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.
21 #include "kvm_cache_regs.h"
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>
39 * When setting this variable to true it enables Two-Dimensional-Paging
40 * where the hardware walks 2 page tables:
41 * 1. the guest-virtual to guest-physical
42 * 2. while doing 1. it walks guest-physical to host-physical
43 * If the hardware supports that we don't need to do shadow paging.
45 bool tdp_enabled
= false;
52 static void kvm_mmu_audit(struct kvm_vcpu
*vcpu
, const char *msg
);
54 static void kvm_mmu_audit(struct kvm_vcpu
*vcpu
, const char *msg
) {}
59 #define pgprintk(x...) do { if (dbg) printk(x); } while (0)
60 #define rmap_printk(x...) do { if (dbg) printk(x); } while (0)
64 #define pgprintk(x...) do { } while (0)
65 #define rmap_printk(x...) do { } while (0)
69 #if defined(MMU_DEBUG) || defined(AUDIT)
71 module_param(dbg
, bool, 0644);
74 static int oos_shadow
= 1;
75 module_param(oos_shadow
, bool, 0644);
78 #define ASSERT(x) do { } while (0)
82 printk(KERN_WARNING "assertion failed %s:%d: %s\n", \
83 __FILE__, __LINE__, #x); \
87 #define PT_FIRST_AVAIL_BITS_SHIFT 9
88 #define PT64_SECOND_AVAIL_BITS_SHIFT 52
90 #define VALID_PAGE(x) ((x) != INVALID_PAGE)
92 #define PT64_LEVEL_BITS 9
94 #define PT64_LEVEL_SHIFT(level) \
95 (PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS)
97 #define PT64_LEVEL_MASK(level) \
98 (((1ULL << PT64_LEVEL_BITS) - 1) << PT64_LEVEL_SHIFT(level))
100 #define PT64_INDEX(address, level)\
101 (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1))
104 #define PT32_LEVEL_BITS 10
106 #define PT32_LEVEL_SHIFT(level) \
107 (PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS)
109 #define PT32_LEVEL_MASK(level) \
110 (((1ULL << PT32_LEVEL_BITS) - 1) << PT32_LEVEL_SHIFT(level))
112 #define PT32_INDEX(address, level)\
113 (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1))
116 #define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1))
117 #define PT64_DIR_BASE_ADDR_MASK \
118 (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1))
120 #define PT32_BASE_ADDR_MASK PAGE_MASK
121 #define PT32_DIR_BASE_ADDR_MASK \
122 (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1))
124 #define PT64_PERM_MASK (PT_PRESENT_MASK | PT_WRITABLE_MASK | PT_USER_MASK \
127 #define PFERR_PRESENT_MASK (1U << 0)
128 #define PFERR_WRITE_MASK (1U << 1)
129 #define PFERR_USER_MASK (1U << 2)
130 #define PFERR_RSVD_MASK (1U << 3)
131 #define PFERR_FETCH_MASK (1U << 4)
133 #define PT_DIRECTORY_LEVEL 2
134 #define PT_PAGE_TABLE_LEVEL 1
138 #define ACC_EXEC_MASK 1
139 #define ACC_WRITE_MASK PT_WRITABLE_MASK
140 #define ACC_USER_MASK PT_USER_MASK
141 #define ACC_ALL (ACC_EXEC_MASK | ACC_WRITE_MASK | ACC_USER_MASK)
143 #define SHADOW_PT_INDEX(addr, level) PT64_INDEX(addr, level)
145 struct kvm_rmap_desc
{
146 u64
*shadow_ptes
[RMAP_EXT
];
147 struct kvm_rmap_desc
*more
;
150 struct kvm_shadow_walk_iterator
{
158 #define for_each_shadow_entry(_vcpu, _addr, _walker) \
159 for (shadow_walk_init(&(_walker), _vcpu, _addr); \
160 shadow_walk_okay(&(_walker)); \
161 shadow_walk_next(&(_walker)))
164 struct kvm_unsync_walk
{
165 int (*entry
) (struct kvm_mmu_page
*sp
, struct kvm_unsync_walk
*walk
);
168 typedef int (*mmu_parent_walk_fn
) (struct kvm_vcpu
*vcpu
, struct kvm_mmu_page
*sp
);
170 static struct kmem_cache
*pte_chain_cache
;
171 static struct kmem_cache
*rmap_desc_cache
;
172 static struct kmem_cache
*mmu_page_header_cache
;
174 static u64 __read_mostly shadow_trap_nonpresent_pte
;
175 static u64 __read_mostly shadow_notrap_nonpresent_pte
;
176 static u64 __read_mostly shadow_base_present_pte
;
177 static u64 __read_mostly shadow_nx_mask
;
178 static u64 __read_mostly shadow_x_mask
; /* mutual exclusive with nx_mask */
179 static u64 __read_mostly shadow_user_mask
;
180 static u64 __read_mostly shadow_accessed_mask
;
181 static u64 __read_mostly shadow_dirty_mask
;
183 static inline u64
rsvd_bits(int s
, int e
)
185 return ((1ULL << (e
- s
+ 1)) - 1) << s
;
188 void kvm_mmu_set_nonpresent_ptes(u64 trap_pte
, u64 notrap_pte
)
190 shadow_trap_nonpresent_pte
= trap_pte
;
191 shadow_notrap_nonpresent_pte
= notrap_pte
;
193 EXPORT_SYMBOL_GPL(kvm_mmu_set_nonpresent_ptes
);
195 void kvm_mmu_set_base_ptes(u64 base_pte
)
197 shadow_base_present_pte
= base_pte
;
199 EXPORT_SYMBOL_GPL(kvm_mmu_set_base_ptes
);
201 void kvm_mmu_set_mask_ptes(u64 user_mask
, u64 accessed_mask
,
202 u64 dirty_mask
, u64 nx_mask
, u64 x_mask
)
204 shadow_user_mask
= user_mask
;
205 shadow_accessed_mask
= accessed_mask
;
206 shadow_dirty_mask
= dirty_mask
;
207 shadow_nx_mask
= nx_mask
;
208 shadow_x_mask
= x_mask
;
210 EXPORT_SYMBOL_GPL(kvm_mmu_set_mask_ptes
);
212 static int is_write_protection(struct kvm_vcpu
*vcpu
)
214 return vcpu
->arch
.cr0
& X86_CR0_WP
;
217 static int is_cpuid_PSE36(void)
222 static int is_nx(struct kvm_vcpu
*vcpu
)
224 return vcpu
->arch
.shadow_efer
& EFER_NX
;
227 static int is_shadow_present_pte(u64 pte
)
229 return pte
!= shadow_trap_nonpresent_pte
230 && pte
!= shadow_notrap_nonpresent_pte
;
233 static int is_large_pte(u64 pte
)
235 return pte
& PT_PAGE_SIZE_MASK
;
238 static int is_writeble_pte(unsigned long pte
)
240 return pte
& PT_WRITABLE_MASK
;
243 static int is_dirty_pte(unsigned long pte
)
245 return pte
& PT_DIRTY_MASK
;
248 static int is_rmap_pte(u64 pte
)
250 return is_shadow_present_pte(pte
);
253 static pfn_t
spte_to_pfn(u64 pte
)
255 return (pte
& PT64_BASE_ADDR_MASK
) >> PAGE_SHIFT
;
258 static gfn_t
pse36_gfn_delta(u32 gpte
)
260 int shift
= 32 - PT32_DIR_PSE36_SHIFT
- PAGE_SHIFT
;
262 return (gpte
& PT32_DIR_PSE36_MASK
) << shift
;
265 static void set_shadow_pte(u64
*sptep
, u64 spte
)
268 set_64bit((unsigned long *)sptep
, spte
);
270 set_64bit((unsigned long long *)sptep
, spte
);
274 static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache
*cache
,
275 struct kmem_cache
*base_cache
, int min
)
279 if (cache
->nobjs
>= min
)
281 while (cache
->nobjs
< ARRAY_SIZE(cache
->objects
)) {
282 obj
= kmem_cache_zalloc(base_cache
, GFP_KERNEL
);
285 cache
->objects
[cache
->nobjs
++] = obj
;
290 static void mmu_free_memory_cache(struct kvm_mmu_memory_cache
*mc
)
293 kfree(mc
->objects
[--mc
->nobjs
]);
296 static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache
*cache
,
301 if (cache
->nobjs
>= min
)
303 while (cache
->nobjs
< ARRAY_SIZE(cache
->objects
)) {
304 page
= alloc_page(GFP_KERNEL
);
307 set_page_private(page
, 0);
308 cache
->objects
[cache
->nobjs
++] = page_address(page
);
313 static void mmu_free_memory_cache_page(struct kvm_mmu_memory_cache
*mc
)
316 free_page((unsigned long)mc
->objects
[--mc
->nobjs
]);
319 static int mmu_topup_memory_caches(struct kvm_vcpu
*vcpu
)
323 r
= mmu_topup_memory_cache(&vcpu
->arch
.mmu_pte_chain_cache
,
327 r
= mmu_topup_memory_cache(&vcpu
->arch
.mmu_rmap_desc_cache
,
331 r
= mmu_topup_memory_cache_page(&vcpu
->arch
.mmu_page_cache
, 8);
334 r
= mmu_topup_memory_cache(&vcpu
->arch
.mmu_page_header_cache
,
335 mmu_page_header_cache
, 4);
340 static void mmu_free_memory_caches(struct kvm_vcpu
*vcpu
)
342 mmu_free_memory_cache(&vcpu
->arch
.mmu_pte_chain_cache
);
343 mmu_free_memory_cache(&vcpu
->arch
.mmu_rmap_desc_cache
);
344 mmu_free_memory_cache_page(&vcpu
->arch
.mmu_page_cache
);
345 mmu_free_memory_cache(&vcpu
->arch
.mmu_page_header_cache
);
348 static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache
*mc
,
354 p
= mc
->objects
[--mc
->nobjs
];
358 static struct kvm_pte_chain
*mmu_alloc_pte_chain(struct kvm_vcpu
*vcpu
)
360 return mmu_memory_cache_alloc(&vcpu
->arch
.mmu_pte_chain_cache
,
361 sizeof(struct kvm_pte_chain
));
364 static void mmu_free_pte_chain(struct kvm_pte_chain
*pc
)
369 static struct kvm_rmap_desc
*mmu_alloc_rmap_desc(struct kvm_vcpu
*vcpu
)
371 return mmu_memory_cache_alloc(&vcpu
->arch
.mmu_rmap_desc_cache
,
372 sizeof(struct kvm_rmap_desc
));
375 static void mmu_free_rmap_desc(struct kvm_rmap_desc
*rd
)
381 * Return the pointer to the largepage write count for a given
382 * gfn, handling slots that are not large page aligned.
384 static int *slot_largepage_idx(gfn_t gfn
, struct kvm_memory_slot
*slot
)
388 idx
= (gfn
/ KVM_PAGES_PER_HPAGE
) -
389 (slot
->base_gfn
/ KVM_PAGES_PER_HPAGE
);
390 return &slot
->lpage_info
[idx
].write_count
;
393 static void account_shadowed(struct kvm
*kvm
, gfn_t gfn
)
397 gfn
= unalias_gfn(kvm
, gfn
);
398 write_count
= slot_largepage_idx(gfn
,
399 gfn_to_memslot_unaliased(kvm
, gfn
));
403 static void unaccount_shadowed(struct kvm
*kvm
, gfn_t gfn
)
407 gfn
= unalias_gfn(kvm
, gfn
);
408 write_count
= slot_largepage_idx(gfn
,
409 gfn_to_memslot_unaliased(kvm
, gfn
));
411 WARN_ON(*write_count
< 0);
414 static int has_wrprotected_page(struct kvm
*kvm
, gfn_t gfn
)
416 struct kvm_memory_slot
*slot
;
419 gfn
= unalias_gfn(kvm
, gfn
);
420 slot
= gfn_to_memslot_unaliased(kvm
, gfn
);
422 largepage_idx
= slot_largepage_idx(gfn
, slot
);
423 return *largepage_idx
;
429 static int host_largepage_backed(struct kvm
*kvm
, gfn_t gfn
)
431 struct vm_area_struct
*vma
;
435 addr
= gfn_to_hva(kvm
, gfn
);
436 if (kvm_is_error_hva(addr
))
439 down_read(¤t
->mm
->mmap_sem
);
440 vma
= find_vma(current
->mm
, addr
);
441 if (vma
&& is_vm_hugetlb_page(vma
))
443 up_read(¤t
->mm
->mmap_sem
);
448 static int is_largepage_backed(struct kvm_vcpu
*vcpu
, gfn_t large_gfn
)
450 struct kvm_memory_slot
*slot
;
452 if (has_wrprotected_page(vcpu
->kvm
, large_gfn
))
455 if (!host_largepage_backed(vcpu
->kvm
, large_gfn
))
458 slot
= gfn_to_memslot(vcpu
->kvm
, large_gfn
);
459 if (slot
&& slot
->dirty_bitmap
)
466 * Take gfn and return the reverse mapping to it.
467 * Note: gfn must be unaliased before this function get called
470 static unsigned long *gfn_to_rmap(struct kvm
*kvm
, gfn_t gfn
, int lpage
)
472 struct kvm_memory_slot
*slot
;
475 slot
= gfn_to_memslot(kvm
, gfn
);
477 return &slot
->rmap
[gfn
- slot
->base_gfn
];
479 idx
= (gfn
/ KVM_PAGES_PER_HPAGE
) -
480 (slot
->base_gfn
/ KVM_PAGES_PER_HPAGE
);
482 return &slot
->lpage_info
[idx
].rmap_pde
;
486 * Reverse mapping data structures:
488 * If rmapp bit zero is zero, then rmapp point to the shadw page table entry
489 * that points to page_address(page).
491 * If rmapp bit zero is one, (then rmap & ~1) points to a struct kvm_rmap_desc
492 * containing more mappings.
494 * Returns the number of rmap entries before the spte was added or zero if
495 * the spte was not added.
498 static int rmap_add(struct kvm_vcpu
*vcpu
, u64
*spte
, gfn_t gfn
, int lpage
)
500 struct kvm_mmu_page
*sp
;
501 struct kvm_rmap_desc
*desc
;
502 unsigned long *rmapp
;
505 if (!is_rmap_pte(*spte
))
507 gfn
= unalias_gfn(vcpu
->kvm
, gfn
);
508 sp
= page_header(__pa(spte
));
509 sp
->gfns
[spte
- sp
->spt
] = gfn
;
510 rmapp
= gfn_to_rmap(vcpu
->kvm
, gfn
, lpage
);
512 rmap_printk("rmap_add: %p %llx 0->1\n", spte
, *spte
);
513 *rmapp
= (unsigned long)spte
;
514 } else if (!(*rmapp
& 1)) {
515 rmap_printk("rmap_add: %p %llx 1->many\n", spte
, *spte
);
516 desc
= mmu_alloc_rmap_desc(vcpu
);
517 desc
->shadow_ptes
[0] = (u64
*)*rmapp
;
518 desc
->shadow_ptes
[1] = spte
;
519 *rmapp
= (unsigned long)desc
| 1;
521 rmap_printk("rmap_add: %p %llx many->many\n", spte
, *spte
);
522 desc
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
523 while (desc
->shadow_ptes
[RMAP_EXT
-1] && desc
->more
) {
527 if (desc
->shadow_ptes
[RMAP_EXT
-1]) {
528 desc
->more
= mmu_alloc_rmap_desc(vcpu
);
531 for (i
= 0; desc
->shadow_ptes
[i
]; ++i
)
533 desc
->shadow_ptes
[i
] = spte
;
538 static void rmap_desc_remove_entry(unsigned long *rmapp
,
539 struct kvm_rmap_desc
*desc
,
541 struct kvm_rmap_desc
*prev_desc
)
545 for (j
= RMAP_EXT
- 1; !desc
->shadow_ptes
[j
] && j
> i
; --j
)
547 desc
->shadow_ptes
[i
] = desc
->shadow_ptes
[j
];
548 desc
->shadow_ptes
[j
] = NULL
;
551 if (!prev_desc
&& !desc
->more
)
552 *rmapp
= (unsigned long)desc
->shadow_ptes
[0];
555 prev_desc
->more
= desc
->more
;
557 *rmapp
= (unsigned long)desc
->more
| 1;
558 mmu_free_rmap_desc(desc
);
561 static void rmap_remove(struct kvm
*kvm
, u64
*spte
)
563 struct kvm_rmap_desc
*desc
;
564 struct kvm_rmap_desc
*prev_desc
;
565 struct kvm_mmu_page
*sp
;
567 unsigned long *rmapp
;
570 if (!is_rmap_pte(*spte
))
572 sp
= page_header(__pa(spte
));
573 pfn
= spte_to_pfn(*spte
);
574 if (*spte
& shadow_accessed_mask
)
575 kvm_set_pfn_accessed(pfn
);
576 if (is_writeble_pte(*spte
))
577 kvm_release_pfn_dirty(pfn
);
579 kvm_release_pfn_clean(pfn
);
580 rmapp
= gfn_to_rmap(kvm
, sp
->gfns
[spte
- sp
->spt
], is_large_pte(*spte
));
582 printk(KERN_ERR
"rmap_remove: %p %llx 0->BUG\n", spte
, *spte
);
584 } else if (!(*rmapp
& 1)) {
585 rmap_printk("rmap_remove: %p %llx 1->0\n", spte
, *spte
);
586 if ((u64
*)*rmapp
!= spte
) {
587 printk(KERN_ERR
"rmap_remove: %p %llx 1->BUG\n",
593 rmap_printk("rmap_remove: %p %llx many->many\n", spte
, *spte
);
594 desc
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
597 for (i
= 0; i
< RMAP_EXT
&& desc
->shadow_ptes
[i
]; ++i
)
598 if (desc
->shadow_ptes
[i
] == spte
) {
599 rmap_desc_remove_entry(rmapp
,
611 static u64
*rmap_next(struct kvm
*kvm
, unsigned long *rmapp
, u64
*spte
)
613 struct kvm_rmap_desc
*desc
;
614 struct kvm_rmap_desc
*prev_desc
;
620 else if (!(*rmapp
& 1)) {
622 return (u64
*)*rmapp
;
625 desc
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
629 for (i
= 0; i
< RMAP_EXT
&& desc
->shadow_ptes
[i
]; ++i
) {
630 if (prev_spte
== spte
)
631 return desc
->shadow_ptes
[i
];
632 prev_spte
= desc
->shadow_ptes
[i
];
639 static int rmap_write_protect(struct kvm
*kvm
, u64 gfn
)
641 unsigned long *rmapp
;
643 int write_protected
= 0;
645 gfn
= unalias_gfn(kvm
, gfn
);
646 rmapp
= gfn_to_rmap(kvm
, gfn
, 0);
648 spte
= rmap_next(kvm
, rmapp
, NULL
);
651 BUG_ON(!(*spte
& PT_PRESENT_MASK
));
652 rmap_printk("rmap_write_protect: spte %p %llx\n", spte
, *spte
);
653 if (is_writeble_pte(*spte
)) {
654 set_shadow_pte(spte
, *spte
& ~PT_WRITABLE_MASK
);
657 spte
= rmap_next(kvm
, rmapp
, spte
);
659 if (write_protected
) {
662 spte
= rmap_next(kvm
, rmapp
, NULL
);
663 pfn
= spte_to_pfn(*spte
);
664 kvm_set_pfn_dirty(pfn
);
667 /* check for huge page mappings */
668 rmapp
= gfn_to_rmap(kvm
, gfn
, 1);
669 spte
= rmap_next(kvm
, rmapp
, NULL
);
672 BUG_ON(!(*spte
& PT_PRESENT_MASK
));
673 BUG_ON((*spte
& (PT_PAGE_SIZE_MASK
|PT_PRESENT_MASK
)) != (PT_PAGE_SIZE_MASK
|PT_PRESENT_MASK
));
674 pgprintk("rmap_write_protect(large): spte %p %llx %lld\n", spte
, *spte
, gfn
);
675 if (is_writeble_pte(*spte
)) {
676 rmap_remove(kvm
, spte
);
678 set_shadow_pte(spte
, shadow_trap_nonpresent_pte
);
682 spte
= rmap_next(kvm
, rmapp
, spte
);
685 return write_protected
;
688 static int kvm_unmap_rmapp(struct kvm
*kvm
, unsigned long *rmapp
)
691 int need_tlb_flush
= 0;
693 while ((spte
= rmap_next(kvm
, rmapp
, NULL
))) {
694 BUG_ON(!(*spte
& PT_PRESENT_MASK
));
695 rmap_printk("kvm_rmap_unmap_hva: spte %p %llx\n", spte
, *spte
);
696 rmap_remove(kvm
, spte
);
697 set_shadow_pte(spte
, shadow_trap_nonpresent_pte
);
700 return need_tlb_flush
;
703 static int kvm_handle_hva(struct kvm
*kvm
, unsigned long hva
,
704 int (*handler
)(struct kvm
*kvm
, unsigned long *rmapp
))
710 * If mmap_sem isn't taken, we can look the memslots with only
711 * the mmu_lock by skipping over the slots with userspace_addr == 0.
713 for (i
= 0; i
< kvm
->nmemslots
; i
++) {
714 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[i
];
715 unsigned long start
= memslot
->userspace_addr
;
718 /* mmu_lock protects userspace_addr */
722 end
= start
+ (memslot
->npages
<< PAGE_SHIFT
);
723 if (hva
>= start
&& hva
< end
) {
724 gfn_t gfn_offset
= (hva
- start
) >> PAGE_SHIFT
;
725 retval
|= handler(kvm
, &memslot
->rmap
[gfn_offset
]);
726 retval
|= handler(kvm
,
727 &memslot
->lpage_info
[
729 KVM_PAGES_PER_HPAGE
].rmap_pde
);
736 int kvm_unmap_hva(struct kvm
*kvm
, unsigned long hva
)
738 return kvm_handle_hva(kvm
, hva
, kvm_unmap_rmapp
);
741 static int kvm_age_rmapp(struct kvm
*kvm
, unsigned long *rmapp
)
746 /* always return old for EPT */
747 if (!shadow_accessed_mask
)
750 spte
= rmap_next(kvm
, rmapp
, NULL
);
754 BUG_ON(!(_spte
& PT_PRESENT_MASK
));
755 _young
= _spte
& PT_ACCESSED_MASK
;
758 clear_bit(PT_ACCESSED_SHIFT
, (unsigned long *)spte
);
760 spte
= rmap_next(kvm
, rmapp
, spte
);
765 #define RMAP_RECYCLE_THRESHOLD 1000
767 static void rmap_recycle(struct kvm_vcpu
*vcpu
, gfn_t gfn
, int lpage
)
769 unsigned long *rmapp
;
771 gfn
= unalias_gfn(vcpu
->kvm
, gfn
);
772 rmapp
= gfn_to_rmap(vcpu
->kvm
, gfn
, lpage
);
774 kvm_unmap_rmapp(vcpu
->kvm
, rmapp
);
775 kvm_flush_remote_tlbs(vcpu
->kvm
);
778 int kvm_age_hva(struct kvm
*kvm
, unsigned long hva
)
780 return kvm_handle_hva(kvm
, hva
, kvm_age_rmapp
);
784 static int is_empty_shadow_page(u64
*spt
)
789 for (pos
= spt
, end
= pos
+ PAGE_SIZE
/ sizeof(u64
); pos
!= end
; pos
++)
790 if (is_shadow_present_pte(*pos
)) {
791 printk(KERN_ERR
"%s: %p %llx\n", __func__
,
799 static void kvm_mmu_free_page(struct kvm
*kvm
, struct kvm_mmu_page
*sp
)
801 ASSERT(is_empty_shadow_page(sp
->spt
));
803 __free_page(virt_to_page(sp
->spt
));
804 __free_page(virt_to_page(sp
->gfns
));
806 ++kvm
->arch
.n_free_mmu_pages
;
809 static unsigned kvm_page_table_hashfn(gfn_t gfn
)
811 return gfn
& ((1 << KVM_MMU_HASH_SHIFT
) - 1);
814 static struct kvm_mmu_page
*kvm_mmu_alloc_page(struct kvm_vcpu
*vcpu
,
817 struct kvm_mmu_page
*sp
;
819 sp
= mmu_memory_cache_alloc(&vcpu
->arch
.mmu_page_header_cache
, sizeof *sp
);
820 sp
->spt
= mmu_memory_cache_alloc(&vcpu
->arch
.mmu_page_cache
, PAGE_SIZE
);
821 sp
->gfns
= mmu_memory_cache_alloc(&vcpu
->arch
.mmu_page_cache
, PAGE_SIZE
);
822 set_page_private(virt_to_page(sp
->spt
), (unsigned long)sp
);
823 list_add(&sp
->link
, &vcpu
->kvm
->arch
.active_mmu_pages
);
824 INIT_LIST_HEAD(&sp
->oos_link
);
825 bitmap_zero(sp
->slot_bitmap
, KVM_MEMORY_SLOTS
+ KVM_PRIVATE_MEM_SLOTS
);
827 sp
->parent_pte
= parent_pte
;
828 --vcpu
->kvm
->arch
.n_free_mmu_pages
;
832 static void mmu_page_add_parent_pte(struct kvm_vcpu
*vcpu
,
833 struct kvm_mmu_page
*sp
, u64
*parent_pte
)
835 struct kvm_pte_chain
*pte_chain
;
836 struct hlist_node
*node
;
841 if (!sp
->multimapped
) {
842 u64
*old
= sp
->parent_pte
;
845 sp
->parent_pte
= parent_pte
;
849 pte_chain
= mmu_alloc_pte_chain(vcpu
);
850 INIT_HLIST_HEAD(&sp
->parent_ptes
);
851 hlist_add_head(&pte_chain
->link
, &sp
->parent_ptes
);
852 pte_chain
->parent_ptes
[0] = old
;
854 hlist_for_each_entry(pte_chain
, node
, &sp
->parent_ptes
, link
) {
855 if (pte_chain
->parent_ptes
[NR_PTE_CHAIN_ENTRIES
-1])
857 for (i
= 0; i
< NR_PTE_CHAIN_ENTRIES
; ++i
)
858 if (!pte_chain
->parent_ptes
[i
]) {
859 pte_chain
->parent_ptes
[i
] = parent_pte
;
863 pte_chain
= mmu_alloc_pte_chain(vcpu
);
865 hlist_add_head(&pte_chain
->link
, &sp
->parent_ptes
);
866 pte_chain
->parent_ptes
[0] = parent_pte
;
869 static void mmu_page_remove_parent_pte(struct kvm_mmu_page
*sp
,
872 struct kvm_pte_chain
*pte_chain
;
873 struct hlist_node
*node
;
876 if (!sp
->multimapped
) {
877 BUG_ON(sp
->parent_pte
!= parent_pte
);
878 sp
->parent_pte
= NULL
;
881 hlist_for_each_entry(pte_chain
, node
, &sp
->parent_ptes
, link
)
882 for (i
= 0; i
< NR_PTE_CHAIN_ENTRIES
; ++i
) {
883 if (!pte_chain
->parent_ptes
[i
])
885 if (pte_chain
->parent_ptes
[i
] != parent_pte
)
887 while (i
+ 1 < NR_PTE_CHAIN_ENTRIES
888 && pte_chain
->parent_ptes
[i
+ 1]) {
889 pte_chain
->parent_ptes
[i
]
890 = pte_chain
->parent_ptes
[i
+ 1];
893 pte_chain
->parent_ptes
[i
] = NULL
;
895 hlist_del(&pte_chain
->link
);
896 mmu_free_pte_chain(pte_chain
);
897 if (hlist_empty(&sp
->parent_ptes
)) {
899 sp
->parent_pte
= NULL
;
908 static void mmu_parent_walk(struct kvm_vcpu
*vcpu
, struct kvm_mmu_page
*sp
,
909 mmu_parent_walk_fn fn
)
911 struct kvm_pte_chain
*pte_chain
;
912 struct hlist_node
*node
;
913 struct kvm_mmu_page
*parent_sp
;
916 if (!sp
->multimapped
&& sp
->parent_pte
) {
917 parent_sp
= page_header(__pa(sp
->parent_pte
));
919 mmu_parent_walk(vcpu
, parent_sp
, fn
);
922 hlist_for_each_entry(pte_chain
, node
, &sp
->parent_ptes
, link
)
923 for (i
= 0; i
< NR_PTE_CHAIN_ENTRIES
; ++i
) {
924 if (!pte_chain
->parent_ptes
[i
])
926 parent_sp
= page_header(__pa(pte_chain
->parent_ptes
[i
]));
928 mmu_parent_walk(vcpu
, parent_sp
, fn
);
932 static void kvm_mmu_update_unsync_bitmap(u64
*spte
)
935 struct kvm_mmu_page
*sp
= page_header(__pa(spte
));
937 index
= spte
- sp
->spt
;
938 if (!__test_and_set_bit(index
, sp
->unsync_child_bitmap
))
939 sp
->unsync_children
++;
940 WARN_ON(!sp
->unsync_children
);
943 static void kvm_mmu_update_parents_unsync(struct kvm_mmu_page
*sp
)
945 struct kvm_pte_chain
*pte_chain
;
946 struct hlist_node
*node
;
952 if (!sp
->multimapped
) {
953 kvm_mmu_update_unsync_bitmap(sp
->parent_pte
);
957 hlist_for_each_entry(pte_chain
, node
, &sp
->parent_ptes
, link
)
958 for (i
= 0; i
< NR_PTE_CHAIN_ENTRIES
; ++i
) {
959 if (!pte_chain
->parent_ptes
[i
])
961 kvm_mmu_update_unsync_bitmap(pte_chain
->parent_ptes
[i
]);
965 static int unsync_walk_fn(struct kvm_vcpu
*vcpu
, struct kvm_mmu_page
*sp
)
967 kvm_mmu_update_parents_unsync(sp
);
971 static void kvm_mmu_mark_parents_unsync(struct kvm_vcpu
*vcpu
,
972 struct kvm_mmu_page
*sp
)
974 mmu_parent_walk(vcpu
, sp
, unsync_walk_fn
);
975 kvm_mmu_update_parents_unsync(sp
);
978 static void nonpaging_prefetch_page(struct kvm_vcpu
*vcpu
,
979 struct kvm_mmu_page
*sp
)
983 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
)
984 sp
->spt
[i
] = shadow_trap_nonpresent_pte
;
987 static int nonpaging_sync_page(struct kvm_vcpu
*vcpu
,
988 struct kvm_mmu_page
*sp
)
993 static void nonpaging_invlpg(struct kvm_vcpu
*vcpu
, gva_t gva
)
997 #define KVM_PAGE_ARRAY_NR 16
999 struct kvm_mmu_pages
{
1000 struct mmu_page_and_offset
{
1001 struct kvm_mmu_page
*sp
;
1003 } page
[KVM_PAGE_ARRAY_NR
];
1007 #define for_each_unsync_children(bitmap, idx) \
1008 for (idx = find_first_bit(bitmap, 512); \
1010 idx = find_next_bit(bitmap, 512, idx+1))
1012 static int mmu_pages_add(struct kvm_mmu_pages
*pvec
, struct kvm_mmu_page
*sp
,
1018 for (i
=0; i
< pvec
->nr
; i
++)
1019 if (pvec
->page
[i
].sp
== sp
)
1022 pvec
->page
[pvec
->nr
].sp
= sp
;
1023 pvec
->page
[pvec
->nr
].idx
= idx
;
1025 return (pvec
->nr
== KVM_PAGE_ARRAY_NR
);
1028 static int __mmu_unsync_walk(struct kvm_mmu_page
*sp
,
1029 struct kvm_mmu_pages
*pvec
)
1031 int i
, ret
, nr_unsync_leaf
= 0;
1033 for_each_unsync_children(sp
->unsync_child_bitmap
, i
) {
1034 u64 ent
= sp
->spt
[i
];
1036 if (is_shadow_present_pte(ent
) && !is_large_pte(ent
)) {
1037 struct kvm_mmu_page
*child
;
1038 child
= page_header(ent
& PT64_BASE_ADDR_MASK
);
1040 if (child
->unsync_children
) {
1041 if (mmu_pages_add(pvec
, child
, i
))
1044 ret
= __mmu_unsync_walk(child
, pvec
);
1046 __clear_bit(i
, sp
->unsync_child_bitmap
);
1048 nr_unsync_leaf
+= ret
;
1053 if (child
->unsync
) {
1055 if (mmu_pages_add(pvec
, child
, i
))
1061 if (find_first_bit(sp
->unsync_child_bitmap
, 512) == 512)
1062 sp
->unsync_children
= 0;
1064 return nr_unsync_leaf
;
1067 static int mmu_unsync_walk(struct kvm_mmu_page
*sp
,
1068 struct kvm_mmu_pages
*pvec
)
1070 if (!sp
->unsync_children
)
1073 mmu_pages_add(pvec
, sp
, 0);
1074 return __mmu_unsync_walk(sp
, pvec
);
1077 static struct kvm_mmu_page
*kvm_mmu_lookup_page(struct kvm
*kvm
, gfn_t gfn
)
1080 struct hlist_head
*bucket
;
1081 struct kvm_mmu_page
*sp
;
1082 struct hlist_node
*node
;
1084 pgprintk("%s: looking for gfn %lx\n", __func__
, gfn
);
1085 index
= kvm_page_table_hashfn(gfn
);
1086 bucket
= &kvm
->arch
.mmu_page_hash
[index
];
1087 hlist_for_each_entry(sp
, node
, bucket
, hash_link
)
1088 if (sp
->gfn
== gfn
&& !sp
->role
.direct
1089 && !sp
->role
.invalid
) {
1090 pgprintk("%s: found role %x\n",
1091 __func__
, sp
->role
.word
);
1097 static void kvm_unlink_unsync_page(struct kvm
*kvm
, struct kvm_mmu_page
*sp
)
1099 WARN_ON(!sp
->unsync
);
1101 --kvm
->stat
.mmu_unsync
;
1104 static int kvm_mmu_zap_page(struct kvm
*kvm
, struct kvm_mmu_page
*sp
);
1106 static int kvm_sync_page(struct kvm_vcpu
*vcpu
, struct kvm_mmu_page
*sp
)
1108 if (sp
->role
.glevels
!= vcpu
->arch
.mmu
.root_level
) {
1109 kvm_mmu_zap_page(vcpu
->kvm
, sp
);
1113 if (rmap_write_protect(vcpu
->kvm
, sp
->gfn
))
1114 kvm_flush_remote_tlbs(vcpu
->kvm
);
1115 kvm_unlink_unsync_page(vcpu
->kvm
, sp
);
1116 if (vcpu
->arch
.mmu
.sync_page(vcpu
, sp
)) {
1117 kvm_mmu_zap_page(vcpu
->kvm
, sp
);
1121 kvm_mmu_flush_tlb(vcpu
);
1125 struct mmu_page_path
{
1126 struct kvm_mmu_page
*parent
[PT64_ROOT_LEVEL
-1];
1127 unsigned int idx
[PT64_ROOT_LEVEL
-1];
1130 #define for_each_sp(pvec, sp, parents, i) \
1131 for (i = mmu_pages_next(&pvec, &parents, -1), \
1132 sp = pvec.page[i].sp; \
1133 i < pvec.nr && ({ sp = pvec.page[i].sp; 1;}); \
1134 i = mmu_pages_next(&pvec, &parents, i))
1136 static int mmu_pages_next(struct kvm_mmu_pages
*pvec
,
1137 struct mmu_page_path
*parents
,
1142 for (n
= i
+1; n
< pvec
->nr
; n
++) {
1143 struct kvm_mmu_page
*sp
= pvec
->page
[n
].sp
;
1145 if (sp
->role
.level
== PT_PAGE_TABLE_LEVEL
) {
1146 parents
->idx
[0] = pvec
->page
[n
].idx
;
1150 parents
->parent
[sp
->role
.level
-2] = sp
;
1151 parents
->idx
[sp
->role
.level
-1] = pvec
->page
[n
].idx
;
1157 static void mmu_pages_clear_parents(struct mmu_page_path
*parents
)
1159 struct kvm_mmu_page
*sp
;
1160 unsigned int level
= 0;
1163 unsigned int idx
= parents
->idx
[level
];
1165 sp
= parents
->parent
[level
];
1169 --sp
->unsync_children
;
1170 WARN_ON((int)sp
->unsync_children
< 0);
1171 __clear_bit(idx
, sp
->unsync_child_bitmap
);
1173 } while (level
< PT64_ROOT_LEVEL
-1 && !sp
->unsync_children
);
1176 static void kvm_mmu_pages_init(struct kvm_mmu_page
*parent
,
1177 struct mmu_page_path
*parents
,
1178 struct kvm_mmu_pages
*pvec
)
1180 parents
->parent
[parent
->role
.level
-1] = NULL
;
1184 static void mmu_sync_children(struct kvm_vcpu
*vcpu
,
1185 struct kvm_mmu_page
*parent
)
1188 struct kvm_mmu_page
*sp
;
1189 struct mmu_page_path parents
;
1190 struct kvm_mmu_pages pages
;
1192 kvm_mmu_pages_init(parent
, &parents
, &pages
);
1193 while (mmu_unsync_walk(parent
, &pages
)) {
1196 for_each_sp(pages
, sp
, parents
, i
)
1197 protected |= rmap_write_protect(vcpu
->kvm
, sp
->gfn
);
1200 kvm_flush_remote_tlbs(vcpu
->kvm
);
1202 for_each_sp(pages
, sp
, parents
, i
) {
1203 kvm_sync_page(vcpu
, sp
);
1204 mmu_pages_clear_parents(&parents
);
1206 cond_resched_lock(&vcpu
->kvm
->mmu_lock
);
1207 kvm_mmu_pages_init(parent
, &parents
, &pages
);
1211 static struct kvm_mmu_page
*kvm_mmu_get_page(struct kvm_vcpu
*vcpu
,
1219 union kvm_mmu_page_role role
;
1222 struct hlist_head
*bucket
;
1223 struct kvm_mmu_page
*sp
;
1224 struct hlist_node
*node
, *tmp
;
1226 role
= vcpu
->arch
.mmu
.base_role
;
1228 role
.direct
= direct
;
1229 role
.access
= access
;
1230 if (vcpu
->arch
.mmu
.root_level
<= PT32_ROOT_LEVEL
) {
1231 quadrant
= gaddr
>> (PAGE_SHIFT
+ (PT64_PT_BITS
* level
));
1232 quadrant
&= (1 << ((PT32_PT_BITS
- PT64_PT_BITS
) * level
)) - 1;
1233 role
.quadrant
= quadrant
;
1235 pgprintk("%s: looking gfn %lx role %x\n", __func__
,
1237 index
= kvm_page_table_hashfn(gfn
);
1238 bucket
= &vcpu
->kvm
->arch
.mmu_page_hash
[index
];
1239 hlist_for_each_entry_safe(sp
, node
, tmp
, bucket
, hash_link
)
1240 if (sp
->gfn
== gfn
) {
1242 if (kvm_sync_page(vcpu
, sp
))
1245 if (sp
->role
.word
!= role
.word
)
1248 mmu_page_add_parent_pte(vcpu
, sp
, parent_pte
);
1249 if (sp
->unsync_children
) {
1250 set_bit(KVM_REQ_MMU_SYNC
, &vcpu
->requests
);
1251 kvm_mmu_mark_parents_unsync(vcpu
, sp
);
1253 pgprintk("%s: found\n", __func__
);
1256 ++vcpu
->kvm
->stat
.mmu_cache_miss
;
1257 sp
= kvm_mmu_alloc_page(vcpu
, parent_pte
);
1260 pgprintk("%s: adding gfn %lx role %x\n", __func__
, gfn
, role
.word
);
1263 hlist_add_head(&sp
->hash_link
, bucket
);
1265 if (rmap_write_protect(vcpu
->kvm
, gfn
))
1266 kvm_flush_remote_tlbs(vcpu
->kvm
);
1267 account_shadowed(vcpu
->kvm
, gfn
);
1269 if (shadow_trap_nonpresent_pte
!= shadow_notrap_nonpresent_pte
)
1270 vcpu
->arch
.mmu
.prefetch_page(vcpu
, sp
);
1272 nonpaging_prefetch_page(vcpu
, sp
);
1276 static void shadow_walk_init(struct kvm_shadow_walk_iterator
*iterator
,
1277 struct kvm_vcpu
*vcpu
, u64 addr
)
1279 iterator
->addr
= addr
;
1280 iterator
->shadow_addr
= vcpu
->arch
.mmu
.root_hpa
;
1281 iterator
->level
= vcpu
->arch
.mmu
.shadow_root_level
;
1282 if (iterator
->level
== PT32E_ROOT_LEVEL
) {
1283 iterator
->shadow_addr
1284 = vcpu
->arch
.mmu
.pae_root
[(addr
>> 30) & 3];
1285 iterator
->shadow_addr
&= PT64_BASE_ADDR_MASK
;
1287 if (!iterator
->shadow_addr
)
1288 iterator
->level
= 0;
1292 static bool shadow_walk_okay(struct kvm_shadow_walk_iterator
*iterator
)
1294 if (iterator
->level
< PT_PAGE_TABLE_LEVEL
)
1296 iterator
->index
= SHADOW_PT_INDEX(iterator
->addr
, iterator
->level
);
1297 iterator
->sptep
= ((u64
*)__va(iterator
->shadow_addr
)) + iterator
->index
;
1301 static void shadow_walk_next(struct kvm_shadow_walk_iterator
*iterator
)
1303 iterator
->shadow_addr
= *iterator
->sptep
& PT64_BASE_ADDR_MASK
;
1307 static void kvm_mmu_page_unlink_children(struct kvm
*kvm
,
1308 struct kvm_mmu_page
*sp
)
1316 if (sp
->role
.level
== PT_PAGE_TABLE_LEVEL
) {
1317 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
1318 if (is_shadow_present_pte(pt
[i
]))
1319 rmap_remove(kvm
, &pt
[i
]);
1320 pt
[i
] = shadow_trap_nonpresent_pte
;
1325 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
1328 if (is_shadow_present_pte(ent
)) {
1329 if (!is_large_pte(ent
)) {
1330 ent
&= PT64_BASE_ADDR_MASK
;
1331 mmu_page_remove_parent_pte(page_header(ent
),
1335 rmap_remove(kvm
, &pt
[i
]);
1338 pt
[i
] = shadow_trap_nonpresent_pte
;
1342 static void kvm_mmu_put_page(struct kvm_mmu_page
*sp
, u64
*parent_pte
)
1344 mmu_page_remove_parent_pte(sp
, parent_pte
);
1347 static void kvm_mmu_reset_last_pte_updated(struct kvm
*kvm
)
1351 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
)
1353 kvm
->vcpus
[i
]->arch
.last_pte_updated
= NULL
;
1356 static void kvm_mmu_unlink_parents(struct kvm
*kvm
, struct kvm_mmu_page
*sp
)
1360 while (sp
->multimapped
|| sp
->parent_pte
) {
1361 if (!sp
->multimapped
)
1362 parent_pte
= sp
->parent_pte
;
1364 struct kvm_pte_chain
*chain
;
1366 chain
= container_of(sp
->parent_ptes
.first
,
1367 struct kvm_pte_chain
, link
);
1368 parent_pte
= chain
->parent_ptes
[0];
1370 BUG_ON(!parent_pte
);
1371 kvm_mmu_put_page(sp
, parent_pte
);
1372 set_shadow_pte(parent_pte
, shadow_trap_nonpresent_pte
);
1376 static int mmu_zap_unsync_children(struct kvm
*kvm
,
1377 struct kvm_mmu_page
*parent
)
1380 struct mmu_page_path parents
;
1381 struct kvm_mmu_pages pages
;
1383 if (parent
->role
.level
== PT_PAGE_TABLE_LEVEL
)
1386 kvm_mmu_pages_init(parent
, &parents
, &pages
);
1387 while (mmu_unsync_walk(parent
, &pages
)) {
1388 struct kvm_mmu_page
*sp
;
1390 for_each_sp(pages
, sp
, parents
, i
) {
1391 kvm_mmu_zap_page(kvm
, sp
);
1392 mmu_pages_clear_parents(&parents
);
1395 kvm_mmu_pages_init(parent
, &parents
, &pages
);
1401 static int kvm_mmu_zap_page(struct kvm
*kvm
, struct kvm_mmu_page
*sp
)
1404 ++kvm
->stat
.mmu_shadow_zapped
;
1405 ret
= mmu_zap_unsync_children(kvm
, sp
);
1406 kvm_mmu_page_unlink_children(kvm
, sp
);
1407 kvm_mmu_unlink_parents(kvm
, sp
);
1408 kvm_flush_remote_tlbs(kvm
);
1409 if (!sp
->role
.invalid
&& !sp
->role
.direct
)
1410 unaccount_shadowed(kvm
, sp
->gfn
);
1412 kvm_unlink_unsync_page(kvm
, sp
);
1413 if (!sp
->root_count
) {
1414 hlist_del(&sp
->hash_link
);
1415 kvm_mmu_free_page(kvm
, sp
);
1417 sp
->role
.invalid
= 1;
1418 list_move(&sp
->link
, &kvm
->arch
.active_mmu_pages
);
1419 kvm_reload_remote_mmus(kvm
);
1421 kvm_mmu_reset_last_pte_updated(kvm
);
1426 * Changing the number of mmu pages allocated to the vm
1427 * Note: if kvm_nr_mmu_pages is too small, you will get dead lock
1429 void kvm_mmu_change_mmu_pages(struct kvm
*kvm
, unsigned int kvm_nr_mmu_pages
)
1433 used_pages
= kvm
->arch
.n_alloc_mmu_pages
- kvm
->arch
.n_free_mmu_pages
;
1434 used_pages
= max(0, used_pages
);
1437 * If we set the number of mmu pages to be smaller be than the
1438 * number of actived pages , we must to free some mmu pages before we
1442 if (used_pages
> kvm_nr_mmu_pages
) {
1443 while (used_pages
> kvm_nr_mmu_pages
) {
1444 struct kvm_mmu_page
*page
;
1446 page
= container_of(kvm
->arch
.active_mmu_pages
.prev
,
1447 struct kvm_mmu_page
, link
);
1448 kvm_mmu_zap_page(kvm
, page
);
1451 kvm
->arch
.n_free_mmu_pages
= 0;
1454 kvm
->arch
.n_free_mmu_pages
+= kvm_nr_mmu_pages
1455 - kvm
->arch
.n_alloc_mmu_pages
;
1457 kvm
->arch
.n_alloc_mmu_pages
= kvm_nr_mmu_pages
;
1460 static int kvm_mmu_unprotect_page(struct kvm
*kvm
, gfn_t gfn
)
1463 struct hlist_head
*bucket
;
1464 struct kvm_mmu_page
*sp
;
1465 struct hlist_node
*node
, *n
;
1468 pgprintk("%s: looking for gfn %lx\n", __func__
, gfn
);
1470 index
= kvm_page_table_hashfn(gfn
);
1471 bucket
= &kvm
->arch
.mmu_page_hash
[index
];
1472 hlist_for_each_entry_safe(sp
, node
, n
, bucket
, hash_link
)
1473 if (sp
->gfn
== gfn
&& !sp
->role
.direct
) {
1474 pgprintk("%s: gfn %lx role %x\n", __func__
, gfn
,
1477 if (kvm_mmu_zap_page(kvm
, sp
))
1483 static void mmu_unshadow(struct kvm
*kvm
, gfn_t gfn
)
1486 struct hlist_head
*bucket
;
1487 struct kvm_mmu_page
*sp
;
1488 struct hlist_node
*node
, *nn
;
1490 index
= kvm_page_table_hashfn(gfn
);
1491 bucket
= &kvm
->arch
.mmu_page_hash
[index
];
1492 hlist_for_each_entry_safe(sp
, node
, nn
, bucket
, hash_link
) {
1493 if (sp
->gfn
== gfn
&& !sp
->role
.direct
1494 && !sp
->role
.invalid
) {
1495 pgprintk("%s: zap %lx %x\n",
1496 __func__
, gfn
, sp
->role
.word
);
1497 kvm_mmu_zap_page(kvm
, sp
);
1502 static void page_header_update_slot(struct kvm
*kvm
, void *pte
, gfn_t gfn
)
1504 int slot
= memslot_id(kvm
, gfn_to_memslot(kvm
, gfn
));
1505 struct kvm_mmu_page
*sp
= page_header(__pa(pte
));
1507 __set_bit(slot
, sp
->slot_bitmap
);
1510 static void mmu_convert_notrap(struct kvm_mmu_page
*sp
)
1515 if (shadow_trap_nonpresent_pte
== shadow_notrap_nonpresent_pte
)
1518 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
1519 if (pt
[i
] == shadow_notrap_nonpresent_pte
)
1520 set_shadow_pte(&pt
[i
], shadow_trap_nonpresent_pte
);
1524 struct page
*gva_to_page(struct kvm_vcpu
*vcpu
, gva_t gva
)
1528 gpa_t gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, gva
);
1530 if (gpa
== UNMAPPED_GVA
)
1533 page
= gfn_to_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
1539 * The function is based on mtrr_type_lookup() in
1540 * arch/x86/kernel/cpu/mtrr/generic.c
1542 static int get_mtrr_type(struct mtrr_state_type
*mtrr_state
,
1547 u8 prev_match
, curr_match
;
1548 int num_var_ranges
= KVM_NR_VAR_MTRR
;
1550 if (!mtrr_state
->enabled
)
1553 /* Make end inclusive end, instead of exclusive */
1556 /* Look in fixed ranges. Just return the type as per start */
1557 if (mtrr_state
->have_fixed
&& (start
< 0x100000)) {
1560 if (start
< 0x80000) {
1562 idx
+= (start
>> 16);
1563 return mtrr_state
->fixed_ranges
[idx
];
1564 } else if (start
< 0xC0000) {
1566 idx
+= ((start
- 0x80000) >> 14);
1567 return mtrr_state
->fixed_ranges
[idx
];
1568 } else if (start
< 0x1000000) {
1570 idx
+= ((start
- 0xC0000) >> 12);
1571 return mtrr_state
->fixed_ranges
[idx
];
1576 * Look in variable ranges
1577 * Look of multiple ranges matching this address and pick type
1578 * as per MTRR precedence
1580 if (!(mtrr_state
->enabled
& 2))
1581 return mtrr_state
->def_type
;
1584 for (i
= 0; i
< num_var_ranges
; ++i
) {
1585 unsigned short start_state
, end_state
;
1587 if (!(mtrr_state
->var_ranges
[i
].mask_lo
& (1 << 11)))
1590 base
= (((u64
)mtrr_state
->var_ranges
[i
].base_hi
) << 32) +
1591 (mtrr_state
->var_ranges
[i
].base_lo
& PAGE_MASK
);
1592 mask
= (((u64
)mtrr_state
->var_ranges
[i
].mask_hi
) << 32) +
1593 (mtrr_state
->var_ranges
[i
].mask_lo
& PAGE_MASK
);
1595 start_state
= ((start
& mask
) == (base
& mask
));
1596 end_state
= ((end
& mask
) == (base
& mask
));
1597 if (start_state
!= end_state
)
1600 if ((start
& mask
) != (base
& mask
))
1603 curr_match
= mtrr_state
->var_ranges
[i
].base_lo
& 0xff;
1604 if (prev_match
== 0xFF) {
1605 prev_match
= curr_match
;
1609 if (prev_match
== MTRR_TYPE_UNCACHABLE
||
1610 curr_match
== MTRR_TYPE_UNCACHABLE
)
1611 return MTRR_TYPE_UNCACHABLE
;
1613 if ((prev_match
== MTRR_TYPE_WRBACK
&&
1614 curr_match
== MTRR_TYPE_WRTHROUGH
) ||
1615 (prev_match
== MTRR_TYPE_WRTHROUGH
&&
1616 curr_match
== MTRR_TYPE_WRBACK
)) {
1617 prev_match
= MTRR_TYPE_WRTHROUGH
;
1618 curr_match
= MTRR_TYPE_WRTHROUGH
;
1621 if (prev_match
!= curr_match
)
1622 return MTRR_TYPE_UNCACHABLE
;
1625 if (prev_match
!= 0xFF)
1628 return mtrr_state
->def_type
;
1631 u8
kvm_get_guest_memory_type(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1635 mtrr
= get_mtrr_type(&vcpu
->arch
.mtrr_state
, gfn
<< PAGE_SHIFT
,
1636 (gfn
<< PAGE_SHIFT
) + PAGE_SIZE
);
1637 if (mtrr
== 0xfe || mtrr
== 0xff)
1638 mtrr
= MTRR_TYPE_WRBACK
;
1641 EXPORT_SYMBOL_GPL(kvm_get_guest_memory_type
);
1643 static int kvm_unsync_page(struct kvm_vcpu
*vcpu
, struct kvm_mmu_page
*sp
)
1646 struct hlist_head
*bucket
;
1647 struct kvm_mmu_page
*s
;
1648 struct hlist_node
*node
, *n
;
1650 index
= kvm_page_table_hashfn(sp
->gfn
);
1651 bucket
= &vcpu
->kvm
->arch
.mmu_page_hash
[index
];
1652 /* don't unsync if pagetable is shadowed with multiple roles */
1653 hlist_for_each_entry_safe(s
, node
, n
, bucket
, hash_link
) {
1654 if (s
->gfn
!= sp
->gfn
|| s
->role
.direct
)
1656 if (s
->role
.word
!= sp
->role
.word
)
1659 ++vcpu
->kvm
->stat
.mmu_unsync
;
1662 kvm_mmu_mark_parents_unsync(vcpu
, sp
);
1664 mmu_convert_notrap(sp
);
1668 static int mmu_need_write_protect(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
1671 struct kvm_mmu_page
*shadow
;
1673 shadow
= kvm_mmu_lookup_page(vcpu
->kvm
, gfn
);
1675 if (shadow
->role
.level
!= PT_PAGE_TABLE_LEVEL
)
1679 if (can_unsync
&& oos_shadow
)
1680 return kvm_unsync_page(vcpu
, shadow
);
1686 static int set_spte(struct kvm_vcpu
*vcpu
, u64
*shadow_pte
,
1687 unsigned pte_access
, int user_fault
,
1688 int write_fault
, int dirty
, int largepage
,
1689 gfn_t gfn
, pfn_t pfn
, bool speculative
,
1696 * We don't set the accessed bit, since we sometimes want to see
1697 * whether the guest actually used the pte (in order to detect
1700 spte
= shadow_base_present_pte
| shadow_dirty_mask
;
1702 spte
|= shadow_accessed_mask
;
1704 pte_access
&= ~ACC_WRITE_MASK
;
1705 if (pte_access
& ACC_EXEC_MASK
)
1706 spte
|= shadow_x_mask
;
1708 spte
|= shadow_nx_mask
;
1709 if (pte_access
& ACC_USER_MASK
)
1710 spte
|= shadow_user_mask
;
1712 spte
|= PT_PAGE_SIZE_MASK
;
1714 spte
|= kvm_x86_ops
->get_mt_mask(vcpu
, gfn
,
1715 kvm_is_mmio_pfn(pfn
));
1717 spte
|= (u64
)pfn
<< PAGE_SHIFT
;
1719 if ((pte_access
& ACC_WRITE_MASK
)
1720 || (write_fault
&& !is_write_protection(vcpu
) && !user_fault
)) {
1722 if (largepage
&& has_wrprotected_page(vcpu
->kvm
, gfn
)) {
1724 spte
= shadow_trap_nonpresent_pte
;
1728 spte
|= PT_WRITABLE_MASK
;
1731 * Optimization: for pte sync, if spte was writable the hash
1732 * lookup is unnecessary (and expensive). Write protection
1733 * is responsibility of mmu_get_page / kvm_sync_page.
1734 * Same reasoning can be applied to dirty page accounting.
1736 if (!can_unsync
&& is_writeble_pte(*shadow_pte
))
1739 if (mmu_need_write_protect(vcpu
, gfn
, can_unsync
)) {
1740 pgprintk("%s: found shadow page for %lx, marking ro\n",
1743 pte_access
&= ~ACC_WRITE_MASK
;
1744 if (is_writeble_pte(spte
))
1745 spte
&= ~PT_WRITABLE_MASK
;
1749 if (pte_access
& ACC_WRITE_MASK
)
1750 mark_page_dirty(vcpu
->kvm
, gfn
);
1753 set_shadow_pte(shadow_pte
, spte
);
1757 static void mmu_set_spte(struct kvm_vcpu
*vcpu
, u64
*shadow_pte
,
1758 unsigned pt_access
, unsigned pte_access
,
1759 int user_fault
, int write_fault
, int dirty
,
1760 int *ptwrite
, int largepage
, gfn_t gfn
,
1761 pfn_t pfn
, bool speculative
)
1763 int was_rmapped
= 0;
1764 int was_writeble
= is_writeble_pte(*shadow_pte
);
1767 pgprintk("%s: spte %llx access %x write_fault %d"
1768 " user_fault %d gfn %lx\n",
1769 __func__
, *shadow_pte
, pt_access
,
1770 write_fault
, user_fault
, gfn
);
1772 if (is_rmap_pte(*shadow_pte
)) {
1774 * If we overwrite a PTE page pointer with a 2MB PMD, unlink
1775 * the parent of the now unreachable PTE.
1777 if (largepage
&& !is_large_pte(*shadow_pte
)) {
1778 struct kvm_mmu_page
*child
;
1779 u64 pte
= *shadow_pte
;
1781 child
= page_header(pte
& PT64_BASE_ADDR_MASK
);
1782 mmu_page_remove_parent_pte(child
, shadow_pte
);
1783 } else if (pfn
!= spte_to_pfn(*shadow_pte
)) {
1784 pgprintk("hfn old %lx new %lx\n",
1785 spte_to_pfn(*shadow_pte
), pfn
);
1786 rmap_remove(vcpu
->kvm
, shadow_pte
);
1790 if (set_spte(vcpu
, shadow_pte
, pte_access
, user_fault
, write_fault
,
1791 dirty
, largepage
, gfn
, pfn
, speculative
, true)) {
1794 kvm_x86_ops
->tlb_flush(vcpu
);
1797 pgprintk("%s: setting spte %llx\n", __func__
, *shadow_pte
);
1798 pgprintk("instantiating %s PTE (%s) at %ld (%llx) addr %p\n",
1799 is_large_pte(*shadow_pte
)? "2MB" : "4kB",
1800 is_present_pte(*shadow_pte
)?"RW":"R", gfn
,
1801 *shadow_pte
, shadow_pte
);
1802 if (!was_rmapped
&& is_large_pte(*shadow_pte
))
1803 ++vcpu
->kvm
->stat
.lpages
;
1805 page_header_update_slot(vcpu
->kvm
, shadow_pte
, gfn
);
1807 rmap_count
= rmap_add(vcpu
, shadow_pte
, gfn
, largepage
);
1808 if (!is_rmap_pte(*shadow_pte
))
1809 kvm_release_pfn_clean(pfn
);
1810 if (rmap_count
> RMAP_RECYCLE_THRESHOLD
)
1811 rmap_recycle(vcpu
, gfn
, largepage
);
1814 kvm_release_pfn_dirty(pfn
);
1816 kvm_release_pfn_clean(pfn
);
1819 vcpu
->arch
.last_pte_updated
= shadow_pte
;
1820 vcpu
->arch
.last_pte_gfn
= gfn
;
1824 static void nonpaging_new_cr3(struct kvm_vcpu
*vcpu
)
1828 static int __direct_map(struct kvm_vcpu
*vcpu
, gpa_t v
, int write
,
1829 int largepage
, gfn_t gfn
, pfn_t pfn
)
1831 struct kvm_shadow_walk_iterator iterator
;
1832 struct kvm_mmu_page
*sp
;
1836 for_each_shadow_entry(vcpu
, (u64
)gfn
<< PAGE_SHIFT
, iterator
) {
1837 if (iterator
.level
== PT_PAGE_TABLE_LEVEL
1838 || (largepage
&& iterator
.level
== PT_DIRECTORY_LEVEL
)) {
1839 mmu_set_spte(vcpu
, iterator
.sptep
, ACC_ALL
, ACC_ALL
,
1840 0, write
, 1, &pt_write
,
1841 largepage
, gfn
, pfn
, false);
1842 ++vcpu
->stat
.pf_fixed
;
1846 if (*iterator
.sptep
== shadow_trap_nonpresent_pte
) {
1847 pseudo_gfn
= (iterator
.addr
& PT64_DIR_BASE_ADDR_MASK
) >> PAGE_SHIFT
;
1848 sp
= kvm_mmu_get_page(vcpu
, pseudo_gfn
, iterator
.addr
,
1850 1, ACC_ALL
, iterator
.sptep
);
1852 pgprintk("nonpaging_map: ENOMEM\n");
1853 kvm_release_pfn_clean(pfn
);
1857 set_shadow_pte(iterator
.sptep
,
1859 | PT_PRESENT_MASK
| PT_WRITABLE_MASK
1860 | shadow_user_mask
| shadow_x_mask
);
1866 static int nonpaging_map(struct kvm_vcpu
*vcpu
, gva_t v
, int write
, gfn_t gfn
)
1871 unsigned long mmu_seq
;
1873 if (is_largepage_backed(vcpu
, gfn
& ~(KVM_PAGES_PER_HPAGE
-1))) {
1874 gfn
&= ~(KVM_PAGES_PER_HPAGE
-1);
1878 mmu_seq
= vcpu
->kvm
->mmu_notifier_seq
;
1880 pfn
= gfn_to_pfn(vcpu
->kvm
, gfn
);
1883 if (is_error_pfn(pfn
)) {
1884 kvm_release_pfn_clean(pfn
);
1888 spin_lock(&vcpu
->kvm
->mmu_lock
);
1889 if (mmu_notifier_retry(vcpu
, mmu_seq
))
1891 kvm_mmu_free_some_pages(vcpu
);
1892 r
= __direct_map(vcpu
, v
, write
, largepage
, gfn
, pfn
);
1893 spin_unlock(&vcpu
->kvm
->mmu_lock
);
1899 spin_unlock(&vcpu
->kvm
->mmu_lock
);
1900 kvm_release_pfn_clean(pfn
);
1905 static void mmu_free_roots(struct kvm_vcpu
*vcpu
)
1908 struct kvm_mmu_page
*sp
;
1910 if (!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
))
1912 spin_lock(&vcpu
->kvm
->mmu_lock
);
1913 if (vcpu
->arch
.mmu
.shadow_root_level
== PT64_ROOT_LEVEL
) {
1914 hpa_t root
= vcpu
->arch
.mmu
.root_hpa
;
1916 sp
= page_header(root
);
1918 if (!sp
->root_count
&& sp
->role
.invalid
)
1919 kvm_mmu_zap_page(vcpu
->kvm
, sp
);
1920 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
1921 spin_unlock(&vcpu
->kvm
->mmu_lock
);
1924 for (i
= 0; i
< 4; ++i
) {
1925 hpa_t root
= vcpu
->arch
.mmu
.pae_root
[i
];
1928 root
&= PT64_BASE_ADDR_MASK
;
1929 sp
= page_header(root
);
1931 if (!sp
->root_count
&& sp
->role
.invalid
)
1932 kvm_mmu_zap_page(vcpu
->kvm
, sp
);
1934 vcpu
->arch
.mmu
.pae_root
[i
] = INVALID_PAGE
;
1936 spin_unlock(&vcpu
->kvm
->mmu_lock
);
1937 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
1940 static int mmu_check_root(struct kvm_vcpu
*vcpu
, gfn_t root_gfn
)
1944 if (!kvm_is_visible_gfn(vcpu
->kvm
, root_gfn
)) {
1945 set_bit(KVM_REQ_TRIPLE_FAULT
, &vcpu
->requests
);
1952 static int mmu_alloc_roots(struct kvm_vcpu
*vcpu
)
1956 struct kvm_mmu_page
*sp
;
1960 root_gfn
= vcpu
->arch
.cr3
>> PAGE_SHIFT
;
1962 if (vcpu
->arch
.mmu
.shadow_root_level
== PT64_ROOT_LEVEL
) {
1963 hpa_t root
= vcpu
->arch
.mmu
.root_hpa
;
1965 ASSERT(!VALID_PAGE(root
));
1968 if (mmu_check_root(vcpu
, root_gfn
))
1970 sp
= kvm_mmu_get_page(vcpu
, root_gfn
, 0,
1971 PT64_ROOT_LEVEL
, direct
,
1973 root
= __pa(sp
->spt
);
1975 vcpu
->arch
.mmu
.root_hpa
= root
;
1978 direct
= !is_paging(vcpu
);
1981 for (i
= 0; i
< 4; ++i
) {
1982 hpa_t root
= vcpu
->arch
.mmu
.pae_root
[i
];
1984 ASSERT(!VALID_PAGE(root
));
1985 if (vcpu
->arch
.mmu
.root_level
== PT32E_ROOT_LEVEL
) {
1986 pdptr
= kvm_pdptr_read(vcpu
, i
);
1987 if (!is_present_pte(pdptr
)) {
1988 vcpu
->arch
.mmu
.pae_root
[i
] = 0;
1991 root_gfn
= pdptr
>> PAGE_SHIFT
;
1992 } else if (vcpu
->arch
.mmu
.root_level
== 0)
1994 if (mmu_check_root(vcpu
, root_gfn
))
1996 sp
= kvm_mmu_get_page(vcpu
, root_gfn
, i
<< 30,
1997 PT32_ROOT_LEVEL
, direct
,
1999 root
= __pa(sp
->spt
);
2001 vcpu
->arch
.mmu
.pae_root
[i
] = root
| PT_PRESENT_MASK
;
2003 vcpu
->arch
.mmu
.root_hpa
= __pa(vcpu
->arch
.mmu
.pae_root
);
2007 static void mmu_sync_roots(struct kvm_vcpu
*vcpu
)
2010 struct kvm_mmu_page
*sp
;
2012 if (!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
))
2014 if (vcpu
->arch
.mmu
.shadow_root_level
== PT64_ROOT_LEVEL
) {
2015 hpa_t root
= vcpu
->arch
.mmu
.root_hpa
;
2016 sp
= page_header(root
);
2017 mmu_sync_children(vcpu
, sp
);
2020 for (i
= 0; i
< 4; ++i
) {
2021 hpa_t root
= vcpu
->arch
.mmu
.pae_root
[i
];
2023 if (root
&& VALID_PAGE(root
)) {
2024 root
&= PT64_BASE_ADDR_MASK
;
2025 sp
= page_header(root
);
2026 mmu_sync_children(vcpu
, sp
);
2031 void kvm_mmu_sync_roots(struct kvm_vcpu
*vcpu
)
2033 spin_lock(&vcpu
->kvm
->mmu_lock
);
2034 mmu_sync_roots(vcpu
);
2035 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2038 static gpa_t
nonpaging_gva_to_gpa(struct kvm_vcpu
*vcpu
, gva_t vaddr
)
2043 static int nonpaging_page_fault(struct kvm_vcpu
*vcpu
, gva_t gva
,
2049 pgprintk("%s: gva %lx error %x\n", __func__
, gva
, error_code
);
2050 r
= mmu_topup_memory_caches(vcpu
);
2055 ASSERT(VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
2057 gfn
= gva
>> PAGE_SHIFT
;
2059 return nonpaging_map(vcpu
, gva
& PAGE_MASK
,
2060 error_code
& PFERR_WRITE_MASK
, gfn
);
2063 static int tdp_page_fault(struct kvm_vcpu
*vcpu
, gva_t gpa
,
2069 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2070 unsigned long mmu_seq
;
2073 ASSERT(VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
2075 r
= mmu_topup_memory_caches(vcpu
);
2079 if (is_largepage_backed(vcpu
, gfn
& ~(KVM_PAGES_PER_HPAGE
-1))) {
2080 gfn
&= ~(KVM_PAGES_PER_HPAGE
-1);
2083 mmu_seq
= vcpu
->kvm
->mmu_notifier_seq
;
2085 pfn
= gfn_to_pfn(vcpu
->kvm
, gfn
);
2086 if (is_error_pfn(pfn
)) {
2087 kvm_release_pfn_clean(pfn
);
2090 spin_lock(&vcpu
->kvm
->mmu_lock
);
2091 if (mmu_notifier_retry(vcpu
, mmu_seq
))
2093 kvm_mmu_free_some_pages(vcpu
);
2094 r
= __direct_map(vcpu
, gpa
, error_code
& PFERR_WRITE_MASK
,
2095 largepage
, gfn
, pfn
);
2096 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2101 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2102 kvm_release_pfn_clean(pfn
);
2106 static void nonpaging_free(struct kvm_vcpu
*vcpu
)
2108 mmu_free_roots(vcpu
);
2111 static int nonpaging_init_context(struct kvm_vcpu
*vcpu
)
2113 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
2115 context
->new_cr3
= nonpaging_new_cr3
;
2116 context
->page_fault
= nonpaging_page_fault
;
2117 context
->gva_to_gpa
= nonpaging_gva_to_gpa
;
2118 context
->free
= nonpaging_free
;
2119 context
->prefetch_page
= nonpaging_prefetch_page
;
2120 context
->sync_page
= nonpaging_sync_page
;
2121 context
->invlpg
= nonpaging_invlpg
;
2122 context
->root_level
= 0;
2123 context
->shadow_root_level
= PT32E_ROOT_LEVEL
;
2124 context
->root_hpa
= INVALID_PAGE
;
2128 void kvm_mmu_flush_tlb(struct kvm_vcpu
*vcpu
)
2130 ++vcpu
->stat
.tlb_flush
;
2131 kvm_x86_ops
->tlb_flush(vcpu
);
2134 static void paging_new_cr3(struct kvm_vcpu
*vcpu
)
2136 pgprintk("%s: cr3 %lx\n", __func__
, vcpu
->arch
.cr3
);
2137 mmu_free_roots(vcpu
);
2140 static void inject_page_fault(struct kvm_vcpu
*vcpu
,
2144 kvm_inject_page_fault(vcpu
, addr
, err_code
);
2147 static void paging_free(struct kvm_vcpu
*vcpu
)
2149 nonpaging_free(vcpu
);
2152 static bool is_rsvd_bits_set(struct kvm_vcpu
*vcpu
, u64 gpte
, int level
)
2156 bit7
= (gpte
>> 7) & 1;
2157 return (gpte
& vcpu
->arch
.mmu
.rsvd_bits_mask
[bit7
][level
-1]) != 0;
2161 #include "paging_tmpl.h"
2165 #include "paging_tmpl.h"
2168 static void reset_rsvds_bits_mask(struct kvm_vcpu
*vcpu
, int level
)
2170 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
2171 int maxphyaddr
= cpuid_maxphyaddr(vcpu
);
2172 u64 exb_bit_rsvd
= 0;
2175 exb_bit_rsvd
= rsvd_bits(63, 63);
2177 case PT32_ROOT_LEVEL
:
2178 /* no rsvd bits for 2 level 4K page table entries */
2179 context
->rsvd_bits_mask
[0][1] = 0;
2180 context
->rsvd_bits_mask
[0][0] = 0;
2181 if (is_cpuid_PSE36())
2182 /* 36bits PSE 4MB page */
2183 context
->rsvd_bits_mask
[1][1] = rsvd_bits(17, 21);
2185 /* 32 bits PSE 4MB page */
2186 context
->rsvd_bits_mask
[1][1] = rsvd_bits(13, 21);
2187 context
->rsvd_bits_mask
[1][0] = context
->rsvd_bits_mask
[1][0];
2189 case PT32E_ROOT_LEVEL
:
2190 context
->rsvd_bits_mask
[0][2] =
2191 rsvd_bits(maxphyaddr
, 63) |
2192 rsvd_bits(7, 8) | rsvd_bits(1, 2); /* PDPTE */
2193 context
->rsvd_bits_mask
[0][1] = exb_bit_rsvd
|
2194 rsvd_bits(maxphyaddr
, 62); /* PDE */
2195 context
->rsvd_bits_mask
[0][0] = exb_bit_rsvd
|
2196 rsvd_bits(maxphyaddr
, 62); /* PTE */
2197 context
->rsvd_bits_mask
[1][1] = exb_bit_rsvd
|
2198 rsvd_bits(maxphyaddr
, 62) |
2199 rsvd_bits(13, 20); /* large page */
2200 context
->rsvd_bits_mask
[1][0] = context
->rsvd_bits_mask
[1][0];
2202 case PT64_ROOT_LEVEL
:
2203 context
->rsvd_bits_mask
[0][3] = exb_bit_rsvd
|
2204 rsvd_bits(maxphyaddr
, 51) | rsvd_bits(7, 8);
2205 context
->rsvd_bits_mask
[0][2] = exb_bit_rsvd
|
2206 rsvd_bits(maxphyaddr
, 51) | rsvd_bits(7, 8);
2207 context
->rsvd_bits_mask
[0][1] = exb_bit_rsvd
|
2208 rsvd_bits(maxphyaddr
, 51);
2209 context
->rsvd_bits_mask
[0][0] = exb_bit_rsvd
|
2210 rsvd_bits(maxphyaddr
, 51);
2211 context
->rsvd_bits_mask
[1][3] = context
->rsvd_bits_mask
[0][3];
2212 context
->rsvd_bits_mask
[1][2] = context
->rsvd_bits_mask
[0][2];
2213 context
->rsvd_bits_mask
[1][1] = exb_bit_rsvd
|
2214 rsvd_bits(maxphyaddr
, 51) |
2215 rsvd_bits(13, 20); /* large page */
2216 context
->rsvd_bits_mask
[1][0] = context
->rsvd_bits_mask
[1][0];
2221 static int paging64_init_context_common(struct kvm_vcpu
*vcpu
, int level
)
2223 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
2225 ASSERT(is_pae(vcpu
));
2226 context
->new_cr3
= paging_new_cr3
;
2227 context
->page_fault
= paging64_page_fault
;
2228 context
->gva_to_gpa
= paging64_gva_to_gpa
;
2229 context
->prefetch_page
= paging64_prefetch_page
;
2230 context
->sync_page
= paging64_sync_page
;
2231 context
->invlpg
= paging64_invlpg
;
2232 context
->free
= paging_free
;
2233 context
->root_level
= level
;
2234 context
->shadow_root_level
= level
;
2235 context
->root_hpa
= INVALID_PAGE
;
2239 static int paging64_init_context(struct kvm_vcpu
*vcpu
)
2241 reset_rsvds_bits_mask(vcpu
, PT64_ROOT_LEVEL
);
2242 return paging64_init_context_common(vcpu
, PT64_ROOT_LEVEL
);
2245 static int paging32_init_context(struct kvm_vcpu
*vcpu
)
2247 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
2249 reset_rsvds_bits_mask(vcpu
, PT32_ROOT_LEVEL
);
2250 context
->new_cr3
= paging_new_cr3
;
2251 context
->page_fault
= paging32_page_fault
;
2252 context
->gva_to_gpa
= paging32_gva_to_gpa
;
2253 context
->free
= paging_free
;
2254 context
->prefetch_page
= paging32_prefetch_page
;
2255 context
->sync_page
= paging32_sync_page
;
2256 context
->invlpg
= paging32_invlpg
;
2257 context
->root_level
= PT32_ROOT_LEVEL
;
2258 context
->shadow_root_level
= PT32E_ROOT_LEVEL
;
2259 context
->root_hpa
= INVALID_PAGE
;
2263 static int paging32E_init_context(struct kvm_vcpu
*vcpu
)
2265 reset_rsvds_bits_mask(vcpu
, PT32E_ROOT_LEVEL
);
2266 return paging64_init_context_common(vcpu
, PT32E_ROOT_LEVEL
);
2269 static int init_kvm_tdp_mmu(struct kvm_vcpu
*vcpu
)
2271 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
2273 context
->new_cr3
= nonpaging_new_cr3
;
2274 context
->page_fault
= tdp_page_fault
;
2275 context
->free
= nonpaging_free
;
2276 context
->prefetch_page
= nonpaging_prefetch_page
;
2277 context
->sync_page
= nonpaging_sync_page
;
2278 context
->invlpg
= nonpaging_invlpg
;
2279 context
->shadow_root_level
= kvm_x86_ops
->get_tdp_level();
2280 context
->root_hpa
= INVALID_PAGE
;
2282 if (!is_paging(vcpu
)) {
2283 context
->gva_to_gpa
= nonpaging_gva_to_gpa
;
2284 context
->root_level
= 0;
2285 } else if (is_long_mode(vcpu
)) {
2286 reset_rsvds_bits_mask(vcpu
, PT64_ROOT_LEVEL
);
2287 context
->gva_to_gpa
= paging64_gva_to_gpa
;
2288 context
->root_level
= PT64_ROOT_LEVEL
;
2289 } else if (is_pae(vcpu
)) {
2290 reset_rsvds_bits_mask(vcpu
, PT32E_ROOT_LEVEL
);
2291 context
->gva_to_gpa
= paging64_gva_to_gpa
;
2292 context
->root_level
= PT32E_ROOT_LEVEL
;
2294 reset_rsvds_bits_mask(vcpu
, PT32_ROOT_LEVEL
);
2295 context
->gva_to_gpa
= paging32_gva_to_gpa
;
2296 context
->root_level
= PT32_ROOT_LEVEL
;
2302 static int init_kvm_softmmu(struct kvm_vcpu
*vcpu
)
2307 ASSERT(!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
2309 if (!is_paging(vcpu
))
2310 r
= nonpaging_init_context(vcpu
);
2311 else if (is_long_mode(vcpu
))
2312 r
= paging64_init_context(vcpu
);
2313 else if (is_pae(vcpu
))
2314 r
= paging32E_init_context(vcpu
);
2316 r
= paging32_init_context(vcpu
);
2318 vcpu
->arch
.mmu
.base_role
.glevels
= vcpu
->arch
.mmu
.root_level
;
2323 static int init_kvm_mmu(struct kvm_vcpu
*vcpu
)
2325 vcpu
->arch
.update_pte
.pfn
= bad_pfn
;
2328 return init_kvm_tdp_mmu(vcpu
);
2330 return init_kvm_softmmu(vcpu
);
2333 static void destroy_kvm_mmu(struct kvm_vcpu
*vcpu
)
2336 if (VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
)) {
2337 vcpu
->arch
.mmu
.free(vcpu
);
2338 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
2342 int kvm_mmu_reset_context(struct kvm_vcpu
*vcpu
)
2344 destroy_kvm_mmu(vcpu
);
2345 return init_kvm_mmu(vcpu
);
2347 EXPORT_SYMBOL_GPL(kvm_mmu_reset_context
);
2349 int kvm_mmu_load(struct kvm_vcpu
*vcpu
)
2353 r
= mmu_topup_memory_caches(vcpu
);
2356 spin_lock(&vcpu
->kvm
->mmu_lock
);
2357 kvm_mmu_free_some_pages(vcpu
);
2358 r
= mmu_alloc_roots(vcpu
);
2359 mmu_sync_roots(vcpu
);
2360 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2363 kvm_x86_ops
->set_cr3(vcpu
, vcpu
->arch
.mmu
.root_hpa
);
2364 kvm_mmu_flush_tlb(vcpu
);
2368 EXPORT_SYMBOL_GPL(kvm_mmu_load
);
2370 void kvm_mmu_unload(struct kvm_vcpu
*vcpu
)
2372 mmu_free_roots(vcpu
);
2375 static void mmu_pte_write_zap_pte(struct kvm_vcpu
*vcpu
,
2376 struct kvm_mmu_page
*sp
,
2380 struct kvm_mmu_page
*child
;
2383 if (is_shadow_present_pte(pte
)) {
2384 if (sp
->role
.level
== PT_PAGE_TABLE_LEVEL
||
2386 rmap_remove(vcpu
->kvm
, spte
);
2388 child
= page_header(pte
& PT64_BASE_ADDR_MASK
);
2389 mmu_page_remove_parent_pte(child
, spte
);
2392 set_shadow_pte(spte
, shadow_trap_nonpresent_pte
);
2393 if (is_large_pte(pte
))
2394 --vcpu
->kvm
->stat
.lpages
;
2397 static void mmu_pte_write_new_pte(struct kvm_vcpu
*vcpu
,
2398 struct kvm_mmu_page
*sp
,
2402 if (sp
->role
.level
!= PT_PAGE_TABLE_LEVEL
) {
2403 if (!vcpu
->arch
.update_pte
.largepage
||
2404 sp
->role
.glevels
== PT32_ROOT_LEVEL
) {
2405 ++vcpu
->kvm
->stat
.mmu_pde_zapped
;
2410 ++vcpu
->kvm
->stat
.mmu_pte_updated
;
2411 if (sp
->role
.glevels
== PT32_ROOT_LEVEL
)
2412 paging32_update_pte(vcpu
, sp
, spte
, new);
2414 paging64_update_pte(vcpu
, sp
, spte
, new);
2417 static bool need_remote_flush(u64 old
, u64
new)
2419 if (!is_shadow_present_pte(old
))
2421 if (!is_shadow_present_pte(new))
2423 if ((old
^ new) & PT64_BASE_ADDR_MASK
)
2425 old
^= PT64_NX_MASK
;
2426 new ^= PT64_NX_MASK
;
2427 return (old
& ~new & PT64_PERM_MASK
) != 0;
2430 static void mmu_pte_write_flush_tlb(struct kvm_vcpu
*vcpu
, u64 old
, u64
new)
2432 if (need_remote_flush(old
, new))
2433 kvm_flush_remote_tlbs(vcpu
->kvm
);
2435 kvm_mmu_flush_tlb(vcpu
);
2438 static bool last_updated_pte_accessed(struct kvm_vcpu
*vcpu
)
2440 u64
*spte
= vcpu
->arch
.last_pte_updated
;
2442 return !!(spte
&& (*spte
& shadow_accessed_mask
));
2445 static void mmu_guess_page_from_pte_write(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
2446 const u8
*new, int bytes
)
2453 vcpu
->arch
.update_pte
.largepage
= 0;
2455 if (bytes
!= 4 && bytes
!= 8)
2459 * Assume that the pte write on a page table of the same type
2460 * as the current vcpu paging mode. This is nearly always true
2461 * (might be false while changing modes). Note it is verified later
2465 /* Handle a 32-bit guest writing two halves of a 64-bit gpte */
2466 if ((bytes
== 4) && (gpa
% 4 == 0)) {
2467 r
= kvm_read_guest(vcpu
->kvm
, gpa
& ~(u64
)7, &gpte
, 8);
2470 memcpy((void *)&gpte
+ (gpa
% 8), new, 4);
2471 } else if ((bytes
== 8) && (gpa
% 8 == 0)) {
2472 memcpy((void *)&gpte
, new, 8);
2475 if ((bytes
== 4) && (gpa
% 4 == 0))
2476 memcpy((void *)&gpte
, new, 4);
2478 if (!is_present_pte(gpte
))
2480 gfn
= (gpte
& PT64_BASE_ADDR_MASK
) >> PAGE_SHIFT
;
2482 if (is_large_pte(gpte
) && is_largepage_backed(vcpu
, gfn
)) {
2483 gfn
&= ~(KVM_PAGES_PER_HPAGE
-1);
2484 vcpu
->arch
.update_pte
.largepage
= 1;
2486 vcpu
->arch
.update_pte
.mmu_seq
= vcpu
->kvm
->mmu_notifier_seq
;
2488 pfn
= gfn_to_pfn(vcpu
->kvm
, gfn
);
2490 if (is_error_pfn(pfn
)) {
2491 kvm_release_pfn_clean(pfn
);
2494 vcpu
->arch
.update_pte
.gfn
= gfn
;
2495 vcpu
->arch
.update_pte
.pfn
= pfn
;
2498 static void kvm_mmu_access_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2500 u64
*spte
= vcpu
->arch
.last_pte_updated
;
2503 && vcpu
->arch
.last_pte_gfn
== gfn
2504 && shadow_accessed_mask
2505 && !(*spte
& shadow_accessed_mask
)
2506 && is_shadow_present_pte(*spte
))
2507 set_bit(PT_ACCESSED_SHIFT
, (unsigned long *)spte
);
2510 void kvm_mmu_pte_write(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
2511 const u8
*new, int bytes
,
2512 bool guest_initiated
)
2514 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2515 struct kvm_mmu_page
*sp
;
2516 struct hlist_node
*node
, *n
;
2517 struct hlist_head
*bucket
;
2521 unsigned offset
= offset_in_page(gpa
);
2523 unsigned page_offset
;
2524 unsigned misaligned
;
2531 pgprintk("%s: gpa %llx bytes %d\n", __func__
, gpa
, bytes
);
2532 mmu_guess_page_from_pte_write(vcpu
, gpa
, new, bytes
);
2533 spin_lock(&vcpu
->kvm
->mmu_lock
);
2534 kvm_mmu_access_page(vcpu
, gfn
);
2535 kvm_mmu_free_some_pages(vcpu
);
2536 ++vcpu
->kvm
->stat
.mmu_pte_write
;
2537 kvm_mmu_audit(vcpu
, "pre pte write");
2538 if (guest_initiated
) {
2539 if (gfn
== vcpu
->arch
.last_pt_write_gfn
2540 && !last_updated_pte_accessed(vcpu
)) {
2541 ++vcpu
->arch
.last_pt_write_count
;
2542 if (vcpu
->arch
.last_pt_write_count
>= 3)
2545 vcpu
->arch
.last_pt_write_gfn
= gfn
;
2546 vcpu
->arch
.last_pt_write_count
= 1;
2547 vcpu
->arch
.last_pte_updated
= NULL
;
2550 index
= kvm_page_table_hashfn(gfn
);
2551 bucket
= &vcpu
->kvm
->arch
.mmu_page_hash
[index
];
2552 hlist_for_each_entry_safe(sp
, node
, n
, bucket
, hash_link
) {
2553 if (sp
->gfn
!= gfn
|| sp
->role
.direct
|| sp
->role
.invalid
)
2555 pte_size
= sp
->role
.glevels
== PT32_ROOT_LEVEL
? 4 : 8;
2556 misaligned
= (offset
^ (offset
+ bytes
- 1)) & ~(pte_size
- 1);
2557 misaligned
|= bytes
< 4;
2558 if (misaligned
|| flooded
) {
2560 * Misaligned accesses are too much trouble to fix
2561 * up; also, they usually indicate a page is not used
2564 * If we're seeing too many writes to a page,
2565 * it may no longer be a page table, or we may be
2566 * forking, in which case it is better to unmap the
2569 pgprintk("misaligned: gpa %llx bytes %d role %x\n",
2570 gpa
, bytes
, sp
->role
.word
);
2571 if (kvm_mmu_zap_page(vcpu
->kvm
, sp
))
2573 ++vcpu
->kvm
->stat
.mmu_flooded
;
2576 page_offset
= offset
;
2577 level
= sp
->role
.level
;
2579 if (sp
->role
.glevels
== PT32_ROOT_LEVEL
) {
2580 page_offset
<<= 1; /* 32->64 */
2582 * A 32-bit pde maps 4MB while the shadow pdes map
2583 * only 2MB. So we need to double the offset again
2584 * and zap two pdes instead of one.
2586 if (level
== PT32_ROOT_LEVEL
) {
2587 page_offset
&= ~7; /* kill rounding error */
2591 quadrant
= page_offset
>> PAGE_SHIFT
;
2592 page_offset
&= ~PAGE_MASK
;
2593 if (quadrant
!= sp
->role
.quadrant
)
2596 spte
= &sp
->spt
[page_offset
/ sizeof(*spte
)];
2597 if ((gpa
& (pte_size
- 1)) || (bytes
< pte_size
)) {
2599 r
= kvm_read_guest_atomic(vcpu
->kvm
,
2600 gpa
& ~(u64
)(pte_size
- 1),
2602 new = (const void *)&gentry
;
2608 mmu_pte_write_zap_pte(vcpu
, sp
, spte
);
2610 mmu_pte_write_new_pte(vcpu
, sp
, spte
, new);
2611 mmu_pte_write_flush_tlb(vcpu
, entry
, *spte
);
2615 kvm_mmu_audit(vcpu
, "post pte write");
2616 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2617 if (!is_error_pfn(vcpu
->arch
.update_pte
.pfn
)) {
2618 kvm_release_pfn_clean(vcpu
->arch
.update_pte
.pfn
);
2619 vcpu
->arch
.update_pte
.pfn
= bad_pfn
;
2623 int kvm_mmu_unprotect_page_virt(struct kvm_vcpu
*vcpu
, gva_t gva
)
2628 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, gva
);
2630 spin_lock(&vcpu
->kvm
->mmu_lock
);
2631 r
= kvm_mmu_unprotect_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
2632 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2635 EXPORT_SYMBOL_GPL(kvm_mmu_unprotect_page_virt
);
2637 void __kvm_mmu_free_some_pages(struct kvm_vcpu
*vcpu
)
2639 while (vcpu
->kvm
->arch
.n_free_mmu_pages
< KVM_REFILL_PAGES
) {
2640 struct kvm_mmu_page
*sp
;
2642 sp
= container_of(vcpu
->kvm
->arch
.active_mmu_pages
.prev
,
2643 struct kvm_mmu_page
, link
);
2644 kvm_mmu_zap_page(vcpu
->kvm
, sp
);
2645 ++vcpu
->kvm
->stat
.mmu_recycled
;
2649 int kvm_mmu_page_fault(struct kvm_vcpu
*vcpu
, gva_t cr2
, u32 error_code
)
2652 enum emulation_result er
;
2654 r
= vcpu
->arch
.mmu
.page_fault(vcpu
, cr2
, error_code
);
2663 r
= mmu_topup_memory_caches(vcpu
);
2667 er
= emulate_instruction(vcpu
, vcpu
->run
, cr2
, error_code
, 0);
2672 case EMULATE_DO_MMIO
:
2673 ++vcpu
->stat
.mmio_exits
;
2676 kvm_report_emulation_failure(vcpu
, "pagetable");
2684 EXPORT_SYMBOL_GPL(kvm_mmu_page_fault
);
2686 void kvm_mmu_invlpg(struct kvm_vcpu
*vcpu
, gva_t gva
)
2688 vcpu
->arch
.mmu
.invlpg(vcpu
, gva
);
2689 kvm_mmu_flush_tlb(vcpu
);
2690 ++vcpu
->stat
.invlpg
;
2692 EXPORT_SYMBOL_GPL(kvm_mmu_invlpg
);
2694 void kvm_enable_tdp(void)
2698 EXPORT_SYMBOL_GPL(kvm_enable_tdp
);
2700 void kvm_disable_tdp(void)
2702 tdp_enabled
= false;
2704 EXPORT_SYMBOL_GPL(kvm_disable_tdp
);
2706 static void free_mmu_pages(struct kvm_vcpu
*vcpu
)
2708 free_page((unsigned long)vcpu
->arch
.mmu
.pae_root
);
2711 static int alloc_mmu_pages(struct kvm_vcpu
*vcpu
)
2718 if (vcpu
->kvm
->arch
.n_requested_mmu_pages
)
2719 vcpu
->kvm
->arch
.n_free_mmu_pages
=
2720 vcpu
->kvm
->arch
.n_requested_mmu_pages
;
2722 vcpu
->kvm
->arch
.n_free_mmu_pages
=
2723 vcpu
->kvm
->arch
.n_alloc_mmu_pages
;
2725 * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64.
2726 * Therefore we need to allocate shadow page tables in the first
2727 * 4GB of memory, which happens to fit the DMA32 zone.
2729 page
= alloc_page(GFP_KERNEL
| __GFP_DMA32
);
2732 vcpu
->arch
.mmu
.pae_root
= page_address(page
);
2733 for (i
= 0; i
< 4; ++i
)
2734 vcpu
->arch
.mmu
.pae_root
[i
] = INVALID_PAGE
;
2739 free_mmu_pages(vcpu
);
2743 int kvm_mmu_create(struct kvm_vcpu
*vcpu
)
2746 ASSERT(!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
2748 return alloc_mmu_pages(vcpu
);
2751 int kvm_mmu_setup(struct kvm_vcpu
*vcpu
)
2754 ASSERT(!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
2756 return init_kvm_mmu(vcpu
);
2759 void kvm_mmu_destroy(struct kvm_vcpu
*vcpu
)
2763 destroy_kvm_mmu(vcpu
);
2764 free_mmu_pages(vcpu
);
2765 mmu_free_memory_caches(vcpu
);
2768 void kvm_mmu_slot_remove_write_access(struct kvm
*kvm
, int slot
)
2770 struct kvm_mmu_page
*sp
;
2772 list_for_each_entry(sp
, &kvm
->arch
.active_mmu_pages
, link
) {
2776 if (!test_bit(slot
, sp
->slot_bitmap
))
2780 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
)
2782 if (pt
[i
] & PT_WRITABLE_MASK
)
2783 pt
[i
] &= ~PT_WRITABLE_MASK
;
2785 kvm_flush_remote_tlbs(kvm
);
2788 void kvm_mmu_zap_all(struct kvm
*kvm
)
2790 struct kvm_mmu_page
*sp
, *node
;
2792 spin_lock(&kvm
->mmu_lock
);
2793 list_for_each_entry_safe(sp
, node
, &kvm
->arch
.active_mmu_pages
, link
)
2794 if (kvm_mmu_zap_page(kvm
, sp
))
2795 node
= container_of(kvm
->arch
.active_mmu_pages
.next
,
2796 struct kvm_mmu_page
, link
);
2797 spin_unlock(&kvm
->mmu_lock
);
2799 kvm_flush_remote_tlbs(kvm
);
2802 static void kvm_mmu_remove_one_alloc_mmu_page(struct kvm
*kvm
)
2804 struct kvm_mmu_page
*page
;
2806 page
= container_of(kvm
->arch
.active_mmu_pages
.prev
,
2807 struct kvm_mmu_page
, link
);
2808 kvm_mmu_zap_page(kvm
, page
);
2811 static int mmu_shrink(int nr_to_scan
, gfp_t gfp_mask
)
2814 struct kvm
*kvm_freed
= NULL
;
2815 int cache_count
= 0;
2817 spin_lock(&kvm_lock
);
2819 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
2822 if (!down_read_trylock(&kvm
->slots_lock
))
2824 spin_lock(&kvm
->mmu_lock
);
2825 npages
= kvm
->arch
.n_alloc_mmu_pages
-
2826 kvm
->arch
.n_free_mmu_pages
;
2827 cache_count
+= npages
;
2828 if (!kvm_freed
&& nr_to_scan
> 0 && npages
> 0) {
2829 kvm_mmu_remove_one_alloc_mmu_page(kvm
);
2835 spin_unlock(&kvm
->mmu_lock
);
2836 up_read(&kvm
->slots_lock
);
2839 list_move_tail(&kvm_freed
->vm_list
, &vm_list
);
2841 spin_unlock(&kvm_lock
);
2846 static struct shrinker mmu_shrinker
= {
2847 .shrink
= mmu_shrink
,
2848 .seeks
= DEFAULT_SEEKS
* 10,
2851 static void mmu_destroy_caches(void)
2853 if (pte_chain_cache
)
2854 kmem_cache_destroy(pte_chain_cache
);
2855 if (rmap_desc_cache
)
2856 kmem_cache_destroy(rmap_desc_cache
);
2857 if (mmu_page_header_cache
)
2858 kmem_cache_destroy(mmu_page_header_cache
);
2861 void kvm_mmu_module_exit(void)
2863 mmu_destroy_caches();
2864 unregister_shrinker(&mmu_shrinker
);
2867 int kvm_mmu_module_init(void)
2869 pte_chain_cache
= kmem_cache_create("kvm_pte_chain",
2870 sizeof(struct kvm_pte_chain
),
2872 if (!pte_chain_cache
)
2874 rmap_desc_cache
= kmem_cache_create("kvm_rmap_desc",
2875 sizeof(struct kvm_rmap_desc
),
2877 if (!rmap_desc_cache
)
2880 mmu_page_header_cache
= kmem_cache_create("kvm_mmu_page_header",
2881 sizeof(struct kvm_mmu_page
),
2883 if (!mmu_page_header_cache
)
2886 register_shrinker(&mmu_shrinker
);
2891 mmu_destroy_caches();
2896 * Caculate mmu pages needed for kvm.
2898 unsigned int kvm_mmu_calculate_mmu_pages(struct kvm
*kvm
)
2901 unsigned int nr_mmu_pages
;
2902 unsigned int nr_pages
= 0;
2904 for (i
= 0; i
< kvm
->nmemslots
; i
++)
2905 nr_pages
+= kvm
->memslots
[i
].npages
;
2907 nr_mmu_pages
= nr_pages
* KVM_PERMILLE_MMU_PAGES
/ 1000;
2908 nr_mmu_pages
= max(nr_mmu_pages
,
2909 (unsigned int) KVM_MIN_ALLOC_MMU_PAGES
);
2911 return nr_mmu_pages
;
2914 static void *pv_mmu_peek_buffer(struct kvm_pv_mmu_op_buffer
*buffer
,
2917 if (len
> buffer
->len
)
2922 static void *pv_mmu_read_buffer(struct kvm_pv_mmu_op_buffer
*buffer
,
2927 ret
= pv_mmu_peek_buffer(buffer
, len
);
2932 buffer
->processed
+= len
;
2936 static int kvm_pv_mmu_write(struct kvm_vcpu
*vcpu
,
2937 gpa_t addr
, gpa_t value
)
2942 if (!is_long_mode(vcpu
) && !is_pae(vcpu
))
2945 r
= mmu_topup_memory_caches(vcpu
);
2949 if (!emulator_write_phys(vcpu
, addr
, &value
, bytes
))
2955 static int kvm_pv_mmu_flush_tlb(struct kvm_vcpu
*vcpu
)
2957 kvm_set_cr3(vcpu
, vcpu
->arch
.cr3
);
2961 static int kvm_pv_mmu_release_pt(struct kvm_vcpu
*vcpu
, gpa_t addr
)
2963 spin_lock(&vcpu
->kvm
->mmu_lock
);
2964 mmu_unshadow(vcpu
->kvm
, addr
>> PAGE_SHIFT
);
2965 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2969 static int kvm_pv_mmu_op_one(struct kvm_vcpu
*vcpu
,
2970 struct kvm_pv_mmu_op_buffer
*buffer
)
2972 struct kvm_mmu_op_header
*header
;
2974 header
= pv_mmu_peek_buffer(buffer
, sizeof *header
);
2977 switch (header
->op
) {
2978 case KVM_MMU_OP_WRITE_PTE
: {
2979 struct kvm_mmu_op_write_pte
*wpte
;
2981 wpte
= pv_mmu_read_buffer(buffer
, sizeof *wpte
);
2984 return kvm_pv_mmu_write(vcpu
, wpte
->pte_phys
,
2987 case KVM_MMU_OP_FLUSH_TLB
: {
2988 struct kvm_mmu_op_flush_tlb
*ftlb
;
2990 ftlb
= pv_mmu_read_buffer(buffer
, sizeof *ftlb
);
2993 return kvm_pv_mmu_flush_tlb(vcpu
);
2995 case KVM_MMU_OP_RELEASE_PT
: {
2996 struct kvm_mmu_op_release_pt
*rpt
;
2998 rpt
= pv_mmu_read_buffer(buffer
, sizeof *rpt
);
3001 return kvm_pv_mmu_release_pt(vcpu
, rpt
->pt_phys
);
3007 int kvm_pv_mmu_op(struct kvm_vcpu
*vcpu
, unsigned long bytes
,
3008 gpa_t addr
, unsigned long *ret
)
3011 struct kvm_pv_mmu_op_buffer
*buffer
= &vcpu
->arch
.mmu_op_buffer
;
3013 buffer
->ptr
= buffer
->buf
;
3014 buffer
->len
= min_t(unsigned long, bytes
, sizeof buffer
->buf
);
3015 buffer
->processed
= 0;
3017 r
= kvm_read_guest(vcpu
->kvm
, addr
, buffer
->buf
, buffer
->len
);
3021 while (buffer
->len
) {
3022 r
= kvm_pv_mmu_op_one(vcpu
, buffer
);
3031 *ret
= buffer
->processed
;
3037 static const char *audit_msg
;
3039 static gva_t
canonicalize(gva_t gva
)
3041 #ifdef CONFIG_X86_64
3042 gva
= (long long)(gva
<< 16) >> 16;
3047 static void audit_mappings_page(struct kvm_vcpu
*vcpu
, u64 page_pte
,
3048 gva_t va
, int level
)
3050 u64
*pt
= __va(page_pte
& PT64_BASE_ADDR_MASK
);
3052 gva_t va_delta
= 1ul << (PAGE_SHIFT
+ 9 * (level
- 1));
3054 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
, va
+= va_delta
) {
3057 if (ent
== shadow_trap_nonpresent_pte
)
3060 va
= canonicalize(va
);
3062 if (ent
== shadow_notrap_nonpresent_pte
)
3063 printk(KERN_ERR
"audit: (%s) nontrapping pte"
3064 " in nonleaf level: levels %d gva %lx"
3065 " level %d pte %llx\n", audit_msg
,
3066 vcpu
->arch
.mmu
.root_level
, va
, level
, ent
);
3068 audit_mappings_page(vcpu
, ent
, va
, level
- 1);
3070 gpa_t gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, va
);
3071 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3072 pfn_t pfn
= gfn_to_pfn(vcpu
->kvm
, gfn
);
3073 hpa_t hpa
= (hpa_t
)pfn
<< PAGE_SHIFT
;
3075 if (is_shadow_present_pte(ent
)
3076 && (ent
& PT64_BASE_ADDR_MASK
) != hpa
)
3077 printk(KERN_ERR
"xx audit error: (%s) levels %d"
3078 " gva %lx gpa %llx hpa %llx ent %llx %d\n",
3079 audit_msg
, vcpu
->arch
.mmu
.root_level
,
3081 is_shadow_present_pte(ent
));
3082 else if (ent
== shadow_notrap_nonpresent_pte
3083 && !is_error_hpa(hpa
))
3084 printk(KERN_ERR
"audit: (%s) notrap shadow,"
3085 " valid guest gva %lx\n", audit_msg
, va
);
3086 kvm_release_pfn_clean(pfn
);
3092 static void audit_mappings(struct kvm_vcpu
*vcpu
)
3096 if (vcpu
->arch
.mmu
.root_level
== 4)
3097 audit_mappings_page(vcpu
, vcpu
->arch
.mmu
.root_hpa
, 0, 4);
3099 for (i
= 0; i
< 4; ++i
)
3100 if (vcpu
->arch
.mmu
.pae_root
[i
] & PT_PRESENT_MASK
)
3101 audit_mappings_page(vcpu
,
3102 vcpu
->arch
.mmu
.pae_root
[i
],
3107 static int count_rmaps(struct kvm_vcpu
*vcpu
)
3112 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
3113 struct kvm_memory_slot
*m
= &vcpu
->kvm
->memslots
[i
];
3114 struct kvm_rmap_desc
*d
;
3116 for (j
= 0; j
< m
->npages
; ++j
) {
3117 unsigned long *rmapp
= &m
->rmap
[j
];
3121 if (!(*rmapp
& 1)) {
3125 d
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
3127 for (k
= 0; k
< RMAP_EXT
; ++k
)
3128 if (d
->shadow_ptes
[k
])
3139 static int count_writable_mappings(struct kvm_vcpu
*vcpu
)
3142 struct kvm_mmu_page
*sp
;
3145 list_for_each_entry(sp
, &vcpu
->kvm
->arch
.active_mmu_pages
, link
) {
3148 if (sp
->role
.level
!= PT_PAGE_TABLE_LEVEL
)
3151 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
3154 if (!(ent
& PT_PRESENT_MASK
))
3156 if (!(ent
& PT_WRITABLE_MASK
))
3164 static void audit_rmap(struct kvm_vcpu
*vcpu
)
3166 int n_rmap
= count_rmaps(vcpu
);
3167 int n_actual
= count_writable_mappings(vcpu
);
3169 if (n_rmap
!= n_actual
)
3170 printk(KERN_ERR
"%s: (%s) rmap %d actual %d\n",
3171 __func__
, audit_msg
, n_rmap
, n_actual
);
3174 static void audit_write_protection(struct kvm_vcpu
*vcpu
)
3176 struct kvm_mmu_page
*sp
;
3177 struct kvm_memory_slot
*slot
;
3178 unsigned long *rmapp
;
3181 list_for_each_entry(sp
, &vcpu
->kvm
->arch
.active_mmu_pages
, link
) {
3182 if (sp
->role
.direct
)
3185 gfn
= unalias_gfn(vcpu
->kvm
, sp
->gfn
);
3186 slot
= gfn_to_memslot_unaliased(vcpu
->kvm
, sp
->gfn
);
3187 rmapp
= &slot
->rmap
[gfn
- slot
->base_gfn
];
3189 printk(KERN_ERR
"%s: (%s) shadow page has writable"
3190 " mappings: gfn %lx role %x\n",
3191 __func__
, audit_msg
, sp
->gfn
,
3196 static void kvm_mmu_audit(struct kvm_vcpu
*vcpu
, const char *msg
)
3203 audit_write_protection(vcpu
);
3204 audit_mappings(vcpu
);