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/types.h>
24 #include <linux/string.h>
26 #include <linux/highmem.h>
27 #include <linux/module.h>
30 #include <asm/cmpxchg.h>
37 static void kvm_mmu_audit(struct kvm_vcpu
*vcpu
, const char *msg
);
39 static void kvm_mmu_audit(struct kvm_vcpu
*vcpu
, const char *msg
) {}
44 #define pgprintk(x...) do { if (dbg) printk(x); } while (0)
45 #define rmap_printk(x...) do { if (dbg) printk(x); } while (0)
49 #define pgprintk(x...) do { } while (0)
50 #define rmap_printk(x...) do { } while (0)
54 #if defined(MMU_DEBUG) || defined(AUDIT)
59 #define ASSERT(x) do { } while (0)
63 printk(KERN_WARNING "assertion failed %s:%d: %s\n", \
64 __FILE__, __LINE__, #x); \
68 #define PT64_PT_BITS 9
69 #define PT64_ENT_PER_PAGE (1 << PT64_PT_BITS)
70 #define PT32_PT_BITS 10
71 #define PT32_ENT_PER_PAGE (1 << PT32_PT_BITS)
73 #define PT_WRITABLE_SHIFT 1
75 #define PT_PRESENT_MASK (1ULL << 0)
76 #define PT_WRITABLE_MASK (1ULL << PT_WRITABLE_SHIFT)
77 #define PT_USER_MASK (1ULL << 2)
78 #define PT_PWT_MASK (1ULL << 3)
79 #define PT_PCD_MASK (1ULL << 4)
80 #define PT_ACCESSED_MASK (1ULL << 5)
81 #define PT_DIRTY_MASK (1ULL << 6)
82 #define PT_PAGE_SIZE_MASK (1ULL << 7)
83 #define PT_PAT_MASK (1ULL << 7)
84 #define PT_GLOBAL_MASK (1ULL << 8)
85 #define PT64_NX_MASK (1ULL << 63)
87 #define PT_PAT_SHIFT 7
88 #define PT_DIR_PAT_SHIFT 12
89 #define PT_DIR_PAT_MASK (1ULL << PT_DIR_PAT_SHIFT)
91 #define PT32_DIR_PSE36_SIZE 4
92 #define PT32_DIR_PSE36_SHIFT 13
93 #define PT32_DIR_PSE36_MASK (((1ULL << PT32_DIR_PSE36_SIZE) - 1) << PT32_DIR_PSE36_SHIFT)
96 #define PT_FIRST_AVAIL_BITS_SHIFT 9
97 #define PT64_SECOND_AVAIL_BITS_SHIFT 52
99 #define PT_SHADOW_IO_MARK (1ULL << PT_FIRST_AVAIL_BITS_SHIFT)
101 #define VALID_PAGE(x) ((x) != INVALID_PAGE)
103 #define PT64_LEVEL_BITS 9
105 #define PT64_LEVEL_SHIFT(level) \
106 ( PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS )
108 #define PT64_LEVEL_MASK(level) \
109 (((1ULL << PT64_LEVEL_BITS) - 1) << PT64_LEVEL_SHIFT(level))
111 #define PT64_INDEX(address, level)\
112 (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1))
115 #define PT32_LEVEL_BITS 10
117 #define PT32_LEVEL_SHIFT(level) \
118 ( PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS )
120 #define PT32_LEVEL_MASK(level) \
121 (((1ULL << PT32_LEVEL_BITS) - 1) << PT32_LEVEL_SHIFT(level))
123 #define PT32_INDEX(address, level)\
124 (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1))
127 #define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1))
128 #define PT64_DIR_BASE_ADDR_MASK \
129 (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1))
131 #define PT32_BASE_ADDR_MASK PAGE_MASK
132 #define PT32_DIR_BASE_ADDR_MASK \
133 (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1))
136 #define PFERR_PRESENT_MASK (1U << 0)
137 #define PFERR_WRITE_MASK (1U << 1)
138 #define PFERR_USER_MASK (1U << 2)
139 #define PFERR_FETCH_MASK (1U << 4)
141 #define PT64_ROOT_LEVEL 4
142 #define PT32_ROOT_LEVEL 2
143 #define PT32E_ROOT_LEVEL 3
145 #define PT_DIRECTORY_LEVEL 2
146 #define PT_PAGE_TABLE_LEVEL 1
150 struct kvm_rmap_desc
{
151 u64
*shadow_ptes
[RMAP_EXT
];
152 struct kvm_rmap_desc
*more
;
155 static struct kmem_cache
*pte_chain_cache
;
156 static struct kmem_cache
*rmap_desc_cache
;
157 static struct kmem_cache
*mmu_page_header_cache
;
159 static int is_write_protection(struct kvm_vcpu
*vcpu
)
161 return vcpu
->cr0
& CR0_WP_MASK
;
164 static int is_cpuid_PSE36(void)
169 static int is_nx(struct kvm_vcpu
*vcpu
)
171 return vcpu
->shadow_efer
& EFER_NX
;
174 static int is_present_pte(unsigned long pte
)
176 return pte
& PT_PRESENT_MASK
;
179 static int is_writeble_pte(unsigned long pte
)
181 return pte
& PT_WRITABLE_MASK
;
184 static int is_io_pte(unsigned long pte
)
186 return pte
& PT_SHADOW_IO_MARK
;
189 static int is_rmap_pte(u64 pte
)
191 return (pte
& (PT_WRITABLE_MASK
| PT_PRESENT_MASK
))
192 == (PT_WRITABLE_MASK
| PT_PRESENT_MASK
);
195 static void set_shadow_pte(u64
*sptep
, u64 spte
)
198 set_64bit((unsigned long *)sptep
, spte
);
200 set_64bit((unsigned long long *)sptep
, spte
);
204 static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache
*cache
,
205 struct kmem_cache
*base_cache
, int min
,
210 if (cache
->nobjs
>= min
)
212 while (cache
->nobjs
< ARRAY_SIZE(cache
->objects
)) {
213 obj
= kmem_cache_zalloc(base_cache
, gfp_flags
);
216 cache
->objects
[cache
->nobjs
++] = obj
;
221 static void mmu_free_memory_cache(struct kvm_mmu_memory_cache
*mc
)
224 kfree(mc
->objects
[--mc
->nobjs
]);
227 static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache
*cache
,
228 int min
, gfp_t gfp_flags
)
232 if (cache
->nobjs
>= min
)
234 while (cache
->nobjs
< ARRAY_SIZE(cache
->objects
)) {
235 page
= alloc_page(gfp_flags
);
238 set_page_private(page
, 0);
239 cache
->objects
[cache
->nobjs
++] = page_address(page
);
244 static void mmu_free_memory_cache_page(struct kvm_mmu_memory_cache
*mc
)
247 free_page((unsigned long)mc
->objects
[--mc
->nobjs
]);
250 static int __mmu_topup_memory_caches(struct kvm_vcpu
*vcpu
, gfp_t gfp_flags
)
254 r
= mmu_topup_memory_cache(&vcpu
->mmu_pte_chain_cache
,
255 pte_chain_cache
, 4, gfp_flags
);
258 r
= mmu_topup_memory_cache(&vcpu
->mmu_rmap_desc_cache
,
259 rmap_desc_cache
, 1, gfp_flags
);
262 r
= mmu_topup_memory_cache_page(&vcpu
->mmu_page_cache
, 4, gfp_flags
);
265 r
= mmu_topup_memory_cache(&vcpu
->mmu_page_header_cache
,
266 mmu_page_header_cache
, 4, gfp_flags
);
271 static int mmu_topup_memory_caches(struct kvm_vcpu
*vcpu
)
275 r
= __mmu_topup_memory_caches(vcpu
, GFP_NOWAIT
);
276 kvm_mmu_free_some_pages(vcpu
);
278 spin_unlock(&vcpu
->kvm
->lock
);
279 kvm_arch_ops
->vcpu_put(vcpu
);
280 r
= __mmu_topup_memory_caches(vcpu
, GFP_KERNEL
);
281 kvm_arch_ops
->vcpu_load(vcpu
);
282 spin_lock(&vcpu
->kvm
->lock
);
283 kvm_mmu_free_some_pages(vcpu
);
288 static void mmu_free_memory_caches(struct kvm_vcpu
*vcpu
)
290 mmu_free_memory_cache(&vcpu
->mmu_pte_chain_cache
);
291 mmu_free_memory_cache(&vcpu
->mmu_rmap_desc_cache
);
292 mmu_free_memory_cache_page(&vcpu
->mmu_page_cache
);
293 mmu_free_memory_cache(&vcpu
->mmu_page_header_cache
);
296 static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache
*mc
,
302 p
= mc
->objects
[--mc
->nobjs
];
307 static struct kvm_pte_chain
*mmu_alloc_pte_chain(struct kvm_vcpu
*vcpu
)
309 return mmu_memory_cache_alloc(&vcpu
->mmu_pte_chain_cache
,
310 sizeof(struct kvm_pte_chain
));
313 static void mmu_free_pte_chain(struct kvm_pte_chain
*pc
)
318 static struct kvm_rmap_desc
*mmu_alloc_rmap_desc(struct kvm_vcpu
*vcpu
)
320 return mmu_memory_cache_alloc(&vcpu
->mmu_rmap_desc_cache
,
321 sizeof(struct kvm_rmap_desc
));
324 static void mmu_free_rmap_desc(struct kvm_rmap_desc
*rd
)
330 * Reverse mapping data structures:
332 * If page->private bit zero is zero, then page->private points to the
333 * shadow page table entry that points to page_address(page).
335 * If page->private bit zero is one, (then page->private & ~1) points
336 * to a struct kvm_rmap_desc containing more mappings.
338 static void rmap_add(struct kvm_vcpu
*vcpu
, u64
*spte
)
341 struct kvm_rmap_desc
*desc
;
344 if (!is_rmap_pte(*spte
))
346 page
= pfn_to_page((*spte
& PT64_BASE_ADDR_MASK
) >> PAGE_SHIFT
);
347 if (!page_private(page
)) {
348 rmap_printk("rmap_add: %p %llx 0->1\n", spte
, *spte
);
349 set_page_private(page
,(unsigned long)spte
);
350 } else if (!(page_private(page
) & 1)) {
351 rmap_printk("rmap_add: %p %llx 1->many\n", spte
, *spte
);
352 desc
= mmu_alloc_rmap_desc(vcpu
);
353 desc
->shadow_ptes
[0] = (u64
*)page_private(page
);
354 desc
->shadow_ptes
[1] = spte
;
355 set_page_private(page
,(unsigned long)desc
| 1);
357 rmap_printk("rmap_add: %p %llx many->many\n", spte
, *spte
);
358 desc
= (struct kvm_rmap_desc
*)(page_private(page
) & ~1ul);
359 while (desc
->shadow_ptes
[RMAP_EXT
-1] && desc
->more
)
361 if (desc
->shadow_ptes
[RMAP_EXT
-1]) {
362 desc
->more
= mmu_alloc_rmap_desc(vcpu
);
365 for (i
= 0; desc
->shadow_ptes
[i
]; ++i
)
367 desc
->shadow_ptes
[i
] = spte
;
371 static void rmap_desc_remove_entry(struct page
*page
,
372 struct kvm_rmap_desc
*desc
,
374 struct kvm_rmap_desc
*prev_desc
)
378 for (j
= RMAP_EXT
- 1; !desc
->shadow_ptes
[j
] && j
> i
; --j
)
380 desc
->shadow_ptes
[i
] = desc
->shadow_ptes
[j
];
381 desc
->shadow_ptes
[j
] = NULL
;
384 if (!prev_desc
&& !desc
->more
)
385 set_page_private(page
,(unsigned long)desc
->shadow_ptes
[0]);
388 prev_desc
->more
= desc
->more
;
390 set_page_private(page
,(unsigned long)desc
->more
| 1);
391 mmu_free_rmap_desc(desc
);
394 static void rmap_remove(u64
*spte
)
397 struct kvm_rmap_desc
*desc
;
398 struct kvm_rmap_desc
*prev_desc
;
401 if (!is_rmap_pte(*spte
))
403 page
= pfn_to_page((*spte
& PT64_BASE_ADDR_MASK
) >> PAGE_SHIFT
);
404 if (!page_private(page
)) {
405 printk(KERN_ERR
"rmap_remove: %p %llx 0->BUG\n", spte
, *spte
);
407 } else if (!(page_private(page
) & 1)) {
408 rmap_printk("rmap_remove: %p %llx 1->0\n", spte
, *spte
);
409 if ((u64
*)page_private(page
) != spte
) {
410 printk(KERN_ERR
"rmap_remove: %p %llx 1->BUG\n",
414 set_page_private(page
,0);
416 rmap_printk("rmap_remove: %p %llx many->many\n", spte
, *spte
);
417 desc
= (struct kvm_rmap_desc
*)(page_private(page
) & ~1ul);
420 for (i
= 0; i
< RMAP_EXT
&& desc
->shadow_ptes
[i
]; ++i
)
421 if (desc
->shadow_ptes
[i
] == spte
) {
422 rmap_desc_remove_entry(page
,
434 static void rmap_write_protect(struct kvm_vcpu
*vcpu
, u64 gfn
)
436 struct kvm
*kvm
= vcpu
->kvm
;
438 struct kvm_rmap_desc
*desc
;
441 page
= gfn_to_page(kvm
, gfn
);
444 while (page_private(page
)) {
445 if (!(page_private(page
) & 1))
446 spte
= (u64
*)page_private(page
);
448 desc
= (struct kvm_rmap_desc
*)(page_private(page
) & ~1ul);
449 spte
= desc
->shadow_ptes
[0];
452 BUG_ON((*spte
& PT64_BASE_ADDR_MASK
) >> PAGE_SHIFT
453 != page_to_pfn(page
));
454 BUG_ON(!(*spte
& PT_PRESENT_MASK
));
455 BUG_ON(!(*spte
& PT_WRITABLE_MASK
));
456 rmap_printk("rmap_write_protect: spte %p %llx\n", spte
, *spte
);
458 set_shadow_pte(spte
, *spte
& ~PT_WRITABLE_MASK
);
459 kvm_flush_remote_tlbs(vcpu
->kvm
);
464 static int is_empty_shadow_page(u64
*spt
)
469 for (pos
= spt
, end
= pos
+ PAGE_SIZE
/ sizeof(u64
); pos
!= end
; pos
++)
471 printk(KERN_ERR
"%s: %p %llx\n", __FUNCTION__
,
479 static void kvm_mmu_free_page(struct kvm
*kvm
,
480 struct kvm_mmu_page
*page_head
)
482 ASSERT(is_empty_shadow_page(page_head
->spt
));
483 list_del(&page_head
->link
);
484 __free_page(virt_to_page(page_head
->spt
));
486 ++kvm
->n_free_mmu_pages
;
489 static unsigned kvm_page_table_hashfn(gfn_t gfn
)
494 static struct kvm_mmu_page
*kvm_mmu_alloc_page(struct kvm_vcpu
*vcpu
,
497 struct kvm_mmu_page
*page
;
499 if (!vcpu
->kvm
->n_free_mmu_pages
)
502 page
= mmu_memory_cache_alloc(&vcpu
->mmu_page_header_cache
,
504 page
->spt
= mmu_memory_cache_alloc(&vcpu
->mmu_page_cache
, PAGE_SIZE
);
505 set_page_private(virt_to_page(page
->spt
), (unsigned long)page
);
506 list_add(&page
->link
, &vcpu
->kvm
->active_mmu_pages
);
507 ASSERT(is_empty_shadow_page(page
->spt
));
508 page
->slot_bitmap
= 0;
509 page
->multimapped
= 0;
510 page
->parent_pte
= parent_pte
;
511 --vcpu
->kvm
->n_free_mmu_pages
;
515 static void mmu_page_add_parent_pte(struct kvm_vcpu
*vcpu
,
516 struct kvm_mmu_page
*page
, u64
*parent_pte
)
518 struct kvm_pte_chain
*pte_chain
;
519 struct hlist_node
*node
;
524 if (!page
->multimapped
) {
525 u64
*old
= page
->parent_pte
;
528 page
->parent_pte
= parent_pte
;
531 page
->multimapped
= 1;
532 pte_chain
= mmu_alloc_pte_chain(vcpu
);
533 INIT_HLIST_HEAD(&page
->parent_ptes
);
534 hlist_add_head(&pte_chain
->link
, &page
->parent_ptes
);
535 pte_chain
->parent_ptes
[0] = old
;
537 hlist_for_each_entry(pte_chain
, node
, &page
->parent_ptes
, link
) {
538 if (pte_chain
->parent_ptes
[NR_PTE_CHAIN_ENTRIES
-1])
540 for (i
= 0; i
< NR_PTE_CHAIN_ENTRIES
; ++i
)
541 if (!pte_chain
->parent_ptes
[i
]) {
542 pte_chain
->parent_ptes
[i
] = parent_pte
;
546 pte_chain
= mmu_alloc_pte_chain(vcpu
);
548 hlist_add_head(&pte_chain
->link
, &page
->parent_ptes
);
549 pte_chain
->parent_ptes
[0] = parent_pte
;
552 static void mmu_page_remove_parent_pte(struct kvm_mmu_page
*page
,
555 struct kvm_pte_chain
*pte_chain
;
556 struct hlist_node
*node
;
559 if (!page
->multimapped
) {
560 BUG_ON(page
->parent_pte
!= parent_pte
);
561 page
->parent_pte
= NULL
;
564 hlist_for_each_entry(pte_chain
, node
, &page
->parent_ptes
, link
)
565 for (i
= 0; i
< NR_PTE_CHAIN_ENTRIES
; ++i
) {
566 if (!pte_chain
->parent_ptes
[i
])
568 if (pte_chain
->parent_ptes
[i
] != parent_pte
)
570 while (i
+ 1 < NR_PTE_CHAIN_ENTRIES
571 && pte_chain
->parent_ptes
[i
+ 1]) {
572 pte_chain
->parent_ptes
[i
]
573 = pte_chain
->parent_ptes
[i
+ 1];
576 pte_chain
->parent_ptes
[i
] = NULL
;
578 hlist_del(&pte_chain
->link
);
579 mmu_free_pte_chain(pte_chain
);
580 if (hlist_empty(&page
->parent_ptes
)) {
581 page
->multimapped
= 0;
582 page
->parent_pte
= NULL
;
590 static struct kvm_mmu_page
*kvm_mmu_lookup_page(struct kvm_vcpu
*vcpu
,
594 struct hlist_head
*bucket
;
595 struct kvm_mmu_page
*page
;
596 struct hlist_node
*node
;
598 pgprintk("%s: looking for gfn %lx\n", __FUNCTION__
, gfn
);
599 index
= kvm_page_table_hashfn(gfn
) % KVM_NUM_MMU_PAGES
;
600 bucket
= &vcpu
->kvm
->mmu_page_hash
[index
];
601 hlist_for_each_entry(page
, node
, bucket
, hash_link
)
602 if (page
->gfn
== gfn
&& !page
->role
.metaphysical
) {
603 pgprintk("%s: found role %x\n",
604 __FUNCTION__
, page
->role
.word
);
610 static struct kvm_mmu_page
*kvm_mmu_get_page(struct kvm_vcpu
*vcpu
,
615 unsigned hugepage_access
,
618 union kvm_mmu_page_role role
;
621 struct hlist_head
*bucket
;
622 struct kvm_mmu_page
*page
;
623 struct hlist_node
*node
;
626 role
.glevels
= vcpu
->mmu
.root_level
;
628 role
.metaphysical
= metaphysical
;
629 role
.hugepage_access
= hugepage_access
;
630 if (vcpu
->mmu
.root_level
<= PT32_ROOT_LEVEL
) {
631 quadrant
= gaddr
>> (PAGE_SHIFT
+ (PT64_PT_BITS
* level
));
632 quadrant
&= (1 << ((PT32_PT_BITS
- PT64_PT_BITS
) * level
)) - 1;
633 role
.quadrant
= quadrant
;
635 pgprintk("%s: looking gfn %lx role %x\n", __FUNCTION__
,
637 index
= kvm_page_table_hashfn(gfn
) % KVM_NUM_MMU_PAGES
;
638 bucket
= &vcpu
->kvm
->mmu_page_hash
[index
];
639 hlist_for_each_entry(page
, node
, bucket
, hash_link
)
640 if (page
->gfn
== gfn
&& page
->role
.word
== role
.word
) {
641 mmu_page_add_parent_pte(vcpu
, page
, parent_pte
);
642 pgprintk("%s: found\n", __FUNCTION__
);
645 page
= kvm_mmu_alloc_page(vcpu
, parent_pte
);
648 pgprintk("%s: adding gfn %lx role %x\n", __FUNCTION__
, gfn
, role
.word
);
651 hlist_add_head(&page
->hash_link
, bucket
);
653 rmap_write_protect(vcpu
, gfn
);
657 static void kvm_mmu_page_unlink_children(struct kvm
*kvm
,
658 struct kvm_mmu_page
*page
)
666 if (page
->role
.level
== PT_PAGE_TABLE_LEVEL
) {
667 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
668 if (pt
[i
] & PT_PRESENT_MASK
)
672 kvm_flush_remote_tlbs(kvm
);
676 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
680 if (!(ent
& PT_PRESENT_MASK
))
682 ent
&= PT64_BASE_ADDR_MASK
;
683 mmu_page_remove_parent_pte(page_header(ent
), &pt
[i
]);
685 kvm_flush_remote_tlbs(kvm
);
688 static void kvm_mmu_put_page(struct kvm_mmu_page
*page
,
691 mmu_page_remove_parent_pte(page
, parent_pte
);
694 static void kvm_mmu_zap_page(struct kvm
*kvm
,
695 struct kvm_mmu_page
*page
)
699 while (page
->multimapped
|| page
->parent_pte
) {
700 if (!page
->multimapped
)
701 parent_pte
= page
->parent_pte
;
703 struct kvm_pte_chain
*chain
;
705 chain
= container_of(page
->parent_ptes
.first
,
706 struct kvm_pte_chain
, link
);
707 parent_pte
= chain
->parent_ptes
[0];
710 kvm_mmu_put_page(page
, parent_pte
);
711 set_shadow_pte(parent_pte
, 0);
713 kvm_mmu_page_unlink_children(kvm
, page
);
714 if (!page
->root_count
) {
715 hlist_del(&page
->hash_link
);
716 kvm_mmu_free_page(kvm
, page
);
718 list_move(&page
->link
, &kvm
->active_mmu_pages
);
721 static int kvm_mmu_unprotect_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
724 struct hlist_head
*bucket
;
725 struct kvm_mmu_page
*page
;
726 struct hlist_node
*node
, *n
;
729 pgprintk("%s: looking for gfn %lx\n", __FUNCTION__
, gfn
);
731 index
= kvm_page_table_hashfn(gfn
) % KVM_NUM_MMU_PAGES
;
732 bucket
= &vcpu
->kvm
->mmu_page_hash
[index
];
733 hlist_for_each_entry_safe(page
, node
, n
, bucket
, hash_link
)
734 if (page
->gfn
== gfn
&& !page
->role
.metaphysical
) {
735 pgprintk("%s: gfn %lx role %x\n", __FUNCTION__
, gfn
,
737 kvm_mmu_zap_page(vcpu
->kvm
, page
);
743 static void mmu_unshadow(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
745 struct kvm_mmu_page
*page
;
747 while ((page
= kvm_mmu_lookup_page(vcpu
, gfn
)) != NULL
) {
748 pgprintk("%s: zap %lx %x\n",
749 __FUNCTION__
, gfn
, page
->role
.word
);
750 kvm_mmu_zap_page(vcpu
->kvm
, page
);
754 static void page_header_update_slot(struct kvm
*kvm
, void *pte
, gpa_t gpa
)
756 int slot
= memslot_id(kvm
, gfn_to_memslot(kvm
, gpa
>> PAGE_SHIFT
));
757 struct kvm_mmu_page
*page_head
= page_header(__pa(pte
));
759 __set_bit(slot
, &page_head
->slot_bitmap
);
762 hpa_t
safe_gpa_to_hpa(struct kvm_vcpu
*vcpu
, gpa_t gpa
)
764 hpa_t hpa
= gpa_to_hpa(vcpu
, gpa
);
766 return is_error_hpa(hpa
) ? bad_page_address
| (gpa
& ~PAGE_MASK
): hpa
;
769 hpa_t
gpa_to_hpa(struct kvm_vcpu
*vcpu
, gpa_t gpa
)
773 ASSERT((gpa
& HPA_ERR_MASK
) == 0);
774 page
= gfn_to_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
776 return gpa
| HPA_ERR_MASK
;
777 return ((hpa_t
)page_to_pfn(page
) << PAGE_SHIFT
)
778 | (gpa
& (PAGE_SIZE
-1));
781 hpa_t
gva_to_hpa(struct kvm_vcpu
*vcpu
, gva_t gva
)
783 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, gva
);
785 if (gpa
== UNMAPPED_GVA
)
787 return gpa_to_hpa(vcpu
, gpa
);
790 struct page
*gva_to_page(struct kvm_vcpu
*vcpu
, gva_t gva
)
792 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, gva
);
794 if (gpa
== UNMAPPED_GVA
)
796 return pfn_to_page(gpa_to_hpa(vcpu
, gpa
) >> PAGE_SHIFT
);
799 static void nonpaging_new_cr3(struct kvm_vcpu
*vcpu
)
803 static int nonpaging_map(struct kvm_vcpu
*vcpu
, gva_t v
, hpa_t p
)
805 int level
= PT32E_ROOT_LEVEL
;
806 hpa_t table_addr
= vcpu
->mmu
.root_hpa
;
809 u32 index
= PT64_INDEX(v
, level
);
813 ASSERT(VALID_PAGE(table_addr
));
814 table
= __va(table_addr
);
818 if (is_present_pte(pte
) && is_writeble_pte(pte
))
820 mark_page_dirty(vcpu
->kvm
, v
>> PAGE_SHIFT
);
821 page_header_update_slot(vcpu
->kvm
, table
, v
);
822 table
[index
] = p
| PT_PRESENT_MASK
| PT_WRITABLE_MASK
|
824 rmap_add(vcpu
, &table
[index
]);
828 if (table
[index
] == 0) {
829 struct kvm_mmu_page
*new_table
;
832 pseudo_gfn
= (v
& PT64_DIR_BASE_ADDR_MASK
)
834 new_table
= kvm_mmu_get_page(vcpu
, pseudo_gfn
,
836 1, 0, &table
[index
]);
838 pgprintk("nonpaging_map: ENOMEM\n");
842 table
[index
] = __pa(new_table
->spt
) | PT_PRESENT_MASK
843 | PT_WRITABLE_MASK
| PT_USER_MASK
;
845 table_addr
= table
[index
] & PT64_BASE_ADDR_MASK
;
849 static void mmu_free_roots(struct kvm_vcpu
*vcpu
)
852 struct kvm_mmu_page
*page
;
854 if (!VALID_PAGE(vcpu
->mmu
.root_hpa
))
857 if (vcpu
->mmu
.shadow_root_level
== PT64_ROOT_LEVEL
) {
858 hpa_t root
= vcpu
->mmu
.root_hpa
;
860 page
= page_header(root
);
862 vcpu
->mmu
.root_hpa
= INVALID_PAGE
;
866 for (i
= 0; i
< 4; ++i
) {
867 hpa_t root
= vcpu
->mmu
.pae_root
[i
];
870 root
&= PT64_BASE_ADDR_MASK
;
871 page
= page_header(root
);
874 vcpu
->mmu
.pae_root
[i
] = INVALID_PAGE
;
876 vcpu
->mmu
.root_hpa
= INVALID_PAGE
;
879 static void mmu_alloc_roots(struct kvm_vcpu
*vcpu
)
883 struct kvm_mmu_page
*page
;
885 root_gfn
= vcpu
->cr3
>> PAGE_SHIFT
;
888 if (vcpu
->mmu
.shadow_root_level
== PT64_ROOT_LEVEL
) {
889 hpa_t root
= vcpu
->mmu
.root_hpa
;
891 ASSERT(!VALID_PAGE(root
));
892 page
= kvm_mmu_get_page(vcpu
, root_gfn
, 0,
893 PT64_ROOT_LEVEL
, 0, 0, NULL
);
894 root
= __pa(page
->spt
);
896 vcpu
->mmu
.root_hpa
= root
;
900 for (i
= 0; i
< 4; ++i
) {
901 hpa_t root
= vcpu
->mmu
.pae_root
[i
];
903 ASSERT(!VALID_PAGE(root
));
904 if (vcpu
->mmu
.root_level
== PT32E_ROOT_LEVEL
) {
905 if (!is_present_pte(vcpu
->pdptrs
[i
])) {
906 vcpu
->mmu
.pae_root
[i
] = 0;
909 root_gfn
= vcpu
->pdptrs
[i
] >> PAGE_SHIFT
;
910 } else if (vcpu
->mmu
.root_level
== 0)
912 page
= kvm_mmu_get_page(vcpu
, root_gfn
, i
<< 30,
913 PT32_ROOT_LEVEL
, !is_paging(vcpu
),
915 root
= __pa(page
->spt
);
917 vcpu
->mmu
.pae_root
[i
] = root
| PT_PRESENT_MASK
;
919 vcpu
->mmu
.root_hpa
= __pa(vcpu
->mmu
.pae_root
);
922 static gpa_t
nonpaging_gva_to_gpa(struct kvm_vcpu
*vcpu
, gva_t vaddr
)
927 static int nonpaging_page_fault(struct kvm_vcpu
*vcpu
, gva_t gva
,
934 r
= mmu_topup_memory_caches(vcpu
);
939 ASSERT(VALID_PAGE(vcpu
->mmu
.root_hpa
));
942 paddr
= gpa_to_hpa(vcpu
, addr
& PT64_BASE_ADDR_MASK
);
944 if (is_error_hpa(paddr
))
947 return nonpaging_map(vcpu
, addr
& PAGE_MASK
, paddr
);
950 static void nonpaging_free(struct kvm_vcpu
*vcpu
)
952 mmu_free_roots(vcpu
);
955 static int nonpaging_init_context(struct kvm_vcpu
*vcpu
)
957 struct kvm_mmu
*context
= &vcpu
->mmu
;
959 context
->new_cr3
= nonpaging_new_cr3
;
960 context
->page_fault
= nonpaging_page_fault
;
961 context
->gva_to_gpa
= nonpaging_gva_to_gpa
;
962 context
->free
= nonpaging_free
;
963 context
->root_level
= 0;
964 context
->shadow_root_level
= PT32E_ROOT_LEVEL
;
965 context
->root_hpa
= INVALID_PAGE
;
969 static void kvm_mmu_flush_tlb(struct kvm_vcpu
*vcpu
)
971 ++vcpu
->stat
.tlb_flush
;
972 kvm_arch_ops
->tlb_flush(vcpu
);
975 static void paging_new_cr3(struct kvm_vcpu
*vcpu
)
977 pgprintk("%s: cr3 %lx\n", __FUNCTION__
, vcpu
->cr3
);
978 mmu_free_roots(vcpu
);
981 static void inject_page_fault(struct kvm_vcpu
*vcpu
,
985 kvm_arch_ops
->inject_page_fault(vcpu
, addr
, err_code
);
988 static void paging_free(struct kvm_vcpu
*vcpu
)
990 nonpaging_free(vcpu
);
994 #include "paging_tmpl.h"
998 #include "paging_tmpl.h"
1001 static int paging64_init_context_common(struct kvm_vcpu
*vcpu
, int level
)
1003 struct kvm_mmu
*context
= &vcpu
->mmu
;
1005 ASSERT(is_pae(vcpu
));
1006 context
->new_cr3
= paging_new_cr3
;
1007 context
->page_fault
= paging64_page_fault
;
1008 context
->gva_to_gpa
= paging64_gva_to_gpa
;
1009 context
->free
= paging_free
;
1010 context
->root_level
= level
;
1011 context
->shadow_root_level
= level
;
1012 context
->root_hpa
= INVALID_PAGE
;
1016 static int paging64_init_context(struct kvm_vcpu
*vcpu
)
1018 return paging64_init_context_common(vcpu
, PT64_ROOT_LEVEL
);
1021 static int paging32_init_context(struct kvm_vcpu
*vcpu
)
1023 struct kvm_mmu
*context
= &vcpu
->mmu
;
1025 context
->new_cr3
= paging_new_cr3
;
1026 context
->page_fault
= paging32_page_fault
;
1027 context
->gva_to_gpa
= paging32_gva_to_gpa
;
1028 context
->free
= paging_free
;
1029 context
->root_level
= PT32_ROOT_LEVEL
;
1030 context
->shadow_root_level
= PT32E_ROOT_LEVEL
;
1031 context
->root_hpa
= INVALID_PAGE
;
1035 static int paging32E_init_context(struct kvm_vcpu
*vcpu
)
1037 return paging64_init_context_common(vcpu
, PT32E_ROOT_LEVEL
);
1040 static int init_kvm_mmu(struct kvm_vcpu
*vcpu
)
1043 ASSERT(!VALID_PAGE(vcpu
->mmu
.root_hpa
));
1045 if (!is_paging(vcpu
))
1046 return nonpaging_init_context(vcpu
);
1047 else if (is_long_mode(vcpu
))
1048 return paging64_init_context(vcpu
);
1049 else if (is_pae(vcpu
))
1050 return paging32E_init_context(vcpu
);
1052 return paging32_init_context(vcpu
);
1055 static void destroy_kvm_mmu(struct kvm_vcpu
*vcpu
)
1058 if (VALID_PAGE(vcpu
->mmu
.root_hpa
)) {
1059 vcpu
->mmu
.free(vcpu
);
1060 vcpu
->mmu
.root_hpa
= INVALID_PAGE
;
1064 int kvm_mmu_reset_context(struct kvm_vcpu
*vcpu
)
1066 destroy_kvm_mmu(vcpu
);
1067 return init_kvm_mmu(vcpu
);
1070 int kvm_mmu_load(struct kvm_vcpu
*vcpu
)
1074 spin_lock(&vcpu
->kvm
->lock
);
1075 r
= mmu_topup_memory_caches(vcpu
);
1078 mmu_alloc_roots(vcpu
);
1079 kvm_arch_ops
->set_cr3(vcpu
, vcpu
->mmu
.root_hpa
);
1080 kvm_mmu_flush_tlb(vcpu
);
1082 spin_unlock(&vcpu
->kvm
->lock
);
1085 EXPORT_SYMBOL_GPL(kvm_mmu_load
);
1087 void kvm_mmu_unload(struct kvm_vcpu
*vcpu
)
1089 mmu_free_roots(vcpu
);
1092 static void mmu_pte_write_zap_pte(struct kvm_vcpu
*vcpu
,
1093 struct kvm_mmu_page
*page
,
1097 struct kvm_mmu_page
*child
;
1100 if (is_present_pte(pte
)) {
1101 if (page
->role
.level
== PT_PAGE_TABLE_LEVEL
)
1104 child
= page_header(pte
& PT64_BASE_ADDR_MASK
);
1105 mmu_page_remove_parent_pte(child
, spte
);
1109 kvm_flush_remote_tlbs(vcpu
->kvm
);
1112 static void mmu_pte_write_new_pte(struct kvm_vcpu
*vcpu
,
1113 struct kvm_mmu_page
*page
,
1115 const void *new, int bytes
)
1117 if (page
->role
.level
!= PT_PAGE_TABLE_LEVEL
)
1120 if (page
->role
.glevels
== PT32_ROOT_LEVEL
)
1121 paging32_update_pte(vcpu
, page
, spte
, new, bytes
);
1123 paging64_update_pte(vcpu
, page
, spte
, new, bytes
);
1126 void kvm_mmu_pte_write(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1127 const u8
*old
, const u8
*new, int bytes
)
1129 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1130 struct kvm_mmu_page
*page
;
1131 struct hlist_node
*node
, *n
;
1132 struct hlist_head
*bucket
;
1135 unsigned offset
= offset_in_page(gpa
);
1137 unsigned page_offset
;
1138 unsigned misaligned
;
1144 pgprintk("%s: gpa %llx bytes %d\n", __FUNCTION__
, gpa
, bytes
);
1145 if (gfn
== vcpu
->last_pt_write_gfn
) {
1146 ++vcpu
->last_pt_write_count
;
1147 if (vcpu
->last_pt_write_count
>= 3)
1150 vcpu
->last_pt_write_gfn
= gfn
;
1151 vcpu
->last_pt_write_count
= 1;
1153 index
= kvm_page_table_hashfn(gfn
) % KVM_NUM_MMU_PAGES
;
1154 bucket
= &vcpu
->kvm
->mmu_page_hash
[index
];
1155 hlist_for_each_entry_safe(page
, node
, n
, bucket
, hash_link
) {
1156 if (page
->gfn
!= gfn
|| page
->role
.metaphysical
)
1158 pte_size
= page
->role
.glevels
== PT32_ROOT_LEVEL
? 4 : 8;
1159 misaligned
= (offset
^ (offset
+ bytes
- 1)) & ~(pte_size
- 1);
1160 misaligned
|= bytes
< 4;
1161 if (misaligned
|| flooded
) {
1163 * Misaligned accesses are too much trouble to fix
1164 * up; also, they usually indicate a page is not used
1167 * If we're seeing too many writes to a page,
1168 * it may no longer be a page table, or we may be
1169 * forking, in which case it is better to unmap the
1172 pgprintk("misaligned: gpa %llx bytes %d role %x\n",
1173 gpa
, bytes
, page
->role
.word
);
1174 kvm_mmu_zap_page(vcpu
->kvm
, page
);
1177 page_offset
= offset
;
1178 level
= page
->role
.level
;
1180 if (page
->role
.glevels
== PT32_ROOT_LEVEL
) {
1181 page_offset
<<= 1; /* 32->64 */
1183 * A 32-bit pde maps 4MB while the shadow pdes map
1184 * only 2MB. So we need to double the offset again
1185 * and zap two pdes instead of one.
1187 if (level
== PT32_ROOT_LEVEL
) {
1188 page_offset
&= ~7; /* kill rounding error */
1192 quadrant
= page_offset
>> PAGE_SHIFT
;
1193 page_offset
&= ~PAGE_MASK
;
1194 if (quadrant
!= page
->role
.quadrant
)
1197 spte
= &page
->spt
[page_offset
/ sizeof(*spte
)];
1199 mmu_pte_write_zap_pte(vcpu
, page
, spte
);
1200 mmu_pte_write_new_pte(vcpu
, page
, spte
, new, bytes
);
1206 int kvm_mmu_unprotect_page_virt(struct kvm_vcpu
*vcpu
, gva_t gva
)
1208 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, gva
);
1210 return kvm_mmu_unprotect_page(vcpu
, gpa
>> PAGE_SHIFT
);
1213 void __kvm_mmu_free_some_pages(struct kvm_vcpu
*vcpu
)
1215 while (vcpu
->kvm
->n_free_mmu_pages
< KVM_REFILL_PAGES
) {
1216 struct kvm_mmu_page
*page
;
1218 page
= container_of(vcpu
->kvm
->active_mmu_pages
.prev
,
1219 struct kvm_mmu_page
, link
);
1220 kvm_mmu_zap_page(vcpu
->kvm
, page
);
1224 static void free_mmu_pages(struct kvm_vcpu
*vcpu
)
1226 struct kvm_mmu_page
*page
;
1228 while (!list_empty(&vcpu
->kvm
->active_mmu_pages
)) {
1229 page
= container_of(vcpu
->kvm
->active_mmu_pages
.next
,
1230 struct kvm_mmu_page
, link
);
1231 kvm_mmu_zap_page(vcpu
->kvm
, page
);
1233 free_page((unsigned long)vcpu
->mmu
.pae_root
);
1236 static int alloc_mmu_pages(struct kvm_vcpu
*vcpu
)
1243 vcpu
->kvm
->n_free_mmu_pages
= KVM_NUM_MMU_PAGES
;
1246 * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64.
1247 * Therefore we need to allocate shadow page tables in the first
1248 * 4GB of memory, which happens to fit the DMA32 zone.
1250 page
= alloc_page(GFP_KERNEL
| __GFP_DMA32
);
1253 vcpu
->mmu
.pae_root
= page_address(page
);
1254 for (i
= 0; i
< 4; ++i
)
1255 vcpu
->mmu
.pae_root
[i
] = INVALID_PAGE
;
1260 free_mmu_pages(vcpu
);
1264 int kvm_mmu_create(struct kvm_vcpu
*vcpu
)
1267 ASSERT(!VALID_PAGE(vcpu
->mmu
.root_hpa
));
1269 return alloc_mmu_pages(vcpu
);
1272 int kvm_mmu_setup(struct kvm_vcpu
*vcpu
)
1275 ASSERT(!VALID_PAGE(vcpu
->mmu
.root_hpa
));
1277 return init_kvm_mmu(vcpu
);
1280 void kvm_mmu_destroy(struct kvm_vcpu
*vcpu
)
1284 destroy_kvm_mmu(vcpu
);
1285 free_mmu_pages(vcpu
);
1286 mmu_free_memory_caches(vcpu
);
1289 void kvm_mmu_slot_remove_write_access(struct kvm
*kvm
, int slot
)
1291 struct kvm_mmu_page
*page
;
1293 list_for_each_entry(page
, &kvm
->active_mmu_pages
, link
) {
1297 if (!test_bit(slot
, &page
->slot_bitmap
))
1301 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
)
1303 if (pt
[i
] & PT_WRITABLE_MASK
) {
1304 rmap_remove(&pt
[i
]);
1305 pt
[i
] &= ~PT_WRITABLE_MASK
;
1310 void kvm_mmu_zap_all(struct kvm
*kvm
)
1312 struct kvm_mmu_page
*page
, *node
;
1314 list_for_each_entry_safe(page
, node
, &kvm
->active_mmu_pages
, link
)
1315 kvm_mmu_zap_page(kvm
, page
);
1317 kvm_flush_remote_tlbs(kvm
);
1320 void kvm_mmu_module_exit(void)
1322 if (pte_chain_cache
)
1323 kmem_cache_destroy(pte_chain_cache
);
1324 if (rmap_desc_cache
)
1325 kmem_cache_destroy(rmap_desc_cache
);
1326 if (mmu_page_header_cache
)
1327 kmem_cache_destroy(mmu_page_header_cache
);
1330 int kvm_mmu_module_init(void)
1332 pte_chain_cache
= kmem_cache_create("kvm_pte_chain",
1333 sizeof(struct kvm_pte_chain
),
1335 if (!pte_chain_cache
)
1337 rmap_desc_cache
= kmem_cache_create("kvm_rmap_desc",
1338 sizeof(struct kvm_rmap_desc
),
1340 if (!rmap_desc_cache
)
1343 mmu_page_header_cache
= kmem_cache_create("kvm_mmu_page_header",
1344 sizeof(struct kvm_mmu_page
),
1346 if (!mmu_page_header_cache
)
1352 kvm_mmu_module_exit();
1358 static const char *audit_msg
;
1360 static gva_t
canonicalize(gva_t gva
)
1362 #ifdef CONFIG_X86_64
1363 gva
= (long long)(gva
<< 16) >> 16;
1368 static void audit_mappings_page(struct kvm_vcpu
*vcpu
, u64 page_pte
,
1369 gva_t va
, int level
)
1371 u64
*pt
= __va(page_pte
& PT64_BASE_ADDR_MASK
);
1373 gva_t va_delta
= 1ul << (PAGE_SHIFT
+ 9 * (level
- 1));
1375 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
, va
+= va_delta
) {
1378 if (!(ent
& PT_PRESENT_MASK
))
1381 va
= canonicalize(va
);
1383 audit_mappings_page(vcpu
, ent
, va
, level
- 1);
1385 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, va
);
1386 hpa_t hpa
= gpa_to_hpa(vcpu
, gpa
);
1388 if ((ent
& PT_PRESENT_MASK
)
1389 && (ent
& PT64_BASE_ADDR_MASK
) != hpa
)
1390 printk(KERN_ERR
"audit error: (%s) levels %d"
1391 " gva %lx gpa %llx hpa %llx ent %llx\n",
1392 audit_msg
, vcpu
->mmu
.root_level
,
1398 static void audit_mappings(struct kvm_vcpu
*vcpu
)
1402 if (vcpu
->mmu
.root_level
== 4)
1403 audit_mappings_page(vcpu
, vcpu
->mmu
.root_hpa
, 0, 4);
1405 for (i
= 0; i
< 4; ++i
)
1406 if (vcpu
->mmu
.pae_root
[i
] & PT_PRESENT_MASK
)
1407 audit_mappings_page(vcpu
,
1408 vcpu
->mmu
.pae_root
[i
],
1413 static int count_rmaps(struct kvm_vcpu
*vcpu
)
1418 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
1419 struct kvm_memory_slot
*m
= &vcpu
->kvm
->memslots
[i
];
1420 struct kvm_rmap_desc
*d
;
1422 for (j
= 0; j
< m
->npages
; ++j
) {
1423 struct page
*page
= m
->phys_mem
[j
];
1427 if (!(page
->private & 1)) {
1431 d
= (struct kvm_rmap_desc
*)(page
->private & ~1ul);
1433 for (k
= 0; k
< RMAP_EXT
; ++k
)
1434 if (d
->shadow_ptes
[k
])
1445 static int count_writable_mappings(struct kvm_vcpu
*vcpu
)
1448 struct kvm_mmu_page
*page
;
1451 list_for_each_entry(page
, &vcpu
->kvm
->active_mmu_pages
, link
) {
1452 u64
*pt
= page
->spt
;
1454 if (page
->role
.level
!= PT_PAGE_TABLE_LEVEL
)
1457 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
1460 if (!(ent
& PT_PRESENT_MASK
))
1462 if (!(ent
& PT_WRITABLE_MASK
))
1470 static void audit_rmap(struct kvm_vcpu
*vcpu
)
1472 int n_rmap
= count_rmaps(vcpu
);
1473 int n_actual
= count_writable_mappings(vcpu
);
1475 if (n_rmap
!= n_actual
)
1476 printk(KERN_ERR
"%s: (%s) rmap %d actual %d\n",
1477 __FUNCTION__
, audit_msg
, n_rmap
, n_actual
);
1480 static void audit_write_protection(struct kvm_vcpu
*vcpu
)
1482 struct kvm_mmu_page
*page
;
1484 list_for_each_entry(page
, &vcpu
->kvm
->active_mmu_pages
, link
) {
1488 if (page
->role
.metaphysical
)
1491 hfn
= gpa_to_hpa(vcpu
, (gpa_t
)page
->gfn
<< PAGE_SHIFT
)
1493 pg
= pfn_to_page(hfn
);
1495 printk(KERN_ERR
"%s: (%s) shadow page has writable"
1496 " mappings: gfn %lx role %x\n",
1497 __FUNCTION__
, audit_msg
, page
->gfn
,
1502 static void kvm_mmu_audit(struct kvm_vcpu
*vcpu
, const char *msg
)
1509 audit_write_protection(vcpu
);
1510 audit_mappings(vcpu
);