linux/arch/x86/kernel/pci-gart_64.c

883 lines
22 KiB
C
Raw Normal View History

/*
* Dynamic DMA mapping support for AMD Hammer.
*
* Use the integrated AGP GART in the Hammer northbridge as an IOMMU for PCI.
* This allows to use PCI devices that only support 32bit addresses on systems
* with more than 4GB.
*
* See Documentation/DMA-mapping.txt for the interface specification.
*
* Copyright 2002 Andi Kleen, SuSE Labs.
* Subject to the GNU General Public License v2 only.
*/
#include <linux/types.h>
#include <linux/ctype.h>
#include <linux/agp_backend.h>
#include <linux/init.h>
#include <linux/mm.h>
#include <linux/string.h>
#include <linux/spinlock.h>
#include <linux/pci.h>
#include <linux/module.h>
#include <linux/topology.h>
#include <linux/interrupt.h>
#include <linux/bitops.h>
#include <linux/kdebug.h>
#include <linux/scatterlist.h>
#include <linux/iommu-helper.h>
#include <linux/sysdev.h>
#include <asm/atomic.h>
#include <asm/io.h>
#include <asm/mtrr.h>
#include <asm/pgtable.h>
#include <asm/proto.h>
#include <asm/iommu.h>
#include <asm/gart.h>
#include <asm/cacheflush.h>
#include <asm/swiotlb.h>
#include <asm/dma.h>
#include <asm/k8.h>
static unsigned long iommu_bus_base; /* GART remapping area (physical) */
static unsigned long iommu_size; /* size of remapping area bytes */
static unsigned long iommu_pages; /* .. and in pages */
static u32 *iommu_gatt_base; /* Remapping table */
/*
* If this is disabled the IOMMU will use an optimized flushing strategy
* of only flushing when an mapping is reused. With it true the GART is
* flushed for every mapping. Problem is that doing the lazy flush seems
* to trigger bugs with some popular PCI cards, in particular 3ware (but
* has been also also seen with Qlogic at least).
*/
int iommu_fullflush = 1;
/* Allocation bitmap for the remapping area: */
static DEFINE_SPINLOCK(iommu_bitmap_lock);
/* Guarded by iommu_bitmap_lock: */
static unsigned long *iommu_gart_bitmap;
static u32 gart_unmapped_entry;
#define GPTE_VALID 1
#define GPTE_COHERENT 2
#define GPTE_ENCODE(x) \
(((x) & 0xfffff000) | (((x) >> 32) << 4) | GPTE_VALID | GPTE_COHERENT)
#define GPTE_DECODE(x) (((x) & 0xfffff000) | (((u64)(x) & 0xff0) << 28))
#define EMERGENCY_PAGES 32 /* = 128KB */
#ifdef CONFIG_AGP
#define AGPEXTERN extern
#else
#define AGPEXTERN
#endif
/* backdoor interface to AGP driver */
AGPEXTERN int agp_memory_reserved;
AGPEXTERN __u32 *agp_gatt_table;
static unsigned long next_bit; /* protected by iommu_bitmap_lock */
static int need_flush; /* global flush state. set for each gart wrap */
static unsigned long alloc_iommu(struct device *dev, int size)
{
unsigned long offset, flags;
unsigned long boundary_size;
unsigned long base_index;
base_index = ALIGN(iommu_bus_base & dma_get_seg_boundary(dev),
PAGE_SIZE) >> PAGE_SHIFT;
boundary_size = ALIGN(dma_get_seg_boundary(dev) + 1,
PAGE_SIZE) >> PAGE_SHIFT;
spin_lock_irqsave(&iommu_bitmap_lock, flags);
offset = iommu_area_alloc(iommu_gart_bitmap, iommu_pages, next_bit,
size, base_index, boundary_size, 0);
if (offset == -1) {
need_flush = 1;
offset = iommu_area_alloc(iommu_gart_bitmap, iommu_pages, 0,
size, base_index, boundary_size, 0);
}
if (offset != -1) {
next_bit = offset+size;
if (next_bit >= iommu_pages) {
next_bit = 0;
need_flush = 1;
}
}
if (iommu_fullflush)
need_flush = 1;
spin_unlock_irqrestore(&iommu_bitmap_lock, flags);
return offset;
}
static void free_iommu(unsigned long offset, int size)
{
unsigned long flags;
spin_lock_irqsave(&iommu_bitmap_lock, flags);
iommu_area_free(iommu_gart_bitmap, offset, size);
spin_unlock_irqrestore(&iommu_bitmap_lock, flags);
}
/*
* Use global flush state to avoid races with multiple flushers.
*/
static void flush_gart(void)
{
unsigned long flags;
spin_lock_irqsave(&iommu_bitmap_lock, flags);
if (need_flush) {
k8_flush_garts();
need_flush = 0;
}
spin_unlock_irqrestore(&iommu_bitmap_lock, flags);
}
#ifdef CONFIG_IOMMU_LEAK
#define SET_LEAK(x) \
do { \
if (iommu_leak_tab) \
iommu_leak_tab[x] = __builtin_return_address(0);\
} while (0)
#define CLEAR_LEAK(x) \
do { \
if (iommu_leak_tab) \
iommu_leak_tab[x] = NULL; \
} while (0)
/* Debugging aid for drivers that don't free their IOMMU tables */
static void **iommu_leak_tab;
static int leak_trace;
static int iommu_leak_pages = 20;
static void dump_leak(void)
{
int i;
static int dump;
if (dump || !iommu_leak_tab)
return;
dump = 1;
show_stack(NULL, NULL);
/* Very crude. dump some from the end of the table too */
printk(KERN_DEBUG "Dumping %d pages from end of IOMMU:\n",
iommu_leak_pages);
for (i = 0; i < iommu_leak_pages; i += 2) {
printk(KERN_DEBUG "%lu: ", iommu_pages-i);
printk_address((unsigned long) iommu_leak_tab[iommu_pages-i], 0);
printk(KERN_CONT "%c", (i+1)%2 == 0 ? '\n' : ' ');
}
printk(KERN_DEBUG "\n");
}
#else
# define SET_LEAK(x)
# define CLEAR_LEAK(x)
#endif
static void iommu_full(struct device *dev, size_t size, int dir)
{
/*
* Ran out of IOMMU space for this operation. This is very bad.
* Unfortunately the drivers cannot handle this operation properly.
* Return some non mapped prereserved space in the aperture and
* let the Northbridge deal with it. This will result in garbage
* in the IO operation. When the size exceeds the prereserved space
* memory corruption will occur or random memory will be DMAed
* out. Hopefully no network devices use single mappings that big.
*/
dev_err(dev, "PCI-DMA: Out of IOMMU space for %lu bytes\n", size);
if (size > PAGE_SIZE*EMERGENCY_PAGES) {
if (dir == PCI_DMA_FROMDEVICE || dir == PCI_DMA_BIDIRECTIONAL)
panic("PCI-DMA: Memory would be corrupted\n");
if (dir == PCI_DMA_TODEVICE || dir == PCI_DMA_BIDIRECTIONAL)
panic(KERN_ERR
"PCI-DMA: Random memory would be DMAed\n");
}
#ifdef CONFIG_IOMMU_LEAK
dump_leak();
#endif
}
static inline int
need_iommu(struct device *dev, unsigned long addr, size_t size)
{
u64 mask = *dev->dma_mask;
int high = addr + size > mask;
int mmu = high;
if (force_iommu)
mmu = 1;
return mmu;
}
static inline int
nonforced_iommu(struct device *dev, unsigned long addr, size_t size)
{
u64 mask = *dev->dma_mask;
int high = addr + size > mask;
int mmu = high;
return mmu;
}
/* Map a single continuous physical area into the IOMMU.
* Caller needs to check if the iommu is needed and flush.
*/
static dma_addr_t dma_map_area(struct device *dev, dma_addr_t phys_mem,
size_t size, int dir)
{
unsigned long npages = iommu_num_pages(phys_mem, size);
unsigned long iommu_page = alloc_iommu(dev, npages);
int i;
if (iommu_page == -1) {
if (!nonforced_iommu(dev, phys_mem, size))
return phys_mem;
if (panic_on_overflow)
panic("dma_map_area overflow %lu bytes\n", size);
iommu_full(dev, size, dir);
return bad_dma_address;
}
for (i = 0; i < npages; i++) {
iommu_gatt_base[iommu_page + i] = GPTE_ENCODE(phys_mem);
SET_LEAK(iommu_page + i);
phys_mem += PAGE_SIZE;
}
return iommu_bus_base + iommu_page*PAGE_SIZE + (phys_mem & ~PAGE_MASK);
}
static dma_addr_t
gart_map_simple(struct device *dev, phys_addr_t paddr, size_t size, int dir)
{
dma_addr_t map = dma_map_area(dev, paddr, size, dir);
flush_gart();
return map;
}
/* Map a single area into the IOMMU */
static dma_addr_t
gart_map_single(struct device *dev, phys_addr_t paddr, size_t size, int dir)
{
unsigned long bus;
if (!dev)
dev = &fallback_dev;
if (!need_iommu(dev, paddr, size))
return paddr;
bus = gart_map_simple(dev, paddr, size, dir);
return bus;
}
/*
* Free a DMA mapping.
*/
static void gart_unmap_single(struct device *dev, dma_addr_t dma_addr,
size_t size, int direction)
{
unsigned long iommu_page;
int npages;
int i;
if (dma_addr < iommu_bus_base + EMERGENCY_PAGES*PAGE_SIZE ||
dma_addr >= iommu_bus_base + iommu_size)
return;
iommu_page = (dma_addr - iommu_bus_base)>>PAGE_SHIFT;
npages = iommu_num_pages(dma_addr, size);
for (i = 0; i < npages; i++) {
iommu_gatt_base[iommu_page + i] = gart_unmapped_entry;
CLEAR_LEAK(iommu_page + i);
}
free_iommu(iommu_page, npages);
}
/*
* Wrapper for pci_unmap_single working with scatterlists.
*/
static void
gart_unmap_sg(struct device *dev, struct scatterlist *sg, int nents, int dir)
{
struct scatterlist *s;
int i;
for_each_sg(sg, s, nents, i) {
if (!s->dma_length || !s->length)
break;
gart_unmap_single(dev, s->dma_address, s->dma_length, dir);
}
}
/* Fallback for dma_map_sg in case of overflow */
static int dma_map_sg_nonforce(struct device *dev, struct scatterlist *sg,
int nents, int dir)
{
struct scatterlist *s;
int i;
#ifdef CONFIG_IOMMU_DEBUG
printk(KERN_DEBUG "dma_map_sg overflow\n");
#endif
for_each_sg(sg, s, nents, i) {
unsigned long addr = sg_phys(s);
if (nonforced_iommu(dev, addr, s->length)) {
addr = dma_map_area(dev, addr, s->length, dir);
if (addr == bad_dma_address) {
if (i > 0)
gart_unmap_sg(dev, sg, i, dir);
nents = 0;
sg[0].dma_length = 0;
break;
}
}
s->dma_address = addr;
s->dma_length = s->length;
}
flush_gart();
return nents;
}
/* Map multiple scatterlist entries continuous into the first. */
static int __dma_map_cont(struct device *dev, struct scatterlist *start,
int nelems, struct scatterlist *sout,
unsigned long pages)
{
unsigned long iommu_start = alloc_iommu(dev, pages);
unsigned long iommu_page = iommu_start;
struct scatterlist *s;
int i;
if (iommu_start == -1)
return -1;
for_each_sg(start, s, nelems, i) {
unsigned long pages, addr;
unsigned long phys_addr = s->dma_address;
BUG_ON(s != start && s->offset);
if (s == start) {
sout->dma_address = iommu_bus_base;
sout->dma_address += iommu_page*PAGE_SIZE + s->offset;
sout->dma_length = s->length;
} else {
sout->dma_length += s->length;
}
addr = phys_addr;
pages = iommu_num_pages(s->offset, s->length);
while (pages--) {
iommu_gatt_base[iommu_page] = GPTE_ENCODE(addr);
SET_LEAK(iommu_page);
addr += PAGE_SIZE;
iommu_page++;
}
}
BUG_ON(iommu_page - iommu_start != pages);
return 0;
}
static inline int
dma_map_cont(struct device *dev, struct scatterlist *start, int nelems,
struct scatterlist *sout, unsigned long pages, int need)
{
if (!need) {
BUG_ON(nelems != 1);
sout->dma_address = start->dma_address;
sout->dma_length = start->length;
return 0;
}
return __dma_map_cont(dev, start, nelems, sout, pages);
}
/*
* DMA map all entries in a scatterlist.
* Merge chunks that have page aligned sizes into a continuous mapping.
*/
static int
gart_map_sg(struct device *dev, struct scatterlist *sg, int nents, int dir)
{
struct scatterlist *s, *ps, *start_sg, *sgmap;
int need = 0, nextneed, i, out, start;
unsigned long pages = 0;
unsigned int seg_size;
unsigned int max_seg_size;
if (nents == 0)
return 0;
if (!dev)
dev = &fallback_dev;
out = 0;
start = 0;
start_sg = sgmap = sg;
seg_size = 0;
max_seg_size = dma_get_max_seg_size(dev);
ps = NULL; /* shut up gcc */
for_each_sg(sg, s, nents, i) {
dma_addr_t addr = sg_phys(s);
s->dma_address = addr;
BUG_ON(s->length == 0);
nextneed = need_iommu(dev, addr, s->length);
/* Handle the previous not yet processed entries */
if (i > start) {
/*
* Can only merge when the last chunk ends on a
* page boundary and the new one doesn't have an
* offset.
*/
if (!iommu_merge || !nextneed || !need || s->offset ||
(s->length + seg_size > max_seg_size) ||
(ps->offset + ps->length) % PAGE_SIZE) {
if (dma_map_cont(dev, start_sg, i - start,
sgmap, pages, need) < 0)
goto error;
out++;
seg_size = 0;
sgmap = sg_next(sgmap);
pages = 0;
start = i;
start_sg = s;
}
}
seg_size += s->length;
need = nextneed;
pages += iommu_num_pages(s->offset, s->length);
ps = s;
}
if (dma_map_cont(dev, start_sg, i - start, sgmap, pages, need) < 0)
goto error;
out++;
flush_gart();
if (out < nents) {
sgmap = sg_next(sgmap);
sgmap->dma_length = 0;
}
return out;
error:
flush_gart();
gart_unmap_sg(dev, sg, out, dir);
/* When it was forced or merged try again in a dumb way */
if (force_iommu || iommu_merge) {
out = dma_map_sg_nonforce(dev, sg, nents, dir);
if (out > 0)
return out;
}
if (panic_on_overflow)
panic("dma_map_sg: overflow on %lu pages\n", pages);
iommu_full(dev, pages << PAGE_SHIFT, dir);
for_each_sg(sg, s, nents, i)
s->dma_address = bad_dma_address;
return 0;
}
static int no_agp;
static __init unsigned long check_iommu_size(unsigned long aper, u64 aper_size)
{
unsigned long a;
if (!iommu_size) {
iommu_size = aper_size;
if (!no_agp)
iommu_size /= 2;
}
a = aper + iommu_size;
iommu_size -= round_up(a, PMD_PAGE_SIZE) - a;
if (iommu_size < 64*1024*1024) {
printk(KERN_WARNING
"PCI-DMA: Warning: Small IOMMU %luMB."
" Consider increasing the AGP aperture in BIOS\n",
iommu_size >> 20);
}
return iommu_size;
}
static __init unsigned read_aperture(struct pci_dev *dev, u32 *size)
{
unsigned aper_size = 0, aper_base_32, aper_order;
u64 aper_base;
pci_read_config_dword(dev, AMD64_GARTAPERTUREBASE, &aper_base_32);
pci_read_config_dword(dev, AMD64_GARTAPERTURECTL, &aper_order);
aper_order = (aper_order >> 1) & 7;
aper_base = aper_base_32 & 0x7fff;
aper_base <<= 25;
aper_size = (32 * 1024 * 1024) << aper_order;
if (aper_base + aper_size > 0x100000000UL || !aper_size)
aper_base = 0;
*size = aper_size;
return aper_base;
}
static void enable_gart_translations(void)
{
int i;
for (i = 0; i < num_k8_northbridges; i++) {
struct pci_dev *dev = k8_northbridges[i];
enable_gart_translation(dev, __pa(agp_gatt_table));
}
}
/*
* If fix_up_north_bridges is set, the north bridges have to be fixed up on
* resume in the same way as they are handled in gart_iommu_hole_init().
*/
static bool fix_up_north_bridges;
static u32 aperture_order;
static u32 aperture_alloc;
void set_up_gart_resume(u32 aper_order, u32 aper_alloc)
{
fix_up_north_bridges = true;
aperture_order = aper_order;
aperture_alloc = aper_alloc;
}
static int gart_resume(struct sys_device *dev)
{
printk(KERN_INFO "PCI-DMA: Resuming GART IOMMU\n");
if (fix_up_north_bridges) {
int i;
printk(KERN_INFO "PCI-DMA: Restoring GART aperture settings\n");
for (i = 0; i < num_k8_northbridges; i++) {
struct pci_dev *dev = k8_northbridges[i];
/*
* Don't enable translations just yet. That is the next
* step. Restore the pre-suspend aperture settings.
*/
pci_write_config_dword(dev, AMD64_GARTAPERTURECTL,
aperture_order << 1);
pci_write_config_dword(dev, AMD64_GARTAPERTUREBASE,
aperture_alloc >> 25);
}
}
enable_gart_translations();
return 0;
}
static int gart_suspend(struct sys_device *dev, pm_message_t state)
{
return 0;
}
static struct sysdev_class gart_sysdev_class = {
.name = "gart",
.suspend = gart_suspend,
.resume = gart_resume,
};
static struct sys_device device_gart = {
.id = 0,
.cls = &gart_sysdev_class,
};
/*
* Private Northbridge GATT initialization in case we cannot use the
* AGP driver for some reason.
*/
static __init int init_k8_gatt(struct agp_kern_info *info)
{
unsigned aper_size, gatt_size, new_aper_size;
unsigned aper_base, new_aper_base;
struct pci_dev *dev;
void *gatt;
int i, error;
x86: max_low_pfn_mapped fix, #1 fix crash on Ingo's big box: calling pci_iommu_init+0x0/0x17 PCI-DMA: Disabling AGP. PCI-DMA: aperture base @ d0000000 size 65536 KB PCI-DMA: using GART IOMMU. PCI-DMA: Reserving 64MB of IOMMU area in the AGP aperture BUG: unable to handle kernel paging request at ffff88000003be88 IP: [<ffffffff8026d377>] __alloc_pages_internal+0xc3/0x3f2 PGD 202063 PUD 206063 PMD 22fc00163 PTE 3b162 Oops: 0000 [1] SMP and e820 is: BIOS-e820: 0000000000000000 - 000000000009ac00 (usable) BIOS-e820: 000000000009ac00 - 00000000000a0000 (reserved) BIOS-e820: 00000000000ca000 - 0000000000100000 (reserved) BIOS-e820: 0000000000100000 - 000000007ff70000 (usable) BIOS-e820: 000000007ff70000 - 000000007ff86000 (ACPI data) BIOS-e820: 000000007ff86000 - 0000000080000000 (ACPI NVS) BIOS-e820: 0000000080000000 - 00000000cfe00000 (usable) BIOS-e820: 00000000cfe00000 - 00000000d0000000 (reserved) BIOS-e820: 00000000e0000000 - 00000000f0000000 (reserved) BIOS-e820: 00000000fec00000 - 00000000fec10000 (reserved) BIOS-e820: 00000000fee00000 - 00000000fee01000 (reserved) BIOS-e820: 00000000fff80000 - 0000000100000000 (reserved) BIOS-e820: 0000000100000000 - 0000000830000000 (usable) system has 32 GB RAM installed. max_low_pfn_mapped is 0xcfe00, and GART aperture is not mapped. So try to use init_memory_mapping to map that area, because the iommu thinks that area is ram ... Signed-off-by: Yinghai Lu <yhlu.kernel@gmail.com> Cc: Suresh Siddha <suresh.b.siddha@intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-07-12 21:30:35 +00:00
unsigned long start_pfn, end_pfn;
printk(KERN_INFO "PCI-DMA: Disabling AGP.\n");
aper_size = aper_base = info->aper_size = 0;
dev = NULL;
for (i = 0; i < num_k8_northbridges; i++) {
dev = k8_northbridges[i];
new_aper_base = read_aperture(dev, &new_aper_size);
if (!new_aper_base)
goto nommu;
if (!aper_base) {
aper_size = new_aper_size;
aper_base = new_aper_base;
}
if (aper_size != new_aper_size || aper_base != new_aper_base)
goto nommu;
}
if (!aper_base)
goto nommu;
info->aper_base = aper_base;
info->aper_size = aper_size >> 20;
gatt_size = (aper_size >> PAGE_SHIFT) * sizeof(u32);
gatt = (void *)__get_free_pages(GFP_KERNEL, get_order(gatt_size));
if (!gatt)
panic("Cannot allocate GATT table");
if (set_memory_uc((unsigned long)gatt, gatt_size >> PAGE_SHIFT))
panic("Could not set GART PTEs to uncacheable pages");
memset(gatt, 0, gatt_size);
agp_gatt_table = gatt;
enable_gart_translations();
error = sysdev_class_register(&gart_sysdev_class);
if (!error)
error = sysdev_register(&device_gart);
if (error)
panic("Could not register gart_sysdev -- would corrupt data on next suspend");
flush_gart();
printk(KERN_INFO "PCI-DMA: aperture base @ %x size %u KB\n",
aper_base, aper_size>>10);
x86: max_low_pfn_mapped fix, #1 fix crash on Ingo's big box: calling pci_iommu_init+0x0/0x17 PCI-DMA: Disabling AGP. PCI-DMA: aperture base @ d0000000 size 65536 KB PCI-DMA: using GART IOMMU. PCI-DMA: Reserving 64MB of IOMMU area in the AGP aperture BUG: unable to handle kernel paging request at ffff88000003be88 IP: [<ffffffff8026d377>] __alloc_pages_internal+0xc3/0x3f2 PGD 202063 PUD 206063 PMD 22fc00163 PTE 3b162 Oops: 0000 [1] SMP and e820 is: BIOS-e820: 0000000000000000 - 000000000009ac00 (usable) BIOS-e820: 000000000009ac00 - 00000000000a0000 (reserved) BIOS-e820: 00000000000ca000 - 0000000000100000 (reserved) BIOS-e820: 0000000000100000 - 000000007ff70000 (usable) BIOS-e820: 000000007ff70000 - 000000007ff86000 (ACPI data) BIOS-e820: 000000007ff86000 - 0000000080000000 (ACPI NVS) BIOS-e820: 0000000080000000 - 00000000cfe00000 (usable) BIOS-e820: 00000000cfe00000 - 00000000d0000000 (reserved) BIOS-e820: 00000000e0000000 - 00000000f0000000 (reserved) BIOS-e820: 00000000fec00000 - 00000000fec10000 (reserved) BIOS-e820: 00000000fee00000 - 00000000fee01000 (reserved) BIOS-e820: 00000000fff80000 - 0000000100000000 (reserved) BIOS-e820: 0000000100000000 - 0000000830000000 (usable) system has 32 GB RAM installed. max_low_pfn_mapped is 0xcfe00, and GART aperture is not mapped. So try to use init_memory_mapping to map that area, because the iommu thinks that area is ram ... Signed-off-by: Yinghai Lu <yhlu.kernel@gmail.com> Cc: Suresh Siddha <suresh.b.siddha@intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-07-12 21:30:35 +00:00
/* need to map that range */
end_pfn = (aper_base>>PAGE_SHIFT) + (aper_size>>PAGE_SHIFT);
if (end_pfn > max_low_pfn_mapped) {
start_pfn = (aper_base>>PAGE_SHIFT);
init_memory_mapping(start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
x86: max_low_pfn_mapped fix, #1 fix crash on Ingo's big box: calling pci_iommu_init+0x0/0x17 PCI-DMA: Disabling AGP. PCI-DMA: aperture base @ d0000000 size 65536 KB PCI-DMA: using GART IOMMU. PCI-DMA: Reserving 64MB of IOMMU area in the AGP aperture BUG: unable to handle kernel paging request at ffff88000003be88 IP: [<ffffffff8026d377>] __alloc_pages_internal+0xc3/0x3f2 PGD 202063 PUD 206063 PMD 22fc00163 PTE 3b162 Oops: 0000 [1] SMP and e820 is: BIOS-e820: 0000000000000000 - 000000000009ac00 (usable) BIOS-e820: 000000000009ac00 - 00000000000a0000 (reserved) BIOS-e820: 00000000000ca000 - 0000000000100000 (reserved) BIOS-e820: 0000000000100000 - 000000007ff70000 (usable) BIOS-e820: 000000007ff70000 - 000000007ff86000 (ACPI data) BIOS-e820: 000000007ff86000 - 0000000080000000 (ACPI NVS) BIOS-e820: 0000000080000000 - 00000000cfe00000 (usable) BIOS-e820: 00000000cfe00000 - 00000000d0000000 (reserved) BIOS-e820: 00000000e0000000 - 00000000f0000000 (reserved) BIOS-e820: 00000000fec00000 - 00000000fec10000 (reserved) BIOS-e820: 00000000fee00000 - 00000000fee01000 (reserved) BIOS-e820: 00000000fff80000 - 0000000100000000 (reserved) BIOS-e820: 0000000100000000 - 0000000830000000 (usable) system has 32 GB RAM installed. max_low_pfn_mapped is 0xcfe00, and GART aperture is not mapped. So try to use init_memory_mapping to map that area, because the iommu thinks that area is ram ... Signed-off-by: Yinghai Lu <yhlu.kernel@gmail.com> Cc: Suresh Siddha <suresh.b.siddha@intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-07-12 21:30:35 +00:00
}
return 0;
nommu:
/* Should not happen anymore */
printk(KERN_WARNING "PCI-DMA: More than 4GB of RAM and no IOMMU\n"
KERN_WARNING "falling back to iommu=soft.\n");
return -1;
}
extern int agp_amd64_init(void);
dma-mapping: add the device argument to dma_mapping_error() Add per-device dma_mapping_ops support for CONFIG_X86_64 as POWER architecture does: This enables us to cleanly fix the Calgary IOMMU issue that some devices are not behind the IOMMU (http://lkml.org/lkml/2008/5/8/423). I think that per-device dma_mapping_ops support would be also helpful for KVM people to support PCI passthrough but Andi thinks that this makes it difficult to support the PCI passthrough (see the above thread). So I CC'ed this to KVM camp. Comments are appreciated. A pointer to dma_mapping_ops to struct dev_archdata is added. If the pointer is non NULL, DMA operations in asm/dma-mapping.h use it. If it's NULL, the system-wide dma_ops pointer is used as before. If it's useful for KVM people, I plan to implement a mechanism to register a hook called when a new pci (or dma capable) device is created (it works with hot plugging). It enables IOMMUs to set up an appropriate dma_mapping_ops per device. The major obstacle is that dma_mapping_error doesn't take a pointer to the device unlike other DMA operations. So x86 can't have dma_mapping_ops per device. Note all the POWER IOMMUs use the same dma_mapping_error function so this is not a problem for POWER but x86 IOMMUs use different dma_mapping_error functions. The first patch adds the device argument to dma_mapping_error. The patch is trivial but large since it touches lots of drivers and dma-mapping.h in all the architecture. This patch: dma_mapping_error() doesn't take a pointer to the device unlike other DMA operations. So we can't have dma_mapping_ops per device. Note that POWER already has dma_mapping_ops per device but all the POWER IOMMUs use the same dma_mapping_error function. x86 IOMMUs use device argument. [akpm@linux-foundation.org: fix sge] [akpm@linux-foundation.org: fix svc_rdma] [akpm@linux-foundation.org: build fix] [akpm@linux-foundation.org: fix bnx2x] [akpm@linux-foundation.org: fix s2io] [akpm@linux-foundation.org: fix pasemi_mac] [akpm@linux-foundation.org: fix sdhci] [akpm@linux-foundation.org: build fix] [akpm@linux-foundation.org: fix sparc] [akpm@linux-foundation.org: fix ibmvscsi] Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp> Cc: Muli Ben-Yehuda <muli@il.ibm.com> Cc: Andi Kleen <andi@firstfloor.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: Avi Kivity <avi@qumranet.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-26 02:44:49 +00:00
static struct dma_mapping_ops gart_dma_ops = {
.map_single = gart_map_single,
.map_simple = gart_map_simple,
.unmap_single = gart_unmap_single,
.sync_single_for_cpu = NULL,
.sync_single_for_device = NULL,
.sync_single_range_for_cpu = NULL,
.sync_single_range_for_device = NULL,
.sync_sg_for_cpu = NULL,
.sync_sg_for_device = NULL,
.map_sg = gart_map_sg,
.unmap_sg = gart_unmap_sg,
};
void gart_iommu_shutdown(void)
{
struct pci_dev *dev;
int i;
if (no_agp && (dma_ops != &gart_dma_ops))
return;
for (i = 0; i < num_k8_northbridges; i++) {
u32 ctl;
dev = k8_northbridges[i];
pci_read_config_dword(dev, AMD64_GARTAPERTURECTL, &ctl);
ctl &= ~GARTEN;
pci_write_config_dword(dev, AMD64_GARTAPERTURECTL, ctl);
}
}
void __init gart_iommu_init(void)
{
struct agp_kern_info info;
unsigned long iommu_start;
unsigned long aper_size;
unsigned long scratch;
long i;
if (cache_k8_northbridges() < 0 || num_k8_northbridges == 0) {
printk(KERN_INFO "PCI-GART: No AMD northbridge found.\n");
return;
}
#ifndef CONFIG_AGP_AMD64
no_agp = 1;
#else
/* Makefile puts PCI initialization via subsys_initcall first. */
/* Add other K8 AGP bridge drivers here */
no_agp = no_agp ||
(agp_amd64_init() < 0) ||
(agp_copy_info(agp_bridge, &info) < 0);
#endif
if (swiotlb)
return;
/* Did we detect a different HW IOMMU? */
if (iommu_detected && !gart_iommu_aperture)
return;
if (no_iommu ||
(!force_iommu && max_pfn <= MAX_DMA32_PFN) ||
!gart_iommu_aperture ||
(no_agp && init_k8_gatt(&info) < 0)) {
if (max_pfn > MAX_DMA32_PFN) {
printk(KERN_WARNING "More than 4GB of memory "
"but GART IOMMU not available.\n"
KERN_WARNING "falling back to iommu=soft.\n");
}
return;
}
printk(KERN_INFO "PCI-DMA: using GART IOMMU.\n");
aper_size = info.aper_size * 1024 * 1024;
iommu_size = check_iommu_size(info.aper_base, aper_size);
iommu_pages = iommu_size >> PAGE_SHIFT;
iommu_gart_bitmap = (void *) __get_free_pages(GFP_KERNEL,
get_order(iommu_pages/8));
if (!iommu_gart_bitmap)
panic("Cannot allocate iommu bitmap\n");
memset(iommu_gart_bitmap, 0, iommu_pages/8);
#ifdef CONFIG_IOMMU_LEAK
if (leak_trace) {
iommu_leak_tab = (void *)__get_free_pages(GFP_KERNEL,
get_order(iommu_pages*sizeof(void *)));
if (iommu_leak_tab)
memset(iommu_leak_tab, 0, iommu_pages * 8);
else
printk(KERN_DEBUG
"PCI-DMA: Cannot allocate leak trace area\n");
}
#endif
/*
* Out of IOMMU space handling.
* Reserve some invalid pages at the beginning of the GART.
*/
set_bit_string(iommu_gart_bitmap, 0, EMERGENCY_PAGES);
agp_memory_reserved = iommu_size;
printk(KERN_INFO
"PCI-DMA: Reserving %luMB of IOMMU area in the AGP aperture\n",
iommu_size >> 20);
iommu_start = aper_size - iommu_size;
iommu_bus_base = info.aper_base + iommu_start;
bad_dma_address = iommu_bus_base;
iommu_gatt_base = agp_gatt_table + (iommu_start>>PAGE_SHIFT);
/*
* Unmap the IOMMU part of the GART. The alias of the page is
* always mapped with cache enabled and there is no full cache
* coherency across the GART remapping. The unmapping avoids
* automatic prefetches from the CPU allocating cache lines in
* there. All CPU accesses are done via the direct mapping to
* the backing memory. The GART address is only used by PCI
* devices.
*/
set_memory_np((unsigned long)__va(iommu_bus_base),
iommu_size >> PAGE_SHIFT);
/*
* Tricky. The GART table remaps the physical memory range,
* so the CPU wont notice potential aliases and if the memory
* is remapped to UC later on, we might surprise the PCI devices
* with a stray writeout of a cacheline. So play it sure and
* do an explicit, full-scale wbinvd() _after_ having marked all
* the pages as Not-Present:
*/
wbinvd();
/*
* Try to workaround a bug (thanks to BenH):
* Set unmapped entries to a scratch page instead of 0.
* Any prefetches that hit unmapped entries won't get an bus abort
* then. (P2P bridge may be prefetching on DMA reads).
*/
scratch = get_zeroed_page(GFP_KERNEL);
if (!scratch)
panic("Cannot allocate iommu scratch page");
gart_unmapped_entry = GPTE_ENCODE(__pa(scratch));
for (i = EMERGENCY_PAGES; i < iommu_pages; i++)
iommu_gatt_base[i] = gart_unmapped_entry;
flush_gart();
dma_ops = &gart_dma_ops;
}
[PATCH] x86-64: fix section mismatch warnings Fix the following section mismatch warnings on x86_64: (build using defconfig) WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.text:mtrr_bp_init from .text between 'identify_cpu' (at offset 0x65eb) and 'IRQ0x20_interrupt' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.data: from .text between 'finish_e820_parsing' (at offset 0x7dc2) and 'early_panic' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.text:e820_print_map from .text between 'finish_e820_parsing' (at offset 0x7de1) and 'early_panic' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.data:num_processors from .text between 'acpi_unmap_lsapic' (at offset 0xc88f) and 'acpi_register_ioapic' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.data:disabled_cpus from .text between 'MP_processor_info' (at offset 0x11f35) and 'mp_register_lapic' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.data:num_processors from .text between 'MP_processor_info' (at offset 0x11f6e) and 'mp_register_lapic' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.data:num_processors from .text between 'MP_processor_info' (at offset 0x11f93) and 'mp_register_lapic' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.data:fix_aperture from .text between 'gart_parse_options' (at offset 0x15517) and 'iommu_full' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.data:fix_aperture from .text between 'gart_parse_options' (at offset 0x1552c) and 'iommu_full' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.data:iommu_aperture_allowed from .text between 'gart_parse_options' (at offset 0x1553d) and 'iommu_full' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.data:iommu_aperture_allowed from .text between 'gart_parse_options' (at offset 0x15552) and 'iommu_full' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.data:iommu_aperture_allowed from .text between 'gart_parse_options' (at offset 0x15561) and 'iommu_full' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.data:iommu_aperture_allowed from .text between 'gart_parse_options' (at offset 0x15577) and 'iommu_full' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.data:fallback_aper_force from .text between 'gart_parse_options' (at offset 0x1558a) and 'iommu_full' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.data:fallback_aper_order from .text between 'gart_parse_options' (at offset 0x155bf) and 'iommu_full' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.data:timer_over_8254 from .text between 'ati_bugs' (at offset 0x16344) and 'via_bugs' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.data:timer_over_8254 from .text between 'ati_bugs' (at offset 0x16356) and 'via_bugs' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.data:iommu_aperture_allowed from .text between 'via_bugs' (at offset 0x16380) and 'nvidia_bugs' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.data:iommu_aperture_disabled from .text between 'via_bugs' (at offset 0x16397) and 'nvidia_bugs' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.data:acpi_use_timer_override from .text between 'nvidia_bugs' (at offset 0x163a7) and 'arch_unregister_cpu' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.text:nvidia_hpet_check from .text between 'nvidia_bugs' (at offset 0x163b1) and 'arch_unregister_cpu' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.data: from .text between 'nvidia_bugs' (at offset 0x163be) and 'arch_unregister_cpu' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.data: from .text between 'nvidia_bugs' (at offset 0x163d1) and 'arch_unregister_cpu' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.data:acpi_skip_timer_override from .text between 'nvidia_bugs' (at offset 0x163e1) and 'arch_unregister_cpu' WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.text:quirk_intel_irqbalance from .text between 'intel_bugs' (at offset 0x1633c) and 'ati_bugs' But adds: WARNING: arch/x86_64/kernel/built-in.o - Section mismatch: reference to .init.text:get_mtrr_state from .text between 'mtrr_bp_init' (at offset 0xb887) and 'ipi_handler' The warnings does not show up during a normal build due to kbuild failing to check for section mismatch in vmlinux. To see these warnings run: scripts/mod/modpost arch/x86_64/kernel/built-in.o kbuild will be fixed but the 'noise-level' had to be decresed first. There remains a few section mismatch warnigns for x86_64 for areas where I did not feel confident. Signed-off-by: Sam Ravnborg <sam@ravnborg.org> Signed-off-by: Andi Kleen <ak@suse.de>
2007-03-16 20:07:36 +00:00
void __init gart_parse_options(char *p)
{
int arg;
#ifdef CONFIG_IOMMU_LEAK
if (!strncmp(p, "leak", 4)) {
leak_trace = 1;
p += 4;
if (*p == '=') ++p;
if (isdigit(*p) && get_option(&p, &arg))
iommu_leak_pages = arg;
}
#endif
if (isdigit(*p) && get_option(&p, &arg))
iommu_size = arg;
if (!strncmp(p, "fullflush", 8))
iommu_fullflush = 1;
if (!strncmp(p, "nofullflush", 11))
iommu_fullflush = 0;
if (!strncmp(p, "noagp", 5))
no_agp = 1;
if (!strncmp(p, "noaperture", 10))
fix_aperture = 0;
/* duplicated from pci-dma.c */
if (!strncmp(p, "force", 5))
gart_iommu_aperture_allowed = 1;
if (!strncmp(p, "allowed", 7))
gart_iommu_aperture_allowed = 1;
if (!strncmp(p, "memaper", 7)) {
fallback_aper_force = 1;
p += 7;
if (*p == '=') {
++p;
if (get_option(&p, &arg))
fallback_aper_order = arg;
}
}
}