Kernel KVM virtualization development
 help / color / mirror / Atom feed
* [PATCH 1/2] VT-d: Support multiple device assignment for KVM
@ 2008-11-26  3:18 Han, Weidong
  2008-11-26 10:53 ` Avi Kivity
  0 siblings, 1 reply; 5+ messages in thread
From: Han, Weidong @ 2008-11-26  3:18 UTC (permalink / raw)
  To: Woodhouse, David, Jesse Barnes, 'Avi Kivity'
  Cc: Kay, Allen M, Yu, Fenghua, kvm@vger.kernel.org,
	iommu@lists.linux-foundation.org

[-- Attachment #1: Type: text/plain, Size: 45273 bytes --]

In order to support multiple device assignment for KVM, this patch does following main changes:
   - extend dmar_domain to own multiple devices from different iommus, use a bitmap of iommus to replace iommu pointer in dmar_domain.
   - add a flag DOMAIN_FLAG_VIRTUAL_MACHINE to represent KVM VT-d usage. Many functions (e.g. intel_map_single() and intel_unmap_single()) won't be used by KVM VT-d. Let them return directly when this flag is set.
   - "SAGAW" capability may be different across iommus, that's to say the VT-d page table levels may be different among iommus. This patch uses a defaut agaw, and skip top levels of page tables for iommus which have smaller agaw than default.
   - rename the APIs for kvm VT-d, make it more readable.


Signed-off-by: Weidong Han <weidong.han@intel.com>
---
 drivers/pci/dmar.c            |   15 +
 drivers/pci/intel-iommu.c     |  601 +++++++++++++++++++++++++++++++----------
 include/linux/dma_remapping.h |   13 +-
 include/linux/intel-iommu.h   |   21 +-
 4 files changed, 503 insertions(+), 147 deletions(-)

diff --git a/drivers/pci/dmar.c b/drivers/pci/dmar.c
index 691b3ad..d6bdced 100644
--- a/drivers/pci/dmar.c
+++ b/drivers/pci/dmar.c
@@ -484,6 +484,7 @@ void __init detect_intel_iommu(void)
        dmar_tbl = NULL;
 }

+extern int width_to_agaw(int width);

 int alloc_iommu(struct dmar_drhd_unit *drhd)
 {
@@ -491,6 +492,8 @@ int alloc_iommu(struct dmar_drhd_unit *drhd)
        int map_size;
        u32 ver;
        static int iommu_allocated = 0;
+       unsigned long sagaw;
+       int agaw;

        iommu = kzalloc(sizeof(*iommu), GFP_KERNEL);
        if (!iommu)
@@ -506,6 +509,18 @@ int alloc_iommu(struct dmar_drhd_unit *drhd)
        iommu->cap = dmar_readq(iommu->reg + DMAR_CAP_REG);
        iommu->ecap = dmar_readq(iommu->reg + DMAR_ECAP_REG);

+       /* set agaw, "SAGAW" may be different across iommus */
+       sagaw = cap_sagaw(iommu->cap);
+       for (agaw = width_to_agaw(DEFAULT_DOMAIN_ADDRESS_WIDTH);
+            agaw >= 0; agaw--)
+               if (test_bit(agaw, &sagaw))
+                       break;
+       if (agaw < 0) {
+               printk(KERN_ERR "IOMMU: unsupported sagaw %lx\n", sagaw);
+               goto error;
+       }
+       iommu->agaw = agaw;
+
        /* the registers might be more than one page */
        map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap),
                cap_max_fault_reg_offset(iommu->cap));
diff --git a/drivers/pci/intel-iommu.c b/drivers/pci/intel-iommu.c
index 5c8baa4..3576ebe 100644
--- a/drivers/pci/intel-iommu.c
+++ b/drivers/pci/intel-iommu.c
@@ -50,8 +50,6 @@
 #define IOAPIC_RANGE_END       (0xfeefffff)
 #define IOVA_START_ADDR                (0x1000)

-#define DEFAULT_DOMAIN_ADDRESS_WIDTH 48
-
 #define DOMAIN_MAX_ADDR(gaw) ((((u64)1) << gaw) - 1)


@@ -64,6 +62,7 @@ struct deferred_flush_tables {
        int next;
        struct iova *iova[HIGH_WATER_MARK];
        struct dmar_domain *domain[HIGH_WATER_MARK];
+       struct intel_iommu *iommu;
 };

 static struct deferred_flush_tables *deferred_flush;
@@ -77,7 +76,7 @@ static LIST_HEAD(unmaps_to_do);
 static int timer_on;
 static long list_size;

-static void domain_remove_dev_info(struct dmar_domain *domain);
+static void domain_remove_all_dev_info(struct dmar_domain *domain);

 int dmar_disabled;
 static int __initdata dmar_map_gfx = 1;
@@ -184,6 +183,72 @@ void free_iova_mem(struct iova *iova)
        kmem_cache_free(iommu_iova_cache, iova);
 }

+/* in native case, each domain is related to only one iommu */
+static struct intel_iommu *domain_get_only_iommu(struct dmar_domain *domain)
+{
+       struct dmar_drhd_unit *drhd;
+
+       if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE)
+               return NULL;
+
+       for_each_drhd_unit(drhd) {
+               if (drhd->ignored)
+                       continue;
+               if (test_bit(drhd->iommu->seq_id, &domain->iommu_bmp))
+                       return drhd->iommu;
+       }
+
+       return NULL;
+}
+
+static void domain_flush_cache(struct dmar_domain *domain,
+                              void *addr, int size)
+{
+       struct intel_iommu *iommu;
+
+       if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE) {
+               struct dmar_drhd_unit *drhd;
+
+               for_each_drhd_unit(drhd) {
+                       if (drhd->ignored)
+                               continue;
+                       iommu = drhd->iommu;
+
+                       if (!test_bit(iommu->seq_id, &domain->iommu_bmp))
+                               continue;
+
+                       if (!ecap_coherent(iommu->ecap))
+                               clflush_cache_range(addr, size);
+               }
+       }
+       else {
+               iommu = domain_get_only_iommu(domain);
+               if (iommu && !ecap_coherent(iommu->ecap))
+                       clflush_cache_range(addr, size);
+       }
+}
+
+static struct intel_iommu *iommu_find_matched_iommu(u8 bus, u8 devfn)
+{
+       struct dmar_drhd_unit *drhd = NULL;
+       int i;
+
+       for_each_drhd_unit(drhd) {
+               if (drhd->ignored)
+                       continue;
+
+               for (i = 0; i < drhd->devices_cnt; i++)
+                       if (drhd->devices[i]->bus->number == bus &&
+                           drhd->devices[i]->devfn == devfn)
+                               return drhd->iommu;
+
+               if (drhd->include_all)
+                       return drhd->iommu;
+       }
+
+       return NULL;
+}
+
 /* Gets context entry for a given bus and devfn */
 static struct context_entry * device_to_context_entry(struct intel_iommu *iommu,
                u8 bus, u8 devfn)
@@ -287,7 +352,7 @@ static inline int agaw_to_width(int agaw)

 }

-static inline int width_to_agaw(int width)
+int width_to_agaw(int width)
 {
        return (width - 30) / LEVEL_STRIDE;
 }
@@ -347,8 +412,7 @@ static struct dma_pte * addr_to_dma_pte(struct dmar_domain *domain, u64 addr)
                                        flags);
                                return NULL;
                        }
-                       __iommu_flush_cache(domain->iommu, tmp_page,
-                                       PAGE_SIZE);
+                       domain_flush_cache(domain, tmp_page, PAGE_SIZE);
                        dma_set_pte_addr(*pte, virt_to_phys(tmp_page));
                        /*
                         * high level table always sets r/w, last level page
@@ -356,7 +420,7 @@ static struct dma_pte * addr_to_dma_pte(struct dmar_domain *domain, u64 addr)
                         */
                        dma_set_pte_readable(*pte);
                        dma_set_pte_writable(*pte);
-                       __iommu_flush_cache(domain->iommu, pte, sizeof(*pte));
+                       domain_flush_cache(domain, pte, sizeof(*pte));
                }
                parent = phys_to_virt(dma_pte_addr(*pte));
                level--;
@@ -399,7 +463,7 @@ static void dma_pte_clear_one(struct dmar_domain *domain, u64 addr)

        if (pte) {
                dma_clear_pte(*pte);
-               __iommu_flush_cache(domain->iommu, pte, sizeof(*pte));
+               domain_flush_cache(domain, pte, sizeof(*pte));
        }
 }

@@ -447,8 +511,7 @@ static void dma_pte_free_pagetable(struct dmar_domain *domain,
                                free_pgtable_page(
                                        phys_to_virt(dma_pte_addr(*pte)));
                                dma_clear_pte(*pte);
-                               __iommu_flush_cache(domain->iommu,
-                                               pte, sizeof(*pte));
+                               domain_flush_cache(domain, pte, sizeof(*pte));
                        }
                        tmp += level_size(level);
                }
@@ -951,6 +1014,37 @@ static int iommu_init_domains(struct intel_iommu *iommu)

 static void domain_exit(struct dmar_domain *domain);

+static int domain_in_other_iommus(struct dmar_domain *domain,
+                                 struct intel_iommu *iommu)
+{
+       struct dmar_drhd_unit *drhd;
+       struct intel_iommu *other_iommu;
+       unsigned long ndomains;
+       unsigned long i;
+
+       if (!domain)
+               return 0;
+
+       if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE) {
+               for_each_drhd_unit(drhd) {
+                       if (drhd->ignored && drhd->iommu == iommu)
+                               continue;
+                       other_iommu = drhd->iommu;
+                       ndomains = cap_ndoms(other_iommu->cap);
+                       i = find_first_bit(other_iommu->domain_ids,
+                                          cap_ndoms(other_iommu->cap));
+                       for (i = 0; i < ndomains; ) {
+                               if (other_iommu->domains[i] == domain)
+                                       return 1;
+                               i = find_next_bit(other_iommu->domain_ids,
+                                                 ndomains, i+1);
+                       }
+               }
+       }
+
+       return 0;
+}
+
 void free_dmar_iommu(struct intel_iommu *iommu)
 {
        struct dmar_domain *domain;
@@ -960,7 +1054,14 @@ void free_dmar_iommu(struct intel_iommu *iommu)
        for (; i < cap_ndoms(iommu->cap); ) {
                domain = iommu->domains[i];
                clear_bit(i, iommu->domain_ids);
-               domain_exit(domain);
+
+               if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE) {
+                       /* domain may be referenced by other iommus */
+                       if (domain_in_other_iommus(domain, iommu) == 0)
+                               domain_exit(domain);
+               }
+               else
+                       domain_exit(domain);
                i = find_next_bit(iommu->domain_ids,
                        cap_ndoms(iommu->cap), i+1);
        }
@@ -1006,7 +1107,7 @@ static struct dmar_domain * iommu_alloc_domain(struct intel_iommu *iommu)

        set_bit(num, iommu->domain_ids);
        domain->id = num;
-       domain->iommu = iommu;
+       set_bit(iommu->seq_id, &domain->iommu_bmp);
        iommu->domains[num] = domain;
        spin_unlock_irqrestore(&iommu->lock, flags);

@@ -1016,10 +1117,39 @@ static struct dmar_domain * iommu_alloc_domain(struct intel_iommu *iommu)
 static void iommu_free_domain(struct dmar_domain *domain)
 {
        unsigned long flags;
+       struct dmar_drhd_unit *drhd;
+       struct intel_iommu *iommu;
+       unsigned long i;
+       unsigned long ndomains;

-       spin_lock_irqsave(&domain->iommu->lock, flags);
-       clear_bit(domain->id, domain->iommu->domain_ids);
-       spin_unlock_irqrestore(&domain->iommu->lock, flags);
+       if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE) {
+               for_each_drhd_unit(drhd) {
+                       if (drhd->ignored)
+                               continue;
+                       iommu = drhd->iommu;
+
+                       ndomains = cap_ndoms(iommu->cap);
+                       i = find_first_bit(iommu->domain_ids, ndomains);
+                       for (; i < ndomains; ) {
+                               if (iommu->domains[i] == domain) {
+                                       spin_lock_irqsave(&iommu->lock, flags);
+                                       clear_bit(i, iommu->domain_ids);
+                                       iommu->domains[i] = NULL;
+                                       spin_unlock_irqrestore(&iommu->lock,
+                                                              flags);
+                                       break;
+                               }
+                               i = find_next_bit(iommu->domain_ids,
+                                                 ndomains, i+1);
+                       }
+               }
+       }
+       else {
+               iommu = domain_get_only_iommu(domain);
+               spin_lock_irqsave(&iommu->lock, flags);
+               clear_bit(domain->id, iommu->domain_ids);
+               spin_unlock_irqrestore(&iommu->lock, flags);
+       }
 }

 static struct iova_domain reserved_iova_list;
@@ -1097,29 +1227,37 @@ static int domain_init(struct dmar_domain *domain, int guest_width)

        domain_reserve_special_ranges(domain);

-       /* calculate AGAW */
-       iommu = domain->iommu;
-       if (guest_width > cap_mgaw(iommu->cap))
-               guest_width = cap_mgaw(iommu->cap);
-       domain->gaw = guest_width;
-       adjust_width = guestwidth_to_adjustwidth(guest_width);
-       agaw = width_to_agaw(adjust_width);
-       sagaw = cap_sagaw(iommu->cap);
-       if (!test_bit(agaw, &sagaw)) {
-               /* hardware doesn't support it, choose a bigger one */
-               pr_debug("IOMMU: hardware doesn't support agaw %d\n", agaw);
-               agaw = find_next_bit(&sagaw, 5, agaw);
-               if (agaw >= 5)
-                       return -ENODEV;
-       }
-       domain->agaw = agaw;
+       /* set AGAW */
+       if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE) {
+               domain->gaw = DEFAULT_DOMAIN_ADDRESS_WIDTH;
+               domain->agaw = width_to_agaw(DEFAULT_DOMAIN_ADDRESS_WIDTH);
+       }
+       else {
+               iommu = domain_get_only_iommu(domain);
+               if (guest_width > cap_mgaw(iommu->cap))
+                       guest_width = cap_mgaw(iommu->cap);
+               domain->gaw = guest_width;
+               adjust_width = guestwidth_to_adjustwidth(guest_width);
+               agaw = width_to_agaw(adjust_width);
+               sagaw = cap_sagaw(iommu->cap);
+               if (!test_bit(agaw, &sagaw)) {
+                       /* hardware doesn't support it, choose a bigger one */
+                       pr_debug("IOMMU: hardware doesn't support agaw %d\n",
+                                agaw);
+                       agaw = find_next_bit(&sagaw, 5, agaw);
+                       if (agaw >= 5)
+                               return -ENODEV;
+               }
+               domain->agaw = agaw;
+       }
+
        INIT_LIST_HEAD(&domain->devices);

        /* always allocate the top pgd */
        domain->pgd = (struct dma_pte *)alloc_pgtable_page();
        if (!domain->pgd)
                return -ENOMEM;
-       __iommu_flush_cache(iommu, domain->pgd, PAGE_SIZE);
+       domain_flush_cache(domain, domain->pgd, PAGE_SIZE);
        return 0;
 }

@@ -1131,7 +1269,7 @@ static void domain_exit(struct dmar_domain *domain)
        if (!domain)
                return;

-       domain_remove_dev_info(domain);
+       domain_remove_all_dev_info(domain);
        /* destroy iovas */
        put_iova_domain(&domain->iovad);
        end = DOMAIN_MAX_ADDR(domain->gaw);
@@ -1148,11 +1286,15 @@ static void domain_exit(struct dmar_domain *domain)
 }

 static int domain_context_mapping_one(struct dmar_domain *domain,
-               u8 bus, u8 devfn)
+               struct intel_iommu *iommu, u8 bus, u8 devfn)
 {
        struct context_entry *context;
-       struct intel_iommu *iommu = domain->iommu;
        unsigned long flags;
+       struct dma_pte *pgd;
+       unsigned long num;
+       unsigned long ndomains;
+       int id;
+       int agaw;

        pr_debug("Set context mapping for %02x:%02x.%d\n",
                bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
@@ -1166,22 +1308,67 @@ static int domain_context_mapping_one(struct dmar_domain *domain,
                return 0;
        }

-       context_set_domain_id(*context, domain->id);
-       context_set_address_width(*context, domain->agaw);
-       context_set_address_root(*context, virt_to_phys(domain->pgd));
+       id = domain->id;
+       pgd = domain->pgd;
+
+       if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE) {
+               int found = 0;
+
+               /* find an available domain id for this device in iommu */
+               ndomains = cap_ndoms(iommu->cap);
+               num = find_first_bit(iommu->domain_ids, ndomains);
+               for (; num < ndomains; ) {
+                       if (iommu->domains[num] == domain) {
+                               id = num;
+                               found = 1;
+                               break;
+                       }
+                       num = find_next_bit(iommu->domain_ids,
+                                           cap_ndoms(iommu->cap), num+1);
+               }
+
+               if (found == 0) {
+                       num = find_first_zero_bit(iommu->domain_ids, ndomains);
+                       if (num >= ndomains) {
+                               spin_unlock_irqrestore(&iommu->lock, flags);
+                               printk(KERN_ERR "IOMMU: no free domain ids\n");
+                               return -EFAULT;
+                       }
+
+                       set_bit(num, iommu->domain_ids);
+                       iommu->domains[num] = domain;
+                       id = num;
+               }
+
+               /* Skip top levels of page tables for
+                * iommu which has less agaw than default.
+                */
+               for (agaw = domain->agaw; agaw != iommu->agaw; agaw--) {
+                       pgd = phys_to_virt(dma_pte_addr(*pgd));
+                       if (!dma_pte_present(*pgd)) {
+                               spin_unlock_irqrestore(&iommu->lock, flags);
+                               return -ENOMEM;
+                       }
+               }
+       }
+
+       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);
        context_set_fault_enable(*context);
        context_set_present(*context);
        __iommu_flush_cache(iommu, context, sizeof(*context));

        /* it's a non-present to present mapping */
-       if (iommu->flush.flush_context(iommu, domain->id,
+       if (iommu->flush.flush_context(iommu, id,
                (((u16)bus) << 8) | devfn, DMA_CCMD_MASK_NOBIT,
                DMA_CCMD_DEVICE_INVL, 1))
                iommu_flush_write_buffer(iommu);
        else
                iommu->flush.flush_iotlb(iommu, 0, 0, 0, DMA_TLB_DSI_FLUSH, 0);

+       set_bit(iommu->seq_id, &domain->iommu_bmp);
        spin_unlock_irqrestore(&iommu->lock, flags);
        return 0;
 }
@@ -1191,9 +1378,15 @@ domain_context_mapping(struct dmar_domain *domain, struct pci_dev *pdev)
 {
        int ret;
        struct pci_dev *tmp, *parent;
+       struct intel_iommu *iommu;

-       ret = domain_context_mapping_one(domain, pdev->bus->number,
-               pdev->devfn);
+       iommu = iommu_find_matched_iommu(pdev->bus->number,
+                                        pdev->devfn);
+       if (!iommu)
+               return  -ENODEV;
+
+       ret = domain_context_mapping_one(domain, iommu,
+               pdev->bus->number, pdev->devfn);
        if (ret)
                return ret;

@@ -1204,27 +1397,32 @@ domain_context_mapping(struct dmar_domain *domain, struct pci_dev *pdev)
        /* Secondary interface's bus number and devfn 0 */
        parent = pdev->bus->self;
        while (parent != tmp) {
-               ret = domain_context_mapping_one(domain, parent->bus->number,
-                       parent->devfn);
+               ret = domain_context_mapping_one(domain, iommu,
+                       parent->bus->number, parent->devfn);
                if (ret)
                        return ret;
                parent = parent->bus->self;
        }
        if (tmp->is_pcie) /* this is a PCIE-to-PCI bridge */
-               return domain_context_mapping_one(domain,
+               return domain_context_mapping_one(domain, iommu,
                        tmp->subordinate->number, 0);
        else /* this is a legacy PCI bridge */
-               return domain_context_mapping_one(domain,
+               return domain_context_mapping_one(domain, iommu,
                        tmp->bus->number, tmp->devfn);
 }

-static int domain_context_mapped(struct dmar_domain *domain,
-       struct pci_dev *pdev)
+static int domain_context_mapped(struct pci_dev *pdev)
 {
        int ret;
        struct pci_dev *tmp, *parent;
+       struct intel_iommu *iommu;
+
+       iommu = iommu_find_matched_iommu(pdev->bus->number,
+                                        pdev->devfn);
+       if (!iommu)
+               return 0;

-       ret = device_context_mapped(domain->iommu,
+       ret = device_context_mapped(iommu,
                pdev->bus->number, pdev->devfn);
        if (!ret)
                return ret;
@@ -1235,17 +1433,17 @@ static int domain_context_mapped(struct dmar_domain *domain,
        /* Secondary interface's bus number and devfn 0 */
        parent = pdev->bus->self;
        while (parent != tmp) {
-               ret = device_context_mapped(domain->iommu, parent->bus->number,
+               ret = device_context_mapped(iommu, parent->bus->number,
                        parent->devfn);
                if (!ret)
                        return ret;
                parent = parent->bus->self;
        }
        if (tmp->is_pcie)
-               return device_context_mapped(domain->iommu,
+               return device_context_mapped(iommu,
                        tmp->subordinate->number, 0);
        else
-               return device_context_mapped(domain->iommu,
+               return device_context_mapped(iommu,
                        tmp->bus->number, tmp->devfn);
 }

@@ -1276,23 +1474,60 @@ domain_page_mapping(struct dmar_domain *domain, dma_addr_t iova,
                BUG_ON(dma_pte_addr(*pte));
                dma_set_pte_addr(*pte, start_pfn << VTD_PAGE_SHIFT);
                dma_set_pte_prot(*pte, prot);
-               __iommu_flush_cache(domain->iommu, pte, sizeof(*pte));
+               domain_flush_cache(domain, pte, sizeof(*pte));
                start_pfn++;
                index++;
        }
        return 0;
 }

-static void detach_domain_for_dev(struct dmar_domain *domain, u8 bus, u8 devfn)
+static void detach_domain_for_dev(struct dmar_domain *domain,
+                                 u8 bus, u8 devfn)
 {
-       clear_context_table(domain->iommu, bus, devfn);
-       domain->iommu->flush.flush_context(domain->iommu, 0, 0, 0,
-                                          DMA_CCMD_GLOBAL_INVL, 0);
-       domain->iommu->flush.flush_iotlb(domain->iommu, 0, 0, 0,
-                                        DMA_TLB_GLOBAL_FLUSH, 0);
+       struct device_domain_info *info;
+       unsigned long flags;
+       struct intel_iommu *iommu;
+
+       iommu = iommu_find_matched_iommu(bus, devfn);
+       if (!iommu)
+               return;
+
+       clear_context_table(iommu, bus, devfn);
+       iommu->flush.flush_context(iommu, 0, 0, 0,
+                                  DMA_CCMD_GLOBAL_INVL, 0);
+       iommu->flush.flush_iotlb(iommu, 0, 0, 0,
+                                DMA_TLB_GLOBAL_FLUSH, 0);
+
+       /* clear this iommu in iommu_bmp if there is no
+        * its device owned by this domain
+        */
+       if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE) {
+               int found = 0;
+
+               while (!list_empty(&domain->devices)) {
+                       info = list_entry(domain->devices.next,
+                               struct device_domain_info, link);
+
+                       if (iommu_find_matched_iommu(info->bus,
+                                       info->devfn) == iommu) {
+                               found = 1;
+                               break;
+                       }
+               }
+               if (found == 0) {
+                       spin_lock_irqsave(&iommu->lock, flags);
+                       clear_bit(iommu->seq_id, &domain->iommu_bmp);
+                       spin_unlock_irqrestore(&iommu->lock, flags);
+               }
+       }
+       else {
+               spin_lock_irqsave(&iommu->lock, flags);
+               clear_bit(iommu->seq_id, &domain->iommu_bmp);
+               spin_unlock_irqrestore(&iommu->lock, flags);
+       }
 }

-static void domain_remove_dev_info(struct dmar_domain *domain)
+static void domain_remove_all_dev_info(struct dmar_domain *domain)
 {
        struct device_domain_info *info;
        unsigned long flags;
@@ -1336,7 +1571,6 @@ static struct dmar_domain *get_domain_for_dev(struct pci_dev *pdev, int gaw)
 {
        struct dmar_domain *domain, *found = NULL;
        struct intel_iommu *iommu;
-       struct dmar_drhd_unit *drhd;
        struct device_domain_info *info, *tmp;
        struct pci_dev *dev_tmp;
        unsigned long flags;
@@ -1371,13 +1605,10 @@ static struct dmar_domain *get_domain_for_dev(struct pci_dev *pdev, int gaw)
        }

        /* Allocate new domain for the device */
-       drhd = dmar_find_matched_drhd_unit(pdev);
-       if (!drhd) {
-               printk(KERN_ERR "IOMMU: can't find DMAR for device %s\n",
-                       pci_name(pdev));
-               return NULL;
-       }
-       iommu = drhd->iommu;
+       iommu = iommu_find_matched_iommu(pdev->bus->number,
+                                        pdev->devfn);
+       if (!iommu)
+               return  NULL;

        domain = iommu_alloc_domain(iommu);
        if (!domain)
@@ -1400,7 +1631,7 @@ static struct dmar_domain *get_domain_for_dev(struct pci_dev *pdev, int gaw)
                info->dev = NULL;
                info->domain = domain;
                /* This domain is shared by devices under p2p bridge */
-               domain->flags |= DOMAIN_FLAG_MULTIPLE_DEVICES;
+               domain->flags |= DOMAIN_FLAG_P2P_MULTIPLE_DEVICES;

                /* pcie-to-pci bridge already has a domain, uses it */
                found = NULL;
@@ -1805,7 +2036,7 @@ get_valid_domain_for_dev(struct pci_dev *pdev)
        }

        /* make sure context mapping is ok */
-       if (unlikely(!domain_context_mapped(domain, pdev))) {
+       if (unlikely(!domain_context_mapped(pdev))) {
                ret = domain_context_mapping(domain, pdev);
                if (ret) {
                        printk(KERN_ERR
@@ -1823,6 +2054,7 @@ static dma_addr_t __intel_map_single(struct device *hwdev, phys_addr_t paddr,
 {
        struct pci_dev *pdev = to_pci_dev(hwdev);
        struct dmar_domain *domain;
+       struct intel_iommu *iommu;
        phys_addr_t start_paddr;
        struct iova *iova;
        int prot = 0;
@@ -1836,6 +2068,14 @@ static dma_addr_t __intel_map_single(struct device *hwdev, phys_addr_t paddr,
        if (!domain)
                return 0;

+       if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE)
+               return 0;
+
+       iommu = iommu_find_matched_iommu(pdev->bus->number,
+                                        pdev->devfn);
+       if (!iommu)
+               return 0;
+
        size = aligned_size((u64)paddr, size);

        iova = __intel_alloc_iova(hwdev, domain, size, pdev->dma_mask);
@@ -1849,7 +2089,7 @@ static dma_addr_t __intel_map_single(struct device *hwdev, phys_addr_t paddr,
         * mappings..
         */
        if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL || \
-                       !cap_zlr(domain->iommu->cap))
+                       !cap_zlr(iommu->cap))
                prot |= DMA_PTE_READ;
        if (dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL)
                prot |= DMA_PTE_WRITE;
@@ -1865,10 +2105,10 @@ static dma_addr_t __intel_map_single(struct device *hwdev, phys_addr_t paddr,
                goto error;

        /* it's a non-present to present mapping */
-       ret = iommu_flush_iotlb_psi(domain->iommu, domain->id,
+       ret = iommu_flush_iotlb_psi(iommu, domain->id,
                        start_paddr, size >> VTD_PAGE_SHIFT, 1);
        if (ret)
-               iommu_flush_write_buffer(domain->iommu);
+               iommu_flush_write_buffer(iommu);

        return start_paddr + ((u64)paddr & (~PAGE_MASK));

@@ -1896,8 +2136,7 @@ static void flush_unmaps(void)
        /* just flush them all */
        for (i = 0; i < g_num_of_iommus; i++) {
                if (deferred_flush[i].next) {
-                       struct intel_iommu *iommu =
-                               deferred_flush[i].domain[0]->iommu;
+                       struct intel_iommu *iommu = deferred_flush[i].iommu;

                        iommu->flush.flush_iotlb(iommu, 0, 0, 0,
                                                 DMA_TLB_GLOBAL_FLUSH, 0);
@@ -1921,7 +2160,8 @@ static void flush_unmaps_timeout(unsigned long data)
        spin_unlock_irqrestore(&async_umap_flush_lock, flags);
 }

-static void add_unmap(struct dmar_domain *dom, struct iova *iova)
+static void add_unmap(struct dmar_domain *dom,
+                     struct intel_iommu *iommu, struct iova *iova)
 {
        unsigned long flags;
        int next, iommu_id;
@@ -1930,11 +2170,12 @@ static void add_unmap(struct dmar_domain *dom, struct iova *iova)
        if (list_size == HIGH_WATER_MARK)
                flush_unmaps();

-       iommu_id = dom->iommu->seq_id;
+       iommu_id = iommu->seq_id;

        next = deferred_flush[iommu_id].next;
        deferred_flush[iommu_id].domain[next] = dom;
        deferred_flush[iommu_id].iova[next] = iova;
+       deferred_flush[iommu_id].iommu = iommu;
        deferred_flush[iommu_id].next++;

        if (!timer_on) {
@@ -1952,12 +2193,21 @@ void intel_unmap_single(struct device *dev, dma_addr_t dev_addr, size_t size,
        struct dmar_domain *domain;
        unsigned long start_addr;
        struct iova *iova;
+       struct intel_iommu *iommu;

        if (pdev->dev.archdata.iommu == DUMMY_DEVICE_DOMAIN_INFO)
                return;
        domain = find_domain(pdev);
        BUG_ON(!domain);

+       if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE)
+               return;
+
+       iommu = iommu_find_matched_iommu(pdev->bus->number,
+                                        pdev->devfn);
+       if (!iommu)
+               return;
+
        iova = find_iova(&domain->iovad, IOVA_PFN(dev_addr));
        if (!iova)
                return;
@@ -1973,13 +2223,13 @@ void intel_unmap_single(struct device *dev, dma_addr_t dev_addr, size_t size,
        /* free page tables */
        dma_pte_free_pagetable(domain, start_addr, start_addr + size);
        if (intel_iommu_strict) {
-               if (iommu_flush_iotlb_psi(domain->iommu,
+               if (iommu_flush_iotlb_psi(iommu,
                        domain->id, start_addr, size >> VTD_PAGE_SHIFT, 0))
-                       iommu_flush_write_buffer(domain->iommu);
+                       iommu_flush_write_buffer(iommu);
                /* free iova */
                __free_iova(&domain->iovad, iova);
        } else {
-               add_unmap(domain, iova);
+               add_unmap(domain, iommu, iova);
                /*
                 * queue up the release of the unmap to save the 1/6th of the
                 * cpu used up by the iotlb flush operation...
@@ -2036,12 +2286,21 @@ void intel_unmap_sg(struct device *hwdev, struct scatterlist *sglist,
        size_t size = 0;
        void *addr;
        struct scatterlist *sg;
+       struct intel_iommu *iommu;

        if (pdev->dev.archdata.iommu == DUMMY_DEVICE_DOMAIN_INFO)
                return;

        domain = find_domain(pdev);

+       if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE)
+               return;
+
+       iommu = iommu_find_matched_iommu(pdev->bus->number,
+                                        pdev->devfn);
+       if (!iommu)
+               return;
+
        iova = find_iova(&domain->iovad, IOVA_PFN(sglist[0].dma_address));
        if (!iova)
                return;
@@ -2057,9 +2316,9 @@ void intel_unmap_sg(struct device *hwdev, struct scatterlist *sglist,
        /* free page tables */
        dma_pte_free_pagetable(domain, start_addr, start_addr + size);

-       if (iommu_flush_iotlb_psi(domain->iommu, domain->id, start_addr,
+       if (iommu_flush_iotlb_psi(iommu, domain->id, start_addr,
                        size >> VTD_PAGE_SHIFT, 0))
-               iommu_flush_write_buffer(domain->iommu);
+               iommu_flush_write_buffer(iommu);

        /* free iova */
        __free_iova(&domain->iovad, iova);
@@ -2093,6 +2352,7 @@ int intel_map_sg(struct device *hwdev, struct scatterlist *sglist, int nelems,
        int ret;
        struct scatterlist *sg;
        unsigned long start_addr;
+       struct intel_iommu *iommu;

        BUG_ON(dir == DMA_NONE);
        if (pdev->dev.archdata.iommu == DUMMY_DEVICE_DOMAIN_INFO)
@@ -2102,6 +2362,14 @@ int intel_map_sg(struct device *hwdev, struct scatterlist *sglist, int nelems,
        if (!domain)
                return 0;

+       if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE)
+               return 0;
+
+       iommu = iommu_find_matched_iommu(pdev->bus->number,
+                                        pdev->devfn);
+       if (!iommu)
+               return 0;
+
        for_each_sg(sglist, sg, nelems, i) {
                addr = SG_ENT_VIRT_ADDRESS(sg);
                addr = (void *)virt_to_phys(addr);
@@ -2119,7 +2387,7 @@ int intel_map_sg(struct device *hwdev, struct scatterlist *sglist, int nelems,
         * mappings..
         */
        if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL || \
-                       !cap_zlr(domain->iommu->cap))
+                       !cap_zlr(iommu->cap))
                prot |= DMA_PTE_READ;
        if (dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL)
                prot |= DMA_PTE_WRITE;
@@ -2151,9 +2419,9 @@ int intel_map_sg(struct device *hwdev, struct scatterlist *sglist, int nelems,
        }

        /* it's a non-present to present mapping */
-       if (iommu_flush_iotlb_psi(domain->iommu, domain->id,
+       if (iommu_flush_iotlb_psi(iommu, domain->id,
                        start_addr, offset >> VTD_PAGE_SHIFT, 1))
-               iommu_flush_write_buffer(domain->iommu);
+               iommu_flush_write_buffer(iommu);
        return nelems;
 }

@@ -2328,47 +2596,28 @@ int __init intel_iommu_init(void)
        return 0;
 }

-void intel_iommu_domain_exit(struct dmar_domain *domain)
+/* domain id for virtual machine, it won't be used to set in context */
+static unsigned long vm_domid;
+
+static struct dmar_domain *iommu_alloc_vm_domain(void)
 {
-       u64 end;
+       struct dmar_domain *domain;

-       /* Domain 0 is reserved, so dont process it */
+       domain = alloc_domain_mem();
        if (!domain)
-               return;
-
-       end = DOMAIN_MAX_ADDR(domain->gaw);
-       end = end & (~VTD_PAGE_MASK);
-
-       /* clear ptes */
-       dma_pte_clear_range(domain, 0, end);
+               return NULL;

-       /* free page tables */
-       dma_pte_free_pagetable(domain, 0, end);
+       domain->id = vm_domid++;
+       domain->flags |= DOMAIN_FLAG_VIRTUAL_MACHINE;

-       iommu_free_domain(domain);
-       free_domain_mem(domain);
+       return domain;
 }
-EXPORT_SYMBOL_GPL(intel_iommu_domain_exit);

-struct dmar_domain *intel_iommu_domain_alloc(struct pci_dev *pdev)
+struct dmar_domain *intel_iommu_alloc_domain(void)
 {
-       struct dmar_drhd_unit *drhd;
        struct dmar_domain *domain;
-       struct intel_iommu *iommu;

-       drhd = dmar_find_matched_drhd_unit(pdev);
-       if (!drhd) {
-               printk(KERN_ERR "intel_iommu_domain_alloc: drhd == NULL\n");
-               return NULL;
-       }
-
-       iommu = drhd->iommu;
-       if (!iommu) {
-               printk(KERN_ERR
-                       "intel_iommu_domain_alloc: iommu == NULL\n");
-               return NULL;
-       }
-       domain = iommu_alloc_domain(iommu);
+       domain = iommu_alloc_vm_domain();
        if (!domain) {
                printk(KERN_ERR
                        "intel_iommu_domain_alloc: domain == NULL\n");
@@ -2377,44 +2626,122 @@ struct dmar_domain *intel_iommu_domain_alloc(struct pci_dev *pdev)
        if (domain_init(domain, DEFAULT_DOMAIN_ADDRESS_WIDTH)) {
                printk(KERN_ERR
                        "intel_iommu_domain_alloc: domain_init() failed\n");
-               intel_iommu_domain_exit(domain);
+               domain_exit(domain);
                return NULL;
        }
+
        return domain;
 }
-EXPORT_SYMBOL_GPL(intel_iommu_domain_alloc);
+EXPORT_SYMBOL_GPL(intel_iommu_alloc_domain);

-int intel_iommu_context_mapping(
-       struct dmar_domain *domain, struct pci_dev *pdev)
+void intel_iommu_free_domain(struct dmar_domain *domain)
 {
-       int rc;
-       rc = domain_context_mapping(domain, pdev);
-       return rc;
+       domain_exit(domain);
+}
+EXPORT_SYMBOL_GPL(intel_iommu_free_domain);
+
+static int domain_add_dev_info(struct dmar_domain *domain,
+                              struct pci_dev *pdev)
+{
+       struct device_domain_info *info;
+       unsigned long flags;
+
+       info = alloc_devinfo_mem();
+       if (!info)
+               return -ENOMEM;
+
+       info->bus = pdev->bus->number;
+       info->devfn = pdev->devfn;
+       info->dev = pdev;
+       info->domain = domain;
+
+       spin_lock_irqsave(&device_domain_lock, flags);
+       list_add(&info->link, &domain->devices);
+       list_add(&info->global, &device_domain_list);
+       pdev->dev.archdata.iommu = info;
+       spin_unlock_irqrestore(&device_domain_lock, flags);
+
+       return 0;
+}
+
+static void domain_remove_dev_info(struct dmar_domain *domain,
+                                  struct pci_dev *pdev)
+{
+       struct device_domain_info *info;
+       unsigned long flags;
+
+       spin_lock_irqsave(&device_domain_lock, flags);
+       while (!list_empty(&domain->devices)) {
+               info = list_entry(domain->devices.next,
+                       struct device_domain_info, link);
+               if (info->bus == pdev->bus->number &&
+                   info->devfn == pdev->devfn) {
+                       list_del(&info->link);
+                       list_del(&info->global);
+                       info->dev->dev.archdata.iommu = NULL;
+                       spin_unlock_irqrestore(&device_domain_lock, flags);
+
+                       detach_domain_for_dev(info->domain,
+                                             info->bus, info->devfn);
+                       free_devinfo_mem(info);
+
+                       spin_lock_irqsave(&device_domain_lock, flags);
+                       break;
+               }
+       }
+       spin_unlock_irqrestore(&device_domain_lock, flags);
+}
+
+int intel_iommu_assign_device(struct dmar_domain *domain,
+                             struct pci_dev *pdev)
+{
+       int ret;
+
+       if (domain_context_mapped(pdev)) {
+               struct dmar_domain *old_domain;
+
+               old_domain = find_domain(pdev);
+               if (old_domain)
+                       domain_remove_dev_info(old_domain, pdev);
+       }
+
+       ret = domain_context_mapping(domain, pdev);
+       if (ret)
+               return ret;
+
+       ret = domain_add_dev_info(domain, pdev);
+       return ret;
 }
-EXPORT_SYMBOL_GPL(intel_iommu_context_mapping);
+EXPORT_SYMBOL_GPL(intel_iommu_assign_device);

-int intel_iommu_page_mapping(
-       struct dmar_domain *domain, dma_addr_t iova,
-       u64 hpa, size_t size, int prot)
+
+void intel_iommu_deassign_device(struct dmar_domain *domain,
+                                struct pci_dev *pdev)
 {
-       int rc;
-       rc = domain_page_mapping(domain, iova, hpa, size, prot);
-       return rc;
+       domain_remove_dev_info(domain, pdev);
 }
-EXPORT_SYMBOL_GPL(intel_iommu_page_mapping);
+EXPORT_SYMBOL_GPL(intel_iommu_deassign_device);

-void intel_iommu_detach_dev(struct dmar_domain *domain, u8 bus, u8 devfn)
+int intel_iommu_map_pages(struct dmar_domain *domain, dma_addr_t iova,
+                         u64 hpa, size_t size, int prot)
 {
-       detach_domain_for_dev(domain, bus, devfn);
+       int ret;
+       ret = domain_page_mapping(domain, iova, hpa, size, prot);
+       return ret;
 }
-EXPORT_SYMBOL_GPL(intel_iommu_detach_dev);
+EXPORT_SYMBOL_GPL(intel_iommu_map_pages);

-struct dmar_domain *
-intel_iommu_find_domain(struct pci_dev *pdev)
+void intel_iommu_unmap_pages(struct dmar_domain *domain,
+                            dma_addr_t iova, size_t size)
 {
-       return find_domain(pdev);
+       dma_addr_t base;
+
+       /* The address might not be aligned */
+       base = iova & PAGE_MASK;
+       size = PAGE_ALIGN(size);
+       dma_pte_clear_range(domain, base, base + size);
 }
-EXPORT_SYMBOL_GPL(intel_iommu_find_domain);
+EXPORT_SYMBOL_GPL(intel_iommu_unmap_pages);

 int intel_iommu_found(void)
 {
diff --git a/include/linux/dma_remapping.h b/include/linux/dma_remapping.h
index 952df39..28ef89d 100644
--- a/include/linux/dma_remapping.h
+++ b/include/linux/dma_remapping.h
@@ -111,11 +111,21 @@ struct dma_pte {
                (p).val |= ((addr) & VTD_PAGE_MASK); } while (0)
 #define dma_pte_present(p) (((p).val & 3) != 0)

+/* domain flags, one domain owns one device by default */
+
+/* devices under the same p2p bridge are owned in one domain */
+#define DOMAIN_FLAG_P2P_MULTIPLE_DEVICES       (1 << 0)
+
+/* domain represents a virtual domain, more than one devices
+ * may be owned in one domain, e.g. kvm guest.
+ */
+#define DOMAIN_FLAG_VIRTUAL_MACHINE            (1 << 1)
+
 struct intel_iommu;

 struct dmar_domain {
        int     id;                     /* domain id */
-       struct intel_iommu *iommu;      /* back pointer to owning iommu */
+       unsigned long iommu_bmp;        /* bitmap of iommus this domain uses*/

        struct list_head devices;       /* all devices' list */
        struct iova_domain iovad;       /* iova's that belong to this domain */
@@ -127,7 +137,6 @@ struct dmar_domain {
        /* adjusted guest address width, 0 is level 2 30-bit */
        int             agaw;

-#define DOMAIN_FLAG_MULTIPLE_DEVICES 1
        int             flags;
 };

diff --git a/include/linux/intel-iommu.h b/include/linux/intel-iommu.h
index 3d017cf..c2f37b8 100644
--- a/include/linux/intel-iommu.h
+++ b/include/linux/intel-iommu.h
@@ -219,6 +219,8 @@ do {                                                                        \
        }                                                               \
 } while (0)

+#define DEFAULT_DOMAIN_ADDRESS_WIDTH 48
+
 #define QI_LENGTH      256     /* queue length */

 enum {
@@ -299,6 +301,7 @@ struct intel_iommu {
        struct dmar_domain **domains; /* ptr to domains */
        spinlock_t      lock; /* protect context, domain ids */
        struct root_entry *root_entry; /* virtual address */
+       int             agaw;

        unsigned int irq;
        unsigned char name[7];    /* Device Name */
@@ -334,14 +337,16 @@ extern int qi_flush_iotlb(struct intel_iommu *iommu, u16 did, u64 addr,

 extern void qi_submit_sync(struct qi_desc *desc, struct intel_iommu *iommu);

-void intel_iommu_domain_exit(struct dmar_domain *domain);
-struct dmar_domain *intel_iommu_domain_alloc(struct pci_dev *pdev);
-int intel_iommu_context_mapping(struct dmar_domain *domain,
-                               struct pci_dev *pdev);
-int intel_iommu_page_mapping(struct dmar_domain *domain, dma_addr_t iova,
-                            u64 hpa, size_t size, int prot);
-void intel_iommu_detach_dev(struct dmar_domain *domain, u8 bus, u8 devfn);
-struct dmar_domain *intel_iommu_find_domain(struct pci_dev *pdev);
+struct dmar_domain *intel_iommu_alloc_domain(void);
+void intel_iommu_free_domain(struct dmar_domain *domain);
+int intel_iommu_assign_device(struct dmar_domain *domain,
+                             struct pci_dev *pdev);
+void intel_iommu_deassign_device(struct dmar_domain *domain,
+                                struct pci_dev *pdev);
+int intel_iommu_map_pages(struct dmar_domain *domain, dma_addr_t iova,
+                         u64 hpa, size_t size, int prot);
+void intel_iommu_unmap_pages(struct dmar_domain *domain,
+                            dma_addr_t iova, size_t size);
 u64 intel_iommu_iova_to_pfn(struct dmar_domain *domain, u64 iova);

 #ifdef CONFIG_DMAR
--
1.5.1


[-- Attachment #2: 0001-VT-d-multiple-device-assignment-for-KVM.patch --]
[-- Type: application/octet-stream, Size: 34464 bytes --]

From 8202dd6dbe68fe2cde052a14071af7109e7b4b2b Mon Sep 17 00:00:00 2001
From: Weidong Han <weidong.han@intel.com>
Date: Wed, 26 Nov 2008 09:22:05 +0800
Subject: [PATCH] VT-d multiple device assignment for KVM


Signed-off-by: Weidong Han <weidong.han@intel.com>
---
 drivers/pci/dmar.c            |   15 +
 drivers/pci/intel-iommu.c     |  601 +++++++++++++++++++++++++++++++----------
 include/linux/dma_remapping.h |   13 +-
 include/linux/intel-iommu.h   |   21 +-
 4 files changed, 503 insertions(+), 147 deletions(-)

diff --git a/drivers/pci/dmar.c b/drivers/pci/dmar.c
index 691b3ad..d6bdced 100644
--- a/drivers/pci/dmar.c
+++ b/drivers/pci/dmar.c
@@ -484,6 +484,7 @@ void __init detect_intel_iommu(void)
 	dmar_tbl = NULL;
 }
 
+extern int width_to_agaw(int width);
 
 int alloc_iommu(struct dmar_drhd_unit *drhd)
 {
@@ -491,6 +492,8 @@ int alloc_iommu(struct dmar_drhd_unit *drhd)
 	int map_size;
 	u32 ver;
 	static int iommu_allocated = 0;
+	unsigned long sagaw;
+	int agaw;
 
 	iommu = kzalloc(sizeof(*iommu), GFP_KERNEL);
 	if (!iommu)
@@ -506,6 +509,18 @@ int alloc_iommu(struct dmar_drhd_unit *drhd)
 	iommu->cap = dmar_readq(iommu->reg + DMAR_CAP_REG);
 	iommu->ecap = dmar_readq(iommu->reg + DMAR_ECAP_REG);
 
+	/* set agaw, "SAGAW" may be different across iommus */
+	sagaw = cap_sagaw(iommu->cap);
+	for (agaw = width_to_agaw(DEFAULT_DOMAIN_ADDRESS_WIDTH);
+	     agaw >= 0; agaw--)
+		if (test_bit(agaw, &sagaw))
+			break;
+	if (agaw < 0) {
+		printk(KERN_ERR "IOMMU: unsupported sagaw %lx\n", sagaw);
+		goto error;
+	}
+	iommu->agaw = agaw;
+
 	/* the registers might be more than one page */
 	map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap),
 		cap_max_fault_reg_offset(iommu->cap));
diff --git a/drivers/pci/intel-iommu.c b/drivers/pci/intel-iommu.c
index 5c8baa4..3576ebe 100644
--- a/drivers/pci/intel-iommu.c
+++ b/drivers/pci/intel-iommu.c
@@ -50,8 +50,6 @@
 #define IOAPIC_RANGE_END	(0xfeefffff)
 #define IOVA_START_ADDR		(0x1000)
 
-#define DEFAULT_DOMAIN_ADDRESS_WIDTH 48
-
 #define DOMAIN_MAX_ADDR(gaw) ((((u64)1) << gaw) - 1)
 
 
@@ -64,6 +62,7 @@ struct deferred_flush_tables {
 	int next;
 	struct iova *iova[HIGH_WATER_MARK];
 	struct dmar_domain *domain[HIGH_WATER_MARK];
+	struct intel_iommu *iommu;
 };
 
 static struct deferred_flush_tables *deferred_flush;
@@ -77,7 +76,7 @@ static LIST_HEAD(unmaps_to_do);
 static int timer_on;
 static long list_size;
 
-static void domain_remove_dev_info(struct dmar_domain *domain);
+static void domain_remove_all_dev_info(struct dmar_domain *domain);
 
 int dmar_disabled;
 static int __initdata dmar_map_gfx = 1;
@@ -184,6 +183,72 @@ void free_iova_mem(struct iova *iova)
 	kmem_cache_free(iommu_iova_cache, iova);
 }
 
+/* in native case, each domain is related to only one iommu */
+static struct intel_iommu *domain_get_only_iommu(struct dmar_domain *domain)
+{
+	struct dmar_drhd_unit *drhd;
+
+	if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE)
+		return NULL;
+
+	for_each_drhd_unit(drhd) {
+		if (drhd->ignored)
+			continue;
+		if (test_bit(drhd->iommu->seq_id, &domain->iommu_bmp))
+			return drhd->iommu;
+	}
+
+	return NULL;
+}
+
+static void domain_flush_cache(struct dmar_domain *domain,
+			       void *addr, int size)
+{
+	struct intel_iommu *iommu;
+
+	if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE) {
+		struct dmar_drhd_unit *drhd;
+
+		for_each_drhd_unit(drhd) {
+			if (drhd->ignored)
+				continue;
+			iommu = drhd->iommu;
+
+			if (!test_bit(iommu->seq_id, &domain->iommu_bmp))
+				continue;
+
+			if (!ecap_coherent(iommu->ecap))
+                		clflush_cache_range(addr, size);
+		}
+	}
+	else {
+		iommu = domain_get_only_iommu(domain);
+		if (iommu && !ecap_coherent(iommu->ecap))
+                	clflush_cache_range(addr, size);
+	}
+}
+
+static struct intel_iommu *iommu_find_matched_iommu(u8 bus, u8 devfn)
+{
+	struct dmar_drhd_unit *drhd = NULL;
+	int i;
+
+	for_each_drhd_unit(drhd) {
+		if (drhd->ignored)
+			continue;
+
+		for (i = 0; i < drhd->devices_cnt; i++)
+			if (drhd->devices[i]->bus->number == bus &&
+			    drhd->devices[i]->devfn == devfn)
+				return drhd->iommu;
+
+		if (drhd->include_all)
+			return drhd->iommu;
+	}
+
+	return NULL;
+}
+
 /* Gets context entry for a given bus and devfn */
 static struct context_entry * device_to_context_entry(struct intel_iommu *iommu,
 		u8 bus, u8 devfn)
@@ -287,7 +352,7 @@ static inline int agaw_to_width(int agaw)
 
 }
 
-static inline int width_to_agaw(int width)
+int width_to_agaw(int width)
 {
 	return (width - 30) / LEVEL_STRIDE;
 }
@@ -347,8 +412,7 @@ static struct dma_pte * addr_to_dma_pte(struct dmar_domain *domain, u64 addr)
 					flags);
 				return NULL;
 			}
-			__iommu_flush_cache(domain->iommu, tmp_page,
-					PAGE_SIZE);
+			domain_flush_cache(domain, tmp_page, PAGE_SIZE);
 			dma_set_pte_addr(*pte, virt_to_phys(tmp_page));
 			/*
 			 * high level table always sets r/w, last level page
@@ -356,7 +420,7 @@ static struct dma_pte * addr_to_dma_pte(struct dmar_domain *domain, u64 addr)
 			 */
 			dma_set_pte_readable(*pte);
 			dma_set_pte_writable(*pte);
-			__iommu_flush_cache(domain->iommu, pte, sizeof(*pte));
+			domain_flush_cache(domain, pte, sizeof(*pte));
 		}
 		parent = phys_to_virt(dma_pte_addr(*pte));
 		level--;
@@ -399,7 +463,7 @@ static void dma_pte_clear_one(struct dmar_domain *domain, u64 addr)
 
 	if (pte) {
 		dma_clear_pte(*pte);
-		__iommu_flush_cache(domain->iommu, pte, sizeof(*pte));
+		domain_flush_cache(domain, pte, sizeof(*pte));
 	}
 }
 
@@ -447,8 +511,7 @@ static void dma_pte_free_pagetable(struct dmar_domain *domain,
 				free_pgtable_page(
 					phys_to_virt(dma_pte_addr(*pte)));
 				dma_clear_pte(*pte);
-				__iommu_flush_cache(domain->iommu,
-						pte, sizeof(*pte));
+				domain_flush_cache(domain, pte, sizeof(*pte));
 			}
 			tmp += level_size(level);
 		}
@@ -951,6 +1014,37 @@ static int iommu_init_domains(struct intel_iommu *iommu)
 
 static void domain_exit(struct dmar_domain *domain);
 
+static int domain_in_other_iommus(struct dmar_domain *domain,
+				  struct intel_iommu *iommu)
+{
+	struct dmar_drhd_unit *drhd;
+	struct intel_iommu *other_iommu;
+	unsigned long ndomains;
+	unsigned long i;
+
+	if (!domain)
+		return 0;
+
+	if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE) {
+		for_each_drhd_unit(drhd) {
+			if (drhd->ignored && drhd->iommu == iommu)
+				continue;
+			other_iommu = drhd->iommu;
+			ndomains = cap_ndoms(other_iommu->cap);
+			i = find_first_bit(other_iommu->domain_ids,
+					   cap_ndoms(other_iommu->cap));
+			for (i = 0; i < ndomains; ) {
+				if (other_iommu->domains[i] == domain)
+					return 1;
+				i = find_next_bit(other_iommu->domain_ids,
+						  ndomains, i+1);
+			}
+		}
+	}
+
+	return 0;
+}
+
 void free_dmar_iommu(struct intel_iommu *iommu)
 {
 	struct dmar_domain *domain;
@@ -960,7 +1054,14 @@ void free_dmar_iommu(struct intel_iommu *iommu)
 	for (; i < cap_ndoms(iommu->cap); ) {
 		domain = iommu->domains[i];
 		clear_bit(i, iommu->domain_ids);
-		domain_exit(domain);
+
+		if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE) {
+			/* domain may be referenced by other iommus */
+			if (domain_in_other_iommus(domain, iommu) == 0)
+				domain_exit(domain);
+		}
+		else
+			domain_exit(domain);
 		i = find_next_bit(iommu->domain_ids,
 			cap_ndoms(iommu->cap), i+1);
 	}
@@ -1006,7 +1107,7 @@ static struct dmar_domain * iommu_alloc_domain(struct intel_iommu *iommu)
 
 	set_bit(num, iommu->domain_ids);
 	domain->id = num;
-	domain->iommu = iommu;
+	set_bit(iommu->seq_id, &domain->iommu_bmp);
 	iommu->domains[num] = domain;
 	spin_unlock_irqrestore(&iommu->lock, flags);
 
@@ -1016,10 +1117,39 @@ static struct dmar_domain * iommu_alloc_domain(struct intel_iommu *iommu)
 static void iommu_free_domain(struct dmar_domain *domain)
 {
 	unsigned long flags;
+	struct dmar_drhd_unit *drhd;
+	struct intel_iommu *iommu;
+	unsigned long i;
+	unsigned long ndomains;
 
-	spin_lock_irqsave(&domain->iommu->lock, flags);
-	clear_bit(domain->id, domain->iommu->domain_ids);
-	spin_unlock_irqrestore(&domain->iommu->lock, flags);
+	if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE) {
+		for_each_drhd_unit(drhd) {
+			if (drhd->ignored)
+				continue;
+			iommu = drhd->iommu;
+
+			ndomains = cap_ndoms(iommu->cap);
+			i = find_first_bit(iommu->domain_ids, ndomains);
+			for (; i < ndomains; ) {
+				if (iommu->domains[i] == domain) {
+					spin_lock_irqsave(&iommu->lock, flags);
+					clear_bit(i, iommu->domain_ids);
+					iommu->domains[i] = NULL;
+					spin_unlock_irqrestore(&iommu->lock,
+							       flags);
+					break;
+				}
+				i = find_next_bit(iommu->domain_ids,
+						  ndomains, i+1);
+			}
+		}
+	}
+	else {
+		iommu = domain_get_only_iommu(domain);
+		spin_lock_irqsave(&iommu->lock, flags);
+		clear_bit(domain->id, iommu->domain_ids);
+		spin_unlock_irqrestore(&iommu->lock, flags);
+	}
 }
 
 static struct iova_domain reserved_iova_list;
@@ -1097,29 +1227,37 @@ static int domain_init(struct dmar_domain *domain, int guest_width)
 
 	domain_reserve_special_ranges(domain);
 
-	/* calculate AGAW */
-	iommu = domain->iommu;
-	if (guest_width > cap_mgaw(iommu->cap))
-		guest_width = cap_mgaw(iommu->cap);
-	domain->gaw = guest_width;
-	adjust_width = guestwidth_to_adjustwidth(guest_width);
-	agaw = width_to_agaw(adjust_width);
-	sagaw = cap_sagaw(iommu->cap);
-	if (!test_bit(agaw, &sagaw)) {
-		/* hardware doesn't support it, choose a bigger one */
-		pr_debug("IOMMU: hardware doesn't support agaw %d\n", agaw);
-		agaw = find_next_bit(&sagaw, 5, agaw);
-		if (agaw >= 5)
-			return -ENODEV;
-	}
-	domain->agaw = agaw;
+	/* set AGAW */
+	if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE) {
+		domain->gaw = DEFAULT_DOMAIN_ADDRESS_WIDTH;
+		domain->agaw = width_to_agaw(DEFAULT_DOMAIN_ADDRESS_WIDTH);
+	}
+	else {
+		iommu = domain_get_only_iommu(domain);
+		if (guest_width > cap_mgaw(iommu->cap))
+			guest_width = cap_mgaw(iommu->cap);
+		domain->gaw = guest_width;
+		adjust_width = guestwidth_to_adjustwidth(guest_width);
+		agaw = width_to_agaw(adjust_width);
+		sagaw = cap_sagaw(iommu->cap);
+		if (!test_bit(agaw, &sagaw)) {
+			/* hardware doesn't support it, choose a bigger one */
+			pr_debug("IOMMU: hardware doesn't support agaw %d\n",
+				 agaw);
+			agaw = find_next_bit(&sagaw, 5, agaw);
+			if (agaw >= 5)
+				return -ENODEV;
+		}
+		domain->agaw = agaw;
+	}
+
 	INIT_LIST_HEAD(&domain->devices);
 
 	/* always allocate the top pgd */
 	domain->pgd = (struct dma_pte *)alloc_pgtable_page();
 	if (!domain->pgd)
 		return -ENOMEM;
-	__iommu_flush_cache(iommu, domain->pgd, PAGE_SIZE);
+	domain_flush_cache(domain, domain->pgd, PAGE_SIZE);
 	return 0;
 }
 
@@ -1131,7 +1269,7 @@ static void domain_exit(struct dmar_domain *domain)
 	if (!domain)
 		return;
 
-	domain_remove_dev_info(domain);
+	domain_remove_all_dev_info(domain);
 	/* destroy iovas */
 	put_iova_domain(&domain->iovad);
 	end = DOMAIN_MAX_ADDR(domain->gaw);
@@ -1148,11 +1286,15 @@ static void domain_exit(struct dmar_domain *domain)
 }
 
 static int domain_context_mapping_one(struct dmar_domain *domain,
-		u8 bus, u8 devfn)
+		struct intel_iommu *iommu, u8 bus, u8 devfn)
 {
 	struct context_entry *context;
-	struct intel_iommu *iommu = domain->iommu;
 	unsigned long flags;
+	struct dma_pte *pgd;
+	unsigned long num;
+	unsigned long ndomains;
+	int id;
+	int agaw;
 
 	pr_debug("Set context mapping for %02x:%02x.%d\n",
 		bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
@@ -1166,22 +1308,67 @@ static int domain_context_mapping_one(struct dmar_domain *domain,
 		return 0;
 	}
 
-	context_set_domain_id(*context, domain->id);
-	context_set_address_width(*context, domain->agaw);
-	context_set_address_root(*context, virt_to_phys(domain->pgd));
+	id = domain->id;
+	pgd = domain->pgd;
+
+	if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE) {
+		int found = 0;
+
+		/* find an available domain id for this device in iommu */
+		ndomains = cap_ndoms(iommu->cap);
+		num = find_first_bit(iommu->domain_ids, ndomains);
+		for (; num < ndomains; ) {
+			if (iommu->domains[num] == domain) {
+				id = num;
+				found = 1;
+				break;
+			}
+			num = find_next_bit(iommu->domain_ids,
+					    cap_ndoms(iommu->cap), num+1);
+		}
+
+		if (found == 0) {
+			num = find_first_zero_bit(iommu->domain_ids, ndomains);
+			if (num >= ndomains) {
+				spin_unlock_irqrestore(&iommu->lock, flags);
+				printk(KERN_ERR "IOMMU: no free domain ids\n");
+				return -EFAULT;
+			}
+
+			set_bit(num, iommu->domain_ids);
+			iommu->domains[num] = domain;
+			id = num;
+		}
+
+		/* Skip top levels of page tables for
+		 * iommu which has less agaw than default.
+		 */
+		for (agaw = domain->agaw; agaw != iommu->agaw; agaw--) {
+			pgd = phys_to_virt(dma_pte_addr(*pgd));
+			if (!dma_pte_present(*pgd)) {
+				spin_unlock_irqrestore(&iommu->lock, flags);
+				return -ENOMEM;
+			}
+		}
+	}
+
+	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);
 	context_set_fault_enable(*context);
 	context_set_present(*context);
 	__iommu_flush_cache(iommu, context, sizeof(*context));
 
 	/* it's a non-present to present mapping */
-	if (iommu->flush.flush_context(iommu, domain->id,
+	if (iommu->flush.flush_context(iommu, id,
 		(((u16)bus) << 8) | devfn, DMA_CCMD_MASK_NOBIT,
 		DMA_CCMD_DEVICE_INVL, 1))
 		iommu_flush_write_buffer(iommu);
 	else
 		iommu->flush.flush_iotlb(iommu, 0, 0, 0, DMA_TLB_DSI_FLUSH, 0);
 
+	set_bit(iommu->seq_id, &domain->iommu_bmp);
 	spin_unlock_irqrestore(&iommu->lock, flags);
 	return 0;
 }
@@ -1191,9 +1378,15 @@ domain_context_mapping(struct dmar_domain *domain, struct pci_dev *pdev)
 {
 	int ret;
 	struct pci_dev *tmp, *parent;
+	struct intel_iommu *iommu;
 
-	ret = domain_context_mapping_one(domain, pdev->bus->number,
-		pdev->devfn);
+	iommu = iommu_find_matched_iommu(pdev->bus->number,
+					 pdev->devfn);
+	if (!iommu)
+		return 	-ENODEV;
+
+	ret = domain_context_mapping_one(domain, iommu,
+		pdev->bus->number, pdev->devfn);
 	if (ret)
 		return ret;
 
@@ -1204,27 +1397,32 @@ domain_context_mapping(struct dmar_domain *domain, struct pci_dev *pdev)
 	/* Secondary interface's bus number and devfn 0 */
 	parent = pdev->bus->self;
 	while (parent != tmp) {
-		ret = domain_context_mapping_one(domain, parent->bus->number,
-			parent->devfn);
+		ret = domain_context_mapping_one(domain, iommu,
+			parent->bus->number, parent->devfn);
 		if (ret)
 			return ret;
 		parent = parent->bus->self;
 	}
 	if (tmp->is_pcie) /* this is a PCIE-to-PCI bridge */
-		return domain_context_mapping_one(domain,
+		return domain_context_mapping_one(domain, iommu,
 			tmp->subordinate->number, 0);
 	else /* this is a legacy PCI bridge */
-		return domain_context_mapping_one(domain,
+		return domain_context_mapping_one(domain, iommu,
 			tmp->bus->number, tmp->devfn);
 }
 
-static int domain_context_mapped(struct dmar_domain *domain,
-	struct pci_dev *pdev)
+static int domain_context_mapped(struct pci_dev *pdev)
 {
 	int ret;
 	struct pci_dev *tmp, *parent;
+	struct intel_iommu *iommu;
+
+	iommu = iommu_find_matched_iommu(pdev->bus->number,
+					 pdev->devfn);
+	if (!iommu)
+		return 0;
 
-	ret = device_context_mapped(domain->iommu,
+	ret = device_context_mapped(iommu,
 		pdev->bus->number, pdev->devfn);
 	if (!ret)
 		return ret;
@@ -1235,17 +1433,17 @@ static int domain_context_mapped(struct dmar_domain *domain,
 	/* Secondary interface's bus number and devfn 0 */
 	parent = pdev->bus->self;
 	while (parent != tmp) {
-		ret = device_context_mapped(domain->iommu, parent->bus->number,
+		ret = device_context_mapped(iommu, parent->bus->number,
 			parent->devfn);
 		if (!ret)
 			return ret;
 		parent = parent->bus->self;
 	}
 	if (tmp->is_pcie)
-		return device_context_mapped(domain->iommu,
+		return device_context_mapped(iommu,
 			tmp->subordinate->number, 0);
 	else
-		return device_context_mapped(domain->iommu,
+		return device_context_mapped(iommu,
 			tmp->bus->number, tmp->devfn);
 }
 
@@ -1276,23 +1474,60 @@ domain_page_mapping(struct dmar_domain *domain, dma_addr_t iova,
 		BUG_ON(dma_pte_addr(*pte));
 		dma_set_pte_addr(*pte, start_pfn << VTD_PAGE_SHIFT);
 		dma_set_pte_prot(*pte, prot);
-		__iommu_flush_cache(domain->iommu, pte, sizeof(*pte));
+		domain_flush_cache(domain, pte, sizeof(*pte));
 		start_pfn++;
 		index++;
 	}
 	return 0;
 }
 
-static void detach_domain_for_dev(struct dmar_domain *domain, u8 bus, u8 devfn)
+static void detach_domain_for_dev(struct dmar_domain *domain,
+				  u8 bus, u8 devfn)
 {
-	clear_context_table(domain->iommu, bus, devfn);
-	domain->iommu->flush.flush_context(domain->iommu, 0, 0, 0,
-					   DMA_CCMD_GLOBAL_INVL, 0);
-	domain->iommu->flush.flush_iotlb(domain->iommu, 0, 0, 0,
-					 DMA_TLB_GLOBAL_FLUSH, 0);
+	struct device_domain_info *info;
+	unsigned long flags;
+	struct intel_iommu *iommu;
+
+	iommu = iommu_find_matched_iommu(bus, devfn);
+	if (!iommu)
+		return;
+
+	clear_context_table(iommu, bus, devfn);
+	iommu->flush.flush_context(iommu, 0, 0, 0,
+				   DMA_CCMD_GLOBAL_INVL, 0);
+	iommu->flush.flush_iotlb(iommu, 0, 0, 0,
+				 DMA_TLB_GLOBAL_FLUSH, 0);
+
+	/* clear this iommu in iommu_bmp if there is no
+	 * its device owned by this domain
+	 */
+	if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE) {
+		int found = 0;
+
+		while (!list_empty(&domain->devices)) {
+			info = list_entry(domain->devices.next,
+				struct device_domain_info, link);
+
+			if (iommu_find_matched_iommu(info->bus,
+					info->devfn) == iommu) {
+				found = 1;
+				break;
+			}
+		}
+		if (found == 0) {
+			spin_lock_irqsave(&iommu->lock, flags);
+			clear_bit(iommu->seq_id, &domain->iommu_bmp);
+			spin_unlock_irqrestore(&iommu->lock, flags);
+		}
+	}
+	else {
+		spin_lock_irqsave(&iommu->lock, flags);
+		clear_bit(iommu->seq_id, &domain->iommu_bmp);
+		spin_unlock_irqrestore(&iommu->lock, flags);
+	}
 }
 
-static void domain_remove_dev_info(struct dmar_domain *domain)
+static void domain_remove_all_dev_info(struct dmar_domain *domain)
 {
 	struct device_domain_info *info;
 	unsigned long flags;
@@ -1336,7 +1571,6 @@ static struct dmar_domain *get_domain_for_dev(struct pci_dev *pdev, int gaw)
 {
 	struct dmar_domain *domain, *found = NULL;
 	struct intel_iommu *iommu;
-	struct dmar_drhd_unit *drhd;
 	struct device_domain_info *info, *tmp;
 	struct pci_dev *dev_tmp;
 	unsigned long flags;
@@ -1371,13 +1605,10 @@ static struct dmar_domain *get_domain_for_dev(struct pci_dev *pdev, int gaw)
 	}
 
 	/* Allocate new domain for the device */
-	drhd = dmar_find_matched_drhd_unit(pdev);
-	if (!drhd) {
-		printk(KERN_ERR "IOMMU: can't find DMAR for device %s\n",
-			pci_name(pdev));
-		return NULL;
-	}
-	iommu = drhd->iommu;
+	iommu = iommu_find_matched_iommu(pdev->bus->number,
+					 pdev->devfn);
+	if (!iommu)
+		return 	NULL;
 
 	domain = iommu_alloc_domain(iommu);
 	if (!domain)
@@ -1400,7 +1631,7 @@ static struct dmar_domain *get_domain_for_dev(struct pci_dev *pdev, int gaw)
 		info->dev = NULL;
 		info->domain = domain;
 		/* This domain is shared by devices under p2p bridge */
-		domain->flags |= DOMAIN_FLAG_MULTIPLE_DEVICES;
+		domain->flags |= DOMAIN_FLAG_P2P_MULTIPLE_DEVICES;
 
 		/* pcie-to-pci bridge already has a domain, uses it */
 		found = NULL;
@@ -1805,7 +2036,7 @@ get_valid_domain_for_dev(struct pci_dev *pdev)
 	}
 
 	/* make sure context mapping is ok */
-	if (unlikely(!domain_context_mapped(domain, pdev))) {
+	if (unlikely(!domain_context_mapped(pdev))) {
 		ret = domain_context_mapping(domain, pdev);
 		if (ret) {
 			printk(KERN_ERR
@@ -1823,6 +2054,7 @@ static dma_addr_t __intel_map_single(struct device *hwdev, phys_addr_t paddr,
 {
 	struct pci_dev *pdev = to_pci_dev(hwdev);
 	struct dmar_domain *domain;
+	struct intel_iommu *iommu;
 	phys_addr_t start_paddr;
 	struct iova *iova;
 	int prot = 0;
@@ -1836,6 +2068,14 @@ static dma_addr_t __intel_map_single(struct device *hwdev, phys_addr_t paddr,
 	if (!domain)
 		return 0;
 
+	if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE)
+		return 0;
+
+	iommu = iommu_find_matched_iommu(pdev->bus->number,
+					 pdev->devfn);
+	if (!iommu)
+		return 0;
+
 	size = aligned_size((u64)paddr, size);
 
 	iova = __intel_alloc_iova(hwdev, domain, size, pdev->dma_mask);
@@ -1849,7 +2089,7 @@ static dma_addr_t __intel_map_single(struct device *hwdev, phys_addr_t paddr,
 	 * mappings..
 	 */
 	if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL || \
-			!cap_zlr(domain->iommu->cap))
+			!cap_zlr(iommu->cap))
 		prot |= DMA_PTE_READ;
 	if (dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL)
 		prot |= DMA_PTE_WRITE;
@@ -1865,10 +2105,10 @@ static dma_addr_t __intel_map_single(struct device *hwdev, phys_addr_t paddr,
 		goto error;
 
 	/* it's a non-present to present mapping */
-	ret = iommu_flush_iotlb_psi(domain->iommu, domain->id,
+	ret = iommu_flush_iotlb_psi(iommu, domain->id,
 			start_paddr, size >> VTD_PAGE_SHIFT, 1);
 	if (ret)
-		iommu_flush_write_buffer(domain->iommu);
+		iommu_flush_write_buffer(iommu);
 
 	return start_paddr + ((u64)paddr & (~PAGE_MASK));
 
@@ -1896,8 +2136,7 @@ static void flush_unmaps(void)
 	/* just flush them all */
 	for (i = 0; i < g_num_of_iommus; i++) {
 		if (deferred_flush[i].next) {
-			struct intel_iommu *iommu =
-				deferred_flush[i].domain[0]->iommu;
+			struct intel_iommu *iommu = deferred_flush[i].iommu;
 
 			iommu->flush.flush_iotlb(iommu, 0, 0, 0,
 						 DMA_TLB_GLOBAL_FLUSH, 0);
@@ -1921,7 +2160,8 @@ static void flush_unmaps_timeout(unsigned long data)
 	spin_unlock_irqrestore(&async_umap_flush_lock, flags);
 }
 
-static void add_unmap(struct dmar_domain *dom, struct iova *iova)
+static void add_unmap(struct dmar_domain *dom,
+		      struct intel_iommu *iommu, struct iova *iova)
 {
 	unsigned long flags;
 	int next, iommu_id;
@@ -1930,11 +2170,12 @@ static void add_unmap(struct dmar_domain *dom, struct iova *iova)
 	if (list_size == HIGH_WATER_MARK)
 		flush_unmaps();
 
-	iommu_id = dom->iommu->seq_id;
+	iommu_id = iommu->seq_id;
 
 	next = deferred_flush[iommu_id].next;
 	deferred_flush[iommu_id].domain[next] = dom;
 	deferred_flush[iommu_id].iova[next] = iova;
+	deferred_flush[iommu_id].iommu = iommu;
 	deferred_flush[iommu_id].next++;
 
 	if (!timer_on) {
@@ -1952,12 +2193,21 @@ void intel_unmap_single(struct device *dev, dma_addr_t dev_addr, size_t size,
 	struct dmar_domain *domain;
 	unsigned long start_addr;
 	struct iova *iova;
+	struct intel_iommu *iommu;
 
 	if (pdev->dev.archdata.iommu == DUMMY_DEVICE_DOMAIN_INFO)
 		return;
 	domain = find_domain(pdev);
 	BUG_ON(!domain);
 
+	if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE)
+		return;
+
+	iommu = iommu_find_matched_iommu(pdev->bus->number,
+					 pdev->devfn);
+	if (!iommu)
+		return;
+
 	iova = find_iova(&domain->iovad, IOVA_PFN(dev_addr));
 	if (!iova)
 		return;
@@ -1973,13 +2223,13 @@ void intel_unmap_single(struct device *dev, dma_addr_t dev_addr, size_t size,
 	/* free page tables */
 	dma_pte_free_pagetable(domain, start_addr, start_addr + size);
 	if (intel_iommu_strict) {
-		if (iommu_flush_iotlb_psi(domain->iommu,
+		if (iommu_flush_iotlb_psi(iommu,
 			domain->id, start_addr, size >> VTD_PAGE_SHIFT, 0))
-			iommu_flush_write_buffer(domain->iommu);
+			iommu_flush_write_buffer(iommu);
 		/* free iova */
 		__free_iova(&domain->iovad, iova);
 	} else {
-		add_unmap(domain, iova);
+		add_unmap(domain, iommu, iova);
 		/*
 		 * queue up the release of the unmap to save the 1/6th of the
 		 * cpu used up by the iotlb flush operation...
@@ -2036,12 +2286,21 @@ void intel_unmap_sg(struct device *hwdev, struct scatterlist *sglist,
 	size_t size = 0;
 	void *addr;
 	struct scatterlist *sg;
+	struct intel_iommu *iommu;
 
 	if (pdev->dev.archdata.iommu == DUMMY_DEVICE_DOMAIN_INFO)
 		return;
 
 	domain = find_domain(pdev);
 
+	if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE)
+		return;
+
+	iommu = iommu_find_matched_iommu(pdev->bus->number,
+					 pdev->devfn);
+	if (!iommu)
+		return;
+
 	iova = find_iova(&domain->iovad, IOVA_PFN(sglist[0].dma_address));
 	if (!iova)
 		return;
@@ -2057,9 +2316,9 @@ void intel_unmap_sg(struct device *hwdev, struct scatterlist *sglist,
 	/* free page tables */
 	dma_pte_free_pagetable(domain, start_addr, start_addr + size);
 
-	if (iommu_flush_iotlb_psi(domain->iommu, domain->id, start_addr,
+	if (iommu_flush_iotlb_psi(iommu, domain->id, start_addr,
 			size >> VTD_PAGE_SHIFT, 0))
-		iommu_flush_write_buffer(domain->iommu);
+		iommu_flush_write_buffer(iommu);
 
 	/* free iova */
 	__free_iova(&domain->iovad, iova);
@@ -2093,6 +2352,7 @@ int intel_map_sg(struct device *hwdev, struct scatterlist *sglist, int nelems,
 	int ret;
 	struct scatterlist *sg;
 	unsigned long start_addr;
+	struct intel_iommu *iommu;
 
 	BUG_ON(dir == DMA_NONE);
 	if (pdev->dev.archdata.iommu == DUMMY_DEVICE_DOMAIN_INFO)
@@ -2102,6 +2362,14 @@ int intel_map_sg(struct device *hwdev, struct scatterlist *sglist, int nelems,
 	if (!domain)
 		return 0;
 
+	if (domain->flags & DOMAIN_FLAG_VIRTUAL_MACHINE)
+		return 0;
+
+	iommu = iommu_find_matched_iommu(pdev->bus->number,
+					 pdev->devfn);
+	if (!iommu)
+		return 0;
+
 	for_each_sg(sglist, sg, nelems, i) {
 		addr = SG_ENT_VIRT_ADDRESS(sg);
 		addr = (void *)virt_to_phys(addr);
@@ -2119,7 +2387,7 @@ int intel_map_sg(struct device *hwdev, struct scatterlist *sglist, int nelems,
 	 * mappings..
 	 */
 	if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL || \
-			!cap_zlr(domain->iommu->cap))
+			!cap_zlr(iommu->cap))
 		prot |= DMA_PTE_READ;
 	if (dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL)
 		prot |= DMA_PTE_WRITE;
@@ -2151,9 +2419,9 @@ int intel_map_sg(struct device *hwdev, struct scatterlist *sglist, int nelems,
 	}
 
 	/* it's a non-present to present mapping */
-	if (iommu_flush_iotlb_psi(domain->iommu, domain->id,
+	if (iommu_flush_iotlb_psi(iommu, domain->id,
 			start_addr, offset >> VTD_PAGE_SHIFT, 1))
-		iommu_flush_write_buffer(domain->iommu);
+		iommu_flush_write_buffer(iommu);
 	return nelems;
 }
 
@@ -2328,47 +2596,28 @@ int __init intel_iommu_init(void)
 	return 0;
 }
 
-void intel_iommu_domain_exit(struct dmar_domain *domain)
+/* domain id for virtual machine, it won't be used to set in context */
+static unsigned long vm_domid;
+
+static struct dmar_domain *iommu_alloc_vm_domain(void)
 {
-	u64 end;
+	struct dmar_domain *domain;
 
-	/* Domain 0 is reserved, so dont process it */
+	domain = alloc_domain_mem();
 	if (!domain)
-		return;
-
-	end = DOMAIN_MAX_ADDR(domain->gaw);
-	end = end & (~VTD_PAGE_MASK);
-
-	/* clear ptes */
-	dma_pte_clear_range(domain, 0, end);
+		return NULL;
 
-	/* free page tables */
-	dma_pte_free_pagetable(domain, 0, end);
+	domain->id = vm_domid++;
+	domain->flags |= DOMAIN_FLAG_VIRTUAL_MACHINE;
 
-	iommu_free_domain(domain);
-	free_domain_mem(domain);
+	return domain;
 }
-EXPORT_SYMBOL_GPL(intel_iommu_domain_exit);
 
-struct dmar_domain *intel_iommu_domain_alloc(struct pci_dev *pdev)
+struct dmar_domain *intel_iommu_alloc_domain(void)
 {
-	struct dmar_drhd_unit *drhd;
 	struct dmar_domain *domain;
-	struct intel_iommu *iommu;
 
-	drhd = dmar_find_matched_drhd_unit(pdev);
-	if (!drhd) {
-		printk(KERN_ERR "intel_iommu_domain_alloc: drhd == NULL\n");
-		return NULL;
-	}
-
-	iommu = drhd->iommu;
-	if (!iommu) {
-		printk(KERN_ERR
-			"intel_iommu_domain_alloc: iommu == NULL\n");
-		return NULL;
-	}
-	domain = iommu_alloc_domain(iommu);
+	domain = iommu_alloc_vm_domain();
 	if (!domain) {
 		printk(KERN_ERR
 			"intel_iommu_domain_alloc: domain == NULL\n");
@@ -2377,44 +2626,122 @@ struct dmar_domain *intel_iommu_domain_alloc(struct pci_dev *pdev)
 	if (domain_init(domain, DEFAULT_DOMAIN_ADDRESS_WIDTH)) {
 		printk(KERN_ERR
 			"intel_iommu_domain_alloc: domain_init() failed\n");
-		intel_iommu_domain_exit(domain);
+		domain_exit(domain);
 		return NULL;
 	}
+
 	return domain;
 }
-EXPORT_SYMBOL_GPL(intel_iommu_domain_alloc);
+EXPORT_SYMBOL_GPL(intel_iommu_alloc_domain);
 
-int intel_iommu_context_mapping(
-	struct dmar_domain *domain, struct pci_dev *pdev)
+void intel_iommu_free_domain(struct dmar_domain *domain)
 {
-	int rc;
-	rc = domain_context_mapping(domain, pdev);
-	return rc;
+	domain_exit(domain);
+}
+EXPORT_SYMBOL_GPL(intel_iommu_free_domain);
+
+static int domain_add_dev_info(struct dmar_domain *domain,
+			       struct pci_dev *pdev)
+{
+	struct device_domain_info *info;
+	unsigned long flags;
+
+	info = alloc_devinfo_mem();
+	if (!info)
+		return -ENOMEM;
+
+	info->bus = pdev->bus->number;
+	info->devfn = pdev->devfn;
+	info->dev = pdev;
+	info->domain = domain;
+
+	spin_lock_irqsave(&device_domain_lock, flags);
+	list_add(&info->link, &domain->devices);
+	list_add(&info->global, &device_domain_list);
+	pdev->dev.archdata.iommu = info;
+	spin_unlock_irqrestore(&device_domain_lock, flags);
+	
+	return 0;
+}
+
+static void domain_remove_dev_info(struct dmar_domain *domain,
+				   struct pci_dev *pdev)
+{
+	struct device_domain_info *info;
+	unsigned long flags;
+
+	spin_lock_irqsave(&device_domain_lock, flags);
+	while (!list_empty(&domain->devices)) {
+		info = list_entry(domain->devices.next,
+			struct device_domain_info, link);
+		if (info->bus == pdev->bus->number &&
+		    info->devfn == pdev->devfn) {
+			list_del(&info->link);
+			list_del(&info->global);
+			info->dev->dev.archdata.iommu = NULL;
+			spin_unlock_irqrestore(&device_domain_lock, flags);
+
+			detach_domain_for_dev(info->domain,
+					      info->bus, info->devfn);
+			free_devinfo_mem(info);
+
+			spin_lock_irqsave(&device_domain_lock, flags);
+			break;
+		}
+	}
+	spin_unlock_irqrestore(&device_domain_lock, flags);
+}
+
+int intel_iommu_assign_device(struct dmar_domain *domain,
+			      struct pci_dev *pdev)
+{
+	int ret;
+
+	if (domain_context_mapped(pdev)) {
+		struct dmar_domain *old_domain;
+
+		old_domain = find_domain(pdev);
+		if (old_domain)
+			domain_remove_dev_info(old_domain, pdev);
+	}
+
+	ret = domain_context_mapping(domain, pdev);
+	if (ret)
+		return ret;
+
+	ret = domain_add_dev_info(domain, pdev);
+	return ret;
 }
-EXPORT_SYMBOL_GPL(intel_iommu_context_mapping);
+EXPORT_SYMBOL_GPL(intel_iommu_assign_device);
 
-int intel_iommu_page_mapping(
-	struct dmar_domain *domain, dma_addr_t iova,
-	u64 hpa, size_t size, int prot)
+
+void intel_iommu_deassign_device(struct dmar_domain *domain,
+				 struct pci_dev *pdev)
 {
-	int rc;
-	rc = domain_page_mapping(domain, iova, hpa, size, prot);
-	return rc;
+	domain_remove_dev_info(domain, pdev);
 }
-EXPORT_SYMBOL_GPL(intel_iommu_page_mapping);
+EXPORT_SYMBOL_GPL(intel_iommu_deassign_device);
 
-void intel_iommu_detach_dev(struct dmar_domain *domain, u8 bus, u8 devfn)
+int intel_iommu_map_pages(struct dmar_domain *domain, dma_addr_t iova,
+			  u64 hpa, size_t size, int prot)
 {
-	detach_domain_for_dev(domain, bus, devfn);
+	int ret;
+	ret = domain_page_mapping(domain, iova, hpa, size, prot);
+	return ret;
 }
-EXPORT_SYMBOL_GPL(intel_iommu_detach_dev);
+EXPORT_SYMBOL_GPL(intel_iommu_map_pages);
 
-struct dmar_domain *
-intel_iommu_find_domain(struct pci_dev *pdev)
+void intel_iommu_unmap_pages(struct dmar_domain *domain,
+			     dma_addr_t iova, size_t size)
 {
-	return find_domain(pdev);
+	dma_addr_t base;
+	
+	/* The address might not be aligned */
+	base = iova & PAGE_MASK;
+	size = PAGE_ALIGN(size);
+	dma_pte_clear_range(domain, base, base + size);
 }
-EXPORT_SYMBOL_GPL(intel_iommu_find_domain);
+EXPORT_SYMBOL_GPL(intel_iommu_unmap_pages);
 
 int intel_iommu_found(void)
 {
diff --git a/include/linux/dma_remapping.h b/include/linux/dma_remapping.h
index 952df39..28ef89d 100644
--- a/include/linux/dma_remapping.h
+++ b/include/linux/dma_remapping.h
@@ -111,11 +111,21 @@ struct dma_pte {
 		(p).val |= ((addr) & VTD_PAGE_MASK); } while (0)
 #define dma_pte_present(p) (((p).val & 3) != 0)
 
+/* domain flags, one domain owns one device by default */
+
+/* devices under the same p2p bridge are owned in one domain */ 
+#define DOMAIN_FLAG_P2P_MULTIPLE_DEVICES	(1 << 0)
+
+/* domain represents a virtual domain, more than one devices
+ * may be owned in one domain, e.g. kvm guest.
+ */
+#define DOMAIN_FLAG_VIRTUAL_MACHINE		(1 << 1)
+
 struct intel_iommu;
 
 struct dmar_domain {
 	int	id;			/* domain id */
-	struct intel_iommu *iommu;	/* back pointer to owning iommu */
+	unsigned long iommu_bmp;	/* bitmap of iommus this domain uses*/
 
 	struct list_head devices; 	/* all devices' list */
 	struct iova_domain iovad;	/* iova's that belong to this domain */
@@ -127,7 +137,6 @@ struct dmar_domain {
 	/* adjusted guest address width, 0 is level 2 30-bit */
 	int		agaw;
 
-#define DOMAIN_FLAG_MULTIPLE_DEVICES 1
 	int		flags;
 };
 
diff --git a/include/linux/intel-iommu.h b/include/linux/intel-iommu.h
index 3d017cf..c2f37b8 100644
--- a/include/linux/intel-iommu.h
+++ b/include/linux/intel-iommu.h
@@ -219,6 +219,8 @@ do {									\
 	}								\
 } while (0)
 
+#define DEFAULT_DOMAIN_ADDRESS_WIDTH 48
+
 #define QI_LENGTH	256	/* queue length */
 
 enum {
@@ -299,6 +301,7 @@ struct intel_iommu {
 	struct dmar_domain **domains; /* ptr to domains */
 	spinlock_t	lock; /* protect context, domain ids */
 	struct root_entry *root_entry; /* virtual address */
+	int		agaw;
 
 	unsigned int irq;
 	unsigned char name[7];    /* Device Name */
@@ -334,14 +337,16 @@ extern int qi_flush_iotlb(struct intel_iommu *iommu, u16 did, u64 addr,
 
 extern void qi_submit_sync(struct qi_desc *desc, struct intel_iommu *iommu);
 
-void intel_iommu_domain_exit(struct dmar_domain *domain);
-struct dmar_domain *intel_iommu_domain_alloc(struct pci_dev *pdev);
-int intel_iommu_context_mapping(struct dmar_domain *domain,
-				struct pci_dev *pdev);
-int intel_iommu_page_mapping(struct dmar_domain *domain, dma_addr_t iova,
-			     u64 hpa, size_t size, int prot);
-void intel_iommu_detach_dev(struct dmar_domain *domain, u8 bus, u8 devfn);
-struct dmar_domain *intel_iommu_find_domain(struct pci_dev *pdev);
+struct dmar_domain *intel_iommu_alloc_domain(void);
+void intel_iommu_free_domain(struct dmar_domain *domain);
+int intel_iommu_assign_device(struct dmar_domain *domain,
+			      struct pci_dev *pdev);
+void intel_iommu_deassign_device(struct dmar_domain *domain,
+				 struct pci_dev *pdev);
+int intel_iommu_map_pages(struct dmar_domain *domain, dma_addr_t iova,
+			  u64 hpa, size_t size, int prot);
+void intel_iommu_unmap_pages(struct dmar_domain *domain,
+			     dma_addr_t iova, size_t size);
 u64 intel_iommu_iova_to_pfn(struct dmar_domain *domain, u64 iova);
 
 #ifdef CONFIG_DMAR
-- 
1.5.1


^ permalink raw reply related	[flat|nested] 5+ messages in thread

end of thread, other threads:[~2008-11-27  3:38 UTC | newest]

Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2008-11-26  3:18 [PATCH 1/2] VT-d: Support multiple device assignment for KVM Han, Weidong
2008-11-26 10:53 ` Avi Kivity
2008-11-26 14:04   ` Han, Weidong
2008-11-26 19:07     ` Yu, Fenghua
2008-11-27  3:38       ` Han, Weidong

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox