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.
22 #include <linux/kvm_host.h>
23 #include <linux/types.h>
24 #include <linux/string.h>
26 #include <linux/highmem.h>
27 #include <linux/module.h>
28 #include <linux/swap.h>
29 #include <linux/hugetlb.h>
30 #include <linux/compiler.h>
33 #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);
73 static int oos_shadow = 1;
74 module_param(oos_shadow, bool, 0644);
77 #define ASSERT(x) do { } while (0)
81 printk(KERN_WARNING "assertion failed %s:%d: %s\n", \
82 __FILE__, __LINE__, #x); \
86 #define PT_FIRST_AVAIL_BITS_SHIFT 9
87 #define PT64_SECOND_AVAIL_BITS_SHIFT 52
89 #define VALID_PAGE(x) ((x) != INVALID_PAGE)
91 #define PT64_LEVEL_BITS 9
93 #define PT64_LEVEL_SHIFT(level) \
94 (PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS)
96 #define PT64_LEVEL_MASK(level) \
97 (((1ULL << PT64_LEVEL_BITS) - 1) << PT64_LEVEL_SHIFT(level))
99 #define PT64_INDEX(address, level)\
100 (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1))
103 #define PT32_LEVEL_BITS 10
105 #define PT32_LEVEL_SHIFT(level) \
106 (PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS)
108 #define PT32_LEVEL_MASK(level) \
109 (((1ULL << PT32_LEVEL_BITS) - 1) << PT32_LEVEL_SHIFT(level))
111 #define PT32_INDEX(address, level)\
112 (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1))
115 #define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1))
116 #define PT64_DIR_BASE_ADDR_MASK \
117 (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1))
119 #define PT32_BASE_ADDR_MASK PAGE_MASK
120 #define PT32_DIR_BASE_ADDR_MASK \
121 (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1))
123 #define PT64_PERM_MASK (PT_PRESENT_MASK | PT_WRITABLE_MASK | PT_USER_MASK \
126 #define PFERR_PRESENT_MASK (1U << 0)
127 #define PFERR_WRITE_MASK (1U << 1)
128 #define PFERR_USER_MASK (1U << 2)
129 #define PFERR_RSVD_MASK (1U << 3)
130 #define PFERR_FETCH_MASK (1U << 4)
132 #define PT_DIRECTORY_LEVEL 2
133 #define PT_PAGE_TABLE_LEVEL 1
137 #define ACC_EXEC_MASK 1
138 #define ACC_WRITE_MASK PT_WRITABLE_MASK
139 #define ACC_USER_MASK PT_USER_MASK
140 #define ACC_ALL (ACC_EXEC_MASK | ACC_WRITE_MASK | ACC_USER_MASK)
142 #define SHADOW_PT_INDEX(addr, level) PT64_INDEX(addr, level)
144 struct kvm_rmap_desc {
145 u64 *shadow_ptes[RMAP_EXT];
146 struct kvm_rmap_desc *more;
149 struct kvm_shadow_walk_iterator {
157 #define for_each_shadow_entry(_vcpu, _addr, _walker) \
158 for (shadow_walk_init(&(_walker), _vcpu, _addr); \
159 shadow_walk_okay(&(_walker)); \
160 shadow_walk_next(&(_walker)))
163 struct kvm_unsync_walk {
164 int (*entry) (struct kvm_mmu_page *sp, struct kvm_unsync_walk *walk);
167 typedef int (*mmu_parent_walk_fn) (struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp);
169 static struct kmem_cache *pte_chain_cache;
170 static struct kmem_cache *rmap_desc_cache;
171 static struct kmem_cache *mmu_page_header_cache;
173 static u64 __read_mostly shadow_trap_nonpresent_pte;
174 static u64 __read_mostly shadow_notrap_nonpresent_pte;
175 static u64 __read_mostly shadow_base_present_pte;
176 static u64 __read_mostly shadow_nx_mask;
177 static u64 __read_mostly shadow_x_mask; /* mutual exclusive with nx_mask */
178 static u64 __read_mostly shadow_user_mask;
179 static u64 __read_mostly shadow_accessed_mask;
180 static u64 __read_mostly shadow_dirty_mask;
182 static inline u64 rsvd_bits(int s, int e)
184 return ((1ULL << (e - s + 1)) - 1) << s;
187 void kvm_mmu_set_nonpresent_ptes(u64 trap_pte, u64 notrap_pte)
189 shadow_trap_nonpresent_pte = trap_pte;
190 shadow_notrap_nonpresent_pte = notrap_pte;
192 EXPORT_SYMBOL_GPL(kvm_mmu_set_nonpresent_ptes);
194 void kvm_mmu_set_base_ptes(u64 base_pte)
196 shadow_base_present_pte = base_pte;
198 EXPORT_SYMBOL_GPL(kvm_mmu_set_base_ptes);
200 void kvm_mmu_set_mask_ptes(u64 user_mask, u64 accessed_mask,
201 u64 dirty_mask, u64 nx_mask, u64 x_mask)
203 shadow_user_mask = user_mask;
204 shadow_accessed_mask = accessed_mask;
205 shadow_dirty_mask = dirty_mask;
206 shadow_nx_mask = nx_mask;
207 shadow_x_mask = x_mask;
209 EXPORT_SYMBOL_GPL(kvm_mmu_set_mask_ptes);
211 static int is_write_protection(struct kvm_vcpu *vcpu)
213 return vcpu->arch.cr0 & X86_CR0_WP;
216 static int is_cpuid_PSE36(void)
221 static int is_nx(struct kvm_vcpu *vcpu)
223 return vcpu->arch.shadow_efer & EFER_NX;
226 static int is_shadow_present_pte(u64 pte)
228 return pte != shadow_trap_nonpresent_pte
229 && pte != shadow_notrap_nonpresent_pte;
232 static int is_large_pte(u64 pte)
234 return pte & PT_PAGE_SIZE_MASK;
237 static int is_writeble_pte(unsigned long pte)
239 return pte & PT_WRITABLE_MASK;
242 static int is_dirty_pte(unsigned long pte)
244 return pte & shadow_dirty_mask;
247 static int is_rmap_pte(u64 pte)
249 return is_shadow_present_pte(pte);
252 static pfn_t spte_to_pfn(u64 pte)
254 return (pte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT;
257 static gfn_t pse36_gfn_delta(u32 gpte)
259 int shift = 32 - PT32_DIR_PSE36_SHIFT - PAGE_SHIFT;
261 return (gpte & PT32_DIR_PSE36_MASK) << shift;
264 static void set_shadow_pte(u64 *sptep, u64 spte)
267 set_64bit((unsigned long *)sptep, spte);
269 set_64bit((unsigned long long *)sptep, spte);
273 static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache,
274 struct kmem_cache *base_cache, int min)
278 if (cache->nobjs >= min)
280 while (cache->nobjs < ARRAY_SIZE(cache->objects)) {
281 obj = kmem_cache_zalloc(base_cache, GFP_KERNEL);
284 cache->objects[cache->nobjs++] = obj;
289 static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
292 kfree(mc->objects[--mc->nobjs]);
295 static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache *cache,
300 if (cache->nobjs >= min)
302 while (cache->nobjs < ARRAY_SIZE(cache->objects)) {
303 page = alloc_page(GFP_KERNEL);
306 set_page_private(page, 0);
307 cache->objects[cache->nobjs++] = page_address(page);
312 static void mmu_free_memory_cache_page(struct kvm_mmu_memory_cache *mc)
315 free_page((unsigned long)mc->objects[--mc->nobjs]);
318 static int mmu_topup_memory_caches(struct kvm_vcpu *vcpu)
322 r = mmu_topup_memory_cache(&vcpu->arch.mmu_pte_chain_cache,
326 r = mmu_topup_memory_cache(&vcpu->arch.mmu_rmap_desc_cache,
330 r = mmu_topup_memory_cache_page(&vcpu->arch.mmu_page_cache, 8);
333 r = mmu_topup_memory_cache(&vcpu->arch.mmu_page_header_cache,
334 mmu_page_header_cache, 4);
339 static void mmu_free_memory_caches(struct kvm_vcpu *vcpu)
341 mmu_free_memory_cache(&vcpu->arch.mmu_pte_chain_cache);
342 mmu_free_memory_cache(&vcpu->arch.mmu_rmap_desc_cache);
343 mmu_free_memory_cache_page(&vcpu->arch.mmu_page_cache);
344 mmu_free_memory_cache(&vcpu->arch.mmu_page_header_cache);
347 static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc,
353 p = mc->objects[--mc->nobjs];
357 static struct kvm_pte_chain *mmu_alloc_pte_chain(struct kvm_vcpu *vcpu)
359 return mmu_memory_cache_alloc(&vcpu->arch.mmu_pte_chain_cache,
360 sizeof(struct kvm_pte_chain));
363 static void mmu_free_pte_chain(struct kvm_pte_chain *pc)
368 static struct kvm_rmap_desc *mmu_alloc_rmap_desc(struct kvm_vcpu *vcpu)
370 return mmu_memory_cache_alloc(&vcpu->arch.mmu_rmap_desc_cache,
371 sizeof(struct kvm_rmap_desc));
374 static void mmu_free_rmap_desc(struct kvm_rmap_desc *rd)
380 * Return the pointer to the largepage write count for a given
381 * gfn, handling slots that are not large page aligned.
383 static int *slot_largepage_idx(gfn_t gfn, struct kvm_memory_slot *slot)
387 idx = (gfn / KVM_PAGES_PER_HPAGE) -
388 (slot->base_gfn / KVM_PAGES_PER_HPAGE);
389 return &slot->lpage_info[idx].write_count;
392 static void account_shadowed(struct kvm *kvm, gfn_t gfn)
396 gfn = unalias_gfn(kvm, gfn);
397 write_count = slot_largepage_idx(gfn,
398 gfn_to_memslot_unaliased(kvm, gfn));
402 static void unaccount_shadowed(struct kvm *kvm, gfn_t gfn)
406 gfn = unalias_gfn(kvm, gfn);
407 write_count = slot_largepage_idx(gfn,
408 gfn_to_memslot_unaliased(kvm, gfn));
410 WARN_ON(*write_count < 0);
413 static int has_wrprotected_page(struct kvm *kvm, gfn_t gfn)
415 struct kvm_memory_slot *slot;
418 gfn = unalias_gfn(kvm, gfn);
419 slot = gfn_to_memslot_unaliased(kvm, gfn);
421 largepage_idx = slot_largepage_idx(gfn, slot);
422 return *largepage_idx;
428 static int host_largepage_backed(struct kvm *kvm, gfn_t gfn)
430 struct vm_area_struct *vma;
434 addr = gfn_to_hva(kvm, gfn);
435 if (kvm_is_error_hva(addr))
438 down_read(¤t->mm->mmap_sem);
439 vma = find_vma(current->mm, addr);
440 if (vma && is_vm_hugetlb_page(vma))
442 up_read(¤t->mm->mmap_sem);
447 static int is_largepage_backed(struct kvm_vcpu *vcpu, gfn_t large_gfn)
449 struct kvm_memory_slot *slot;
451 if (has_wrprotected_page(vcpu->kvm, large_gfn))
454 if (!host_largepage_backed(vcpu->kvm, large_gfn))
457 slot = gfn_to_memslot(vcpu->kvm, large_gfn);
458 if (slot && slot->dirty_bitmap)
465 * Take gfn and return the reverse mapping to it.
466 * Note: gfn must be unaliased before this function get called
469 static unsigned long *gfn_to_rmap(struct kvm *kvm, gfn_t gfn, int lpage)
471 struct kvm_memory_slot *slot;
474 slot = gfn_to_memslot(kvm, gfn);
476 return &slot->rmap[gfn - slot->base_gfn];
478 idx = (gfn / KVM_PAGES_PER_HPAGE) -
479 (slot->base_gfn / KVM_PAGES_PER_HPAGE);
481 return &slot->lpage_info[idx].rmap_pde;
485 * Reverse mapping data structures:
487 * If rmapp bit zero is zero, then rmapp point to the shadw page table entry
488 * that points to page_address(page).
490 * If rmapp bit zero is one, (then rmap & ~1) points to a struct kvm_rmap_desc
491 * containing more mappings.
493 * Returns the number of rmap entries before the spte was added or zero if
494 * the spte was not added.
497 static int rmap_add(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn, int lpage)
499 struct kvm_mmu_page *sp;
500 struct kvm_rmap_desc *desc;
501 unsigned long *rmapp;
504 if (!is_rmap_pte(*spte))
506 gfn = unalias_gfn(vcpu->kvm, gfn);
507 sp = page_header(__pa(spte));
508 sp->gfns[spte - sp->spt] = gfn;
509 rmapp = gfn_to_rmap(vcpu->kvm, gfn, lpage);
511 rmap_printk("rmap_add: %p %llx 0->1\n", spte, *spte);
512 *rmapp = (unsigned long)spte;
513 } else if (!(*rmapp & 1)) {
514 rmap_printk("rmap_add: %p %llx 1->many\n", spte, *spte);
515 desc = mmu_alloc_rmap_desc(vcpu);
516 desc->shadow_ptes[0] = (u64 *)*rmapp;
517 desc->shadow_ptes[1] = spte;
518 *rmapp = (unsigned long)desc | 1;
520 rmap_printk("rmap_add: %p %llx many->many\n", spte, *spte);
521 desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
522 while (desc->shadow_ptes[RMAP_EXT-1] && desc->more) {
526 if (desc->shadow_ptes[RMAP_EXT-1]) {
527 desc->more = mmu_alloc_rmap_desc(vcpu);
530 for (i = 0; desc->shadow_ptes[i]; ++i)
532 desc->shadow_ptes[i] = spte;
537 static void rmap_desc_remove_entry(unsigned long *rmapp,
538 struct kvm_rmap_desc *desc,
540 struct kvm_rmap_desc *prev_desc)
544 for (j = RMAP_EXT - 1; !desc->shadow_ptes[j] && j > i; --j)
546 desc->shadow_ptes[i] = desc->shadow_ptes[j];
547 desc->shadow_ptes[j] = NULL;
550 if (!prev_desc && !desc->more)
551 *rmapp = (unsigned long)desc->shadow_ptes[0];
554 prev_desc->more = desc->more;
556 *rmapp = (unsigned long)desc->more | 1;
557 mmu_free_rmap_desc(desc);
560 static void rmap_remove(struct kvm *kvm, u64 *spte)
562 struct kvm_rmap_desc *desc;
563 struct kvm_rmap_desc *prev_desc;
564 struct kvm_mmu_page *sp;
566 unsigned long *rmapp;
569 if (!is_rmap_pte(*spte))
571 sp = page_header(__pa(spte));
572 pfn = spte_to_pfn(*spte);
573 if (*spte & shadow_accessed_mask)
574 kvm_set_pfn_accessed(pfn);
575 if (is_writeble_pte(*spte))
576 kvm_release_pfn_dirty(pfn);
578 kvm_release_pfn_clean(pfn);
579 rmapp = gfn_to_rmap(kvm, sp->gfns[spte - sp->spt], is_large_pte(*spte));
581 printk(KERN_ERR "rmap_remove: %p %llx 0->BUG\n", spte, *spte);
583 } else if (!(*rmapp & 1)) {
584 rmap_printk("rmap_remove: %p %llx 1->0\n", spte, *spte);
585 if ((u64 *)*rmapp != spte) {
586 printk(KERN_ERR "rmap_remove: %p %llx 1->BUG\n",
592 rmap_printk("rmap_remove: %p %llx many->many\n", spte, *spte);
593 desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
596 for (i = 0; i < RMAP_EXT && desc->shadow_ptes[i]; ++i)
597 if (desc->shadow_ptes[i] == spte) {
598 rmap_desc_remove_entry(rmapp,
610 static u64 *rmap_next(struct kvm *kvm, unsigned long *rmapp, u64 *spte)
612 struct kvm_rmap_desc *desc;
613 struct kvm_rmap_desc *prev_desc;
619 else if (!(*rmapp & 1)) {
621 return (u64 *)*rmapp;
624 desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
628 for (i = 0; i < RMAP_EXT && desc->shadow_ptes[i]; ++i) {
629 if (prev_spte == spte)
630 return desc->shadow_ptes[i];
631 prev_spte = desc->shadow_ptes[i];
638 static int rmap_write_protect(struct kvm *kvm, u64 gfn)
640 unsigned long *rmapp;
642 int write_protected = 0;
644 gfn = unalias_gfn(kvm, gfn);
645 rmapp = gfn_to_rmap(kvm, gfn, 0);
647 spte = rmap_next(kvm, rmapp, NULL);
650 BUG_ON(!(*spte & PT_PRESENT_MASK));
651 rmap_printk("rmap_write_protect: spte %p %llx\n", spte, *spte);
652 if (is_writeble_pte(*spte)) {
653 set_shadow_pte(spte, *spte & ~PT_WRITABLE_MASK);
656 spte = rmap_next(kvm, rmapp, spte);
658 if (write_protected) {
661 spte = rmap_next(kvm, rmapp, NULL);
662 pfn = spte_to_pfn(*spte);
663 kvm_set_pfn_dirty(pfn);
666 /* check for huge page mappings */
667 rmapp = gfn_to_rmap(kvm, gfn, 1);
668 spte = rmap_next(kvm, rmapp, NULL);
671 BUG_ON(!(*spte & PT_PRESENT_MASK));
672 BUG_ON((*spte & (PT_PAGE_SIZE_MASK|PT_PRESENT_MASK)) != (PT_PAGE_SIZE_MASK|PT_PRESENT_MASK));
673 pgprintk("rmap_write_protect(large): spte %p %llx %lld\n", spte, *spte, gfn);
674 if (is_writeble_pte(*spte)) {
675 rmap_remove(kvm, spte);
677 set_shadow_pte(spte, shadow_trap_nonpresent_pte);
681 spte = rmap_next(kvm, rmapp, spte);
684 return write_protected;
687 static int kvm_unmap_rmapp(struct kvm *kvm, unsigned long *rmapp)
690 int need_tlb_flush = 0;
692 while ((spte = rmap_next(kvm, rmapp, NULL))) {
693 BUG_ON(!(*spte & PT_PRESENT_MASK));
694 rmap_printk("kvm_rmap_unmap_hva: spte %p %llx\n", spte, *spte);
695 rmap_remove(kvm, spte);
696 set_shadow_pte(spte, shadow_trap_nonpresent_pte);
699 return need_tlb_flush;
702 static int kvm_handle_hva(struct kvm *kvm, unsigned long hva,
703 int (*handler)(struct kvm *kvm, unsigned long *rmapp))
709 * If mmap_sem isn't taken, we can look the memslots with only
710 * the mmu_lock by skipping over the slots with userspace_addr == 0.
712 for (i = 0; i < kvm->nmemslots; i++) {
713 struct kvm_memory_slot *memslot = &kvm->memslots[i];
714 unsigned long start = memslot->userspace_addr;
717 /* mmu_lock protects userspace_addr */
721 end = start + (memslot->npages << PAGE_SHIFT);
722 if (hva >= start && hva < end) {
723 gfn_t gfn_offset = (hva - start) >> PAGE_SHIFT;
724 retval |= handler(kvm, &memslot->rmap[gfn_offset]);
725 retval |= handler(kvm,
726 &memslot->lpage_info[
728 KVM_PAGES_PER_HPAGE].rmap_pde);
735 int kvm_unmap_hva(struct kvm *kvm, unsigned long hva)
737 return kvm_handle_hva(kvm, hva, kvm_unmap_rmapp);
740 static int kvm_age_rmapp(struct kvm *kvm, unsigned long *rmapp)
745 /* always return old for EPT */
746 if (!shadow_accessed_mask)
749 spte = rmap_next(kvm, rmapp, NULL);
753 BUG_ON(!(_spte & PT_PRESENT_MASK));
754 _young = _spte & PT_ACCESSED_MASK;
757 clear_bit(PT_ACCESSED_SHIFT, (unsigned long *)spte);
759 spte = rmap_next(kvm, rmapp, spte);
764 #define RMAP_RECYCLE_THRESHOLD 1000
766 static void rmap_recycle(struct kvm_vcpu *vcpu, gfn_t gfn, int lpage)
768 unsigned long *rmapp;
770 gfn = unalias_gfn(vcpu->kvm, gfn);
771 rmapp = gfn_to_rmap(vcpu->kvm, gfn, lpage);
773 kvm_unmap_rmapp(vcpu->kvm, rmapp);
774 kvm_flush_remote_tlbs(vcpu->kvm);
777 int kvm_age_hva(struct kvm *kvm, unsigned long hva)
779 return kvm_handle_hva(kvm, hva, kvm_age_rmapp);
783 static int is_empty_shadow_page(u64 *spt)
788 for (pos = spt, end = pos + PAGE_SIZE / sizeof(u64); pos != end; pos++)
789 if (is_shadow_present_pte(*pos)) {
790 printk(KERN_ERR "%s: %p %llx\n", __func__,
798 static void kvm_mmu_free_page(struct kvm *kvm, struct kvm_mmu_page *sp)
800 ASSERT(is_empty_shadow_page(sp->spt));
802 __free_page(virt_to_page(sp->spt));
803 __free_page(virt_to_page(sp->gfns));
805 ++kvm->arch.n_free_mmu_pages;
808 static unsigned kvm_page_table_hashfn(gfn_t gfn)
810 return gfn & ((1 << KVM_MMU_HASH_SHIFT) - 1);
813 static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu,
816 struct kvm_mmu_page *sp;
818 sp = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_header_cache, sizeof *sp);
819 sp->spt = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_cache, PAGE_SIZE);
820 sp->gfns = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_cache, PAGE_SIZE);
821 set_page_private(virt_to_page(sp->spt), (unsigned long)sp);
822 list_add(&sp->link, &vcpu->kvm->arch.active_mmu_pages);
823 INIT_LIST_HEAD(&sp->oos_link);
824 bitmap_zero(sp->slot_bitmap, KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS);
826 sp->parent_pte = parent_pte;
827 --vcpu->kvm->arch.n_free_mmu_pages;
831 static void mmu_page_add_parent_pte(struct kvm_vcpu *vcpu,
832 struct kvm_mmu_page *sp, u64 *parent_pte)
834 struct kvm_pte_chain *pte_chain;
835 struct hlist_node *node;
840 if (!sp->multimapped) {
841 u64 *old = sp->parent_pte;
844 sp->parent_pte = parent_pte;
848 pte_chain = mmu_alloc_pte_chain(vcpu);
849 INIT_HLIST_HEAD(&sp->parent_ptes);
850 hlist_add_head(&pte_chain->link, &sp->parent_ptes);
851 pte_chain->parent_ptes[0] = old;
853 hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link) {
854 if (pte_chain->parent_ptes[NR_PTE_CHAIN_ENTRIES-1])
856 for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i)
857 if (!pte_chain->parent_ptes[i]) {
858 pte_chain->parent_ptes[i] = parent_pte;
862 pte_chain = mmu_alloc_pte_chain(vcpu);
864 hlist_add_head(&pte_chain->link, &sp->parent_ptes);
865 pte_chain->parent_ptes[0] = parent_pte;
868 static void mmu_page_remove_parent_pte(struct kvm_mmu_page *sp,
871 struct kvm_pte_chain *pte_chain;
872 struct hlist_node *node;
875 if (!sp->multimapped) {
876 BUG_ON(sp->parent_pte != parent_pte);
877 sp->parent_pte = NULL;
880 hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link)
881 for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) {
882 if (!pte_chain->parent_ptes[i])
884 if (pte_chain->parent_ptes[i] != parent_pte)
886 while (i + 1 < NR_PTE_CHAIN_ENTRIES
887 && pte_chain->parent_ptes[i + 1]) {
888 pte_chain->parent_ptes[i]
889 = pte_chain->parent_ptes[i + 1];
892 pte_chain->parent_ptes[i] = NULL;
894 hlist_del(&pte_chain->link);
895 mmu_free_pte_chain(pte_chain);
896 if (hlist_empty(&sp->parent_ptes)) {
898 sp->parent_pte = NULL;
907 static void mmu_parent_walk(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
908 mmu_parent_walk_fn fn)
910 struct kvm_pte_chain *pte_chain;
911 struct hlist_node *node;
912 struct kvm_mmu_page *parent_sp;
915 if (!sp->multimapped && sp->parent_pte) {
916 parent_sp = page_header(__pa(sp->parent_pte));
918 mmu_parent_walk(vcpu, parent_sp, fn);
921 hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link)
922 for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) {
923 if (!pte_chain->parent_ptes[i])
925 parent_sp = page_header(__pa(pte_chain->parent_ptes[i]));
927 mmu_parent_walk(vcpu, parent_sp, fn);
931 static void kvm_mmu_update_unsync_bitmap(u64 *spte)
934 struct kvm_mmu_page *sp = page_header(__pa(spte));
936 index = spte - sp->spt;
937 if (!__test_and_set_bit(index, sp->unsync_child_bitmap))
938 sp->unsync_children++;
939 WARN_ON(!sp->unsync_children);
942 static void kvm_mmu_update_parents_unsync(struct kvm_mmu_page *sp)
944 struct kvm_pte_chain *pte_chain;
945 struct hlist_node *node;
951 if (!sp->multimapped) {
952 kvm_mmu_update_unsync_bitmap(sp->parent_pte);
956 hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link)
957 for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) {
958 if (!pte_chain->parent_ptes[i])
960 kvm_mmu_update_unsync_bitmap(pte_chain->parent_ptes[i]);
964 static int unsync_walk_fn(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
966 kvm_mmu_update_parents_unsync(sp);
970 static void kvm_mmu_mark_parents_unsync(struct kvm_vcpu *vcpu,
971 struct kvm_mmu_page *sp)
973 mmu_parent_walk(vcpu, sp, unsync_walk_fn);
974 kvm_mmu_update_parents_unsync(sp);
977 static void nonpaging_prefetch_page(struct kvm_vcpu *vcpu,
978 struct kvm_mmu_page *sp)
982 for (i = 0; i < PT64_ENT_PER_PAGE; ++i)
983 sp->spt[i] = shadow_trap_nonpresent_pte;
986 static int nonpaging_sync_page(struct kvm_vcpu *vcpu,
987 struct kvm_mmu_page *sp)
992 static void nonpaging_invlpg(struct kvm_vcpu *vcpu, gva_t gva)
996 #define KVM_PAGE_ARRAY_NR 16
998 struct kvm_mmu_pages {
999 struct mmu_page_and_offset {
1000 struct kvm_mmu_page *sp;
1002 } page[KVM_PAGE_ARRAY_NR];
1006 #define for_each_unsync_children(bitmap, idx) \
1007 for (idx = find_first_bit(bitmap, 512); \
1009 idx = find_next_bit(bitmap, 512, idx+1))
1011 static int mmu_pages_add(struct kvm_mmu_pages *pvec, struct kvm_mmu_page *sp,
1017 for (i=0; i < pvec->nr; i++)
1018 if (pvec->page[i].sp == sp)
1021 pvec->page[pvec->nr].sp = sp;
1022 pvec->page[pvec->nr].idx = idx;
1024 return (pvec->nr == KVM_PAGE_ARRAY_NR);
1027 static int __mmu_unsync_walk(struct kvm_mmu_page *sp,
1028 struct kvm_mmu_pages *pvec)
1030 int i, ret, nr_unsync_leaf = 0;
1032 for_each_unsync_children(sp->unsync_child_bitmap, i) {
1033 u64 ent = sp->spt[i];
1035 if (is_shadow_present_pte(ent) && !is_large_pte(ent)) {
1036 struct kvm_mmu_page *child;
1037 child = page_header(ent & PT64_BASE_ADDR_MASK);
1039 if (child->unsync_children) {
1040 if (mmu_pages_add(pvec, child, i))
1043 ret = __mmu_unsync_walk(child, pvec);
1045 __clear_bit(i, sp->unsync_child_bitmap);
1047 nr_unsync_leaf += ret;
1052 if (child->unsync) {
1054 if (mmu_pages_add(pvec, child, i))
1060 if (find_first_bit(sp->unsync_child_bitmap, 512) == 512)
1061 sp->unsync_children = 0;
1063 return nr_unsync_leaf;
1066 static int mmu_unsync_walk(struct kvm_mmu_page *sp,
1067 struct kvm_mmu_pages *pvec)
1069 if (!sp->unsync_children)
1072 mmu_pages_add(pvec, sp, 0);
1073 return __mmu_unsync_walk(sp, pvec);
1076 static struct kvm_mmu_page *kvm_mmu_lookup_page(struct kvm *kvm, gfn_t gfn)
1079 struct hlist_head *bucket;
1080 struct kvm_mmu_page *sp;
1081 struct hlist_node *node;
1083 pgprintk("%s: looking for gfn %lx\n", __func__, gfn);
1084 index = kvm_page_table_hashfn(gfn);
1085 bucket = &kvm->arch.mmu_page_hash[index];
1086 hlist_for_each_entry(sp, node, bucket, hash_link)
1087 if (sp->gfn == gfn && !sp->role.direct
1088 && !sp->role.invalid) {
1089 pgprintk("%s: found role %x\n",
1090 __func__, sp->role.word);
1096 static void kvm_unlink_unsync_page(struct kvm *kvm, struct kvm_mmu_page *sp)
1098 WARN_ON(!sp->unsync);
1100 --kvm->stat.mmu_unsync;
1103 static int kvm_mmu_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp);
1105 static int kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
1107 if (sp->role.glevels != vcpu->arch.mmu.root_level) {
1108 kvm_mmu_zap_page(vcpu->kvm, sp);
1112 if (rmap_write_protect(vcpu->kvm, sp->gfn))
1113 kvm_flush_remote_tlbs(vcpu->kvm);
1114 kvm_unlink_unsync_page(vcpu->kvm, sp);
1115 if (vcpu->arch.mmu.sync_page(vcpu, sp)) {
1116 kvm_mmu_zap_page(vcpu->kvm, sp);
1120 kvm_mmu_flush_tlb(vcpu);
1124 struct mmu_page_path {
1125 struct kvm_mmu_page *parent[PT64_ROOT_LEVEL-1];
1126 unsigned int idx[PT64_ROOT_LEVEL-1];
1129 #define for_each_sp(pvec, sp, parents, i) \
1130 for (i = mmu_pages_next(&pvec, &parents, -1), \
1131 sp = pvec.page[i].sp; \
1132 i < pvec.nr && ({ sp = pvec.page[i].sp; 1;}); \
1133 i = mmu_pages_next(&pvec, &parents, i))
1135 static int mmu_pages_next(struct kvm_mmu_pages *pvec,
1136 struct mmu_page_path *parents,
1141 for (n = i+1; n < pvec->nr; n++) {
1142 struct kvm_mmu_page *sp = pvec->page[n].sp;
1144 if (sp->role.level == PT_PAGE_TABLE_LEVEL) {
1145 parents->idx[0] = pvec->page[n].idx;
1149 parents->parent[sp->role.level-2] = sp;
1150 parents->idx[sp->role.level-1] = pvec->page[n].idx;
1156 static void mmu_pages_clear_parents(struct mmu_page_path *parents)
1158 struct kvm_mmu_page *sp;
1159 unsigned int level = 0;
1162 unsigned int idx = parents->idx[level];
1164 sp = parents->parent[level];
1168 --sp->unsync_children;
1169 WARN_ON((int)sp->unsync_children < 0);
1170 __clear_bit(idx, sp->unsync_child_bitmap);
1172 } while (level < PT64_ROOT_LEVEL-1 && !sp->unsync_children);
1175 static void kvm_mmu_pages_init(struct kvm_mmu_page *parent,
1176 struct mmu_page_path *parents,
1177 struct kvm_mmu_pages *pvec)
1179 parents->parent[parent->role.level-1] = NULL;
1183 static void mmu_sync_children(struct kvm_vcpu *vcpu,
1184 struct kvm_mmu_page *parent)
1187 struct kvm_mmu_page *sp;
1188 struct mmu_page_path parents;
1189 struct kvm_mmu_pages pages;
1191 kvm_mmu_pages_init(parent, &parents, &pages);
1192 while (mmu_unsync_walk(parent, &pages)) {
1195 for_each_sp(pages, sp, parents, i)
1196 protected |= rmap_write_protect(vcpu->kvm, sp->gfn);
1199 kvm_flush_remote_tlbs(vcpu->kvm);
1201 for_each_sp(pages, sp, parents, i) {
1202 kvm_sync_page(vcpu, sp);
1203 mmu_pages_clear_parents(&parents);
1205 cond_resched_lock(&vcpu->kvm->mmu_lock);
1206 kvm_mmu_pages_init(parent, &parents, &pages);
1210 static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
1218 union kvm_mmu_page_role role;
1221 struct hlist_head *bucket;
1222 struct kvm_mmu_page *sp;
1223 struct hlist_node *node, *tmp;
1225 role = vcpu->arch.mmu.base_role;
1227 role.direct = direct;
1228 role.access = access;
1229 if (vcpu->arch.mmu.root_level <= PT32_ROOT_LEVEL) {
1230 quadrant = gaddr >> (PAGE_SHIFT + (PT64_PT_BITS * level));
1231 quadrant &= (1 << ((PT32_PT_BITS - PT64_PT_BITS) * level)) - 1;
1232 role.quadrant = quadrant;
1234 pgprintk("%s: looking gfn %lx role %x\n", __func__,
1236 index = kvm_page_table_hashfn(gfn);
1237 bucket = &vcpu->kvm->arch.mmu_page_hash[index];
1238 hlist_for_each_entry_safe(sp, node, tmp, bucket, hash_link)
1239 if (sp->gfn == gfn) {
1241 if (kvm_sync_page(vcpu, sp))
1244 if (sp->role.word != role.word)
1247 mmu_page_add_parent_pte(vcpu, sp, parent_pte);
1248 if (sp->unsync_children) {
1249 set_bit(KVM_REQ_MMU_SYNC, &vcpu->requests);
1250 kvm_mmu_mark_parents_unsync(vcpu, sp);
1252 pgprintk("%s: found\n", __func__);
1255 ++vcpu->kvm->stat.mmu_cache_miss;
1256 sp = kvm_mmu_alloc_page(vcpu, parent_pte);
1259 pgprintk("%s: adding gfn %lx role %x\n", __func__, gfn, role.word);
1262 hlist_add_head(&sp->hash_link, bucket);
1264 if (rmap_write_protect(vcpu->kvm, gfn))
1265 kvm_flush_remote_tlbs(vcpu->kvm);
1266 account_shadowed(vcpu->kvm, gfn);
1268 if (shadow_trap_nonpresent_pte != shadow_notrap_nonpresent_pte)
1269 vcpu->arch.mmu.prefetch_page(vcpu, sp);
1271 nonpaging_prefetch_page(vcpu, sp);
1275 static void shadow_walk_init(struct kvm_shadow_walk_iterator *iterator,
1276 struct kvm_vcpu *vcpu, u64 addr)
1278 iterator->addr = addr;
1279 iterator->shadow_addr = vcpu->arch.mmu.root_hpa;
1280 iterator->level = vcpu->arch.mmu.shadow_root_level;
1281 if (iterator->level == PT32E_ROOT_LEVEL) {
1282 iterator->shadow_addr
1283 = vcpu->arch.mmu.pae_root[(addr >> 30) & 3];
1284 iterator->shadow_addr &= PT64_BASE_ADDR_MASK;
1286 if (!iterator->shadow_addr)
1287 iterator->level = 0;
1291 static bool shadow_walk_okay(struct kvm_shadow_walk_iterator *iterator)
1293 if (iterator->level < PT_PAGE_TABLE_LEVEL)
1295 iterator->index = SHADOW_PT_INDEX(iterator->addr, iterator->level);
1296 iterator->sptep = ((u64 *)__va(iterator->shadow_addr)) + iterator->index;
1300 static void shadow_walk_next(struct kvm_shadow_walk_iterator *iterator)
1302 iterator->shadow_addr = *iterator->sptep & PT64_BASE_ADDR_MASK;
1306 static void kvm_mmu_page_unlink_children(struct kvm *kvm,
1307 struct kvm_mmu_page *sp)
1315 if (sp->role.level == PT_PAGE_TABLE_LEVEL) {
1316 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
1317 if (is_shadow_present_pte(pt[i]))
1318 rmap_remove(kvm, &pt[i]);
1319 pt[i] = shadow_trap_nonpresent_pte;
1324 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
1327 if (is_shadow_present_pte(ent)) {
1328 if (!is_large_pte(ent)) {
1329 ent &= PT64_BASE_ADDR_MASK;
1330 mmu_page_remove_parent_pte(page_header(ent),
1334 rmap_remove(kvm, &pt[i]);
1337 pt[i] = shadow_trap_nonpresent_pte;
1341 static void kvm_mmu_put_page(struct kvm_mmu_page *sp, u64 *parent_pte)
1343 mmu_page_remove_parent_pte(sp, parent_pte);
1346 static void kvm_mmu_reset_last_pte_updated(struct kvm *kvm)
1350 for (i = 0; i < KVM_MAX_VCPUS; ++i)
1352 kvm->vcpus[i]->arch.last_pte_updated = NULL;
1355 static void kvm_mmu_unlink_parents(struct kvm *kvm, struct kvm_mmu_page *sp)
1359 while (sp->multimapped || sp->parent_pte) {
1360 if (!sp->multimapped)
1361 parent_pte = sp->parent_pte;
1363 struct kvm_pte_chain *chain;
1365 chain = container_of(sp->parent_ptes.first,
1366 struct kvm_pte_chain, link);
1367 parent_pte = chain->parent_ptes[0];
1369 BUG_ON(!parent_pte);
1370 kvm_mmu_put_page(sp, parent_pte);
1371 set_shadow_pte(parent_pte, shadow_trap_nonpresent_pte);
1375 static int mmu_zap_unsync_children(struct kvm *kvm,
1376 struct kvm_mmu_page *parent)
1379 struct mmu_page_path parents;
1380 struct kvm_mmu_pages pages;
1382 if (parent->role.level == PT_PAGE_TABLE_LEVEL)
1385 kvm_mmu_pages_init(parent, &parents, &pages);
1386 while (mmu_unsync_walk(parent, &pages)) {
1387 struct kvm_mmu_page *sp;
1389 for_each_sp(pages, sp, parents, i) {
1390 kvm_mmu_zap_page(kvm, sp);
1391 mmu_pages_clear_parents(&parents);
1394 kvm_mmu_pages_init(parent, &parents, &pages);
1400 static int kvm_mmu_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp)
1403 ++kvm->stat.mmu_shadow_zapped;
1404 ret = mmu_zap_unsync_children(kvm, sp);
1405 kvm_mmu_page_unlink_children(kvm, sp);
1406 kvm_mmu_unlink_parents(kvm, sp);
1407 kvm_flush_remote_tlbs(kvm);
1408 if (!sp->role.invalid && !sp->role.direct)
1409 unaccount_shadowed(kvm, sp->gfn);
1411 kvm_unlink_unsync_page(kvm, sp);
1412 if (!sp->root_count) {
1413 hlist_del(&sp->hash_link);
1414 kvm_mmu_free_page(kvm, sp);
1416 sp->role.invalid = 1;
1417 list_move(&sp->link, &kvm->arch.active_mmu_pages);
1418 kvm_reload_remote_mmus(kvm);
1420 kvm_mmu_reset_last_pte_updated(kvm);
1425 * Changing the number of mmu pages allocated to the vm
1426 * Note: if kvm_nr_mmu_pages is too small, you will get dead lock
1428 void kvm_mmu_change_mmu_pages(struct kvm *kvm, unsigned int kvm_nr_mmu_pages)
1432 used_pages = kvm->arch.n_alloc_mmu_pages - kvm->arch.n_free_mmu_pages;
1433 used_pages = max(0, used_pages);
1436 * If we set the number of mmu pages to be smaller be than the
1437 * number of actived pages , we must to free some mmu pages before we
1441 if (used_pages > kvm_nr_mmu_pages) {
1442 while (used_pages > kvm_nr_mmu_pages) {
1443 struct kvm_mmu_page *page;
1445 page = container_of(kvm->arch.active_mmu_pages.prev,
1446 struct kvm_mmu_page, link);
1447 kvm_mmu_zap_page(kvm, page);
1450 kvm->arch.n_free_mmu_pages = 0;
1453 kvm->arch.n_free_mmu_pages += kvm_nr_mmu_pages
1454 - kvm->arch.n_alloc_mmu_pages;
1456 kvm->arch.n_alloc_mmu_pages = kvm_nr_mmu_pages;
1459 static int kvm_mmu_unprotect_page(struct kvm *kvm, gfn_t gfn)
1462 struct hlist_head *bucket;
1463 struct kvm_mmu_page *sp;
1464 struct hlist_node *node, *n;
1467 pgprintk("%s: looking for gfn %lx\n", __func__, gfn);
1469 index = kvm_page_table_hashfn(gfn);
1470 bucket = &kvm->arch.mmu_page_hash[index];
1471 hlist_for_each_entry_safe(sp, node, n, bucket, hash_link)
1472 if (sp->gfn == gfn && !sp->role.direct) {
1473 pgprintk("%s: gfn %lx role %x\n", __func__, gfn,
1476 if (kvm_mmu_zap_page(kvm, sp))
1482 static void mmu_unshadow(struct kvm *kvm, gfn_t gfn)
1485 struct hlist_head *bucket;
1486 struct kvm_mmu_page *sp;
1487 struct hlist_node *node, *nn;
1489 index = kvm_page_table_hashfn(gfn);
1490 bucket = &kvm->arch.mmu_page_hash[index];
1491 hlist_for_each_entry_safe(sp, node, nn, bucket, hash_link) {
1492 if (sp->gfn == gfn && !sp->role.direct
1493 && !sp->role.invalid) {
1494 pgprintk("%s: zap %lx %x\n",
1495 __func__, gfn, sp->role.word);
1496 kvm_mmu_zap_page(kvm, sp);
1501 static void page_header_update_slot(struct kvm *kvm, void *pte, gfn_t gfn)
1503 int slot = memslot_id(kvm, gfn_to_memslot(kvm, gfn));
1504 struct kvm_mmu_page *sp = page_header(__pa(pte));
1506 __set_bit(slot, sp->slot_bitmap);
1509 static void mmu_convert_notrap(struct kvm_mmu_page *sp)
1514 if (shadow_trap_nonpresent_pte == shadow_notrap_nonpresent_pte)
1517 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
1518 if (pt[i] == shadow_notrap_nonpresent_pte)
1519 set_shadow_pte(&pt[i], shadow_trap_nonpresent_pte);
1523 struct page *gva_to_page(struct kvm_vcpu *vcpu, gva_t gva)
1527 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gva);
1529 if (gpa == UNMAPPED_GVA)
1532 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1538 * The function is based on mtrr_type_lookup() in
1539 * arch/x86/kernel/cpu/mtrr/generic.c
1541 static int get_mtrr_type(struct mtrr_state_type *mtrr_state,
1546 u8 prev_match, curr_match;
1547 int num_var_ranges = KVM_NR_VAR_MTRR;
1549 if (!mtrr_state->enabled)
1552 /* Make end inclusive end, instead of exclusive */
1555 /* Look in fixed ranges. Just return the type as per start */
1556 if (mtrr_state->have_fixed && (start < 0x100000)) {
1559 if (start < 0x80000) {
1561 idx += (start >> 16);
1562 return mtrr_state->fixed_ranges[idx];
1563 } else if (start < 0xC0000) {
1565 idx += ((start - 0x80000) >> 14);
1566 return mtrr_state->fixed_ranges[idx];
1567 } else if (start < 0x1000000) {
1569 idx += ((start - 0xC0000) >> 12);
1570 return mtrr_state->fixed_ranges[idx];
1575 * Look in variable ranges
1576 * Look of multiple ranges matching this address and pick type
1577 * as per MTRR precedence
1579 if (!(mtrr_state->enabled & 2))
1580 return mtrr_state->def_type;
1583 for (i = 0; i < num_var_ranges; ++i) {
1584 unsigned short start_state, end_state;
1586 if (!(mtrr_state->var_ranges[i].mask_lo & (1 << 11)))
1589 base = (((u64)mtrr_state->var_ranges[i].base_hi) << 32) +
1590 (mtrr_state->var_ranges[i].base_lo & PAGE_MASK);
1591 mask = (((u64)mtrr_state->var_ranges[i].mask_hi) << 32) +
1592 (mtrr_state->var_ranges[i].mask_lo & PAGE_MASK);
1594 start_state = ((start & mask) == (base & mask));
1595 end_state = ((end & mask) == (base & mask));
1596 if (start_state != end_state)
1599 if ((start & mask) != (base & mask))
1602 curr_match = mtrr_state->var_ranges[i].base_lo & 0xff;
1603 if (prev_match == 0xFF) {
1604 prev_match = curr_match;
1608 if (prev_match == MTRR_TYPE_UNCACHABLE ||
1609 curr_match == MTRR_TYPE_UNCACHABLE)
1610 return MTRR_TYPE_UNCACHABLE;
1612 if ((prev_match == MTRR_TYPE_WRBACK &&
1613 curr_match == MTRR_TYPE_WRTHROUGH) ||
1614 (prev_match == MTRR_TYPE_WRTHROUGH &&
1615 curr_match == MTRR_TYPE_WRBACK)) {
1616 prev_match = MTRR_TYPE_WRTHROUGH;
1617 curr_match = MTRR_TYPE_WRTHROUGH;
1620 if (prev_match != curr_match)
1621 return MTRR_TYPE_UNCACHABLE;
1624 if (prev_match != 0xFF)
1627 return mtrr_state->def_type;
1630 u8 kvm_get_guest_memory_type(struct kvm_vcpu *vcpu, gfn_t gfn)
1634 mtrr = get_mtrr_type(&vcpu->arch.mtrr_state, gfn << PAGE_SHIFT,
1635 (gfn << PAGE_SHIFT) + PAGE_SIZE);
1636 if (mtrr == 0xfe || mtrr == 0xff)
1637 mtrr = MTRR_TYPE_WRBACK;
1640 EXPORT_SYMBOL_GPL(kvm_get_guest_memory_type);
1642 static int kvm_unsync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
1645 struct hlist_head *bucket;
1646 struct kvm_mmu_page *s;
1647 struct hlist_node *node, *n;
1649 index = kvm_page_table_hashfn(sp->gfn);
1650 bucket = &vcpu->kvm->arch.mmu_page_hash[index];
1651 /* don't unsync if pagetable is shadowed with multiple roles */
1652 hlist_for_each_entry_safe(s, node, n, bucket, hash_link) {
1653 if (s->gfn != sp->gfn || s->role.direct)
1655 if (s->role.word != sp->role.word)
1658 ++vcpu->kvm->stat.mmu_unsync;
1661 kvm_mmu_mark_parents_unsync(vcpu, sp);
1663 mmu_convert_notrap(sp);
1667 static int mmu_need_write_protect(struct kvm_vcpu *vcpu, gfn_t gfn,
1670 struct kvm_mmu_page *shadow;
1672 shadow = kvm_mmu_lookup_page(vcpu->kvm, gfn);
1674 if (shadow->role.level != PT_PAGE_TABLE_LEVEL)
1678 if (can_unsync && oos_shadow)
1679 return kvm_unsync_page(vcpu, shadow);
1685 static int set_spte(struct kvm_vcpu *vcpu, u64 *shadow_pte,
1686 unsigned pte_access, int user_fault,
1687 int write_fault, int dirty, int largepage,
1688 gfn_t gfn, pfn_t pfn, bool speculative,
1695 * We don't set the accessed bit, since we sometimes want to see
1696 * whether the guest actually used the pte (in order to detect
1699 spte = shadow_base_present_pte | shadow_dirty_mask;
1701 spte |= shadow_accessed_mask;
1703 pte_access &= ~ACC_WRITE_MASK;
1704 if (pte_access & ACC_EXEC_MASK)
1705 spte |= shadow_x_mask;
1707 spte |= shadow_nx_mask;
1708 if (pte_access & ACC_USER_MASK)
1709 spte |= shadow_user_mask;
1711 spte |= PT_PAGE_SIZE_MASK;
1713 spte |= kvm_x86_ops->get_mt_mask(vcpu, gfn,
1714 kvm_is_mmio_pfn(pfn));
1716 spte |= (u64)pfn << PAGE_SHIFT;
1718 if ((pte_access & ACC_WRITE_MASK)
1719 || (write_fault && !is_write_protection(vcpu) && !user_fault)) {
1721 if (largepage && has_wrprotected_page(vcpu->kvm, gfn)) {
1723 spte = shadow_trap_nonpresent_pte;
1727 spte |= PT_WRITABLE_MASK;
1730 * Optimization: for pte sync, if spte was writable the hash
1731 * lookup is unnecessary (and expensive). Write protection
1732 * is responsibility of mmu_get_page / kvm_sync_page.
1733 * Same reasoning can be applied to dirty page accounting.
1735 if (!can_unsync && is_writeble_pte(*shadow_pte))
1738 if (mmu_need_write_protect(vcpu, gfn, can_unsync)) {
1739 pgprintk("%s: found shadow page for %lx, marking ro\n",
1742 pte_access &= ~ACC_WRITE_MASK;
1743 if (is_writeble_pte(spte))
1744 spte &= ~PT_WRITABLE_MASK;
1748 if (pte_access & ACC_WRITE_MASK)
1749 mark_page_dirty(vcpu->kvm, gfn);
1752 set_shadow_pte(shadow_pte, spte);
1756 static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *shadow_pte,
1757 unsigned pt_access, unsigned pte_access,
1758 int user_fault, int write_fault, int dirty,
1759 int *ptwrite, int largepage, gfn_t gfn,
1760 pfn_t pfn, bool speculative)
1762 int was_rmapped = 0;
1763 int was_writeble = is_writeble_pte(*shadow_pte);
1766 pgprintk("%s: spte %llx access %x write_fault %d"
1767 " user_fault %d gfn %lx\n",
1768 __func__, *shadow_pte, pt_access,
1769 write_fault, user_fault, gfn);
1771 if (is_rmap_pte(*shadow_pte)) {
1773 * If we overwrite a PTE page pointer with a 2MB PMD, unlink
1774 * the parent of the now unreachable PTE.
1776 if (largepage && !is_large_pte(*shadow_pte)) {
1777 struct kvm_mmu_page *child;
1778 u64 pte = *shadow_pte;
1780 child = page_header(pte & PT64_BASE_ADDR_MASK);
1781 mmu_page_remove_parent_pte(child, shadow_pte);
1782 } else if (pfn != spte_to_pfn(*shadow_pte)) {
1783 pgprintk("hfn old %lx new %lx\n",
1784 spte_to_pfn(*shadow_pte), pfn);
1785 rmap_remove(vcpu->kvm, shadow_pte);
1789 if (set_spte(vcpu, shadow_pte, pte_access, user_fault, write_fault,
1790 dirty, largepage, gfn, pfn, speculative, true)) {
1793 kvm_x86_ops->tlb_flush(vcpu);
1796 pgprintk("%s: setting spte %llx\n", __func__, *shadow_pte);
1797 pgprintk("instantiating %s PTE (%s) at %ld (%llx) addr %p\n",
1798 is_large_pte(*shadow_pte)? "2MB" : "4kB",
1799 is_present_pte(*shadow_pte)?"RW":"R", gfn,
1800 *shadow_pte, shadow_pte);
1801 if (!was_rmapped && is_large_pte(*shadow_pte))
1802 ++vcpu->kvm->stat.lpages;
1804 page_header_update_slot(vcpu->kvm, shadow_pte, gfn);
1806 rmap_count = rmap_add(vcpu, shadow_pte, gfn, largepage);
1807 if (!is_rmap_pte(*shadow_pte))
1808 kvm_release_pfn_clean(pfn);
1809 if (rmap_count > RMAP_RECYCLE_THRESHOLD)
1810 rmap_recycle(vcpu, gfn, largepage);
1813 kvm_release_pfn_dirty(pfn);
1815 kvm_release_pfn_clean(pfn);
1818 vcpu->arch.last_pte_updated = shadow_pte;
1819 vcpu->arch.last_pte_gfn = gfn;
1823 static void nonpaging_new_cr3(struct kvm_vcpu *vcpu)
1827 static int __direct_map(struct kvm_vcpu *vcpu, gpa_t v, int write,
1828 int largepage, gfn_t gfn, pfn_t pfn)
1830 struct kvm_shadow_walk_iterator iterator;
1831 struct kvm_mmu_page *sp;
1835 for_each_shadow_entry(vcpu, (u64)gfn << PAGE_SHIFT, iterator) {
1836 if (iterator.level == PT_PAGE_TABLE_LEVEL
1837 || (largepage && iterator.level == PT_DIRECTORY_LEVEL)) {
1838 mmu_set_spte(vcpu, iterator.sptep, ACC_ALL, ACC_ALL,
1839 0, write, 1, &pt_write,
1840 largepage, gfn, pfn, false);
1841 ++vcpu->stat.pf_fixed;
1845 if (*iterator.sptep == shadow_trap_nonpresent_pte) {
1846 pseudo_gfn = (iterator.addr & PT64_DIR_BASE_ADDR_MASK) >> PAGE_SHIFT;
1847 sp = kvm_mmu_get_page(vcpu, pseudo_gfn, iterator.addr,
1849 1, ACC_ALL, iterator.sptep);
1851 pgprintk("nonpaging_map: ENOMEM\n");
1852 kvm_release_pfn_clean(pfn);
1856 set_shadow_pte(iterator.sptep,
1858 | PT_PRESENT_MASK | PT_WRITABLE_MASK
1859 | shadow_user_mask | shadow_x_mask);
1865 static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, int write, gfn_t gfn)
1870 unsigned long mmu_seq;
1872 if (is_largepage_backed(vcpu, gfn & ~(KVM_PAGES_PER_HPAGE-1))) {
1873 gfn &= ~(KVM_PAGES_PER_HPAGE-1);
1877 mmu_seq = vcpu->kvm->mmu_notifier_seq;
1879 pfn = gfn_to_pfn(vcpu->kvm, gfn);
1882 if (is_error_pfn(pfn)) {
1883 kvm_release_pfn_clean(pfn);
1887 spin_lock(&vcpu->kvm->mmu_lock);
1888 if (mmu_notifier_retry(vcpu, mmu_seq))
1890 kvm_mmu_free_some_pages(vcpu);
1891 r = __direct_map(vcpu, v, write, largepage, gfn, pfn);
1892 spin_unlock(&vcpu->kvm->mmu_lock);
1898 spin_unlock(&vcpu->kvm->mmu_lock);
1899 kvm_release_pfn_clean(pfn);
1904 static void mmu_free_roots(struct kvm_vcpu *vcpu)
1907 struct kvm_mmu_page *sp;
1909 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
1911 spin_lock(&vcpu->kvm->mmu_lock);
1912 if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) {
1913 hpa_t root = vcpu->arch.mmu.root_hpa;
1915 sp = page_header(root);
1917 if (!sp->root_count && sp->role.invalid)
1918 kvm_mmu_zap_page(vcpu->kvm, sp);
1919 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
1920 spin_unlock(&vcpu->kvm->mmu_lock);
1923 for (i = 0; i < 4; ++i) {
1924 hpa_t root = vcpu->arch.mmu.pae_root[i];
1927 root &= PT64_BASE_ADDR_MASK;
1928 sp = page_header(root);
1930 if (!sp->root_count && sp->role.invalid)
1931 kvm_mmu_zap_page(vcpu->kvm, sp);
1933 vcpu->arch.mmu.pae_root[i] = INVALID_PAGE;
1935 spin_unlock(&vcpu->kvm->mmu_lock);
1936 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
1939 static int mmu_check_root(struct kvm_vcpu *vcpu, gfn_t root_gfn)
1943 if (!kvm_is_visible_gfn(vcpu->kvm, root_gfn)) {
1944 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
1951 static int mmu_alloc_roots(struct kvm_vcpu *vcpu)
1955 struct kvm_mmu_page *sp;
1958 root_gfn = vcpu->arch.cr3 >> PAGE_SHIFT;
1960 if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) {
1961 hpa_t root = vcpu->arch.mmu.root_hpa;
1963 ASSERT(!VALID_PAGE(root));
1966 if (mmu_check_root(vcpu, root_gfn))
1968 sp = kvm_mmu_get_page(vcpu, root_gfn, 0,
1969 PT64_ROOT_LEVEL, direct,
1971 root = __pa(sp->spt);
1973 vcpu->arch.mmu.root_hpa = root;
1976 direct = !is_paging(vcpu);
1979 for (i = 0; i < 4; ++i) {
1980 hpa_t root = vcpu->arch.mmu.pae_root[i];
1982 ASSERT(!VALID_PAGE(root));
1983 if (vcpu->arch.mmu.root_level == PT32E_ROOT_LEVEL) {
1984 if (!is_present_pte(vcpu->arch.pdptrs[i])) {
1985 vcpu->arch.mmu.pae_root[i] = 0;
1988 root_gfn = vcpu->arch.pdptrs[i] >> PAGE_SHIFT;
1989 } else if (vcpu->arch.mmu.root_level == 0)
1991 if (mmu_check_root(vcpu, root_gfn))
1993 sp = kvm_mmu_get_page(vcpu, root_gfn, i << 30,
1994 PT32_ROOT_LEVEL, direct,
1996 root = __pa(sp->spt);
1998 vcpu->arch.mmu.pae_root[i] = root | PT_PRESENT_MASK;
2000 vcpu->arch.mmu.root_hpa = __pa(vcpu->arch.mmu.pae_root);
2004 static void mmu_sync_roots(struct kvm_vcpu *vcpu)
2007 struct kvm_mmu_page *sp;
2009 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
2011 if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) {
2012 hpa_t root = vcpu->arch.mmu.root_hpa;
2013 sp = page_header(root);
2014 mmu_sync_children(vcpu, sp);
2017 for (i = 0; i < 4; ++i) {
2018 hpa_t root = vcpu->arch.mmu.pae_root[i];
2020 if (root && VALID_PAGE(root)) {
2021 root &= PT64_BASE_ADDR_MASK;
2022 sp = page_header(root);
2023 mmu_sync_children(vcpu, sp);
2028 void kvm_mmu_sync_roots(struct kvm_vcpu *vcpu)
2030 spin_lock(&vcpu->kvm->mmu_lock);
2031 mmu_sync_roots(vcpu);
2032 spin_unlock(&vcpu->kvm->mmu_lock);
2035 static gpa_t nonpaging_gva_to_gpa(struct kvm_vcpu *vcpu, gva_t vaddr)
2040 static int nonpaging_page_fault(struct kvm_vcpu *vcpu, gva_t gva,
2046 pgprintk("%s: gva %lx error %x\n", __func__, gva, error_code);
2047 r = mmu_topup_memory_caches(vcpu);
2052 ASSERT(VALID_PAGE(vcpu->arch.mmu.root_hpa));
2054 gfn = gva >> PAGE_SHIFT;
2056 return nonpaging_map(vcpu, gva & PAGE_MASK,
2057 error_code & PFERR_WRITE_MASK, gfn);
2060 static int tdp_page_fault(struct kvm_vcpu *vcpu, gva_t gpa,
2066 gfn_t gfn = gpa >> PAGE_SHIFT;
2067 unsigned long mmu_seq;
2070 ASSERT(VALID_PAGE(vcpu->arch.mmu.root_hpa));
2072 r = mmu_topup_memory_caches(vcpu);
2076 if (is_largepage_backed(vcpu, gfn & ~(KVM_PAGES_PER_HPAGE-1))) {
2077 gfn &= ~(KVM_PAGES_PER_HPAGE-1);
2080 mmu_seq = vcpu->kvm->mmu_notifier_seq;
2082 pfn = gfn_to_pfn(vcpu->kvm, gfn);
2083 if (is_error_pfn(pfn)) {
2084 kvm_release_pfn_clean(pfn);
2087 spin_lock(&vcpu->kvm->mmu_lock);
2088 if (mmu_notifier_retry(vcpu, mmu_seq))
2090 kvm_mmu_free_some_pages(vcpu);
2091 r = __direct_map(vcpu, gpa, error_code & PFERR_WRITE_MASK,
2092 largepage, gfn, pfn);
2093 spin_unlock(&vcpu->kvm->mmu_lock);
2098 spin_unlock(&vcpu->kvm->mmu_lock);
2099 kvm_release_pfn_clean(pfn);
2103 static void nonpaging_free(struct kvm_vcpu *vcpu)
2105 mmu_free_roots(vcpu);
2108 static int nonpaging_init_context(struct kvm_vcpu *vcpu)
2110 struct kvm_mmu *context = &vcpu->arch.mmu;
2112 context->new_cr3 = nonpaging_new_cr3;
2113 context->page_fault = nonpaging_page_fault;
2114 context->gva_to_gpa = nonpaging_gva_to_gpa;
2115 context->free = nonpaging_free;
2116 context->prefetch_page = nonpaging_prefetch_page;
2117 context->sync_page = nonpaging_sync_page;
2118 context->invlpg = nonpaging_invlpg;
2119 context->root_level = 0;
2120 context->shadow_root_level = PT32E_ROOT_LEVEL;
2121 context->root_hpa = INVALID_PAGE;
2125 void kvm_mmu_flush_tlb(struct kvm_vcpu *vcpu)
2127 ++vcpu->stat.tlb_flush;
2128 kvm_x86_ops->tlb_flush(vcpu);
2131 static void paging_new_cr3(struct kvm_vcpu *vcpu)
2133 pgprintk("%s: cr3 %lx\n", __func__, vcpu->arch.cr3);
2134 mmu_free_roots(vcpu);
2137 static void inject_page_fault(struct kvm_vcpu *vcpu,
2141 kvm_inject_page_fault(vcpu, addr, err_code);
2144 static void paging_free(struct kvm_vcpu *vcpu)
2146 nonpaging_free(vcpu);
2149 static bool is_rsvd_bits_set(struct kvm_vcpu *vcpu, u64 gpte, int level)
2153 bit7 = (gpte >> 7) & 1;
2154 return (gpte & vcpu->arch.mmu.rsvd_bits_mask[bit7][level-1]) != 0;
2158 #include "paging_tmpl.h"
2162 #include "paging_tmpl.h"
2165 static void reset_rsvds_bits_mask(struct kvm_vcpu *vcpu, int level)
2167 struct kvm_mmu *context = &vcpu->arch.mmu;
2168 int maxphyaddr = cpuid_maxphyaddr(vcpu);
2169 u64 exb_bit_rsvd = 0;
2172 exb_bit_rsvd = rsvd_bits(63, 63);
2174 case PT32_ROOT_LEVEL:
2175 /* no rsvd bits for 2 level 4K page table entries */
2176 context->rsvd_bits_mask[0][1] = 0;
2177 context->rsvd_bits_mask[0][0] = 0;
2178 if (is_cpuid_PSE36())
2179 /* 36bits PSE 4MB page */
2180 context->rsvd_bits_mask[1][1] = rsvd_bits(17, 21);
2182 /* 32 bits PSE 4MB page */
2183 context->rsvd_bits_mask[1][1] = rsvd_bits(13, 21);
2184 context->rsvd_bits_mask[1][0] = context->rsvd_bits_mask[1][0];
2186 case PT32E_ROOT_LEVEL:
2187 context->rsvd_bits_mask[0][2] =
2188 rsvd_bits(maxphyaddr, 63) |
2189 rsvd_bits(7, 8) | rsvd_bits(1, 2); /* PDPTE */
2190 context->rsvd_bits_mask[0][1] = exb_bit_rsvd |
2191 rsvd_bits(maxphyaddr, 62); /* PDE */
2192 context->rsvd_bits_mask[0][0] = exb_bit_rsvd |
2193 rsvd_bits(maxphyaddr, 62); /* PTE */
2194 context->rsvd_bits_mask[1][1] = exb_bit_rsvd |
2195 rsvd_bits(maxphyaddr, 62) |
2196 rsvd_bits(13, 20); /* large page */
2197 context->rsvd_bits_mask[1][0] = context->rsvd_bits_mask[1][0];
2199 case PT64_ROOT_LEVEL:
2200 context->rsvd_bits_mask[0][3] = exb_bit_rsvd |
2201 rsvd_bits(maxphyaddr, 51) | rsvd_bits(7, 8);
2202 context->rsvd_bits_mask[0][2] = exb_bit_rsvd |
2203 rsvd_bits(maxphyaddr, 51) | rsvd_bits(7, 8);
2204 context->rsvd_bits_mask[0][1] = exb_bit_rsvd |
2205 rsvd_bits(maxphyaddr, 51);
2206 context->rsvd_bits_mask[0][0] = exb_bit_rsvd |
2207 rsvd_bits(maxphyaddr, 51);
2208 context->rsvd_bits_mask[1][3] = context->rsvd_bits_mask[0][3];
2209 context->rsvd_bits_mask[1][2] = context->rsvd_bits_mask[0][2];
2210 context->rsvd_bits_mask[1][1] = exb_bit_rsvd |
2211 rsvd_bits(maxphyaddr, 51) |
2212 rsvd_bits(13, 20); /* large page */
2213 context->rsvd_bits_mask[1][0] = context->rsvd_bits_mask[1][0];
2218 static int paging64_init_context_common(struct kvm_vcpu *vcpu, int level)
2220 struct kvm_mmu *context = &vcpu->arch.mmu;
2222 ASSERT(is_pae(vcpu));
2223 context->new_cr3 = paging_new_cr3;
2224 context->page_fault = paging64_page_fault;
2225 context->gva_to_gpa = paging64_gva_to_gpa;
2226 context->prefetch_page = paging64_prefetch_page;
2227 context->sync_page = paging64_sync_page;
2228 context->invlpg = paging64_invlpg;
2229 context->free = paging_free;
2230 context->root_level = level;
2231 context->shadow_root_level = level;
2232 context->root_hpa = INVALID_PAGE;
2236 static int paging64_init_context(struct kvm_vcpu *vcpu)
2238 reset_rsvds_bits_mask(vcpu, PT64_ROOT_LEVEL);
2239 return paging64_init_context_common(vcpu, PT64_ROOT_LEVEL);
2242 static int paging32_init_context(struct kvm_vcpu *vcpu)
2244 struct kvm_mmu *context = &vcpu->arch.mmu;
2246 reset_rsvds_bits_mask(vcpu, PT32_ROOT_LEVEL);
2247 context->new_cr3 = paging_new_cr3;
2248 context->page_fault = paging32_page_fault;
2249 context->gva_to_gpa = paging32_gva_to_gpa;
2250 context->free = paging_free;
2251 context->prefetch_page = paging32_prefetch_page;
2252 context->sync_page = paging32_sync_page;
2253 context->invlpg = paging32_invlpg;
2254 context->root_level = PT32_ROOT_LEVEL;
2255 context->shadow_root_level = PT32E_ROOT_LEVEL;
2256 context->root_hpa = INVALID_PAGE;
2260 static int paging32E_init_context(struct kvm_vcpu *vcpu)
2262 reset_rsvds_bits_mask(vcpu, PT32E_ROOT_LEVEL);
2263 return paging64_init_context_common(vcpu, PT32E_ROOT_LEVEL);
2266 static int init_kvm_tdp_mmu(struct kvm_vcpu *vcpu)
2268 struct kvm_mmu *context = &vcpu->arch.mmu;
2270 context->new_cr3 = nonpaging_new_cr3;
2271 context->page_fault = tdp_page_fault;
2272 context->free = nonpaging_free;
2273 context->prefetch_page = nonpaging_prefetch_page;
2274 context->sync_page = nonpaging_sync_page;
2275 context->invlpg = nonpaging_invlpg;
2276 context->shadow_root_level = kvm_x86_ops->get_tdp_level();
2277 context->root_hpa = INVALID_PAGE;
2279 if (!is_paging(vcpu)) {
2280 context->gva_to_gpa = nonpaging_gva_to_gpa;
2281 context->root_level = 0;
2282 } else if (is_long_mode(vcpu)) {
2283 reset_rsvds_bits_mask(vcpu, PT64_ROOT_LEVEL);
2284 context->gva_to_gpa = paging64_gva_to_gpa;
2285 context->root_level = PT64_ROOT_LEVEL;
2286 } else if (is_pae(vcpu)) {
2287 reset_rsvds_bits_mask(vcpu, PT32E_ROOT_LEVEL);
2288 context->gva_to_gpa = paging64_gva_to_gpa;
2289 context->root_level = PT32E_ROOT_LEVEL;
2291 reset_rsvds_bits_mask(vcpu, PT32_ROOT_LEVEL);
2292 context->gva_to_gpa = paging32_gva_to_gpa;
2293 context->root_level = PT32_ROOT_LEVEL;
2299 static int init_kvm_softmmu(struct kvm_vcpu *vcpu)
2304 ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa));
2306 if (!is_paging(vcpu))
2307 r = nonpaging_init_context(vcpu);
2308 else if (is_long_mode(vcpu))
2309 r = paging64_init_context(vcpu);
2310 else if (is_pae(vcpu))
2311 r = paging32E_init_context(vcpu);
2313 r = paging32_init_context(vcpu);
2315 vcpu->arch.mmu.base_role.glevels = vcpu->arch.mmu.root_level;
2320 static int init_kvm_mmu(struct kvm_vcpu *vcpu)
2322 vcpu->arch.update_pte.pfn = bad_pfn;
2325 return init_kvm_tdp_mmu(vcpu);
2327 return init_kvm_softmmu(vcpu);
2330 static void destroy_kvm_mmu(struct kvm_vcpu *vcpu)
2333 if (VALID_PAGE(vcpu->arch.mmu.root_hpa)) {
2334 vcpu->arch.mmu.free(vcpu);
2335 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
2339 int kvm_mmu_reset_context(struct kvm_vcpu *vcpu)
2341 destroy_kvm_mmu(vcpu);
2342 return init_kvm_mmu(vcpu);
2344 EXPORT_SYMBOL_GPL(kvm_mmu_reset_context);
2346 int kvm_mmu_load(struct kvm_vcpu *vcpu)
2350 r = mmu_topup_memory_caches(vcpu);
2353 spin_lock(&vcpu->kvm->mmu_lock);
2354 kvm_mmu_free_some_pages(vcpu);
2355 r = mmu_alloc_roots(vcpu);
2356 mmu_sync_roots(vcpu);
2357 spin_unlock(&vcpu->kvm->mmu_lock);
2360 kvm_x86_ops->set_cr3(vcpu, vcpu->arch.mmu.root_hpa);
2361 kvm_mmu_flush_tlb(vcpu);
2365 EXPORT_SYMBOL_GPL(kvm_mmu_load);
2367 void kvm_mmu_unload(struct kvm_vcpu *vcpu)
2369 mmu_free_roots(vcpu);
2372 static void mmu_pte_write_zap_pte(struct kvm_vcpu *vcpu,
2373 struct kvm_mmu_page *sp,
2377 struct kvm_mmu_page *child;
2380 if (is_shadow_present_pte(pte)) {
2381 if (sp->role.level == PT_PAGE_TABLE_LEVEL ||
2383 rmap_remove(vcpu->kvm, spte);
2385 child = page_header(pte & PT64_BASE_ADDR_MASK);
2386 mmu_page_remove_parent_pte(child, spte);
2389 set_shadow_pte(spte, shadow_trap_nonpresent_pte);
2390 if (is_large_pte(pte))
2391 --vcpu->kvm->stat.lpages;
2394 static void mmu_pte_write_new_pte(struct kvm_vcpu *vcpu,
2395 struct kvm_mmu_page *sp,
2399 if (sp->role.level != PT_PAGE_TABLE_LEVEL) {
2400 if (!vcpu->arch.update_pte.largepage ||
2401 sp->role.glevels == PT32_ROOT_LEVEL) {
2402 ++vcpu->kvm->stat.mmu_pde_zapped;
2407 ++vcpu->kvm->stat.mmu_pte_updated;
2408 if (sp->role.glevels == PT32_ROOT_LEVEL)
2409 paging32_update_pte(vcpu, sp, spte, new);
2411 paging64_update_pte(vcpu, sp, spte, new);
2414 static bool need_remote_flush(u64 old, u64 new)
2416 if (!is_shadow_present_pte(old))
2418 if (!is_shadow_present_pte(new))
2420 if ((old ^ new) & PT64_BASE_ADDR_MASK)
2422 old ^= PT64_NX_MASK;
2423 new ^= PT64_NX_MASK;
2424 return (old & ~new & PT64_PERM_MASK) != 0;
2427 static void mmu_pte_write_flush_tlb(struct kvm_vcpu *vcpu, u64 old, u64 new)
2429 if (need_remote_flush(old, new))
2430 kvm_flush_remote_tlbs(vcpu->kvm);
2432 kvm_mmu_flush_tlb(vcpu);
2435 static bool last_updated_pte_accessed(struct kvm_vcpu *vcpu)
2437 u64 *spte = vcpu->arch.last_pte_updated;
2439 return !!(spte && (*spte & shadow_accessed_mask));
2442 static void mmu_guess_page_from_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
2443 const u8 *new, int bytes)
2450 vcpu->arch.update_pte.largepage = 0;
2452 if (bytes != 4 && bytes != 8)
2456 * Assume that the pte write on a page table of the same type
2457 * as the current vcpu paging mode. This is nearly always true
2458 * (might be false while changing modes). Note it is verified later
2462 /* Handle a 32-bit guest writing two halves of a 64-bit gpte */
2463 if ((bytes == 4) && (gpa % 4 == 0)) {
2464 r = kvm_read_guest(vcpu->kvm, gpa & ~(u64)7, &gpte, 8);
2467 memcpy((void *)&gpte + (gpa % 8), new, 4);
2468 } else if ((bytes == 8) && (gpa % 8 == 0)) {
2469 memcpy((void *)&gpte, new, 8);
2472 if ((bytes == 4) && (gpa % 4 == 0))
2473 memcpy((void *)&gpte, new, 4);
2475 if (!is_present_pte(gpte))
2477 gfn = (gpte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT;
2479 if (is_large_pte(gpte) && is_largepage_backed(vcpu, gfn)) {
2480 gfn &= ~(KVM_PAGES_PER_HPAGE-1);
2481 vcpu->arch.update_pte.largepage = 1;
2483 vcpu->arch.update_pte.mmu_seq = vcpu->kvm->mmu_notifier_seq;
2485 pfn = gfn_to_pfn(vcpu->kvm, gfn);
2487 if (is_error_pfn(pfn)) {
2488 kvm_release_pfn_clean(pfn);
2491 vcpu->arch.update_pte.gfn = gfn;
2492 vcpu->arch.update_pte.pfn = pfn;
2495 static void kvm_mmu_access_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2497 u64 *spte = vcpu->arch.last_pte_updated;
2500 && vcpu->arch.last_pte_gfn == gfn
2501 && shadow_accessed_mask
2502 && !(*spte & shadow_accessed_mask)
2503 && is_shadow_present_pte(*spte))
2504 set_bit(PT_ACCESSED_SHIFT, (unsigned long *)spte);
2507 void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
2508 const u8 *new, int bytes,
2509 bool guest_initiated)
2511 gfn_t gfn = gpa >> PAGE_SHIFT;
2512 struct kvm_mmu_page *sp;
2513 struct hlist_node *node, *n;
2514 struct hlist_head *bucket;
2518 unsigned offset = offset_in_page(gpa);
2520 unsigned page_offset;
2521 unsigned misaligned;
2528 pgprintk("%s: gpa %llx bytes %d\n", __func__, gpa, bytes);
2529 mmu_guess_page_from_pte_write(vcpu, gpa, new, bytes);
2530 spin_lock(&vcpu->kvm->mmu_lock);
2531 kvm_mmu_access_page(vcpu, gfn);
2532 kvm_mmu_free_some_pages(vcpu);
2533 ++vcpu->kvm->stat.mmu_pte_write;
2534 kvm_mmu_audit(vcpu, "pre pte write");
2535 if (guest_initiated) {
2536 if (gfn == vcpu->arch.last_pt_write_gfn
2537 && !last_updated_pte_accessed(vcpu)) {
2538 ++vcpu->arch.last_pt_write_count;
2539 if (vcpu->arch.last_pt_write_count >= 3)
2542 vcpu->arch.last_pt_write_gfn = gfn;
2543 vcpu->arch.last_pt_write_count = 1;
2544 vcpu->arch.last_pte_updated = NULL;
2547 index = kvm_page_table_hashfn(gfn);
2548 bucket = &vcpu->kvm->arch.mmu_page_hash[index];
2549 hlist_for_each_entry_safe(sp, node, n, bucket, hash_link) {
2550 if (sp->gfn != gfn || sp->role.direct || sp->role.invalid)
2552 pte_size = sp->role.glevels == PT32_ROOT_LEVEL ? 4 : 8;
2553 misaligned = (offset ^ (offset + bytes - 1)) & ~(pte_size - 1);
2554 misaligned |= bytes < 4;
2555 if (misaligned || flooded) {
2557 * Misaligned accesses are too much trouble to fix
2558 * up; also, they usually indicate a page is not used
2561 * If we're seeing too many writes to a page,
2562 * it may no longer be a page table, or we may be
2563 * forking, in which case it is better to unmap the
2566 pgprintk("misaligned: gpa %llx bytes %d role %x\n",
2567 gpa, bytes, sp->role.word);
2568 if (kvm_mmu_zap_page(vcpu->kvm, sp))
2570 ++vcpu->kvm->stat.mmu_flooded;
2573 page_offset = offset;
2574 level = sp->role.level;
2576 if (sp->role.glevels == PT32_ROOT_LEVEL) {
2577 page_offset <<= 1; /* 32->64 */
2579 * A 32-bit pde maps 4MB while the shadow pdes map
2580 * only 2MB. So we need to double the offset again
2581 * and zap two pdes instead of one.
2583 if (level == PT32_ROOT_LEVEL) {
2584 page_offset &= ~7; /* kill rounding error */
2588 quadrant = page_offset >> PAGE_SHIFT;
2589 page_offset &= ~PAGE_MASK;
2590 if (quadrant != sp->role.quadrant)
2593 spte = &sp->spt[page_offset / sizeof(*spte)];
2594 if ((gpa & (pte_size - 1)) || (bytes < pte_size)) {
2596 r = kvm_read_guest_atomic(vcpu->kvm,
2597 gpa & ~(u64)(pte_size - 1),
2599 new = (const void *)&gentry;
2605 mmu_pte_write_zap_pte(vcpu, sp, spte);
2607 mmu_pte_write_new_pte(vcpu, sp, spte, new);
2608 mmu_pte_write_flush_tlb(vcpu, entry, *spte);
2612 kvm_mmu_audit(vcpu, "post pte write");
2613 spin_unlock(&vcpu->kvm->mmu_lock);
2614 if (!is_error_pfn(vcpu->arch.update_pte.pfn)) {
2615 kvm_release_pfn_clean(vcpu->arch.update_pte.pfn);
2616 vcpu->arch.update_pte.pfn = bad_pfn;
2620 int kvm_mmu_unprotect_page_virt(struct kvm_vcpu *vcpu, gva_t gva)
2625 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gva);
2627 spin_lock(&vcpu->kvm->mmu_lock);
2628 r = kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2629 spin_unlock(&vcpu->kvm->mmu_lock);
2632 EXPORT_SYMBOL_GPL(kvm_mmu_unprotect_page_virt);
2634 void __kvm_mmu_free_some_pages(struct kvm_vcpu *vcpu)
2636 while (vcpu->kvm->arch.n_free_mmu_pages < KVM_REFILL_PAGES) {
2637 struct kvm_mmu_page *sp;
2639 sp = container_of(vcpu->kvm->arch.active_mmu_pages.prev,
2640 struct kvm_mmu_page, link);
2641 kvm_mmu_zap_page(vcpu->kvm, sp);
2642 ++vcpu->kvm->stat.mmu_recycled;
2646 int kvm_mmu_page_fault(struct kvm_vcpu *vcpu, gva_t cr2, u32 error_code)
2649 enum emulation_result er;
2651 r = vcpu->arch.mmu.page_fault(vcpu, cr2, error_code);
2660 r = mmu_topup_memory_caches(vcpu);
2664 er = emulate_instruction(vcpu, vcpu->run, cr2, error_code, 0);
2669 case EMULATE_DO_MMIO:
2670 ++vcpu->stat.mmio_exits;
2673 kvm_report_emulation_failure(vcpu, "pagetable");
2681 EXPORT_SYMBOL_GPL(kvm_mmu_page_fault);
2683 void kvm_mmu_invlpg(struct kvm_vcpu *vcpu, gva_t gva)
2685 vcpu->arch.mmu.invlpg(vcpu, gva);
2686 kvm_mmu_flush_tlb(vcpu);
2687 ++vcpu->stat.invlpg;
2689 EXPORT_SYMBOL_GPL(kvm_mmu_invlpg);
2691 void kvm_enable_tdp(void)
2695 EXPORT_SYMBOL_GPL(kvm_enable_tdp);
2697 void kvm_disable_tdp(void)
2699 tdp_enabled = false;
2701 EXPORT_SYMBOL_GPL(kvm_disable_tdp);
2703 static void free_mmu_pages(struct kvm_vcpu *vcpu)
2705 free_page((unsigned long)vcpu->arch.mmu.pae_root);
2708 static int alloc_mmu_pages(struct kvm_vcpu *vcpu)
2715 if (vcpu->kvm->arch.n_requested_mmu_pages)
2716 vcpu->kvm->arch.n_free_mmu_pages =
2717 vcpu->kvm->arch.n_requested_mmu_pages;
2719 vcpu->kvm->arch.n_free_mmu_pages =
2720 vcpu->kvm->arch.n_alloc_mmu_pages;
2722 * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64.
2723 * Therefore we need to allocate shadow page tables in the first
2724 * 4GB of memory, which happens to fit the DMA32 zone.
2726 page = alloc_page(GFP_KERNEL | __GFP_DMA32);
2729 vcpu->arch.mmu.pae_root = page_address(page);
2730 for (i = 0; i < 4; ++i)
2731 vcpu->arch.mmu.pae_root[i] = INVALID_PAGE;
2736 free_mmu_pages(vcpu);
2740 int kvm_mmu_create(struct kvm_vcpu *vcpu)
2743 ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa));
2745 return alloc_mmu_pages(vcpu);
2748 int kvm_mmu_setup(struct kvm_vcpu *vcpu)
2751 ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa));
2753 return init_kvm_mmu(vcpu);
2756 void kvm_mmu_destroy(struct kvm_vcpu *vcpu)
2760 destroy_kvm_mmu(vcpu);
2761 free_mmu_pages(vcpu);
2762 mmu_free_memory_caches(vcpu);
2765 void kvm_mmu_slot_remove_write_access(struct kvm *kvm, int slot)
2767 struct kvm_mmu_page *sp;
2769 list_for_each_entry(sp, &kvm->arch.active_mmu_pages, link) {
2773 if (!test_bit(slot, sp->slot_bitmap))
2777 for (i = 0; i < PT64_ENT_PER_PAGE; ++i)
2779 if (pt[i] & PT_WRITABLE_MASK)
2780 pt[i] &= ~PT_WRITABLE_MASK;
2782 kvm_flush_remote_tlbs(kvm);
2785 void kvm_mmu_zap_all(struct kvm *kvm)
2787 struct kvm_mmu_page *sp, *node;
2789 spin_lock(&kvm->mmu_lock);
2790 list_for_each_entry_safe(sp, node, &kvm->arch.active_mmu_pages, link)
2791 if (kvm_mmu_zap_page(kvm, sp))
2792 node = container_of(kvm->arch.active_mmu_pages.next,
2793 struct kvm_mmu_page, link);
2794 spin_unlock(&kvm->mmu_lock);
2796 kvm_flush_remote_tlbs(kvm);
2799 static void kvm_mmu_remove_one_alloc_mmu_page(struct kvm *kvm)
2801 struct kvm_mmu_page *page;
2803 page = container_of(kvm->arch.active_mmu_pages.prev,
2804 struct kvm_mmu_page, link);
2805 kvm_mmu_zap_page(kvm, page);
2808 static int mmu_shrink(int nr_to_scan, gfp_t gfp_mask)
2811 struct kvm *kvm_freed = NULL;
2812 int cache_count = 0;
2814 spin_lock(&kvm_lock);
2816 list_for_each_entry(kvm, &vm_list, vm_list) {
2819 if (!down_read_trylock(&kvm->slots_lock))
2821 spin_lock(&kvm->mmu_lock);
2822 npages = kvm->arch.n_alloc_mmu_pages -
2823 kvm->arch.n_free_mmu_pages;
2824 cache_count += npages;
2825 if (!kvm_freed && nr_to_scan > 0 && npages > 0) {
2826 kvm_mmu_remove_one_alloc_mmu_page(kvm);
2832 spin_unlock(&kvm->mmu_lock);
2833 up_read(&kvm->slots_lock);
2836 list_move_tail(&kvm_freed->vm_list, &vm_list);
2838 spin_unlock(&kvm_lock);
2843 static struct shrinker mmu_shrinker = {
2844 .shrink = mmu_shrink,
2845 .seeks = DEFAULT_SEEKS * 10,
2848 static void mmu_destroy_caches(void)
2850 if (pte_chain_cache)
2851 kmem_cache_destroy(pte_chain_cache);
2852 if (rmap_desc_cache)
2853 kmem_cache_destroy(rmap_desc_cache);
2854 if (mmu_page_header_cache)
2855 kmem_cache_destroy(mmu_page_header_cache);
2858 void kvm_mmu_module_exit(void)
2860 mmu_destroy_caches();
2861 unregister_shrinker(&mmu_shrinker);
2864 int kvm_mmu_module_init(void)
2866 pte_chain_cache = kmem_cache_create("kvm_pte_chain",
2867 sizeof(struct kvm_pte_chain),
2869 if (!pte_chain_cache)
2871 rmap_desc_cache = kmem_cache_create("kvm_rmap_desc",
2872 sizeof(struct kvm_rmap_desc),
2874 if (!rmap_desc_cache)
2877 mmu_page_header_cache = kmem_cache_create("kvm_mmu_page_header",
2878 sizeof(struct kvm_mmu_page),
2880 if (!mmu_page_header_cache)
2883 register_shrinker(&mmu_shrinker);
2888 mmu_destroy_caches();
2893 * Caculate mmu pages needed for kvm.
2895 unsigned int kvm_mmu_calculate_mmu_pages(struct kvm *kvm)
2898 unsigned int nr_mmu_pages;
2899 unsigned int nr_pages = 0;
2901 for (i = 0; i < kvm->nmemslots; i++)
2902 nr_pages += kvm->memslots[i].npages;
2904 nr_mmu_pages = nr_pages * KVM_PERMILLE_MMU_PAGES / 1000;
2905 nr_mmu_pages = max(nr_mmu_pages,
2906 (unsigned int) KVM_MIN_ALLOC_MMU_PAGES);
2908 return nr_mmu_pages;
2911 static void *pv_mmu_peek_buffer(struct kvm_pv_mmu_op_buffer *buffer,
2914 if (len > buffer->len)
2919 static void *pv_mmu_read_buffer(struct kvm_pv_mmu_op_buffer *buffer,
2924 ret = pv_mmu_peek_buffer(buffer, len);
2929 buffer->processed += len;
2933 static int kvm_pv_mmu_write(struct kvm_vcpu *vcpu,
2934 gpa_t addr, gpa_t value)
2939 if (!is_long_mode(vcpu) && !is_pae(vcpu))
2942 r = mmu_topup_memory_caches(vcpu);
2946 if (!emulator_write_phys(vcpu, addr, &value, bytes))
2952 static int kvm_pv_mmu_flush_tlb(struct kvm_vcpu *vcpu)
2954 kvm_set_cr3(vcpu, vcpu->arch.cr3);
2958 static int kvm_pv_mmu_release_pt(struct kvm_vcpu *vcpu, gpa_t addr)
2960 spin_lock(&vcpu->kvm->mmu_lock);
2961 mmu_unshadow(vcpu->kvm, addr >> PAGE_SHIFT);
2962 spin_unlock(&vcpu->kvm->mmu_lock);
2966 static int kvm_pv_mmu_op_one(struct kvm_vcpu *vcpu,
2967 struct kvm_pv_mmu_op_buffer *buffer)
2969 struct kvm_mmu_op_header *header;
2971 header = pv_mmu_peek_buffer(buffer, sizeof *header);
2974 switch (header->op) {
2975 case KVM_MMU_OP_WRITE_PTE: {
2976 struct kvm_mmu_op_write_pte *wpte;
2978 wpte = pv_mmu_read_buffer(buffer, sizeof *wpte);
2981 return kvm_pv_mmu_write(vcpu, wpte->pte_phys,
2984 case KVM_MMU_OP_FLUSH_TLB: {
2985 struct kvm_mmu_op_flush_tlb *ftlb;
2987 ftlb = pv_mmu_read_buffer(buffer, sizeof *ftlb);
2990 return kvm_pv_mmu_flush_tlb(vcpu);
2992 case KVM_MMU_OP_RELEASE_PT: {
2993 struct kvm_mmu_op_release_pt *rpt;
2995 rpt = pv_mmu_read_buffer(buffer, sizeof *rpt);
2998 return kvm_pv_mmu_release_pt(vcpu, rpt->pt_phys);
3004 int kvm_pv_mmu_op(struct kvm_vcpu *vcpu, unsigned long bytes,
3005 gpa_t addr, unsigned long *ret)
3008 struct kvm_pv_mmu_op_buffer *buffer = &vcpu->arch.mmu_op_buffer;
3010 buffer->ptr = buffer->buf;
3011 buffer->len = min_t(unsigned long, bytes, sizeof buffer->buf);
3012 buffer->processed = 0;
3014 r = kvm_read_guest(vcpu->kvm, addr, buffer->buf, buffer->len);
3018 while (buffer->len) {
3019 r = kvm_pv_mmu_op_one(vcpu, buffer);
3028 *ret = buffer->processed;
3034 static const char *audit_msg;
3036 static gva_t canonicalize(gva_t gva)
3038 #ifdef CONFIG_X86_64
3039 gva = (long long)(gva << 16) >> 16;
3044 static void audit_mappings_page(struct kvm_vcpu *vcpu, u64 page_pte,
3045 gva_t va, int level)
3047 u64 *pt = __va(page_pte & PT64_BASE_ADDR_MASK);
3049 gva_t va_delta = 1ul << (PAGE_SHIFT + 9 * (level - 1));
3051 for (i = 0; i < PT64_ENT_PER_PAGE; ++i, va += va_delta) {
3054 if (ent == shadow_trap_nonpresent_pte)
3057 va = canonicalize(va);
3059 if (ent == shadow_notrap_nonpresent_pte)
3060 printk(KERN_ERR "audit: (%s) nontrapping pte"
3061 " in nonleaf level: levels %d gva %lx"
3062 " level %d pte %llx\n", audit_msg,
3063 vcpu->arch.mmu.root_level, va, level, ent);
3065 audit_mappings_page(vcpu, ent, va, level - 1);
3067 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, va);
3068 gfn_t gfn = gpa >> PAGE_SHIFT;
3069 pfn_t pfn = gfn_to_pfn(vcpu->kvm, gfn);
3070 hpa_t hpa = (hpa_t)pfn << PAGE_SHIFT;
3072 if (is_shadow_present_pte(ent)
3073 && (ent & PT64_BASE_ADDR_MASK) != hpa)
3074 printk(KERN_ERR "xx audit error: (%s) levels %d"
3075 " gva %lx gpa %llx hpa %llx ent %llx %d\n",
3076 audit_msg, vcpu->arch.mmu.root_level,
3078 is_shadow_present_pte(ent));
3079 else if (ent == shadow_notrap_nonpresent_pte
3080 && !is_error_hpa(hpa))
3081 printk(KERN_ERR "audit: (%s) notrap shadow,"
3082 " valid guest gva %lx\n", audit_msg, va);
3083 kvm_release_pfn_clean(pfn);
3089 static void audit_mappings(struct kvm_vcpu *vcpu)
3093 if (vcpu->arch.mmu.root_level == 4)
3094 audit_mappings_page(vcpu, vcpu->arch.mmu.root_hpa, 0, 4);
3096 for (i = 0; i < 4; ++i)
3097 if (vcpu->arch.mmu.pae_root[i] & PT_PRESENT_MASK)
3098 audit_mappings_page(vcpu,
3099 vcpu->arch.mmu.pae_root[i],
3104 static int count_rmaps(struct kvm_vcpu *vcpu)
3109 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
3110 struct kvm_memory_slot *m = &vcpu->kvm->memslots[i];
3111 struct kvm_rmap_desc *d;
3113 for (j = 0; j < m->npages; ++j) {
3114 unsigned long *rmapp = &m->rmap[j];
3118 if (!(*rmapp & 1)) {
3122 d = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
3124 for (k = 0; k < RMAP_EXT; ++k)
3125 if (d->shadow_ptes[k])
3136 static int count_writable_mappings(struct kvm_vcpu *vcpu)
3139 struct kvm_mmu_page *sp;
3142 list_for_each_entry(sp, &vcpu->kvm->arch.active_mmu_pages, link) {
3145 if (sp->role.level != PT_PAGE_TABLE_LEVEL)
3148 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
3151 if (!(ent & PT_PRESENT_MASK))
3153 if (!(ent & PT_WRITABLE_MASK))
3161 static void audit_rmap(struct kvm_vcpu *vcpu)
3163 int n_rmap = count_rmaps(vcpu);
3164 int n_actual = count_writable_mappings(vcpu);
3166 if (n_rmap != n_actual)
3167 printk(KERN_ERR "%s: (%s) rmap %d actual %d\n",
3168 __func__, audit_msg, n_rmap, n_actual);
3171 static void audit_write_protection(struct kvm_vcpu *vcpu)
3173 struct kvm_mmu_page *sp;
3174 struct kvm_memory_slot *slot;
3175 unsigned long *rmapp;
3178 list_for_each_entry(sp, &vcpu->kvm->arch.active_mmu_pages, link) {
3179 if (sp->role.direct)
3182 gfn = unalias_gfn(vcpu->kvm, sp->gfn);
3183 slot = gfn_to_memslot_unaliased(vcpu->kvm, sp->gfn);
3184 rmapp = &slot->rmap[gfn - slot->base_gfn];
3186 printk(KERN_ERR "%s: (%s) shadow page has writable"
3187 " mappings: gfn %lx role %x\n",
3188 __func__, audit_msg, sp->gfn,
3193 static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg)
3200 audit_write_protection(vcpu);
3201 audit_mappings(vcpu);