2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
9 * Copyright (C) 2006 Qumranet, Inc.
12 * Yaniv Kamay <yaniv@qumranet.com>
13 * Avi Kivity <avi@qumranet.com>
15 * This work is licensed under the terms of the GNU GPL, version 2. See
16 * the COPYING file in the top-level directory.
23 #include <linux/kvm_host.h>
24 #include <linux/types.h>
25 #include <linux/string.h>
27 #include <linux/highmem.h>
28 #include <linux/module.h>
29 #include <linux/swap.h>
30 #include <linux/hugetlb.h>
31 #include <linux/compiler.h>
34 #include <asm/cmpxchg.h>
38 * When setting this variable to true it enables Two-Dimensional-Paging
39 * where the hardware walks 2 page tables:
40 * 1. the guest-virtual to guest-physical
41 * 2. while doing 1. it walks guest-physical to host-physical
42 * If the hardware supports that we don't need to do shadow paging.
44 bool tdp_enabled
= false;
51 static void kvm_mmu_audit(struct kvm_vcpu
*vcpu
, const char *msg
);
53 static void kvm_mmu_audit(struct kvm_vcpu
*vcpu
, const char *msg
) {}
58 #define pgprintk(x...) do { if (dbg) printk(x); } while (0)
59 #define rmap_printk(x...) do { if (dbg) printk(x); } while (0)
63 #define pgprintk(x...) do { } while (0)
64 #define rmap_printk(x...) do { } while (0)
68 #if defined(MMU_DEBUG) || defined(AUDIT)
70 module_param(dbg
, bool, 0644);
74 #define ASSERT(x) do { } while (0)
78 printk(KERN_WARNING "assertion failed %s:%d: %s\n", \
79 __FILE__, __LINE__, #x); \
83 #define PT_FIRST_AVAIL_BITS_SHIFT 9
84 #define PT64_SECOND_AVAIL_BITS_SHIFT 52
86 #define VALID_PAGE(x) ((x) != INVALID_PAGE)
88 #define PT64_LEVEL_BITS 9
90 #define PT64_LEVEL_SHIFT(level) \
91 (PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS)
93 #define PT64_LEVEL_MASK(level) \
94 (((1ULL << PT64_LEVEL_BITS) - 1) << PT64_LEVEL_SHIFT(level))
96 #define PT64_INDEX(address, level)\
97 (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1))
100 #define PT32_LEVEL_BITS 10
102 #define PT32_LEVEL_SHIFT(level) \
103 (PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS)
105 #define PT32_LEVEL_MASK(level) \
106 (((1ULL << PT32_LEVEL_BITS) - 1) << PT32_LEVEL_SHIFT(level))
108 #define PT32_INDEX(address, level)\
109 (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1))
112 #define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1))
113 #define PT64_DIR_BASE_ADDR_MASK \
114 (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1))
116 #define PT32_BASE_ADDR_MASK PAGE_MASK
117 #define PT32_DIR_BASE_ADDR_MASK \
118 (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1))
120 #define PT64_PERM_MASK (PT_PRESENT_MASK | PT_WRITABLE_MASK | PT_USER_MASK \
123 #define PFERR_PRESENT_MASK (1U << 0)
124 #define PFERR_WRITE_MASK (1U << 1)
125 #define PFERR_USER_MASK (1U << 2)
126 #define PFERR_FETCH_MASK (1U << 4)
128 #define PT_DIRECTORY_LEVEL 2
129 #define PT_PAGE_TABLE_LEVEL 1
133 #define ACC_EXEC_MASK 1
134 #define ACC_WRITE_MASK PT_WRITABLE_MASK
135 #define ACC_USER_MASK PT_USER_MASK
136 #define ACC_ALL (ACC_EXEC_MASK | ACC_WRITE_MASK | ACC_USER_MASK)
138 #define SHADOW_PT_INDEX(addr, level) PT64_INDEX(addr, level)
140 struct kvm_rmap_desc
{
141 u64
*shadow_ptes
[RMAP_EXT
];
142 struct kvm_rmap_desc
*more
;
145 struct kvm_shadow_walk
{
146 int (*entry
)(struct kvm_shadow_walk
*walk
, struct kvm_vcpu
*vcpu
,
147 gva_t addr
, u64
*spte
, int level
);
150 static struct kmem_cache
*pte_chain_cache
;
151 static struct kmem_cache
*rmap_desc_cache
;
152 static struct kmem_cache
*mmu_page_header_cache
;
154 static u64 __read_mostly shadow_trap_nonpresent_pte
;
155 static u64 __read_mostly shadow_notrap_nonpresent_pte
;
156 static u64 __read_mostly shadow_base_present_pte
;
157 static u64 __read_mostly shadow_nx_mask
;
158 static u64 __read_mostly shadow_x_mask
; /* mutual exclusive with nx_mask */
159 static u64 __read_mostly shadow_user_mask
;
160 static u64 __read_mostly shadow_accessed_mask
;
161 static u64 __read_mostly shadow_dirty_mask
;
163 void kvm_mmu_set_nonpresent_ptes(u64 trap_pte
, u64 notrap_pte
)
165 shadow_trap_nonpresent_pte
= trap_pte
;
166 shadow_notrap_nonpresent_pte
= notrap_pte
;
168 EXPORT_SYMBOL_GPL(kvm_mmu_set_nonpresent_ptes
);
170 void kvm_mmu_set_base_ptes(u64 base_pte
)
172 shadow_base_present_pte
= base_pte
;
174 EXPORT_SYMBOL_GPL(kvm_mmu_set_base_ptes
);
176 void kvm_mmu_set_mask_ptes(u64 user_mask
, u64 accessed_mask
,
177 u64 dirty_mask
, u64 nx_mask
, u64 x_mask
)
179 shadow_user_mask
= user_mask
;
180 shadow_accessed_mask
= accessed_mask
;
181 shadow_dirty_mask
= dirty_mask
;
182 shadow_nx_mask
= nx_mask
;
183 shadow_x_mask
= x_mask
;
185 EXPORT_SYMBOL_GPL(kvm_mmu_set_mask_ptes
);
187 static int is_write_protection(struct kvm_vcpu
*vcpu
)
189 return vcpu
->arch
.cr0
& X86_CR0_WP
;
192 static int is_cpuid_PSE36(void)
197 static int is_nx(struct kvm_vcpu
*vcpu
)
199 return vcpu
->arch
.shadow_efer
& EFER_NX
;
202 static int is_present_pte(unsigned long pte
)
204 return pte
& PT_PRESENT_MASK
;
207 static int is_shadow_present_pte(u64 pte
)
209 return pte
!= shadow_trap_nonpresent_pte
210 && pte
!= shadow_notrap_nonpresent_pte
;
213 static int is_large_pte(u64 pte
)
215 return pte
& PT_PAGE_SIZE_MASK
;
218 static int is_writeble_pte(unsigned long pte
)
220 return pte
& PT_WRITABLE_MASK
;
223 static int is_dirty_pte(unsigned long pte
)
225 return pte
& shadow_dirty_mask
;
228 static int is_rmap_pte(u64 pte
)
230 return is_shadow_present_pte(pte
);
233 static pfn_t
spte_to_pfn(u64 pte
)
235 return (pte
& PT64_BASE_ADDR_MASK
) >> PAGE_SHIFT
;
238 static gfn_t
pse36_gfn_delta(u32 gpte
)
240 int shift
= 32 - PT32_DIR_PSE36_SHIFT
- PAGE_SHIFT
;
242 return (gpte
& PT32_DIR_PSE36_MASK
) << shift
;
245 static void set_shadow_pte(u64
*sptep
, u64 spte
)
248 set_64bit((unsigned long *)sptep
, spte
);
250 set_64bit((unsigned long long *)sptep
, spte
);
254 static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache
*cache
,
255 struct kmem_cache
*base_cache
, int min
)
259 if (cache
->nobjs
>= min
)
261 while (cache
->nobjs
< ARRAY_SIZE(cache
->objects
)) {
262 obj
= kmem_cache_zalloc(base_cache
, GFP_KERNEL
);
265 cache
->objects
[cache
->nobjs
++] = obj
;
270 static void mmu_free_memory_cache(struct kvm_mmu_memory_cache
*mc
)
273 kfree(mc
->objects
[--mc
->nobjs
]);
276 static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache
*cache
,
281 if (cache
->nobjs
>= min
)
283 while (cache
->nobjs
< ARRAY_SIZE(cache
->objects
)) {
284 page
= alloc_page(GFP_KERNEL
);
287 set_page_private(page
, 0);
288 cache
->objects
[cache
->nobjs
++] = page_address(page
);
293 static void mmu_free_memory_cache_page(struct kvm_mmu_memory_cache
*mc
)
296 free_page((unsigned long)mc
->objects
[--mc
->nobjs
]);
299 static int mmu_topup_memory_caches(struct kvm_vcpu
*vcpu
)
303 r
= mmu_topup_memory_cache(&vcpu
->arch
.mmu_pte_chain_cache
,
307 r
= mmu_topup_memory_cache(&vcpu
->arch
.mmu_rmap_desc_cache
,
311 r
= mmu_topup_memory_cache_page(&vcpu
->arch
.mmu_page_cache
, 8);
314 r
= mmu_topup_memory_cache(&vcpu
->arch
.mmu_page_header_cache
,
315 mmu_page_header_cache
, 4);
320 static void mmu_free_memory_caches(struct kvm_vcpu
*vcpu
)
322 mmu_free_memory_cache(&vcpu
->arch
.mmu_pte_chain_cache
);
323 mmu_free_memory_cache(&vcpu
->arch
.mmu_rmap_desc_cache
);
324 mmu_free_memory_cache_page(&vcpu
->arch
.mmu_page_cache
);
325 mmu_free_memory_cache(&vcpu
->arch
.mmu_page_header_cache
);
328 static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache
*mc
,
334 p
= mc
->objects
[--mc
->nobjs
];
339 static struct kvm_pte_chain
*mmu_alloc_pte_chain(struct kvm_vcpu
*vcpu
)
341 return mmu_memory_cache_alloc(&vcpu
->arch
.mmu_pte_chain_cache
,
342 sizeof(struct kvm_pte_chain
));
345 static void mmu_free_pte_chain(struct kvm_pte_chain
*pc
)
350 static struct kvm_rmap_desc
*mmu_alloc_rmap_desc(struct kvm_vcpu
*vcpu
)
352 return mmu_memory_cache_alloc(&vcpu
->arch
.mmu_rmap_desc_cache
,
353 sizeof(struct kvm_rmap_desc
));
356 static void mmu_free_rmap_desc(struct kvm_rmap_desc
*rd
)
362 * Return the pointer to the largepage write count for a given
363 * gfn, handling slots that are not large page aligned.
365 static int *slot_largepage_idx(gfn_t gfn
, struct kvm_memory_slot
*slot
)
369 idx
= (gfn
/ KVM_PAGES_PER_HPAGE
) -
370 (slot
->base_gfn
/ KVM_PAGES_PER_HPAGE
);
371 return &slot
->lpage_info
[idx
].write_count
;
374 static void account_shadowed(struct kvm
*kvm
, gfn_t gfn
)
378 write_count
= slot_largepage_idx(gfn
, gfn_to_memslot(kvm
, gfn
));
382 static void unaccount_shadowed(struct kvm
*kvm
, gfn_t gfn
)
386 write_count
= slot_largepage_idx(gfn
, gfn_to_memslot(kvm
, gfn
));
388 WARN_ON(*write_count
< 0);
391 static int has_wrprotected_page(struct kvm
*kvm
, gfn_t gfn
)
393 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
397 largepage_idx
= slot_largepage_idx(gfn
, slot
);
398 return *largepage_idx
;
404 static int host_largepage_backed(struct kvm
*kvm
, gfn_t gfn
)
406 struct vm_area_struct
*vma
;
409 addr
= gfn_to_hva(kvm
, gfn
);
410 if (kvm_is_error_hva(addr
))
413 vma
= find_vma(current
->mm
, addr
);
414 if (vma
&& is_vm_hugetlb_page(vma
))
420 static int is_largepage_backed(struct kvm_vcpu
*vcpu
, gfn_t large_gfn
)
422 struct kvm_memory_slot
*slot
;
424 if (has_wrprotected_page(vcpu
->kvm
, large_gfn
))
427 if (!host_largepage_backed(vcpu
->kvm
, large_gfn
))
430 slot
= gfn_to_memslot(vcpu
->kvm
, large_gfn
);
431 if (slot
&& slot
->dirty_bitmap
)
438 * Take gfn and return the reverse mapping to it.
439 * Note: gfn must be unaliased before this function get called
442 static unsigned long *gfn_to_rmap(struct kvm
*kvm
, gfn_t gfn
, int lpage
)
444 struct kvm_memory_slot
*slot
;
447 slot
= gfn_to_memslot(kvm
, gfn
);
449 return &slot
->rmap
[gfn
- slot
->base_gfn
];
451 idx
= (gfn
/ KVM_PAGES_PER_HPAGE
) -
452 (slot
->base_gfn
/ KVM_PAGES_PER_HPAGE
);
454 return &slot
->lpage_info
[idx
].rmap_pde
;
458 * Reverse mapping data structures:
460 * If rmapp bit zero is zero, then rmapp point to the shadw page table entry
461 * that points to page_address(page).
463 * If rmapp bit zero is one, (then rmap & ~1) points to a struct kvm_rmap_desc
464 * containing more mappings.
466 static void rmap_add(struct kvm_vcpu
*vcpu
, u64
*spte
, gfn_t gfn
, int lpage
)
468 struct kvm_mmu_page
*sp
;
469 struct kvm_rmap_desc
*desc
;
470 unsigned long *rmapp
;
473 if (!is_rmap_pte(*spte
))
475 gfn
= unalias_gfn(vcpu
->kvm
, gfn
);
476 sp
= page_header(__pa(spte
));
477 sp
->gfns
[spte
- sp
->spt
] = gfn
;
478 rmapp
= gfn_to_rmap(vcpu
->kvm
, gfn
, lpage
);
480 rmap_printk("rmap_add: %p %llx 0->1\n", spte
, *spte
);
481 *rmapp
= (unsigned long)spte
;
482 } else if (!(*rmapp
& 1)) {
483 rmap_printk("rmap_add: %p %llx 1->many\n", spte
, *spte
);
484 desc
= mmu_alloc_rmap_desc(vcpu
);
485 desc
->shadow_ptes
[0] = (u64
*)*rmapp
;
486 desc
->shadow_ptes
[1] = spte
;
487 *rmapp
= (unsigned long)desc
| 1;
489 rmap_printk("rmap_add: %p %llx many->many\n", spte
, *spte
);
490 desc
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
491 while (desc
->shadow_ptes
[RMAP_EXT
-1] && desc
->more
)
493 if (desc
->shadow_ptes
[RMAP_EXT
-1]) {
494 desc
->more
= mmu_alloc_rmap_desc(vcpu
);
497 for (i
= 0; desc
->shadow_ptes
[i
]; ++i
)
499 desc
->shadow_ptes
[i
] = spte
;
503 static void rmap_desc_remove_entry(unsigned long *rmapp
,
504 struct kvm_rmap_desc
*desc
,
506 struct kvm_rmap_desc
*prev_desc
)
510 for (j
= RMAP_EXT
- 1; !desc
->shadow_ptes
[j
] && j
> i
; --j
)
512 desc
->shadow_ptes
[i
] = desc
->shadow_ptes
[j
];
513 desc
->shadow_ptes
[j
] = NULL
;
516 if (!prev_desc
&& !desc
->more
)
517 *rmapp
= (unsigned long)desc
->shadow_ptes
[0];
520 prev_desc
->more
= desc
->more
;
522 *rmapp
= (unsigned long)desc
->more
| 1;
523 mmu_free_rmap_desc(desc
);
526 static void rmap_remove(struct kvm
*kvm
, u64
*spte
)
528 struct kvm_rmap_desc
*desc
;
529 struct kvm_rmap_desc
*prev_desc
;
530 struct kvm_mmu_page
*sp
;
532 unsigned long *rmapp
;
535 if (!is_rmap_pte(*spte
))
537 sp
= page_header(__pa(spte
));
538 pfn
= spte_to_pfn(*spte
);
539 if (*spte
& shadow_accessed_mask
)
540 kvm_set_pfn_accessed(pfn
);
541 if (is_writeble_pte(*spte
))
542 kvm_release_pfn_dirty(pfn
);
544 kvm_release_pfn_clean(pfn
);
545 rmapp
= gfn_to_rmap(kvm
, sp
->gfns
[spte
- sp
->spt
], is_large_pte(*spte
));
547 printk(KERN_ERR
"rmap_remove: %p %llx 0->BUG\n", spte
, *spte
);
549 } else if (!(*rmapp
& 1)) {
550 rmap_printk("rmap_remove: %p %llx 1->0\n", spte
, *spte
);
551 if ((u64
*)*rmapp
!= spte
) {
552 printk(KERN_ERR
"rmap_remove: %p %llx 1->BUG\n",
558 rmap_printk("rmap_remove: %p %llx many->many\n", spte
, *spte
);
559 desc
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
562 for (i
= 0; i
< RMAP_EXT
&& desc
->shadow_ptes
[i
]; ++i
)
563 if (desc
->shadow_ptes
[i
] == spte
) {
564 rmap_desc_remove_entry(rmapp
,
576 static u64
*rmap_next(struct kvm
*kvm
, unsigned long *rmapp
, u64
*spte
)
578 struct kvm_rmap_desc
*desc
;
579 struct kvm_rmap_desc
*prev_desc
;
585 else if (!(*rmapp
& 1)) {
587 return (u64
*)*rmapp
;
590 desc
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
594 for (i
= 0; i
< RMAP_EXT
&& desc
->shadow_ptes
[i
]; ++i
) {
595 if (prev_spte
== spte
)
596 return desc
->shadow_ptes
[i
];
597 prev_spte
= desc
->shadow_ptes
[i
];
604 static void rmap_write_protect(struct kvm
*kvm
, u64 gfn
)
606 unsigned long *rmapp
;
608 int write_protected
= 0;
610 gfn
= unalias_gfn(kvm
, gfn
);
611 rmapp
= gfn_to_rmap(kvm
, gfn
, 0);
613 spte
= rmap_next(kvm
, rmapp
, NULL
);
616 BUG_ON(!(*spte
& PT_PRESENT_MASK
));
617 rmap_printk("rmap_write_protect: spte %p %llx\n", spte
, *spte
);
618 if (is_writeble_pte(*spte
)) {
619 set_shadow_pte(spte
, *spte
& ~PT_WRITABLE_MASK
);
622 spte
= rmap_next(kvm
, rmapp
, spte
);
624 if (write_protected
) {
627 spte
= rmap_next(kvm
, rmapp
, NULL
);
628 pfn
= spte_to_pfn(*spte
);
629 kvm_set_pfn_dirty(pfn
);
632 /* check for huge page mappings */
633 rmapp
= gfn_to_rmap(kvm
, gfn
, 1);
634 spte
= rmap_next(kvm
, rmapp
, NULL
);
637 BUG_ON(!(*spte
& PT_PRESENT_MASK
));
638 BUG_ON((*spte
& (PT_PAGE_SIZE_MASK
|PT_PRESENT_MASK
)) != (PT_PAGE_SIZE_MASK
|PT_PRESENT_MASK
));
639 pgprintk("rmap_write_protect(large): spte %p %llx %lld\n", spte
, *spte
, gfn
);
640 if (is_writeble_pte(*spte
)) {
641 rmap_remove(kvm
, spte
);
643 set_shadow_pte(spte
, shadow_trap_nonpresent_pte
);
647 spte
= rmap_next(kvm
, rmapp
, spte
);
651 kvm_flush_remote_tlbs(kvm
);
653 account_shadowed(kvm
, gfn
);
656 static int kvm_unmap_rmapp(struct kvm
*kvm
, unsigned long *rmapp
)
659 int need_tlb_flush
= 0;
661 while ((spte
= rmap_next(kvm
, rmapp
, NULL
))) {
662 BUG_ON(!(*spte
& PT_PRESENT_MASK
));
663 rmap_printk("kvm_rmap_unmap_hva: spte %p %llx\n", spte
, *spte
);
664 rmap_remove(kvm
, spte
);
665 set_shadow_pte(spte
, shadow_trap_nonpresent_pte
);
668 return need_tlb_flush
;
671 static int kvm_handle_hva(struct kvm
*kvm
, unsigned long hva
,
672 int (*handler
)(struct kvm
*kvm
, unsigned long *rmapp
))
678 * If mmap_sem isn't taken, we can look the memslots with only
679 * the mmu_lock by skipping over the slots with userspace_addr == 0.
681 for (i
= 0; i
< kvm
->nmemslots
; i
++) {
682 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[i
];
683 unsigned long start
= memslot
->userspace_addr
;
686 /* mmu_lock protects userspace_addr */
690 end
= start
+ (memslot
->npages
<< PAGE_SHIFT
);
691 if (hva
>= start
&& hva
< end
) {
692 gfn_t gfn_offset
= (hva
- start
) >> PAGE_SHIFT
;
693 retval
|= handler(kvm
, &memslot
->rmap
[gfn_offset
]);
694 retval
|= handler(kvm
,
695 &memslot
->lpage_info
[
697 KVM_PAGES_PER_HPAGE
].rmap_pde
);
704 int kvm_unmap_hva(struct kvm
*kvm
, unsigned long hva
)
706 return kvm_handle_hva(kvm
, hva
, kvm_unmap_rmapp
);
709 static int kvm_age_rmapp(struct kvm
*kvm
, unsigned long *rmapp
)
714 /* always return old for EPT */
715 if (!shadow_accessed_mask
)
718 spte
= rmap_next(kvm
, rmapp
, NULL
);
722 BUG_ON(!(_spte
& PT_PRESENT_MASK
));
723 _young
= _spte
& PT_ACCESSED_MASK
;
726 clear_bit(PT_ACCESSED_SHIFT
, (unsigned long *)spte
);
728 spte
= rmap_next(kvm
, rmapp
, spte
);
733 int kvm_age_hva(struct kvm
*kvm
, unsigned long hva
)
735 return kvm_handle_hva(kvm
, hva
, kvm_age_rmapp
);
739 static int is_empty_shadow_page(u64
*spt
)
744 for (pos
= spt
, end
= pos
+ PAGE_SIZE
/ sizeof(u64
); pos
!= end
; pos
++)
745 if (is_shadow_present_pte(*pos
)) {
746 printk(KERN_ERR
"%s: %p %llx\n", __func__
,
754 static void kvm_mmu_free_page(struct kvm
*kvm
, struct kvm_mmu_page
*sp
)
756 ASSERT(is_empty_shadow_page(sp
->spt
));
758 __free_page(virt_to_page(sp
->spt
));
759 __free_page(virt_to_page(sp
->gfns
));
761 ++kvm
->arch
.n_free_mmu_pages
;
764 static unsigned kvm_page_table_hashfn(gfn_t gfn
)
766 return gfn
& ((1 << KVM_MMU_HASH_SHIFT
) - 1);
769 static struct kvm_mmu_page
*kvm_mmu_alloc_page(struct kvm_vcpu
*vcpu
,
772 struct kvm_mmu_page
*sp
;
774 sp
= mmu_memory_cache_alloc(&vcpu
->arch
.mmu_page_header_cache
, sizeof *sp
);
775 sp
->spt
= mmu_memory_cache_alloc(&vcpu
->arch
.mmu_page_cache
, PAGE_SIZE
);
776 sp
->gfns
= mmu_memory_cache_alloc(&vcpu
->arch
.mmu_page_cache
, PAGE_SIZE
);
777 set_page_private(virt_to_page(sp
->spt
), (unsigned long)sp
);
778 list_add(&sp
->link
, &vcpu
->kvm
->arch
.active_mmu_pages
);
779 ASSERT(is_empty_shadow_page(sp
->spt
));
782 sp
->parent_pte
= parent_pte
;
783 --vcpu
->kvm
->arch
.n_free_mmu_pages
;
787 static void mmu_page_add_parent_pte(struct kvm_vcpu
*vcpu
,
788 struct kvm_mmu_page
*sp
, u64
*parent_pte
)
790 struct kvm_pte_chain
*pte_chain
;
791 struct hlist_node
*node
;
796 if (!sp
->multimapped
) {
797 u64
*old
= sp
->parent_pte
;
800 sp
->parent_pte
= parent_pte
;
804 pte_chain
= mmu_alloc_pte_chain(vcpu
);
805 INIT_HLIST_HEAD(&sp
->parent_ptes
);
806 hlist_add_head(&pte_chain
->link
, &sp
->parent_ptes
);
807 pte_chain
->parent_ptes
[0] = old
;
809 hlist_for_each_entry(pte_chain
, node
, &sp
->parent_ptes
, link
) {
810 if (pte_chain
->parent_ptes
[NR_PTE_CHAIN_ENTRIES
-1])
812 for (i
= 0; i
< NR_PTE_CHAIN_ENTRIES
; ++i
)
813 if (!pte_chain
->parent_ptes
[i
]) {
814 pte_chain
->parent_ptes
[i
] = parent_pte
;
818 pte_chain
= mmu_alloc_pte_chain(vcpu
);
820 hlist_add_head(&pte_chain
->link
, &sp
->parent_ptes
);
821 pte_chain
->parent_ptes
[0] = parent_pte
;
824 static void mmu_page_remove_parent_pte(struct kvm_mmu_page
*sp
,
827 struct kvm_pte_chain
*pte_chain
;
828 struct hlist_node
*node
;
831 if (!sp
->multimapped
) {
832 BUG_ON(sp
->parent_pte
!= parent_pte
);
833 sp
->parent_pte
= NULL
;
836 hlist_for_each_entry(pte_chain
, node
, &sp
->parent_ptes
, link
)
837 for (i
= 0; i
< NR_PTE_CHAIN_ENTRIES
; ++i
) {
838 if (!pte_chain
->parent_ptes
[i
])
840 if (pte_chain
->parent_ptes
[i
] != parent_pte
)
842 while (i
+ 1 < NR_PTE_CHAIN_ENTRIES
843 && pte_chain
->parent_ptes
[i
+ 1]) {
844 pte_chain
->parent_ptes
[i
]
845 = pte_chain
->parent_ptes
[i
+ 1];
848 pte_chain
->parent_ptes
[i
] = NULL
;
850 hlist_del(&pte_chain
->link
);
851 mmu_free_pte_chain(pte_chain
);
852 if (hlist_empty(&sp
->parent_ptes
)) {
854 sp
->parent_pte
= NULL
;
862 static void nonpaging_prefetch_page(struct kvm_vcpu
*vcpu
,
863 struct kvm_mmu_page
*sp
)
867 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
)
868 sp
->spt
[i
] = shadow_trap_nonpresent_pte
;
871 static struct kvm_mmu_page
*kvm_mmu_lookup_page(struct kvm
*kvm
, gfn_t gfn
)
874 struct hlist_head
*bucket
;
875 struct kvm_mmu_page
*sp
;
876 struct hlist_node
*node
;
878 pgprintk("%s: looking for gfn %lx\n", __func__
, gfn
);
879 index
= kvm_page_table_hashfn(gfn
);
880 bucket
= &kvm
->arch
.mmu_page_hash
[index
];
881 hlist_for_each_entry(sp
, node
, bucket
, hash_link
)
882 if (sp
->gfn
== gfn
&& !sp
->role
.metaphysical
883 && !sp
->role
.invalid
) {
884 pgprintk("%s: found role %x\n",
885 __func__
, sp
->role
.word
);
891 static struct kvm_mmu_page
*kvm_mmu_get_page(struct kvm_vcpu
*vcpu
,
899 union kvm_mmu_page_role role
;
902 struct hlist_head
*bucket
;
903 struct kvm_mmu_page
*sp
;
904 struct hlist_node
*node
;
907 role
.glevels
= vcpu
->arch
.mmu
.root_level
;
909 role
.metaphysical
= metaphysical
;
910 role
.access
= access
;
911 if (vcpu
->arch
.mmu
.root_level
<= PT32_ROOT_LEVEL
) {
912 quadrant
= gaddr
>> (PAGE_SHIFT
+ (PT64_PT_BITS
* level
));
913 quadrant
&= (1 << ((PT32_PT_BITS
- PT64_PT_BITS
) * level
)) - 1;
914 role
.quadrant
= quadrant
;
916 pgprintk("%s: looking gfn %lx role %x\n", __func__
,
918 index
= kvm_page_table_hashfn(gfn
);
919 bucket
= &vcpu
->kvm
->arch
.mmu_page_hash
[index
];
920 hlist_for_each_entry(sp
, node
, bucket
, hash_link
)
921 if (sp
->gfn
== gfn
&& sp
->role
.word
== role
.word
) {
922 mmu_page_add_parent_pte(vcpu
, sp
, parent_pte
);
923 pgprintk("%s: found\n", __func__
);
926 ++vcpu
->kvm
->stat
.mmu_cache_miss
;
927 sp
= kvm_mmu_alloc_page(vcpu
, parent_pte
);
930 pgprintk("%s: adding gfn %lx role %x\n", __func__
, gfn
, role
.word
);
933 hlist_add_head(&sp
->hash_link
, bucket
);
935 rmap_write_protect(vcpu
->kvm
, gfn
);
936 if (shadow_trap_nonpresent_pte
!= shadow_notrap_nonpresent_pte
)
937 vcpu
->arch
.mmu
.prefetch_page(vcpu
, sp
);
939 nonpaging_prefetch_page(vcpu
, sp
);
943 static int walk_shadow(struct kvm_shadow_walk
*walker
,
944 struct kvm_vcpu
*vcpu
, gva_t addr
)
952 shadow_addr
= vcpu
->arch
.mmu
.root_hpa
;
953 level
= vcpu
->arch
.mmu
.shadow_root_level
;
954 if (level
== PT32E_ROOT_LEVEL
) {
955 shadow_addr
= vcpu
->arch
.mmu
.pae_root
[(addr
>> 30) & 3];
956 shadow_addr
&= PT64_BASE_ADDR_MASK
;
960 while (level
>= PT_PAGE_TABLE_LEVEL
) {
961 index
= SHADOW_PT_INDEX(addr
, level
);
962 sptep
= ((u64
*)__va(shadow_addr
)) + index
;
963 r
= walker
->entry(walker
, vcpu
, addr
, sptep
, level
);
966 shadow_addr
= *sptep
& PT64_BASE_ADDR_MASK
;
972 static void kvm_mmu_page_unlink_children(struct kvm
*kvm
,
973 struct kvm_mmu_page
*sp
)
981 if (sp
->role
.level
== PT_PAGE_TABLE_LEVEL
) {
982 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
983 if (is_shadow_present_pte(pt
[i
]))
984 rmap_remove(kvm
, &pt
[i
]);
985 pt
[i
] = shadow_trap_nonpresent_pte
;
990 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
993 if (is_shadow_present_pte(ent
)) {
994 if (!is_large_pte(ent
)) {
995 ent
&= PT64_BASE_ADDR_MASK
;
996 mmu_page_remove_parent_pte(page_header(ent
),
1000 rmap_remove(kvm
, &pt
[i
]);
1003 pt
[i
] = shadow_trap_nonpresent_pte
;
1007 static void kvm_mmu_put_page(struct kvm_mmu_page
*sp
, u64
*parent_pte
)
1009 mmu_page_remove_parent_pte(sp
, parent_pte
);
1012 static void kvm_mmu_reset_last_pte_updated(struct kvm
*kvm
)
1016 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
)
1018 kvm
->vcpus
[i
]->arch
.last_pte_updated
= NULL
;
1021 static void kvm_mmu_unlink_parents(struct kvm
*kvm
, struct kvm_mmu_page
*sp
)
1025 while (sp
->multimapped
|| sp
->parent_pte
) {
1026 if (!sp
->multimapped
)
1027 parent_pte
= sp
->parent_pte
;
1029 struct kvm_pte_chain
*chain
;
1031 chain
= container_of(sp
->parent_ptes
.first
,
1032 struct kvm_pte_chain
, link
);
1033 parent_pte
= chain
->parent_ptes
[0];
1035 BUG_ON(!parent_pte
);
1036 kvm_mmu_put_page(sp
, parent_pte
);
1037 set_shadow_pte(parent_pte
, shadow_trap_nonpresent_pte
);
1041 static void kvm_mmu_zap_page(struct kvm
*kvm
, struct kvm_mmu_page
*sp
)
1043 ++kvm
->stat
.mmu_shadow_zapped
;
1044 kvm_mmu_page_unlink_children(kvm
, sp
);
1045 kvm_mmu_unlink_parents(kvm
, sp
);
1046 kvm_flush_remote_tlbs(kvm
);
1047 if (!sp
->role
.invalid
&& !sp
->role
.metaphysical
)
1048 unaccount_shadowed(kvm
, sp
->gfn
);
1049 if (!sp
->root_count
) {
1050 hlist_del(&sp
->hash_link
);
1051 kvm_mmu_free_page(kvm
, sp
);
1053 sp
->role
.invalid
= 1;
1054 list_move(&sp
->link
, &kvm
->arch
.active_mmu_pages
);
1055 kvm_reload_remote_mmus(kvm
);
1057 kvm_mmu_reset_last_pte_updated(kvm
);
1061 * Changing the number of mmu pages allocated to the vm
1062 * Note: if kvm_nr_mmu_pages is too small, you will get dead lock
1064 void kvm_mmu_change_mmu_pages(struct kvm
*kvm
, unsigned int kvm_nr_mmu_pages
)
1067 * If we set the number of mmu pages to be smaller be than the
1068 * number of actived pages , we must to free some mmu pages before we
1072 if ((kvm
->arch
.n_alloc_mmu_pages
- kvm
->arch
.n_free_mmu_pages
) >
1074 int n_used_mmu_pages
= kvm
->arch
.n_alloc_mmu_pages
1075 - kvm
->arch
.n_free_mmu_pages
;
1077 while (n_used_mmu_pages
> kvm_nr_mmu_pages
) {
1078 struct kvm_mmu_page
*page
;
1080 page
= container_of(kvm
->arch
.active_mmu_pages
.prev
,
1081 struct kvm_mmu_page
, link
);
1082 kvm_mmu_zap_page(kvm
, page
);
1085 kvm
->arch
.n_free_mmu_pages
= 0;
1088 kvm
->arch
.n_free_mmu_pages
+= kvm_nr_mmu_pages
1089 - kvm
->arch
.n_alloc_mmu_pages
;
1091 kvm
->arch
.n_alloc_mmu_pages
= kvm_nr_mmu_pages
;
1094 static int kvm_mmu_unprotect_page(struct kvm
*kvm
, gfn_t gfn
)
1097 struct hlist_head
*bucket
;
1098 struct kvm_mmu_page
*sp
;
1099 struct hlist_node
*node
, *n
;
1102 pgprintk("%s: looking for gfn %lx\n", __func__
, gfn
);
1104 index
= kvm_page_table_hashfn(gfn
);
1105 bucket
= &kvm
->arch
.mmu_page_hash
[index
];
1106 hlist_for_each_entry_safe(sp
, node
, n
, bucket
, hash_link
)
1107 if (sp
->gfn
== gfn
&& !sp
->role
.metaphysical
) {
1108 pgprintk("%s: gfn %lx role %x\n", __func__
, gfn
,
1110 kvm_mmu_zap_page(kvm
, sp
);
1116 static void mmu_unshadow(struct kvm
*kvm
, gfn_t gfn
)
1118 struct kvm_mmu_page
*sp
;
1120 while ((sp
= kvm_mmu_lookup_page(kvm
, gfn
)) != NULL
) {
1121 pgprintk("%s: zap %lx %x\n", __func__
, gfn
, sp
->role
.word
);
1122 kvm_mmu_zap_page(kvm
, sp
);
1126 static void page_header_update_slot(struct kvm
*kvm
, void *pte
, gfn_t gfn
)
1128 int slot
= memslot_id(kvm
, gfn_to_memslot(kvm
, gfn
));
1129 struct kvm_mmu_page
*sp
= page_header(__pa(pte
));
1131 __set_bit(slot
, &sp
->slot_bitmap
);
1134 struct page
*gva_to_page(struct kvm_vcpu
*vcpu
, gva_t gva
)
1138 gpa_t gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, gva
);
1140 if (gpa
== UNMAPPED_GVA
)
1143 down_read(¤t
->mm
->mmap_sem
);
1144 page
= gfn_to_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
1145 up_read(¤t
->mm
->mmap_sem
);
1150 static void mmu_set_spte(struct kvm_vcpu
*vcpu
, u64
*shadow_pte
,
1151 unsigned pt_access
, unsigned pte_access
,
1152 int user_fault
, int write_fault
, int dirty
,
1153 int *ptwrite
, int largepage
, gfn_t gfn
,
1154 pfn_t pfn
, bool speculative
)
1157 int was_rmapped
= 0;
1158 int was_writeble
= is_writeble_pte(*shadow_pte
);
1160 pgprintk("%s: spte %llx access %x write_fault %d"
1161 " user_fault %d gfn %lx\n",
1162 __func__
, *shadow_pte
, pt_access
,
1163 write_fault
, user_fault
, gfn
);
1165 if (is_rmap_pte(*shadow_pte
)) {
1167 * If we overwrite a PTE page pointer with a 2MB PMD, unlink
1168 * the parent of the now unreachable PTE.
1170 if (largepage
&& !is_large_pte(*shadow_pte
)) {
1171 struct kvm_mmu_page
*child
;
1172 u64 pte
= *shadow_pte
;
1174 child
= page_header(pte
& PT64_BASE_ADDR_MASK
);
1175 mmu_page_remove_parent_pte(child
, shadow_pte
);
1176 } else if (pfn
!= spte_to_pfn(*shadow_pte
)) {
1177 pgprintk("hfn old %lx new %lx\n",
1178 spte_to_pfn(*shadow_pte
), pfn
);
1179 rmap_remove(vcpu
->kvm
, shadow_pte
);
1182 was_rmapped
= is_large_pte(*shadow_pte
);
1189 * We don't set the accessed bit, since we sometimes want to see
1190 * whether the guest actually used the pte (in order to detect
1193 spte
= shadow_base_present_pte
| shadow_dirty_mask
;
1195 pte_access
|= PT_ACCESSED_MASK
;
1197 pte_access
&= ~ACC_WRITE_MASK
;
1198 if (pte_access
& ACC_EXEC_MASK
)
1199 spte
|= shadow_x_mask
;
1201 spte
|= shadow_nx_mask
;
1202 if (pte_access
& ACC_USER_MASK
)
1203 spte
|= shadow_user_mask
;
1205 spte
|= PT_PAGE_SIZE_MASK
;
1207 spte
|= (u64
)pfn
<< PAGE_SHIFT
;
1209 if ((pte_access
& ACC_WRITE_MASK
)
1210 || (write_fault
&& !is_write_protection(vcpu
) && !user_fault
)) {
1211 struct kvm_mmu_page
*shadow
;
1213 spte
|= PT_WRITABLE_MASK
;
1215 shadow
= kvm_mmu_lookup_page(vcpu
->kvm
, gfn
);
1217 (largepage
&& has_wrprotected_page(vcpu
->kvm
, gfn
))) {
1218 pgprintk("%s: found shadow page for %lx, marking ro\n",
1220 pte_access
&= ~ACC_WRITE_MASK
;
1221 if (is_writeble_pte(spte
)) {
1222 spte
&= ~PT_WRITABLE_MASK
;
1223 kvm_x86_ops
->tlb_flush(vcpu
);
1230 if (pte_access
& ACC_WRITE_MASK
)
1231 mark_page_dirty(vcpu
->kvm
, gfn
);
1233 pgprintk("%s: setting spte %llx\n", __func__
, spte
);
1234 pgprintk("instantiating %s PTE (%s) at %ld (%llx) addr %p\n",
1235 (spte
&PT_PAGE_SIZE_MASK
)? "2MB" : "4kB",
1236 (spte
&PT_WRITABLE_MASK
)?"RW":"R", gfn
, spte
, shadow_pte
);
1237 set_shadow_pte(shadow_pte
, spte
);
1238 if (!was_rmapped
&& (spte
& PT_PAGE_SIZE_MASK
)
1239 && (spte
& PT_PRESENT_MASK
))
1240 ++vcpu
->kvm
->stat
.lpages
;
1242 page_header_update_slot(vcpu
->kvm
, shadow_pte
, gfn
);
1244 rmap_add(vcpu
, shadow_pte
, gfn
, largepage
);
1245 if (!is_rmap_pte(*shadow_pte
))
1246 kvm_release_pfn_clean(pfn
);
1249 kvm_release_pfn_dirty(pfn
);
1251 kvm_release_pfn_clean(pfn
);
1254 vcpu
->arch
.last_pte_updated
= shadow_pte
;
1255 vcpu
->arch
.last_pte_gfn
= gfn
;
1259 static void nonpaging_new_cr3(struct kvm_vcpu
*vcpu
)
1263 struct direct_shadow_walk
{
1264 struct kvm_shadow_walk walker
;
1271 static int direct_map_entry(struct kvm_shadow_walk
*_walk
,
1272 struct kvm_vcpu
*vcpu
,
1273 gva_t addr
, u64
*sptep
, int level
)
1275 struct direct_shadow_walk
*walk
=
1276 container_of(_walk
, struct direct_shadow_walk
, walker
);
1277 struct kvm_mmu_page
*sp
;
1279 gfn_t gfn
= addr
>> PAGE_SHIFT
;
1281 if (level
== PT_PAGE_TABLE_LEVEL
1282 || (walk
->largepage
&& level
== PT_DIRECTORY_LEVEL
)) {
1283 mmu_set_spte(vcpu
, sptep
, ACC_ALL
, ACC_ALL
,
1284 0, walk
->write
, 1, &walk
->pt_write
,
1285 walk
->largepage
, gfn
, walk
->pfn
, false);
1286 ++vcpu
->stat
.pf_fixed
;
1290 if (*sptep
== shadow_trap_nonpresent_pte
) {
1291 pseudo_gfn
= (addr
& PT64_DIR_BASE_ADDR_MASK
) >> PAGE_SHIFT
;
1292 sp
= kvm_mmu_get_page(vcpu
, pseudo_gfn
, addr
, level
- 1,
1295 pgprintk("nonpaging_map: ENOMEM\n");
1296 kvm_release_pfn_clean(walk
->pfn
);
1300 set_shadow_pte(sptep
,
1302 | PT_PRESENT_MASK
| PT_WRITABLE_MASK
1303 | shadow_user_mask
| shadow_x_mask
);
1308 static int __direct_map(struct kvm_vcpu
*vcpu
, gpa_t v
, int write
,
1309 int largepage
, gfn_t gfn
, pfn_t pfn
)
1312 struct direct_shadow_walk walker
= {
1313 .walker
= { .entry
= direct_map_entry
, },
1315 .largepage
= largepage
,
1320 r
= walk_shadow(&walker
.walker
, vcpu
, (gva_t
)gfn
<< PAGE_SHIFT
);
1323 return walker
.pt_write
;
1326 static int nonpaging_map(struct kvm_vcpu
*vcpu
, gva_t v
, int write
, gfn_t gfn
)
1331 unsigned long mmu_seq
;
1333 down_read(¤t
->mm
->mmap_sem
);
1334 if (is_largepage_backed(vcpu
, gfn
& ~(KVM_PAGES_PER_HPAGE
-1))) {
1335 gfn
&= ~(KVM_PAGES_PER_HPAGE
-1);
1339 mmu_seq
= vcpu
->kvm
->mmu_notifier_seq
;
1340 /* implicit mb(), we'll read before PT lock is unlocked */
1341 pfn
= gfn_to_pfn(vcpu
->kvm
, gfn
);
1342 up_read(¤t
->mm
->mmap_sem
);
1345 if (is_error_pfn(pfn
)) {
1346 kvm_release_pfn_clean(pfn
);
1350 spin_lock(&vcpu
->kvm
->mmu_lock
);
1351 if (mmu_notifier_retry(vcpu
, mmu_seq
))
1353 kvm_mmu_free_some_pages(vcpu
);
1354 r
= __direct_map(vcpu
, v
, write
, largepage
, gfn
, pfn
);
1355 spin_unlock(&vcpu
->kvm
->mmu_lock
);
1361 spin_unlock(&vcpu
->kvm
->mmu_lock
);
1362 kvm_release_pfn_clean(pfn
);
1367 static void mmu_free_roots(struct kvm_vcpu
*vcpu
)
1370 struct kvm_mmu_page
*sp
;
1372 if (!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
))
1374 spin_lock(&vcpu
->kvm
->mmu_lock
);
1375 if (vcpu
->arch
.mmu
.shadow_root_level
== PT64_ROOT_LEVEL
) {
1376 hpa_t root
= vcpu
->arch
.mmu
.root_hpa
;
1378 sp
= page_header(root
);
1380 if (!sp
->root_count
&& sp
->role
.invalid
)
1381 kvm_mmu_zap_page(vcpu
->kvm
, sp
);
1382 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
1383 spin_unlock(&vcpu
->kvm
->mmu_lock
);
1386 for (i
= 0; i
< 4; ++i
) {
1387 hpa_t root
= vcpu
->arch
.mmu
.pae_root
[i
];
1390 root
&= PT64_BASE_ADDR_MASK
;
1391 sp
= page_header(root
);
1393 if (!sp
->root_count
&& sp
->role
.invalid
)
1394 kvm_mmu_zap_page(vcpu
->kvm
, sp
);
1396 vcpu
->arch
.mmu
.pae_root
[i
] = INVALID_PAGE
;
1398 spin_unlock(&vcpu
->kvm
->mmu_lock
);
1399 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
1402 static void mmu_alloc_roots(struct kvm_vcpu
*vcpu
)
1406 struct kvm_mmu_page
*sp
;
1407 int metaphysical
= 0;
1409 root_gfn
= vcpu
->arch
.cr3
>> PAGE_SHIFT
;
1411 if (vcpu
->arch
.mmu
.shadow_root_level
== PT64_ROOT_LEVEL
) {
1412 hpa_t root
= vcpu
->arch
.mmu
.root_hpa
;
1414 ASSERT(!VALID_PAGE(root
));
1417 sp
= kvm_mmu_get_page(vcpu
, root_gfn
, 0,
1418 PT64_ROOT_LEVEL
, metaphysical
,
1420 root
= __pa(sp
->spt
);
1422 vcpu
->arch
.mmu
.root_hpa
= root
;
1425 metaphysical
= !is_paging(vcpu
);
1428 for (i
= 0; i
< 4; ++i
) {
1429 hpa_t root
= vcpu
->arch
.mmu
.pae_root
[i
];
1431 ASSERT(!VALID_PAGE(root
));
1432 if (vcpu
->arch
.mmu
.root_level
== PT32E_ROOT_LEVEL
) {
1433 if (!is_present_pte(vcpu
->arch
.pdptrs
[i
])) {
1434 vcpu
->arch
.mmu
.pae_root
[i
] = 0;
1437 root_gfn
= vcpu
->arch
.pdptrs
[i
] >> PAGE_SHIFT
;
1438 } else if (vcpu
->arch
.mmu
.root_level
== 0)
1440 sp
= kvm_mmu_get_page(vcpu
, root_gfn
, i
<< 30,
1441 PT32_ROOT_LEVEL
, metaphysical
,
1443 root
= __pa(sp
->spt
);
1445 vcpu
->arch
.mmu
.pae_root
[i
] = root
| PT_PRESENT_MASK
;
1447 vcpu
->arch
.mmu
.root_hpa
= __pa(vcpu
->arch
.mmu
.pae_root
);
1450 static gpa_t
nonpaging_gva_to_gpa(struct kvm_vcpu
*vcpu
, gva_t vaddr
)
1455 static int nonpaging_page_fault(struct kvm_vcpu
*vcpu
, gva_t gva
,
1461 pgprintk("%s: gva %lx error %x\n", __func__
, gva
, error_code
);
1462 r
= mmu_topup_memory_caches(vcpu
);
1467 ASSERT(VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
1469 gfn
= gva
>> PAGE_SHIFT
;
1471 return nonpaging_map(vcpu
, gva
& PAGE_MASK
,
1472 error_code
& PFERR_WRITE_MASK
, gfn
);
1475 static int tdp_page_fault(struct kvm_vcpu
*vcpu
, gva_t gpa
,
1481 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1482 unsigned long mmu_seq
;
1485 ASSERT(VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
1487 r
= mmu_topup_memory_caches(vcpu
);
1491 down_read(¤t
->mm
->mmap_sem
);
1492 if (is_largepage_backed(vcpu
, gfn
& ~(KVM_PAGES_PER_HPAGE
-1))) {
1493 gfn
&= ~(KVM_PAGES_PER_HPAGE
-1);
1496 mmu_seq
= vcpu
->kvm
->mmu_notifier_seq
;
1497 /* implicit mb(), we'll read before PT lock is unlocked */
1498 pfn
= gfn_to_pfn(vcpu
->kvm
, gfn
);
1499 up_read(¤t
->mm
->mmap_sem
);
1500 if (is_error_pfn(pfn
)) {
1501 kvm_release_pfn_clean(pfn
);
1504 spin_lock(&vcpu
->kvm
->mmu_lock
);
1505 if (mmu_notifier_retry(vcpu
, mmu_seq
))
1507 kvm_mmu_free_some_pages(vcpu
);
1508 r
= __direct_map(vcpu
, gpa
, error_code
& PFERR_WRITE_MASK
,
1509 largepage
, gfn
, pfn
);
1510 spin_unlock(&vcpu
->kvm
->mmu_lock
);
1515 spin_unlock(&vcpu
->kvm
->mmu_lock
);
1516 kvm_release_pfn_clean(pfn
);
1520 static void nonpaging_free(struct kvm_vcpu
*vcpu
)
1522 mmu_free_roots(vcpu
);
1525 static int nonpaging_init_context(struct kvm_vcpu
*vcpu
)
1527 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
1529 context
->new_cr3
= nonpaging_new_cr3
;
1530 context
->page_fault
= nonpaging_page_fault
;
1531 context
->gva_to_gpa
= nonpaging_gva_to_gpa
;
1532 context
->free
= nonpaging_free
;
1533 context
->prefetch_page
= nonpaging_prefetch_page
;
1534 context
->root_level
= 0;
1535 context
->shadow_root_level
= PT32E_ROOT_LEVEL
;
1536 context
->root_hpa
= INVALID_PAGE
;
1540 void kvm_mmu_flush_tlb(struct kvm_vcpu
*vcpu
)
1542 ++vcpu
->stat
.tlb_flush
;
1543 kvm_x86_ops
->tlb_flush(vcpu
);
1546 static void paging_new_cr3(struct kvm_vcpu
*vcpu
)
1548 pgprintk("%s: cr3 %lx\n", __func__
, vcpu
->arch
.cr3
);
1549 mmu_free_roots(vcpu
);
1552 static void inject_page_fault(struct kvm_vcpu
*vcpu
,
1556 kvm_inject_page_fault(vcpu
, addr
, err_code
);
1559 static void paging_free(struct kvm_vcpu
*vcpu
)
1561 nonpaging_free(vcpu
);
1565 #include "paging_tmpl.h"
1569 #include "paging_tmpl.h"
1572 static int paging64_init_context_common(struct kvm_vcpu
*vcpu
, int level
)
1574 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
1576 ASSERT(is_pae(vcpu
));
1577 context
->new_cr3
= paging_new_cr3
;
1578 context
->page_fault
= paging64_page_fault
;
1579 context
->gva_to_gpa
= paging64_gva_to_gpa
;
1580 context
->prefetch_page
= paging64_prefetch_page
;
1581 context
->free
= paging_free
;
1582 context
->root_level
= level
;
1583 context
->shadow_root_level
= level
;
1584 context
->root_hpa
= INVALID_PAGE
;
1588 static int paging64_init_context(struct kvm_vcpu
*vcpu
)
1590 return paging64_init_context_common(vcpu
, PT64_ROOT_LEVEL
);
1593 static int paging32_init_context(struct kvm_vcpu
*vcpu
)
1595 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
1597 context
->new_cr3
= paging_new_cr3
;
1598 context
->page_fault
= paging32_page_fault
;
1599 context
->gva_to_gpa
= paging32_gva_to_gpa
;
1600 context
->free
= paging_free
;
1601 context
->prefetch_page
= paging32_prefetch_page
;
1602 context
->root_level
= PT32_ROOT_LEVEL
;
1603 context
->shadow_root_level
= PT32E_ROOT_LEVEL
;
1604 context
->root_hpa
= INVALID_PAGE
;
1608 static int paging32E_init_context(struct kvm_vcpu
*vcpu
)
1610 return paging64_init_context_common(vcpu
, PT32E_ROOT_LEVEL
);
1613 static int init_kvm_tdp_mmu(struct kvm_vcpu
*vcpu
)
1615 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
1617 context
->new_cr3
= nonpaging_new_cr3
;
1618 context
->page_fault
= tdp_page_fault
;
1619 context
->free
= nonpaging_free
;
1620 context
->prefetch_page
= nonpaging_prefetch_page
;
1621 context
->shadow_root_level
= kvm_x86_ops
->get_tdp_level();
1622 context
->root_hpa
= INVALID_PAGE
;
1624 if (!is_paging(vcpu
)) {
1625 context
->gva_to_gpa
= nonpaging_gva_to_gpa
;
1626 context
->root_level
= 0;
1627 } else if (is_long_mode(vcpu
)) {
1628 context
->gva_to_gpa
= paging64_gva_to_gpa
;
1629 context
->root_level
= PT64_ROOT_LEVEL
;
1630 } else if (is_pae(vcpu
)) {
1631 context
->gva_to_gpa
= paging64_gva_to_gpa
;
1632 context
->root_level
= PT32E_ROOT_LEVEL
;
1634 context
->gva_to_gpa
= paging32_gva_to_gpa
;
1635 context
->root_level
= PT32_ROOT_LEVEL
;
1641 static int init_kvm_softmmu(struct kvm_vcpu
*vcpu
)
1644 ASSERT(!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
1646 if (!is_paging(vcpu
))
1647 return nonpaging_init_context(vcpu
);
1648 else if (is_long_mode(vcpu
))
1649 return paging64_init_context(vcpu
);
1650 else if (is_pae(vcpu
))
1651 return paging32E_init_context(vcpu
);
1653 return paging32_init_context(vcpu
);
1656 static int init_kvm_mmu(struct kvm_vcpu
*vcpu
)
1658 vcpu
->arch
.update_pte
.pfn
= bad_pfn
;
1661 return init_kvm_tdp_mmu(vcpu
);
1663 return init_kvm_softmmu(vcpu
);
1666 static void destroy_kvm_mmu(struct kvm_vcpu
*vcpu
)
1669 if (VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
)) {
1670 vcpu
->arch
.mmu
.free(vcpu
);
1671 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
1675 int kvm_mmu_reset_context(struct kvm_vcpu
*vcpu
)
1677 destroy_kvm_mmu(vcpu
);
1678 return init_kvm_mmu(vcpu
);
1680 EXPORT_SYMBOL_GPL(kvm_mmu_reset_context
);
1682 int kvm_mmu_load(struct kvm_vcpu
*vcpu
)
1686 r
= mmu_topup_memory_caches(vcpu
);
1689 spin_lock(&vcpu
->kvm
->mmu_lock
);
1690 kvm_mmu_free_some_pages(vcpu
);
1691 mmu_alloc_roots(vcpu
);
1692 spin_unlock(&vcpu
->kvm
->mmu_lock
);
1693 kvm_x86_ops
->set_cr3(vcpu
, vcpu
->arch
.mmu
.root_hpa
);
1694 kvm_mmu_flush_tlb(vcpu
);
1698 EXPORT_SYMBOL_GPL(kvm_mmu_load
);
1700 void kvm_mmu_unload(struct kvm_vcpu
*vcpu
)
1702 mmu_free_roots(vcpu
);
1705 static void mmu_pte_write_zap_pte(struct kvm_vcpu
*vcpu
,
1706 struct kvm_mmu_page
*sp
,
1710 struct kvm_mmu_page
*child
;
1713 if (is_shadow_present_pte(pte
)) {
1714 if (sp
->role
.level
== PT_PAGE_TABLE_LEVEL
||
1716 rmap_remove(vcpu
->kvm
, spte
);
1718 child
= page_header(pte
& PT64_BASE_ADDR_MASK
);
1719 mmu_page_remove_parent_pte(child
, spte
);
1722 set_shadow_pte(spte
, shadow_trap_nonpresent_pte
);
1723 if (is_large_pte(pte
))
1724 --vcpu
->kvm
->stat
.lpages
;
1727 static void mmu_pte_write_new_pte(struct kvm_vcpu
*vcpu
,
1728 struct kvm_mmu_page
*sp
,
1732 if (sp
->role
.level
!= PT_PAGE_TABLE_LEVEL
) {
1733 if (!vcpu
->arch
.update_pte
.largepage
||
1734 sp
->role
.glevels
== PT32_ROOT_LEVEL
) {
1735 ++vcpu
->kvm
->stat
.mmu_pde_zapped
;
1740 ++vcpu
->kvm
->stat
.mmu_pte_updated
;
1741 if (sp
->role
.glevels
== PT32_ROOT_LEVEL
)
1742 paging32_update_pte(vcpu
, sp
, spte
, new);
1744 paging64_update_pte(vcpu
, sp
, spte
, new);
1747 static bool need_remote_flush(u64 old
, u64
new)
1749 if (!is_shadow_present_pte(old
))
1751 if (!is_shadow_present_pte(new))
1753 if ((old
^ new) & PT64_BASE_ADDR_MASK
)
1755 old
^= PT64_NX_MASK
;
1756 new ^= PT64_NX_MASK
;
1757 return (old
& ~new & PT64_PERM_MASK
) != 0;
1760 static void mmu_pte_write_flush_tlb(struct kvm_vcpu
*vcpu
, u64 old
, u64
new)
1762 if (need_remote_flush(old
, new))
1763 kvm_flush_remote_tlbs(vcpu
->kvm
);
1765 kvm_mmu_flush_tlb(vcpu
);
1768 static bool last_updated_pte_accessed(struct kvm_vcpu
*vcpu
)
1770 u64
*spte
= vcpu
->arch
.last_pte_updated
;
1772 return !!(spte
&& (*spte
& shadow_accessed_mask
));
1775 static void mmu_guess_page_from_pte_write(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1776 const u8
*new, int bytes
)
1783 vcpu
->arch
.update_pte
.largepage
= 0;
1785 if (bytes
!= 4 && bytes
!= 8)
1789 * Assume that the pte write on a page table of the same type
1790 * as the current vcpu paging mode. This is nearly always true
1791 * (might be false while changing modes). Note it is verified later
1795 /* Handle a 32-bit guest writing two halves of a 64-bit gpte */
1796 if ((bytes
== 4) && (gpa
% 4 == 0)) {
1797 r
= kvm_read_guest(vcpu
->kvm
, gpa
& ~(u64
)7, &gpte
, 8);
1800 memcpy((void *)&gpte
+ (gpa
% 8), new, 4);
1801 } else if ((bytes
== 8) && (gpa
% 8 == 0)) {
1802 memcpy((void *)&gpte
, new, 8);
1805 if ((bytes
== 4) && (gpa
% 4 == 0))
1806 memcpy((void *)&gpte
, new, 4);
1808 if (!is_present_pte(gpte
))
1810 gfn
= (gpte
& PT64_BASE_ADDR_MASK
) >> PAGE_SHIFT
;
1812 down_read(¤t
->mm
->mmap_sem
);
1813 if (is_large_pte(gpte
) && is_largepage_backed(vcpu
, gfn
)) {
1814 gfn
&= ~(KVM_PAGES_PER_HPAGE
-1);
1815 vcpu
->arch
.update_pte
.largepage
= 1;
1817 vcpu
->arch
.update_pte
.mmu_seq
= vcpu
->kvm
->mmu_notifier_seq
;
1818 /* implicit mb(), we'll read before PT lock is unlocked */
1819 pfn
= gfn_to_pfn(vcpu
->kvm
, gfn
);
1820 up_read(¤t
->mm
->mmap_sem
);
1822 if (is_error_pfn(pfn
)) {
1823 kvm_release_pfn_clean(pfn
);
1826 vcpu
->arch
.update_pte
.gfn
= gfn
;
1827 vcpu
->arch
.update_pte
.pfn
= pfn
;
1830 static void kvm_mmu_access_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1832 u64
*spte
= vcpu
->arch
.last_pte_updated
;
1835 && vcpu
->arch
.last_pte_gfn
== gfn
1836 && shadow_accessed_mask
1837 && !(*spte
& shadow_accessed_mask
)
1838 && is_shadow_present_pte(*spte
))
1839 set_bit(PT_ACCESSED_SHIFT
, (unsigned long *)spte
);
1842 void kvm_mmu_pte_write(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1843 const u8
*new, int bytes
)
1845 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1846 struct kvm_mmu_page
*sp
;
1847 struct hlist_node
*node
, *n
;
1848 struct hlist_head
*bucket
;
1852 unsigned offset
= offset_in_page(gpa
);
1854 unsigned page_offset
;
1855 unsigned misaligned
;
1862 pgprintk("%s: gpa %llx bytes %d\n", __func__
, gpa
, bytes
);
1863 mmu_guess_page_from_pte_write(vcpu
, gpa
, new, bytes
);
1864 spin_lock(&vcpu
->kvm
->mmu_lock
);
1865 kvm_mmu_access_page(vcpu
, gfn
);
1866 kvm_mmu_free_some_pages(vcpu
);
1867 ++vcpu
->kvm
->stat
.mmu_pte_write
;
1868 kvm_mmu_audit(vcpu
, "pre pte write");
1869 if (gfn
== vcpu
->arch
.last_pt_write_gfn
1870 && !last_updated_pte_accessed(vcpu
)) {
1871 ++vcpu
->arch
.last_pt_write_count
;
1872 if (vcpu
->arch
.last_pt_write_count
>= 3)
1875 vcpu
->arch
.last_pt_write_gfn
= gfn
;
1876 vcpu
->arch
.last_pt_write_count
= 1;
1877 vcpu
->arch
.last_pte_updated
= NULL
;
1879 index
= kvm_page_table_hashfn(gfn
);
1880 bucket
= &vcpu
->kvm
->arch
.mmu_page_hash
[index
];
1881 hlist_for_each_entry_safe(sp
, node
, n
, bucket
, hash_link
) {
1882 if (sp
->gfn
!= gfn
|| sp
->role
.metaphysical
|| sp
->role
.invalid
)
1884 pte_size
= sp
->role
.glevels
== PT32_ROOT_LEVEL
? 4 : 8;
1885 misaligned
= (offset
^ (offset
+ bytes
- 1)) & ~(pte_size
- 1);
1886 misaligned
|= bytes
< 4;
1887 if (misaligned
|| flooded
) {
1889 * Misaligned accesses are too much trouble to fix
1890 * up; also, they usually indicate a page is not used
1893 * If we're seeing too many writes to a page,
1894 * it may no longer be a page table, or we may be
1895 * forking, in which case it is better to unmap the
1898 pgprintk("misaligned: gpa %llx bytes %d role %x\n",
1899 gpa
, bytes
, sp
->role
.word
);
1900 kvm_mmu_zap_page(vcpu
->kvm
, sp
);
1901 ++vcpu
->kvm
->stat
.mmu_flooded
;
1904 page_offset
= offset
;
1905 level
= sp
->role
.level
;
1907 if (sp
->role
.glevels
== PT32_ROOT_LEVEL
) {
1908 page_offset
<<= 1; /* 32->64 */
1910 * A 32-bit pde maps 4MB while the shadow pdes map
1911 * only 2MB. So we need to double the offset again
1912 * and zap two pdes instead of one.
1914 if (level
== PT32_ROOT_LEVEL
) {
1915 page_offset
&= ~7; /* kill rounding error */
1919 quadrant
= page_offset
>> PAGE_SHIFT
;
1920 page_offset
&= ~PAGE_MASK
;
1921 if (quadrant
!= sp
->role
.quadrant
)
1924 spte
= &sp
->spt
[page_offset
/ sizeof(*spte
)];
1925 if ((gpa
& (pte_size
- 1)) || (bytes
< pte_size
)) {
1927 r
= kvm_read_guest_atomic(vcpu
->kvm
,
1928 gpa
& ~(u64
)(pte_size
- 1),
1930 new = (const void *)&gentry
;
1936 mmu_pte_write_zap_pte(vcpu
, sp
, spte
);
1938 mmu_pte_write_new_pte(vcpu
, sp
, spte
, new);
1939 mmu_pte_write_flush_tlb(vcpu
, entry
, *spte
);
1943 kvm_mmu_audit(vcpu
, "post pte write");
1944 spin_unlock(&vcpu
->kvm
->mmu_lock
);
1945 if (!is_error_pfn(vcpu
->arch
.update_pte
.pfn
)) {
1946 kvm_release_pfn_clean(vcpu
->arch
.update_pte
.pfn
);
1947 vcpu
->arch
.update_pte
.pfn
= bad_pfn
;
1951 int kvm_mmu_unprotect_page_virt(struct kvm_vcpu
*vcpu
, gva_t gva
)
1956 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, gva
);
1958 spin_lock(&vcpu
->kvm
->mmu_lock
);
1959 r
= kvm_mmu_unprotect_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
1960 spin_unlock(&vcpu
->kvm
->mmu_lock
);
1963 EXPORT_SYMBOL_GPL(kvm_mmu_unprotect_page_virt
);
1965 void __kvm_mmu_free_some_pages(struct kvm_vcpu
*vcpu
)
1967 while (vcpu
->kvm
->arch
.n_free_mmu_pages
< KVM_REFILL_PAGES
) {
1968 struct kvm_mmu_page
*sp
;
1970 sp
= container_of(vcpu
->kvm
->arch
.active_mmu_pages
.prev
,
1971 struct kvm_mmu_page
, link
);
1972 kvm_mmu_zap_page(vcpu
->kvm
, sp
);
1973 ++vcpu
->kvm
->stat
.mmu_recycled
;
1977 int kvm_mmu_page_fault(struct kvm_vcpu
*vcpu
, gva_t cr2
, u32 error_code
)
1980 enum emulation_result er
;
1982 r
= vcpu
->arch
.mmu
.page_fault(vcpu
, cr2
, error_code
);
1991 r
= mmu_topup_memory_caches(vcpu
);
1995 er
= emulate_instruction(vcpu
, vcpu
->run
, cr2
, error_code
, 0);
2000 case EMULATE_DO_MMIO
:
2001 ++vcpu
->stat
.mmio_exits
;
2004 kvm_report_emulation_failure(vcpu
, "pagetable");
2012 EXPORT_SYMBOL_GPL(kvm_mmu_page_fault
);
2014 void kvm_enable_tdp(void)
2018 EXPORT_SYMBOL_GPL(kvm_enable_tdp
);
2020 void kvm_disable_tdp(void)
2022 tdp_enabled
= false;
2024 EXPORT_SYMBOL_GPL(kvm_disable_tdp
);
2026 static void free_mmu_pages(struct kvm_vcpu
*vcpu
)
2028 struct kvm_mmu_page
*sp
;
2030 while (!list_empty(&vcpu
->kvm
->arch
.active_mmu_pages
)) {
2031 sp
= container_of(vcpu
->kvm
->arch
.active_mmu_pages
.next
,
2032 struct kvm_mmu_page
, link
);
2033 kvm_mmu_zap_page(vcpu
->kvm
, sp
);
2036 free_page((unsigned long)vcpu
->arch
.mmu
.pae_root
);
2039 static int alloc_mmu_pages(struct kvm_vcpu
*vcpu
)
2046 if (vcpu
->kvm
->arch
.n_requested_mmu_pages
)
2047 vcpu
->kvm
->arch
.n_free_mmu_pages
=
2048 vcpu
->kvm
->arch
.n_requested_mmu_pages
;
2050 vcpu
->kvm
->arch
.n_free_mmu_pages
=
2051 vcpu
->kvm
->arch
.n_alloc_mmu_pages
;
2053 * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64.
2054 * Therefore we need to allocate shadow page tables in the first
2055 * 4GB of memory, which happens to fit the DMA32 zone.
2057 page
= alloc_page(GFP_KERNEL
| __GFP_DMA32
);
2060 vcpu
->arch
.mmu
.pae_root
= page_address(page
);
2061 for (i
= 0; i
< 4; ++i
)
2062 vcpu
->arch
.mmu
.pae_root
[i
] = INVALID_PAGE
;
2067 free_mmu_pages(vcpu
);
2071 int kvm_mmu_create(struct kvm_vcpu
*vcpu
)
2074 ASSERT(!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
2076 return alloc_mmu_pages(vcpu
);
2079 int kvm_mmu_setup(struct kvm_vcpu
*vcpu
)
2082 ASSERT(!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
2084 return init_kvm_mmu(vcpu
);
2087 void kvm_mmu_destroy(struct kvm_vcpu
*vcpu
)
2091 destroy_kvm_mmu(vcpu
);
2092 free_mmu_pages(vcpu
);
2093 mmu_free_memory_caches(vcpu
);
2096 void kvm_mmu_slot_remove_write_access(struct kvm
*kvm
, int slot
)
2098 struct kvm_mmu_page
*sp
;
2100 list_for_each_entry(sp
, &kvm
->arch
.active_mmu_pages
, link
) {
2104 if (!test_bit(slot
, &sp
->slot_bitmap
))
2108 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
)
2110 if (pt
[i
] & PT_WRITABLE_MASK
)
2111 pt
[i
] &= ~PT_WRITABLE_MASK
;
2115 void kvm_mmu_zap_all(struct kvm
*kvm
)
2117 struct kvm_mmu_page
*sp
, *node
;
2119 spin_lock(&kvm
->mmu_lock
);
2120 list_for_each_entry_safe(sp
, node
, &kvm
->arch
.active_mmu_pages
, link
)
2121 kvm_mmu_zap_page(kvm
, sp
);
2122 spin_unlock(&kvm
->mmu_lock
);
2124 kvm_flush_remote_tlbs(kvm
);
2127 static void kvm_mmu_remove_one_alloc_mmu_page(struct kvm
*kvm
)
2129 struct kvm_mmu_page
*page
;
2131 page
= container_of(kvm
->arch
.active_mmu_pages
.prev
,
2132 struct kvm_mmu_page
, link
);
2133 kvm_mmu_zap_page(kvm
, page
);
2136 static int mmu_shrink(int nr_to_scan
, gfp_t gfp_mask
)
2139 struct kvm
*kvm_freed
= NULL
;
2140 int cache_count
= 0;
2142 spin_lock(&kvm_lock
);
2144 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
2147 if (!down_read_trylock(&kvm
->slots_lock
))
2149 spin_lock(&kvm
->mmu_lock
);
2150 npages
= kvm
->arch
.n_alloc_mmu_pages
-
2151 kvm
->arch
.n_free_mmu_pages
;
2152 cache_count
+= npages
;
2153 if (!kvm_freed
&& nr_to_scan
> 0 && npages
> 0) {
2154 kvm_mmu_remove_one_alloc_mmu_page(kvm
);
2160 spin_unlock(&kvm
->mmu_lock
);
2161 up_read(&kvm
->slots_lock
);
2164 list_move_tail(&kvm_freed
->vm_list
, &vm_list
);
2166 spin_unlock(&kvm_lock
);
2171 static struct shrinker mmu_shrinker
= {
2172 .shrink
= mmu_shrink
,
2173 .seeks
= DEFAULT_SEEKS
* 10,
2176 static void mmu_destroy_caches(void)
2178 if (pte_chain_cache
)
2179 kmem_cache_destroy(pte_chain_cache
);
2180 if (rmap_desc_cache
)
2181 kmem_cache_destroy(rmap_desc_cache
);
2182 if (mmu_page_header_cache
)
2183 kmem_cache_destroy(mmu_page_header_cache
);
2186 void kvm_mmu_module_exit(void)
2188 mmu_destroy_caches();
2189 unregister_shrinker(&mmu_shrinker
);
2192 int kvm_mmu_module_init(void)
2194 pte_chain_cache
= kmem_cache_create("kvm_pte_chain",
2195 sizeof(struct kvm_pte_chain
),
2197 if (!pte_chain_cache
)
2199 rmap_desc_cache
= kmem_cache_create("kvm_rmap_desc",
2200 sizeof(struct kvm_rmap_desc
),
2202 if (!rmap_desc_cache
)
2205 mmu_page_header_cache
= kmem_cache_create("kvm_mmu_page_header",
2206 sizeof(struct kvm_mmu_page
),
2208 if (!mmu_page_header_cache
)
2211 register_shrinker(&mmu_shrinker
);
2216 mmu_destroy_caches();
2221 * Caculate mmu pages needed for kvm.
2223 unsigned int kvm_mmu_calculate_mmu_pages(struct kvm
*kvm
)
2226 unsigned int nr_mmu_pages
;
2227 unsigned int nr_pages
= 0;
2229 for (i
= 0; i
< kvm
->nmemslots
; i
++)
2230 nr_pages
+= kvm
->memslots
[i
].npages
;
2232 nr_mmu_pages
= nr_pages
* KVM_PERMILLE_MMU_PAGES
/ 1000;
2233 nr_mmu_pages
= max(nr_mmu_pages
,
2234 (unsigned int) KVM_MIN_ALLOC_MMU_PAGES
);
2236 return nr_mmu_pages
;
2239 static void *pv_mmu_peek_buffer(struct kvm_pv_mmu_op_buffer
*buffer
,
2242 if (len
> buffer
->len
)
2247 static void *pv_mmu_read_buffer(struct kvm_pv_mmu_op_buffer
*buffer
,
2252 ret
= pv_mmu_peek_buffer(buffer
, len
);
2257 buffer
->processed
+= len
;
2261 static int kvm_pv_mmu_write(struct kvm_vcpu
*vcpu
,
2262 gpa_t addr
, gpa_t value
)
2267 if (!is_long_mode(vcpu
) && !is_pae(vcpu
))
2270 r
= mmu_topup_memory_caches(vcpu
);
2274 if (!emulator_write_phys(vcpu
, addr
, &value
, bytes
))
2280 static int kvm_pv_mmu_flush_tlb(struct kvm_vcpu
*vcpu
)
2282 kvm_x86_ops
->tlb_flush(vcpu
);
2286 static int kvm_pv_mmu_release_pt(struct kvm_vcpu
*vcpu
, gpa_t addr
)
2288 spin_lock(&vcpu
->kvm
->mmu_lock
);
2289 mmu_unshadow(vcpu
->kvm
, addr
>> PAGE_SHIFT
);
2290 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2294 static int kvm_pv_mmu_op_one(struct kvm_vcpu
*vcpu
,
2295 struct kvm_pv_mmu_op_buffer
*buffer
)
2297 struct kvm_mmu_op_header
*header
;
2299 header
= pv_mmu_peek_buffer(buffer
, sizeof *header
);
2302 switch (header
->op
) {
2303 case KVM_MMU_OP_WRITE_PTE
: {
2304 struct kvm_mmu_op_write_pte
*wpte
;
2306 wpte
= pv_mmu_read_buffer(buffer
, sizeof *wpte
);
2309 return kvm_pv_mmu_write(vcpu
, wpte
->pte_phys
,
2312 case KVM_MMU_OP_FLUSH_TLB
: {
2313 struct kvm_mmu_op_flush_tlb
*ftlb
;
2315 ftlb
= pv_mmu_read_buffer(buffer
, sizeof *ftlb
);
2318 return kvm_pv_mmu_flush_tlb(vcpu
);
2320 case KVM_MMU_OP_RELEASE_PT
: {
2321 struct kvm_mmu_op_release_pt
*rpt
;
2323 rpt
= pv_mmu_read_buffer(buffer
, sizeof *rpt
);
2326 return kvm_pv_mmu_release_pt(vcpu
, rpt
->pt_phys
);
2332 int kvm_pv_mmu_op(struct kvm_vcpu
*vcpu
, unsigned long bytes
,
2333 gpa_t addr
, unsigned long *ret
)
2336 struct kvm_pv_mmu_op_buffer
*buffer
= &vcpu
->arch
.mmu_op_buffer
;
2338 buffer
->ptr
= buffer
->buf
;
2339 buffer
->len
= min_t(unsigned long, bytes
, sizeof buffer
->buf
);
2340 buffer
->processed
= 0;
2342 r
= kvm_read_guest(vcpu
->kvm
, addr
, buffer
->buf
, buffer
->len
);
2346 while (buffer
->len
) {
2347 r
= kvm_pv_mmu_op_one(vcpu
, buffer
);
2356 *ret
= buffer
->processed
;
2362 static const char *audit_msg
;
2364 static gva_t
canonicalize(gva_t gva
)
2366 #ifdef CONFIG_X86_64
2367 gva
= (long long)(gva
<< 16) >> 16;
2372 static void audit_mappings_page(struct kvm_vcpu
*vcpu
, u64 page_pte
,
2373 gva_t va
, int level
)
2375 u64
*pt
= __va(page_pte
& PT64_BASE_ADDR_MASK
);
2377 gva_t va_delta
= 1ul << (PAGE_SHIFT
+ 9 * (level
- 1));
2379 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
, va
+= va_delta
) {
2382 if (ent
== shadow_trap_nonpresent_pte
)
2385 va
= canonicalize(va
);
2387 if (ent
== shadow_notrap_nonpresent_pte
)
2388 printk(KERN_ERR
"audit: (%s) nontrapping pte"
2389 " in nonleaf level: levels %d gva %lx"
2390 " level %d pte %llx\n", audit_msg
,
2391 vcpu
->arch
.mmu
.root_level
, va
, level
, ent
);
2393 audit_mappings_page(vcpu
, ent
, va
, level
- 1);
2395 gpa_t gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, va
);
2396 hpa_t hpa
= (hpa_t
)gpa_to_pfn(vcpu
, gpa
) << PAGE_SHIFT
;
2398 if (is_shadow_present_pte(ent
)
2399 && (ent
& PT64_BASE_ADDR_MASK
) != hpa
)
2400 printk(KERN_ERR
"xx audit error: (%s) levels %d"
2401 " gva %lx gpa %llx hpa %llx ent %llx %d\n",
2402 audit_msg
, vcpu
->arch
.mmu
.root_level
,
2404 is_shadow_present_pte(ent
));
2405 else if (ent
== shadow_notrap_nonpresent_pte
2406 && !is_error_hpa(hpa
))
2407 printk(KERN_ERR
"audit: (%s) notrap shadow,"
2408 " valid guest gva %lx\n", audit_msg
, va
);
2409 kvm_release_pfn_clean(pfn
);
2415 static void audit_mappings(struct kvm_vcpu
*vcpu
)
2419 if (vcpu
->arch
.mmu
.root_level
== 4)
2420 audit_mappings_page(vcpu
, vcpu
->arch
.mmu
.root_hpa
, 0, 4);
2422 for (i
= 0; i
< 4; ++i
)
2423 if (vcpu
->arch
.mmu
.pae_root
[i
] & PT_PRESENT_MASK
)
2424 audit_mappings_page(vcpu
,
2425 vcpu
->arch
.mmu
.pae_root
[i
],
2430 static int count_rmaps(struct kvm_vcpu
*vcpu
)
2435 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
2436 struct kvm_memory_slot
*m
= &vcpu
->kvm
->memslots
[i
];
2437 struct kvm_rmap_desc
*d
;
2439 for (j
= 0; j
< m
->npages
; ++j
) {
2440 unsigned long *rmapp
= &m
->rmap
[j
];
2444 if (!(*rmapp
& 1)) {
2448 d
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
2450 for (k
= 0; k
< RMAP_EXT
; ++k
)
2451 if (d
->shadow_ptes
[k
])
2462 static int count_writable_mappings(struct kvm_vcpu
*vcpu
)
2465 struct kvm_mmu_page
*sp
;
2468 list_for_each_entry(sp
, &vcpu
->kvm
->arch
.active_mmu_pages
, link
) {
2471 if (sp
->role
.level
!= PT_PAGE_TABLE_LEVEL
)
2474 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
2477 if (!(ent
& PT_PRESENT_MASK
))
2479 if (!(ent
& PT_WRITABLE_MASK
))
2487 static void audit_rmap(struct kvm_vcpu
*vcpu
)
2489 int n_rmap
= count_rmaps(vcpu
);
2490 int n_actual
= count_writable_mappings(vcpu
);
2492 if (n_rmap
!= n_actual
)
2493 printk(KERN_ERR
"%s: (%s) rmap %d actual %d\n",
2494 __func__
, audit_msg
, n_rmap
, n_actual
);
2497 static void audit_write_protection(struct kvm_vcpu
*vcpu
)
2499 struct kvm_mmu_page
*sp
;
2500 struct kvm_memory_slot
*slot
;
2501 unsigned long *rmapp
;
2504 list_for_each_entry(sp
, &vcpu
->kvm
->arch
.active_mmu_pages
, link
) {
2505 if (sp
->role
.metaphysical
)
2508 slot
= gfn_to_memslot(vcpu
->kvm
, sp
->gfn
);
2509 gfn
= unalias_gfn(vcpu
->kvm
, sp
->gfn
);
2510 rmapp
= &slot
->rmap
[gfn
- slot
->base_gfn
];
2512 printk(KERN_ERR
"%s: (%s) shadow page has writable"
2513 " mappings: gfn %lx role %x\n",
2514 __func__
, audit_msg
, sp
->gfn
,
2519 static void kvm_mmu_audit(struct kvm_vcpu
*vcpu
, const char *msg
)
2526 audit_write_protection(vcpu
);
2527 audit_mappings(vcpu
);