Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1756143AbZDPAUO (ORCPT ); Wed, 15 Apr 2009 20:20:14 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1751932AbZDPAUA (ORCPT ); Wed, 15 Apr 2009 20:20:00 -0400 Received: from mga03.intel.com ([143.182.124.21]:32982 "EHLO mga03.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751814AbZDPAT7 (ORCPT ); Wed, 15 Apr 2009 20:19:59 -0400 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="4.40,194,1239001200"; d="scan'208";a="131963729" Date: Wed, 15 Apr 2009 17:19:57 -0700 From: Fenghua Yu To: David Woodhouse , Ingo Molnar , Linus Torvalds , Weidong Han Cc: LKML , IOMMU Subject: [PATCH] Intel IOMMU Pass Through Support Message-ID: <20090416001957.GA1527@linux-os.sc.intel.com> References: <20090327212241.234500000@intel.com>> <20090327212321.070229000@intel.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20090327212321.070229000@intel.com> User-Agent: Mutt/1.4.1i Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 17403 Lines: 544 The patch adds kernel parameter intel_iommu=pt to set up pass through mode in context mapping entry. This disables DMAR in linux kernel; but KVM still runs on VT-d and interrupt remapping still works. In this mode, kernel uses swiotlb for DMA API functions but other VT-d functionalities are enabled for KVM. KVM always uses multi level translation page table in VT-d. By default, pass though mode is disabled in kernel. This is useful when people don't want to enable VT-d DMAR in kernel but still want to use KVM and interrupt remapping for reasons like DMAR performance concern or debug purpose. Thanks. -Fenghua Signed-off-by: Fenghua Yu --- Documentation/kernel-parameters.txt | 5 arch/ia64/include/asm/iommu.h | 1 arch/ia64/kernel/pci-swiotlb.c | 2 arch/x86/include/asm/iommu.h | 1 arch/x86/kernel/pci-swiotlb.c | 3 drivers/pci/dmar.c | 9 + drivers/pci/intel-iommu.c | 187 ++++++++++++++++++++++++++---------- include/linux/dma_remapping.h | 8 + include/linux/intel-iommu.h | 2 9 files changed, 167 insertions(+), 51 deletions(-) diff --git a/Documentation/kernel-parameters.txt b/Documentation/kernel-parameters.txt index 6172e43..5594cdb 100644 --- a/Documentation/kernel-parameters.txt +++ b/Documentation/kernel-parameters.txt @@ -915,6 +915,11 @@ and is between 256 and 4096 characters. It is defined in the file With this option on every unmap_single operation will result in a hardware IOTLB flush operation as opposed to batching them for performance. + pt [Default no Pass Through] + This option enables Pass Through in context mapping if + Pass Through is supported in hardware. With this option + DMAR is disabled in kernel and kernel uses swiotlb, but + KVM can still uses VT-d IOTLB hardware. inttest= [IA64] diff --git a/arch/ia64/include/asm/iommu.h b/arch/ia64/include/asm/iommu.h index 0490794..37d41ca 100644 --- a/arch/ia64/include/asm/iommu.h +++ b/arch/ia64/include/asm/iommu.h @@ -9,6 +9,7 @@ extern void pci_iommu_shutdown(void); extern void no_iommu_init(void); extern int force_iommu, no_iommu; extern int iommu_detected; +extern int iommu_pass_through; extern void iommu_dma_init(void); extern void machvec_init(const char *name); diff --git a/arch/ia64/kernel/pci-swiotlb.c b/arch/ia64/kernel/pci-swiotlb.c index 285aae8..223abb1 100644 --- a/arch/ia64/kernel/pci-swiotlb.c +++ b/arch/ia64/kernel/pci-swiotlb.c @@ -46,7 +46,7 @@ void __init swiotlb_dma_init(void) void __init pci_swiotlb_init(void) { - if (!iommu_detected) { + if (!iommu_detected || iommu_pass_through) { #ifdef CONFIG_IA64_GENERIC swiotlb = 1; printk(KERN_INFO "PCI-DMA: Re-initialize machine vector.\n"); diff --git a/arch/x86/include/asm/iommu.h b/arch/x86/include/asm/iommu.h index af326a2..fd6d21b 100644 --- a/arch/x86/include/asm/iommu.h +++ b/arch/x86/include/asm/iommu.h @@ -6,6 +6,7 @@ extern void no_iommu_init(void); extern struct dma_map_ops nommu_dma_ops; extern int force_iommu, no_iommu; extern int iommu_detected; +extern int iommu_pass_through; /* 10 seconds */ #define DMAR_OPERATION_TIMEOUT ((cycles_t) tsc_khz*10*1000) diff --git a/arch/x86/kernel/pci-swiotlb.c b/arch/x86/kernel/pci-swiotlb.c index 34f12e9..42a0eb1 100644 --- a/arch/x86/kernel/pci-swiotlb.c +++ b/arch/x86/kernel/pci-swiotlb.c @@ -71,7 +71,8 @@ void __init pci_swiotlb_init(void) { /* don't initialize swiotlb if iommu=off (no_iommu=1) */ #ifdef CONFIG_X86_64 - if (!iommu_detected && !no_iommu && max_pfn > MAX_DMA32_PFN) + if ((!iommu_detected && !no_iommu && max_pfn > MAX_DMA32_PFN) || + iommu_pass_through) swiotlb = 1; #endif if (swiotlb_force) diff --git a/drivers/pci/dmar.c b/drivers/pci/dmar.c index fa3a113..1ef1a19 100644 --- a/drivers/pci/dmar.c +++ b/drivers/pci/dmar.c @@ -515,6 +515,7 @@ int alloc_iommu(struct dmar_drhd_unit *drhd) u32 ver; static int iommu_allocated = 0; int agaw = 0; + int msagaw = 0; iommu = kzalloc(sizeof(*iommu), GFP_KERNEL); if (!iommu) @@ -539,8 +540,16 @@ int alloc_iommu(struct dmar_drhd_unit *drhd) iommu->seq_id); goto error; } + msagaw = iommu_calculate_max_sagaw(iommu); + if (msagaw < 0) { + printk(KERN_ERR + "Cannot get a valid max agaw for iommu (seq_id = %d)\n", + iommu->seq_id); + goto error; + } #endif iommu->agaw = agaw; + iommu->msagaw = msagaw; /* the registers might be more than one page */ map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap), diff --git a/drivers/pci/intel-iommu.c b/drivers/pci/intel-iommu.c index 001b328..205e4a1 100644 --- a/drivers/pci/intel-iommu.c +++ b/drivers/pci/intel-iommu.c @@ -53,6 +53,8 @@ #define DEFAULT_DOMAIN_ADDRESS_WIDTH 48 +#define MAX_AGAW_WIDTH 64 + #define DOMAIN_MAX_ADDR(gaw) ((((u64)1) << gaw) - 1) #define IOVA_PFN(addr) ((addr) >> PAGE_SHIFT) @@ -127,8 +129,6 @@ static inline void context_set_fault_enable(struct context_entry *context) context->lo &= (((u64)-1) << 2) | 1; } -#define CONTEXT_TT_MULTI_LEVEL 0 - static inline void context_set_translation_type(struct context_entry *context, unsigned long value) { @@ -288,6 +288,7 @@ int dmar_disabled = 1; static int __initdata dmar_map_gfx = 1; static int dmar_forcedac; static int intel_iommu_strict; +int iommu_pass_through; #define DUMMY_DEVICE_DOMAIN_INFO ((struct device_domain_info *)(-1)) static DEFINE_SPINLOCK(device_domain_lock); @@ -318,6 +319,9 @@ static int __init intel_iommu_setup(char *str) printk(KERN_INFO "Intel-IOMMU: disable batched IOTLB flush\n"); intel_iommu_strict = 1; + } else if (!strncmp(str, "pt", 2)) { + iommu_pass_through = 1; + printk(KERN_INFO "Intel-IOMMU: Pass Through enabled\n"); } str += strcspn(str, ","); @@ -397,17 +401,13 @@ void free_iova_mem(struct iova *iova) static inline int width_to_agaw(int width); -/* calculate agaw for each iommu. - * "SAGAW" may be different across iommus, use a default agaw, and - * get a supported less agaw for iommus that don't support the default agaw. - */ -int iommu_calculate_agaw(struct intel_iommu *iommu) +static int __iommu_calculate_agaw(struct intel_iommu *iommu, int max_gaw) { unsigned long sagaw; int agaw = -1; sagaw = cap_sagaw(iommu->cap); - for (agaw = width_to_agaw(DEFAULT_DOMAIN_ADDRESS_WIDTH); + for (agaw = width_to_agaw(max_gaw); agaw >= 0; agaw--) { if (test_bit(agaw, &sagaw)) break; @@ -416,6 +416,24 @@ int iommu_calculate_agaw(struct intel_iommu *iommu) return agaw; } +/* + * Calculate max SAGAW for each iommu. + */ +int iommu_calculate_max_sagaw(struct intel_iommu *iommu) +{ + return __iommu_calculate_agaw(iommu, MAX_AGAW_WIDTH); +} + +/* + * calculate agaw for each iommu. + * "SAGAW" may be different across iommus, use a default agaw, and + * get a supported less agaw for iommus that don't support the default agaw. + */ +int iommu_calculate_agaw(struct intel_iommu *iommu) +{ + return __iommu_calculate_agaw(iommu, DEFAULT_DOMAIN_ADDRESS_WIDTH); +} + /* in native case, each domain is related to only one iommu */ static struct intel_iommu *domain_get_iommu(struct dmar_domain *domain) { @@ -1321,8 +1339,8 @@ static void domain_exit(struct dmar_domain *domain) free_domain_mem(domain); } -static int domain_context_mapping_one(struct dmar_domain *domain, - int segment, u8 bus, u8 devfn) +static int domain_context_mapping_one(struct dmar_domain *domain, int segment, + u8 bus, u8 devfn, int translation) { struct context_entry *context; unsigned long flags; @@ -1335,7 +1353,10 @@ static int domain_context_mapping_one(struct dmar_domain *domain, pr_debug("Set context mapping for %02x:%02x.%d\n", bus, PCI_SLOT(devfn), PCI_FUNC(devfn)); + BUG_ON(!domain->pgd); + BUG_ON(translation != CONTEXT_TT_PASS_THROUGH && + translation != CONTEXT_TT_MULTI_LEVEL); iommu = device_to_iommu(segment, bus, devfn); if (!iommu) @@ -1395,9 +1416,18 @@ static int domain_context_mapping_one(struct dmar_domain *domain, } context_set_domain_id(context, id); - context_set_address_width(context, iommu->agaw); - context_set_address_root(context, virt_to_phys(pgd)); - context_set_translation_type(context, CONTEXT_TT_MULTI_LEVEL); + + /* + * In pass through mode, AW must be programmed to indicate the largest + * AGAW value supported by hardware. And ASR is ignored by hardware. + */ + if (likely(translation == CONTEXT_TT_MULTI_LEVEL)) { + context_set_address_width(context, iommu->agaw); + context_set_address_root(context, virt_to_phys(pgd)); + } else + context_set_address_width(context, iommu->msagaw); + + context_set_translation_type(context, translation); context_set_fault_enable(context); context_set_present(context); domain_flush_cache(domain, context, sizeof(*context)); @@ -1422,13 +1452,15 @@ static int domain_context_mapping_one(struct dmar_domain *domain, } static int -domain_context_mapping(struct dmar_domain *domain, struct pci_dev *pdev) +domain_context_mapping(struct dmar_domain *domain, struct pci_dev *pdev, + int translation) { int ret; struct pci_dev *tmp, *parent; ret = domain_context_mapping_one(domain, pci_domain_nr(pdev->bus), - pdev->bus->number, pdev->devfn); + pdev->bus->number, pdev->devfn, + translation); if (ret) return ret; @@ -1440,9 +1472,9 @@ domain_context_mapping(struct dmar_domain *domain, struct pci_dev *pdev) parent = pdev->bus->self; while (parent != tmp) { ret = domain_context_mapping_one(domain, - pci_domain_nr(parent->bus), - parent->bus->number, - parent->devfn); + pci_domain_nr(parent->bus), + parent->bus->number, + parent->devfn, translation); if (ret) return ret; parent = parent->bus->self; @@ -1450,12 +1482,14 @@ domain_context_mapping(struct dmar_domain *domain, struct pci_dev *pdev) if (tmp->is_pcie) /* this is a PCIE-to-PCI bridge */ return domain_context_mapping_one(domain, pci_domain_nr(tmp->subordinate), - tmp->subordinate->number, 0); + tmp->subordinate->number, 0, + translation); else /* this is a legacy PCI bridge */ return domain_context_mapping_one(domain, pci_domain_nr(tmp->bus), tmp->bus->number, - tmp->devfn); + tmp->devfn, + translation); } static int domain_context_mapped(struct pci_dev *pdev) @@ -1752,7 +1786,7 @@ static int iommu_prepare_identity_map(struct pci_dev *pdev, goto error; /* context entry init */ - ret = domain_context_mapping(domain, pdev); + ret = domain_context_mapping(domain, pdev, CONTEXT_TT_MULTI_LEVEL); if (!ret) return 0; error: @@ -1853,6 +1887,23 @@ static inline void iommu_prepare_isa(void) } #endif /* !CONFIG_DMAR_FLPY_WA */ +/* Initialize each context entry as pass through.*/ +static int __init init_context_pass_through(void) +{ + struct pci_dev *pdev = NULL; + struct dmar_domain *domain; + int ret; + + for_each_pci_dev(pdev) { + domain = get_domain_for_dev(pdev, DEFAULT_DOMAIN_ADDRESS_WIDTH); + ret = domain_context_mapping(domain, pdev, + CONTEXT_TT_PASS_THROUGH); + if (ret) + return ret; + } + return 0; +} + static int __init init_dmars(void) { struct dmar_drhd_unit *drhd; @@ -1860,6 +1911,7 @@ static int __init init_dmars(void) struct pci_dev *pdev; struct intel_iommu *iommu; int i, ret; + int pass_through = 1; /* * for each drhd @@ -1913,7 +1965,15 @@ static int __init init_dmars(void) printk(KERN_ERR "IOMMU: allocate root entry failed\n"); goto error; } + if (!ecap_pass_through(iommu->ecap)) + pass_through = 0; } + if (iommu_pass_through) + if (!pass_through) { + printk(KERN_INFO + "Pass Through is not supported by hardware.\n"); + iommu_pass_through = 0; + } /* * Start from the sane iommu hardware state. @@ -1976,37 +2036,57 @@ static int __init init_dmars(void) "IOMMU: enable interrupt remapping failed\n"); } #endif + /* + * If pass through is set and enabled, context entries of all pci + * devices are intialized by pass through translation type. + */ + if (iommu_pass_through) { + ret = init_context_pass_through(); + if (ret) { + printk(KERN_ERR "IOMMU: Pass through init failed.\n"); + iommu_pass_through = 0; + } + } /* - * For each rmrr - * for each dev attached to rmrr - * do - * locate drhd for dev, alloc domain for dev - * allocate free domain - * allocate page table entries for rmrr - * if context not allocated for bus - * allocate and init context - * set present in root table for this bus - * init context with domain, translation etc - * endfor - * endfor + * If pass through is not set or not enabled, setup context entries for + * identity mappings for rmrr, gfx, and isa. */ - for_each_rmrr_units(rmrr) { - for (i = 0; i < rmrr->devices_cnt; i++) { - pdev = rmrr->devices[i]; - /* some BIOS lists non-exist devices in DMAR table */ - if (!pdev) - continue; - ret = iommu_prepare_rmrr_dev(rmrr, pdev); - if (ret) - printk(KERN_ERR + if (!iommu_pass_through) { + /* + * For each rmrr + * for each dev attached to rmrr + * do + * locate drhd for dev, alloc domain for dev + * allocate free domain + * allocate page table entries for rmrr + * if context not allocated for bus + * allocate and init context + * set present in root table for this bus + * init context with domain, translation etc + * endfor + * endfor + */ + for_each_rmrr_units(rmrr) { + for (i = 0; i < rmrr->devices_cnt; i++) { + pdev = rmrr->devices[i]; + /* + * some BIOS lists non-exist devices in DMAR + * table. + */ + if (!pdev) + continue; + ret = iommu_prepare_rmrr_dev(rmrr, pdev); + if (ret) + printk(KERN_ERR "IOMMU: mapping reserved region failed\n"); + } } - } - iommu_prepare_gfx_mapping(); + iommu_prepare_gfx_mapping(); - iommu_prepare_isa(); + iommu_prepare_isa(); + } /* * for each drhd @@ -2117,7 +2197,8 @@ get_valid_domain_for_dev(struct pci_dev *pdev) /* make sure context mapping is ok */ if (unlikely(!domain_context_mapped(pdev))) { - ret = domain_context_mapping(domain, pdev); + ret = domain_context_mapping(domain, pdev, + CONTEXT_TT_MULTI_LEVEL); if (ret) { printk(KERN_ERR "Domain context map for %s failed", @@ -2786,7 +2867,7 @@ int __init intel_iommu_init(void) * Check the need for DMA-remapping initialization now. * Above initialization will also be used by Interrupt-remapping. */ - if (no_iommu || swiotlb || dmar_disabled) + if (no_iommu || (swiotlb && !iommu_pass_through) || dmar_disabled) return -ENODEV; iommu_init_mempool(); @@ -2806,7 +2887,15 @@ int __init intel_iommu_init(void) init_timer(&unmap_timer); force_iommu = 1; - dma_ops = &intel_dma_ops; + + if (!iommu_pass_through) { + printk(KERN_INFO + "Multi-level page-table translation for DMAR.\n"); + dma_ops = &intel_dma_ops; + } else + printk(KERN_INFO + "DMAR: Pass through translation for DMAR.\n"); + init_iommu_sysfs(); register_iommu(&intel_iommu_ops); @@ -3146,7 +3235,7 @@ static int intel_iommu_attach_device(struct iommu_domain *domain, return -EFAULT; } - ret = domain_context_mapping(dmar_domain, pdev); + ret = domain_context_mapping(dmar_domain, pdev, CONTEXT_TT_MULTI_LEVEL); if (ret) return ret; diff --git a/include/linux/dma_remapping.h b/include/linux/dma_remapping.h index 1a455f1..e0a03af 100644 --- a/include/linux/dma_remapping.h +++ b/include/linux/dma_remapping.h @@ -13,6 +13,9 @@ #define DMA_PTE_WRITE (2) #define DMA_PTE_SNP (1 << 11) +#define CONTEXT_TT_MULTI_LEVEL 0 +#define CONTEXT_TT_PASS_THROUGH 2 + struct intel_iommu; struct dmar_domain; struct root_entry; @@ -21,11 +24,16 @@ extern void free_dmar_iommu(struct intel_iommu *iommu); #ifdef CONFIG_DMAR extern int iommu_calculate_agaw(struct intel_iommu *iommu); +extern int iommu_calculate_max_sagaw(struct intel_iommu *iommu); #else static inline int iommu_calculate_agaw(struct intel_iommu *iommu) { return 0; } +static inline int iommu_calculate_max_sagaw(struct intel_iommu *iommu) +{ + return 0; +} #endif extern int dmar_disabled; diff --git a/include/linux/intel-iommu.h b/include/linux/intel-iommu.h index aa8c531..7246971 100644 --- a/include/linux/intel-iommu.h +++ b/include/linux/intel-iommu.h @@ -120,6 +120,7 @@ static inline void dmar_writeq(void __iomem *addr, u64 val) (ecap_iotlb_offset(e) + ecap_niotlb_iunits(e) * 16) #define ecap_coherent(e) ((e) & 0x1) #define ecap_qis(e) ((e) & 0x2) +#define ecap_pass_through(e) ((e >> 6) & 0x1) #define ecap_eim_support(e) ((e >> 4) & 0x1) #define ecap_ir_support(e) ((e >> 3) & 0x1) #define ecap_max_handle_mask(e) ((e >> 20) & 0xf) @@ -302,6 +303,7 @@ struct intel_iommu { spinlock_t register_lock; /* protect register handling */ int seq_id; /* sequence id of the iommu */ int agaw; /* agaw of this iommu */ + int msagaw; /* max sagaw of this iommu */ unsigned int irq; unsigned char name[13]; /* Device Name */ -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/