From 6c505ce3930c6a6b455cda53fab3e88ae44f8221 Mon Sep 17 00:00:00 2001 From: Joerg Roedel Date: Tue, 19 Aug 2008 16:32:45 +0200 Subject: x86: move dma_*_coherent functions to include file All the x86 DMA-API functions are defined in asm/dma-mapping.h. This patch moves the dma_*_coherent functions also to this header file because they are now small enough to do so. This is done as a separate patch because it also includes some renaming and restructuring of the dma-mapping.h file. Signed-off-by: Joerg Roedel Signed-off-by: Ingo Molnar --- include/asm-x86/dma-mapping.h | 47 ++++++++++++++++++++++++++++++++++--------- 1 file changed, 37 insertions(+), 10 deletions(-) (limited to 'include') diff --git a/include/asm-x86/dma-mapping.h b/include/asm-x86/dma-mapping.h index ad9cd6d49bfc..8e16095d1fa9 100644 --- a/include/asm-x86/dma-mapping.h +++ b/include/asm-x86/dma-mapping.h @@ -9,10 +9,11 @@ #include #include #include +#include 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; @@ -87,13 +88,7 @@ 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); @@ -247,7 +242,39 @@ static inline int dma_get_cache_alignment(void) return boot_cpu_data.x86_clflush_size; } -#define dma_is_consistent(d, h) (1) +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; + + if (dma_alloc_from_coherent(dev, size, dma_handle, &memory)) + return memory; + + if (!dev) { + dev = &x86_dma_fallback_dev; + gfp |= GFP_DMA; + } + + if (ops->alloc_coherent) + return ops->alloc_coherent(dev, size, + dma_handle, gfp); + return NULL; +} + +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); +} -#include #endif -- cgit v1.2.3 From 766af9fa812f49feb4a3e62cf92f3d37f33c7fb6 Mon Sep 17 00:00:00 2001 From: FUJITA Tomonori Date: Fri, 22 Aug 2008 00:12:09 +0900 Subject: dma-mapping.h, x86: remove last user of dma_mapping_ops->map_simple pci-dma.c doesn't use map_simple hook any more so we can remove it from struct dma_mapping_ops now. Signed-off-by: FUJITA Tomonori Signed-off-by: Ingo Molnar --- arch/x86/kernel/pci-gart_64.c | 1 - include/asm-x86/dma-mapping.h | 3 --- 2 files changed, 4 deletions(-) (limited to 'include') diff --git a/arch/x86/kernel/pci-gart_64.c b/arch/x86/kernel/pci-gart_64.c index e81df25dc067..cfd7ec25e37e 100644 --- a/arch/x86/kernel/pci-gart_64.c +++ b/arch/x86/kernel/pci-gart_64.c @@ -720,7 +720,6 @@ 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, diff --git a/include/asm-x86/dma-mapping.h b/include/asm-x86/dma-mapping.h index 8e16095d1fa9..3a9a6f5e681a 100644 --- a/include/asm-x86/dma-mapping.h +++ b/include/asm-x86/dma-mapping.h @@ -26,9 +26,6 @@ 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, -- cgit v1.2.3 From ca1af29a733629b9158a4a32a927d16ff9009a95 Mon Sep 17 00:00:00 2001 From: Joerg Roedel Date: Tue, 2 Sep 2008 13:13:39 +0200 Subject: x86, pci: add northbridge pci ids for fam 0x11 processors The PCI device ids for AMD family 0x11 processors are missing in pci_ids.h. This patch adds them. Signed-off-by: Joerg Roedel Signed-off-by: Ingo Molnar --- include/linux/pci_ids.h | 5 +++++ 1 file changed, 5 insertions(+) (limited to 'include') diff --git a/include/linux/pci_ids.h b/include/linux/pci_ids.h index 9ec2bcce8e83..0d0b314476c9 100644 --- a/include/linux/pci_ids.h +++ b/include/linux/pci_ids.h @@ -497,6 +497,11 @@ #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_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 -- cgit v1.2.3 From e51af6630848406fc97adbd71443818cdcda297b Mon Sep 17 00:00:00 2001 From: David Woodhouse Date: Thu, 4 Sep 2008 09:54:37 +0100 Subject: x86: blacklist DMAR on Intel G31/G33 chipsets Some BIOSes (the Intel DG33BU, for example) wrongly claim to have DMAR when they don't. Avoid the resulting crashes when it doesn't work as expected. I'd still be grateful if someone could test it on a DG33BU with the old BIOS though, since I've killed mine. I tested the DMI version, but not this one. Signed-off-by: David Woodhouse Signed-off-by: Ingo Molnar --- arch/x86/kernel/early-quirks.c | 18 ++++++++++++++++++ drivers/pci/intel-iommu.c | 2 +- include/asm-x86/iommu.h | 1 + 3 files changed, 20 insertions(+), 1 deletion(-) (limited to 'include') diff --git a/arch/x86/kernel/early-quirks.c b/arch/x86/kernel/early-quirks.c index 4353cf5e6fac..24bb5faf5efa 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/drivers/pci/intel-iommu.c b/drivers/pci/intel-iommu.c index 8d0e60ac849c..eaba6ecc2adb 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/iommu.h b/include/asm-x86/iommu.h index 5f888cc5be49..621a1af94c4c 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); -- cgit v1.2.3 From de9f521fb72dd091aa4989fe2e004ecf4785a850 Mon Sep 17 00:00:00 2001 From: FUJITA Tomonori Date: Mon, 8 Sep 2008 18:10:11 +0900 Subject: x86: move pci-nommu's dma_mask check to common code The check to see if dev->dma_mask is NULL in pci-nommu is more appropriate for dma_alloc_coherent(). Signed-off-by: FUJITA Tomonori Acked-by: Joerg Roedel Signed-off-by: Ingo Molnar --- arch/x86/kernel/pci-nommu.c | 3 --- include/asm-x86/dma-mapping.h | 3 +++ 2 files changed, 3 insertions(+), 3 deletions(-) (limited to 'include') diff --git a/arch/x86/kernel/pci-nommu.c b/arch/x86/kernel/pci-nommu.c index 73853d3fdcac..0f51883cc6a8 100644 --- a/arch/x86/kernel/pci-nommu.c +++ b/arch/x86/kernel/pci-nommu.c @@ -80,9 +80,6 @@ nommu_alloc_coherent(struct device *hwdev, size_t size, int node; struct page *page; - if (hwdev->dma_mask == NULL) - return NULL; - gfp &= ~(__GFP_DMA | __GFP_HIGHMEM | __GFP_DMA32); gfp |= __GFP_ZERO; diff --git a/include/asm-x86/dma-mapping.h b/include/asm-x86/dma-mapping.h index 3a9a6f5e681a..088c56814aa3 100644 --- a/include/asm-x86/dma-mapping.h +++ b/include/asm-x86/dma-mapping.h @@ -254,6 +254,9 @@ dma_alloc_coherent(struct device *dev, size_t size, dma_addr_t *dma_handle, gfp |= GFP_DMA; } + if (!dev->dma_mask) + return NULL; + if (ops->alloc_coherent) return ops->alloc_coherent(dev, size, dma_handle, gfp); -- cgit v1.2.3 From 8a53ad675f86ee003482b557da944e070d3c4859 Mon Sep 17 00:00:00 2001 From: FUJITA Tomonori Date: Mon, 8 Sep 2008 18:10:12 +0900 Subject: x86: fix nommu_alloc_coherent allocation with NULL device argument We need to use __GFP_DMA for NULL device argument (fallback_dev) with pci-nommu. It's a hack for ISA (and some old code) so we need to use GFP_DMA. Signed-off-by: FUJITA Tomonori Acked-by: Joerg Roedel Signed-off-by: Ingo Molnar --- arch/x86/kernel/pci-nommu.c | 3 +-- include/asm-x86/dma-mapping.h | 2 ++ 2 files changed, 3 insertions(+), 2 deletions(-) (limited to 'include') diff --git a/arch/x86/kernel/pci-nommu.c b/arch/x86/kernel/pci-nommu.c index 0f51883cc6a8..ada1c87cafc2 100644 --- a/arch/x86/kernel/pci-nommu.c +++ b/arch/x86/kernel/pci-nommu.c @@ -80,7 +80,6 @@ nommu_alloc_coherent(struct device *hwdev, size_t size, int node; struct page *page; - gfp &= ~(__GFP_DMA | __GFP_HIGHMEM | __GFP_DMA32); gfp |= __GFP_ZERO; dma_mask = hwdev->coherent_dma_mask; @@ -93,7 +92,7 @@ nommu_alloc_coherent(struct device *hwdev, size_t size, node = dev_to_node(hwdev); #ifdef CONFIG_X86_64 - if (dma_mask <= DMA_32BIT_MASK) + if (dma_mask <= DMA_32BIT_MASK && !(gfp & GFP_DMA)) gfp |= GFP_DMA32; #endif diff --git a/include/asm-x86/dma-mapping.h b/include/asm-x86/dma-mapping.h index 088c56814aa3..ad8b49032d18 100644 --- a/include/asm-x86/dma-mapping.h +++ b/include/asm-x86/dma-mapping.h @@ -246,6 +246,8 @@ dma_alloc_coherent(struct device *dev, size_t size, dma_addr_t *dma_handle, 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; -- cgit v1.2.3 From 823e7e8c6ef12cd1943dc42fe7595ca74e8cc3d7 Mon Sep 17 00:00:00 2001 From: FUJITA Tomonori Date: Mon, 8 Sep 2008 18:10:13 +0900 Subject: x86: dma_alloc_coherent sets gfp flags properly Non real IOMMU implemenations (which doesn't do virtual mappings, e.g. swiotlb, pci-nommu, etc) need to use proper gfp flags and dma_mask to allocate pages in their own dma_alloc_coherent() (allocated page need to be suitable for device's coherent_dma_mask). This patch makes dma_alloc_coherent do this job so that IOMMUs don't need to take care of it any more. Real IOMMU implemenataions can simply ignore the gfp flags. Signed-off-by: FUJITA Tomonori Acked-by: Joerg Roedel Signed-off-by: Ingo Molnar --- arch/x86/kernel/pci-nommu.c | 19 ++----------------- include/asm-x86/dma-mapping.h | 32 ++++++++++++++++++++++++++++---- 2 files changed, 30 insertions(+), 21 deletions(-) (limited to 'include') diff --git a/arch/x86/kernel/pci-nommu.c b/arch/x86/kernel/pci-nommu.c index ada1c87cafc2..8e398b56f50b 100644 --- a/arch/x86/kernel/pci-nommu.c +++ b/arch/x86/kernel/pci-nommu.c @@ -80,26 +80,11 @@ nommu_alloc_coherent(struct device *hwdev, size_t size, int node; struct page *page; - gfp |= __GFP_ZERO; - - dma_mask = hwdev->coherent_dma_mask; - if (!dma_mask) - dma_mask = *(hwdev->dma_mask); + dma_mask = dma_alloc_coherent_mask(hwdev, gfp); - if (dma_mask < DMA_24BIT_MASK) - return NULL; + gfp |= __GFP_ZERO; node = dev_to_node(hwdev); - -#ifdef CONFIG_X86_64 - if (dma_mask <= DMA_32BIT_MASK && !(gfp & GFP_DMA)) - gfp |= GFP_DMA32; -#endif - - /* No alloc-free penalty for ISA devices */ - if (dma_mask == DMA_24BIT_MASK) - gfp |= GFP_DMA; - again: page = alloc_pages_node(node, gfp, get_order(size)); if (!page) diff --git a/include/asm-x86/dma-mapping.h b/include/asm-x86/dma-mapping.h index ad8b49032d18..0cc022b9a4a1 100644 --- a/include/asm-x86/dma-mapping.h +++ b/include/asm-x86/dma-mapping.h @@ -239,6 +239,29 @@ static inline int dma_get_cache_alignment(void) return boot_cpu_data.x86_clflush_size; } +static inline unsigned long dma_alloc_coherent_mask(struct device *dev, + gfp_t gfp) +{ + unsigned long dma_mask = 0; + + 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) +{ + unsigned long dma_mask = dma_alloc_coherent_mask(dev, gfp); + +#ifdef CONFIG_X86_64 + 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) @@ -259,10 +282,11 @@ dma_alloc_coherent(struct device *dev, size_t size, dma_addr_t *dma_handle, if (!dev->dma_mask) return NULL; - if (ops->alloc_coherent) - return ops->alloc_coherent(dev, size, - dma_handle, gfp); - 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, -- cgit v1.2.3 From 9fcaff0e660d886e9a766460adbe558dd25de31b Mon Sep 17 00:00:00 2001 From: Steven Noonan Date: Mon, 8 Sep 2008 16:19:11 -0700 Subject: x86: unused variable in dma_alloc_coherent_gfp_flags() Fixed a warning caused by a badly placed ifdef. Signed-off-by: Steven Noonan Signed-off-by: Ingo Molnar --- include/asm-x86/dma-mapping.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'include') diff --git a/include/asm-x86/dma-mapping.h b/include/asm-x86/dma-mapping.h index 0cc022b9a4a1..56075320b813 100644 --- a/include/asm-x86/dma-mapping.h +++ b/include/asm-x86/dma-mapping.h @@ -253,9 +253,9 @@ static inline unsigned long dma_alloc_coherent_mask(struct device *dev, 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); -#ifdef CONFIG_X86_64 if (dma_mask <= DMA_32BIT_MASK && !(gfp & GFP_DMA)) gfp |= GFP_DMA32; #endif -- cgit v1.2.3 From 636dc67cbf8c481a996faf6c23f0532d0f02ebad Mon Sep 17 00:00:00 2001 From: FUJITA Tomonori Date: Wed, 10 Sep 2008 01:06:46 +0900 Subject: add is_buffer_dma_capable helper function is_buffer_dma_capable helper function is to see if a memory region is DMA-capable or not. The arugments are the dma_mask (or coherent_dma_mask) of a device and the address and size of a memory region. Signed-off-by: FUJITA Tomonori Acked-by: Joerg Roedel Signed-off-by: Ingo Molnar --- include/linux/dma-mapping.h | 5 +++++ 1 file changed, 5 insertions(+) (limited to 'include') diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h index 952e0f857ac9..6ed50c1642f1 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 #else -- cgit v1.2.3 From 8a493d37f049b631e19584d1cb84cec88cbcf8fc Mon Sep 17 00:00:00 2001 From: FUJITA Tomonori Date: Wed, 10 Sep 2008 00:49:47 +0900 Subject: x86: remove duplicated extern force_iommu Both iommu.h and dma-mapping.h have extern force_iommu. The latter doesn't need to do. Signed-off-by: FUJITA Tomonori Signed-off-by: Ingo Molnar --- include/asm-x86/dma-mapping.h | 1 - 1 file changed, 1 deletion(-) (limited to 'include') diff --git a/include/asm-x86/dma-mapping.h b/include/asm-x86/dma-mapping.h index 56075320b813..f43420b8c573 100644 --- a/include/asm-x86/dma-mapping.h +++ b/include/asm-x86/dma-mapping.h @@ -15,7 +15,6 @@ extern dma_addr_t bad_dma_address; extern int iommu_merge; 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, -- cgit v1.2.3 From 982162602b31041b426edec6480d327743abcbcc Mon Sep 17 00:00:00 2001 From: FUJITA Tomonori Date: Wed, 10 Sep 2008 00:49:48 +0900 Subject: x86: convert dma_alloc_coherent to use is_device_dma_capable Signed-off-by: FUJITA Tomonori Signed-off-by: Ingo Molnar --- include/asm-x86/dma-mapping.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'include') diff --git a/include/asm-x86/dma-mapping.h b/include/asm-x86/dma-mapping.h index f43420b8c573..f408e6dd1778 100644 --- a/include/asm-x86/dma-mapping.h +++ b/include/asm-x86/dma-mapping.h @@ -278,7 +278,7 @@ dma_alloc_coherent(struct device *dev, size_t size, dma_addr_t *dma_handle, gfp |= GFP_DMA; } - if (!dev->dma_mask) + if (!is_device_dma_capable(dev)) return NULL; if (!ops->alloc_coherent) -- cgit v1.2.3 From eecfffc154ffbfe70686a9905c090b488778c28e Mon Sep 17 00:00:00 2001 From: FUJITA Tomonori Date: Fri, 12 Sep 2008 19:42:33 +0900 Subject: iommu: add iommu_device_max_index IOMMU helper function This function helps IOMMUs to know the highest address that a device can access to. Signed-off-by: FUJITA Tomonori Signed-off-by: Ingo Molnar --- include/linux/iommu-helper.h | 10 ++++++++++ 1 file changed, 10 insertions(+) (limited to 'include') diff --git a/include/linux/iommu-helper.h b/include/linux/iommu-helper.h index c975caf75385..58f41107e4ae 100644 --- a/include/linux/iommu-helper.h +++ b/include/linux/iommu-helper.h @@ -1,3 +1,13 @@ +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); -- cgit v1.2.3 From 589fc9a6e2102b498978f6350581ec7fa5aeb032 Mon Sep 17 00:00:00 2001 From: FUJITA Tomonori Date: Fri, 12 Sep 2008 19:42:34 +0900 Subject: iommu: add dma_get_mask helper function Several IOMMUs do the same thing to get the dma_mask of a device. This adds a helper function to do the same thing to sweep them. Signed-off-by: FUJITA Tomonori Signed-off-by: Ingo Molnar --- include/linux/dma-mapping.h | 7 +++++++ 1 file changed, 7 insertions(+) (limited to 'include') diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h index 6ed50c1642f1..0dba7433af18 100644 --- a/include/linux/dma-mapping.h +++ b/include/linux/dma-mapping.h @@ -63,6 +63,13 @@ static inline int is_buffer_dma_capable(u64 mask, dma_addr_t addr, size_t size) #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->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) -- cgit v1.2.3 From 07a2c01a0c2a0cb4581a67d50d4f17cb4d2457c4 Mon Sep 17 00:00:00 2001 From: FUJITA Tomonori Date: Fri, 19 Sep 2008 02:02:05 +0900 Subject: convert swiotlb to use dma_get_mask swiotlb can use dma_get_mask() instead of the homegrown function. Signed-off-by: FUJITA Tomonori Cc: tony.luck@intel.com Signed-off-by: Ingo Molnar --- include/linux/dma-mapping.h | 2 +- lib/swiotlb.c | 6 +----- 2 files changed, 2 insertions(+), 6 deletions(-) (limited to 'include') diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h index 0dba7433af18..ba9114ec5d3a 100644 --- a/include/linux/dma-mapping.h +++ b/include/linux/dma-mapping.h @@ -65,7 +65,7 @@ static inline int is_buffer_dma_capable(u64 mask, dma_addr_t addr, size_t size) static inline u64 dma_get_mask(struct device *dev) { - if (dev->dma_mask && *dev->dma_mask) + if (dev && dev->dma_mask && *dev->dma_mask) return *dev->dma_mask; return DMA_32BIT_MASK; } diff --git a/lib/swiotlb.c b/lib/swiotlb.c index 240a67c2c979..f8eebd489149 100644 --- a/lib/swiotlb.c +++ b/lib/swiotlb.c @@ -276,11 +276,7 @@ cleanup1: static int 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 !is_buffer_dma_capable(mask, addr, size); + return !is_buffer_dma_capable(dma_get_mask(hwdev), addr, size); } static int is_swiotlb_buffer(char *addr) -- cgit v1.2.3 From 2842e5bf3115193f05dc9dac20f940e7abf44c1a Mon Sep 17 00:00:00 2001 From: Joerg Roedel Date: Thu, 18 Sep 2008 15:23:43 +0200 Subject: x86: move GART TLB flushing options to generic code The GART currently implements the iommu=[no]fullflush command line parameters which influence its IO/TLB flushing strategy. This patch makes these parameters generic so that they can be used by the AMD IOMMU too. Signed-off-by: Joerg Roedel Signed-off-by: Ingo Molnar --- Documentation/kernel-parameters.txt | 4 ++++ Documentation/x86/x86_64/boot-options.txt | 2 -- arch/x86/kernel/pci-dma.c | 13 +++++++++++++ arch/x86/kernel/pci-gart_64.c | 13 ------------- include/asm-x86/iommu.h | 1 + 5 files changed, 18 insertions(+), 15 deletions(-) (limited to 'include') diff --git a/Documentation/kernel-parameters.txt b/Documentation/kernel-parameters.txt index 1150444a21ab..40066ceb48fe 100644 --- a/Documentation/kernel-parameters.txt +++ b/Documentation/kernel-parameters.txt @@ -893,6 +893,10 @@ and is between 256 and 4096 characters. It is defined in the file nomerge forcesac soft + fullflush + Flush IO/TLB at every deallocation + nofullflush + Flush IO/TLB only when addresses are reused (default) intel_iommu= [DMAR] Intel IOMMU driver (DMAR) option diff --git a/Documentation/x86/x86_64/boot-options.txt b/Documentation/x86/x86_64/boot-options.txt index b0c7b6c4abda..c83c8e4bc8e5 100644 --- a/Documentation/x86/x86_64/boot-options.txt +++ b/Documentation/x86/x86_64/boot-options.txt @@ -233,8 +233,6 @@ IOMMU (input/output memory management unit) iommu options only relevant to the AMD GART hardware IOMMU: Set the size of the remapping area in bytes. allowed Overwrite iommu off workarounds for specific chipsets. - fullflush Flush IOMMU on each allocation (default). - nofullflush Don't use IOMMU fullflush. leak Turn on simple iommu leak tracing (only when CONFIG_IOMMU_LEAK is on). Default number of leak pages is 20. diff --git a/arch/x86/kernel/pci-dma.c b/arch/x86/kernel/pci-dma.c index 0a1408abcc62..d2f2c0158dc1 100644 --- a/arch/x86/kernel/pci-dma.c +++ b/arch/x86/kernel/pci-dma.c @@ -16,6 +16,15 @@ EXPORT_SYMBOL(dma_ops); static int iommu_sac_force __read_mostly; +/* + * 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; + #ifdef CONFIG_IOMMU_DEBUG int panic_on_overflow __read_mostly = 1; int force_iommu __read_mostly = 1; @@ -171,6 +180,10 @@ static __init int iommu_setup(char *p) } if (!strncmp(p, "nomerge", 7)) iommu_merge = 0; + if (!strncmp(p, "fullflush", 8)) + iommu_fullflush = 1; + if (!strncmp(p, "nofullflush", 11)) + iommu_fullflush = 0; if (!strncmp(p, "forcesac", 8)) iommu_sac_force = 1; if (!strncmp(p, "allowdac", 8)) diff --git a/arch/x86/kernel/pci-gart_64.c b/arch/x86/kernel/pci-gart_64.c index 9739d5682093..508ef470b27f 100644 --- a/arch/x86/kernel/pci-gart_64.c +++ b/arch/x86/kernel/pci-gart_64.c @@ -45,15 +45,6 @@ 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: */ @@ -901,10 +892,6 @@ void __init gart_parse_options(char *p) #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)) diff --git a/include/asm-x86/iommu.h b/include/asm-x86/iommu.h index 621a1af94c4c..67b2fd56c6da 100644 --- a/include/asm-x86/iommu.h +++ b/include/asm-x86/iommu.h @@ -7,6 +7,7 @@ extern struct dma_mapping_ops nommu_dma_ops; extern int force_iommu, no_iommu; extern int iommu_detected; extern int dmar_disabled; +extern int iommu_fullflush; extern unsigned long iommu_num_pages(unsigned long addr, unsigned long len); -- cgit v1.2.3 From 1c65577398589bb44ab0980f9b9d30804b48a5db Mon Sep 17 00:00:00 2001 From: Joerg Roedel Date: Thu, 4 Sep 2008 18:40:05 +0200 Subject: AMD IOMMU: implement lazy IO/TLB flushing The IO/TLB flushing on every unmaping operation is the most expensive part in AMD IOMMU code and not strictly necessary. It is sufficient to do the flush before any entries are reused. This is patch implements lazy IO/TLB flushing which does exactly this. Signed-off-by: Joerg Roedel Signed-off-by: Ingo Molnar --- arch/x86/kernel/amd_iommu.c | 26 ++++++++++++++++++++++---- arch/x86/kernel/amd_iommu_init.c | 7 ++++++- include/asm-x86/amd_iommu_types.h | 3 +++ 3 files changed, 31 insertions(+), 5 deletions(-) (limited to 'include') diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c index 691e023695ad..679f2a8e22ee 100644 --- a/arch/x86/kernel/amd_iommu.c +++ b/arch/x86/kernel/amd_iommu.c @@ -203,6 +203,14 @@ 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 @@ -386,14 +394,18 @@ static unsigned long dma_ops_alloc_addresses(struct device *dev, PAGE_SIZE) >> PAGE_SHIFT; limit = limit < size ? limit : size; - 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) + if (address == -1) { address = iommu_area_alloc(dom->bitmap, limit, 0, pages, 0, boundary_size, 0); + dom->need_flush = true; + } if (likely(address != -1)) { dom->next_bit = address + pages; @@ -553,6 +565,8 @@ 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; + /* Intialize the exclusion range if necessary */ if (iommu->exclusion_start && iommu->exclusion_start < dma_dom->aperture_size) { @@ -795,7 +809,10 @@ static dma_addr_t __map_single(struct device *dev, } address += offset; - if (unlikely(iommu_has_npcache(iommu))) + if (unlikely(dma_dom->need_flush && !iommu_fullflush)) { + 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: @@ -829,7 +846,8 @@ static void __unmap_single(struct amd_iommu *iommu, dma_ops_free_addresses(dma_dom, dma_addr, pages); - iommu_flush_pages(iommu, dma_dom->domain.id, dma_addr, size); + if (iommu_fullflush) + iommu_flush_pages(iommu, dma_dom->domain.id, dma_addr, size); } /* diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c index a69cc0f52042..f2fa8dc81beb 100644 --- a/arch/x86/kernel/amd_iommu_init.c +++ b/arch/x86/kernel/amd_iommu_init.c @@ -995,6 +995,11 @@ int __init amd_iommu_init(void) else printk("disabled\n"); + if (iommu_fullflush) + 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; @@ -1057,7 +1062,7 @@ 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; } diff --git a/include/asm-x86/amd_iommu_types.h b/include/asm-x86/amd_iommu_types.h index dcc812067394..dcc472445ffd 100644 --- a/include/asm-x86/amd_iommu_types.h +++ b/include/asm-x86/amd_iommu_types.h @@ -196,6 +196,9 @@ 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; }; /* -- cgit v1.2.3 From 335503e57b6b8de04cec5d27eb2c3d09ff98905b Mon Sep 17 00:00:00 2001 From: Joerg Roedel Date: Fri, 5 Sep 2008 14:29:07 +0200 Subject: AMD IOMMU: add event buffer allocation This patch adds the allocation of a event buffer for each AMD IOMMU in the system. The hardware will log events like device page faults or other errors to this buffer once this is enabled. Signed-off-by: Joerg Roedel Signed-off-by: Ingo Molnar --- arch/x86/kernel/amd_iommu_init.c | 29 +++++++++++++++++++++++++++++ include/asm-x86/amd_iommu_types.h | 9 +++++++++ 2 files changed, 38 insertions(+) (limited to 'include') diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c index f2fa8dc81beb..41ce8d5d626e 100644 --- a/arch/x86/kernel/amd_iommu_init.c +++ b/arch/x86/kernel/amd_iommu_init.c @@ -417,6 +417,30 @@ static void __init free_command_buffer(struct amd_iommu *iommu) free_pages((unsigned long)iommu->cmd_buf, get_order(CMD_BUFFER_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. */ static void set_dev_entry_bit(u16 devid, u8 bit) { @@ -622,6 +646,7 @@ 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); } @@ -661,6 +686,10 @@ 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; + init_iommu_from_pci(iommu); init_iommu_from_acpi(iommu, h); init_iommu_devices(iommu); diff --git a/include/asm-x86/amd_iommu_types.h b/include/asm-x86/amd_iommu_types.h index dcc472445ffd..8b8cd0c60b3c 100644 --- a/include/asm-x86/amd_iommu_types.h +++ b/include/asm-x86/amd_iommu_types.h @@ -116,6 +116,10 @@ #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 @@ -243,6 +247,11 @@ 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; + /* if one, we need to send a completion wait command */ int need_sync; -- cgit v1.2.3 From ee893c24edb8ebab9a3fb66566855572579ad616 Mon Sep 17 00:00:00 2001 From: Joerg Roedel Date: Mon, 8 Sep 2008 14:48:04 +0200 Subject: AMD IOMMU: save pci segment from ACPI tables This patch adds the pci_seg field to the amd_iommu structure and fills it with the corresponding value from the ACPI table. Signed-off-by: Joerg Roedel Signed-off-by: Ingo Molnar --- arch/x86/kernel/amd_iommu_init.c | 1 + include/asm-x86/amd_iommu_types.h | 3 +++ 2 files changed, 4 insertions(+) (limited to 'include') diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c index 41ce8d5d626e..b50234ef91ed 100644 --- a/arch/x86/kernel/amd_iommu_init.c +++ b/arch/x86/kernel/amd_iommu_init.c @@ -676,6 +676,7 @@ static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h) */ iommu->devid = h->devid; 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) diff --git a/include/asm-x86/amd_iommu_types.h b/include/asm-x86/amd_iommu_types.h index 8b8cd0c60b3c..20814b85bbc3 100644 --- a/include/asm-x86/amd_iommu_types.h +++ b/include/asm-x86/amd_iommu_types.h @@ -232,6 +232,9 @@ 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 */ -- cgit v1.2.3 From 3eaf28a1cd2686aaa185b54d5a5e18e91b41f7f2 Mon Sep 17 00:00:00 2001 From: Joerg Roedel Date: Mon, 8 Sep 2008 15:55:10 +0200 Subject: AMD IOMMU: save pci_dev instead of devid We need the pci_dev later anyways to enable MSI for the IOMMU hardware. So remove the devid pointing to the BDF and replace it with the pci_dev structure where the IOMMU is implemented. Signed-off-by: Joerg Roedel Signed-off-by: Ingo Molnar --- arch/x86/kernel/amd_iommu_init.c | 25 ++++++++++++++++--------- include/asm-x86/amd_iommu_types.h | 5 +++-- 2 files changed, 19 insertions(+), 11 deletions(-) (limited to 'include') diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c index b50234ef91ed..a7eb89d8923d 100644 --- a/arch/x86/kernel/amd_iommu_init.c +++ b/arch/x86/kernel/amd_iommu_init.c @@ -242,9 +242,12 @@ 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); } @@ -511,15 +514,14 @@ 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; - 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); - 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), @@ -674,7 +676,10 @@ 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; @@ -695,6 +700,8 @@ static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h) init_iommu_from_acpi(iommu, h); init_iommu_devices(iommu); + pci_enable_device(iommu->dev); + return 0; } diff --git a/include/asm-x86/amd_iommu_types.h b/include/asm-x86/amd_iommu_types.h index 20814b85bbc3..a5629a21557c 100644 --- a/include/asm-x86/amd_iommu_types.h +++ b/include/asm-x86/amd_iommu_types.h @@ -215,8 +215,9 @@ 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 -- cgit v1.2.3 From a80dc3e0e0dc8393158de317d66ae0f345dc58f9 Mon Sep 17 00:00:00 2001 From: Joerg Roedel Date: Thu, 11 Sep 2008 16:51:41 +0200 Subject: AMD IOMMU: add MSI interrupt support The AMD IOMMU can generate interrupts for various reasons. This patch adds the basic interrupt enabling infrastructure to the driver. Signed-off-by: Joerg Roedel Signed-off-by: Ingo Molnar --- arch/x86/Kconfig | 1 + arch/x86/kernel/amd_iommu.c | 11 +++++ arch/x86/kernel/amd_iommu_init.c | 99 ++++++++++++++++++++++++++++++++++++++- include/asm-x86/amd_iommu.h | 3 ++ include/asm-x86/amd_iommu_types.h | 7 +++ 5 files changed, 120 insertions(+), 1 deletion(-) (limited to 'include') diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig index ed92864d1325..39fd3f42696d 100644 --- a/arch/x86/Kconfig +++ b/arch/x86/Kconfig @@ -553,6 +553,7 @@ 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 15792ed082e0..0e494b9d5f20 100644 --- a/arch/x86/kernel/amd_iommu.c +++ b/arch/x86/kernel/amd_iommu.c @@ -49,6 +49,17 @@ static int iommu_has_npcache(struct amd_iommu *iommu) return iommu->cap & IOMMU_CAP_NPCACHE; } +/**************************************************************************** + * + * Interrupt handling functions + * + ****************************************************************************/ + +irqreturn_t amd_iommu_int_handler(int irq, void *data) +{ + return IRQ_NONE; +} + /**************************************************************************** * * IOMMU command queuing functions diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c index a7eb89d8923d..14a06464a694 100644 --- a/arch/x86/kernel/amd_iommu_init.c +++ b/arch/x86/kernel/amd_iommu_init.c @@ -22,6 +22,8 @@ #include #include #include +#include +#include #include #include #include @@ -515,17 +517,20 @@ 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 cap_ptr = iommu->cap_ptr; - u32 range; + u32 range, misc; 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); 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); } /* @@ -696,6 +701,8 @@ static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h) 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); @@ -741,6 +748,95 @@ static int __init init_iommu_all(struct acpi_table_header *table) return 0; } +/**************************************************************************** + * + * 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 @@ -862,6 +958,7 @@ 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(iommu); } } diff --git a/include/asm-x86/amd_iommu.h b/include/asm-x86/amd_iommu.h index 30a12049353b..2fd97cb250c7 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 + #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 a5629a21557c..8533f09b34b7 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 @@ -255,10 +257,15 @@ struct amd_iommu { 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; }; -- cgit v1.2.3 From 90008ee4b811c944455752dcb72b291a5ba81b53 Mon Sep 17 00:00:00 2001 From: Joerg Roedel Date: Tue, 9 Sep 2008 16:41:05 +0200 Subject: AMD IOMMU: add event handling code This patch adds code for polling and printing out events generated by the AMD IOMMU. Signed-off-by: Joerg Roedel Signed-off-by: Ingo Molnar --- arch/x86/kernel/amd_iommu.c | 87 ++++++++++++++++++++++++++++++++++++++- arch/x86/kernel/amd_iommu_init.c | 1 - include/asm-x86/amd_iommu_types.h | 22 ++++++++++ 3 files changed, 108 insertions(+), 2 deletions(-) (limited to 'include') diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c index 0e494b9d5f20..0cb8fd2359f5 100644 --- a/arch/x86/kernel/amd_iommu.c +++ b/arch/x86/kernel/amd_iommu.c @@ -55,9 +55,94 @@ static int iommu_has_npcache(struct amd_iommu *iommu) * ****************************************************************************/ +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) { - return IRQ_NONE; + struct amd_iommu *iommu; + + list_for_each_entry(iommu, &amd_iommu_list, list) + iommu_poll_events(iommu); + + return IRQ_HANDLED; } /**************************************************************************** diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c index 14a06464a694..eed488892c01 100644 --- a/arch/x86/kernel/amd_iommu_init.c +++ b/arch/x86/kernel/amd_iommu_init.c @@ -32,7 +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 diff --git a/include/asm-x86/amd_iommu_types.h b/include/asm-x86/amd_iommu_types.h index 8533f09b34b7..d8c5a6c69955 100644 --- a/include/asm-x86/amd_iommu_types.h +++ b/include/asm-x86/amd_iommu_types.h @@ -71,6 +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 @@ -165,6 +184,9 @@ #define MAX_DOMAIN_ID 65536 +/* FIXME: move this macro to */ +#define PCI_BUS(x) (((x) >> 8) & 0xff) + /* * This structure contains generic data for IOMMU protection domains * independent of their use. -- cgit v1.2.3 From bd60b735c658e6e8c656e89771d281bcfcf51279 Mon Sep 17 00:00:00 2001 From: Joerg Roedel Date: Thu, 11 Sep 2008 10:24:48 +0200 Subject: AMD IOMMU: don't assign preallocated protection domains to devices In isolation mode the protection domains for the devices are preallocated and preassigned. This is bad if a device should be passed to a virtualization guest because the IOMMU code does not know if it is in use by a driver. This patch changes the code to assign the device to the preallocated domain only if there are dma mapping requests for it. Signed-off-by: Joerg Roedel Signed-off-by: Ingo Molnar --- arch/x86/kernel/amd_iommu.c | 43 ++++++++++++++++++++++++++++++++++----- include/asm-x86/amd_iommu_types.h | 6 ++++++ 2 files changed, 44 insertions(+), 5 deletions(-) (limited to 'include') diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c index a6a6f8ed1cf5..7c1791447451 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 */ @@ -663,6 +667,7 @@ static struct dma_ops_domain *dma_ops_domain_alloc(struct amd_iommu *iommu, 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 && @@ -768,6 +773,33 @@ static bool check_device(struct device *dev) 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 @@ -803,9 +835,11 @@ static int get_device_resources(struct device *dev, *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 " @@ -1257,10 +1291,9 @@ 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); } } diff --git a/include/asm-x86/amd_iommu_types.h b/include/asm-x86/amd_iommu_types.h index d8c5a6c69955..9aa22ead22f3 100644 --- a/include/asm-x86/amd_iommu_types.h +++ b/include/asm-x86/amd_iommu_types.h @@ -227,6 +227,12 @@ struct dma_ops_domain { /* 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; }; /* -- cgit v1.2.3 From 38ddf41b198e21d3ecbe5752e875857b7ce7589e Mon Sep 17 00:00:00 2001 From: Joerg Roedel Date: Thu, 11 Sep 2008 10:38:32 +0200 Subject: AMD IOMMU: some set_device_domain cleanups Remove some magic numbers and split the pte_root using standard functions. Signed-off-by: Joerg Roedel Signed-off-by: Ingo Molnar --- arch/x86/kernel/amd_iommu.c | 9 +++++---- include/asm-x86/amd_iommu_types.h | 3 +++ 2 files changed, 8 insertions(+), 4 deletions(-) (limited to 'include') diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c index 7c1791447451..a34d8e915e3a 100644 --- a/arch/x86/kernel/amd_iommu.c +++ b/arch/x86/kernel/amd_iommu.c @@ -739,12 +739,13 @@ 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; diff --git a/include/asm-x86/amd_iommu_types.h b/include/asm-x86/amd_iommu_types.h index 9aa22ead22f3..f953309a636e 100644 --- a/include/asm-x86/amd_iommu_types.h +++ b/include/asm-x86/amd_iommu_types.h @@ -130,6 +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 @@ -159,6 +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) -- cgit v1.2.3 From d26dbc5cf94b0a28acc947285c3b54814a73cb2e Mon Sep 17 00:00:00 2001 From: FUJITA Tomonori Date: Mon, 22 Sep 2008 22:35:07 +0900 Subject: iommu: export iommu_area_reserve helper function x86 has set_bit_string() that does the exact same thing that set_bit_area() in lib/iommu-helper.c does. This patch exports set_bit_area() in lib/iommu-helper.c as iommu_area_reserve(), converts GART, Calgary, and AMD IOMMU to use it. Signed-off-by: FUJITA Tomonori Acked-by: Joerg Roedel Signed-off-by: Ingo Molnar --- arch/x86/kernel/amd_iommu.c | 2 +- arch/x86/kernel/pci-calgary_64.c | 2 +- arch/x86/kernel/pci-gart_64.c | 2 +- include/linux/iommu-helper.h | 1 + lib/iommu-helper.c | 5 ++--- 5 files changed, 6 insertions(+), 6 deletions(-) (limited to 'include') diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c index 6f7b97445738..70537d117a96 100644 --- a/arch/x86/kernel/amd_iommu.c +++ b/arch/x86/kernel/amd_iommu.c @@ -572,7 +572,7 @@ 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) diff --git a/arch/x86/kernel/pci-calgary_64.c b/arch/x86/kernel/pci-calgary_64.c index fe7695e4caae..080d1d27f37a 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); } diff --git a/arch/x86/kernel/pci-gart_64.c b/arch/x86/kernel/pci-gart_64.c index 508ef470b27f..3dcb1ad86e38 100644 --- a/arch/x86/kernel/pci-gart_64.c +++ b/arch/x86/kernel/pci-gart_64.c @@ -827,7 +827,7 @@ 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 diff --git a/include/linux/iommu-helper.h b/include/linux/iommu-helper.h index 58f41107e4ae..786539e432d7 100644 --- a/include/linux/iommu-helper.h +++ b/include/linux/iommu-helper.h @@ -11,6 +11,7 @@ static inline unsigned long iommu_device_max_index(unsigned long 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, diff --git a/lib/iommu-helper.c b/lib/iommu-helper.c index a3b8d4c3f77a..5d90074dca75 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; } -- cgit v1.2.3 From ed6dc4981368aa8ac89b0ea61535cfa2b03533cb Mon Sep 17 00:00:00 2001 From: FUJITA Tomonori Date: Tue, 23 Sep 2008 01:15:10 +0900 Subject: x86: remove set_bit_string() "export iommu_area_reserve helper funciton" patch converted all the users of set_bit_string, GART, Calgary and AMD IOMMU drivers, to use iommu_area_reserve helper function. Now we can remove unused set_bit_string function. Signed-off-by: FUJITA Tomonori Signed-off-by: Ingo Molnar --- include/asm-x86/bitops.h | 10 ---------- 1 file changed, 10 deletions(-) (limited to 'include') diff --git a/include/asm-x86/bitops.h b/include/asm-x86/bitops.h index cfb2b64f76e7..28ea8742b6e1 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 -- cgit v1.2.3 From afa9fdc2f5f8e4d98f3e77bfa204412cbc181346 Mon Sep 17 00:00:00 2001 From: FUJITA Tomonori Date: Sat, 20 Sep 2008 01:23:30 +0900 Subject: iommu: remove fullflush and nofullflush in IOMMU generic option This patch against tip/x86/iommu virtually reverts 2842e5bf3115193f05dc9dac20f940e7abf44c1a. But just reverting the commit breaks AMD IOMMU so this patch also includes some fixes. The above commit adds new two options to x86 IOMMU generic kernel boot options, fullflush and nofullflush. But such change that affects all the IOMMUs needs more discussion (all IOMMU parties need the chance to discuss it): http://lkml.org/lkml/2008/9/19/106 Signed-off-by: FUJITA Tomonori Acked-by: Joerg Roedel Signed-off-by: Ingo Molnar --- Documentation/kernel-parameters.txt | 9 +++++---- Documentation/x86/x86_64/boot-options.txt | 2 ++ arch/x86/kernel/amd_iommu.c | 4 ++-- arch/x86/kernel/amd_iommu_init.c | 5 ++++- arch/x86/kernel/pci-dma.c | 13 ------------- arch/x86/kernel/pci-gart_64.c | 13 +++++++++++++ include/asm-x86/amd_iommu_types.h | 6 ++++++ include/asm-x86/iommu.h | 1 - 8 files changed, 32 insertions(+), 21 deletions(-) (limited to 'include') diff --git a/Documentation/kernel-parameters.txt b/Documentation/kernel-parameters.txt index 40066ceb48fe..040ce30632b5 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: @@ -893,10 +898,6 @@ and is between 256 and 4096 characters. It is defined in the file nomerge forcesac soft - fullflush - Flush IO/TLB at every deallocation - nofullflush - Flush IO/TLB only when addresses are reused (default) intel_iommu= [DMAR] Intel IOMMU driver (DMAR) option diff --git a/Documentation/x86/x86_64/boot-options.txt b/Documentation/x86/x86_64/boot-options.txt index c83c8e4bc8e5..b0c7b6c4abda 100644 --- a/Documentation/x86/x86_64/boot-options.txt +++ b/Documentation/x86/x86_64/boot-options.txt @@ -233,6 +233,8 @@ IOMMU (input/output memory management unit) iommu options only relevant to the AMD GART hardware IOMMU: Set the size of the remapping area in bytes. allowed Overwrite iommu off workarounds for specific chipsets. + fullflush Flush IOMMU on each allocation (default). + nofullflush Don't use IOMMU fullflush. leak Turn on simple iommu leak tracing (only when CONFIG_IOMMU_LEAK is on). Default number of leak pages is 20. diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c index 70537d117a96..c19212191c98 100644 --- a/arch/x86/kernel/amd_iommu.c +++ b/arch/x86/kernel/amd_iommu.c @@ -948,7 +948,7 @@ static dma_addr_t __map_single(struct device *dev, } address += offset; - if (unlikely(dma_dom->need_flush && !iommu_fullflush)) { + 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))) @@ -985,7 +985,7 @@ static void __unmap_single(struct amd_iommu *iommu, dma_ops_free_addresses(dma_dom, dma_addr, pages); - if (iommu_fullflush) + if (amd_iommu_unmap_flush) iommu_flush_pages(iommu, dma_dom->domain.id, dma_addr, size); } diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c index db0c83af44de..148fcfe22f17 100644 --- a/arch/x86/kernel/amd_iommu_init.c +++ b/arch/x86/kernel/amd_iommu_init.c @@ -122,6 +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 */ @@ -1144,7 +1145,7 @@ int __init amd_iommu_init(void) else printk("disabled\n"); - if (iommu_fullflush) + 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"); @@ -1214,6 +1215,8 @@ static int __init parse_amd_iommu_options(char *str) for (; *str; ++str) { 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/pci-dma.c b/arch/x86/kernel/pci-dma.c index d2f2c0158dc1..0a1408abcc62 100644 --- a/arch/x86/kernel/pci-dma.c +++ b/arch/x86/kernel/pci-dma.c @@ -16,15 +16,6 @@ EXPORT_SYMBOL(dma_ops); static int iommu_sac_force __read_mostly; -/* - * 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; - #ifdef CONFIG_IOMMU_DEBUG int panic_on_overflow __read_mostly = 1; int force_iommu __read_mostly = 1; @@ -180,10 +171,6 @@ static __init int iommu_setup(char *p) } if (!strncmp(p, "nomerge", 7)) iommu_merge = 0; - if (!strncmp(p, "fullflush", 8)) - iommu_fullflush = 1; - if (!strncmp(p, "nofullflush", 11)) - iommu_fullflush = 0; if (!strncmp(p, "forcesac", 8)) iommu_sac_force = 1; if (!strncmp(p, "allowdac", 8)) diff --git a/arch/x86/kernel/pci-gart_64.c b/arch/x86/kernel/pci-gart_64.c index 3dcb1ad86e38..9e390f1bd46a 100644 --- a/arch/x86/kernel/pci-gart_64.c +++ b/arch/x86/kernel/pci-gart_64.c @@ -45,6 +45,15 @@ 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: */ @@ -892,6 +901,10 @@ void __init gart_parse_options(char *p) #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)) diff --git a/include/asm-x86/amd_iommu_types.h b/include/asm-x86/amd_iommu_types.h index f953309a636e..4ff892f3b0ad 100644 --- a/include/asm-x86/amd_iommu_types.h +++ b/include/asm-x86/amd_iommu_types.h @@ -376,6 +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/iommu.h b/include/asm-x86/iommu.h index 67b2fd56c6da..621a1af94c4c 100644 --- a/include/asm-x86/iommu.h +++ b/include/asm-x86/iommu.h @@ -7,7 +7,6 @@ extern struct dma_mapping_ops nommu_dma_ops; extern int force_iommu, no_iommu; extern int iommu_detected; extern int dmar_disabled; -extern int iommu_fullflush; extern unsigned long iommu_num_pages(unsigned long addr, unsigned long len); -- cgit v1.2.3 From 9f6ac57729724b58df81ca5dc005326759a806fe Mon Sep 17 00:00:00 2001 From: FUJITA Tomonori Date: Wed, 24 Sep 2008 20:48:35 +0900 Subject: x86: export pci-nommu's alloc_coherent This patch exports nommu_alloc_coherent (renamed dma_generic_alloc_coherent). GART needs this function. Signed-off-by: FUJITA Tomonori Signed-off-by: Ingo Molnar --- arch/x86/kernel/pci-dma.c | 31 +++++++++++++++++++++++++++++++ arch/x86/kernel/pci-nommu.c | 39 +-------------------------------------- include/asm-x86/dma-mapping.h | 3 +++ 3 files changed, 35 insertions(+), 38 deletions(-) (limited to 'include') diff --git a/arch/x86/kernel/pci-dma.c b/arch/x86/kernel/pci-dma.c index 0a1408abcc62..4e612d20170a 100644 --- a/arch/x86/kernel/pci-dma.c +++ b/arch/x86/kernel/pci-dma.c @@ -134,6 +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 for the iommu kernel parameter * documentation. diff --git a/arch/x86/kernel/pci-nommu.c b/arch/x86/kernel/pci-nommu.c index 1c1c98a31d57..c70ab5a5d4c8 100644 --- a/arch/x86/kernel/pci-nommu.c +++ b/arch/x86/kernel/pci-nommu.c @@ -72,43 +72,6 @@ static int nommu_map_sg(struct device *hwdev, struct scatterlist *sg, return nents; } -static void * -nommu_alloc_coherent(struct device *hwdev, size_t size, - dma_addr_t *dma_addr, gfp_t gfp) -{ - unsigned long dma_mask; - int node; - struct page *page; - dma_addr_t addr; - - dma_mask = dma_alloc_coherent_mask(hwdev, gfp); - - gfp |= __GFP_ZERO; - - node = dev_to_node(hwdev); -again: - page = alloc_pages_node(node, gfp, get_order(size)); - if (!page) - return NULL; - - addr = page_to_phys(page); - if (!is_buffer_dma_capable(dma_mask, addr, size) && !(gfp & GFP_DMA)) { - free_pages((unsigned long)page_address(page), get_order(size)); - gfp |= GFP_DMA; - goto again; - } - - if (check_addr("alloc_coherent", hwdev, addr, size)) { - *dma_addr = addr; - flush_write_buffers(); - return page_address(page); - } - - free_pages((unsigned long)page_address(page), get_order(size)); - - return NULL; -} - static void nommu_free_coherent(struct device *dev, size_t size, void *vaddr, dma_addr_t dma_addr) { @@ -116,7 +79,7 @@ static void nommu_free_coherent(struct device *dev, size_t size, void *vaddr, } struct dma_mapping_ops nommu_dma_ops = { - .alloc_coherent = nommu_alloc_coherent, + .alloc_coherent = dma_generic_alloc_coherent, .free_coherent = nommu_free_coherent, .map_single = nommu_map_single, .map_sg = nommu_map_sg, diff --git a/include/asm-x86/dma-mapping.h b/include/asm-x86/dma-mapping.h index f408e6dd1778..3b808e9bb72c 100644 --- a/include/asm-x86/dma-mapping.h +++ b/include/asm-x86/dma-mapping.h @@ -89,6 +89,9 @@ static inline int dma_mapping_error(struct device *dev, dma_addr_t dma_addr) 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) -- cgit v1.2.3 From 237a62247c2879331986a300d6ab36ad21264c68 Mon Sep 17 00:00:00 2001 From: Joerg Roedel Date: Thu, 25 Sep 2008 12:13:53 +0200 Subject: x86/iommu: make GART driver checkpatch clean Signed-off-by: Joerg Roedel Signed-off-by: Ingo Molnar --- arch/x86/kernel/pci-gart_64.c | 17 +++++++++-------- include/asm-x86/gart.h | 2 ++ 2 files changed, 11 insertions(+), 8 deletions(-) (limited to 'include') diff --git a/arch/x86/kernel/pci-gart_64.c b/arch/x86/kernel/pci-gart_64.c index b85a2f9bb340..aa569db73a2a 100644 --- a/arch/x86/kernel/pci-gart_64.c +++ b/arch/x86/kernel/pci-gart_64.c @@ -27,8 +27,8 @@ #include #include #include +#include #include -#include #include #include #include @@ -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"); @@ -688,7 +689,8 @@ 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(); @@ -710,8 +712,6 @@ 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, .unmap_single = gart_unmap_single, @@ -777,8 +777,8 @@ 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; } @@ -868,7 +868,8 @@ 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/include/asm-x86/gart.h b/include/asm-x86/gart.h index 3f62a83887f3..cdf4f78e081b 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; -- cgit v1.2.3 From 24268245d8ba9270152b2281666099ddc8ca389d Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Wed, 1 Oct 2008 08:59:40 +0200 Subject: x86: add PCI IDs for AMD Barcelona PCI devices Signed-off-by: Robert Richter Cc: oprofile-list Cc: Barry Kasindorf Signed-off-by: Ingo Molnar --- include/linux/pci_ids.h | 5 +++++ 1 file changed, 5 insertions(+) (limited to 'include') diff --git a/include/linux/pci_ids.h b/include/linux/pci_ids.h index 2886b0eb53ec..c114103af987 100644 --- a/include/linux/pci_ids.h +++ b/include/linux/pci_ids.h @@ -497,6 +497,11 @@ #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 -- cgit v1.2.3 From 1e19b16a30c34c042f1eaa23db4c99bfad1dac0e Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Mon, 6 Oct 2008 14:15:24 +0200 Subject: AMD IOMMU: use iommu_device_max_index, fix include/linux/iommu-helper.h has no header guards, which breaks sparc64 build. Add them. Signed-off-by: Thomas Gleixner Cc: Joerg Roedel Cc: FUJITA Tomonori Signed-off-by: Ingo Molnar --- include/linux/iommu-helper.h | 5 +++++ 1 file changed, 5 insertions(+) (limited to 'include') diff --git a/include/linux/iommu-helper.h b/include/linux/iommu-helper.h index 786539e432d7..a6d0586e2bf7 100644 --- a/include/linux/iommu-helper.h +++ b/include/linux/iommu-helper.h @@ -1,3 +1,6 @@ +#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) @@ -19,3 +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 -- cgit v1.2.3