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)
73 #define ASSERT(x) do { } while (0)
77 printk(KERN_WARNING "assertion failed %s:%d: %s\n", \
78 __FILE__, __LINE__, #x); \
82 #define PT64_PT_BITS 9
83 #define PT64_ENT_PER_PAGE (1 << PT64_PT_BITS)
84 #define PT32_PT_BITS 10
85 #define PT32_ENT_PER_PAGE (1 << PT32_PT_BITS)
87 #define PT_WRITABLE_SHIFT 1
89 #define PT_PRESENT_MASK (1ULL << 0)
90 #define PT_WRITABLE_MASK (1ULL << PT_WRITABLE_SHIFT)
91 #define PT_USER_MASK (1ULL << 2)
92 #define PT_PWT_MASK (1ULL << 3)
93 #define PT_PCD_MASK (1ULL << 4)
94 #define PT_ACCESSED_MASK (1ULL << 5)
95 #define PT_DIRTY_MASK (1ULL << 6)
96 #define PT_PAGE_SIZE_MASK (1ULL << 7)
97 #define PT_PAT_MASK (1ULL << 7)
98 #define PT_GLOBAL_MASK (1ULL << 8)
99 #define PT64_NX_SHIFT 63
100 #define PT64_NX_MASK (1ULL << PT64_NX_SHIFT)
102 #define PT_PAT_SHIFT 7
103 #define PT_DIR_PAT_SHIFT 12
104 #define PT_DIR_PAT_MASK (1ULL << PT_DIR_PAT_SHIFT)
106 #define PT32_DIR_PSE36_SIZE 4
107 #define PT32_DIR_PSE36_SHIFT 13
108 #define PT32_DIR_PSE36_MASK \
109 (((1ULL << PT32_DIR_PSE36_SIZE) - 1) << PT32_DIR_PSE36_SHIFT)
112 #define PT_FIRST_AVAIL_BITS_SHIFT 9
113 #define PT64_SECOND_AVAIL_BITS_SHIFT 52
115 #define VALID_PAGE(x) ((x) != INVALID_PAGE)
117 #define PT64_LEVEL_BITS 9
119 #define PT64_LEVEL_SHIFT(level) \
120 (PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS)
122 #define PT64_LEVEL_MASK(level) \
123 (((1ULL << PT64_LEVEL_BITS) - 1) << PT64_LEVEL_SHIFT(level))
125 #define PT64_INDEX(address, level)\
126 (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1))
129 #define PT32_LEVEL_BITS 10
131 #define PT32_LEVEL_SHIFT(level) \
132 (PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS)
134 #define PT32_LEVEL_MASK(level) \
135 (((1ULL << PT32_LEVEL_BITS) - 1) << PT32_LEVEL_SHIFT(level))
137 #define PT32_INDEX(address, level)\
138 (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1))
141 #define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1))
142 #define PT64_DIR_BASE_ADDR_MASK \
143 (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1))
145 #define PT32_BASE_ADDR_MASK PAGE_MASK
146 #define PT32_DIR_BASE_ADDR_MASK \
147 (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1))
149 #define PT64_PERM_MASK (PT_PRESENT_MASK | PT_WRITABLE_MASK | PT_USER_MASK \
152 #define PFERR_PRESENT_MASK (1U << 0)
153 #define PFERR_WRITE_MASK (1U << 1)
154 #define PFERR_USER_MASK (1U << 2)
155 #define PFERR_FETCH_MASK (1U << 4)
157 #define PT64_ROOT_LEVEL 4
158 #define PT32_ROOT_LEVEL 2
159 #define PT32E_ROOT_LEVEL 3
161 #define PT_DIRECTORY_LEVEL 2
162 #define PT_PAGE_TABLE_LEVEL 1
166 #define ACC_EXEC_MASK 1
167 #define ACC_WRITE_MASK PT_WRITABLE_MASK
168 #define ACC_USER_MASK PT_USER_MASK
169 #define ACC_ALL (ACC_EXEC_MASK | ACC_WRITE_MASK | ACC_USER_MASK)
171 struct kvm_pv_mmu_op_buffer
{
175 char buf
[512] __aligned(sizeof(long));
178 struct kvm_rmap_desc
{
179 u64
*shadow_ptes
[RMAP_EXT
];
180 struct kvm_rmap_desc
*more
;
183 static struct kmem_cache
*pte_chain_cache
;
184 static struct kmem_cache
*rmap_desc_cache
;
185 static struct kmem_cache
*mmu_page_header_cache
;
187 static u64 __read_mostly shadow_trap_nonpresent_pte
;
188 static u64 __read_mostly shadow_notrap_nonpresent_pte
;
190 void kvm_mmu_set_nonpresent_ptes(u64 trap_pte
, u64 notrap_pte
)
192 shadow_trap_nonpresent_pte
= trap_pte
;
193 shadow_notrap_nonpresent_pte
= notrap_pte
;
195 EXPORT_SYMBOL_GPL(kvm_mmu_set_nonpresent_ptes
);
197 static int is_write_protection(struct kvm_vcpu
*vcpu
)
199 return vcpu
->arch
.cr0
& X86_CR0_WP
;
202 static int is_cpuid_PSE36(void)
207 static int is_nx(struct kvm_vcpu
*vcpu
)
209 return vcpu
->arch
.shadow_efer
& EFER_NX
;
212 static int is_present_pte(unsigned long pte
)
214 return pte
& PT_PRESENT_MASK
;
217 static int is_shadow_present_pte(u64 pte
)
219 return pte
!= shadow_trap_nonpresent_pte
220 && pte
!= shadow_notrap_nonpresent_pte
;
223 static int is_large_pte(u64 pte
)
225 return pte
& PT_PAGE_SIZE_MASK
;
228 static int is_writeble_pte(unsigned long pte
)
230 return pte
& PT_WRITABLE_MASK
;
233 static int is_dirty_pte(unsigned long pte
)
235 return pte
& PT_DIRTY_MASK
;
238 static int is_rmap_pte(u64 pte
)
240 return is_shadow_present_pte(pte
);
243 static struct page
*spte_to_page(u64 pte
)
245 hfn_t hfn
= (pte
& PT64_BASE_ADDR_MASK
) >> PAGE_SHIFT
;
247 return pfn_to_page(hfn
);
250 static gfn_t
pse36_gfn_delta(u32 gpte
)
252 int shift
= 32 - PT32_DIR_PSE36_SHIFT
- PAGE_SHIFT
;
254 return (gpte
& PT32_DIR_PSE36_MASK
) << shift
;
257 static void set_shadow_pte(u64
*sptep
, u64 spte
)
260 set_64bit((unsigned long *)sptep
, spte
);
262 set_64bit((unsigned long long *)sptep
, spte
);
266 static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache
*cache
,
267 struct kmem_cache
*base_cache
, int min
)
271 if (cache
->nobjs
>= min
)
273 while (cache
->nobjs
< ARRAY_SIZE(cache
->objects
)) {
274 obj
= kmem_cache_zalloc(base_cache
, GFP_KERNEL
);
277 cache
->objects
[cache
->nobjs
++] = obj
;
282 static void mmu_free_memory_cache(struct kvm_mmu_memory_cache
*mc
)
285 kfree(mc
->objects
[--mc
->nobjs
]);
288 static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache
*cache
,
293 if (cache
->nobjs
>= min
)
295 while (cache
->nobjs
< ARRAY_SIZE(cache
->objects
)) {
296 page
= alloc_page(GFP_KERNEL
);
299 set_page_private(page
, 0);
300 cache
->objects
[cache
->nobjs
++] = page_address(page
);
305 static void mmu_free_memory_cache_page(struct kvm_mmu_memory_cache
*mc
)
308 free_page((unsigned long)mc
->objects
[--mc
->nobjs
]);
311 static int mmu_topup_memory_caches(struct kvm_vcpu
*vcpu
)
315 r
= mmu_topup_memory_cache(&vcpu
->arch
.mmu_pte_chain_cache
,
319 r
= mmu_topup_memory_cache(&vcpu
->arch
.mmu_rmap_desc_cache
,
323 r
= mmu_topup_memory_cache_page(&vcpu
->arch
.mmu_page_cache
, 8);
326 r
= mmu_topup_memory_cache(&vcpu
->arch
.mmu_page_header_cache
,
327 mmu_page_header_cache
, 4);
332 static void mmu_free_memory_caches(struct kvm_vcpu
*vcpu
)
334 mmu_free_memory_cache(&vcpu
->arch
.mmu_pte_chain_cache
);
335 mmu_free_memory_cache(&vcpu
->arch
.mmu_rmap_desc_cache
);
336 mmu_free_memory_cache_page(&vcpu
->arch
.mmu_page_cache
);
337 mmu_free_memory_cache(&vcpu
->arch
.mmu_page_header_cache
);
340 static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache
*mc
,
346 p
= mc
->objects
[--mc
->nobjs
];
351 static struct kvm_pte_chain
*mmu_alloc_pte_chain(struct kvm_vcpu
*vcpu
)
353 return mmu_memory_cache_alloc(&vcpu
->arch
.mmu_pte_chain_cache
,
354 sizeof(struct kvm_pte_chain
));
357 static void mmu_free_pte_chain(struct kvm_pte_chain
*pc
)
362 static struct kvm_rmap_desc
*mmu_alloc_rmap_desc(struct kvm_vcpu
*vcpu
)
364 return mmu_memory_cache_alloc(&vcpu
->arch
.mmu_rmap_desc_cache
,
365 sizeof(struct kvm_rmap_desc
));
368 static void mmu_free_rmap_desc(struct kvm_rmap_desc
*rd
)
374 * Return the pointer to the largepage write count for a given
375 * gfn, handling slots that are not large page aligned.
377 static int *slot_largepage_idx(gfn_t gfn
, struct kvm_memory_slot
*slot
)
381 idx
= (gfn
/ KVM_PAGES_PER_HPAGE
) -
382 (slot
->base_gfn
/ KVM_PAGES_PER_HPAGE
);
383 return &slot
->lpage_info
[idx
].write_count
;
386 static void account_shadowed(struct kvm
*kvm
, gfn_t gfn
)
390 write_count
= slot_largepage_idx(gfn
, gfn_to_memslot(kvm
, gfn
));
392 WARN_ON(*write_count
> KVM_PAGES_PER_HPAGE
);
395 static void unaccount_shadowed(struct kvm
*kvm
, gfn_t gfn
)
399 write_count
= slot_largepage_idx(gfn
, gfn_to_memslot(kvm
, gfn
));
401 WARN_ON(*write_count
< 0);
404 static int has_wrprotected_page(struct kvm
*kvm
, gfn_t gfn
)
406 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
410 largepage_idx
= slot_largepage_idx(gfn
, slot
);
411 return *largepage_idx
;
417 static int host_largepage_backed(struct kvm
*kvm
, gfn_t gfn
)
419 struct vm_area_struct
*vma
;
422 addr
= gfn_to_hva(kvm
, gfn
);
423 if (kvm_is_error_hva(addr
))
426 vma
= find_vma(current
->mm
, addr
);
427 if (vma
&& is_vm_hugetlb_page(vma
))
433 static int is_largepage_backed(struct kvm_vcpu
*vcpu
, gfn_t large_gfn
)
435 struct kvm_memory_slot
*slot
;
437 if (has_wrprotected_page(vcpu
->kvm
, large_gfn
))
440 if (!host_largepage_backed(vcpu
->kvm
, large_gfn
))
443 slot
= gfn_to_memslot(vcpu
->kvm
, large_gfn
);
444 if (slot
&& slot
->dirty_bitmap
)
451 * Take gfn and return the reverse mapping to it.
452 * Note: gfn must be unaliased before this function get called
455 static unsigned long *gfn_to_rmap(struct kvm
*kvm
, gfn_t gfn
, int lpage
)
457 struct kvm_memory_slot
*slot
;
460 slot
= gfn_to_memslot(kvm
, gfn
);
462 return &slot
->rmap
[gfn
- slot
->base_gfn
];
464 idx
= (gfn
/ KVM_PAGES_PER_HPAGE
) -
465 (slot
->base_gfn
/ KVM_PAGES_PER_HPAGE
);
467 return &slot
->lpage_info
[idx
].rmap_pde
;
471 * Reverse mapping data structures:
473 * If rmapp bit zero is zero, then rmapp point to the shadw page table entry
474 * that points to page_address(page).
476 * If rmapp bit zero is one, (then rmap & ~1) points to a struct kvm_rmap_desc
477 * containing more mappings.
479 static void rmap_add(struct kvm_vcpu
*vcpu
, u64
*spte
, gfn_t gfn
, int lpage
)
481 struct kvm_mmu_page
*sp
;
482 struct kvm_rmap_desc
*desc
;
483 unsigned long *rmapp
;
486 if (!is_rmap_pte(*spte
))
488 gfn
= unalias_gfn(vcpu
->kvm
, gfn
);
489 sp
= page_header(__pa(spte
));
490 sp
->gfns
[spte
- sp
->spt
] = gfn
;
491 rmapp
= gfn_to_rmap(vcpu
->kvm
, gfn
, lpage
);
493 rmap_printk("rmap_add: %p %llx 0->1\n", spte
, *spte
);
494 *rmapp
= (unsigned long)spte
;
495 } else if (!(*rmapp
& 1)) {
496 rmap_printk("rmap_add: %p %llx 1->many\n", spte
, *spte
);
497 desc
= mmu_alloc_rmap_desc(vcpu
);
498 desc
->shadow_ptes
[0] = (u64
*)*rmapp
;
499 desc
->shadow_ptes
[1] = spte
;
500 *rmapp
= (unsigned long)desc
| 1;
502 rmap_printk("rmap_add: %p %llx many->many\n", spte
, *spte
);
503 desc
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
504 while (desc
->shadow_ptes
[RMAP_EXT
-1] && desc
->more
)
506 if (desc
->shadow_ptes
[RMAP_EXT
-1]) {
507 desc
->more
= mmu_alloc_rmap_desc(vcpu
);
510 for (i
= 0; desc
->shadow_ptes
[i
]; ++i
)
512 desc
->shadow_ptes
[i
] = spte
;
516 static void rmap_desc_remove_entry(unsigned long *rmapp
,
517 struct kvm_rmap_desc
*desc
,
519 struct kvm_rmap_desc
*prev_desc
)
523 for (j
= RMAP_EXT
- 1; !desc
->shadow_ptes
[j
] && j
> i
; --j
)
525 desc
->shadow_ptes
[i
] = desc
->shadow_ptes
[j
];
526 desc
->shadow_ptes
[j
] = NULL
;
529 if (!prev_desc
&& !desc
->more
)
530 *rmapp
= (unsigned long)desc
->shadow_ptes
[0];
533 prev_desc
->more
= desc
->more
;
535 *rmapp
= (unsigned long)desc
->more
| 1;
536 mmu_free_rmap_desc(desc
);
539 static void rmap_remove(struct kvm
*kvm
, u64
*spte
)
541 struct kvm_rmap_desc
*desc
;
542 struct kvm_rmap_desc
*prev_desc
;
543 struct kvm_mmu_page
*sp
;
545 unsigned long *rmapp
;
548 if (!is_rmap_pte(*spte
))
550 sp
= page_header(__pa(spte
));
551 page
= spte_to_page(*spte
);
552 if (*spte
& PT_ACCESSED_MASK
)
553 mark_page_accessed(page
);
554 if (is_writeble_pte(*spte
))
555 kvm_release_page_dirty(page
);
557 kvm_release_page_clean(page
);
558 rmapp
= gfn_to_rmap(kvm
, sp
->gfns
[spte
- sp
->spt
], is_large_pte(*spte
));
560 printk(KERN_ERR
"rmap_remove: %p %llx 0->BUG\n", spte
, *spte
);
562 } else if (!(*rmapp
& 1)) {
563 rmap_printk("rmap_remove: %p %llx 1->0\n", spte
, *spte
);
564 if ((u64
*)*rmapp
!= spte
) {
565 printk(KERN_ERR
"rmap_remove: %p %llx 1->BUG\n",
571 rmap_printk("rmap_remove: %p %llx many->many\n", spte
, *spte
);
572 desc
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
575 for (i
= 0; i
< RMAP_EXT
&& desc
->shadow_ptes
[i
]; ++i
)
576 if (desc
->shadow_ptes
[i
] == spte
) {
577 rmap_desc_remove_entry(rmapp
,
589 static u64
*rmap_next(struct kvm
*kvm
, unsigned long *rmapp
, u64
*spte
)
591 struct kvm_rmap_desc
*desc
;
592 struct kvm_rmap_desc
*prev_desc
;
598 else if (!(*rmapp
& 1)) {
600 return (u64
*)*rmapp
;
603 desc
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
607 for (i
= 0; i
< RMAP_EXT
&& desc
->shadow_ptes
[i
]; ++i
) {
608 if (prev_spte
== spte
)
609 return desc
->shadow_ptes
[i
];
610 prev_spte
= desc
->shadow_ptes
[i
];
617 static void rmap_write_protect(struct kvm
*kvm
, u64 gfn
)
619 unsigned long *rmapp
;
621 int write_protected
= 0;
623 gfn
= unalias_gfn(kvm
, gfn
);
624 rmapp
= gfn_to_rmap(kvm
, gfn
, 0);
626 spte
= rmap_next(kvm
, rmapp
, NULL
);
629 BUG_ON(!(*spte
& PT_PRESENT_MASK
));
630 rmap_printk("rmap_write_protect: spte %p %llx\n", spte
, *spte
);
631 if (is_writeble_pte(*spte
)) {
632 set_shadow_pte(spte
, *spte
& ~PT_WRITABLE_MASK
);
635 spte
= rmap_next(kvm
, rmapp
, spte
);
637 if (write_protected
) {
640 spte
= rmap_next(kvm
, rmapp
, NULL
);
641 page
= spte_to_page(*spte
);
645 /* check for huge page mappings */
646 rmapp
= gfn_to_rmap(kvm
, gfn
, 1);
647 spte
= rmap_next(kvm
, rmapp
, NULL
);
650 BUG_ON(!(*spte
& PT_PRESENT_MASK
));
651 BUG_ON((*spte
& (PT_PAGE_SIZE_MASK
|PT_PRESENT_MASK
)) != (PT_PAGE_SIZE_MASK
|PT_PRESENT_MASK
));
652 pgprintk("rmap_write_protect(large): spte %p %llx %lld\n", spte
, *spte
, gfn
);
653 if (is_writeble_pte(*spte
)) {
654 rmap_remove(kvm
, spte
);
656 set_shadow_pte(spte
, shadow_trap_nonpresent_pte
);
659 spte
= rmap_next(kvm
, rmapp
, spte
);
663 kvm_flush_remote_tlbs(kvm
);
665 account_shadowed(kvm
, gfn
);
669 static int is_empty_shadow_page(u64
*spt
)
674 for (pos
= spt
, end
= pos
+ PAGE_SIZE
/ sizeof(u64
); pos
!= end
; pos
++)
675 if (*pos
!= shadow_trap_nonpresent_pte
) {
676 printk(KERN_ERR
"%s: %p %llx\n", __func__
,
684 static void kvm_mmu_free_page(struct kvm
*kvm
, struct kvm_mmu_page
*sp
)
686 ASSERT(is_empty_shadow_page(sp
->spt
));
688 __free_page(virt_to_page(sp
->spt
));
689 __free_page(virt_to_page(sp
->gfns
));
691 ++kvm
->arch
.n_free_mmu_pages
;
694 static unsigned kvm_page_table_hashfn(gfn_t gfn
)
696 return gfn
& ((1 << KVM_MMU_HASH_SHIFT
) - 1);
699 static struct kvm_mmu_page
*kvm_mmu_alloc_page(struct kvm_vcpu
*vcpu
,
702 struct kvm_mmu_page
*sp
;
704 sp
= mmu_memory_cache_alloc(&vcpu
->arch
.mmu_page_header_cache
, sizeof *sp
);
705 sp
->spt
= mmu_memory_cache_alloc(&vcpu
->arch
.mmu_page_cache
, PAGE_SIZE
);
706 sp
->gfns
= mmu_memory_cache_alloc(&vcpu
->arch
.mmu_page_cache
, PAGE_SIZE
);
707 set_page_private(virt_to_page(sp
->spt
), (unsigned long)sp
);
708 list_add(&sp
->link
, &vcpu
->kvm
->arch
.active_mmu_pages
);
709 ASSERT(is_empty_shadow_page(sp
->spt
));
712 sp
->parent_pte
= parent_pte
;
713 --vcpu
->kvm
->arch
.n_free_mmu_pages
;
717 static void mmu_page_add_parent_pte(struct kvm_vcpu
*vcpu
,
718 struct kvm_mmu_page
*sp
, u64
*parent_pte
)
720 struct kvm_pte_chain
*pte_chain
;
721 struct hlist_node
*node
;
726 if (!sp
->multimapped
) {
727 u64
*old
= sp
->parent_pte
;
730 sp
->parent_pte
= parent_pte
;
734 pte_chain
= mmu_alloc_pte_chain(vcpu
);
735 INIT_HLIST_HEAD(&sp
->parent_ptes
);
736 hlist_add_head(&pte_chain
->link
, &sp
->parent_ptes
);
737 pte_chain
->parent_ptes
[0] = old
;
739 hlist_for_each_entry(pte_chain
, node
, &sp
->parent_ptes
, link
) {
740 if (pte_chain
->parent_ptes
[NR_PTE_CHAIN_ENTRIES
-1])
742 for (i
= 0; i
< NR_PTE_CHAIN_ENTRIES
; ++i
)
743 if (!pte_chain
->parent_ptes
[i
]) {
744 pte_chain
->parent_ptes
[i
] = parent_pte
;
748 pte_chain
= mmu_alloc_pte_chain(vcpu
);
750 hlist_add_head(&pte_chain
->link
, &sp
->parent_ptes
);
751 pte_chain
->parent_ptes
[0] = parent_pte
;
754 static void mmu_page_remove_parent_pte(struct kvm_mmu_page
*sp
,
757 struct kvm_pte_chain
*pte_chain
;
758 struct hlist_node
*node
;
761 if (!sp
->multimapped
) {
762 BUG_ON(sp
->parent_pte
!= parent_pte
);
763 sp
->parent_pte
= NULL
;
766 hlist_for_each_entry(pte_chain
, node
, &sp
->parent_ptes
, link
)
767 for (i
= 0; i
< NR_PTE_CHAIN_ENTRIES
; ++i
) {
768 if (!pte_chain
->parent_ptes
[i
])
770 if (pte_chain
->parent_ptes
[i
] != parent_pte
)
772 while (i
+ 1 < NR_PTE_CHAIN_ENTRIES
773 && pte_chain
->parent_ptes
[i
+ 1]) {
774 pte_chain
->parent_ptes
[i
]
775 = pte_chain
->parent_ptes
[i
+ 1];
778 pte_chain
->parent_ptes
[i
] = NULL
;
780 hlist_del(&pte_chain
->link
);
781 mmu_free_pte_chain(pte_chain
);
782 if (hlist_empty(&sp
->parent_ptes
)) {
784 sp
->parent_pte
= NULL
;
792 static struct kvm_mmu_page
*kvm_mmu_lookup_page(struct kvm
*kvm
, gfn_t gfn
)
795 struct hlist_head
*bucket
;
796 struct kvm_mmu_page
*sp
;
797 struct hlist_node
*node
;
799 pgprintk("%s: looking for gfn %lx\n", __func__
, gfn
);
800 index
= kvm_page_table_hashfn(gfn
);
801 bucket
= &kvm
->arch
.mmu_page_hash
[index
];
802 hlist_for_each_entry(sp
, node
, bucket
, hash_link
)
803 if (sp
->gfn
== gfn
&& !sp
->role
.metaphysical
804 && !sp
->role
.invalid
) {
805 pgprintk("%s: found role %x\n",
806 __func__
, sp
->role
.word
);
812 static struct kvm_mmu_page
*kvm_mmu_get_page(struct kvm_vcpu
*vcpu
,
820 union kvm_mmu_page_role role
;
823 struct hlist_head
*bucket
;
824 struct kvm_mmu_page
*sp
;
825 struct hlist_node
*node
;
828 role
.glevels
= vcpu
->arch
.mmu
.root_level
;
830 role
.metaphysical
= metaphysical
;
831 role
.access
= access
;
832 if (vcpu
->arch
.mmu
.root_level
<= PT32_ROOT_LEVEL
) {
833 quadrant
= gaddr
>> (PAGE_SHIFT
+ (PT64_PT_BITS
* level
));
834 quadrant
&= (1 << ((PT32_PT_BITS
- PT64_PT_BITS
) * level
)) - 1;
835 role
.quadrant
= quadrant
;
837 pgprintk("%s: looking gfn %lx role %x\n", __func__
,
839 index
= kvm_page_table_hashfn(gfn
);
840 bucket
= &vcpu
->kvm
->arch
.mmu_page_hash
[index
];
841 hlist_for_each_entry(sp
, node
, bucket
, hash_link
)
842 if (sp
->gfn
== gfn
&& sp
->role
.word
== role
.word
) {
843 mmu_page_add_parent_pte(vcpu
, sp
, parent_pte
);
844 pgprintk("%s: found\n", __func__
);
847 ++vcpu
->kvm
->stat
.mmu_cache_miss
;
848 sp
= kvm_mmu_alloc_page(vcpu
, parent_pte
);
851 pgprintk("%s: adding gfn %lx role %x\n", __func__
, gfn
, role
.word
);
854 hlist_add_head(&sp
->hash_link
, bucket
);
855 vcpu
->arch
.mmu
.prefetch_page(vcpu
, sp
);
857 rmap_write_protect(vcpu
->kvm
, gfn
);
861 static void kvm_mmu_page_unlink_children(struct kvm
*kvm
,
862 struct kvm_mmu_page
*sp
)
870 if (sp
->role
.level
== PT_PAGE_TABLE_LEVEL
) {
871 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
872 if (is_shadow_present_pte(pt
[i
]))
873 rmap_remove(kvm
, &pt
[i
]);
874 pt
[i
] = shadow_trap_nonpresent_pte
;
876 kvm_flush_remote_tlbs(kvm
);
880 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
883 if (is_shadow_present_pte(ent
)) {
884 if (!is_large_pte(ent
)) {
885 ent
&= PT64_BASE_ADDR_MASK
;
886 mmu_page_remove_parent_pte(page_header(ent
),
890 rmap_remove(kvm
, &pt
[i
]);
893 pt
[i
] = shadow_trap_nonpresent_pte
;
895 kvm_flush_remote_tlbs(kvm
);
898 static void kvm_mmu_put_page(struct kvm_mmu_page
*sp
, u64
*parent_pte
)
900 mmu_page_remove_parent_pte(sp
, parent_pte
);
903 static void kvm_mmu_reset_last_pte_updated(struct kvm
*kvm
)
907 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
)
909 kvm
->vcpus
[i
]->arch
.last_pte_updated
= NULL
;
912 static void kvm_mmu_zap_page(struct kvm
*kvm
, struct kvm_mmu_page
*sp
)
916 ++kvm
->stat
.mmu_shadow_zapped
;
917 while (sp
->multimapped
|| sp
->parent_pte
) {
918 if (!sp
->multimapped
)
919 parent_pte
= sp
->parent_pte
;
921 struct kvm_pte_chain
*chain
;
923 chain
= container_of(sp
->parent_ptes
.first
,
924 struct kvm_pte_chain
, link
);
925 parent_pte
= chain
->parent_ptes
[0];
928 kvm_mmu_put_page(sp
, parent_pte
);
929 set_shadow_pte(parent_pte
, shadow_trap_nonpresent_pte
);
931 kvm_mmu_page_unlink_children(kvm
, sp
);
932 if (!sp
->root_count
) {
933 if (!sp
->role
.metaphysical
)
934 unaccount_shadowed(kvm
, sp
->gfn
);
935 hlist_del(&sp
->hash_link
);
936 kvm_mmu_free_page(kvm
, sp
);
938 list_move(&sp
->link
, &kvm
->arch
.active_mmu_pages
);
939 sp
->role
.invalid
= 1;
940 kvm_reload_remote_mmus(kvm
);
942 kvm_mmu_reset_last_pte_updated(kvm
);
946 * Changing the number of mmu pages allocated to the vm
947 * Note: if kvm_nr_mmu_pages is too small, you will get dead lock
949 void kvm_mmu_change_mmu_pages(struct kvm
*kvm
, unsigned int kvm_nr_mmu_pages
)
952 * If we set the number of mmu pages to be smaller be than the
953 * number of actived pages , we must to free some mmu pages before we
957 if ((kvm
->arch
.n_alloc_mmu_pages
- kvm
->arch
.n_free_mmu_pages
) >
959 int n_used_mmu_pages
= kvm
->arch
.n_alloc_mmu_pages
960 - kvm
->arch
.n_free_mmu_pages
;
962 while (n_used_mmu_pages
> kvm_nr_mmu_pages
) {
963 struct kvm_mmu_page
*page
;
965 page
= container_of(kvm
->arch
.active_mmu_pages
.prev
,
966 struct kvm_mmu_page
, link
);
967 kvm_mmu_zap_page(kvm
, page
);
970 kvm
->arch
.n_free_mmu_pages
= 0;
973 kvm
->arch
.n_free_mmu_pages
+= kvm_nr_mmu_pages
974 - kvm
->arch
.n_alloc_mmu_pages
;
976 kvm
->arch
.n_alloc_mmu_pages
= kvm_nr_mmu_pages
;
979 static int kvm_mmu_unprotect_page(struct kvm
*kvm
, gfn_t gfn
)
982 struct hlist_head
*bucket
;
983 struct kvm_mmu_page
*sp
;
984 struct hlist_node
*node
, *n
;
987 pgprintk("%s: looking for gfn %lx\n", __func__
, gfn
);
989 index
= kvm_page_table_hashfn(gfn
);
990 bucket
= &kvm
->arch
.mmu_page_hash
[index
];
991 hlist_for_each_entry_safe(sp
, node
, n
, bucket
, hash_link
)
992 if (sp
->gfn
== gfn
&& !sp
->role
.metaphysical
) {
993 pgprintk("%s: gfn %lx role %x\n", __func__
, gfn
,
995 kvm_mmu_zap_page(kvm
, sp
);
1001 static void mmu_unshadow(struct kvm
*kvm
, gfn_t gfn
)
1003 struct kvm_mmu_page
*sp
;
1005 while ((sp
= kvm_mmu_lookup_page(kvm
, gfn
)) != NULL
) {
1006 pgprintk("%s: zap %lx %x\n", __func__
, gfn
, sp
->role
.word
);
1007 kvm_mmu_zap_page(kvm
, sp
);
1011 static void page_header_update_slot(struct kvm
*kvm
, void *pte
, gfn_t gfn
)
1013 int slot
= memslot_id(kvm
, gfn_to_memslot(kvm
, gfn
));
1014 struct kvm_mmu_page
*sp
= page_header(__pa(pte
));
1016 __set_bit(slot
, &sp
->slot_bitmap
);
1019 struct page
*gva_to_page(struct kvm_vcpu
*vcpu
, gva_t gva
)
1023 gpa_t gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, gva
);
1025 if (gpa
== UNMAPPED_GVA
)
1028 down_read(¤t
->mm
->mmap_sem
);
1029 page
= gfn_to_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
1030 up_read(¤t
->mm
->mmap_sem
);
1035 static void mmu_set_spte(struct kvm_vcpu
*vcpu
, u64
*shadow_pte
,
1036 unsigned pt_access
, unsigned pte_access
,
1037 int user_fault
, int write_fault
, int dirty
,
1038 int *ptwrite
, int largepage
, gfn_t gfn
,
1039 struct page
*page
, bool speculative
)
1042 int was_rmapped
= 0;
1043 int was_writeble
= is_writeble_pte(*shadow_pte
);
1045 pgprintk("%s: spte %llx access %x write_fault %d"
1046 " user_fault %d gfn %lx\n",
1047 __func__
, *shadow_pte
, pt_access
,
1048 write_fault
, user_fault
, gfn
);
1050 if (is_rmap_pte(*shadow_pte
)) {
1052 * If we overwrite a PTE page pointer with a 2MB PMD, unlink
1053 * the parent of the now unreachable PTE.
1055 if (largepage
&& !is_large_pte(*shadow_pte
)) {
1056 struct kvm_mmu_page
*child
;
1057 u64 pte
= *shadow_pte
;
1059 child
= page_header(pte
& PT64_BASE_ADDR_MASK
);
1060 mmu_page_remove_parent_pte(child
, shadow_pte
);
1061 } else if (page
!= spte_to_page(*shadow_pte
)) {
1062 pgprintk("hfn old %lx new %lx\n",
1063 page_to_pfn(spte_to_page(*shadow_pte
)),
1065 rmap_remove(vcpu
->kvm
, shadow_pte
);
1068 was_rmapped
= is_large_pte(*shadow_pte
);
1075 * We don't set the accessed bit, since we sometimes want to see
1076 * whether the guest actually used the pte (in order to detect
1079 spte
= PT_PRESENT_MASK
| PT_DIRTY_MASK
;
1081 pte_access
|= PT_ACCESSED_MASK
;
1083 pte_access
&= ~ACC_WRITE_MASK
;
1084 if (!(pte_access
& ACC_EXEC_MASK
))
1085 spte
|= PT64_NX_MASK
;
1087 spte
|= PT_PRESENT_MASK
;
1088 if (pte_access
& ACC_USER_MASK
)
1089 spte
|= PT_USER_MASK
;
1091 spte
|= PT_PAGE_SIZE_MASK
;
1093 spte
|= page_to_phys(page
);
1095 if ((pte_access
& ACC_WRITE_MASK
)
1096 || (write_fault
&& !is_write_protection(vcpu
) && !user_fault
)) {
1097 struct kvm_mmu_page
*shadow
;
1099 spte
|= PT_WRITABLE_MASK
;
1101 mmu_unshadow(vcpu
->kvm
, gfn
);
1105 shadow
= kvm_mmu_lookup_page(vcpu
->kvm
, gfn
);
1107 (largepage
&& has_wrprotected_page(vcpu
->kvm
, gfn
))) {
1108 pgprintk("%s: found shadow page for %lx, marking ro\n",
1110 pte_access
&= ~ACC_WRITE_MASK
;
1111 if (is_writeble_pte(spte
)) {
1112 spte
&= ~PT_WRITABLE_MASK
;
1113 kvm_x86_ops
->tlb_flush(vcpu
);
1122 if (pte_access
& ACC_WRITE_MASK
)
1123 mark_page_dirty(vcpu
->kvm
, gfn
);
1125 pgprintk("%s: setting spte %llx\n", __func__
, spte
);
1126 pgprintk("instantiating %s PTE (%s) at %d (%llx) addr %llx\n",
1127 (spte
&PT_PAGE_SIZE_MASK
)? "2MB" : "4kB",
1128 (spte
&PT_WRITABLE_MASK
)?"RW":"R", gfn
, spte
, shadow_pte
);
1129 set_shadow_pte(shadow_pte
, spte
);
1130 if (!was_rmapped
&& (spte
& PT_PAGE_SIZE_MASK
)
1131 && (spte
& PT_PRESENT_MASK
))
1132 ++vcpu
->kvm
->stat
.lpages
;
1134 page_header_update_slot(vcpu
->kvm
, shadow_pte
, gfn
);
1136 rmap_add(vcpu
, shadow_pte
, gfn
, largepage
);
1137 if (!is_rmap_pte(*shadow_pte
))
1138 kvm_release_page_clean(page
);
1141 kvm_release_page_dirty(page
);
1143 kvm_release_page_clean(page
);
1145 if (!ptwrite
|| !*ptwrite
)
1146 vcpu
->arch
.last_pte_updated
= shadow_pte
;
1149 static void nonpaging_new_cr3(struct kvm_vcpu
*vcpu
)
1153 static int __direct_map(struct kvm_vcpu
*vcpu
, gpa_t v
, int write
,
1154 int largepage
, gfn_t gfn
, struct page
*page
,
1157 hpa_t table_addr
= vcpu
->arch
.mmu
.root_hpa
;
1161 u32 index
= PT64_INDEX(v
, level
);
1164 ASSERT(VALID_PAGE(table_addr
));
1165 table
= __va(table_addr
);
1168 mmu_set_spte(vcpu
, &table
[index
], ACC_ALL
, ACC_ALL
,
1169 0, write
, 1, &pt_write
, 0, gfn
, page
, false);
1173 if (largepage
&& level
== 2) {
1174 mmu_set_spte(vcpu
, &table
[index
], ACC_ALL
, ACC_ALL
,
1175 0, write
, 1, &pt_write
, 1, gfn
, page
, false);
1179 if (table
[index
] == shadow_trap_nonpresent_pte
) {
1180 struct kvm_mmu_page
*new_table
;
1183 pseudo_gfn
= (v
& PT64_DIR_BASE_ADDR_MASK
)
1185 new_table
= kvm_mmu_get_page(vcpu
, pseudo_gfn
,
1187 1, ACC_ALL
, &table
[index
]);
1189 pgprintk("nonpaging_map: ENOMEM\n");
1190 kvm_release_page_clean(page
);
1194 table
[index
] = __pa(new_table
->spt
) | PT_PRESENT_MASK
1195 | PT_WRITABLE_MASK
| PT_USER_MASK
;
1197 table_addr
= table
[index
] & PT64_BASE_ADDR_MASK
;
1201 static int nonpaging_map(struct kvm_vcpu
*vcpu
, gva_t v
, int write
, gfn_t gfn
)
1208 down_read(¤t
->mm
->mmap_sem
);
1209 if (is_largepage_backed(vcpu
, gfn
& ~(KVM_PAGES_PER_HPAGE
-1))) {
1210 gfn
&= ~(KVM_PAGES_PER_HPAGE
-1);
1214 page
= gfn_to_page(vcpu
->kvm
, gfn
);
1215 up_read(¤t
->mm
->mmap_sem
);
1218 if (is_error_page(page
)) {
1219 kvm_release_page_clean(page
);
1223 spin_lock(&vcpu
->kvm
->mmu_lock
);
1224 kvm_mmu_free_some_pages(vcpu
);
1225 r
= __direct_map(vcpu
, v
, write
, largepage
, gfn
, page
,
1227 spin_unlock(&vcpu
->kvm
->mmu_lock
);
1234 static void nonpaging_prefetch_page(struct kvm_vcpu
*vcpu
,
1235 struct kvm_mmu_page
*sp
)
1239 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
)
1240 sp
->spt
[i
] = shadow_trap_nonpresent_pte
;
1243 static void mmu_free_roots(struct kvm_vcpu
*vcpu
)
1246 struct kvm_mmu_page
*sp
;
1248 if (!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
))
1250 spin_lock(&vcpu
->kvm
->mmu_lock
);
1251 #ifdef CONFIG_X86_64
1252 if (vcpu
->arch
.mmu
.shadow_root_level
== PT64_ROOT_LEVEL
) {
1253 hpa_t root
= vcpu
->arch
.mmu
.root_hpa
;
1255 sp
= page_header(root
);
1257 if (!sp
->root_count
&& sp
->role
.invalid
)
1258 kvm_mmu_zap_page(vcpu
->kvm
, sp
);
1259 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
1260 spin_unlock(&vcpu
->kvm
->mmu_lock
);
1264 for (i
= 0; i
< 4; ++i
) {
1265 hpa_t root
= vcpu
->arch
.mmu
.pae_root
[i
];
1268 root
&= PT64_BASE_ADDR_MASK
;
1269 sp
= page_header(root
);
1271 if (!sp
->root_count
&& sp
->role
.invalid
)
1272 kvm_mmu_zap_page(vcpu
->kvm
, sp
);
1274 vcpu
->arch
.mmu
.pae_root
[i
] = INVALID_PAGE
;
1276 spin_unlock(&vcpu
->kvm
->mmu_lock
);
1277 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
1280 static void mmu_alloc_roots(struct kvm_vcpu
*vcpu
)
1284 struct kvm_mmu_page
*sp
;
1285 int metaphysical
= 0;
1287 root_gfn
= vcpu
->arch
.cr3
>> PAGE_SHIFT
;
1289 #ifdef CONFIG_X86_64
1290 if (vcpu
->arch
.mmu
.shadow_root_level
== PT64_ROOT_LEVEL
) {
1291 hpa_t root
= vcpu
->arch
.mmu
.root_hpa
;
1293 ASSERT(!VALID_PAGE(root
));
1296 sp
= kvm_mmu_get_page(vcpu
, root_gfn
, 0,
1297 PT64_ROOT_LEVEL
, metaphysical
,
1299 root
= __pa(sp
->spt
);
1301 vcpu
->arch
.mmu
.root_hpa
= root
;
1305 metaphysical
= !is_paging(vcpu
);
1308 for (i
= 0; i
< 4; ++i
) {
1309 hpa_t root
= vcpu
->arch
.mmu
.pae_root
[i
];
1311 ASSERT(!VALID_PAGE(root
));
1312 if (vcpu
->arch
.mmu
.root_level
== PT32E_ROOT_LEVEL
) {
1313 if (!is_present_pte(vcpu
->arch
.pdptrs
[i
])) {
1314 vcpu
->arch
.mmu
.pae_root
[i
] = 0;
1317 root_gfn
= vcpu
->arch
.pdptrs
[i
] >> PAGE_SHIFT
;
1318 } else if (vcpu
->arch
.mmu
.root_level
== 0)
1320 sp
= kvm_mmu_get_page(vcpu
, root_gfn
, i
<< 30,
1321 PT32_ROOT_LEVEL
, metaphysical
,
1323 root
= __pa(sp
->spt
);
1325 vcpu
->arch
.mmu
.pae_root
[i
] = root
| PT_PRESENT_MASK
;
1327 vcpu
->arch
.mmu
.root_hpa
= __pa(vcpu
->arch
.mmu
.pae_root
);
1330 static gpa_t
nonpaging_gva_to_gpa(struct kvm_vcpu
*vcpu
, gva_t vaddr
)
1335 static int nonpaging_page_fault(struct kvm_vcpu
*vcpu
, gva_t gva
,
1341 pgprintk("%s: gva %lx error %x\n", __func__
, gva
, error_code
);
1342 r
= mmu_topup_memory_caches(vcpu
);
1347 ASSERT(VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
1349 gfn
= gva
>> PAGE_SHIFT
;
1351 return nonpaging_map(vcpu
, gva
& PAGE_MASK
,
1352 error_code
& PFERR_WRITE_MASK
, gfn
);
1355 static int tdp_page_fault(struct kvm_vcpu
*vcpu
, gva_t gpa
,
1361 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1364 ASSERT(VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
1366 r
= mmu_topup_memory_caches(vcpu
);
1370 down_read(¤t
->mm
->mmap_sem
);
1371 if (is_largepage_backed(vcpu
, gfn
& ~(KVM_PAGES_PER_HPAGE
-1))) {
1372 gfn
&= ~(KVM_PAGES_PER_HPAGE
-1);
1375 page
= gfn_to_page(vcpu
->kvm
, gfn
);
1376 up_read(¤t
->mm
->mmap_sem
);
1377 if (is_error_page(page
)) {
1378 kvm_release_page_clean(page
);
1381 spin_lock(&vcpu
->kvm
->mmu_lock
);
1382 kvm_mmu_free_some_pages(vcpu
);
1383 r
= __direct_map(vcpu
, gpa
, error_code
& PFERR_WRITE_MASK
,
1384 largepage
, gfn
, page
, TDP_ROOT_LEVEL
);
1385 spin_unlock(&vcpu
->kvm
->mmu_lock
);
1390 static void nonpaging_free(struct kvm_vcpu
*vcpu
)
1392 mmu_free_roots(vcpu
);
1395 static int nonpaging_init_context(struct kvm_vcpu
*vcpu
)
1397 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
1399 context
->new_cr3
= nonpaging_new_cr3
;
1400 context
->page_fault
= nonpaging_page_fault
;
1401 context
->gva_to_gpa
= nonpaging_gva_to_gpa
;
1402 context
->free
= nonpaging_free
;
1403 context
->prefetch_page
= nonpaging_prefetch_page
;
1404 context
->root_level
= 0;
1405 context
->shadow_root_level
= PT32E_ROOT_LEVEL
;
1406 context
->root_hpa
= INVALID_PAGE
;
1410 void kvm_mmu_flush_tlb(struct kvm_vcpu
*vcpu
)
1412 ++vcpu
->stat
.tlb_flush
;
1413 kvm_x86_ops
->tlb_flush(vcpu
);
1416 static void paging_new_cr3(struct kvm_vcpu
*vcpu
)
1418 pgprintk("%s: cr3 %lx\n", __func__
, vcpu
->arch
.cr3
);
1419 mmu_free_roots(vcpu
);
1422 static void inject_page_fault(struct kvm_vcpu
*vcpu
,
1426 kvm_inject_page_fault(vcpu
, addr
, err_code
);
1429 static void paging_free(struct kvm_vcpu
*vcpu
)
1431 nonpaging_free(vcpu
);
1435 #include "paging_tmpl.h"
1439 #include "paging_tmpl.h"
1442 static int paging64_init_context_common(struct kvm_vcpu
*vcpu
, int level
)
1444 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
1446 ASSERT(is_pae(vcpu
));
1447 context
->new_cr3
= paging_new_cr3
;
1448 context
->page_fault
= paging64_page_fault
;
1449 context
->gva_to_gpa
= paging64_gva_to_gpa
;
1450 context
->prefetch_page
= paging64_prefetch_page
;
1451 context
->free
= paging_free
;
1452 context
->root_level
= level
;
1453 context
->shadow_root_level
= level
;
1454 context
->root_hpa
= INVALID_PAGE
;
1458 static int paging64_init_context(struct kvm_vcpu
*vcpu
)
1460 return paging64_init_context_common(vcpu
, PT64_ROOT_LEVEL
);
1463 static int paging32_init_context(struct kvm_vcpu
*vcpu
)
1465 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
1467 context
->new_cr3
= paging_new_cr3
;
1468 context
->page_fault
= paging32_page_fault
;
1469 context
->gva_to_gpa
= paging32_gva_to_gpa
;
1470 context
->free
= paging_free
;
1471 context
->prefetch_page
= paging32_prefetch_page
;
1472 context
->root_level
= PT32_ROOT_LEVEL
;
1473 context
->shadow_root_level
= PT32E_ROOT_LEVEL
;
1474 context
->root_hpa
= INVALID_PAGE
;
1478 static int paging32E_init_context(struct kvm_vcpu
*vcpu
)
1480 return paging64_init_context_common(vcpu
, PT32E_ROOT_LEVEL
);
1483 static int init_kvm_tdp_mmu(struct kvm_vcpu
*vcpu
)
1485 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
1487 context
->new_cr3
= nonpaging_new_cr3
;
1488 context
->page_fault
= tdp_page_fault
;
1489 context
->free
= nonpaging_free
;
1490 context
->prefetch_page
= nonpaging_prefetch_page
;
1491 context
->shadow_root_level
= TDP_ROOT_LEVEL
;
1492 context
->root_hpa
= INVALID_PAGE
;
1494 if (!is_paging(vcpu
)) {
1495 context
->gva_to_gpa
= nonpaging_gva_to_gpa
;
1496 context
->root_level
= 0;
1497 } else if (is_long_mode(vcpu
)) {
1498 context
->gva_to_gpa
= paging64_gva_to_gpa
;
1499 context
->root_level
= PT64_ROOT_LEVEL
;
1500 } else if (is_pae(vcpu
)) {
1501 context
->gva_to_gpa
= paging64_gva_to_gpa
;
1502 context
->root_level
= PT32E_ROOT_LEVEL
;
1504 context
->gva_to_gpa
= paging32_gva_to_gpa
;
1505 context
->root_level
= PT32_ROOT_LEVEL
;
1511 static int init_kvm_softmmu(struct kvm_vcpu
*vcpu
)
1514 ASSERT(!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
1516 if (!is_paging(vcpu
))
1517 return nonpaging_init_context(vcpu
);
1518 else if (is_long_mode(vcpu
))
1519 return paging64_init_context(vcpu
);
1520 else if (is_pae(vcpu
))
1521 return paging32E_init_context(vcpu
);
1523 return paging32_init_context(vcpu
);
1526 static int init_kvm_mmu(struct kvm_vcpu
*vcpu
)
1529 return init_kvm_tdp_mmu(vcpu
);
1531 return init_kvm_softmmu(vcpu
);
1534 static void destroy_kvm_mmu(struct kvm_vcpu
*vcpu
)
1537 if (VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
)) {
1538 vcpu
->arch
.mmu
.free(vcpu
);
1539 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
1543 int kvm_mmu_reset_context(struct kvm_vcpu
*vcpu
)
1545 destroy_kvm_mmu(vcpu
);
1546 return init_kvm_mmu(vcpu
);
1548 EXPORT_SYMBOL_GPL(kvm_mmu_reset_context
);
1550 int kvm_mmu_load(struct kvm_vcpu
*vcpu
)
1554 r
= mmu_topup_memory_caches(vcpu
);
1557 spin_lock(&vcpu
->kvm
->mmu_lock
);
1558 kvm_mmu_free_some_pages(vcpu
);
1559 mmu_alloc_roots(vcpu
);
1560 spin_unlock(&vcpu
->kvm
->mmu_lock
);
1561 kvm_x86_ops
->set_cr3(vcpu
, vcpu
->arch
.mmu
.root_hpa
);
1562 kvm_mmu_flush_tlb(vcpu
);
1566 EXPORT_SYMBOL_GPL(kvm_mmu_load
);
1568 void kvm_mmu_unload(struct kvm_vcpu
*vcpu
)
1570 mmu_free_roots(vcpu
);
1573 static void mmu_pte_write_zap_pte(struct kvm_vcpu
*vcpu
,
1574 struct kvm_mmu_page
*sp
,
1578 struct kvm_mmu_page
*child
;
1581 if (is_shadow_present_pte(pte
)) {
1582 if (sp
->role
.level
== PT_PAGE_TABLE_LEVEL
||
1584 rmap_remove(vcpu
->kvm
, spte
);
1586 child
= page_header(pte
& PT64_BASE_ADDR_MASK
);
1587 mmu_page_remove_parent_pte(child
, spte
);
1590 set_shadow_pte(spte
, shadow_trap_nonpresent_pte
);
1591 if (is_large_pte(pte
))
1592 --vcpu
->kvm
->stat
.lpages
;
1595 static void mmu_pte_write_new_pte(struct kvm_vcpu
*vcpu
,
1596 struct kvm_mmu_page
*sp
,
1600 if ((sp
->role
.level
!= PT_PAGE_TABLE_LEVEL
)
1601 && !vcpu
->arch
.update_pte
.largepage
) {
1602 ++vcpu
->kvm
->stat
.mmu_pde_zapped
;
1606 ++vcpu
->kvm
->stat
.mmu_pte_updated
;
1607 if (sp
->role
.glevels
== PT32_ROOT_LEVEL
)
1608 paging32_update_pte(vcpu
, sp
, spte
, new);
1610 paging64_update_pte(vcpu
, sp
, spte
, new);
1613 static bool need_remote_flush(u64 old
, u64
new)
1615 if (!is_shadow_present_pte(old
))
1617 if (!is_shadow_present_pte(new))
1619 if ((old
^ new) & PT64_BASE_ADDR_MASK
)
1621 old
^= PT64_NX_MASK
;
1622 new ^= PT64_NX_MASK
;
1623 return (old
& ~new & PT64_PERM_MASK
) != 0;
1626 static void mmu_pte_write_flush_tlb(struct kvm_vcpu
*vcpu
, u64 old
, u64
new)
1628 if (need_remote_flush(old
, new))
1629 kvm_flush_remote_tlbs(vcpu
->kvm
);
1631 kvm_mmu_flush_tlb(vcpu
);
1634 static bool last_updated_pte_accessed(struct kvm_vcpu
*vcpu
)
1636 u64
*spte
= vcpu
->arch
.last_pte_updated
;
1638 return !!(spte
&& (*spte
& PT_ACCESSED_MASK
));
1641 static void mmu_guess_page_from_pte_write(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1642 const u8
*new, int bytes
)
1649 vcpu
->arch
.update_pte
.largepage
= 0;
1651 if (bytes
!= 4 && bytes
!= 8)
1655 * Assume that the pte write on a page table of the same type
1656 * as the current vcpu paging mode. This is nearly always true
1657 * (might be false while changing modes). Note it is verified later
1661 /* Handle a 32-bit guest writing two halves of a 64-bit gpte */
1662 if ((bytes
== 4) && (gpa
% 4 == 0)) {
1663 r
= kvm_read_guest(vcpu
->kvm
, gpa
& ~(u64
)7, &gpte
, 8);
1666 memcpy((void *)&gpte
+ (gpa
% 8), new, 4);
1667 } else if ((bytes
== 8) && (gpa
% 8 == 0)) {
1668 memcpy((void *)&gpte
, new, 8);
1671 if ((bytes
== 4) && (gpa
% 4 == 0))
1672 memcpy((void *)&gpte
, new, 4);
1674 if (!is_present_pte(gpte
))
1676 gfn
= (gpte
& PT64_BASE_ADDR_MASK
) >> PAGE_SHIFT
;
1678 down_read(¤t
->mm
->mmap_sem
);
1679 if (is_large_pte(gpte
) && is_largepage_backed(vcpu
, gfn
)) {
1680 gfn
&= ~(KVM_PAGES_PER_HPAGE
-1);
1681 vcpu
->arch
.update_pte
.largepage
= 1;
1683 page
= gfn_to_page(vcpu
->kvm
, gfn
);
1684 up_read(¤t
->mm
->mmap_sem
);
1686 if (is_error_page(page
)) {
1687 kvm_release_page_clean(page
);
1690 vcpu
->arch
.update_pte
.gfn
= gfn
;
1691 vcpu
->arch
.update_pte
.page
= page
;
1694 void kvm_mmu_pte_write(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1695 const u8
*new, int bytes
)
1697 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1698 struct kvm_mmu_page
*sp
;
1699 struct hlist_node
*node
, *n
;
1700 struct hlist_head
*bucket
;
1704 unsigned offset
= offset_in_page(gpa
);
1706 unsigned page_offset
;
1707 unsigned misaligned
;
1714 pgprintk("%s: gpa %llx bytes %d\n", __func__
, gpa
, bytes
);
1715 mmu_guess_page_from_pte_write(vcpu
, gpa
, new, bytes
);
1716 spin_lock(&vcpu
->kvm
->mmu_lock
);
1717 kvm_mmu_free_some_pages(vcpu
);
1718 ++vcpu
->kvm
->stat
.mmu_pte_write
;
1719 kvm_mmu_audit(vcpu
, "pre pte write");
1720 if (gfn
== vcpu
->arch
.last_pt_write_gfn
1721 && !last_updated_pte_accessed(vcpu
)) {
1722 ++vcpu
->arch
.last_pt_write_count
;
1723 if (vcpu
->arch
.last_pt_write_count
>= 3)
1726 vcpu
->arch
.last_pt_write_gfn
= gfn
;
1727 vcpu
->arch
.last_pt_write_count
= 1;
1728 vcpu
->arch
.last_pte_updated
= NULL
;
1730 index
= kvm_page_table_hashfn(gfn
);
1731 bucket
= &vcpu
->kvm
->arch
.mmu_page_hash
[index
];
1732 hlist_for_each_entry_safe(sp
, node
, n
, bucket
, hash_link
) {
1733 if (sp
->gfn
!= gfn
|| sp
->role
.metaphysical
)
1735 pte_size
= sp
->role
.glevels
== PT32_ROOT_LEVEL
? 4 : 8;
1736 misaligned
= (offset
^ (offset
+ bytes
- 1)) & ~(pte_size
- 1);
1737 misaligned
|= bytes
< 4;
1738 if (misaligned
|| flooded
) {
1740 * Misaligned accesses are too much trouble to fix
1741 * up; also, they usually indicate a page is not used
1744 * If we're seeing too many writes to a page,
1745 * it may no longer be a page table, or we may be
1746 * forking, in which case it is better to unmap the
1749 pgprintk("misaligned: gpa %llx bytes %d role %x\n",
1750 gpa
, bytes
, sp
->role
.word
);
1751 kvm_mmu_zap_page(vcpu
->kvm
, sp
);
1752 ++vcpu
->kvm
->stat
.mmu_flooded
;
1755 page_offset
= offset
;
1756 level
= sp
->role
.level
;
1758 if (sp
->role
.glevels
== PT32_ROOT_LEVEL
) {
1759 page_offset
<<= 1; /* 32->64 */
1761 * A 32-bit pde maps 4MB while the shadow pdes map
1762 * only 2MB. So we need to double the offset again
1763 * and zap two pdes instead of one.
1765 if (level
== PT32_ROOT_LEVEL
) {
1766 page_offset
&= ~7; /* kill rounding error */
1770 quadrant
= page_offset
>> PAGE_SHIFT
;
1771 page_offset
&= ~PAGE_MASK
;
1772 if (quadrant
!= sp
->role
.quadrant
)
1775 spte
= &sp
->spt
[page_offset
/ sizeof(*spte
)];
1776 if ((gpa
& (pte_size
- 1)) || (bytes
< pte_size
)) {
1778 r
= kvm_read_guest_atomic(vcpu
->kvm
,
1779 gpa
& ~(u64
)(pte_size
- 1),
1781 new = (const void *)&gentry
;
1787 mmu_pte_write_zap_pte(vcpu
, sp
, spte
);
1789 mmu_pte_write_new_pte(vcpu
, sp
, spte
, new);
1790 mmu_pte_write_flush_tlb(vcpu
, entry
, *spte
);
1794 kvm_mmu_audit(vcpu
, "post pte write");
1795 spin_unlock(&vcpu
->kvm
->mmu_lock
);
1796 if (vcpu
->arch
.update_pte
.page
) {
1797 kvm_release_page_clean(vcpu
->arch
.update_pte
.page
);
1798 vcpu
->arch
.update_pte
.page
= NULL
;
1802 int kvm_mmu_unprotect_page_virt(struct kvm_vcpu
*vcpu
, gva_t gva
)
1807 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, gva
);
1809 spin_lock(&vcpu
->kvm
->mmu_lock
);
1810 r
= kvm_mmu_unprotect_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
1811 spin_unlock(&vcpu
->kvm
->mmu_lock
);
1815 void __kvm_mmu_free_some_pages(struct kvm_vcpu
*vcpu
)
1817 while (vcpu
->kvm
->arch
.n_free_mmu_pages
< KVM_REFILL_PAGES
) {
1818 struct kvm_mmu_page
*sp
;
1820 sp
= container_of(vcpu
->kvm
->arch
.active_mmu_pages
.prev
,
1821 struct kvm_mmu_page
, link
);
1822 kvm_mmu_zap_page(vcpu
->kvm
, sp
);
1823 ++vcpu
->kvm
->stat
.mmu_recycled
;
1827 int kvm_mmu_page_fault(struct kvm_vcpu
*vcpu
, gva_t cr2
, u32 error_code
)
1830 enum emulation_result er
;
1832 r
= vcpu
->arch
.mmu
.page_fault(vcpu
, cr2
, error_code
);
1841 r
= mmu_topup_memory_caches(vcpu
);
1845 er
= emulate_instruction(vcpu
, vcpu
->run
, cr2
, error_code
, 0);
1850 case EMULATE_DO_MMIO
:
1851 ++vcpu
->stat
.mmio_exits
;
1854 kvm_report_emulation_failure(vcpu
, "pagetable");
1862 EXPORT_SYMBOL_GPL(kvm_mmu_page_fault
);
1864 void kvm_enable_tdp(void)
1868 EXPORT_SYMBOL_GPL(kvm_enable_tdp
);
1870 static void free_mmu_pages(struct kvm_vcpu
*vcpu
)
1872 struct kvm_mmu_page
*sp
;
1874 while (!list_empty(&vcpu
->kvm
->arch
.active_mmu_pages
)) {
1875 sp
= container_of(vcpu
->kvm
->arch
.active_mmu_pages
.next
,
1876 struct kvm_mmu_page
, link
);
1877 kvm_mmu_zap_page(vcpu
->kvm
, sp
);
1879 free_page((unsigned long)vcpu
->arch
.mmu
.pae_root
);
1882 static int alloc_mmu_pages(struct kvm_vcpu
*vcpu
)
1889 if (vcpu
->kvm
->arch
.n_requested_mmu_pages
)
1890 vcpu
->kvm
->arch
.n_free_mmu_pages
=
1891 vcpu
->kvm
->arch
.n_requested_mmu_pages
;
1893 vcpu
->kvm
->arch
.n_free_mmu_pages
=
1894 vcpu
->kvm
->arch
.n_alloc_mmu_pages
;
1896 * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64.
1897 * Therefore we need to allocate shadow page tables in the first
1898 * 4GB of memory, which happens to fit the DMA32 zone.
1900 page
= alloc_page(GFP_KERNEL
| __GFP_DMA32
);
1903 vcpu
->arch
.mmu
.pae_root
= page_address(page
);
1904 for (i
= 0; i
< 4; ++i
)
1905 vcpu
->arch
.mmu
.pae_root
[i
] = INVALID_PAGE
;
1910 free_mmu_pages(vcpu
);
1914 int kvm_mmu_create(struct kvm_vcpu
*vcpu
)
1917 ASSERT(!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
1919 return alloc_mmu_pages(vcpu
);
1922 int kvm_mmu_setup(struct kvm_vcpu
*vcpu
)
1925 ASSERT(!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
1927 return init_kvm_mmu(vcpu
);
1930 void kvm_mmu_destroy(struct kvm_vcpu
*vcpu
)
1934 destroy_kvm_mmu(vcpu
);
1935 free_mmu_pages(vcpu
);
1936 mmu_free_memory_caches(vcpu
);
1939 void kvm_mmu_slot_remove_write_access(struct kvm
*kvm
, int slot
)
1941 struct kvm_mmu_page
*sp
;
1943 list_for_each_entry(sp
, &kvm
->arch
.active_mmu_pages
, link
) {
1947 if (!test_bit(slot
, &sp
->slot_bitmap
))
1951 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
)
1953 if (pt
[i
] & PT_WRITABLE_MASK
)
1954 pt
[i
] &= ~PT_WRITABLE_MASK
;
1958 void kvm_mmu_zap_all(struct kvm
*kvm
)
1960 struct kvm_mmu_page
*sp
, *node
;
1962 spin_lock(&kvm
->mmu_lock
);
1963 list_for_each_entry_safe(sp
, node
, &kvm
->arch
.active_mmu_pages
, link
)
1964 kvm_mmu_zap_page(kvm
, sp
);
1965 spin_unlock(&kvm
->mmu_lock
);
1967 kvm_flush_remote_tlbs(kvm
);
1970 void kvm_mmu_remove_one_alloc_mmu_page(struct kvm
*kvm
)
1972 struct kvm_mmu_page
*page
;
1974 page
= container_of(kvm
->arch
.active_mmu_pages
.prev
,
1975 struct kvm_mmu_page
, link
);
1976 kvm_mmu_zap_page(kvm
, page
);
1979 static int mmu_shrink(int nr_to_scan
, gfp_t gfp_mask
)
1982 struct kvm
*kvm_freed
= NULL
;
1983 int cache_count
= 0;
1985 spin_lock(&kvm_lock
);
1987 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
1990 spin_lock(&kvm
->mmu_lock
);
1991 npages
= kvm
->arch
.n_alloc_mmu_pages
-
1992 kvm
->arch
.n_free_mmu_pages
;
1993 cache_count
+= npages
;
1994 if (!kvm_freed
&& nr_to_scan
> 0 && npages
> 0) {
1995 kvm_mmu_remove_one_alloc_mmu_page(kvm
);
2001 spin_unlock(&kvm
->mmu_lock
);
2004 list_move_tail(&kvm_freed
->vm_list
, &vm_list
);
2006 spin_unlock(&kvm_lock
);
2011 static struct shrinker mmu_shrinker
= {
2012 .shrink
= mmu_shrink
,
2013 .seeks
= DEFAULT_SEEKS
* 10,
2016 void mmu_destroy_caches(void)
2018 if (pte_chain_cache
)
2019 kmem_cache_destroy(pte_chain_cache
);
2020 if (rmap_desc_cache
)
2021 kmem_cache_destroy(rmap_desc_cache
);
2022 if (mmu_page_header_cache
)
2023 kmem_cache_destroy(mmu_page_header_cache
);
2026 void kvm_mmu_module_exit(void)
2028 mmu_destroy_caches();
2029 unregister_shrinker(&mmu_shrinker
);
2032 int kvm_mmu_module_init(void)
2034 pte_chain_cache
= kmem_cache_create("kvm_pte_chain",
2035 sizeof(struct kvm_pte_chain
),
2037 if (!pte_chain_cache
)
2039 rmap_desc_cache
= kmem_cache_create("kvm_rmap_desc",
2040 sizeof(struct kvm_rmap_desc
),
2042 if (!rmap_desc_cache
)
2045 mmu_page_header_cache
= kmem_cache_create("kvm_mmu_page_header",
2046 sizeof(struct kvm_mmu_page
),
2048 if (!mmu_page_header_cache
)
2051 register_shrinker(&mmu_shrinker
);
2056 mmu_destroy_caches();
2061 * Caculate mmu pages needed for kvm.
2063 unsigned int kvm_mmu_calculate_mmu_pages(struct kvm
*kvm
)
2066 unsigned int nr_mmu_pages
;
2067 unsigned int nr_pages
= 0;
2069 for (i
= 0; i
< kvm
->nmemslots
; i
++)
2070 nr_pages
+= kvm
->memslots
[i
].npages
;
2072 nr_mmu_pages
= nr_pages
* KVM_PERMILLE_MMU_PAGES
/ 1000;
2073 nr_mmu_pages
= max(nr_mmu_pages
,
2074 (unsigned int) KVM_MIN_ALLOC_MMU_PAGES
);
2076 return nr_mmu_pages
;
2079 static void *pv_mmu_peek_buffer(struct kvm_pv_mmu_op_buffer
*buffer
,
2082 if (len
> buffer
->len
)
2087 static void *pv_mmu_read_buffer(struct kvm_pv_mmu_op_buffer
*buffer
,
2092 ret
= pv_mmu_peek_buffer(buffer
, len
);
2097 buffer
->processed
+= len
;
2101 static int kvm_pv_mmu_write(struct kvm_vcpu
*vcpu
,
2102 gpa_t addr
, gpa_t value
)
2107 if (!is_long_mode(vcpu
) && !is_pae(vcpu
))
2110 r
= mmu_topup_memory_caches(vcpu
);
2114 if (!emulator_write_phys(vcpu
, addr
, &value
, bytes
))
2120 static int kvm_pv_mmu_flush_tlb(struct kvm_vcpu
*vcpu
)
2122 kvm_x86_ops
->tlb_flush(vcpu
);
2126 static int kvm_pv_mmu_release_pt(struct kvm_vcpu
*vcpu
, gpa_t addr
)
2128 spin_lock(&vcpu
->kvm
->mmu_lock
);
2129 mmu_unshadow(vcpu
->kvm
, addr
>> PAGE_SHIFT
);
2130 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2134 static int kvm_pv_mmu_op_one(struct kvm_vcpu
*vcpu
,
2135 struct kvm_pv_mmu_op_buffer
*buffer
)
2137 struct kvm_mmu_op_header
*header
;
2139 header
= pv_mmu_peek_buffer(buffer
, sizeof *header
);
2142 switch (header
->op
) {
2143 case KVM_MMU_OP_WRITE_PTE
: {
2144 struct kvm_mmu_op_write_pte
*wpte
;
2146 wpte
= pv_mmu_read_buffer(buffer
, sizeof *wpte
);
2149 return kvm_pv_mmu_write(vcpu
, wpte
->pte_phys
,
2152 case KVM_MMU_OP_FLUSH_TLB
: {
2153 struct kvm_mmu_op_flush_tlb
*ftlb
;
2155 ftlb
= pv_mmu_read_buffer(buffer
, sizeof *ftlb
);
2158 return kvm_pv_mmu_flush_tlb(vcpu
);
2160 case KVM_MMU_OP_RELEASE_PT
: {
2161 struct kvm_mmu_op_release_pt
*rpt
;
2163 rpt
= pv_mmu_read_buffer(buffer
, sizeof *rpt
);
2166 return kvm_pv_mmu_release_pt(vcpu
, rpt
->pt_phys
);
2172 int kvm_pv_mmu_op(struct kvm_vcpu
*vcpu
, unsigned long bytes
,
2173 gpa_t addr
, unsigned long *ret
)
2176 struct kvm_pv_mmu_op_buffer buffer
;
2178 down_read(¤t
->mm
->mmap_sem
);
2180 buffer
.ptr
= buffer
.buf
;
2181 buffer
.len
= min_t(unsigned long, bytes
, sizeof buffer
.buf
);
2182 buffer
.processed
= 0;
2184 r
= kvm_read_guest(vcpu
->kvm
, addr
, buffer
.buf
, buffer
.len
);
2188 while (buffer
.len
) {
2189 r
= kvm_pv_mmu_op_one(vcpu
, &buffer
);
2198 *ret
= buffer
.processed
;
2199 up_read(¤t
->mm
->mmap_sem
);
2205 static const char *audit_msg
;
2207 static gva_t
canonicalize(gva_t gva
)
2209 #ifdef CONFIG_X86_64
2210 gva
= (long long)(gva
<< 16) >> 16;
2215 static void audit_mappings_page(struct kvm_vcpu
*vcpu
, u64 page_pte
,
2216 gva_t va
, int level
)
2218 u64
*pt
= __va(page_pte
& PT64_BASE_ADDR_MASK
);
2220 gva_t va_delta
= 1ul << (PAGE_SHIFT
+ 9 * (level
- 1));
2222 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
, va
+= va_delta
) {
2225 if (ent
== shadow_trap_nonpresent_pte
)
2228 va
= canonicalize(va
);
2230 if (ent
== shadow_notrap_nonpresent_pte
)
2231 printk(KERN_ERR
"audit: (%s) nontrapping pte"
2232 " in nonleaf level: levels %d gva %lx"
2233 " level %d pte %llx\n", audit_msg
,
2234 vcpu
->arch
.mmu
.root_level
, va
, level
, ent
);
2236 audit_mappings_page(vcpu
, ent
, va
, level
- 1);
2238 gpa_t gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, va
);
2239 struct page
*page
= gpa_to_page(vcpu
, gpa
);
2240 hpa_t hpa
= page_to_phys(page
);
2242 if (is_shadow_present_pte(ent
)
2243 && (ent
& PT64_BASE_ADDR_MASK
) != hpa
)
2244 printk(KERN_ERR
"xx audit error: (%s) levels %d"
2245 " gva %lx gpa %llx hpa %llx ent %llx %d\n",
2246 audit_msg
, vcpu
->arch
.mmu
.root_level
,
2248 is_shadow_present_pte(ent
));
2249 else if (ent
== shadow_notrap_nonpresent_pte
2250 && !is_error_hpa(hpa
))
2251 printk(KERN_ERR
"audit: (%s) notrap shadow,"
2252 " valid guest gva %lx\n", audit_msg
, va
);
2253 kvm_release_page_clean(page
);
2259 static void audit_mappings(struct kvm_vcpu
*vcpu
)
2263 if (vcpu
->arch
.mmu
.root_level
== 4)
2264 audit_mappings_page(vcpu
, vcpu
->arch
.mmu
.root_hpa
, 0, 4);
2266 for (i
= 0; i
< 4; ++i
)
2267 if (vcpu
->arch
.mmu
.pae_root
[i
] & PT_PRESENT_MASK
)
2268 audit_mappings_page(vcpu
,
2269 vcpu
->arch
.mmu
.pae_root
[i
],
2274 static int count_rmaps(struct kvm_vcpu
*vcpu
)
2279 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
2280 struct kvm_memory_slot
*m
= &vcpu
->kvm
->memslots
[i
];
2281 struct kvm_rmap_desc
*d
;
2283 for (j
= 0; j
< m
->npages
; ++j
) {
2284 unsigned long *rmapp
= &m
->rmap
[j
];
2288 if (!(*rmapp
& 1)) {
2292 d
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
2294 for (k
= 0; k
< RMAP_EXT
; ++k
)
2295 if (d
->shadow_ptes
[k
])
2306 static int count_writable_mappings(struct kvm_vcpu
*vcpu
)
2309 struct kvm_mmu_page
*sp
;
2312 list_for_each_entry(sp
, &vcpu
->kvm
->arch
.active_mmu_pages
, link
) {
2315 if (sp
->role
.level
!= PT_PAGE_TABLE_LEVEL
)
2318 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
2321 if (!(ent
& PT_PRESENT_MASK
))
2323 if (!(ent
& PT_WRITABLE_MASK
))
2331 static void audit_rmap(struct kvm_vcpu
*vcpu
)
2333 int n_rmap
= count_rmaps(vcpu
);
2334 int n_actual
= count_writable_mappings(vcpu
);
2336 if (n_rmap
!= n_actual
)
2337 printk(KERN_ERR
"%s: (%s) rmap %d actual %d\n",
2338 __func__
, audit_msg
, n_rmap
, n_actual
);
2341 static void audit_write_protection(struct kvm_vcpu
*vcpu
)
2343 struct kvm_mmu_page
*sp
;
2344 struct kvm_memory_slot
*slot
;
2345 unsigned long *rmapp
;
2348 list_for_each_entry(sp
, &vcpu
->kvm
->arch
.active_mmu_pages
, link
) {
2349 if (sp
->role
.metaphysical
)
2352 slot
= gfn_to_memslot(vcpu
->kvm
, sp
->gfn
);
2353 gfn
= unalias_gfn(vcpu
->kvm
, sp
->gfn
);
2354 rmapp
= &slot
->rmap
[gfn
- slot
->base_gfn
];
2356 printk(KERN_ERR
"%s: (%s) shadow page has writable"
2357 " mappings: gfn %lx role %x\n",
2358 __func__
, audit_msg
, sp
->gfn
,
2363 static void kvm_mmu_audit(struct kvm_vcpu
*vcpu
, const char *msg
)
2370 audit_write_protection(vcpu
);
2371 audit_mappings(vcpu
);