diff options
Diffstat (limited to 'arch/x86/kernel')
-rw-r--r-- | arch/x86/kernel/amd_iommu.c | 489 | ||||
-rw-r--r-- | arch/x86/kernel/amd_iommu_init.c | 42 | ||||
-rw-r--r-- | arch/x86/kernel/aperture_64.c | 6 | ||||
-rw-r--r-- | arch/x86/kernel/apic/nmi.c | 20 | ||||
-rw-r--r-- | arch/x86/kernel/apic/probe_64.c | 10 | ||||
-rw-r--r-- | arch/x86/kernel/cpu/Makefile | 4 | ||||
-rw-r--r-- | arch/x86/kernel/head_32.S | 8 | ||||
-rw-r--r-- | arch/x86/kernel/pci-dma.c | 17 | ||||
-rw-r--r-- | arch/x86/kernel/pci-gart_64.c | 5 | ||||
-rw-r--r-- | arch/x86/kernel/pci-nommu.c | 29 | ||||
-rw-r--r-- | arch/x86/kernel/pci-swiotlb.c | 25 | ||||
-rw-r--r-- | arch/x86/kernel/process.c | 6 | ||||
-rw-r--r-- | arch/x86/kernel/signal.c | 2 | ||||
-rw-r--r-- | arch/x86/kernel/vmlinux.lds.S | 126 |
14 files changed, 513 insertions, 276 deletions
diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c index 6c99f5037801..98f230f6a28d 100644 --- a/arch/x86/kernel/amd_iommu.c +++ b/arch/x86/kernel/amd_iommu.c | |||
@@ -41,9 +41,13 @@ static DEFINE_RWLOCK(amd_iommu_devtable_lock); | |||
41 | static LIST_HEAD(iommu_pd_list); | 41 | static LIST_HEAD(iommu_pd_list); |
42 | static DEFINE_SPINLOCK(iommu_pd_list_lock); | 42 | static DEFINE_SPINLOCK(iommu_pd_list_lock); |
43 | 43 | ||
44 | #ifdef CONFIG_IOMMU_API | 44 | /* |
45 | * Domain for untranslated devices - only allocated | ||
46 | * if iommu=pt passed on kernel cmd line. | ||
47 | */ | ||
48 | static struct protection_domain *pt_domain; | ||
49 | |||
45 | static struct iommu_ops amd_iommu_ops; | 50 | static struct iommu_ops amd_iommu_ops; |
46 | #endif | ||
47 | 51 | ||
48 | /* | 52 | /* |
49 | * general struct to manage commands send to an IOMMU | 53 | * general struct to manage commands send to an IOMMU |
@@ -55,16 +59,16 @@ struct iommu_cmd { | |||
55 | static int dma_ops_unity_map(struct dma_ops_domain *dma_dom, | 59 | static int dma_ops_unity_map(struct dma_ops_domain *dma_dom, |
56 | struct unity_map_entry *e); | 60 | struct unity_map_entry *e); |
57 | static struct dma_ops_domain *find_protection_domain(u16 devid); | 61 | static struct dma_ops_domain *find_protection_domain(u16 devid); |
58 | static u64* alloc_pte(struct protection_domain *dom, | 62 | static u64 *alloc_pte(struct protection_domain *domain, |
59 | unsigned long address, u64 | 63 | unsigned long address, int end_lvl, |
60 | **pte_page, gfp_t gfp); | 64 | u64 **pte_page, gfp_t gfp); |
61 | static void dma_ops_reserve_addresses(struct dma_ops_domain *dom, | 65 | static void dma_ops_reserve_addresses(struct dma_ops_domain *dom, |
62 | unsigned long start_page, | 66 | unsigned long start_page, |
63 | unsigned int pages); | 67 | unsigned int pages); |
64 | 68 | static void reset_iommu_command_buffer(struct amd_iommu *iommu); | |
65 | #ifndef BUS_NOTIFY_UNBOUND_DRIVER | 69 | static u64 *fetch_pte(struct protection_domain *domain, |
66 | #define BUS_NOTIFY_UNBOUND_DRIVER 0x0005 | 70 | unsigned long address, int map_size); |
67 | #endif | 71 | static void update_domain(struct protection_domain *domain); |
68 | 72 | ||
69 | #ifdef CONFIG_AMD_IOMMU_STATS | 73 | #ifdef CONFIG_AMD_IOMMU_STATS |
70 | 74 | ||
@@ -138,7 +142,25 @@ static int iommu_has_npcache(struct amd_iommu *iommu) | |||
138 | * | 142 | * |
139 | ****************************************************************************/ | 143 | ****************************************************************************/ |
140 | 144 | ||
141 | static void iommu_print_event(void *__evt) | 145 | static void dump_dte_entry(u16 devid) |
146 | { | ||
147 | int i; | ||
148 | |||
149 | for (i = 0; i < 8; ++i) | ||
150 | pr_err("AMD-Vi: DTE[%d]: %08x\n", i, | ||
151 | amd_iommu_dev_table[devid].data[i]); | ||
152 | } | ||
153 | |||
154 | static void dump_command(unsigned long phys_addr) | ||
155 | { | ||
156 | struct iommu_cmd *cmd = phys_to_virt(phys_addr); | ||
157 | int i; | ||
158 | |||
159 | for (i = 0; i < 4; ++i) | ||
160 | pr_err("AMD-Vi: CMD[%d]: %08x\n", i, cmd->data[i]); | ||
161 | } | ||
162 | |||
163 | static void iommu_print_event(struct amd_iommu *iommu, void *__evt) | ||
142 | { | 164 | { |
143 | u32 *event = __evt; | 165 | u32 *event = __evt; |
144 | int type = (event[1] >> EVENT_TYPE_SHIFT) & EVENT_TYPE_MASK; | 166 | int type = (event[1] >> EVENT_TYPE_SHIFT) & EVENT_TYPE_MASK; |
@@ -147,7 +169,7 @@ static void iommu_print_event(void *__evt) | |||
147 | int flags = (event[1] >> EVENT_FLAGS_SHIFT) & EVENT_FLAGS_MASK; | 169 | int flags = (event[1] >> EVENT_FLAGS_SHIFT) & EVENT_FLAGS_MASK; |
148 | u64 address = (u64)(((u64)event[3]) << 32) | event[2]; | 170 | u64 address = (u64)(((u64)event[3]) << 32) | event[2]; |
149 | 171 | ||
150 | printk(KERN_ERR "AMD IOMMU: Event logged ["); | 172 | printk(KERN_ERR "AMD-Vi: Event logged ["); |
151 | 173 | ||
152 | switch (type) { | 174 | switch (type) { |
153 | case EVENT_TYPE_ILL_DEV: | 175 | case EVENT_TYPE_ILL_DEV: |
@@ -155,6 +177,7 @@ static void iommu_print_event(void *__evt) | |||
155 | "address=0x%016llx flags=0x%04x]\n", | 177 | "address=0x%016llx flags=0x%04x]\n", |
156 | PCI_BUS(devid), PCI_SLOT(devid), PCI_FUNC(devid), | 178 | PCI_BUS(devid), PCI_SLOT(devid), PCI_FUNC(devid), |
157 | address, flags); | 179 | address, flags); |
180 | dump_dte_entry(devid); | ||
158 | break; | 181 | break; |
159 | case EVENT_TYPE_IO_FAULT: | 182 | case EVENT_TYPE_IO_FAULT: |
160 | printk("IO_PAGE_FAULT device=%02x:%02x.%x " | 183 | printk("IO_PAGE_FAULT device=%02x:%02x.%x " |
@@ -176,6 +199,8 @@ static void iommu_print_event(void *__evt) | |||
176 | break; | 199 | break; |
177 | case EVENT_TYPE_ILL_CMD: | 200 | case EVENT_TYPE_ILL_CMD: |
178 | printk("ILLEGAL_COMMAND_ERROR address=0x%016llx]\n", address); | 201 | printk("ILLEGAL_COMMAND_ERROR address=0x%016llx]\n", address); |
202 | reset_iommu_command_buffer(iommu); | ||
203 | dump_command(address); | ||
179 | break; | 204 | break; |
180 | case EVENT_TYPE_CMD_HARD_ERR: | 205 | case EVENT_TYPE_CMD_HARD_ERR: |
181 | printk("COMMAND_HARDWARE_ERROR address=0x%016llx " | 206 | printk("COMMAND_HARDWARE_ERROR address=0x%016llx " |
@@ -209,7 +234,7 @@ static void iommu_poll_events(struct amd_iommu *iommu) | |||
209 | tail = readl(iommu->mmio_base + MMIO_EVT_TAIL_OFFSET); | 234 | tail = readl(iommu->mmio_base + MMIO_EVT_TAIL_OFFSET); |
210 | 235 | ||
211 | while (head != tail) { | 236 | while (head != tail) { |
212 | iommu_print_event(iommu->evt_buf + head); | 237 | iommu_print_event(iommu, iommu->evt_buf + head); |
213 | head = (head + EVENT_ENTRY_SIZE) % iommu->evt_buf_size; | 238 | head = (head + EVENT_ENTRY_SIZE) % iommu->evt_buf_size; |
214 | } | 239 | } |
215 | 240 | ||
@@ -296,8 +321,11 @@ static void __iommu_wait_for_completion(struct amd_iommu *iommu) | |||
296 | status &= ~MMIO_STATUS_COM_WAIT_INT_MASK; | 321 | status &= ~MMIO_STATUS_COM_WAIT_INT_MASK; |
297 | writel(status, iommu->mmio_base + MMIO_STATUS_OFFSET); | 322 | writel(status, iommu->mmio_base + MMIO_STATUS_OFFSET); |
298 | 323 | ||
299 | if (unlikely(i == EXIT_LOOP_COUNT)) | 324 | if (unlikely(i == EXIT_LOOP_COUNT)) { |
300 | panic("AMD IOMMU: Completion wait loop failed\n"); | 325 | spin_unlock(&iommu->lock); |
326 | reset_iommu_command_buffer(iommu); | ||
327 | spin_lock(&iommu->lock); | ||
328 | } | ||
301 | } | 329 | } |
302 | 330 | ||
303 | /* | 331 | /* |
@@ -445,47 +473,78 @@ static void iommu_flush_tlb_pde(struct amd_iommu *iommu, u16 domid) | |||
445 | } | 473 | } |
446 | 474 | ||
447 | /* | 475 | /* |
476 | * This function flushes one domain on one IOMMU | ||
477 | */ | ||
478 | static void flush_domain_on_iommu(struct amd_iommu *iommu, u16 domid) | ||
479 | { | ||
480 | struct iommu_cmd cmd; | ||
481 | unsigned long flags; | ||
482 | |||
483 | __iommu_build_inv_iommu_pages(&cmd, CMD_INV_IOMMU_ALL_PAGES_ADDRESS, | ||
484 | domid, 1, 1); | ||
485 | |||
486 | spin_lock_irqsave(&iommu->lock, flags); | ||
487 | __iommu_queue_command(iommu, &cmd); | ||
488 | __iommu_completion_wait(iommu); | ||
489 | __iommu_wait_for_completion(iommu); | ||
490 | spin_unlock_irqrestore(&iommu->lock, flags); | ||
491 | } | ||
492 | |||
493 | static void flush_all_domains_on_iommu(struct amd_iommu *iommu) | ||
494 | { | ||
495 | int i; | ||
496 | |||
497 | for (i = 1; i < MAX_DOMAIN_ID; ++i) { | ||
498 | if (!test_bit(i, amd_iommu_pd_alloc_bitmap)) | ||
499 | continue; | ||
500 | flush_domain_on_iommu(iommu, i); | ||
501 | } | ||
502 | |||
503 | } | ||
504 | |||
505 | /* | ||
448 | * This function is used to flush the IO/TLB for a given protection domain | 506 | * This function is used to flush the IO/TLB for a given protection domain |
449 | * on every IOMMU in the system | 507 | * on every IOMMU in the system |
450 | */ | 508 | */ |
451 | static void iommu_flush_domain(u16 domid) | 509 | static void iommu_flush_domain(u16 domid) |
452 | { | 510 | { |
453 | unsigned long flags; | ||
454 | struct amd_iommu *iommu; | 511 | struct amd_iommu *iommu; |
455 | struct iommu_cmd cmd; | ||
456 | 512 | ||
457 | INC_STATS_COUNTER(domain_flush_all); | 513 | INC_STATS_COUNTER(domain_flush_all); |
458 | 514 | ||
459 | __iommu_build_inv_iommu_pages(&cmd, CMD_INV_IOMMU_ALL_PAGES_ADDRESS, | 515 | for_each_iommu(iommu) |
460 | domid, 1, 1); | 516 | flush_domain_on_iommu(iommu, domid); |
461 | |||
462 | for_each_iommu(iommu) { | ||
463 | spin_lock_irqsave(&iommu->lock, flags); | ||
464 | __iommu_queue_command(iommu, &cmd); | ||
465 | __iommu_completion_wait(iommu); | ||
466 | __iommu_wait_for_completion(iommu); | ||
467 | spin_unlock_irqrestore(&iommu->lock, flags); | ||
468 | } | ||
469 | } | 517 | } |
470 | 518 | ||
471 | void amd_iommu_flush_all_domains(void) | 519 | void amd_iommu_flush_all_domains(void) |
472 | { | 520 | { |
521 | struct amd_iommu *iommu; | ||
522 | |||
523 | for_each_iommu(iommu) | ||
524 | flush_all_domains_on_iommu(iommu); | ||
525 | } | ||
526 | |||
527 | static void flush_all_devices_for_iommu(struct amd_iommu *iommu) | ||
528 | { | ||
473 | int i; | 529 | int i; |
474 | 530 | ||
475 | for (i = 1; i < MAX_DOMAIN_ID; ++i) { | 531 | for (i = 0; i <= amd_iommu_last_bdf; ++i) { |
476 | if (!test_bit(i, amd_iommu_pd_alloc_bitmap)) | 532 | if (iommu != amd_iommu_rlookup_table[i]) |
477 | continue; | 533 | continue; |
478 | iommu_flush_domain(i); | 534 | |
535 | iommu_queue_inv_dev_entry(iommu, i); | ||
536 | iommu_completion_wait(iommu); | ||
479 | } | 537 | } |
480 | } | 538 | } |
481 | 539 | ||
482 | void amd_iommu_flush_all_devices(void) | 540 | static void flush_devices_by_domain(struct protection_domain *domain) |
483 | { | 541 | { |
484 | struct amd_iommu *iommu; | 542 | struct amd_iommu *iommu; |
485 | int i; | 543 | int i; |
486 | 544 | ||
487 | for (i = 0; i <= amd_iommu_last_bdf; ++i) { | 545 | for (i = 0; i <= amd_iommu_last_bdf; ++i) { |
488 | if (amd_iommu_pd_table[i] == NULL) | 546 | if ((domain == NULL && amd_iommu_pd_table[i] == NULL) || |
547 | (amd_iommu_pd_table[i] != domain)) | ||
489 | continue; | 548 | continue; |
490 | 549 | ||
491 | iommu = amd_iommu_rlookup_table[i]; | 550 | iommu = amd_iommu_rlookup_table[i]; |
@@ -497,6 +556,27 @@ void amd_iommu_flush_all_devices(void) | |||
497 | } | 556 | } |
498 | } | 557 | } |
499 | 558 | ||
559 | static void reset_iommu_command_buffer(struct amd_iommu *iommu) | ||
560 | { | ||
561 | pr_err("AMD-Vi: Resetting IOMMU command buffer\n"); | ||
562 | |||
563 | if (iommu->reset_in_progress) | ||
564 | panic("AMD-Vi: ILLEGAL_COMMAND_ERROR while resetting command buffer\n"); | ||
565 | |||
566 | iommu->reset_in_progress = true; | ||
567 | |||
568 | amd_iommu_reset_cmd_buffer(iommu); | ||
569 | flush_all_devices_for_iommu(iommu); | ||
570 | flush_all_domains_on_iommu(iommu); | ||
571 | |||
572 | iommu->reset_in_progress = false; | ||
573 | } | ||
574 | |||
575 | void amd_iommu_flush_all_devices(void) | ||
576 | { | ||
577 | flush_devices_by_domain(NULL); | ||
578 | } | ||
579 | |||
500 | /**************************************************************************** | 580 | /**************************************************************************** |
501 | * | 581 | * |
502 | * The functions below are used the create the page table mappings for | 582 | * The functions below are used the create the page table mappings for |
@@ -514,18 +594,21 @@ void amd_iommu_flush_all_devices(void) | |||
514 | static int iommu_map_page(struct protection_domain *dom, | 594 | static int iommu_map_page(struct protection_domain *dom, |
515 | unsigned long bus_addr, | 595 | unsigned long bus_addr, |
516 | unsigned long phys_addr, | 596 | unsigned long phys_addr, |
517 | int prot) | 597 | int prot, |
598 | int map_size) | ||
518 | { | 599 | { |
519 | u64 __pte, *pte; | 600 | u64 __pte, *pte; |
520 | 601 | ||
521 | bus_addr = PAGE_ALIGN(bus_addr); | 602 | bus_addr = PAGE_ALIGN(bus_addr); |
522 | phys_addr = PAGE_ALIGN(phys_addr); | 603 | phys_addr = PAGE_ALIGN(phys_addr); |
523 | 604 | ||
524 | /* only support 512GB address spaces for now */ | 605 | BUG_ON(!PM_ALIGNED(map_size, bus_addr)); |
525 | if (bus_addr > IOMMU_MAP_SIZE_L3 || !(prot & IOMMU_PROT_MASK)) | 606 | BUG_ON(!PM_ALIGNED(map_size, phys_addr)); |
607 | |||
608 | if (!(prot & IOMMU_PROT_MASK)) | ||
526 | return -EINVAL; | 609 | return -EINVAL; |
527 | 610 | ||
528 | pte = alloc_pte(dom, bus_addr, NULL, GFP_KERNEL); | 611 | pte = alloc_pte(dom, bus_addr, map_size, NULL, GFP_KERNEL); |
529 | 612 | ||
530 | if (IOMMU_PTE_PRESENT(*pte)) | 613 | if (IOMMU_PTE_PRESENT(*pte)) |
531 | return -EBUSY; | 614 | return -EBUSY; |
@@ -538,29 +621,18 @@ static int iommu_map_page(struct protection_domain *dom, | |||
538 | 621 | ||
539 | *pte = __pte; | 622 | *pte = __pte; |
540 | 623 | ||
624 | update_domain(dom); | ||
625 | |||
541 | return 0; | 626 | return 0; |
542 | } | 627 | } |
543 | 628 | ||
544 | static void iommu_unmap_page(struct protection_domain *dom, | 629 | static void iommu_unmap_page(struct protection_domain *dom, |
545 | unsigned long bus_addr) | 630 | unsigned long bus_addr, int map_size) |
546 | { | 631 | { |
547 | u64 *pte; | 632 | u64 *pte = fetch_pte(dom, bus_addr, map_size); |
548 | |||
549 | pte = &dom->pt_root[IOMMU_PTE_L2_INDEX(bus_addr)]; | ||
550 | |||
551 | if (!IOMMU_PTE_PRESENT(*pte)) | ||
552 | return; | ||
553 | |||
554 | pte = IOMMU_PTE_PAGE(*pte); | ||
555 | pte = &pte[IOMMU_PTE_L1_INDEX(bus_addr)]; | ||
556 | 633 | ||
557 | if (!IOMMU_PTE_PRESENT(*pte)) | 634 | if (pte) |
558 | return; | 635 | *pte = 0; |
559 | |||
560 | pte = IOMMU_PTE_PAGE(*pte); | ||
561 | pte = &pte[IOMMU_PTE_L1_INDEX(bus_addr)]; | ||
562 | |||
563 | *pte = 0; | ||
564 | } | 636 | } |
565 | 637 | ||
566 | /* | 638 | /* |
@@ -615,7 +687,8 @@ static int dma_ops_unity_map(struct dma_ops_domain *dma_dom, | |||
615 | 687 | ||
616 | for (addr = e->address_start; addr < e->address_end; | 688 | for (addr = e->address_start; addr < e->address_end; |
617 | addr += PAGE_SIZE) { | 689 | addr += PAGE_SIZE) { |
618 | ret = iommu_map_page(&dma_dom->domain, addr, addr, e->prot); | 690 | ret = iommu_map_page(&dma_dom->domain, addr, addr, e->prot, |
691 | PM_MAP_4k); | ||
619 | if (ret) | 692 | if (ret) |
620 | return ret; | 693 | return ret; |
621 | /* | 694 | /* |
@@ -670,24 +743,29 @@ static int init_unity_mappings_for_device(struct dma_ops_domain *dma_dom, | |||
670 | * This function checks if there is a PTE for a given dma address. If | 743 | * This function checks if there is a PTE for a given dma address. If |
671 | * there is one, it returns the pointer to it. | 744 | * there is one, it returns the pointer to it. |
672 | */ | 745 | */ |
673 | static u64* fetch_pte(struct protection_domain *domain, | 746 | static u64 *fetch_pte(struct protection_domain *domain, |
674 | unsigned long address) | 747 | unsigned long address, int map_size) |
675 | { | 748 | { |
749 | int level; | ||
676 | u64 *pte; | 750 | u64 *pte; |
677 | 751 | ||
678 | pte = &domain->pt_root[IOMMU_PTE_L2_INDEX(address)]; | 752 | level = domain->mode - 1; |
753 | pte = &domain->pt_root[PM_LEVEL_INDEX(level, address)]; | ||
679 | 754 | ||
680 | if (!IOMMU_PTE_PRESENT(*pte)) | 755 | while (level > map_size) { |
681 | return NULL; | 756 | if (!IOMMU_PTE_PRESENT(*pte)) |
757 | return NULL; | ||
682 | 758 | ||
683 | pte = IOMMU_PTE_PAGE(*pte); | 759 | level -= 1; |
684 | pte = &pte[IOMMU_PTE_L1_INDEX(address)]; | ||
685 | 760 | ||
686 | if (!IOMMU_PTE_PRESENT(*pte)) | 761 | pte = IOMMU_PTE_PAGE(*pte); |
687 | return NULL; | 762 | pte = &pte[PM_LEVEL_INDEX(level, address)]; |
688 | 763 | ||
689 | pte = IOMMU_PTE_PAGE(*pte); | 764 | if ((PM_PTE_LEVEL(*pte) == 0) && level != map_size) { |
690 | pte = &pte[IOMMU_PTE_L0_INDEX(address)]; | 765 | pte = NULL; |
766 | break; | ||
767 | } | ||
768 | } | ||
691 | 769 | ||
692 | return pte; | 770 | return pte; |
693 | } | 771 | } |
@@ -727,7 +805,7 @@ static int alloc_new_range(struct amd_iommu *iommu, | |||
727 | u64 *pte, *pte_page; | 805 | u64 *pte, *pte_page; |
728 | 806 | ||
729 | for (i = 0; i < num_ptes; ++i) { | 807 | for (i = 0; i < num_ptes; ++i) { |
730 | pte = alloc_pte(&dma_dom->domain, address, | 808 | pte = alloc_pte(&dma_dom->domain, address, PM_MAP_4k, |
731 | &pte_page, gfp); | 809 | &pte_page, gfp); |
732 | if (!pte) | 810 | if (!pte) |
733 | goto out_free; | 811 | goto out_free; |
@@ -760,16 +838,20 @@ static int alloc_new_range(struct amd_iommu *iommu, | |||
760 | for (i = dma_dom->aperture[index]->offset; | 838 | for (i = dma_dom->aperture[index]->offset; |
761 | i < dma_dom->aperture_size; | 839 | i < dma_dom->aperture_size; |
762 | i += PAGE_SIZE) { | 840 | i += PAGE_SIZE) { |
763 | u64 *pte = fetch_pte(&dma_dom->domain, i); | 841 | u64 *pte = fetch_pte(&dma_dom->domain, i, PM_MAP_4k); |
764 | if (!pte || !IOMMU_PTE_PRESENT(*pte)) | 842 | if (!pte || !IOMMU_PTE_PRESENT(*pte)) |
765 | continue; | 843 | continue; |
766 | 844 | ||
767 | dma_ops_reserve_addresses(dma_dom, i << PAGE_SHIFT, 1); | 845 | dma_ops_reserve_addresses(dma_dom, i << PAGE_SHIFT, 1); |
768 | } | 846 | } |
769 | 847 | ||
848 | update_domain(&dma_dom->domain); | ||
849 | |||
770 | return 0; | 850 | return 0; |
771 | 851 | ||
772 | out_free: | 852 | out_free: |
853 | update_domain(&dma_dom->domain); | ||
854 | |||
773 | free_page((unsigned long)dma_dom->aperture[index]->bitmap); | 855 | free_page((unsigned long)dma_dom->aperture[index]->bitmap); |
774 | 856 | ||
775 | kfree(dma_dom->aperture[index]); | 857 | kfree(dma_dom->aperture[index]); |
@@ -1009,7 +1091,7 @@ static struct dma_ops_domain *dma_ops_domain_alloc(struct amd_iommu *iommu) | |||
1009 | dma_dom->domain.id = domain_id_alloc(); | 1091 | dma_dom->domain.id = domain_id_alloc(); |
1010 | if (dma_dom->domain.id == 0) | 1092 | if (dma_dom->domain.id == 0) |
1011 | goto free_dma_dom; | 1093 | goto free_dma_dom; |
1012 | dma_dom->domain.mode = PAGE_MODE_3_LEVEL; | 1094 | dma_dom->domain.mode = PAGE_MODE_2_LEVEL; |
1013 | dma_dom->domain.pt_root = (void *)get_zeroed_page(GFP_KERNEL); | 1095 | dma_dom->domain.pt_root = (void *)get_zeroed_page(GFP_KERNEL); |
1014 | dma_dom->domain.flags = PD_DMA_OPS_MASK; | 1096 | dma_dom->domain.flags = PD_DMA_OPS_MASK; |
1015 | dma_dom->domain.priv = dma_dom; | 1097 | dma_dom->domain.priv = dma_dom; |
@@ -1063,6 +1145,41 @@ static struct protection_domain *domain_for_device(u16 devid) | |||
1063 | return dom; | 1145 | return dom; |
1064 | } | 1146 | } |
1065 | 1147 | ||
1148 | static void set_dte_entry(u16 devid, struct protection_domain *domain) | ||
1149 | { | ||
1150 | u64 pte_root = virt_to_phys(domain->pt_root); | ||
1151 | |||
1152 | pte_root |= (domain->mode & DEV_ENTRY_MODE_MASK) | ||
1153 | << DEV_ENTRY_MODE_SHIFT; | ||
1154 | pte_root |= IOMMU_PTE_IR | IOMMU_PTE_IW | IOMMU_PTE_P | IOMMU_PTE_TV; | ||
1155 | |||
1156 | amd_iommu_dev_table[devid].data[2] = domain->id; | ||
1157 | amd_iommu_dev_table[devid].data[1] = upper_32_bits(pte_root); | ||
1158 | amd_iommu_dev_table[devid].data[0] = lower_32_bits(pte_root); | ||
1159 | |||
1160 | amd_iommu_pd_table[devid] = domain; | ||
1161 | } | ||
1162 | |||
1163 | /* | ||
1164 | * If a device is not yet associated with a domain, this function does | ||
1165 | * assigns it visible for the hardware | ||
1166 | */ | ||
1167 | static void __attach_device(struct amd_iommu *iommu, | ||
1168 | struct protection_domain *domain, | ||
1169 | u16 devid) | ||
1170 | { | ||
1171 | /* lock domain */ | ||
1172 | spin_lock(&domain->lock); | ||
1173 | |||
1174 | /* update DTE entry */ | ||
1175 | set_dte_entry(devid, domain); | ||
1176 | |||
1177 | domain->dev_cnt += 1; | ||
1178 | |||
1179 | /* ready */ | ||
1180 | spin_unlock(&domain->lock); | ||
1181 | } | ||
1182 | |||
1066 | /* | 1183 | /* |
1067 | * If a device is not yet associated with a domain, this function does | 1184 | * If a device is not yet associated with a domain, this function does |
1068 | * assigns it visible for the hardware | 1185 | * assigns it visible for the hardware |
@@ -1072,27 +1189,16 @@ static void attach_device(struct amd_iommu *iommu, | |||
1072 | u16 devid) | 1189 | u16 devid) |
1073 | { | 1190 | { |
1074 | unsigned long flags; | 1191 | unsigned long flags; |
1075 | u64 pte_root = virt_to_phys(domain->pt_root); | ||
1076 | |||
1077 | domain->dev_cnt += 1; | ||
1078 | |||
1079 | pte_root |= (domain->mode & DEV_ENTRY_MODE_MASK) | ||
1080 | << DEV_ENTRY_MODE_SHIFT; | ||
1081 | pte_root |= IOMMU_PTE_IR | IOMMU_PTE_IW | IOMMU_PTE_P | IOMMU_PTE_TV; | ||
1082 | 1192 | ||
1083 | write_lock_irqsave(&amd_iommu_devtable_lock, flags); | 1193 | write_lock_irqsave(&amd_iommu_devtable_lock, flags); |
1084 | amd_iommu_dev_table[devid].data[0] = lower_32_bits(pte_root); | 1194 | __attach_device(iommu, domain, devid); |
1085 | amd_iommu_dev_table[devid].data[1] = upper_32_bits(pte_root); | ||
1086 | amd_iommu_dev_table[devid].data[2] = domain->id; | ||
1087 | |||
1088 | amd_iommu_pd_table[devid] = domain; | ||
1089 | write_unlock_irqrestore(&amd_iommu_devtable_lock, flags); | 1195 | write_unlock_irqrestore(&amd_iommu_devtable_lock, flags); |
1090 | 1196 | ||
1091 | /* | 1197 | /* |
1092 | * We might boot into a crash-kernel here. The crashed kernel | 1198 | * We might boot into a crash-kernel here. The crashed kernel |
1093 | * left the caches in the IOMMU dirty. So we have to flush | 1199 | * left the caches in the IOMMU dirty. So we have to flush |
1094 | * here to evict all dirty stuff. | 1200 | * here to evict all dirty stuff. |
1095 | */ | 1201 | */ |
1096 | iommu_queue_inv_dev_entry(iommu, devid); | 1202 | iommu_queue_inv_dev_entry(iommu, devid); |
1097 | iommu_flush_tlb_pde(iommu, domain->id); | 1203 | iommu_flush_tlb_pde(iommu, domain->id); |
1098 | } | 1204 | } |
@@ -1119,6 +1225,15 @@ static void __detach_device(struct protection_domain *domain, u16 devid) | |||
1119 | 1225 | ||
1120 | /* ready */ | 1226 | /* ready */ |
1121 | spin_unlock(&domain->lock); | 1227 | spin_unlock(&domain->lock); |
1228 | |||
1229 | /* | ||
1230 | * If we run in passthrough mode the device must be assigned to the | ||
1231 | * passthrough domain if it is detached from any other domain | ||
1232 | */ | ||
1233 | if (iommu_pass_through) { | ||
1234 | struct amd_iommu *iommu = amd_iommu_rlookup_table[devid]; | ||
1235 | __attach_device(iommu, pt_domain, devid); | ||
1236 | } | ||
1122 | } | 1237 | } |
1123 | 1238 | ||
1124 | /* | 1239 | /* |
@@ -1164,6 +1279,8 @@ static int device_change_notifier(struct notifier_block *nb, | |||
1164 | case BUS_NOTIFY_UNBOUND_DRIVER: | 1279 | case BUS_NOTIFY_UNBOUND_DRIVER: |
1165 | if (!domain) | 1280 | if (!domain) |
1166 | goto out; | 1281 | goto out; |
1282 | if (iommu_pass_through) | ||
1283 | break; | ||
1167 | detach_device(domain, devid); | 1284 | detach_device(domain, devid); |
1168 | break; | 1285 | break; |
1169 | case BUS_NOTIFY_ADD_DEVICE: | 1286 | case BUS_NOTIFY_ADD_DEVICE: |
@@ -1292,39 +1409,91 @@ static int get_device_resources(struct device *dev, | |||
1292 | return 1; | 1409 | return 1; |
1293 | } | 1410 | } |
1294 | 1411 | ||
1412 | static void update_device_table(struct protection_domain *domain) | ||
1413 | { | ||
1414 | unsigned long flags; | ||
1415 | int i; | ||
1416 | |||
1417 | for (i = 0; i <= amd_iommu_last_bdf; ++i) { | ||
1418 | if (amd_iommu_pd_table[i] != domain) | ||
1419 | continue; | ||
1420 | write_lock_irqsave(&amd_iommu_devtable_lock, flags); | ||
1421 | set_dte_entry(i, domain); | ||
1422 | write_unlock_irqrestore(&amd_iommu_devtable_lock, flags); | ||
1423 | } | ||
1424 | } | ||
1425 | |||
1426 | static void update_domain(struct protection_domain *domain) | ||
1427 | { | ||
1428 | if (!domain->updated) | ||
1429 | return; | ||
1430 | |||
1431 | update_device_table(domain); | ||
1432 | flush_devices_by_domain(domain); | ||
1433 | iommu_flush_domain(domain->id); | ||
1434 | |||
1435 | domain->updated = false; | ||
1436 | } | ||
1437 | |||
1295 | /* | 1438 | /* |
1296 | * If the pte_page is not yet allocated this function is called | 1439 | * This function is used to add another level to an IO page table. Adding |
1440 | * another level increases the size of the address space by 9 bits to a size up | ||
1441 | * to 64 bits. | ||
1297 | */ | 1442 | */ |
1298 | static u64* alloc_pte(struct protection_domain *dom, | 1443 | static bool increase_address_space(struct protection_domain *domain, |
1299 | unsigned long address, u64 **pte_page, gfp_t gfp) | 1444 | gfp_t gfp) |
1445 | { | ||
1446 | u64 *pte; | ||
1447 | |||
1448 | if (domain->mode == PAGE_MODE_6_LEVEL) | ||
1449 | /* address space already 64 bit large */ | ||
1450 | return false; | ||
1451 | |||
1452 | pte = (void *)get_zeroed_page(gfp); | ||
1453 | if (!pte) | ||
1454 | return false; | ||
1455 | |||
1456 | *pte = PM_LEVEL_PDE(domain->mode, | ||
1457 | virt_to_phys(domain->pt_root)); | ||
1458 | domain->pt_root = pte; | ||
1459 | domain->mode += 1; | ||
1460 | domain->updated = true; | ||
1461 | |||
1462 | return true; | ||
1463 | } | ||
1464 | |||
1465 | static u64 *alloc_pte(struct protection_domain *domain, | ||
1466 | unsigned long address, | ||
1467 | int end_lvl, | ||
1468 | u64 **pte_page, | ||
1469 | gfp_t gfp) | ||
1300 | { | 1470 | { |
1301 | u64 *pte, *page; | 1471 | u64 *pte, *page; |
1472 | int level; | ||
1302 | 1473 | ||
1303 | pte = &dom->pt_root[IOMMU_PTE_L2_INDEX(address)]; | 1474 | while (address > PM_LEVEL_SIZE(domain->mode)) |
1475 | increase_address_space(domain, gfp); | ||
1304 | 1476 | ||
1305 | if (!IOMMU_PTE_PRESENT(*pte)) { | 1477 | level = domain->mode - 1; |
1306 | page = (u64 *)get_zeroed_page(gfp); | 1478 | pte = &domain->pt_root[PM_LEVEL_INDEX(level, address)]; |
1307 | if (!page) | ||
1308 | return NULL; | ||
1309 | *pte = IOMMU_L2_PDE(virt_to_phys(page)); | ||
1310 | } | ||
1311 | 1479 | ||
1312 | pte = IOMMU_PTE_PAGE(*pte); | 1480 | while (level > end_lvl) { |
1313 | pte = &pte[IOMMU_PTE_L1_INDEX(address)]; | 1481 | if (!IOMMU_PTE_PRESENT(*pte)) { |
1482 | page = (u64 *)get_zeroed_page(gfp); | ||
1483 | if (!page) | ||
1484 | return NULL; | ||
1485 | *pte = PM_LEVEL_PDE(level, virt_to_phys(page)); | ||
1486 | } | ||
1314 | 1487 | ||
1315 | if (!IOMMU_PTE_PRESENT(*pte)) { | 1488 | level -= 1; |
1316 | page = (u64 *)get_zeroed_page(gfp); | ||
1317 | if (!page) | ||
1318 | return NULL; | ||
1319 | *pte = IOMMU_L1_PDE(virt_to_phys(page)); | ||
1320 | } | ||
1321 | 1489 | ||
1322 | pte = IOMMU_PTE_PAGE(*pte); | 1490 | pte = IOMMU_PTE_PAGE(*pte); |
1323 | 1491 | ||
1324 | if (pte_page) | 1492 | if (pte_page && level == end_lvl) |
1325 | *pte_page = pte; | 1493 | *pte_page = pte; |
1326 | 1494 | ||
1327 | pte = &pte[IOMMU_PTE_L0_INDEX(address)]; | 1495 | pte = &pte[PM_LEVEL_INDEX(level, address)]; |
1496 | } | ||
1328 | 1497 | ||
1329 | return pte; | 1498 | return pte; |
1330 | } | 1499 | } |
@@ -1344,10 +1513,13 @@ static u64* dma_ops_get_pte(struct dma_ops_domain *dom, | |||
1344 | 1513 | ||
1345 | pte = aperture->pte_pages[APERTURE_PAGE_INDEX(address)]; | 1514 | pte = aperture->pte_pages[APERTURE_PAGE_INDEX(address)]; |
1346 | if (!pte) { | 1515 | if (!pte) { |
1347 | pte = alloc_pte(&dom->domain, address, &pte_page, GFP_ATOMIC); | 1516 | pte = alloc_pte(&dom->domain, address, PM_MAP_4k, &pte_page, |
1517 | GFP_ATOMIC); | ||
1348 | aperture->pte_pages[APERTURE_PAGE_INDEX(address)] = pte_page; | 1518 | aperture->pte_pages[APERTURE_PAGE_INDEX(address)] = pte_page; |
1349 | } else | 1519 | } else |
1350 | pte += IOMMU_PTE_L0_INDEX(address); | 1520 | pte += PM_LEVEL_INDEX(0, address); |
1521 | |||
1522 | update_domain(&dom->domain); | ||
1351 | 1523 | ||
1352 | return pte; | 1524 | return pte; |
1353 | } | 1525 | } |
@@ -1409,7 +1581,7 @@ static void dma_ops_domain_unmap(struct amd_iommu *iommu, | |||
1409 | if (!pte) | 1581 | if (!pte) |
1410 | return; | 1582 | return; |
1411 | 1583 | ||
1412 | pte += IOMMU_PTE_L0_INDEX(address); | 1584 | pte += PM_LEVEL_INDEX(0, address); |
1413 | 1585 | ||
1414 | WARN_ON(!*pte); | 1586 | WARN_ON(!*pte); |
1415 | 1587 | ||
@@ -1988,19 +2160,47 @@ static void cleanup_domain(struct protection_domain *domain) | |||
1988 | write_unlock_irqrestore(&amd_iommu_devtable_lock, flags); | 2160 | write_unlock_irqrestore(&amd_iommu_devtable_lock, flags); |
1989 | } | 2161 | } |
1990 | 2162 | ||
1991 | static int amd_iommu_domain_init(struct iommu_domain *dom) | 2163 | static void protection_domain_free(struct protection_domain *domain) |
2164 | { | ||
2165 | if (!domain) | ||
2166 | return; | ||
2167 | |||
2168 | if (domain->id) | ||
2169 | domain_id_free(domain->id); | ||
2170 | |||
2171 | kfree(domain); | ||
2172 | } | ||
2173 | |||
2174 | static struct protection_domain *protection_domain_alloc(void) | ||
1992 | { | 2175 | { |
1993 | struct protection_domain *domain; | 2176 | struct protection_domain *domain; |
1994 | 2177 | ||
1995 | domain = kzalloc(sizeof(*domain), GFP_KERNEL); | 2178 | domain = kzalloc(sizeof(*domain), GFP_KERNEL); |
1996 | if (!domain) | 2179 | if (!domain) |
1997 | return -ENOMEM; | 2180 | return NULL; |
1998 | 2181 | ||
1999 | spin_lock_init(&domain->lock); | 2182 | spin_lock_init(&domain->lock); |
2000 | domain->mode = PAGE_MODE_3_LEVEL; | ||
2001 | domain->id = domain_id_alloc(); | 2183 | domain->id = domain_id_alloc(); |
2002 | if (!domain->id) | 2184 | if (!domain->id) |
2185 | goto out_err; | ||
2186 | |||
2187 | return domain; | ||
2188 | |||
2189 | out_err: | ||
2190 | kfree(domain); | ||
2191 | |||
2192 | return NULL; | ||
2193 | } | ||
2194 | |||
2195 | static int amd_iommu_domain_init(struct iommu_domain *dom) | ||
2196 | { | ||
2197 | struct protection_domain *domain; | ||
2198 | |||
2199 | domain = protection_domain_alloc(); | ||
2200 | if (!domain) | ||
2003 | goto out_free; | 2201 | goto out_free; |
2202 | |||
2203 | domain->mode = PAGE_MODE_3_LEVEL; | ||
2004 | domain->pt_root = (void *)get_zeroed_page(GFP_KERNEL); | 2204 | domain->pt_root = (void *)get_zeroed_page(GFP_KERNEL); |
2005 | if (!domain->pt_root) | 2205 | if (!domain->pt_root) |
2006 | goto out_free; | 2206 | goto out_free; |
@@ -2010,7 +2210,7 @@ static int amd_iommu_domain_init(struct iommu_domain *dom) | |||
2010 | return 0; | 2210 | return 0; |
2011 | 2211 | ||
2012 | out_free: | 2212 | out_free: |
2013 | kfree(domain); | 2213 | protection_domain_free(domain); |
2014 | 2214 | ||
2015 | return -ENOMEM; | 2215 | return -ENOMEM; |
2016 | } | 2216 | } |
@@ -2115,7 +2315,7 @@ static int amd_iommu_map_range(struct iommu_domain *dom, | |||
2115 | paddr &= PAGE_MASK; | 2315 | paddr &= PAGE_MASK; |
2116 | 2316 | ||
2117 | for (i = 0; i < npages; ++i) { | 2317 | for (i = 0; i < npages; ++i) { |
2118 | ret = iommu_map_page(domain, iova, paddr, prot); | 2318 | ret = iommu_map_page(domain, iova, paddr, prot, PM_MAP_4k); |
2119 | if (ret) | 2319 | if (ret) |
2120 | return ret; | 2320 | return ret; |
2121 | 2321 | ||
@@ -2136,7 +2336,7 @@ static void amd_iommu_unmap_range(struct iommu_domain *dom, | |||
2136 | iova &= PAGE_MASK; | 2336 | iova &= PAGE_MASK; |
2137 | 2337 | ||
2138 | for (i = 0; i < npages; ++i) { | 2338 | for (i = 0; i < npages; ++i) { |
2139 | iommu_unmap_page(domain, iova); | 2339 | iommu_unmap_page(domain, iova, PM_MAP_4k); |
2140 | iova += PAGE_SIZE; | 2340 | iova += PAGE_SIZE; |
2141 | } | 2341 | } |
2142 | 2342 | ||
@@ -2151,21 +2351,9 @@ static phys_addr_t amd_iommu_iova_to_phys(struct iommu_domain *dom, | |||
2151 | phys_addr_t paddr; | 2351 | phys_addr_t paddr; |
2152 | u64 *pte; | 2352 | u64 *pte; |
2153 | 2353 | ||
2154 | pte = &domain->pt_root[IOMMU_PTE_L2_INDEX(iova)]; | 2354 | pte = fetch_pte(domain, iova, PM_MAP_4k); |
2155 | |||
2156 | if (!IOMMU_PTE_PRESENT(*pte)) | ||
2157 | return 0; | ||
2158 | |||
2159 | pte = IOMMU_PTE_PAGE(*pte); | ||
2160 | pte = &pte[IOMMU_PTE_L1_INDEX(iova)]; | ||
2161 | |||
2162 | if (!IOMMU_PTE_PRESENT(*pte)) | ||
2163 | return 0; | ||
2164 | |||
2165 | pte = IOMMU_PTE_PAGE(*pte); | ||
2166 | pte = &pte[IOMMU_PTE_L0_INDEX(iova)]; | ||
2167 | 2355 | ||
2168 | if (!IOMMU_PTE_PRESENT(*pte)) | 2356 | if (!pte || !IOMMU_PTE_PRESENT(*pte)) |
2169 | return 0; | 2357 | return 0; |
2170 | 2358 | ||
2171 | paddr = *pte & IOMMU_PAGE_MASK; | 2359 | paddr = *pte & IOMMU_PAGE_MASK; |
@@ -2191,3 +2379,46 @@ static struct iommu_ops amd_iommu_ops = { | |||
2191 | .domain_has_cap = amd_iommu_domain_has_cap, | 2379 | .domain_has_cap = amd_iommu_domain_has_cap, |
2192 | }; | 2380 | }; |
2193 | 2381 | ||
2382 | /***************************************************************************** | ||
2383 | * | ||
2384 | * The next functions do a basic initialization of IOMMU for pass through | ||
2385 | * mode | ||
2386 | * | ||
2387 | * In passthrough mode the IOMMU is initialized and enabled but not used for | ||
2388 | * DMA-API translation. | ||
2389 | * | ||
2390 | *****************************************************************************/ | ||
2391 | |||
2392 | int __init amd_iommu_init_passthrough(void) | ||
2393 | { | ||
2394 | struct pci_dev *dev = NULL; | ||
2395 | u16 devid, devid2; | ||
2396 | |||
2397 | /* allocate passthroug domain */ | ||
2398 | pt_domain = protection_domain_alloc(); | ||
2399 | if (!pt_domain) | ||
2400 | return -ENOMEM; | ||
2401 | |||
2402 | pt_domain->mode |= PAGE_MODE_NONE; | ||
2403 | |||
2404 | while ((dev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, dev)) != NULL) { | ||
2405 | struct amd_iommu *iommu; | ||
2406 | |||
2407 | devid = calc_devid(dev->bus->number, dev->devfn); | ||
2408 | if (devid > amd_iommu_last_bdf) | ||
2409 | continue; | ||
2410 | |||
2411 | devid2 = amd_iommu_alias_table[devid]; | ||
2412 | |||
2413 | iommu = amd_iommu_rlookup_table[devid2]; | ||
2414 | if (!iommu) | ||
2415 | continue; | ||
2416 | |||
2417 | __attach_device(iommu, pt_domain, devid); | ||
2418 | __attach_device(iommu, pt_domain, devid2); | ||
2419 | } | ||
2420 | |||
2421 | pr_info("AMD-Vi: Initialized for Passthrough Mode\n"); | ||
2422 | |||
2423 | return 0; | ||
2424 | } | ||
diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c index c1b17e97252e..b4b61d462dcc 100644 --- a/arch/x86/kernel/amd_iommu_init.c +++ b/arch/x86/kernel/amd_iommu_init.c | |||
@@ -252,7 +252,7 @@ static void __init iommu_feature_disable(struct amd_iommu *iommu, u8 bit) | |||
252 | /* Function to enable the hardware */ | 252 | /* Function to enable the hardware */ |
253 | static void iommu_enable(struct amd_iommu *iommu) | 253 | static void iommu_enable(struct amd_iommu *iommu) |
254 | { | 254 | { |
255 | printk(KERN_INFO "AMD IOMMU: Enabling IOMMU at %s cap 0x%hx\n", | 255 | printk(KERN_INFO "AMD-Vi: Enabling IOMMU at %s cap 0x%hx\n", |
256 | dev_name(&iommu->dev->dev), iommu->cap_ptr); | 256 | dev_name(&iommu->dev->dev), iommu->cap_ptr); |
257 | 257 | ||
258 | iommu_feature_enable(iommu, CONTROL_IOMMU_EN); | 258 | iommu_feature_enable(iommu, CONTROL_IOMMU_EN); |
@@ -435,6 +435,20 @@ static u8 * __init alloc_command_buffer(struct amd_iommu *iommu) | |||
435 | } | 435 | } |
436 | 436 | ||
437 | /* | 437 | /* |
438 | * This function resets the command buffer if the IOMMU stopped fetching | ||
439 | * commands from it. | ||
440 | */ | ||
441 | void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu) | ||
442 | { | ||
443 | iommu_feature_disable(iommu, CONTROL_CMDBUF_EN); | ||
444 | |||
445 | writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET); | ||
446 | writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET); | ||
447 | |||
448 | iommu_feature_enable(iommu, CONTROL_CMDBUF_EN); | ||
449 | } | ||
450 | |||
451 | /* | ||
438 | * This function writes the command buffer address to the hardware and | 452 | * This function writes the command buffer address to the hardware and |
439 | * enables it. | 453 | * enables it. |
440 | */ | 454 | */ |
@@ -450,11 +464,7 @@ static void iommu_enable_command_buffer(struct amd_iommu *iommu) | |||
450 | memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET, | 464 | memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET, |
451 | &entry, sizeof(entry)); | 465 | &entry, sizeof(entry)); |
452 | 466 | ||
453 | /* set head and tail to zero manually */ | 467 | amd_iommu_reset_cmd_buffer(iommu); |
454 | writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET); | ||
455 | writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET); | ||
456 | |||
457 | iommu_feature_enable(iommu, CONTROL_CMDBUF_EN); | ||
458 | } | 468 | } |
459 | 469 | ||
460 | static void __init free_command_buffer(struct amd_iommu *iommu) | 470 | static void __init free_command_buffer(struct amd_iommu *iommu) |
@@ -858,7 +868,7 @@ static int __init init_iommu_all(struct acpi_table_header *table) | |||
858 | switch (*p) { | 868 | switch (*p) { |
859 | case ACPI_IVHD_TYPE: | 869 | case ACPI_IVHD_TYPE: |
860 | 870 | ||
861 | DUMP_printk("IOMMU: device: %02x:%02x.%01x cap: %04x " | 871 | DUMP_printk("device: %02x:%02x.%01x cap: %04x " |
862 | "seg: %d flags: %01x info %04x\n", | 872 | "seg: %d flags: %01x info %04x\n", |
863 | PCI_BUS(h->devid), PCI_SLOT(h->devid), | 873 | PCI_BUS(h->devid), PCI_SLOT(h->devid), |
864 | PCI_FUNC(h->devid), h->cap_ptr, | 874 | PCI_FUNC(h->devid), h->cap_ptr, |
@@ -902,7 +912,7 @@ static int __init iommu_setup_msi(struct amd_iommu *iommu) | |||
902 | 912 | ||
903 | r = request_irq(iommu->dev->irq, amd_iommu_int_handler, | 913 | r = request_irq(iommu->dev->irq, amd_iommu_int_handler, |
904 | IRQF_SAMPLE_RANDOM, | 914 | IRQF_SAMPLE_RANDOM, |
905 | "AMD IOMMU", | 915 | "AMD-Vi", |
906 | NULL); | 916 | NULL); |
907 | 917 | ||
908 | if (r) { | 918 | if (r) { |
@@ -1150,7 +1160,7 @@ int __init amd_iommu_init(void) | |||
1150 | 1160 | ||
1151 | 1161 | ||
1152 | if (no_iommu) { | 1162 | if (no_iommu) { |
1153 | printk(KERN_INFO "AMD IOMMU disabled by kernel command line\n"); | 1163 | printk(KERN_INFO "AMD-Vi disabled by kernel command line\n"); |
1154 | return 0; | 1164 | return 0; |
1155 | } | 1165 | } |
1156 | 1166 | ||
@@ -1242,22 +1252,28 @@ int __init amd_iommu_init(void) | |||
1242 | if (ret) | 1252 | if (ret) |
1243 | goto free; | 1253 | goto free; |
1244 | 1254 | ||
1245 | ret = amd_iommu_init_dma_ops(); | 1255 | if (iommu_pass_through) |
1256 | ret = amd_iommu_init_passthrough(); | ||
1257 | else | ||
1258 | ret = amd_iommu_init_dma_ops(); | ||
1246 | if (ret) | 1259 | if (ret) |
1247 | goto free; | 1260 | goto free; |
1248 | 1261 | ||
1249 | enable_iommus(); | 1262 | enable_iommus(); |
1250 | 1263 | ||
1251 | printk(KERN_INFO "AMD IOMMU: device isolation "); | 1264 | if (iommu_pass_through) |
1265 | goto out; | ||
1266 | |||
1267 | printk(KERN_INFO "AMD-Vi: device isolation "); | ||
1252 | if (amd_iommu_isolate) | 1268 | if (amd_iommu_isolate) |
1253 | printk("enabled\n"); | 1269 | printk("enabled\n"); |
1254 | else | 1270 | else |
1255 | printk("disabled\n"); | 1271 | printk("disabled\n"); |
1256 | 1272 | ||
1257 | if (amd_iommu_unmap_flush) | 1273 | if (amd_iommu_unmap_flush) |
1258 | printk(KERN_INFO "AMD IOMMU: IO/TLB flush on unmap enabled\n"); | 1274 | printk(KERN_INFO "AMD-Vi: IO/TLB flush on unmap enabled\n"); |
1259 | else | 1275 | else |
1260 | printk(KERN_INFO "AMD IOMMU: Lazy IO/TLB flushing enabled\n"); | 1276 | printk(KERN_INFO "AMD-Vi: Lazy IO/TLB flushing enabled\n"); |
1261 | 1277 | ||
1262 | out: | 1278 | out: |
1263 | return ret; | 1279 | return ret; |
diff --git a/arch/x86/kernel/aperture_64.c b/arch/x86/kernel/aperture_64.c index 676debfc1702..128111d8ffe0 100644 --- a/arch/x86/kernel/aperture_64.c +++ b/arch/x86/kernel/aperture_64.c | |||
@@ -20,6 +20,7 @@ | |||
20 | #include <linux/bitops.h> | 20 | #include <linux/bitops.h> |
21 | #include <linux/ioport.h> | 21 | #include <linux/ioport.h> |
22 | #include <linux/suspend.h> | 22 | #include <linux/suspend.h> |
23 | #include <linux/kmemleak.h> | ||
23 | #include <asm/e820.h> | 24 | #include <asm/e820.h> |
24 | #include <asm/io.h> | 25 | #include <asm/io.h> |
25 | #include <asm/iommu.h> | 26 | #include <asm/iommu.h> |
@@ -94,6 +95,11 @@ static u32 __init allocate_aperture(void) | |||
94 | * code for safe | 95 | * code for safe |
95 | */ | 96 | */ |
96 | p = __alloc_bootmem_nopanic(aper_size, aper_size, 512ULL<<20); | 97 | p = __alloc_bootmem_nopanic(aper_size, aper_size, 512ULL<<20); |
98 | /* | ||
99 | * Kmemleak should not scan this block as it may not be mapped via the | ||
100 | * kernel direct mapping. | ||
101 | */ | ||
102 | kmemleak_ignore(p); | ||
97 | if (!p || __pa(p)+aper_size > 0xffffffff) { | 103 | if (!p || __pa(p)+aper_size > 0xffffffff) { |
98 | printk(KERN_ERR | 104 | printk(KERN_ERR |
99 | "Cannot allocate aperture memory hole (%p,%uK)\n", | 105 | "Cannot allocate aperture memory hole (%p,%uK)\n", |
diff --git a/arch/x86/kernel/apic/nmi.c b/arch/x86/kernel/apic/nmi.c index b3025b43b63a..db7220220d09 100644 --- a/arch/x86/kernel/apic/nmi.c +++ b/arch/x86/kernel/apic/nmi.c | |||
@@ -39,7 +39,7 @@ | |||
39 | int unknown_nmi_panic; | 39 | int unknown_nmi_panic; |
40 | int nmi_watchdog_enabled; | 40 | int nmi_watchdog_enabled; |
41 | 41 | ||
42 | static cpumask_var_t backtrace_mask; | 42 | static cpumask_t backtrace_mask __read_mostly; |
43 | 43 | ||
44 | /* nmi_active: | 44 | /* nmi_active: |
45 | * >0: the lapic NMI watchdog is active, but can be disabled | 45 | * >0: the lapic NMI watchdog is active, but can be disabled |
@@ -138,7 +138,6 @@ int __init check_nmi_watchdog(void) | |||
138 | if (!prev_nmi_count) | 138 | if (!prev_nmi_count) |
139 | goto error; | 139 | goto error; |
140 | 140 | ||
141 | alloc_cpumask_var(&backtrace_mask, GFP_KERNEL|__GFP_ZERO); | ||
142 | printk(KERN_INFO "Testing NMI watchdog ... "); | 141 | printk(KERN_INFO "Testing NMI watchdog ... "); |
143 | 142 | ||
144 | #ifdef CONFIG_SMP | 143 | #ifdef CONFIG_SMP |
@@ -415,14 +414,17 @@ nmi_watchdog_tick(struct pt_regs *regs, unsigned reason) | |||
415 | } | 414 | } |
416 | 415 | ||
417 | /* We can be called before check_nmi_watchdog, hence NULL check. */ | 416 | /* We can be called before check_nmi_watchdog, hence NULL check. */ |
418 | if (backtrace_mask != NULL && cpumask_test_cpu(cpu, backtrace_mask)) { | 417 | if (cpumask_test_cpu(cpu, &backtrace_mask)) { |
419 | static DEFINE_SPINLOCK(lock); /* Serialise the printks */ | 418 | static DEFINE_SPINLOCK(lock); /* Serialise the printks */ |
420 | 419 | ||
421 | spin_lock(&lock); | 420 | spin_lock(&lock); |
422 | printk(KERN_WARNING "NMI backtrace for cpu %d\n", cpu); | 421 | printk(KERN_WARNING "NMI backtrace for cpu %d\n", cpu); |
422 | show_regs(regs); | ||
423 | dump_stack(); | 423 | dump_stack(); |
424 | spin_unlock(&lock); | 424 | spin_unlock(&lock); |
425 | cpumask_clear_cpu(cpu, backtrace_mask); | 425 | cpumask_clear_cpu(cpu, &backtrace_mask); |
426 | |||
427 | rc = 1; | ||
426 | } | 428 | } |
427 | 429 | ||
428 | /* Could check oops_in_progress here too, but it's safer not to */ | 430 | /* Could check oops_in_progress here too, but it's safer not to */ |
@@ -552,14 +554,18 @@ int do_nmi_callback(struct pt_regs *regs, int cpu) | |||
552 | return 0; | 554 | return 0; |
553 | } | 555 | } |
554 | 556 | ||
555 | void __trigger_all_cpu_backtrace(void) | 557 | void arch_trigger_all_cpu_backtrace(void) |
556 | { | 558 | { |
557 | int i; | 559 | int i; |
558 | 560 | ||
559 | cpumask_copy(backtrace_mask, cpu_online_mask); | 561 | cpumask_copy(&backtrace_mask, cpu_online_mask); |
562 | |||
563 | printk(KERN_INFO "sending NMI to all CPUs:\n"); | ||
564 | apic->send_IPI_all(NMI_VECTOR); | ||
565 | |||
560 | /* Wait for up to 10 seconds for all CPUs to do the backtrace */ | 566 | /* Wait for up to 10 seconds for all CPUs to do the backtrace */ |
561 | for (i = 0; i < 10 * 1000; i++) { | 567 | for (i = 0; i < 10 * 1000; i++) { |
562 | if (cpumask_empty(backtrace_mask)) | 568 | if (cpumask_empty(&backtrace_mask)) |
563 | break; | 569 | break; |
564 | mdelay(1); | 570 | mdelay(1); |
565 | } | 571 | } |
diff --git a/arch/x86/kernel/apic/probe_64.c b/arch/x86/kernel/apic/probe_64.c index bc3e880f9b82..fcec2f1d34a1 100644 --- a/arch/x86/kernel/apic/probe_64.c +++ b/arch/x86/kernel/apic/probe_64.c | |||
@@ -44,6 +44,11 @@ static struct apic *apic_probe[] __initdata = { | |||
44 | NULL, | 44 | NULL, |
45 | }; | 45 | }; |
46 | 46 | ||
47 | static int apicid_phys_pkg_id(int initial_apic_id, int index_msb) | ||
48 | { | ||
49 | return hard_smp_processor_id() >> index_msb; | ||
50 | } | ||
51 | |||
47 | /* | 52 | /* |
48 | * Check the APIC IDs in bios_cpu_apicid and choose the APIC mode. | 53 | * Check the APIC IDs in bios_cpu_apicid and choose the APIC mode. |
49 | */ | 54 | */ |
@@ -69,6 +74,11 @@ void __init default_setup_apic_routing(void) | |||
69 | printk(KERN_INFO "Setting APIC routing to %s\n", apic->name); | 74 | printk(KERN_INFO "Setting APIC routing to %s\n", apic->name); |
70 | } | 75 | } |
71 | 76 | ||
77 | if (is_vsmp_box()) { | ||
78 | /* need to update phys_pkg_id */ | ||
79 | apic->phys_pkg_id = apicid_phys_pkg_id; | ||
80 | } | ||
81 | |||
72 | /* | 82 | /* |
73 | * Now that apic routing model is selected, configure the | 83 | * Now that apic routing model is selected, configure the |
74 | * fault handling for intr remapping. | 84 | * fault handling for intr remapping. |
diff --git a/arch/x86/kernel/cpu/Makefile b/arch/x86/kernel/cpu/Makefile index 3efcb2b96a15..c1f253dac155 100644 --- a/arch/x86/kernel/cpu/Makefile +++ b/arch/x86/kernel/cpu/Makefile | |||
@@ -7,6 +7,10 @@ ifdef CONFIG_FUNCTION_TRACER | |||
7 | CFLAGS_REMOVE_common.o = -pg | 7 | CFLAGS_REMOVE_common.o = -pg |
8 | endif | 8 | endif |
9 | 9 | ||
10 | # Make sure load_percpu_segment has no stackprotector | ||
11 | nostackp := $(call cc-option, -fno-stack-protector) | ||
12 | CFLAGS_common.o := $(nostackp) | ||
13 | |||
10 | obj-y := intel_cacheinfo.o addon_cpuid_features.o | 14 | obj-y := intel_cacheinfo.o addon_cpuid_features.o |
11 | obj-y += proc.o capflags.o powerflags.o common.o | 15 | obj-y += proc.o capflags.o powerflags.o common.o |
12 | obj-y += vmware.o hypervisor.o | 16 | obj-y += vmware.o hypervisor.o |
diff --git a/arch/x86/kernel/head_32.S b/arch/x86/kernel/head_32.S index 0d98a01cbdb2..cc827ac9e8d3 100644 --- a/arch/x86/kernel/head_32.S +++ b/arch/x86/kernel/head_32.S | |||
@@ -261,9 +261,7 @@ page_pde_offset = (__PAGE_OFFSET >> 20); | |||
261 | * which will be freed later | 261 | * which will be freed later |
262 | */ | 262 | */ |
263 | 263 | ||
264 | #ifndef CONFIG_HOTPLUG_CPU | 264 | __CPUINIT |
265 | .section .init.text,"ax",@progbits | ||
266 | #endif | ||
267 | 265 | ||
268 | #ifdef CONFIG_SMP | 266 | #ifdef CONFIG_SMP |
269 | ENTRY(startup_32_smp) | 267 | ENTRY(startup_32_smp) |
@@ -602,11 +600,7 @@ ignore_int: | |||
602 | #endif | 600 | #endif |
603 | iret | 601 | iret |
604 | 602 | ||
605 | #ifndef CONFIG_HOTPLUG_CPU | ||
606 | __CPUINITDATA | ||
607 | #else | ||
608 | __REFDATA | 603 | __REFDATA |
609 | #endif | ||
610 | .align 4 | 604 | .align 4 |
611 | ENTRY(initial_code) | 605 | ENTRY(initial_code) |
612 | .long i386_start_kernel | 606 | .long i386_start_kernel |
diff --git a/arch/x86/kernel/pci-dma.c b/arch/x86/kernel/pci-dma.c index 1a041bcf506b..d71c8655905b 100644 --- a/arch/x86/kernel/pci-dma.c +++ b/arch/x86/kernel/pci-dma.c | |||
@@ -3,6 +3,7 @@ | |||
3 | #include <linux/dmar.h> | 3 | #include <linux/dmar.h> |
4 | #include <linux/bootmem.h> | 4 | #include <linux/bootmem.h> |
5 | #include <linux/pci.h> | 5 | #include <linux/pci.h> |
6 | #include <linux/kmemleak.h> | ||
6 | 7 | ||
7 | #include <asm/proto.h> | 8 | #include <asm/proto.h> |
8 | #include <asm/dma.h> | 9 | #include <asm/dma.h> |
@@ -32,7 +33,14 @@ int no_iommu __read_mostly; | |||
32 | /* Set this to 1 if there is a HW IOMMU in the system */ | 33 | /* Set this to 1 if there is a HW IOMMU in the system */ |
33 | int iommu_detected __read_mostly = 0; | 34 | int iommu_detected __read_mostly = 0; |
34 | 35 | ||
35 | int iommu_pass_through; | 36 | /* |
37 | * This variable becomes 1 if iommu=pt is passed on the kernel command line. | ||
38 | * If this variable is 1, IOMMU implementations do no DMA ranslation for | ||
39 | * devices and allow every device to access to whole physical memory. This is | ||
40 | * useful if a user want to use an IOMMU only for KVM device assignment to | ||
41 | * guests and not for driver dma translation. | ||
42 | */ | ||
43 | int iommu_pass_through __read_mostly; | ||
36 | 44 | ||
37 | dma_addr_t bad_dma_address __read_mostly = 0; | 45 | dma_addr_t bad_dma_address __read_mostly = 0; |
38 | EXPORT_SYMBOL(bad_dma_address); | 46 | EXPORT_SYMBOL(bad_dma_address); |
@@ -88,6 +96,11 @@ void __init dma32_reserve_bootmem(void) | |||
88 | size = roundup(dma32_bootmem_size, align); | 96 | size = roundup(dma32_bootmem_size, align); |
89 | dma32_bootmem_ptr = __alloc_bootmem_nopanic(size, align, | 97 | dma32_bootmem_ptr = __alloc_bootmem_nopanic(size, align, |
90 | 512ULL<<20); | 98 | 512ULL<<20); |
99 | /* | ||
100 | * Kmemleak should not scan this block as it may not be mapped via the | ||
101 | * kernel direct mapping. | ||
102 | */ | ||
103 | kmemleak_ignore(dma32_bootmem_ptr); | ||
91 | if (dma32_bootmem_ptr) | 104 | if (dma32_bootmem_ptr) |
92 | dma32_bootmem_size = size; | 105 | dma32_bootmem_size = size; |
93 | else | 106 | else |
@@ -147,7 +160,7 @@ again: | |||
147 | return NULL; | 160 | return NULL; |
148 | 161 | ||
149 | addr = page_to_phys(page); | 162 | addr = page_to_phys(page); |
150 | if (!is_buffer_dma_capable(dma_mask, addr, size)) { | 163 | if (addr + size > dma_mask) { |
151 | __free_pages(page, get_order(size)); | 164 | __free_pages(page, get_order(size)); |
152 | 165 | ||
153 | if (dma_mask < DMA_BIT_MASK(32) && !(flag & GFP_DMA)) { | 166 | if (dma_mask < DMA_BIT_MASK(32) && !(flag & GFP_DMA)) { |
diff --git a/arch/x86/kernel/pci-gart_64.c b/arch/x86/kernel/pci-gart_64.c index d2e56b8f48e7..98a827ee9ed7 100644 --- a/arch/x86/kernel/pci-gart_64.c +++ b/arch/x86/kernel/pci-gart_64.c | |||
@@ -190,14 +190,13 @@ static void iommu_full(struct device *dev, size_t size, int dir) | |||
190 | static inline int | 190 | static inline int |
191 | need_iommu(struct device *dev, unsigned long addr, size_t size) | 191 | need_iommu(struct device *dev, unsigned long addr, size_t size) |
192 | { | 192 | { |
193 | return force_iommu || | 193 | return force_iommu || !dma_capable(dev, addr, size); |
194 | !is_buffer_dma_capable(*dev->dma_mask, addr, size); | ||
195 | } | 194 | } |
196 | 195 | ||
197 | static inline int | 196 | static inline int |
198 | nonforced_iommu(struct device *dev, unsigned long addr, size_t size) | 197 | nonforced_iommu(struct device *dev, unsigned long addr, size_t size) |
199 | { | 198 | { |
200 | return !is_buffer_dma_capable(*dev->dma_mask, addr, size); | 199 | return !dma_capable(dev, addr, size); |
201 | } | 200 | } |
202 | 201 | ||
203 | /* Map a single continuous physical area into the IOMMU. | 202 | /* Map a single continuous physical area into the IOMMU. |
diff --git a/arch/x86/kernel/pci-nommu.c b/arch/x86/kernel/pci-nommu.c index 71d412a09f30..a3933d4330cd 100644 --- a/arch/x86/kernel/pci-nommu.c +++ b/arch/x86/kernel/pci-nommu.c | |||
@@ -14,7 +14,7 @@ | |||
14 | static int | 14 | static int |
15 | check_addr(char *name, struct device *hwdev, dma_addr_t bus, size_t size) | 15 | check_addr(char *name, struct device *hwdev, dma_addr_t bus, size_t size) |
16 | { | 16 | { |
17 | if (hwdev && !is_buffer_dma_capable(*hwdev->dma_mask, bus, size)) { | 17 | if (hwdev && !dma_capable(hwdev, bus, size)) { |
18 | if (*hwdev->dma_mask >= DMA_BIT_MASK(32)) | 18 | if (*hwdev->dma_mask >= DMA_BIT_MASK(32)) |
19 | printk(KERN_ERR | 19 | printk(KERN_ERR |
20 | "nommu_%s: overflow %Lx+%zu of device mask %Lx\n", | 20 | "nommu_%s: overflow %Lx+%zu of device mask %Lx\n", |
@@ -79,12 +79,29 @@ static void nommu_free_coherent(struct device *dev, size_t size, void *vaddr, | |||
79 | free_pages((unsigned long)vaddr, get_order(size)); | 79 | free_pages((unsigned long)vaddr, get_order(size)); |
80 | } | 80 | } |
81 | 81 | ||
82 | static void nommu_sync_single_for_device(struct device *dev, | ||
83 | dma_addr_t addr, size_t size, | ||
84 | enum dma_data_direction dir) | ||
85 | { | ||
86 | flush_write_buffers(); | ||
87 | } | ||
88 | |||
89 | |||
90 | static void nommu_sync_sg_for_device(struct device *dev, | ||
91 | struct scatterlist *sg, int nelems, | ||
92 | enum dma_data_direction dir) | ||
93 | { | ||
94 | flush_write_buffers(); | ||
95 | } | ||
96 | |||
82 | struct dma_map_ops nommu_dma_ops = { | 97 | struct dma_map_ops nommu_dma_ops = { |
83 | .alloc_coherent = dma_generic_alloc_coherent, | 98 | .alloc_coherent = dma_generic_alloc_coherent, |
84 | .free_coherent = nommu_free_coherent, | 99 | .free_coherent = nommu_free_coherent, |
85 | .map_sg = nommu_map_sg, | 100 | .map_sg = nommu_map_sg, |
86 | .map_page = nommu_map_page, | 101 | .map_page = nommu_map_page, |
87 | .is_phys = 1, | 102 | .sync_single_for_device = nommu_sync_single_for_device, |
103 | .sync_sg_for_device = nommu_sync_sg_for_device, | ||
104 | .is_phys = 1, | ||
88 | }; | 105 | }; |
89 | 106 | ||
90 | void __init no_iommu_init(void) | 107 | void __init no_iommu_init(void) |
diff --git a/arch/x86/kernel/pci-swiotlb.c b/arch/x86/kernel/pci-swiotlb.c index 6af96ee44200..e8a35016115f 100644 --- a/arch/x86/kernel/pci-swiotlb.c +++ b/arch/x86/kernel/pci-swiotlb.c | |||
@@ -13,31 +13,6 @@ | |||
13 | 13 | ||
14 | int swiotlb __read_mostly; | 14 | int swiotlb __read_mostly; |
15 | 15 | ||
16 | void * __init swiotlb_alloc_boot(size_t size, unsigned long nslabs) | ||
17 | { | ||
18 | return alloc_bootmem_low_pages(size); | ||
19 | } | ||
20 | |||
21 | void *swiotlb_alloc(unsigned order, unsigned long nslabs) | ||
22 | { | ||
23 | return (void *)__get_free_pages(GFP_DMA | __GFP_NOWARN, order); | ||
24 | } | ||
25 | |||
26 | dma_addr_t swiotlb_phys_to_bus(struct device *hwdev, phys_addr_t paddr) | ||
27 | { | ||
28 | return paddr; | ||
29 | } | ||
30 | |||
31 | phys_addr_t swiotlb_bus_to_phys(struct device *hwdev, dma_addr_t baddr) | ||
32 | { | ||
33 | return baddr; | ||
34 | } | ||
35 | |||
36 | int __weak swiotlb_arch_range_needs_mapping(phys_addr_t paddr, size_t size) | ||
37 | { | ||
38 | return 0; | ||
39 | } | ||
40 | |||
41 | static void *x86_swiotlb_alloc_coherent(struct device *hwdev, size_t size, | 16 | static void *x86_swiotlb_alloc_coherent(struct device *hwdev, size_t size, |
42 | dma_addr_t *dma_handle, gfp_t flags) | 17 | dma_addr_t *dma_handle, gfp_t flags) |
43 | { | 18 | { |
diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c index 994dd6a4a2a0..071166a4ba83 100644 --- a/arch/x86/kernel/process.c +++ b/arch/x86/kernel/process.c | |||
@@ -519,16 +519,12 @@ static void c1e_idle(void) | |||
519 | if (!cpumask_test_cpu(cpu, c1e_mask)) { | 519 | if (!cpumask_test_cpu(cpu, c1e_mask)) { |
520 | cpumask_set_cpu(cpu, c1e_mask); | 520 | cpumask_set_cpu(cpu, c1e_mask); |
521 | /* | 521 | /* |
522 | * Force broadcast so ACPI can not interfere. Needs | 522 | * Force broadcast so ACPI can not interfere. |
523 | * to run with interrupts enabled as it uses | ||
524 | * smp_function_call. | ||
525 | */ | 523 | */ |
526 | local_irq_enable(); | ||
527 | clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE, | 524 | clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE, |
528 | &cpu); | 525 | &cpu); |
529 | printk(KERN_INFO "Switch to broadcast mode on CPU%d\n", | 526 | printk(KERN_INFO "Switch to broadcast mode on CPU%d\n", |
530 | cpu); | 527 | cpu); |
531 | local_irq_disable(); | ||
532 | } | 528 | } |
533 | clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu); | 529 | clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu); |
534 | 530 | ||
diff --git a/arch/x86/kernel/signal.c b/arch/x86/kernel/signal.c index 4c578751e94e..81e58238c4ce 100644 --- a/arch/x86/kernel/signal.c +++ b/arch/x86/kernel/signal.c | |||
@@ -869,6 +869,8 @@ do_notify_resume(struct pt_regs *regs, void *unused, __u32 thread_info_flags) | |||
869 | if (thread_info_flags & _TIF_NOTIFY_RESUME) { | 869 | if (thread_info_flags & _TIF_NOTIFY_RESUME) { |
870 | clear_thread_flag(TIF_NOTIFY_RESUME); | 870 | clear_thread_flag(TIF_NOTIFY_RESUME); |
871 | tracehook_notify_resume(regs); | 871 | tracehook_notify_resume(regs); |
872 | if (current->replacement_session_keyring) | ||
873 | key_replace_session_keyring(); | ||
872 | } | 874 | } |
873 | 875 | ||
874 | #ifdef CONFIG_X86_32 | 876 | #ifdef CONFIG_X86_32 |
diff --git a/arch/x86/kernel/vmlinux.lds.S b/arch/x86/kernel/vmlinux.lds.S index 78d185d797de..9fc178255c04 100644 --- a/arch/x86/kernel/vmlinux.lds.S +++ b/arch/x86/kernel/vmlinux.lds.S | |||
@@ -46,11 +46,10 @@ PHDRS { | |||
46 | data PT_LOAD FLAGS(7); /* RWE */ | 46 | data PT_LOAD FLAGS(7); /* RWE */ |
47 | #ifdef CONFIG_X86_64 | 47 | #ifdef CONFIG_X86_64 |
48 | user PT_LOAD FLAGS(7); /* RWE */ | 48 | user PT_LOAD FLAGS(7); /* RWE */ |
49 | data.init PT_LOAD FLAGS(7); /* RWE */ | ||
50 | #ifdef CONFIG_SMP | 49 | #ifdef CONFIG_SMP |
51 | percpu PT_LOAD FLAGS(7); /* RWE */ | 50 | percpu PT_LOAD FLAGS(7); /* RWE */ |
52 | #endif | 51 | #endif |
53 | data.init2 PT_LOAD FLAGS(7); /* RWE */ | 52 | init PT_LOAD FLAGS(7); /* RWE */ |
54 | #endif | 53 | #endif |
55 | note PT_NOTE FLAGS(0); /* ___ */ | 54 | note PT_NOTE FLAGS(0); /* ___ */ |
56 | } | 55 | } |
@@ -103,65 +102,43 @@ SECTIONS | |||
103 | __stop___ex_table = .; | 102 | __stop___ex_table = .; |
104 | } :text = 0x9090 | 103 | } :text = 0x9090 |
105 | 104 | ||
106 | RODATA | 105 | RO_DATA(PAGE_SIZE) |
107 | 106 | ||
108 | /* Data */ | 107 | /* Data */ |
109 | . = ALIGN(PAGE_SIZE); | ||
110 | .data : AT(ADDR(.data) - LOAD_OFFSET) { | 108 | .data : AT(ADDR(.data) - LOAD_OFFSET) { |
111 | /* Start of data section */ | 109 | /* Start of data section */ |
112 | _sdata = .; | 110 | _sdata = .; |
113 | DATA_DATA | 111 | |
114 | CONSTRUCTORS | 112 | /* init_task */ |
115 | } :data | 113 | INIT_TASK_DATA(THREAD_SIZE) |
116 | 114 | ||
117 | #ifdef CONFIG_X86_32 | 115 | #ifdef CONFIG_X86_32 |
118 | /* 32 bit has nosave before _edata */ | 116 | /* 32 bit has nosave before _edata */ |
119 | . = ALIGN(PAGE_SIZE); | 117 | NOSAVE_DATA |
120 | .data_nosave : AT(ADDR(.data_nosave) - LOAD_OFFSET) { | ||
121 | __nosave_begin = .; | ||
122 | *(.data.nosave) | ||
123 | . = ALIGN(PAGE_SIZE); | ||
124 | __nosave_end = .; | ||
125 | } | ||
126 | #endif | 118 | #endif |
127 | 119 | ||
128 | . = ALIGN(PAGE_SIZE); | 120 | PAGE_ALIGNED_DATA(PAGE_SIZE) |
129 | .data.page_aligned : AT(ADDR(.data.page_aligned) - LOAD_OFFSET) { | ||
130 | *(.data.page_aligned) | ||
131 | *(.data.idt) | 121 | *(.data.idt) |
132 | } | ||
133 | 122 | ||
134 | #ifdef CONFIG_X86_32 | 123 | CACHELINE_ALIGNED_DATA(CONFIG_X86_L1_CACHE_BYTES) |
135 | . = ALIGN(32); | ||
136 | #else | ||
137 | . = ALIGN(PAGE_SIZE); | ||
138 | . = ALIGN(CONFIG_X86_L1_CACHE_BYTES); | ||
139 | #endif | ||
140 | .data.cacheline_aligned : | ||
141 | AT(ADDR(.data.cacheline_aligned) - LOAD_OFFSET) { | ||
142 | *(.data.cacheline_aligned) | ||
143 | } | ||
144 | 124 | ||
145 | /* rarely changed data like cpu maps */ | 125 | DATA_DATA |
146 | #ifdef CONFIG_X86_32 | 126 | CONSTRUCTORS |
147 | . = ALIGN(32); | 127 | |
148 | #else | 128 | /* rarely changed data like cpu maps */ |
149 | . = ALIGN(CONFIG_X86_INTERNODE_CACHE_BYTES); | 129 | READ_MOSTLY_DATA(CONFIG_X86_INTERNODE_CACHE_BYTES) |
150 | #endif | ||
151 | .data.read_mostly : AT(ADDR(.data.read_mostly) - LOAD_OFFSET) { | ||
152 | *(.data.read_mostly) | ||
153 | 130 | ||
154 | /* End of data section */ | 131 | /* End of data section */ |
155 | _edata = .; | 132 | _edata = .; |
156 | } | 133 | } :data |
157 | 134 | ||
158 | #ifdef CONFIG_X86_64 | 135 | #ifdef CONFIG_X86_64 |
159 | 136 | ||
160 | #define VSYSCALL_ADDR (-10*1024*1024) | 137 | #define VSYSCALL_ADDR (-10*1024*1024) |
161 | #define VSYSCALL_PHYS_ADDR ((LOADADDR(.data.read_mostly) + \ | 138 | #define VSYSCALL_PHYS_ADDR ((LOADADDR(.data) + SIZEOF(.data) + \ |
162 | SIZEOF(.data.read_mostly) + 4095) & ~(4095)) | 139 | PAGE_SIZE - 1) & ~(PAGE_SIZE - 1)) |
163 | #define VSYSCALL_VIRT_ADDR ((ADDR(.data.read_mostly) + \ | 140 | #define VSYSCALL_VIRT_ADDR ((ADDR(.data) + SIZEOF(.data) + \ |
164 | SIZEOF(.data.read_mostly) + 4095) & ~(4095)) | 141 | PAGE_SIZE - 1) & ~(PAGE_SIZE - 1)) |
165 | 142 | ||
166 | #define VLOAD_OFFSET (VSYSCALL_ADDR - VSYSCALL_PHYS_ADDR) | 143 | #define VLOAD_OFFSET (VSYSCALL_ADDR - VSYSCALL_PHYS_ADDR) |
167 | #define VLOAD(x) (ADDR(x) - VLOAD_OFFSET) | 144 | #define VLOAD(x) (ADDR(x) - VLOAD_OFFSET) |
@@ -227,35 +204,29 @@ SECTIONS | |||
227 | 204 | ||
228 | #endif /* CONFIG_X86_64 */ | 205 | #endif /* CONFIG_X86_64 */ |
229 | 206 | ||
230 | /* init_task */ | 207 | /* Init code and data - will be freed after init */ |
231 | . = ALIGN(THREAD_SIZE); | 208 | . = ALIGN(PAGE_SIZE); |
232 | .data.init_task : AT(ADDR(.data.init_task) - LOAD_OFFSET) { | 209 | .init.begin : AT(ADDR(.init.begin) - LOAD_OFFSET) { |
233 | *(.data.init_task) | 210 | __init_begin = .; /* paired with __init_end */ |
234 | } | 211 | } |
235 | #ifdef CONFIG_X86_64 | ||
236 | :data.init | ||
237 | #endif | ||
238 | 212 | ||
213 | #if defined(CONFIG_X86_64) && defined(CONFIG_SMP) | ||
239 | /* | 214 | /* |
240 | * smp_locks might be freed after init | 215 | * percpu offsets are zero-based on SMP. PERCPU_VADDR() changes the |
241 | * start/end must be page aligned | 216 | * output PHDR, so the next output section - .init.text - should |
217 | * start another segment - init. | ||
242 | */ | 218 | */ |
243 | . = ALIGN(PAGE_SIZE); | 219 | PERCPU_VADDR(0, :percpu) |
244 | .smp_locks : AT(ADDR(.smp_locks) - LOAD_OFFSET) { | 220 | #endif |
245 | __smp_locks = .; | ||
246 | *(.smp_locks) | ||
247 | __smp_locks_end = .; | ||
248 | . = ALIGN(PAGE_SIZE); | ||
249 | } | ||
250 | 221 | ||
251 | /* Init code and data - will be freed after init */ | ||
252 | . = ALIGN(PAGE_SIZE); | ||
253 | .init.text : AT(ADDR(.init.text) - LOAD_OFFSET) { | 222 | .init.text : AT(ADDR(.init.text) - LOAD_OFFSET) { |
254 | __init_begin = .; /* paired with __init_end */ | ||
255 | _sinittext = .; | 223 | _sinittext = .; |
256 | INIT_TEXT | 224 | INIT_TEXT |
257 | _einittext = .; | 225 | _einittext = .; |
258 | } | 226 | } |
227 | #ifdef CONFIG_X86_64 | ||
228 | :init | ||
229 | #endif | ||
259 | 230 | ||
260 | .init.data : AT(ADDR(.init.data) - LOAD_OFFSET) { | 231 | .init.data : AT(ADDR(.init.data) - LOAD_OFFSET) { |
261 | INIT_DATA | 232 | INIT_DATA |
@@ -326,17 +297,7 @@ SECTIONS | |||
326 | } | 297 | } |
327 | #endif | 298 | #endif |
328 | 299 | ||
329 | #if defined(CONFIG_X86_64) && defined(CONFIG_SMP) | 300 | #if !defined(CONFIG_X86_64) || !defined(CONFIG_SMP) |
330 | /* | ||
331 | * percpu offsets are zero-based on SMP. PERCPU_VADDR() changes the | ||
332 | * output PHDR, so the next output section - __data_nosave - should | ||
333 | * start another section data.init2. Also, pda should be at the head of | ||
334 | * percpu area. Preallocate it and define the percpu offset symbol | ||
335 | * so that it can be accessed as a percpu variable. | ||
336 | */ | ||
337 | . = ALIGN(PAGE_SIZE); | ||
338 | PERCPU_VADDR(0, :percpu) | ||
339 | #else | ||
340 | PERCPU(PAGE_SIZE) | 301 | PERCPU(PAGE_SIZE) |
341 | #endif | 302 | #endif |
342 | 303 | ||
@@ -347,15 +308,22 @@ SECTIONS | |||
347 | __init_end = .; | 308 | __init_end = .; |
348 | } | 309 | } |
349 | 310 | ||
311 | /* | ||
312 | * smp_locks might be freed after init | ||
313 | * start/end must be page aligned | ||
314 | */ | ||
315 | . = ALIGN(PAGE_SIZE); | ||
316 | .smp_locks : AT(ADDR(.smp_locks) - LOAD_OFFSET) { | ||
317 | __smp_locks = .; | ||
318 | *(.smp_locks) | ||
319 | __smp_locks_end = .; | ||
320 | . = ALIGN(PAGE_SIZE); | ||
321 | } | ||
322 | |||
350 | #ifdef CONFIG_X86_64 | 323 | #ifdef CONFIG_X86_64 |
351 | .data_nosave : AT(ADDR(.data_nosave) - LOAD_OFFSET) { | 324 | .data_nosave : AT(ADDR(.data_nosave) - LOAD_OFFSET) { |
352 | . = ALIGN(PAGE_SIZE); | 325 | NOSAVE_DATA |
353 | __nosave_begin = .; | 326 | } |
354 | *(.data.nosave) | ||
355 | . = ALIGN(PAGE_SIZE); | ||
356 | __nosave_end = .; | ||
357 | } :data.init2 | ||
358 | /* use another section data.init2, see PERCPU_VADDR() above */ | ||
359 | #endif | 327 | #endif |
360 | 328 | ||
361 | /* BSS */ | 329 | /* BSS */ |