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.
19 #include <linux/types.h>
20 #include <linux/string.h>
23 #include <linux/highmem.h>
24 #include <linux/module.h>
29 #define pgprintk(x...) do { printk(x); } while (0)
30 #define rmap_printk(x...) do { printk(x); } while (0)
34 printk(KERN_WARNING "assertion failed %s:%d: %s\n", \
35 __FILE__, __LINE__, #x); \
38 #define PT64_PT_BITS 9
39 #define PT64_ENT_PER_PAGE (1 << PT64_PT_BITS)
40 #define PT32_PT_BITS 10
41 #define PT32_ENT_PER_PAGE (1 << PT32_PT_BITS)
43 #define PT_WRITABLE_SHIFT 1
45 #define PT_PRESENT_MASK (1ULL << 0)
46 #define PT_WRITABLE_MASK (1ULL << PT_WRITABLE_SHIFT)
47 #define PT_USER_MASK (1ULL << 2)
48 #define PT_PWT_MASK (1ULL << 3)
49 #define PT_PCD_MASK (1ULL << 4)
50 #define PT_ACCESSED_MASK (1ULL << 5)
51 #define PT_DIRTY_MASK (1ULL << 6)
52 #define PT_PAGE_SIZE_MASK (1ULL << 7)
53 #define PT_PAT_MASK (1ULL << 7)
54 #define PT_GLOBAL_MASK (1ULL << 8)
55 #define PT64_NX_MASK (1ULL << 63)
57 #define PT_PAT_SHIFT 7
58 #define PT_DIR_PAT_SHIFT 12
59 #define PT_DIR_PAT_MASK (1ULL << PT_DIR_PAT_SHIFT)
61 #define PT32_DIR_PSE36_SIZE 4
62 #define PT32_DIR_PSE36_SHIFT 13
63 #define PT32_DIR_PSE36_MASK (((1ULL << PT32_DIR_PSE36_SIZE) - 1) << PT32_DIR_PSE36_SHIFT)
66 #define PT32_PTE_COPY_MASK \
67 (PT_PRESENT_MASK | PT_ACCESSED_MASK | PT_DIRTY_MASK | PT_GLOBAL_MASK)
69 #define PT64_PTE_COPY_MASK (PT64_NX_MASK | PT32_PTE_COPY_MASK)
71 #define PT_FIRST_AVAIL_BITS_SHIFT 9
72 #define PT64_SECOND_AVAIL_BITS_SHIFT 52
74 #define PT_SHADOW_PS_MARK (1ULL << PT_FIRST_AVAIL_BITS_SHIFT)
75 #define PT_SHADOW_IO_MARK (1ULL << PT_FIRST_AVAIL_BITS_SHIFT)
77 #define PT_SHADOW_WRITABLE_SHIFT (PT_FIRST_AVAIL_BITS_SHIFT + 1)
78 #define PT_SHADOW_WRITABLE_MASK (1ULL << PT_SHADOW_WRITABLE_SHIFT)
80 #define PT_SHADOW_USER_SHIFT (PT_SHADOW_WRITABLE_SHIFT + 1)
81 #define PT_SHADOW_USER_MASK (1ULL << (PT_SHADOW_USER_SHIFT))
83 #define PT_SHADOW_BITS_OFFSET (PT_SHADOW_WRITABLE_SHIFT - PT_WRITABLE_SHIFT)
85 #define VALID_PAGE(x) ((x) != INVALID_PAGE)
87 #define PT64_LEVEL_BITS 9
89 #define PT64_LEVEL_SHIFT(level) \
90 ( PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS )
92 #define PT64_LEVEL_MASK(level) \
93 (((1ULL << PT64_LEVEL_BITS) - 1) << PT64_LEVEL_SHIFT(level))
95 #define PT64_INDEX(address, level)\
96 (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1))
99 #define PT32_LEVEL_BITS 10
101 #define PT32_LEVEL_SHIFT(level) \
102 ( PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS )
104 #define PT32_LEVEL_MASK(level) \
105 (((1ULL << PT32_LEVEL_BITS) - 1) << PT32_LEVEL_SHIFT(level))
107 #define PT32_INDEX(address, level)\
108 (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1))
111 #define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & PAGE_MASK)
112 #define PT64_DIR_BASE_ADDR_MASK \
113 (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1))
115 #define PT32_BASE_ADDR_MASK PAGE_MASK
116 #define PT32_DIR_BASE_ADDR_MASK \
117 (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1))
120 #define PFERR_PRESENT_MASK (1U << 0)
121 #define PFERR_WRITE_MASK (1U << 1)
122 #define PFERR_USER_MASK (1U << 2)
124 #define PT64_ROOT_LEVEL 4
125 #define PT32_ROOT_LEVEL 2
126 #define PT32E_ROOT_LEVEL 3
128 #define PT_DIRECTORY_LEVEL 2
129 #define PT_PAGE_TABLE_LEVEL 1
133 struct kvm_rmap_desc
{
134 u64
*shadow_ptes
[RMAP_EXT
];
135 struct kvm_rmap_desc
*more
;
138 static int is_write_protection(struct kvm_vcpu
*vcpu
)
140 return vcpu
->cr0
& CR0_WP_MASK
;
143 static int is_cpuid_PSE36(void)
148 static int is_present_pte(unsigned long pte
)
150 return pte
& PT_PRESENT_MASK
;
153 static int is_writeble_pte(unsigned long pte
)
155 return pte
& PT_WRITABLE_MASK
;
158 static int is_io_pte(unsigned long pte
)
160 return pte
& PT_SHADOW_IO_MARK
;
163 static int is_rmap_pte(u64 pte
)
165 return (pte
& (PT_WRITABLE_MASK
| PT_PRESENT_MASK
))
166 == (PT_WRITABLE_MASK
| PT_PRESENT_MASK
);
170 * Reverse mapping data structures:
172 * If page->private bit zero is zero, then page->private points to the
173 * shadow page table entry that points to page_address(page).
175 * If page->private bit zero is one, (then page->private & ~1) points
176 * to a struct kvm_rmap_desc containing more mappings.
178 static void rmap_add(struct kvm
*kvm
, u64
*spte
)
181 struct kvm_rmap_desc
*desc
;
184 if (!is_rmap_pte(*spte
))
186 page
= pfn_to_page((*spte
& PT64_BASE_ADDR_MASK
) >> PAGE_SHIFT
);
187 if (!page
->private) {
188 rmap_printk("rmap_add: %p %llx 0->1\n", spte
, *spte
);
189 page
->private = (unsigned long)spte
;
190 } else if (!(page
->private & 1)) {
191 rmap_printk("rmap_add: %p %llx 1->many\n", spte
, *spte
);
192 desc
= kzalloc(sizeof *desc
, GFP_NOWAIT
);
194 BUG(); /* FIXME: return error */
195 desc
->shadow_ptes
[0] = (u64
*)page
->private;
196 desc
->shadow_ptes
[1] = spte
;
197 page
->private = (unsigned long)desc
| 1;
199 rmap_printk("rmap_add: %p %llx many->many\n", spte
, *spte
);
200 desc
= (struct kvm_rmap_desc
*)(page
->private & ~1ul);
201 while (desc
->shadow_ptes
[RMAP_EXT
-1] && desc
->more
)
203 if (desc
->shadow_ptes
[RMAP_EXT
-1]) {
204 desc
->more
= kzalloc(sizeof *desc
->more
, GFP_NOWAIT
);
206 BUG(); /* FIXME: return error */
209 for (i
= 0; desc
->shadow_ptes
[i
]; ++i
)
211 desc
->shadow_ptes
[i
] = spte
;
215 static void rmap_desc_remove_entry(struct page
*page
,
216 struct kvm_rmap_desc
*desc
,
218 struct kvm_rmap_desc
*prev_desc
)
222 for (j
= RMAP_EXT
- 1; !desc
->shadow_ptes
[j
] && j
> i
; --j
)
224 desc
->shadow_ptes
[i
] = desc
->shadow_ptes
[j
];
225 desc
->shadow_ptes
[j
] = 0;
228 if (!prev_desc
&& !desc
->more
)
229 page
->private = (unsigned long)desc
->shadow_ptes
[0];
232 prev_desc
->more
= desc
->more
;
234 page
->private = (unsigned long)desc
->more
| 1;
238 static void rmap_remove(struct kvm
*kvm
, u64
*spte
)
241 struct kvm_rmap_desc
*desc
;
242 struct kvm_rmap_desc
*prev_desc
;
245 if (!is_rmap_pte(*spte
))
247 page
= pfn_to_page((*spte
& PT64_BASE_ADDR_MASK
) >> PAGE_SHIFT
);
248 if (!page
->private) {
249 printk(KERN_ERR
"rmap_remove: %p %llx 0->BUG\n", spte
, *spte
);
251 } else if (!(page
->private & 1)) {
252 rmap_printk("rmap_remove: %p %llx 1->0\n", spte
, *spte
);
253 if ((u64
*)page
->private != spte
) {
254 printk(KERN_ERR
"rmap_remove: %p %llx 1->BUG\n",
260 rmap_printk("rmap_remove: %p %llx many->many\n", spte
, *spte
);
261 desc
= (struct kvm_rmap_desc
*)(page
->private & ~1ul);
264 for (i
= 0; i
< RMAP_EXT
&& desc
->shadow_ptes
[i
]; ++i
)
265 if (desc
->shadow_ptes
[i
] == spte
) {
266 rmap_desc_remove_entry(page
, desc
, i
,
277 static void rmap_write_protect(struct kvm
*kvm
, u64 gfn
)
280 struct kvm_memory_slot
*slot
;
281 struct kvm_rmap_desc
*desc
;
284 slot
= gfn_to_memslot(kvm
, gfn
);
286 page
= gfn_to_page(slot
, gfn
);
288 while (page
->private) {
289 if (!(page
->private & 1))
290 spte
= (u64
*)page
->private;
292 desc
= (struct kvm_rmap_desc
*)(page
->private & ~1ul);
293 spte
= desc
->shadow_ptes
[0];
296 BUG_ON((*spte
& PT64_BASE_ADDR_MASK
) !=
297 page_to_pfn(page
) << PAGE_SHIFT
);
298 BUG_ON(!(*spte
& PT_PRESENT_MASK
));
299 BUG_ON(!(*spte
& PT_WRITABLE_MASK
));
300 rmap_printk("rmap_write_protect: spte %p %llx\n", spte
, *spte
);
301 rmap_remove(kvm
, spte
);
302 *spte
&= ~(u64
)PT_WRITABLE_MASK
;
306 static void kvm_mmu_free_page(struct kvm_vcpu
*vcpu
, hpa_t page_hpa
)
308 struct kvm_mmu_page
*page_head
= page_header(page_hpa
);
310 list_del(&page_head
->link
);
311 page_head
->page_hpa
= page_hpa
;
312 list_add(&page_head
->link
, &vcpu
->free_pages
);
313 ++vcpu
->kvm
->n_free_mmu_pages
;
316 static int is_empty_shadow_page(hpa_t page_hpa
)
320 for (pos
= __va(page_hpa
), end
= pos
+ PAGE_SIZE
/ sizeof(u32
);
327 static unsigned kvm_page_table_hashfn(gfn_t gfn
)
332 static struct kvm_mmu_page
*kvm_mmu_alloc_page(struct kvm_vcpu
*vcpu
,
335 struct kvm_mmu_page
*page
;
337 if (list_empty(&vcpu
->free_pages
))
340 page
= list_entry(vcpu
->free_pages
.next
, struct kvm_mmu_page
, link
);
341 list_del(&page
->link
);
342 list_add(&page
->link
, &vcpu
->kvm
->active_mmu_pages
);
343 ASSERT(is_empty_shadow_page(page
->page_hpa
));
344 page
->slot_bitmap
= 0;
346 page
->multimapped
= 0;
347 page
->parent_pte
= parent_pte
;
348 --vcpu
->kvm
->n_free_mmu_pages
;
352 static void mmu_page_add_parent_pte(struct kvm_mmu_page
*page
, u64
*parent_pte
)
354 struct kvm_pte_chain
*pte_chain
;
355 struct hlist_node
*node
;
360 if (!page
->multimapped
) {
361 u64
*old
= page
->parent_pte
;
364 page
->parent_pte
= parent_pte
;
367 page
->multimapped
= 1;
368 pte_chain
= kzalloc(sizeof(struct kvm_pte_chain
), GFP_NOWAIT
);
370 INIT_HLIST_HEAD(&page
->parent_ptes
);
371 hlist_add_head(&pte_chain
->link
, &page
->parent_ptes
);
372 pte_chain
->parent_ptes
[0] = old
;
374 hlist_for_each_entry(pte_chain
, node
, &page
->parent_ptes
, link
) {
375 if (pte_chain
->parent_ptes
[NR_PTE_CHAIN_ENTRIES
-1])
377 for (i
= 0; i
< NR_PTE_CHAIN_ENTRIES
; ++i
)
378 if (!pte_chain
->parent_ptes
[i
]) {
379 pte_chain
->parent_ptes
[i
] = parent_pte
;
383 pte_chain
= kzalloc(sizeof(struct kvm_pte_chain
), GFP_NOWAIT
);
385 hlist_add_head(&pte_chain
->link
, &page
->parent_ptes
);
386 pte_chain
->parent_ptes
[0] = parent_pte
;
389 static void mmu_page_remove_parent_pte(struct kvm_mmu_page
*page
,
392 struct kvm_pte_chain
*pte_chain
;
393 struct hlist_node
*node
;
396 if (!page
->multimapped
) {
397 BUG_ON(page
->parent_pte
!= parent_pte
);
398 page
->parent_pte
= NULL
;
401 hlist_for_each_entry(pte_chain
, node
, &page
->parent_ptes
, link
)
402 for (i
= 0; i
< NR_PTE_CHAIN_ENTRIES
; ++i
) {
403 if (!pte_chain
->parent_ptes
[i
])
405 if (pte_chain
->parent_ptes
[i
] != parent_pte
)
407 while (i
+ 1 < NR_PTE_CHAIN_ENTRIES
408 && pte_chain
->parent_ptes
[i
+ 1]) {
409 pte_chain
->parent_ptes
[i
]
410 = pte_chain
->parent_ptes
[i
+ 1];
413 pte_chain
->parent_ptes
[i
] = NULL
;
415 hlist_del(&pte_chain
->link
);
417 if (hlist_empty(&page
->parent_ptes
)) {
418 page
->multimapped
= 0;
419 page
->parent_pte
= NULL
;
427 static struct kvm_mmu_page
*kvm_mmu_lookup_page(struct kvm_vcpu
*vcpu
,
431 struct hlist_head
*bucket
;
432 struct kvm_mmu_page
*page
;
433 struct hlist_node
*node
;
435 pgprintk("%s: looking for gfn %lx\n", __FUNCTION__
, gfn
);
436 index
= kvm_page_table_hashfn(gfn
) % KVM_NUM_MMU_PAGES
;
437 bucket
= &vcpu
->kvm
->mmu_page_hash
[index
];
438 hlist_for_each_entry(page
, node
, bucket
, hash_link
)
439 if (page
->gfn
== gfn
&& !page
->role
.metaphysical
) {
440 pgprintk("%s: found role %x\n",
441 __FUNCTION__
, page
->role
.word
);
447 static struct kvm_mmu_page
*kvm_mmu_get_page(struct kvm_vcpu
*vcpu
,
454 union kvm_mmu_page_role role
;
457 struct hlist_head
*bucket
;
458 struct kvm_mmu_page
*page
;
459 struct hlist_node
*node
;
462 role
.glevels
= vcpu
->mmu
.root_level
;
464 role
.metaphysical
= metaphysical
;
465 if (vcpu
->mmu
.root_level
<= PT32_ROOT_LEVEL
) {
466 quadrant
= gaddr
>> (PAGE_SHIFT
+ (PT64_PT_BITS
* level
));
467 quadrant
&= (1 << ((PT32_PT_BITS
- PT64_PT_BITS
) * level
)) - 1;
468 role
.quadrant
= quadrant
;
470 pgprintk("%s: looking gfn %lx role %x\n", __FUNCTION__
,
472 index
= kvm_page_table_hashfn(gfn
) % KVM_NUM_MMU_PAGES
;
473 bucket
= &vcpu
->kvm
->mmu_page_hash
[index
];
474 hlist_for_each_entry(page
, node
, bucket
, hash_link
)
475 if (page
->gfn
== gfn
&& page
->role
.word
== role
.word
) {
476 mmu_page_add_parent_pte(page
, parent_pte
);
477 pgprintk("%s: found\n", __FUNCTION__
);
480 page
= kvm_mmu_alloc_page(vcpu
, parent_pte
);
483 pgprintk("%s: adding gfn %lx role %x\n", __FUNCTION__
, gfn
, role
.word
);
486 hlist_add_head(&page
->hash_link
, bucket
);
488 rmap_write_protect(vcpu
->kvm
, gfn
);
492 static void kvm_mmu_page_unlink_children(struct kvm_vcpu
*vcpu
,
493 struct kvm_mmu_page
*page
)
499 pt
= __va(page
->page_hpa
);
501 if (page
->role
.level
== PT_PAGE_TABLE_LEVEL
) {
502 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
503 if (pt
[i
] & PT_PRESENT_MASK
)
504 rmap_remove(vcpu
->kvm
, &pt
[i
]);
510 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
514 if (!(ent
& PT_PRESENT_MASK
))
516 ent
&= PT64_BASE_ADDR_MASK
;
517 mmu_page_remove_parent_pte(page_header(ent
), &pt
[i
]);
521 static void kvm_mmu_put_page(struct kvm_vcpu
*vcpu
,
522 struct kvm_mmu_page
*page
,
525 mmu_page_remove_parent_pte(page
, parent_pte
);
528 static void kvm_mmu_zap_page(struct kvm_vcpu
*vcpu
,
529 struct kvm_mmu_page
*page
)
533 while (page
->multimapped
|| page
->parent_pte
) {
534 if (!page
->multimapped
)
535 parent_pte
= page
->parent_pte
;
537 struct kvm_pte_chain
*chain
;
539 chain
= container_of(page
->parent_ptes
.first
,
540 struct kvm_pte_chain
, link
);
541 parent_pte
= chain
->parent_ptes
[0];
544 kvm_mmu_put_page(vcpu
, page
, parent_pte
);
547 kvm_mmu_page_unlink_children(vcpu
, page
);
548 hlist_del(&page
->hash_link
);
549 kvm_mmu_free_page(vcpu
, page
->page_hpa
);
552 static int kvm_mmu_unprotect_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
555 struct hlist_head
*bucket
;
556 struct kvm_mmu_page
*page
;
557 struct hlist_node
*node
, *n
;
560 pgprintk("%s: looking for gfn %lx\n", __FUNCTION__
, gfn
);
562 index
= kvm_page_table_hashfn(gfn
) % KVM_NUM_MMU_PAGES
;
563 bucket
= &vcpu
->kvm
->mmu_page_hash
[index
];
564 hlist_for_each_entry_safe(page
, node
, n
, bucket
, hash_link
)
565 if (page
->gfn
== gfn
&& !page
->role
.metaphysical
) {
566 pgprintk("%s: gfn %lx role %x\n", __FUNCTION__
, gfn
,
568 kvm_mmu_zap_page(vcpu
, page
);
574 static void page_header_update_slot(struct kvm
*kvm
, void *pte
, gpa_t gpa
)
576 int slot
= memslot_id(kvm
, gfn_to_memslot(kvm
, gpa
>> PAGE_SHIFT
));
577 struct kvm_mmu_page
*page_head
= page_header(__pa(pte
));
579 __set_bit(slot
, &page_head
->slot_bitmap
);
582 hpa_t
safe_gpa_to_hpa(struct kvm_vcpu
*vcpu
, gpa_t gpa
)
584 hpa_t hpa
= gpa_to_hpa(vcpu
, gpa
);
586 return is_error_hpa(hpa
) ? bad_page_address
| (gpa
& ~PAGE_MASK
): hpa
;
589 hpa_t
gpa_to_hpa(struct kvm_vcpu
*vcpu
, gpa_t gpa
)
591 struct kvm_memory_slot
*slot
;
594 ASSERT((gpa
& HPA_ERR_MASK
) == 0);
595 slot
= gfn_to_memslot(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
597 return gpa
| HPA_ERR_MASK
;
598 page
= gfn_to_page(slot
, gpa
>> PAGE_SHIFT
);
599 return ((hpa_t
)page_to_pfn(page
) << PAGE_SHIFT
)
600 | (gpa
& (PAGE_SIZE
-1));
603 hpa_t
gva_to_hpa(struct kvm_vcpu
*vcpu
, gva_t gva
)
605 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, gva
);
607 if (gpa
== UNMAPPED_GVA
)
609 return gpa_to_hpa(vcpu
, gpa
);
612 static void nonpaging_new_cr3(struct kvm_vcpu
*vcpu
)
616 static int nonpaging_map(struct kvm_vcpu
*vcpu
, gva_t v
, hpa_t p
)
618 int level
= PT32E_ROOT_LEVEL
;
619 hpa_t table_addr
= vcpu
->mmu
.root_hpa
;
622 u32 index
= PT64_INDEX(v
, level
);
626 ASSERT(VALID_PAGE(table_addr
));
627 table
= __va(table_addr
);
631 if (is_present_pte(pte
) && is_writeble_pte(pte
))
633 mark_page_dirty(vcpu
->kvm
, v
>> PAGE_SHIFT
);
634 page_header_update_slot(vcpu
->kvm
, table
, v
);
635 table
[index
] = p
| PT_PRESENT_MASK
| PT_WRITABLE_MASK
|
637 rmap_add(vcpu
->kvm
, &table
[index
]);
641 if (table
[index
] == 0) {
642 struct kvm_mmu_page
*new_table
;
645 pseudo_gfn
= (v
& PT64_DIR_BASE_ADDR_MASK
)
647 new_table
= kvm_mmu_get_page(vcpu
, pseudo_gfn
,
651 pgprintk("nonpaging_map: ENOMEM\n");
655 table
[index
] = new_table
->page_hpa
| PT_PRESENT_MASK
656 | PT_WRITABLE_MASK
| PT_USER_MASK
;
658 table_addr
= table
[index
] & PT64_BASE_ADDR_MASK
;
662 static void mmu_free_roots(struct kvm_vcpu
*vcpu
)
667 if (vcpu
->mmu
.shadow_root_level
== PT64_ROOT_LEVEL
) {
668 hpa_t root
= vcpu
->mmu
.root_hpa
;
670 ASSERT(VALID_PAGE(root
));
671 vcpu
->mmu
.root_hpa
= INVALID_PAGE
;
675 for (i
= 0; i
< 4; ++i
) {
676 hpa_t root
= vcpu
->mmu
.pae_root
[i
];
678 ASSERT(VALID_PAGE(root
));
679 root
&= PT64_BASE_ADDR_MASK
;
680 vcpu
->mmu
.pae_root
[i
] = INVALID_PAGE
;
682 vcpu
->mmu
.root_hpa
= INVALID_PAGE
;
685 static void mmu_alloc_roots(struct kvm_vcpu
*vcpu
)
689 root_gfn
= vcpu
->cr3
>> PAGE_SHIFT
;
692 if (vcpu
->mmu
.shadow_root_level
== PT64_ROOT_LEVEL
) {
693 hpa_t root
= vcpu
->mmu
.root_hpa
;
695 ASSERT(!VALID_PAGE(root
));
696 root
= kvm_mmu_get_page(vcpu
, root_gfn
, 0,
697 PT64_ROOT_LEVEL
, 0, NULL
)->page_hpa
;
698 vcpu
->mmu
.root_hpa
= root
;
702 for (i
= 0; i
< 4; ++i
) {
703 hpa_t root
= vcpu
->mmu
.pae_root
[i
];
705 ASSERT(!VALID_PAGE(root
));
706 if (vcpu
->mmu
.root_level
== PT32E_ROOT_LEVEL
)
707 root_gfn
= vcpu
->pdptrs
[i
] >> PAGE_SHIFT
;
708 else if (vcpu
->mmu
.root_level
== 0)
710 root
= kvm_mmu_get_page(vcpu
, root_gfn
, i
<< 30,
711 PT32_ROOT_LEVEL
, !is_paging(vcpu
),
713 vcpu
->mmu
.pae_root
[i
] = root
| PT_PRESENT_MASK
;
715 vcpu
->mmu
.root_hpa
= __pa(vcpu
->mmu
.pae_root
);
718 static gpa_t
nonpaging_gva_to_gpa(struct kvm_vcpu
*vcpu
, gva_t vaddr
)
723 static int nonpaging_page_fault(struct kvm_vcpu
*vcpu
, gva_t gva
,
730 ASSERT(VALID_PAGE(vcpu
->mmu
.root_hpa
));
733 paddr
= gpa_to_hpa(vcpu
, addr
& PT64_BASE_ADDR_MASK
);
735 if (is_error_hpa(paddr
))
738 return nonpaging_map(vcpu
, addr
& PAGE_MASK
, paddr
);
741 static void nonpaging_free(struct kvm_vcpu
*vcpu
)
743 mmu_free_roots(vcpu
);
746 static int nonpaging_init_context(struct kvm_vcpu
*vcpu
)
748 struct kvm_mmu
*context
= &vcpu
->mmu
;
750 context
->new_cr3
= nonpaging_new_cr3
;
751 context
->page_fault
= nonpaging_page_fault
;
752 context
->gva_to_gpa
= nonpaging_gva_to_gpa
;
753 context
->free
= nonpaging_free
;
754 context
->root_level
= 0;
755 context
->shadow_root_level
= PT32E_ROOT_LEVEL
;
756 mmu_alloc_roots(vcpu
);
757 ASSERT(VALID_PAGE(context
->root_hpa
));
758 kvm_arch_ops
->set_cr3(vcpu
, context
->root_hpa
);
762 static void kvm_mmu_flush_tlb(struct kvm_vcpu
*vcpu
)
764 ++kvm_stat
.tlb_flush
;
765 kvm_arch_ops
->tlb_flush(vcpu
);
768 static void paging_new_cr3(struct kvm_vcpu
*vcpu
)
770 pgprintk("%s: cr3 %lx\n", __FUNCTION__
, vcpu
->cr3
);
771 mmu_free_roots(vcpu
);
772 mmu_alloc_roots(vcpu
);
773 kvm_mmu_flush_tlb(vcpu
);
774 kvm_arch_ops
->set_cr3(vcpu
, vcpu
->mmu
.root_hpa
);
777 static void mark_pagetable_nonglobal(void *shadow_pte
)
779 page_header(__pa(shadow_pte
))->global
= 0;
782 static inline void set_pte_common(struct kvm_vcpu
*vcpu
,
791 *shadow_pte
|= access_bits
<< PT_SHADOW_BITS_OFFSET
;
793 access_bits
&= ~PT_WRITABLE_MASK
;
795 paddr
= gpa_to_hpa(vcpu
, gaddr
& PT64_BASE_ADDR_MASK
);
797 *shadow_pte
|= access_bits
;
799 if (!(*shadow_pte
& PT_GLOBAL_MASK
))
800 mark_pagetable_nonglobal(shadow_pte
);
802 if (is_error_hpa(paddr
)) {
803 *shadow_pte
|= gaddr
;
804 *shadow_pte
|= PT_SHADOW_IO_MARK
;
805 *shadow_pte
&= ~PT_PRESENT_MASK
;
809 *shadow_pte
|= paddr
;
811 if (access_bits
& PT_WRITABLE_MASK
) {
812 struct kvm_mmu_page
*shadow
;
814 shadow
= kvm_mmu_lookup_page(vcpu
, gfn
);
816 pgprintk("%s: found shadow page for %lx, marking ro\n",
818 access_bits
&= ~PT_WRITABLE_MASK
;
819 *shadow_pte
&= ~PT_WRITABLE_MASK
;
823 if (access_bits
& PT_WRITABLE_MASK
)
824 mark_page_dirty(vcpu
->kvm
, gaddr
>> PAGE_SHIFT
);
826 page_header_update_slot(vcpu
->kvm
, shadow_pte
, gaddr
);
827 rmap_add(vcpu
->kvm
, shadow_pte
);
830 static void inject_page_fault(struct kvm_vcpu
*vcpu
,
834 kvm_arch_ops
->inject_page_fault(vcpu
, addr
, err_code
);
837 static inline int fix_read_pf(u64
*shadow_ent
)
839 if ((*shadow_ent
& PT_SHADOW_USER_MASK
) &&
840 !(*shadow_ent
& PT_USER_MASK
)) {
842 * If supervisor write protect is disabled, we shadow kernel
843 * pages as user pages so we can trap the write access.
845 *shadow_ent
|= PT_USER_MASK
;
846 *shadow_ent
&= ~PT_WRITABLE_MASK
;
854 static int may_access(u64 pte
, int write
, int user
)
857 if (user
&& !(pte
& PT_USER_MASK
))
859 if (write
&& !(pte
& PT_WRITABLE_MASK
))
864 static void paging_free(struct kvm_vcpu
*vcpu
)
866 nonpaging_free(vcpu
);
870 #include "paging_tmpl.h"
874 #include "paging_tmpl.h"
877 static int paging64_init_context_common(struct kvm_vcpu
*vcpu
, int level
)
879 struct kvm_mmu
*context
= &vcpu
->mmu
;
881 ASSERT(is_pae(vcpu
));
882 context
->new_cr3
= paging_new_cr3
;
883 context
->page_fault
= paging64_page_fault
;
884 context
->gva_to_gpa
= paging64_gva_to_gpa
;
885 context
->free
= paging_free
;
886 context
->root_level
= level
;
887 context
->shadow_root_level
= level
;
888 mmu_alloc_roots(vcpu
);
889 ASSERT(VALID_PAGE(context
->root_hpa
));
890 kvm_arch_ops
->set_cr3(vcpu
, context
->root_hpa
|
891 (vcpu
->cr3
& (CR3_PCD_MASK
| CR3_WPT_MASK
)));
895 static int paging64_init_context(struct kvm_vcpu
*vcpu
)
897 return paging64_init_context_common(vcpu
, PT64_ROOT_LEVEL
);
900 static int paging32_init_context(struct kvm_vcpu
*vcpu
)
902 struct kvm_mmu
*context
= &vcpu
->mmu
;
904 context
->new_cr3
= paging_new_cr3
;
905 context
->page_fault
= paging32_page_fault
;
906 context
->gva_to_gpa
= paging32_gva_to_gpa
;
907 context
->free
= paging_free
;
908 context
->root_level
= PT32_ROOT_LEVEL
;
909 context
->shadow_root_level
= PT32E_ROOT_LEVEL
;
910 mmu_alloc_roots(vcpu
);
911 ASSERT(VALID_PAGE(context
->root_hpa
));
912 kvm_arch_ops
->set_cr3(vcpu
, context
->root_hpa
|
913 (vcpu
->cr3
& (CR3_PCD_MASK
| CR3_WPT_MASK
)));
917 static int paging32E_init_context(struct kvm_vcpu
*vcpu
)
919 return paging64_init_context_common(vcpu
, PT32E_ROOT_LEVEL
);
922 static int init_kvm_mmu(struct kvm_vcpu
*vcpu
)
925 ASSERT(!VALID_PAGE(vcpu
->mmu
.root_hpa
));
927 if (!is_paging(vcpu
))
928 return nonpaging_init_context(vcpu
);
929 else if (is_long_mode(vcpu
))
930 return paging64_init_context(vcpu
);
931 else if (is_pae(vcpu
))
932 return paging32E_init_context(vcpu
);
934 return paging32_init_context(vcpu
);
937 static void destroy_kvm_mmu(struct kvm_vcpu
*vcpu
)
940 if (VALID_PAGE(vcpu
->mmu
.root_hpa
)) {
941 vcpu
->mmu
.free(vcpu
);
942 vcpu
->mmu
.root_hpa
= INVALID_PAGE
;
946 int kvm_mmu_reset_context(struct kvm_vcpu
*vcpu
)
948 destroy_kvm_mmu(vcpu
);
949 return init_kvm_mmu(vcpu
);
952 void kvm_mmu_pre_write(struct kvm_vcpu
*vcpu
, gpa_t gpa
, int bytes
)
954 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
955 struct kvm_mmu_page
*page
;
956 struct kvm_mmu_page
*child
;
957 struct hlist_node
*node
;
958 struct hlist_head
*bucket
;
962 unsigned offset
= offset_in_page(gpa
);
963 unsigned page_offset
;
966 pgprintk("%s: gpa %llx bytes %d\n", __FUNCTION__
, gpa
, bytes
);
967 index
= kvm_page_table_hashfn(gfn
) % KVM_NUM_MMU_PAGES
;
968 bucket
= &vcpu
->kvm
->mmu_page_hash
[index
];
969 hlist_for_each_entry(page
, node
, bucket
, hash_link
) {
970 if (page
->gfn
!= gfn
|| page
->role
.metaphysical
)
972 page_offset
= offset
;
973 level
= page
->role
.level
;
974 if (page
->role
.glevels
== PT32_ROOT_LEVEL
) {
975 page_offset
<<= 1; /* 32->64 */
976 page_offset
&= ~PAGE_MASK
;
978 spte
= __va(page
->page_hpa
);
979 spte
+= page_offset
/ sizeof(*spte
);
981 if (is_present_pte(pte
)) {
982 if (level
== PT_PAGE_TABLE_LEVEL
)
983 rmap_remove(vcpu
->kvm
, spte
);
985 child
= page_header(pte
& PT64_BASE_ADDR_MASK
);
986 mmu_page_remove_parent_pte(child
, spte
);
993 void kvm_mmu_post_write(struct kvm_vcpu
*vcpu
, gpa_t gpa
, int bytes
)
997 int kvm_mmu_unprotect_page_virt(struct kvm_vcpu
*vcpu
, gva_t gva
)
999 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, gva
);
1001 return kvm_mmu_unprotect_page(vcpu
, gpa
>> PAGE_SHIFT
);
1004 void kvm_mmu_free_some_pages(struct kvm_vcpu
*vcpu
)
1006 while (vcpu
->kvm
->n_free_mmu_pages
< KVM_REFILL_PAGES
) {
1007 struct kvm_mmu_page
*page
;
1009 page
= container_of(vcpu
->kvm
->active_mmu_pages
.prev
,
1010 struct kvm_mmu_page
, link
);
1011 kvm_mmu_zap_page(vcpu
, page
);
1014 EXPORT_SYMBOL_GPL(kvm_mmu_free_some_pages
);
1016 static void free_mmu_pages(struct kvm_vcpu
*vcpu
)
1018 while (!list_empty(&vcpu
->free_pages
)) {
1019 struct kvm_mmu_page
*page
;
1021 page
= list_entry(vcpu
->free_pages
.next
,
1022 struct kvm_mmu_page
, link
);
1023 list_del(&page
->link
);
1024 __free_page(pfn_to_page(page
->page_hpa
>> PAGE_SHIFT
));
1025 page
->page_hpa
= INVALID_PAGE
;
1027 free_page((unsigned long)vcpu
->mmu
.pae_root
);
1030 static int alloc_mmu_pages(struct kvm_vcpu
*vcpu
)
1037 for (i
= 0; i
< KVM_NUM_MMU_PAGES
; i
++) {
1038 struct kvm_mmu_page
*page_header
= &vcpu
->page_header_buf
[i
];
1040 INIT_LIST_HEAD(&page_header
->link
);
1041 if ((page
= alloc_page(GFP_KERNEL
)) == NULL
)
1043 page
->private = (unsigned long)page_header
;
1044 page_header
->page_hpa
= (hpa_t
)page_to_pfn(page
) << PAGE_SHIFT
;
1045 memset(__va(page_header
->page_hpa
), 0, PAGE_SIZE
);
1046 list_add(&page_header
->link
, &vcpu
->free_pages
);
1047 ++vcpu
->kvm
->n_free_mmu_pages
;
1051 * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64.
1052 * Therefore we need to allocate shadow page tables in the first
1053 * 4GB of memory, which happens to fit the DMA32 zone.
1055 page
= alloc_page(GFP_KERNEL
| __GFP_DMA32
);
1058 vcpu
->mmu
.pae_root
= page_address(page
);
1059 for (i
= 0; i
< 4; ++i
)
1060 vcpu
->mmu
.pae_root
[i
] = INVALID_PAGE
;
1065 free_mmu_pages(vcpu
);
1069 int kvm_mmu_create(struct kvm_vcpu
*vcpu
)
1072 ASSERT(!VALID_PAGE(vcpu
->mmu
.root_hpa
));
1073 ASSERT(list_empty(&vcpu
->free_pages
));
1075 return alloc_mmu_pages(vcpu
);
1078 int kvm_mmu_setup(struct kvm_vcpu
*vcpu
)
1081 ASSERT(!VALID_PAGE(vcpu
->mmu
.root_hpa
));
1082 ASSERT(!list_empty(&vcpu
->free_pages
));
1084 return init_kvm_mmu(vcpu
);
1087 void kvm_mmu_destroy(struct kvm_vcpu
*vcpu
)
1091 destroy_kvm_mmu(vcpu
);
1092 free_mmu_pages(vcpu
);
1095 void kvm_mmu_slot_remove_write_access(struct kvm
*kvm
, int slot
)
1097 struct kvm_mmu_page
*page
;
1099 list_for_each_entry(page
, &kvm
->active_mmu_pages
, link
) {
1103 if (!test_bit(slot
, &page
->slot_bitmap
))
1106 pt
= __va(page
->page_hpa
);
1107 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
)
1109 if (pt
[i
] & PT_WRITABLE_MASK
) {
1110 rmap_remove(kvm
, &pt
[i
]);
1111 pt
[i
] &= ~PT_WRITABLE_MASK
;