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.
25 #include <linux/types.h>
26 #include <linux/string.h>
28 #include <linux/highmem.h>
29 #include <linux/module.h>
30 #include <linux/swap.h>
33 #include <asm/cmpxchg.h>
41 static void kvm_mmu_audit(struct kvm_vcpu
*vcpu
, const char *msg
);
43 static void kvm_mmu_audit(struct kvm_vcpu
*vcpu
, const char *msg
) {}
48 #define pgprintk(x...) do { if (dbg) printk(x); } while (0)
49 #define rmap_printk(x...) do { if (dbg) printk(x); } while (0)
53 #define pgprintk(x...) do { } while (0)
54 #define rmap_printk(x...) do { } while (0)
58 #if defined(MMU_DEBUG) || defined(AUDIT)
63 #define ASSERT(x) do { } while (0)
67 printk(KERN_WARNING "assertion failed %s:%d: %s\n", \
68 __FILE__, __LINE__, #x); \
72 #define PT64_PT_BITS 9
73 #define PT64_ENT_PER_PAGE (1 << PT64_PT_BITS)
74 #define PT32_PT_BITS 10
75 #define PT32_ENT_PER_PAGE (1 << PT32_PT_BITS)
77 #define PT_WRITABLE_SHIFT 1
79 #define PT_PRESENT_MASK (1ULL << 0)
80 #define PT_WRITABLE_MASK (1ULL << PT_WRITABLE_SHIFT)
81 #define PT_USER_MASK (1ULL << 2)
82 #define PT_PWT_MASK (1ULL << 3)
83 #define PT_PCD_MASK (1ULL << 4)
84 #define PT_ACCESSED_MASK (1ULL << 5)
85 #define PT_DIRTY_MASK (1ULL << 6)
86 #define PT_PAGE_SIZE_MASK (1ULL << 7)
87 #define PT_PAT_MASK (1ULL << 7)
88 #define PT_GLOBAL_MASK (1ULL << 8)
89 #define PT64_NX_SHIFT 63
90 #define PT64_NX_MASK (1ULL << PT64_NX_SHIFT)
92 #define PT_PAT_SHIFT 7
93 #define PT_DIR_PAT_SHIFT 12
94 #define PT_DIR_PAT_MASK (1ULL << PT_DIR_PAT_SHIFT)
96 #define PT32_DIR_PSE36_SIZE 4
97 #define PT32_DIR_PSE36_SHIFT 13
98 #define PT32_DIR_PSE36_MASK \
99 (((1ULL << PT32_DIR_PSE36_SIZE) - 1) << PT32_DIR_PSE36_SHIFT)
102 #define PT_FIRST_AVAIL_BITS_SHIFT 9
103 #define PT64_SECOND_AVAIL_BITS_SHIFT 52
105 #define PT_SHADOW_IO_MARK (1ULL << PT_FIRST_AVAIL_BITS_SHIFT)
107 #define VALID_PAGE(x) ((x) != INVALID_PAGE)
109 #define PT64_LEVEL_BITS 9
111 #define PT64_LEVEL_SHIFT(level) \
112 (PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS)
114 #define PT64_LEVEL_MASK(level) \
115 (((1ULL << PT64_LEVEL_BITS) - 1) << PT64_LEVEL_SHIFT(level))
117 #define PT64_INDEX(address, level)\
118 (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1))
121 #define PT32_LEVEL_BITS 10
123 #define PT32_LEVEL_SHIFT(level) \
124 (PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS)
126 #define PT32_LEVEL_MASK(level) \
127 (((1ULL << PT32_LEVEL_BITS) - 1) << PT32_LEVEL_SHIFT(level))
129 #define PT32_INDEX(address, level)\
130 (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1))
133 #define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1))
134 #define PT64_DIR_BASE_ADDR_MASK \
135 (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1))
137 #define PT32_BASE_ADDR_MASK PAGE_MASK
138 #define PT32_DIR_BASE_ADDR_MASK \
139 (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1))
141 #define PT64_PERM_MASK (PT_PRESENT_MASK | PT_WRITABLE_MASK | PT_USER_MASK \
144 #define PFERR_PRESENT_MASK (1U << 0)
145 #define PFERR_WRITE_MASK (1U << 1)
146 #define PFERR_USER_MASK (1U << 2)
147 #define PFERR_FETCH_MASK (1U << 4)
149 #define PT64_ROOT_LEVEL 4
150 #define PT32_ROOT_LEVEL 2
151 #define PT32E_ROOT_LEVEL 3
153 #define PT_DIRECTORY_LEVEL 2
154 #define PT_PAGE_TABLE_LEVEL 1
158 #define ACC_EXEC_MASK 1
159 #define ACC_WRITE_MASK PT_WRITABLE_MASK
160 #define ACC_USER_MASK PT_USER_MASK
161 #define ACC_ALL (ACC_EXEC_MASK | ACC_WRITE_MASK | ACC_USER_MASK)
163 struct kvm_rmap_desc
{
164 u64
*shadow_ptes
[RMAP_EXT
];
165 struct kvm_rmap_desc
*more
;
168 static struct kmem_cache
*pte_chain_cache
;
169 static struct kmem_cache
*rmap_desc_cache
;
170 static struct kmem_cache
*mmu_page_header_cache
;
172 static u64 __read_mostly shadow_trap_nonpresent_pte
;
173 static u64 __read_mostly shadow_notrap_nonpresent_pte
;
175 void kvm_mmu_set_nonpresent_ptes(u64 trap_pte
, u64 notrap_pte
)
177 shadow_trap_nonpresent_pte
= trap_pte
;
178 shadow_notrap_nonpresent_pte
= notrap_pte
;
180 EXPORT_SYMBOL_GPL(kvm_mmu_set_nonpresent_ptes
);
182 static int is_write_protection(struct kvm_vcpu
*vcpu
)
184 return vcpu
->arch
.cr0
& X86_CR0_WP
;
187 static int is_cpuid_PSE36(void)
192 static int is_nx(struct kvm_vcpu
*vcpu
)
194 return vcpu
->arch
.shadow_efer
& EFER_NX
;
197 static int is_present_pte(unsigned long pte
)
199 return pte
& PT_PRESENT_MASK
;
202 static int is_shadow_present_pte(u64 pte
)
204 pte
&= ~PT_SHADOW_IO_MARK
;
205 return pte
!= shadow_trap_nonpresent_pte
206 && pte
!= shadow_notrap_nonpresent_pte
;
209 static int is_writeble_pte(unsigned long pte
)
211 return pte
& PT_WRITABLE_MASK
;
214 static int is_dirty_pte(unsigned long pte
)
216 return pte
& PT_DIRTY_MASK
;
219 static int is_io_pte(unsigned long pte
)
221 return pte
& PT_SHADOW_IO_MARK
;
224 static int is_rmap_pte(u64 pte
)
226 return pte
!= shadow_trap_nonpresent_pte
227 && pte
!= shadow_notrap_nonpresent_pte
;
230 static gfn_t
pse36_gfn_delta(u32 gpte
)
232 int shift
= 32 - PT32_DIR_PSE36_SHIFT
- PAGE_SHIFT
;
234 return (gpte
& PT32_DIR_PSE36_MASK
) << shift
;
237 static void set_shadow_pte(u64
*sptep
, u64 spte
)
240 set_64bit((unsigned long *)sptep
, spte
);
242 set_64bit((unsigned long long *)sptep
, spte
);
246 static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache
*cache
,
247 struct kmem_cache
*base_cache
, int min
)
251 if (cache
->nobjs
>= min
)
253 while (cache
->nobjs
< ARRAY_SIZE(cache
->objects
)) {
254 obj
= kmem_cache_zalloc(base_cache
, GFP_KERNEL
);
257 cache
->objects
[cache
->nobjs
++] = obj
;
262 static void mmu_free_memory_cache(struct kvm_mmu_memory_cache
*mc
)
265 kfree(mc
->objects
[--mc
->nobjs
]);
268 static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache
*cache
,
273 if (cache
->nobjs
>= min
)
275 while (cache
->nobjs
< ARRAY_SIZE(cache
->objects
)) {
276 page
= alloc_page(GFP_KERNEL
);
279 set_page_private(page
, 0);
280 cache
->objects
[cache
->nobjs
++] = page_address(page
);
285 static void mmu_free_memory_cache_page(struct kvm_mmu_memory_cache
*mc
)
288 free_page((unsigned long)mc
->objects
[--mc
->nobjs
]);
291 static int mmu_topup_memory_caches(struct kvm_vcpu
*vcpu
)
295 kvm_mmu_free_some_pages(vcpu
);
296 r
= mmu_topup_memory_cache(&vcpu
->arch
.mmu_pte_chain_cache
,
300 r
= mmu_topup_memory_cache(&vcpu
->arch
.mmu_rmap_desc_cache
,
304 r
= mmu_topup_memory_cache_page(&vcpu
->arch
.mmu_page_cache
, 8);
307 r
= mmu_topup_memory_cache(&vcpu
->arch
.mmu_page_header_cache
,
308 mmu_page_header_cache
, 4);
313 static void mmu_free_memory_caches(struct kvm_vcpu
*vcpu
)
315 mmu_free_memory_cache(&vcpu
->arch
.mmu_pte_chain_cache
);
316 mmu_free_memory_cache(&vcpu
->arch
.mmu_rmap_desc_cache
);
317 mmu_free_memory_cache_page(&vcpu
->arch
.mmu_page_cache
);
318 mmu_free_memory_cache(&vcpu
->arch
.mmu_page_header_cache
);
321 static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache
*mc
,
327 p
= mc
->objects
[--mc
->nobjs
];
332 static struct kvm_pte_chain
*mmu_alloc_pte_chain(struct kvm_vcpu
*vcpu
)
334 return mmu_memory_cache_alloc(&vcpu
->arch
.mmu_pte_chain_cache
,
335 sizeof(struct kvm_pte_chain
));
338 static void mmu_free_pte_chain(struct kvm_pte_chain
*pc
)
343 static struct kvm_rmap_desc
*mmu_alloc_rmap_desc(struct kvm_vcpu
*vcpu
)
345 return mmu_memory_cache_alloc(&vcpu
->arch
.mmu_rmap_desc_cache
,
346 sizeof(struct kvm_rmap_desc
));
349 static void mmu_free_rmap_desc(struct kvm_rmap_desc
*rd
)
355 * Take gfn and return the reverse mapping to it.
356 * Note: gfn must be unaliased before this function get called
359 static unsigned long *gfn_to_rmap(struct kvm
*kvm
, gfn_t gfn
)
361 struct kvm_memory_slot
*slot
;
363 slot
= gfn_to_memslot(kvm
, gfn
);
364 return &slot
->rmap
[gfn
- slot
->base_gfn
];
368 * Reverse mapping data structures:
370 * If rmapp bit zero is zero, then rmapp point to the shadw page table entry
371 * that points to page_address(page).
373 * If rmapp bit zero is one, (then rmap & ~1) points to a struct kvm_rmap_desc
374 * containing more mappings.
376 static void rmap_add(struct kvm_vcpu
*vcpu
, u64
*spte
, gfn_t gfn
)
378 struct kvm_mmu_page
*sp
;
379 struct kvm_rmap_desc
*desc
;
380 unsigned long *rmapp
;
383 if (!is_rmap_pte(*spte
))
385 gfn
= unalias_gfn(vcpu
->kvm
, gfn
);
386 sp
= page_header(__pa(spte
));
387 sp
->gfns
[spte
- sp
->spt
] = gfn
;
388 rmapp
= gfn_to_rmap(vcpu
->kvm
, gfn
);
390 rmap_printk("rmap_add: %p %llx 0->1\n", spte
, *spte
);
391 *rmapp
= (unsigned long)spte
;
392 } else if (!(*rmapp
& 1)) {
393 rmap_printk("rmap_add: %p %llx 1->many\n", spte
, *spte
);
394 desc
= mmu_alloc_rmap_desc(vcpu
);
395 desc
->shadow_ptes
[0] = (u64
*)*rmapp
;
396 desc
->shadow_ptes
[1] = spte
;
397 *rmapp
= (unsigned long)desc
| 1;
399 rmap_printk("rmap_add: %p %llx many->many\n", spte
, *spte
);
400 desc
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
401 while (desc
->shadow_ptes
[RMAP_EXT
-1] && desc
->more
)
403 if (desc
->shadow_ptes
[RMAP_EXT
-1]) {
404 desc
->more
= mmu_alloc_rmap_desc(vcpu
);
407 for (i
= 0; desc
->shadow_ptes
[i
]; ++i
)
409 desc
->shadow_ptes
[i
] = spte
;
413 static void rmap_desc_remove_entry(unsigned long *rmapp
,
414 struct kvm_rmap_desc
*desc
,
416 struct kvm_rmap_desc
*prev_desc
)
420 for (j
= RMAP_EXT
- 1; !desc
->shadow_ptes
[j
] && j
> i
; --j
)
422 desc
->shadow_ptes
[i
] = desc
->shadow_ptes
[j
];
423 desc
->shadow_ptes
[j
] = NULL
;
426 if (!prev_desc
&& !desc
->more
)
427 *rmapp
= (unsigned long)desc
->shadow_ptes
[0];
430 prev_desc
->more
= desc
->more
;
432 *rmapp
= (unsigned long)desc
->more
| 1;
433 mmu_free_rmap_desc(desc
);
436 static void rmap_remove(struct kvm
*kvm
, u64
*spte
)
438 struct kvm_rmap_desc
*desc
;
439 struct kvm_rmap_desc
*prev_desc
;
440 struct kvm_mmu_page
*sp
;
442 unsigned long *rmapp
;
445 if (!is_rmap_pte(*spte
))
447 sp
= page_header(__pa(spte
));
448 page
= pfn_to_page((*spte
& PT64_BASE_ADDR_MASK
) >> PAGE_SHIFT
);
449 mark_page_accessed(page
);
450 if (is_writeble_pte(*spte
))
451 kvm_release_page_dirty(page
);
453 kvm_release_page_clean(page
);
454 rmapp
= gfn_to_rmap(kvm
, sp
->gfns
[spte
- sp
->spt
]);
456 printk(KERN_ERR
"rmap_remove: %p %llx 0->BUG\n", spte
, *spte
);
458 } else if (!(*rmapp
& 1)) {
459 rmap_printk("rmap_remove: %p %llx 1->0\n", spte
, *spte
);
460 if ((u64
*)*rmapp
!= spte
) {
461 printk(KERN_ERR
"rmap_remove: %p %llx 1->BUG\n",
467 rmap_printk("rmap_remove: %p %llx many->many\n", spte
, *spte
);
468 desc
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
471 for (i
= 0; i
< RMAP_EXT
&& desc
->shadow_ptes
[i
]; ++i
)
472 if (desc
->shadow_ptes
[i
] == spte
) {
473 rmap_desc_remove_entry(rmapp
,
485 static u64
*rmap_next(struct kvm
*kvm
, unsigned long *rmapp
, u64
*spte
)
487 struct kvm_rmap_desc
*desc
;
488 struct kvm_rmap_desc
*prev_desc
;
494 else if (!(*rmapp
& 1)) {
496 return (u64
*)*rmapp
;
499 desc
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
503 for (i
= 0; i
< RMAP_EXT
&& desc
->shadow_ptes
[i
]; ++i
) {
504 if (prev_spte
== spte
)
505 return desc
->shadow_ptes
[i
];
506 prev_spte
= desc
->shadow_ptes
[i
];
513 static void rmap_write_protect(struct kvm
*kvm
, u64 gfn
)
515 unsigned long *rmapp
;
518 gfn
= unalias_gfn(kvm
, gfn
);
519 rmapp
= gfn_to_rmap(kvm
, gfn
);
521 spte
= rmap_next(kvm
, rmapp
, NULL
);
524 BUG_ON(!(*spte
& PT_PRESENT_MASK
));
525 rmap_printk("rmap_write_protect: spte %p %llx\n", spte
, *spte
);
526 if (is_writeble_pte(*spte
))
527 set_shadow_pte(spte
, *spte
& ~PT_WRITABLE_MASK
);
528 kvm_flush_remote_tlbs(kvm
);
529 spte
= rmap_next(kvm
, rmapp
, spte
);
534 static int is_empty_shadow_page(u64
*spt
)
539 for (pos
= spt
, end
= pos
+ PAGE_SIZE
/ sizeof(u64
); pos
!= end
; pos
++)
540 if ((*pos
& ~PT_SHADOW_IO_MARK
) != shadow_trap_nonpresent_pte
) {
541 printk(KERN_ERR
"%s: %p %llx\n", __FUNCTION__
,
549 static void kvm_mmu_free_page(struct kvm
*kvm
, struct kvm_mmu_page
*sp
)
551 ASSERT(is_empty_shadow_page(sp
->spt
));
553 __free_page(virt_to_page(sp
->spt
));
554 __free_page(virt_to_page(sp
->gfns
));
556 ++kvm
->n_free_mmu_pages
;
559 static unsigned kvm_page_table_hashfn(gfn_t gfn
)
564 static struct kvm_mmu_page
*kvm_mmu_alloc_page(struct kvm_vcpu
*vcpu
,
567 struct kvm_mmu_page
*sp
;
569 if (!vcpu
->kvm
->n_free_mmu_pages
)
572 sp
= mmu_memory_cache_alloc(&vcpu
->arch
.mmu_page_header_cache
, sizeof *sp
);
573 sp
->spt
= mmu_memory_cache_alloc(&vcpu
->arch
.mmu_page_cache
, PAGE_SIZE
);
574 sp
->gfns
= mmu_memory_cache_alloc(&vcpu
->arch
.mmu_page_cache
, PAGE_SIZE
);
575 set_page_private(virt_to_page(sp
->spt
), (unsigned long)sp
);
576 list_add(&sp
->link
, &vcpu
->kvm
->active_mmu_pages
);
577 ASSERT(is_empty_shadow_page(sp
->spt
));
580 sp
->parent_pte
= parent_pte
;
581 --vcpu
->kvm
->n_free_mmu_pages
;
585 static void mmu_page_add_parent_pte(struct kvm_vcpu
*vcpu
,
586 struct kvm_mmu_page
*sp
, u64
*parent_pte
)
588 struct kvm_pte_chain
*pte_chain
;
589 struct hlist_node
*node
;
594 if (!sp
->multimapped
) {
595 u64
*old
= sp
->parent_pte
;
598 sp
->parent_pte
= parent_pte
;
602 pte_chain
= mmu_alloc_pte_chain(vcpu
);
603 INIT_HLIST_HEAD(&sp
->parent_ptes
);
604 hlist_add_head(&pte_chain
->link
, &sp
->parent_ptes
);
605 pte_chain
->parent_ptes
[0] = old
;
607 hlist_for_each_entry(pte_chain
, node
, &sp
->parent_ptes
, link
) {
608 if (pte_chain
->parent_ptes
[NR_PTE_CHAIN_ENTRIES
-1])
610 for (i
= 0; i
< NR_PTE_CHAIN_ENTRIES
; ++i
)
611 if (!pte_chain
->parent_ptes
[i
]) {
612 pte_chain
->parent_ptes
[i
] = parent_pte
;
616 pte_chain
= mmu_alloc_pte_chain(vcpu
);
618 hlist_add_head(&pte_chain
->link
, &sp
->parent_ptes
);
619 pte_chain
->parent_ptes
[0] = parent_pte
;
622 static void mmu_page_remove_parent_pte(struct kvm_mmu_page
*sp
,
625 struct kvm_pte_chain
*pte_chain
;
626 struct hlist_node
*node
;
629 if (!sp
->multimapped
) {
630 BUG_ON(sp
->parent_pte
!= parent_pte
);
631 sp
->parent_pte
= NULL
;
634 hlist_for_each_entry(pte_chain
, node
, &sp
->parent_ptes
, link
)
635 for (i
= 0; i
< NR_PTE_CHAIN_ENTRIES
; ++i
) {
636 if (!pte_chain
->parent_ptes
[i
])
638 if (pte_chain
->parent_ptes
[i
] != parent_pte
)
640 while (i
+ 1 < NR_PTE_CHAIN_ENTRIES
641 && pte_chain
->parent_ptes
[i
+ 1]) {
642 pte_chain
->parent_ptes
[i
]
643 = pte_chain
->parent_ptes
[i
+ 1];
646 pte_chain
->parent_ptes
[i
] = NULL
;
648 hlist_del(&pte_chain
->link
);
649 mmu_free_pte_chain(pte_chain
);
650 if (hlist_empty(&sp
->parent_ptes
)) {
652 sp
->parent_pte
= NULL
;
660 static struct kvm_mmu_page
*kvm_mmu_lookup_page(struct kvm
*kvm
, gfn_t gfn
)
663 struct hlist_head
*bucket
;
664 struct kvm_mmu_page
*sp
;
665 struct hlist_node
*node
;
667 pgprintk("%s: looking for gfn %lx\n", __FUNCTION__
, gfn
);
668 index
= kvm_page_table_hashfn(gfn
) % KVM_NUM_MMU_PAGES
;
669 bucket
= &kvm
->mmu_page_hash
[index
];
670 hlist_for_each_entry(sp
, node
, bucket
, hash_link
)
671 if (sp
->gfn
== gfn
&& !sp
->role
.metaphysical
) {
672 pgprintk("%s: found role %x\n",
673 __FUNCTION__
, sp
->role
.word
);
679 static struct kvm_mmu_page
*kvm_mmu_get_page(struct kvm_vcpu
*vcpu
,
688 union kvm_mmu_page_role role
;
691 struct hlist_head
*bucket
;
692 struct kvm_mmu_page
*sp
;
693 struct hlist_node
*node
;
696 role
.glevels
= vcpu
->arch
.mmu
.root_level
;
698 role
.metaphysical
= metaphysical
;
699 role
.access
= access
;
700 if (vcpu
->arch
.mmu
.root_level
<= PT32_ROOT_LEVEL
) {
701 quadrant
= gaddr
>> (PAGE_SHIFT
+ (PT64_PT_BITS
* level
));
702 quadrant
&= (1 << ((PT32_PT_BITS
- PT64_PT_BITS
) * level
)) - 1;
703 role
.quadrant
= quadrant
;
705 pgprintk("%s: looking gfn %lx role %x\n", __FUNCTION__
,
707 index
= kvm_page_table_hashfn(gfn
) % KVM_NUM_MMU_PAGES
;
708 bucket
= &vcpu
->kvm
->mmu_page_hash
[index
];
709 hlist_for_each_entry(sp
, node
, bucket
, hash_link
)
710 if (sp
->gfn
== gfn
&& sp
->role
.word
== role
.word
) {
711 mmu_page_add_parent_pte(vcpu
, sp
, parent_pte
);
712 pgprintk("%s: found\n", __FUNCTION__
);
715 sp
= kvm_mmu_alloc_page(vcpu
, parent_pte
);
718 pgprintk("%s: adding gfn %lx role %x\n", __FUNCTION__
, gfn
, role
.word
);
721 hlist_add_head(&sp
->hash_link
, bucket
);
722 vcpu
->arch
.mmu
.prefetch_page(vcpu
, sp
);
724 rmap_write_protect(vcpu
->kvm
, gfn
);
730 static void kvm_mmu_page_unlink_children(struct kvm
*kvm
,
731 struct kvm_mmu_page
*sp
)
739 if (sp
->role
.level
== PT_PAGE_TABLE_LEVEL
) {
740 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
741 if (is_shadow_present_pte(pt
[i
]))
742 rmap_remove(kvm
, &pt
[i
]);
743 pt
[i
] = shadow_trap_nonpresent_pte
;
745 kvm_flush_remote_tlbs(kvm
);
749 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
752 pt
[i
] = shadow_trap_nonpresent_pte
;
753 if (!is_shadow_present_pte(ent
))
755 ent
&= PT64_BASE_ADDR_MASK
;
756 mmu_page_remove_parent_pte(page_header(ent
), &pt
[i
]);
758 kvm_flush_remote_tlbs(kvm
);
761 static void kvm_mmu_put_page(struct kvm_mmu_page
*sp
, u64
*parent_pte
)
763 mmu_page_remove_parent_pte(sp
, parent_pte
);
766 static void kvm_mmu_reset_last_pte_updated(struct kvm
*kvm
)
770 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
)
772 kvm
->vcpus
[i
]->arch
.last_pte_updated
= NULL
;
775 static void kvm_mmu_zap_page(struct kvm
*kvm
, struct kvm_mmu_page
*sp
)
779 ++kvm
->stat
.mmu_shadow_zapped
;
780 while (sp
->multimapped
|| sp
->parent_pte
) {
781 if (!sp
->multimapped
)
782 parent_pte
= sp
->parent_pte
;
784 struct kvm_pte_chain
*chain
;
786 chain
= container_of(sp
->parent_ptes
.first
,
787 struct kvm_pte_chain
, link
);
788 parent_pte
= chain
->parent_ptes
[0];
791 kvm_mmu_put_page(sp
, parent_pte
);
792 set_shadow_pte(parent_pte
, shadow_trap_nonpresent_pte
);
794 kvm_mmu_page_unlink_children(kvm
, sp
);
795 if (!sp
->root_count
) {
796 hlist_del(&sp
->hash_link
);
797 kvm_mmu_free_page(kvm
, sp
);
799 list_move(&sp
->link
, &kvm
->active_mmu_pages
);
800 kvm_mmu_reset_last_pte_updated(kvm
);
804 * Changing the number of mmu pages allocated to the vm
805 * Note: if kvm_nr_mmu_pages is too small, you will get dead lock
807 void kvm_mmu_change_mmu_pages(struct kvm
*kvm
, unsigned int kvm_nr_mmu_pages
)
810 * If we set the number of mmu pages to be smaller be than the
811 * number of actived pages , we must to free some mmu pages before we
815 if ((kvm
->n_alloc_mmu_pages
- kvm
->n_free_mmu_pages
) >
817 int n_used_mmu_pages
= kvm
->n_alloc_mmu_pages
818 - kvm
->n_free_mmu_pages
;
820 while (n_used_mmu_pages
> kvm_nr_mmu_pages
) {
821 struct kvm_mmu_page
*page
;
823 page
= container_of(kvm
->active_mmu_pages
.prev
,
824 struct kvm_mmu_page
, link
);
825 kvm_mmu_zap_page(kvm
, page
);
828 kvm
->n_free_mmu_pages
= 0;
831 kvm
->n_free_mmu_pages
+= kvm_nr_mmu_pages
832 - kvm
->n_alloc_mmu_pages
;
834 kvm
->n_alloc_mmu_pages
= kvm_nr_mmu_pages
;
837 static int kvm_mmu_unprotect_page(struct kvm
*kvm
, gfn_t gfn
)
840 struct hlist_head
*bucket
;
841 struct kvm_mmu_page
*sp
;
842 struct hlist_node
*node
, *n
;
845 pgprintk("%s: looking for gfn %lx\n", __FUNCTION__
, gfn
);
847 index
= kvm_page_table_hashfn(gfn
) % KVM_NUM_MMU_PAGES
;
848 bucket
= &kvm
->mmu_page_hash
[index
];
849 hlist_for_each_entry_safe(sp
, node
, n
, bucket
, hash_link
)
850 if (sp
->gfn
== gfn
&& !sp
->role
.metaphysical
) {
851 pgprintk("%s: gfn %lx role %x\n", __FUNCTION__
, gfn
,
853 kvm_mmu_zap_page(kvm
, sp
);
859 static void mmu_unshadow(struct kvm
*kvm
, gfn_t gfn
)
861 struct kvm_mmu_page
*sp
;
863 while ((sp
= kvm_mmu_lookup_page(kvm
, gfn
)) != NULL
) {
864 pgprintk("%s: zap %lx %x\n", __FUNCTION__
, gfn
, sp
->role
.word
);
865 kvm_mmu_zap_page(kvm
, sp
);
869 static void page_header_update_slot(struct kvm
*kvm
, void *pte
, gfn_t gfn
)
871 int slot
= memslot_id(kvm
, gfn_to_memslot(kvm
, gfn
));
872 struct kvm_mmu_page
*sp
= page_header(__pa(pte
));
874 __set_bit(slot
, &sp
->slot_bitmap
);
877 struct page
*gva_to_page(struct kvm_vcpu
*vcpu
, gva_t gva
)
879 gpa_t gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, gva
);
881 if (gpa
== UNMAPPED_GVA
)
883 return gfn_to_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
886 static void mmu_set_spte(struct kvm_vcpu
*vcpu
, u64
*shadow_pte
,
887 unsigned pt_access
, unsigned pte_access
,
888 int user_fault
, int write_fault
, int dirty
,
889 int *ptwrite
, gfn_t gfn
)
892 int was_rmapped
= is_rmap_pte(*shadow_pte
);
895 pgprintk("%s: spte %llx access %x write_fault %d"
896 " user_fault %d gfn %lx\n",
897 __FUNCTION__
, *shadow_pte
, pt_access
,
898 write_fault
, user_fault
, gfn
);
901 * We don't set the accessed bit, since we sometimes want to see
902 * whether the guest actually used the pte (in order to detect
905 spte
= PT_PRESENT_MASK
| PT_DIRTY_MASK
;
907 pte_access
&= ~ACC_WRITE_MASK
;
908 if (!(pte_access
& ACC_EXEC_MASK
))
909 spte
|= PT64_NX_MASK
;
911 page
= gfn_to_page(vcpu
->kvm
, gfn
);
913 spte
|= PT_PRESENT_MASK
;
914 if (pte_access
& ACC_USER_MASK
)
915 spte
|= PT_USER_MASK
;
917 if (is_error_page(page
)) {
918 set_shadow_pte(shadow_pte
,
919 shadow_trap_nonpresent_pte
| PT_SHADOW_IO_MARK
);
920 kvm_release_page_clean(page
);
924 spte
|= page_to_phys(page
);
926 if ((pte_access
& ACC_WRITE_MASK
)
927 || (write_fault
&& !is_write_protection(vcpu
) && !user_fault
)) {
928 struct kvm_mmu_page
*shadow
;
930 spte
|= PT_WRITABLE_MASK
;
932 mmu_unshadow(vcpu
->kvm
, gfn
);
936 shadow
= kvm_mmu_lookup_page(vcpu
->kvm
, gfn
);
938 pgprintk("%s: found shadow page for %lx, marking ro\n",
940 pte_access
&= ~ACC_WRITE_MASK
;
941 if (is_writeble_pte(spte
)) {
942 spte
&= ~PT_WRITABLE_MASK
;
943 kvm_x86_ops
->tlb_flush(vcpu
);
952 if (pte_access
& ACC_WRITE_MASK
)
953 mark_page_dirty(vcpu
->kvm
, gfn
);
955 pgprintk("%s: setting spte %llx\n", __FUNCTION__
, spte
);
956 set_shadow_pte(shadow_pte
, spte
);
957 page_header_update_slot(vcpu
->kvm
, shadow_pte
, gfn
);
959 rmap_add(vcpu
, shadow_pte
, gfn
);
960 if (!is_rmap_pte(*shadow_pte
))
961 kvm_release_page_clean(page
);
964 kvm_release_page_clean(page
);
965 if (!ptwrite
|| !*ptwrite
)
966 vcpu
->arch
.last_pte_updated
= shadow_pte
;
969 static void nonpaging_new_cr3(struct kvm_vcpu
*vcpu
)
973 static int nonpaging_map(struct kvm_vcpu
*vcpu
, gva_t v
, int write
, gfn_t gfn
)
975 int level
= PT32E_ROOT_LEVEL
;
976 hpa_t table_addr
= vcpu
->arch
.mmu
.root_hpa
;
980 u32 index
= PT64_INDEX(v
, level
);
983 ASSERT(VALID_PAGE(table_addr
));
984 table
= __va(table_addr
);
987 mmu_set_spte(vcpu
, &table
[index
], ACC_ALL
, ACC_ALL
,
988 0, write
, 1, &pt_write
, gfn
);
989 return pt_write
|| is_io_pte(table
[index
]);
992 if (table
[index
] == shadow_trap_nonpresent_pte
) {
993 struct kvm_mmu_page
*new_table
;
996 pseudo_gfn
= (v
& PT64_DIR_BASE_ADDR_MASK
)
998 new_table
= kvm_mmu_get_page(vcpu
, pseudo_gfn
,
1000 1, ACC_ALL
, &table
[index
],
1003 pgprintk("nonpaging_map: ENOMEM\n");
1007 table
[index
] = __pa(new_table
->spt
) | PT_PRESENT_MASK
1008 | PT_WRITABLE_MASK
| PT_USER_MASK
;
1010 table_addr
= table
[index
] & PT64_BASE_ADDR_MASK
;
1014 static void nonpaging_prefetch_page(struct kvm_vcpu
*vcpu
,
1015 struct kvm_mmu_page
*sp
)
1019 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
)
1020 sp
->spt
[i
] = shadow_trap_nonpresent_pte
;
1023 static void mmu_free_roots(struct kvm_vcpu
*vcpu
)
1026 struct kvm_mmu_page
*sp
;
1028 if (!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
))
1030 #ifdef CONFIG_X86_64
1031 if (vcpu
->arch
.mmu
.shadow_root_level
== PT64_ROOT_LEVEL
) {
1032 hpa_t root
= vcpu
->arch
.mmu
.root_hpa
;
1034 sp
= page_header(root
);
1036 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
1040 for (i
= 0; i
< 4; ++i
) {
1041 hpa_t root
= vcpu
->arch
.mmu
.pae_root
[i
];
1044 root
&= PT64_BASE_ADDR_MASK
;
1045 sp
= page_header(root
);
1048 vcpu
->arch
.mmu
.pae_root
[i
] = INVALID_PAGE
;
1050 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
1053 static void mmu_alloc_roots(struct kvm_vcpu
*vcpu
)
1057 struct kvm_mmu_page
*sp
;
1059 root_gfn
= vcpu
->arch
.cr3
>> PAGE_SHIFT
;
1061 #ifdef CONFIG_X86_64
1062 if (vcpu
->arch
.mmu
.shadow_root_level
== PT64_ROOT_LEVEL
) {
1063 hpa_t root
= vcpu
->arch
.mmu
.root_hpa
;
1065 ASSERT(!VALID_PAGE(root
));
1066 sp
= kvm_mmu_get_page(vcpu
, root_gfn
, 0,
1067 PT64_ROOT_LEVEL
, 0, ACC_ALL
, NULL
, NULL
);
1068 root
= __pa(sp
->spt
);
1070 vcpu
->arch
.mmu
.root_hpa
= root
;
1074 for (i
= 0; i
< 4; ++i
) {
1075 hpa_t root
= vcpu
->arch
.mmu
.pae_root
[i
];
1077 ASSERT(!VALID_PAGE(root
));
1078 if (vcpu
->arch
.mmu
.root_level
== PT32E_ROOT_LEVEL
) {
1079 if (!is_present_pte(vcpu
->arch
.pdptrs
[i
])) {
1080 vcpu
->arch
.mmu
.pae_root
[i
] = 0;
1083 root_gfn
= vcpu
->arch
.pdptrs
[i
] >> PAGE_SHIFT
;
1084 } else if (vcpu
->arch
.mmu
.root_level
== 0)
1086 sp
= kvm_mmu_get_page(vcpu
, root_gfn
, i
<< 30,
1087 PT32_ROOT_LEVEL
, !is_paging(vcpu
),
1088 ACC_ALL
, NULL
, NULL
);
1089 root
= __pa(sp
->spt
);
1091 vcpu
->arch
.mmu
.pae_root
[i
] = root
| PT_PRESENT_MASK
;
1093 vcpu
->arch
.mmu
.root_hpa
= __pa(vcpu
->arch
.mmu
.pae_root
);
1096 static gpa_t
nonpaging_gva_to_gpa(struct kvm_vcpu
*vcpu
, gva_t vaddr
)
1101 static int nonpaging_page_fault(struct kvm_vcpu
*vcpu
, gva_t gva
,
1107 pgprintk("%s: gva %lx error %x\n", __FUNCTION__
, gva
, error_code
);
1108 r
= mmu_topup_memory_caches(vcpu
);
1113 ASSERT(VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
1115 gfn
= gva
>> PAGE_SHIFT
;
1117 return nonpaging_map(vcpu
, gva
& PAGE_MASK
,
1118 error_code
& PFERR_WRITE_MASK
, gfn
);
1121 static void nonpaging_free(struct kvm_vcpu
*vcpu
)
1123 mmu_free_roots(vcpu
);
1126 static int nonpaging_init_context(struct kvm_vcpu
*vcpu
)
1128 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
1130 context
->new_cr3
= nonpaging_new_cr3
;
1131 context
->page_fault
= nonpaging_page_fault
;
1132 context
->gva_to_gpa
= nonpaging_gva_to_gpa
;
1133 context
->free
= nonpaging_free
;
1134 context
->prefetch_page
= nonpaging_prefetch_page
;
1135 context
->root_level
= 0;
1136 context
->shadow_root_level
= PT32E_ROOT_LEVEL
;
1137 context
->root_hpa
= INVALID_PAGE
;
1141 void kvm_mmu_flush_tlb(struct kvm_vcpu
*vcpu
)
1143 ++vcpu
->stat
.tlb_flush
;
1144 kvm_x86_ops
->tlb_flush(vcpu
);
1147 static void paging_new_cr3(struct kvm_vcpu
*vcpu
)
1149 pgprintk("%s: cr3 %lx\n", __FUNCTION__
, vcpu
->cr3
);
1150 mmu_free_roots(vcpu
);
1153 static void inject_page_fault(struct kvm_vcpu
*vcpu
,
1157 kvm_inject_page_fault(vcpu
, addr
, err_code
);
1160 static void paging_free(struct kvm_vcpu
*vcpu
)
1162 nonpaging_free(vcpu
);
1166 #include "paging_tmpl.h"
1170 #include "paging_tmpl.h"
1173 static int paging64_init_context_common(struct kvm_vcpu
*vcpu
, int level
)
1175 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
1177 ASSERT(is_pae(vcpu
));
1178 context
->new_cr3
= paging_new_cr3
;
1179 context
->page_fault
= paging64_page_fault
;
1180 context
->gva_to_gpa
= paging64_gva_to_gpa
;
1181 context
->prefetch_page
= paging64_prefetch_page
;
1182 context
->free
= paging_free
;
1183 context
->root_level
= level
;
1184 context
->shadow_root_level
= level
;
1185 context
->root_hpa
= INVALID_PAGE
;
1189 static int paging64_init_context(struct kvm_vcpu
*vcpu
)
1191 return paging64_init_context_common(vcpu
, PT64_ROOT_LEVEL
);
1194 static int paging32_init_context(struct kvm_vcpu
*vcpu
)
1196 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
1198 context
->new_cr3
= paging_new_cr3
;
1199 context
->page_fault
= paging32_page_fault
;
1200 context
->gva_to_gpa
= paging32_gva_to_gpa
;
1201 context
->free
= paging_free
;
1202 context
->prefetch_page
= paging32_prefetch_page
;
1203 context
->root_level
= PT32_ROOT_LEVEL
;
1204 context
->shadow_root_level
= PT32E_ROOT_LEVEL
;
1205 context
->root_hpa
= INVALID_PAGE
;
1209 static int paging32E_init_context(struct kvm_vcpu
*vcpu
)
1211 return paging64_init_context_common(vcpu
, PT32E_ROOT_LEVEL
);
1214 static int init_kvm_mmu(struct kvm_vcpu
*vcpu
)
1217 ASSERT(!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
1219 if (!is_paging(vcpu
))
1220 return nonpaging_init_context(vcpu
);
1221 else if (is_long_mode(vcpu
))
1222 return paging64_init_context(vcpu
);
1223 else if (is_pae(vcpu
))
1224 return paging32E_init_context(vcpu
);
1226 return paging32_init_context(vcpu
);
1229 static void destroy_kvm_mmu(struct kvm_vcpu
*vcpu
)
1232 if (VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
)) {
1233 vcpu
->arch
.mmu
.free(vcpu
);
1234 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
1238 int kvm_mmu_reset_context(struct kvm_vcpu
*vcpu
)
1240 destroy_kvm_mmu(vcpu
);
1241 return init_kvm_mmu(vcpu
);
1243 EXPORT_SYMBOL_GPL(kvm_mmu_reset_context
);
1245 int kvm_mmu_load(struct kvm_vcpu
*vcpu
)
1249 mutex_lock(&vcpu
->kvm
->lock
);
1250 r
= mmu_topup_memory_caches(vcpu
);
1253 mmu_alloc_roots(vcpu
);
1254 kvm_x86_ops
->set_cr3(vcpu
, vcpu
->arch
.mmu
.root_hpa
);
1255 kvm_mmu_flush_tlb(vcpu
);
1257 mutex_unlock(&vcpu
->kvm
->lock
);
1260 EXPORT_SYMBOL_GPL(kvm_mmu_load
);
1262 void kvm_mmu_unload(struct kvm_vcpu
*vcpu
)
1264 mmu_free_roots(vcpu
);
1267 static void mmu_pte_write_zap_pte(struct kvm_vcpu
*vcpu
,
1268 struct kvm_mmu_page
*sp
,
1272 struct kvm_mmu_page
*child
;
1275 if (is_shadow_present_pte(pte
)) {
1276 if (sp
->role
.level
== PT_PAGE_TABLE_LEVEL
)
1277 rmap_remove(vcpu
->kvm
, spte
);
1279 child
= page_header(pte
& PT64_BASE_ADDR_MASK
);
1280 mmu_page_remove_parent_pte(child
, spte
);
1283 set_shadow_pte(spte
, shadow_trap_nonpresent_pte
);
1286 static void mmu_pte_write_new_pte(struct kvm_vcpu
*vcpu
,
1287 struct kvm_mmu_page
*sp
,
1289 const void *new, int bytes
,
1292 if (sp
->role
.level
!= PT_PAGE_TABLE_LEVEL
) {
1293 ++vcpu
->kvm
->stat
.mmu_pde_zapped
;
1297 ++vcpu
->kvm
->stat
.mmu_pte_updated
;
1298 if (sp
->role
.glevels
== PT32_ROOT_LEVEL
)
1299 paging32_update_pte(vcpu
, sp
, spte
, new, bytes
, offset_in_pte
);
1301 paging64_update_pte(vcpu
, sp
, spte
, new, bytes
, offset_in_pte
);
1304 static bool need_remote_flush(u64 old
, u64
new)
1306 if (!is_shadow_present_pte(old
))
1308 if (!is_shadow_present_pte(new))
1310 if ((old
^ new) & PT64_BASE_ADDR_MASK
)
1312 old
^= PT64_NX_MASK
;
1313 new ^= PT64_NX_MASK
;
1314 return (old
& ~new & PT64_PERM_MASK
) != 0;
1317 static void mmu_pte_write_flush_tlb(struct kvm_vcpu
*vcpu
, u64 old
, u64
new)
1319 if (need_remote_flush(old
, new))
1320 kvm_flush_remote_tlbs(vcpu
->kvm
);
1322 kvm_mmu_flush_tlb(vcpu
);
1325 static bool last_updated_pte_accessed(struct kvm_vcpu
*vcpu
)
1327 u64
*spte
= vcpu
->arch
.last_pte_updated
;
1329 return !!(spte
&& (*spte
& PT_ACCESSED_MASK
));
1332 void kvm_mmu_pte_write(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1333 const u8
*new, int bytes
)
1335 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1336 struct kvm_mmu_page
*sp
;
1337 struct hlist_node
*node
, *n
;
1338 struct hlist_head
*bucket
;
1342 unsigned offset
= offset_in_page(gpa
);
1344 unsigned page_offset
;
1345 unsigned misaligned
;
1351 pgprintk("%s: gpa %llx bytes %d\n", __FUNCTION__
, gpa
, bytes
);
1352 ++vcpu
->kvm
->stat
.mmu_pte_write
;
1353 kvm_mmu_audit(vcpu
, "pre pte write");
1354 if (gfn
== vcpu
->arch
.last_pt_write_gfn
1355 && !last_updated_pte_accessed(vcpu
)) {
1356 ++vcpu
->arch
.last_pt_write_count
;
1357 if (vcpu
->arch
.last_pt_write_count
>= 3)
1360 vcpu
->arch
.last_pt_write_gfn
= gfn
;
1361 vcpu
->arch
.last_pt_write_count
= 1;
1362 vcpu
->arch
.last_pte_updated
= NULL
;
1364 index
= kvm_page_table_hashfn(gfn
) % KVM_NUM_MMU_PAGES
;
1365 bucket
= &vcpu
->kvm
->mmu_page_hash
[index
];
1366 hlist_for_each_entry_safe(sp
, node
, n
, bucket
, hash_link
) {
1367 if (sp
->gfn
!= gfn
|| sp
->role
.metaphysical
)
1369 pte_size
= sp
->role
.glevels
== PT32_ROOT_LEVEL
? 4 : 8;
1370 misaligned
= (offset
^ (offset
+ bytes
- 1)) & ~(pte_size
- 1);
1371 misaligned
|= bytes
< 4;
1372 if (misaligned
|| flooded
) {
1374 * Misaligned accesses are too much trouble to fix
1375 * up; also, they usually indicate a page is not used
1378 * If we're seeing too many writes to a page,
1379 * it may no longer be a page table, or we may be
1380 * forking, in which case it is better to unmap the
1383 pgprintk("misaligned: gpa %llx bytes %d role %x\n",
1384 gpa
, bytes
, sp
->role
.word
);
1385 kvm_mmu_zap_page(vcpu
->kvm
, sp
);
1386 ++vcpu
->kvm
->stat
.mmu_flooded
;
1389 page_offset
= offset
;
1390 level
= sp
->role
.level
;
1392 if (sp
->role
.glevels
== PT32_ROOT_LEVEL
) {
1393 page_offset
<<= 1; /* 32->64 */
1395 * A 32-bit pde maps 4MB while the shadow pdes map
1396 * only 2MB. So we need to double the offset again
1397 * and zap two pdes instead of one.
1399 if (level
== PT32_ROOT_LEVEL
) {
1400 page_offset
&= ~7; /* kill rounding error */
1404 quadrant
= page_offset
>> PAGE_SHIFT
;
1405 page_offset
&= ~PAGE_MASK
;
1406 if (quadrant
!= sp
->role
.quadrant
)
1409 spte
= &sp
->spt
[page_offset
/ sizeof(*spte
)];
1412 mmu_pte_write_zap_pte(vcpu
, sp
, spte
);
1413 mmu_pte_write_new_pte(vcpu
, sp
, spte
, new, bytes
,
1414 page_offset
& (pte_size
- 1));
1415 mmu_pte_write_flush_tlb(vcpu
, entry
, *spte
);
1419 kvm_mmu_audit(vcpu
, "post pte write");
1422 int kvm_mmu_unprotect_page_virt(struct kvm_vcpu
*vcpu
, gva_t gva
)
1424 gpa_t gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, gva
);
1426 return kvm_mmu_unprotect_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
1429 void __kvm_mmu_free_some_pages(struct kvm_vcpu
*vcpu
)
1431 while (vcpu
->kvm
->n_free_mmu_pages
< KVM_REFILL_PAGES
) {
1432 struct kvm_mmu_page
*sp
;
1434 sp
= container_of(vcpu
->kvm
->active_mmu_pages
.prev
,
1435 struct kvm_mmu_page
, link
);
1436 kvm_mmu_zap_page(vcpu
->kvm
, sp
);
1437 ++vcpu
->kvm
->stat
.mmu_recycled
;
1441 int kvm_mmu_page_fault(struct kvm_vcpu
*vcpu
, gva_t cr2
, u32 error_code
)
1444 enum emulation_result er
;
1446 mutex_lock(&vcpu
->kvm
->lock
);
1447 r
= vcpu
->arch
.mmu
.page_fault(vcpu
, cr2
, error_code
);
1456 r
= mmu_topup_memory_caches(vcpu
);
1460 er
= emulate_instruction(vcpu
, vcpu
->run
, cr2
, error_code
, 0);
1461 mutex_unlock(&vcpu
->kvm
->lock
);
1466 case EMULATE_DO_MMIO
:
1467 ++vcpu
->stat
.mmio_exits
;
1470 kvm_report_emulation_failure(vcpu
, "pagetable");
1476 mutex_unlock(&vcpu
->kvm
->lock
);
1479 EXPORT_SYMBOL_GPL(kvm_mmu_page_fault
);
1481 static void free_mmu_pages(struct kvm_vcpu
*vcpu
)
1483 struct kvm_mmu_page
*sp
;
1485 while (!list_empty(&vcpu
->kvm
->active_mmu_pages
)) {
1486 sp
= container_of(vcpu
->kvm
->active_mmu_pages
.next
,
1487 struct kvm_mmu_page
, link
);
1488 kvm_mmu_zap_page(vcpu
->kvm
, sp
);
1490 free_page((unsigned long)vcpu
->arch
.mmu
.pae_root
);
1493 static int alloc_mmu_pages(struct kvm_vcpu
*vcpu
)
1500 if (vcpu
->kvm
->n_requested_mmu_pages
)
1501 vcpu
->kvm
->n_free_mmu_pages
= vcpu
->kvm
->n_requested_mmu_pages
;
1503 vcpu
->kvm
->n_free_mmu_pages
= vcpu
->kvm
->n_alloc_mmu_pages
;
1505 * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64.
1506 * Therefore we need to allocate shadow page tables in the first
1507 * 4GB of memory, which happens to fit the DMA32 zone.
1509 page
= alloc_page(GFP_KERNEL
| __GFP_DMA32
);
1512 vcpu
->arch
.mmu
.pae_root
= page_address(page
);
1513 for (i
= 0; i
< 4; ++i
)
1514 vcpu
->arch
.mmu
.pae_root
[i
] = INVALID_PAGE
;
1519 free_mmu_pages(vcpu
);
1523 int kvm_mmu_create(struct kvm_vcpu
*vcpu
)
1526 ASSERT(!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
1528 return alloc_mmu_pages(vcpu
);
1531 int kvm_mmu_setup(struct kvm_vcpu
*vcpu
)
1534 ASSERT(!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
1536 return init_kvm_mmu(vcpu
);
1539 void kvm_mmu_destroy(struct kvm_vcpu
*vcpu
)
1543 destroy_kvm_mmu(vcpu
);
1544 free_mmu_pages(vcpu
);
1545 mmu_free_memory_caches(vcpu
);
1548 void kvm_mmu_slot_remove_write_access(struct kvm
*kvm
, int slot
)
1550 struct kvm_mmu_page
*sp
;
1552 list_for_each_entry(sp
, &kvm
->active_mmu_pages
, link
) {
1556 if (!test_bit(slot
, &sp
->slot_bitmap
))
1560 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
)
1562 if (pt
[i
] & PT_WRITABLE_MASK
)
1563 pt
[i
] &= ~PT_WRITABLE_MASK
;
1567 void kvm_mmu_zap_all(struct kvm
*kvm
)
1569 struct kvm_mmu_page
*sp
, *node
;
1571 list_for_each_entry_safe(sp
, node
, &kvm
->active_mmu_pages
, link
)
1572 kvm_mmu_zap_page(kvm
, sp
);
1574 kvm_flush_remote_tlbs(kvm
);
1577 void kvm_mmu_module_exit(void)
1579 if (pte_chain_cache
)
1580 kmem_cache_destroy(pte_chain_cache
);
1581 if (rmap_desc_cache
)
1582 kmem_cache_destroy(rmap_desc_cache
);
1583 if (mmu_page_header_cache
)
1584 kmem_cache_destroy(mmu_page_header_cache
);
1587 int kvm_mmu_module_init(void)
1589 pte_chain_cache
= kmem_cache_create("kvm_pte_chain",
1590 sizeof(struct kvm_pte_chain
),
1592 if (!pte_chain_cache
)
1594 rmap_desc_cache
= kmem_cache_create("kvm_rmap_desc",
1595 sizeof(struct kvm_rmap_desc
),
1597 if (!rmap_desc_cache
)
1600 mmu_page_header_cache
= kmem_cache_create("kvm_mmu_page_header",
1601 sizeof(struct kvm_mmu_page
),
1603 if (!mmu_page_header_cache
)
1609 kvm_mmu_module_exit();
1614 * Caculate mmu pages needed for kvm.
1616 unsigned int kvm_mmu_calculate_mmu_pages(struct kvm
*kvm
)
1619 unsigned int nr_mmu_pages
;
1620 unsigned int nr_pages
= 0;
1622 for (i
= 0; i
< kvm
->nmemslots
; i
++)
1623 nr_pages
+= kvm
->memslots
[i
].npages
;
1625 nr_mmu_pages
= nr_pages
* KVM_PERMILLE_MMU_PAGES
/ 1000;
1626 nr_mmu_pages
= max(nr_mmu_pages
,
1627 (unsigned int) KVM_MIN_ALLOC_MMU_PAGES
);
1629 return nr_mmu_pages
;
1634 static const char *audit_msg
;
1636 static gva_t
canonicalize(gva_t gva
)
1638 #ifdef CONFIG_X86_64
1639 gva
= (long long)(gva
<< 16) >> 16;
1644 static void audit_mappings_page(struct kvm_vcpu
*vcpu
, u64 page_pte
,
1645 gva_t va
, int level
)
1647 u64
*pt
= __va(page_pte
& PT64_BASE_ADDR_MASK
);
1649 gva_t va_delta
= 1ul << (PAGE_SHIFT
+ 9 * (level
- 1));
1651 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
, va
+= va_delta
) {
1654 if (ent
== shadow_trap_nonpresent_pte
)
1657 va
= canonicalize(va
);
1659 if (ent
== shadow_notrap_nonpresent_pte
)
1660 printk(KERN_ERR
"audit: (%s) nontrapping pte"
1661 " in nonleaf level: levels %d gva %lx"
1662 " level %d pte %llx\n", audit_msg
,
1663 vcpu
->arch
.mmu
.root_level
, va
, level
, ent
);
1665 audit_mappings_page(vcpu
, ent
, va
, level
- 1);
1667 gpa_t gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, va
);
1668 struct page
*page
= gpa_to_page(vcpu
, gpa
);
1669 hpa_t hpa
= page_to_phys(page
);
1671 if (is_shadow_present_pte(ent
)
1672 && (ent
& PT64_BASE_ADDR_MASK
) != hpa
)
1673 printk(KERN_ERR
"xx audit error: (%s) levels %d"
1674 " gva %lx gpa %llx hpa %llx ent %llx %d\n",
1675 audit_msg
, vcpu
->arch
.mmu
.root_level
,
1677 is_shadow_present_pte(ent
));
1678 else if (ent
== shadow_notrap_nonpresent_pte
1679 && !is_error_hpa(hpa
))
1680 printk(KERN_ERR
"audit: (%s) notrap shadow,"
1681 " valid guest gva %lx\n", audit_msg
, va
);
1682 kvm_release_page_clean(page
);
1688 static void audit_mappings(struct kvm_vcpu
*vcpu
)
1692 if (vcpu
->arch
.mmu
.root_level
== 4)
1693 audit_mappings_page(vcpu
, vcpu
->arch
.mmu
.root_hpa
, 0, 4);
1695 for (i
= 0; i
< 4; ++i
)
1696 if (vcpu
->arch
.mmu
.pae_root
[i
] & PT_PRESENT_MASK
)
1697 audit_mappings_page(vcpu
,
1698 vcpu
->arch
.mmu
.pae_root
[i
],
1703 static int count_rmaps(struct kvm_vcpu
*vcpu
)
1708 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
1709 struct kvm_memory_slot
*m
= &vcpu
->kvm
->memslots
[i
];
1710 struct kvm_rmap_desc
*d
;
1712 for (j
= 0; j
< m
->npages
; ++j
) {
1713 unsigned long *rmapp
= &m
->rmap
[j
];
1717 if (!(*rmapp
& 1)) {
1721 d
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
1723 for (k
= 0; k
< RMAP_EXT
; ++k
)
1724 if (d
->shadow_ptes
[k
])
1735 static int count_writable_mappings(struct kvm_vcpu
*vcpu
)
1738 struct kvm_mmu_page
*sp
;
1741 list_for_each_entry(sp
, &vcpu
->kvm
->active_mmu_pages
, link
) {
1744 if (sp
->role
.level
!= PT_PAGE_TABLE_LEVEL
)
1747 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
1750 if (!(ent
& PT_PRESENT_MASK
))
1752 if (!(ent
& PT_WRITABLE_MASK
))
1760 static void audit_rmap(struct kvm_vcpu
*vcpu
)
1762 int n_rmap
= count_rmaps(vcpu
);
1763 int n_actual
= count_writable_mappings(vcpu
);
1765 if (n_rmap
!= n_actual
)
1766 printk(KERN_ERR
"%s: (%s) rmap %d actual %d\n",
1767 __FUNCTION__
, audit_msg
, n_rmap
, n_actual
);
1770 static void audit_write_protection(struct kvm_vcpu
*vcpu
)
1772 struct kvm_mmu_page
*sp
;
1773 struct kvm_memory_slot
*slot
;
1774 unsigned long *rmapp
;
1777 list_for_each_entry(sp
, &vcpu
->kvm
->active_mmu_pages
, link
) {
1778 if (sp
->role
.metaphysical
)
1781 slot
= gfn_to_memslot(vcpu
->kvm
, sp
->gfn
);
1782 gfn
= unalias_gfn(vcpu
->kvm
, sp
->gfn
);
1783 rmapp
= &slot
->rmap
[gfn
- slot
->base_gfn
];
1785 printk(KERN_ERR
"%s: (%s) shadow page has writable"
1786 " mappings: gfn %lx role %x\n",
1787 __FUNCTION__
, audit_msg
, sp
->gfn
,
1792 static void kvm_mmu_audit(struct kvm_vcpu
*vcpu
, const char *msg
)
1799 audit_write_protection(vcpu
);
1800 audit_mappings(vcpu
);