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 void (*mmu_parent_walk_fn
) (struct kvm_mmu_page
*sp
, u64
*spte
);
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
);
400 static gfn_t
kvm_mmu_page_get_gfn(struct kvm_mmu_page
*sp
, int index
)
402 if (!sp
->role
.direct
)
403 return sp
->gfns
[index
];
405 return sp
->gfn
+ (index
<< ((sp
->role
.level
- 1) * PT64_LEVEL_BITS
));
408 static void kvm_mmu_page_set_gfn(struct kvm_mmu_page
*sp
, int index
, gfn_t gfn
)
411 BUG_ON(gfn
!= kvm_mmu_page_get_gfn(sp
, index
));
413 sp
->gfns
[index
] = gfn
;
417 * Return the pointer to the largepage write count for a given
418 * gfn, handling slots that are not large page aligned.
420 static int *slot_largepage_idx(gfn_t gfn
,
421 struct kvm_memory_slot
*slot
,
426 idx
= (gfn
>> KVM_HPAGE_GFN_SHIFT(level
)) -
427 (slot
->base_gfn
>> KVM_HPAGE_GFN_SHIFT(level
));
428 return &slot
->lpage_info
[level
- 2][idx
].write_count
;
431 static void account_shadowed(struct kvm
*kvm
, gfn_t gfn
)
433 struct kvm_memory_slot
*slot
;
437 slot
= gfn_to_memslot(kvm
, gfn
);
438 for (i
= PT_DIRECTORY_LEVEL
;
439 i
< PT_PAGE_TABLE_LEVEL
+ KVM_NR_PAGE_SIZES
; ++i
) {
440 write_count
= slot_largepage_idx(gfn
, slot
, i
);
445 static void unaccount_shadowed(struct kvm
*kvm
, gfn_t gfn
)
447 struct kvm_memory_slot
*slot
;
451 slot
= gfn_to_memslot(kvm
, gfn
);
452 for (i
= PT_DIRECTORY_LEVEL
;
453 i
< PT_PAGE_TABLE_LEVEL
+ KVM_NR_PAGE_SIZES
; ++i
) {
454 write_count
= slot_largepage_idx(gfn
, slot
, i
);
456 WARN_ON(*write_count
< 0);
460 static int has_wrprotected_page(struct kvm
*kvm
,
464 struct kvm_memory_slot
*slot
;
467 slot
= gfn_to_memslot(kvm
, gfn
);
469 largepage_idx
= slot_largepage_idx(gfn
, slot
, level
);
470 return *largepage_idx
;
476 static int host_mapping_level(struct kvm
*kvm
, gfn_t gfn
)
478 unsigned long page_size
;
481 page_size
= kvm_host_page_size(kvm
, gfn
);
483 for (i
= PT_PAGE_TABLE_LEVEL
;
484 i
< (PT_PAGE_TABLE_LEVEL
+ KVM_NR_PAGE_SIZES
); ++i
) {
485 if (page_size
>= KVM_HPAGE_SIZE(i
))
494 static int mapping_level(struct kvm_vcpu
*vcpu
, gfn_t large_gfn
)
496 struct kvm_memory_slot
*slot
;
497 int host_level
, level
, max_level
;
499 slot
= gfn_to_memslot(vcpu
->kvm
, large_gfn
);
500 if (slot
&& slot
->dirty_bitmap
)
501 return PT_PAGE_TABLE_LEVEL
;
503 host_level
= host_mapping_level(vcpu
->kvm
, large_gfn
);
505 if (host_level
== PT_PAGE_TABLE_LEVEL
)
508 max_level
= kvm_x86_ops
->get_lpage_level() < host_level
?
509 kvm_x86_ops
->get_lpage_level() : host_level
;
511 for (level
= PT_DIRECTORY_LEVEL
; level
<= max_level
; ++level
)
512 if (has_wrprotected_page(vcpu
->kvm
, large_gfn
, level
))
519 * Take gfn and return the reverse mapping to it.
522 static unsigned long *gfn_to_rmap(struct kvm
*kvm
, gfn_t gfn
, int level
)
524 struct kvm_memory_slot
*slot
;
527 slot
= gfn_to_memslot(kvm
, gfn
);
528 if (likely(level
== PT_PAGE_TABLE_LEVEL
))
529 return &slot
->rmap
[gfn
- slot
->base_gfn
];
531 idx
= (gfn
>> KVM_HPAGE_GFN_SHIFT(level
)) -
532 (slot
->base_gfn
>> KVM_HPAGE_GFN_SHIFT(level
));
534 return &slot
->lpage_info
[level
- 2][idx
].rmap_pde
;
538 * Reverse mapping data structures:
540 * If rmapp bit zero is zero, then rmapp point to the shadw page table entry
541 * that points to page_address(page).
543 * If rmapp bit zero is one, (then rmap & ~1) points to a struct kvm_rmap_desc
544 * containing more mappings.
546 * Returns the number of rmap entries before the spte was added or zero if
547 * the spte was not added.
550 static int rmap_add(struct kvm_vcpu
*vcpu
, u64
*spte
, gfn_t gfn
)
552 struct kvm_mmu_page
*sp
;
553 struct kvm_rmap_desc
*desc
;
554 unsigned long *rmapp
;
557 if (!is_rmap_spte(*spte
))
559 sp
= page_header(__pa(spte
));
560 kvm_mmu_page_set_gfn(sp
, spte
- sp
->spt
, gfn
);
561 rmapp
= gfn_to_rmap(vcpu
->kvm
, gfn
, sp
->role
.level
);
563 rmap_printk("rmap_add: %p %llx 0->1\n", spte
, *spte
);
564 *rmapp
= (unsigned long)spte
;
565 } else if (!(*rmapp
& 1)) {
566 rmap_printk("rmap_add: %p %llx 1->many\n", spte
, *spte
);
567 desc
= mmu_alloc_rmap_desc(vcpu
);
568 desc
->sptes
[0] = (u64
*)*rmapp
;
569 desc
->sptes
[1] = spte
;
570 *rmapp
= (unsigned long)desc
| 1;
572 rmap_printk("rmap_add: %p %llx many->many\n", spte
, *spte
);
573 desc
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
574 while (desc
->sptes
[RMAP_EXT
-1] && desc
->more
) {
578 if (desc
->sptes
[RMAP_EXT
-1]) {
579 desc
->more
= mmu_alloc_rmap_desc(vcpu
);
582 for (i
= 0; desc
->sptes
[i
]; ++i
)
584 desc
->sptes
[i
] = spte
;
589 static void rmap_desc_remove_entry(unsigned long *rmapp
,
590 struct kvm_rmap_desc
*desc
,
592 struct kvm_rmap_desc
*prev_desc
)
596 for (j
= RMAP_EXT
- 1; !desc
->sptes
[j
] && j
> i
; --j
)
598 desc
->sptes
[i
] = desc
->sptes
[j
];
599 desc
->sptes
[j
] = NULL
;
602 if (!prev_desc
&& !desc
->more
)
603 *rmapp
= (unsigned long)desc
->sptes
[0];
606 prev_desc
->more
= desc
->more
;
608 *rmapp
= (unsigned long)desc
->more
| 1;
609 mmu_free_rmap_desc(desc
);
612 static void rmap_remove(struct kvm
*kvm
, u64
*spte
)
614 struct kvm_rmap_desc
*desc
;
615 struct kvm_rmap_desc
*prev_desc
;
616 struct kvm_mmu_page
*sp
;
619 unsigned long *rmapp
;
622 if (!is_rmap_spte(*spte
))
624 sp
= page_header(__pa(spte
));
625 pfn
= spte_to_pfn(*spte
);
626 if (*spte
& shadow_accessed_mask
)
627 kvm_set_pfn_accessed(pfn
);
628 if (is_writable_pte(*spte
))
629 kvm_set_pfn_dirty(pfn
);
630 gfn
= kvm_mmu_page_get_gfn(sp
, spte
- sp
->spt
);
631 rmapp
= gfn_to_rmap(kvm
, gfn
, sp
->role
.level
);
633 printk(KERN_ERR
"rmap_remove: %p %llx 0->BUG\n", spte
, *spte
);
635 } else if (!(*rmapp
& 1)) {
636 rmap_printk("rmap_remove: %p %llx 1->0\n", spte
, *spte
);
637 if ((u64
*)*rmapp
!= spte
) {
638 printk(KERN_ERR
"rmap_remove: %p %llx 1->BUG\n",
644 rmap_printk("rmap_remove: %p %llx many->many\n", spte
, *spte
);
645 desc
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
648 for (i
= 0; i
< RMAP_EXT
&& desc
->sptes
[i
]; ++i
)
649 if (desc
->sptes
[i
] == spte
) {
650 rmap_desc_remove_entry(rmapp
,
658 pr_err("rmap_remove: %p %llx many->many\n", spte
, *spte
);
663 static u64
*rmap_next(struct kvm
*kvm
, unsigned long *rmapp
, u64
*spte
)
665 struct kvm_rmap_desc
*desc
;
671 else if (!(*rmapp
& 1)) {
673 return (u64
*)*rmapp
;
676 desc
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
679 for (i
= 0; i
< RMAP_EXT
&& desc
->sptes
[i
]; ++i
) {
680 if (prev_spte
== spte
)
681 return desc
->sptes
[i
];
682 prev_spte
= desc
->sptes
[i
];
689 static int rmap_write_protect(struct kvm
*kvm
, u64 gfn
)
691 unsigned long *rmapp
;
693 int i
, write_protected
= 0;
695 rmapp
= gfn_to_rmap(kvm
, gfn
, PT_PAGE_TABLE_LEVEL
);
697 spte
= rmap_next(kvm
, rmapp
, NULL
);
700 BUG_ON(!(*spte
& PT_PRESENT_MASK
));
701 rmap_printk("rmap_write_protect: spte %p %llx\n", spte
, *spte
);
702 if (is_writable_pte(*spte
)) {
703 __set_spte(spte
, *spte
& ~PT_WRITABLE_MASK
);
706 spte
= rmap_next(kvm
, rmapp
, spte
);
708 if (write_protected
) {
711 spte
= rmap_next(kvm
, rmapp
, NULL
);
712 pfn
= spte_to_pfn(*spte
);
713 kvm_set_pfn_dirty(pfn
);
716 /* check for huge page mappings */
717 for (i
= PT_DIRECTORY_LEVEL
;
718 i
< PT_PAGE_TABLE_LEVEL
+ KVM_NR_PAGE_SIZES
; ++i
) {
719 rmapp
= gfn_to_rmap(kvm
, gfn
, i
);
720 spte
= rmap_next(kvm
, rmapp
, NULL
);
723 BUG_ON(!(*spte
& PT_PRESENT_MASK
));
724 BUG_ON((*spte
& (PT_PAGE_SIZE_MASK
|PT_PRESENT_MASK
)) != (PT_PAGE_SIZE_MASK
|PT_PRESENT_MASK
));
725 pgprintk("rmap_write_protect(large): spte %p %llx %lld\n", spte
, *spte
, gfn
);
726 if (is_writable_pte(*spte
)) {
727 rmap_remove(kvm
, spte
);
729 __set_spte(spte
, shadow_trap_nonpresent_pte
);
733 spte
= rmap_next(kvm
, rmapp
, spte
);
737 return write_protected
;
740 static int kvm_unmap_rmapp(struct kvm
*kvm
, unsigned long *rmapp
,
744 int need_tlb_flush
= 0;
746 while ((spte
= rmap_next(kvm
, rmapp
, NULL
))) {
747 BUG_ON(!(*spte
& PT_PRESENT_MASK
));
748 rmap_printk("kvm_rmap_unmap_hva: spte %p %llx\n", spte
, *spte
);
749 rmap_remove(kvm
, spte
);
750 __set_spte(spte
, shadow_trap_nonpresent_pte
);
753 return need_tlb_flush
;
756 static int kvm_set_pte_rmapp(struct kvm
*kvm
, unsigned long *rmapp
,
761 pte_t
*ptep
= (pte_t
*)data
;
764 WARN_ON(pte_huge(*ptep
));
765 new_pfn
= pte_pfn(*ptep
);
766 spte
= rmap_next(kvm
, rmapp
, NULL
);
768 BUG_ON(!is_shadow_present_pte(*spte
));
769 rmap_printk("kvm_set_pte_rmapp: spte %p %llx\n", spte
, *spte
);
771 if (pte_write(*ptep
)) {
772 rmap_remove(kvm
, spte
);
773 __set_spte(spte
, shadow_trap_nonpresent_pte
);
774 spte
= rmap_next(kvm
, rmapp
, NULL
);
776 new_spte
= *spte
&~ (PT64_BASE_ADDR_MASK
);
777 new_spte
|= (u64
)new_pfn
<< PAGE_SHIFT
;
779 new_spte
&= ~PT_WRITABLE_MASK
;
780 new_spte
&= ~SPTE_HOST_WRITEABLE
;
781 if (is_writable_pte(*spte
))
782 kvm_set_pfn_dirty(spte_to_pfn(*spte
));
783 __set_spte(spte
, new_spte
);
784 spte
= rmap_next(kvm
, rmapp
, spte
);
788 kvm_flush_remote_tlbs(kvm
);
793 static int kvm_handle_hva(struct kvm
*kvm
, unsigned long hva
,
795 int (*handler
)(struct kvm
*kvm
, unsigned long *rmapp
,
801 struct kvm_memslots
*slots
;
803 slots
= kvm_memslots(kvm
);
805 for (i
= 0; i
< slots
->nmemslots
; i
++) {
806 struct kvm_memory_slot
*memslot
= &slots
->memslots
[i
];
807 unsigned long start
= memslot
->userspace_addr
;
810 end
= start
+ (memslot
->npages
<< PAGE_SHIFT
);
811 if (hva
>= start
&& hva
< end
) {
812 gfn_t gfn_offset
= (hva
- start
) >> PAGE_SHIFT
;
814 ret
= handler(kvm
, &memslot
->rmap
[gfn_offset
], data
);
816 for (j
= 0; j
< KVM_NR_PAGE_SIZES
- 1; ++j
) {
817 int idx
= gfn_offset
;
818 idx
/= KVM_PAGES_PER_HPAGE(PT_DIRECTORY_LEVEL
+ j
);
820 &memslot
->lpage_info
[j
][idx
].rmap_pde
,
823 trace_kvm_age_page(hva
, memslot
, ret
);
831 int kvm_unmap_hva(struct kvm
*kvm
, unsigned long hva
)
833 return kvm_handle_hva(kvm
, hva
, 0, kvm_unmap_rmapp
);
836 void kvm_set_spte_hva(struct kvm
*kvm
, unsigned long hva
, pte_t pte
)
838 kvm_handle_hva(kvm
, hva
, (unsigned long)&pte
, kvm_set_pte_rmapp
);
841 static int kvm_age_rmapp(struct kvm
*kvm
, unsigned long *rmapp
,
848 * Emulate the accessed bit for EPT, by checking if this page has
849 * an EPT mapping, and clearing it if it does. On the next access,
850 * a new EPT mapping will be established.
851 * This has some overhead, but not as much as the cost of swapping
852 * out actively used pages or breaking up actively used hugepages.
854 if (!shadow_accessed_mask
)
855 return kvm_unmap_rmapp(kvm
, rmapp
, data
);
857 spte
= rmap_next(kvm
, rmapp
, NULL
);
861 BUG_ON(!(_spte
& PT_PRESENT_MASK
));
862 _young
= _spte
& PT_ACCESSED_MASK
;
865 clear_bit(PT_ACCESSED_SHIFT
, (unsigned long *)spte
);
867 spte
= rmap_next(kvm
, rmapp
, spte
);
872 #define RMAP_RECYCLE_THRESHOLD 1000
874 static void rmap_recycle(struct kvm_vcpu
*vcpu
, u64
*spte
, gfn_t gfn
)
876 unsigned long *rmapp
;
877 struct kvm_mmu_page
*sp
;
879 sp
= page_header(__pa(spte
));
881 rmapp
= gfn_to_rmap(vcpu
->kvm
, gfn
, sp
->role
.level
);
883 kvm_unmap_rmapp(vcpu
->kvm
, rmapp
, 0);
884 kvm_flush_remote_tlbs(vcpu
->kvm
);
887 int kvm_age_hva(struct kvm
*kvm
, unsigned long hva
)
889 return kvm_handle_hva(kvm
, hva
, 0, kvm_age_rmapp
);
893 static int is_empty_shadow_page(u64
*spt
)
898 for (pos
= spt
, end
= pos
+ PAGE_SIZE
/ sizeof(u64
); pos
!= end
; pos
++)
899 if (is_shadow_present_pte(*pos
)) {
900 printk(KERN_ERR
"%s: %p %llx\n", __func__
,
908 static void kvm_mmu_free_page(struct kvm
*kvm
, struct kvm_mmu_page
*sp
)
910 ASSERT(is_empty_shadow_page(sp
->spt
));
911 hlist_del(&sp
->hash_link
);
913 __free_page(virt_to_page(sp
->spt
));
914 if (!sp
->role
.direct
)
915 __free_page(virt_to_page(sp
->gfns
));
916 kmem_cache_free(mmu_page_header_cache
, sp
);
917 ++kvm
->arch
.n_free_mmu_pages
;
920 static unsigned kvm_page_table_hashfn(gfn_t gfn
)
922 return gfn
& ((1 << KVM_MMU_HASH_SHIFT
) - 1);
925 static struct kvm_mmu_page
*kvm_mmu_alloc_page(struct kvm_vcpu
*vcpu
,
926 u64
*parent_pte
, int direct
)
928 struct kvm_mmu_page
*sp
;
930 sp
= mmu_memory_cache_alloc(&vcpu
->arch
.mmu_page_header_cache
, sizeof *sp
);
931 sp
->spt
= mmu_memory_cache_alloc(&vcpu
->arch
.mmu_page_cache
, PAGE_SIZE
);
933 sp
->gfns
= mmu_memory_cache_alloc(&vcpu
->arch
.mmu_page_cache
,
935 set_page_private(virt_to_page(sp
->spt
), (unsigned long)sp
);
936 list_add(&sp
->link
, &vcpu
->kvm
->arch
.active_mmu_pages
);
937 bitmap_zero(sp
->slot_bitmap
, KVM_MEMORY_SLOTS
+ KVM_PRIVATE_MEM_SLOTS
);
939 sp
->parent_pte
= parent_pte
;
940 --vcpu
->kvm
->arch
.n_free_mmu_pages
;
944 static void mmu_page_add_parent_pte(struct kvm_vcpu
*vcpu
,
945 struct kvm_mmu_page
*sp
, u64
*parent_pte
)
947 struct kvm_pte_chain
*pte_chain
;
948 struct hlist_node
*node
;
953 if (!sp
->multimapped
) {
954 u64
*old
= sp
->parent_pte
;
957 sp
->parent_pte
= parent_pte
;
961 pte_chain
= mmu_alloc_pte_chain(vcpu
);
962 INIT_HLIST_HEAD(&sp
->parent_ptes
);
963 hlist_add_head(&pte_chain
->link
, &sp
->parent_ptes
);
964 pte_chain
->parent_ptes
[0] = old
;
966 hlist_for_each_entry(pte_chain
, node
, &sp
->parent_ptes
, link
) {
967 if (pte_chain
->parent_ptes
[NR_PTE_CHAIN_ENTRIES
-1])
969 for (i
= 0; i
< NR_PTE_CHAIN_ENTRIES
; ++i
)
970 if (!pte_chain
->parent_ptes
[i
]) {
971 pte_chain
->parent_ptes
[i
] = parent_pte
;
975 pte_chain
= mmu_alloc_pte_chain(vcpu
);
977 hlist_add_head(&pte_chain
->link
, &sp
->parent_ptes
);
978 pte_chain
->parent_ptes
[0] = parent_pte
;
981 static void mmu_page_remove_parent_pte(struct kvm_mmu_page
*sp
,
984 struct kvm_pte_chain
*pte_chain
;
985 struct hlist_node
*node
;
988 if (!sp
->multimapped
) {
989 BUG_ON(sp
->parent_pte
!= parent_pte
);
990 sp
->parent_pte
= NULL
;
993 hlist_for_each_entry(pte_chain
, node
, &sp
->parent_ptes
, link
)
994 for (i
= 0; i
< NR_PTE_CHAIN_ENTRIES
; ++i
) {
995 if (!pte_chain
->parent_ptes
[i
])
997 if (pte_chain
->parent_ptes
[i
] != parent_pte
)
999 while (i
+ 1 < NR_PTE_CHAIN_ENTRIES
1000 && pte_chain
->parent_ptes
[i
+ 1]) {
1001 pte_chain
->parent_ptes
[i
]
1002 = pte_chain
->parent_ptes
[i
+ 1];
1005 pte_chain
->parent_ptes
[i
] = NULL
;
1007 hlist_del(&pte_chain
->link
);
1008 mmu_free_pte_chain(pte_chain
);
1009 if (hlist_empty(&sp
->parent_ptes
)) {
1010 sp
->multimapped
= 0;
1011 sp
->parent_pte
= NULL
;
1019 static void mmu_parent_walk(struct kvm_mmu_page
*sp
, mmu_parent_walk_fn fn
)
1021 struct kvm_pte_chain
*pte_chain
;
1022 struct hlist_node
*node
;
1023 struct kvm_mmu_page
*parent_sp
;
1026 if (!sp
->multimapped
&& sp
->parent_pte
) {
1027 parent_sp
= page_header(__pa(sp
->parent_pte
));
1028 fn(parent_sp
, sp
->parent_pte
);
1032 hlist_for_each_entry(pte_chain
, node
, &sp
->parent_ptes
, link
)
1033 for (i
= 0; i
< NR_PTE_CHAIN_ENTRIES
; ++i
) {
1034 u64
*spte
= pte_chain
->parent_ptes
[i
];
1038 parent_sp
= page_header(__pa(spte
));
1039 fn(parent_sp
, spte
);
1043 static void mark_unsync(struct kvm_mmu_page
*sp
, u64
*spte
);
1044 static void kvm_mmu_mark_parents_unsync(struct kvm_mmu_page
*sp
)
1046 mmu_parent_walk(sp
, mark_unsync
);
1049 static void mark_unsync(struct kvm_mmu_page
*sp
, u64
*spte
)
1053 index
= spte
- sp
->spt
;
1054 if (__test_and_set_bit(index
, sp
->unsync_child_bitmap
))
1056 if (sp
->unsync_children
++)
1058 kvm_mmu_mark_parents_unsync(sp
);
1061 static void nonpaging_prefetch_page(struct kvm_vcpu
*vcpu
,
1062 struct kvm_mmu_page
*sp
)
1066 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
)
1067 sp
->spt
[i
] = shadow_trap_nonpresent_pte
;
1070 static int nonpaging_sync_page(struct kvm_vcpu
*vcpu
,
1071 struct kvm_mmu_page
*sp
, bool clear_unsync
)
1076 static void nonpaging_invlpg(struct kvm_vcpu
*vcpu
, gva_t gva
)
1080 #define KVM_PAGE_ARRAY_NR 16
1082 struct kvm_mmu_pages
{
1083 struct mmu_page_and_offset
{
1084 struct kvm_mmu_page
*sp
;
1086 } page
[KVM_PAGE_ARRAY_NR
];
1090 #define for_each_unsync_children(bitmap, idx) \
1091 for (idx = find_first_bit(bitmap, 512); \
1093 idx = find_next_bit(bitmap, 512, idx+1))
1095 static int mmu_pages_add(struct kvm_mmu_pages
*pvec
, struct kvm_mmu_page
*sp
,
1101 for (i
=0; i
< pvec
->nr
; i
++)
1102 if (pvec
->page
[i
].sp
== sp
)
1105 pvec
->page
[pvec
->nr
].sp
= sp
;
1106 pvec
->page
[pvec
->nr
].idx
= idx
;
1108 return (pvec
->nr
== KVM_PAGE_ARRAY_NR
);
1111 static int __mmu_unsync_walk(struct kvm_mmu_page
*sp
,
1112 struct kvm_mmu_pages
*pvec
)
1114 int i
, ret
, nr_unsync_leaf
= 0;
1116 for_each_unsync_children(sp
->unsync_child_bitmap
, i
) {
1117 struct kvm_mmu_page
*child
;
1118 u64 ent
= sp
->spt
[i
];
1120 if (!is_shadow_present_pte(ent
) || is_large_pte(ent
))
1121 goto clear_child_bitmap
;
1123 child
= page_header(ent
& PT64_BASE_ADDR_MASK
);
1125 if (child
->unsync_children
) {
1126 if (mmu_pages_add(pvec
, child
, i
))
1129 ret
= __mmu_unsync_walk(child
, pvec
);
1131 goto clear_child_bitmap
;
1133 nr_unsync_leaf
+= ret
;
1136 } else if (child
->unsync
) {
1138 if (mmu_pages_add(pvec
, child
, i
))
1141 goto clear_child_bitmap
;
1146 __clear_bit(i
, sp
->unsync_child_bitmap
);
1147 sp
->unsync_children
--;
1148 WARN_ON((int)sp
->unsync_children
< 0);
1152 return nr_unsync_leaf
;
1155 static int mmu_unsync_walk(struct kvm_mmu_page
*sp
,
1156 struct kvm_mmu_pages
*pvec
)
1158 if (!sp
->unsync_children
)
1161 mmu_pages_add(pvec
, sp
, 0);
1162 return __mmu_unsync_walk(sp
, pvec
);
1165 static void kvm_unlink_unsync_page(struct kvm
*kvm
, struct kvm_mmu_page
*sp
)
1167 WARN_ON(!sp
->unsync
);
1168 trace_kvm_mmu_sync_page(sp
);
1170 --kvm
->stat
.mmu_unsync
;
1173 static int kvm_mmu_prepare_zap_page(struct kvm
*kvm
, struct kvm_mmu_page
*sp
,
1174 struct list_head
*invalid_list
);
1175 static void kvm_mmu_commit_zap_page(struct kvm
*kvm
,
1176 struct list_head
*invalid_list
);
1178 #define for_each_gfn_sp(kvm, sp, gfn, pos) \
1179 hlist_for_each_entry(sp, pos, \
1180 &(kvm)->arch.mmu_page_hash[kvm_page_table_hashfn(gfn)], hash_link) \
1181 if ((sp)->gfn != (gfn)) {} else
1183 #define for_each_gfn_indirect_valid_sp(kvm, sp, gfn, pos) \
1184 hlist_for_each_entry(sp, pos, \
1185 &(kvm)->arch.mmu_page_hash[kvm_page_table_hashfn(gfn)], hash_link) \
1186 if ((sp)->gfn != (gfn) || (sp)->role.direct || \
1187 (sp)->role.invalid) {} else
1189 /* @sp->gfn should be write-protected at the call site */
1190 static int __kvm_sync_page(struct kvm_vcpu
*vcpu
, struct kvm_mmu_page
*sp
,
1191 struct list_head
*invalid_list
, bool clear_unsync
)
1193 if (sp
->role
.cr4_pae
!= !!is_pae(vcpu
)) {
1194 kvm_mmu_prepare_zap_page(vcpu
->kvm
, sp
, invalid_list
);
1199 kvm_unlink_unsync_page(vcpu
->kvm
, sp
);
1201 if (vcpu
->arch
.mmu
.sync_page(vcpu
, sp
, clear_unsync
)) {
1202 kvm_mmu_prepare_zap_page(vcpu
->kvm
, sp
, invalid_list
);
1206 kvm_mmu_flush_tlb(vcpu
);
1210 static int kvm_sync_page_transient(struct kvm_vcpu
*vcpu
,
1211 struct kvm_mmu_page
*sp
)
1213 LIST_HEAD(invalid_list
);
1216 ret
= __kvm_sync_page(vcpu
, sp
, &invalid_list
, false);
1218 kvm_mmu_commit_zap_page(vcpu
->kvm
, &invalid_list
);
1223 static int kvm_sync_page(struct kvm_vcpu
*vcpu
, struct kvm_mmu_page
*sp
,
1224 struct list_head
*invalid_list
)
1226 return __kvm_sync_page(vcpu
, sp
, invalid_list
, true);
1229 /* @gfn should be write-protected at the call site */
1230 static void kvm_sync_pages(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1232 struct kvm_mmu_page
*s
;
1233 struct hlist_node
*node
;
1234 LIST_HEAD(invalid_list
);
1237 for_each_gfn_indirect_valid_sp(vcpu
->kvm
, s
, gfn
, node
) {
1241 WARN_ON(s
->role
.level
!= PT_PAGE_TABLE_LEVEL
);
1242 if ((s
->role
.cr4_pae
!= !!is_pae(vcpu
)) ||
1243 (vcpu
->arch
.mmu
.sync_page(vcpu
, s
, true))) {
1244 kvm_mmu_prepare_zap_page(vcpu
->kvm
, s
, &invalid_list
);
1247 kvm_unlink_unsync_page(vcpu
->kvm
, s
);
1251 kvm_mmu_commit_zap_page(vcpu
->kvm
, &invalid_list
);
1253 kvm_mmu_flush_tlb(vcpu
);
1256 struct mmu_page_path
{
1257 struct kvm_mmu_page
*parent
[PT64_ROOT_LEVEL
-1];
1258 unsigned int idx
[PT64_ROOT_LEVEL
-1];
1261 #define for_each_sp(pvec, sp, parents, i) \
1262 for (i = mmu_pages_next(&pvec, &parents, -1), \
1263 sp = pvec.page[i].sp; \
1264 i < pvec.nr && ({ sp = pvec.page[i].sp; 1;}); \
1265 i = mmu_pages_next(&pvec, &parents, i))
1267 static int mmu_pages_next(struct kvm_mmu_pages
*pvec
,
1268 struct mmu_page_path
*parents
,
1273 for (n
= i
+1; n
< pvec
->nr
; n
++) {
1274 struct kvm_mmu_page
*sp
= pvec
->page
[n
].sp
;
1276 if (sp
->role
.level
== PT_PAGE_TABLE_LEVEL
) {
1277 parents
->idx
[0] = pvec
->page
[n
].idx
;
1281 parents
->parent
[sp
->role
.level
-2] = sp
;
1282 parents
->idx
[sp
->role
.level
-1] = pvec
->page
[n
].idx
;
1288 static void mmu_pages_clear_parents(struct mmu_page_path
*parents
)
1290 struct kvm_mmu_page
*sp
;
1291 unsigned int level
= 0;
1294 unsigned int idx
= parents
->idx
[level
];
1296 sp
= parents
->parent
[level
];
1300 --sp
->unsync_children
;
1301 WARN_ON((int)sp
->unsync_children
< 0);
1302 __clear_bit(idx
, sp
->unsync_child_bitmap
);
1304 } while (level
< PT64_ROOT_LEVEL
-1 && !sp
->unsync_children
);
1307 static void kvm_mmu_pages_init(struct kvm_mmu_page
*parent
,
1308 struct mmu_page_path
*parents
,
1309 struct kvm_mmu_pages
*pvec
)
1311 parents
->parent
[parent
->role
.level
-1] = NULL
;
1315 static void mmu_sync_children(struct kvm_vcpu
*vcpu
,
1316 struct kvm_mmu_page
*parent
)
1319 struct kvm_mmu_page
*sp
;
1320 struct mmu_page_path parents
;
1321 struct kvm_mmu_pages pages
;
1322 LIST_HEAD(invalid_list
);
1324 kvm_mmu_pages_init(parent
, &parents
, &pages
);
1325 while (mmu_unsync_walk(parent
, &pages
)) {
1328 for_each_sp(pages
, sp
, parents
, i
)
1329 protected |= rmap_write_protect(vcpu
->kvm
, sp
->gfn
);
1332 kvm_flush_remote_tlbs(vcpu
->kvm
);
1334 for_each_sp(pages
, sp
, parents
, i
) {
1335 kvm_sync_page(vcpu
, sp
, &invalid_list
);
1336 mmu_pages_clear_parents(&parents
);
1338 kvm_mmu_commit_zap_page(vcpu
->kvm
, &invalid_list
);
1339 cond_resched_lock(&vcpu
->kvm
->mmu_lock
);
1340 kvm_mmu_pages_init(parent
, &parents
, &pages
);
1344 static struct kvm_mmu_page
*kvm_mmu_get_page(struct kvm_vcpu
*vcpu
,
1352 union kvm_mmu_page_role role
;
1354 struct kvm_mmu_page
*sp
;
1355 struct hlist_node
*node
;
1356 bool need_sync
= false;
1358 role
= vcpu
->arch
.mmu
.base_role
;
1360 role
.direct
= direct
;
1363 role
.access
= access
;
1364 if (!tdp_enabled
&& vcpu
->arch
.mmu
.root_level
<= PT32_ROOT_LEVEL
) {
1365 quadrant
= gaddr
>> (PAGE_SHIFT
+ (PT64_PT_BITS
* level
));
1366 quadrant
&= (1 << ((PT32_PT_BITS
- PT64_PT_BITS
) * level
)) - 1;
1367 role
.quadrant
= quadrant
;
1369 for_each_gfn_sp(vcpu
->kvm
, sp
, gfn
, node
) {
1370 if (!need_sync
&& sp
->unsync
)
1373 if (sp
->role
.word
!= role
.word
)
1376 if (sp
->unsync
&& kvm_sync_page_transient(vcpu
, sp
))
1379 mmu_page_add_parent_pte(vcpu
, sp
, parent_pte
);
1380 if (sp
->unsync_children
) {
1381 kvm_make_request(KVM_REQ_MMU_SYNC
, vcpu
);
1382 kvm_mmu_mark_parents_unsync(sp
);
1383 } else if (sp
->unsync
)
1384 kvm_mmu_mark_parents_unsync(sp
);
1386 trace_kvm_mmu_get_page(sp
, false);
1389 ++vcpu
->kvm
->stat
.mmu_cache_miss
;
1390 sp
= kvm_mmu_alloc_page(vcpu
, parent_pte
, direct
);
1395 hlist_add_head(&sp
->hash_link
,
1396 &vcpu
->kvm
->arch
.mmu_page_hash
[kvm_page_table_hashfn(gfn
)]);
1398 if (rmap_write_protect(vcpu
->kvm
, gfn
))
1399 kvm_flush_remote_tlbs(vcpu
->kvm
);
1400 if (level
> PT_PAGE_TABLE_LEVEL
&& need_sync
)
1401 kvm_sync_pages(vcpu
, gfn
);
1403 account_shadowed(vcpu
->kvm
, gfn
);
1405 if (shadow_trap_nonpresent_pte
!= shadow_notrap_nonpresent_pte
)
1406 vcpu
->arch
.mmu
.prefetch_page(vcpu
, sp
);
1408 nonpaging_prefetch_page(vcpu
, sp
);
1409 trace_kvm_mmu_get_page(sp
, true);
1413 static void shadow_walk_init(struct kvm_shadow_walk_iterator
*iterator
,
1414 struct kvm_vcpu
*vcpu
, u64 addr
)
1416 iterator
->addr
= addr
;
1417 iterator
->shadow_addr
= vcpu
->arch
.mmu
.root_hpa
;
1418 iterator
->level
= vcpu
->arch
.mmu
.shadow_root_level
;
1419 if (iterator
->level
== PT32E_ROOT_LEVEL
) {
1420 iterator
->shadow_addr
1421 = vcpu
->arch
.mmu
.pae_root
[(addr
>> 30) & 3];
1422 iterator
->shadow_addr
&= PT64_BASE_ADDR_MASK
;
1424 if (!iterator
->shadow_addr
)
1425 iterator
->level
= 0;
1429 static bool shadow_walk_okay(struct kvm_shadow_walk_iterator
*iterator
)
1431 if (iterator
->level
< PT_PAGE_TABLE_LEVEL
)
1434 if (iterator
->level
== PT_PAGE_TABLE_LEVEL
)
1435 if (is_large_pte(*iterator
->sptep
))
1438 iterator
->index
= SHADOW_PT_INDEX(iterator
->addr
, iterator
->level
);
1439 iterator
->sptep
= ((u64
*)__va(iterator
->shadow_addr
)) + iterator
->index
;
1443 static void shadow_walk_next(struct kvm_shadow_walk_iterator
*iterator
)
1445 iterator
->shadow_addr
= *iterator
->sptep
& PT64_BASE_ADDR_MASK
;
1449 static void kvm_mmu_page_unlink_children(struct kvm
*kvm
,
1450 struct kvm_mmu_page
*sp
)
1458 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
1461 if (is_shadow_present_pte(ent
)) {
1462 if (!is_last_spte(ent
, sp
->role
.level
)) {
1463 ent
&= PT64_BASE_ADDR_MASK
;
1464 mmu_page_remove_parent_pte(page_header(ent
),
1467 if (is_large_pte(ent
))
1469 rmap_remove(kvm
, &pt
[i
]);
1472 pt
[i
] = shadow_trap_nonpresent_pte
;
1476 static void kvm_mmu_put_page(struct kvm_mmu_page
*sp
, u64
*parent_pte
)
1478 mmu_page_remove_parent_pte(sp
, parent_pte
);
1481 static void kvm_mmu_reset_last_pte_updated(struct kvm
*kvm
)
1484 struct kvm_vcpu
*vcpu
;
1486 kvm_for_each_vcpu(i
, vcpu
, kvm
)
1487 vcpu
->arch
.last_pte_updated
= NULL
;
1490 static void kvm_mmu_unlink_parents(struct kvm
*kvm
, struct kvm_mmu_page
*sp
)
1494 while (sp
->multimapped
|| sp
->parent_pte
) {
1495 if (!sp
->multimapped
)
1496 parent_pte
= sp
->parent_pte
;
1498 struct kvm_pte_chain
*chain
;
1500 chain
= container_of(sp
->parent_ptes
.first
,
1501 struct kvm_pte_chain
, link
);
1502 parent_pte
= chain
->parent_ptes
[0];
1504 BUG_ON(!parent_pte
);
1505 kvm_mmu_put_page(sp
, parent_pte
);
1506 __set_spte(parent_pte
, shadow_trap_nonpresent_pte
);
1510 static int mmu_zap_unsync_children(struct kvm
*kvm
,
1511 struct kvm_mmu_page
*parent
,
1512 struct list_head
*invalid_list
)
1515 struct mmu_page_path parents
;
1516 struct kvm_mmu_pages pages
;
1518 if (parent
->role
.level
== PT_PAGE_TABLE_LEVEL
)
1521 kvm_mmu_pages_init(parent
, &parents
, &pages
);
1522 while (mmu_unsync_walk(parent
, &pages
)) {
1523 struct kvm_mmu_page
*sp
;
1525 for_each_sp(pages
, sp
, parents
, i
) {
1526 kvm_mmu_prepare_zap_page(kvm
, sp
, invalid_list
);
1527 mmu_pages_clear_parents(&parents
);
1530 kvm_mmu_pages_init(parent
, &parents
, &pages
);
1536 static int kvm_mmu_prepare_zap_page(struct kvm
*kvm
, struct kvm_mmu_page
*sp
,
1537 struct list_head
*invalid_list
)
1541 trace_kvm_mmu_prepare_zap_page(sp
);
1542 ++kvm
->stat
.mmu_shadow_zapped
;
1543 ret
= mmu_zap_unsync_children(kvm
, sp
, invalid_list
);
1544 kvm_mmu_page_unlink_children(kvm
, sp
);
1545 kvm_mmu_unlink_parents(kvm
, sp
);
1546 if (!sp
->role
.invalid
&& !sp
->role
.direct
)
1547 unaccount_shadowed(kvm
, sp
->gfn
);
1549 kvm_unlink_unsync_page(kvm
, sp
);
1550 if (!sp
->root_count
) {
1553 list_move(&sp
->link
, invalid_list
);
1555 list_move(&sp
->link
, &kvm
->arch
.active_mmu_pages
);
1556 kvm_reload_remote_mmus(kvm
);
1559 sp
->role
.invalid
= 1;
1560 kvm_mmu_reset_last_pte_updated(kvm
);
1564 static void kvm_mmu_commit_zap_page(struct kvm
*kvm
,
1565 struct list_head
*invalid_list
)
1567 struct kvm_mmu_page
*sp
;
1569 if (list_empty(invalid_list
))
1572 kvm_flush_remote_tlbs(kvm
);
1575 sp
= list_first_entry(invalid_list
, struct kvm_mmu_page
, link
);
1576 WARN_ON(!sp
->role
.invalid
|| sp
->root_count
);
1577 kvm_mmu_free_page(kvm
, sp
);
1578 } while (!list_empty(invalid_list
));
1583 * Changing the number of mmu pages allocated to the vm
1584 * Note: if kvm_nr_mmu_pages is too small, you will get dead lock
1586 void kvm_mmu_change_mmu_pages(struct kvm
*kvm
, unsigned int kvm_nr_mmu_pages
)
1589 LIST_HEAD(invalid_list
);
1591 used_pages
= kvm
->arch
.n_alloc_mmu_pages
- kvm
->arch
.n_free_mmu_pages
;
1592 used_pages
= max(0, used_pages
);
1595 * If we set the number of mmu pages to be smaller be than the
1596 * number of actived pages , we must to free some mmu pages before we
1600 if (used_pages
> kvm_nr_mmu_pages
) {
1601 while (used_pages
> kvm_nr_mmu_pages
&&
1602 !list_empty(&kvm
->arch
.active_mmu_pages
)) {
1603 struct kvm_mmu_page
*page
;
1605 page
= container_of(kvm
->arch
.active_mmu_pages
.prev
,
1606 struct kvm_mmu_page
, link
);
1607 used_pages
-= kvm_mmu_prepare_zap_page(kvm
, page
,
1610 kvm_mmu_commit_zap_page(kvm
, &invalid_list
);
1611 kvm_nr_mmu_pages
= used_pages
;
1612 kvm
->arch
.n_free_mmu_pages
= 0;
1615 kvm
->arch
.n_free_mmu_pages
+= kvm_nr_mmu_pages
1616 - kvm
->arch
.n_alloc_mmu_pages
;
1618 kvm
->arch
.n_alloc_mmu_pages
= kvm_nr_mmu_pages
;
1621 static int kvm_mmu_unprotect_page(struct kvm
*kvm
, gfn_t gfn
)
1623 struct kvm_mmu_page
*sp
;
1624 struct hlist_node
*node
;
1625 LIST_HEAD(invalid_list
);
1628 pgprintk("%s: looking for gfn %lx\n", __func__
, gfn
);
1631 for_each_gfn_indirect_valid_sp(kvm
, sp
, gfn
, node
) {
1632 pgprintk("%s: gfn %lx role %x\n", __func__
, gfn
,
1635 kvm_mmu_prepare_zap_page(kvm
, sp
, &invalid_list
);
1637 kvm_mmu_commit_zap_page(kvm
, &invalid_list
);
1641 static void mmu_unshadow(struct kvm
*kvm
, gfn_t gfn
)
1643 struct kvm_mmu_page
*sp
;
1644 struct hlist_node
*node
;
1645 LIST_HEAD(invalid_list
);
1647 for_each_gfn_indirect_valid_sp(kvm
, sp
, gfn
, node
) {
1648 pgprintk("%s: zap %lx %x\n",
1649 __func__
, gfn
, sp
->role
.word
);
1650 kvm_mmu_prepare_zap_page(kvm
, sp
, &invalid_list
);
1652 kvm_mmu_commit_zap_page(kvm
, &invalid_list
);
1655 static void page_header_update_slot(struct kvm
*kvm
, void *pte
, gfn_t gfn
)
1657 int slot
= memslot_id(kvm
, gfn
);
1658 struct kvm_mmu_page
*sp
= page_header(__pa(pte
));
1660 __set_bit(slot
, sp
->slot_bitmap
);
1663 static void mmu_convert_notrap(struct kvm_mmu_page
*sp
)
1668 if (shadow_trap_nonpresent_pte
== shadow_notrap_nonpresent_pte
)
1671 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
1672 if (pt
[i
] == shadow_notrap_nonpresent_pte
)
1673 __set_spte(&pt
[i
], shadow_trap_nonpresent_pte
);
1678 * The function is based on mtrr_type_lookup() in
1679 * arch/x86/kernel/cpu/mtrr/generic.c
1681 static int get_mtrr_type(struct mtrr_state_type
*mtrr_state
,
1686 u8 prev_match
, curr_match
;
1687 int num_var_ranges
= KVM_NR_VAR_MTRR
;
1689 if (!mtrr_state
->enabled
)
1692 /* Make end inclusive end, instead of exclusive */
1695 /* Look in fixed ranges. Just return the type as per start */
1696 if (mtrr_state
->have_fixed
&& (start
< 0x100000)) {
1699 if (start
< 0x80000) {
1701 idx
+= (start
>> 16);
1702 return mtrr_state
->fixed_ranges
[idx
];
1703 } else if (start
< 0xC0000) {
1705 idx
+= ((start
- 0x80000) >> 14);
1706 return mtrr_state
->fixed_ranges
[idx
];
1707 } else if (start
< 0x1000000) {
1709 idx
+= ((start
- 0xC0000) >> 12);
1710 return mtrr_state
->fixed_ranges
[idx
];
1715 * Look in variable ranges
1716 * Look of multiple ranges matching this address and pick type
1717 * as per MTRR precedence
1719 if (!(mtrr_state
->enabled
& 2))
1720 return mtrr_state
->def_type
;
1723 for (i
= 0; i
< num_var_ranges
; ++i
) {
1724 unsigned short start_state
, end_state
;
1726 if (!(mtrr_state
->var_ranges
[i
].mask_lo
& (1 << 11)))
1729 base
= (((u64
)mtrr_state
->var_ranges
[i
].base_hi
) << 32) +
1730 (mtrr_state
->var_ranges
[i
].base_lo
& PAGE_MASK
);
1731 mask
= (((u64
)mtrr_state
->var_ranges
[i
].mask_hi
) << 32) +
1732 (mtrr_state
->var_ranges
[i
].mask_lo
& PAGE_MASK
);
1734 start_state
= ((start
& mask
) == (base
& mask
));
1735 end_state
= ((end
& mask
) == (base
& mask
));
1736 if (start_state
!= end_state
)
1739 if ((start
& mask
) != (base
& mask
))
1742 curr_match
= mtrr_state
->var_ranges
[i
].base_lo
& 0xff;
1743 if (prev_match
== 0xFF) {
1744 prev_match
= curr_match
;
1748 if (prev_match
== MTRR_TYPE_UNCACHABLE
||
1749 curr_match
== MTRR_TYPE_UNCACHABLE
)
1750 return MTRR_TYPE_UNCACHABLE
;
1752 if ((prev_match
== MTRR_TYPE_WRBACK
&&
1753 curr_match
== MTRR_TYPE_WRTHROUGH
) ||
1754 (prev_match
== MTRR_TYPE_WRTHROUGH
&&
1755 curr_match
== MTRR_TYPE_WRBACK
)) {
1756 prev_match
= MTRR_TYPE_WRTHROUGH
;
1757 curr_match
= MTRR_TYPE_WRTHROUGH
;
1760 if (prev_match
!= curr_match
)
1761 return MTRR_TYPE_UNCACHABLE
;
1764 if (prev_match
!= 0xFF)
1767 return mtrr_state
->def_type
;
1770 u8
kvm_get_guest_memory_type(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1774 mtrr
= get_mtrr_type(&vcpu
->arch
.mtrr_state
, gfn
<< PAGE_SHIFT
,
1775 (gfn
<< PAGE_SHIFT
) + PAGE_SIZE
);
1776 if (mtrr
== 0xfe || mtrr
== 0xff)
1777 mtrr
= MTRR_TYPE_WRBACK
;
1780 EXPORT_SYMBOL_GPL(kvm_get_guest_memory_type
);
1782 static void __kvm_unsync_page(struct kvm_vcpu
*vcpu
, struct kvm_mmu_page
*sp
)
1784 trace_kvm_mmu_unsync_page(sp
);
1785 ++vcpu
->kvm
->stat
.mmu_unsync
;
1788 kvm_mmu_mark_parents_unsync(sp
);
1789 mmu_convert_notrap(sp
);
1792 static void kvm_unsync_pages(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1794 struct kvm_mmu_page
*s
;
1795 struct hlist_node
*node
;
1797 for_each_gfn_indirect_valid_sp(vcpu
->kvm
, s
, gfn
, node
) {
1800 WARN_ON(s
->role
.level
!= PT_PAGE_TABLE_LEVEL
);
1801 __kvm_unsync_page(vcpu
, s
);
1805 static int mmu_need_write_protect(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
1808 struct kvm_mmu_page
*s
;
1809 struct hlist_node
*node
;
1810 bool need_unsync
= false;
1812 for_each_gfn_indirect_valid_sp(vcpu
->kvm
, s
, gfn
, node
) {
1816 if (s
->role
.level
!= PT_PAGE_TABLE_LEVEL
)
1819 if (!need_unsync
&& !s
->unsync
) {
1826 kvm_unsync_pages(vcpu
, gfn
);
1830 static int set_spte(struct kvm_vcpu
*vcpu
, u64
*sptep
,
1831 unsigned pte_access
, int user_fault
,
1832 int write_fault
, int dirty
, int level
,
1833 gfn_t gfn
, pfn_t pfn
, bool speculative
,
1834 bool can_unsync
, bool reset_host_protection
)
1840 * We don't set the accessed bit, since we sometimes want to see
1841 * whether the guest actually used the pte (in order to detect
1844 spte
= shadow_base_present_pte
| shadow_dirty_mask
;
1846 spte
|= shadow_accessed_mask
;
1848 pte_access
&= ~ACC_WRITE_MASK
;
1849 if (pte_access
& ACC_EXEC_MASK
)
1850 spte
|= shadow_x_mask
;
1852 spte
|= shadow_nx_mask
;
1853 if (pte_access
& ACC_USER_MASK
)
1854 spte
|= shadow_user_mask
;
1855 if (level
> PT_PAGE_TABLE_LEVEL
)
1856 spte
|= PT_PAGE_SIZE_MASK
;
1858 spte
|= kvm_x86_ops
->get_mt_mask(vcpu
, gfn
,
1859 kvm_is_mmio_pfn(pfn
));
1861 if (reset_host_protection
)
1862 spte
|= SPTE_HOST_WRITEABLE
;
1864 spte
|= (u64
)pfn
<< PAGE_SHIFT
;
1866 if ((pte_access
& ACC_WRITE_MASK
)
1867 || (!tdp_enabled
&& write_fault
&& !is_write_protection(vcpu
)
1870 if (level
> PT_PAGE_TABLE_LEVEL
&&
1871 has_wrprotected_page(vcpu
->kvm
, gfn
, level
)) {
1873 rmap_remove(vcpu
->kvm
, sptep
);
1874 spte
= shadow_trap_nonpresent_pte
;
1878 spte
|= PT_WRITABLE_MASK
;
1880 if (!tdp_enabled
&& !(pte_access
& ACC_WRITE_MASK
))
1881 spte
&= ~PT_USER_MASK
;
1884 * Optimization: for pte sync, if spte was writable the hash
1885 * lookup is unnecessary (and expensive). Write protection
1886 * is responsibility of mmu_get_page / kvm_sync_page.
1887 * Same reasoning can be applied to dirty page accounting.
1889 if (!can_unsync
&& is_writable_pte(*sptep
))
1892 if (mmu_need_write_protect(vcpu
, gfn
, can_unsync
)) {
1893 pgprintk("%s: found shadow page for %lx, marking ro\n",
1896 pte_access
&= ~ACC_WRITE_MASK
;
1897 if (is_writable_pte(spte
))
1898 spte
&= ~PT_WRITABLE_MASK
;
1902 if (pte_access
& ACC_WRITE_MASK
)
1903 mark_page_dirty(vcpu
->kvm
, gfn
);
1906 __set_spte(sptep
, spte
);
1910 static void mmu_set_spte(struct kvm_vcpu
*vcpu
, u64
*sptep
,
1911 unsigned pt_access
, unsigned pte_access
,
1912 int user_fault
, int write_fault
, int dirty
,
1913 int *ptwrite
, int level
, gfn_t gfn
,
1914 pfn_t pfn
, bool speculative
,
1915 bool reset_host_protection
)
1917 int was_rmapped
= 0;
1918 int was_writable
= is_writable_pte(*sptep
);
1921 pgprintk("%s: spte %llx access %x write_fault %d"
1922 " user_fault %d gfn %lx\n",
1923 __func__
, *sptep
, pt_access
,
1924 write_fault
, user_fault
, gfn
);
1926 if (is_rmap_spte(*sptep
)) {
1928 * If we overwrite a PTE page pointer with a 2MB PMD, unlink
1929 * the parent of the now unreachable PTE.
1931 if (level
> PT_PAGE_TABLE_LEVEL
&&
1932 !is_large_pte(*sptep
)) {
1933 struct kvm_mmu_page
*child
;
1936 child
= page_header(pte
& PT64_BASE_ADDR_MASK
);
1937 mmu_page_remove_parent_pte(child
, sptep
);
1938 __set_spte(sptep
, shadow_trap_nonpresent_pte
);
1939 kvm_flush_remote_tlbs(vcpu
->kvm
);
1940 } else if (pfn
!= spte_to_pfn(*sptep
)) {
1941 pgprintk("hfn old %lx new %lx\n",
1942 spte_to_pfn(*sptep
), pfn
);
1943 rmap_remove(vcpu
->kvm
, sptep
);
1944 __set_spte(sptep
, shadow_trap_nonpresent_pte
);
1945 kvm_flush_remote_tlbs(vcpu
->kvm
);
1950 if (set_spte(vcpu
, sptep
, pte_access
, user_fault
, write_fault
,
1951 dirty
, level
, gfn
, pfn
, speculative
, true,
1952 reset_host_protection
)) {
1955 kvm_mmu_flush_tlb(vcpu
);
1958 pgprintk("%s: setting spte %llx\n", __func__
, *sptep
);
1959 pgprintk("instantiating %s PTE (%s) at %ld (%llx) addr %p\n",
1960 is_large_pte(*sptep
)? "2MB" : "4kB",
1961 *sptep
& PT_PRESENT_MASK
?"RW":"R", gfn
,
1963 if (!was_rmapped
&& is_large_pte(*sptep
))
1964 ++vcpu
->kvm
->stat
.lpages
;
1966 page_header_update_slot(vcpu
->kvm
, sptep
, gfn
);
1968 rmap_count
= rmap_add(vcpu
, sptep
, gfn
);
1969 kvm_release_pfn_clean(pfn
);
1970 if (rmap_count
> RMAP_RECYCLE_THRESHOLD
)
1971 rmap_recycle(vcpu
, sptep
, gfn
);
1974 kvm_release_pfn_dirty(pfn
);
1976 kvm_release_pfn_clean(pfn
);
1979 vcpu
->arch
.last_pte_updated
= sptep
;
1980 vcpu
->arch
.last_pte_gfn
= gfn
;
1984 static void nonpaging_new_cr3(struct kvm_vcpu
*vcpu
)
1988 static int __direct_map(struct kvm_vcpu
*vcpu
, gpa_t v
, int write
,
1989 int level
, gfn_t gfn
, pfn_t pfn
)
1991 struct kvm_shadow_walk_iterator iterator
;
1992 struct kvm_mmu_page
*sp
;
1996 for_each_shadow_entry(vcpu
, (u64
)gfn
<< PAGE_SHIFT
, iterator
) {
1997 if (iterator
.level
== level
) {
1998 mmu_set_spte(vcpu
, iterator
.sptep
, ACC_ALL
, ACC_ALL
,
1999 0, write
, 1, &pt_write
,
2000 level
, gfn
, pfn
, false, true);
2001 ++vcpu
->stat
.pf_fixed
;
2005 if (*iterator
.sptep
== shadow_trap_nonpresent_pte
) {
2006 u64 base_addr
= iterator
.addr
;
2008 base_addr
&= PT64_LVL_ADDR_MASK(iterator
.level
);
2009 pseudo_gfn
= base_addr
>> PAGE_SHIFT
;
2010 sp
= kvm_mmu_get_page(vcpu
, pseudo_gfn
, iterator
.addr
,
2012 1, ACC_ALL
, iterator
.sptep
);
2014 pgprintk("nonpaging_map: ENOMEM\n");
2015 kvm_release_pfn_clean(pfn
);
2019 __set_spte(iterator
.sptep
,
2021 | PT_PRESENT_MASK
| PT_WRITABLE_MASK
2022 | shadow_user_mask
| shadow_x_mask
);
2028 static void kvm_send_hwpoison_signal(struct kvm
*kvm
, gfn_t gfn
)
2034 /* Touch the page, so send SIGBUS */
2035 hva
= (void __user
*)gfn_to_hva(kvm
, gfn
);
2036 r
= copy_from_user(buf
, hva
, 1);
2039 static int kvm_handle_bad_page(struct kvm
*kvm
, gfn_t gfn
, pfn_t pfn
)
2041 kvm_release_pfn_clean(pfn
);
2042 if (is_hwpoison_pfn(pfn
)) {
2043 kvm_send_hwpoison_signal(kvm
, gfn
);
2049 static int nonpaging_map(struct kvm_vcpu
*vcpu
, gva_t v
, int write
, gfn_t gfn
)
2054 unsigned long mmu_seq
;
2056 level
= mapping_level(vcpu
, gfn
);
2059 * This path builds a PAE pagetable - so we can map 2mb pages at
2060 * maximum. Therefore check if the level is larger than that.
2062 if (level
> PT_DIRECTORY_LEVEL
)
2063 level
= PT_DIRECTORY_LEVEL
;
2065 gfn
&= ~(KVM_PAGES_PER_HPAGE(level
) - 1);
2067 mmu_seq
= vcpu
->kvm
->mmu_notifier_seq
;
2069 pfn
= gfn_to_pfn(vcpu
->kvm
, gfn
);
2072 if (is_error_pfn(pfn
))
2073 return kvm_handle_bad_page(vcpu
->kvm
, gfn
, pfn
);
2075 spin_lock(&vcpu
->kvm
->mmu_lock
);
2076 if (mmu_notifier_retry(vcpu
, mmu_seq
))
2078 kvm_mmu_free_some_pages(vcpu
);
2079 r
= __direct_map(vcpu
, v
, write
, level
, gfn
, pfn
);
2080 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2086 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2087 kvm_release_pfn_clean(pfn
);
2092 static void mmu_free_roots(struct kvm_vcpu
*vcpu
)
2095 struct kvm_mmu_page
*sp
;
2096 LIST_HEAD(invalid_list
);
2098 if (!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
))
2100 spin_lock(&vcpu
->kvm
->mmu_lock
);
2101 if (vcpu
->arch
.mmu
.shadow_root_level
== PT64_ROOT_LEVEL
) {
2102 hpa_t root
= vcpu
->arch
.mmu
.root_hpa
;
2104 sp
= page_header(root
);
2106 if (!sp
->root_count
&& sp
->role
.invalid
) {
2107 kvm_mmu_prepare_zap_page(vcpu
->kvm
, sp
, &invalid_list
);
2108 kvm_mmu_commit_zap_page(vcpu
->kvm
, &invalid_list
);
2110 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
2111 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2114 for (i
= 0; i
< 4; ++i
) {
2115 hpa_t root
= vcpu
->arch
.mmu
.pae_root
[i
];
2118 root
&= PT64_BASE_ADDR_MASK
;
2119 sp
= page_header(root
);
2121 if (!sp
->root_count
&& sp
->role
.invalid
)
2122 kvm_mmu_prepare_zap_page(vcpu
->kvm
, sp
,
2125 vcpu
->arch
.mmu
.pae_root
[i
] = INVALID_PAGE
;
2127 kvm_mmu_commit_zap_page(vcpu
->kvm
, &invalid_list
);
2128 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2129 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
2132 static int mmu_check_root(struct kvm_vcpu
*vcpu
, gfn_t root_gfn
)
2136 if (!kvm_is_visible_gfn(vcpu
->kvm
, root_gfn
)) {
2137 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
2144 static int mmu_alloc_roots(struct kvm_vcpu
*vcpu
)
2148 struct kvm_mmu_page
*sp
;
2152 root_gfn
= vcpu
->arch
.cr3
>> PAGE_SHIFT
;
2154 if (vcpu
->arch
.mmu
.shadow_root_level
== PT64_ROOT_LEVEL
) {
2155 hpa_t root
= vcpu
->arch
.mmu
.root_hpa
;
2157 ASSERT(!VALID_PAGE(root
));
2158 if (mmu_check_root(vcpu
, root_gfn
))
2164 spin_lock(&vcpu
->kvm
->mmu_lock
);
2165 kvm_mmu_free_some_pages(vcpu
);
2166 sp
= kvm_mmu_get_page(vcpu
, root_gfn
, 0,
2167 PT64_ROOT_LEVEL
, direct
,
2169 root
= __pa(sp
->spt
);
2171 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2172 vcpu
->arch
.mmu
.root_hpa
= root
;
2175 direct
= !is_paging(vcpu
);
2176 for (i
= 0; i
< 4; ++i
) {
2177 hpa_t root
= vcpu
->arch
.mmu
.pae_root
[i
];
2179 ASSERT(!VALID_PAGE(root
));
2180 if (vcpu
->arch
.mmu
.root_level
== PT32E_ROOT_LEVEL
) {
2181 pdptr
= kvm_pdptr_read(vcpu
, i
);
2182 if (!is_present_gpte(pdptr
)) {
2183 vcpu
->arch
.mmu
.pae_root
[i
] = 0;
2186 root_gfn
= pdptr
>> PAGE_SHIFT
;
2187 } else if (vcpu
->arch
.mmu
.root_level
== 0)
2189 if (mmu_check_root(vcpu
, root_gfn
))
2195 spin_lock(&vcpu
->kvm
->mmu_lock
);
2196 kvm_mmu_free_some_pages(vcpu
);
2197 sp
= kvm_mmu_get_page(vcpu
, root_gfn
, i
<< 30,
2198 PT32_ROOT_LEVEL
, direct
,
2200 root
= __pa(sp
->spt
);
2202 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2204 vcpu
->arch
.mmu
.pae_root
[i
] = root
| PT_PRESENT_MASK
;
2206 vcpu
->arch
.mmu
.root_hpa
= __pa(vcpu
->arch
.mmu
.pae_root
);
2210 static void mmu_sync_roots(struct kvm_vcpu
*vcpu
)
2213 struct kvm_mmu_page
*sp
;
2215 if (!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
))
2217 if (vcpu
->arch
.mmu
.shadow_root_level
== PT64_ROOT_LEVEL
) {
2218 hpa_t root
= vcpu
->arch
.mmu
.root_hpa
;
2219 sp
= page_header(root
);
2220 mmu_sync_children(vcpu
, sp
);
2223 for (i
= 0; i
< 4; ++i
) {
2224 hpa_t root
= vcpu
->arch
.mmu
.pae_root
[i
];
2226 if (root
&& VALID_PAGE(root
)) {
2227 root
&= PT64_BASE_ADDR_MASK
;
2228 sp
= page_header(root
);
2229 mmu_sync_children(vcpu
, sp
);
2234 void kvm_mmu_sync_roots(struct kvm_vcpu
*vcpu
)
2236 spin_lock(&vcpu
->kvm
->mmu_lock
);
2237 mmu_sync_roots(vcpu
);
2238 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2241 static gpa_t
nonpaging_gva_to_gpa(struct kvm_vcpu
*vcpu
, gva_t vaddr
,
2242 u32 access
, u32
*error
)
2249 static int nonpaging_page_fault(struct kvm_vcpu
*vcpu
, gva_t gva
,
2255 pgprintk("%s: gva %lx error %x\n", __func__
, gva
, error_code
);
2256 r
= mmu_topup_memory_caches(vcpu
);
2261 ASSERT(VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
2263 gfn
= gva
>> PAGE_SHIFT
;
2265 return nonpaging_map(vcpu
, gva
& PAGE_MASK
,
2266 error_code
& PFERR_WRITE_MASK
, gfn
);
2269 static int tdp_page_fault(struct kvm_vcpu
*vcpu
, gva_t gpa
,
2275 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2276 unsigned long mmu_seq
;
2279 ASSERT(VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
2281 r
= mmu_topup_memory_caches(vcpu
);
2285 level
= mapping_level(vcpu
, gfn
);
2287 gfn
&= ~(KVM_PAGES_PER_HPAGE(level
) - 1);
2289 mmu_seq
= vcpu
->kvm
->mmu_notifier_seq
;
2291 pfn
= gfn_to_pfn(vcpu
->kvm
, gfn
);
2292 if (is_error_pfn(pfn
))
2293 return kvm_handle_bad_page(vcpu
->kvm
, gfn
, pfn
);
2294 spin_lock(&vcpu
->kvm
->mmu_lock
);
2295 if (mmu_notifier_retry(vcpu
, mmu_seq
))
2297 kvm_mmu_free_some_pages(vcpu
);
2298 r
= __direct_map(vcpu
, gpa
, error_code
& PFERR_WRITE_MASK
,
2300 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2305 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2306 kvm_release_pfn_clean(pfn
);
2310 static void nonpaging_free(struct kvm_vcpu
*vcpu
)
2312 mmu_free_roots(vcpu
);
2315 static int nonpaging_init_context(struct kvm_vcpu
*vcpu
)
2317 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
2319 context
->new_cr3
= nonpaging_new_cr3
;
2320 context
->page_fault
= nonpaging_page_fault
;
2321 context
->gva_to_gpa
= nonpaging_gva_to_gpa
;
2322 context
->free
= nonpaging_free
;
2323 context
->prefetch_page
= nonpaging_prefetch_page
;
2324 context
->sync_page
= nonpaging_sync_page
;
2325 context
->invlpg
= nonpaging_invlpg
;
2326 context
->root_level
= 0;
2327 context
->shadow_root_level
= PT32E_ROOT_LEVEL
;
2328 context
->root_hpa
= INVALID_PAGE
;
2332 void kvm_mmu_flush_tlb(struct kvm_vcpu
*vcpu
)
2334 ++vcpu
->stat
.tlb_flush
;
2335 kvm_make_request(KVM_REQ_TLB_FLUSH
, vcpu
);
2338 static void paging_new_cr3(struct kvm_vcpu
*vcpu
)
2340 pgprintk("%s: cr3 %lx\n", __func__
, vcpu
->arch
.cr3
);
2341 mmu_free_roots(vcpu
);
2344 static void inject_page_fault(struct kvm_vcpu
*vcpu
,
2348 kvm_inject_page_fault(vcpu
, addr
, err_code
);
2351 static void paging_free(struct kvm_vcpu
*vcpu
)
2353 nonpaging_free(vcpu
);
2356 static bool is_rsvd_bits_set(struct kvm_vcpu
*vcpu
, u64 gpte
, int level
)
2360 bit7
= (gpte
>> 7) & 1;
2361 return (gpte
& vcpu
->arch
.mmu
.rsvd_bits_mask
[bit7
][level
-1]) != 0;
2365 #include "paging_tmpl.h"
2369 #include "paging_tmpl.h"
2372 static void reset_rsvds_bits_mask(struct kvm_vcpu
*vcpu
, int level
)
2374 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
2375 int maxphyaddr
= cpuid_maxphyaddr(vcpu
);
2376 u64 exb_bit_rsvd
= 0;
2379 exb_bit_rsvd
= rsvd_bits(63, 63);
2381 case PT32_ROOT_LEVEL
:
2382 /* no rsvd bits for 2 level 4K page table entries */
2383 context
->rsvd_bits_mask
[0][1] = 0;
2384 context
->rsvd_bits_mask
[0][0] = 0;
2385 context
->rsvd_bits_mask
[1][0] = context
->rsvd_bits_mask
[0][0];
2387 if (!is_pse(vcpu
)) {
2388 context
->rsvd_bits_mask
[1][1] = 0;
2392 if (is_cpuid_PSE36())
2393 /* 36bits PSE 4MB page */
2394 context
->rsvd_bits_mask
[1][1] = rsvd_bits(17, 21);
2396 /* 32 bits PSE 4MB page */
2397 context
->rsvd_bits_mask
[1][1] = rsvd_bits(13, 21);
2399 case PT32E_ROOT_LEVEL
:
2400 context
->rsvd_bits_mask
[0][2] =
2401 rsvd_bits(maxphyaddr
, 63) |
2402 rsvd_bits(7, 8) | rsvd_bits(1, 2); /* PDPTE */
2403 context
->rsvd_bits_mask
[0][1] = exb_bit_rsvd
|
2404 rsvd_bits(maxphyaddr
, 62); /* PDE */
2405 context
->rsvd_bits_mask
[0][0] = exb_bit_rsvd
|
2406 rsvd_bits(maxphyaddr
, 62); /* PTE */
2407 context
->rsvd_bits_mask
[1][1] = exb_bit_rsvd
|
2408 rsvd_bits(maxphyaddr
, 62) |
2409 rsvd_bits(13, 20); /* large page */
2410 context
->rsvd_bits_mask
[1][0] = context
->rsvd_bits_mask
[0][0];
2412 case PT64_ROOT_LEVEL
:
2413 context
->rsvd_bits_mask
[0][3] = exb_bit_rsvd
|
2414 rsvd_bits(maxphyaddr
, 51) | rsvd_bits(7, 8);
2415 context
->rsvd_bits_mask
[0][2] = exb_bit_rsvd
|
2416 rsvd_bits(maxphyaddr
, 51) | rsvd_bits(7, 8);
2417 context
->rsvd_bits_mask
[0][1] = exb_bit_rsvd
|
2418 rsvd_bits(maxphyaddr
, 51);
2419 context
->rsvd_bits_mask
[0][0] = exb_bit_rsvd
|
2420 rsvd_bits(maxphyaddr
, 51);
2421 context
->rsvd_bits_mask
[1][3] = context
->rsvd_bits_mask
[0][3];
2422 context
->rsvd_bits_mask
[1][2] = exb_bit_rsvd
|
2423 rsvd_bits(maxphyaddr
, 51) |
2425 context
->rsvd_bits_mask
[1][1] = exb_bit_rsvd
|
2426 rsvd_bits(maxphyaddr
, 51) |
2427 rsvd_bits(13, 20); /* large page */
2428 context
->rsvd_bits_mask
[1][0] = context
->rsvd_bits_mask
[0][0];
2433 static int paging64_init_context_common(struct kvm_vcpu
*vcpu
, int level
)
2435 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
2437 ASSERT(is_pae(vcpu
));
2438 context
->new_cr3
= paging_new_cr3
;
2439 context
->page_fault
= paging64_page_fault
;
2440 context
->gva_to_gpa
= paging64_gva_to_gpa
;
2441 context
->prefetch_page
= paging64_prefetch_page
;
2442 context
->sync_page
= paging64_sync_page
;
2443 context
->invlpg
= paging64_invlpg
;
2444 context
->free
= paging_free
;
2445 context
->root_level
= level
;
2446 context
->shadow_root_level
= level
;
2447 context
->root_hpa
= INVALID_PAGE
;
2451 static int paging64_init_context(struct kvm_vcpu
*vcpu
)
2453 reset_rsvds_bits_mask(vcpu
, PT64_ROOT_LEVEL
);
2454 return paging64_init_context_common(vcpu
, PT64_ROOT_LEVEL
);
2457 static int paging32_init_context(struct kvm_vcpu
*vcpu
)
2459 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
2461 reset_rsvds_bits_mask(vcpu
, PT32_ROOT_LEVEL
);
2462 context
->new_cr3
= paging_new_cr3
;
2463 context
->page_fault
= paging32_page_fault
;
2464 context
->gva_to_gpa
= paging32_gva_to_gpa
;
2465 context
->free
= paging_free
;
2466 context
->prefetch_page
= paging32_prefetch_page
;
2467 context
->sync_page
= paging32_sync_page
;
2468 context
->invlpg
= paging32_invlpg
;
2469 context
->root_level
= PT32_ROOT_LEVEL
;
2470 context
->shadow_root_level
= PT32E_ROOT_LEVEL
;
2471 context
->root_hpa
= INVALID_PAGE
;
2475 static int paging32E_init_context(struct kvm_vcpu
*vcpu
)
2477 reset_rsvds_bits_mask(vcpu
, PT32E_ROOT_LEVEL
);
2478 return paging64_init_context_common(vcpu
, PT32E_ROOT_LEVEL
);
2481 static int init_kvm_tdp_mmu(struct kvm_vcpu
*vcpu
)
2483 struct kvm_mmu
*context
= &vcpu
->arch
.mmu
;
2485 context
->new_cr3
= nonpaging_new_cr3
;
2486 context
->page_fault
= tdp_page_fault
;
2487 context
->free
= nonpaging_free
;
2488 context
->prefetch_page
= nonpaging_prefetch_page
;
2489 context
->sync_page
= nonpaging_sync_page
;
2490 context
->invlpg
= nonpaging_invlpg
;
2491 context
->shadow_root_level
= kvm_x86_ops
->get_tdp_level();
2492 context
->root_hpa
= INVALID_PAGE
;
2494 if (!is_paging(vcpu
)) {
2495 context
->gva_to_gpa
= nonpaging_gva_to_gpa
;
2496 context
->root_level
= 0;
2497 } else if (is_long_mode(vcpu
)) {
2498 reset_rsvds_bits_mask(vcpu
, PT64_ROOT_LEVEL
);
2499 context
->gva_to_gpa
= paging64_gva_to_gpa
;
2500 context
->root_level
= PT64_ROOT_LEVEL
;
2501 } else if (is_pae(vcpu
)) {
2502 reset_rsvds_bits_mask(vcpu
, PT32E_ROOT_LEVEL
);
2503 context
->gva_to_gpa
= paging64_gva_to_gpa
;
2504 context
->root_level
= PT32E_ROOT_LEVEL
;
2506 reset_rsvds_bits_mask(vcpu
, PT32_ROOT_LEVEL
);
2507 context
->gva_to_gpa
= paging32_gva_to_gpa
;
2508 context
->root_level
= PT32_ROOT_LEVEL
;
2514 static int init_kvm_softmmu(struct kvm_vcpu
*vcpu
)
2519 ASSERT(!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
2521 if (!is_paging(vcpu
))
2522 r
= nonpaging_init_context(vcpu
);
2523 else if (is_long_mode(vcpu
))
2524 r
= paging64_init_context(vcpu
);
2525 else if (is_pae(vcpu
))
2526 r
= paging32E_init_context(vcpu
);
2528 r
= paging32_init_context(vcpu
);
2530 vcpu
->arch
.mmu
.base_role
.cr4_pae
= !!is_pae(vcpu
);
2531 vcpu
->arch
.mmu
.base_role
.cr0_wp
= is_write_protection(vcpu
);
2536 static int init_kvm_mmu(struct kvm_vcpu
*vcpu
)
2538 vcpu
->arch
.update_pte
.pfn
= bad_pfn
;
2541 return init_kvm_tdp_mmu(vcpu
);
2543 return init_kvm_softmmu(vcpu
);
2546 static void destroy_kvm_mmu(struct kvm_vcpu
*vcpu
)
2549 if (VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
))
2550 /* mmu.free() should set root_hpa = INVALID_PAGE */
2551 vcpu
->arch
.mmu
.free(vcpu
);
2554 int kvm_mmu_reset_context(struct kvm_vcpu
*vcpu
)
2556 destroy_kvm_mmu(vcpu
);
2557 return init_kvm_mmu(vcpu
);
2559 EXPORT_SYMBOL_GPL(kvm_mmu_reset_context
);
2561 int kvm_mmu_load(struct kvm_vcpu
*vcpu
)
2565 r
= mmu_topup_memory_caches(vcpu
);
2568 r
= mmu_alloc_roots(vcpu
);
2569 spin_lock(&vcpu
->kvm
->mmu_lock
);
2570 mmu_sync_roots(vcpu
);
2571 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2574 /* set_cr3() should ensure TLB has been flushed */
2575 kvm_x86_ops
->set_cr3(vcpu
, vcpu
->arch
.mmu
.root_hpa
);
2579 EXPORT_SYMBOL_GPL(kvm_mmu_load
);
2581 void kvm_mmu_unload(struct kvm_vcpu
*vcpu
)
2583 mmu_free_roots(vcpu
);
2586 static void mmu_pte_write_zap_pte(struct kvm_vcpu
*vcpu
,
2587 struct kvm_mmu_page
*sp
,
2591 struct kvm_mmu_page
*child
;
2594 if (is_shadow_present_pte(pte
)) {
2595 if (is_last_spte(pte
, sp
->role
.level
))
2596 rmap_remove(vcpu
->kvm
, spte
);
2598 child
= page_header(pte
& PT64_BASE_ADDR_MASK
);
2599 mmu_page_remove_parent_pte(child
, spte
);
2602 __set_spte(spte
, shadow_trap_nonpresent_pte
);
2603 if (is_large_pte(pte
))
2604 --vcpu
->kvm
->stat
.lpages
;
2607 static void mmu_pte_write_new_pte(struct kvm_vcpu
*vcpu
,
2608 struct kvm_mmu_page
*sp
,
2612 if (sp
->role
.level
!= PT_PAGE_TABLE_LEVEL
) {
2613 ++vcpu
->kvm
->stat
.mmu_pde_zapped
;
2617 ++vcpu
->kvm
->stat
.mmu_pte_updated
;
2618 if (!sp
->role
.cr4_pae
)
2619 paging32_update_pte(vcpu
, sp
, spte
, new);
2621 paging64_update_pte(vcpu
, sp
, spte
, new);
2624 static bool need_remote_flush(u64 old
, u64
new)
2626 if (!is_shadow_present_pte(old
))
2628 if (!is_shadow_present_pte(new))
2630 if ((old
^ new) & PT64_BASE_ADDR_MASK
)
2632 old
^= PT64_NX_MASK
;
2633 new ^= PT64_NX_MASK
;
2634 return (old
& ~new & PT64_PERM_MASK
) != 0;
2637 static void mmu_pte_write_flush_tlb(struct kvm_vcpu
*vcpu
, bool zap_page
,
2638 bool remote_flush
, bool local_flush
)
2644 kvm_flush_remote_tlbs(vcpu
->kvm
);
2645 else if (local_flush
)
2646 kvm_mmu_flush_tlb(vcpu
);
2649 static bool last_updated_pte_accessed(struct kvm_vcpu
*vcpu
)
2651 u64
*spte
= vcpu
->arch
.last_pte_updated
;
2653 return !!(spte
&& (*spte
& shadow_accessed_mask
));
2656 static void mmu_guess_page_from_pte_write(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
2662 if (!is_present_gpte(gpte
))
2664 gfn
= (gpte
& PT64_BASE_ADDR_MASK
) >> PAGE_SHIFT
;
2666 vcpu
->arch
.update_pte
.mmu_seq
= vcpu
->kvm
->mmu_notifier_seq
;
2668 pfn
= gfn_to_pfn(vcpu
->kvm
, gfn
);
2670 if (is_error_pfn(pfn
)) {
2671 kvm_release_pfn_clean(pfn
);
2674 vcpu
->arch
.update_pte
.gfn
= gfn
;
2675 vcpu
->arch
.update_pte
.pfn
= pfn
;
2678 static void kvm_mmu_access_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2680 u64
*spte
= vcpu
->arch
.last_pte_updated
;
2683 && vcpu
->arch
.last_pte_gfn
== gfn
2684 && shadow_accessed_mask
2685 && !(*spte
& shadow_accessed_mask
)
2686 && is_shadow_present_pte(*spte
))
2687 set_bit(PT_ACCESSED_SHIFT
, (unsigned long *)spte
);
2690 void kvm_mmu_pte_write(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
2691 const u8
*new, int bytes
,
2692 bool guest_initiated
)
2694 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2695 struct kvm_mmu_page
*sp
;
2696 struct hlist_node
*node
;
2697 LIST_HEAD(invalid_list
);
2700 unsigned offset
= offset_in_page(gpa
);
2702 unsigned page_offset
;
2703 unsigned misaligned
;
2710 bool remote_flush
, local_flush
, zap_page
;
2712 zap_page
= remote_flush
= local_flush
= false;
2714 pgprintk("%s: gpa %llx bytes %d\n", __func__
, gpa
, bytes
);
2716 invlpg_counter
= atomic_read(&vcpu
->kvm
->arch
.invlpg_counter
);
2719 * Assume that the pte write on a page table of the same type
2720 * as the current vcpu paging mode. This is nearly always true
2721 * (might be false while changing modes). Note it is verified later
2724 if ((is_pae(vcpu
) && bytes
== 4) || !new) {
2725 /* Handle a 32-bit guest writing two halves of a 64-bit gpte */
2730 r
= kvm_read_guest(vcpu
->kvm
, gpa
, &gentry
, min(bytes
, 8));
2733 new = (const u8
*)&gentry
;
2738 gentry
= *(const u32
*)new;
2741 gentry
= *(const u64
*)new;
2748 mmu_guess_page_from_pte_write(vcpu
, gpa
, gentry
);
2749 spin_lock(&vcpu
->kvm
->mmu_lock
);
2750 if (atomic_read(&vcpu
->kvm
->arch
.invlpg_counter
) != invlpg_counter
)
2752 kvm_mmu_access_page(vcpu
, gfn
);
2753 kvm_mmu_free_some_pages(vcpu
);
2754 ++vcpu
->kvm
->stat
.mmu_pte_write
;
2755 kvm_mmu_audit(vcpu
, "pre pte write");
2756 if (guest_initiated
) {
2757 if (gfn
== vcpu
->arch
.last_pt_write_gfn
2758 && !last_updated_pte_accessed(vcpu
)) {
2759 ++vcpu
->arch
.last_pt_write_count
;
2760 if (vcpu
->arch
.last_pt_write_count
>= 3)
2763 vcpu
->arch
.last_pt_write_gfn
= gfn
;
2764 vcpu
->arch
.last_pt_write_count
= 1;
2765 vcpu
->arch
.last_pte_updated
= NULL
;
2769 for_each_gfn_indirect_valid_sp(vcpu
->kvm
, sp
, gfn
, node
) {
2770 pte_size
= sp
->role
.cr4_pae
? 8 : 4;
2771 misaligned
= (offset
^ (offset
+ bytes
- 1)) & ~(pte_size
- 1);
2772 misaligned
|= bytes
< 4;
2773 if (misaligned
|| flooded
) {
2775 * Misaligned accesses are too much trouble to fix
2776 * up; also, they usually indicate a page is not used
2779 * If we're seeing too many writes to a page,
2780 * it may no longer be a page table, or we may be
2781 * forking, in which case it is better to unmap the
2784 pgprintk("misaligned: gpa %llx bytes %d role %x\n",
2785 gpa
, bytes
, sp
->role
.word
);
2786 zap_page
|= !!kvm_mmu_prepare_zap_page(vcpu
->kvm
, sp
,
2788 ++vcpu
->kvm
->stat
.mmu_flooded
;
2791 page_offset
= offset
;
2792 level
= sp
->role
.level
;
2794 if (!sp
->role
.cr4_pae
) {
2795 page_offset
<<= 1; /* 32->64 */
2797 * A 32-bit pde maps 4MB while the shadow pdes map
2798 * only 2MB. So we need to double the offset again
2799 * and zap two pdes instead of one.
2801 if (level
== PT32_ROOT_LEVEL
) {
2802 page_offset
&= ~7; /* kill rounding error */
2806 quadrant
= page_offset
>> PAGE_SHIFT
;
2807 page_offset
&= ~PAGE_MASK
;
2808 if (quadrant
!= sp
->role
.quadrant
)
2812 spte
= &sp
->spt
[page_offset
/ sizeof(*spte
)];
2815 mmu_pte_write_zap_pte(vcpu
, sp
, spte
);
2817 mmu_pte_write_new_pte(vcpu
, sp
, spte
, &gentry
);
2818 if (!remote_flush
&& need_remote_flush(entry
, *spte
))
2819 remote_flush
= true;
2823 mmu_pte_write_flush_tlb(vcpu
, zap_page
, remote_flush
, local_flush
);
2824 kvm_mmu_commit_zap_page(vcpu
->kvm
, &invalid_list
);
2825 kvm_mmu_audit(vcpu
, "post pte write");
2826 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2827 if (!is_error_pfn(vcpu
->arch
.update_pte
.pfn
)) {
2828 kvm_release_pfn_clean(vcpu
->arch
.update_pte
.pfn
);
2829 vcpu
->arch
.update_pte
.pfn
= bad_pfn
;
2833 int kvm_mmu_unprotect_page_virt(struct kvm_vcpu
*vcpu
, gva_t gva
)
2841 gpa
= kvm_mmu_gva_to_gpa_read(vcpu
, gva
, NULL
);
2843 spin_lock(&vcpu
->kvm
->mmu_lock
);
2844 r
= kvm_mmu_unprotect_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
2845 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2848 EXPORT_SYMBOL_GPL(kvm_mmu_unprotect_page_virt
);
2850 void __kvm_mmu_free_some_pages(struct kvm_vcpu
*vcpu
)
2853 LIST_HEAD(invalid_list
);
2855 free_pages
= vcpu
->kvm
->arch
.n_free_mmu_pages
;
2856 while (free_pages
< KVM_REFILL_PAGES
&&
2857 !list_empty(&vcpu
->kvm
->arch
.active_mmu_pages
)) {
2858 struct kvm_mmu_page
*sp
;
2860 sp
= container_of(vcpu
->kvm
->arch
.active_mmu_pages
.prev
,
2861 struct kvm_mmu_page
, link
);
2862 free_pages
+= kvm_mmu_prepare_zap_page(vcpu
->kvm
, sp
,
2864 ++vcpu
->kvm
->stat
.mmu_recycled
;
2866 kvm_mmu_commit_zap_page(vcpu
->kvm
, &invalid_list
);
2869 int kvm_mmu_page_fault(struct kvm_vcpu
*vcpu
, gva_t cr2
, u32 error_code
)
2872 enum emulation_result er
;
2874 r
= vcpu
->arch
.mmu
.page_fault(vcpu
, cr2
, error_code
);
2883 r
= mmu_topup_memory_caches(vcpu
);
2887 er
= emulate_instruction(vcpu
, cr2
, error_code
, 0);
2892 case EMULATE_DO_MMIO
:
2893 ++vcpu
->stat
.mmio_exits
;
2903 EXPORT_SYMBOL_GPL(kvm_mmu_page_fault
);
2905 void kvm_mmu_invlpg(struct kvm_vcpu
*vcpu
, gva_t gva
)
2907 vcpu
->arch
.mmu
.invlpg(vcpu
, gva
);
2908 kvm_mmu_flush_tlb(vcpu
);
2909 ++vcpu
->stat
.invlpg
;
2911 EXPORT_SYMBOL_GPL(kvm_mmu_invlpg
);
2913 void kvm_enable_tdp(void)
2917 EXPORT_SYMBOL_GPL(kvm_enable_tdp
);
2919 void kvm_disable_tdp(void)
2921 tdp_enabled
= false;
2923 EXPORT_SYMBOL_GPL(kvm_disable_tdp
);
2925 static void free_mmu_pages(struct kvm_vcpu
*vcpu
)
2927 free_page((unsigned long)vcpu
->arch
.mmu
.pae_root
);
2930 static int alloc_mmu_pages(struct kvm_vcpu
*vcpu
)
2938 * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64.
2939 * Therefore we need to allocate shadow page tables in the first
2940 * 4GB of memory, which happens to fit the DMA32 zone.
2942 page
= alloc_page(GFP_KERNEL
| __GFP_DMA32
);
2946 vcpu
->arch
.mmu
.pae_root
= page_address(page
);
2947 for (i
= 0; i
< 4; ++i
)
2948 vcpu
->arch
.mmu
.pae_root
[i
] = INVALID_PAGE
;
2953 int kvm_mmu_create(struct kvm_vcpu
*vcpu
)
2956 ASSERT(!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
2958 return alloc_mmu_pages(vcpu
);
2961 int kvm_mmu_setup(struct kvm_vcpu
*vcpu
)
2964 ASSERT(!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
2966 return init_kvm_mmu(vcpu
);
2969 void kvm_mmu_destroy(struct kvm_vcpu
*vcpu
)
2973 destroy_kvm_mmu(vcpu
);
2974 free_mmu_pages(vcpu
);
2975 mmu_free_memory_caches(vcpu
);
2978 void kvm_mmu_slot_remove_write_access(struct kvm
*kvm
, int slot
)
2980 struct kvm_mmu_page
*sp
;
2982 list_for_each_entry(sp
, &kvm
->arch
.active_mmu_pages
, link
) {
2986 if (!test_bit(slot
, sp
->slot_bitmap
))
2990 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
)
2992 if (is_writable_pte(pt
[i
]))
2993 pt
[i
] &= ~PT_WRITABLE_MASK
;
2995 kvm_flush_remote_tlbs(kvm
);
2998 void kvm_mmu_zap_all(struct kvm
*kvm
)
3000 struct kvm_mmu_page
*sp
, *node
;
3001 LIST_HEAD(invalid_list
);
3003 spin_lock(&kvm
->mmu_lock
);
3005 list_for_each_entry_safe(sp
, node
, &kvm
->arch
.active_mmu_pages
, link
)
3006 if (kvm_mmu_prepare_zap_page(kvm
, sp
, &invalid_list
))
3009 kvm_mmu_commit_zap_page(kvm
, &invalid_list
);
3010 spin_unlock(&kvm
->mmu_lock
);
3013 static int kvm_mmu_remove_some_alloc_mmu_pages(struct kvm
*kvm
,
3014 struct list_head
*invalid_list
)
3016 struct kvm_mmu_page
*page
;
3018 page
= container_of(kvm
->arch
.active_mmu_pages
.prev
,
3019 struct kvm_mmu_page
, link
);
3020 return kvm_mmu_prepare_zap_page(kvm
, page
, invalid_list
);
3023 static int mmu_shrink(struct shrinker
*shrink
, int nr_to_scan
, gfp_t gfp_mask
)
3026 struct kvm
*kvm_freed
= NULL
;
3027 int cache_count
= 0;
3029 spin_lock(&kvm_lock
);
3031 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3032 int npages
, idx
, freed_pages
;
3033 LIST_HEAD(invalid_list
);
3035 idx
= srcu_read_lock(&kvm
->srcu
);
3036 spin_lock(&kvm
->mmu_lock
);
3037 npages
= kvm
->arch
.n_alloc_mmu_pages
-
3038 kvm
->arch
.n_free_mmu_pages
;
3039 cache_count
+= npages
;
3040 if (!kvm_freed
&& nr_to_scan
> 0 && npages
> 0) {
3041 freed_pages
= kvm_mmu_remove_some_alloc_mmu_pages(kvm
,
3043 cache_count
-= freed_pages
;
3048 kvm_mmu_commit_zap_page(kvm
, &invalid_list
);
3049 spin_unlock(&kvm
->mmu_lock
);
3050 srcu_read_unlock(&kvm
->srcu
, idx
);
3053 list_move_tail(&kvm_freed
->vm_list
, &vm_list
);
3055 spin_unlock(&kvm_lock
);
3060 static struct shrinker mmu_shrinker
= {
3061 .shrink
= mmu_shrink
,
3062 .seeks
= DEFAULT_SEEKS
* 10,
3065 static void mmu_destroy_caches(void)
3067 if (pte_chain_cache
)
3068 kmem_cache_destroy(pte_chain_cache
);
3069 if (rmap_desc_cache
)
3070 kmem_cache_destroy(rmap_desc_cache
);
3071 if (mmu_page_header_cache
)
3072 kmem_cache_destroy(mmu_page_header_cache
);
3075 void kvm_mmu_module_exit(void)
3077 mmu_destroy_caches();
3078 unregister_shrinker(&mmu_shrinker
);
3081 int kvm_mmu_module_init(void)
3083 pte_chain_cache
= kmem_cache_create("kvm_pte_chain",
3084 sizeof(struct kvm_pte_chain
),
3086 if (!pte_chain_cache
)
3088 rmap_desc_cache
= kmem_cache_create("kvm_rmap_desc",
3089 sizeof(struct kvm_rmap_desc
),
3091 if (!rmap_desc_cache
)
3094 mmu_page_header_cache
= kmem_cache_create("kvm_mmu_page_header",
3095 sizeof(struct kvm_mmu_page
),
3097 if (!mmu_page_header_cache
)
3100 register_shrinker(&mmu_shrinker
);
3105 mmu_destroy_caches();
3110 * Caculate mmu pages needed for kvm.
3112 unsigned int kvm_mmu_calculate_mmu_pages(struct kvm
*kvm
)
3115 unsigned int nr_mmu_pages
;
3116 unsigned int nr_pages
= 0;
3117 struct kvm_memslots
*slots
;
3119 slots
= kvm_memslots(kvm
);
3121 for (i
= 0; i
< slots
->nmemslots
; i
++)
3122 nr_pages
+= slots
->memslots
[i
].npages
;
3124 nr_mmu_pages
= nr_pages
* KVM_PERMILLE_MMU_PAGES
/ 1000;
3125 nr_mmu_pages
= max(nr_mmu_pages
,
3126 (unsigned int) KVM_MIN_ALLOC_MMU_PAGES
);
3128 return nr_mmu_pages
;
3131 static void *pv_mmu_peek_buffer(struct kvm_pv_mmu_op_buffer
*buffer
,
3134 if (len
> buffer
->len
)
3139 static void *pv_mmu_read_buffer(struct kvm_pv_mmu_op_buffer
*buffer
,
3144 ret
= pv_mmu_peek_buffer(buffer
, len
);
3149 buffer
->processed
+= len
;
3153 static int kvm_pv_mmu_write(struct kvm_vcpu
*vcpu
,
3154 gpa_t addr
, gpa_t value
)
3159 if (!is_long_mode(vcpu
) && !is_pae(vcpu
))
3162 r
= mmu_topup_memory_caches(vcpu
);
3166 if (!emulator_write_phys(vcpu
, addr
, &value
, bytes
))
3172 static int kvm_pv_mmu_flush_tlb(struct kvm_vcpu
*vcpu
)
3174 (void)kvm_set_cr3(vcpu
, vcpu
->arch
.cr3
);
3178 static int kvm_pv_mmu_release_pt(struct kvm_vcpu
*vcpu
, gpa_t addr
)
3180 spin_lock(&vcpu
->kvm
->mmu_lock
);
3181 mmu_unshadow(vcpu
->kvm
, addr
>> PAGE_SHIFT
);
3182 spin_unlock(&vcpu
->kvm
->mmu_lock
);
3186 static int kvm_pv_mmu_op_one(struct kvm_vcpu
*vcpu
,
3187 struct kvm_pv_mmu_op_buffer
*buffer
)
3189 struct kvm_mmu_op_header
*header
;
3191 header
= pv_mmu_peek_buffer(buffer
, sizeof *header
);
3194 switch (header
->op
) {
3195 case KVM_MMU_OP_WRITE_PTE
: {
3196 struct kvm_mmu_op_write_pte
*wpte
;
3198 wpte
= pv_mmu_read_buffer(buffer
, sizeof *wpte
);
3201 return kvm_pv_mmu_write(vcpu
, wpte
->pte_phys
,
3204 case KVM_MMU_OP_FLUSH_TLB
: {
3205 struct kvm_mmu_op_flush_tlb
*ftlb
;
3207 ftlb
= pv_mmu_read_buffer(buffer
, sizeof *ftlb
);
3210 return kvm_pv_mmu_flush_tlb(vcpu
);
3212 case KVM_MMU_OP_RELEASE_PT
: {
3213 struct kvm_mmu_op_release_pt
*rpt
;
3215 rpt
= pv_mmu_read_buffer(buffer
, sizeof *rpt
);
3218 return kvm_pv_mmu_release_pt(vcpu
, rpt
->pt_phys
);
3224 int kvm_pv_mmu_op(struct kvm_vcpu
*vcpu
, unsigned long bytes
,
3225 gpa_t addr
, unsigned long *ret
)
3228 struct kvm_pv_mmu_op_buffer
*buffer
= &vcpu
->arch
.mmu_op_buffer
;
3230 buffer
->ptr
= buffer
->buf
;
3231 buffer
->len
= min_t(unsigned long, bytes
, sizeof buffer
->buf
);
3232 buffer
->processed
= 0;
3234 r
= kvm_read_guest(vcpu
->kvm
, addr
, buffer
->buf
, buffer
->len
);
3238 while (buffer
->len
) {
3239 r
= kvm_pv_mmu_op_one(vcpu
, buffer
);
3248 *ret
= buffer
->processed
;
3252 int kvm_mmu_get_spte_hierarchy(struct kvm_vcpu
*vcpu
, u64 addr
, u64 sptes
[4])
3254 struct kvm_shadow_walk_iterator iterator
;
3257 spin_lock(&vcpu
->kvm
->mmu_lock
);
3258 for_each_shadow_entry(vcpu
, addr
, iterator
) {
3259 sptes
[iterator
.level
-1] = *iterator
.sptep
;
3261 if (!is_shadow_present_pte(*iterator
.sptep
))
3264 spin_unlock(&vcpu
->kvm
->mmu_lock
);
3268 EXPORT_SYMBOL_GPL(kvm_mmu_get_spte_hierarchy
);
3272 static const char *audit_msg
;
3274 static gva_t
canonicalize(gva_t gva
)
3276 #ifdef CONFIG_X86_64
3277 gva
= (long long)(gva
<< 16) >> 16;
3283 typedef void (*inspect_spte_fn
) (struct kvm
*kvm
, u64
*sptep
);
3285 static void __mmu_spte_walk(struct kvm
*kvm
, struct kvm_mmu_page
*sp
,
3290 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
3291 u64 ent
= sp
->spt
[i
];
3293 if (is_shadow_present_pte(ent
)) {
3294 if (!is_last_spte(ent
, sp
->role
.level
)) {
3295 struct kvm_mmu_page
*child
;
3296 child
= page_header(ent
& PT64_BASE_ADDR_MASK
);
3297 __mmu_spte_walk(kvm
, child
, fn
);
3299 fn(kvm
, &sp
->spt
[i
]);
3304 static void mmu_spte_walk(struct kvm_vcpu
*vcpu
, inspect_spte_fn fn
)
3307 struct kvm_mmu_page
*sp
;
3309 if (!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
))
3311 if (vcpu
->arch
.mmu
.shadow_root_level
== PT64_ROOT_LEVEL
) {
3312 hpa_t root
= vcpu
->arch
.mmu
.root_hpa
;
3313 sp
= page_header(root
);
3314 __mmu_spte_walk(vcpu
->kvm
, sp
, fn
);
3317 for (i
= 0; i
< 4; ++i
) {
3318 hpa_t root
= vcpu
->arch
.mmu
.pae_root
[i
];
3320 if (root
&& VALID_PAGE(root
)) {
3321 root
&= PT64_BASE_ADDR_MASK
;
3322 sp
= page_header(root
);
3323 __mmu_spte_walk(vcpu
->kvm
, sp
, fn
);
3329 static void audit_mappings_page(struct kvm_vcpu
*vcpu
, u64 page_pte
,
3330 gva_t va
, int level
)
3332 u64
*pt
= __va(page_pte
& PT64_BASE_ADDR_MASK
);
3334 gva_t va_delta
= 1ul << (PAGE_SHIFT
+ 9 * (level
- 1));
3336 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
, va
+= va_delta
) {
3339 if (ent
== shadow_trap_nonpresent_pte
)
3342 va
= canonicalize(va
);
3343 if (is_shadow_present_pte(ent
) && !is_last_spte(ent
, level
))
3344 audit_mappings_page(vcpu
, ent
, va
, level
- 1);
3346 gpa_t gpa
= kvm_mmu_gva_to_gpa_read(vcpu
, va
, NULL
);
3347 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3348 pfn_t pfn
= gfn_to_pfn(vcpu
->kvm
, gfn
);
3349 hpa_t hpa
= (hpa_t
)pfn
<< PAGE_SHIFT
;
3351 if (is_error_pfn(pfn
)) {
3352 kvm_release_pfn_clean(pfn
);
3356 if (is_shadow_present_pte(ent
)
3357 && (ent
& PT64_BASE_ADDR_MASK
) != hpa
)
3358 printk(KERN_ERR
"xx audit error: (%s) levels %d"
3359 " gva %lx gpa %llx hpa %llx ent %llx %d\n",
3360 audit_msg
, vcpu
->arch
.mmu
.root_level
,
3362 is_shadow_present_pte(ent
));
3363 else if (ent
== shadow_notrap_nonpresent_pte
3364 && !is_error_hpa(hpa
))
3365 printk(KERN_ERR
"audit: (%s) notrap shadow,"
3366 " valid guest gva %lx\n", audit_msg
, va
);
3367 kvm_release_pfn_clean(pfn
);
3373 static void audit_mappings(struct kvm_vcpu
*vcpu
)
3377 if (vcpu
->arch
.mmu
.root_level
== 4)
3378 audit_mappings_page(vcpu
, vcpu
->arch
.mmu
.root_hpa
, 0, 4);
3380 for (i
= 0; i
< 4; ++i
)
3381 if (vcpu
->arch
.mmu
.pae_root
[i
] & PT_PRESENT_MASK
)
3382 audit_mappings_page(vcpu
,
3383 vcpu
->arch
.mmu
.pae_root
[i
],
3388 static int count_rmaps(struct kvm_vcpu
*vcpu
)
3390 struct kvm
*kvm
= vcpu
->kvm
;
3391 struct kvm_memslots
*slots
;
3395 idx
= srcu_read_lock(&kvm
->srcu
);
3396 slots
= kvm_memslots(kvm
);
3397 for (i
= 0; i
< KVM_MEMORY_SLOTS
; ++i
) {
3398 struct kvm_memory_slot
*m
= &slots
->memslots
[i
];
3399 struct kvm_rmap_desc
*d
;
3401 for (j
= 0; j
< m
->npages
; ++j
) {
3402 unsigned long *rmapp
= &m
->rmap
[j
];
3406 if (!(*rmapp
& 1)) {
3410 d
= (struct kvm_rmap_desc
*)(*rmapp
& ~1ul);
3412 for (k
= 0; k
< RMAP_EXT
; ++k
)
3421 srcu_read_unlock(&kvm
->srcu
, idx
);
3425 void inspect_spte_has_rmap(struct kvm
*kvm
, u64
*sptep
)
3427 unsigned long *rmapp
;
3428 struct kvm_mmu_page
*rev_sp
;
3431 if (is_writable_pte(*sptep
)) {
3432 rev_sp
= page_header(__pa(sptep
));
3433 gfn
= kvm_mmu_page_get_gfn(rev_sp
, sptep
- rev_sp
->spt
);
3435 if (!gfn_to_memslot(kvm
, gfn
)) {
3436 if (!printk_ratelimit())
3438 printk(KERN_ERR
"%s: no memslot for gfn %ld\n",
3440 printk(KERN_ERR
"%s: index %ld of sp (gfn=%lx)\n",
3441 audit_msg
, (long int)(sptep
- rev_sp
->spt
),
3447 rmapp
= gfn_to_rmap(kvm
, gfn
, rev_sp
->role
.level
);
3449 if (!printk_ratelimit())
3451 printk(KERN_ERR
"%s: no rmap for writable spte %llx\n",
3459 void audit_writable_sptes_have_rmaps(struct kvm_vcpu
*vcpu
)
3461 mmu_spte_walk(vcpu
, inspect_spte_has_rmap
);
3464 static void check_writable_mappings_rmap(struct kvm_vcpu
*vcpu
)
3466 struct kvm_mmu_page
*sp
;
3469 list_for_each_entry(sp
, &vcpu
->kvm
->arch
.active_mmu_pages
, link
) {
3472 if (sp
->role
.level
!= PT_PAGE_TABLE_LEVEL
)
3475 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
3478 if (!(ent
& PT_PRESENT_MASK
))
3480 if (!is_writable_pte(ent
))
3482 inspect_spte_has_rmap(vcpu
->kvm
, &pt
[i
]);
3488 static void audit_rmap(struct kvm_vcpu
*vcpu
)
3490 check_writable_mappings_rmap(vcpu
);
3494 static void audit_write_protection(struct kvm_vcpu
*vcpu
)
3496 struct kvm_mmu_page
*sp
;
3497 struct kvm_memory_slot
*slot
;
3498 unsigned long *rmapp
;
3502 list_for_each_entry(sp
, &vcpu
->kvm
->arch
.active_mmu_pages
, link
) {
3503 if (sp
->role
.direct
)
3508 slot
= gfn_to_memslot(vcpu
->kvm
, sp
->gfn
);
3509 rmapp
= &slot
->rmap
[gfn
- slot
->base_gfn
];
3511 spte
= rmap_next(vcpu
->kvm
, rmapp
, NULL
);
3513 if (is_writable_pte(*spte
))
3514 printk(KERN_ERR
"%s: (%s) shadow page has "
3515 "writable mappings: gfn %lx role %x\n",
3516 __func__
, audit_msg
, sp
->gfn
,
3518 spte
= rmap_next(vcpu
->kvm
, rmapp
, spte
);
3523 static void kvm_mmu_audit(struct kvm_vcpu
*vcpu
, const char *msg
)
3530 audit_write_protection(vcpu
);
3531 if (strcmp("pre pte write", audit_msg
) != 0)
3532 audit_mappings(vcpu
);
3533 audit_writable_sptes_have_rmaps(vcpu
);