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 struct kvm_pv_mmu_op_buffer {
142 char buf[512] __aligned(sizeof(long));
145 struct kvm_rmap_desc {
146 u64 *shadow_ptes[RMAP_EXT];
147 struct kvm_rmap_desc *more;
150 static struct kmem_cache *pte_chain_cache;
151 static struct kmem_cache *rmap_desc_cache;
152 static struct kmem_cache *mmu_page_header_cache;
154 static u64 __read_mostly shadow_trap_nonpresent_pte;
155 static u64 __read_mostly shadow_notrap_nonpresent_pte;
156 static u64 __read_mostly shadow_base_present_pte;
157 static u64 __read_mostly shadow_nx_mask;
158 static u64 __read_mostly shadow_x_mask; /* mutual exclusive with nx_mask */
159 static u64 __read_mostly shadow_user_mask;
160 static u64 __read_mostly shadow_accessed_mask;
161 static u64 __read_mostly shadow_dirty_mask;
163 void kvm_mmu_set_nonpresent_ptes(u64 trap_pte, u64 notrap_pte)
165 shadow_trap_nonpresent_pte = trap_pte;
166 shadow_notrap_nonpresent_pte = notrap_pte;
168 EXPORT_SYMBOL_GPL(kvm_mmu_set_nonpresent_ptes);
170 void kvm_mmu_set_base_ptes(u64 base_pte)
172 shadow_base_present_pte = base_pte;
174 EXPORT_SYMBOL_GPL(kvm_mmu_set_base_ptes);
176 void kvm_mmu_set_mask_ptes(u64 user_mask, u64 accessed_mask,
177 u64 dirty_mask, u64 nx_mask, u64 x_mask)
179 shadow_user_mask = user_mask;
180 shadow_accessed_mask = accessed_mask;
181 shadow_dirty_mask = dirty_mask;
182 shadow_nx_mask = nx_mask;
183 shadow_x_mask = x_mask;
185 EXPORT_SYMBOL_GPL(kvm_mmu_set_mask_ptes);
187 static int is_write_protection(struct kvm_vcpu *vcpu)
189 return vcpu->arch.cr0 & X86_CR0_WP;
192 static int is_cpuid_PSE36(void)
197 static int is_nx(struct kvm_vcpu *vcpu)
199 return vcpu->arch.shadow_efer & EFER_NX;
202 static int is_present_pte(unsigned long pte)
204 return pte & PT_PRESENT_MASK;
207 static int is_shadow_present_pte(u64 pte)
209 return pte != shadow_trap_nonpresent_pte
210 && pte != shadow_notrap_nonpresent_pte;
213 static int is_large_pte(u64 pte)
215 return pte & PT_PAGE_SIZE_MASK;
218 static int is_writeble_pte(unsigned long pte)
220 return pte & PT_WRITABLE_MASK;
223 static int is_dirty_pte(unsigned long pte)
225 return pte & shadow_dirty_mask;
228 static int is_rmap_pte(u64 pte)
230 return is_shadow_present_pte(pte);
233 static pfn_t spte_to_pfn(u64 pte)
235 return (pte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT;
238 static gfn_t pse36_gfn_delta(u32 gpte)
240 int shift = 32 - PT32_DIR_PSE36_SHIFT - PAGE_SHIFT;
242 return (gpte & PT32_DIR_PSE36_MASK) << shift;
245 static void set_shadow_pte(u64 *sptep, u64 spte)
248 set_64bit((unsigned long *)sptep, spte);
250 set_64bit((unsigned long long *)sptep, spte);
254 static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache,
255 struct kmem_cache *base_cache, int min)
259 if (cache->nobjs >= min)
261 while (cache->nobjs < ARRAY_SIZE(cache->objects)) {
262 obj = kmem_cache_zalloc(base_cache, GFP_KERNEL);
265 cache->objects[cache->nobjs++] = obj;
270 static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
273 kfree(mc->objects[--mc->nobjs]);
276 static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache *cache,
281 if (cache->nobjs >= min)
283 while (cache->nobjs < ARRAY_SIZE(cache->objects)) {
284 page = alloc_page(GFP_KERNEL);
287 set_page_private(page, 0);
288 cache->objects[cache->nobjs++] = page_address(page);
293 static void mmu_free_memory_cache_page(struct kvm_mmu_memory_cache *mc)
296 free_page((unsigned long)mc->objects[--mc->nobjs]);
299 static int mmu_topup_memory_caches(struct kvm_vcpu *vcpu)
303 r = mmu_topup_memory_cache(&vcpu->arch.mmu_pte_chain_cache,
307 r = mmu_topup_memory_cache(&vcpu->arch.mmu_rmap_desc_cache,
311 r = mmu_topup_memory_cache_page(&vcpu->arch.mmu_page_cache, 8);
314 r = mmu_topup_memory_cache(&vcpu->arch.mmu_page_header_cache,
315 mmu_page_header_cache, 4);
320 static void mmu_free_memory_caches(struct kvm_vcpu *vcpu)
322 mmu_free_memory_cache(&vcpu->arch.mmu_pte_chain_cache);
323 mmu_free_memory_cache(&vcpu->arch.mmu_rmap_desc_cache);
324 mmu_free_memory_cache_page(&vcpu->arch.mmu_page_cache);
325 mmu_free_memory_cache(&vcpu->arch.mmu_page_header_cache);
328 static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc,
334 p = mc->objects[--mc->nobjs];
339 static struct kvm_pte_chain *mmu_alloc_pte_chain(struct kvm_vcpu *vcpu)
341 return mmu_memory_cache_alloc(&vcpu->arch.mmu_pte_chain_cache,
342 sizeof(struct kvm_pte_chain));
345 static void mmu_free_pte_chain(struct kvm_pte_chain *pc)
350 static struct kvm_rmap_desc *mmu_alloc_rmap_desc(struct kvm_vcpu *vcpu)
352 return mmu_memory_cache_alloc(&vcpu->arch.mmu_rmap_desc_cache,
353 sizeof(struct kvm_rmap_desc));
356 static void mmu_free_rmap_desc(struct kvm_rmap_desc *rd)
362 * Return the pointer to the largepage write count for a given
363 * gfn, handling slots that are not large page aligned.
365 static int *slot_largepage_idx(gfn_t gfn, struct kvm_memory_slot *slot)
369 idx = (gfn / KVM_PAGES_PER_HPAGE) -
370 (slot->base_gfn / KVM_PAGES_PER_HPAGE);
371 return &slot->lpage_info[idx].write_count;
374 static void account_shadowed(struct kvm *kvm, gfn_t gfn)
378 write_count = slot_largepage_idx(gfn, gfn_to_memslot(kvm, gfn));
382 static void unaccount_shadowed(struct kvm *kvm, gfn_t gfn)
386 write_count = slot_largepage_idx(gfn, gfn_to_memslot(kvm, gfn));
388 WARN_ON(*write_count < 0);
391 static int has_wrprotected_page(struct kvm *kvm, gfn_t gfn)
393 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
397 largepage_idx = slot_largepage_idx(gfn, slot);
398 return *largepage_idx;
404 static int host_largepage_backed(struct kvm *kvm, gfn_t gfn)
406 struct vm_area_struct *vma;
409 addr = gfn_to_hva(kvm, gfn);
410 if (kvm_is_error_hva(addr))
413 vma = find_vma(current->mm, addr);
414 if (vma && is_vm_hugetlb_page(vma))
420 static int is_largepage_backed(struct kvm_vcpu *vcpu, gfn_t large_gfn)
422 struct kvm_memory_slot *slot;
424 if (has_wrprotected_page(vcpu->kvm, large_gfn))
427 if (!host_largepage_backed(vcpu->kvm, large_gfn))
430 slot = gfn_to_memslot(vcpu->kvm, large_gfn);
431 if (slot && slot->dirty_bitmap)
438 * Take gfn and return the reverse mapping to it.
439 * Note: gfn must be unaliased before this function get called
442 static unsigned long *gfn_to_rmap(struct kvm *kvm, gfn_t gfn, int lpage)
444 struct kvm_memory_slot *slot;
447 slot = gfn_to_memslot(kvm, gfn);
449 return &slot->rmap[gfn - slot->base_gfn];
451 idx = (gfn / KVM_PAGES_PER_HPAGE) -
452 (slot->base_gfn / KVM_PAGES_PER_HPAGE);
454 return &slot->lpage_info[idx].rmap_pde;
458 * Reverse mapping data structures:
460 * If rmapp bit zero is zero, then rmapp point to the shadw page table entry
461 * that points to page_address(page).
463 * If rmapp bit zero is one, (then rmap & ~1) points to a struct kvm_rmap_desc
464 * containing more mappings.
466 static void rmap_add(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn, int lpage)
468 struct kvm_mmu_page *sp;
469 struct kvm_rmap_desc *desc;
470 unsigned long *rmapp;
473 if (!is_rmap_pte(*spte))
475 gfn = unalias_gfn(vcpu->kvm, gfn);
476 sp = page_header(__pa(spte));
477 sp->gfns[spte - sp->spt] = gfn;
478 rmapp = gfn_to_rmap(vcpu->kvm, gfn, lpage);
480 rmap_printk("rmap_add: %p %llx 0->1\n", spte, *spte);
481 *rmapp = (unsigned long)spte;
482 } else if (!(*rmapp & 1)) {
483 rmap_printk("rmap_add: %p %llx 1->many\n", spte, *spte);
484 desc = mmu_alloc_rmap_desc(vcpu);
485 desc->shadow_ptes[0] = (u64 *)*rmapp;
486 desc->shadow_ptes[1] = spte;
487 *rmapp = (unsigned long)desc | 1;
489 rmap_printk("rmap_add: %p %llx many->many\n", spte, *spte);
490 desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
491 while (desc->shadow_ptes[RMAP_EXT-1] && desc->more)
493 if (desc->shadow_ptes[RMAP_EXT-1]) {
494 desc->more = mmu_alloc_rmap_desc(vcpu);
497 for (i = 0; desc->shadow_ptes[i]; ++i)
499 desc->shadow_ptes[i] = spte;
503 static void rmap_desc_remove_entry(unsigned long *rmapp,
504 struct kvm_rmap_desc *desc,
506 struct kvm_rmap_desc *prev_desc)
510 for (j = RMAP_EXT - 1; !desc->shadow_ptes[j] && j > i; --j)
512 desc->shadow_ptes[i] = desc->shadow_ptes[j];
513 desc->shadow_ptes[j] = NULL;
516 if (!prev_desc && !desc->more)
517 *rmapp = (unsigned long)desc->shadow_ptes[0];
520 prev_desc->more = desc->more;
522 *rmapp = (unsigned long)desc->more | 1;
523 mmu_free_rmap_desc(desc);
526 static void rmap_remove(struct kvm *kvm, u64 *spte)
528 struct kvm_rmap_desc *desc;
529 struct kvm_rmap_desc *prev_desc;
530 struct kvm_mmu_page *sp;
532 unsigned long *rmapp;
535 if (!is_rmap_pte(*spte))
537 sp = page_header(__pa(spte));
538 pfn = spte_to_pfn(*spte);
539 if (*spte & shadow_accessed_mask)
540 kvm_set_pfn_accessed(pfn);
541 if (is_writeble_pte(*spte))
542 kvm_release_pfn_dirty(pfn);
544 kvm_release_pfn_clean(pfn);
545 rmapp = gfn_to_rmap(kvm, sp->gfns[spte - sp->spt], is_large_pte(*spte));
547 printk(KERN_ERR "rmap_remove: %p %llx 0->BUG\n", spte, *spte);
549 } else if (!(*rmapp & 1)) {
550 rmap_printk("rmap_remove: %p %llx 1->0\n", spte, *spte);
551 if ((u64 *)*rmapp != spte) {
552 printk(KERN_ERR "rmap_remove: %p %llx 1->BUG\n",
558 rmap_printk("rmap_remove: %p %llx many->many\n", spte, *spte);
559 desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
562 for (i = 0; i < RMAP_EXT && desc->shadow_ptes[i]; ++i)
563 if (desc->shadow_ptes[i] == spte) {
564 rmap_desc_remove_entry(rmapp,
576 static u64 *rmap_next(struct kvm *kvm, unsigned long *rmapp, u64 *spte)
578 struct kvm_rmap_desc *desc;
579 struct kvm_rmap_desc *prev_desc;
585 else if (!(*rmapp & 1)) {
587 return (u64 *)*rmapp;
590 desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
594 for (i = 0; i < RMAP_EXT && desc->shadow_ptes[i]; ++i) {
595 if (prev_spte == spte)
596 return desc->shadow_ptes[i];
597 prev_spte = desc->shadow_ptes[i];
604 static void rmap_write_protect(struct kvm *kvm, u64 gfn)
606 unsigned long *rmapp;
608 int write_protected = 0;
610 gfn = unalias_gfn(kvm, gfn);
611 rmapp = gfn_to_rmap(kvm, gfn, 0);
613 spte = rmap_next(kvm, rmapp, NULL);
616 BUG_ON(!(*spte & PT_PRESENT_MASK));
617 rmap_printk("rmap_write_protect: spte %p %llx\n", spte, *spte);
618 if (is_writeble_pte(*spte)) {
619 set_shadow_pte(spte, *spte & ~PT_WRITABLE_MASK);
622 spte = rmap_next(kvm, rmapp, spte);
624 if (write_protected) {
627 spte = rmap_next(kvm, rmapp, NULL);
628 pfn = spte_to_pfn(*spte);
629 kvm_set_pfn_dirty(pfn);
632 /* check for huge page mappings */
633 rmapp = gfn_to_rmap(kvm, gfn, 1);
634 spte = rmap_next(kvm, rmapp, NULL);
637 BUG_ON(!(*spte & PT_PRESENT_MASK));
638 BUG_ON((*spte & (PT_PAGE_SIZE_MASK|PT_PRESENT_MASK)) != (PT_PAGE_SIZE_MASK|PT_PRESENT_MASK));
639 pgprintk("rmap_write_protect(large): spte %p %llx %lld\n", spte, *spte, gfn);
640 if (is_writeble_pte(*spte)) {
641 rmap_remove(kvm, spte);
643 set_shadow_pte(spte, shadow_trap_nonpresent_pte);
647 spte = rmap_next(kvm, rmapp, spte);
651 kvm_flush_remote_tlbs(kvm);
653 account_shadowed(kvm, gfn);
656 static int kvm_unmap_rmapp(struct kvm *kvm, unsigned long *rmapp)
659 int need_tlb_flush = 0;
661 while ((spte = rmap_next(kvm, rmapp, NULL))) {
662 BUG_ON(!(*spte & PT_PRESENT_MASK));
663 rmap_printk("kvm_rmap_unmap_hva: spte %p %llx\n", spte, *spte);
664 rmap_remove(kvm, spte);
665 set_shadow_pte(spte, shadow_trap_nonpresent_pte);
668 return need_tlb_flush;
671 static int kvm_handle_hva(struct kvm *kvm, unsigned long hva,
672 int (*handler)(struct kvm *kvm, unsigned long *rmapp))
678 * If mmap_sem isn't taken, we can look the memslots with only
679 * the mmu_lock by skipping over the slots with userspace_addr == 0.
681 for (i = 0; i < kvm->nmemslots; i++) {
682 struct kvm_memory_slot *memslot = &kvm->memslots[i];
683 unsigned long start = memslot->userspace_addr;
686 /* mmu_lock protects userspace_addr */
690 end = start + (memslot->npages << PAGE_SHIFT);
691 if (hva >= start && hva < end) {
692 gfn_t gfn_offset = (hva - start) >> PAGE_SHIFT;
693 retval |= handler(kvm, &memslot->rmap[gfn_offset]);
694 retval |= handler(kvm,
695 &memslot->lpage_info[
697 KVM_PAGES_PER_HPAGE].rmap_pde);
704 int kvm_unmap_hva(struct kvm *kvm, unsigned long hva)
706 return kvm_handle_hva(kvm, hva, kvm_unmap_rmapp);
709 static int kvm_age_rmapp(struct kvm *kvm, unsigned long *rmapp)
714 /* always return old for EPT */
715 if (!shadow_accessed_mask)
718 spte = rmap_next(kvm, rmapp, NULL);
722 BUG_ON(!(_spte & PT_PRESENT_MASK));
723 _young = _spte & PT_ACCESSED_MASK;
726 clear_bit(PT_ACCESSED_SHIFT, (unsigned long *)spte);
728 spte = rmap_next(kvm, rmapp, spte);
733 int kvm_age_hva(struct kvm *kvm, unsigned long hva)
735 return kvm_handle_hva(kvm, hva, kvm_age_rmapp);
739 static int is_empty_shadow_page(u64 *spt)
744 for (pos = spt, end = pos + PAGE_SIZE / sizeof(u64); pos != end; pos++)
745 if (is_shadow_present_pte(*pos)) {
746 printk(KERN_ERR "%s: %p %llx\n", __func__,
754 static void kvm_mmu_free_page(struct kvm *kvm, struct kvm_mmu_page *sp)
756 ASSERT(is_empty_shadow_page(sp->spt));
758 __free_page(virt_to_page(sp->spt));
759 __free_page(virt_to_page(sp->gfns));
761 ++kvm->arch.n_free_mmu_pages;
764 static unsigned kvm_page_table_hashfn(gfn_t gfn)
766 return gfn & ((1 << KVM_MMU_HASH_SHIFT) - 1);
769 static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu,
772 struct kvm_mmu_page *sp;
774 sp = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_header_cache, sizeof *sp);
775 sp->spt = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_cache, PAGE_SIZE);
776 sp->gfns = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_cache, PAGE_SIZE);
777 set_page_private(virt_to_page(sp->spt), (unsigned long)sp);
778 list_add(&sp->link, &vcpu->kvm->arch.active_mmu_pages);
779 ASSERT(is_empty_shadow_page(sp->spt));
782 sp->parent_pte = parent_pte;
783 --vcpu->kvm->arch.n_free_mmu_pages;
787 static void mmu_page_add_parent_pte(struct kvm_vcpu *vcpu,
788 struct kvm_mmu_page *sp, u64 *parent_pte)
790 struct kvm_pte_chain *pte_chain;
791 struct hlist_node *node;
796 if (!sp->multimapped) {
797 u64 *old = sp->parent_pte;
800 sp->parent_pte = parent_pte;
804 pte_chain = mmu_alloc_pte_chain(vcpu);
805 INIT_HLIST_HEAD(&sp->parent_ptes);
806 hlist_add_head(&pte_chain->link, &sp->parent_ptes);
807 pte_chain->parent_ptes[0] = old;
809 hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link) {
810 if (pte_chain->parent_ptes[NR_PTE_CHAIN_ENTRIES-1])
812 for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i)
813 if (!pte_chain->parent_ptes[i]) {
814 pte_chain->parent_ptes[i] = parent_pte;
818 pte_chain = mmu_alloc_pte_chain(vcpu);
820 hlist_add_head(&pte_chain->link, &sp->parent_ptes);
821 pte_chain->parent_ptes[0] = parent_pte;
824 static void mmu_page_remove_parent_pte(struct kvm_mmu_page *sp,
827 struct kvm_pte_chain *pte_chain;
828 struct hlist_node *node;
831 if (!sp->multimapped) {
832 BUG_ON(sp->parent_pte != parent_pte);
833 sp->parent_pte = NULL;
836 hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link)
837 for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) {
838 if (!pte_chain->parent_ptes[i])
840 if (pte_chain->parent_ptes[i] != parent_pte)
842 while (i + 1 < NR_PTE_CHAIN_ENTRIES
843 && pte_chain->parent_ptes[i + 1]) {
844 pte_chain->parent_ptes[i]
845 = pte_chain->parent_ptes[i + 1];
848 pte_chain->parent_ptes[i] = NULL;
850 hlist_del(&pte_chain->link);
851 mmu_free_pte_chain(pte_chain);
852 if (hlist_empty(&sp->parent_ptes)) {
854 sp->parent_pte = NULL;
862 static void nonpaging_prefetch_page(struct kvm_vcpu *vcpu,
863 struct kvm_mmu_page *sp)
867 for (i = 0; i < PT64_ENT_PER_PAGE; ++i)
868 sp->spt[i] = shadow_trap_nonpresent_pte;
871 static struct kvm_mmu_page *kvm_mmu_lookup_page(struct kvm *kvm, gfn_t gfn)
874 struct hlist_head *bucket;
875 struct kvm_mmu_page *sp;
876 struct hlist_node *node;
878 pgprintk("%s: looking for gfn %lx\n", __func__, gfn);
879 index = kvm_page_table_hashfn(gfn);
880 bucket = &kvm->arch.mmu_page_hash[index];
881 hlist_for_each_entry(sp, node, bucket, hash_link)
882 if (sp->gfn == gfn && !sp->role.metaphysical
883 && !sp->role.invalid) {
884 pgprintk("%s: found role %x\n",
885 __func__, sp->role.word);
891 static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
899 union kvm_mmu_page_role role;
902 struct hlist_head *bucket;
903 struct kvm_mmu_page *sp;
904 struct hlist_node *node;
907 role.glevels = vcpu->arch.mmu.root_level;
909 role.metaphysical = metaphysical;
910 role.access = access;
911 if (vcpu->arch.mmu.root_level <= PT32_ROOT_LEVEL) {
912 quadrant = gaddr >> (PAGE_SHIFT + (PT64_PT_BITS * level));
913 quadrant &= (1 << ((PT32_PT_BITS - PT64_PT_BITS) * level)) - 1;
914 role.quadrant = quadrant;
916 pgprintk("%s: looking gfn %lx role %x\n", __func__,
918 index = kvm_page_table_hashfn(gfn);
919 bucket = &vcpu->kvm->arch.mmu_page_hash[index];
920 hlist_for_each_entry(sp, node, bucket, hash_link)
921 if (sp->gfn == gfn && sp->role.word == role.word) {
922 mmu_page_add_parent_pte(vcpu, sp, parent_pte);
923 pgprintk("%s: found\n", __func__);
926 ++vcpu->kvm->stat.mmu_cache_miss;
927 sp = kvm_mmu_alloc_page(vcpu, parent_pte);
930 pgprintk("%s: adding gfn %lx role %x\n", __func__, gfn, role.word);
933 hlist_add_head(&sp->hash_link, bucket);
935 rmap_write_protect(vcpu->kvm, gfn);
936 if (shadow_trap_nonpresent_pte != shadow_notrap_nonpresent_pte)
937 vcpu->arch.mmu.prefetch_page(vcpu, sp);
939 nonpaging_prefetch_page(vcpu, sp);
943 static void kvm_mmu_page_unlink_children(struct kvm *kvm,
944 struct kvm_mmu_page *sp)
952 if (sp->role.level == PT_PAGE_TABLE_LEVEL) {
953 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
954 if (is_shadow_present_pte(pt[i]))
955 rmap_remove(kvm, &pt[i]);
956 pt[i] = shadow_trap_nonpresent_pte;
958 kvm_flush_remote_tlbs(kvm);
962 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
965 if (is_shadow_present_pte(ent)) {
966 if (!is_large_pte(ent)) {
967 ent &= PT64_BASE_ADDR_MASK;
968 mmu_page_remove_parent_pte(page_header(ent),
972 rmap_remove(kvm, &pt[i]);
975 pt[i] = shadow_trap_nonpresent_pte;
977 kvm_flush_remote_tlbs(kvm);
980 static void kvm_mmu_put_page(struct kvm_mmu_page *sp, u64 *parent_pte)
982 mmu_page_remove_parent_pte(sp, parent_pte);
985 static void kvm_mmu_reset_last_pte_updated(struct kvm *kvm)
989 for (i = 0; i < KVM_MAX_VCPUS; ++i)
991 kvm->vcpus[i]->arch.last_pte_updated = NULL;
994 static void kvm_mmu_unlink_parents(struct kvm *kvm, struct kvm_mmu_page *sp)
998 while (sp->multimapped || sp->parent_pte) {
999 if (!sp->multimapped)
1000 parent_pte = sp->parent_pte;
1002 struct kvm_pte_chain *chain;
1004 chain = container_of(sp->parent_ptes.first,
1005 struct kvm_pte_chain, link);
1006 parent_pte = chain->parent_ptes[0];
1008 BUG_ON(!parent_pte);
1009 kvm_mmu_put_page(sp, parent_pte);
1010 set_shadow_pte(parent_pte, shadow_trap_nonpresent_pte);
1014 static void kvm_mmu_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp)
1016 ++kvm->stat.mmu_shadow_zapped;
1017 kvm_mmu_page_unlink_children(kvm, sp);
1018 kvm_mmu_unlink_parents(kvm, sp);
1019 if (!sp->root_count) {
1020 if (!sp->role.metaphysical && !sp->role.invalid)
1021 unaccount_shadowed(kvm, sp->gfn);
1022 hlist_del(&sp->hash_link);
1023 kvm_mmu_free_page(kvm, sp);
1025 int invalid = sp->role.invalid;
1026 list_move(&sp->link, &kvm->arch.active_mmu_pages);
1027 sp->role.invalid = 1;
1028 kvm_reload_remote_mmus(kvm);
1029 if (!sp->role.metaphysical && !invalid)
1030 unaccount_shadowed(kvm, sp->gfn);
1032 kvm_mmu_reset_last_pte_updated(kvm);
1036 * Changing the number of mmu pages allocated to the vm
1037 * Note: if kvm_nr_mmu_pages is too small, you will get dead lock
1039 void kvm_mmu_change_mmu_pages(struct kvm *kvm, unsigned int kvm_nr_mmu_pages)
1042 * If we set the number of mmu pages to be smaller be than the
1043 * number of actived pages , we must to free some mmu pages before we
1047 if ((kvm->arch.n_alloc_mmu_pages - kvm->arch.n_free_mmu_pages) >
1049 int n_used_mmu_pages = kvm->arch.n_alloc_mmu_pages
1050 - kvm->arch.n_free_mmu_pages;
1052 while (n_used_mmu_pages > kvm_nr_mmu_pages) {
1053 struct kvm_mmu_page *page;
1055 page = container_of(kvm->arch.active_mmu_pages.prev,
1056 struct kvm_mmu_page, link);
1057 kvm_mmu_zap_page(kvm, page);
1060 kvm->arch.n_free_mmu_pages = 0;
1063 kvm->arch.n_free_mmu_pages += kvm_nr_mmu_pages
1064 - kvm->arch.n_alloc_mmu_pages;
1066 kvm->arch.n_alloc_mmu_pages = kvm_nr_mmu_pages;
1069 static int kvm_mmu_unprotect_page(struct kvm *kvm, gfn_t gfn)
1072 struct hlist_head *bucket;
1073 struct kvm_mmu_page *sp;
1074 struct hlist_node *node, *n;
1077 pgprintk("%s: looking for gfn %lx\n", __func__, gfn);
1079 index = kvm_page_table_hashfn(gfn);
1080 bucket = &kvm->arch.mmu_page_hash[index];
1081 hlist_for_each_entry_safe(sp, node, n, bucket, hash_link)
1082 if (sp->gfn == gfn && !sp->role.metaphysical) {
1083 pgprintk("%s: gfn %lx role %x\n", __func__, gfn,
1085 kvm_mmu_zap_page(kvm, sp);
1091 static void mmu_unshadow(struct kvm *kvm, gfn_t gfn)
1093 struct kvm_mmu_page *sp;
1095 while ((sp = kvm_mmu_lookup_page(kvm, gfn)) != NULL) {
1096 pgprintk("%s: zap %lx %x\n", __func__, gfn, sp->role.word);
1097 kvm_mmu_zap_page(kvm, sp);
1101 static void page_header_update_slot(struct kvm *kvm, void *pte, gfn_t gfn)
1103 int slot = memslot_id(kvm, gfn_to_memslot(kvm, gfn));
1104 struct kvm_mmu_page *sp = page_header(__pa(pte));
1106 __set_bit(slot, &sp->slot_bitmap);
1109 struct page *gva_to_page(struct kvm_vcpu *vcpu, gva_t gva)
1113 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gva);
1115 if (gpa == UNMAPPED_GVA)
1118 down_read(¤t->mm->mmap_sem);
1119 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1120 up_read(¤t->mm->mmap_sem);
1125 static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *shadow_pte,
1126 unsigned pt_access, unsigned pte_access,
1127 int user_fault, int write_fault, int dirty,
1128 int *ptwrite, int largepage, gfn_t gfn,
1129 pfn_t pfn, bool speculative)
1132 int was_rmapped = 0;
1133 int was_writeble = is_writeble_pte(*shadow_pte);
1135 pgprintk("%s: spte %llx access %x write_fault %d"
1136 " user_fault %d gfn %lx\n",
1137 __func__, *shadow_pte, pt_access,
1138 write_fault, user_fault, gfn);
1140 if (is_rmap_pte(*shadow_pte)) {
1142 * If we overwrite a PTE page pointer with a 2MB PMD, unlink
1143 * the parent of the now unreachable PTE.
1145 if (largepage && !is_large_pte(*shadow_pte)) {
1146 struct kvm_mmu_page *child;
1147 u64 pte = *shadow_pte;
1149 child = page_header(pte & PT64_BASE_ADDR_MASK);
1150 mmu_page_remove_parent_pte(child, shadow_pte);
1151 } else if (pfn != spte_to_pfn(*shadow_pte)) {
1152 pgprintk("hfn old %lx new %lx\n",
1153 spte_to_pfn(*shadow_pte), pfn);
1154 rmap_remove(vcpu->kvm, shadow_pte);
1157 was_rmapped = is_large_pte(*shadow_pte);
1164 * We don't set the accessed bit, since we sometimes want to see
1165 * whether the guest actually used the pte (in order to detect
1168 spte = shadow_base_present_pte | shadow_dirty_mask;
1170 pte_access |= PT_ACCESSED_MASK;
1172 pte_access &= ~ACC_WRITE_MASK;
1173 if (pte_access & ACC_EXEC_MASK)
1174 spte |= shadow_x_mask;
1176 spte |= shadow_nx_mask;
1177 if (pte_access & ACC_USER_MASK)
1178 spte |= shadow_user_mask;
1180 spte |= PT_PAGE_SIZE_MASK;
1182 spte |= (u64)pfn << PAGE_SHIFT;
1184 if ((pte_access & ACC_WRITE_MASK)
1185 || (write_fault && !is_write_protection(vcpu) && !user_fault)) {
1186 struct kvm_mmu_page *shadow;
1188 spte |= PT_WRITABLE_MASK;
1190 shadow = kvm_mmu_lookup_page(vcpu->kvm, gfn);
1192 (largepage && has_wrprotected_page(vcpu->kvm, gfn))) {
1193 pgprintk("%s: found shadow page for %lx, marking ro\n",
1195 pte_access &= ~ACC_WRITE_MASK;
1196 if (is_writeble_pte(spte)) {
1197 spte &= ~PT_WRITABLE_MASK;
1198 kvm_x86_ops->tlb_flush(vcpu);
1205 if (pte_access & ACC_WRITE_MASK)
1206 mark_page_dirty(vcpu->kvm, gfn);
1208 pgprintk("%s: setting spte %llx\n", __func__, spte);
1209 pgprintk("instantiating %s PTE (%s) at %ld (%llx) addr %p\n",
1210 (spte&PT_PAGE_SIZE_MASK)? "2MB" : "4kB",
1211 (spte&PT_WRITABLE_MASK)?"RW":"R", gfn, spte, shadow_pte);
1212 set_shadow_pte(shadow_pte, spte);
1213 if (!was_rmapped && (spte & PT_PAGE_SIZE_MASK)
1214 && (spte & PT_PRESENT_MASK))
1215 ++vcpu->kvm->stat.lpages;
1217 page_header_update_slot(vcpu->kvm, shadow_pte, gfn);
1219 rmap_add(vcpu, shadow_pte, gfn, largepage);
1220 if (!is_rmap_pte(*shadow_pte))
1221 kvm_release_pfn_clean(pfn);
1224 kvm_release_pfn_dirty(pfn);
1226 kvm_release_pfn_clean(pfn);
1229 vcpu->arch.last_pte_updated = shadow_pte;
1230 vcpu->arch.last_pte_gfn = gfn;
1234 static void nonpaging_new_cr3(struct kvm_vcpu *vcpu)
1238 static int __direct_map(struct kvm_vcpu *vcpu, gpa_t v, int write,
1239 int largepage, gfn_t gfn, pfn_t pfn,
1242 hpa_t table_addr = vcpu->arch.mmu.root_hpa;
1246 u32 index = PT64_INDEX(v, level);
1249 ASSERT(VALID_PAGE(table_addr));
1250 table = __va(table_addr);
1253 mmu_set_spte(vcpu, &table[index], ACC_ALL, ACC_ALL,
1254 0, write, 1, &pt_write, 0, gfn, pfn, false);
1258 if (largepage && level == 2) {
1259 mmu_set_spte(vcpu, &table[index], ACC_ALL, ACC_ALL,
1260 0, write, 1, &pt_write, 1, gfn, pfn, false);
1264 if (table[index] == shadow_trap_nonpresent_pte) {
1265 struct kvm_mmu_page *new_table;
1268 pseudo_gfn = (v & PT64_DIR_BASE_ADDR_MASK)
1270 new_table = kvm_mmu_get_page(vcpu, pseudo_gfn,
1272 1, ACC_ALL, &table[index]);
1274 pgprintk("nonpaging_map: ENOMEM\n");
1275 kvm_release_pfn_clean(pfn);
1279 set_shadow_pte(&table[index],
1280 __pa(new_table->spt)
1281 | PT_PRESENT_MASK | PT_WRITABLE_MASK
1282 | shadow_user_mask | shadow_x_mask);
1284 table_addr = table[index] & PT64_BASE_ADDR_MASK;
1288 static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, int write, gfn_t gfn)
1293 unsigned long mmu_seq;
1295 down_read(¤t->mm->mmap_sem);
1296 if (is_largepage_backed(vcpu, gfn & ~(KVM_PAGES_PER_HPAGE-1))) {
1297 gfn &= ~(KVM_PAGES_PER_HPAGE-1);
1301 mmu_seq = vcpu->kvm->mmu_notifier_seq;
1302 /* implicit mb(), we'll read before PT lock is unlocked */
1303 pfn = gfn_to_pfn(vcpu->kvm, gfn);
1304 up_read(¤t->mm->mmap_sem);
1307 if (is_error_pfn(pfn)) {
1308 kvm_release_pfn_clean(pfn);
1312 spin_lock(&vcpu->kvm->mmu_lock);
1313 if (mmu_notifier_retry(vcpu, mmu_seq))
1315 kvm_mmu_free_some_pages(vcpu);
1316 r = __direct_map(vcpu, v, write, largepage, gfn, pfn,
1318 spin_unlock(&vcpu->kvm->mmu_lock);
1324 spin_unlock(&vcpu->kvm->mmu_lock);
1325 kvm_release_pfn_clean(pfn);
1330 static void mmu_free_roots(struct kvm_vcpu *vcpu)
1333 struct kvm_mmu_page *sp;
1335 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
1337 spin_lock(&vcpu->kvm->mmu_lock);
1338 if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) {
1339 hpa_t root = vcpu->arch.mmu.root_hpa;
1341 sp = page_header(root);
1343 if (!sp->root_count && sp->role.invalid)
1344 kvm_mmu_zap_page(vcpu->kvm, sp);
1345 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
1346 spin_unlock(&vcpu->kvm->mmu_lock);
1349 for (i = 0; i < 4; ++i) {
1350 hpa_t root = vcpu->arch.mmu.pae_root[i];
1353 root &= PT64_BASE_ADDR_MASK;
1354 sp = page_header(root);
1356 if (!sp->root_count && sp->role.invalid)
1357 kvm_mmu_zap_page(vcpu->kvm, sp);
1359 vcpu->arch.mmu.pae_root[i] = INVALID_PAGE;
1361 spin_unlock(&vcpu->kvm->mmu_lock);
1362 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
1365 static void mmu_alloc_roots(struct kvm_vcpu *vcpu)
1369 struct kvm_mmu_page *sp;
1370 int metaphysical = 0;
1372 root_gfn = vcpu->arch.cr3 >> PAGE_SHIFT;
1374 if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) {
1375 hpa_t root = vcpu->arch.mmu.root_hpa;
1377 ASSERT(!VALID_PAGE(root));
1380 sp = kvm_mmu_get_page(vcpu, root_gfn, 0,
1381 PT64_ROOT_LEVEL, metaphysical,
1383 root = __pa(sp->spt);
1385 vcpu->arch.mmu.root_hpa = root;
1388 metaphysical = !is_paging(vcpu);
1391 for (i = 0; i < 4; ++i) {
1392 hpa_t root = vcpu->arch.mmu.pae_root[i];
1394 ASSERT(!VALID_PAGE(root));
1395 if (vcpu->arch.mmu.root_level == PT32E_ROOT_LEVEL) {
1396 if (!is_present_pte(vcpu->arch.pdptrs[i])) {
1397 vcpu->arch.mmu.pae_root[i] = 0;
1400 root_gfn = vcpu->arch.pdptrs[i] >> PAGE_SHIFT;
1401 } else if (vcpu->arch.mmu.root_level == 0)
1403 sp = kvm_mmu_get_page(vcpu, root_gfn, i << 30,
1404 PT32_ROOT_LEVEL, metaphysical,
1406 root = __pa(sp->spt);
1408 vcpu->arch.mmu.pae_root[i] = root | PT_PRESENT_MASK;
1410 vcpu->arch.mmu.root_hpa = __pa(vcpu->arch.mmu.pae_root);
1413 static gpa_t nonpaging_gva_to_gpa(struct kvm_vcpu *vcpu, gva_t vaddr)
1418 static int nonpaging_page_fault(struct kvm_vcpu *vcpu, gva_t gva,
1424 pgprintk("%s: gva %lx error %x\n", __func__, gva, error_code);
1425 r = mmu_topup_memory_caches(vcpu);
1430 ASSERT(VALID_PAGE(vcpu->arch.mmu.root_hpa));
1432 gfn = gva >> PAGE_SHIFT;
1434 return nonpaging_map(vcpu, gva & PAGE_MASK,
1435 error_code & PFERR_WRITE_MASK, gfn);
1438 static int tdp_page_fault(struct kvm_vcpu *vcpu, gva_t gpa,
1444 gfn_t gfn = gpa >> PAGE_SHIFT;
1445 unsigned long mmu_seq;
1448 ASSERT(VALID_PAGE(vcpu->arch.mmu.root_hpa));
1450 r = mmu_topup_memory_caches(vcpu);
1454 down_read(¤t->mm->mmap_sem);
1455 if (is_largepage_backed(vcpu, gfn & ~(KVM_PAGES_PER_HPAGE-1))) {
1456 gfn &= ~(KVM_PAGES_PER_HPAGE-1);
1459 mmu_seq = vcpu->kvm->mmu_notifier_seq;
1460 /* implicit mb(), we'll read before PT lock is unlocked */
1461 pfn = gfn_to_pfn(vcpu->kvm, gfn);
1462 up_read(¤t->mm->mmap_sem);
1463 if (is_error_pfn(pfn)) {
1464 kvm_release_pfn_clean(pfn);
1467 spin_lock(&vcpu->kvm->mmu_lock);
1468 if (mmu_notifier_retry(vcpu, mmu_seq))
1470 kvm_mmu_free_some_pages(vcpu);
1471 r = __direct_map(vcpu, gpa, error_code & PFERR_WRITE_MASK,
1472 largepage, gfn, pfn, kvm_x86_ops->get_tdp_level());
1473 spin_unlock(&vcpu->kvm->mmu_lock);
1478 spin_unlock(&vcpu->kvm->mmu_lock);
1479 kvm_release_pfn_clean(pfn);
1483 static void nonpaging_free(struct kvm_vcpu *vcpu)
1485 mmu_free_roots(vcpu);
1488 static int nonpaging_init_context(struct kvm_vcpu *vcpu)
1490 struct kvm_mmu *context = &vcpu->arch.mmu;
1492 context->new_cr3 = nonpaging_new_cr3;
1493 context->page_fault = nonpaging_page_fault;
1494 context->gva_to_gpa = nonpaging_gva_to_gpa;
1495 context->free = nonpaging_free;
1496 context->prefetch_page = nonpaging_prefetch_page;
1497 context->root_level = 0;
1498 context->shadow_root_level = PT32E_ROOT_LEVEL;
1499 context->root_hpa = INVALID_PAGE;
1503 void kvm_mmu_flush_tlb(struct kvm_vcpu *vcpu)
1505 ++vcpu->stat.tlb_flush;
1506 kvm_x86_ops->tlb_flush(vcpu);
1509 static void paging_new_cr3(struct kvm_vcpu *vcpu)
1511 pgprintk("%s: cr3 %lx\n", __func__, vcpu->arch.cr3);
1512 mmu_free_roots(vcpu);
1515 static void inject_page_fault(struct kvm_vcpu *vcpu,
1519 kvm_inject_page_fault(vcpu, addr, err_code);
1522 static void paging_free(struct kvm_vcpu *vcpu)
1524 nonpaging_free(vcpu);
1528 #include "paging_tmpl.h"
1532 #include "paging_tmpl.h"
1535 static int paging64_init_context_common(struct kvm_vcpu *vcpu, int level)
1537 struct kvm_mmu *context = &vcpu->arch.mmu;
1539 ASSERT(is_pae(vcpu));
1540 context->new_cr3 = paging_new_cr3;
1541 context->page_fault = paging64_page_fault;
1542 context->gva_to_gpa = paging64_gva_to_gpa;
1543 context->prefetch_page = paging64_prefetch_page;
1544 context->free = paging_free;
1545 context->root_level = level;
1546 context->shadow_root_level = level;
1547 context->root_hpa = INVALID_PAGE;
1551 static int paging64_init_context(struct kvm_vcpu *vcpu)
1553 return paging64_init_context_common(vcpu, PT64_ROOT_LEVEL);
1556 static int paging32_init_context(struct kvm_vcpu *vcpu)
1558 struct kvm_mmu *context = &vcpu->arch.mmu;
1560 context->new_cr3 = paging_new_cr3;
1561 context->page_fault = paging32_page_fault;
1562 context->gva_to_gpa = paging32_gva_to_gpa;
1563 context->free = paging_free;
1564 context->prefetch_page = paging32_prefetch_page;
1565 context->root_level = PT32_ROOT_LEVEL;
1566 context->shadow_root_level = PT32E_ROOT_LEVEL;
1567 context->root_hpa = INVALID_PAGE;
1571 static int paging32E_init_context(struct kvm_vcpu *vcpu)
1573 return paging64_init_context_common(vcpu, PT32E_ROOT_LEVEL);
1576 static int init_kvm_tdp_mmu(struct kvm_vcpu *vcpu)
1578 struct kvm_mmu *context = &vcpu->arch.mmu;
1580 context->new_cr3 = nonpaging_new_cr3;
1581 context->page_fault = tdp_page_fault;
1582 context->free = nonpaging_free;
1583 context->prefetch_page = nonpaging_prefetch_page;
1584 context->shadow_root_level = kvm_x86_ops->get_tdp_level();
1585 context->root_hpa = INVALID_PAGE;
1587 if (!is_paging(vcpu)) {
1588 context->gva_to_gpa = nonpaging_gva_to_gpa;
1589 context->root_level = 0;
1590 } else if (is_long_mode(vcpu)) {
1591 context->gva_to_gpa = paging64_gva_to_gpa;
1592 context->root_level = PT64_ROOT_LEVEL;
1593 } else if (is_pae(vcpu)) {
1594 context->gva_to_gpa = paging64_gva_to_gpa;
1595 context->root_level = PT32E_ROOT_LEVEL;
1597 context->gva_to_gpa = paging32_gva_to_gpa;
1598 context->root_level = PT32_ROOT_LEVEL;
1604 static int init_kvm_softmmu(struct kvm_vcpu *vcpu)
1607 ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa));
1609 if (!is_paging(vcpu))
1610 return nonpaging_init_context(vcpu);
1611 else if (is_long_mode(vcpu))
1612 return paging64_init_context(vcpu);
1613 else if (is_pae(vcpu))
1614 return paging32E_init_context(vcpu);
1616 return paging32_init_context(vcpu);
1619 static int init_kvm_mmu(struct kvm_vcpu *vcpu)
1621 vcpu->arch.update_pte.pfn = bad_pfn;
1624 return init_kvm_tdp_mmu(vcpu);
1626 return init_kvm_softmmu(vcpu);
1629 static void destroy_kvm_mmu(struct kvm_vcpu *vcpu)
1632 if (VALID_PAGE(vcpu->arch.mmu.root_hpa)) {
1633 vcpu->arch.mmu.free(vcpu);
1634 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
1638 int kvm_mmu_reset_context(struct kvm_vcpu *vcpu)
1640 destroy_kvm_mmu(vcpu);
1641 return init_kvm_mmu(vcpu);
1643 EXPORT_SYMBOL_GPL(kvm_mmu_reset_context);
1645 int kvm_mmu_load(struct kvm_vcpu *vcpu)
1649 r = mmu_topup_memory_caches(vcpu);
1652 spin_lock(&vcpu->kvm->mmu_lock);
1653 kvm_mmu_free_some_pages(vcpu);
1654 mmu_alloc_roots(vcpu);
1655 spin_unlock(&vcpu->kvm->mmu_lock);
1656 kvm_x86_ops->set_cr3(vcpu, vcpu->arch.mmu.root_hpa);
1657 kvm_mmu_flush_tlb(vcpu);
1661 EXPORT_SYMBOL_GPL(kvm_mmu_load);
1663 void kvm_mmu_unload(struct kvm_vcpu *vcpu)
1665 mmu_free_roots(vcpu);
1668 static void mmu_pte_write_zap_pte(struct kvm_vcpu *vcpu,
1669 struct kvm_mmu_page *sp,
1673 struct kvm_mmu_page *child;
1676 if (is_shadow_present_pte(pte)) {
1677 if (sp->role.level == PT_PAGE_TABLE_LEVEL ||
1679 rmap_remove(vcpu->kvm, spte);
1681 child = page_header(pte & PT64_BASE_ADDR_MASK);
1682 mmu_page_remove_parent_pte(child, spte);
1685 set_shadow_pte(spte, shadow_trap_nonpresent_pte);
1686 if (is_large_pte(pte))
1687 --vcpu->kvm->stat.lpages;
1690 static void mmu_pte_write_new_pte(struct kvm_vcpu *vcpu,
1691 struct kvm_mmu_page *sp,
1695 if (sp->role.level != PT_PAGE_TABLE_LEVEL) {
1696 if (!vcpu->arch.update_pte.largepage ||
1697 sp->role.glevels == PT32_ROOT_LEVEL) {
1698 ++vcpu->kvm->stat.mmu_pde_zapped;
1703 ++vcpu->kvm->stat.mmu_pte_updated;
1704 if (sp->role.glevels == PT32_ROOT_LEVEL)
1705 paging32_update_pte(vcpu, sp, spte, new);
1707 paging64_update_pte(vcpu, sp, spte, new);
1710 static bool need_remote_flush(u64 old, u64 new)
1712 if (!is_shadow_present_pte(old))
1714 if (!is_shadow_present_pte(new))
1716 if ((old ^ new) & PT64_BASE_ADDR_MASK)
1718 old ^= PT64_NX_MASK;
1719 new ^= PT64_NX_MASK;
1720 return (old & ~new & PT64_PERM_MASK) != 0;
1723 static void mmu_pte_write_flush_tlb(struct kvm_vcpu *vcpu, u64 old, u64 new)
1725 if (need_remote_flush(old, new))
1726 kvm_flush_remote_tlbs(vcpu->kvm);
1728 kvm_mmu_flush_tlb(vcpu);
1731 static bool last_updated_pte_accessed(struct kvm_vcpu *vcpu)
1733 u64 *spte = vcpu->arch.last_pte_updated;
1735 return !!(spte && (*spte & shadow_accessed_mask));
1738 static void mmu_guess_page_from_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
1739 const u8 *new, int bytes)
1746 vcpu->arch.update_pte.largepage = 0;
1748 if (bytes != 4 && bytes != 8)
1752 * Assume that the pte write on a page table of the same type
1753 * as the current vcpu paging mode. This is nearly always true
1754 * (might be false while changing modes). Note it is verified later
1758 /* Handle a 32-bit guest writing two halves of a 64-bit gpte */
1759 if ((bytes == 4) && (gpa % 4 == 0)) {
1760 r = kvm_read_guest(vcpu->kvm, gpa & ~(u64)7, &gpte, 8);
1763 memcpy((void *)&gpte + (gpa % 8), new, 4);
1764 } else if ((bytes == 8) && (gpa % 8 == 0)) {
1765 memcpy((void *)&gpte, new, 8);
1768 if ((bytes == 4) && (gpa % 4 == 0))
1769 memcpy((void *)&gpte, new, 4);
1771 if (!is_present_pte(gpte))
1773 gfn = (gpte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT;
1775 down_read(¤t->mm->mmap_sem);
1776 if (is_large_pte(gpte) && is_largepage_backed(vcpu, gfn)) {
1777 gfn &= ~(KVM_PAGES_PER_HPAGE-1);
1778 vcpu->arch.update_pte.largepage = 1;
1780 vcpu->arch.update_pte.mmu_seq = vcpu->kvm->mmu_notifier_seq;
1781 /* implicit mb(), we'll read before PT lock is unlocked */
1782 pfn = gfn_to_pfn(vcpu->kvm, gfn);
1783 up_read(¤t->mm->mmap_sem);
1785 if (is_error_pfn(pfn)) {
1786 kvm_release_pfn_clean(pfn);
1789 vcpu->arch.update_pte.gfn = gfn;
1790 vcpu->arch.update_pte.pfn = pfn;
1793 static void kvm_mmu_access_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1795 u64 *spte = vcpu->arch.last_pte_updated;
1798 && vcpu->arch.last_pte_gfn == gfn
1799 && shadow_accessed_mask
1800 && !(*spte & shadow_accessed_mask)
1801 && is_shadow_present_pte(*spte))
1802 set_bit(PT_ACCESSED_SHIFT, (unsigned long *)spte);
1805 void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
1806 const u8 *new, int bytes)
1808 gfn_t gfn = gpa >> PAGE_SHIFT;
1809 struct kvm_mmu_page *sp;
1810 struct hlist_node *node, *n;
1811 struct hlist_head *bucket;
1815 unsigned offset = offset_in_page(gpa);
1817 unsigned page_offset;
1818 unsigned misaligned;
1825 pgprintk("%s: gpa %llx bytes %d\n", __func__, gpa, bytes);
1826 mmu_guess_page_from_pte_write(vcpu, gpa, new, bytes);
1827 spin_lock(&vcpu->kvm->mmu_lock);
1828 kvm_mmu_access_page(vcpu, gfn);
1829 kvm_mmu_free_some_pages(vcpu);
1830 ++vcpu->kvm->stat.mmu_pte_write;
1831 kvm_mmu_audit(vcpu, "pre pte write");
1832 if (gfn == vcpu->arch.last_pt_write_gfn
1833 && !last_updated_pte_accessed(vcpu)) {
1834 ++vcpu->arch.last_pt_write_count;
1835 if (vcpu->arch.last_pt_write_count >= 3)
1838 vcpu->arch.last_pt_write_gfn = gfn;
1839 vcpu->arch.last_pt_write_count = 1;
1840 vcpu->arch.last_pte_updated = NULL;
1842 index = kvm_page_table_hashfn(gfn);
1843 bucket = &vcpu->kvm->arch.mmu_page_hash[index];
1844 hlist_for_each_entry_safe(sp, node, n, bucket, hash_link) {
1845 if (sp->gfn != gfn || sp->role.metaphysical)
1847 pte_size = sp->role.glevels == PT32_ROOT_LEVEL ? 4 : 8;
1848 misaligned = (offset ^ (offset + bytes - 1)) & ~(pte_size - 1);
1849 misaligned |= bytes < 4;
1850 if (misaligned || flooded) {
1852 * Misaligned accesses are too much trouble to fix
1853 * up; also, they usually indicate a page is not used
1856 * If we're seeing too many writes to a page,
1857 * it may no longer be a page table, or we may be
1858 * forking, in which case it is better to unmap the
1861 pgprintk("misaligned: gpa %llx bytes %d role %x\n",
1862 gpa, bytes, sp->role.word);
1863 kvm_mmu_zap_page(vcpu->kvm, sp);
1864 ++vcpu->kvm->stat.mmu_flooded;
1867 page_offset = offset;
1868 level = sp->role.level;
1870 if (sp->role.glevels == PT32_ROOT_LEVEL) {
1871 page_offset <<= 1; /* 32->64 */
1873 * A 32-bit pde maps 4MB while the shadow pdes map
1874 * only 2MB. So we need to double the offset again
1875 * and zap two pdes instead of one.
1877 if (level == PT32_ROOT_LEVEL) {
1878 page_offset &= ~7; /* kill rounding error */
1882 quadrant = page_offset >> PAGE_SHIFT;
1883 page_offset &= ~PAGE_MASK;
1884 if (quadrant != sp->role.quadrant)
1887 spte = &sp->spt[page_offset / sizeof(*spte)];
1888 if ((gpa & (pte_size - 1)) || (bytes < pte_size)) {
1890 r = kvm_read_guest_atomic(vcpu->kvm,
1891 gpa & ~(u64)(pte_size - 1),
1893 new = (const void *)&gentry;
1899 mmu_pte_write_zap_pte(vcpu, sp, spte);
1901 mmu_pte_write_new_pte(vcpu, sp, spte, new);
1902 mmu_pte_write_flush_tlb(vcpu, entry, *spte);
1906 kvm_mmu_audit(vcpu, "post pte write");
1907 spin_unlock(&vcpu->kvm->mmu_lock);
1908 if (!is_error_pfn(vcpu->arch.update_pte.pfn)) {
1909 kvm_release_pfn_clean(vcpu->arch.update_pte.pfn);
1910 vcpu->arch.update_pte.pfn = bad_pfn;
1914 int kvm_mmu_unprotect_page_virt(struct kvm_vcpu *vcpu, gva_t gva)
1919 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gva);
1921 spin_lock(&vcpu->kvm->mmu_lock);
1922 r = kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1923 spin_unlock(&vcpu->kvm->mmu_lock);
1926 EXPORT_SYMBOL_GPL(kvm_mmu_unprotect_page_virt);
1928 void __kvm_mmu_free_some_pages(struct kvm_vcpu *vcpu)
1930 while (vcpu->kvm->arch.n_free_mmu_pages < KVM_REFILL_PAGES) {
1931 struct kvm_mmu_page *sp;
1933 sp = container_of(vcpu->kvm->arch.active_mmu_pages.prev,
1934 struct kvm_mmu_page, link);
1935 kvm_mmu_zap_page(vcpu->kvm, sp);
1936 ++vcpu->kvm->stat.mmu_recycled;
1940 int kvm_mmu_page_fault(struct kvm_vcpu *vcpu, gva_t cr2, u32 error_code)
1943 enum emulation_result er;
1945 r = vcpu->arch.mmu.page_fault(vcpu, cr2, error_code);
1954 r = mmu_topup_memory_caches(vcpu);
1958 er = emulate_instruction(vcpu, vcpu->run, cr2, error_code, 0);
1963 case EMULATE_DO_MMIO:
1964 ++vcpu->stat.mmio_exits;
1967 kvm_report_emulation_failure(vcpu, "pagetable");
1975 EXPORT_SYMBOL_GPL(kvm_mmu_page_fault);
1977 void kvm_enable_tdp(void)
1981 EXPORT_SYMBOL_GPL(kvm_enable_tdp);
1983 void kvm_disable_tdp(void)
1985 tdp_enabled = false;
1987 EXPORT_SYMBOL_GPL(kvm_disable_tdp);
1989 static void free_mmu_pages(struct kvm_vcpu *vcpu)
1991 struct kvm_mmu_page *sp;
1993 while (!list_empty(&vcpu->kvm->arch.active_mmu_pages)) {
1994 sp = container_of(vcpu->kvm->arch.active_mmu_pages.next,
1995 struct kvm_mmu_page, link);
1996 kvm_mmu_zap_page(vcpu->kvm, sp);
1999 free_page((unsigned long)vcpu->arch.mmu.pae_root);
2002 static int alloc_mmu_pages(struct kvm_vcpu *vcpu)
2009 if (vcpu->kvm->arch.n_requested_mmu_pages)
2010 vcpu->kvm->arch.n_free_mmu_pages =
2011 vcpu->kvm->arch.n_requested_mmu_pages;
2013 vcpu->kvm->arch.n_free_mmu_pages =
2014 vcpu->kvm->arch.n_alloc_mmu_pages;
2016 * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64.
2017 * Therefore we need to allocate shadow page tables in the first
2018 * 4GB of memory, which happens to fit the DMA32 zone.
2020 page = alloc_page(GFP_KERNEL | __GFP_DMA32);
2023 vcpu->arch.mmu.pae_root = page_address(page);
2024 for (i = 0; i < 4; ++i)
2025 vcpu->arch.mmu.pae_root[i] = INVALID_PAGE;
2030 free_mmu_pages(vcpu);
2034 int kvm_mmu_create(struct kvm_vcpu *vcpu)
2037 ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa));
2039 return alloc_mmu_pages(vcpu);
2042 int kvm_mmu_setup(struct kvm_vcpu *vcpu)
2045 ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa));
2047 return init_kvm_mmu(vcpu);
2050 void kvm_mmu_destroy(struct kvm_vcpu *vcpu)
2054 destroy_kvm_mmu(vcpu);
2055 free_mmu_pages(vcpu);
2056 mmu_free_memory_caches(vcpu);
2059 void kvm_mmu_slot_remove_write_access(struct kvm *kvm, int slot)
2061 struct kvm_mmu_page *sp;
2063 list_for_each_entry(sp, &kvm->arch.active_mmu_pages, link) {
2067 if (!test_bit(slot, &sp->slot_bitmap))
2071 for (i = 0; i < PT64_ENT_PER_PAGE; ++i)
2073 if (pt[i] & PT_WRITABLE_MASK)
2074 pt[i] &= ~PT_WRITABLE_MASK;
2078 void kvm_mmu_zap_all(struct kvm *kvm)
2080 struct kvm_mmu_page *sp, *node;
2082 spin_lock(&kvm->mmu_lock);
2083 list_for_each_entry_safe(sp, node, &kvm->arch.active_mmu_pages, link)
2084 kvm_mmu_zap_page(kvm, sp);
2085 spin_unlock(&kvm->mmu_lock);
2087 kvm_flush_remote_tlbs(kvm);
2090 static void kvm_mmu_remove_one_alloc_mmu_page(struct kvm *kvm)
2092 struct kvm_mmu_page *page;
2094 page = container_of(kvm->arch.active_mmu_pages.prev,
2095 struct kvm_mmu_page, link);
2096 kvm_mmu_zap_page(kvm, page);
2099 static int mmu_shrink(int nr_to_scan, gfp_t gfp_mask)
2102 struct kvm *kvm_freed = NULL;
2103 int cache_count = 0;
2105 spin_lock(&kvm_lock);
2107 list_for_each_entry(kvm, &vm_list, vm_list) {
2110 if (!down_read_trylock(&kvm->slots_lock))
2112 spin_lock(&kvm->mmu_lock);
2113 npages = kvm->arch.n_alloc_mmu_pages -
2114 kvm->arch.n_free_mmu_pages;
2115 cache_count += npages;
2116 if (!kvm_freed && nr_to_scan > 0 && npages > 0) {
2117 kvm_mmu_remove_one_alloc_mmu_page(kvm);
2123 spin_unlock(&kvm->mmu_lock);
2124 up_read(&kvm->slots_lock);
2127 list_move_tail(&kvm_freed->vm_list, &vm_list);
2129 spin_unlock(&kvm_lock);
2134 static struct shrinker mmu_shrinker = {
2135 .shrink = mmu_shrink,
2136 .seeks = DEFAULT_SEEKS * 10,
2139 static void mmu_destroy_caches(void)
2141 if (pte_chain_cache)
2142 kmem_cache_destroy(pte_chain_cache);
2143 if (rmap_desc_cache)
2144 kmem_cache_destroy(rmap_desc_cache);
2145 if (mmu_page_header_cache)
2146 kmem_cache_destroy(mmu_page_header_cache);
2149 void kvm_mmu_module_exit(void)
2151 mmu_destroy_caches();
2152 unregister_shrinker(&mmu_shrinker);
2155 int kvm_mmu_module_init(void)
2157 pte_chain_cache = kmem_cache_create("kvm_pte_chain",
2158 sizeof(struct kvm_pte_chain),
2160 if (!pte_chain_cache)
2162 rmap_desc_cache = kmem_cache_create("kvm_rmap_desc",
2163 sizeof(struct kvm_rmap_desc),
2165 if (!rmap_desc_cache)
2168 mmu_page_header_cache = kmem_cache_create("kvm_mmu_page_header",
2169 sizeof(struct kvm_mmu_page),
2171 if (!mmu_page_header_cache)
2174 register_shrinker(&mmu_shrinker);
2179 mmu_destroy_caches();
2184 * Caculate mmu pages needed for kvm.
2186 unsigned int kvm_mmu_calculate_mmu_pages(struct kvm *kvm)
2189 unsigned int nr_mmu_pages;
2190 unsigned int nr_pages = 0;
2192 for (i = 0; i < kvm->nmemslots; i++)
2193 nr_pages += kvm->memslots[i].npages;
2195 nr_mmu_pages = nr_pages * KVM_PERMILLE_MMU_PAGES / 1000;
2196 nr_mmu_pages = max(nr_mmu_pages,
2197 (unsigned int) KVM_MIN_ALLOC_MMU_PAGES);
2199 return nr_mmu_pages;
2202 static void *pv_mmu_peek_buffer(struct kvm_pv_mmu_op_buffer *buffer,
2205 if (len > buffer->len)
2210 static void *pv_mmu_read_buffer(struct kvm_pv_mmu_op_buffer *buffer,
2215 ret = pv_mmu_peek_buffer(buffer, len);
2220 buffer->processed += len;
2224 static int kvm_pv_mmu_write(struct kvm_vcpu *vcpu,
2225 gpa_t addr, gpa_t value)
2230 if (!is_long_mode(vcpu) && !is_pae(vcpu))
2233 r = mmu_topup_memory_caches(vcpu);
2237 if (!emulator_write_phys(vcpu, addr, &value, bytes))
2243 static int kvm_pv_mmu_flush_tlb(struct kvm_vcpu *vcpu)
2245 kvm_x86_ops->tlb_flush(vcpu);
2249 static int kvm_pv_mmu_release_pt(struct kvm_vcpu *vcpu, gpa_t addr)
2251 spin_lock(&vcpu->kvm->mmu_lock);
2252 mmu_unshadow(vcpu->kvm, addr >> PAGE_SHIFT);
2253 spin_unlock(&vcpu->kvm->mmu_lock);
2257 static int kvm_pv_mmu_op_one(struct kvm_vcpu *vcpu,
2258 struct kvm_pv_mmu_op_buffer *buffer)
2260 struct kvm_mmu_op_header *header;
2262 header = pv_mmu_peek_buffer(buffer, sizeof *header);
2265 switch (header->op) {
2266 case KVM_MMU_OP_WRITE_PTE: {
2267 struct kvm_mmu_op_write_pte *wpte;
2269 wpte = pv_mmu_read_buffer(buffer, sizeof *wpte);
2272 return kvm_pv_mmu_write(vcpu, wpte->pte_phys,
2275 case KVM_MMU_OP_FLUSH_TLB: {
2276 struct kvm_mmu_op_flush_tlb *ftlb;
2278 ftlb = pv_mmu_read_buffer(buffer, sizeof *ftlb);
2281 return kvm_pv_mmu_flush_tlb(vcpu);
2283 case KVM_MMU_OP_RELEASE_PT: {
2284 struct kvm_mmu_op_release_pt *rpt;
2286 rpt = pv_mmu_read_buffer(buffer, sizeof *rpt);
2289 return kvm_pv_mmu_release_pt(vcpu, rpt->pt_phys);
2295 int kvm_pv_mmu_op(struct kvm_vcpu *vcpu, unsigned long bytes,
2296 gpa_t addr, unsigned long *ret)
2299 struct kvm_pv_mmu_op_buffer buffer;
2301 buffer.ptr = buffer.buf;
2302 buffer.len = min_t(unsigned long, bytes, sizeof buffer.buf);
2303 buffer.processed = 0;
2305 r = kvm_read_guest(vcpu->kvm, addr, buffer.buf, buffer.len);
2309 while (buffer.len) {
2310 r = kvm_pv_mmu_op_one(vcpu, &buffer);
2319 *ret = buffer.processed;
2325 static const char *audit_msg;
2327 static gva_t canonicalize(gva_t gva)
2329 #ifdef CONFIG_X86_64
2330 gva = (long long)(gva << 16) >> 16;
2335 static void audit_mappings_page(struct kvm_vcpu *vcpu, u64 page_pte,
2336 gva_t va, int level)
2338 u64 *pt = __va(page_pte & PT64_BASE_ADDR_MASK);
2340 gva_t va_delta = 1ul << (PAGE_SHIFT + 9 * (level - 1));
2342 for (i = 0; i < PT64_ENT_PER_PAGE; ++i, va += va_delta) {
2345 if (ent == shadow_trap_nonpresent_pte)
2348 va = canonicalize(va);
2350 if (ent == shadow_notrap_nonpresent_pte)
2351 printk(KERN_ERR "audit: (%s) nontrapping pte"
2352 " in nonleaf level: levels %d gva %lx"
2353 " level %d pte %llx\n", audit_msg,
2354 vcpu->arch.mmu.root_level, va, level, ent);
2356 audit_mappings_page(vcpu, ent, va, level - 1);
2358 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, va);
2359 hpa_t hpa = (hpa_t)gpa_to_pfn(vcpu, gpa) << PAGE_SHIFT;
2361 if (is_shadow_present_pte(ent)
2362 && (ent & PT64_BASE_ADDR_MASK) != hpa)
2363 printk(KERN_ERR "xx audit error: (%s) levels %d"
2364 " gva %lx gpa %llx hpa %llx ent %llx %d\n",
2365 audit_msg, vcpu->arch.mmu.root_level,
2367 is_shadow_present_pte(ent));
2368 else if (ent == shadow_notrap_nonpresent_pte
2369 && !is_error_hpa(hpa))
2370 printk(KERN_ERR "audit: (%s) notrap shadow,"
2371 " valid guest gva %lx\n", audit_msg, va);
2372 kvm_release_pfn_clean(pfn);
2378 static void audit_mappings(struct kvm_vcpu *vcpu)
2382 if (vcpu->arch.mmu.root_level == 4)
2383 audit_mappings_page(vcpu, vcpu->arch.mmu.root_hpa, 0, 4);
2385 for (i = 0; i < 4; ++i)
2386 if (vcpu->arch.mmu.pae_root[i] & PT_PRESENT_MASK)
2387 audit_mappings_page(vcpu,
2388 vcpu->arch.mmu.pae_root[i],
2393 static int count_rmaps(struct kvm_vcpu *vcpu)
2398 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
2399 struct kvm_memory_slot *m = &vcpu->kvm->memslots[i];
2400 struct kvm_rmap_desc *d;
2402 for (j = 0; j < m->npages; ++j) {
2403 unsigned long *rmapp = &m->rmap[j];
2407 if (!(*rmapp & 1)) {
2411 d = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
2413 for (k = 0; k < RMAP_EXT; ++k)
2414 if (d->shadow_ptes[k])
2425 static int count_writable_mappings(struct kvm_vcpu *vcpu)
2428 struct kvm_mmu_page *sp;
2431 list_for_each_entry(sp, &vcpu->kvm->arch.active_mmu_pages, link) {
2434 if (sp->role.level != PT_PAGE_TABLE_LEVEL)
2437 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
2440 if (!(ent & PT_PRESENT_MASK))
2442 if (!(ent & PT_WRITABLE_MASK))
2450 static void audit_rmap(struct kvm_vcpu *vcpu)
2452 int n_rmap = count_rmaps(vcpu);
2453 int n_actual = count_writable_mappings(vcpu);
2455 if (n_rmap != n_actual)
2456 printk(KERN_ERR "%s: (%s) rmap %d actual %d\n",
2457 __func__, audit_msg, n_rmap, n_actual);
2460 static void audit_write_protection(struct kvm_vcpu *vcpu)
2462 struct kvm_mmu_page *sp;
2463 struct kvm_memory_slot *slot;
2464 unsigned long *rmapp;
2467 list_for_each_entry(sp, &vcpu->kvm->arch.active_mmu_pages, link) {
2468 if (sp->role.metaphysical)
2471 slot = gfn_to_memslot(vcpu->kvm, sp->gfn);
2472 gfn = unalias_gfn(vcpu->kvm, sp->gfn);
2473 rmapp = &slot->rmap[gfn - slot->base_gfn];
2475 printk(KERN_ERR "%s: (%s) shadow page has writable"
2476 " mappings: gfn %lx role %x\n",
2477 __func__, audit_msg, sp->gfn,
2482 static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg)
2489 audit_write_protection(vcpu);
2490 audit_mappings(vcpu);