Re: [git pull] x86 updates for v2.6.28, phase #3 - IOMMU

From: Ingo Molnar
Date: Fri Oct 10 2008 - 13:49:20 EST



* Ingo Molnar <mingo@xxxxxxx> wrote:

> Linus,
>
> Please pull the latest x86-v28-for-linus-phase3 git tree from:
>
> IOMMU updates.

Linus, please find the updated pull request below - relative to the
respun x86-v28-for-linus-phase2-B.

Ingo

-------------->
Linus,

Please pull the latest x86-v28-for-linus-phase3-B git tree from:

git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip.git x86-v28-for-linus-phase3-B


out-of-topic modifications in x86-v28-for-linus-phase3-B:
---------------------------------------------------------
MAINTAINERS # bbd001c: add AMD IOMMU tree to MAINTAINERS
arch/ia64/include/asm/dma-mapping.h# 3a80b6a: ia64: dma_alloc_coherent always u
include/linux/dma-mapping.h # 07a2c01: convert swiotlb to use dma_get_ma
# 589fc9a6: iommu: add dma_get_mask helper f
# 636dc67: add is_buffer_dma_capable helper
include/linux/iommu-helper.h # 1e19b16: AMD IOMMU: use iommu_device_max_i
# d26dbc5: iommu: export iommu_area_reserve
# eecfffc: iommu: add iommu_device_max_index
include/linux/pci_ids.h # 2426824: x86: add PCI IDs for AMD Barcelon
# ca1af29: x86, pci: add northbridge pci ids
kernel/dma-coherent.c # a38409f: dma-coherent: export dma_[alloc|r
lib/iommu-helper.c # d26dbc5: iommu: export iommu_area_reserve
lib/swiotlb.c # 07a2c01: convert swiotlb to use dma_get_ma
# 2797982: swiotlb: convert swiotlb to use i
# 640aebf: swiotlb: add is_swiotlb_buffer he
# 21f6c4d: swiotlb: use unmap_single instead
# 9dfda12: swiotlb: use map_single instead o
# 68e91d6: swiotlb: remove GFP_DMA hack in s

Thanks,

Ingo

------------------>
David Woodhouse (1):
x86: blacklist DMAR on Intel G31/G33 chipsets

FUJITA Tomonori (32):
dma-mapping.h, x86: remove last user of dma_mapping_ops->map_simple
x86: dma_*_coherent rework patchset v2, fix
x86: gart alloc_coherent doesn't need to check NULL device argument
ia64: dma_alloc_coherent always use GFP_DMA
x86: move pci-nommu's dma_mask check to common code
x86: fix nommu_alloc_coherent allocation with NULL device argument
x86: dma_alloc_coherent sets gfp flags properly
swiotlb: remove GFP_DMA hack in swiotlb_alloc_coherent
swiotlb: use map_single instead of swiotlb_map_single in swiotlb_alloc_coherent
swiotlb: use unmap_single instead of swiotlb_unmap_single in swiotlb_free_coherent
swiotlb: add is_swiotlb_buffer helper function
add is_buffer_dma_capable helper function
x86: convert gart to use is_buffer_dma_capable helper function
x86: convert pci-nommu to use is_buffer_dma_capable helper function
swiotlb: convert swiotlb to use is_buffer_dma_capable helper function
x86: remove duplicated extern force_iommu
x86: convert dma_alloc_coherent to use is_device_dma_capable
iommu: add iommu_device_max_index IOMMU helper function
iommu: add dma_get_mask helper function
x86: make GART to respect device's dma_mask about virtual mappings
x86: avoid unnecessary low zone allocation in Calgary's alloc_coherent
x86: gart alloc_coherent does virtual mapppings only when necessary
convert swiotlb to use dma_get_mask
AMD IOMMU: avoid unnecessary low zone allocation in alloc_coherent
iommu: export iommu_area_reserve helper function
x86: remove set_bit_string()
iommu: remove fullflush and nofullflush in IOMMU generic option
x86: export pci-nommu's alloc_coherent
revert "x86: make GART to respect device's dma_mask about virtual mappings"
x86: restore old GART alloc_coherent behavior
x86 gart: remove unnecessary initialization
AMD IOMMU: use iommu_device_max_index

Ingo Molnar (2):
dma-coherent: export dma_[alloc|release]_from_coherent methods
x86: add PCI IDs for AMD Barcelona PCI devices

Joerg Roedel (37):
x86: add alloc_coherent dma_ops callback to GART driver
x86: add free_coherent dma_ops callback to GART driver
x86: add free_coherent dma_ops callback to Calgary IOMMU driver
x86: add alloc_coherent dma_ops callback to NOMMU driver
x86: add free_coherent dma_ops callback to NOMMU driver
x86: cleanup dma_*_coherent functions
x86: move dma_*_coherent functions to include file
x86, AMD IOMMU: remove obsolete FIXME comment
x86, pci: add northbridge pci ids for fam 0x11 processors
x86, gart: add detection of AMD family 0x11 northbridges
AMD IOMMU: check for invalid device pointers
AMD IOMMU: move TLB flushing to the map/unmap helper functions
x86: move GART TLB flushing options to generic code
AMD IOMMU: implement lazy IO/TLB flushing
AMD IOMMU: add branch hints to completion wait checks
AMD IOMMU: align alloc_coherent addresses properly
AMD IOMMU: add event buffer allocation
AMD IOMMU: save pci segment from ACPI tables
AMD IOMMU: save pci_dev instead of devid
AMD IOMMU: add MSI interrupt support
AMD IOMMU: add event handling code
AMD IOMMU: enable event logging
AMD IOMMU: allow IO page faults from devices
AMD IOMMU: add dma_supported callback
AMD IOMMU: don't assign preallocated protection domains to devices
AMD IOMMU: some set_device_domain cleanups
AMD IOMMU: replace memset with __GFP_ZERO in alloc_coherent
AMD IOMMU: simplify dma_mask_to_pages
AMD IOMMU: free domain bitmap with its allocation order
AMD IOMMU: remove unnecessary cast to u64 in the init code
AMD IOMMU: calculate IVHD size with a function
AMD IOMMU: use cmd_buf_size when freeing the command buffer
add AMD IOMMU tree to MAINTAINERS file
AMD IOMMU: use coherent_dma_mask in alloc_coherent
x86/iommu: make GART driver checkpatch clean
x86/iommu: convert GART need_flush to bool
x86/iommu: use __GFP_ZERO instead of memset for GART

Steven Noonan (1):
x86: unused variable in dma_alloc_coherent_gfp_flags()

Thomas Gleixner (1):
AMD IOMMU: use iommu_device_max_index, fix


Documentation/kernel-parameters.txt | 5 +
MAINTAINERS | 1 +
arch/ia64/include/asm/dma-mapping.h | 4 +-
arch/x86/Kconfig | 1 +
arch/x86/kernel/amd_iommu.c | 328 ++++++++++++++++++++++++++++-------
arch/x86/kernel/amd_iommu_init.c | 194 +++++++++++++++++++--
arch/x86/kernel/early-quirks.c | 18 ++
arch/x86/kernel/k8.c | 5 +-
arch/x86/kernel/pci-calgary_64.c | 18 ++-
arch/x86/kernel/pci-dma.c | 177 ++++---------------
arch/x86/kernel/pci-gart_64.c | 125 +++++++------
arch/x86/kernel/pci-nommu.c | 10 +-
drivers/pci/intel-iommu.c | 2 +-
include/asm-x86/amd_iommu.h | 3 +
include/asm-x86/amd_iommu_types.h | 64 +++++++-
include/asm-x86/bitops.h | 10 -
include/asm-x86/dma-mapping.h | 87 ++++++++--
include/asm-x86/gart.h | 2 +
include/asm-x86/iommu.h | 1 +
include/linux/dma-mapping.h | 12 ++
include/linux/iommu-helper.h | 16 ++
include/linux/pci_ids.h | 10 +
kernel/dma-coherent.c | 2 +
lib/iommu-helper.c | 5 +-
lib/swiotlb.c | 49 ++---
25 files changed, 803 insertions(+), 346 deletions(-)

diff --git a/Documentation/kernel-parameters.txt b/Documentation/kernel-parameters.txt
index 329dcab..a2701cb 100644
--- a/Documentation/kernel-parameters.txt
+++ b/Documentation/kernel-parameters.txt
@@ -284,6 +284,11 @@ and is between 256 and 4096 characters. It is defined in the file
isolate - enable device isolation (each device, as far
as possible, will get its own protection
domain)
+ fullflush - enable flushing of IO/TLB entries when
+ they are unmapped. Otherwise they are
+ flushed before they will be reused, which
+ is a lot of faster
+
amd_iommu_size= [HW,X86-64]
Define the size of the aperture for the AMD IOMMU
driver. Possible values are:
diff --git a/MAINTAINERS b/MAINTAINERS
index 8dae455..3c124d7 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -387,6 +387,7 @@ AMD IOMMU (AMD-VI)
P: Joerg Roedel
M: joerg.roedel@xxxxxxx
L: iommu@xxxxxxxxxxxxxxxxxxxxxxxxxx
+T: git://git.kernel.org/pub/scm/linux/kernel/git/joro/linux-2.6-iommu.git
S: Supported

AMS (Apple Motion Sensor) DRIVER
diff --git a/arch/ia64/include/asm/dma-mapping.h b/arch/ia64/include/asm/dma-mapping.h
index 9f0df9b..06ff1ba 100644
--- a/arch/ia64/include/asm/dma-mapping.h
+++ b/arch/ia64/include/asm/dma-mapping.h
@@ -8,7 +8,9 @@
#include <asm/machvec.h>
#include <linux/scatterlist.h>

-#define dma_alloc_coherent platform_dma_alloc_coherent
+#define dma_alloc_coherent(dev, size, handle, gfp) \
+ platform_dma_alloc_coherent(dev, size, handle, (gfp) | GFP_DMA)
+
/* coherent mem. is cheap */
static inline void *
dma_alloc_noncoherent(struct device *dev, size_t size, dma_addr_t *dma_handle,
diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 97f0d2b..0d7cdbb 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -554,6 +554,7 @@ config CALGARY_IOMMU_ENABLED_BY_DEFAULT
config AMD_IOMMU
bool "AMD IOMMU support"
select SWIOTLB
+ select PCI_MSI
depends on X86_64 && PCI && ACPI
help
With this option you can enable support for AMD IOMMU hardware in
diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 042fdc2..34e4d11 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -33,6 +33,10 @@

static DEFINE_RWLOCK(amd_iommu_devtable_lock);

+/* A list of preallocated protection domains */
+static LIST_HEAD(iommu_pd_list);
+static DEFINE_SPINLOCK(iommu_pd_list_lock);
+
/*
* general struct to manage commands send to an IOMMU
*/
@@ -51,6 +55,102 @@ static int iommu_has_npcache(struct amd_iommu *iommu)

/****************************************************************************
*
+ * Interrupt handling functions
+ *
+ ****************************************************************************/
+
+static void iommu_print_event(void *__evt)
+{
+ u32 *event = __evt;
+ int type = (event[1] >> EVENT_TYPE_SHIFT) & EVENT_TYPE_MASK;
+ int devid = (event[0] >> EVENT_DEVID_SHIFT) & EVENT_DEVID_MASK;
+ int domid = (event[1] >> EVENT_DOMID_SHIFT) & EVENT_DOMID_MASK;
+ int flags = (event[1] >> EVENT_FLAGS_SHIFT) & EVENT_FLAGS_MASK;
+ u64 address = (u64)(((u64)event[3]) << 32) | event[2];
+
+ printk(KERN_ERR "AMD IOMMU: Event logged [");
+
+ switch (type) {
+ case EVENT_TYPE_ILL_DEV:
+ printk("ILLEGAL_DEV_TABLE_ENTRY device=%02x:%02x.%x "
+ "address=0x%016llx flags=0x%04x]\n",
+ PCI_BUS(devid), PCI_SLOT(devid), PCI_FUNC(devid),
+ address, flags);
+ break;
+ case EVENT_TYPE_IO_FAULT:
+ printk("IO_PAGE_FAULT device=%02x:%02x.%x "
+ "domain=0x%04x address=0x%016llx flags=0x%04x]\n",
+ PCI_BUS(devid), PCI_SLOT(devid), PCI_FUNC(devid),
+ domid, address, flags);
+ break;
+ case EVENT_TYPE_DEV_TAB_ERR:
+ printk("DEV_TAB_HARDWARE_ERROR device=%02x:%02x.%x "
+ "address=0x%016llx flags=0x%04x]\n",
+ PCI_BUS(devid), PCI_SLOT(devid), PCI_FUNC(devid),
+ address, flags);
+ break;
+ case EVENT_TYPE_PAGE_TAB_ERR:
+ printk("PAGE_TAB_HARDWARE_ERROR device=%02x:%02x.%x "
+ "domain=0x%04x address=0x%016llx flags=0x%04x]\n",
+ PCI_BUS(devid), PCI_SLOT(devid), PCI_FUNC(devid),
+ domid, address, flags);
+ break;
+ case EVENT_TYPE_ILL_CMD:
+ printk("ILLEGAL_COMMAND_ERROR address=0x%016llx]\n", address);
+ break;
+ case EVENT_TYPE_CMD_HARD_ERR:
+ printk("COMMAND_HARDWARE_ERROR address=0x%016llx "
+ "flags=0x%04x]\n", address, flags);
+ break;
+ case EVENT_TYPE_IOTLB_INV_TO:
+ printk("IOTLB_INV_TIMEOUT device=%02x:%02x.%x "
+ "address=0x%016llx]\n",
+ PCI_BUS(devid), PCI_SLOT(devid), PCI_FUNC(devid),
+ address);
+ break;
+ case EVENT_TYPE_INV_DEV_REQ:
+ printk("INVALID_DEVICE_REQUEST device=%02x:%02x.%x "
+ "address=0x%016llx flags=0x%04x]\n",
+ PCI_BUS(devid), PCI_SLOT(devid), PCI_FUNC(devid),
+ address, flags);
+ break;
+ default:
+ printk(KERN_ERR "UNKNOWN type=0x%02x]\n", type);
+ }
+}
+
+static void iommu_poll_events(struct amd_iommu *iommu)
+{
+ u32 head, tail;
+ unsigned long flags;
+
+ spin_lock_irqsave(&iommu->lock, flags);
+
+ head = readl(iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
+ tail = readl(iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
+
+ while (head != tail) {
+ iommu_print_event(iommu->evt_buf + head);
+ head = (head + EVENT_ENTRY_SIZE) % iommu->evt_buf_size;
+ }
+
+ writel(head, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
+
+ spin_unlock_irqrestore(&iommu->lock, flags);
+}
+
+irqreturn_t amd_iommu_int_handler(int irq, void *data)
+{
+ struct amd_iommu *iommu;
+
+ list_for_each_entry(iommu, &amd_iommu_list, list)
+ iommu_poll_events(iommu);
+
+ return IRQ_HANDLED;
+}
+
+/****************************************************************************
+ *
* IOMMU command queuing functions
*
****************************************************************************/
@@ -213,6 +313,14 @@ static int iommu_flush_pages(struct amd_iommu *iommu, u16 domid,
return 0;
}

+/* Flush the whole IO/TLB for a given protection domain */
+static void iommu_flush_tlb(struct amd_iommu *iommu, u16 domid)
+{
+ u64 address = CMD_INV_IOMMU_ALL_PAGES_ADDRESS;
+
+ iommu_queue_inv_iommu_pages(iommu, address, domid, 0, 1);
+}
+
/****************************************************************************
*
* The functions below are used the create the page table mappings for
@@ -372,11 +480,6 @@ static int init_unity_mappings_for_device(struct dma_ops_domain *dma_dom,
* efficient allocator.
*
****************************************************************************/
-static unsigned long dma_mask_to_pages(unsigned long mask)
-{
- return (mask >> PAGE_SHIFT) +
- (PAGE_ALIGN(mask & ~PAGE_MASK) >> PAGE_SHIFT);
-}

/*
* The address allocator core function.
@@ -385,25 +488,31 @@ static unsigned long dma_mask_to_pages(unsigned long mask)
*/
static unsigned long dma_ops_alloc_addresses(struct device *dev,
struct dma_ops_domain *dom,
- unsigned int pages)
+ unsigned int pages,
+ unsigned long align_mask,
+ u64 dma_mask)
{
- unsigned long limit = dma_mask_to_pages(*dev->dma_mask);
+ unsigned long limit;
unsigned long address;
- unsigned long size = dom->aperture_size >> PAGE_SHIFT;
unsigned long boundary_size;

boundary_size = ALIGN(dma_get_seg_boundary(dev) + 1,
PAGE_SIZE) >> PAGE_SHIFT;
- limit = limit < size ? limit : size;
+ limit = iommu_device_max_index(dom->aperture_size >> PAGE_SHIFT, 0,
+ dma_mask >> PAGE_SHIFT);

- if (dom->next_bit >= limit)
+ if (dom->next_bit >= limit) {
dom->next_bit = 0;
+ dom->need_flush = true;
+ }

address = iommu_area_alloc(dom->bitmap, limit, dom->next_bit, pages,
- 0 , boundary_size, 0);
- if (address == -1)
+ 0 , boundary_size, align_mask);
+ if (address == -1) {
address = iommu_area_alloc(dom->bitmap, limit, 0, pages,
- 0, boundary_size, 0);
+ 0, boundary_size, align_mask);
+ dom->need_flush = true;
+ }

if (likely(address != -1)) {
dom->next_bit = address + pages;
@@ -469,7 +578,7 @@ static void dma_ops_reserve_addresses(struct dma_ops_domain *dom,
if (start_page + pages > last_page)
pages = last_page - start_page;

- set_bit_string(dom->bitmap, start_page, pages);
+ iommu_area_reserve(dom->bitmap, start_page, pages);
}

static void dma_ops_free_pagetable(struct dma_ops_domain *dma_dom)
@@ -563,6 +672,9 @@ static struct dma_ops_domain *dma_ops_domain_alloc(struct amd_iommu *iommu,
dma_dom->bitmap[0] = 1;
dma_dom->next_bit = 0;

+ dma_dom->need_flush = false;
+ dma_dom->target_dev = 0xffff;
+
/* Intialize the exclusion range if necessary */
if (iommu->exclusion_start &&
iommu->exclusion_start < dma_dom->aperture_size) {
@@ -633,12 +745,13 @@ static void set_device_domain(struct amd_iommu *iommu,

u64 pte_root = virt_to_phys(domain->pt_root);

- pte_root |= (domain->mode & 0x07) << 9;
- pte_root |= IOMMU_PTE_IR | IOMMU_PTE_IW | IOMMU_PTE_P | 2;
+ pte_root |= (domain->mode & DEV_ENTRY_MODE_MASK)
+ << DEV_ENTRY_MODE_SHIFT;
+ pte_root |= IOMMU_PTE_IR | IOMMU_PTE_IW | IOMMU_PTE_P | IOMMU_PTE_TV;

write_lock_irqsave(&amd_iommu_devtable_lock, flags);
- amd_iommu_dev_table[devid].data[0] = pte_root;
- amd_iommu_dev_table[devid].data[1] = pte_root >> 32;
+ amd_iommu_dev_table[devid].data[0] = lower_32_bits(pte_root);
+ amd_iommu_dev_table[devid].data[1] = upper_32_bits(pte_root);
amd_iommu_dev_table[devid].data[2] = domain->id;

amd_iommu_pd_table[devid] = domain;
@@ -656,6 +769,45 @@ static void set_device_domain(struct amd_iommu *iommu,
*****************************************************************************/

/*
+ * This function checks if the driver got a valid device from the caller to
+ * avoid dereferencing invalid pointers.
+ */
+static bool check_device(struct device *dev)
+{
+ if (!dev || !dev->dma_mask)
+ return false;
+
+ return true;
+}
+
+/*
+ * In this function the list of preallocated protection domains is traversed to
+ * find the domain for a specific device
+ */
+static struct dma_ops_domain *find_protection_domain(u16 devid)
+{
+ struct dma_ops_domain *entry, *ret = NULL;
+ unsigned long flags;
+
+ if (list_empty(&iommu_pd_list))
+ return NULL;
+
+ spin_lock_irqsave(&iommu_pd_list_lock, flags);
+
+ list_for_each_entry(entry, &iommu_pd_list, list) {
+ if (entry->target_dev == devid) {
+ ret = entry;
+ list_del(&ret->list);
+ break;
+ }
+ }
+
+ spin_unlock_irqrestore(&iommu_pd_list_lock, flags);
+
+ return ret;
+}
+
+/*
* In the dma_ops path we only have the struct device. This function
* finds the corresponding IOMMU, the protection domain and the
* requestor id for a given device.
@@ -671,27 +823,30 @@ static int get_device_resources(struct device *dev,
struct pci_dev *pcidev;
u16 _bdf;

- BUG_ON(!dev || dev->bus != &pci_bus_type || !dev->dma_mask);
+ *iommu = NULL;
+ *domain = NULL;
+ *bdf = 0xffff;
+
+ if (dev->bus != &pci_bus_type)
+ return 0;

pcidev = to_pci_dev(dev);
_bdf = calc_devid(pcidev->bus->number, pcidev->devfn);

/* device not translated by any IOMMU in the system? */
- if (_bdf > amd_iommu_last_bdf) {
- *iommu = NULL;
- *domain = NULL;
- *bdf = 0xffff;
+ if (_bdf > amd_iommu_last_bdf)
return 0;
- }

*bdf = amd_iommu_alias_table[_bdf];

*iommu = amd_iommu_rlookup_table[*bdf];
if (*iommu == NULL)
return 0;
- dma_dom = (*iommu)->default_dom;
*domain = domain_for_device(*bdf);
if (*domain == NULL) {
+ dma_dom = find_protection_domain(*bdf);
+ if (!dma_dom)
+ dma_dom = (*iommu)->default_dom;
*domain = &dma_dom->domain;
set_device_domain(*iommu, *domain, *bdf);
printk(KERN_INFO "AMD IOMMU: Using protection domain %d for "
@@ -770,17 +925,24 @@ static dma_addr_t __map_single(struct device *dev,
struct dma_ops_domain *dma_dom,
phys_addr_t paddr,
size_t size,
- int dir)
+ int dir,
+ bool align,
+ u64 dma_mask)
{
dma_addr_t offset = paddr & ~PAGE_MASK;
dma_addr_t address, start;
unsigned int pages;
+ unsigned long align_mask = 0;
int i;

pages = iommu_num_pages(paddr, size);
paddr &= PAGE_MASK;

- address = dma_ops_alloc_addresses(dev, dma_dom, pages);
+ if (align)
+ align_mask = (1UL << get_order(size)) - 1;
+
+ address = dma_ops_alloc_addresses(dev, dma_dom, pages, align_mask,
+ dma_mask);
if (unlikely(address == bad_dma_address))
goto out;

@@ -792,6 +954,12 @@ static dma_addr_t __map_single(struct device *dev,
}
address += offset;

+ if (unlikely(dma_dom->need_flush && !amd_iommu_unmap_flush)) {
+ iommu_flush_tlb(iommu, dma_dom->domain.id);
+ dma_dom->need_flush = false;
+ } else if (unlikely(iommu_has_npcache(iommu)))
+ iommu_flush_pages(iommu, dma_dom->domain.id, address, size);
+
out:
return address;
}
@@ -822,6 +990,9 @@ static void __unmap_single(struct amd_iommu *iommu,
}

dma_ops_free_addresses(dma_dom, dma_addr, pages);
+
+ if (amd_iommu_unmap_flush)
+ iommu_flush_pages(iommu, dma_dom->domain.id, dma_addr, size);
}

/*
@@ -835,6 +1006,12 @@ static dma_addr_t map_single(struct device *dev, phys_addr_t paddr,
struct protection_domain *domain;
u16 devid;
dma_addr_t addr;
+ u64 dma_mask;
+
+ if (!check_device(dev))
+ return bad_dma_address;
+
+ dma_mask = *dev->dma_mask;

get_device_resources(dev, &iommu, &domain, &devid);

@@ -843,14 +1020,12 @@ static dma_addr_t map_single(struct device *dev, phys_addr_t paddr,
return (dma_addr_t)paddr;

spin_lock_irqsave(&domain->lock, flags);
- addr = __map_single(dev, iommu, domain->priv, paddr, size, dir);
+ addr = __map_single(dev, iommu, domain->priv, paddr, size, dir, false,
+ dma_mask);
if (addr == bad_dma_address)
goto out;

- if (iommu_has_npcache(iommu))
- iommu_flush_pages(iommu, domain->id, addr, size);
-
- if (iommu->need_sync)
+ if (unlikely(iommu->need_sync))
iommu_completion_wait(iommu);

out:
@@ -870,7 +1045,8 @@ static void unmap_single(struct device *dev, dma_addr_t dma_addr,
struct protection_domain *domain;
u16 devid;

- if (!get_device_resources(dev, &iommu, &domain, &devid))
+ if (!check_device(dev) ||
+ !get_device_resources(dev, &iommu, &domain, &devid))
/* device not handled by any AMD IOMMU */
return;

@@ -878,9 +1054,7 @@ static void unmap_single(struct device *dev, dma_addr_t dma_addr,

__unmap_single(iommu, domain->priv, dma_addr, size, dir);

- iommu_flush_pages(iommu, domain->id, dma_addr, size);
-
- if (iommu->need_sync)
+ if (unlikely(iommu->need_sync))
iommu_completion_wait(iommu);

spin_unlock_irqrestore(&domain->lock, flags);
@@ -919,6 +1093,12 @@ static int map_sg(struct device *dev, struct scatterlist *sglist,
struct scatterlist *s;
phys_addr_t paddr;
int mapped_elems = 0;
+ u64 dma_mask;
+
+ if (!check_device(dev))
+ return 0;
+
+ dma_mask = *dev->dma_mask;

get_device_resources(dev, &iommu, &domain, &devid);

@@ -931,19 +1111,17 @@ static int map_sg(struct device *dev, struct scatterlist *sglist,
paddr = sg_phys(s);

s->dma_address = __map_single(dev, iommu, domain->priv,
- paddr, s->length, dir);
+ paddr, s->length, dir, false,
+ dma_mask);

if (s->dma_address) {
s->dma_length = s->length;
mapped_elems++;
} else
goto unmap;
- if (iommu_has_npcache(iommu))
- iommu_flush_pages(iommu, domain->id, s->dma_address,
- s->dma_length);
}

- if (iommu->need_sync)
+ if (unlikely(iommu->need_sync))
iommu_completion_wait(iommu);

out:
@@ -977,7 +1155,8 @@ static void unmap_sg(struct device *dev, struct scatterlist *sglist,
u16 devid;
int i;

- if (!get_device_resources(dev, &iommu, &domain, &devid))
+ if (!check_device(dev) ||
+ !get_device_resources(dev, &iommu, &domain, &devid))
return;

spin_lock_irqsave(&domain->lock, flags);
@@ -985,12 +1164,10 @@ static void unmap_sg(struct device *dev, struct scatterlist *sglist,
for_each_sg(sglist, s, nelems, i) {
__unmap_single(iommu, domain->priv, s->dma_address,
s->dma_length, dir);
- iommu_flush_pages(iommu, domain->id, s->dma_address,
- s->dma_length);
s->dma_address = s->dma_length = 0;
}

- if (iommu->need_sync)
+ if (unlikely(iommu->need_sync))
iommu_completion_wait(iommu);

spin_unlock_irqrestore(&domain->lock, flags);
@@ -1008,25 +1185,33 @@ static void *alloc_coherent(struct device *dev, size_t size,
struct protection_domain *domain;
u16 devid;
phys_addr_t paddr;
+ u64 dma_mask = dev->coherent_dma_mask;
+
+ if (!check_device(dev))
+ return NULL;
+
+ if (!get_device_resources(dev, &iommu, &domain, &devid))
+ flag &= ~(__GFP_DMA | __GFP_HIGHMEM | __GFP_DMA32);

+ flag |= __GFP_ZERO;
virt_addr = (void *)__get_free_pages(flag, get_order(size));
if (!virt_addr)
return 0;

- memset(virt_addr, 0, size);
paddr = virt_to_phys(virt_addr);

- get_device_resources(dev, &iommu, &domain, &devid);
-
if (!iommu || !domain) {
*dma_addr = (dma_addr_t)paddr;
return virt_addr;
}

+ if (!dma_mask)
+ dma_mask = *dev->dma_mask;
+
spin_lock_irqsave(&domain->lock, flags);

*dma_addr = __map_single(dev, iommu, domain->priv, paddr,
- size, DMA_BIDIRECTIONAL);
+ size, DMA_BIDIRECTIONAL, true, dma_mask);

if (*dma_addr == bad_dma_address) {
free_pages((unsigned long)virt_addr, get_order(size));
@@ -1034,10 +1219,7 @@ static void *alloc_coherent(struct device *dev, size_t size,
goto out;
}

- if (iommu_has_npcache(iommu))
- iommu_flush_pages(iommu, domain->id, *dma_addr, size);
-
- if (iommu->need_sync)
+ if (unlikely(iommu->need_sync))
iommu_completion_wait(iommu);

out:
@@ -1048,8 +1230,6 @@ out:

/*
* The exported free_coherent function for dma_ops.
- * FIXME: fix the generic x86 DMA layer so that it actually calls that
- * function.
*/
static void free_coherent(struct device *dev, size_t size,
void *virt_addr, dma_addr_t dma_addr)
@@ -1059,6 +1239,9 @@ static void free_coherent(struct device *dev, size_t size,
struct protection_domain *domain;
u16 devid;

+ if (!check_device(dev))
+ return;
+
get_device_resources(dev, &iommu, &domain, &devid);

if (!iommu || !domain)
@@ -1067,9 +1250,8 @@ static void free_coherent(struct device *dev, size_t size,
spin_lock_irqsave(&domain->lock, flags);

__unmap_single(iommu, domain->priv, dma_addr, size, DMA_BIDIRECTIONAL);
- iommu_flush_pages(iommu, domain->id, dma_addr, size);

- if (iommu->need_sync)
+ if (unlikely(iommu->need_sync))
iommu_completion_wait(iommu);

spin_unlock_irqrestore(&domain->lock, flags);
@@ -1079,6 +1261,30 @@ free_mem:
}

/*
+ * This function is called by the DMA layer to find out if we can handle a
+ * particular device. It is part of the dma_ops.
+ */
+static int amd_iommu_dma_supported(struct device *dev, u64 mask)
+{
+ u16 bdf;
+ struct pci_dev *pcidev;
+
+ /* No device or no PCI device */
+ if (!dev || dev->bus != &pci_bus_type)
+ return 0;
+
+ pcidev = to_pci_dev(dev);
+
+ bdf = calc_devid(pcidev->bus->number, pcidev->devfn);
+
+ /* Out of our scope? */
+ if (bdf > amd_iommu_last_bdf)
+ return 0;
+
+ return 1;
+}
+
+/*
* The function for pre-allocating protection domains.
*
* If the driver core informs the DMA layer if a driver grabs a device
@@ -1107,10 +1313,9 @@ void prealloc_protection_domains(void)
if (!dma_dom)
continue;
init_unity_mappings_for_device(dma_dom, devid);
- set_device_domain(iommu, &dma_dom->domain, devid);
- printk(KERN_INFO "AMD IOMMU: Allocated domain %d for device ",
- dma_dom->domain.id);
- print_devid(devid, 1);
+ dma_dom->target_dev = devid;
+
+ list_add_tail(&dma_dom->list, &iommu_pd_list);
}
}

@@ -1121,6 +1326,7 @@ static struct dma_mapping_ops amd_iommu_dma_ops = {
.unmap_single = unmap_single,
.map_sg = map_sg,
.unmap_sg = unmap_sg,
+ .dma_supported = amd_iommu_dma_supported,
};

/*
diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c
index a69cc0f..148fcfe 100644
--- a/arch/x86/kernel/amd_iommu_init.c
+++ b/arch/x86/kernel/amd_iommu_init.c
@@ -22,6 +22,8 @@
#include <linux/gfp.h>
#include <linux/list.h>
#include <linux/sysdev.h>
+#include <linux/interrupt.h>
+#include <linux/msi.h>
#include <asm/pci-direct.h>
#include <asm/amd_iommu_types.h>
#include <asm/amd_iommu.h>
@@ -30,7 +32,6 @@
/*
* definitions for the ACPI scanning code
*/
-#define PCI_BUS(x) (((x) >> 8) & 0xff)
#define IVRS_HEADER_LENGTH 48

#define ACPI_IVHD_TYPE 0x10
@@ -121,6 +122,7 @@ LIST_HEAD(amd_iommu_unity_map); /* a list of required unity mappings
we find in ACPI */
unsigned amd_iommu_aperture_order = 26; /* size of aperture in power of 2 */
int amd_iommu_isolate; /* if 1, device isolation is enabled */
+bool amd_iommu_unmap_flush; /* if true, flush on every unmap */

LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
system */
@@ -234,7 +236,7 @@ static void __init iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
{
u32 ctrl;

- ctrl = (u64)readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
+ ctrl = readl(iommu->mmio_base + MMIO_CONTROL_OFFSET);
ctrl &= ~(1 << bit);
writel(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
}
@@ -242,13 +244,23 @@ static void __init iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
/* Function to enable the hardware */
void __init iommu_enable(struct amd_iommu *iommu)
{
- printk(KERN_INFO "AMD IOMMU: Enabling IOMMU at ");
- print_devid(iommu->devid, 0);
- printk(" cap 0x%hx\n", iommu->cap_ptr);
+ printk(KERN_INFO "AMD IOMMU: Enabling IOMMU "
+ "at %02x:%02x.%x cap 0x%hx\n",
+ iommu->dev->bus->number,
+ PCI_SLOT(iommu->dev->devfn),
+ PCI_FUNC(iommu->dev->devfn),
+ iommu->cap_ptr);

iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
}

+/* Function to enable IOMMU event logging and event interrupts */
+void __init iommu_enable_event_logging(struct amd_iommu *iommu)
+{
+ iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
+ iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
+}
+
/*
* mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
* the system has one.
@@ -286,6 +298,14 @@ static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
****************************************************************************/

/*
+ * This function calculates the length of a given IVHD entry
+ */
+static inline int ivhd_entry_length(u8 *ivhd)
+{
+ return 0x04 << (*ivhd >> 6);
+}
+
+/*
* This function reads the last device id the IOMMU has to handle from the PCI
* capability header for this IOMMU
*/
@@ -329,7 +349,7 @@ static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
default:
break;
}
- p += 0x04 << (*p >> 6);
+ p += ivhd_entry_length(p);
}

WARN_ON(p != end);
@@ -414,7 +434,32 @@ static u8 * __init alloc_command_buffer(struct amd_iommu *iommu)

static void __init free_command_buffer(struct amd_iommu *iommu)
{
- free_pages((unsigned long)iommu->cmd_buf, get_order(CMD_BUFFER_SIZE));
+ free_pages((unsigned long)iommu->cmd_buf,
+ get_order(iommu->cmd_buf_size));
+}
+
+/* allocates the memory where the IOMMU will log its events to */
+static u8 * __init alloc_event_buffer(struct amd_iommu *iommu)
+{
+ u64 entry;
+ iommu->evt_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
+ get_order(EVT_BUFFER_SIZE));
+
+ if (iommu->evt_buf == NULL)
+ return NULL;
+
+ entry = (u64)virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
+ memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
+ &entry, sizeof(entry));
+
+ iommu->evt_buf_size = EVT_BUFFER_SIZE;
+
+ return iommu->evt_buf;
+}
+
+static void __init free_event_buffer(struct amd_iommu *iommu)
+{
+ free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
}

/* sets a specific bit in the device table entry. */
@@ -487,19 +532,21 @@ static void __init set_device_exclusion_range(u16 devid, struct ivmd_header *m)
*/
static void __init init_iommu_from_pci(struct amd_iommu *iommu)
{
- int bus = PCI_BUS(iommu->devid);
- int dev = PCI_SLOT(iommu->devid);
- int fn = PCI_FUNC(iommu->devid);
int cap_ptr = iommu->cap_ptr;
- u32 range;
+ u32 range, misc;

- iommu->cap = read_pci_config(bus, dev, fn, cap_ptr+MMIO_CAP_HDR_OFFSET);
+ pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
+ &iommu->cap);
+ pci_read_config_dword(iommu->dev, cap_ptr + MMIO_RANGE_OFFSET,
+ &range);
+ pci_read_config_dword(iommu->dev, cap_ptr + MMIO_MISC_OFFSET,
+ &misc);

- range = read_pci_config(bus, dev, fn, cap_ptr+MMIO_RANGE_OFFSET);
iommu->first_device = calc_devid(MMIO_GET_BUS(range),
MMIO_GET_FD(range));
iommu->last_device = calc_devid(MMIO_GET_BUS(range),
MMIO_GET_LD(range));
+ iommu->evt_msi_num = MMIO_MSI_NUM(misc);
}

/*
@@ -604,7 +651,7 @@ static void __init init_iommu_from_acpi(struct amd_iommu *iommu,
break;
}

- p += 0x04 << (e->type >> 6);
+ p += ivhd_entry_length(p);
}
}

@@ -622,6 +669,7 @@ static int __init init_iommu_devices(struct amd_iommu *iommu)
static void __init free_iommu_one(struct amd_iommu *iommu)
{
free_command_buffer(iommu);
+ free_event_buffer(iommu);
iommu_unmap_mmio_space(iommu);
}

@@ -649,8 +697,12 @@ static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
/*
* Copy data from ACPI table entry to the iommu struct
*/
- iommu->devid = h->devid;
+ iommu->dev = pci_get_bus_and_slot(PCI_BUS(h->devid), h->devid & 0xff);
+ if (!iommu->dev)
+ return 1;
+
iommu->cap_ptr = h->cap_ptr;
+ iommu->pci_seg = h->pci_seg;
iommu->mmio_phys = h->mmio_phys;
iommu->mmio_base = iommu_map_mmio_space(h->mmio_phys);
if (!iommu->mmio_base)
@@ -661,10 +713,18 @@ static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
if (!iommu->cmd_buf)
return -ENOMEM;

+ iommu->evt_buf = alloc_event_buffer(iommu);
+ if (!iommu->evt_buf)
+ return -ENOMEM;
+
+ iommu->int_enabled = false;
+
init_iommu_from_pci(iommu);
init_iommu_from_acpi(iommu, h);
init_iommu_devices(iommu);

+ pci_enable_device(iommu->dev);
+
return 0;
}

@@ -706,6 +766,95 @@ static int __init init_iommu_all(struct acpi_table_header *table)

/****************************************************************************
*
+ * The following functions initialize the MSI interrupts for all IOMMUs
+ * in the system. Its a bit challenging because there could be multiple
+ * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
+ * pci_dev.
+ *
+ ****************************************************************************/
+
+static int __init iommu_setup_msix(struct amd_iommu *iommu)
+{
+ struct amd_iommu *curr;
+ struct msix_entry entries[32]; /* only 32 supported by AMD IOMMU */
+ int nvec = 0, i;
+
+ list_for_each_entry(curr, &amd_iommu_list, list) {
+ if (curr->dev == iommu->dev) {
+ entries[nvec].entry = curr->evt_msi_num;
+ entries[nvec].vector = 0;
+ curr->int_enabled = true;
+ nvec++;
+ }
+ }
+
+ if (pci_enable_msix(iommu->dev, entries, nvec)) {
+ pci_disable_msix(iommu->dev);
+ return 1;
+ }
+
+ for (i = 0; i < nvec; ++i) {
+ int r = request_irq(entries->vector, amd_iommu_int_handler,
+ IRQF_SAMPLE_RANDOM,
+ "AMD IOMMU",
+ NULL);
+ if (r)
+ goto out_free;
+ }
+
+ return 0;
+
+out_free:
+ for (i -= 1; i >= 0; --i)
+ free_irq(entries->vector, NULL);
+
+ pci_disable_msix(iommu->dev);
+
+ return 1;
+}
+
+static int __init iommu_setup_msi(struct amd_iommu *iommu)
+{
+ int r;
+ struct amd_iommu *curr;
+
+ list_for_each_entry(curr, &amd_iommu_list, list) {
+ if (curr->dev == iommu->dev)
+ curr->int_enabled = true;
+ }
+
+
+ if (pci_enable_msi(iommu->dev))
+ return 1;
+
+ r = request_irq(iommu->dev->irq, amd_iommu_int_handler,
+ IRQF_SAMPLE_RANDOM,
+ "AMD IOMMU",
+ NULL);
+
+ if (r) {
+ pci_disable_msi(iommu->dev);
+ return 1;
+ }
+
+ return 0;
+}
+
+static int __init iommu_init_msi(struct amd_iommu *iommu)
+{
+ if (iommu->int_enabled)
+ return 0;
+
+ if (pci_find_capability(iommu->dev, PCI_CAP_ID_MSIX))
+ return iommu_setup_msix(iommu);
+ else if (pci_find_capability(iommu->dev, PCI_CAP_ID_MSI))
+ return iommu_setup_msi(iommu);
+
+ return 1;
+}
+
+/****************************************************************************
+ *
* The next functions belong to the third pass of parsing the ACPI
* table. In this last pass the memory mapping requirements are
* gathered (like exclusion and unity mapping reanges).
@@ -811,7 +960,6 @@ static void init_device_table(void)
for (devid = 0; devid <= amd_iommu_last_bdf; ++devid) {
set_dev_entry_bit(devid, DEV_ENTRY_VALID);
set_dev_entry_bit(devid, DEV_ENTRY_TRANSLATION);
- set_dev_entry_bit(devid, DEV_ENTRY_NO_PAGE_FAULT);
}
}

@@ -825,6 +973,8 @@ static void __init enable_iommus(void)

list_for_each_entry(iommu, &amd_iommu_list, list) {
iommu_set_exclusion_range(iommu);
+ iommu_init_msi(iommu);
+ iommu_enable_event_logging(iommu);
iommu_enable(iommu);
}
}
@@ -995,11 +1145,17 @@ int __init amd_iommu_init(void)
else
printk("disabled\n");

+ if (amd_iommu_unmap_flush)
+ printk(KERN_INFO "AMD IOMMU: IO/TLB flush on unmap enabled\n");
+ else
+ printk(KERN_INFO "AMD IOMMU: Lazy IO/TLB flushing enabled\n");
+
out:
return ret;

free:
- free_pages((unsigned long)amd_iommu_pd_alloc_bitmap, 1);
+ free_pages((unsigned long)amd_iommu_pd_alloc_bitmap,
+ get_order(MAX_DOMAIN_ID/8));

free_pages((unsigned long)amd_iommu_pd_table,
get_order(rlookup_table_size));
@@ -1057,8 +1213,10 @@ void __init amd_iommu_detect(void)
static int __init parse_amd_iommu_options(char *str)
{
for (; *str; ++str) {
- if (strcmp(str, "isolate") == 0)
+ if (strncmp(str, "isolate", 7) == 0)
amd_iommu_isolate = 1;
+ if (strncmp(str, "fullflush", 11) == 0)
+ amd_iommu_unmap_flush = true;
}

return 1;
diff --git a/arch/x86/kernel/early-quirks.c b/arch/x86/kernel/early-quirks.c
index 4353cf5..24bb5fa 100644
--- a/arch/x86/kernel/early-quirks.c
+++ b/arch/x86/kernel/early-quirks.c
@@ -95,6 +95,20 @@ static void __init nvidia_bugs(int num, int slot, int func)

}

+#ifdef CONFIG_DMAR
+static void __init intel_g33_dmar(int num, int slot, int func)
+{
+ struct acpi_table_header *dmar_tbl;
+ acpi_status status;
+
+ status = acpi_get_table(ACPI_SIG_DMAR, 0, &dmar_tbl);
+ if (ACPI_SUCCESS(status)) {
+ printk(KERN_INFO "BIOS BUG: DMAR advertised on Intel G31/G33 chipset -- ignoring\n");
+ dmar_disabled = 1;
+ }
+}
+#endif
+
#define QFLAG_APPLY_ONCE 0x1
#define QFLAG_APPLIED 0x2
#define QFLAG_DONE (QFLAG_APPLY_ONCE|QFLAG_APPLIED)
@@ -114,6 +128,10 @@ static struct chipset early_qrk[] __initdata = {
PCI_CLASS_BRIDGE_PCI, PCI_ANY_ID, QFLAG_APPLY_ONCE, via_bugs },
{ PCI_VENDOR_ID_AMD, PCI_DEVICE_ID_AMD_K8_NB,
PCI_CLASS_BRIDGE_HOST, PCI_ANY_ID, 0, fix_hypertransport_config },
+#ifdef CONFIG_DMAR
+ { PCI_VENDOR_ID_INTEL, 0x29c0,
+ PCI_CLASS_BRIDGE_HOST, PCI_ANY_ID, 0, intel_g33_dmar },
+#endif
{}
};

diff --git a/arch/x86/kernel/k8.c b/arch/x86/kernel/k8.c
index 7377ccb..304d8ba 100644
--- a/arch/x86/kernel/k8.c
+++ b/arch/x86/kernel/k8.c
@@ -16,8 +16,9 @@ EXPORT_SYMBOL(num_k8_northbridges);
static u32 *flush_words;

struct pci_device_id k8_nb_ids[] = {
- { PCI_DEVICE(PCI_VENDOR_ID_AMD, 0x1103) },
- { PCI_DEVICE(PCI_VENDOR_ID_AMD, 0x1203) },
+ { PCI_DEVICE(PCI_VENDOR_ID_AMD, PCI_DEVICE_ID_AMD_K8_NB_MISC) },
+ { PCI_DEVICE(PCI_VENDOR_ID_AMD, PCI_DEVICE_ID_AMD_10H_NB_MISC) },
+ { PCI_DEVICE(PCI_VENDOR_ID_AMD, PCI_DEVICE_ID_AMD_11H_NB_MISC) },
{}
};
EXPORT_SYMBOL(k8_nb_ids);
diff --git a/arch/x86/kernel/pci-calgary_64.c b/arch/x86/kernel/pci-calgary_64.c
index dcdac6c..080d1d2 100644
--- a/arch/x86/kernel/pci-calgary_64.c
+++ b/arch/x86/kernel/pci-calgary_64.c
@@ -261,7 +261,7 @@ static void iommu_range_reserve(struct iommu_table *tbl,
badbit, tbl, start_addr, npages);
}

- set_bit_string(tbl->it_map, index, npages);
+ iommu_area_reserve(tbl->it_map, index, npages);

spin_unlock_irqrestore(&tbl->it_lock, flags);
}
@@ -491,6 +491,8 @@ static void* calgary_alloc_coherent(struct device *dev, size_t size,
npages = size >> PAGE_SHIFT;
order = get_order(size);

+ flag &= ~(__GFP_DMA | __GFP_HIGHMEM | __GFP_DMA32);
+
/* alloc enough pages (and possibly more) */
ret = (void *)__get_free_pages(flag, order);
if (!ret)
@@ -510,8 +512,22 @@ error:
return ret;
}

+static void calgary_free_coherent(struct device *dev, size_t size,
+ void *vaddr, dma_addr_t dma_handle)
+{
+ unsigned int npages;
+ struct iommu_table *tbl = find_iommu_table(dev);
+
+ size = PAGE_ALIGN(size);
+ npages = size >> PAGE_SHIFT;
+
+ iommu_free(tbl, dma_handle, npages);
+ free_pages((unsigned long)vaddr, get_order(size));
+}
+
static struct dma_mapping_ops calgary_dma_ops = {
.alloc_coherent = calgary_alloc_coherent,
+ .free_coherent = calgary_free_coherent,
.map_single = calgary_map_single,
.unmap_single = calgary_unmap_single,
.map_sg = calgary_map_sg,
diff --git a/arch/x86/kernel/pci-dma.c b/arch/x86/kernel/pci-dma.c
index f704cb5..0a3824e 100644
--- a/arch/x86/kernel/pci-dma.c
+++ b/arch/x86/kernel/pci-dma.c
@@ -41,11 +41,12 @@ EXPORT_SYMBOL(bad_dma_address);
/* Dummy device used for NULL arguments (normally ISA). Better would
be probably a smaller DMA mask, but this is bug-to-bug compatible
to older i386. */
-struct device fallback_dev = {
+struct device x86_dma_fallback_dev = {
.bus_id = "fallback device",
.coherent_dma_mask = DMA_32BIT_MASK,
- .dma_mask = &fallback_dev.coherent_dma_mask,
+ .dma_mask = &x86_dma_fallback_dev.coherent_dma_mask,
};
+EXPORT_SYMBOL(x86_dma_fallback_dev);

int dma_set_mask(struct device *dev, u64 mask)
{
@@ -133,6 +134,37 @@ unsigned long iommu_num_pages(unsigned long addr, unsigned long len)
EXPORT_SYMBOL(iommu_num_pages);
#endif

+void *dma_generic_alloc_coherent(struct device *dev, size_t size,
+ dma_addr_t *dma_addr, gfp_t flag)
+{
+ unsigned long dma_mask;
+ struct page *page;
+ dma_addr_t addr;
+
+ dma_mask = dma_alloc_coherent_mask(dev, flag);
+
+ flag |= __GFP_ZERO;
+again:
+ page = alloc_pages_node(dev_to_node(dev), flag, get_order(size));
+ if (!page)
+ return NULL;
+
+ addr = page_to_phys(page);
+ if (!is_buffer_dma_capable(dma_mask, addr, size)) {
+ __free_pages(page, get_order(size));
+
+ if (dma_mask < DMA_32BIT_MASK && !(flag & GFP_DMA)) {
+ flag = (flag & ~GFP_DMA32) | GFP_DMA;
+ goto again;
+ }
+
+ return NULL;
+ }
+
+ *dma_addr = addr;
+ return page_address(page);
+}
+
/*
* See <Documentation/x86_64/boot-options.txt> for the iommu kernel parameter
* documentation.
@@ -241,147 +273,6 @@ int dma_supported(struct device *dev, u64 mask)
}
EXPORT_SYMBOL(dma_supported);

-/* Allocate DMA memory on node near device */
-static noinline struct page *
-dma_alloc_pages(struct device *dev, gfp_t gfp, unsigned order)
-{
- int node;
-
- node = dev_to_node(dev);
-
- return alloc_pages_node(node, gfp, order);
-}
-
-/*
- * Allocate memory for a coherent mapping.
- */
-void *
-dma_alloc_coherent(struct device *dev, size_t size, dma_addr_t *dma_handle,
- gfp_t gfp)
-{
- struct dma_mapping_ops *ops = get_dma_ops(dev);
- void *memory = NULL;
- struct page *page;
- unsigned long dma_mask = 0;
- dma_addr_t bus;
- int noretry = 0;
-
- /* ignore region specifiers */
- gfp &= ~(__GFP_DMA | __GFP_HIGHMEM | __GFP_DMA32);
-
- if (dma_alloc_from_coherent(dev, size, dma_handle, &memory))
- return memory;
-
- if (!dev) {
- dev = &fallback_dev;
- gfp |= GFP_DMA;
- }
- dma_mask = dev->coherent_dma_mask;
- if (dma_mask == 0)
- dma_mask = (gfp & GFP_DMA) ? DMA_24BIT_MASK : DMA_32BIT_MASK;
-
- /* Device not DMA able */
- if (dev->dma_mask == NULL)
- return NULL;
-
- /* Don't invoke OOM killer or retry in lower 16MB DMA zone */
- if (gfp & __GFP_DMA)
- noretry = 1;
-
-#ifdef CONFIG_X86_64
- /* Why <=? Even when the mask is smaller than 4GB it is often
- larger than 16MB and in this case we have a chance of
- finding fitting memory in the next higher zone first. If
- not retry with true GFP_DMA. -AK */
- if (dma_mask <= DMA_32BIT_MASK && !(gfp & GFP_DMA)) {
- gfp |= GFP_DMA32;
- if (dma_mask < DMA_32BIT_MASK)
- noretry = 1;
- }
-#endif
-
- again:
- page = dma_alloc_pages(dev,
- noretry ? gfp | __GFP_NORETRY : gfp, get_order(size));
- if (page == NULL)
- return NULL;
-
- {
- int high, mmu;
- bus = page_to_phys(page);
- memory = page_address(page);
- high = (bus + size) >= dma_mask;
- mmu = high;
- if (force_iommu && !(gfp & GFP_DMA))
- mmu = 1;
- else if (high) {
- free_pages((unsigned long)memory,
- get_order(size));
-
- /* Don't use the 16MB ZONE_DMA unless absolutely
- needed. It's better to use remapping first. */
- if (dma_mask < DMA_32BIT_MASK && !(gfp & GFP_DMA)) {
- gfp = (gfp & ~GFP_DMA32) | GFP_DMA;
- goto again;
- }
-
- /* Let low level make its own zone decisions */
- gfp &= ~(GFP_DMA32|GFP_DMA);
-
- if (ops->alloc_coherent)
- return ops->alloc_coherent(dev, size,
- dma_handle, gfp);
- return NULL;
- }
-
- memset(memory, 0, size);
- if (!mmu) {
- *dma_handle = bus;
- return memory;
- }
- }
-
- if (ops->alloc_coherent) {
- free_pages((unsigned long)memory, get_order(size));
- gfp &= ~(GFP_DMA|GFP_DMA32);
- return ops->alloc_coherent(dev, size, dma_handle, gfp);
- }
-
- if (ops->map_simple) {
- *dma_handle = ops->map_simple(dev, virt_to_phys(memory),
- size,
- PCI_DMA_BIDIRECTIONAL);
- if (*dma_handle != bad_dma_address)
- return memory;
- }
-
- if (panic_on_overflow)
- panic("dma_alloc_coherent: IOMMU overflow by %lu bytes\n",
- (unsigned long)size);
- free_pages((unsigned long)memory, get_order(size));
- return NULL;
-}
-EXPORT_SYMBOL(dma_alloc_coherent);
-
-/*
- * Unmap coherent memory.
- * The caller must ensure that the device has finished accessing the mapping.
- */
-void dma_free_coherent(struct device *dev, size_t size,
- void *vaddr, dma_addr_t bus)
-{
- struct dma_mapping_ops *ops = get_dma_ops(dev);
-
- int order = get_order(size);
- WARN_ON(irqs_disabled()); /* for portability */
- if (dma_release_from_coherent(dev, order, vaddr))
- return;
- if (ops->unmap_single)
- ops->unmap_single(dev, bus, size, 0);
- free_pages((unsigned long)vaddr, order);
-}
-EXPORT_SYMBOL(dma_free_coherent);
-
static int __init pci_iommu_init(void)
{
calgary_iommu_init();
diff --git a/arch/x86/kernel/pci-gart_64.c b/arch/x86/kernel/pci-gart_64.c
index 1a895a5..145f1c8 100644
--- a/arch/x86/kernel/pci-gart_64.c
+++ b/arch/x86/kernel/pci-gart_64.c
@@ -27,8 +27,8 @@
#include <linux/scatterlist.h>
#include <linux/iommu-helper.h>
#include <linux/sysdev.h>
+#include <linux/io.h>
#include <asm/atomic.h>
-#include <asm/io.h>
#include <asm/mtrr.h>
#include <asm/pgtable.h>
#include <asm/proto.h>
@@ -80,7 +80,7 @@ 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 bool need_flush; /* global flush state. set for each gart wrap */

static unsigned long alloc_iommu(struct device *dev, int size,
unsigned long align_mask)
@@ -98,7 +98,7 @@ static unsigned long alloc_iommu(struct device *dev, int size,
offset = iommu_area_alloc(iommu_gart_bitmap, iommu_pages, next_bit,
size, base_index, boundary_size, align_mask);
if (offset == -1) {
- need_flush = 1;
+ need_flush = true;
offset = iommu_area_alloc(iommu_gart_bitmap, iommu_pages, 0,
size, base_index, boundary_size,
align_mask);
@@ -107,11 +107,11 @@ static unsigned long alloc_iommu(struct device *dev, int size,
next_bit = offset+size;
if (next_bit >= iommu_pages) {
next_bit = 0;
- need_flush = 1;
+ need_flush = true;
}
}
if (iommu_fullflush)
- need_flush = 1;
+ need_flush = true;
spin_unlock_irqrestore(&iommu_bitmap_lock, flags);

return offset;
@@ -136,7 +136,7 @@ static void flush_gart(void)
spin_lock_irqsave(&iommu_bitmap_lock, flags);
if (need_flush) {
k8_flush_garts();
- need_flush = 0;
+ need_flush = false;
}
spin_unlock_irqrestore(&iommu_bitmap_lock, flags);
}
@@ -175,7 +175,8 @@ static void dump_leak(void)
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_address((unsigned long) iommu_leak_tab[iommu_pages-i],
+ 0);
printk(KERN_CONT "%c", (i+1)%2 == 0 ? '\n' : ' ');
}
printk(KERN_DEBUG "\n");
@@ -214,24 +215,14 @@ static void iommu_full(struct device *dev, size_t size, int dir)
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;
+ return force_iommu ||
+ !is_buffer_dma_capable(*dev->dma_mask, addr, size);
}

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;
+ return !is_buffer_dma_capable(*dev->dma_mask, addr, size);
}

/* Map a single continuous physical area into the IOMMU.
@@ -261,20 +252,6 @@ static dma_addr_t dma_map_area(struct device *dev, dma_addr_t phys_mem,
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;
- unsigned long align_mask;
-
- align_mask = (1UL << get_order(size)) - 1;
- map = dma_map_area(dev, paddr, size, dir, align_mask);
-
- 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)
@@ -282,7 +259,7 @@ gart_map_single(struct device *dev, phys_addr_t paddr, size_t size, int dir)
unsigned long bus;

if (!dev)
- dev = &fallback_dev;
+ dev = &x86_dma_fallback_dev;

if (!need_iommu(dev, paddr, size))
return paddr;
@@ -434,7 +411,7 @@ gart_map_sg(struct device *dev, struct scatterlist *sg, int nents, int dir)
return 0;

if (!dev)
- dev = &fallback_dev;
+ dev = &x86_dma_fallback_dev;

out = 0;
start = 0;
@@ -506,6 +483,46 @@ error:
return 0;
}

+/* allocate and map a coherent mapping */
+static void *
+gart_alloc_coherent(struct device *dev, size_t size, dma_addr_t *dma_addr,
+ gfp_t flag)
+{
+ dma_addr_t paddr;
+ unsigned long align_mask;
+ struct page *page;
+
+ if (force_iommu && !(flag & GFP_DMA)) {
+ flag &= ~(__GFP_DMA | __GFP_HIGHMEM | __GFP_DMA32);
+ page = alloc_pages(flag | __GFP_ZERO, get_order(size));
+ if (!page)
+ return NULL;
+
+ align_mask = (1UL << get_order(size)) - 1;
+ paddr = dma_map_area(dev, page_to_phys(page), size,
+ DMA_BIDIRECTIONAL, align_mask);
+
+ flush_gart();
+ if (paddr != bad_dma_address) {
+ *dma_addr = paddr;
+ return page_address(page);
+ }
+ __free_pages(page, get_order(size));
+ } else
+ return dma_generic_alloc_coherent(dev, size, dma_addr, flag);
+
+ return NULL;
+}
+
+/* free a coherent mapping */
+static void
+gart_free_coherent(struct device *dev, size_t size, void *vaddr,
+ dma_addr_t dma_addr)
+{
+ gart_unmap_single(dev, dma_addr, size, DMA_BIDIRECTIONAL);
+ free_pages((unsigned long)vaddr, get_order(size));
+}
+
static int no_agp;

static __init unsigned long check_iommu_size(unsigned long aper, u64 aper_size)
@@ -656,13 +673,13 @@ static __init int init_k8_gatt(struct agp_kern_info *info)
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));
+ gatt = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
+ 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();
@@ -671,7 +688,8 @@ static __init int init_k8_gatt(struct agp_kern_info *info)
if (!error)
error = sysdev_register(&device_gart);
if (error)
- panic("Could not register gart_sysdev -- would corrupt data on next suspend");
+ panic("Could not register gart_sysdev -- "
+ "would corrupt data on next suspend");

flush_gart();

@@ -687,20 +705,13 @@ static __init int init_k8_gatt(struct agp_kern_info *info)
return -1;
}

-extern int agp_amd64_init(void);
-
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,
+ .alloc_coherent = gart_alloc_coherent,
+ .free_coherent = gart_free_coherent,
};

void gart_iommu_shutdown(void)
@@ -760,8 +771,8 @@ void __init gart_iommu_init(void)
(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");
+ "but GART IOMMU not available.\n");
+ printk(KERN_WARNING "falling back to iommu=soft.\n");
}
return;
}
@@ -779,19 +790,16 @@ void __init gart_iommu_init(void)
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,
+ iommu_gart_bitmap = (void *) __get_free_pages(GFP_KERNEL | __GFP_ZERO,
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,
+ iommu_leak_tab = (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
get_order(iommu_pages*sizeof(void *)));
- if (iommu_leak_tab)
- memset(iommu_leak_tab, 0, iommu_pages * 8);
- else
+ if (!iommu_leak_tab)
printk(KERN_DEBUG
"PCI-DMA: Cannot allocate leak trace area\n");
}
@@ -801,7 +809,7 @@ void __init gart_iommu_init(void)
* Out of IOMMU space handling.
* Reserve some invalid pages at the beginning of the GART.
*/
- set_bit_string(iommu_gart_bitmap, 0, EMERGENCY_PAGES);
+ iommu_area_reserve(iommu_gart_bitmap, 0, EMERGENCY_PAGES);

agp_memory_reserved = iommu_size;
printk(KERN_INFO
@@ -859,7 +867,8 @@ void __init gart_parse_options(char *p)
if (!strncmp(p, "leak", 4)) {
leak_trace = 1;
p += 4;
- if (*p == '=') ++p;
+ if (*p == '=')
+ ++p;
if (isdigit(*p) && get_option(&p, &arg))
iommu_leak_pages = arg;
}
diff --git a/arch/x86/kernel/pci-nommu.c b/arch/x86/kernel/pci-nommu.c
index 3f91f71..c70ab5a 100644
--- a/arch/x86/kernel/pci-nommu.c
+++ b/arch/x86/kernel/pci-nommu.c
@@ -14,7 +14,7 @@
static int
check_addr(char *name, struct device *hwdev, dma_addr_t bus, size_t size)
{
- if (hwdev && bus + size > *hwdev->dma_mask) {
+ if (hwdev && !is_buffer_dma_capable(*hwdev->dma_mask, bus, size)) {
if (*hwdev->dma_mask >= DMA_32BIT_MASK)
printk(KERN_ERR
"nommu_%s: overflow %Lx+%zu of device mask %Lx\n",
@@ -72,7 +72,15 @@ static int nommu_map_sg(struct device *hwdev, struct scatterlist *sg,
return nents;
}

+static void nommu_free_coherent(struct device *dev, size_t size, void *vaddr,
+ dma_addr_t dma_addr)
+{
+ free_pages((unsigned long)vaddr, get_order(size));
+}
+
struct dma_mapping_ops nommu_dma_ops = {
+ .alloc_coherent = dma_generic_alloc_coherent,
+ .free_coherent = nommu_free_coherent,
.map_single = nommu_map_single,
.map_sg = nommu_map_sg,
.is_phys = 1,
diff --git a/drivers/pci/intel-iommu.c b/drivers/pci/intel-iommu.c
index c3edcdc..6c4c1c3 100644
--- a/drivers/pci/intel-iommu.c
+++ b/drivers/pci/intel-iommu.c
@@ -80,7 +80,7 @@ static long list_size;

static void domain_remove_dev_info(struct dmar_domain *domain);

-static int dmar_disabled;
+int dmar_disabled;
static int __initdata dmar_map_gfx = 1;
static int dmar_forcedac;
static int intel_iommu_strict;
diff --git a/include/asm-x86/amd_iommu.h b/include/asm-x86/amd_iommu.h
index 783f43e..041d0db 100644
--- a/include/asm-x86/amd_iommu.h
+++ b/include/asm-x86/amd_iommu.h
@@ -20,10 +20,13 @@
#ifndef ASM_X86__AMD_IOMMU_H
#define ASM_X86__AMD_IOMMU_H

+#include <linux/irqreturn.h>
+
#ifdef CONFIG_AMD_IOMMU
extern int amd_iommu_init(void);
extern int amd_iommu_init_dma_ops(void);
extern void amd_iommu_detect(void);
+extern irqreturn_t amd_iommu_int_handler(int irq, void *data);
#else
static inline int amd_iommu_init(void) { return -ENODEV; }
static inline void amd_iommu_detect(void) { }
diff --git a/include/asm-x86/amd_iommu_types.h b/include/asm-x86/amd_iommu_types.h
index 1ffa4e5..b308586 100644
--- a/include/asm-x86/amd_iommu_types.h
+++ b/include/asm-x86/amd_iommu_types.h
@@ -37,6 +37,7 @@
/* Capability offsets used by the driver */
#define MMIO_CAP_HDR_OFFSET 0x00
#define MMIO_RANGE_OFFSET 0x0c
+#define MMIO_MISC_OFFSET 0x10

/* Masks, shifts and macros to parse the device range capability */
#define MMIO_RANGE_LD_MASK 0xff000000
@@ -48,6 +49,7 @@
#define MMIO_GET_LD(x) (((x) & MMIO_RANGE_LD_MASK) >> MMIO_RANGE_LD_SHIFT)
#define MMIO_GET_FD(x) (((x) & MMIO_RANGE_FD_MASK) >> MMIO_RANGE_FD_SHIFT)
#define MMIO_GET_BUS(x) (((x) & MMIO_RANGE_BUS_MASK) >> MMIO_RANGE_BUS_SHIFT)
+#define MMIO_MSI_NUM(x) ((x) & 0x1f)

/* Flag masks for the AMD IOMMU exclusion range */
#define MMIO_EXCL_ENABLE_MASK 0x01ULL
@@ -69,6 +71,25 @@
/* MMIO status bits */
#define MMIO_STATUS_COM_WAIT_INT_MASK 0x04

+/* event logging constants */
+#define EVENT_ENTRY_SIZE 0x10
+#define EVENT_TYPE_SHIFT 28
+#define EVENT_TYPE_MASK 0xf
+#define EVENT_TYPE_ILL_DEV 0x1
+#define EVENT_TYPE_IO_FAULT 0x2
+#define EVENT_TYPE_DEV_TAB_ERR 0x3
+#define EVENT_TYPE_PAGE_TAB_ERR 0x4
+#define EVENT_TYPE_ILL_CMD 0x5
+#define EVENT_TYPE_CMD_HARD_ERR 0x6
+#define EVENT_TYPE_IOTLB_INV_TO 0x7
+#define EVENT_TYPE_INV_DEV_REQ 0x8
+#define EVENT_DEVID_MASK 0xffff
+#define EVENT_DEVID_SHIFT 0
+#define EVENT_DOMID_MASK 0xffff
+#define EVENT_DOMID_SHIFT 0
+#define EVENT_FLAGS_MASK 0xfff
+#define EVENT_FLAGS_SHIFT 0x10
+
/* feature control bits */
#define CONTROL_IOMMU_EN 0x00ULL
#define CONTROL_HT_TUN_EN 0x01ULL
@@ -109,6 +130,8 @@
#define DEV_ENTRY_NMI_PASS 0xba
#define DEV_ENTRY_LINT0_PASS 0xbe
#define DEV_ENTRY_LINT1_PASS 0xbf
+#define DEV_ENTRY_MODE_MASK 0x07
+#define DEV_ENTRY_MODE_SHIFT 0x09

/* constants to configure the command buffer */
#define CMD_BUFFER_SIZE 8192
@@ -116,6 +139,10 @@
#define MMIO_CMD_SIZE_SHIFT 56
#define MMIO_CMD_SIZE_512 (0x9ULL << MMIO_CMD_SIZE_SHIFT)

+/* constants for event buffer handling */
+#define EVT_BUFFER_SIZE 8192 /* 512 entries */
+#define EVT_LEN_MASK (0x9ULL << 56)
+
#define PAGE_MODE_1_LEVEL 0x01
#define PAGE_MODE_2_LEVEL 0x02
#define PAGE_MODE_3_LEVEL 0x03
@@ -134,6 +161,7 @@
#define IOMMU_MAP_SIZE_L3 (1ULL << 39)

#define IOMMU_PTE_P (1ULL << 0)
+#define IOMMU_PTE_TV (1ULL << 1)
#define IOMMU_PTE_U (1ULL << 59)
#define IOMMU_PTE_FC (1ULL << 60)
#define IOMMU_PTE_IR (1ULL << 61)
@@ -159,6 +187,9 @@

#define MAX_DOMAIN_ID 65536

+/* FIXME: move this macro to <linux/pci.h> */
+#define PCI_BUS(x) (((x) >> 8) & 0xff)
+
/*
* This structure contains generic data for IOMMU protection domains
* independent of their use.
@@ -196,6 +227,15 @@ struct dma_ops_domain {
* just calculate its address in constant time.
*/
u64 **pte_pages;
+
+ /* This will be set to true when TLB needs to be flushed */
+ bool need_flush;
+
+ /*
+ * if this is a preallocated domain, keep the device for which it was
+ * preallocated in this variable
+ */
+ u16 target_dev;
};

/*
@@ -208,8 +248,9 @@ struct amd_iommu {
/* locks the accesses to the hardware */
spinlock_t lock;

- /* device id of this IOMMU */
- u16 devid;
+ /* Pointer to PCI device of this IOMMU */
+ struct pci_dev *dev;
+
/*
* Capability pointer. There could be more than one IOMMU per PCI
* device function if there are more than one AMD IOMMU capability
@@ -225,6 +266,9 @@ struct amd_iommu {
/* capabilities of that IOMMU read from ACPI */
u32 cap;

+ /* pci domain of this IOMMU */
+ u16 pci_seg;
+
/* first device this IOMMU handles. read from PCI */
u16 first_device;
/* last device this IOMMU handles. read from PCI */
@@ -240,9 +284,19 @@ struct amd_iommu {
/* size of command buffer */
u32 cmd_buf_size;

+ /* event buffer virtual address */
+ u8 *evt_buf;
+ /* size of event buffer */
+ u32 evt_buf_size;
+ /* MSI number for event interrupt */
+ u16 evt_msi_num;
+
/* if one, we need to send a completion wait command */
int need_sync;

+ /* true if interrupts for this IOMMU are already enabled */
+ bool int_enabled;
+
/* default dma_ops domain for that IOMMU */
struct dma_ops_domain *default_dom;
};
@@ -322,6 +376,12 @@ extern unsigned long *amd_iommu_pd_alloc_bitmap;
/* will be 1 if device isolation is enabled */
extern int amd_iommu_isolate;

+/*
+ * If true, the addresses will be flushed on unmap time, not when
+ * they are reused
+ */
+extern bool amd_iommu_unmap_flush;
+
/* takes a PCI device id and prints it out in a readable form */
static inline void print_devid(u16 devid, int nl)
{
diff --git a/include/asm-x86/bitops.h b/include/asm-x86/bitops.h
index 61989b9..451a747 100644
--- a/include/asm-x86/bitops.h
+++ b/include/asm-x86/bitops.h
@@ -424,16 +424,6 @@ static inline int fls(int x)

#undef ADDR

-static inline void set_bit_string(unsigned long *bitmap,
- unsigned long i, int len)
-{
- unsigned long end = i + len;
- while (i < end) {
- __set_bit(i, bitmap);
- i++;
- }
-}
-
#ifdef __KERNEL__

#include <asm-generic/bitops/sched.h>
diff --git a/include/asm-x86/dma-mapping.h b/include/asm-x86/dma-mapping.h
index 5d200e7..219c33d 100644
--- a/include/asm-x86/dma-mapping.h
+++ b/include/asm-x86/dma-mapping.h
@@ -9,12 +9,12 @@
#include <linux/scatterlist.h>
#include <asm/io.h>
#include <asm/swiotlb.h>
+#include <asm-generic/dma-coherent.h>

extern dma_addr_t bad_dma_address;
extern int iommu_merge;
-extern struct device fallback_dev;
+extern struct device x86_dma_fallback_dev;
extern int panic_on_overflow;
-extern int force_iommu;

struct dma_mapping_ops {
int (*mapping_error)(struct device *dev,
@@ -25,9 +25,6 @@ struct dma_mapping_ops {
void *vaddr, dma_addr_t dma_handle);
dma_addr_t (*map_single)(struct device *hwdev, phys_addr_t ptr,
size_t size, int direction);
- /* like map_single, but doesn't check the device mask */
- dma_addr_t (*map_simple)(struct device *hwdev, phys_addr_t ptr,
- size_t size, int direction);
void (*unmap_single)(struct device *dev, dma_addr_t addr,
size_t size, int direction);
void (*sync_single_for_cpu)(struct device *hwdev,
@@ -68,7 +65,7 @@ static inline struct dma_mapping_ops *get_dma_ops(struct device *dev)
return dma_ops;
else
return dev->archdata.dma_ops;
-#endif
+#endif /* ASM_X86__DMA_MAPPING_H */
}

/* Make sure we keep the same behaviour */
@@ -87,17 +84,14 @@ static inline int dma_mapping_error(struct device *dev, dma_addr_t dma_addr)

#define dma_alloc_noncoherent(d, s, h, f) dma_alloc_coherent(d, s, h, f)
#define dma_free_noncoherent(d, s, v, h) dma_free_coherent(d, s, v, h)
-
-void *dma_alloc_coherent(struct device *dev, size_t size,
- dma_addr_t *dma_handle, gfp_t flag);
-
-void dma_free_coherent(struct device *dev, size_t size,
- void *vaddr, dma_addr_t dma_handle);
-
+#define dma_is_consistent(d, h) (1)

extern int dma_supported(struct device *hwdev, u64 mask);
extern int dma_set_mask(struct device *dev, u64 mask);

+extern void *dma_generic_alloc_coherent(struct device *dev, size_t size,
+ dma_addr_t *dma_addr, gfp_t flag);
+
static inline dma_addr_t
dma_map_single(struct device *hwdev, void *ptr, size_t size,
int direction)
@@ -247,7 +241,68 @@ static inline int dma_get_cache_alignment(void)
return boot_cpu_data.x86_clflush_size;
}

-#define dma_is_consistent(d, h) (1)
+static inline unsigned long dma_alloc_coherent_mask(struct device *dev,
+ gfp_t gfp)
+{
+ unsigned long dma_mask = 0;

-#include <asm-generic/dma-coherent.h>
-#endif /* ASM_X86__DMA_MAPPING_H */
+ dma_mask = dev->coherent_dma_mask;
+ if (!dma_mask)
+ dma_mask = (gfp & GFP_DMA) ? DMA_24BIT_MASK : DMA_32BIT_MASK;
+
+ return dma_mask;
+}
+
+static inline gfp_t dma_alloc_coherent_gfp_flags(struct device *dev, gfp_t gfp)
+{
+#ifdef CONFIG_X86_64
+ unsigned long dma_mask = dma_alloc_coherent_mask(dev, gfp);
+
+ if (dma_mask <= DMA_32BIT_MASK && !(gfp & GFP_DMA))
+ gfp |= GFP_DMA32;
+#endif
+ return gfp;
+}
+
+static inline void *
+dma_alloc_coherent(struct device *dev, size_t size, dma_addr_t *dma_handle,
+ gfp_t gfp)
+{
+ struct dma_mapping_ops *ops = get_dma_ops(dev);
+ void *memory;
+
+ gfp &= ~(__GFP_DMA | __GFP_HIGHMEM | __GFP_DMA32);
+
+ if (dma_alloc_from_coherent(dev, size, dma_handle, &memory))
+ return memory;
+
+ if (!dev) {
+ dev = &x86_dma_fallback_dev;
+ gfp |= GFP_DMA;
+ }
+
+ if (!is_device_dma_capable(dev))
+ return NULL;
+
+ if (!ops->alloc_coherent)
+ return NULL;
+
+ return ops->alloc_coherent(dev, size, dma_handle,
+ dma_alloc_coherent_gfp_flags(dev, gfp));
+}
+
+static inline void dma_free_coherent(struct device *dev, size_t size,
+ void *vaddr, dma_addr_t bus)
+{
+ struct dma_mapping_ops *ops = get_dma_ops(dev);
+
+ WARN_ON(irqs_disabled()); /* for portability */
+
+ if (dma_release_from_coherent(dev, get_order(size), vaddr))
+ return;
+
+ if (ops->free_coherent)
+ ops->free_coherent(dev, size, vaddr, bus);
+}
+
+#endif
diff --git a/include/asm-x86/gart.h b/include/asm-x86/gart.h
index baa54fa..605edb3 100644
--- a/include/asm-x86/gart.h
+++ b/include/asm-x86/gart.h
@@ -29,6 +29,8 @@ extern int fix_aperture;
#define AMD64_GARTCACHECTL 0x9c
#define AMD64_GARTEN (1<<0)

+extern int agp_amd64_init(void);
+
static inline void enable_gart_translation(struct pci_dev *dev, u64 addr)
{
u32 tmp, ctl;
diff --git a/include/asm-x86/iommu.h b/include/asm-x86/iommu.h
index e86f441..546ad31 100644
--- a/include/asm-x86/iommu.h
+++ b/include/asm-x86/iommu.h
@@ -6,6 +6,7 @@ extern void no_iommu_init(void);
extern struct dma_mapping_ops nommu_dma_ops;
extern int force_iommu, no_iommu;
extern int iommu_detected;
+extern int dmar_disabled;

extern unsigned long iommu_num_pages(unsigned long addr, unsigned long len);

diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h
index 952e0f8..ba9114e 100644
--- a/include/linux/dma-mapping.h
+++ b/include/linux/dma-mapping.h
@@ -48,6 +48,11 @@ static inline int is_device_dma_capable(struct device *dev)
return dev->dma_mask != NULL && *dev->dma_mask != DMA_MASK_NONE;
}

+static inline int is_buffer_dma_capable(u64 mask, dma_addr_t addr, size_t size)
+{
+ return addr + size <= mask;
+}
+
#ifdef CONFIG_HAS_DMA
#include <asm/dma-mapping.h>
#else
@@ -58,6 +63,13 @@ static inline int is_device_dma_capable(struct device *dev)
#define dma_sync_single dma_sync_single_for_cpu
#define dma_sync_sg dma_sync_sg_for_cpu

+static inline u64 dma_get_mask(struct device *dev)
+{
+ if (dev && dev->dma_mask && *dev->dma_mask)
+ return *dev->dma_mask;
+ return DMA_32BIT_MASK;
+}
+
extern u64 dma_get_required_mask(struct device *dev);

static inline unsigned int dma_get_max_seg_size(struct device *dev)
diff --git a/include/linux/iommu-helper.h b/include/linux/iommu-helper.h
index c975caf..a6d0586 100644
--- a/include/linux/iommu-helper.h
+++ b/include/linux/iommu-helper.h
@@ -1,6 +1,20 @@
+#ifndef _LINUX_IOMMU_HELPER_H
+#define _LINUX_IOMMU_HELPER_H
+
+static inline unsigned long iommu_device_max_index(unsigned long size,
+ unsigned long offset,
+ u64 dma_mask)
+{
+ if (size + offset > dma_mask)
+ return dma_mask - offset + 1;
+ else
+ return size;
+}
+
extern int iommu_is_span_boundary(unsigned int index, unsigned int nr,
unsigned long shift,
unsigned long boundary_size);
+extern void iommu_area_reserve(unsigned long *map, unsigned long i, int len);
extern unsigned long iommu_area_alloc(unsigned long *map, unsigned long size,
unsigned long start, unsigned int nr,
unsigned long shift,
@@ -8,3 +22,5 @@ extern unsigned long iommu_area_alloc(unsigned long *map, unsigned long size,
unsigned long align_mask);
extern void iommu_area_free(unsigned long *map, unsigned long start,
unsigned int nr);
+
+#endif
diff --git a/include/linux/pci_ids.h b/include/linux/pci_ids.h
index f1624b3..c114103 100644
--- a/include/linux/pci_ids.h
+++ b/include/linux/pci_ids.h
@@ -497,6 +497,16 @@
#define PCI_DEVICE_ID_AMD_K8_NB_ADDRMAP 0x1101
#define PCI_DEVICE_ID_AMD_K8_NB_MEMCTL 0x1102
#define PCI_DEVICE_ID_AMD_K8_NB_MISC 0x1103
+#define PCI_DEVICE_ID_AMD_10H_NB_HT 0x1200
+#define PCI_DEVICE_ID_AMD_10H_NB_MAP 0x1201
+#define PCI_DEVICE_ID_AMD_10H_NB_DRAM 0x1202
+#define PCI_DEVICE_ID_AMD_10H_NB_MISC 0x1203
+#define PCI_DEVICE_ID_AMD_10H_NB_LINK 0x1204
+#define PCI_DEVICE_ID_AMD_11H_NB_HT 0x1300
+#define PCI_DEVICE_ID_AMD_11H_NB_MAP 0x1301
+#define PCI_DEVICE_ID_AMD_11H_NB_DRAM 0x1302
+#define PCI_DEVICE_ID_AMD_11H_NB_MISC 0x1303
+#define PCI_DEVICE_ID_AMD_11H_NB_LINK 0x1304
#define PCI_DEVICE_ID_AMD_LANCE 0x2000
#define PCI_DEVICE_ID_AMD_LANCE_HOME 0x2001
#define PCI_DEVICE_ID_AMD_SCSI 0x2020
diff --git a/kernel/dma-coherent.c b/kernel/dma-coherent.c
index c1d4d5b..f013a0c 100644
--- a/kernel/dma-coherent.c
+++ b/kernel/dma-coherent.c
@@ -124,6 +124,7 @@ int dma_alloc_from_coherent(struct device *dev, ssize_t size,
}
return (mem != NULL);
}
+EXPORT_SYMBOL(dma_alloc_from_coherent);

/**
* dma_release_from_coherent() - try to free the memory allocated from per-device coherent memory pool
@@ -151,3 +152,4 @@ int dma_release_from_coherent(struct device *dev, int order, void *vaddr)
}
return 0;
}
+EXPORT_SYMBOL(dma_release_from_coherent);
diff --git a/lib/iommu-helper.c b/lib/iommu-helper.c
index a3b8d4c..5d90074 100644
--- a/lib/iommu-helper.c
+++ b/lib/iommu-helper.c
@@ -30,8 +30,7 @@ again:
return index;
}

-static inline void set_bit_area(unsigned long *map, unsigned long i,
- int len)
+void iommu_area_reserve(unsigned long *map, unsigned long i, int len)
{
unsigned long end = i + len;
while (i < end) {
@@ -64,7 +63,7 @@ again:
start = index + 1;
goto again;
}
- set_bit_area(map, index, nr);
+ iommu_area_reserve(map, index, nr);
}
return index;
}
diff --git a/lib/swiotlb.c b/lib/swiotlb.c
index 8826fdf..f8eebd4 100644
--- a/lib/swiotlb.c
+++ b/lib/swiotlb.c
@@ -274,13 +274,14 @@ cleanup1:
}

static int
-address_needs_mapping(struct device *hwdev, dma_addr_t addr)
+address_needs_mapping(struct device *hwdev, dma_addr_t addr, size_t size)
{
- dma_addr_t mask = 0xffffffff;
- /* If the device has a mask, use it, otherwise default to 32 bits */
- if (hwdev && hwdev->dma_mask)
- mask = *hwdev->dma_mask;
- return (addr & ~mask) != 0;
+ return !is_buffer_dma_capable(dma_get_mask(hwdev), addr, size);
+}
+
+static int is_swiotlb_buffer(char *addr)
+{
+ return addr >= io_tlb_start && addr < io_tlb_end;
}

/*
@@ -467,15 +468,8 @@ swiotlb_alloc_coherent(struct device *hwdev, size_t size,
void *ret;
int order = get_order(size);

- /*
- * XXX fix me: the DMA API should pass us an explicit DMA mask
- * instead, or use ZONE_DMA32 (ia64 overloads ZONE_DMA to be a ~32
- * bit range instead of a 16MB one).
- */
- flags |= GFP_DMA;
-
ret = (void *)__get_free_pages(flags, order);
- if (ret && address_needs_mapping(hwdev, virt_to_bus(ret))) {
+ if (ret && address_needs_mapping(hwdev, virt_to_bus(ret), size)) {
/*
* The allocated memory isn't reachable by the device.
* Fall back on swiotlb_map_single().
@@ -490,19 +484,16 @@ swiotlb_alloc_coherent(struct device *hwdev, size_t size,
* swiotlb_map_single(), which will grab memory from
* the lowest available address range.
*/
- dma_addr_t handle;
- handle = swiotlb_map_single(hwdev, NULL, size, DMA_FROM_DEVICE);
- if (swiotlb_dma_mapping_error(hwdev, handle))
+ ret = map_single(hwdev, NULL, size, DMA_FROM_DEVICE);
+ if (!ret)
return NULL;
-
- ret = bus_to_virt(handle);
}

memset(ret, 0, size);
dev_addr = virt_to_bus(ret);

/* Confirm address can be DMA'd by device */
- if (address_needs_mapping(hwdev, dev_addr)) {
+ if (address_needs_mapping(hwdev, dev_addr, size)) {
printk("hwdev DMA mask = 0x%016Lx, dev_addr = 0x%016Lx\n",
(unsigned long long)*hwdev->dma_mask,
(unsigned long long)dev_addr);
@@ -518,12 +509,11 @@ swiotlb_free_coherent(struct device *hwdev, size_t size, void *vaddr,
dma_addr_t dma_handle)
{
WARN_ON(irqs_disabled());
- if (!(vaddr >= (void *)io_tlb_start
- && vaddr < (void *)io_tlb_end))
+ if (!is_swiotlb_buffer(vaddr))
free_pages((unsigned long) vaddr, get_order(size));
else
/* DMA_TO_DEVICE to avoid memcpy in unmap_single */
- swiotlb_unmap_single (hwdev, dma_handle, size, DMA_TO_DEVICE);
+ unmap_single(hwdev, vaddr, size, DMA_TO_DEVICE);
}

static void
@@ -567,7 +557,7 @@ swiotlb_map_single_attrs(struct device *hwdev, void *ptr, size_t size,
* we can safely return the device addr and not worry about bounce
* buffering it.
*/
- if (!address_needs_mapping(hwdev, dev_addr) && !swiotlb_force)
+ if (!address_needs_mapping(hwdev, dev_addr, size) && !swiotlb_force)
return dev_addr;

/*
@@ -584,7 +574,7 @@ swiotlb_map_single_attrs(struct device *hwdev, void *ptr, size_t size,
/*
* Ensure that the address returned is DMA'ble
*/
- if (address_needs_mapping(hwdev, dev_addr))
+ if (address_needs_mapping(hwdev, dev_addr, size))
panic("map_single: bounce buffer is not DMA'ble");

return dev_addr;
@@ -612,7 +602,7 @@ swiotlb_unmap_single_attrs(struct device *hwdev, dma_addr_t dev_addr,
char *dma_addr = bus_to_virt(dev_addr);

BUG_ON(dir == DMA_NONE);
- if (dma_addr >= io_tlb_start && dma_addr < io_tlb_end)
+ if (is_swiotlb_buffer(dma_addr))
unmap_single(hwdev, dma_addr, size, dir);
else if (dir == DMA_FROM_DEVICE)
dma_mark_clean(dma_addr, size);
@@ -642,7 +632,7 @@ swiotlb_sync_single(struct device *hwdev, dma_addr_t dev_addr,
char *dma_addr = bus_to_virt(dev_addr);

BUG_ON(dir == DMA_NONE);
- if (dma_addr >= io_tlb_start && dma_addr < io_tlb_end)
+ if (is_swiotlb_buffer(dma_addr))
sync_single(hwdev, dma_addr, size, dir, target);
else if (dir == DMA_FROM_DEVICE)
dma_mark_clean(dma_addr, size);
@@ -673,7 +663,7 @@ swiotlb_sync_single_range(struct device *hwdev, dma_addr_t dev_addr,
char *dma_addr = bus_to_virt(dev_addr) + offset;

BUG_ON(dir == DMA_NONE);
- if (dma_addr >= io_tlb_start && dma_addr < io_tlb_end)
+ if (is_swiotlb_buffer(dma_addr))
sync_single(hwdev, dma_addr, size, dir, target);
else if (dir == DMA_FROM_DEVICE)
dma_mark_clean(dma_addr, size);
@@ -727,7 +717,8 @@ swiotlb_map_sg_attrs(struct device *hwdev, struct scatterlist *sgl, int nelems,
for_each_sg(sgl, sg, nelems, i) {
addr = SG_ENT_VIRT_ADDRESS(sg);
dev_addr = virt_to_bus(addr);
- if (swiotlb_force || address_needs_mapping(hwdev, dev_addr)) {
+ if (swiotlb_force ||
+ address_needs_mapping(hwdev, dev_addr, sg->length)) {
void *map = map_single(hwdev, addr, sg->length, dir);
if (!map) {
/* Don't panic here, we expect map_sg users
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/