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.
10 * Copyright 2010 Red Hat, Inc. and/or its affilates.
13 * Yaniv Kamay <yaniv@qumranet.com>
14 * Avi Kivity <avi@qumranet.com>
16 * This work is licensed under the terms of the GNU GPL, version 2. See
17 * the COPYING file in the top-level directory.
23 #include "kvm_cache_regs.h"
25 #include <linux/kvm_host.h>
26 #include <linux/types.h>
27 #include <linux/string.h>
29 #include <linux/highmem.h>
30 #include <linux/module.h>
31 #include <linux/swap.h>
32 #include <linux/hugetlb.h>
33 #include <linux/compiler.h>
34 #include <linux/srcu.h>
35 #include <linux/slab.h>
36 #include <linux/uaccess.h>
39 #include <asm/cmpxchg.h>
44 * When setting this variable to true it enables Two-Dimensional-Paging
45 * where the hardware walks 2 page tables:
46 * 1. the guest-virtual to guest-physical
47 * 2. while doing 1. it walks guest-physical to host-physical
48 * If the hardware supports that we don't need to do shadow paging.
50 bool tdp_enabled
= false;
57 static void kvm_mmu_audit(struct kvm_vcpu
*vcpu
, const char *msg
);
59 static void kvm_mmu_audit(struct kvm_vcpu
*vcpu
, const char *msg
) {}
64 #define pgprintk(x...) do { if (dbg) printk(x); } while (0)
65 #define rmap_printk(x...) do { if (dbg) printk(x); } while (0)
69 #define pgprintk(x...) do { } while (0)
70 #define rmap_printk(x...) do { } while (0)
74 #if defined(MMU_DEBUG) || defined(AUDIT)
76 module_param(dbg
, bool, 0644);
79 static int oos_shadow
= 1;
80 module_param(oos_shadow
, bool, 0644);
83 #define ASSERT(x) do { } while (0)
87 printk(KERN_WARNING "assertion failed %s:%d: %s\n", \
88 __FILE__, __LINE__, #x); \
92 #define PT_FIRST_AVAIL_BITS_SHIFT 9
93 #define PT64_SECOND_AVAIL_BITS_SHIFT 52
95 #define VALID_PAGE(x) ((x) != INVALID_PAGE)
97 #define PT64_LEVEL_BITS 9
99 #define PT64_LEVEL_SHIFT(level) \
100 (PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS)
102 #define PT64_LEVEL_MASK(level) \
103 (((1ULL << PT64_LEVEL_BITS) - 1) << PT64_LEVEL_SHIFT(level))
105 #define PT64_INDEX(address, level)\
106 (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1))
109 #define PT32_LEVEL_BITS 10
111 #define PT32_LEVEL_SHIFT(level) \
112 (PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS)
114 #define PT32_LEVEL_MASK(level) \
115 (((1ULL << PT32_LEVEL_BITS) - 1) << PT32_LEVEL_SHIFT(level))
116 #define PT32_LVL_OFFSET_MASK(level) \
117 (PT32_BASE_ADDR_MASK & ((1ULL << (PAGE_SHIFT + (((level) - 1) \
118 * PT32_LEVEL_BITS))) - 1))
120 #define PT32_INDEX(address, level)\
121 (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1))
124 #define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1))
125 #define PT64_DIR_BASE_ADDR_MASK \
126 (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1))
127 #define PT64_LVL_ADDR_MASK(level) \
128 (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + (((level) - 1) \
129 * PT64_LEVEL_BITS))) - 1))
130 #define PT64_LVL_OFFSET_MASK(level) \
131 (PT64_BASE_ADDR_MASK & ((1ULL << (PAGE_SHIFT + (((level) - 1) \
132 * PT64_LEVEL_BITS))) - 1))
134 #define PT32_BASE_ADDR_MASK PAGE_MASK
135 #define PT32_DIR_BASE_ADDR_MASK \
136 (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1))
137 #define PT32_LVL_ADDR_MASK(level) \
138 (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + (((level) - 1) \
139 * PT32_LEVEL_BITS))) - 1))
141 #define PT64_PERM_MASK (PT_PRESENT_MASK | PT_WRITABLE_MASK | PT_USER_MASK \
146 #define ACC_EXEC_MASK 1
147 #define ACC_WRITE_MASK PT_WRITABLE_MASK
148 #define ACC_USER_MASK PT_USER_MASK
149 #define ACC_ALL (ACC_EXEC_MASK | ACC_WRITE_MASK | ACC_USER_MASK)
151 #include <trace/events/kvm.h>
153 #define CREATE_TRACE_POINTS
154 #include "mmutrace.h"
156 #define SPTE_HOST_WRITEABLE (1ULL << PT_FIRST_AVAIL_BITS_SHIFT)
158 #define SHADOW_PT_INDEX(addr, level) PT64_INDEX(addr, level)
160 struct kvm_rmap_desc
{
161 u64
*sptes
[RMAP_EXT
];
162 struct kvm_rmap_desc
*more
;
165 struct kvm_shadow_walk_iterator
{
173 #define for_each_shadow_entry(_vcpu, _addr, _walker) \
174 for (shadow_walk_init(&(_walker), _vcpu, _addr); \
175 shadow_walk_okay(&(_walker)); \
176 shadow_walk_next(&(_walker)))
178 typedef int (*mmu_parent_walk_fn
) (struct kvm_mmu_page
*sp
);
180 static struct kmem_cache
*pte_chain_cache
;
181 static struct kmem_cache
*rmap_desc_cache
;
182 static struct kmem_cache
*mmu_page_header_cache
;
184 static u64 __read_mostly shadow_trap_nonpresent_pte
;
185 static u64 __read_mostly shadow_notrap_nonpresent_pte
;
186 static u64 __read_mostly shadow_base_present_pte
;
187 static u64 __read_mostly shadow_nx_mask
;
188 static u64 __read_mostly shadow_x_mask
; /* mutual exclusive with nx_mask */
189 static u64 __read_mostly shadow_user_mask
;
190 static u64 __read_mostly shadow_accessed_mask
;
191 static u64 __read_mostly shadow_dirty_mask
;
193 static inline u64
rsvd_bits(int s
, int e
)
195 return ((1ULL << (e
- s
+ 1)) - 1) << s
;
198 void kvm_mmu_set_nonpresent_ptes(u64 trap_pte
, u64 notrap_pte
)
200 shadow_trap_nonpresent_pte
= trap_pte
;
201 shadow_notrap_nonpresent_pte
= notrap_pte
;
203 EXPORT_SYMBOL_GPL(kvm_mmu_set_nonpresent_ptes
);
205 void kvm_mmu_set_base_ptes(u64 base_pte
)
207 shadow_base_present_pte
= base_pte
;
209 EXPORT_SYMBOL_GPL(kvm_mmu_set_base_ptes
);
211 void kvm_mmu_set_mask_ptes(u64 user_mask
, u64 accessed_mask
,
212 u64 dirty_mask
, u64 nx_mask
, u64 x_mask
)
214 shadow_user_mask
= user_mask
;
215 shadow_accessed_mask
= accessed_mask
;
216 shadow_dirty_mask
= dirty_mask
;
217 shadow_nx_mask
= nx_mask
;
218 shadow_x_mask
= x_mask
;
220 EXPORT_SYMBOL_GPL(kvm_mmu_set_mask_ptes
);
222 static bool is_write_protection(struct kvm_vcpu
*vcpu
)
224 return kvm_read_cr0_bits(vcpu
, X86_CR0_WP
);
227 static int is_cpuid_PSE36(void)
232 static int is_nx(struct kvm_vcpu
*vcpu
)
234 return vcpu
->arch
.efer
& EFER_NX
;
237 static int is_shadow_present_pte(u64 pte
)
239 return pte
!= shadow_trap_nonpresent_pte
240 && pte
!= shadow_notrap_nonpresent_pte
;
243 static int is_large_pte(u64 pte
)
245 return pte
& PT_PAGE_SIZE_MASK
;
248 static int is_writable_pte(unsigned long pte
)
250 return pte
& PT_WRITABLE_MASK
;
253 static int is_dirty_gpte(unsigned long pte
)
255 return pte
& PT_DIRTY_MASK
;
258 static int is_rmap_spte(u64 pte
)
260 return is_shadow_present_pte(pte
);
263 static int is_last_spte(u64 pte
, int level
)
265 if (level
== PT_PAGE_TABLE_LEVEL
)
267 if (is_large_pte(pte
))
272 static pfn_t
spte_to_pfn(u64 pte
)
274 return (pte
& PT64_BASE_ADDR_MASK
) >> PAGE_SHIFT
;
277 static gfn_t
pse36_gfn_delta(u32 gpte
)
279 int shift
= 32 - PT32_DIR_PSE36_SHIFT
- PAGE_SHIFT
;
281 return (gpte
& PT32_DIR_PSE36_MASK
) << shift
;
284 static void __set_spte(u64
*sptep
, u64 spte
)
287 set_64bit((unsigned long *)sptep
, spte
);
289 set_64bit((unsigned long long *)sptep
, spte
);
293 static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache
*cache
,
294 struct kmem_cache
*base_cache
, int min
)
298 if (cache
->nobjs
>= min
)
300 while (cache
->nobjs
< ARRAY_SIZE(cache
->objects
)) {
301 obj
= kmem_cache_zalloc(base_cache
, GFP_KERNEL
);
304 cache
->objects
[cache
->nobjs
++] = obj
;
309 static void mmu_free_memory_cache(struct kvm_mmu_memory_cache
*mc
,
310 struct kmem_cache
*cache
)
313 kmem_cache_free(cache
, mc
->objects
[--mc
->nobjs
]);
316 static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache
*cache
,
321 if (cache
->nobjs
>= min
)
323 while (cache
->nobjs
< ARRAY_SIZE(cache
->objects
)) {
324 page
= alloc_page(GFP_KERNEL
);
327 cache
->objects
[cache
->nobjs
++] = page_address(page
);
332 static void mmu_free_memory_cache_page(struct kvm_mmu_memory_cache
*mc
)
335 free_page((unsigned long)mc
->objects
[--mc
->nobjs
]);
338 static int mmu_topup_memory_caches(struct kvm_vcpu
*vcpu
)
342 r
= mmu_topup_memory_cache(&vcpu
->arch
.mmu_pte_chain_cache
,
346 r
= mmu_topup_memory_cache(&vcpu
->arch
.mmu_rmap_desc_cache
,
350 r
= mmu_topup_memory_cache_page(&vcpu
->arch
.mmu_page_cache
, 8);
353 r
= mmu_topup_memory_cache(&vcpu
->arch
.mmu_page_header_cache
,
354 mmu_page_header_cache
, 4);
359 static void mmu_free_memory_caches(struct kvm_vcpu
*vcpu
)
361 mmu_free_memory_cache(&vcpu
->arch
.mmu_pte_chain_cache
, pte_chain_cache
);
362 mmu_free_memory_cache(&vcpu
->arch
.mmu_rmap_desc_cache
, rmap_desc_cache
);
363 mmu_free_memory_cache_page(&vcpu
->arch
.mmu_page_cache
);
364 mmu_free_memory_cache(&vcpu
->arch
.mmu_page_header_cache
,
365 mmu_page_header_cache
);
368 static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache
*mc
,
374 p
= mc
->objects
[--mc
->nobjs
];
378 static struct kvm_pte_chain
*mmu_alloc_pte_chain(struct kvm_vcpu
*vcpu
)
380 return mmu_memory_cache_alloc(&vcpu
->arch
.mmu_pte_chain_cache
,
381 sizeof(struct kvm_pte_chain
));
384 static void mmu_free_pte_chain(struct kvm_pte_chain
*pc
)
386 kmem_cache_free(pte_chain_cache
, pc
);
389 static struct kvm_rmap_desc
*mmu_alloc_rmap_desc(struct kvm_vcpu
*vcpu
)
391 return mmu_memory_cache_alloc(&vcpu
->arch
.mmu_rmap_desc_cache
,
392 sizeof(struct kvm_rmap_desc
));
395 static void mmu_free_rmap_desc(struct kvm_rmap_desc
*rd
)
397 kmem_cache_free(rmap_desc_cache
, rd
);
401 * Return the pointer to the largepage write count for a given
402 * gfn, handling slots that are not large page aligned.
404 static int *slot_largepage_idx(gfn_t gfn
,
405 struct kvm_memory_slot
*slot
,
410 idx
= (gfn
/ KVM_PAGES_PER_HPAGE(level
)) -
411 (slot
->base_gfn
/ KVM_PAGES_PER_HPAGE(level
));
412 return &slot
->lpage_info
[level
- 2][idx
].write_count
;
415 static void account_shadowed(struct kvm
*kvm
, gfn_t gfn
)
417 struct kvm_memory_slot
*slot
;
421 gfn
= unalias_gfn(kvm
, gfn
);
423 slot
= gfn_to_memslot_unaliased(kvm
, gfn
);
424 for (i
= PT_DIRECTORY_LEVEL
;
425 i
< PT_PAGE_TABLE_LEVEL
+ KVM_NR_PAGE_SIZES
; ++i
) {
426 write_count
= slot_largepage_idx(gfn
, slot
, i
);
431 static void unaccount_shadowed(struct kvm
*kvm
, gfn_t gfn
)
433 struct kvm_memory_slot
*slot
;
437 gfn
= unalias_gfn(kvm
, gfn
);
438 slot
= gfn_to_memslot_unaliased(kvm
, gfn
);
439 for (i
= PT_DIRECTORY_LEVEL
;
440 i
< PT_PAGE_TABLE_LEVEL
+ KVM_NR_PAGE_SIZES
; ++i
) {
441 write_count
= slot_largepage_idx(gfn
, slot
, i
);
443 WARN_ON(*write_count
< 0);
447 static int has_wrprotected_page(struct kvm
*kvm
,
451 struct kvm_memory_slot
*slot
;
454 gfn
= unalias_gfn(kvm
, gfn
);
455 slot
= gfn_to_memslot_unaliased(kvm
, gfn
);
457 largepage_idx
= slot_largepage_idx(gfn
, slot
, level
);
458 return *largepage_idx
;
464 static int host_mapping_level(struct kvm
*kvm
, gfn_t gfn
)
466 unsigned long page_size
;
469 page_size
= kvm_host_page_size(kvm
, gfn
);
471 for (i
= PT_PAGE_TABLE_LEVEL
;
472 i
< (PT_PAGE_TABLE_LEVEL
+ KVM_NR_PAGE_SIZES
); ++i
) {
473 if (page_size
>= KVM_HPAGE_SIZE(i
))
482 static int mapping_level(struct kvm_vcpu
*vcpu
, gfn_t large_gfn
)
484 struct kvm_memory_slot
*slot
;
485 int host_level
, level
, max_level
;
487 slot
= gfn_to_memslot(vcpu
->kvm
, large_gfn
);
488 if (slot
&& slot
->dirty_bitmap
)
489 return PT_PAGE_TABLE_LEVEL
;
491 host_level
= host_mapping_level(vcpu
->kvm
, large_gfn
);
493 if (host_level
== PT_PAGE_TABLE_LEVEL
)
496 max_level
= kvm_x86_ops
->get_lpage_level() < host_level
?
497 kvm_x86_ops
->get_lpage_level() : host_level
;
499 for (level
= PT_DIRECTORY_LEVEL
; level
<= max_level
; ++level
)
500 if (has_wrprotected_page(vcpu
->kvm
, large_gfn
, level
))
507 * Take gfn and return the reverse mapping to it.
508 * Note: gfn must be unaliased before this function get called
511 static unsigned long *gfn_to_rmap(struct kvm
*kvm
, gfn_t gfn
, int level
)
513 struct kvm_memory_slot
*slot
;
516 slot
= gfn_to_memslot(kvm
, gfn
);
517 if (likely(level
== PT_PAGE_TABLE_LEVEL
))
518 return &slot
->rmap
[gfn
- slot
->base_gfn
];
520 idx
= (gfn
/ KVM_PAGES_PER_HPAGE(level
)) -
521 (slot
->base_gfn
/ KVM_PAGES_PER_HPAGE(level
));
523 return &slot
->lpage_info
[level
- 2][idx
].rmap_pde
;
527 * Reverse mapping data structures:
529 * If rmapp bit zero is zero, then rmapp point to the shadw page table entry
530 * that points to page_address(page).
532 * If rmapp bit zero is one, (then rmap & ~1) points to a struct kvm_rmap_desc
533 * containing more mappings.
535 * Returns the number of rmap entries before the spte was added or zero if
536 * the spte was not added.
539 static int rmap_add(struct kvm_vcpu
*vcpu
, u64
*spte
, gfn_t gfn
)
541 struct kvm_mmu_page
*sp
;
542 struct kvm_rmap_desc
*desc
;
543 unsigned long *rmapp
;
546 if (!is_rmap_spte(*spte
))
548 gfn
= unalias_gfn(vcpu
->kvm
, gfn
);
549 sp
= page_header(__pa(spte
));
550 sp
->gfns
[spte
- sp
->spt
] = gfn
;
551 rmapp
= gfn_to_rmap(vcpu
->kvm
, gfn
, sp
->role
.level
);
553 rmap_printk("rmap_add: %p %llx 0->1\n", spte
, *spte
);
554 *rmapp
= (unsigned long)spte
;
555 } else if (!(*rmapp
& 1)) {
556 rmap_printk("rmap_add: %p %llx 1->many\n", spte
, *spte
);
557 desc
= mmu_alloc_rmap_desc(vcpu
);
558 desc
->sptes
[0] = (u64
*)*rmapp
;
559 desc
->sptes
[1] = spte
;
560 *rmapp
= (unsigned long)desc
| 1;
562 rmap_printk("rmap_add: %p %llx many->many\n", spte
, *spte
);
563 desc
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
564 while (desc
->sptes
[RMAP_EXT
-1] && desc
->more
) {
568 if (desc
->sptes
[RMAP_EXT
-1]) {
569 desc
->more
= mmu_alloc_rmap_desc(vcpu
);
572 for (i
= 0; desc
->sptes
[i
]; ++i
)
574 desc
->sptes
[i
] = spte
;
579 static void rmap_desc_remove_entry(unsigned long *rmapp
,
580 struct kvm_rmap_desc
*desc
,
582 struct kvm_rmap_desc
*prev_desc
)
586 for (j
= RMAP_EXT
- 1; !desc
->sptes
[j
] && j
> i
; --j
)
588 desc
->sptes
[i
] = desc
->sptes
[j
];
589 desc
->sptes
[j
] = NULL
;
592 if (!prev_desc
&& !desc
->more
)
593 *rmapp
= (unsigned long)desc
->sptes
[0];
596 prev_desc
->more
= desc
->more
;
598 *rmapp
= (unsigned long)desc
->more
| 1;
599 mmu_free_rmap_desc(desc
);
602 static void rmap_remove(struct kvm
*kvm
, u64
*spte
)
604 struct kvm_rmap_desc
*desc
;
605 struct kvm_rmap_desc
*prev_desc
;
606 struct kvm_mmu_page
*sp
;
608 unsigned long *rmapp
;
611 if (!is_rmap_spte(*spte
))
613 sp
= page_header(__pa(spte
));
614 pfn
= spte_to_pfn(*spte
);
615 if (*spte
& shadow_accessed_mask
)
616 kvm_set_pfn_accessed(pfn
);
617 if (is_writable_pte(*spte
))
618 kvm_set_pfn_dirty(pfn
);
619 rmapp
= gfn_to_rmap(kvm
, sp
->gfns
[spte
- sp
->spt
], sp
->role
.level
);
621 printk(KERN_ERR
"rmap_remove: %p %llx 0->BUG\n", spte
, *spte
);
623 } else if (!(*rmapp
& 1)) {
624 rmap_printk("rmap_remove: %p %llx 1->0\n", spte
, *spte
);
625 if ((u64
*)*rmapp
!= spte
) {
626 printk(KERN_ERR
"rmap_remove: %p %llx 1->BUG\n",
632 rmap_printk("rmap_remove: %p %llx many->many\n", spte
, *spte
);
633 desc
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
636 for (i
= 0; i
< RMAP_EXT
&& desc
->sptes
[i
]; ++i
)
637 if (desc
->sptes
[i
] == spte
) {
638 rmap_desc_remove_entry(rmapp
,
646 pr_err("rmap_remove: %p %llx many->many\n", spte
, *spte
);
651 static u64
*rmap_next(struct kvm
*kvm
, unsigned long *rmapp
, u64
*spte
)
653 struct kvm_rmap_desc
*desc
;
659 else if (!(*rmapp
& 1)) {
661 return (u64
*)*rmapp
;
664 desc
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
667 for (i
= 0; i
< RMAP_EXT
&& desc
->sptes
[i
]; ++i
) {
668 if (prev_spte
== spte
)
669 return desc
->sptes
[i
];
670 prev_spte
= desc
->sptes
[i
];
677 static int rmap_write_protect(struct kvm
*kvm
, u64 gfn
)
679 unsigned long *rmapp
;
681 int i
, write_protected
= 0;
683 gfn
= unalias_gfn(kvm
, gfn
);
684 rmapp
= gfn_to_rmap(kvm
, gfn
, PT_PAGE_TABLE_LEVEL
);
686 spte
= rmap_next(kvm
, rmapp
, NULL
);
689 BUG_ON(!(*spte
& PT_PRESENT_MASK
));
690 rmap_printk("rmap_write_protect: spte %p %llx\n", spte
, *spte
);
691 if (is_writable_pte(*spte
)) {
692 __set_spte(spte
, *spte
& ~PT_WRITABLE_MASK
);
695 spte
= rmap_next(kvm
, rmapp
, spte
);
697 if (write_protected
) {
700 spte
= rmap_next(kvm
, rmapp
, NULL
);
701 pfn
= spte_to_pfn(*spte
);
702 kvm_set_pfn_dirty(pfn
);
705 /* check for huge page mappings */
706 for (i
= PT_DIRECTORY_LEVEL
;
707 i
< PT_PAGE_TABLE_LEVEL
+ KVM_NR_PAGE_SIZES
; ++i
) {
708 rmapp
= gfn_to_rmap(kvm
, gfn
, i
);
709 spte
= rmap_next(kvm
, rmapp
, NULL
);
712 BUG_ON(!(*spte
& PT_PRESENT_MASK
));
713 BUG_ON((*spte
& (PT_PAGE_SIZE_MASK
|PT_PRESENT_MASK
)) != (PT_PAGE_SIZE_MASK
|PT_PRESENT_MASK
));
714 pgprintk("rmap_write_protect(large): spte %p %llx %lld\n", spte
, *spte
, gfn
);
715 if (is_writable_pte(*spte
)) {
716 rmap_remove(kvm
, spte
);
718 __set_spte(spte
, shadow_trap_nonpresent_pte
);
722 spte
= rmap_next(kvm
, rmapp
, spte
);
726 return write_protected
;
729 static int kvm_unmap_rmapp(struct kvm
*kvm
, unsigned long *rmapp
,
733 int need_tlb_flush
= 0;
735 while ((spte
= rmap_next(kvm
, rmapp
, NULL
))) {
736 BUG_ON(!(*spte
& PT_PRESENT_MASK
));
737 rmap_printk("kvm_rmap_unmap_hva: spte %p %llx\n", spte
, *spte
);
738 rmap_remove(kvm
, spte
);
739 __set_spte(spte
, shadow_trap_nonpresent_pte
);
742 return need_tlb_flush
;
745 static int kvm_set_pte_rmapp(struct kvm
*kvm
, unsigned long *rmapp
,
750 pte_t
*ptep
= (pte_t
*)data
;
753 WARN_ON(pte_huge(*ptep
));
754 new_pfn
= pte_pfn(*ptep
);
755 spte
= rmap_next(kvm
, rmapp
, NULL
);
757 BUG_ON(!is_shadow_present_pte(*spte
));
758 rmap_printk("kvm_set_pte_rmapp: spte %p %llx\n", spte
, *spte
);
760 if (pte_write(*ptep
)) {
761 rmap_remove(kvm
, spte
);
762 __set_spte(spte
, shadow_trap_nonpresent_pte
);
763 spte
= rmap_next(kvm
, rmapp
, NULL
);
765 new_spte
= *spte
&~ (PT64_BASE_ADDR_MASK
);
766 new_spte
|= (u64
)new_pfn
<< PAGE_SHIFT
;
768 new_spte
&= ~PT_WRITABLE_MASK
;
769 new_spte
&= ~SPTE_HOST_WRITEABLE
;
770 if (is_writable_pte(*spte
))
771 kvm_set_pfn_dirty(spte_to_pfn(*spte
));
772 __set_spte(spte
, new_spte
);
773 spte
= rmap_next(kvm
, rmapp
, spte
);
777 kvm_flush_remote_tlbs(kvm
);
782 static int kvm_handle_hva(struct kvm
*kvm
, unsigned long hva
,
784 int (*handler
)(struct kvm
*kvm
, unsigned long *rmapp
,
790 struct kvm_memslots
*slots
;
792 slots
= kvm_memslots(kvm
);
794 for (i
= 0; i
< slots
->nmemslots
; i
++) {
795 struct kvm_memory_slot
*memslot
= &slots
->memslots
[i
];
796 unsigned long start
= memslot
->userspace_addr
;
799 end
= start
+ (memslot
->npages
<< PAGE_SHIFT
);
800 if (hva
>= start
&& hva
< end
) {
801 gfn_t gfn_offset
= (hva
- start
) >> PAGE_SHIFT
;
803 ret
= handler(kvm
, &memslot
->rmap
[gfn_offset
], data
);
805 for (j
= 0; j
< KVM_NR_PAGE_SIZES
- 1; ++j
) {
806 int idx
= gfn_offset
;
807 idx
/= KVM_PAGES_PER_HPAGE(PT_DIRECTORY_LEVEL
+ j
);
809 &memslot
->lpage_info
[j
][idx
].rmap_pde
,
812 trace_kvm_age_page(hva
, memslot
, ret
);
820 int kvm_unmap_hva(struct kvm
*kvm
, unsigned long hva
)
822 return kvm_handle_hva(kvm
, hva
, 0, kvm_unmap_rmapp
);
825 void kvm_set_spte_hva(struct kvm
*kvm
, unsigned long hva
, pte_t pte
)
827 kvm_handle_hva(kvm
, hva
, (unsigned long)&pte
, kvm_set_pte_rmapp
);
830 static int kvm_age_rmapp(struct kvm
*kvm
, unsigned long *rmapp
,
837 * Emulate the accessed bit for EPT, by checking if this page has
838 * an EPT mapping, and clearing it if it does. On the next access,
839 * a new EPT mapping will be established.
840 * This has some overhead, but not as much as the cost of swapping
841 * out actively used pages or breaking up actively used hugepages.
843 if (!shadow_accessed_mask
)
844 return kvm_unmap_rmapp(kvm
, rmapp
, data
);
846 spte
= rmap_next(kvm
, rmapp
, NULL
);
850 BUG_ON(!(_spte
& PT_PRESENT_MASK
));
851 _young
= _spte
& PT_ACCESSED_MASK
;
854 clear_bit(PT_ACCESSED_SHIFT
, (unsigned long *)spte
);
856 spte
= rmap_next(kvm
, rmapp
, spte
);
861 #define RMAP_RECYCLE_THRESHOLD 1000
863 static void rmap_recycle(struct kvm_vcpu
*vcpu
, u64
*spte
, gfn_t gfn
)
865 unsigned long *rmapp
;
866 struct kvm_mmu_page
*sp
;
868 sp
= page_header(__pa(spte
));
870 gfn
= unalias_gfn(vcpu
->kvm
, gfn
);
871 rmapp
= gfn_to_rmap(vcpu
->kvm
, gfn
, sp
->role
.level
);
873 kvm_unmap_rmapp(vcpu
->kvm
, rmapp
, 0);
874 kvm_flush_remote_tlbs(vcpu
->kvm
);
877 int kvm_age_hva(struct kvm
*kvm
, unsigned long hva
)
879 return kvm_handle_hva(kvm
, hva
, 0, kvm_age_rmapp
);
883 static int is_empty_shadow_page(u64
*spt
)
888 for (pos
= spt
, end
= pos
+ PAGE_SIZE
/ sizeof(u64
); pos
!= end
; pos
++)
889 if (is_shadow_present_pte(*pos
)) {
890 printk(KERN_ERR
"%s: %p %llx\n", __func__
,
898 static void kvm_mmu_free_page(struct kvm
*kvm
, struct kvm_mmu_page
*sp
)
900 ASSERT(is_empty_shadow_page(sp
->spt
));
902 __free_page(virt_to_page(sp
->spt
));
903 __free_page(virt_to_page(sp
->gfns
));
904 kmem_cache_free(mmu_page_header_cache
, sp
);
905 ++kvm
->arch
.n_free_mmu_pages
;
908 static unsigned kvm_page_table_hashfn(gfn_t gfn
)
910 return gfn
& ((1 << KVM_MMU_HASH_SHIFT
) - 1);
913 static struct kvm_mmu_page
*kvm_mmu_alloc_page(struct kvm_vcpu
*vcpu
,
916 struct kvm_mmu_page
*sp
;
918 sp
= mmu_memory_cache_alloc(&vcpu
->arch
.mmu_page_header_cache
, sizeof *sp
);
919 sp
->spt
= mmu_memory_cache_alloc(&vcpu
->arch
.mmu_page_cache
, PAGE_SIZE
);
920 sp
->gfns
= mmu_memory_cache_alloc(&vcpu
->arch
.mmu_page_cache
, PAGE_SIZE
);
921 set_page_private(virt_to_page(sp
->spt
), (unsigned long)sp
);
922 list_add(&sp
->link
, &vcpu
->kvm
->arch
.active_mmu_pages
);
923 bitmap_zero(sp
->slot_bitmap
, KVM_MEMORY_SLOTS
+ KVM_PRIVATE_MEM_SLOTS
);
925 sp
->parent_pte
= parent_pte
;
926 --vcpu
->kvm
->arch
.n_free_mmu_pages
;
930 static void mmu_page_add_parent_pte(struct kvm_vcpu
*vcpu
,
931 struct kvm_mmu_page
*sp
, u64
*parent_pte
)
933 struct kvm_pte_chain
*pte_chain
;
934 struct hlist_node
*node
;
939 if (!sp
->multimapped
) {
940 u64
*old
= sp
->parent_pte
;
943 sp
->parent_pte
= parent_pte
;
947 pte_chain
= mmu_alloc_pte_chain(vcpu
);
948 INIT_HLIST_HEAD(&sp
->parent_ptes
);
949 hlist_add_head(&pte_chain
->link
, &sp
->parent_ptes
);
950 pte_chain
->parent_ptes
[0] = old
;
952 hlist_for_each_entry(pte_chain
, node
, &sp
->parent_ptes
, link
) {
953 if (pte_chain
->parent_ptes
[NR_PTE_CHAIN_ENTRIES
-1])
955 for (i
= 0; i
< NR_PTE_CHAIN_ENTRIES
; ++i
)
956 if (!pte_chain
->parent_ptes
[i
]) {
957 pte_chain
->parent_ptes
[i
] = parent_pte
;
961 pte_chain
= mmu_alloc_pte_chain(vcpu
);
963 hlist_add_head(&pte_chain
->link
, &sp
->parent_ptes
);
964 pte_chain
->parent_ptes
[0] = parent_pte
;
967 static void mmu_page_remove_parent_pte(struct kvm_mmu_page
*sp
,
970 struct kvm_pte_chain
*pte_chain
;
971 struct hlist_node
*node
;
974 if (!sp
->multimapped
) {
975 BUG_ON(sp
->parent_pte
!= parent_pte
);
976 sp
->parent_pte
= NULL
;
979 hlist_for_each_entry(pte_chain
, node
, &sp
->parent_ptes
, link
)
980 for (i
= 0; i
< NR_PTE_CHAIN_ENTRIES
; ++i
) {
981 if (!pte_chain
->parent_ptes
[i
])
983 if (pte_chain
->parent_ptes
[i
] != parent_pte
)
985 while (i
+ 1 < NR_PTE_CHAIN_ENTRIES
986 && pte_chain
->parent_ptes
[i
+ 1]) {
987 pte_chain
->parent_ptes
[i
]
988 = pte_chain
->parent_ptes
[i
+ 1];
991 pte_chain
->parent_ptes
[i
] = NULL
;
993 hlist_del(&pte_chain
->link
);
994 mmu_free_pte_chain(pte_chain
);
995 if (hlist_empty(&sp
->parent_ptes
)) {
997 sp
->parent_pte
= NULL
;
1006 static void mmu_parent_walk(struct kvm_mmu_page
*sp
, mmu_parent_walk_fn fn
)
1008 struct kvm_pte_chain
*pte_chain
;
1009 struct hlist_node
*node
;
1010 struct kvm_mmu_page
*parent_sp
;
1013 if (!sp
->multimapped
&& sp
->parent_pte
) {
1014 parent_sp
= page_header(__pa(sp
->parent_pte
));
1016 mmu_parent_walk(parent_sp
, fn
);
1019 hlist_for_each_entry(pte_chain
, node
, &sp
->parent_ptes
, link
)
1020 for (i
= 0; i
< NR_PTE_CHAIN_ENTRIES
; ++i
) {
1021 if (!pte_chain
->parent_ptes
[i
])
1023 parent_sp
= page_header(__pa(pte_chain
->parent_ptes
[i
]));
1025 mmu_parent_walk(parent_sp
, fn
);
1029 static void kvm_mmu_update_unsync_bitmap(u64
*spte
)
1032 struct kvm_mmu_page
*sp
= page_header(__pa(spte
));
1034 index
= spte
- sp
->spt
;
1035 if (!__test_and_set_bit(index
, sp
->unsync_child_bitmap
))
1036 sp
->unsync_children
++;
1037 WARN_ON(!sp
->unsync_children
);
1040 static void kvm_mmu_update_parents_unsync(struct kvm_mmu_page
*sp
)
1042 struct kvm_pte_chain
*pte_chain
;
1043 struct hlist_node
*node
;
1046 if (!sp
->parent_pte
)
1049 if (!sp
->multimapped
) {
1050 kvm_mmu_update_unsync_bitmap(sp
->parent_pte
);
1054 hlist_for_each_entry(pte_chain
, node
, &sp
->parent_ptes
, link
)
1055 for (i
= 0; i
< NR_PTE_CHAIN_ENTRIES
; ++i
) {
1056 if (!pte_chain
->parent_ptes
[i
])
1058 kvm_mmu_update_unsync_bitmap(pte_chain
->parent_ptes
[i
]);
1062 static int unsync_walk_fn(struct kvm_mmu_page
*sp
)
1064 kvm_mmu_update_parents_unsync(sp
);
1068 static void kvm_mmu_mark_parents_unsync(struct kvm_mmu_page
*sp
)
1070 mmu_parent_walk(sp
, unsync_walk_fn
);
1071 kvm_mmu_update_parents_unsync(sp
);
1074 static void nonpaging_prefetch_page(struct kvm_vcpu
*vcpu
,
1075 struct kvm_mmu_page
*sp
)
1079 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
)
1080 sp
->spt
[i
] = shadow_trap_nonpresent_pte
;
1083 static int nonpaging_sync_page(struct kvm_vcpu
*vcpu
,
1084 struct kvm_mmu_page
*sp
)
1089 static void nonpaging_invlpg(struct kvm_vcpu
*vcpu
, gva_t gva
)
1093 #define KVM_PAGE_ARRAY_NR 16
1095 struct kvm_mmu_pages
{
1096 struct mmu_page_and_offset
{
1097 struct kvm_mmu_page
*sp
;
1099 } page
[KVM_PAGE_ARRAY_NR
];
1103 #define for_each_unsync_children(bitmap, idx) \
1104 for (idx = find_first_bit(bitmap, 512); \
1106 idx = find_next_bit(bitmap, 512, idx+1))
1108 static int mmu_pages_add(struct kvm_mmu_pages
*pvec
, struct kvm_mmu_page
*sp
,
1114 for (i
=0; i
< pvec
->nr
; i
++)
1115 if (pvec
->page
[i
].sp
== sp
)
1118 pvec
->page
[pvec
->nr
].sp
= sp
;
1119 pvec
->page
[pvec
->nr
].idx
= idx
;
1121 return (pvec
->nr
== KVM_PAGE_ARRAY_NR
);
1124 static int __mmu_unsync_walk(struct kvm_mmu_page
*sp
,
1125 struct kvm_mmu_pages
*pvec
)
1127 int i
, ret
, nr_unsync_leaf
= 0;
1129 for_each_unsync_children(sp
->unsync_child_bitmap
, i
) {
1130 u64 ent
= sp
->spt
[i
];
1132 if (is_shadow_present_pte(ent
) && !is_large_pte(ent
)) {
1133 struct kvm_mmu_page
*child
;
1134 child
= page_header(ent
& PT64_BASE_ADDR_MASK
);
1136 if (child
->unsync_children
) {
1137 if (mmu_pages_add(pvec
, child
, i
))
1140 ret
= __mmu_unsync_walk(child
, pvec
);
1142 __clear_bit(i
, sp
->unsync_child_bitmap
);
1144 nr_unsync_leaf
+= ret
;
1149 if (child
->unsync
) {
1151 if (mmu_pages_add(pvec
, child
, i
))
1157 if (find_first_bit(sp
->unsync_child_bitmap
, 512) == 512)
1158 sp
->unsync_children
= 0;
1160 return nr_unsync_leaf
;
1163 static int mmu_unsync_walk(struct kvm_mmu_page
*sp
,
1164 struct kvm_mmu_pages
*pvec
)
1166 if (!sp
->unsync_children
)
1169 mmu_pages_add(pvec
, sp
, 0);
1170 return __mmu_unsync_walk(sp
, pvec
);
1173 static void kvm_unlink_unsync_page(struct kvm
*kvm
, struct kvm_mmu_page
*sp
)
1175 WARN_ON(!sp
->unsync
);
1176 trace_kvm_mmu_sync_page(sp
);
1178 --kvm
->stat
.mmu_unsync
;
1181 static int kvm_mmu_zap_page(struct kvm
*kvm
, struct kvm_mmu_page
*sp
);
1183 static int __kvm_sync_page(struct kvm_vcpu
*vcpu
, struct kvm_mmu_page
*sp
,
1186 if (sp
->role
.cr4_pae
!= !!is_pae(vcpu
)) {
1187 kvm_mmu_zap_page(vcpu
->kvm
, sp
);
1192 if (rmap_write_protect(vcpu
->kvm
, sp
->gfn
))
1193 kvm_flush_remote_tlbs(vcpu
->kvm
);
1194 kvm_unlink_unsync_page(vcpu
->kvm
, sp
);
1197 if (vcpu
->arch
.mmu
.sync_page(vcpu
, sp
)) {
1198 kvm_mmu_zap_page(vcpu
->kvm
, sp
);
1202 kvm_mmu_flush_tlb(vcpu
);
1206 static void mmu_convert_notrap(struct kvm_mmu_page
*sp
);
1207 static int kvm_sync_page_transient(struct kvm_vcpu
*vcpu
,
1208 struct kvm_mmu_page
*sp
)
1212 ret
= __kvm_sync_page(vcpu
, sp
, false);
1214 mmu_convert_notrap(sp
);
1218 static int kvm_sync_page(struct kvm_vcpu
*vcpu
, struct kvm_mmu_page
*sp
)
1220 return __kvm_sync_page(vcpu
, sp
, true);
1223 /* @gfn should be write-protected at the call site */
1224 static void kvm_sync_pages(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1226 struct hlist_head
*bucket
;
1227 struct kvm_mmu_page
*s
;
1228 struct hlist_node
*node
, *n
;
1232 index
= kvm_page_table_hashfn(gfn
);
1233 bucket
= &vcpu
->kvm
->arch
.mmu_page_hash
[index
];
1234 hlist_for_each_entry_safe(s
, node
, n
, bucket
, hash_link
) {
1235 if (s
->gfn
!= gfn
|| !s
->unsync
|| s
->role
.invalid
)
1238 WARN_ON(s
->role
.level
!= PT_PAGE_TABLE_LEVEL
);
1239 if ((s
->role
.cr4_pae
!= !!is_pae(vcpu
)) ||
1240 (vcpu
->arch
.mmu
.sync_page(vcpu
, s
))) {
1241 kvm_mmu_zap_page(vcpu
->kvm
, s
);
1244 kvm_unlink_unsync_page(vcpu
->kvm
, s
);
1249 kvm_mmu_flush_tlb(vcpu
);
1252 struct mmu_page_path
{
1253 struct kvm_mmu_page
*parent
[PT64_ROOT_LEVEL
-1];
1254 unsigned int idx
[PT64_ROOT_LEVEL
-1];
1257 #define for_each_sp(pvec, sp, parents, i) \
1258 for (i = mmu_pages_next(&pvec, &parents, -1), \
1259 sp = pvec.page[i].sp; \
1260 i < pvec.nr && ({ sp = pvec.page[i].sp; 1;}); \
1261 i = mmu_pages_next(&pvec, &parents, i))
1263 static int mmu_pages_next(struct kvm_mmu_pages
*pvec
,
1264 struct mmu_page_path
*parents
,
1269 for (n
= i
+1; n
< pvec
->nr
; n
++) {
1270 struct kvm_mmu_page
*sp
= pvec
->page
[n
].sp
;
1272 if (sp
->role
.level
== PT_PAGE_TABLE_LEVEL
) {
1273 parents
->idx
[0] = pvec
->page
[n
].idx
;
1277 parents
->parent
[sp
->role
.level
-2] = sp
;
1278 parents
->idx
[sp
->role
.level
-1] = pvec
->page
[n
].idx
;
1284 static void mmu_pages_clear_parents(struct mmu_page_path
*parents
)
1286 struct kvm_mmu_page
*sp
;
1287 unsigned int level
= 0;
1290 unsigned int idx
= parents
->idx
[level
];
1292 sp
= parents
->parent
[level
];
1296 --sp
->unsync_children
;
1297 WARN_ON((int)sp
->unsync_children
< 0);
1298 __clear_bit(idx
, sp
->unsync_child_bitmap
);
1300 } while (level
< PT64_ROOT_LEVEL
-1 && !sp
->unsync_children
);
1303 static void kvm_mmu_pages_init(struct kvm_mmu_page
*parent
,
1304 struct mmu_page_path
*parents
,
1305 struct kvm_mmu_pages
*pvec
)
1307 parents
->parent
[parent
->role
.level
-1] = NULL
;
1311 static void mmu_sync_children(struct kvm_vcpu
*vcpu
,
1312 struct kvm_mmu_page
*parent
)
1315 struct kvm_mmu_page
*sp
;
1316 struct mmu_page_path parents
;
1317 struct kvm_mmu_pages pages
;
1319 kvm_mmu_pages_init(parent
, &parents
, &pages
);
1320 while (mmu_unsync_walk(parent
, &pages
)) {
1323 for_each_sp(pages
, sp
, parents
, i
)
1324 protected |= rmap_write_protect(vcpu
->kvm
, sp
->gfn
);
1327 kvm_flush_remote_tlbs(vcpu
->kvm
);
1329 for_each_sp(pages
, sp
, parents
, i
) {
1330 kvm_sync_page(vcpu
, sp
);
1331 mmu_pages_clear_parents(&parents
);
1333 cond_resched_lock(&vcpu
->kvm
->mmu_lock
);
1334 kvm_mmu_pages_init(parent
, &parents
, &pages
);
1338 static struct kvm_mmu_page
*kvm_mmu_get_page(struct kvm_vcpu
*vcpu
,
1346 union kvm_mmu_page_role role
;
1349 struct hlist_head
*bucket
;
1350 struct kvm_mmu_page
*sp
;
1351 struct hlist_node
*node
, *tmp
;
1352 bool need_sync
= false;
1354 role
= vcpu
->arch
.mmu
.base_role
;
1356 role
.direct
= direct
;
1359 role
.access
= access
;
1360 if (vcpu
->arch
.mmu
.root_level
<= PT32_ROOT_LEVEL
) {
1361 quadrant
= gaddr
>> (PAGE_SHIFT
+ (PT64_PT_BITS
* level
));
1362 quadrant
&= (1 << ((PT32_PT_BITS
- PT64_PT_BITS
) * level
)) - 1;
1363 role
.quadrant
= quadrant
;
1365 index
= kvm_page_table_hashfn(gfn
);
1366 bucket
= &vcpu
->kvm
->arch
.mmu_page_hash
[index
];
1367 hlist_for_each_entry_safe(sp
, node
, tmp
, bucket
, hash_link
)
1368 if (sp
->gfn
== gfn
) {
1369 if (!need_sync
&& sp
->unsync
)
1372 if (sp
->role
.word
!= role
.word
)
1375 if (sp
->unsync
&& kvm_sync_page_transient(vcpu
, sp
))
1378 mmu_page_add_parent_pte(vcpu
, sp
, parent_pte
);
1379 if (sp
->unsync_children
) {
1380 set_bit(KVM_REQ_MMU_SYNC
, &vcpu
->requests
);
1381 kvm_mmu_mark_parents_unsync(sp
);
1382 } else if (sp
->unsync
)
1383 kvm_mmu_mark_parents_unsync(sp
);
1385 trace_kvm_mmu_get_page(sp
, false);
1388 ++vcpu
->kvm
->stat
.mmu_cache_miss
;
1389 sp
= kvm_mmu_alloc_page(vcpu
, parent_pte
);
1394 hlist_add_head(&sp
->hash_link
, bucket
);
1396 if (rmap_write_protect(vcpu
->kvm
, gfn
))
1397 kvm_flush_remote_tlbs(vcpu
->kvm
);
1398 if (level
> PT_PAGE_TABLE_LEVEL
&& need_sync
)
1399 kvm_sync_pages(vcpu
, gfn
);
1401 account_shadowed(vcpu
->kvm
, gfn
);
1403 if (shadow_trap_nonpresent_pte
!= shadow_notrap_nonpresent_pte
)
1404 vcpu
->arch
.mmu
.prefetch_page(vcpu
, sp
);
1406 nonpaging_prefetch_page(vcpu
, sp
);
1407 trace_kvm_mmu_get_page(sp
, true);
1411 static void shadow_walk_init(struct kvm_shadow_walk_iterator
*iterator
,
1412 struct kvm_vcpu
*vcpu
, u64 addr
)
1414 iterator
->addr
= addr
;
1415 iterator
->shadow_addr
= vcpu
->arch
.mmu
.root_hpa
;
1416 iterator
->level
= vcpu
->arch
.mmu
.shadow_root_level
;
1417 if (iterator
->level
== PT32E_ROOT_LEVEL
) {
1418 iterator
->shadow_addr
1419 = vcpu
->arch
.mmu
.pae_root
[(addr
>> 30) & 3];
1420 iterator
->shadow_addr
&= PT64_BASE_ADDR_MASK
;
1422 if (!iterator
->shadow_addr
)
1423 iterator
->level
= 0;
1427 static bool shadow_walk_okay(struct kvm_shadow_walk_iterator
*iterator
)
1429 if (iterator
->level
< PT_PAGE_TABLE_LEVEL
)
1432 if (iterator
->level
== PT_PAGE_TABLE_LEVEL
)
1433 if (is_large_pte(*iterator
->sptep
))
1436 iterator
->index
= SHADOW_PT_INDEX(iterator
->addr
, iterator
->level
);
1437 iterator
->sptep
= ((u64
*)__va(iterator
->shadow_addr
)) + iterator
->index
;
1441 static void shadow_walk_next(struct kvm_shadow_walk_iterator
*iterator
)
1443 iterator
->shadow_addr
= *iterator
->sptep
& PT64_BASE_ADDR_MASK
;
1447 static void kvm_mmu_page_unlink_children(struct kvm
*kvm
,
1448 struct kvm_mmu_page
*sp
)
1456 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
1459 if (is_shadow_present_pte(ent
)) {
1460 if (!is_last_spte(ent
, sp
->role
.level
)) {
1461 ent
&= PT64_BASE_ADDR_MASK
;
1462 mmu_page_remove_parent_pte(page_header(ent
),
1465 if (is_large_pte(ent
))
1467 rmap_remove(kvm
, &pt
[i
]);
1470 pt
[i
] = shadow_trap_nonpresent_pte
;
1474 static void kvm_mmu_put_page(struct kvm_mmu_page
*sp
, u64
*parent_pte
)
1476 mmu_page_remove_parent_pte(sp
, parent_pte
);
1479 static void kvm_mmu_reset_last_pte_updated(struct kvm
*kvm
)
1482 struct kvm_vcpu
*vcpu
;
1484 kvm_for_each_vcpu(i
, vcpu
, kvm
)
1485 vcpu
->arch
.last_pte_updated
= NULL
;
1488 static void kvm_mmu_unlink_parents(struct kvm
*kvm
, struct kvm_mmu_page
*sp
)
1492 while (sp
->multimapped
|| sp
->parent_pte
) {
1493 if (!sp
->multimapped
)
1494 parent_pte
= sp
->parent_pte
;
1496 struct kvm_pte_chain
*chain
;
1498 chain
= container_of(sp
->parent_ptes
.first
,
1499 struct kvm_pte_chain
, link
);
1500 parent_pte
= chain
->parent_ptes
[0];
1502 BUG_ON(!parent_pte
);
1503 kvm_mmu_put_page(sp
, parent_pte
);
1504 __set_spte(parent_pte
, shadow_trap_nonpresent_pte
);
1508 static int mmu_zap_unsync_children(struct kvm
*kvm
,
1509 struct kvm_mmu_page
*parent
)
1512 struct mmu_page_path parents
;
1513 struct kvm_mmu_pages pages
;
1515 if (parent
->role
.level
== PT_PAGE_TABLE_LEVEL
)
1518 kvm_mmu_pages_init(parent
, &parents
, &pages
);
1519 while (mmu_unsync_walk(parent
, &pages
)) {
1520 struct kvm_mmu_page
*sp
;
1522 for_each_sp(pages
, sp
, parents
, i
) {
1523 kvm_mmu_zap_page(kvm
, sp
);
1524 mmu_pages_clear_parents(&parents
);
1527 kvm_mmu_pages_init(parent
, &parents
, &pages
);
1533 static int kvm_mmu_zap_page(struct kvm
*kvm
, struct kvm_mmu_page
*sp
)
1537 trace_kvm_mmu_zap_page(sp
);
1538 ++kvm
->stat
.mmu_shadow_zapped
;
1539 ret
= mmu_zap_unsync_children(kvm
, sp
);
1540 kvm_mmu_page_unlink_children(kvm
, sp
);
1541 kvm_mmu_unlink_parents(kvm
, sp
);
1542 kvm_flush_remote_tlbs(kvm
);
1543 if (!sp
->role
.invalid
&& !sp
->role
.direct
)
1544 unaccount_shadowed(kvm
, sp
->gfn
);
1546 kvm_unlink_unsync_page(kvm
, sp
);
1547 if (!sp
->root_count
) {
1550 hlist_del(&sp
->hash_link
);
1551 kvm_mmu_free_page(kvm
, sp
);
1553 sp
->role
.invalid
= 1;
1554 list_move(&sp
->link
, &kvm
->arch
.active_mmu_pages
);
1555 kvm_reload_remote_mmus(kvm
);
1557 kvm_mmu_reset_last_pte_updated(kvm
);
1562 * Changing the number of mmu pages allocated to the vm
1563 * Note: if kvm_nr_mmu_pages is too small, you will get dead lock
1565 void kvm_mmu_change_mmu_pages(struct kvm
*kvm
, unsigned int kvm_nr_mmu_pages
)
1569 used_pages
= kvm
->arch
.n_alloc_mmu_pages
- kvm
->arch
.n_free_mmu_pages
;
1570 used_pages
= max(0, used_pages
);
1573 * If we set the number of mmu pages to be smaller be than the
1574 * number of actived pages , we must to free some mmu pages before we
1578 if (used_pages
> kvm_nr_mmu_pages
) {
1579 while (used_pages
> kvm_nr_mmu_pages
&&
1580 !list_empty(&kvm
->arch
.active_mmu_pages
)) {
1581 struct kvm_mmu_page
*page
;
1583 page
= container_of(kvm
->arch
.active_mmu_pages
.prev
,
1584 struct kvm_mmu_page
, link
);
1585 used_pages
-= kvm_mmu_zap_page(kvm
, page
);
1587 kvm_nr_mmu_pages
= used_pages
;
1588 kvm
->arch
.n_free_mmu_pages
= 0;
1591 kvm
->arch
.n_free_mmu_pages
+= kvm_nr_mmu_pages
1592 - kvm
->arch
.n_alloc_mmu_pages
;
1594 kvm
->arch
.n_alloc_mmu_pages
= kvm_nr_mmu_pages
;
1597 static int kvm_mmu_unprotect_page(struct kvm
*kvm
, gfn_t gfn
)
1600 struct hlist_head
*bucket
;
1601 struct kvm_mmu_page
*sp
;
1602 struct hlist_node
*node
, *n
;
1605 pgprintk("%s: looking for gfn %lx\n", __func__
, gfn
);
1607 index
= kvm_page_table_hashfn(gfn
);
1608 bucket
= &kvm
->arch
.mmu_page_hash
[index
];
1610 hlist_for_each_entry_safe(sp
, node
, n
, bucket
, hash_link
)
1611 if (sp
->gfn
== gfn
&& !sp
->role
.direct
) {
1612 pgprintk("%s: gfn %lx role %x\n", __func__
, gfn
,
1615 if (kvm_mmu_zap_page(kvm
, sp
))
1621 static void mmu_unshadow(struct kvm
*kvm
, gfn_t gfn
)
1624 struct hlist_head
*bucket
;
1625 struct kvm_mmu_page
*sp
;
1626 struct hlist_node
*node
, *nn
;
1628 index
= kvm_page_table_hashfn(gfn
);
1629 bucket
= &kvm
->arch
.mmu_page_hash
[index
];
1631 hlist_for_each_entry_safe(sp
, node
, nn
, bucket
, hash_link
) {
1632 if (sp
->gfn
== gfn
&& !sp
->role
.direct
1633 && !sp
->role
.invalid
) {
1634 pgprintk("%s: zap %lx %x\n",
1635 __func__
, gfn
, sp
->role
.word
);
1636 if (kvm_mmu_zap_page(kvm
, sp
))
1642 static void page_header_update_slot(struct kvm
*kvm
, void *pte
, gfn_t gfn
)
1644 int slot
= memslot_id(kvm
, gfn
);
1645 struct kvm_mmu_page
*sp
= page_header(__pa(pte
));
1647 __set_bit(slot
, sp
->slot_bitmap
);
1650 static void mmu_convert_notrap(struct kvm_mmu_page
*sp
)
1655 if (shadow_trap_nonpresent_pte
== shadow_notrap_nonpresent_pte
)
1658 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
1659 if (pt
[i
] == shadow_notrap_nonpresent_pte
)
1660 __set_spte(&pt
[i
], shadow_trap_nonpresent_pte
);
1665 * The function is based on mtrr_type_lookup() in
1666 * arch/x86/kernel/cpu/mtrr/generic.c
1668 static int get_mtrr_type(struct mtrr_state_type
*mtrr_state
,
1673 u8 prev_match
, curr_match
;
1674 int num_var_ranges
= KVM_NR_VAR_MTRR
;
1676 if (!mtrr_state
->enabled
)
1679 /* Make end inclusive end, instead of exclusive */
1682 /* Look in fixed ranges. Just return the type as per start */
1683 if (mtrr_state
->have_fixed
&& (start
< 0x100000)) {
1686 if (start
< 0x80000) {
1688 idx
+= (start
>> 16);
1689 return mtrr_state
->fixed_ranges
[idx
];
1690 } else if (start
< 0xC0000) {
1692 idx
+= ((start
- 0x80000) >> 14);
1693 return mtrr_state
->fixed_ranges
[idx
];
1694 } else if (start
< 0x1000000) {
1696 idx
+= ((start
- 0xC0000) >> 12);
1697 return mtrr_state
->fixed_ranges
[idx
];
1702 * Look in variable ranges
1703 * Look of multiple ranges matching this address and pick type
1704 * as per MTRR precedence
1706 if (!(mtrr_state
->enabled
& 2))
1707 return mtrr_state
->def_type
;
1710 for (i
= 0; i
< num_var_ranges
; ++i
) {
1711 unsigned short start_state
, end_state
;
1713 if (!(mtrr_state
->var_ranges
[i
].mask_lo
& (1 << 11)))
1716 base
= (((u64
)mtrr_state
->var_ranges
[i
].base_hi
) << 32) +
1717 (mtrr_state
->var_ranges
[i
].base_lo
& PAGE_MASK
);
1718 mask
= (((u64
)mtrr_state
->var_ranges
[i
].mask_hi
) << 32) +
1719 (mtrr_state
->var_ranges
[i
].mask_lo
& PAGE_MASK
);
1721 start_state
= ((start
& mask
) == (base
& mask
));
1722 end_state
= ((end
& mask
) == (base
& mask
));
1723 if (start_state
!= end_state
)
1726 if ((start
& mask
) != (base
& mask
))
1729 curr_match
= mtrr_state
->var_ranges
[i
].base_lo
& 0xff;
1730 if (prev_match
== 0xFF) {
1731 prev_match
= curr_match
;
1735 if (prev_match
== MTRR_TYPE_UNCACHABLE
||
1736 curr_match
== MTRR_TYPE_UNCACHABLE
)
1737 return MTRR_TYPE_UNCACHABLE
;
1739 if ((prev_match
== MTRR_TYPE_WRBACK
&&
1740 curr_match
== MTRR_TYPE_WRTHROUGH
) ||
1741 (prev_match
== MTRR_TYPE_WRTHROUGH
&&
1742 curr_match
== MTRR_TYPE_WRBACK
)) {
1743 prev_match
= MTRR_TYPE_WRTHROUGH
;
1744 curr_match
= MTRR_TYPE_WRTHROUGH
;
1747 if (prev_match
!= curr_match
)
1748 return MTRR_TYPE_UNCACHABLE
;
1751 if (prev_match
!= 0xFF)
1754 return mtrr_state
->def_type
;
1757 u8
kvm_get_guest_memory_type(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1761 mtrr
= get_mtrr_type(&vcpu
->arch
.mtrr_state
, gfn
<< PAGE_SHIFT
,
1762 (gfn
<< PAGE_SHIFT
) + PAGE_SIZE
);
1763 if (mtrr
== 0xfe || mtrr
== 0xff)
1764 mtrr
= MTRR_TYPE_WRBACK
;
1767 EXPORT_SYMBOL_GPL(kvm_get_guest_memory_type
);
1769 static void __kvm_unsync_page(struct kvm_vcpu
*vcpu
, struct kvm_mmu_page
*sp
)
1771 trace_kvm_mmu_unsync_page(sp
);
1772 ++vcpu
->kvm
->stat
.mmu_unsync
;
1775 kvm_mmu_mark_parents_unsync(sp
);
1776 mmu_convert_notrap(sp
);
1779 static void kvm_unsync_pages(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1781 struct hlist_head
*bucket
;
1782 struct kvm_mmu_page
*s
;
1783 struct hlist_node
*node
, *n
;
1786 index
= kvm_page_table_hashfn(gfn
);
1787 bucket
= &vcpu
->kvm
->arch
.mmu_page_hash
[index
];
1789 hlist_for_each_entry_safe(s
, node
, n
, bucket
, hash_link
) {
1790 if (s
->gfn
!= gfn
|| s
->role
.direct
|| s
->unsync
||
1793 WARN_ON(s
->role
.level
!= PT_PAGE_TABLE_LEVEL
);
1794 __kvm_unsync_page(vcpu
, s
);
1798 static int mmu_need_write_protect(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
1802 struct hlist_head
*bucket
;
1803 struct kvm_mmu_page
*s
;
1804 struct hlist_node
*node
, *n
;
1805 bool need_unsync
= false;
1807 index
= kvm_page_table_hashfn(gfn
);
1808 bucket
= &vcpu
->kvm
->arch
.mmu_page_hash
[index
];
1809 hlist_for_each_entry_safe(s
, node
, n
, bucket
, hash_link
) {
1810 if (s
->gfn
!= gfn
|| s
->role
.direct
|| s
->role
.invalid
)
1813 if (s
->role
.level
!= PT_PAGE_TABLE_LEVEL
)
1816 if (!need_unsync
&& !s
->unsync
) {
1817 if (!can_unsync
|| !oos_shadow
)
1823 kvm_unsync_pages(vcpu
, gfn
);
1827 static int set_spte(struct kvm_vcpu
*vcpu
, u64
*sptep
,
1828 unsigned pte_access
, int user_fault
,
1829 int write_fault
, int dirty
, int level
,
1830 gfn_t gfn
, pfn_t pfn
, bool speculative
,
1831 bool can_unsync
, bool reset_host_protection
)
1837 * We don't set the accessed bit, since we sometimes want to see
1838 * whether the guest actually used the pte (in order to detect
1841 spte
= shadow_base_present_pte
| shadow_dirty_mask
;
1843 spte
|= shadow_accessed_mask
;
1845 pte_access
&= ~ACC_WRITE_MASK
;
1846 if (pte_access
& ACC_EXEC_MASK
)
1847 spte
|= shadow_x_mask
;
1849 spte
|= shadow_nx_mask
;
1850 if (pte_access
& ACC_USER_MASK
)
1851 spte
|= shadow_user_mask
;
1852 if (level
> PT_PAGE_TABLE_LEVEL
)
1853 spte
|= PT_PAGE_SIZE_MASK
;
1855 spte
|= kvm_x86_ops
->get_mt_mask(vcpu
, gfn
,
1856 kvm_is_mmio_pfn(pfn
));
1858 if (reset_host_protection
)
1859 spte
|= SPTE_HOST_WRITEABLE
;
1861 spte
|= (u64
)pfn
<< PAGE_SHIFT
;
1863 if ((pte_access
& ACC_WRITE_MASK
)
1864 || (write_fault
&& !is_write_protection(vcpu
) && !user_fault
)) {
1866 if (level
> PT_PAGE_TABLE_LEVEL
&&
1867 has_wrprotected_page(vcpu
->kvm
, gfn
, level
)) {
1869 rmap_remove(vcpu
->kvm
, sptep
);
1870 spte
= shadow_trap_nonpresent_pte
;
1874 spte
|= PT_WRITABLE_MASK
;
1876 if (!tdp_enabled
&& !(pte_access
& ACC_WRITE_MASK
))
1877 spte
&= ~PT_USER_MASK
;
1880 * Optimization: for pte sync, if spte was writable the hash
1881 * lookup is unnecessary (and expensive). Write protection
1882 * is responsibility of mmu_get_page / kvm_sync_page.
1883 * Same reasoning can be applied to dirty page accounting.
1885 if (!can_unsync
&& is_writable_pte(*sptep
))
1888 if (mmu_need_write_protect(vcpu
, gfn
, can_unsync
)) {
1889 pgprintk("%s: found shadow page for %lx, marking ro\n",
1892 pte_access
&= ~ACC_WRITE_MASK
;
1893 if (is_writable_pte(spte
))
1894 spte
&= ~PT_WRITABLE_MASK
;
1898 if (pte_access
& ACC_WRITE_MASK
)
1899 mark_page_dirty(vcpu
->kvm
, gfn
);
1902 __set_spte(sptep
, spte
);
1906 static void mmu_set_spte(struct kvm_vcpu
*vcpu
, u64
*sptep
,
1907 unsigned pt_access
, unsigned pte_access
,
1908 int user_fault
, int write_fault
, int dirty
,
1909 int *ptwrite
, int level
, gfn_t gfn
,
1910 pfn_t pfn
, bool speculative
,
1911 bool reset_host_protection
)
1913 int was_rmapped
= 0;
1914 int was_writable
= is_writable_pte(*sptep
);
1917 pgprintk("%s: spte %llx access %x write_fault %d"
1918 " user_fault %d gfn %lx\n",
1919 __func__
, *sptep
, pt_access
,
1920 write_fault
, user_fault
, gfn
);
1922 if (is_rmap_spte(*sptep
)) {
1924 * If we overwrite a PTE page pointer with a 2MB PMD, unlink
1925 * the parent of the now unreachable PTE.
1927 if (level
> PT_PAGE_TABLE_LEVEL
&&
1928 !is_large_pte(*sptep
)) {
1929 struct kvm_mmu_page
*child
;
1932 child
= page_header(pte
& PT64_BASE_ADDR_MASK
);
1933 mmu_page_remove_parent_pte(child
, sptep
);
1934 __set_spte(sptep
, shadow_trap_nonpresent_pte
);
1935 kvm_flush_remote_tlbs(vcpu
->kvm
);
1936 } else if (pfn
!= spte_to_pfn(*sptep
)) {
1937 pgprintk("hfn old %lx new %lx\n",
1938 spte_to_pfn(*sptep
), pfn
);
1939 rmap_remove(vcpu
->kvm
, sptep
);
1940 __set_spte(sptep
, shadow_trap_nonpresent_pte
);
1941 kvm_flush_remote_tlbs(vcpu
->kvm
);
1946 if (set_spte(vcpu
, sptep
, pte_access
, user_fault
, write_fault
,
1947 dirty
, level
, gfn
, pfn
, speculative
, true,
1948 reset_host_protection
)) {
1951 kvm_x86_ops
->tlb_flush(vcpu
);
1954 pgprintk("%s: setting spte %llx\n", __func__
, *sptep
);
1955 pgprintk("instantiating %s PTE (%s) at %ld (%llx) addr %p\n",
1956 is_large_pte(*sptep
)? "2MB" : "4kB",
1957 *sptep
& PT_PRESENT_MASK
?"RW":"R", gfn
,
1959 if (!was_rmapped
&& is_large_pte(*sptep
))
1960 ++vcpu
->kvm
->stat
.lpages
;
1962 page_header_update_slot(vcpu
->kvm
, sptep
, gfn
);
1964 rmap_count
= rmap_add(vcpu
, sptep
, gfn
);
1965 kvm_release_pfn_clean(pfn
);
1966 if (rmap_count
> RMAP_RECYCLE_THRESHOLD
)
1967 rmap_recycle(vcpu
, sptep
, gfn
);
1970 kvm_release_pfn_dirty(pfn
);
1972 kvm_release_pfn_clean(pfn
);
1975 vcpu
->arch
.last_pte_updated
= sptep
;
1976 vcpu
->arch
.last_pte_gfn
= gfn
;
1980 static void nonpaging_new_cr3(struct kvm_vcpu
*vcpu
)
1984 static int __direct_map(struct kvm_vcpu
*vcpu
, gpa_t v
, int write
,
1985 int level
, gfn_t gfn
, pfn_t pfn
)
1987 struct kvm_shadow_walk_iterator iterator
;
1988 struct kvm_mmu_page
*sp
;
1992 for_each_shadow_entry(vcpu
, (u64
)gfn
<< PAGE_SHIFT
, iterator
) {
1993 if (iterator
.level
== level
) {
1994 mmu_set_spte(vcpu
, iterator
.sptep
, ACC_ALL
, ACC_ALL
,
1995 0, write
, 1, &pt_write
,
1996 level
, gfn
, pfn
, false, true);
1997 ++vcpu
->stat
.pf_fixed
;
2001 if (*iterator
.sptep
== shadow_trap_nonpresent_pte
) {
2002 pseudo_gfn
= (iterator
.addr
& PT64_DIR_BASE_ADDR_MASK
) >> PAGE_SHIFT
;
2003 sp
= kvm_mmu_get_page(vcpu
, pseudo_gfn
, iterator
.addr
,
2005 1, ACC_ALL
, iterator
.sptep
);
2007 pgprintk("nonpaging_map: ENOMEM\n");
2008 kvm_release_pfn_clean(pfn
);
2012 __set_spte(iterator
.sptep
,
2014 | PT_PRESENT_MASK
| PT_WRITABLE_MASK
2015 | shadow_user_mask
| shadow_x_mask
);
2021 static void kvm_send_hwpoison_signal(struct kvm
*kvm
, gfn_t gfn
)
2027 /* Touch the page, so send SIGBUS */
2028 hva
= (void __user
*)gfn_to_hva(kvm
, gfn
);
2029 r
= copy_from_user(buf
, hva
, 1);
2032 static int kvm_handle_bad_page(struct kvm
*kvm
, gfn_t gfn
, pfn_t pfn
)
2034 kvm_release_pfn_clean(pfn
);
2035 if (is_hwpoison_pfn(pfn
)) {
2036 kvm_send_hwpoison_signal(kvm
, gfn
);
2042 static int nonpaging_map(struct kvm_vcpu
*vcpu
, gva_t v
, int write
, gfn_t gfn
)
2047 unsigned long mmu_seq
;
2049 level
= mapping_level(vcpu
, gfn
);
2052 * This path builds a PAE pagetable - so we can map 2mb pages at
2053 * maximum. Therefore check if the level is larger than that.
2055 if (level
> PT_DIRECTORY_LEVEL
)
2056 level
= PT_DIRECTORY_LEVEL
;
2058 gfn
&= ~(KVM_PAGES_PER_HPAGE(level
) - 1);
2060 mmu_seq
= vcpu
->kvm
->mmu_notifier_seq
;
2062 pfn
= gfn_to_pfn(vcpu
->kvm
, gfn
);
2065 if (is_error_pfn(pfn
))
2066 return kvm_handle_bad_page(vcpu
->kvm
, gfn
, pfn
);
2068 spin_lock(&vcpu
->kvm
->mmu_lock
);
2069 if (mmu_notifier_retry(vcpu
, mmu_seq
))
2071 kvm_mmu_free_some_pages(vcpu
);
2072 r
= __direct_map(vcpu
, v
, write
, level
, gfn
, pfn
);
2073 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2079 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2080 kvm_release_pfn_clean(pfn
);
2085 static void mmu_free_roots(struct kvm_vcpu
*vcpu
)
2088 struct kvm_mmu_page
*sp
;
2090 if (!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
))
2092 spin_lock(&vcpu
->kvm
->mmu_lock
);
2093 if (vcpu
->arch
.mmu
.shadow_root_level
== PT64_ROOT_LEVEL
) {
2094 hpa_t root
= vcpu
->arch
.mmu
.root_hpa
;
2096 sp
= page_header(root
);
2098 if (!sp
->root_count
&& sp
->role
.invalid
)
2099 kvm_mmu_zap_page(vcpu
->kvm
, sp
);
2100 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
2101 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2104 for (i
= 0; i
< 4; ++i
) {
2105 hpa_t root
= vcpu
->arch
.mmu
.pae_root
[i
];
2108 root
&= PT64_BASE_ADDR_MASK
;
2109 sp
= page_header(root
);
2111 if (!sp
->root_count
&& sp
->role
.invalid
)
2112 kvm_mmu_zap_page(vcpu
->kvm
, sp
);
2114 vcpu
->arch
.mmu
.pae_root
[i
] = INVALID_PAGE
;
2116 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2117 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
2120 static int mmu_check_root(struct kvm_vcpu
*vcpu
, gfn_t root_gfn
)
2124 if (!kvm_is_visible_gfn(vcpu
->kvm
, root_gfn
)) {
2125 set_bit(KVM_REQ_TRIPLE_FAULT
, &vcpu
->requests
);
2132 static int mmu_alloc_roots(struct kvm_vcpu
*vcpu
)
2136 struct kvm_mmu_page
*sp
;
2140 root_gfn
= vcpu
->arch
.cr3
>> PAGE_SHIFT
;
2142 if (vcpu
->arch
.mmu
.shadow_root_level
== PT64_ROOT_LEVEL
) {
2143 hpa_t root
= vcpu
->arch
.mmu
.root_hpa
;
2145 ASSERT(!VALID_PAGE(root
));
2146 if (mmu_check_root(vcpu
, root_gfn
))
2152 spin_lock(&vcpu
->kvm
->mmu_lock
);
2153 kvm_mmu_free_some_pages(vcpu
);
2154 sp
= kvm_mmu_get_page(vcpu
, root_gfn
, 0,
2155 PT64_ROOT_LEVEL
, direct
,
2157 root
= __pa(sp
->spt
);
2159 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2160 vcpu
->arch
.mmu
.root_hpa
= root
;
2163 direct
= !is_paging(vcpu
);
2164 for (i
= 0; i
< 4; ++i
) {
2165 hpa_t root
= vcpu
->arch
.mmu
.pae_root
[i
];
2167 ASSERT(!VALID_PAGE(root
));
2168 if (vcpu
->arch
.mmu
.root_level
== PT32E_ROOT_LEVEL
) {
2169 pdptr
= kvm_pdptr_read(vcpu
, i
);
2170 if (!is_present_gpte(pdptr
)) {
2171 vcpu
->arch
.mmu
.pae_root
[i
] = 0;
2174 root_gfn
= pdptr
>> PAGE_SHIFT
;
2175 } else if (vcpu
->arch
.mmu
.root_level
== 0)
2177 if (mmu_check_root(vcpu
, root_gfn
))
2183 spin_lock(&vcpu
->kvm
->mmu_lock
);
2184 kvm_mmu_free_some_pages(vcpu
);
2185 sp
= kvm_mmu_get_page(vcpu
, root_gfn
, i
<< 30,
2186 PT32_ROOT_LEVEL
, direct
,
2188 root
= __pa(sp
->spt
);
2190 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2192 vcpu
->arch
.mmu
.pae_root
[i
] = root
| PT_PRESENT_MASK
;
2194 vcpu
->arch
.mmu
.root_hpa
= __pa(vcpu
->arch
.mmu
.pae_root
);
2198 static void mmu_sync_roots(struct kvm_vcpu
*vcpu
)
2201 struct kvm_mmu_page
*sp
;
2203 if (!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
))
2205 if (vcpu
->arch
.mmu
.shadow_root_level
== PT64_ROOT_LEVEL
) {
2206 hpa_t root
= vcpu
->arch
.mmu
.root_hpa
;
2207 sp
= page_header(root
);
2208 mmu_sync_children(vcpu
, sp
);
2211 for (i
= 0; i
< 4; ++i
) {
2212 hpa_t root
= vcpu
->arch
.mmu
.pae_root
[i
];
2214 if (root
&& VALID_PAGE(root
)) {
2215 root
&= PT64_BASE_ADDR_MASK
;
2216 sp
= page_header(root
);
2217 mmu_sync_children(vcpu
, sp
);
2222 void kvm_mmu_sync_roots(struct kvm_vcpu
*vcpu
)
2224 spin_lock(&vcpu
->kvm
->mmu_lock
);
2225 mmu_sync_roots(vcpu
);
2226 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2229 static gpa_t
nonpaging_gva_to_gpa(struct kvm_vcpu
*vcpu
, gva_t vaddr
,
2230 u32 access
, u32
*error
)
2237 static int nonpaging_page_fault(struct kvm_vcpu
*vcpu
, gva_t gva
,
2243 pgprintk("%s: gva %lx error %x\n", __func__
, gva
, error_code
);
2244 r
= mmu_topup_memory_caches(vcpu
);
2249 ASSERT(VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
2251 gfn
= gva
>> PAGE_SHIFT
;
2253 return nonpaging_map(vcpu
, gva
& PAGE_MASK
,
2254 error_code
& PFERR_WRITE_MASK
, gfn
);
2257 static int tdp_page_fault(struct kvm_vcpu
*vcpu
, gva_t gpa
,
2263 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2264 unsigned long mmu_seq
;
2267 ASSERT(VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
2269 r
= mmu_topup_memory_caches(vcpu
);
2273 level
= mapping_level(vcpu
, gfn
);
2275 gfn
&= ~(KVM_PAGES_PER_HPAGE(level
) - 1);
2277 mmu_seq
= vcpu
->kvm
->mmu_notifier_seq
;
2279 pfn
= gfn_to_pfn(vcpu
->kvm
, gfn
);
2280 if (is_error_pfn(pfn
))
2281 return kvm_handle_bad_page(vcpu
->kvm
, gfn
, pfn
);
2282 spin_lock(&vcpu
->kvm
->mmu_lock
);
2283 if (mmu_notifier_retry(vcpu
, mmu_seq
))
2285 kvm_mmu_free_some_pages(vcpu
);
2286 r
= __direct_map(vcpu
, gpa
, error_code
& PFERR_WRITE_MASK
,
2288 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2293 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2294 kvm_release_pfn_clean(pfn
);
2298 static void nonpaging_free(struct kvm_vcpu
*vcpu
)
2300 mmu_free_roots(vcpu
);
2303 static int nonpaging_init_context(struct kvm_vcpu
*vcpu
)
2305 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
2307 context
->new_cr3
= nonpaging_new_cr3
;
2308 context
->page_fault
= nonpaging_page_fault
;
2309 context
->gva_to_gpa
= nonpaging_gva_to_gpa
;
2310 context
->free
= nonpaging_free
;
2311 context
->prefetch_page
= nonpaging_prefetch_page
;
2312 context
->sync_page
= nonpaging_sync_page
;
2313 context
->invlpg
= nonpaging_invlpg
;
2314 context
->root_level
= 0;
2315 context
->shadow_root_level
= PT32E_ROOT_LEVEL
;
2316 context
->root_hpa
= INVALID_PAGE
;
2320 void kvm_mmu_flush_tlb(struct kvm_vcpu
*vcpu
)
2322 ++vcpu
->stat
.tlb_flush
;
2323 kvm_x86_ops
->tlb_flush(vcpu
);
2326 static void paging_new_cr3(struct kvm_vcpu
*vcpu
)
2328 pgprintk("%s: cr3 %lx\n", __func__
, vcpu
->arch
.cr3
);
2329 mmu_free_roots(vcpu
);
2332 static void inject_page_fault(struct kvm_vcpu
*vcpu
,
2336 kvm_inject_page_fault(vcpu
, addr
, err_code
);
2339 static void paging_free(struct kvm_vcpu
*vcpu
)
2341 nonpaging_free(vcpu
);
2344 static bool is_rsvd_bits_set(struct kvm_vcpu
*vcpu
, u64 gpte
, int level
)
2348 bit7
= (gpte
>> 7) & 1;
2349 return (gpte
& vcpu
->arch
.mmu
.rsvd_bits_mask
[bit7
][level
-1]) != 0;
2353 #include "paging_tmpl.h"
2357 #include "paging_tmpl.h"
2360 static void reset_rsvds_bits_mask(struct kvm_vcpu
*vcpu
, int level
)
2362 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
2363 int maxphyaddr
= cpuid_maxphyaddr(vcpu
);
2364 u64 exb_bit_rsvd
= 0;
2367 exb_bit_rsvd
= rsvd_bits(63, 63);
2369 case PT32_ROOT_LEVEL
:
2370 /* no rsvd bits for 2 level 4K page table entries */
2371 context
->rsvd_bits_mask
[0][1] = 0;
2372 context
->rsvd_bits_mask
[0][0] = 0;
2373 context
->rsvd_bits_mask
[1][0] = context
->rsvd_bits_mask
[0][0];
2375 if (!is_pse(vcpu
)) {
2376 context
->rsvd_bits_mask
[1][1] = 0;
2380 if (is_cpuid_PSE36())
2381 /* 36bits PSE 4MB page */
2382 context
->rsvd_bits_mask
[1][1] = rsvd_bits(17, 21);
2384 /* 32 bits PSE 4MB page */
2385 context
->rsvd_bits_mask
[1][1] = rsvd_bits(13, 21);
2387 case PT32E_ROOT_LEVEL
:
2388 context
->rsvd_bits_mask
[0][2] =
2389 rsvd_bits(maxphyaddr
, 63) |
2390 rsvd_bits(7, 8) | rsvd_bits(1, 2); /* PDPTE */
2391 context
->rsvd_bits_mask
[0][1] = exb_bit_rsvd
|
2392 rsvd_bits(maxphyaddr
, 62); /* PDE */
2393 context
->rsvd_bits_mask
[0][0] = exb_bit_rsvd
|
2394 rsvd_bits(maxphyaddr
, 62); /* PTE */
2395 context
->rsvd_bits_mask
[1][1] = exb_bit_rsvd
|
2396 rsvd_bits(maxphyaddr
, 62) |
2397 rsvd_bits(13, 20); /* large page */
2398 context
->rsvd_bits_mask
[1][0] = context
->rsvd_bits_mask
[0][0];
2400 case PT64_ROOT_LEVEL
:
2401 context
->rsvd_bits_mask
[0][3] = exb_bit_rsvd
|
2402 rsvd_bits(maxphyaddr
, 51) | rsvd_bits(7, 8);
2403 context
->rsvd_bits_mask
[0][2] = exb_bit_rsvd
|
2404 rsvd_bits(maxphyaddr
, 51) | rsvd_bits(7, 8);
2405 context
->rsvd_bits_mask
[0][1] = exb_bit_rsvd
|
2406 rsvd_bits(maxphyaddr
, 51);
2407 context
->rsvd_bits_mask
[0][0] = exb_bit_rsvd
|
2408 rsvd_bits(maxphyaddr
, 51);
2409 context
->rsvd_bits_mask
[1][3] = context
->rsvd_bits_mask
[0][3];
2410 context
->rsvd_bits_mask
[1][2] = exb_bit_rsvd
|
2411 rsvd_bits(maxphyaddr
, 51) |
2413 context
->rsvd_bits_mask
[1][1] = exb_bit_rsvd
|
2414 rsvd_bits(maxphyaddr
, 51) |
2415 rsvd_bits(13, 20); /* large page */
2416 context
->rsvd_bits_mask
[1][0] = context
->rsvd_bits_mask
[0][0];
2421 static int paging64_init_context_common(struct kvm_vcpu
*vcpu
, int level
)
2423 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
2425 ASSERT(is_pae(vcpu
));
2426 context
->new_cr3
= paging_new_cr3
;
2427 context
->page_fault
= paging64_page_fault
;
2428 context
->gva_to_gpa
= paging64_gva_to_gpa
;
2429 context
->prefetch_page
= paging64_prefetch_page
;
2430 context
->sync_page
= paging64_sync_page
;
2431 context
->invlpg
= paging64_invlpg
;
2432 context
->free
= paging_free
;
2433 context
->root_level
= level
;
2434 context
->shadow_root_level
= level
;
2435 context
->root_hpa
= INVALID_PAGE
;
2439 static int paging64_init_context(struct kvm_vcpu
*vcpu
)
2441 reset_rsvds_bits_mask(vcpu
, PT64_ROOT_LEVEL
);
2442 return paging64_init_context_common(vcpu
, PT64_ROOT_LEVEL
);
2445 static int paging32_init_context(struct kvm_vcpu
*vcpu
)
2447 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
2449 reset_rsvds_bits_mask(vcpu
, PT32_ROOT_LEVEL
);
2450 context
->new_cr3
= paging_new_cr3
;
2451 context
->page_fault
= paging32_page_fault
;
2452 context
->gva_to_gpa
= paging32_gva_to_gpa
;
2453 context
->free
= paging_free
;
2454 context
->prefetch_page
= paging32_prefetch_page
;
2455 context
->sync_page
= paging32_sync_page
;
2456 context
->invlpg
= paging32_invlpg
;
2457 context
->root_level
= PT32_ROOT_LEVEL
;
2458 context
->shadow_root_level
= PT32E_ROOT_LEVEL
;
2459 context
->root_hpa
= INVALID_PAGE
;
2463 static int paging32E_init_context(struct kvm_vcpu
*vcpu
)
2465 reset_rsvds_bits_mask(vcpu
, PT32E_ROOT_LEVEL
);
2466 return paging64_init_context_common(vcpu
, PT32E_ROOT_LEVEL
);
2469 static int init_kvm_tdp_mmu(struct kvm_vcpu
*vcpu
)
2471 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
2473 context
->new_cr3
= nonpaging_new_cr3
;
2474 context
->page_fault
= tdp_page_fault
;
2475 context
->free
= nonpaging_free
;
2476 context
->prefetch_page
= nonpaging_prefetch_page
;
2477 context
->sync_page
= nonpaging_sync_page
;
2478 context
->invlpg
= nonpaging_invlpg
;
2479 context
->shadow_root_level
= kvm_x86_ops
->get_tdp_level();
2480 context
->root_hpa
= INVALID_PAGE
;
2482 if (!is_paging(vcpu
)) {
2483 context
->gva_to_gpa
= nonpaging_gva_to_gpa
;
2484 context
->root_level
= 0;
2485 } else if (is_long_mode(vcpu
)) {
2486 reset_rsvds_bits_mask(vcpu
, PT64_ROOT_LEVEL
);
2487 context
->gva_to_gpa
= paging64_gva_to_gpa
;
2488 context
->root_level
= PT64_ROOT_LEVEL
;
2489 } else if (is_pae(vcpu
)) {
2490 reset_rsvds_bits_mask(vcpu
, PT32E_ROOT_LEVEL
);
2491 context
->gva_to_gpa
= paging64_gva_to_gpa
;
2492 context
->root_level
= PT32E_ROOT_LEVEL
;
2494 reset_rsvds_bits_mask(vcpu
, PT32_ROOT_LEVEL
);
2495 context
->gva_to_gpa
= paging32_gva_to_gpa
;
2496 context
->root_level
= PT32_ROOT_LEVEL
;
2502 static int init_kvm_softmmu(struct kvm_vcpu
*vcpu
)
2507 ASSERT(!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
2509 if (!is_paging(vcpu
))
2510 r
= nonpaging_init_context(vcpu
);
2511 else if (is_long_mode(vcpu
))
2512 r
= paging64_init_context(vcpu
);
2513 else if (is_pae(vcpu
))
2514 r
= paging32E_init_context(vcpu
);
2516 r
= paging32_init_context(vcpu
);
2518 vcpu
->arch
.mmu
.base_role
.cr4_pae
= !!is_pae(vcpu
);
2519 vcpu
->arch
.mmu
.base_role
.cr0_wp
= is_write_protection(vcpu
);
2524 static int init_kvm_mmu(struct kvm_vcpu
*vcpu
)
2526 vcpu
->arch
.update_pte
.pfn
= bad_pfn
;
2529 return init_kvm_tdp_mmu(vcpu
);
2531 return init_kvm_softmmu(vcpu
);
2534 static void destroy_kvm_mmu(struct kvm_vcpu
*vcpu
)
2537 if (VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
))
2538 /* mmu.free() should set root_hpa = INVALID_PAGE */
2539 vcpu
->arch
.mmu
.free(vcpu
);
2542 int kvm_mmu_reset_context(struct kvm_vcpu
*vcpu
)
2544 destroy_kvm_mmu(vcpu
);
2545 return init_kvm_mmu(vcpu
);
2547 EXPORT_SYMBOL_GPL(kvm_mmu_reset_context
);
2549 int kvm_mmu_load(struct kvm_vcpu
*vcpu
)
2553 r
= mmu_topup_memory_caches(vcpu
);
2556 r
= mmu_alloc_roots(vcpu
);
2557 spin_lock(&vcpu
->kvm
->mmu_lock
);
2558 mmu_sync_roots(vcpu
);
2559 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2562 /* set_cr3() should ensure TLB has been flushed */
2563 kvm_x86_ops
->set_cr3(vcpu
, vcpu
->arch
.mmu
.root_hpa
);
2567 EXPORT_SYMBOL_GPL(kvm_mmu_load
);
2569 void kvm_mmu_unload(struct kvm_vcpu
*vcpu
)
2571 mmu_free_roots(vcpu
);
2574 static void mmu_pte_write_zap_pte(struct kvm_vcpu
*vcpu
,
2575 struct kvm_mmu_page
*sp
,
2579 struct kvm_mmu_page
*child
;
2582 if (is_shadow_present_pte(pte
)) {
2583 if (is_last_spte(pte
, sp
->role
.level
))
2584 rmap_remove(vcpu
->kvm
, spte
);
2586 child
= page_header(pte
& PT64_BASE_ADDR_MASK
);
2587 mmu_page_remove_parent_pte(child
, spte
);
2590 __set_spte(spte
, shadow_trap_nonpresent_pte
);
2591 if (is_large_pte(pte
))
2592 --vcpu
->kvm
->stat
.lpages
;
2595 static void mmu_pte_write_new_pte(struct kvm_vcpu
*vcpu
,
2596 struct kvm_mmu_page
*sp
,
2600 if (sp
->role
.level
!= PT_PAGE_TABLE_LEVEL
) {
2601 ++vcpu
->kvm
->stat
.mmu_pde_zapped
;
2605 ++vcpu
->kvm
->stat
.mmu_pte_updated
;
2606 if (!sp
->role
.cr4_pae
)
2607 paging32_update_pte(vcpu
, sp
, spte
, new);
2609 paging64_update_pte(vcpu
, sp
, spte
, new);
2612 static bool need_remote_flush(u64 old
, u64
new)
2614 if (!is_shadow_present_pte(old
))
2616 if (!is_shadow_present_pte(new))
2618 if ((old
^ new) & PT64_BASE_ADDR_MASK
)
2620 old
^= PT64_NX_MASK
;
2621 new ^= PT64_NX_MASK
;
2622 return (old
& ~new & PT64_PERM_MASK
) != 0;
2625 static void mmu_pte_write_flush_tlb(struct kvm_vcpu
*vcpu
, u64 old
, u64
new)
2627 if (need_remote_flush(old
, new))
2628 kvm_flush_remote_tlbs(vcpu
->kvm
);
2630 kvm_mmu_flush_tlb(vcpu
);
2633 static bool last_updated_pte_accessed(struct kvm_vcpu
*vcpu
)
2635 u64
*spte
= vcpu
->arch
.last_pte_updated
;
2637 return !!(spte
&& (*spte
& shadow_accessed_mask
));
2640 static void mmu_guess_page_from_pte_write(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
2646 if (!is_present_gpte(gpte
))
2648 gfn
= (gpte
& PT64_BASE_ADDR_MASK
) >> PAGE_SHIFT
;
2650 vcpu
->arch
.update_pte
.mmu_seq
= vcpu
->kvm
->mmu_notifier_seq
;
2652 pfn
= gfn_to_pfn(vcpu
->kvm
, gfn
);
2654 if (is_error_pfn(pfn
)) {
2655 kvm_release_pfn_clean(pfn
);
2658 vcpu
->arch
.update_pte
.gfn
= gfn
;
2659 vcpu
->arch
.update_pte
.pfn
= pfn
;
2662 static void kvm_mmu_access_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2664 u64
*spte
= vcpu
->arch
.last_pte_updated
;
2667 && vcpu
->arch
.last_pte_gfn
== gfn
2668 && shadow_accessed_mask
2669 && !(*spte
& shadow_accessed_mask
)
2670 && is_shadow_present_pte(*spte
))
2671 set_bit(PT_ACCESSED_SHIFT
, (unsigned long *)spte
);
2674 void kvm_mmu_pte_write(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
2675 const u8
*new, int bytes
,
2676 bool guest_initiated
)
2678 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2679 struct kvm_mmu_page
*sp
;
2680 struct hlist_node
*node
, *n
;
2681 struct hlist_head
*bucket
;
2685 unsigned offset
= offset_in_page(gpa
);
2687 unsigned page_offset
;
2688 unsigned misaligned
;
2696 pgprintk("%s: gpa %llx bytes %d\n", __func__
, gpa
, bytes
);
2698 invlpg_counter
= atomic_read(&vcpu
->kvm
->arch
.invlpg_counter
);
2701 * Assume that the pte write on a page table of the same type
2702 * as the current vcpu paging mode. This is nearly always true
2703 * (might be false while changing modes). Note it is verified later
2706 if ((is_pae(vcpu
) && bytes
== 4) || !new) {
2707 /* Handle a 32-bit guest writing two halves of a 64-bit gpte */
2712 r
= kvm_read_guest(vcpu
->kvm
, gpa
, &gentry
, min(bytes
, 8));
2715 new = (const u8
*)&gentry
;
2720 gentry
= *(const u32
*)new;
2723 gentry
= *(const u64
*)new;
2730 mmu_guess_page_from_pte_write(vcpu
, gpa
, gentry
);
2731 spin_lock(&vcpu
->kvm
->mmu_lock
);
2732 if (atomic_read(&vcpu
->kvm
->arch
.invlpg_counter
) != invlpg_counter
)
2734 kvm_mmu_access_page(vcpu
, gfn
);
2735 kvm_mmu_free_some_pages(vcpu
);
2736 ++vcpu
->kvm
->stat
.mmu_pte_write
;
2737 kvm_mmu_audit(vcpu
, "pre pte write");
2738 if (guest_initiated
) {
2739 if (gfn
== vcpu
->arch
.last_pt_write_gfn
2740 && !last_updated_pte_accessed(vcpu
)) {
2741 ++vcpu
->arch
.last_pt_write_count
;
2742 if (vcpu
->arch
.last_pt_write_count
>= 3)
2745 vcpu
->arch
.last_pt_write_gfn
= gfn
;
2746 vcpu
->arch
.last_pt_write_count
= 1;
2747 vcpu
->arch
.last_pte_updated
= NULL
;
2750 index
= kvm_page_table_hashfn(gfn
);
2751 bucket
= &vcpu
->kvm
->arch
.mmu_page_hash
[index
];
2754 hlist_for_each_entry_safe(sp
, node
, n
, bucket
, hash_link
) {
2755 if (sp
->gfn
!= gfn
|| sp
->role
.direct
|| sp
->role
.invalid
)
2757 pte_size
= sp
->role
.cr4_pae
? 8 : 4;
2758 misaligned
= (offset
^ (offset
+ bytes
- 1)) & ~(pte_size
- 1);
2759 misaligned
|= bytes
< 4;
2760 if (misaligned
|| flooded
) {
2762 * Misaligned accesses are too much trouble to fix
2763 * up; also, they usually indicate a page is not used
2766 * If we're seeing too many writes to a page,
2767 * it may no longer be a page table, or we may be
2768 * forking, in which case it is better to unmap the
2771 pgprintk("misaligned: gpa %llx bytes %d role %x\n",
2772 gpa
, bytes
, sp
->role
.word
);
2773 if (kvm_mmu_zap_page(vcpu
->kvm
, sp
))
2775 ++vcpu
->kvm
->stat
.mmu_flooded
;
2778 page_offset
= offset
;
2779 level
= sp
->role
.level
;
2781 if (!sp
->role
.cr4_pae
) {
2782 page_offset
<<= 1; /* 32->64 */
2784 * A 32-bit pde maps 4MB while the shadow pdes map
2785 * only 2MB. So we need to double the offset again
2786 * and zap two pdes instead of one.
2788 if (level
== PT32_ROOT_LEVEL
) {
2789 page_offset
&= ~7; /* kill rounding error */
2793 quadrant
= page_offset
>> PAGE_SHIFT
;
2794 page_offset
&= ~PAGE_MASK
;
2795 if (quadrant
!= sp
->role
.quadrant
)
2798 spte
= &sp
->spt
[page_offset
/ sizeof(*spte
)];
2801 mmu_pte_write_zap_pte(vcpu
, sp
, spte
);
2803 mmu_pte_write_new_pte(vcpu
, sp
, spte
, &gentry
);
2804 mmu_pte_write_flush_tlb(vcpu
, entry
, *spte
);
2808 kvm_mmu_audit(vcpu
, "post pte write");
2809 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2810 if (!is_error_pfn(vcpu
->arch
.update_pte
.pfn
)) {
2811 kvm_release_pfn_clean(vcpu
->arch
.update_pte
.pfn
);
2812 vcpu
->arch
.update_pte
.pfn
= bad_pfn
;
2816 int kvm_mmu_unprotect_page_virt(struct kvm_vcpu
*vcpu
, gva_t gva
)
2824 gpa
= kvm_mmu_gva_to_gpa_read(vcpu
, gva
, NULL
);
2826 spin_lock(&vcpu
->kvm
->mmu_lock
);
2827 r
= kvm_mmu_unprotect_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
2828 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2831 EXPORT_SYMBOL_GPL(kvm_mmu_unprotect_page_virt
);
2833 void __kvm_mmu_free_some_pages(struct kvm_vcpu
*vcpu
)
2835 while (vcpu
->kvm
->arch
.n_free_mmu_pages
< KVM_REFILL_PAGES
&&
2836 !list_empty(&vcpu
->kvm
->arch
.active_mmu_pages
)) {
2837 struct kvm_mmu_page
*sp
;
2839 sp
= container_of(vcpu
->kvm
->arch
.active_mmu_pages
.prev
,
2840 struct kvm_mmu_page
, link
);
2841 kvm_mmu_zap_page(vcpu
->kvm
, sp
);
2842 ++vcpu
->kvm
->stat
.mmu_recycled
;
2846 int kvm_mmu_page_fault(struct kvm_vcpu
*vcpu
, gva_t cr2
, u32 error_code
)
2849 enum emulation_result er
;
2851 r
= vcpu
->arch
.mmu
.page_fault(vcpu
, cr2
, error_code
);
2860 r
= mmu_topup_memory_caches(vcpu
);
2864 er
= emulate_instruction(vcpu
, cr2
, error_code
, 0);
2869 case EMULATE_DO_MMIO
:
2870 ++vcpu
->stat
.mmio_exits
;
2880 EXPORT_SYMBOL_GPL(kvm_mmu_page_fault
);
2882 void kvm_mmu_invlpg(struct kvm_vcpu
*vcpu
, gva_t gva
)
2884 vcpu
->arch
.mmu
.invlpg(vcpu
, gva
);
2885 kvm_mmu_flush_tlb(vcpu
);
2886 ++vcpu
->stat
.invlpg
;
2888 EXPORT_SYMBOL_GPL(kvm_mmu_invlpg
);
2890 void kvm_enable_tdp(void)
2894 EXPORT_SYMBOL_GPL(kvm_enable_tdp
);
2896 void kvm_disable_tdp(void)
2898 tdp_enabled
= false;
2900 EXPORT_SYMBOL_GPL(kvm_disable_tdp
);
2902 static void free_mmu_pages(struct kvm_vcpu
*vcpu
)
2904 free_page((unsigned long)vcpu
->arch
.mmu
.pae_root
);
2907 static int alloc_mmu_pages(struct kvm_vcpu
*vcpu
)
2915 * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64.
2916 * Therefore we need to allocate shadow page tables in the first
2917 * 4GB of memory, which happens to fit the DMA32 zone.
2919 page
= alloc_page(GFP_KERNEL
| __GFP_DMA32
);
2923 vcpu
->arch
.mmu
.pae_root
= page_address(page
);
2924 for (i
= 0; i
< 4; ++i
)
2925 vcpu
->arch
.mmu
.pae_root
[i
] = INVALID_PAGE
;
2930 int kvm_mmu_create(struct kvm_vcpu
*vcpu
)
2933 ASSERT(!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
2935 return alloc_mmu_pages(vcpu
);
2938 int kvm_mmu_setup(struct kvm_vcpu
*vcpu
)
2941 ASSERT(!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
2943 return init_kvm_mmu(vcpu
);
2946 void kvm_mmu_destroy(struct kvm_vcpu
*vcpu
)
2950 destroy_kvm_mmu(vcpu
);
2951 free_mmu_pages(vcpu
);
2952 mmu_free_memory_caches(vcpu
);
2955 void kvm_mmu_slot_remove_write_access(struct kvm
*kvm
, int slot
)
2957 struct kvm_mmu_page
*sp
;
2959 list_for_each_entry(sp
, &kvm
->arch
.active_mmu_pages
, link
) {
2963 if (!test_bit(slot
, sp
->slot_bitmap
))
2967 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
)
2969 if (pt
[i
] & PT_WRITABLE_MASK
)
2970 pt
[i
] &= ~PT_WRITABLE_MASK
;
2972 kvm_flush_remote_tlbs(kvm
);
2975 void kvm_mmu_zap_all(struct kvm
*kvm
)
2977 struct kvm_mmu_page
*sp
, *node
;
2979 spin_lock(&kvm
->mmu_lock
);
2981 list_for_each_entry_safe(sp
, node
, &kvm
->arch
.active_mmu_pages
, link
)
2982 if (kvm_mmu_zap_page(kvm
, sp
))
2985 spin_unlock(&kvm
->mmu_lock
);
2987 kvm_flush_remote_tlbs(kvm
);
2990 static int kvm_mmu_remove_some_alloc_mmu_pages(struct kvm
*kvm
)
2992 struct kvm_mmu_page
*page
;
2994 page
= container_of(kvm
->arch
.active_mmu_pages
.prev
,
2995 struct kvm_mmu_page
, link
);
2996 return kvm_mmu_zap_page(kvm
, page
);
2999 static int mmu_shrink(struct shrinker
*shrink
, int nr_to_scan
, gfp_t gfp_mask
)
3002 struct kvm
*kvm_freed
= NULL
;
3003 int cache_count
= 0;
3005 spin_lock(&kvm_lock
);
3007 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3008 int npages
, idx
, freed_pages
;
3010 idx
= srcu_read_lock(&kvm
->srcu
);
3011 spin_lock(&kvm
->mmu_lock
);
3012 npages
= kvm
->arch
.n_alloc_mmu_pages
-
3013 kvm
->arch
.n_free_mmu_pages
;
3014 cache_count
+= npages
;
3015 if (!kvm_freed
&& nr_to_scan
> 0 && npages
> 0) {
3016 freed_pages
= kvm_mmu_remove_some_alloc_mmu_pages(kvm
);
3017 cache_count
-= freed_pages
;
3022 spin_unlock(&kvm
->mmu_lock
);
3023 srcu_read_unlock(&kvm
->srcu
, idx
);
3026 list_move_tail(&kvm_freed
->vm_list
, &vm_list
);
3028 spin_unlock(&kvm_lock
);
3033 static struct shrinker mmu_shrinker
= {
3034 .shrink
= mmu_shrink
,
3035 .seeks
= DEFAULT_SEEKS
* 10,
3038 static void mmu_destroy_caches(void)
3040 if (pte_chain_cache
)
3041 kmem_cache_destroy(pte_chain_cache
);
3042 if (rmap_desc_cache
)
3043 kmem_cache_destroy(rmap_desc_cache
);
3044 if (mmu_page_header_cache
)
3045 kmem_cache_destroy(mmu_page_header_cache
);
3048 void kvm_mmu_module_exit(void)
3050 mmu_destroy_caches();
3051 unregister_shrinker(&mmu_shrinker
);
3054 int kvm_mmu_module_init(void)
3056 pte_chain_cache
= kmem_cache_create("kvm_pte_chain",
3057 sizeof(struct kvm_pte_chain
),
3059 if (!pte_chain_cache
)
3061 rmap_desc_cache
= kmem_cache_create("kvm_rmap_desc",
3062 sizeof(struct kvm_rmap_desc
),
3064 if (!rmap_desc_cache
)
3067 mmu_page_header_cache
= kmem_cache_create("kvm_mmu_page_header",
3068 sizeof(struct kvm_mmu_page
),
3070 if (!mmu_page_header_cache
)
3073 register_shrinker(&mmu_shrinker
);
3078 mmu_destroy_caches();
3083 * Caculate mmu pages needed for kvm.
3085 unsigned int kvm_mmu_calculate_mmu_pages(struct kvm
*kvm
)
3088 unsigned int nr_mmu_pages
;
3089 unsigned int nr_pages
= 0;
3090 struct kvm_memslots
*slots
;
3092 slots
= kvm_memslots(kvm
);
3094 for (i
= 0; i
< slots
->nmemslots
; i
++)
3095 nr_pages
+= slots
->memslots
[i
].npages
;
3097 nr_mmu_pages
= nr_pages
* KVM_PERMILLE_MMU_PAGES
/ 1000;
3098 nr_mmu_pages
= max(nr_mmu_pages
,
3099 (unsigned int) KVM_MIN_ALLOC_MMU_PAGES
);
3101 return nr_mmu_pages
;
3104 static void *pv_mmu_peek_buffer(struct kvm_pv_mmu_op_buffer
*buffer
,
3107 if (len
> buffer
->len
)
3112 static void *pv_mmu_read_buffer(struct kvm_pv_mmu_op_buffer
*buffer
,
3117 ret
= pv_mmu_peek_buffer(buffer
, len
);
3122 buffer
->processed
+= len
;
3126 static int kvm_pv_mmu_write(struct kvm_vcpu
*vcpu
,
3127 gpa_t addr
, gpa_t value
)
3132 if (!is_long_mode(vcpu
) && !is_pae(vcpu
))
3135 r
= mmu_topup_memory_caches(vcpu
);
3139 if (!emulator_write_phys(vcpu
, addr
, &value
, bytes
))
3145 static int kvm_pv_mmu_flush_tlb(struct kvm_vcpu
*vcpu
)
3147 kvm_set_cr3(vcpu
, vcpu
->arch
.cr3
);
3151 static int kvm_pv_mmu_release_pt(struct kvm_vcpu
*vcpu
, gpa_t addr
)
3153 spin_lock(&vcpu
->kvm
->mmu_lock
);
3154 mmu_unshadow(vcpu
->kvm
, addr
>> PAGE_SHIFT
);
3155 spin_unlock(&vcpu
->kvm
->mmu_lock
);
3159 static int kvm_pv_mmu_op_one(struct kvm_vcpu
*vcpu
,
3160 struct kvm_pv_mmu_op_buffer
*buffer
)
3162 struct kvm_mmu_op_header
*header
;
3164 header
= pv_mmu_peek_buffer(buffer
, sizeof *header
);
3167 switch (header
->op
) {
3168 case KVM_MMU_OP_WRITE_PTE
: {
3169 struct kvm_mmu_op_write_pte
*wpte
;
3171 wpte
= pv_mmu_read_buffer(buffer
, sizeof *wpte
);
3174 return kvm_pv_mmu_write(vcpu
, wpte
->pte_phys
,
3177 case KVM_MMU_OP_FLUSH_TLB
: {
3178 struct kvm_mmu_op_flush_tlb
*ftlb
;
3180 ftlb
= pv_mmu_read_buffer(buffer
, sizeof *ftlb
);
3183 return kvm_pv_mmu_flush_tlb(vcpu
);
3185 case KVM_MMU_OP_RELEASE_PT
: {
3186 struct kvm_mmu_op_release_pt
*rpt
;
3188 rpt
= pv_mmu_read_buffer(buffer
, sizeof *rpt
);
3191 return kvm_pv_mmu_release_pt(vcpu
, rpt
->pt_phys
);
3197 int kvm_pv_mmu_op(struct kvm_vcpu
*vcpu
, unsigned long bytes
,
3198 gpa_t addr
, unsigned long *ret
)
3201 struct kvm_pv_mmu_op_buffer
*buffer
= &vcpu
->arch
.mmu_op_buffer
;
3203 buffer
->ptr
= buffer
->buf
;
3204 buffer
->len
= min_t(unsigned long, bytes
, sizeof buffer
->buf
);
3205 buffer
->processed
= 0;
3207 r
= kvm_read_guest(vcpu
->kvm
, addr
, buffer
->buf
, buffer
->len
);
3211 while (buffer
->len
) {
3212 r
= kvm_pv_mmu_op_one(vcpu
, buffer
);
3221 *ret
= buffer
->processed
;
3225 int kvm_mmu_get_spte_hierarchy(struct kvm_vcpu
*vcpu
, u64 addr
, u64 sptes
[4])
3227 struct kvm_shadow_walk_iterator iterator
;
3230 spin_lock(&vcpu
->kvm
->mmu_lock
);
3231 for_each_shadow_entry(vcpu
, addr
, iterator
) {
3232 sptes
[iterator
.level
-1] = *iterator
.sptep
;
3234 if (!is_shadow_present_pte(*iterator
.sptep
))
3237 spin_unlock(&vcpu
->kvm
->mmu_lock
);
3241 EXPORT_SYMBOL_GPL(kvm_mmu_get_spte_hierarchy
);
3245 static const char *audit_msg
;
3247 static gva_t
canonicalize(gva_t gva
)
3249 #ifdef CONFIG_X86_64
3250 gva
= (long long)(gva
<< 16) >> 16;
3256 typedef void (*inspect_spte_fn
) (struct kvm
*kvm
, u64
*sptep
);
3258 static void __mmu_spte_walk(struct kvm
*kvm
, struct kvm_mmu_page
*sp
,
3263 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
3264 u64 ent
= sp
->spt
[i
];
3266 if (is_shadow_present_pte(ent
)) {
3267 if (!is_last_spte(ent
, sp
->role
.level
)) {
3268 struct kvm_mmu_page
*child
;
3269 child
= page_header(ent
& PT64_BASE_ADDR_MASK
);
3270 __mmu_spte_walk(kvm
, child
, fn
);
3272 fn(kvm
, &sp
->spt
[i
]);
3277 static void mmu_spte_walk(struct kvm_vcpu
*vcpu
, inspect_spte_fn fn
)
3280 struct kvm_mmu_page
*sp
;
3282 if (!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
))
3284 if (vcpu
->arch
.mmu
.shadow_root_level
== PT64_ROOT_LEVEL
) {
3285 hpa_t root
= vcpu
->arch
.mmu
.root_hpa
;
3286 sp
= page_header(root
);
3287 __mmu_spte_walk(vcpu
->kvm
, sp
, fn
);
3290 for (i
= 0; i
< 4; ++i
) {
3291 hpa_t root
= vcpu
->arch
.mmu
.pae_root
[i
];
3293 if (root
&& VALID_PAGE(root
)) {
3294 root
&= PT64_BASE_ADDR_MASK
;
3295 sp
= page_header(root
);
3296 __mmu_spte_walk(vcpu
->kvm
, sp
, fn
);
3302 static void audit_mappings_page(struct kvm_vcpu
*vcpu
, u64 page_pte
,
3303 gva_t va
, int level
)
3305 u64
*pt
= __va(page_pte
& PT64_BASE_ADDR_MASK
);
3307 gva_t va_delta
= 1ul << (PAGE_SHIFT
+ 9 * (level
- 1));
3309 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
, va
+= va_delta
) {
3312 if (ent
== shadow_trap_nonpresent_pte
)
3315 va
= canonicalize(va
);
3316 if (is_shadow_present_pte(ent
) && !is_last_spte(ent
, level
))
3317 audit_mappings_page(vcpu
, ent
, va
, level
- 1);
3319 gpa_t gpa
= kvm_mmu_gva_to_gpa_read(vcpu
, va
, NULL
);
3320 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3321 pfn_t pfn
= gfn_to_pfn(vcpu
->kvm
, gfn
);
3322 hpa_t hpa
= (hpa_t
)pfn
<< PAGE_SHIFT
;
3324 if (is_error_pfn(pfn
)) {
3325 kvm_release_pfn_clean(pfn
);
3329 if (is_shadow_present_pte(ent
)
3330 && (ent
& PT64_BASE_ADDR_MASK
) != hpa
)
3331 printk(KERN_ERR
"xx audit error: (%s) levels %d"
3332 " gva %lx gpa %llx hpa %llx ent %llx %d\n",
3333 audit_msg
, vcpu
->arch
.mmu
.root_level
,
3335 is_shadow_present_pte(ent
));
3336 else if (ent
== shadow_notrap_nonpresent_pte
3337 && !is_error_hpa(hpa
))
3338 printk(KERN_ERR
"audit: (%s) notrap shadow,"
3339 " valid guest gva %lx\n", audit_msg
, va
);
3340 kvm_release_pfn_clean(pfn
);
3346 static void audit_mappings(struct kvm_vcpu
*vcpu
)
3350 if (vcpu
->arch
.mmu
.root_level
== 4)
3351 audit_mappings_page(vcpu
, vcpu
->arch
.mmu
.root_hpa
, 0, 4);
3353 for (i
= 0; i
< 4; ++i
)
3354 if (vcpu
->arch
.mmu
.pae_root
[i
] & PT_PRESENT_MASK
)
3355 audit_mappings_page(vcpu
,
3356 vcpu
->arch
.mmu
.pae_root
[i
],
3361 static int count_rmaps(struct kvm_vcpu
*vcpu
)
3363 struct kvm
*kvm
= vcpu
->kvm
;
3364 struct kvm_memslots
*slots
;
3368 idx
= srcu_read_lock(&kvm
->srcu
);
3369 slots
= kvm_memslots(kvm
);
3370 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
3371 struct kvm_memory_slot
*m
= &slots
->memslots
[i
];
3372 struct kvm_rmap_desc
*d
;
3374 for (j
= 0; j
< m
->npages
; ++j
) {
3375 unsigned long *rmapp
= &m
->rmap
[j
];
3379 if (!(*rmapp
& 1)) {
3383 d
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
3385 for (k
= 0; k
< RMAP_EXT
; ++k
)
3394 srcu_read_unlock(&kvm
->srcu
, idx
);
3398 void inspect_spte_has_rmap(struct kvm
*kvm
, u64
*sptep
)
3400 unsigned long *rmapp
;
3401 struct kvm_mmu_page
*rev_sp
;
3404 if (*sptep
& PT_WRITABLE_MASK
) {
3405 rev_sp
= page_header(__pa(sptep
));
3406 gfn
= rev_sp
->gfns
[sptep
- rev_sp
->spt
];
3408 if (!gfn_to_memslot(kvm
, gfn
)) {
3409 if (!printk_ratelimit())
3411 printk(KERN_ERR
"%s: no memslot for gfn %ld\n",
3413 printk(KERN_ERR
"%s: index %ld of sp (gfn=%lx)\n",
3414 audit_msg
, (long int)(sptep
- rev_sp
->spt
),
3420 rmapp
= gfn_to_rmap(kvm
, rev_sp
->gfns
[sptep
- rev_sp
->spt
],
3421 rev_sp
->role
.level
);
3423 if (!printk_ratelimit())
3425 printk(KERN_ERR
"%s: no rmap for writable spte %llx\n",
3433 void audit_writable_sptes_have_rmaps(struct kvm_vcpu
*vcpu
)
3435 mmu_spte_walk(vcpu
, inspect_spte_has_rmap
);
3438 static void check_writable_mappings_rmap(struct kvm_vcpu
*vcpu
)
3440 struct kvm_mmu_page
*sp
;
3443 list_for_each_entry(sp
, &vcpu
->kvm
->arch
.active_mmu_pages
, link
) {
3446 if (sp
->role
.level
!= PT_PAGE_TABLE_LEVEL
)
3449 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
3452 if (!(ent
& PT_PRESENT_MASK
))
3454 if (!(ent
& PT_WRITABLE_MASK
))
3456 inspect_spte_has_rmap(vcpu
->kvm
, &pt
[i
]);
3462 static void audit_rmap(struct kvm_vcpu
*vcpu
)
3464 check_writable_mappings_rmap(vcpu
);
3468 static void audit_write_protection(struct kvm_vcpu
*vcpu
)
3470 struct kvm_mmu_page
*sp
;
3471 struct kvm_memory_slot
*slot
;
3472 unsigned long *rmapp
;
3476 list_for_each_entry(sp
, &vcpu
->kvm
->arch
.active_mmu_pages
, link
) {
3477 if (sp
->role
.direct
)
3482 gfn
= unalias_gfn(vcpu
->kvm
, sp
->gfn
);
3483 slot
= gfn_to_memslot_unaliased(vcpu
->kvm
, sp
->gfn
);
3484 rmapp
= &slot
->rmap
[gfn
- slot
->base_gfn
];
3486 spte
= rmap_next(vcpu
->kvm
, rmapp
, NULL
);
3488 if (*spte
& PT_WRITABLE_MASK
)
3489 printk(KERN_ERR
"%s: (%s) shadow page has "
3490 "writable mappings: gfn %lx role %x\n",
3491 __func__
, audit_msg
, sp
->gfn
,
3493 spte
= rmap_next(vcpu
->kvm
, rmapp
, spte
);
3498 static void kvm_mmu_audit(struct kvm_vcpu
*vcpu
, const char *msg
)
3505 audit_write_protection(vcpu
);
3506 if (strcmp("pre pte write", audit_msg
) != 0)
3507 audit_mappings(vcpu
);
3508 audit_writable_sptes_have_rmaps(vcpu
);