aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/pci/intel-iommu.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/pci/intel-iommu.c')
-rw-r--r--drivers/pci/intel-iommu.c250
1 files changed, 129 insertions, 121 deletions
diff --git a/drivers/pci/intel-iommu.c b/drivers/pci/intel-iommu.c
index 8b51e10b7783..a2692724b68f 100644
--- a/drivers/pci/intel-iommu.c
+++ b/drivers/pci/intel-iommu.c
@@ -18,6 +18,7 @@
18 * Author: Ashok Raj <ashok.raj@intel.com> 18 * Author: Ashok Raj <ashok.raj@intel.com>
19 * Author: Shaohua Li <shaohua.li@intel.com> 19 * Author: Shaohua Li <shaohua.li@intel.com>
20 * Author: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com> 20 * Author: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
21 * Author: Fenghua Yu <fenghua.yu@intel.com>
21 */ 22 */
22 23
23#include <linux/init.h> 24#include <linux/init.h>
@@ -35,11 +36,13 @@
35#include <linux/timer.h> 36#include <linux/timer.h>
36#include <linux/iova.h> 37#include <linux/iova.h>
37#include <linux/intel-iommu.h> 38#include <linux/intel-iommu.h>
38#include <asm/proto.h> /* force_iommu in this header in x86-64*/
39#include <asm/cacheflush.h> 39#include <asm/cacheflush.h>
40#include <asm/iommu.h> 40#include <asm/iommu.h>
41#include "pci.h" 41#include "pci.h"
42 42
43#define ROOT_SIZE VTD_PAGE_SIZE
44#define CONTEXT_SIZE VTD_PAGE_SIZE
45
43#define IS_GFX_DEVICE(pdev) ((pdev->class >> 16) == PCI_BASE_CLASS_DISPLAY) 46#define IS_GFX_DEVICE(pdev) ((pdev->class >> 16) == PCI_BASE_CLASS_DISPLAY)
44#define IS_ISA_DEVICE(pdev) ((pdev->class >> 8) == PCI_CLASS_BRIDGE_ISA) 47#define IS_ISA_DEVICE(pdev) ((pdev->class >> 8) == PCI_CLASS_BRIDGE_ISA)
45 48
@@ -199,7 +202,7 @@ static struct context_entry * device_to_context_entry(struct intel_iommu *iommu,
199 spin_unlock_irqrestore(&iommu->lock, flags); 202 spin_unlock_irqrestore(&iommu->lock, flags);
200 return NULL; 203 return NULL;
201 } 204 }
202 __iommu_flush_cache(iommu, (void *)context, PAGE_SIZE_4K); 205 __iommu_flush_cache(iommu, (void *)context, CONTEXT_SIZE);
203 phy_addr = virt_to_phys((void *)context); 206 phy_addr = virt_to_phys((void *)context);
204 set_root_value(root, phy_addr); 207 set_root_value(root, phy_addr);
205 set_root_present(root); 208 set_root_present(root);
@@ -345,7 +348,7 @@ static struct dma_pte * addr_to_dma_pte(struct dmar_domain *domain, u64 addr)
345 return NULL; 348 return NULL;
346 } 349 }
347 __iommu_flush_cache(domain->iommu, tmp_page, 350 __iommu_flush_cache(domain->iommu, tmp_page,
348 PAGE_SIZE_4K); 351 PAGE_SIZE);
349 dma_set_pte_addr(*pte, virt_to_phys(tmp_page)); 352 dma_set_pte_addr(*pte, virt_to_phys(tmp_page));
350 /* 353 /*
351 * high level table always sets r/w, last level page 354 * high level table always sets r/w, last level page
@@ -408,13 +411,13 @@ static void dma_pte_clear_range(struct dmar_domain *domain, u64 start, u64 end)
408 start &= (((u64)1) << addr_width) - 1; 411 start &= (((u64)1) << addr_width) - 1;
409 end &= (((u64)1) << addr_width) - 1; 412 end &= (((u64)1) << addr_width) - 1;
410 /* in case it's partial page */ 413 /* in case it's partial page */
411 start = PAGE_ALIGN_4K(start); 414 start = PAGE_ALIGN(start);
412 end &= PAGE_MASK_4K; 415 end &= PAGE_MASK;
413 416
414 /* we don't need lock here, nobody else touches the iova range */ 417 /* we don't need lock here, nobody else touches the iova range */
415 while (start < end) { 418 while (start < end) {
416 dma_pte_clear_one(domain, start); 419 dma_pte_clear_one(domain, start);
417 start += PAGE_SIZE_4K; 420 start += VTD_PAGE_SIZE;
418 } 421 }
419} 422}
420 423
@@ -468,7 +471,7 @@ static int iommu_alloc_root_entry(struct intel_iommu *iommu)
468 if (!root) 471 if (!root)
469 return -ENOMEM; 472 return -ENOMEM;
470 473
471 __iommu_flush_cache(iommu, root, PAGE_SIZE_4K); 474 __iommu_flush_cache(iommu, root, ROOT_SIZE);
472 475
473 spin_lock_irqsave(&iommu->lock, flags); 476 spin_lock_irqsave(&iommu->lock, flags);
474 iommu->root_entry = root; 477 iommu->root_entry = root;
@@ -567,27 +570,6 @@ static int __iommu_flush_context(struct intel_iommu *iommu,
567 return 0; 570 return 0;
568} 571}
569 572
570static int inline iommu_flush_context_global(struct intel_iommu *iommu,
571 int non_present_entry_flush)
572{
573 return __iommu_flush_context(iommu, 0, 0, 0, DMA_CCMD_GLOBAL_INVL,
574 non_present_entry_flush);
575}
576
577static int inline iommu_flush_context_domain(struct intel_iommu *iommu, u16 did,
578 int non_present_entry_flush)
579{
580 return __iommu_flush_context(iommu, did, 0, 0, DMA_CCMD_DOMAIN_INVL,
581 non_present_entry_flush);
582}
583
584static int inline iommu_flush_context_device(struct intel_iommu *iommu,
585 u16 did, u16 source_id, u8 function_mask, int non_present_entry_flush)
586{
587 return __iommu_flush_context(iommu, did, source_id, function_mask,
588 DMA_CCMD_DEVICE_INVL, non_present_entry_flush);
589}
590
591/* return value determine if we need a write buffer flush */ 573/* return value determine if we need a write buffer flush */
592static int __iommu_flush_iotlb(struct intel_iommu *iommu, u16 did, 574static int __iommu_flush_iotlb(struct intel_iommu *iommu, u16 did,
593 u64 addr, unsigned int size_order, u64 type, 575 u64 addr, unsigned int size_order, u64 type,
@@ -655,37 +637,25 @@ static int __iommu_flush_iotlb(struct intel_iommu *iommu, u16 did,
655 printk(KERN_ERR"IOMMU: flush IOTLB failed\n"); 637 printk(KERN_ERR"IOMMU: flush IOTLB failed\n");
656 if (DMA_TLB_IAIG(val) != DMA_TLB_IIRG(type)) 638 if (DMA_TLB_IAIG(val) != DMA_TLB_IIRG(type))
657 pr_debug("IOMMU: tlb flush request %Lx, actual %Lx\n", 639 pr_debug("IOMMU: tlb flush request %Lx, actual %Lx\n",
658 DMA_TLB_IIRG(type), DMA_TLB_IAIG(val)); 640 (unsigned long long)DMA_TLB_IIRG(type),
641 (unsigned long long)DMA_TLB_IAIG(val));
659 /* flush iotlb entry will implicitly flush write buffer */ 642 /* flush iotlb entry will implicitly flush write buffer */
660 return 0; 643 return 0;
661} 644}
662 645
663static int inline iommu_flush_iotlb_global(struct intel_iommu *iommu,
664 int non_present_entry_flush)
665{
666 return __iommu_flush_iotlb(iommu, 0, 0, 0, DMA_TLB_GLOBAL_FLUSH,
667 non_present_entry_flush);
668}
669
670static int inline iommu_flush_iotlb_dsi(struct intel_iommu *iommu, u16 did,
671 int non_present_entry_flush)
672{
673 return __iommu_flush_iotlb(iommu, did, 0, 0, DMA_TLB_DSI_FLUSH,
674 non_present_entry_flush);
675}
676
677static int iommu_flush_iotlb_psi(struct intel_iommu *iommu, u16 did, 646static int iommu_flush_iotlb_psi(struct intel_iommu *iommu, u16 did,
678 u64 addr, unsigned int pages, int non_present_entry_flush) 647 u64 addr, unsigned int pages, int non_present_entry_flush)
679{ 648{
680 unsigned int mask; 649 unsigned int mask;
681 650
682 BUG_ON(addr & (~PAGE_MASK_4K)); 651 BUG_ON(addr & (~VTD_PAGE_MASK));
683 BUG_ON(pages == 0); 652 BUG_ON(pages == 0);
684 653
685 /* Fallback to domain selective flush if no PSI support */ 654 /* Fallback to domain selective flush if no PSI support */
686 if (!cap_pgsel_inv(iommu->cap)) 655 if (!cap_pgsel_inv(iommu->cap))
687 return iommu_flush_iotlb_dsi(iommu, did, 656 return iommu->flush.flush_iotlb(iommu, did, 0, 0,
688 non_present_entry_flush); 657 DMA_TLB_DSI_FLUSH,
658 non_present_entry_flush);
689 659
690 /* 660 /*
691 * PSI requires page size to be 2 ^ x, and the base address is naturally 661 * PSI requires page size to be 2 ^ x, and the base address is naturally
@@ -694,11 +664,12 @@ static int iommu_flush_iotlb_psi(struct intel_iommu *iommu, u16 did,
694 mask = ilog2(__roundup_pow_of_two(pages)); 664 mask = ilog2(__roundup_pow_of_two(pages));
695 /* Fallback to domain selective flush if size is too big */ 665 /* Fallback to domain selective flush if size is too big */
696 if (mask > cap_max_amask_val(iommu->cap)) 666 if (mask > cap_max_amask_val(iommu->cap))
697 return iommu_flush_iotlb_dsi(iommu, did, 667 return iommu->flush.flush_iotlb(iommu, did, 0, 0,
698 non_present_entry_flush); 668 DMA_TLB_DSI_FLUSH, non_present_entry_flush);
699 669
700 return __iommu_flush_iotlb(iommu, did, addr, mask, 670 return iommu->flush.flush_iotlb(iommu, did, addr, mask,
701 DMA_TLB_PSI_FLUSH, non_present_entry_flush); 671 DMA_TLB_PSI_FLUSH,
672 non_present_entry_flush);
702} 673}
703 674
704static void iommu_disable_protect_mem_regions(struct intel_iommu *iommu) 675static void iommu_disable_protect_mem_regions(struct intel_iommu *iommu)
@@ -831,7 +802,7 @@ void dmar_msi_read(int irq, struct msi_msg *msg)
831} 802}
832 803
833static int iommu_page_fault_do_one(struct intel_iommu *iommu, int type, 804static int iommu_page_fault_do_one(struct intel_iommu *iommu, int type,
834 u8 fault_reason, u16 source_id, u64 addr) 805 u8 fault_reason, u16 source_id, unsigned long long addr)
835{ 806{
836 const char *reason; 807 const char *reason;
837 808
@@ -1084,9 +1055,9 @@ static void dmar_init_reserved_ranges(void)
1084 if (!r->flags || !(r->flags & IORESOURCE_MEM)) 1055 if (!r->flags || !(r->flags & IORESOURCE_MEM))
1085 continue; 1056 continue;
1086 addr = r->start; 1057 addr = r->start;
1087 addr &= PAGE_MASK_4K; 1058 addr &= PAGE_MASK;
1088 size = r->end - addr; 1059 size = r->end - addr;
1089 size = PAGE_ALIGN_4K(size); 1060 size = PAGE_ALIGN(size);
1090 iova = reserve_iova(&reserved_iova_list, IOVA_PFN(addr), 1061 iova = reserve_iova(&reserved_iova_list, IOVA_PFN(addr),
1091 IOVA_PFN(size + addr) - 1); 1062 IOVA_PFN(size + addr) - 1);
1092 if (!iova) 1063 if (!iova)
@@ -1148,7 +1119,7 @@ static int domain_init(struct dmar_domain *domain, int guest_width)
1148 domain->pgd = (struct dma_pte *)alloc_pgtable_page(); 1119 domain->pgd = (struct dma_pte *)alloc_pgtable_page();
1149 if (!domain->pgd) 1120 if (!domain->pgd)
1150 return -ENOMEM; 1121 return -ENOMEM;
1151 __iommu_flush_cache(iommu, domain->pgd, PAGE_SIZE_4K); 1122 __iommu_flush_cache(iommu, domain->pgd, PAGE_SIZE);
1152 return 0; 1123 return 0;
1153} 1124}
1154 1125
@@ -1164,7 +1135,7 @@ static void domain_exit(struct dmar_domain *domain)
1164 /* destroy iovas */ 1135 /* destroy iovas */
1165 put_iova_domain(&domain->iovad); 1136 put_iova_domain(&domain->iovad);
1166 end = DOMAIN_MAX_ADDR(domain->gaw); 1137 end = DOMAIN_MAX_ADDR(domain->gaw);
1167 end = end & (~PAGE_MASK_4K); 1138 end = end & (~PAGE_MASK);
1168 1139
1169 /* clear ptes */ 1140 /* clear ptes */
1170 dma_pte_clear_range(domain, 0, end); 1141 dma_pte_clear_range(domain, 0, end);
@@ -1204,11 +1175,13 @@ static int domain_context_mapping_one(struct dmar_domain *domain,
1204 __iommu_flush_cache(iommu, context, sizeof(*context)); 1175 __iommu_flush_cache(iommu, context, sizeof(*context));
1205 1176
1206 /* it's a non-present to present mapping */ 1177 /* it's a non-present to present mapping */
1207 if (iommu_flush_context_device(iommu, domain->id, 1178 if (iommu->flush.flush_context(iommu, domain->id,
1208 (((u16)bus) << 8) | devfn, DMA_CCMD_MASK_NOBIT, 1)) 1179 (((u16)bus) << 8) | devfn, DMA_CCMD_MASK_NOBIT,
1180 DMA_CCMD_DEVICE_INVL, 1))
1209 iommu_flush_write_buffer(iommu); 1181 iommu_flush_write_buffer(iommu);
1210 else 1182 else
1211 iommu_flush_iotlb_dsi(iommu, 0, 0); 1183 iommu->flush.flush_iotlb(iommu, 0, 0, 0, DMA_TLB_DSI_FLUSH, 0);
1184
1212 spin_unlock_irqrestore(&iommu->lock, flags); 1185 spin_unlock_irqrestore(&iommu->lock, flags);
1213 return 0; 1186 return 0;
1214} 1187}
@@ -1283,22 +1256,25 @@ domain_page_mapping(struct dmar_domain *domain, dma_addr_t iova,
1283 u64 start_pfn, end_pfn; 1256 u64 start_pfn, end_pfn;
1284 struct dma_pte *pte; 1257 struct dma_pte *pte;
1285 int index; 1258 int index;
1259 int addr_width = agaw_to_width(domain->agaw);
1260
1261 hpa &= (((u64)1) << addr_width) - 1;
1286 1262
1287 if ((prot & (DMA_PTE_READ|DMA_PTE_WRITE)) == 0) 1263 if ((prot & (DMA_PTE_READ|DMA_PTE_WRITE)) == 0)
1288 return -EINVAL; 1264 return -EINVAL;
1289 iova &= PAGE_MASK_4K; 1265 iova &= PAGE_MASK;
1290 start_pfn = ((u64)hpa) >> PAGE_SHIFT_4K; 1266 start_pfn = ((u64)hpa) >> VTD_PAGE_SHIFT;
1291 end_pfn = (PAGE_ALIGN_4K(((u64)hpa) + size)) >> PAGE_SHIFT_4K; 1267 end_pfn = (VTD_PAGE_ALIGN(((u64)hpa) + size)) >> VTD_PAGE_SHIFT;
1292 index = 0; 1268 index = 0;
1293 while (start_pfn < end_pfn) { 1269 while (start_pfn < end_pfn) {
1294 pte = addr_to_dma_pte(domain, iova + PAGE_SIZE_4K * index); 1270 pte = addr_to_dma_pte(domain, iova + VTD_PAGE_SIZE * index);
1295 if (!pte) 1271 if (!pte)
1296 return -ENOMEM; 1272 return -ENOMEM;
1297 /* We don't need lock here, nobody else 1273 /* We don't need lock here, nobody else
1298 * touches the iova range 1274 * touches the iova range
1299 */ 1275 */
1300 BUG_ON(dma_pte_addr(*pte)); 1276 BUG_ON(dma_pte_addr(*pte));
1301 dma_set_pte_addr(*pte, start_pfn << PAGE_SHIFT_4K); 1277 dma_set_pte_addr(*pte, start_pfn << VTD_PAGE_SHIFT);
1302 dma_set_pte_prot(*pte, prot); 1278 dma_set_pte_prot(*pte, prot);
1303 __iommu_flush_cache(domain->iommu, pte, sizeof(*pte)); 1279 __iommu_flush_cache(domain->iommu, pte, sizeof(*pte));
1304 start_pfn++; 1280 start_pfn++;
@@ -1310,8 +1286,10 @@ domain_page_mapping(struct dmar_domain *domain, dma_addr_t iova,
1310static void detach_domain_for_dev(struct dmar_domain *domain, u8 bus, u8 devfn) 1286static void detach_domain_for_dev(struct dmar_domain *domain, u8 bus, u8 devfn)
1311{ 1287{
1312 clear_context_table(domain->iommu, bus, devfn); 1288 clear_context_table(domain->iommu, bus, devfn);
1313 iommu_flush_context_global(domain->iommu, 0); 1289 domain->iommu->flush.flush_context(domain->iommu, 0, 0, 0,
1314 iommu_flush_iotlb_global(domain->iommu, 0); 1290 DMA_CCMD_GLOBAL_INVL, 0);
1291 domain->iommu->flush.flush_iotlb(domain->iommu, 0, 0, 0,
1292 DMA_TLB_GLOBAL_FLUSH, 0);
1315} 1293}
1316 1294
1317static void domain_remove_dev_info(struct dmar_domain *domain) 1295static void domain_remove_dev_info(struct dmar_domain *domain)
@@ -1474,11 +1452,13 @@ error:
1474 return find_domain(pdev); 1452 return find_domain(pdev);
1475} 1453}
1476 1454
1477static int iommu_prepare_identity_map(struct pci_dev *pdev, u64 start, u64 end) 1455static int iommu_prepare_identity_map(struct pci_dev *pdev,
1456 unsigned long long start,
1457 unsigned long long end)
1478{ 1458{
1479 struct dmar_domain *domain; 1459 struct dmar_domain *domain;
1480 unsigned long size; 1460 unsigned long size;
1481 u64 base; 1461 unsigned long long base;
1482 int ret; 1462 int ret;
1483 1463
1484 printk(KERN_INFO 1464 printk(KERN_INFO
@@ -1490,9 +1470,9 @@ static int iommu_prepare_identity_map(struct pci_dev *pdev, u64 start, u64 end)
1490 return -ENOMEM; 1470 return -ENOMEM;
1491 1471
1492 /* The address might not be aligned */ 1472 /* The address might not be aligned */
1493 base = start & PAGE_MASK_4K; 1473 base = start & PAGE_MASK;
1494 size = end - base; 1474 size = end - base;
1495 size = PAGE_ALIGN_4K(size); 1475 size = PAGE_ALIGN(size);
1496 if (!reserve_iova(&domain->iovad, IOVA_PFN(base), 1476 if (!reserve_iova(&domain->iovad, IOVA_PFN(base),
1497 IOVA_PFN(base + size) - 1)) { 1477 IOVA_PFN(base + size) - 1)) {
1498 printk(KERN_ERR "IOMMU: reserve iova failed\n"); 1478 printk(KERN_ERR "IOMMU: reserve iova failed\n");
@@ -1662,6 +1642,28 @@ int __init init_dmars(void)
1662 } 1642 }
1663 } 1643 }
1664 1644
1645 for_each_drhd_unit(drhd) {
1646 if (drhd->ignored)
1647 continue;
1648
1649 iommu = drhd->iommu;
1650 if (dmar_enable_qi(iommu)) {
1651 /*
1652 * Queued Invalidate not enabled, use Register Based
1653 * Invalidate
1654 */
1655 iommu->flush.flush_context = __iommu_flush_context;
1656 iommu->flush.flush_iotlb = __iommu_flush_iotlb;
1657 printk(KERN_INFO "IOMMU 0x%Lx: using Register based "
1658 "invalidation\n", drhd->reg_base_addr);
1659 } else {
1660 iommu->flush.flush_context = qi_flush_context;
1661 iommu->flush.flush_iotlb = qi_flush_iotlb;
1662 printk(KERN_INFO "IOMMU 0x%Lx: using Queued "
1663 "invalidation\n", drhd->reg_base_addr);
1664 }
1665 }
1666
1665 /* 1667 /*
1666 * For each rmrr 1668 * For each rmrr
1667 * for each dev attached to rmrr 1669 * for each dev attached to rmrr
@@ -1714,9 +1716,10 @@ int __init init_dmars(void)
1714 1716
1715 iommu_set_root_entry(iommu); 1717 iommu_set_root_entry(iommu);
1716 1718
1717 iommu_flush_context_global(iommu, 0); 1719 iommu->flush.flush_context(iommu, 0, 0, 0, DMA_CCMD_GLOBAL_INVL,
1718 iommu_flush_iotlb_global(iommu, 0); 1720 0);
1719 1721 iommu->flush.flush_iotlb(iommu, 0, 0, 0, DMA_TLB_GLOBAL_FLUSH,
1722 0);
1720 iommu_disable_protect_mem_regions(iommu); 1723 iommu_disable_protect_mem_regions(iommu);
1721 1724
1722 ret = iommu_enable_translation(iommu); 1725 ret = iommu_enable_translation(iommu);
@@ -1738,8 +1741,8 @@ error:
1738static inline u64 aligned_size(u64 host_addr, size_t size) 1741static inline u64 aligned_size(u64 host_addr, size_t size)
1739{ 1742{
1740 u64 addr; 1743 u64 addr;
1741 addr = (host_addr & (~PAGE_MASK_4K)) + size; 1744 addr = (host_addr & (~PAGE_MASK)) + size;
1742 return PAGE_ALIGN_4K(addr); 1745 return PAGE_ALIGN(addr);
1743} 1746}
1744 1747
1745struct iova * 1748struct iova *
@@ -1753,20 +1756,20 @@ iommu_alloc_iova(struct dmar_domain *domain, size_t size, u64 end)
1753 return NULL; 1756 return NULL;
1754 1757
1755 piova = alloc_iova(&domain->iovad, 1758 piova = alloc_iova(&domain->iovad,
1756 size >> PAGE_SHIFT_4K, IOVA_PFN(end), 1); 1759 size >> PAGE_SHIFT, IOVA_PFN(end), 1);
1757 return piova; 1760 return piova;
1758} 1761}
1759 1762
1760static struct iova * 1763static struct iova *
1761__intel_alloc_iova(struct device *dev, struct dmar_domain *domain, 1764__intel_alloc_iova(struct device *dev, struct dmar_domain *domain,
1762 size_t size) 1765 size_t size, u64 dma_mask)
1763{ 1766{
1764 struct pci_dev *pdev = to_pci_dev(dev); 1767 struct pci_dev *pdev = to_pci_dev(dev);
1765 struct iova *iova = NULL; 1768 struct iova *iova = NULL;
1766 1769
1767 if ((pdev->dma_mask <= DMA_32BIT_MASK) || (dmar_forcedac)) { 1770 if (dma_mask <= DMA_32BIT_MASK || dmar_forcedac)
1768 iova = iommu_alloc_iova(domain, size, pdev->dma_mask); 1771 iova = iommu_alloc_iova(domain, size, dma_mask);
1769 } else { 1772 else {
1770 /* 1773 /*
1771 * First try to allocate an io virtual address in 1774 * First try to allocate an io virtual address in
1772 * DMA_32BIT_MASK and if that fails then try allocating 1775 * DMA_32BIT_MASK and if that fails then try allocating
@@ -1774,7 +1777,7 @@ __intel_alloc_iova(struct device *dev, struct dmar_domain *domain,
1774 */ 1777 */
1775 iova = iommu_alloc_iova(domain, size, DMA_32BIT_MASK); 1778 iova = iommu_alloc_iova(domain, size, DMA_32BIT_MASK);
1776 if (!iova) 1779 if (!iova)
1777 iova = iommu_alloc_iova(domain, size, pdev->dma_mask); 1780 iova = iommu_alloc_iova(domain, size, dma_mask);
1778 } 1781 }
1779 1782
1780 if (!iova) { 1783 if (!iova) {
@@ -1813,12 +1816,12 @@ get_valid_domain_for_dev(struct pci_dev *pdev)
1813 return domain; 1816 return domain;
1814} 1817}
1815 1818
1816static dma_addr_t 1819static dma_addr_t __intel_map_single(struct device *hwdev, phys_addr_t paddr,
1817intel_map_single(struct device *hwdev, phys_addr_t paddr, size_t size, int dir) 1820 size_t size, int dir, u64 dma_mask)
1818{ 1821{
1819 struct pci_dev *pdev = to_pci_dev(hwdev); 1822 struct pci_dev *pdev = to_pci_dev(hwdev);
1820 struct dmar_domain *domain; 1823 struct dmar_domain *domain;
1821 unsigned long start_paddr; 1824 phys_addr_t start_paddr;
1822 struct iova *iova; 1825 struct iova *iova;
1823 int prot = 0; 1826 int prot = 0;
1824 int ret; 1827 int ret;
@@ -1833,11 +1836,11 @@ intel_map_single(struct device *hwdev, phys_addr_t paddr, size_t size, int dir)
1833 1836
1834 size = aligned_size((u64)paddr, size); 1837 size = aligned_size((u64)paddr, size);
1835 1838
1836 iova = __intel_alloc_iova(hwdev, domain, size); 1839 iova = __intel_alloc_iova(hwdev, domain, size, pdev->dma_mask);
1837 if (!iova) 1840 if (!iova)
1838 goto error; 1841 goto error;
1839 1842
1840 start_paddr = iova->pfn_lo << PAGE_SHIFT_4K; 1843 start_paddr = (phys_addr_t)iova->pfn_lo << PAGE_SHIFT;
1841 1844
1842 /* 1845 /*
1843 * Check if DMAR supports zero-length reads on write only 1846 * Check if DMAR supports zero-length reads on write only
@@ -1855,30 +1858,33 @@ intel_map_single(struct device *hwdev, phys_addr_t paddr, size_t size, int dir)
1855 * is not a big problem 1858 * is not a big problem
1856 */ 1859 */
1857 ret = domain_page_mapping(domain, start_paddr, 1860 ret = domain_page_mapping(domain, start_paddr,
1858 ((u64)paddr) & PAGE_MASK_4K, size, prot); 1861 ((u64)paddr) & PAGE_MASK, size, prot);
1859 if (ret) 1862 if (ret)
1860 goto error; 1863 goto error;
1861 1864
1862 pr_debug("Device %s request: %lx@%llx mapping: %lx@%llx, dir %d\n",
1863 pci_name(pdev), size, (u64)paddr,
1864 size, (u64)start_paddr, dir);
1865
1866 /* it's a non-present to present mapping */ 1865 /* it's a non-present to present mapping */
1867 ret = iommu_flush_iotlb_psi(domain->iommu, domain->id, 1866 ret = iommu_flush_iotlb_psi(domain->iommu, domain->id,
1868 start_paddr, size >> PAGE_SHIFT_4K, 1); 1867 start_paddr, size >> VTD_PAGE_SHIFT, 1);
1869 if (ret) 1868 if (ret)
1870 iommu_flush_write_buffer(domain->iommu); 1869 iommu_flush_write_buffer(domain->iommu);
1871 1870
1872 return (start_paddr + ((u64)paddr & (~PAGE_MASK_4K))); 1871 return start_paddr + ((u64)paddr & (~PAGE_MASK));
1873 1872
1874error: 1873error:
1875 if (iova) 1874 if (iova)
1876 __free_iova(&domain->iovad, iova); 1875 __free_iova(&domain->iovad, iova);
1877 printk(KERN_ERR"Device %s request: %lx@%llx dir %d --- failed\n", 1876 printk(KERN_ERR"Device %s request: %lx@%llx dir %d --- failed\n",
1878 pci_name(pdev), size, (u64)paddr, dir); 1877 pci_name(pdev), size, (unsigned long long)paddr, dir);
1879 return 0; 1878 return 0;
1880} 1879}
1881 1880
1881dma_addr_t intel_map_single(struct device *hwdev, phys_addr_t paddr,
1882 size_t size, int dir)
1883{
1884 return __intel_map_single(hwdev, paddr, size, dir,
1885 to_pci_dev(hwdev)->dma_mask);
1886}
1887
1882static void flush_unmaps(void) 1888static void flush_unmaps(void)
1883{ 1889{
1884 int i, j; 1890 int i, j;
@@ -1891,7 +1897,8 @@ static void flush_unmaps(void)
1891 struct intel_iommu *iommu = 1897 struct intel_iommu *iommu =
1892 deferred_flush[i].domain[0]->iommu; 1898 deferred_flush[i].domain[0]->iommu;
1893 1899
1894 iommu_flush_iotlb_global(iommu, 0); 1900 iommu->flush.flush_iotlb(iommu, 0, 0, 0,
1901 DMA_TLB_GLOBAL_FLUSH, 0);
1895 for (j = 0; j < deferred_flush[i].next; j++) { 1902 for (j = 0; j < deferred_flush[i].next; j++) {
1896 __free_iova(&deferred_flush[i].domain[j]->iovad, 1903 __free_iova(&deferred_flush[i].domain[j]->iovad,
1897 deferred_flush[i].iova[j]); 1904 deferred_flush[i].iova[j]);
@@ -1936,8 +1943,8 @@ static void add_unmap(struct dmar_domain *dom, struct iova *iova)
1936 spin_unlock_irqrestore(&async_umap_flush_lock, flags); 1943 spin_unlock_irqrestore(&async_umap_flush_lock, flags);
1937} 1944}
1938 1945
1939static void intel_unmap_single(struct device *dev, dma_addr_t dev_addr, 1946void intel_unmap_single(struct device *dev, dma_addr_t dev_addr, size_t size,
1940 size_t size, int dir) 1947 int dir)
1941{ 1948{
1942 struct pci_dev *pdev = to_pci_dev(dev); 1949 struct pci_dev *pdev = to_pci_dev(dev);
1943 struct dmar_domain *domain; 1950 struct dmar_domain *domain;
@@ -1953,11 +1960,11 @@ static void intel_unmap_single(struct device *dev, dma_addr_t dev_addr,
1953 if (!iova) 1960 if (!iova)
1954 return; 1961 return;
1955 1962
1956 start_addr = iova->pfn_lo << PAGE_SHIFT_4K; 1963 start_addr = iova->pfn_lo << PAGE_SHIFT;
1957 size = aligned_size((u64)dev_addr, size); 1964 size = aligned_size((u64)dev_addr, size);
1958 1965
1959 pr_debug("Device %s unmapping: %lx@%llx\n", 1966 pr_debug("Device %s unmapping: %lx@%llx\n",
1960 pci_name(pdev), size, (u64)start_addr); 1967 pci_name(pdev), size, (unsigned long long)start_addr);
1961 1968
1962 /* clear the whole page */ 1969 /* clear the whole page */
1963 dma_pte_clear_range(domain, start_addr, start_addr + size); 1970 dma_pte_clear_range(domain, start_addr, start_addr + size);
@@ -1965,7 +1972,7 @@ static void intel_unmap_single(struct device *dev, dma_addr_t dev_addr,
1965 dma_pte_free_pagetable(domain, start_addr, start_addr + size); 1972 dma_pte_free_pagetable(domain, start_addr, start_addr + size);
1966 if (intel_iommu_strict) { 1973 if (intel_iommu_strict) {
1967 if (iommu_flush_iotlb_psi(domain->iommu, 1974 if (iommu_flush_iotlb_psi(domain->iommu,
1968 domain->id, start_addr, size >> PAGE_SHIFT_4K, 0)) 1975 domain->id, start_addr, size >> VTD_PAGE_SHIFT, 0))
1969 iommu_flush_write_buffer(domain->iommu); 1976 iommu_flush_write_buffer(domain->iommu);
1970 /* free iova */ 1977 /* free iova */
1971 __free_iova(&domain->iovad, iova); 1978 __free_iova(&domain->iovad, iova);
@@ -1978,13 +1985,13 @@ static void intel_unmap_single(struct device *dev, dma_addr_t dev_addr,
1978 } 1985 }
1979} 1986}
1980 1987
1981static void * intel_alloc_coherent(struct device *hwdev, size_t size, 1988void *intel_alloc_coherent(struct device *hwdev, size_t size,
1982 dma_addr_t *dma_handle, gfp_t flags) 1989 dma_addr_t *dma_handle, gfp_t flags)
1983{ 1990{
1984 void *vaddr; 1991 void *vaddr;
1985 int order; 1992 int order;
1986 1993
1987 size = PAGE_ALIGN_4K(size); 1994 size = PAGE_ALIGN(size);
1988 order = get_order(size); 1995 order = get_order(size);
1989 flags &= ~(GFP_DMA | GFP_DMA32); 1996 flags &= ~(GFP_DMA | GFP_DMA32);
1990 1997
@@ -1993,19 +2000,21 @@ static void * intel_alloc_coherent(struct device *hwdev, size_t size,
1993 return NULL; 2000 return NULL;
1994 memset(vaddr, 0, size); 2001 memset(vaddr, 0, size);
1995 2002
1996 *dma_handle = intel_map_single(hwdev, virt_to_bus(vaddr), size, DMA_BIDIRECTIONAL); 2003 *dma_handle = __intel_map_single(hwdev, virt_to_bus(vaddr), size,
2004 DMA_BIDIRECTIONAL,
2005 hwdev->coherent_dma_mask);
1997 if (*dma_handle) 2006 if (*dma_handle)
1998 return vaddr; 2007 return vaddr;
1999 free_pages((unsigned long)vaddr, order); 2008 free_pages((unsigned long)vaddr, order);
2000 return NULL; 2009 return NULL;
2001} 2010}
2002 2011
2003static void intel_free_coherent(struct device *hwdev, size_t size, 2012void intel_free_coherent(struct device *hwdev, size_t size, void *vaddr,
2004 void *vaddr, dma_addr_t dma_handle) 2013 dma_addr_t dma_handle)
2005{ 2014{
2006 int order; 2015 int order;
2007 2016
2008 size = PAGE_ALIGN_4K(size); 2017 size = PAGE_ALIGN(size);
2009 order = get_order(size); 2018 order = get_order(size);
2010 2019
2011 intel_unmap_single(hwdev, dma_handle, size, DMA_BIDIRECTIONAL); 2020 intel_unmap_single(hwdev, dma_handle, size, DMA_BIDIRECTIONAL);
@@ -2013,8 +2022,9 @@ static void intel_free_coherent(struct device *hwdev, size_t size,
2013} 2022}
2014 2023
2015#define SG_ENT_VIRT_ADDRESS(sg) (sg_virt((sg))) 2024#define SG_ENT_VIRT_ADDRESS(sg) (sg_virt((sg)))
2016static void intel_unmap_sg(struct device *hwdev, struct scatterlist *sglist, 2025
2017 int nelems, int dir) 2026void intel_unmap_sg(struct device *hwdev, struct scatterlist *sglist,
2027 int nelems, int dir)
2018{ 2028{
2019 int i; 2029 int i;
2020 struct pci_dev *pdev = to_pci_dev(hwdev); 2030 struct pci_dev *pdev = to_pci_dev(hwdev);
@@ -2038,7 +2048,7 @@ static void intel_unmap_sg(struct device *hwdev, struct scatterlist *sglist,
2038 size += aligned_size((u64)addr, sg->length); 2048 size += aligned_size((u64)addr, sg->length);
2039 } 2049 }
2040 2050
2041 start_addr = iova->pfn_lo << PAGE_SHIFT_4K; 2051 start_addr = iova->pfn_lo << PAGE_SHIFT;
2042 2052
2043 /* clear the whole page */ 2053 /* clear the whole page */
2044 dma_pte_clear_range(domain, start_addr, start_addr + size); 2054 dma_pte_clear_range(domain, start_addr, start_addr + size);
@@ -2046,7 +2056,7 @@ static void intel_unmap_sg(struct device *hwdev, struct scatterlist *sglist,
2046 dma_pte_free_pagetable(domain, start_addr, start_addr + size); 2056 dma_pte_free_pagetable(domain, start_addr, start_addr + size);
2047 2057
2048 if (iommu_flush_iotlb_psi(domain->iommu, domain->id, start_addr, 2058 if (iommu_flush_iotlb_psi(domain->iommu, domain->id, start_addr,
2049 size >> PAGE_SHIFT_4K, 0)) 2059 size >> VTD_PAGE_SHIFT, 0))
2050 iommu_flush_write_buffer(domain->iommu); 2060 iommu_flush_write_buffer(domain->iommu);
2051 2061
2052 /* free iova */ 2062 /* free iova */
@@ -2067,8 +2077,8 @@ static int intel_nontranslate_map_sg(struct device *hddev,
2067 return nelems; 2077 return nelems;
2068} 2078}
2069 2079
2070static int intel_map_sg(struct device *hwdev, struct scatterlist *sglist, 2080int intel_map_sg(struct device *hwdev, struct scatterlist *sglist, int nelems,
2071 int nelems, int dir) 2081 int dir)
2072{ 2082{
2073 void *addr; 2083 void *addr;
2074 int i; 2084 int i;
@@ -2096,7 +2106,7 @@ static int intel_map_sg(struct device *hwdev, struct scatterlist *sglist,
2096 size += aligned_size((u64)addr, sg->length); 2106 size += aligned_size((u64)addr, sg->length);
2097 } 2107 }
2098 2108
2099 iova = __intel_alloc_iova(hwdev, domain, size); 2109 iova = __intel_alloc_iova(hwdev, domain, size, pdev->dma_mask);
2100 if (!iova) { 2110 if (!iova) {
2101 sglist->dma_length = 0; 2111 sglist->dma_length = 0;
2102 return 0; 2112 return 0;
@@ -2112,14 +2122,14 @@ static int intel_map_sg(struct device *hwdev, struct scatterlist *sglist,
2112 if (dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL) 2122 if (dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL)
2113 prot |= DMA_PTE_WRITE; 2123 prot |= DMA_PTE_WRITE;
2114 2124
2115 start_addr = iova->pfn_lo << PAGE_SHIFT_4K; 2125 start_addr = iova->pfn_lo << PAGE_SHIFT;
2116 offset = 0; 2126 offset = 0;
2117 for_each_sg(sglist, sg, nelems, i) { 2127 for_each_sg(sglist, sg, nelems, i) {
2118 addr = SG_ENT_VIRT_ADDRESS(sg); 2128 addr = SG_ENT_VIRT_ADDRESS(sg);
2119 addr = (void *)virt_to_phys(addr); 2129 addr = (void *)virt_to_phys(addr);
2120 size = aligned_size((u64)addr, sg->length); 2130 size = aligned_size((u64)addr, sg->length);
2121 ret = domain_page_mapping(domain, start_addr + offset, 2131 ret = domain_page_mapping(domain, start_addr + offset,
2122 ((u64)addr) & PAGE_MASK_4K, 2132 ((u64)addr) & PAGE_MASK,
2123 size, prot); 2133 size, prot);
2124 if (ret) { 2134 if (ret) {
2125 /* clear the page */ 2135 /* clear the page */
@@ -2133,14 +2143,14 @@ static int intel_map_sg(struct device *hwdev, struct scatterlist *sglist,
2133 return 0; 2143 return 0;
2134 } 2144 }
2135 sg->dma_address = start_addr + offset + 2145 sg->dma_address = start_addr + offset +
2136 ((u64)addr & (~PAGE_MASK_4K)); 2146 ((u64)addr & (~PAGE_MASK));
2137 sg->dma_length = sg->length; 2147 sg->dma_length = sg->length;
2138 offset += size; 2148 offset += size;
2139 } 2149 }
2140 2150
2141 /* it's a non-present to present mapping */ 2151 /* it's a non-present to present mapping */
2142 if (iommu_flush_iotlb_psi(domain->iommu, domain->id, 2152 if (iommu_flush_iotlb_psi(domain->iommu, domain->id,
2143 start_addr, offset >> PAGE_SHIFT_4K, 1)) 2153 start_addr, offset >> VTD_PAGE_SHIFT, 1))
2144 iommu_flush_write_buffer(domain->iommu); 2154 iommu_flush_write_buffer(domain->iommu);
2145 return nelems; 2155 return nelems;
2146} 2156}
@@ -2180,7 +2190,6 @@ static inline int iommu_devinfo_cache_init(void)
2180 sizeof(struct device_domain_info), 2190 sizeof(struct device_domain_info),
2181 0, 2191 0,
2182 SLAB_HWCACHE_ALIGN, 2192 SLAB_HWCACHE_ALIGN,
2183
2184 NULL); 2193 NULL);
2185 if (!iommu_devinfo_cache) { 2194 if (!iommu_devinfo_cache) {
2186 printk(KERN_ERR "Couldn't create devinfo cache\n"); 2195 printk(KERN_ERR "Couldn't create devinfo cache\n");
@@ -2198,7 +2207,6 @@ static inline int iommu_iova_cache_init(void)
2198 sizeof(struct iova), 2207 sizeof(struct iova),
2199 0, 2208 0,
2200 SLAB_HWCACHE_ALIGN, 2209 SLAB_HWCACHE_ALIGN,
2201
2202 NULL); 2210 NULL);
2203 if (!iommu_iova_cache) { 2211 if (!iommu_iova_cache) {
2204 printk(KERN_ERR "Couldn't create iova cache\n"); 2212 printk(KERN_ERR "Couldn't create iova cache\n");
@@ -2327,7 +2335,7 @@ void intel_iommu_domain_exit(struct dmar_domain *domain)
2327 return; 2335 return;
2328 2336
2329 end = DOMAIN_MAX_ADDR(domain->gaw); 2337 end = DOMAIN_MAX_ADDR(domain->gaw);
2330 end = end & (~PAGE_MASK_4K); 2338 end = end & (~VTD_PAGE_MASK);
2331 2339
2332 /* clear ptes */ 2340 /* clear ptes */
2333 dma_pte_clear_range(domain, 0, end); 2341 dma_pte_clear_range(domain, 0, end);
@@ -2423,6 +2431,6 @@ u64 intel_iommu_iova_to_pfn(struct dmar_domain *domain, u64 iova)
2423 if (pte) 2431 if (pte)
2424 pfn = dma_pte_addr(*pte); 2432 pfn = dma_pte_addr(*pte);
2425 2433
2426 return pfn >> PAGE_SHIFT_4K; 2434 return pfn >> VTD_PAGE_SHIFT;
2427} 2435}
2428EXPORT_SYMBOL_GPL(intel_iommu_iova_to_pfn); 2436EXPORT_SYMBOL_GPL(intel_iommu_iova_to_pfn);