swap and uffd page faults to be handled under per-VMA locks by dropping per-VMA and retrying, the same way it's done under mmap_lock. Naturally, once VMA lock is dropped that VMA should be assumed unstable and can't be used. This patch (of 6): Commit [1] introduced IO polling support duding swapin to reduce swap read latency for block devices that can be polled. However later commit [2] removed polling support. Therefore it seems safe to remove do_poll parameter in read_swap_cache_async and always call swap_readpage with synchronous=false waiting for IO completion in folio_lock_or_retry. [1] commit 23955622ff8d ("swap: add block io poll in swapin path") [2] commit 9650b453a3d4 ("block: ignore RWF_HIPRI hint for sync dio") Link: https://lkml.kernel.org/r/20230630211957.1341547-1-surenb@google.com Link: https://lkml.kernel.org/r/20230630211957.1341547-2-surenb@google.com Signed-off-by: Suren Baghdasaryan <surenb@google.com> Suggested-by: "Huang, Ying" <ying.huang@intel.com> Reviewed-by: "Huang, Ying" <ying.huang@intel.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Cc: Alistair Popple <apopple@nvidia.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Christian Brauner <brauner@kernel.org> Cc: David Hildenbrand <david@redhat.com> Cc: David Howells <dhowells@redhat.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Hillf Danton <hdanton@sina.com> Cc: Hugh Dickins <hughd@google.com> Cc: Jan Kara <jack@suse.cz> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Josef Bacik <josef@toxicpanda.com> Cc: Laurent Dufour <ldufour@linux.ibm.com> Cc: Liam R. Howlett <Liam.Howlett@oracle.com> Cc: Lorenzo Stoakes <lstoakes@gmail.com> Cc: Matthew Wilcox <willy@infradead.org> Cc: Michal Hocko <mhocko@suse.com> Cc: Michel Lespinasse <michel@lespinasse.org> Cc: Minchan Kim <minchan@google.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Xu <peterx@redhat.com> Cc: Punit Agrawal <punit.agrawal@bytedance.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Yu Zhao <yuzhao@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-06-30 21:19:52 +00:00
bool page_allocated;
swap: add a simple detector for inappropriate swapin readahead This is a patch to improve swap readahead algorithm. It's from Hugh and I slightly changed it. Hugh's original changelog: swapin readahead does a blind readahead, whether or not the swapin is sequential. This may be ok on harddisk, because large reads have relatively small costs, and if the readahead pages are unneeded they can be reclaimed easily - though, what if their allocation forced reclaim of useful pages? But on SSD devices large reads are more expensive than small ones: if the readahead pages are unneeded, reading them in caused significant overhead. This patch adds very simplistic random read detection. Stealing the PageReadahead technique from Konstantin Khlebnikov's patch, avoiding the vma/anon_vma sophistications of Shaohua Li's patch, swapin_nr_pages() simply looks at readahead's current success rate, and narrows or widens its readahead window accordingly. There is little science to its heuristic: it's about as stupid as can be whilst remaining effective. The table below shows elapsed times (in centiseconds) when running a single repetitive swapping load across a 1000MB mapping in 900MB ram with 1GB swap (the harddisk tests had taken painfully too long when I used mem=500M, but SSD shows similar results for that). Vanilla is the 3.6-rc7 kernel on which I started; Shaohua denotes his Sep 3 patch in mmotm and linux-next; HughOld denotes my Oct 1 patch which Shaohua showed to be defective; HughNew this Nov 14 patch, with page_cluster as usual at default of 3 (8-page reads); HughPC4 this same patch with page_cluster 4 (16-page reads); HughPC0 with page_cluster 0 (1-page reads: no readahead). HDD for swapping to harddisk, SSD for swapping to VertexII SSD. Seq for sequential access to the mapping, cycling five times around; Rand for the same number of random touches. Anon for a MAP_PRIVATE anon mapping; Shmem for a MAP_SHARED anon mapping, equivalent to tmpfs. One weakness of Shaohua's vma/anon_vma approach was that it did not optimize Shmem: seen below. Konstantin's approach was perhaps mistuned, 50% slower on Seq: did not compete and is not shown below. HDD Vanilla Shaohua HughOld HughNew HughPC4 HughPC0 Seq Anon 73921 76210 75611 76904 78191 121542 Seq Shmem 73601 73176 73855 72947 74543 118322 Rand Anon 895392 831243 871569 845197 846496 841680 Rand Shmem 1058375 1053486 827935 764955 764376 756489 SSD Vanilla Shaohua HughOld HughNew HughPC4 HughPC0 Seq Anon 24634 24198 24673 25107 21614 70018 Seq Shmem 24959 24932 25052 25703 22030 69678 Rand Anon 43014 26146 28075 25989 26935 25901 Rand Shmem 45349 45215 28249 24268 24138 24332 These tests are, of course, two extremes of a very simple case: under heavier mixed loads I've not yet observed any consistent improvement or degradation, and wider testing would be welcome. Shaohua Li: Test shows Vanilla is slightly better in sequential workload than Hugh's patch. I observed with Hugh's patch sometimes the readahead size is shrinked too fast (from 8 to 1 immediately) in sequential workload if there is no hit. And in such case, continuing doing readahead is good actually. I don't prepare a sophisticated algorithm for the sequential workload because so far we can't guarantee sequential accessed pages are swap out sequentially. So I slightly change Hugh's heuristic - don't shrink readahead size too fast. Here is my test result (unit second, 3 runs average): Vanilla Hugh New Seq 356 370 360 Random 4525 2447 2444 Attached graph is the swapin/swapout throughput I collected with 'vmstat 2'. The first part is running a random workload (till around 1200 of the x-axis) and the second part is running a sequential workload. swapin and swapout throughput are almost identical in steady state in both workloads. These are expected behavior. while in Vanilla, swapin is much bigger than swapout especially in random workload (because wrong readahead). Original patches by: Shaohua Li and Konstantin Khlebnikov. [fengguang.wu@intel.com: swapin_nr_pages() can be static] Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: Fengguang Wu <fengguang.wu@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Konstantin Khlebnikov <khlebnikov@openvz.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-02-06 20:04:21 +00:00
mask = swapin_nr_pages(offset) - 1;
if (!mask)
goto skip;
/* Read a page_cluster sized and aligned cluster around offset. */
start_offset = offset & ~mask;
end_offset = offset | mask;
if (!start_offset) /* First page is swap header. */
start_offset++;
mm, swap: fix false error message in __swp_swapcount() When a page fault occurs for a swap entry, the physical swap readahead (not the VMA base swap readahead) may readahead several swap entries after the fault swap entry. The readahead algorithm calculates some of the swap entries to readahead via increasing the offset of the fault swap entry without checking whether they are beyond the end of the swap device and it relys on the __swp_swapcount() and swapcache_prepare() to check it. Although __swp_swapcount() checks for the swap entry passed in, it will complain with the error message as follow for the expected invalid swap entry. This may make the end users confused. swap_info_get: Bad swap offset entry 0200f8a7 To fix the false error message, the swap entry checking is added in swapin_readahead() to avoid to pass the out-of-bound swap entries and the swap entry reserved for the swap header to __swp_swapcount() and swapcache_prepare(). Link: http://lkml.kernel.org/r/20171102054225.22897-1-ying.huang@intel.com Fixes: e8c26ab60598 ("mm/swap: skip readahead for unreferenced swap slots") Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Reported-by: Christian Kujau <lists@nerdbynature.de> Acked-by: Minchan Kim <minchan@kernel.org> Suggested-by: Minchan Kim <minchan@kernel.org> Cc: Tim Chen <tim.c.chen@linux.intel.com> Cc: Michal Hocko <mhocko@suse.com> Cc: Hugh Dickins <hughd@google.com> Cc: <stable@vger.kernel.org> [4.11+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-11-16 01:33:15 +00:00
if (end_offset >= si->max)
end_offset = si->max - 1;
swap: allow swap readahead to be merged Swap readahead works fine, but the I/O to disk is almost always done in page size requests, despite the fact that readahead submits 1<<page-cluster pages at a time. On older kernels the old per device plugging behavior might have captured this and merged the requests, but currently all comes down to much more I/Os than required. On a single device this might not be an issue, but as soon as a server runs on shared san resources savin I/Os not only improves swapin throughput but also provides a lower resource utilization. With a load running KVM in a lot of memory overcommitment (the hot memory is 1.5 times the host memory) swapping throughput improves significantly and the lead feels more responsive as well as achieves more throughput. In a test setup with 16 swap disks running blocktrace on one of those disks shows the improved merging: Prior: Reads Queued: 560,888, 2,243MiB Writes Queued: 226,242, 904,968KiB Read Dispatches: 544,701, 2,243MiB Write Dispatches: 159,318, 904,968KiB Reads Requeued: 0 Writes Requeued: 0 Reads Completed: 544,716, 2,243MiB Writes Completed: 159,321, 904,980KiB Read Merges: 16,187, 64,748KiB Write Merges: 61,744, 246,976KiB IO unplugs: 149,614 Timer unplugs: 2,940 With the patch: Reads Queued: 734,315, 2,937MiB Writes Queued: 300,188, 1,200MiB Read Dispatches: 214,972, 2,937MiB Write Dispatches: 215,176, 1,200MiB Reads Requeued: 0 Writes Requeued: 0 Reads Completed: 214,971, 2,937MiB Writes Completed: 215,177, 1,200MiB Read Merges: 519,343, 2,077MiB Write Merges: 73,325, 293,300KiB IO unplugs: 337,130 Timer unplugs: 11,184 I got ~10% to ~40% more throughput in my cases and at the same time much lower cpu consumption when broken down per transferred kilobyte (the majority of that due to saved interrupts and better cache handling). In a shared SAN others might get an additional benefit as well, because this now causes less protocol overhead. Signed-off-by: Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com> Acked-by: Rik van Riel <riel@redhat.com> Acked-by: Jens Axboe <axboe@kernel.dk> Reviewed-by: Minchan Kim <minchan@kernel.org> Cc: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-07-31 23:41:44 +00:00
blk_start_plug(&plug);
for (offset = start_offset; offset <= end_offset ; offset++) {
/* Ok, do the async read-ahead now */
folio = __read_swap_cache_async(
mempolicy: alloc_pages_mpol() for NUMA policy without vma Shrink shmem's stack usage by eliminating the pseudo-vma from its folio allocation. alloc_pages_mpol(gfp, order, pol, ilx, nid) becomes the principal actor for passing mempolicy choice down to __alloc_pages(), rather than vma_alloc_folio(gfp, order, vma, addr, hugepage). vma_alloc_folio() and alloc_pages() remain, but as wrappers around alloc_pages_mpol(). alloc_pages_bulk_*() untouched, except to provide the additional args to policy_nodemask(), which subsumes policy_node(). Cleanup throughout, cutting out some unhelpful "helpers". It would all be much simpler without MPOL_INTERLEAVE, but that adds a dynamic to the constant mpol: complicated by v3.6 commit 09c231cb8bfd ("tmpfs: distribute interleave better across nodes"), which added ino bias to the interleave, hidden from mm/mempolicy.c until this commit. Hence "ilx" throughout, the "interleave index". Originally I thought it could be done just with nid, but that's wrong: the nodemask may come from the shared policy layer below a shmem vma, or it may come from the task layer above a shmem vma; and without the final nodemask then nodeid cannot be decided. And how ilx is applied depends also on page order. The interleave index is almost always irrelevant unless MPOL_INTERLEAVE: with one exception in alloc_pages_mpol(), where the NO_INTERLEAVE_INDEX passed down from vma-less alloc_pages() is also used as hint not to use THP-style hugepage allocation - to avoid the overhead of a hugepage arg (though I don't understand why we never just added a GFP bit for THP - if it actually needs a different allocation strategy from other pages of the same order). vma_alloc_folio() still carries its hugepage arg here, but it is not used, and should be removed when agreed. get_vma_policy() no longer allows a NULL vma: over time I believe we've eradicated all the places which used to need it e.g. swapoff and madvise used to pass NULL vma to read_swap_cache_async(), but now know the vma. [hughd@google.com: handle NULL mpol being passed to __read_swap_cache_async()] Link: https://lkml.kernel.org/r/ea419956-4751-0102-21f7-9c93cb957892@google.com Link: https://lkml.kernel.org/r/74e34633-6060-f5e3-aee-7040d43f2e93@google.com Link: https://lkml.kernel.org/r/1738368e-bac0-fd11-ed7f-b87142a939fe@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Hildenbrand <david@redhat.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Huang Ying <ying.huang@intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Nhat Pham <nphamcs@gmail.com> Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Tejun heo <tj@kernel.org> Cc: Vishal Moola (Oracle) <vishal.moola@gmail.com> Cc: Yang Shi <shy828301@gmail.com> Cc: Yosry Ahmed <yosryahmed@google.com> Cc: Domenico Cerasuolo <mimmocerasuolo@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-10-19 20:39:08 +00:00
swp_entry(swp_type(entry), offset),
zswap: make shrinking memcg-aware Currently, we only have a single global LRU for zswap. This makes it impossible to perform worload-specific shrinking - an memcg cannot determine which pages in the pool it owns, and often ends up writing pages from other memcgs. This issue has been previously observed in practice and mitigated by simply disabling memcg-initiated shrinking: https://lore.kernel.org/all/20230530232435.3097106-1-nphamcs@gmail.com/T/#u This patch fully resolves the issue by replacing the global zswap LRU with memcg- and NUMA-specific LRUs, and modify the reclaim logic: a) When a store attempt hits an memcg limit, it now triggers a synchronous reclaim attempt that, if successful, allows the new hotter page to be accepted by zswap. b) If the store attempt instead hits the global zswap limit, it will trigger an asynchronous reclaim attempt, in which an memcg is selected for reclaim in a round-robin-like fashion. [nphamcs@gmail.com: use correct function for the onlineness check, use mem_cgroup_iter_break()] Link: https://lkml.kernel.org/r/20231205195419.2563217-1-nphamcs@gmail.com [nphamcs@gmail.com: drop the pool's reference at the end of the writeback step] Link: https://lkml.kernel.org/r/20231206030627.4155634-1-nphamcs@gmail.com Link: https://lkml.kernel.org/r/20231130194023.4102148-4-nphamcs@gmail.com Signed-off-by: Domenico Cerasuolo <cerasuolodomenico@gmail.com> Co-developed-by: Nhat Pham <nphamcs@gmail.com> Signed-off-by: Nhat Pham <nphamcs@gmail.com> Tested-by: Bagas Sanjaya <bagasdotme@gmail.com> Cc: Chris Li <chrisl@kernel.org> Cc: Dan Streetman <ddstreet@ieee.org> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Muchun Song <muchun.song@linux.dev> Cc: Roman Gushchin <roman.gushchin@linux.dev> Cc: Seth Jennings <sjenning@redhat.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Shuah Khan <shuah@kernel.org> Cc: Vitaly Wool <vitaly.wool@konsulko.com> Cc: Yosry Ahmed <yosryahmed@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-11-30 19:40:20 +00:00
gfp_mask, mpol, ilx, &page_allocated, false);
if (!folio)
continue;
if (page_allocated) {
swap_read_folio(folio, false, &splug);
if (offset != entry_offset) {
folio_set_readahead(folio);
count_vm_event(SWAP_RA);
}
mm, swap: add swap readahead hit statistics Patch series "mm, swap: VMA based swap readahead", v4. The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory space. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patchset, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. In addition to the swap readahead changes, some new sysfs interface is added to show the efficiency of the readahead algorithm and some other swap statistics. This new implementation will incur more small random read, on SSD, the improved correctness of estimation and readahead target should beat the potential increased overhead, this is also illustrated in the test results below. But on HDD, the overhead may beat the benefit, so the original implementation will be used by default. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. This patch (of 5): The statistics for total readahead pages and total readahead hits are recorded and exported via the following sysfs interface. /sys/kernel/mm/swap/ra_hits /sys/kernel/mm/swap/ra_total With them, the efficiency of the swap readahead could be measured, so that the swap readahead algorithm and parameters could be tuned accordingly. [akpm@linux-foundation.org: don't display swap stats if CONFIG_SWAP=n] Link: http://lkml.kernel.org/r/20170807054038.1843-2-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:29 +00:00
}
folio_put(folio);
}
swap: allow swap readahead to be merged Swap readahead works fine, but the I/O to disk is almost always done in page size requests, despite the fact that readahead submits 1<<page-cluster pages at a time. On older kernels the old per device plugging behavior might have captured this and merged the requests, but currently all comes down to much more I/Os than required. On a single device this might not be an issue, but as soon as a server runs on shared san resources savin I/Os not only improves swapin throughput but also provides a lower resource utilization. With a load running KVM in a lot of memory overcommitment (the hot memory is 1.5 times the host memory) swapping throughput improves significantly and the lead feels more responsive as well as achieves more throughput. In a test setup with 16 swap disks running blocktrace on one of those disks shows the improved merging: Prior: Reads Queued: 560,888, 2,243MiB Writes Queued: 226,242, 904,968KiB Read Dispatches: 544,701, 2,243MiB Write Dispatches: 159,318, 904,968KiB Reads Requeued: 0 Writes Requeued: 0 Reads Completed: 544,716, 2,243MiB Writes Completed: 159,321, 904,980KiB Read Merges: 16,187, 64,748KiB Write Merges: 61,744, 246,976KiB IO unplugs: 149,614 Timer unplugs: 2,940 With the patch: Reads Queued: 734,315, 2,937MiB Writes Queued: 300,188, 1,200MiB Read Dispatches: 214,972, 2,937MiB Write Dispatches: 215,176, 1,200MiB Reads Requeued: 0 Writes Requeued: 0 Reads Completed: 214,971, 2,937MiB Writes Completed: 215,177, 1,200MiB Read Merges: 519,343, 2,077MiB Write Merges: 73,325, 293,300KiB IO unplugs: 337,130 Timer unplugs: 11,184 I got ~10% to ~40% more throughput in my cases and at the same time much lower cpu consumption when broken down per transferred kilobyte (the majority of that due to saved interrupts and better cache handling). In a shared SAN others might get an additional benefit as well, because this now causes less protocol overhead. Signed-off-by: Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com> Acked-by: Rik van Riel <riel@redhat.com> Acked-by: Jens Axboe <axboe@kernel.dk> Reviewed-by: Minchan Kim <minchan@kernel.org> Cc: Hugh Dickins <hughd@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-07-31 23:41:44 +00:00
blk_finish_plug(&plug);
swap_read_unplug(splug);
lru_add_drain(); /* Push any new pages onto the LRU now */
swap: add a simple detector for inappropriate swapin readahead This is a patch to improve swap readahead algorithm. It's from Hugh and I slightly changed it. Hugh's original changelog: swapin readahead does a blind readahead, whether or not the swapin is sequential. This may be ok on harddisk, because large reads have relatively small costs, and if the readahead pages are unneeded they can be reclaimed easily - though, what if their allocation forced reclaim of useful pages? But on SSD devices large reads are more expensive than small ones: if the readahead pages are unneeded, reading them in caused significant overhead. This patch adds very simplistic random read detection. Stealing the PageReadahead technique from Konstantin Khlebnikov's patch, avoiding the vma/anon_vma sophistications of Shaohua Li's patch, swapin_nr_pages() simply looks at readahead's current success rate, and narrows or widens its readahead window accordingly. There is little science to its heuristic: it's about as stupid as can be whilst remaining effective. The table below shows elapsed times (in centiseconds) when running a single repetitive swapping load across a 1000MB mapping in 900MB ram with 1GB swap (the harddisk tests had taken painfully too long when I used mem=500M, but SSD shows similar results for that). Vanilla is the 3.6-rc7 kernel on which I started; Shaohua denotes his Sep 3 patch in mmotm and linux-next; HughOld denotes my Oct 1 patch which Shaohua showed to be defective; HughNew this Nov 14 patch, with page_cluster as usual at default of 3 (8-page reads); HughPC4 this same patch with page_cluster 4 (16-page reads); HughPC0 with page_cluster 0 (1-page reads: no readahead). HDD for swapping to harddisk, SSD for swapping to VertexII SSD. Seq for sequential access to the mapping, cycling five times around; Rand for the same number of random touches. Anon for a MAP_PRIVATE anon mapping; Shmem for a MAP_SHARED anon mapping, equivalent to tmpfs. One weakness of Shaohua's vma/anon_vma approach was that it did not optimize Shmem: seen below. Konstantin's approach was perhaps mistuned, 50% slower on Seq: did not compete and is not shown below. HDD Vanilla Shaohua HughOld HughNew HughPC4 HughPC0 Seq Anon 73921 76210 75611 76904 78191 121542 Seq Shmem 73601 73176 73855 72947 74543 118322 Rand Anon 895392 831243 871569 845197 846496 841680 Rand Shmem 1058375 1053486 827935 764955 764376 756489 SSD Vanilla Shaohua HughOld HughNew HughPC4 HughPC0 Seq Anon 24634 24198 24673 25107 21614 70018 Seq Shmem 24959 24932 25052 25703 22030 69678 Rand Anon 43014 26146 28075 25989 26935 25901 Rand Shmem 45349 45215 28249 24268 24138 24332 These tests are, of course, two extremes of a very simple case: under heavier mixed loads I've not yet observed any consistent improvement or degradation, and wider testing would be welcome. Shaohua Li: Test shows Vanilla is slightly better in sequential workload than Hugh's patch. I observed with Hugh's patch sometimes the readahead size is shrinked too fast (from 8 to 1 immediately) in sequential workload if there is no hit. And in such case, continuing doing readahead is good actually. I don't prepare a sophisticated algorithm for the sequential workload because so far we can't guarantee sequential accessed pages are swap out sequentially. So I slightly change Hugh's heuristic - don't shrink readahead size too fast. Here is my test result (unit second, 3 runs average): Vanilla Hugh New Seq 356 370 360 Random 4525 2447 2444 Attached graph is the swapin/swapout throughput I collected with 'vmstat 2'. The first part is running a random workload (till around 1200 of the x-axis) and the second part is running a sequential workload. swapin and swapout throughput are almost identical in steady state in both workloads. These are expected behavior. while in Vanilla, swapin is much bigger than swapout especially in random workload (because wrong readahead). Original patches by: Shaohua Li and Konstantin Khlebnikov. [fengguang.wu@intel.com: swapin_nr_pages() can be static] Signed-off-by: Hugh Dickins <hughd@google.com> Signed-off-by: Shaohua Li <shli@fusionio.com> Signed-off-by: Fengguang Wu <fengguang.wu@intel.com> Cc: Rik van Riel <riel@redhat.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Konstantin Khlebnikov <khlebnikov@openvz.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-02-06 20:04:21 +00:00
skip:
/* The page was likely read above, so no need for plugging here */
folio = __read_swap_cache_async(entry, gfp_mask, mpol, ilx,
zswap: make shrinking memcg-aware Currently, we only have a single global LRU for zswap. This makes it impossible to perform worload-specific shrinking - an memcg cannot determine which pages in the pool it owns, and often ends up writing pages from other memcgs. This issue has been previously observed in practice and mitigated by simply disabling memcg-initiated shrinking: https://lore.kernel.org/all/20230530232435.3097106-1-nphamcs@gmail.com/T/#u This patch fully resolves the issue by replacing the global zswap LRU with memcg- and NUMA-specific LRUs, and modify the reclaim logic: a) When a store attempt hits an memcg limit, it now triggers a synchronous reclaim attempt that, if successful, allows the new hotter page to be accepted by zswap. b) If the store attempt instead hits the global zswap limit, it will trigger an asynchronous reclaim attempt, in which an memcg is selected for reclaim in a round-robin-like fashion. [nphamcs@gmail.com: use correct function for the onlineness check, use mem_cgroup_iter_break()] Link: https://lkml.kernel.org/r/20231205195419.2563217-1-nphamcs@gmail.com [nphamcs@gmail.com: drop the pool's reference at the end of the writeback step] Link: https://lkml.kernel.org/r/20231206030627.4155634-1-nphamcs@gmail.com Link: https://lkml.kernel.org/r/20231130194023.4102148-4-nphamcs@gmail.com Signed-off-by: Domenico Cerasuolo <cerasuolodomenico@gmail.com> Co-developed-by: Nhat Pham <nphamcs@gmail.com> Signed-off-by: Nhat Pham <nphamcs@gmail.com> Tested-by: Bagas Sanjaya <bagasdotme@gmail.com> Cc: Chris Li <chrisl@kernel.org> Cc: Dan Streetman <ddstreet@ieee.org> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Muchun Song <muchun.song@linux.dev> Cc: Roman Gushchin <roman.gushchin@linux.dev> Cc: Seth Jennings <sjenning@redhat.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Shuah Khan <shuah@kernel.org> Cc: Vitaly Wool <vitaly.wool@konsulko.com> Cc: Yosry Ahmed <yosryahmed@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-11-30 19:40:20 +00:00
&page_allocated, false);
mm/swap_state: update zswap LRU's protection range with the folio locked When a folio is swapped in, the protection size of the corresponding zswap LRU is incremented, so that the zswap shrinker is more conservative with its reclaiming action. This field is embedded within the struct lruvec, so updating it requires looking up the folio's memcg and lruvec. However, currently this lookup can happen after the folio is unlocked, for instance if a new folio is allocated, and swap_read_folio() unlocks the folio before returning. In this scenario, there is no stability guarantee for the binding between a folio and its memcg and lruvec: * A folio's memcg and lruvec can be freed between the lookup and the update, leading to a UAF. * Folio migration can clear the now-unlocked folio's memcg_data, which directs the zswap LRU protection size update towards the root memcg instead of the original memcg. This was recently picked up by the syzbot thanks to a warning in the inlined folio_lruvec() call. Move the zswap LRU protection range update above the swap_read_folio() call, and only when a new page is allocated, to prevent this. [nphamcs@gmail.com: add VM_WARN_ON_ONCE() to zswap_folio_swapin()] Link: https://lkml.kernel.org/r/20240206180855.3987204-1-nphamcs@gmail.com [nphamcs@gmail.com: remove unneeded if (folio) checks] Link: https://lkml.kernel.org/r/20240206191355.83755-1-nphamcs@gmail.com Link: https://lkml.kernel.org/r/20240205232442.3240571-1-nphamcs@gmail.com Fixes: b5ba474f3f51 ("zswap: shrink zswap pool based on memory pressure") Reported-by: syzbot+17a611d10af7d18a7092@syzkaller.appspotmail.com Closes: https://lore.kernel.org/all/000000000000ae47f90610803260@google.com/ Signed-off-by: Nhat Pham <nphamcs@gmail.com> Reviewed-by: Chengming Zhou <zhouchengming@bytedance.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Cc: Yosry Ahmed <yosryahmed@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-02-05 23:24:42 +00:00
if (unlikely(page_allocated)) {
zswap_folio_swapin(folio);
swap_read_folio(folio, false, NULL);
mm/swap_state: update zswap LRU's protection range with the folio locked When a folio is swapped in, the protection size of the corresponding zswap LRU is incremented, so that the zswap shrinker is more conservative with its reclaiming action. This field is embedded within the struct lruvec, so updating it requires looking up the folio's memcg and lruvec. However, currently this lookup can happen after the folio is unlocked, for instance if a new folio is allocated, and swap_read_folio() unlocks the folio before returning. In this scenario, there is no stability guarantee for the binding between a folio and its memcg and lruvec: * A folio's memcg and lruvec can be freed between the lookup and the update, leading to a UAF. * Folio migration can clear the now-unlocked folio's memcg_data, which directs the zswap LRU protection size update towards the root memcg instead of the original memcg. This was recently picked up by the syzbot thanks to a warning in the inlined folio_lruvec() call. Move the zswap LRU protection range update above the swap_read_folio() call, and only when a new page is allocated, to prevent this. [nphamcs@gmail.com: add VM_WARN_ON_ONCE() to zswap_folio_swapin()] Link: https://lkml.kernel.org/r/20240206180855.3987204-1-nphamcs@gmail.com [nphamcs@gmail.com: remove unneeded if (folio) checks] Link: https://lkml.kernel.org/r/20240206191355.83755-1-nphamcs@gmail.com Link: https://lkml.kernel.org/r/20240205232442.3240571-1-nphamcs@gmail.com Fixes: b5ba474f3f51 ("zswap: shrink zswap pool based on memory pressure") Reported-by: syzbot+17a611d10af7d18a7092@syzkaller.appspotmail.com Closes: https://lore.kernel.org/all/000000000000ae47f90610803260@google.com/ Signed-off-by: Nhat Pham <nphamcs@gmail.com> Reviewed-by: Chengming Zhou <zhouchengming@bytedance.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Cc: Yosry Ahmed <yosryahmed@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-02-05 23:24:42 +00:00
}
return folio;
}
mm/swap: split swap cache into 64MB trunks The patch is to improve the scalability of the swap out/in via using fine grained locks for the swap cache. In current kernel, one address space will be used for each swap device. And in the common configuration, the number of the swap device is very small (one is typical). This causes the heavy lock contention on the radix tree of the address space if multiple tasks swap out/in concurrently. But in fact, there is no dependency between pages in the swap cache. So that, we can split the one shared address space for each swap device into several address spaces to reduce the lock contention. In the patch, the shared address space is split into 64MB trunks. 64MB is chosen to balance the memory space usage and effect of lock contention reduction. The size of struct address_space on x86_64 architecture is 408B, so with the patch, 6528B more memory will be used for every 1GB swap space on x86_64 architecture. One address space is still shared for the swap entries in the same 64M trunks. To avoid lock contention for the first round of swap space allocation, the order of the swap clusters in the initial free clusters list is changed. The swap space distance between the consecutive swap clusters in the free cluster list is at least 64M. After the first round of allocation, the swap clusters are expected to be freed randomly, so the lock contention should be reduced effectively. Link: http://lkml.kernel.org/r/735bab895e64c930581ffb0a05b661e01da82bc5.1484082593.git.tim.c.chen@linux.intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com> Cc: Aaron Lu <aaron.lu@intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Christian Borntraeger <borntraeger@de.ibm.com> Cc: Dave Hansen <dave.hansen@intel.com> Cc: Hillf Danton <hillf.zj@alibaba-inc.com> Cc: Huang Ying <ying.huang@intel.com> Cc: Hugh Dickins <hughd@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Jonathan Corbet <corbet@lwn.net> escreveu: Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Vladimir Davydov <vdavydov.dev@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-02-22 23:45:26 +00:00
int init_swap_address_space(unsigned int type, unsigned long nr_pages)
{
struct address_space *spaces, *space;
unsigned int i, nr;
nr = DIV_ROUND_UP(nr_pages, SWAP_ADDRESS_SPACE_PAGES);
treewide: kvzalloc() -> kvcalloc() The kvzalloc() function has a 2-factor argument form, kvcalloc(). This patch replaces cases of: kvzalloc(a * b, gfp) with: kvcalloc(a * b, gfp) as well as handling cases of: kvzalloc(a * b * c, gfp) with: kvzalloc(array3_size(a, b, c), gfp) as it's slightly less ugly than: kvcalloc(array_size(a, b), c, gfp) This does, however, attempt to ignore constant size factors like: kvzalloc(4 * 1024, gfp) though any constants defined via macros get caught up in the conversion. Any factors with a sizeof() of "unsigned char", "char", and "u8" were dropped, since they're redundant. The Coccinelle script used for this was: // Fix redundant parens around sizeof(). @@ type TYPE; expression THING, E; @@ ( kvzalloc( - (sizeof(TYPE)) * E + sizeof(TYPE) * E , ...) | kvzalloc( - (sizeof(THING)) * E + sizeof(THING) * E , ...) ) // Drop single-byte sizes and redundant parens. @@ expression COUNT; typedef u8; typedef __u8; @@ ( kvzalloc( - sizeof(u8) * (COUNT) + COUNT , ...) | kvzalloc( - sizeof(__u8) * (COUNT) + COUNT , ...) | kvzalloc( - sizeof(char) * (COUNT) + COUNT , ...) | kvzalloc( - sizeof(unsigned char) * (COUNT) + COUNT , ...) | kvzalloc( - sizeof(u8) * COUNT + COUNT , ...) | kvzalloc( - sizeof(__u8) * COUNT + COUNT , ...) | kvzalloc( - sizeof(char) * COUNT + COUNT , ...) | kvzalloc( - sizeof(unsigned char) * COUNT + COUNT , ...) ) // 2-factor product with sizeof(type/expression) and identifier or constant. @@ type TYPE; expression THING; identifier COUNT_ID; constant COUNT_CONST; @@ ( - kvzalloc + kvcalloc ( - sizeof(TYPE) * (COUNT_ID) + COUNT_ID, sizeof(TYPE) , ...) | - kvzalloc + kvcalloc ( - sizeof(TYPE) * COUNT_ID + COUNT_ID, sizeof(TYPE) , ...) | - kvzalloc + kvcalloc ( - sizeof(TYPE) * (COUNT_CONST) + COUNT_CONST, sizeof(TYPE) , ...) | - kvzalloc + kvcalloc ( - sizeof(TYPE) * COUNT_CONST + COUNT_CONST, sizeof(TYPE) , ...) | - kvzalloc + kvcalloc ( - sizeof(THING) * (COUNT_ID) + COUNT_ID, sizeof(THING) , ...) | - kvzalloc + kvcalloc ( - sizeof(THING) * COUNT_ID + COUNT_ID, sizeof(THING) , ...) | - kvzalloc + kvcalloc ( - sizeof(THING) * (COUNT_CONST) + COUNT_CONST, sizeof(THING) , ...) | - kvzalloc + kvcalloc ( - sizeof(THING) * COUNT_CONST + COUNT_CONST, sizeof(THING) , ...) ) // 2-factor product, only identifiers. @@ identifier SIZE, COUNT; @@ - kvzalloc + kvcalloc ( - SIZE * COUNT + COUNT, SIZE , ...) // 3-factor product with 1 sizeof(type) or sizeof(expression), with // redundant parens removed. @@ expression THING; identifier STRIDE, COUNT; type TYPE; @@ ( kvzalloc( - sizeof(TYPE) * (COUNT) * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | kvzalloc( - sizeof(TYPE) * (COUNT) * STRIDE + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | kvzalloc( - sizeof(TYPE) * COUNT * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | kvzalloc( - sizeof(TYPE) * COUNT * STRIDE + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | kvzalloc( - sizeof(THING) * (COUNT) * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) | kvzalloc( - sizeof(THING) * (COUNT) * STRIDE + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) | kvzalloc( - sizeof(THING) * COUNT * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) | kvzalloc( - sizeof(THING) * COUNT * STRIDE + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) ) // 3-factor product with 2 sizeof(variable), with redundant parens removed. @@ expression THING1, THING2; identifier COUNT; type TYPE1, TYPE2; @@ ( kvzalloc( - sizeof(TYPE1) * sizeof(TYPE2) * COUNT + array3_size(COUNT, sizeof(TYPE1), sizeof(TYPE2)) , ...) | kvzalloc( - sizeof(TYPE1) * sizeof(THING2) * (COUNT) + array3_size(COUNT, sizeof(TYPE1), sizeof(TYPE2)) , ...) | kvzalloc( - sizeof(THING1) * sizeof(THING2) * COUNT + array3_size(COUNT, sizeof(THING1), sizeof(THING2)) , ...) | kvzalloc( - sizeof(THING1) * sizeof(THING2) * (COUNT) + array3_size(COUNT, sizeof(THING1), sizeof(THING2)) , ...) | kvzalloc( - sizeof(TYPE1) * sizeof(THING2) * COUNT + array3_size(COUNT, sizeof(TYPE1), sizeof(THING2)) , ...) | kvzalloc( - sizeof(TYPE1) * sizeof(THING2) * (COUNT) + array3_size(COUNT, sizeof(TYPE1), sizeof(THING2)) , ...) ) // 3-factor product, only identifiers, with redundant parens removed. @@ identifier STRIDE, SIZE, COUNT; @@ ( kvzalloc( - (COUNT) * STRIDE * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) | kvzalloc( - COUNT * (STRIDE) * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) | kvzalloc( - COUNT * STRIDE * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | kvzalloc( - (COUNT) * (STRIDE) * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) | kvzalloc( - COUNT * (STRIDE) * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | kvzalloc( - (COUNT) * STRIDE * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | kvzalloc( - (COUNT) * (STRIDE) * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | kvzalloc( - COUNT * STRIDE * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) ) // Any remaining multi-factor products, first at least 3-factor products, // when they're not all constants... @@ expression E1, E2, E3; constant C1, C2, C3; @@ ( kvzalloc(C1 * C2 * C3, ...) | kvzalloc( - (E1) * E2 * E3 + array3_size(E1, E2, E3) , ...) | kvzalloc( - (E1) * (E2) * E3 + array3_size(E1, E2, E3) , ...) | kvzalloc( - (E1) * (E2) * (E3) + array3_size(E1, E2, E3) , ...) | kvzalloc( - E1 * E2 * E3 + array3_size(E1, E2, E3) , ...) ) // And then all remaining 2 factors products when they're not all constants, // keeping sizeof() as the second factor argument. @@ expression THING, E1, E2; type TYPE; constant C1, C2, C3; @@ ( kvzalloc(sizeof(THING) * C2, ...) | kvzalloc(sizeof(TYPE) * C2, ...) | kvzalloc(C1 * C2 * C3, ...) | kvzalloc(C1 * C2, ...) | - kvzalloc + kvcalloc ( - sizeof(TYPE) * (E2) + E2, sizeof(TYPE) , ...) | - kvzalloc + kvcalloc ( - sizeof(TYPE) * E2 + E2, sizeof(TYPE) , ...) | - kvzalloc + kvcalloc ( - sizeof(THING) * (E2) + E2, sizeof(THING) , ...) | - kvzalloc + kvcalloc ( - sizeof(THING) * E2 + E2, sizeof(THING) , ...) | - kvzalloc + kvcalloc ( - (E1) * E2 + E1, E2 , ...) | - kvzalloc + kvcalloc ( - (E1) * (E2) + E1, E2 , ...) | - kvzalloc + kvcalloc ( - E1 * E2 + E1, E2 , ...) ) Signed-off-by: Kees Cook <keescook@chromium.org>
2018-06-12 21:04:48 +00:00
spaces = kvcalloc(nr, sizeof(struct address_space), GFP_KERNEL);
mm/swap: split swap cache into 64MB trunks The patch is to improve the scalability of the swap out/in via using fine grained locks for the swap cache. In current kernel, one address space will be used for each swap device. And in the common configuration, the number of the swap device is very small (one is typical). This causes the heavy lock contention on the radix tree of the address space if multiple tasks swap out/in concurrently. But in fact, there is no dependency between pages in the swap cache. So that, we can split the one shared address space for each swap device into several address spaces to reduce the lock contention. In the patch, the shared address space is split into 64MB trunks. 64MB is chosen to balance the memory space usage and effect of lock contention reduction. The size of struct address_space on x86_64 architecture is 408B, so with the patch, 6528B more memory will be used for every 1GB swap space on x86_64 architecture. One address space is still shared for the swap entries in the same 64M trunks. To avoid lock contention for the first round of swap space allocation, the order of the swap clusters in the initial free clusters list is changed. The swap space distance between the consecutive swap clusters in the free cluster list is at least 64M. After the first round of allocation, the swap clusters are expected to be freed randomly, so the lock contention should be reduced effectively. Link: http://lkml.kernel.org/r/735bab895e64c930581ffb0a05b661e01da82bc5.1484082593.git.tim.c.chen@linux.intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com> Cc: Aaron Lu <aaron.lu@intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Christian Borntraeger <borntraeger@de.ibm.com> Cc: Dave Hansen <dave.hansen@intel.com> Cc: Hillf Danton <hillf.zj@alibaba-inc.com> Cc: Huang Ying <ying.huang@intel.com> Cc: Hugh Dickins <hughd@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Jonathan Corbet <corbet@lwn.net> escreveu: Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Vladimir Davydov <vdavydov.dev@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-02-22 23:45:26 +00:00
if (!spaces)
return -ENOMEM;
for (i = 0; i < nr; i++) {
space = spaces + i;
xa_init_flags(&space->i_pages, XA_FLAGS_LOCK_IRQ);
mm/swap: split swap cache into 64MB trunks The patch is to improve the scalability of the swap out/in via using fine grained locks for the swap cache. In current kernel, one address space will be used for each swap device. And in the common configuration, the number of the swap device is very small (one is typical). This causes the heavy lock contention on the radix tree of the address space if multiple tasks swap out/in concurrently. But in fact, there is no dependency between pages in the swap cache. So that, we can split the one shared address space for each swap device into several address spaces to reduce the lock contention. In the patch, the shared address space is split into 64MB trunks. 64MB is chosen to balance the memory space usage and effect of lock contention reduction. The size of struct address_space on x86_64 architecture is 408B, so with the patch, 6528B more memory will be used for every 1GB swap space on x86_64 architecture. One address space is still shared for the swap entries in the same 64M trunks. To avoid lock contention for the first round of swap space allocation, the order of the swap clusters in the initial free clusters list is changed. The swap space distance between the consecutive swap clusters in the free cluster list is at least 64M. After the first round of allocation, the swap clusters are expected to be freed randomly, so the lock contention should be reduced effectively. Link: http://lkml.kernel.org/r/735bab895e64c930581ffb0a05b661e01da82bc5.1484082593.git.tim.c.chen@linux.intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com> Cc: Aaron Lu <aaron.lu@intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Christian Borntraeger <borntraeger@de.ibm.com> Cc: Dave Hansen <dave.hansen@intel.com> Cc: Hillf Danton <hillf.zj@alibaba-inc.com> Cc: Huang Ying <ying.huang@intel.com> Cc: Hugh Dickins <hughd@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Jonathan Corbet <corbet@lwn.net> escreveu: Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Vladimir Davydov <vdavydov.dev@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-02-22 23:45:26 +00:00
atomic_set(&space->i_mmap_writable, 0);
space->a_ops = &swap_aops;
/* swap cache doesn't use writeback related tags */
mapping_set_no_writeback_tags(space);
}
nr_swapper_spaces[type] = nr;
mm/swap_state.c: simplify total_swapcache_pages() with get_swap_device() total_swapcache_pages() may race with swapper_spaces[] allocation and freeing. Previously, this is protected with a swapper_spaces[] specific RCU mechanism. To simplify the logic/code complexity, it is replaced with get/put_swap_device(). The code line number is reduced too. Although not so important, the swapoff() performance improves too because one synchronize_rcu() call during swapoff() is deleted. [ying.huang@intel.com: fix bad swap file entry warning] Link: http://lkml.kernel.org/r/20190531024102.21723-1-ying.huang@intel.com Link: http://lkml.kernel.org/r/20190527082714.12151-1-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Reviewed-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Mike Kravetz <mike.kravetz@oracle.com> Cc: Hugh Dickins <hughd@google.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Tim Chen <tim.c.chen@linux.intel.com> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Jérôme Glisse <jglisse@redhat.com> Cc: Michal Hocko <mhocko@suse.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Yang Shi <yang.shi@linux.alibaba.com> Cc: David Rientjes <rientjes@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Jan Kara <jack@suse.cz> Cc: Dave Jiang <dave.jiang@intel.com> Cc: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Andrea Parri <andrea.parri@amarulasolutions.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-07-12 03:55:37 +00:00
swapper_spaces[type] = spaces;
mm/swap: split swap cache into 64MB trunks The patch is to improve the scalability of the swap out/in via using fine grained locks for the swap cache. In current kernel, one address space will be used for each swap device. And in the common configuration, the number of the swap device is very small (one is typical). This causes the heavy lock contention on the radix tree of the address space if multiple tasks swap out/in concurrently. But in fact, there is no dependency between pages in the swap cache. So that, we can split the one shared address space for each swap device into several address spaces to reduce the lock contention. In the patch, the shared address space is split into 64MB trunks. 64MB is chosen to balance the memory space usage and effect of lock contention reduction. The size of struct address_space on x86_64 architecture is 408B, so with the patch, 6528B more memory will be used for every 1GB swap space on x86_64 architecture. One address space is still shared for the swap entries in the same 64M trunks. To avoid lock contention for the first round of swap space allocation, the order of the swap clusters in the initial free clusters list is changed. The swap space distance between the consecutive swap clusters in the free cluster list is at least 64M. After the first round of allocation, the swap clusters are expected to be freed randomly, so the lock contention should be reduced effectively. Link: http://lkml.kernel.org/r/735bab895e64c930581ffb0a05b661e01da82bc5.1484082593.git.tim.c.chen@linux.intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com> Cc: Aaron Lu <aaron.lu@intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Christian Borntraeger <borntraeger@de.ibm.com> Cc: Dave Hansen <dave.hansen@intel.com> Cc: Hillf Danton <hillf.zj@alibaba-inc.com> Cc: Huang Ying <ying.huang@intel.com> Cc: Hugh Dickins <hughd@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Jonathan Corbet <corbet@lwn.net> escreveu: Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Vladimir Davydov <vdavydov.dev@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-02-22 23:45:26 +00:00
return 0;
}
void exit_swap_address_space(unsigned int type)
{
int i;
struct address_space *spaces = swapper_spaces[type];
for (i = 0; i < nr_swapper_spaces[type]; i++)
VM_WARN_ON_ONCE(!mapping_empty(&spaces[i]));
kvfree(spaces);
mm/swap: split swap cache into 64MB trunks The patch is to improve the scalability of the swap out/in via using fine grained locks for the swap cache. In current kernel, one address space will be used for each swap device. And in the common configuration, the number of the swap device is very small (one is typical). This causes the heavy lock contention on the radix tree of the address space if multiple tasks swap out/in concurrently. But in fact, there is no dependency between pages in the swap cache. So that, we can split the one shared address space for each swap device into several address spaces to reduce the lock contention. In the patch, the shared address space is split into 64MB trunks. 64MB is chosen to balance the memory space usage and effect of lock contention reduction. The size of struct address_space on x86_64 architecture is 408B, so with the patch, 6528B more memory will be used for every 1GB swap space on x86_64 architecture. One address space is still shared for the swap entries in the same 64M trunks. To avoid lock contention for the first round of swap space allocation, the order of the swap clusters in the initial free clusters list is changed. The swap space distance between the consecutive swap clusters in the free cluster list is at least 64M. After the first round of allocation, the swap clusters are expected to be freed randomly, so the lock contention should be reduced effectively. Link: http://lkml.kernel.org/r/735bab895e64c930581ffb0a05b661e01da82bc5.1484082593.git.tim.c.chen@linux.intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com> Cc: Aaron Lu <aaron.lu@intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Christian Borntraeger <borntraeger@de.ibm.com> Cc: Dave Hansen <dave.hansen@intel.com> Cc: Hillf Danton <hillf.zj@alibaba-inc.com> Cc: Huang Ying <ying.huang@intel.com> Cc: Hugh Dickins <hughd@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Jonathan Corbet <corbet@lwn.net> escreveu: Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Vladimir Davydov <vdavydov.dev@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-02-22 23:45:26 +00:00
nr_swapper_spaces[type] = 0;
mm/swap_state.c: simplify total_swapcache_pages() with get_swap_device() total_swapcache_pages() may race with swapper_spaces[] allocation and freeing. Previously, this is protected with a swapper_spaces[] specific RCU mechanism. To simplify the logic/code complexity, it is replaced with get/put_swap_device(). The code line number is reduced too. Although not so important, the swapoff() performance improves too because one synchronize_rcu() call during swapoff() is deleted. [ying.huang@intel.com: fix bad swap file entry warning] Link: http://lkml.kernel.org/r/20190531024102.21723-1-ying.huang@intel.com Link: http://lkml.kernel.org/r/20190527082714.12151-1-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Reviewed-by: Andrew Morton <akpm@linux-foundation.org> Tested-by: Mike Kravetz <mike.kravetz@oracle.com> Cc: Hugh Dickins <hughd@google.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Tim Chen <tim.c.chen@linux.intel.com> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Jérôme Glisse <jglisse@redhat.com> Cc: Michal Hocko <mhocko@suse.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Yang Shi <yang.shi@linux.alibaba.com> Cc: David Rientjes <rientjes@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Jan Kara <jack@suse.cz> Cc: Dave Jiang <dave.jiang@intel.com> Cc: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Andrea Parri <andrea.parri@amarulasolutions.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-07-12 03:55:37 +00:00
swapper_spaces[type] = NULL;
mm/swap: split swap cache into 64MB trunks The patch is to improve the scalability of the swap out/in via using fine grained locks for the swap cache. In current kernel, one address space will be used for each swap device. And in the common configuration, the number of the swap device is very small (one is typical). This causes the heavy lock contention on the radix tree of the address space if multiple tasks swap out/in concurrently. But in fact, there is no dependency between pages in the swap cache. So that, we can split the one shared address space for each swap device into several address spaces to reduce the lock contention. In the patch, the shared address space is split into 64MB trunks. 64MB is chosen to balance the memory space usage and effect of lock contention reduction. The size of struct address_space on x86_64 architecture is 408B, so with the patch, 6528B more memory will be used for every 1GB swap space on x86_64 architecture. One address space is still shared for the swap entries in the same 64M trunks. To avoid lock contention for the first round of swap space allocation, the order of the swap clusters in the initial free clusters list is changed. The swap space distance between the consecutive swap clusters in the free cluster list is at least 64M. After the first round of allocation, the swap clusters are expected to be freed randomly, so the lock contention should be reduced effectively. Link: http://lkml.kernel.org/r/735bab895e64c930581ffb0a05b661e01da82bc5.1484082593.git.tim.c.chen@linux.intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com> Cc: Aaron Lu <aaron.lu@intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Christian Borntraeger <borntraeger@de.ibm.com> Cc: Dave Hansen <dave.hansen@intel.com> Cc: Hillf Danton <hillf.zj@alibaba-inc.com> Cc: Huang Ying <ying.huang@intel.com> Cc: Hugh Dickins <hughd@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Jonathan Corbet <corbet@lwn.net> escreveu: Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Vladimir Davydov <vdavydov.dev@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-02-22 23:45:26 +00:00
}
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
mm/swap: swap_vma_readahead() do the pte_offset_map() swap_vma_readahead() has been proceeding in an unconventional way, its preliminary swap_ra_info() doing the pte_offset_map() and pte_unmap(), then relying on that pte pointer even after the pte_unmap() - in its CONFIG_64BIT case (I think !CONFIG_HIGHPTE was intended; whereas 32-bit copied ptes to stack while they were mapped, but had to limit how many). Though it would be difficult to construct a failing testcase, accessing page table after pte_unmap() will become bad practice, even on 64-bit: an rcu_read_unlock() in pte_unmap() will allow page table to be freed. Move relevant definitions from include/linux/swap.h to mm/swap_state.c, nothing else used them. Delete the CONFIG_64BIT distinction and buffer, delete all reference to ptes from swap_ra_info(), use pte_offset_map() repeatedly in swap_vma_readahead(), breaking from the loop if it fails. (Will the repeated "map" and "unmap" show up as a slowdown anywhere? If so, maybe modify __read_swap_cache_async() to do the pte_unmap() only when it does not find the page already in the swapcache.) Use ptep_get_lockless(), mainly for its READ_ONCE(). Correctly advance the address passed down to each call of __read__swap_cache_async(). Link: https://lkml.kernel.org/r/b7c64ab3-9e44-aac0-d2b-c57de578af1c@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Reviewed-by: "Huang, Ying" <ying.huang@intel.com> Cc: Alistair Popple <apopple@nvidia.com> Cc: Anshuman Khandual <anshuman.khandual@arm.com> Cc: Axel Rasmussen <axelrasmussen@google.com> Cc: Christophe Leroy <christophe.leroy@csgroup.eu> Cc: Christoph Hellwig <hch@infradead.org> Cc: David Hildenbrand <david@redhat.com> Cc: Ira Weiny <ira.weiny@intel.com> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Lorenzo Stoakes <lstoakes@gmail.com> Cc: Matthew Wilcox <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Miaohe Lin <linmiaohe@huawei.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Mike Rapoport (IBM) <rppt@kernel.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Naoya Horiguchi <naoya.horiguchi@nec.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Xu <peterx@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Qi Zheng <zhengqi.arch@bytedance.com> Cc: Ralph Campbell <rcampbell@nvidia.com> Cc: Ryan Roberts <ryan.roberts@arm.com> Cc: SeongJae Park <sj@kernel.org> Cc: Song Liu <song@kernel.org> Cc: Steven Price <steven.price@arm.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com> Cc: Will Deacon <will@kernel.org> Cc: Yang Shi <shy828301@gmail.com> Cc: Yu Zhao <yuzhao@google.com> Cc: Zack Rusin <zackr@vmware.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-06-09 01:52:17 +00:00
#define SWAP_RA_ORDER_CEILING 5
struct vma_swap_readahead {
unsigned short win;
unsigned short offset;
unsigned short nr_pte;
};
static void swap_ra_info(struct vm_fault *vmf,
struct vma_swap_readahead *ra_info)
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
{
struct vm_area_struct *vma = vmf->vma;
unsigned long ra_val;
unsigned long faddr, pfn, fpfn, lpfn, rpfn;
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
unsigned long start, end;
unsigned int max_win, hits, prev_win, win;
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
mm, swap: use page-cluster as max window of VMA based swap readahead When the VMA based swap readahead was introduced, a new knob /sys/kernel/mm/swap/vma_ra_max_order was added as the max window of VMA swap readahead. This is to make it possible to use different max window for VMA based readahead and original physical readahead. But Minchan Kim pointed out that this will cause a regression because setting page-cluster sysctl to zero cannot disable swap readahead with the change. To fix the regression, the page-cluster sysctl is used as the max window of both the VMA based swap readahead and original physical swap readahead. If more fine grained control is needed in the future, more knobs can be added as the subordinate knobs of the page-cluster sysctl. The vma_ra_max_order knob is deleted. Because the knob was introduced in v4.14-rc1, and this patch is targeting being merged before v4.14 releasing, there should be no existing users of this newly added ABI. Link: http://lkml.kernel.org/r/20171011070847.16003-1-ying.huang@intel.com Fixes: ec560175c0b6fce ("mm, swap: VMA based swap readahead") Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Reported-by: Minchan Kim <minchan@kernel.org> Acked-by: Minchan Kim <minchan@kernel.org> Acked-by: Michal Hocko <mhocko@suse.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-10-13 22:58:29 +00:00
max_win = 1 << min_t(unsigned int, READ_ONCE(page_cluster),
SWAP_RA_ORDER_CEILING);
if (max_win == 1) {
ra_info->win = 1;
return;
mm, swap: use page-cluster as max window of VMA based swap readahead When the VMA based swap readahead was introduced, a new knob /sys/kernel/mm/swap/vma_ra_max_order was added as the max window of VMA swap readahead. This is to make it possible to use different max window for VMA based readahead and original physical readahead. But Minchan Kim pointed out that this will cause a regression because setting page-cluster sysctl to zero cannot disable swap readahead with the change. To fix the regression, the page-cluster sysctl is used as the max window of both the VMA based swap readahead and original physical swap readahead. If more fine grained control is needed in the future, more knobs can be added as the subordinate knobs of the page-cluster sysctl. The vma_ra_max_order knob is deleted. Because the knob was introduced in v4.14-rc1, and this patch is targeting being merged before v4.14 releasing, there should be no existing users of this newly added ABI. Link: http://lkml.kernel.org/r/20171011070847.16003-1-ying.huang@intel.com Fixes: ec560175c0b6fce ("mm, swap: VMA based swap readahead") Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Reported-by: Minchan Kim <minchan@kernel.org> Acked-by: Minchan Kim <minchan@kernel.org> Acked-by: Michal Hocko <mhocko@suse.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-10-13 22:58:29 +00:00
}
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
faddr = vmf->address;
fpfn = PFN_DOWN(faddr);
ra_val = GET_SWAP_RA_VAL(vma);
pfn = PFN_DOWN(SWAP_RA_ADDR(ra_val));
prev_win = SWAP_RA_WIN(ra_val);
hits = SWAP_RA_HITS(ra_val);
ra_info->win = win = __swapin_nr_pages(pfn, fpfn, hits,
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
max_win, prev_win);
atomic_long_set(&vma->swap_readahead_info,
SWAP_RA_VAL(faddr, win, 0));
if (win == 1)
return;
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
if (fpfn == pfn + 1) {
lpfn = fpfn;
rpfn = fpfn + win;
} else if (pfn == fpfn + 1) {
lpfn = fpfn - win + 1;
rpfn = fpfn + 1;
} else {
unsigned int left = (win - 1) / 2;
lpfn = fpfn - left;
rpfn = fpfn + win - left;
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
}
start = max3(lpfn, PFN_DOWN(vma->vm_start),
PFN_DOWN(faddr & PMD_MASK));
end = min3(rpfn, PFN_DOWN(vma->vm_end),
PFN_DOWN((faddr & PMD_MASK) + PMD_SIZE));
ra_info->nr_pte = end - start;
ra_info->offset = fpfn - start;
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
}
/**
* swap_vma_readahead - swap in pages in hope we need them soon
mempolicy: alloc_pages_mpol() for NUMA policy without vma Shrink shmem's stack usage by eliminating the pseudo-vma from its folio allocation. alloc_pages_mpol(gfp, order, pol, ilx, nid) becomes the principal actor for passing mempolicy choice down to __alloc_pages(), rather than vma_alloc_folio(gfp, order, vma, addr, hugepage). vma_alloc_folio() and alloc_pages() remain, but as wrappers around alloc_pages_mpol(). alloc_pages_bulk_*() untouched, except to provide the additional args to policy_nodemask(), which subsumes policy_node(). Cleanup throughout, cutting out some unhelpful "helpers". It would all be much simpler without MPOL_INTERLEAVE, but that adds a dynamic to the constant mpol: complicated by v3.6 commit 09c231cb8bfd ("tmpfs: distribute interleave better across nodes"), which added ino bias to the interleave, hidden from mm/mempolicy.c until this commit. Hence "ilx" throughout, the "interleave index". Originally I thought it could be done just with nid, but that's wrong: the nodemask may come from the shared policy layer below a shmem vma, or it may come from the task layer above a shmem vma; and without the final nodemask then nodeid cannot be decided. And how ilx is applied depends also on page order. The interleave index is almost always irrelevant unless MPOL_INTERLEAVE: with one exception in alloc_pages_mpol(), where the NO_INTERLEAVE_INDEX passed down from vma-less alloc_pages() is also used as hint not to use THP-style hugepage allocation - to avoid the overhead of a hugepage arg (though I don't understand why we never just added a GFP bit for THP - if it actually needs a different allocation strategy from other pages of the same order). vma_alloc_folio() still carries its hugepage arg here, but it is not used, and should be removed when agreed. get_vma_policy() no longer allows a NULL vma: over time I believe we've eradicated all the places which used to need it e.g. swapoff and madvise used to pass NULL vma to read_swap_cache_async(), but now know the vma. [hughd@google.com: handle NULL mpol being passed to __read_swap_cache_async()] Link: https://lkml.kernel.org/r/ea419956-4751-0102-21f7-9c93cb957892@google.com Link: https://lkml.kernel.org/r/74e34633-6060-f5e3-aee-7040d43f2e93@google.com Link: https://lkml.kernel.org/r/1738368e-bac0-fd11-ed7f-b87142a939fe@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Hildenbrand <david@redhat.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Huang Ying <ying.huang@intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Nhat Pham <nphamcs@gmail.com> Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Tejun heo <tj@kernel.org> Cc: Vishal Moola (Oracle) <vishal.moola@gmail.com> Cc: Yang Shi <shy828301@gmail.com> Cc: Yosry Ahmed <yosryahmed@google.com> Cc: Domenico Cerasuolo <mimmocerasuolo@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-10-19 20:39:08 +00:00
* @targ_entry: swap entry of the targeted memory
* @gfp_mask: memory allocation flags
mempolicy: alloc_pages_mpol() for NUMA policy without vma Shrink shmem's stack usage by eliminating the pseudo-vma from its folio allocation. alloc_pages_mpol(gfp, order, pol, ilx, nid) becomes the principal actor for passing mempolicy choice down to __alloc_pages(), rather than vma_alloc_folio(gfp, order, vma, addr, hugepage). vma_alloc_folio() and alloc_pages() remain, but as wrappers around alloc_pages_mpol(). alloc_pages_bulk_*() untouched, except to provide the additional args to policy_nodemask(), which subsumes policy_node(). Cleanup throughout, cutting out some unhelpful "helpers". It would all be much simpler without MPOL_INTERLEAVE, but that adds a dynamic to the constant mpol: complicated by v3.6 commit 09c231cb8bfd ("tmpfs: distribute interleave better across nodes"), which added ino bias to the interleave, hidden from mm/mempolicy.c until this commit. Hence "ilx" throughout, the "interleave index". Originally I thought it could be done just with nid, but that's wrong: the nodemask may come from the shared policy layer below a shmem vma, or it may come from the task layer above a shmem vma; and without the final nodemask then nodeid cannot be decided. And how ilx is applied depends also on page order. The interleave index is almost always irrelevant unless MPOL_INTERLEAVE: with one exception in alloc_pages_mpol(), where the NO_INTERLEAVE_INDEX passed down from vma-less alloc_pages() is also used as hint not to use THP-style hugepage allocation - to avoid the overhead of a hugepage arg (though I don't understand why we never just added a GFP bit for THP - if it actually needs a different allocation strategy from other pages of the same order). vma_alloc_folio() still carries its hugepage arg here, but it is not used, and should be removed when agreed. get_vma_policy() no longer allows a NULL vma: over time I believe we've eradicated all the places which used to need it e.g. swapoff and madvise used to pass NULL vma to read_swap_cache_async(), but now know the vma. [hughd@google.com: handle NULL mpol being passed to __read_swap_cache_async()] Link: https://lkml.kernel.org/r/ea419956-4751-0102-21f7-9c93cb957892@google.com Link: https://lkml.kernel.org/r/74e34633-6060-f5e3-aee-7040d43f2e93@google.com Link: https://lkml.kernel.org/r/1738368e-bac0-fd11-ed7f-b87142a939fe@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Hildenbrand <david@redhat.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Huang Ying <ying.huang@intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Nhat Pham <nphamcs@gmail.com> Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Tejun heo <tj@kernel.org> Cc: Vishal Moola (Oracle) <vishal.moola@gmail.com> Cc: Yang Shi <shy828301@gmail.com> Cc: Yosry Ahmed <yosryahmed@google.com> Cc: Domenico Cerasuolo <mimmocerasuolo@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-10-19 20:39:08 +00:00
* @mpol: NUMA memory allocation policy to be applied
* @targ_ilx: NUMA interleave index, for use only when MPOL_INTERLEAVE
* @vmf: fault information
*
* Returns the struct folio for entry and addr, after queueing swapin.
*
* Primitive swap readahead code. We simply read in a few pages whose
* virtual addresses are around the fault address in the same vma.
*
* Caller must hold read mmap_lock if vmf->vma is not NULL.
*
*/
static struct folio *swap_vma_readahead(swp_entry_t targ_entry, gfp_t gfp_mask,
struct mempolicy *mpol, pgoff_t targ_ilx, struct vm_fault *vmf)
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
{
struct blk_plug plug;
struct swap_iocb *splug = NULL;
struct folio *folio;
mm/swap: swap_vma_readahead() do the pte_offset_map() swap_vma_readahead() has been proceeding in an unconventional way, its preliminary swap_ra_info() doing the pte_offset_map() and pte_unmap(), then relying on that pte pointer even after the pte_unmap() - in its CONFIG_64BIT case (I think !CONFIG_HIGHPTE was intended; whereas 32-bit copied ptes to stack while they were mapped, but had to limit how many). Though it would be difficult to construct a failing testcase, accessing page table after pte_unmap() will become bad practice, even on 64-bit: an rcu_read_unlock() in pte_unmap() will allow page table to be freed. Move relevant definitions from include/linux/swap.h to mm/swap_state.c, nothing else used them. Delete the CONFIG_64BIT distinction and buffer, delete all reference to ptes from swap_ra_info(), use pte_offset_map() repeatedly in swap_vma_readahead(), breaking from the loop if it fails. (Will the repeated "map" and "unmap" show up as a slowdown anywhere? If so, maybe modify __read_swap_cache_async() to do the pte_unmap() only when it does not find the page already in the swapcache.) Use ptep_get_lockless(), mainly for its READ_ONCE(). Correctly advance the address passed down to each call of __read__swap_cache_async(). Link: https://lkml.kernel.org/r/b7c64ab3-9e44-aac0-d2b-c57de578af1c@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Reviewed-by: "Huang, Ying" <ying.huang@intel.com> Cc: Alistair Popple <apopple@nvidia.com> Cc: Anshuman Khandual <anshuman.khandual@arm.com> Cc: Axel Rasmussen <axelrasmussen@google.com> Cc: Christophe Leroy <christophe.leroy@csgroup.eu> Cc: Christoph Hellwig <hch@infradead.org> Cc: David Hildenbrand <david@redhat.com> Cc: Ira Weiny <ira.weiny@intel.com> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Lorenzo Stoakes <lstoakes@gmail.com> Cc: Matthew Wilcox <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Miaohe Lin <linmiaohe@huawei.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Mike Rapoport (IBM) <rppt@kernel.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Naoya Horiguchi <naoya.horiguchi@nec.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Xu <peterx@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Qi Zheng <zhengqi.arch@bytedance.com> Cc: Ralph Campbell <rcampbell@nvidia.com> Cc: Ryan Roberts <ryan.roberts@arm.com> Cc: SeongJae Park <sj@kernel.org> Cc: Song Liu <song@kernel.org> Cc: Steven Price <steven.price@arm.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com> Cc: Will Deacon <will@kernel.org> Cc: Yang Shi <shy828301@gmail.com> Cc: Yu Zhao <yuzhao@google.com> Cc: Zack Rusin <zackr@vmware.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-06-09 01:52:17 +00:00
pte_t *pte = NULL, pentry;
unsigned long addr;
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
swp_entry_t entry;
mempolicy: alloc_pages_mpol() for NUMA policy without vma Shrink shmem's stack usage by eliminating the pseudo-vma from its folio allocation. alloc_pages_mpol(gfp, order, pol, ilx, nid) becomes the principal actor for passing mempolicy choice down to __alloc_pages(), rather than vma_alloc_folio(gfp, order, vma, addr, hugepage). vma_alloc_folio() and alloc_pages() remain, but as wrappers around alloc_pages_mpol(). alloc_pages_bulk_*() untouched, except to provide the additional args to policy_nodemask(), which subsumes policy_node(). Cleanup throughout, cutting out some unhelpful "helpers". It would all be much simpler without MPOL_INTERLEAVE, but that adds a dynamic to the constant mpol: complicated by v3.6 commit 09c231cb8bfd ("tmpfs: distribute interleave better across nodes"), which added ino bias to the interleave, hidden from mm/mempolicy.c until this commit. Hence "ilx" throughout, the "interleave index". Originally I thought it could be done just with nid, but that's wrong: the nodemask may come from the shared policy layer below a shmem vma, or it may come from the task layer above a shmem vma; and without the final nodemask then nodeid cannot be decided. And how ilx is applied depends also on page order. The interleave index is almost always irrelevant unless MPOL_INTERLEAVE: with one exception in alloc_pages_mpol(), where the NO_INTERLEAVE_INDEX passed down from vma-less alloc_pages() is also used as hint not to use THP-style hugepage allocation - to avoid the overhead of a hugepage arg (though I don't understand why we never just added a GFP bit for THP - if it actually needs a different allocation strategy from other pages of the same order). vma_alloc_folio() still carries its hugepage arg here, but it is not used, and should be removed when agreed. get_vma_policy() no longer allows a NULL vma: over time I believe we've eradicated all the places which used to need it e.g. swapoff and madvise used to pass NULL vma to read_swap_cache_async(), but now know the vma. [hughd@google.com: handle NULL mpol being passed to __read_swap_cache_async()] Link: https://lkml.kernel.org/r/ea419956-4751-0102-21f7-9c93cb957892@google.com Link: https://lkml.kernel.org/r/74e34633-6060-f5e3-aee-7040d43f2e93@google.com Link: https://lkml.kernel.org/r/1738368e-bac0-fd11-ed7f-b87142a939fe@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Hildenbrand <david@redhat.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Huang Ying <ying.huang@intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Nhat Pham <nphamcs@gmail.com> Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Tejun heo <tj@kernel.org> Cc: Vishal Moola (Oracle) <vishal.moola@gmail.com> Cc: Yang Shi <shy828301@gmail.com> Cc: Yosry Ahmed <yosryahmed@google.com> Cc: Domenico Cerasuolo <mimmocerasuolo@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-10-19 20:39:08 +00:00
pgoff_t ilx;
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
unsigned int i;
bool page_allocated;
struct vma_swap_readahead ra_info = {
.win = 1,
};
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
swap_ra_info(vmf, &ra_info);
if (ra_info.win == 1)
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
goto skip;
mm/swap: swap_vma_readahead() do the pte_offset_map() swap_vma_readahead() has been proceeding in an unconventional way, its preliminary swap_ra_info() doing the pte_offset_map() and pte_unmap(), then relying on that pte pointer even after the pte_unmap() - in its CONFIG_64BIT case (I think !CONFIG_HIGHPTE was intended; whereas 32-bit copied ptes to stack while they were mapped, but had to limit how many). Though it would be difficult to construct a failing testcase, accessing page table after pte_unmap() will become bad practice, even on 64-bit: an rcu_read_unlock() in pte_unmap() will allow page table to be freed. Move relevant definitions from include/linux/swap.h to mm/swap_state.c, nothing else used them. Delete the CONFIG_64BIT distinction and buffer, delete all reference to ptes from swap_ra_info(), use pte_offset_map() repeatedly in swap_vma_readahead(), breaking from the loop if it fails. (Will the repeated "map" and "unmap" show up as a slowdown anywhere? If so, maybe modify __read_swap_cache_async() to do the pte_unmap() only when it does not find the page already in the swapcache.) Use ptep_get_lockless(), mainly for its READ_ONCE(). Correctly advance the address passed down to each call of __read__swap_cache_async(). Link: https://lkml.kernel.org/r/b7c64ab3-9e44-aac0-d2b-c57de578af1c@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Reviewed-by: "Huang, Ying" <ying.huang@intel.com> Cc: Alistair Popple <apopple@nvidia.com> Cc: Anshuman Khandual <anshuman.khandual@arm.com> Cc: Axel Rasmussen <axelrasmussen@google.com> Cc: Christophe Leroy <christophe.leroy@csgroup.eu> Cc: Christoph Hellwig <hch@infradead.org> Cc: David Hildenbrand <david@redhat.com> Cc: Ira Weiny <ira.weiny@intel.com> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Lorenzo Stoakes <lstoakes@gmail.com> Cc: Matthew Wilcox <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Miaohe Lin <linmiaohe@huawei.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Mike Rapoport (IBM) <rppt@kernel.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Naoya Horiguchi <naoya.horiguchi@nec.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Xu <peterx@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Qi Zheng <zhengqi.arch@bytedance.com> Cc: Ralph Campbell <rcampbell@nvidia.com> Cc: Ryan Roberts <ryan.roberts@arm.com> Cc: SeongJae Park <sj@kernel.org> Cc: Song Liu <song@kernel.org> Cc: Steven Price <steven.price@arm.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com> Cc: Will Deacon <will@kernel.org> Cc: Yang Shi <shy828301@gmail.com> Cc: Yu Zhao <yuzhao@google.com> Cc: Zack Rusin <zackr@vmware.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-06-09 01:52:17 +00:00
addr = vmf->address - (ra_info.offset * PAGE_SIZE);
mempolicy: alloc_pages_mpol() for NUMA policy without vma Shrink shmem's stack usage by eliminating the pseudo-vma from its folio allocation. alloc_pages_mpol(gfp, order, pol, ilx, nid) becomes the principal actor for passing mempolicy choice down to __alloc_pages(), rather than vma_alloc_folio(gfp, order, vma, addr, hugepage). vma_alloc_folio() and alloc_pages() remain, but as wrappers around alloc_pages_mpol(). alloc_pages_bulk_*() untouched, except to provide the additional args to policy_nodemask(), which subsumes policy_node(). Cleanup throughout, cutting out some unhelpful "helpers". It would all be much simpler without MPOL_INTERLEAVE, but that adds a dynamic to the constant mpol: complicated by v3.6 commit 09c231cb8bfd ("tmpfs: distribute interleave better across nodes"), which added ino bias to the interleave, hidden from mm/mempolicy.c until this commit. Hence "ilx" throughout, the "interleave index". Originally I thought it could be done just with nid, but that's wrong: the nodemask may come from the shared policy layer below a shmem vma, or it may come from the task layer above a shmem vma; and without the final nodemask then nodeid cannot be decided. And how ilx is applied depends also on page order. The interleave index is almost always irrelevant unless MPOL_INTERLEAVE: with one exception in alloc_pages_mpol(), where the NO_INTERLEAVE_INDEX passed down from vma-less alloc_pages() is also used as hint not to use THP-style hugepage allocation - to avoid the overhead of a hugepage arg (though I don't understand why we never just added a GFP bit for THP - if it actually needs a different allocation strategy from other pages of the same order). vma_alloc_folio() still carries its hugepage arg here, but it is not used, and should be removed when agreed. get_vma_policy() no longer allows a NULL vma: over time I believe we've eradicated all the places which used to need it e.g. swapoff and madvise used to pass NULL vma to read_swap_cache_async(), but now know the vma. [hughd@google.com: handle NULL mpol being passed to __read_swap_cache_async()] Link: https://lkml.kernel.org/r/ea419956-4751-0102-21f7-9c93cb957892@google.com Link: https://lkml.kernel.org/r/74e34633-6060-f5e3-aee-7040d43f2e93@google.com Link: https://lkml.kernel.org/r/1738368e-bac0-fd11-ed7f-b87142a939fe@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Hildenbrand <david@redhat.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Huang Ying <ying.huang@intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Nhat Pham <nphamcs@gmail.com> Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Tejun heo <tj@kernel.org> Cc: Vishal Moola (Oracle) <vishal.moola@gmail.com> Cc: Yang Shi <shy828301@gmail.com> Cc: Yosry Ahmed <yosryahmed@google.com> Cc: Domenico Cerasuolo <mimmocerasuolo@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-10-19 20:39:08 +00:00
ilx = targ_ilx - ra_info.offset;
mm/swap: swap_vma_readahead() do the pte_offset_map() swap_vma_readahead() has been proceeding in an unconventional way, its preliminary swap_ra_info() doing the pte_offset_map() and pte_unmap(), then relying on that pte pointer even after the pte_unmap() - in its CONFIG_64BIT case (I think !CONFIG_HIGHPTE was intended; whereas 32-bit copied ptes to stack while they were mapped, but had to limit how many). Though it would be difficult to construct a failing testcase, accessing page table after pte_unmap() will become bad practice, even on 64-bit: an rcu_read_unlock() in pte_unmap() will allow page table to be freed. Move relevant definitions from include/linux/swap.h to mm/swap_state.c, nothing else used them. Delete the CONFIG_64BIT distinction and buffer, delete all reference to ptes from swap_ra_info(), use pte_offset_map() repeatedly in swap_vma_readahead(), breaking from the loop if it fails. (Will the repeated "map" and "unmap" show up as a slowdown anywhere? If so, maybe modify __read_swap_cache_async() to do the pte_unmap() only when it does not find the page already in the swapcache.) Use ptep_get_lockless(), mainly for its READ_ONCE(). Correctly advance the address passed down to each call of __read__swap_cache_async(). Link: https://lkml.kernel.org/r/b7c64ab3-9e44-aac0-d2b-c57de578af1c@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Reviewed-by: "Huang, Ying" <ying.huang@intel.com> Cc: Alistair Popple <apopple@nvidia.com> Cc: Anshuman Khandual <anshuman.khandual@arm.com> Cc: Axel Rasmussen <axelrasmussen@google.com> Cc: Christophe Leroy <christophe.leroy@csgroup.eu> Cc: Christoph Hellwig <hch@infradead.org> Cc: David Hildenbrand <david@redhat.com> Cc: Ira Weiny <ira.weiny@intel.com> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Lorenzo Stoakes <lstoakes@gmail.com> Cc: Matthew Wilcox <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Miaohe Lin <linmiaohe@huawei.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Mike Rapoport (IBM) <rppt@kernel.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Naoya Horiguchi <naoya.horiguchi@nec.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Xu <peterx@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Qi Zheng <zhengqi.arch@bytedance.com> Cc: Ralph Campbell <rcampbell@nvidia.com> Cc: Ryan Roberts <ryan.roberts@arm.com> Cc: SeongJae Park <sj@kernel.org> Cc: Song Liu <song@kernel.org> Cc: Steven Price <steven.price@arm.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com> Cc: Will Deacon <will@kernel.org> Cc: Yang Shi <shy828301@gmail.com> Cc: Yu Zhao <yuzhao@google.com> Cc: Zack Rusin <zackr@vmware.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-06-09 01:52:17 +00:00
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
blk_start_plug(&plug);
mempolicy: alloc_pages_mpol() for NUMA policy without vma Shrink shmem's stack usage by eliminating the pseudo-vma from its folio allocation. alloc_pages_mpol(gfp, order, pol, ilx, nid) becomes the principal actor for passing mempolicy choice down to __alloc_pages(), rather than vma_alloc_folio(gfp, order, vma, addr, hugepage). vma_alloc_folio() and alloc_pages() remain, but as wrappers around alloc_pages_mpol(). alloc_pages_bulk_*() untouched, except to provide the additional args to policy_nodemask(), which subsumes policy_node(). Cleanup throughout, cutting out some unhelpful "helpers". It would all be much simpler without MPOL_INTERLEAVE, but that adds a dynamic to the constant mpol: complicated by v3.6 commit 09c231cb8bfd ("tmpfs: distribute interleave better across nodes"), which added ino bias to the interleave, hidden from mm/mempolicy.c until this commit. Hence "ilx" throughout, the "interleave index". Originally I thought it could be done just with nid, but that's wrong: the nodemask may come from the shared policy layer below a shmem vma, or it may come from the task layer above a shmem vma; and without the final nodemask then nodeid cannot be decided. And how ilx is applied depends also on page order. The interleave index is almost always irrelevant unless MPOL_INTERLEAVE: with one exception in alloc_pages_mpol(), where the NO_INTERLEAVE_INDEX passed down from vma-less alloc_pages() is also used as hint not to use THP-style hugepage allocation - to avoid the overhead of a hugepage arg (though I don't understand why we never just added a GFP bit for THP - if it actually needs a different allocation strategy from other pages of the same order). vma_alloc_folio() still carries its hugepage arg here, but it is not used, and should be removed when agreed. get_vma_policy() no longer allows a NULL vma: over time I believe we've eradicated all the places which used to need it e.g. swapoff and madvise used to pass NULL vma to read_swap_cache_async(), but now know the vma. [hughd@google.com: handle NULL mpol being passed to __read_swap_cache_async()] Link: https://lkml.kernel.org/r/ea419956-4751-0102-21f7-9c93cb957892@google.com Link: https://lkml.kernel.org/r/74e34633-6060-f5e3-aee-7040d43f2e93@google.com Link: https://lkml.kernel.org/r/1738368e-bac0-fd11-ed7f-b87142a939fe@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Hildenbrand <david@redhat.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Huang Ying <ying.huang@intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Nhat Pham <nphamcs@gmail.com> Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Tejun heo <tj@kernel.org> Cc: Vishal Moola (Oracle) <vishal.moola@gmail.com> Cc: Yang Shi <shy828301@gmail.com> Cc: Yosry Ahmed <yosryahmed@google.com> Cc: Domenico Cerasuolo <mimmocerasuolo@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-10-19 20:39:08 +00:00
for (i = 0; i < ra_info.nr_pte; i++, ilx++, addr += PAGE_SIZE) {
mm/swap: swap_vma_readahead() do the pte_offset_map() swap_vma_readahead() has been proceeding in an unconventional way, its preliminary swap_ra_info() doing the pte_offset_map() and pte_unmap(), then relying on that pte pointer even after the pte_unmap() - in its CONFIG_64BIT case (I think !CONFIG_HIGHPTE was intended; whereas 32-bit copied ptes to stack while they were mapped, but had to limit how many). Though it would be difficult to construct a failing testcase, accessing page table after pte_unmap() will become bad practice, even on 64-bit: an rcu_read_unlock() in pte_unmap() will allow page table to be freed. Move relevant definitions from include/linux/swap.h to mm/swap_state.c, nothing else used them. Delete the CONFIG_64BIT distinction and buffer, delete all reference to ptes from swap_ra_info(), use pte_offset_map() repeatedly in swap_vma_readahead(), breaking from the loop if it fails. (Will the repeated "map" and "unmap" show up as a slowdown anywhere? If so, maybe modify __read_swap_cache_async() to do the pte_unmap() only when it does not find the page already in the swapcache.) Use ptep_get_lockless(), mainly for its READ_ONCE(). Correctly advance the address passed down to each call of __read__swap_cache_async(). Link: https://lkml.kernel.org/r/b7c64ab3-9e44-aac0-d2b-c57de578af1c@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Reviewed-by: "Huang, Ying" <ying.huang@intel.com> Cc: Alistair Popple <apopple@nvidia.com> Cc: Anshuman Khandual <anshuman.khandual@arm.com> Cc: Axel Rasmussen <axelrasmussen@google.com> Cc: Christophe Leroy <christophe.leroy@csgroup.eu> Cc: Christoph Hellwig <hch@infradead.org> Cc: David Hildenbrand <david@redhat.com> Cc: Ira Weiny <ira.weiny@intel.com> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Lorenzo Stoakes <lstoakes@gmail.com> Cc: Matthew Wilcox <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Miaohe Lin <linmiaohe@huawei.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Mike Rapoport (IBM) <rppt@kernel.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Naoya Horiguchi <naoya.horiguchi@nec.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Xu <peterx@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Qi Zheng <zhengqi.arch@bytedance.com> Cc: Ralph Campbell <rcampbell@nvidia.com> Cc: Ryan Roberts <ryan.roberts@arm.com> Cc: SeongJae Park <sj@kernel.org> Cc: Song Liu <song@kernel.org> Cc: Steven Price <steven.price@arm.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com> Cc: Will Deacon <will@kernel.org> Cc: Yang Shi <shy828301@gmail.com> Cc: Yu Zhao <yuzhao@google.com> Cc: Zack Rusin <zackr@vmware.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-06-09 01:52:17 +00:00
if (!pte++) {
pte = pte_offset_map(vmf->pmd, addr);
if (!pte)
break;
}
pentry = ptep_get_lockless(pte);
if (!is_swap_pte(pentry))
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
continue;
entry = pte_to_swp_entry(pentry);
if (unlikely(non_swap_entry(entry)))
continue;
mm/swap: swap_vma_readahead() do the pte_offset_map() swap_vma_readahead() has been proceeding in an unconventional way, its preliminary swap_ra_info() doing the pte_offset_map() and pte_unmap(), then relying on that pte pointer even after the pte_unmap() - in its CONFIG_64BIT case (I think !CONFIG_HIGHPTE was intended; whereas 32-bit copied ptes to stack while they were mapped, but had to limit how many). Though it would be difficult to construct a failing testcase, accessing page table after pte_unmap() will become bad practice, even on 64-bit: an rcu_read_unlock() in pte_unmap() will allow page table to be freed. Move relevant definitions from include/linux/swap.h to mm/swap_state.c, nothing else used them. Delete the CONFIG_64BIT distinction and buffer, delete all reference to ptes from swap_ra_info(), use pte_offset_map() repeatedly in swap_vma_readahead(), breaking from the loop if it fails. (Will the repeated "map" and "unmap" show up as a slowdown anywhere? If so, maybe modify __read_swap_cache_async() to do the pte_unmap() only when it does not find the page already in the swapcache.) Use ptep_get_lockless(), mainly for its READ_ONCE(). Correctly advance the address passed down to each call of __read__swap_cache_async(). Link: https://lkml.kernel.org/r/b7c64ab3-9e44-aac0-d2b-c57de578af1c@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Reviewed-by: "Huang, Ying" <ying.huang@intel.com> Cc: Alistair Popple <apopple@nvidia.com> Cc: Anshuman Khandual <anshuman.khandual@arm.com> Cc: Axel Rasmussen <axelrasmussen@google.com> Cc: Christophe Leroy <christophe.leroy@csgroup.eu> Cc: Christoph Hellwig <hch@infradead.org> Cc: David Hildenbrand <david@redhat.com> Cc: Ira Weiny <ira.weiny@intel.com> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Lorenzo Stoakes <lstoakes@gmail.com> Cc: Matthew Wilcox <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Miaohe Lin <linmiaohe@huawei.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Mike Rapoport (IBM) <rppt@kernel.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Naoya Horiguchi <naoya.horiguchi@nec.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Xu <peterx@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Qi Zheng <zhengqi.arch@bytedance.com> Cc: Ralph Campbell <rcampbell@nvidia.com> Cc: Ryan Roberts <ryan.roberts@arm.com> Cc: SeongJae Park <sj@kernel.org> Cc: Song Liu <song@kernel.org> Cc: Steven Price <steven.price@arm.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com> Cc: Will Deacon <will@kernel.org> Cc: Yang Shi <shy828301@gmail.com> Cc: Yu Zhao <yuzhao@google.com> Cc: Zack Rusin <zackr@vmware.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-06-09 01:52:17 +00:00
pte_unmap(pte);
pte = NULL;
folio = __read_swap_cache_async(entry, gfp_mask, mpol, ilx,
zswap: make shrinking memcg-aware Currently, we only have a single global LRU for zswap. This makes it impossible to perform worload-specific shrinking - an memcg cannot determine which pages in the pool it owns, and often ends up writing pages from other memcgs. This issue has been previously observed in practice and mitigated by simply disabling memcg-initiated shrinking: https://lore.kernel.org/all/20230530232435.3097106-1-nphamcs@gmail.com/T/#u This patch fully resolves the issue by replacing the global zswap LRU with memcg- and NUMA-specific LRUs, and modify the reclaim logic: a) When a store attempt hits an memcg limit, it now triggers a synchronous reclaim attempt that, if successful, allows the new hotter page to be accepted by zswap. b) If the store attempt instead hits the global zswap limit, it will trigger an asynchronous reclaim attempt, in which an memcg is selected for reclaim in a round-robin-like fashion. [nphamcs@gmail.com: use correct function for the onlineness check, use mem_cgroup_iter_break()] Link: https://lkml.kernel.org/r/20231205195419.2563217-1-nphamcs@gmail.com [nphamcs@gmail.com: drop the pool's reference at the end of the writeback step] Link: https://lkml.kernel.org/r/20231206030627.4155634-1-nphamcs@gmail.com Link: https://lkml.kernel.org/r/20231130194023.4102148-4-nphamcs@gmail.com Signed-off-by: Domenico Cerasuolo <cerasuolodomenico@gmail.com> Co-developed-by: Nhat Pham <nphamcs@gmail.com> Signed-off-by: Nhat Pham <nphamcs@gmail.com> Tested-by: Bagas Sanjaya <bagasdotme@gmail.com> Cc: Chris Li <chrisl@kernel.org> Cc: Dan Streetman <ddstreet@ieee.org> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Muchun Song <muchun.song@linux.dev> Cc: Roman Gushchin <roman.gushchin@linux.dev> Cc: Seth Jennings <sjenning@redhat.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Shuah Khan <shuah@kernel.org> Cc: Vitaly Wool <vitaly.wool@konsulko.com> Cc: Yosry Ahmed <yosryahmed@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-11-30 19:40:20 +00:00
&page_allocated, false);
if (!folio)
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
continue;
if (page_allocated) {
swap_read_folio(folio, false, &splug);
if (i != ra_info.offset) {
folio_set_readahead(folio);
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
count_vm_event(SWAP_RA);
}
}
folio_put(folio);
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
}
mm/swap: swap_vma_readahead() do the pte_offset_map() swap_vma_readahead() has been proceeding in an unconventional way, its preliminary swap_ra_info() doing the pte_offset_map() and pte_unmap(), then relying on that pte pointer even after the pte_unmap() - in its CONFIG_64BIT case (I think !CONFIG_HIGHPTE was intended; whereas 32-bit copied ptes to stack while they were mapped, but had to limit how many). Though it would be difficult to construct a failing testcase, accessing page table after pte_unmap() will become bad practice, even on 64-bit: an rcu_read_unlock() in pte_unmap() will allow page table to be freed. Move relevant definitions from include/linux/swap.h to mm/swap_state.c, nothing else used them. Delete the CONFIG_64BIT distinction and buffer, delete all reference to ptes from swap_ra_info(), use pte_offset_map() repeatedly in swap_vma_readahead(), breaking from the loop if it fails. (Will the repeated "map" and "unmap" show up as a slowdown anywhere? If so, maybe modify __read_swap_cache_async() to do the pte_unmap() only when it does not find the page already in the swapcache.) Use ptep_get_lockless(), mainly for its READ_ONCE(). Correctly advance the address passed down to each call of __read__swap_cache_async(). Link: https://lkml.kernel.org/r/b7c64ab3-9e44-aac0-d2b-c57de578af1c@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Reviewed-by: "Huang, Ying" <ying.huang@intel.com> Cc: Alistair Popple <apopple@nvidia.com> Cc: Anshuman Khandual <anshuman.khandual@arm.com> Cc: Axel Rasmussen <axelrasmussen@google.com> Cc: Christophe Leroy <christophe.leroy@csgroup.eu> Cc: Christoph Hellwig <hch@infradead.org> Cc: David Hildenbrand <david@redhat.com> Cc: Ira Weiny <ira.weiny@intel.com> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Lorenzo Stoakes <lstoakes@gmail.com> Cc: Matthew Wilcox <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Miaohe Lin <linmiaohe@huawei.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Mike Rapoport (IBM) <rppt@kernel.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Naoya Horiguchi <naoya.horiguchi@nec.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Peter Xu <peterx@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Qi Zheng <zhengqi.arch@bytedance.com> Cc: Ralph Campbell <rcampbell@nvidia.com> Cc: Ryan Roberts <ryan.roberts@arm.com> Cc: SeongJae Park <sj@kernel.org> Cc: Song Liu <song@kernel.org> Cc: Steven Price <steven.price@arm.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com> Cc: Will Deacon <will@kernel.org> Cc: Yang Shi <shy828301@gmail.com> Cc: Yu Zhao <yuzhao@google.com> Cc: Zack Rusin <zackr@vmware.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-06-09 01:52:17 +00:00
if (pte)
pte_unmap(pte);
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
blk_finish_plug(&plug);
swap_read_unplug(splug);
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
lru_add_drain();
skip:
/* The folio was likely read above, so no need for plugging here */
folio = __read_swap_cache_async(targ_entry, gfp_mask, mpol, targ_ilx,
zswap: make shrinking memcg-aware Currently, we only have a single global LRU for zswap. This makes it impossible to perform worload-specific shrinking - an memcg cannot determine which pages in the pool it owns, and often ends up writing pages from other memcgs. This issue has been previously observed in practice and mitigated by simply disabling memcg-initiated shrinking: https://lore.kernel.org/all/20230530232435.3097106-1-nphamcs@gmail.com/T/#u This patch fully resolves the issue by replacing the global zswap LRU with memcg- and NUMA-specific LRUs, and modify the reclaim logic: a) When a store attempt hits an memcg limit, it now triggers a synchronous reclaim attempt that, if successful, allows the new hotter page to be accepted by zswap. b) If the store attempt instead hits the global zswap limit, it will trigger an asynchronous reclaim attempt, in which an memcg is selected for reclaim in a round-robin-like fashion. [nphamcs@gmail.com: use correct function for the onlineness check, use mem_cgroup_iter_break()] Link: https://lkml.kernel.org/r/20231205195419.2563217-1-nphamcs@gmail.com [nphamcs@gmail.com: drop the pool's reference at the end of the writeback step] Link: https://lkml.kernel.org/r/20231206030627.4155634-1-nphamcs@gmail.com Link: https://lkml.kernel.org/r/20231130194023.4102148-4-nphamcs@gmail.com Signed-off-by: Domenico Cerasuolo <cerasuolodomenico@gmail.com> Co-developed-by: Nhat Pham <nphamcs@gmail.com> Signed-off-by: Nhat Pham <nphamcs@gmail.com> Tested-by: Bagas Sanjaya <bagasdotme@gmail.com> Cc: Chris Li <chrisl@kernel.org> Cc: Dan Streetman <ddstreet@ieee.org> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Muchun Song <muchun.song@linux.dev> Cc: Roman Gushchin <roman.gushchin@linux.dev> Cc: Seth Jennings <sjenning@redhat.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Shuah Khan <shuah@kernel.org> Cc: Vitaly Wool <vitaly.wool@konsulko.com> Cc: Yosry Ahmed <yosryahmed@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-11-30 19:40:20 +00:00
&page_allocated, false);
mm/swap_state: update zswap LRU's protection range with the folio locked When a folio is swapped in, the protection size of the corresponding zswap LRU is incremented, so that the zswap shrinker is more conservative with its reclaiming action. This field is embedded within the struct lruvec, so updating it requires looking up the folio's memcg and lruvec. However, currently this lookup can happen after the folio is unlocked, for instance if a new folio is allocated, and swap_read_folio() unlocks the folio before returning. In this scenario, there is no stability guarantee for the binding between a folio and its memcg and lruvec: * A folio's memcg and lruvec can be freed between the lookup and the update, leading to a UAF. * Folio migration can clear the now-unlocked folio's memcg_data, which directs the zswap LRU protection size update towards the root memcg instead of the original memcg. This was recently picked up by the syzbot thanks to a warning in the inlined folio_lruvec() call. Move the zswap LRU protection range update above the swap_read_folio() call, and only when a new page is allocated, to prevent this. [nphamcs@gmail.com: add VM_WARN_ON_ONCE() to zswap_folio_swapin()] Link: https://lkml.kernel.org/r/20240206180855.3987204-1-nphamcs@gmail.com [nphamcs@gmail.com: remove unneeded if (folio) checks] Link: https://lkml.kernel.org/r/20240206191355.83755-1-nphamcs@gmail.com Link: https://lkml.kernel.org/r/20240205232442.3240571-1-nphamcs@gmail.com Fixes: b5ba474f3f51 ("zswap: shrink zswap pool based on memory pressure") Reported-by: syzbot+17a611d10af7d18a7092@syzkaller.appspotmail.com Closes: https://lore.kernel.org/all/000000000000ae47f90610803260@google.com/ Signed-off-by: Nhat Pham <nphamcs@gmail.com> Reviewed-by: Chengming Zhou <zhouchengming@bytedance.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Cc: Yosry Ahmed <yosryahmed@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-02-05 23:24:42 +00:00
if (unlikely(page_allocated)) {
zswap_folio_swapin(folio);
swap_read_folio(folio, false, NULL);
mm/swap_state: update zswap LRU's protection range with the folio locked When a folio is swapped in, the protection size of the corresponding zswap LRU is incremented, so that the zswap shrinker is more conservative with its reclaiming action. This field is embedded within the struct lruvec, so updating it requires looking up the folio's memcg and lruvec. However, currently this lookup can happen after the folio is unlocked, for instance if a new folio is allocated, and swap_read_folio() unlocks the folio before returning. In this scenario, there is no stability guarantee for the binding between a folio and its memcg and lruvec: * A folio's memcg and lruvec can be freed between the lookup and the update, leading to a UAF. * Folio migration can clear the now-unlocked folio's memcg_data, which directs the zswap LRU protection size update towards the root memcg instead of the original memcg. This was recently picked up by the syzbot thanks to a warning in the inlined folio_lruvec() call. Move the zswap LRU protection range update above the swap_read_folio() call, and only when a new page is allocated, to prevent this. [nphamcs@gmail.com: add VM_WARN_ON_ONCE() to zswap_folio_swapin()] Link: https://lkml.kernel.org/r/20240206180855.3987204-1-nphamcs@gmail.com [nphamcs@gmail.com: remove unneeded if (folio) checks] Link: https://lkml.kernel.org/r/20240206191355.83755-1-nphamcs@gmail.com Link: https://lkml.kernel.org/r/20240205232442.3240571-1-nphamcs@gmail.com Fixes: b5ba474f3f51 ("zswap: shrink zswap pool based on memory pressure") Reported-by: syzbot+17a611d10af7d18a7092@syzkaller.appspotmail.com Closes: https://lore.kernel.org/all/000000000000ae47f90610803260@google.com/ Signed-off-by: Nhat Pham <nphamcs@gmail.com> Reviewed-by: Chengming Zhou <zhouchengming@bytedance.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Cc: Yosry Ahmed <yosryahmed@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-02-05 23:24:42 +00:00
}
return folio;
mm, swap: VMA based swap readahead The swap readahead is an important mechanism to reduce the swap in latency. Although pure sequential memory access pattern isn't very popular for anonymous memory, the space locality is still considered valid. In the original swap readahead implementation, the consecutive blocks in swap device are readahead based on the global space locality estimation. But the consecutive blocks in swap device just reflect the order of page reclaiming, don't necessarily reflect the access pattern in virtual memory. And the different tasks in the system may have different access patterns, which makes the global space locality estimation incorrect. In this patch, when page fault occurs, the virtual pages near the fault address will be readahead instead of the swap slots near the fault swap slot in swap device. This avoid to readahead the unrelated swap slots. At the same time, the swap readahead is changed to work on per-VMA from globally. So that the different access patterns of the different VMAs could be distinguished, and the different readahead policy could be applied accordingly. The original core readahead detection and scaling algorithm is reused, because it is an effect algorithm to detect the space locality. The test and result is as follow, Common test condition ===================== Test Machine: Xeon E5 v3 (2 sockets, 72 threads, 32G RAM) Swap device: NVMe disk Micro-benchmark with combined access pattern ============================================ vm-scalability, sequential swap test case, 4 processes to eat 50G virtual memory space, repeat the sequential memory writing until 300 seconds. The first round writing will trigger swap out, the following rounds will trigger sequential swap in and out. At the same time, run vm-scalability random swap test case in background, 8 processes to eat 30G virtual memory space, repeat the random memory write until 300 seconds. This will trigger random swap-in in the background. This is a combined workload with sequential and random memory accessing at the same time. The result (for sequential workload) is as follow, Base Optimized ---- --------- throughput 345413 KB/s 414029 KB/s (+19.9%) latency.average 97.14 us 61.06 us (-37.1%) latency.50th 2 us 1 us latency.60th 2 us 1 us latency.70th 98 us 2 us latency.80th 160 us 2 us latency.90th 260 us 217 us latency.95th 346 us 369 us latency.99th 1.34 ms 1.09 ms ra_hit% 52.69% 99.98% The original swap readahead algorithm is confused by the background random access workload, so readahead hit rate is lower. The VMA-base readahead algorithm works much better. Linpack ======= The test memory size is bigger than RAM to trigger swapping. Base Optimized ---- --------- elapsed_time 393.49 s 329.88 s (-16.2%) ra_hit% 86.21% 98.82% The score of base and optimized kernel hasn't visible changes. But the elapsed time reduced and readahead hit rate improved, so the optimized kernel runs better for startup and tear down stages. And the absolute value of readahead hit rate is high, shows that the space locality is still valid in some practical workloads. Link: http://lkml.kernel.org/r/20170807054038.1843-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Shaohua Li <shli@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Tim Chen <tim.c.chen@intel.com> Cc: Dave Hansen <dave.hansen@intel.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 23:24:36 +00:00
}
/**
* swapin_readahead - swap in pages in hope we need them soon
* @entry: swap entry of this memory
* @gfp_mask: memory allocation flags
* @vmf: fault information
*
* Returns the struct page for entry and addr, after queueing swapin.
*
* It's a main entry function for swap readahead. By the configuration,
* it will read ahead blocks by cluster-based(ie, physical disk based)
* or vma-based(ie, virtual address based on faulty address) readahead.
*/
struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
struct vm_fault *vmf)
{
mempolicy: alloc_pages_mpol() for NUMA policy without vma Shrink shmem's stack usage by eliminating the pseudo-vma from its folio allocation. alloc_pages_mpol(gfp, order, pol, ilx, nid) becomes the principal actor for passing mempolicy choice down to __alloc_pages(), rather than vma_alloc_folio(gfp, order, vma, addr, hugepage). vma_alloc_folio() and alloc_pages() remain, but as wrappers around alloc_pages_mpol(). alloc_pages_bulk_*() untouched, except to provide the additional args to policy_nodemask(), which subsumes policy_node(). Cleanup throughout, cutting out some unhelpful "helpers". It would all be much simpler without MPOL_INTERLEAVE, but that adds a dynamic to the constant mpol: complicated by v3.6 commit 09c231cb8bfd ("tmpfs: distribute interleave better across nodes"), which added ino bias to the interleave, hidden from mm/mempolicy.c until this commit. Hence "ilx" throughout, the "interleave index". Originally I thought it could be done just with nid, but that's wrong: the nodemask may come from the shared policy layer below a shmem vma, or it may come from the task layer above a shmem vma; and without the final nodemask then nodeid cannot be decided. And how ilx is applied depends also on page order. The interleave index is almost always irrelevant unless MPOL_INTERLEAVE: with one exception in alloc_pages_mpol(), where the NO_INTERLEAVE_INDEX passed down from vma-less alloc_pages() is also used as hint not to use THP-style hugepage allocation - to avoid the overhead of a hugepage arg (though I don't understand why we never just added a GFP bit for THP - if it actually needs a different allocation strategy from other pages of the same order). vma_alloc_folio() still carries its hugepage arg here, but it is not used, and should be removed when agreed. get_vma_policy() no longer allows a NULL vma: over time I believe we've eradicated all the places which used to need it e.g. swapoff and madvise used to pass NULL vma to read_swap_cache_async(), but now know the vma. [hughd@google.com: handle NULL mpol being passed to __read_swap_cache_async()] Link: https://lkml.kernel.org/r/ea419956-4751-0102-21f7-9c93cb957892@google.com Link: https://lkml.kernel.org/r/74e34633-6060-f5e3-aee-7040d43f2e93@google.com Link: https://lkml.kernel.org/r/1738368e-bac0-fd11-ed7f-b87142a939fe@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Hildenbrand <david@redhat.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Huang Ying <ying.huang@intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Nhat Pham <nphamcs@gmail.com> Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Tejun heo <tj@kernel.org> Cc: Vishal Moola (Oracle) <vishal.moola@gmail.com> Cc: Yang Shi <shy828301@gmail.com> Cc: Yosry Ahmed <yosryahmed@google.com> Cc: Domenico Cerasuolo <mimmocerasuolo@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-10-19 20:39:08 +00:00
struct mempolicy *mpol;
pgoff_t ilx;
struct folio *folio;
mempolicy: alloc_pages_mpol() for NUMA policy without vma Shrink shmem's stack usage by eliminating the pseudo-vma from its folio allocation. alloc_pages_mpol(gfp, order, pol, ilx, nid) becomes the principal actor for passing mempolicy choice down to __alloc_pages(), rather than vma_alloc_folio(gfp, order, vma, addr, hugepage). vma_alloc_folio() and alloc_pages() remain, but as wrappers around alloc_pages_mpol(). alloc_pages_bulk_*() untouched, except to provide the additional args to policy_nodemask(), which subsumes policy_node(). Cleanup throughout, cutting out some unhelpful "helpers". It would all be much simpler without MPOL_INTERLEAVE, but that adds a dynamic to the constant mpol: complicated by v3.6 commit 09c231cb8bfd ("tmpfs: distribute interleave better across nodes"), which added ino bias to the interleave, hidden from mm/mempolicy.c until this commit. Hence "ilx" throughout, the "interleave index". Originally I thought it could be done just with nid, but that's wrong: the nodemask may come from the shared policy layer below a shmem vma, or it may come from the task layer above a shmem vma; and without the final nodemask then nodeid cannot be decided. And how ilx is applied depends also on page order. The interleave index is almost always irrelevant unless MPOL_INTERLEAVE: with one exception in alloc_pages_mpol(), where the NO_INTERLEAVE_INDEX passed down from vma-less alloc_pages() is also used as hint not to use THP-style hugepage allocation - to avoid the overhead of a hugepage arg (though I don't understand why we never just added a GFP bit for THP - if it actually needs a different allocation strategy from other pages of the same order). vma_alloc_folio() still carries its hugepage arg here, but it is not used, and should be removed when agreed. get_vma_policy() no longer allows a NULL vma: over time I believe we've eradicated all the places which used to need it e.g. swapoff and madvise used to pass NULL vma to read_swap_cache_async(), but now know the vma. [hughd@google.com: handle NULL mpol being passed to __read_swap_cache_async()] Link: https://lkml.kernel.org/r/ea419956-4751-0102-21f7-9c93cb957892@google.com Link: https://lkml.kernel.org/r/74e34633-6060-f5e3-aee-7040d43f2e93@google.com Link: https://lkml.kernel.org/r/1738368e-bac0-fd11-ed7f-b87142a939fe@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Hildenbrand <david@redhat.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Huang Ying <ying.huang@intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Nhat Pham <nphamcs@gmail.com> Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Tejun heo <tj@kernel.org> Cc: Vishal Moola (Oracle) <vishal.moola@gmail.com> Cc: Yang Shi <shy828301@gmail.com> Cc: Yosry Ahmed <yosryahmed@google.com> Cc: Domenico Cerasuolo <mimmocerasuolo@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-10-19 20:39:08 +00:00
mpol = get_vma_policy(vmf->vma, vmf->address, 0, &ilx);
folio = swap_use_vma_readahead() ?
mempolicy: alloc_pages_mpol() for NUMA policy without vma Shrink shmem's stack usage by eliminating the pseudo-vma from its folio allocation. alloc_pages_mpol(gfp, order, pol, ilx, nid) becomes the principal actor for passing mempolicy choice down to __alloc_pages(), rather than vma_alloc_folio(gfp, order, vma, addr, hugepage). vma_alloc_folio() and alloc_pages() remain, but as wrappers around alloc_pages_mpol(). alloc_pages_bulk_*() untouched, except to provide the additional args to policy_nodemask(), which subsumes policy_node(). Cleanup throughout, cutting out some unhelpful "helpers". It would all be much simpler without MPOL_INTERLEAVE, but that adds a dynamic to the constant mpol: complicated by v3.6 commit 09c231cb8bfd ("tmpfs: distribute interleave better across nodes"), which added ino bias to the interleave, hidden from mm/mempolicy.c until this commit. Hence "ilx" throughout, the "interleave index". Originally I thought it could be done just with nid, but that's wrong: the nodemask may come from the shared policy layer below a shmem vma, or it may come from the task layer above a shmem vma; and without the final nodemask then nodeid cannot be decided. And how ilx is applied depends also on page order. The interleave index is almost always irrelevant unless MPOL_INTERLEAVE: with one exception in alloc_pages_mpol(), where the NO_INTERLEAVE_INDEX passed down from vma-less alloc_pages() is also used as hint not to use THP-style hugepage allocation - to avoid the overhead of a hugepage arg (though I don't understand why we never just added a GFP bit for THP - if it actually needs a different allocation strategy from other pages of the same order). vma_alloc_folio() still carries its hugepage arg here, but it is not used, and should be removed when agreed. get_vma_policy() no longer allows a NULL vma: over time I believe we've eradicated all the places which used to need it e.g. swapoff and madvise used to pass NULL vma to read_swap_cache_async(), but now know the vma. [hughd@google.com: handle NULL mpol being passed to __read_swap_cache_async()] Link: https://lkml.kernel.org/r/ea419956-4751-0102-21f7-9c93cb957892@google.com Link: https://lkml.kernel.org/r/74e34633-6060-f5e3-aee-7040d43f2e93@google.com Link: https://lkml.kernel.org/r/1738368e-bac0-fd11-ed7f-b87142a939fe@google.com Signed-off-by: Hugh Dickins <hughd@google.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Hildenbrand <david@redhat.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Huang Ying <ying.huang@intel.com> Cc: Kefeng Wang <wangkefeng.wang@huawei.com> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Michal Hocko <mhocko@suse.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Nhat Pham <nphamcs@gmail.com> Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Tejun heo <tj@kernel.org> Cc: Vishal Moola (Oracle) <vishal.moola@gmail.com> Cc: Yang Shi <shy828301@gmail.com> Cc: Yosry Ahmed <yosryahmed@google.com> Cc: Domenico Cerasuolo <mimmocerasuolo@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-10-19 20:39:08 +00:00
swap_vma_readahead(entry, gfp_mask, mpol, ilx, vmf) :
swap_cluster_readahead(entry, gfp_mask, mpol, ilx);
mpol_cond_put(mpol);
if (!folio)
return NULL;
return folio_file_page(folio, swp_offset(entry));
}
#ifdef CONFIG_SYSFS
static ssize_t vma_ra_enabled_show(struct kobject *kobj,
struct kobj_attribute *attr, char *buf)
{
return sysfs_emit(buf, "%s\n",
enable_vma_readahead ? "true" : "false");
}
static ssize_t vma_ra_enabled_store(struct kobject *kobj,
struct kobj_attribute *attr,
const char *buf, size_t count)
{
ssize_t ret;
ret = kstrtobool(buf, &enable_vma_readahead);
if (ret)
return ret;
return count;
}
static struct kobj_attribute vma_ra_enabled_attr = __ATTR_RW(vma_ra_enabled);
static struct attribute *swap_attrs[] = {
&vma_ra_enabled_attr.attr,
NULL,
};
static const struct attribute_group swap_attr_group = {
.attrs = swap_attrs,
};
static int __init swap_init_sysfs(void)
{
int err;
struct kobject *swap_kobj;
swap_kobj = kobject_create_and_add("swap", mm_kobj);
if (!swap_kobj) {
pr_err("failed to create swap kobject\n");
return -ENOMEM;
}
err = sysfs_create_group(swap_kobj, &swap_attr_group);
if (err) {
pr_err("failed to register swap group\n");
goto delete_obj;
}
return 0;
delete_obj:
kobject_put(swap_kobj);
return err;
}
subsys_initcall(swap_init_sysfs);
#endif