2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
9 * Copyright (C) 2006 Qumranet, Inc.
12 * Yaniv Kamay <yaniv@qumranet.com>
13 * Avi Kivity <avi@qumranet.com>
15 * This work is licensed under the terms of the GNU GPL, version 2. See
16 * the COPYING file in the top-level directory.
23 #include <linux/kvm_host.h>
24 #include <linux/types.h>
25 #include <linux/string.h>
27 #include <linux/highmem.h>
28 #include <linux/module.h>
29 #include <linux/swap.h>
30 #include <linux/hugetlb.h>
31 #include <linux/compiler.h>
34 #include <asm/cmpxchg.h>
38 * When setting this variable to true it enables Two-Dimensional-Paging
39 * where the hardware walks 2 page tables:
40 * 1. the guest-virtual to guest-physical
41 * 2. while doing 1. it walks guest-physical to host-physical
42 * If the hardware supports that we don't need to do shadow paging.
44 bool tdp_enabled = false;
51 static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg);
53 static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg) {}
58 #define pgprintk(x...) do { if (dbg) printk(x); } while (0)
59 #define rmap_printk(x...) do { if (dbg) printk(x); } while (0)
63 #define pgprintk(x...) do { } while (0)
64 #define rmap_printk(x...) do { } while (0)
68 #if defined(MMU_DEBUG) || defined(AUDIT)
70 module_param(dbg, bool, 0644);
74 #define ASSERT(x) do { } while (0)
78 printk(KERN_WARNING "assertion failed %s:%d: %s\n", \
79 __FILE__, __LINE__, #x); \
83 #define PT_FIRST_AVAIL_BITS_SHIFT 9
84 #define PT64_SECOND_AVAIL_BITS_SHIFT 52
86 #define VALID_PAGE(x) ((x) != INVALID_PAGE)
88 #define PT64_LEVEL_BITS 9
90 #define PT64_LEVEL_SHIFT(level) \
91 (PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS)
93 #define PT64_LEVEL_MASK(level) \
94 (((1ULL << PT64_LEVEL_BITS) - 1) << PT64_LEVEL_SHIFT(level))
96 #define PT64_INDEX(address, level)\
97 (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1))
100 #define PT32_LEVEL_BITS 10
102 #define PT32_LEVEL_SHIFT(level) \
103 (PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS)
105 #define PT32_LEVEL_MASK(level) \
106 (((1ULL << PT32_LEVEL_BITS) - 1) << PT32_LEVEL_SHIFT(level))
108 #define PT32_INDEX(address, level)\
109 (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1))
112 #define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1))
113 #define PT64_DIR_BASE_ADDR_MASK \
114 (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1))
116 #define PT32_BASE_ADDR_MASK PAGE_MASK
117 #define PT32_DIR_BASE_ADDR_MASK \
118 (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1))
120 #define PT64_PERM_MASK (PT_PRESENT_MASK | PT_WRITABLE_MASK | PT_USER_MASK \
123 #define PFERR_PRESENT_MASK (1U << 0)
124 #define PFERR_WRITE_MASK (1U << 1)
125 #define PFERR_USER_MASK (1U << 2)
126 #define PFERR_FETCH_MASK (1U << 4)
128 #define PT_DIRECTORY_LEVEL 2
129 #define PT_PAGE_TABLE_LEVEL 1
133 #define ACC_EXEC_MASK 1
134 #define ACC_WRITE_MASK PT_WRITABLE_MASK
135 #define ACC_USER_MASK PT_USER_MASK
136 #define ACC_ALL (ACC_EXEC_MASK | ACC_WRITE_MASK | ACC_USER_MASK)
138 #define SHADOW_PT_INDEX(addr, level) PT64_INDEX(addr, level)
140 struct kvm_rmap_desc {
141 u64 *shadow_ptes[RMAP_EXT];
142 struct kvm_rmap_desc *more;
145 static struct kmem_cache *pte_chain_cache;
146 static struct kmem_cache *rmap_desc_cache;
147 static struct kmem_cache *mmu_page_header_cache;
149 static u64 __read_mostly shadow_trap_nonpresent_pte;
150 static u64 __read_mostly shadow_notrap_nonpresent_pte;
151 static u64 __read_mostly shadow_base_present_pte;
152 static u64 __read_mostly shadow_nx_mask;
153 static u64 __read_mostly shadow_x_mask; /* mutual exclusive with nx_mask */
154 static u64 __read_mostly shadow_user_mask;
155 static u64 __read_mostly shadow_accessed_mask;
156 static u64 __read_mostly shadow_dirty_mask;
158 void kvm_mmu_set_nonpresent_ptes(u64 trap_pte, u64 notrap_pte)
160 shadow_trap_nonpresent_pte = trap_pte;
161 shadow_notrap_nonpresent_pte = notrap_pte;
163 EXPORT_SYMBOL_GPL(kvm_mmu_set_nonpresent_ptes);
165 void kvm_mmu_set_base_ptes(u64 base_pte)
167 shadow_base_present_pte = base_pte;
169 EXPORT_SYMBOL_GPL(kvm_mmu_set_base_ptes);
171 void kvm_mmu_set_mask_ptes(u64 user_mask, u64 accessed_mask,
172 u64 dirty_mask, u64 nx_mask, u64 x_mask)
174 shadow_user_mask = user_mask;
175 shadow_accessed_mask = accessed_mask;
176 shadow_dirty_mask = dirty_mask;
177 shadow_nx_mask = nx_mask;
178 shadow_x_mask = x_mask;
180 EXPORT_SYMBOL_GPL(kvm_mmu_set_mask_ptes);
182 static int is_write_protection(struct kvm_vcpu *vcpu)
184 return vcpu->arch.cr0 & X86_CR0_WP;
187 static int is_cpuid_PSE36(void)
192 static int is_nx(struct kvm_vcpu *vcpu)
194 return vcpu->arch.shadow_efer & EFER_NX;
197 static int is_present_pte(unsigned long pte)
199 return pte & PT_PRESENT_MASK;
202 static int is_shadow_present_pte(u64 pte)
204 return pte != shadow_trap_nonpresent_pte
205 && pte != shadow_notrap_nonpresent_pte;
208 static int is_large_pte(u64 pte)
210 return pte & PT_PAGE_SIZE_MASK;
213 static int is_writeble_pte(unsigned long pte)
215 return pte & PT_WRITABLE_MASK;
218 static int is_dirty_pte(unsigned long pte)
220 return pte & shadow_dirty_mask;
223 static int is_rmap_pte(u64 pte)
225 return is_shadow_present_pte(pte);
228 static pfn_t spte_to_pfn(u64 pte)
230 return (pte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT;
233 static gfn_t pse36_gfn_delta(u32 gpte)
235 int shift = 32 - PT32_DIR_PSE36_SHIFT - PAGE_SHIFT;
237 return (gpte & PT32_DIR_PSE36_MASK) << shift;
240 static void set_shadow_pte(u64 *sptep, u64 spte)
243 set_64bit((unsigned long *)sptep, spte);
245 set_64bit((unsigned long long *)sptep, spte);
249 static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache,
250 struct kmem_cache *base_cache, int min)
254 if (cache->nobjs >= min)
256 while (cache->nobjs < ARRAY_SIZE(cache->objects)) {
257 obj = kmem_cache_zalloc(base_cache, GFP_KERNEL);
260 cache->objects[cache->nobjs++] = obj;
265 static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
268 kfree(mc->objects[--mc->nobjs]);
271 static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache *cache,
276 if (cache->nobjs >= min)
278 while (cache->nobjs < ARRAY_SIZE(cache->objects)) {
279 page = alloc_page(GFP_KERNEL);
282 set_page_private(page, 0);
283 cache->objects[cache->nobjs++] = page_address(page);
288 static void mmu_free_memory_cache_page(struct kvm_mmu_memory_cache *mc)
291 free_page((unsigned long)mc->objects[--mc->nobjs]);
294 static int mmu_topup_memory_caches(struct kvm_vcpu *vcpu)
298 r = mmu_topup_memory_cache(&vcpu->arch.mmu_pte_chain_cache,
302 r = mmu_topup_memory_cache(&vcpu->arch.mmu_rmap_desc_cache,
306 r = mmu_topup_memory_cache_page(&vcpu->arch.mmu_page_cache, 8);
309 r = mmu_topup_memory_cache(&vcpu->arch.mmu_page_header_cache,
310 mmu_page_header_cache, 4);
315 static void mmu_free_memory_caches(struct kvm_vcpu *vcpu)
317 mmu_free_memory_cache(&vcpu->arch.mmu_pte_chain_cache);
318 mmu_free_memory_cache(&vcpu->arch.mmu_rmap_desc_cache);
319 mmu_free_memory_cache_page(&vcpu->arch.mmu_page_cache);
320 mmu_free_memory_cache(&vcpu->arch.mmu_page_header_cache);
323 static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc,
329 p = mc->objects[--mc->nobjs];
334 static struct kvm_pte_chain *mmu_alloc_pte_chain(struct kvm_vcpu *vcpu)
336 return mmu_memory_cache_alloc(&vcpu->arch.mmu_pte_chain_cache,
337 sizeof(struct kvm_pte_chain));
340 static void mmu_free_pte_chain(struct kvm_pte_chain *pc)
345 static struct kvm_rmap_desc *mmu_alloc_rmap_desc(struct kvm_vcpu *vcpu)
347 return mmu_memory_cache_alloc(&vcpu->arch.mmu_rmap_desc_cache,
348 sizeof(struct kvm_rmap_desc));
351 static void mmu_free_rmap_desc(struct kvm_rmap_desc *rd)
357 * Return the pointer to the largepage write count for a given
358 * gfn, handling slots that are not large page aligned.
360 static int *slot_largepage_idx(gfn_t gfn, struct kvm_memory_slot *slot)
364 idx = (gfn / KVM_PAGES_PER_HPAGE) -
365 (slot->base_gfn / KVM_PAGES_PER_HPAGE);
366 return &slot->lpage_info[idx].write_count;
369 static void account_shadowed(struct kvm *kvm, gfn_t gfn)
373 write_count = slot_largepage_idx(gfn, gfn_to_memslot(kvm, gfn));
377 static void unaccount_shadowed(struct kvm *kvm, gfn_t gfn)
381 write_count = slot_largepage_idx(gfn, gfn_to_memslot(kvm, gfn));
383 WARN_ON(*write_count < 0);
386 static int has_wrprotected_page(struct kvm *kvm, gfn_t gfn)
388 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
392 largepage_idx = slot_largepage_idx(gfn, slot);
393 return *largepage_idx;
399 static int host_largepage_backed(struct kvm *kvm, gfn_t gfn)
401 struct vm_area_struct *vma;
404 addr = gfn_to_hva(kvm, gfn);
405 if (kvm_is_error_hva(addr))
408 vma = find_vma(current->mm, addr);
409 if (vma && is_vm_hugetlb_page(vma))
415 static int is_largepage_backed(struct kvm_vcpu *vcpu, gfn_t large_gfn)
417 struct kvm_memory_slot *slot;
419 if (has_wrprotected_page(vcpu->kvm, large_gfn))
422 if (!host_largepage_backed(vcpu->kvm, large_gfn))
425 slot = gfn_to_memslot(vcpu->kvm, large_gfn);
426 if (slot && slot->dirty_bitmap)
433 * Take gfn and return the reverse mapping to it.
434 * Note: gfn must be unaliased before this function get called
437 static unsigned long *gfn_to_rmap(struct kvm *kvm, gfn_t gfn, int lpage)
439 struct kvm_memory_slot *slot;
442 slot = gfn_to_memslot(kvm, gfn);
444 return &slot->rmap[gfn - slot->base_gfn];
446 idx = (gfn / KVM_PAGES_PER_HPAGE) -
447 (slot->base_gfn / KVM_PAGES_PER_HPAGE);
449 return &slot->lpage_info[idx].rmap_pde;
453 * Reverse mapping data structures:
455 * If rmapp bit zero is zero, then rmapp point to the shadw page table entry
456 * that points to page_address(page).
458 * If rmapp bit zero is one, (then rmap & ~1) points to a struct kvm_rmap_desc
459 * containing more mappings.
461 static void rmap_add(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn, int lpage)
463 struct kvm_mmu_page *sp;
464 struct kvm_rmap_desc *desc;
465 unsigned long *rmapp;
468 if (!is_rmap_pte(*spte))
470 gfn = unalias_gfn(vcpu->kvm, gfn);
471 sp = page_header(__pa(spte));
472 sp->gfns[spte - sp->spt] = gfn;
473 rmapp = gfn_to_rmap(vcpu->kvm, gfn, lpage);
475 rmap_printk("rmap_add: %p %llx 0->1\n", spte, *spte);
476 *rmapp = (unsigned long)spte;
477 } else if (!(*rmapp & 1)) {
478 rmap_printk("rmap_add: %p %llx 1->many\n", spte, *spte);
479 desc = mmu_alloc_rmap_desc(vcpu);
480 desc->shadow_ptes[0] = (u64 *)*rmapp;
481 desc->shadow_ptes[1] = spte;
482 *rmapp = (unsigned long)desc | 1;
484 rmap_printk("rmap_add: %p %llx many->many\n", spte, *spte);
485 desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
486 while (desc->shadow_ptes[RMAP_EXT-1] && desc->more)
488 if (desc->shadow_ptes[RMAP_EXT-1]) {
489 desc->more = mmu_alloc_rmap_desc(vcpu);
492 for (i = 0; desc->shadow_ptes[i]; ++i)
494 desc->shadow_ptes[i] = spte;
498 static void rmap_desc_remove_entry(unsigned long *rmapp,
499 struct kvm_rmap_desc *desc,
501 struct kvm_rmap_desc *prev_desc)
505 for (j = RMAP_EXT - 1; !desc->shadow_ptes[j] && j > i; --j)
507 desc->shadow_ptes[i] = desc->shadow_ptes[j];
508 desc->shadow_ptes[j] = NULL;
511 if (!prev_desc && !desc->more)
512 *rmapp = (unsigned long)desc->shadow_ptes[0];
515 prev_desc->more = desc->more;
517 *rmapp = (unsigned long)desc->more | 1;
518 mmu_free_rmap_desc(desc);
521 static void rmap_remove(struct kvm *kvm, u64 *spte)
523 struct kvm_rmap_desc *desc;
524 struct kvm_rmap_desc *prev_desc;
525 struct kvm_mmu_page *sp;
527 unsigned long *rmapp;
530 if (!is_rmap_pte(*spte))
532 sp = page_header(__pa(spte));
533 pfn = spte_to_pfn(*spte);
534 if (*spte & shadow_accessed_mask)
535 kvm_set_pfn_accessed(pfn);
536 if (is_writeble_pte(*spte))
537 kvm_release_pfn_dirty(pfn);
539 kvm_release_pfn_clean(pfn);
540 rmapp = gfn_to_rmap(kvm, sp->gfns[spte - sp->spt], is_large_pte(*spte));
542 printk(KERN_ERR "rmap_remove: %p %llx 0->BUG\n", spte, *spte);
544 } else if (!(*rmapp & 1)) {
545 rmap_printk("rmap_remove: %p %llx 1->0\n", spte, *spte);
546 if ((u64 *)*rmapp != spte) {
547 printk(KERN_ERR "rmap_remove: %p %llx 1->BUG\n",
553 rmap_printk("rmap_remove: %p %llx many->many\n", spte, *spte);
554 desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
557 for (i = 0; i < RMAP_EXT && desc->shadow_ptes[i]; ++i)
558 if (desc->shadow_ptes[i] == spte) {
559 rmap_desc_remove_entry(rmapp,
571 static u64 *rmap_next(struct kvm *kvm, unsigned long *rmapp, u64 *spte)
573 struct kvm_rmap_desc *desc;
574 struct kvm_rmap_desc *prev_desc;
580 else if (!(*rmapp & 1)) {
582 return (u64 *)*rmapp;
585 desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
589 for (i = 0; i < RMAP_EXT && desc->shadow_ptes[i]; ++i) {
590 if (prev_spte == spte)
591 return desc->shadow_ptes[i];
592 prev_spte = desc->shadow_ptes[i];
599 static void rmap_write_protect(struct kvm *kvm, u64 gfn)
601 unsigned long *rmapp;
603 int write_protected = 0;
605 gfn = unalias_gfn(kvm, gfn);
606 rmapp = gfn_to_rmap(kvm, gfn, 0);
608 spte = rmap_next(kvm, rmapp, NULL);
611 BUG_ON(!(*spte & PT_PRESENT_MASK));
612 rmap_printk("rmap_write_protect: spte %p %llx\n", spte, *spte);
613 if (is_writeble_pte(*spte)) {
614 set_shadow_pte(spte, *spte & ~PT_WRITABLE_MASK);
617 spte = rmap_next(kvm, rmapp, spte);
619 if (write_protected) {
622 spte = rmap_next(kvm, rmapp, NULL);
623 pfn = spte_to_pfn(*spte);
624 kvm_set_pfn_dirty(pfn);
627 /* check for huge page mappings */
628 rmapp = gfn_to_rmap(kvm, gfn, 1);
629 spte = rmap_next(kvm, rmapp, NULL);
632 BUG_ON(!(*spte & PT_PRESENT_MASK));
633 BUG_ON((*spte & (PT_PAGE_SIZE_MASK|PT_PRESENT_MASK)) != (PT_PAGE_SIZE_MASK|PT_PRESENT_MASK));
634 pgprintk("rmap_write_protect(large): spte %p %llx %lld\n", spte, *spte, gfn);
635 if (is_writeble_pte(*spte)) {
636 rmap_remove(kvm, spte);
638 set_shadow_pte(spte, shadow_trap_nonpresent_pte);
642 spte = rmap_next(kvm, rmapp, spte);
646 kvm_flush_remote_tlbs(kvm);
648 account_shadowed(kvm, gfn);
651 static int kvm_unmap_rmapp(struct kvm *kvm, unsigned long *rmapp)
654 int need_tlb_flush = 0;
656 while ((spte = rmap_next(kvm, rmapp, NULL))) {
657 BUG_ON(!(*spte & PT_PRESENT_MASK));
658 rmap_printk("kvm_rmap_unmap_hva: spte %p %llx\n", spte, *spte);
659 rmap_remove(kvm, spte);
660 set_shadow_pte(spte, shadow_trap_nonpresent_pte);
663 return need_tlb_flush;
666 static int kvm_handle_hva(struct kvm *kvm, unsigned long hva,
667 int (*handler)(struct kvm *kvm, unsigned long *rmapp))
673 * If mmap_sem isn't taken, we can look the memslots with only
674 * the mmu_lock by skipping over the slots with userspace_addr == 0.
676 for (i = 0; i < kvm->nmemslots; i++) {
677 struct kvm_memory_slot *memslot = &kvm->memslots[i];
678 unsigned long start = memslot->userspace_addr;
681 /* mmu_lock protects userspace_addr */
685 end = start + (memslot->npages << PAGE_SHIFT);
686 if (hva >= start && hva < end) {
687 gfn_t gfn_offset = (hva - start) >> PAGE_SHIFT;
688 retval |= handler(kvm, &memslot->rmap[gfn_offset]);
689 retval |= handler(kvm,
690 &memslot->lpage_info[
692 KVM_PAGES_PER_HPAGE].rmap_pde);
699 int kvm_unmap_hva(struct kvm *kvm, unsigned long hva)
701 return kvm_handle_hva(kvm, hva, kvm_unmap_rmapp);
704 static int kvm_age_rmapp(struct kvm *kvm, unsigned long *rmapp)
709 /* always return old for EPT */
710 if (!shadow_accessed_mask)
713 spte = rmap_next(kvm, rmapp, NULL);
717 BUG_ON(!(_spte & PT_PRESENT_MASK));
718 _young = _spte & PT_ACCESSED_MASK;
721 clear_bit(PT_ACCESSED_SHIFT, (unsigned long *)spte);
723 spte = rmap_next(kvm, rmapp, spte);
728 int kvm_age_hva(struct kvm *kvm, unsigned long hva)
730 return kvm_handle_hva(kvm, hva, kvm_age_rmapp);
734 static int is_empty_shadow_page(u64 *spt)
739 for (pos = spt, end = pos + PAGE_SIZE / sizeof(u64); pos != end; pos++)
740 if (is_shadow_present_pte(*pos)) {
741 printk(KERN_ERR "%s: %p %llx\n", __func__,
749 static void kvm_mmu_free_page(struct kvm *kvm, struct kvm_mmu_page *sp)
751 ASSERT(is_empty_shadow_page(sp->spt));
753 __free_page(virt_to_page(sp->spt));
754 __free_page(virt_to_page(sp->gfns));
756 ++kvm->arch.n_free_mmu_pages;
759 static unsigned kvm_page_table_hashfn(gfn_t gfn)
761 return gfn & ((1 << KVM_MMU_HASH_SHIFT) - 1);
764 static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu,
767 struct kvm_mmu_page *sp;
769 sp = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_header_cache, sizeof *sp);
770 sp->spt = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_cache, PAGE_SIZE);
771 sp->gfns = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_cache, PAGE_SIZE);
772 set_page_private(virt_to_page(sp->spt), (unsigned long)sp);
773 list_add(&sp->link, &vcpu->kvm->arch.active_mmu_pages);
774 ASSERT(is_empty_shadow_page(sp->spt));
777 sp->parent_pte = parent_pte;
778 --vcpu->kvm->arch.n_free_mmu_pages;
782 static void mmu_page_add_parent_pte(struct kvm_vcpu *vcpu,
783 struct kvm_mmu_page *sp, u64 *parent_pte)
785 struct kvm_pte_chain *pte_chain;
786 struct hlist_node *node;
791 if (!sp->multimapped) {
792 u64 *old = sp->parent_pte;
795 sp->parent_pte = parent_pte;
799 pte_chain = mmu_alloc_pte_chain(vcpu);
800 INIT_HLIST_HEAD(&sp->parent_ptes);
801 hlist_add_head(&pte_chain->link, &sp->parent_ptes);
802 pte_chain->parent_ptes[0] = old;
804 hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link) {
805 if (pte_chain->parent_ptes[NR_PTE_CHAIN_ENTRIES-1])
807 for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i)
808 if (!pte_chain->parent_ptes[i]) {
809 pte_chain->parent_ptes[i] = parent_pte;
813 pte_chain = mmu_alloc_pte_chain(vcpu);
815 hlist_add_head(&pte_chain->link, &sp->parent_ptes);
816 pte_chain->parent_ptes[0] = parent_pte;
819 static void mmu_page_remove_parent_pte(struct kvm_mmu_page *sp,
822 struct kvm_pte_chain *pte_chain;
823 struct hlist_node *node;
826 if (!sp->multimapped) {
827 BUG_ON(sp->parent_pte != parent_pte);
828 sp->parent_pte = NULL;
831 hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link)
832 for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) {
833 if (!pte_chain->parent_ptes[i])
835 if (pte_chain->parent_ptes[i] != parent_pte)
837 while (i + 1 < NR_PTE_CHAIN_ENTRIES
838 && pte_chain->parent_ptes[i + 1]) {
839 pte_chain->parent_ptes[i]
840 = pte_chain->parent_ptes[i + 1];
843 pte_chain->parent_ptes[i] = NULL;
845 hlist_del(&pte_chain->link);
846 mmu_free_pte_chain(pte_chain);
847 if (hlist_empty(&sp->parent_ptes)) {
849 sp->parent_pte = NULL;
857 static void nonpaging_prefetch_page(struct kvm_vcpu *vcpu,
858 struct kvm_mmu_page *sp)
862 for (i = 0; i < PT64_ENT_PER_PAGE; ++i)
863 sp->spt[i] = shadow_trap_nonpresent_pte;
866 static struct kvm_mmu_page *kvm_mmu_lookup_page(struct kvm *kvm, gfn_t gfn)
869 struct hlist_head *bucket;
870 struct kvm_mmu_page *sp;
871 struct hlist_node *node;
873 pgprintk("%s: looking for gfn %lx\n", __func__, gfn);
874 index = kvm_page_table_hashfn(gfn);
875 bucket = &kvm->arch.mmu_page_hash[index];
876 hlist_for_each_entry(sp, node, bucket, hash_link)
877 if (sp->gfn == gfn && !sp->role.metaphysical
878 && !sp->role.invalid) {
879 pgprintk("%s: found role %x\n",
880 __func__, sp->role.word);
886 static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
894 union kvm_mmu_page_role role;
897 struct hlist_head *bucket;
898 struct kvm_mmu_page *sp;
899 struct hlist_node *node;
902 role.glevels = vcpu->arch.mmu.root_level;
904 role.metaphysical = metaphysical;
905 role.access = access;
906 if (vcpu->arch.mmu.root_level <= PT32_ROOT_LEVEL) {
907 quadrant = gaddr >> (PAGE_SHIFT + (PT64_PT_BITS * level));
908 quadrant &= (1 << ((PT32_PT_BITS - PT64_PT_BITS) * level)) - 1;
909 role.quadrant = quadrant;
911 pgprintk("%s: looking gfn %lx role %x\n", __func__,
913 index = kvm_page_table_hashfn(gfn);
914 bucket = &vcpu->kvm->arch.mmu_page_hash[index];
915 hlist_for_each_entry(sp, node, bucket, hash_link)
916 if (sp->gfn == gfn && sp->role.word == role.word) {
917 mmu_page_add_parent_pte(vcpu, sp, parent_pte);
918 pgprintk("%s: found\n", __func__);
921 ++vcpu->kvm->stat.mmu_cache_miss;
922 sp = kvm_mmu_alloc_page(vcpu, parent_pte);
925 pgprintk("%s: adding gfn %lx role %x\n", __func__, gfn, role.word);
928 hlist_add_head(&sp->hash_link, bucket);
930 rmap_write_protect(vcpu->kvm, gfn);
931 if (shadow_trap_nonpresent_pte != shadow_notrap_nonpresent_pte)
932 vcpu->arch.mmu.prefetch_page(vcpu, sp);
934 nonpaging_prefetch_page(vcpu, sp);
938 static void kvm_mmu_page_unlink_children(struct kvm *kvm,
939 struct kvm_mmu_page *sp)
947 if (sp->role.level == PT_PAGE_TABLE_LEVEL) {
948 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
949 if (is_shadow_present_pte(pt[i]))
950 rmap_remove(kvm, &pt[i]);
951 pt[i] = shadow_trap_nonpresent_pte;
956 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
959 if (is_shadow_present_pte(ent)) {
960 if (!is_large_pte(ent)) {
961 ent &= PT64_BASE_ADDR_MASK;
962 mmu_page_remove_parent_pte(page_header(ent),
966 rmap_remove(kvm, &pt[i]);
969 pt[i] = shadow_trap_nonpresent_pte;
973 static void kvm_mmu_put_page(struct kvm_mmu_page *sp, u64 *parent_pte)
975 mmu_page_remove_parent_pte(sp, parent_pte);
978 static void kvm_mmu_reset_last_pte_updated(struct kvm *kvm)
982 for (i = 0; i < KVM_MAX_VCPUS; ++i)
984 kvm->vcpus[i]->arch.last_pte_updated = NULL;
987 static void kvm_mmu_unlink_parents(struct kvm *kvm, struct kvm_mmu_page *sp)
991 while (sp->multimapped || sp->parent_pte) {
992 if (!sp->multimapped)
993 parent_pte = sp->parent_pte;
995 struct kvm_pte_chain *chain;
997 chain = container_of(sp->parent_ptes.first,
998 struct kvm_pte_chain, link);
999 parent_pte = chain->parent_ptes[0];
1001 BUG_ON(!parent_pte);
1002 kvm_mmu_put_page(sp, parent_pte);
1003 set_shadow_pte(parent_pte, shadow_trap_nonpresent_pte);
1007 static void kvm_mmu_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp)
1009 ++kvm->stat.mmu_shadow_zapped;
1010 kvm_mmu_page_unlink_children(kvm, sp);
1011 kvm_mmu_unlink_parents(kvm, sp);
1012 kvm_flush_remote_tlbs(kvm);
1013 if (!sp->role.invalid && !sp->role.metaphysical)
1014 unaccount_shadowed(kvm, sp->gfn);
1015 if (!sp->root_count) {
1016 hlist_del(&sp->hash_link);
1017 kvm_mmu_free_page(kvm, sp);
1019 sp->role.invalid = 1;
1020 list_move(&sp->link, &kvm->arch.active_mmu_pages);
1021 kvm_reload_remote_mmus(kvm);
1023 kvm_mmu_reset_last_pte_updated(kvm);
1027 * Changing the number of mmu pages allocated to the vm
1028 * Note: if kvm_nr_mmu_pages is too small, you will get dead lock
1030 void kvm_mmu_change_mmu_pages(struct kvm *kvm, unsigned int kvm_nr_mmu_pages)
1033 * If we set the number of mmu pages to be smaller be than the
1034 * number of actived pages , we must to free some mmu pages before we
1038 if ((kvm->arch.n_alloc_mmu_pages - kvm->arch.n_free_mmu_pages) >
1040 int n_used_mmu_pages = kvm->arch.n_alloc_mmu_pages
1041 - kvm->arch.n_free_mmu_pages;
1043 while (n_used_mmu_pages > kvm_nr_mmu_pages) {
1044 struct kvm_mmu_page *page;
1046 page = container_of(kvm->arch.active_mmu_pages.prev,
1047 struct kvm_mmu_page, link);
1048 kvm_mmu_zap_page(kvm, page);
1051 kvm->arch.n_free_mmu_pages = 0;
1054 kvm->arch.n_free_mmu_pages += kvm_nr_mmu_pages
1055 - kvm->arch.n_alloc_mmu_pages;
1057 kvm->arch.n_alloc_mmu_pages = kvm_nr_mmu_pages;
1060 static int kvm_mmu_unprotect_page(struct kvm *kvm, gfn_t gfn)
1063 struct hlist_head *bucket;
1064 struct kvm_mmu_page *sp;
1065 struct hlist_node *node, *n;
1068 pgprintk("%s: looking for gfn %lx\n", __func__, gfn);
1070 index = kvm_page_table_hashfn(gfn);
1071 bucket = &kvm->arch.mmu_page_hash[index];
1072 hlist_for_each_entry_safe(sp, node, n, bucket, hash_link)
1073 if (sp->gfn == gfn && !sp->role.metaphysical) {
1074 pgprintk("%s: gfn %lx role %x\n", __func__, gfn,
1076 kvm_mmu_zap_page(kvm, sp);
1082 static void mmu_unshadow(struct kvm *kvm, gfn_t gfn)
1084 struct kvm_mmu_page *sp;
1086 while ((sp = kvm_mmu_lookup_page(kvm, gfn)) != NULL) {
1087 pgprintk("%s: zap %lx %x\n", __func__, gfn, sp->role.word);
1088 kvm_mmu_zap_page(kvm, sp);
1092 static void page_header_update_slot(struct kvm *kvm, void *pte, gfn_t gfn)
1094 int slot = memslot_id(kvm, gfn_to_memslot(kvm, gfn));
1095 struct kvm_mmu_page *sp = page_header(__pa(pte));
1097 __set_bit(slot, &sp->slot_bitmap);
1100 struct page *gva_to_page(struct kvm_vcpu *vcpu, gva_t gva)
1104 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gva);
1106 if (gpa == UNMAPPED_GVA)
1109 down_read(¤t->mm->mmap_sem);
1110 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1111 up_read(¤t->mm->mmap_sem);
1116 static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *shadow_pte,
1117 unsigned pt_access, unsigned pte_access,
1118 int user_fault, int write_fault, int dirty,
1119 int *ptwrite, int largepage, gfn_t gfn,
1120 pfn_t pfn, bool speculative)
1123 int was_rmapped = 0;
1124 int was_writeble = is_writeble_pte(*shadow_pte);
1126 pgprintk("%s: spte %llx access %x write_fault %d"
1127 " user_fault %d gfn %lx\n",
1128 __func__, *shadow_pte, pt_access,
1129 write_fault, user_fault, gfn);
1131 if (is_rmap_pte(*shadow_pte)) {
1133 * If we overwrite a PTE page pointer with a 2MB PMD, unlink
1134 * the parent of the now unreachable PTE.
1136 if (largepage && !is_large_pte(*shadow_pte)) {
1137 struct kvm_mmu_page *child;
1138 u64 pte = *shadow_pte;
1140 child = page_header(pte & PT64_BASE_ADDR_MASK);
1141 mmu_page_remove_parent_pte(child, shadow_pte);
1142 } else if (pfn != spte_to_pfn(*shadow_pte)) {
1143 pgprintk("hfn old %lx new %lx\n",
1144 spte_to_pfn(*shadow_pte), pfn);
1145 rmap_remove(vcpu->kvm, shadow_pte);
1148 was_rmapped = is_large_pte(*shadow_pte);
1155 * We don't set the accessed bit, since we sometimes want to see
1156 * whether the guest actually used the pte (in order to detect
1159 spte = shadow_base_present_pte | shadow_dirty_mask;
1161 pte_access |= PT_ACCESSED_MASK;
1163 pte_access &= ~ACC_WRITE_MASK;
1164 if (pte_access & ACC_EXEC_MASK)
1165 spte |= shadow_x_mask;
1167 spte |= shadow_nx_mask;
1168 if (pte_access & ACC_USER_MASK)
1169 spte |= shadow_user_mask;
1171 spte |= PT_PAGE_SIZE_MASK;
1173 spte |= (u64)pfn << PAGE_SHIFT;
1175 if ((pte_access & ACC_WRITE_MASK)
1176 || (write_fault && !is_write_protection(vcpu) && !user_fault)) {
1177 struct kvm_mmu_page *shadow;
1179 spte |= PT_WRITABLE_MASK;
1181 shadow = kvm_mmu_lookup_page(vcpu->kvm, gfn);
1183 (largepage && has_wrprotected_page(vcpu->kvm, gfn))) {
1184 pgprintk("%s: found shadow page for %lx, marking ro\n",
1186 pte_access &= ~ACC_WRITE_MASK;
1187 if (is_writeble_pte(spte)) {
1188 spte &= ~PT_WRITABLE_MASK;
1189 kvm_x86_ops->tlb_flush(vcpu);
1196 if (pte_access & ACC_WRITE_MASK)
1197 mark_page_dirty(vcpu->kvm, gfn);
1199 pgprintk("%s: setting spte %llx\n", __func__, spte);
1200 pgprintk("instantiating %s PTE (%s) at %ld (%llx) addr %p\n",
1201 (spte&PT_PAGE_SIZE_MASK)? "2MB" : "4kB",
1202 (spte&PT_WRITABLE_MASK)?"RW":"R", gfn, spte, shadow_pte);
1203 set_shadow_pte(shadow_pte, spte);
1204 if (!was_rmapped && (spte & PT_PAGE_SIZE_MASK)
1205 && (spte & PT_PRESENT_MASK))
1206 ++vcpu->kvm->stat.lpages;
1208 page_header_update_slot(vcpu->kvm, shadow_pte, gfn);
1210 rmap_add(vcpu, shadow_pte, gfn, largepage);
1211 if (!is_rmap_pte(*shadow_pte))
1212 kvm_release_pfn_clean(pfn);
1215 kvm_release_pfn_dirty(pfn);
1217 kvm_release_pfn_clean(pfn);
1220 vcpu->arch.last_pte_updated = shadow_pte;
1221 vcpu->arch.last_pte_gfn = gfn;
1225 static void nonpaging_new_cr3(struct kvm_vcpu *vcpu)
1229 static int __direct_map(struct kvm_vcpu *vcpu, gpa_t v, int write,
1230 int largepage, gfn_t gfn, pfn_t pfn,
1233 hpa_t table_addr = vcpu->arch.mmu.root_hpa;
1237 u32 index = PT64_INDEX(v, level);
1240 ASSERT(VALID_PAGE(table_addr));
1241 table = __va(table_addr);
1243 if (level == 1 || (largepage && level == 2)) {
1244 mmu_set_spte(vcpu, &table[index], ACC_ALL, ACC_ALL,
1245 0, write, 1, &pt_write, largepage,
1250 if (table[index] == shadow_trap_nonpresent_pte) {
1251 struct kvm_mmu_page *new_table;
1254 pseudo_gfn = (v & PT64_DIR_BASE_ADDR_MASK)
1256 new_table = kvm_mmu_get_page(vcpu, pseudo_gfn,
1258 1, ACC_ALL, &table[index]);
1260 pgprintk("nonpaging_map: ENOMEM\n");
1261 kvm_release_pfn_clean(pfn);
1265 set_shadow_pte(&table[index],
1266 __pa(new_table->spt)
1267 | PT_PRESENT_MASK | PT_WRITABLE_MASK
1268 | shadow_user_mask | shadow_x_mask);
1270 table_addr = table[index] & PT64_BASE_ADDR_MASK;
1274 static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, int write, gfn_t gfn)
1279 unsigned long mmu_seq;
1281 down_read(¤t->mm->mmap_sem);
1282 if (is_largepage_backed(vcpu, gfn & ~(KVM_PAGES_PER_HPAGE-1))) {
1283 gfn &= ~(KVM_PAGES_PER_HPAGE-1);
1287 mmu_seq = vcpu->kvm->mmu_notifier_seq;
1288 /* implicit mb(), we'll read before PT lock is unlocked */
1289 pfn = gfn_to_pfn(vcpu->kvm, gfn);
1290 up_read(¤t->mm->mmap_sem);
1293 if (is_error_pfn(pfn)) {
1294 kvm_release_pfn_clean(pfn);
1298 spin_lock(&vcpu->kvm->mmu_lock);
1299 if (mmu_notifier_retry(vcpu, mmu_seq))
1301 kvm_mmu_free_some_pages(vcpu);
1302 r = __direct_map(vcpu, v, write, largepage, gfn, pfn,
1304 spin_unlock(&vcpu->kvm->mmu_lock);
1310 spin_unlock(&vcpu->kvm->mmu_lock);
1311 kvm_release_pfn_clean(pfn);
1316 static void mmu_free_roots(struct kvm_vcpu *vcpu)
1319 struct kvm_mmu_page *sp;
1321 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
1323 spin_lock(&vcpu->kvm->mmu_lock);
1324 if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) {
1325 hpa_t root = vcpu->arch.mmu.root_hpa;
1327 sp = page_header(root);
1329 if (!sp->root_count && sp->role.invalid)
1330 kvm_mmu_zap_page(vcpu->kvm, sp);
1331 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
1332 spin_unlock(&vcpu->kvm->mmu_lock);
1335 for (i = 0; i < 4; ++i) {
1336 hpa_t root = vcpu->arch.mmu.pae_root[i];
1339 root &= PT64_BASE_ADDR_MASK;
1340 sp = page_header(root);
1342 if (!sp->root_count && sp->role.invalid)
1343 kvm_mmu_zap_page(vcpu->kvm, sp);
1345 vcpu->arch.mmu.pae_root[i] = INVALID_PAGE;
1347 spin_unlock(&vcpu->kvm->mmu_lock);
1348 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
1351 static void mmu_alloc_roots(struct kvm_vcpu *vcpu)
1355 struct kvm_mmu_page *sp;
1356 int metaphysical = 0;
1358 root_gfn = vcpu->arch.cr3 >> PAGE_SHIFT;
1360 if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) {
1361 hpa_t root = vcpu->arch.mmu.root_hpa;
1363 ASSERT(!VALID_PAGE(root));
1366 sp = kvm_mmu_get_page(vcpu, root_gfn, 0,
1367 PT64_ROOT_LEVEL, metaphysical,
1369 root = __pa(sp->spt);
1371 vcpu->arch.mmu.root_hpa = root;
1374 metaphysical = !is_paging(vcpu);
1377 for (i = 0; i < 4; ++i) {
1378 hpa_t root = vcpu->arch.mmu.pae_root[i];
1380 ASSERT(!VALID_PAGE(root));
1381 if (vcpu->arch.mmu.root_level == PT32E_ROOT_LEVEL) {
1382 if (!is_present_pte(vcpu->arch.pdptrs[i])) {
1383 vcpu->arch.mmu.pae_root[i] = 0;
1386 root_gfn = vcpu->arch.pdptrs[i] >> PAGE_SHIFT;
1387 } else if (vcpu->arch.mmu.root_level == 0)
1389 sp = kvm_mmu_get_page(vcpu, root_gfn, i << 30,
1390 PT32_ROOT_LEVEL, metaphysical,
1392 root = __pa(sp->spt);
1394 vcpu->arch.mmu.pae_root[i] = root | PT_PRESENT_MASK;
1396 vcpu->arch.mmu.root_hpa = __pa(vcpu->arch.mmu.pae_root);
1399 static gpa_t nonpaging_gva_to_gpa(struct kvm_vcpu *vcpu, gva_t vaddr)
1404 static int nonpaging_page_fault(struct kvm_vcpu *vcpu, gva_t gva,
1410 pgprintk("%s: gva %lx error %x\n", __func__, gva, error_code);
1411 r = mmu_topup_memory_caches(vcpu);
1416 ASSERT(VALID_PAGE(vcpu->arch.mmu.root_hpa));
1418 gfn = gva >> PAGE_SHIFT;
1420 return nonpaging_map(vcpu, gva & PAGE_MASK,
1421 error_code & PFERR_WRITE_MASK, gfn);
1424 static int tdp_page_fault(struct kvm_vcpu *vcpu, gva_t gpa,
1430 gfn_t gfn = gpa >> PAGE_SHIFT;
1431 unsigned long mmu_seq;
1434 ASSERT(VALID_PAGE(vcpu->arch.mmu.root_hpa));
1436 r = mmu_topup_memory_caches(vcpu);
1440 down_read(¤t->mm->mmap_sem);
1441 if (is_largepage_backed(vcpu, gfn & ~(KVM_PAGES_PER_HPAGE-1))) {
1442 gfn &= ~(KVM_PAGES_PER_HPAGE-1);
1445 mmu_seq = vcpu->kvm->mmu_notifier_seq;
1446 /* implicit mb(), we'll read before PT lock is unlocked */
1447 pfn = gfn_to_pfn(vcpu->kvm, gfn);
1448 up_read(¤t->mm->mmap_sem);
1449 if (is_error_pfn(pfn)) {
1450 kvm_release_pfn_clean(pfn);
1453 spin_lock(&vcpu->kvm->mmu_lock);
1454 if (mmu_notifier_retry(vcpu, mmu_seq))
1456 kvm_mmu_free_some_pages(vcpu);
1457 r = __direct_map(vcpu, gpa, error_code & PFERR_WRITE_MASK,
1458 largepage, gfn, pfn, kvm_x86_ops->get_tdp_level());
1459 spin_unlock(&vcpu->kvm->mmu_lock);
1464 spin_unlock(&vcpu->kvm->mmu_lock);
1465 kvm_release_pfn_clean(pfn);
1469 static void nonpaging_free(struct kvm_vcpu *vcpu)
1471 mmu_free_roots(vcpu);
1474 static int nonpaging_init_context(struct kvm_vcpu *vcpu)
1476 struct kvm_mmu *context = &vcpu->arch.mmu;
1478 context->new_cr3 = nonpaging_new_cr3;
1479 context->page_fault = nonpaging_page_fault;
1480 context->gva_to_gpa = nonpaging_gva_to_gpa;
1481 context->free = nonpaging_free;
1482 context->prefetch_page = nonpaging_prefetch_page;
1483 context->root_level = 0;
1484 context->shadow_root_level = PT32E_ROOT_LEVEL;
1485 context->root_hpa = INVALID_PAGE;
1489 void kvm_mmu_flush_tlb(struct kvm_vcpu *vcpu)
1491 ++vcpu->stat.tlb_flush;
1492 kvm_x86_ops->tlb_flush(vcpu);
1495 static void paging_new_cr3(struct kvm_vcpu *vcpu)
1497 pgprintk("%s: cr3 %lx\n", __func__, vcpu->arch.cr3);
1498 mmu_free_roots(vcpu);
1501 static void inject_page_fault(struct kvm_vcpu *vcpu,
1505 kvm_inject_page_fault(vcpu, addr, err_code);
1508 static void paging_free(struct kvm_vcpu *vcpu)
1510 nonpaging_free(vcpu);
1514 #include "paging_tmpl.h"
1518 #include "paging_tmpl.h"
1521 static int paging64_init_context_common(struct kvm_vcpu *vcpu, int level)
1523 struct kvm_mmu *context = &vcpu->arch.mmu;
1525 ASSERT(is_pae(vcpu));
1526 context->new_cr3 = paging_new_cr3;
1527 context->page_fault = paging64_page_fault;
1528 context->gva_to_gpa = paging64_gva_to_gpa;
1529 context->prefetch_page = paging64_prefetch_page;
1530 context->free = paging_free;
1531 context->root_level = level;
1532 context->shadow_root_level = level;
1533 context->root_hpa = INVALID_PAGE;
1537 static int paging64_init_context(struct kvm_vcpu *vcpu)
1539 return paging64_init_context_common(vcpu, PT64_ROOT_LEVEL);
1542 static int paging32_init_context(struct kvm_vcpu *vcpu)
1544 struct kvm_mmu *context = &vcpu->arch.mmu;
1546 context->new_cr3 = paging_new_cr3;
1547 context->page_fault = paging32_page_fault;
1548 context->gva_to_gpa = paging32_gva_to_gpa;
1549 context->free = paging_free;
1550 context->prefetch_page = paging32_prefetch_page;
1551 context->root_level = PT32_ROOT_LEVEL;
1552 context->shadow_root_level = PT32E_ROOT_LEVEL;
1553 context->root_hpa = INVALID_PAGE;
1557 static int paging32E_init_context(struct kvm_vcpu *vcpu)
1559 return paging64_init_context_common(vcpu, PT32E_ROOT_LEVEL);
1562 static int init_kvm_tdp_mmu(struct kvm_vcpu *vcpu)
1564 struct kvm_mmu *context = &vcpu->arch.mmu;
1566 context->new_cr3 = nonpaging_new_cr3;
1567 context->page_fault = tdp_page_fault;
1568 context->free = nonpaging_free;
1569 context->prefetch_page = nonpaging_prefetch_page;
1570 context->shadow_root_level = kvm_x86_ops->get_tdp_level();
1571 context->root_hpa = INVALID_PAGE;
1573 if (!is_paging(vcpu)) {
1574 context->gva_to_gpa = nonpaging_gva_to_gpa;
1575 context->root_level = 0;
1576 } else if (is_long_mode(vcpu)) {
1577 context->gva_to_gpa = paging64_gva_to_gpa;
1578 context->root_level = PT64_ROOT_LEVEL;
1579 } else if (is_pae(vcpu)) {
1580 context->gva_to_gpa = paging64_gva_to_gpa;
1581 context->root_level = PT32E_ROOT_LEVEL;
1583 context->gva_to_gpa = paging32_gva_to_gpa;
1584 context->root_level = PT32_ROOT_LEVEL;
1590 static int init_kvm_softmmu(struct kvm_vcpu *vcpu)
1593 ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa));
1595 if (!is_paging(vcpu))
1596 return nonpaging_init_context(vcpu);
1597 else if (is_long_mode(vcpu))
1598 return paging64_init_context(vcpu);
1599 else if (is_pae(vcpu))
1600 return paging32E_init_context(vcpu);
1602 return paging32_init_context(vcpu);
1605 static int init_kvm_mmu(struct kvm_vcpu *vcpu)
1607 vcpu->arch.update_pte.pfn = bad_pfn;
1610 return init_kvm_tdp_mmu(vcpu);
1612 return init_kvm_softmmu(vcpu);
1615 static void destroy_kvm_mmu(struct kvm_vcpu *vcpu)
1618 if (VALID_PAGE(vcpu->arch.mmu.root_hpa)) {
1619 vcpu->arch.mmu.free(vcpu);
1620 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
1624 int kvm_mmu_reset_context(struct kvm_vcpu *vcpu)
1626 destroy_kvm_mmu(vcpu);
1627 return init_kvm_mmu(vcpu);
1629 EXPORT_SYMBOL_GPL(kvm_mmu_reset_context);
1631 int kvm_mmu_load(struct kvm_vcpu *vcpu)
1635 r = mmu_topup_memory_caches(vcpu);
1638 spin_lock(&vcpu->kvm->mmu_lock);
1639 kvm_mmu_free_some_pages(vcpu);
1640 mmu_alloc_roots(vcpu);
1641 spin_unlock(&vcpu->kvm->mmu_lock);
1642 kvm_x86_ops->set_cr3(vcpu, vcpu->arch.mmu.root_hpa);
1643 kvm_mmu_flush_tlb(vcpu);
1647 EXPORT_SYMBOL_GPL(kvm_mmu_load);
1649 void kvm_mmu_unload(struct kvm_vcpu *vcpu)
1651 mmu_free_roots(vcpu);
1654 static void mmu_pte_write_zap_pte(struct kvm_vcpu *vcpu,
1655 struct kvm_mmu_page *sp,
1659 struct kvm_mmu_page *child;
1662 if (is_shadow_present_pte(pte)) {
1663 if (sp->role.level == PT_PAGE_TABLE_LEVEL ||
1665 rmap_remove(vcpu->kvm, spte);
1667 child = page_header(pte & PT64_BASE_ADDR_MASK);
1668 mmu_page_remove_parent_pte(child, spte);
1671 set_shadow_pte(spte, shadow_trap_nonpresent_pte);
1672 if (is_large_pte(pte))
1673 --vcpu->kvm->stat.lpages;
1676 static void mmu_pte_write_new_pte(struct kvm_vcpu *vcpu,
1677 struct kvm_mmu_page *sp,
1681 if (sp->role.level != PT_PAGE_TABLE_LEVEL) {
1682 if (!vcpu->arch.update_pte.largepage ||
1683 sp->role.glevels == PT32_ROOT_LEVEL) {
1684 ++vcpu->kvm->stat.mmu_pde_zapped;
1689 ++vcpu->kvm->stat.mmu_pte_updated;
1690 if (sp->role.glevels == PT32_ROOT_LEVEL)
1691 paging32_update_pte(vcpu, sp, spte, new);
1693 paging64_update_pte(vcpu, sp, spte, new);
1696 static bool need_remote_flush(u64 old, u64 new)
1698 if (!is_shadow_present_pte(old))
1700 if (!is_shadow_present_pte(new))
1702 if ((old ^ new) & PT64_BASE_ADDR_MASK)
1704 old ^= PT64_NX_MASK;
1705 new ^= PT64_NX_MASK;
1706 return (old & ~new & PT64_PERM_MASK) != 0;
1709 static void mmu_pte_write_flush_tlb(struct kvm_vcpu *vcpu, u64 old, u64 new)
1711 if (need_remote_flush(old, new))
1712 kvm_flush_remote_tlbs(vcpu->kvm);
1714 kvm_mmu_flush_tlb(vcpu);
1717 static bool last_updated_pte_accessed(struct kvm_vcpu *vcpu)
1719 u64 *spte = vcpu->arch.last_pte_updated;
1721 return !!(spte && (*spte & shadow_accessed_mask));
1724 static void mmu_guess_page_from_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
1725 const u8 *new, int bytes)
1732 vcpu->arch.update_pte.largepage = 0;
1734 if (bytes != 4 && bytes != 8)
1738 * Assume that the pte write on a page table of the same type
1739 * as the current vcpu paging mode. This is nearly always true
1740 * (might be false while changing modes). Note it is verified later
1744 /* Handle a 32-bit guest writing two halves of a 64-bit gpte */
1745 if ((bytes == 4) && (gpa % 4 == 0)) {
1746 r = kvm_read_guest(vcpu->kvm, gpa & ~(u64)7, &gpte, 8);
1749 memcpy((void *)&gpte + (gpa % 8), new, 4);
1750 } else if ((bytes == 8) && (gpa % 8 == 0)) {
1751 memcpy((void *)&gpte, new, 8);
1754 if ((bytes == 4) && (gpa % 4 == 0))
1755 memcpy((void *)&gpte, new, 4);
1757 if (!is_present_pte(gpte))
1759 gfn = (gpte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT;
1761 down_read(¤t->mm->mmap_sem);
1762 if (is_large_pte(gpte) && is_largepage_backed(vcpu, gfn)) {
1763 gfn &= ~(KVM_PAGES_PER_HPAGE-1);
1764 vcpu->arch.update_pte.largepage = 1;
1766 vcpu->arch.update_pte.mmu_seq = vcpu->kvm->mmu_notifier_seq;
1767 /* implicit mb(), we'll read before PT lock is unlocked */
1768 pfn = gfn_to_pfn(vcpu->kvm, gfn);
1769 up_read(¤t->mm->mmap_sem);
1771 if (is_error_pfn(pfn)) {
1772 kvm_release_pfn_clean(pfn);
1775 vcpu->arch.update_pte.gfn = gfn;
1776 vcpu->arch.update_pte.pfn = pfn;
1779 static void kvm_mmu_access_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1781 u64 *spte = vcpu->arch.last_pte_updated;
1784 && vcpu->arch.last_pte_gfn == gfn
1785 && shadow_accessed_mask
1786 && !(*spte & shadow_accessed_mask)
1787 && is_shadow_present_pte(*spte))
1788 set_bit(PT_ACCESSED_SHIFT, (unsigned long *)spte);
1791 void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
1792 const u8 *new, int bytes)
1794 gfn_t gfn = gpa >> PAGE_SHIFT;
1795 struct kvm_mmu_page *sp;
1796 struct hlist_node *node, *n;
1797 struct hlist_head *bucket;
1801 unsigned offset = offset_in_page(gpa);
1803 unsigned page_offset;
1804 unsigned misaligned;
1811 pgprintk("%s: gpa %llx bytes %d\n", __func__, gpa, bytes);
1812 mmu_guess_page_from_pte_write(vcpu, gpa, new, bytes);
1813 spin_lock(&vcpu->kvm->mmu_lock);
1814 kvm_mmu_access_page(vcpu, gfn);
1815 kvm_mmu_free_some_pages(vcpu);
1816 ++vcpu->kvm->stat.mmu_pte_write;
1817 kvm_mmu_audit(vcpu, "pre pte write");
1818 if (gfn == vcpu->arch.last_pt_write_gfn
1819 && !last_updated_pte_accessed(vcpu)) {
1820 ++vcpu->arch.last_pt_write_count;
1821 if (vcpu->arch.last_pt_write_count >= 3)
1824 vcpu->arch.last_pt_write_gfn = gfn;
1825 vcpu->arch.last_pt_write_count = 1;
1826 vcpu->arch.last_pte_updated = NULL;
1828 index = kvm_page_table_hashfn(gfn);
1829 bucket = &vcpu->kvm->arch.mmu_page_hash[index];
1830 hlist_for_each_entry_safe(sp, node, n, bucket, hash_link) {
1831 if (sp->gfn != gfn || sp->role.metaphysical || sp->role.invalid)
1833 pte_size = sp->role.glevels == PT32_ROOT_LEVEL ? 4 : 8;
1834 misaligned = (offset ^ (offset + bytes - 1)) & ~(pte_size - 1);
1835 misaligned |= bytes < 4;
1836 if (misaligned || flooded) {
1838 * Misaligned accesses are too much trouble to fix
1839 * up; also, they usually indicate a page is not used
1842 * If we're seeing too many writes to a page,
1843 * it may no longer be a page table, or we may be
1844 * forking, in which case it is better to unmap the
1847 pgprintk("misaligned: gpa %llx bytes %d role %x\n",
1848 gpa, bytes, sp->role.word);
1849 kvm_mmu_zap_page(vcpu->kvm, sp);
1850 ++vcpu->kvm->stat.mmu_flooded;
1853 page_offset = offset;
1854 level = sp->role.level;
1856 if (sp->role.glevels == PT32_ROOT_LEVEL) {
1857 page_offset <<= 1; /* 32->64 */
1859 * A 32-bit pde maps 4MB while the shadow pdes map
1860 * only 2MB. So we need to double the offset again
1861 * and zap two pdes instead of one.
1863 if (level == PT32_ROOT_LEVEL) {
1864 page_offset &= ~7; /* kill rounding error */
1868 quadrant = page_offset >> PAGE_SHIFT;
1869 page_offset &= ~PAGE_MASK;
1870 if (quadrant != sp->role.quadrant)
1873 spte = &sp->spt[page_offset / sizeof(*spte)];
1874 if ((gpa & (pte_size - 1)) || (bytes < pte_size)) {
1876 r = kvm_read_guest_atomic(vcpu->kvm,
1877 gpa & ~(u64)(pte_size - 1),
1879 new = (const void *)&gentry;
1885 mmu_pte_write_zap_pte(vcpu, sp, spte);
1887 mmu_pte_write_new_pte(vcpu, sp, spte, new);
1888 mmu_pte_write_flush_tlb(vcpu, entry, *spte);
1892 kvm_mmu_audit(vcpu, "post pte write");
1893 spin_unlock(&vcpu->kvm->mmu_lock);
1894 if (!is_error_pfn(vcpu->arch.update_pte.pfn)) {
1895 kvm_release_pfn_clean(vcpu->arch.update_pte.pfn);
1896 vcpu->arch.update_pte.pfn = bad_pfn;
1900 int kvm_mmu_unprotect_page_virt(struct kvm_vcpu *vcpu, gva_t gva)
1905 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gva);
1907 spin_lock(&vcpu->kvm->mmu_lock);
1908 r = kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1909 spin_unlock(&vcpu->kvm->mmu_lock);
1912 EXPORT_SYMBOL_GPL(kvm_mmu_unprotect_page_virt);
1914 void __kvm_mmu_free_some_pages(struct kvm_vcpu *vcpu)
1916 while (vcpu->kvm->arch.n_free_mmu_pages < KVM_REFILL_PAGES) {
1917 struct kvm_mmu_page *sp;
1919 sp = container_of(vcpu->kvm->arch.active_mmu_pages.prev,
1920 struct kvm_mmu_page, link);
1921 kvm_mmu_zap_page(vcpu->kvm, sp);
1922 ++vcpu->kvm->stat.mmu_recycled;
1926 int kvm_mmu_page_fault(struct kvm_vcpu *vcpu, gva_t cr2, u32 error_code)
1929 enum emulation_result er;
1931 r = vcpu->arch.mmu.page_fault(vcpu, cr2, error_code);
1940 r = mmu_topup_memory_caches(vcpu);
1944 er = emulate_instruction(vcpu, vcpu->run, cr2, error_code, 0);
1949 case EMULATE_DO_MMIO:
1950 ++vcpu->stat.mmio_exits;
1953 kvm_report_emulation_failure(vcpu, "pagetable");
1961 EXPORT_SYMBOL_GPL(kvm_mmu_page_fault);
1963 void kvm_enable_tdp(void)
1967 EXPORT_SYMBOL_GPL(kvm_enable_tdp);
1969 void kvm_disable_tdp(void)
1971 tdp_enabled = false;
1973 EXPORT_SYMBOL_GPL(kvm_disable_tdp);
1975 static void free_mmu_pages(struct kvm_vcpu *vcpu)
1977 struct kvm_mmu_page *sp;
1979 while (!list_empty(&vcpu->kvm->arch.active_mmu_pages)) {
1980 sp = container_of(vcpu->kvm->arch.active_mmu_pages.next,
1981 struct kvm_mmu_page, link);
1982 kvm_mmu_zap_page(vcpu->kvm, sp);
1985 free_page((unsigned long)vcpu->arch.mmu.pae_root);
1988 static int alloc_mmu_pages(struct kvm_vcpu *vcpu)
1995 if (vcpu->kvm->arch.n_requested_mmu_pages)
1996 vcpu->kvm->arch.n_free_mmu_pages =
1997 vcpu->kvm->arch.n_requested_mmu_pages;
1999 vcpu->kvm->arch.n_free_mmu_pages =
2000 vcpu->kvm->arch.n_alloc_mmu_pages;
2002 * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64.
2003 * Therefore we need to allocate shadow page tables in the first
2004 * 4GB of memory, which happens to fit the DMA32 zone.
2006 page = alloc_page(GFP_KERNEL | __GFP_DMA32);
2009 vcpu->arch.mmu.pae_root = page_address(page);
2010 for (i = 0; i < 4; ++i)
2011 vcpu->arch.mmu.pae_root[i] = INVALID_PAGE;
2016 free_mmu_pages(vcpu);
2020 int kvm_mmu_create(struct kvm_vcpu *vcpu)
2023 ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa));
2025 return alloc_mmu_pages(vcpu);
2028 int kvm_mmu_setup(struct kvm_vcpu *vcpu)
2031 ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa));
2033 return init_kvm_mmu(vcpu);
2036 void kvm_mmu_destroy(struct kvm_vcpu *vcpu)
2040 destroy_kvm_mmu(vcpu);
2041 free_mmu_pages(vcpu);
2042 mmu_free_memory_caches(vcpu);
2045 void kvm_mmu_slot_remove_write_access(struct kvm *kvm, int slot)
2047 struct kvm_mmu_page *sp;
2049 list_for_each_entry(sp, &kvm->arch.active_mmu_pages, link) {
2053 if (!test_bit(slot, &sp->slot_bitmap))
2057 for (i = 0; i < PT64_ENT_PER_PAGE; ++i)
2059 if (pt[i] & PT_WRITABLE_MASK)
2060 pt[i] &= ~PT_WRITABLE_MASK;
2064 void kvm_mmu_zap_all(struct kvm *kvm)
2066 struct kvm_mmu_page *sp, *node;
2068 spin_lock(&kvm->mmu_lock);
2069 list_for_each_entry_safe(sp, node, &kvm->arch.active_mmu_pages, link)
2070 kvm_mmu_zap_page(kvm, sp);
2071 spin_unlock(&kvm->mmu_lock);
2073 kvm_flush_remote_tlbs(kvm);
2076 static void kvm_mmu_remove_one_alloc_mmu_page(struct kvm *kvm)
2078 struct kvm_mmu_page *page;
2080 page = container_of(kvm->arch.active_mmu_pages.prev,
2081 struct kvm_mmu_page, link);
2082 kvm_mmu_zap_page(kvm, page);
2085 static int mmu_shrink(int nr_to_scan, gfp_t gfp_mask)
2088 struct kvm *kvm_freed = NULL;
2089 int cache_count = 0;
2091 spin_lock(&kvm_lock);
2093 list_for_each_entry(kvm, &vm_list, vm_list) {
2096 if (!down_read_trylock(&kvm->slots_lock))
2098 spin_lock(&kvm->mmu_lock);
2099 npages = kvm->arch.n_alloc_mmu_pages -
2100 kvm->arch.n_free_mmu_pages;
2101 cache_count += npages;
2102 if (!kvm_freed && nr_to_scan > 0 && npages > 0) {
2103 kvm_mmu_remove_one_alloc_mmu_page(kvm);
2109 spin_unlock(&kvm->mmu_lock);
2110 up_read(&kvm->slots_lock);
2113 list_move_tail(&kvm_freed->vm_list, &vm_list);
2115 spin_unlock(&kvm_lock);
2120 static struct shrinker mmu_shrinker = {
2121 .shrink = mmu_shrink,
2122 .seeks = DEFAULT_SEEKS * 10,
2125 static void mmu_destroy_caches(void)
2127 if (pte_chain_cache)
2128 kmem_cache_destroy(pte_chain_cache);
2129 if (rmap_desc_cache)
2130 kmem_cache_destroy(rmap_desc_cache);
2131 if (mmu_page_header_cache)
2132 kmem_cache_destroy(mmu_page_header_cache);
2135 void kvm_mmu_module_exit(void)
2137 mmu_destroy_caches();
2138 unregister_shrinker(&mmu_shrinker);
2141 int kvm_mmu_module_init(void)
2143 pte_chain_cache = kmem_cache_create("kvm_pte_chain",
2144 sizeof(struct kvm_pte_chain),
2146 if (!pte_chain_cache)
2148 rmap_desc_cache = kmem_cache_create("kvm_rmap_desc",
2149 sizeof(struct kvm_rmap_desc),
2151 if (!rmap_desc_cache)
2154 mmu_page_header_cache = kmem_cache_create("kvm_mmu_page_header",
2155 sizeof(struct kvm_mmu_page),
2157 if (!mmu_page_header_cache)
2160 register_shrinker(&mmu_shrinker);
2165 mmu_destroy_caches();
2170 * Caculate mmu pages needed for kvm.
2172 unsigned int kvm_mmu_calculate_mmu_pages(struct kvm *kvm)
2175 unsigned int nr_mmu_pages;
2176 unsigned int nr_pages = 0;
2178 for (i = 0; i < kvm->nmemslots; i++)
2179 nr_pages += kvm->memslots[i].npages;
2181 nr_mmu_pages = nr_pages * KVM_PERMILLE_MMU_PAGES / 1000;
2182 nr_mmu_pages = max(nr_mmu_pages,
2183 (unsigned int) KVM_MIN_ALLOC_MMU_PAGES);
2185 return nr_mmu_pages;
2188 static void *pv_mmu_peek_buffer(struct kvm_pv_mmu_op_buffer *buffer,
2191 if (len > buffer->len)
2196 static void *pv_mmu_read_buffer(struct kvm_pv_mmu_op_buffer *buffer,
2201 ret = pv_mmu_peek_buffer(buffer, len);
2206 buffer->processed += len;
2210 static int kvm_pv_mmu_write(struct kvm_vcpu *vcpu,
2211 gpa_t addr, gpa_t value)
2216 if (!is_long_mode(vcpu) && !is_pae(vcpu))
2219 r = mmu_topup_memory_caches(vcpu);
2223 if (!emulator_write_phys(vcpu, addr, &value, bytes))
2229 static int kvm_pv_mmu_flush_tlb(struct kvm_vcpu *vcpu)
2231 kvm_x86_ops->tlb_flush(vcpu);
2235 static int kvm_pv_mmu_release_pt(struct kvm_vcpu *vcpu, gpa_t addr)
2237 spin_lock(&vcpu->kvm->mmu_lock);
2238 mmu_unshadow(vcpu->kvm, addr >> PAGE_SHIFT);
2239 spin_unlock(&vcpu->kvm->mmu_lock);
2243 static int kvm_pv_mmu_op_one(struct kvm_vcpu *vcpu,
2244 struct kvm_pv_mmu_op_buffer *buffer)
2246 struct kvm_mmu_op_header *header;
2248 header = pv_mmu_peek_buffer(buffer, sizeof *header);
2251 switch (header->op) {
2252 case KVM_MMU_OP_WRITE_PTE: {
2253 struct kvm_mmu_op_write_pte *wpte;
2255 wpte = pv_mmu_read_buffer(buffer, sizeof *wpte);
2258 return kvm_pv_mmu_write(vcpu, wpte->pte_phys,
2261 case KVM_MMU_OP_FLUSH_TLB: {
2262 struct kvm_mmu_op_flush_tlb *ftlb;
2264 ftlb = pv_mmu_read_buffer(buffer, sizeof *ftlb);
2267 return kvm_pv_mmu_flush_tlb(vcpu);
2269 case KVM_MMU_OP_RELEASE_PT: {
2270 struct kvm_mmu_op_release_pt *rpt;
2272 rpt = pv_mmu_read_buffer(buffer, sizeof *rpt);
2275 return kvm_pv_mmu_release_pt(vcpu, rpt->pt_phys);
2281 int kvm_pv_mmu_op(struct kvm_vcpu *vcpu, unsigned long bytes,
2282 gpa_t addr, unsigned long *ret)
2285 struct kvm_pv_mmu_op_buffer *buffer = &vcpu->arch.mmu_op_buffer;
2287 buffer->ptr = buffer->buf;
2288 buffer->len = min_t(unsigned long, bytes, sizeof buffer->buf);
2289 buffer->processed = 0;
2291 r = kvm_read_guest(vcpu->kvm, addr, buffer->buf, buffer->len);
2295 while (buffer->len) {
2296 r = kvm_pv_mmu_op_one(vcpu, buffer);
2305 *ret = buffer->processed;
2311 static const char *audit_msg;
2313 static gva_t canonicalize(gva_t gva)
2315 #ifdef CONFIG_X86_64
2316 gva = (long long)(gva << 16) >> 16;
2321 static void audit_mappings_page(struct kvm_vcpu *vcpu, u64 page_pte,
2322 gva_t va, int level)
2324 u64 *pt = __va(page_pte & PT64_BASE_ADDR_MASK);
2326 gva_t va_delta = 1ul << (PAGE_SHIFT + 9 * (level - 1));
2328 for (i = 0; i < PT64_ENT_PER_PAGE; ++i, va += va_delta) {
2331 if (ent == shadow_trap_nonpresent_pte)
2334 va = canonicalize(va);
2336 if (ent == shadow_notrap_nonpresent_pte)
2337 printk(KERN_ERR "audit: (%s) nontrapping pte"
2338 " in nonleaf level: levels %d gva %lx"
2339 " level %d pte %llx\n", audit_msg,
2340 vcpu->arch.mmu.root_level, va, level, ent);
2342 audit_mappings_page(vcpu, ent, va, level - 1);
2344 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, va);
2345 hpa_t hpa = (hpa_t)gpa_to_pfn(vcpu, gpa) << PAGE_SHIFT;
2347 if (is_shadow_present_pte(ent)
2348 && (ent & PT64_BASE_ADDR_MASK) != hpa)
2349 printk(KERN_ERR "xx audit error: (%s) levels %d"
2350 " gva %lx gpa %llx hpa %llx ent %llx %d\n",
2351 audit_msg, vcpu->arch.mmu.root_level,
2353 is_shadow_present_pte(ent));
2354 else if (ent == shadow_notrap_nonpresent_pte
2355 && !is_error_hpa(hpa))
2356 printk(KERN_ERR "audit: (%s) notrap shadow,"
2357 " valid guest gva %lx\n", audit_msg, va);
2358 kvm_release_pfn_clean(pfn);
2364 static void audit_mappings(struct kvm_vcpu *vcpu)
2368 if (vcpu->arch.mmu.root_level == 4)
2369 audit_mappings_page(vcpu, vcpu->arch.mmu.root_hpa, 0, 4);
2371 for (i = 0; i < 4; ++i)
2372 if (vcpu->arch.mmu.pae_root[i] & PT_PRESENT_MASK)
2373 audit_mappings_page(vcpu,
2374 vcpu->arch.mmu.pae_root[i],
2379 static int count_rmaps(struct kvm_vcpu *vcpu)
2384 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
2385 struct kvm_memory_slot *m = &vcpu->kvm->memslots[i];
2386 struct kvm_rmap_desc *d;
2388 for (j = 0; j < m->npages; ++j) {
2389 unsigned long *rmapp = &m->rmap[j];
2393 if (!(*rmapp & 1)) {
2397 d = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
2399 for (k = 0; k < RMAP_EXT; ++k)
2400 if (d->shadow_ptes[k])
2411 static int count_writable_mappings(struct kvm_vcpu *vcpu)
2414 struct kvm_mmu_page *sp;
2417 list_for_each_entry(sp, &vcpu->kvm->arch.active_mmu_pages, link) {
2420 if (sp->role.level != PT_PAGE_TABLE_LEVEL)
2423 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
2426 if (!(ent & PT_PRESENT_MASK))
2428 if (!(ent & PT_WRITABLE_MASK))
2436 static void audit_rmap(struct kvm_vcpu *vcpu)
2438 int n_rmap = count_rmaps(vcpu);
2439 int n_actual = count_writable_mappings(vcpu);
2441 if (n_rmap != n_actual)
2442 printk(KERN_ERR "%s: (%s) rmap %d actual %d\n",
2443 __func__, audit_msg, n_rmap, n_actual);
2446 static void audit_write_protection(struct kvm_vcpu *vcpu)
2448 struct kvm_mmu_page *sp;
2449 struct kvm_memory_slot *slot;
2450 unsigned long *rmapp;
2453 list_for_each_entry(sp, &vcpu->kvm->arch.active_mmu_pages, link) {
2454 if (sp->role.metaphysical)
2457 slot = gfn_to_memslot(vcpu->kvm, sp->gfn);
2458 gfn = unalias_gfn(vcpu->kvm, sp->gfn);
2459 rmapp = &slot->rmap[gfn - slot->base_gfn];
2461 printk(KERN_ERR "%s: (%s) shadow page has writable"
2462 " mappings: gfn %lx role %x\n",
2463 __func__, audit_msg, sp->gfn,
2468 static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg)
2475 audit_write_protection(vcpu);
2476 audit_mappings(vcpu);