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 affiliates.
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.
24 #include "kvm_cache_regs.h"
27 #include <linux/kvm_host.h>
28 #include <linux/types.h>
29 #include <linux/string.h>
31 #include <linux/highmem.h>
32 #include <linux/module.h>
33 #include <linux/swap.h>
34 #include <linux/hugetlb.h>
35 #include <linux/compiler.h>
36 #include <linux/srcu.h>
37 #include <linux/slab.h>
38 #include <linux/uaccess.h>
41 #include <asm/cmpxchg.h>
46 * When setting this variable to true it enables Two-Dimensional-Paging
47 * where the hardware walks 2 page tables:
48 * 1. the guest-virtual to guest-physical
49 * 2. while doing 1. it walks guest-physical to host-physical
50 * If the hardware supports that we don't need to do shadow paging.
52 bool tdp_enabled
= false;
56 AUDIT_POST_PAGE_FAULT
,
63 char *audit_point_name
[] = {
76 #define pgprintk(x...) do { if (dbg) printk(x); } while (0)
77 #define rmap_printk(x...) do { if (dbg) printk(x); } while (0)
81 #define pgprintk(x...) do { } while (0)
82 #define rmap_printk(x...) do { } while (0)
88 module_param(dbg
, bool, 0644);
91 static int oos_shadow
= 1;
92 module_param(oos_shadow
, bool, 0644);
95 #define ASSERT(x) do { } while (0)
99 printk(KERN_WARNING "assertion failed %s:%d: %s\n", \
100 __FILE__, __LINE__, #x); \
104 #define PTE_PREFETCH_NUM 8
106 #define PT_FIRST_AVAIL_BITS_SHIFT 9
107 #define PT64_SECOND_AVAIL_BITS_SHIFT 52
109 #define PT64_LEVEL_BITS 9
111 #define PT64_LEVEL_SHIFT(level) \
112 (PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS)
114 #define PT64_INDEX(address, level)\
115 (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1))
118 #define PT32_LEVEL_BITS 10
120 #define PT32_LEVEL_SHIFT(level) \
121 (PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS)
123 #define PT32_LVL_OFFSET_MASK(level) \
124 (PT32_BASE_ADDR_MASK & ((1ULL << (PAGE_SHIFT + (((level) - 1) \
125 * PT32_LEVEL_BITS))) - 1))
127 #define PT32_INDEX(address, level)\
128 (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1))
131 #define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1))
132 #define PT64_DIR_BASE_ADDR_MASK \
133 (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1))
134 #define PT64_LVL_ADDR_MASK(level) \
135 (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + (((level) - 1) \
136 * PT64_LEVEL_BITS))) - 1))
137 #define PT64_LVL_OFFSET_MASK(level) \
138 (PT64_BASE_ADDR_MASK & ((1ULL << (PAGE_SHIFT + (((level) - 1) \
139 * PT64_LEVEL_BITS))) - 1))
141 #define PT32_BASE_ADDR_MASK PAGE_MASK
142 #define PT32_DIR_BASE_ADDR_MASK \
143 (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1))
144 #define PT32_LVL_ADDR_MASK(level) \
145 (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + (((level) - 1) \
146 * PT32_LEVEL_BITS))) - 1))
148 #define PT64_PERM_MASK (PT_PRESENT_MASK | PT_WRITABLE_MASK | PT_USER_MASK \
151 #define PTE_LIST_EXT 4
153 #define ACC_EXEC_MASK 1
154 #define ACC_WRITE_MASK PT_WRITABLE_MASK
155 #define ACC_USER_MASK PT_USER_MASK
156 #define ACC_ALL (ACC_EXEC_MASK | ACC_WRITE_MASK | ACC_USER_MASK)
158 #include <trace/events/kvm.h>
160 #define CREATE_TRACE_POINTS
161 #include "mmutrace.h"
163 #define SPTE_HOST_WRITEABLE (1ULL << PT_FIRST_AVAIL_BITS_SHIFT)
165 #define SHADOW_PT_INDEX(addr, level) PT64_INDEX(addr, level)
167 struct pte_list_desc
{
168 u64
*sptes
[PTE_LIST_EXT
];
169 struct pte_list_desc
*more
;
172 struct kvm_shadow_walk_iterator
{
180 #define for_each_shadow_entry(_vcpu, _addr, _walker) \
181 for (shadow_walk_init(&(_walker), _vcpu, _addr); \
182 shadow_walk_okay(&(_walker)); \
183 shadow_walk_next(&(_walker)))
185 static struct kmem_cache
*pte_list_desc_cache
;
186 static struct kmem_cache
*mmu_page_header_cache
;
187 static struct percpu_counter kvm_total_used_mmu_pages
;
189 static u64 __read_mostly shadow_trap_nonpresent_pte
;
190 static u64 __read_mostly shadow_notrap_nonpresent_pte
;
191 static u64 __read_mostly shadow_nx_mask
;
192 static u64 __read_mostly shadow_x_mask
; /* mutual exclusive with nx_mask */
193 static u64 __read_mostly shadow_user_mask
;
194 static u64 __read_mostly shadow_accessed_mask
;
195 static u64 __read_mostly shadow_dirty_mask
;
197 static inline u64
rsvd_bits(int s
, int e
)
199 return ((1ULL << (e
- s
+ 1)) - 1) << s
;
202 void kvm_mmu_set_nonpresent_ptes(u64 trap_pte
, u64 notrap_pte
)
204 shadow_trap_nonpresent_pte
= trap_pte
;
205 shadow_notrap_nonpresent_pte
= notrap_pte
;
207 EXPORT_SYMBOL_GPL(kvm_mmu_set_nonpresent_ptes
);
209 void kvm_mmu_set_mask_ptes(u64 user_mask
, u64 accessed_mask
,
210 u64 dirty_mask
, u64 nx_mask
, u64 x_mask
)
212 shadow_user_mask
= user_mask
;
213 shadow_accessed_mask
= accessed_mask
;
214 shadow_dirty_mask
= dirty_mask
;
215 shadow_nx_mask
= nx_mask
;
216 shadow_x_mask
= x_mask
;
218 EXPORT_SYMBOL_GPL(kvm_mmu_set_mask_ptes
);
220 static bool is_write_protection(struct kvm_vcpu
*vcpu
)
222 return kvm_read_cr0_bits(vcpu
, X86_CR0_WP
);
225 static int is_cpuid_PSE36(void)
230 static int is_nx(struct kvm_vcpu
*vcpu
)
232 return vcpu
->arch
.efer
& EFER_NX
;
235 static int is_shadow_present_pte(u64 pte
)
237 return pte
!= shadow_trap_nonpresent_pte
238 && pte
!= shadow_notrap_nonpresent_pte
;
241 static int is_large_pte(u64 pte
)
243 return pte
& PT_PAGE_SIZE_MASK
;
246 static int is_writable_pte(unsigned long pte
)
248 return pte
& PT_WRITABLE_MASK
;
251 static int is_dirty_gpte(unsigned long pte
)
253 return pte
& PT_DIRTY_MASK
;
256 static int is_rmap_spte(u64 pte
)
258 return is_shadow_present_pte(pte
);
261 static int is_last_spte(u64 pte
, int level
)
263 if (level
== PT_PAGE_TABLE_LEVEL
)
265 if (is_large_pte(pte
))
270 static pfn_t
spte_to_pfn(u64 pte
)
272 return (pte
& PT64_BASE_ADDR_MASK
) >> PAGE_SHIFT
;
275 static gfn_t
pse36_gfn_delta(u32 gpte
)
277 int shift
= 32 - PT32_DIR_PSE36_SHIFT
- PAGE_SHIFT
;
279 return (gpte
& PT32_DIR_PSE36_MASK
) << shift
;
282 static void __set_spte(u64
*sptep
, u64 spte
)
284 set_64bit(sptep
, spte
);
287 static u64
__xchg_spte(u64
*sptep
, u64 new_spte
)
290 return xchg(sptep
, new_spte
);
296 } while (cmpxchg64(sptep
, old_spte
, new_spte
) != old_spte
);
302 static bool spte_has_volatile_bits(u64 spte
)
304 if (!shadow_accessed_mask
)
307 if (!is_shadow_present_pte(spte
))
310 if ((spte
& shadow_accessed_mask
) &&
311 (!is_writable_pte(spte
) || (spte
& shadow_dirty_mask
)))
317 static bool spte_is_bit_cleared(u64 old_spte
, u64 new_spte
, u64 bit_mask
)
319 return (old_spte
& bit_mask
) && !(new_spte
& bit_mask
);
322 static void update_spte(u64
*sptep
, u64 new_spte
)
324 u64 mask
, old_spte
= *sptep
;
326 WARN_ON(!is_rmap_spte(new_spte
));
328 new_spte
|= old_spte
& shadow_dirty_mask
;
330 mask
= shadow_accessed_mask
;
331 if (is_writable_pte(old_spte
))
332 mask
|= shadow_dirty_mask
;
334 if (!spte_has_volatile_bits(old_spte
) || (new_spte
& mask
) == mask
)
335 __set_spte(sptep
, new_spte
);
337 old_spte
= __xchg_spte(sptep
, new_spte
);
339 if (!shadow_accessed_mask
)
342 if (spte_is_bit_cleared(old_spte
, new_spte
, shadow_accessed_mask
))
343 kvm_set_pfn_accessed(spte_to_pfn(old_spte
));
344 if (spte_is_bit_cleared(old_spte
, new_spte
, shadow_dirty_mask
))
345 kvm_set_pfn_dirty(spte_to_pfn(old_spte
));
348 static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache
*cache
,
349 struct kmem_cache
*base_cache
, int min
)
353 if (cache
->nobjs
>= min
)
355 while (cache
->nobjs
< ARRAY_SIZE(cache
->objects
)) {
356 obj
= kmem_cache_zalloc(base_cache
, GFP_KERNEL
);
359 cache
->objects
[cache
->nobjs
++] = obj
;
364 static void mmu_free_memory_cache(struct kvm_mmu_memory_cache
*mc
,
365 struct kmem_cache
*cache
)
368 kmem_cache_free(cache
, mc
->objects
[--mc
->nobjs
]);
371 static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache
*cache
,
376 if (cache
->nobjs
>= min
)
378 while (cache
->nobjs
< ARRAY_SIZE(cache
->objects
)) {
379 page
= (void *)__get_free_page(GFP_KERNEL
);
382 cache
->objects
[cache
->nobjs
++] = page
;
387 static void mmu_free_memory_cache_page(struct kvm_mmu_memory_cache
*mc
)
390 free_page((unsigned long)mc
->objects
[--mc
->nobjs
]);
393 static int mmu_topup_memory_caches(struct kvm_vcpu
*vcpu
)
397 r
= mmu_topup_memory_cache(&vcpu
->arch
.mmu_pte_list_desc_cache
,
398 pte_list_desc_cache
, 8 + PTE_PREFETCH_NUM
);
401 r
= mmu_topup_memory_cache_page(&vcpu
->arch
.mmu_page_cache
, 8);
404 r
= mmu_topup_memory_cache(&vcpu
->arch
.mmu_page_header_cache
,
405 mmu_page_header_cache
, 4);
410 static void mmu_free_memory_caches(struct kvm_vcpu
*vcpu
)
412 mmu_free_memory_cache(&vcpu
->arch
.mmu_pte_list_desc_cache
,
413 pte_list_desc_cache
);
414 mmu_free_memory_cache_page(&vcpu
->arch
.mmu_page_cache
);
415 mmu_free_memory_cache(&vcpu
->arch
.mmu_page_header_cache
,
416 mmu_page_header_cache
);
419 static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache
*mc
,
425 p
= mc
->objects
[--mc
->nobjs
];
429 static struct pte_list_desc
*mmu_alloc_pte_list_desc(struct kvm_vcpu
*vcpu
)
431 return mmu_memory_cache_alloc(&vcpu
->arch
.mmu_pte_list_desc_cache
,
432 sizeof(struct pte_list_desc
));
435 static void mmu_free_pte_list_desc(struct pte_list_desc
*pte_list_desc
)
437 kmem_cache_free(pte_list_desc_cache
, pte_list_desc
);
440 static gfn_t
kvm_mmu_page_get_gfn(struct kvm_mmu_page
*sp
, int index
)
442 if (!sp
->role
.direct
)
443 return sp
->gfns
[index
];
445 return sp
->gfn
+ (index
<< ((sp
->role
.level
- 1) * PT64_LEVEL_BITS
));
448 static void kvm_mmu_page_set_gfn(struct kvm_mmu_page
*sp
, int index
, gfn_t gfn
)
451 BUG_ON(gfn
!= kvm_mmu_page_get_gfn(sp
, index
));
453 sp
->gfns
[index
] = gfn
;
457 * Return the pointer to the large page information for a given gfn,
458 * handling slots that are not large page aligned.
460 static struct kvm_lpage_info
*lpage_info_slot(gfn_t gfn
,
461 struct kvm_memory_slot
*slot
,
466 idx
= (gfn
>> KVM_HPAGE_GFN_SHIFT(level
)) -
467 (slot
->base_gfn
>> KVM_HPAGE_GFN_SHIFT(level
));
468 return &slot
->lpage_info
[level
- 2][idx
];
471 static void account_shadowed(struct kvm
*kvm
, gfn_t gfn
)
473 struct kvm_memory_slot
*slot
;
474 struct kvm_lpage_info
*linfo
;
477 slot
= gfn_to_memslot(kvm
, gfn
);
478 for (i
= PT_DIRECTORY_LEVEL
;
479 i
< PT_PAGE_TABLE_LEVEL
+ KVM_NR_PAGE_SIZES
; ++i
) {
480 linfo
= lpage_info_slot(gfn
, slot
, i
);
481 linfo
->write_count
+= 1;
483 kvm
->arch
.indirect_shadow_pages
++;
486 static void unaccount_shadowed(struct kvm
*kvm
, gfn_t gfn
)
488 struct kvm_memory_slot
*slot
;
489 struct kvm_lpage_info
*linfo
;
492 slot
= gfn_to_memslot(kvm
, gfn
);
493 for (i
= PT_DIRECTORY_LEVEL
;
494 i
< PT_PAGE_TABLE_LEVEL
+ KVM_NR_PAGE_SIZES
; ++i
) {
495 linfo
= lpage_info_slot(gfn
, slot
, i
);
496 linfo
->write_count
-= 1;
497 WARN_ON(linfo
->write_count
< 0);
499 kvm
->arch
.indirect_shadow_pages
--;
502 static int has_wrprotected_page(struct kvm
*kvm
,
506 struct kvm_memory_slot
*slot
;
507 struct kvm_lpage_info
*linfo
;
509 slot
= gfn_to_memslot(kvm
, gfn
);
511 linfo
= lpage_info_slot(gfn
, slot
, level
);
512 return linfo
->write_count
;
518 static int host_mapping_level(struct kvm
*kvm
, gfn_t gfn
)
520 unsigned long page_size
;
523 page_size
= kvm_host_page_size(kvm
, gfn
);
525 for (i
= PT_PAGE_TABLE_LEVEL
;
526 i
< (PT_PAGE_TABLE_LEVEL
+ KVM_NR_PAGE_SIZES
); ++i
) {
527 if (page_size
>= KVM_HPAGE_SIZE(i
))
536 static struct kvm_memory_slot
*
537 gfn_to_memslot_dirty_bitmap(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
540 struct kvm_memory_slot
*slot
;
542 slot
= gfn_to_memslot(vcpu
->kvm
, gfn
);
543 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
||
544 (no_dirty_log
&& slot
->dirty_bitmap
))
550 static bool mapping_level_dirty_bitmap(struct kvm_vcpu
*vcpu
, gfn_t large_gfn
)
552 return !gfn_to_memslot_dirty_bitmap(vcpu
, large_gfn
, true);
555 static int mapping_level(struct kvm_vcpu
*vcpu
, gfn_t large_gfn
)
557 int host_level
, level
, max_level
;
559 host_level
= host_mapping_level(vcpu
->kvm
, large_gfn
);
561 if (host_level
== PT_PAGE_TABLE_LEVEL
)
564 max_level
= kvm_x86_ops
->get_lpage_level() < host_level
?
565 kvm_x86_ops
->get_lpage_level() : host_level
;
567 for (level
= PT_DIRECTORY_LEVEL
; level
<= max_level
; ++level
)
568 if (has_wrprotected_page(vcpu
->kvm
, large_gfn
, level
))
575 * Pte mapping structures:
577 * If pte_list bit zero is zero, then pte_list point to the spte.
579 * If pte_list bit zero is one, (then pte_list & ~1) points to a struct
580 * pte_list_desc containing more mappings.
582 * Returns the number of pte entries before the spte was added or zero if
583 * the spte was not added.
586 static int pte_list_add(struct kvm_vcpu
*vcpu
, u64
*spte
,
587 unsigned long *pte_list
)
589 struct pte_list_desc
*desc
;
593 rmap_printk("pte_list_add: %p %llx 0->1\n", spte
, *spte
);
594 *pte_list
= (unsigned long)spte
;
595 } else if (!(*pte_list
& 1)) {
596 rmap_printk("pte_list_add: %p %llx 1->many\n", spte
, *spte
);
597 desc
= mmu_alloc_pte_list_desc(vcpu
);
598 desc
->sptes
[0] = (u64
*)*pte_list
;
599 desc
->sptes
[1] = spte
;
600 *pte_list
= (unsigned long)desc
| 1;
603 rmap_printk("pte_list_add: %p %llx many->many\n", spte
, *spte
);
604 desc
= (struct pte_list_desc
*)(*pte_list
& ~1ul);
605 while (desc
->sptes
[PTE_LIST_EXT
-1] && desc
->more
) {
607 count
+= PTE_LIST_EXT
;
609 if (desc
->sptes
[PTE_LIST_EXT
-1]) {
610 desc
->more
= mmu_alloc_pte_list_desc(vcpu
);
613 for (i
= 0; desc
->sptes
[i
]; ++i
)
615 desc
->sptes
[i
] = spte
;
620 static u64
*pte_list_next(unsigned long *pte_list
, u64
*spte
)
622 struct pte_list_desc
*desc
;
628 else if (!(*pte_list
& 1)) {
630 return (u64
*)*pte_list
;
633 desc
= (struct pte_list_desc
*)(*pte_list
& ~1ul);
636 for (i
= 0; i
< PTE_LIST_EXT
&& desc
->sptes
[i
]; ++i
) {
637 if (prev_spte
== spte
)
638 return desc
->sptes
[i
];
639 prev_spte
= desc
->sptes
[i
];
647 pte_list_desc_remove_entry(unsigned long *pte_list
, struct pte_list_desc
*desc
,
648 int i
, struct pte_list_desc
*prev_desc
)
652 for (j
= PTE_LIST_EXT
- 1; !desc
->sptes
[j
] && j
> i
; --j
)
654 desc
->sptes
[i
] = desc
->sptes
[j
];
655 desc
->sptes
[j
] = NULL
;
658 if (!prev_desc
&& !desc
->more
)
659 *pte_list
= (unsigned long)desc
->sptes
[0];
662 prev_desc
->more
= desc
->more
;
664 *pte_list
= (unsigned long)desc
->more
| 1;
665 mmu_free_pte_list_desc(desc
);
668 static void pte_list_remove(u64
*spte
, unsigned long *pte_list
)
670 struct pte_list_desc
*desc
;
671 struct pte_list_desc
*prev_desc
;
675 printk(KERN_ERR
"pte_list_remove: %p 0->BUG\n", spte
);
677 } else if (!(*pte_list
& 1)) {
678 rmap_printk("pte_list_remove: %p 1->0\n", spte
);
679 if ((u64
*)*pte_list
!= spte
) {
680 printk(KERN_ERR
"pte_list_remove: %p 1->BUG\n", spte
);
685 rmap_printk("pte_list_remove: %p many->many\n", spte
);
686 desc
= (struct pte_list_desc
*)(*pte_list
& ~1ul);
689 for (i
= 0; i
< PTE_LIST_EXT
&& desc
->sptes
[i
]; ++i
)
690 if (desc
->sptes
[i
] == spte
) {
691 pte_list_desc_remove_entry(pte_list
,
699 pr_err("pte_list_remove: %p many->many\n", spte
);
704 typedef void (*pte_list_walk_fn
) (u64
*spte
);
705 static void pte_list_walk(unsigned long *pte_list
, pte_list_walk_fn fn
)
707 struct pte_list_desc
*desc
;
713 if (!(*pte_list
& 1))
714 return fn((u64
*)*pte_list
);
716 desc
= (struct pte_list_desc
*)(*pte_list
& ~1ul);
718 for (i
= 0; i
< PTE_LIST_EXT
&& desc
->sptes
[i
]; ++i
)
725 * Take gfn and return the reverse mapping to it.
727 static unsigned long *gfn_to_rmap(struct kvm
*kvm
, gfn_t gfn
, int level
)
729 struct kvm_memory_slot
*slot
;
730 struct kvm_lpage_info
*linfo
;
732 slot
= gfn_to_memslot(kvm
, gfn
);
733 if (likely(level
== PT_PAGE_TABLE_LEVEL
))
734 return &slot
->rmap
[gfn
- slot
->base_gfn
];
736 linfo
= lpage_info_slot(gfn
, slot
, level
);
738 return &linfo
->rmap_pde
;
741 static int rmap_add(struct kvm_vcpu
*vcpu
, u64
*spte
, gfn_t gfn
)
743 struct kvm_mmu_page
*sp
;
744 unsigned long *rmapp
;
746 if (!is_rmap_spte(*spte
))
749 sp
= page_header(__pa(spte
));
750 kvm_mmu_page_set_gfn(sp
, spte
- sp
->spt
, gfn
);
751 rmapp
= gfn_to_rmap(vcpu
->kvm
, gfn
, sp
->role
.level
);
752 return pte_list_add(vcpu
, spte
, rmapp
);
755 static u64
*rmap_next(struct kvm
*kvm
, unsigned long *rmapp
, u64
*spte
)
757 return pte_list_next(rmapp
, spte
);
760 static void rmap_remove(struct kvm
*kvm
, u64
*spte
)
762 struct kvm_mmu_page
*sp
;
764 unsigned long *rmapp
;
766 sp
= page_header(__pa(spte
));
767 gfn
= kvm_mmu_page_get_gfn(sp
, spte
- sp
->spt
);
768 rmapp
= gfn_to_rmap(kvm
, gfn
, sp
->role
.level
);
769 pte_list_remove(spte
, rmapp
);
772 static int set_spte_track_bits(u64
*sptep
, u64 new_spte
)
775 u64 old_spte
= *sptep
;
777 if (!spte_has_volatile_bits(old_spte
))
778 __set_spte(sptep
, new_spte
);
780 old_spte
= __xchg_spte(sptep
, new_spte
);
782 if (!is_rmap_spte(old_spte
))
785 pfn
= spte_to_pfn(old_spte
);
786 if (!shadow_accessed_mask
|| old_spte
& shadow_accessed_mask
)
787 kvm_set_pfn_accessed(pfn
);
788 if (!shadow_dirty_mask
|| (old_spte
& shadow_dirty_mask
))
789 kvm_set_pfn_dirty(pfn
);
793 static void drop_spte(struct kvm
*kvm
, u64
*sptep
, u64 new_spte
)
795 if (set_spte_track_bits(sptep
, new_spte
))
796 rmap_remove(kvm
, sptep
);
799 static int rmap_write_protect(struct kvm
*kvm
, u64 gfn
)
801 unsigned long *rmapp
;
803 int i
, write_protected
= 0;
805 rmapp
= gfn_to_rmap(kvm
, gfn
, PT_PAGE_TABLE_LEVEL
);
807 spte
= rmap_next(kvm
, rmapp
, NULL
);
810 BUG_ON(!(*spte
& PT_PRESENT_MASK
));
811 rmap_printk("rmap_write_protect: spte %p %llx\n", spte
, *spte
);
812 if (is_writable_pte(*spte
)) {
813 update_spte(spte
, *spte
& ~PT_WRITABLE_MASK
);
816 spte
= rmap_next(kvm
, rmapp
, spte
);
819 /* check for huge page mappings */
820 for (i
= PT_DIRECTORY_LEVEL
;
821 i
< PT_PAGE_TABLE_LEVEL
+ KVM_NR_PAGE_SIZES
; ++i
) {
822 rmapp
= gfn_to_rmap(kvm
, gfn
, i
);
823 spte
= rmap_next(kvm
, rmapp
, NULL
);
826 BUG_ON(!(*spte
& PT_PRESENT_MASK
));
827 BUG_ON((*spte
& (PT_PAGE_SIZE_MASK
|PT_PRESENT_MASK
)) != (PT_PAGE_SIZE_MASK
|PT_PRESENT_MASK
));
828 pgprintk("rmap_write_protect(large): spte %p %llx %lld\n", spte
, *spte
, gfn
);
829 if (is_writable_pte(*spte
)) {
831 shadow_trap_nonpresent_pte
);
836 spte
= rmap_next(kvm
, rmapp
, spte
);
840 return write_protected
;
843 static int kvm_unmap_rmapp(struct kvm
*kvm
, unsigned long *rmapp
,
847 int need_tlb_flush
= 0;
849 while ((spte
= rmap_next(kvm
, rmapp
, NULL
))) {
850 BUG_ON(!(*spte
& PT_PRESENT_MASK
));
851 rmap_printk("kvm_rmap_unmap_hva: spte %p %llx\n", spte
, *spte
);
852 drop_spte(kvm
, spte
, shadow_trap_nonpresent_pte
);
855 return need_tlb_flush
;
858 static int kvm_set_pte_rmapp(struct kvm
*kvm
, unsigned long *rmapp
,
863 pte_t
*ptep
= (pte_t
*)data
;
866 WARN_ON(pte_huge(*ptep
));
867 new_pfn
= pte_pfn(*ptep
);
868 spte
= rmap_next(kvm
, rmapp
, NULL
);
870 BUG_ON(!is_shadow_present_pte(*spte
));
871 rmap_printk("kvm_set_pte_rmapp: spte %p %llx\n", spte
, *spte
);
873 if (pte_write(*ptep
)) {
874 drop_spte(kvm
, spte
, shadow_trap_nonpresent_pte
);
875 spte
= rmap_next(kvm
, rmapp
, NULL
);
877 new_spte
= *spte
&~ (PT64_BASE_ADDR_MASK
);
878 new_spte
|= (u64
)new_pfn
<< PAGE_SHIFT
;
880 new_spte
&= ~PT_WRITABLE_MASK
;
881 new_spte
&= ~SPTE_HOST_WRITEABLE
;
882 new_spte
&= ~shadow_accessed_mask
;
883 set_spte_track_bits(spte
, new_spte
);
884 spte
= rmap_next(kvm
, rmapp
, spte
);
888 kvm_flush_remote_tlbs(kvm
);
893 static int kvm_handle_hva(struct kvm
*kvm
, unsigned long hva
,
895 int (*handler
)(struct kvm
*kvm
, unsigned long *rmapp
,
901 struct kvm_memslots
*slots
;
903 slots
= kvm_memslots(kvm
);
905 for (i
= 0; i
< slots
->nmemslots
; i
++) {
906 struct kvm_memory_slot
*memslot
= &slots
->memslots
[i
];
907 unsigned long start
= memslot
->userspace_addr
;
910 end
= start
+ (memslot
->npages
<< PAGE_SHIFT
);
911 if (hva
>= start
&& hva
< end
) {
912 gfn_t gfn_offset
= (hva
- start
) >> PAGE_SHIFT
;
913 gfn_t gfn
= memslot
->base_gfn
+ gfn_offset
;
915 ret
= handler(kvm
, &memslot
->rmap
[gfn_offset
], data
);
917 for (j
= 0; j
< KVM_NR_PAGE_SIZES
- 1; ++j
) {
918 struct kvm_lpage_info
*linfo
;
920 linfo
= lpage_info_slot(gfn
, memslot
,
921 PT_DIRECTORY_LEVEL
+ j
);
922 ret
|= handler(kvm
, &linfo
->rmap_pde
, data
);
924 trace_kvm_age_page(hva
, memslot
, ret
);
932 int kvm_unmap_hva(struct kvm
*kvm
, unsigned long hva
)
934 return kvm_handle_hva(kvm
, hva
, 0, kvm_unmap_rmapp
);
937 void kvm_set_spte_hva(struct kvm
*kvm
, unsigned long hva
, pte_t pte
)
939 kvm_handle_hva(kvm
, hva
, (unsigned long)&pte
, kvm_set_pte_rmapp
);
942 static int kvm_age_rmapp(struct kvm
*kvm
, unsigned long *rmapp
,
949 * Emulate the accessed bit for EPT, by checking if this page has
950 * an EPT mapping, and clearing it if it does. On the next access,
951 * a new EPT mapping will be established.
952 * This has some overhead, but not as much as the cost of swapping
953 * out actively used pages or breaking up actively used hugepages.
955 if (!shadow_accessed_mask
)
956 return kvm_unmap_rmapp(kvm
, rmapp
, data
);
958 spte
= rmap_next(kvm
, rmapp
, NULL
);
962 BUG_ON(!(_spte
& PT_PRESENT_MASK
));
963 _young
= _spte
& PT_ACCESSED_MASK
;
966 clear_bit(PT_ACCESSED_SHIFT
, (unsigned long *)spte
);
968 spte
= rmap_next(kvm
, rmapp
, spte
);
973 static int kvm_test_age_rmapp(struct kvm
*kvm
, unsigned long *rmapp
,
980 * If there's no access bit in the secondary pte set by the
981 * hardware it's up to gup-fast/gup to set the access bit in
982 * the primary pte or in the page structure.
984 if (!shadow_accessed_mask
)
987 spte
= rmap_next(kvm
, rmapp
, NULL
);
990 BUG_ON(!(_spte
& PT_PRESENT_MASK
));
991 young
= _spte
& PT_ACCESSED_MASK
;
996 spte
= rmap_next(kvm
, rmapp
, spte
);
1002 #define RMAP_RECYCLE_THRESHOLD 1000
1004 static void rmap_recycle(struct kvm_vcpu
*vcpu
, u64
*spte
, gfn_t gfn
)
1006 unsigned long *rmapp
;
1007 struct kvm_mmu_page
*sp
;
1009 sp
= page_header(__pa(spte
));
1011 rmapp
= gfn_to_rmap(vcpu
->kvm
, gfn
, sp
->role
.level
);
1013 kvm_unmap_rmapp(vcpu
->kvm
, rmapp
, 0);
1014 kvm_flush_remote_tlbs(vcpu
->kvm
);
1017 int kvm_age_hva(struct kvm
*kvm
, unsigned long hva
)
1019 return kvm_handle_hva(kvm
, hva
, 0, kvm_age_rmapp
);
1022 int kvm_test_age_hva(struct kvm
*kvm
, unsigned long hva
)
1024 return kvm_handle_hva(kvm
, hva
, 0, kvm_test_age_rmapp
);
1028 static int is_empty_shadow_page(u64
*spt
)
1033 for (pos
= spt
, end
= pos
+ PAGE_SIZE
/ sizeof(u64
); pos
!= end
; pos
++)
1034 if (is_shadow_present_pte(*pos
)) {
1035 printk(KERN_ERR
"%s: %p %llx\n", __func__
,
1044 * This value is the sum of all of the kvm instances's
1045 * kvm->arch.n_used_mmu_pages values. We need a global,
1046 * aggregate version in order to make the slab shrinker
1049 static inline void kvm_mod_used_mmu_pages(struct kvm
*kvm
, int nr
)
1051 kvm
->arch
.n_used_mmu_pages
+= nr
;
1052 percpu_counter_add(&kvm_total_used_mmu_pages
, nr
);
1055 static void kvm_mmu_free_page(struct kvm
*kvm
, struct kvm_mmu_page
*sp
)
1057 ASSERT(is_empty_shadow_page(sp
->spt
));
1058 hlist_del(&sp
->hash_link
);
1059 list_del(&sp
->link
);
1060 free_page((unsigned long)sp
->spt
);
1061 if (!sp
->role
.direct
)
1062 free_page((unsigned long)sp
->gfns
);
1063 kmem_cache_free(mmu_page_header_cache
, sp
);
1064 kvm_mod_used_mmu_pages(kvm
, -1);
1067 static unsigned kvm_page_table_hashfn(gfn_t gfn
)
1069 return gfn
& ((1 << KVM_MMU_HASH_SHIFT
) - 1);
1072 static void mmu_page_add_parent_pte(struct kvm_vcpu
*vcpu
,
1073 struct kvm_mmu_page
*sp
, u64
*parent_pte
)
1078 pte_list_add(vcpu
, parent_pte
, &sp
->parent_ptes
);
1081 static void mmu_page_remove_parent_pte(struct kvm_mmu_page
*sp
,
1084 pte_list_remove(parent_pte
, &sp
->parent_ptes
);
1087 static void drop_parent_pte(struct kvm_mmu_page
*sp
,
1090 mmu_page_remove_parent_pte(sp
, parent_pte
);
1091 __set_spte(parent_pte
, shadow_trap_nonpresent_pte
);
1094 static struct kvm_mmu_page
*kvm_mmu_alloc_page(struct kvm_vcpu
*vcpu
,
1095 u64
*parent_pte
, int direct
)
1097 struct kvm_mmu_page
*sp
;
1098 sp
= mmu_memory_cache_alloc(&vcpu
->arch
.mmu_page_header_cache
,
1100 sp
->spt
= mmu_memory_cache_alloc(&vcpu
->arch
.mmu_page_cache
, PAGE_SIZE
);
1102 sp
->gfns
= mmu_memory_cache_alloc(&vcpu
->arch
.mmu_page_cache
,
1104 set_page_private(virt_to_page(sp
->spt
), (unsigned long)sp
);
1105 list_add(&sp
->link
, &vcpu
->kvm
->arch
.active_mmu_pages
);
1106 bitmap_zero(sp
->slot_bitmap
, KVM_MEMORY_SLOTS
+ KVM_PRIVATE_MEM_SLOTS
);
1107 sp
->parent_ptes
= 0;
1108 mmu_page_add_parent_pte(vcpu
, sp
, parent_pte
);
1109 kvm_mod_used_mmu_pages(vcpu
->kvm
, +1);
1113 static void mark_unsync(u64
*spte
);
1114 static void kvm_mmu_mark_parents_unsync(struct kvm_mmu_page
*sp
)
1116 pte_list_walk(&sp
->parent_ptes
, mark_unsync
);
1119 static void mark_unsync(u64
*spte
)
1121 struct kvm_mmu_page
*sp
;
1124 sp
= page_header(__pa(spte
));
1125 index
= spte
- sp
->spt
;
1126 if (__test_and_set_bit(index
, sp
->unsync_child_bitmap
))
1128 if (sp
->unsync_children
++)
1130 kvm_mmu_mark_parents_unsync(sp
);
1133 static void nonpaging_prefetch_page(struct kvm_vcpu
*vcpu
,
1134 struct kvm_mmu_page
*sp
)
1138 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
)
1139 sp
->spt
[i
] = shadow_trap_nonpresent_pte
;
1142 static int nonpaging_sync_page(struct kvm_vcpu
*vcpu
,
1143 struct kvm_mmu_page
*sp
)
1148 static void nonpaging_invlpg(struct kvm_vcpu
*vcpu
, gva_t gva
)
1152 static void nonpaging_update_pte(struct kvm_vcpu
*vcpu
,
1153 struct kvm_mmu_page
*sp
, u64
*spte
,
1159 #define KVM_PAGE_ARRAY_NR 16
1161 struct kvm_mmu_pages
{
1162 struct mmu_page_and_offset
{
1163 struct kvm_mmu_page
*sp
;
1165 } page
[KVM_PAGE_ARRAY_NR
];
1169 #define for_each_unsync_children(bitmap, idx) \
1170 for (idx = find_first_bit(bitmap, 512); \
1172 idx = find_next_bit(bitmap, 512, idx+1))
1174 static int mmu_pages_add(struct kvm_mmu_pages
*pvec
, struct kvm_mmu_page
*sp
,
1180 for (i
=0; i
< pvec
->nr
; i
++)
1181 if (pvec
->page
[i
].sp
== sp
)
1184 pvec
->page
[pvec
->nr
].sp
= sp
;
1185 pvec
->page
[pvec
->nr
].idx
= idx
;
1187 return (pvec
->nr
== KVM_PAGE_ARRAY_NR
);
1190 static int __mmu_unsync_walk(struct kvm_mmu_page
*sp
,
1191 struct kvm_mmu_pages
*pvec
)
1193 int i
, ret
, nr_unsync_leaf
= 0;
1195 for_each_unsync_children(sp
->unsync_child_bitmap
, i
) {
1196 struct kvm_mmu_page
*child
;
1197 u64 ent
= sp
->spt
[i
];
1199 if (!is_shadow_present_pte(ent
) || is_large_pte(ent
))
1200 goto clear_child_bitmap
;
1202 child
= page_header(ent
& PT64_BASE_ADDR_MASK
);
1204 if (child
->unsync_children
) {
1205 if (mmu_pages_add(pvec
, child
, i
))
1208 ret
= __mmu_unsync_walk(child
, pvec
);
1210 goto clear_child_bitmap
;
1212 nr_unsync_leaf
+= ret
;
1215 } else if (child
->unsync
) {
1217 if (mmu_pages_add(pvec
, child
, i
))
1220 goto clear_child_bitmap
;
1225 __clear_bit(i
, sp
->unsync_child_bitmap
);
1226 sp
->unsync_children
--;
1227 WARN_ON((int)sp
->unsync_children
< 0);
1231 return nr_unsync_leaf
;
1234 static int mmu_unsync_walk(struct kvm_mmu_page
*sp
,
1235 struct kvm_mmu_pages
*pvec
)
1237 if (!sp
->unsync_children
)
1240 mmu_pages_add(pvec
, sp
, 0);
1241 return __mmu_unsync_walk(sp
, pvec
);
1244 static void kvm_unlink_unsync_page(struct kvm
*kvm
, struct kvm_mmu_page
*sp
)
1246 WARN_ON(!sp
->unsync
);
1247 trace_kvm_mmu_sync_page(sp
);
1249 --kvm
->stat
.mmu_unsync
;
1252 static int kvm_mmu_prepare_zap_page(struct kvm
*kvm
, struct kvm_mmu_page
*sp
,
1253 struct list_head
*invalid_list
);
1254 static void kvm_mmu_commit_zap_page(struct kvm
*kvm
,
1255 struct list_head
*invalid_list
);
1257 #define for_each_gfn_sp(kvm, sp, gfn, pos) \
1258 hlist_for_each_entry(sp, pos, \
1259 &(kvm)->arch.mmu_page_hash[kvm_page_table_hashfn(gfn)], hash_link) \
1260 if ((sp)->gfn != (gfn)) {} else
1262 #define for_each_gfn_indirect_valid_sp(kvm, sp, gfn, pos) \
1263 hlist_for_each_entry(sp, pos, \
1264 &(kvm)->arch.mmu_page_hash[kvm_page_table_hashfn(gfn)], hash_link) \
1265 if ((sp)->gfn != (gfn) || (sp)->role.direct || \
1266 (sp)->role.invalid) {} else
1268 /* @sp->gfn should be write-protected at the call site */
1269 static int __kvm_sync_page(struct kvm_vcpu
*vcpu
, struct kvm_mmu_page
*sp
,
1270 struct list_head
*invalid_list
, bool clear_unsync
)
1272 if (sp
->role
.cr4_pae
!= !!is_pae(vcpu
)) {
1273 kvm_mmu_prepare_zap_page(vcpu
->kvm
, sp
, invalid_list
);
1278 kvm_unlink_unsync_page(vcpu
->kvm
, sp
);
1280 if (vcpu
->arch
.mmu
.sync_page(vcpu
, sp
)) {
1281 kvm_mmu_prepare_zap_page(vcpu
->kvm
, sp
, invalid_list
);
1285 kvm_mmu_flush_tlb(vcpu
);
1289 static int kvm_sync_page_transient(struct kvm_vcpu
*vcpu
,
1290 struct kvm_mmu_page
*sp
)
1292 LIST_HEAD(invalid_list
);
1295 ret
= __kvm_sync_page(vcpu
, sp
, &invalid_list
, false);
1297 kvm_mmu_commit_zap_page(vcpu
->kvm
, &invalid_list
);
1302 static int kvm_sync_page(struct kvm_vcpu
*vcpu
, struct kvm_mmu_page
*sp
,
1303 struct list_head
*invalid_list
)
1305 return __kvm_sync_page(vcpu
, sp
, invalid_list
, true);
1308 /* @gfn should be write-protected at the call site */
1309 static void kvm_sync_pages(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1311 struct kvm_mmu_page
*s
;
1312 struct hlist_node
*node
;
1313 LIST_HEAD(invalid_list
);
1316 for_each_gfn_indirect_valid_sp(vcpu
->kvm
, s
, gfn
, node
) {
1320 WARN_ON(s
->role
.level
!= PT_PAGE_TABLE_LEVEL
);
1321 kvm_unlink_unsync_page(vcpu
->kvm
, s
);
1322 if ((s
->role
.cr4_pae
!= !!is_pae(vcpu
)) ||
1323 (vcpu
->arch
.mmu
.sync_page(vcpu
, s
))) {
1324 kvm_mmu_prepare_zap_page(vcpu
->kvm
, s
, &invalid_list
);
1330 kvm_mmu_commit_zap_page(vcpu
->kvm
, &invalid_list
);
1332 kvm_mmu_flush_tlb(vcpu
);
1335 struct mmu_page_path
{
1336 struct kvm_mmu_page
*parent
[PT64_ROOT_LEVEL
-1];
1337 unsigned int idx
[PT64_ROOT_LEVEL
-1];
1340 #define for_each_sp(pvec, sp, parents, i) \
1341 for (i = mmu_pages_next(&pvec, &parents, -1), \
1342 sp = pvec.page[i].sp; \
1343 i < pvec.nr && ({ sp = pvec.page[i].sp; 1;}); \
1344 i = mmu_pages_next(&pvec, &parents, i))
1346 static int mmu_pages_next(struct kvm_mmu_pages
*pvec
,
1347 struct mmu_page_path
*parents
,
1352 for (n
= i
+1; n
< pvec
->nr
; n
++) {
1353 struct kvm_mmu_page
*sp
= pvec
->page
[n
].sp
;
1355 if (sp
->role
.level
== PT_PAGE_TABLE_LEVEL
) {
1356 parents
->idx
[0] = pvec
->page
[n
].idx
;
1360 parents
->parent
[sp
->role
.level
-2] = sp
;
1361 parents
->idx
[sp
->role
.level
-1] = pvec
->page
[n
].idx
;
1367 static void mmu_pages_clear_parents(struct mmu_page_path
*parents
)
1369 struct kvm_mmu_page
*sp
;
1370 unsigned int level
= 0;
1373 unsigned int idx
= parents
->idx
[level
];
1375 sp
= parents
->parent
[level
];
1379 --sp
->unsync_children
;
1380 WARN_ON((int)sp
->unsync_children
< 0);
1381 __clear_bit(idx
, sp
->unsync_child_bitmap
);
1383 } while (level
< PT64_ROOT_LEVEL
-1 && !sp
->unsync_children
);
1386 static void kvm_mmu_pages_init(struct kvm_mmu_page
*parent
,
1387 struct mmu_page_path
*parents
,
1388 struct kvm_mmu_pages
*pvec
)
1390 parents
->parent
[parent
->role
.level
-1] = NULL
;
1394 static void mmu_sync_children(struct kvm_vcpu
*vcpu
,
1395 struct kvm_mmu_page
*parent
)
1398 struct kvm_mmu_page
*sp
;
1399 struct mmu_page_path parents
;
1400 struct kvm_mmu_pages pages
;
1401 LIST_HEAD(invalid_list
);
1403 kvm_mmu_pages_init(parent
, &parents
, &pages
);
1404 while (mmu_unsync_walk(parent
, &pages
)) {
1407 for_each_sp(pages
, sp
, parents
, i
)
1408 protected |= rmap_write_protect(vcpu
->kvm
, sp
->gfn
);
1411 kvm_flush_remote_tlbs(vcpu
->kvm
);
1413 for_each_sp(pages
, sp
, parents
, i
) {
1414 kvm_sync_page(vcpu
, sp
, &invalid_list
);
1415 mmu_pages_clear_parents(&parents
);
1417 kvm_mmu_commit_zap_page(vcpu
->kvm
, &invalid_list
);
1418 cond_resched_lock(&vcpu
->kvm
->mmu_lock
);
1419 kvm_mmu_pages_init(parent
, &parents
, &pages
);
1423 static struct kvm_mmu_page
*kvm_mmu_get_page(struct kvm_vcpu
*vcpu
,
1431 union kvm_mmu_page_role role
;
1433 struct kvm_mmu_page
*sp
;
1434 struct hlist_node
*node
;
1435 bool need_sync
= false;
1437 role
= vcpu
->arch
.mmu
.base_role
;
1439 role
.direct
= direct
;
1442 role
.access
= access
;
1443 if (!vcpu
->arch
.mmu
.direct_map
1444 && vcpu
->arch
.mmu
.root_level
<= PT32_ROOT_LEVEL
) {
1445 quadrant
= gaddr
>> (PAGE_SHIFT
+ (PT64_PT_BITS
* level
));
1446 quadrant
&= (1 << ((PT32_PT_BITS
- PT64_PT_BITS
) * level
)) - 1;
1447 role
.quadrant
= quadrant
;
1449 for_each_gfn_sp(vcpu
->kvm
, sp
, gfn
, node
) {
1450 if (!need_sync
&& sp
->unsync
)
1453 if (sp
->role
.word
!= role
.word
)
1456 if (sp
->unsync
&& kvm_sync_page_transient(vcpu
, sp
))
1459 mmu_page_add_parent_pte(vcpu
, sp
, parent_pte
);
1460 if (sp
->unsync_children
) {
1461 kvm_make_request(KVM_REQ_MMU_SYNC
, vcpu
);
1462 kvm_mmu_mark_parents_unsync(sp
);
1463 } else if (sp
->unsync
)
1464 kvm_mmu_mark_parents_unsync(sp
);
1466 trace_kvm_mmu_get_page(sp
, false);
1469 ++vcpu
->kvm
->stat
.mmu_cache_miss
;
1470 sp
= kvm_mmu_alloc_page(vcpu
, parent_pte
, direct
);
1475 hlist_add_head(&sp
->hash_link
,
1476 &vcpu
->kvm
->arch
.mmu_page_hash
[kvm_page_table_hashfn(gfn
)]);
1478 if (rmap_write_protect(vcpu
->kvm
, gfn
))
1479 kvm_flush_remote_tlbs(vcpu
->kvm
);
1480 if (level
> PT_PAGE_TABLE_LEVEL
&& need_sync
)
1481 kvm_sync_pages(vcpu
, gfn
);
1483 account_shadowed(vcpu
->kvm
, gfn
);
1485 if (shadow_trap_nonpresent_pte
!= shadow_notrap_nonpresent_pte
)
1486 vcpu
->arch
.mmu
.prefetch_page(vcpu
, sp
);
1488 nonpaging_prefetch_page(vcpu
, sp
);
1489 trace_kvm_mmu_get_page(sp
, true);
1493 static void shadow_walk_init(struct kvm_shadow_walk_iterator
*iterator
,
1494 struct kvm_vcpu
*vcpu
, u64 addr
)
1496 iterator
->addr
= addr
;
1497 iterator
->shadow_addr
= vcpu
->arch
.mmu
.root_hpa
;
1498 iterator
->level
= vcpu
->arch
.mmu
.shadow_root_level
;
1500 if (iterator
->level
== PT64_ROOT_LEVEL
&&
1501 vcpu
->arch
.mmu
.root_level
< PT64_ROOT_LEVEL
&&
1502 !vcpu
->arch
.mmu
.direct_map
)
1505 if (iterator
->level
== PT32E_ROOT_LEVEL
) {
1506 iterator
->shadow_addr
1507 = vcpu
->arch
.mmu
.pae_root
[(addr
>> 30) & 3];
1508 iterator
->shadow_addr
&= PT64_BASE_ADDR_MASK
;
1510 if (!iterator
->shadow_addr
)
1511 iterator
->level
= 0;
1515 static bool shadow_walk_okay(struct kvm_shadow_walk_iterator
*iterator
)
1517 if (iterator
->level
< PT_PAGE_TABLE_LEVEL
)
1520 if (iterator
->level
== PT_PAGE_TABLE_LEVEL
)
1521 if (is_large_pte(*iterator
->sptep
))
1524 iterator
->index
= SHADOW_PT_INDEX(iterator
->addr
, iterator
->level
);
1525 iterator
->sptep
= ((u64
*)__va(iterator
->shadow_addr
)) + iterator
->index
;
1529 static void shadow_walk_next(struct kvm_shadow_walk_iterator
*iterator
)
1531 iterator
->shadow_addr
= *iterator
->sptep
& PT64_BASE_ADDR_MASK
;
1535 static void link_shadow_page(u64
*sptep
, struct kvm_mmu_page
*sp
)
1539 spte
= __pa(sp
->spt
)
1540 | PT_PRESENT_MASK
| PT_ACCESSED_MASK
1541 | PT_WRITABLE_MASK
| PT_USER_MASK
;
1542 __set_spte(sptep
, spte
);
1545 static void drop_large_spte(struct kvm_vcpu
*vcpu
, u64
*sptep
)
1547 if (is_large_pte(*sptep
)) {
1548 drop_spte(vcpu
->kvm
, sptep
, shadow_trap_nonpresent_pte
);
1549 kvm_flush_remote_tlbs(vcpu
->kvm
);
1553 static void validate_direct_spte(struct kvm_vcpu
*vcpu
, u64
*sptep
,
1554 unsigned direct_access
)
1556 if (is_shadow_present_pte(*sptep
) && !is_large_pte(*sptep
)) {
1557 struct kvm_mmu_page
*child
;
1560 * For the direct sp, if the guest pte's dirty bit
1561 * changed form clean to dirty, it will corrupt the
1562 * sp's access: allow writable in the read-only sp,
1563 * so we should update the spte at this point to get
1564 * a new sp with the correct access.
1566 child
= page_header(*sptep
& PT64_BASE_ADDR_MASK
);
1567 if (child
->role
.access
== direct_access
)
1570 drop_parent_pte(child
, sptep
);
1571 kvm_flush_remote_tlbs(vcpu
->kvm
);
1575 static void mmu_page_zap_pte(struct kvm
*kvm
, struct kvm_mmu_page
*sp
,
1579 struct kvm_mmu_page
*child
;
1582 if (is_shadow_present_pte(pte
)) {
1583 if (is_last_spte(pte
, sp
->role
.level
))
1584 drop_spte(kvm
, spte
, shadow_trap_nonpresent_pte
);
1586 child
= page_header(pte
& PT64_BASE_ADDR_MASK
);
1587 drop_parent_pte(child
, spte
);
1590 __set_spte(spte
, shadow_trap_nonpresent_pte
);
1591 if (is_large_pte(pte
))
1595 static void kvm_mmu_page_unlink_children(struct kvm
*kvm
,
1596 struct kvm_mmu_page
*sp
)
1600 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
)
1601 mmu_page_zap_pte(kvm
, sp
, sp
->spt
+ i
);
1604 static void kvm_mmu_put_page(struct kvm_mmu_page
*sp
, u64
*parent_pte
)
1606 mmu_page_remove_parent_pte(sp
, parent_pte
);
1609 static void kvm_mmu_reset_last_pte_updated(struct kvm
*kvm
)
1612 struct kvm_vcpu
*vcpu
;
1614 kvm_for_each_vcpu(i
, vcpu
, kvm
)
1615 vcpu
->arch
.last_pte_updated
= NULL
;
1618 static void kvm_mmu_unlink_parents(struct kvm
*kvm
, struct kvm_mmu_page
*sp
)
1622 while ((parent_pte
= pte_list_next(&sp
->parent_ptes
, NULL
)))
1623 drop_parent_pte(sp
, parent_pte
);
1626 static int mmu_zap_unsync_children(struct kvm
*kvm
,
1627 struct kvm_mmu_page
*parent
,
1628 struct list_head
*invalid_list
)
1631 struct mmu_page_path parents
;
1632 struct kvm_mmu_pages pages
;
1634 if (parent
->role
.level
== PT_PAGE_TABLE_LEVEL
)
1637 kvm_mmu_pages_init(parent
, &parents
, &pages
);
1638 while (mmu_unsync_walk(parent
, &pages
)) {
1639 struct kvm_mmu_page
*sp
;
1641 for_each_sp(pages
, sp
, parents
, i
) {
1642 kvm_mmu_prepare_zap_page(kvm
, sp
, invalid_list
);
1643 mmu_pages_clear_parents(&parents
);
1646 kvm_mmu_pages_init(parent
, &parents
, &pages
);
1652 static int kvm_mmu_prepare_zap_page(struct kvm
*kvm
, struct kvm_mmu_page
*sp
,
1653 struct list_head
*invalid_list
)
1657 trace_kvm_mmu_prepare_zap_page(sp
);
1658 ++kvm
->stat
.mmu_shadow_zapped
;
1659 ret
= mmu_zap_unsync_children(kvm
, sp
, invalid_list
);
1660 kvm_mmu_page_unlink_children(kvm
, sp
);
1661 kvm_mmu_unlink_parents(kvm
, sp
);
1662 if (!sp
->role
.invalid
&& !sp
->role
.direct
)
1663 unaccount_shadowed(kvm
, sp
->gfn
);
1665 kvm_unlink_unsync_page(kvm
, sp
);
1666 if (!sp
->root_count
) {
1669 list_move(&sp
->link
, invalid_list
);
1671 list_move(&sp
->link
, &kvm
->arch
.active_mmu_pages
);
1672 kvm_reload_remote_mmus(kvm
);
1675 sp
->role
.invalid
= 1;
1676 kvm_mmu_reset_last_pte_updated(kvm
);
1680 static void kvm_mmu_commit_zap_page(struct kvm
*kvm
,
1681 struct list_head
*invalid_list
)
1683 struct kvm_mmu_page
*sp
;
1685 if (list_empty(invalid_list
))
1688 kvm_flush_remote_tlbs(kvm
);
1691 sp
= list_first_entry(invalid_list
, struct kvm_mmu_page
, link
);
1692 WARN_ON(!sp
->role
.invalid
|| sp
->root_count
);
1693 kvm_mmu_free_page(kvm
, sp
);
1694 } while (!list_empty(invalid_list
));
1699 * Changing the number of mmu pages allocated to the vm
1700 * Note: if goal_nr_mmu_pages is too small, you will get dead lock
1702 void kvm_mmu_change_mmu_pages(struct kvm
*kvm
, unsigned int goal_nr_mmu_pages
)
1704 LIST_HEAD(invalid_list
);
1706 * If we set the number of mmu pages to be smaller be than the
1707 * number of actived pages , we must to free some mmu pages before we
1711 if (kvm
->arch
.n_used_mmu_pages
> goal_nr_mmu_pages
) {
1712 while (kvm
->arch
.n_used_mmu_pages
> goal_nr_mmu_pages
&&
1713 !list_empty(&kvm
->arch
.active_mmu_pages
)) {
1714 struct kvm_mmu_page
*page
;
1716 page
= container_of(kvm
->arch
.active_mmu_pages
.prev
,
1717 struct kvm_mmu_page
, link
);
1718 kvm_mmu_prepare_zap_page(kvm
, page
, &invalid_list
);
1719 kvm_mmu_commit_zap_page(kvm
, &invalid_list
);
1721 goal_nr_mmu_pages
= kvm
->arch
.n_used_mmu_pages
;
1724 kvm
->arch
.n_max_mmu_pages
= goal_nr_mmu_pages
;
1727 static int kvm_mmu_unprotect_page(struct kvm
*kvm
, gfn_t gfn
)
1729 struct kvm_mmu_page
*sp
;
1730 struct hlist_node
*node
;
1731 LIST_HEAD(invalid_list
);
1734 pgprintk("%s: looking for gfn %llx\n", __func__
, gfn
);
1737 for_each_gfn_indirect_valid_sp(kvm
, sp
, gfn
, node
) {
1738 pgprintk("%s: gfn %llx role %x\n", __func__
, gfn
,
1741 kvm_mmu_prepare_zap_page(kvm
, sp
, &invalid_list
);
1743 kvm_mmu_commit_zap_page(kvm
, &invalid_list
);
1747 static void mmu_unshadow(struct kvm
*kvm
, gfn_t gfn
)
1749 struct kvm_mmu_page
*sp
;
1750 struct hlist_node
*node
;
1751 LIST_HEAD(invalid_list
);
1753 for_each_gfn_indirect_valid_sp(kvm
, sp
, gfn
, node
) {
1754 pgprintk("%s: zap %llx %x\n",
1755 __func__
, gfn
, sp
->role
.word
);
1756 kvm_mmu_prepare_zap_page(kvm
, sp
, &invalid_list
);
1758 kvm_mmu_commit_zap_page(kvm
, &invalid_list
);
1761 static void page_header_update_slot(struct kvm
*kvm
, void *pte
, gfn_t gfn
)
1763 int slot
= memslot_id(kvm
, gfn
);
1764 struct kvm_mmu_page
*sp
= page_header(__pa(pte
));
1766 __set_bit(slot
, sp
->slot_bitmap
);
1769 static void mmu_convert_notrap(struct kvm_mmu_page
*sp
)
1774 if (shadow_trap_nonpresent_pte
== shadow_notrap_nonpresent_pte
)
1777 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
1778 if (pt
[i
] == shadow_notrap_nonpresent_pte
)
1779 __set_spte(&pt
[i
], shadow_trap_nonpresent_pte
);
1784 * The function is based on mtrr_type_lookup() in
1785 * arch/x86/kernel/cpu/mtrr/generic.c
1787 static int get_mtrr_type(struct mtrr_state_type
*mtrr_state
,
1792 u8 prev_match
, curr_match
;
1793 int num_var_ranges
= KVM_NR_VAR_MTRR
;
1795 if (!mtrr_state
->enabled
)
1798 /* Make end inclusive end, instead of exclusive */
1801 /* Look in fixed ranges. Just return the type as per start */
1802 if (mtrr_state
->have_fixed
&& (start
< 0x100000)) {
1805 if (start
< 0x80000) {
1807 idx
+= (start
>> 16);
1808 return mtrr_state
->fixed_ranges
[idx
];
1809 } else if (start
< 0xC0000) {
1811 idx
+= ((start
- 0x80000) >> 14);
1812 return mtrr_state
->fixed_ranges
[idx
];
1813 } else if (start
< 0x1000000) {
1815 idx
+= ((start
- 0xC0000) >> 12);
1816 return mtrr_state
->fixed_ranges
[idx
];
1821 * Look in variable ranges
1822 * Look of multiple ranges matching this address and pick type
1823 * as per MTRR precedence
1825 if (!(mtrr_state
->enabled
& 2))
1826 return mtrr_state
->def_type
;
1829 for (i
= 0; i
< num_var_ranges
; ++i
) {
1830 unsigned short start_state
, end_state
;
1832 if (!(mtrr_state
->var_ranges
[i
].mask_lo
& (1 << 11)))
1835 base
= (((u64
)mtrr_state
->var_ranges
[i
].base_hi
) << 32) +
1836 (mtrr_state
->var_ranges
[i
].base_lo
& PAGE_MASK
);
1837 mask
= (((u64
)mtrr_state
->var_ranges
[i
].mask_hi
) << 32) +
1838 (mtrr_state
->var_ranges
[i
].mask_lo
& PAGE_MASK
);
1840 start_state
= ((start
& mask
) == (base
& mask
));
1841 end_state
= ((end
& mask
) == (base
& mask
));
1842 if (start_state
!= end_state
)
1845 if ((start
& mask
) != (base
& mask
))
1848 curr_match
= mtrr_state
->var_ranges
[i
].base_lo
& 0xff;
1849 if (prev_match
== 0xFF) {
1850 prev_match
= curr_match
;
1854 if (prev_match
== MTRR_TYPE_UNCACHABLE
||
1855 curr_match
== MTRR_TYPE_UNCACHABLE
)
1856 return MTRR_TYPE_UNCACHABLE
;
1858 if ((prev_match
== MTRR_TYPE_WRBACK
&&
1859 curr_match
== MTRR_TYPE_WRTHROUGH
) ||
1860 (prev_match
== MTRR_TYPE_WRTHROUGH
&&
1861 curr_match
== MTRR_TYPE_WRBACK
)) {
1862 prev_match
= MTRR_TYPE_WRTHROUGH
;
1863 curr_match
= MTRR_TYPE_WRTHROUGH
;
1866 if (prev_match
!= curr_match
)
1867 return MTRR_TYPE_UNCACHABLE
;
1870 if (prev_match
!= 0xFF)
1873 return mtrr_state
->def_type
;
1876 u8
kvm_get_guest_memory_type(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1880 mtrr
= get_mtrr_type(&vcpu
->arch
.mtrr_state
, gfn
<< PAGE_SHIFT
,
1881 (gfn
<< PAGE_SHIFT
) + PAGE_SIZE
);
1882 if (mtrr
== 0xfe || mtrr
== 0xff)
1883 mtrr
= MTRR_TYPE_WRBACK
;
1886 EXPORT_SYMBOL_GPL(kvm_get_guest_memory_type
);
1888 static void __kvm_unsync_page(struct kvm_vcpu
*vcpu
, struct kvm_mmu_page
*sp
)
1890 trace_kvm_mmu_unsync_page(sp
);
1891 ++vcpu
->kvm
->stat
.mmu_unsync
;
1894 kvm_mmu_mark_parents_unsync(sp
);
1895 mmu_convert_notrap(sp
);
1898 static void kvm_unsync_pages(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1900 struct kvm_mmu_page
*s
;
1901 struct hlist_node
*node
;
1903 for_each_gfn_indirect_valid_sp(vcpu
->kvm
, s
, gfn
, node
) {
1906 WARN_ON(s
->role
.level
!= PT_PAGE_TABLE_LEVEL
);
1907 __kvm_unsync_page(vcpu
, s
);
1911 static int mmu_need_write_protect(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
1914 struct kvm_mmu_page
*s
;
1915 struct hlist_node
*node
;
1916 bool need_unsync
= false;
1918 for_each_gfn_indirect_valid_sp(vcpu
->kvm
, s
, gfn
, node
) {
1922 if (s
->role
.level
!= PT_PAGE_TABLE_LEVEL
)
1925 if (!need_unsync
&& !s
->unsync
) {
1932 kvm_unsync_pages(vcpu
, gfn
);
1936 static int set_spte(struct kvm_vcpu
*vcpu
, u64
*sptep
,
1937 unsigned pte_access
, int user_fault
,
1938 int write_fault
, int dirty
, int level
,
1939 gfn_t gfn
, pfn_t pfn
, bool speculative
,
1940 bool can_unsync
, bool host_writable
)
1942 u64 spte
, entry
= *sptep
;
1946 * We don't set the accessed bit, since we sometimes want to see
1947 * whether the guest actually used the pte (in order to detect
1950 spte
= PT_PRESENT_MASK
;
1952 spte
|= shadow_accessed_mask
;
1954 pte_access
&= ~ACC_WRITE_MASK
;
1955 if (pte_access
& ACC_EXEC_MASK
)
1956 spte
|= shadow_x_mask
;
1958 spte
|= shadow_nx_mask
;
1959 if (pte_access
& ACC_USER_MASK
)
1960 spte
|= shadow_user_mask
;
1961 if (level
> PT_PAGE_TABLE_LEVEL
)
1962 spte
|= PT_PAGE_SIZE_MASK
;
1964 spte
|= kvm_x86_ops
->get_mt_mask(vcpu
, gfn
,
1965 kvm_is_mmio_pfn(pfn
));
1968 spte
|= SPTE_HOST_WRITEABLE
;
1970 pte_access
&= ~ACC_WRITE_MASK
;
1972 spte
|= (u64
)pfn
<< PAGE_SHIFT
;
1974 if ((pte_access
& ACC_WRITE_MASK
)
1975 || (!vcpu
->arch
.mmu
.direct_map
&& write_fault
1976 && !is_write_protection(vcpu
) && !user_fault
)) {
1978 if (level
> PT_PAGE_TABLE_LEVEL
&&
1979 has_wrprotected_page(vcpu
->kvm
, gfn
, level
)) {
1981 drop_spte(vcpu
->kvm
, sptep
, shadow_trap_nonpresent_pte
);
1985 spte
|= PT_WRITABLE_MASK
;
1987 if (!vcpu
->arch
.mmu
.direct_map
1988 && !(pte_access
& ACC_WRITE_MASK
)) {
1989 spte
&= ~PT_USER_MASK
;
1991 * If we converted a user page to a kernel page,
1992 * so that the kernel can write to it when cr0.wp=0,
1993 * then we should prevent the kernel from executing it
1994 * if SMEP is enabled.
1996 if (kvm_read_cr4_bits(vcpu
, X86_CR4_SMEP
))
1997 spte
|= PT64_NX_MASK
;
2001 * Optimization: for pte sync, if spte was writable the hash
2002 * lookup is unnecessary (and expensive). Write protection
2003 * is responsibility of mmu_get_page / kvm_sync_page.
2004 * Same reasoning can be applied to dirty page accounting.
2006 if (!can_unsync
&& is_writable_pte(*sptep
))
2009 if (mmu_need_write_protect(vcpu
, gfn
, can_unsync
)) {
2010 pgprintk("%s: found shadow page for %llx, marking ro\n",
2013 pte_access
&= ~ACC_WRITE_MASK
;
2014 if (is_writable_pte(spte
))
2015 spte
&= ~PT_WRITABLE_MASK
;
2019 if (pte_access
& ACC_WRITE_MASK
)
2020 mark_page_dirty(vcpu
->kvm
, gfn
);
2023 update_spte(sptep
, spte
);
2025 * If we overwrite a writable spte with a read-only one we
2026 * should flush remote TLBs. Otherwise rmap_write_protect
2027 * will find a read-only spte, even though the writable spte
2028 * might be cached on a CPU's TLB.
2030 if (is_writable_pte(entry
) && !is_writable_pte(*sptep
))
2031 kvm_flush_remote_tlbs(vcpu
->kvm
);
2036 static void mmu_set_spte(struct kvm_vcpu
*vcpu
, u64
*sptep
,
2037 unsigned pt_access
, unsigned pte_access
,
2038 int user_fault
, int write_fault
, int dirty
,
2039 int *ptwrite
, int level
, gfn_t gfn
,
2040 pfn_t pfn
, bool speculative
,
2043 int was_rmapped
= 0;
2046 pgprintk("%s: spte %llx access %x write_fault %d"
2047 " user_fault %d gfn %llx\n",
2048 __func__
, *sptep
, pt_access
,
2049 write_fault
, user_fault
, gfn
);
2051 if (is_rmap_spte(*sptep
)) {
2053 * If we overwrite a PTE page pointer with a 2MB PMD, unlink
2054 * the parent of the now unreachable PTE.
2056 if (level
> PT_PAGE_TABLE_LEVEL
&&
2057 !is_large_pte(*sptep
)) {
2058 struct kvm_mmu_page
*child
;
2061 child
= page_header(pte
& PT64_BASE_ADDR_MASK
);
2062 drop_parent_pte(child
, sptep
);
2063 kvm_flush_remote_tlbs(vcpu
->kvm
);
2064 } else if (pfn
!= spte_to_pfn(*sptep
)) {
2065 pgprintk("hfn old %llx new %llx\n",
2066 spte_to_pfn(*sptep
), pfn
);
2067 drop_spte(vcpu
->kvm
, sptep
, shadow_trap_nonpresent_pte
);
2068 kvm_flush_remote_tlbs(vcpu
->kvm
);
2073 if (set_spte(vcpu
, sptep
, pte_access
, user_fault
, write_fault
,
2074 dirty
, level
, gfn
, pfn
, speculative
, true,
2078 kvm_mmu_flush_tlb(vcpu
);
2081 pgprintk("%s: setting spte %llx\n", __func__
, *sptep
);
2082 pgprintk("instantiating %s PTE (%s) at %llx (%llx) addr %p\n",
2083 is_large_pte(*sptep
)? "2MB" : "4kB",
2084 *sptep
& PT_PRESENT_MASK
?"RW":"R", gfn
,
2086 if (!was_rmapped
&& is_large_pte(*sptep
))
2087 ++vcpu
->kvm
->stat
.lpages
;
2089 page_header_update_slot(vcpu
->kvm
, sptep
, gfn
);
2091 rmap_count
= rmap_add(vcpu
, sptep
, gfn
);
2092 if (rmap_count
> RMAP_RECYCLE_THRESHOLD
)
2093 rmap_recycle(vcpu
, sptep
, gfn
);
2095 kvm_release_pfn_clean(pfn
);
2097 vcpu
->arch
.last_pte_updated
= sptep
;
2098 vcpu
->arch
.last_pte_gfn
= gfn
;
2102 static void nonpaging_new_cr3(struct kvm_vcpu
*vcpu
)
2106 static pfn_t
pte_prefetch_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
2109 struct kvm_memory_slot
*slot
;
2112 slot
= gfn_to_memslot_dirty_bitmap(vcpu
, gfn
, no_dirty_log
);
2115 return page_to_pfn(bad_page
);
2118 hva
= gfn_to_hva_memslot(slot
, gfn
);
2120 return hva_to_pfn_atomic(vcpu
->kvm
, hva
);
2123 static int direct_pte_prefetch_many(struct kvm_vcpu
*vcpu
,
2124 struct kvm_mmu_page
*sp
,
2125 u64
*start
, u64
*end
)
2127 struct page
*pages
[PTE_PREFETCH_NUM
];
2128 unsigned access
= sp
->role
.access
;
2132 gfn
= kvm_mmu_page_get_gfn(sp
, start
- sp
->spt
);
2133 if (!gfn_to_memslot_dirty_bitmap(vcpu
, gfn
, access
& ACC_WRITE_MASK
))
2136 ret
= gfn_to_page_many_atomic(vcpu
->kvm
, gfn
, pages
, end
- start
);
2140 for (i
= 0; i
< ret
; i
++, gfn
++, start
++)
2141 mmu_set_spte(vcpu
, start
, ACC_ALL
,
2142 access
, 0, 0, 1, NULL
,
2143 sp
->role
.level
, gfn
,
2144 page_to_pfn(pages
[i
]), true, true);
2149 static void __direct_pte_prefetch(struct kvm_vcpu
*vcpu
,
2150 struct kvm_mmu_page
*sp
, u64
*sptep
)
2152 u64
*spte
, *start
= NULL
;
2155 WARN_ON(!sp
->role
.direct
);
2157 i
= (sptep
- sp
->spt
) & ~(PTE_PREFETCH_NUM
- 1);
2160 for (i
= 0; i
< PTE_PREFETCH_NUM
; i
++, spte
++) {
2161 if (*spte
!= shadow_trap_nonpresent_pte
|| spte
== sptep
) {
2164 if (direct_pte_prefetch_many(vcpu
, sp
, start
, spte
) < 0)
2172 static void direct_pte_prefetch(struct kvm_vcpu
*vcpu
, u64
*sptep
)
2174 struct kvm_mmu_page
*sp
;
2177 * Since it's no accessed bit on EPT, it's no way to
2178 * distinguish between actually accessed translations
2179 * and prefetched, so disable pte prefetch if EPT is
2182 if (!shadow_accessed_mask
)
2185 sp
= page_header(__pa(sptep
));
2186 if (sp
->role
.level
> PT_PAGE_TABLE_LEVEL
)
2189 __direct_pte_prefetch(vcpu
, sp
, sptep
);
2192 static int __direct_map(struct kvm_vcpu
*vcpu
, gpa_t v
, int write
,
2193 int map_writable
, int level
, gfn_t gfn
, pfn_t pfn
,
2196 struct kvm_shadow_walk_iterator iterator
;
2197 struct kvm_mmu_page
*sp
;
2201 for_each_shadow_entry(vcpu
, (u64
)gfn
<< PAGE_SHIFT
, iterator
) {
2202 if (iterator
.level
== level
) {
2203 unsigned pte_access
= ACC_ALL
;
2205 mmu_set_spte(vcpu
, iterator
.sptep
, ACC_ALL
, pte_access
,
2206 0, write
, 1, &pt_write
,
2207 level
, gfn
, pfn
, prefault
, map_writable
);
2208 direct_pte_prefetch(vcpu
, iterator
.sptep
);
2209 ++vcpu
->stat
.pf_fixed
;
2213 if (*iterator
.sptep
== shadow_trap_nonpresent_pte
) {
2214 u64 base_addr
= iterator
.addr
;
2216 base_addr
&= PT64_LVL_ADDR_MASK(iterator
.level
);
2217 pseudo_gfn
= base_addr
>> PAGE_SHIFT
;
2218 sp
= kvm_mmu_get_page(vcpu
, pseudo_gfn
, iterator
.addr
,
2220 1, ACC_ALL
, iterator
.sptep
);
2222 pgprintk("nonpaging_map: ENOMEM\n");
2223 kvm_release_pfn_clean(pfn
);
2227 __set_spte(iterator
.sptep
,
2229 | PT_PRESENT_MASK
| PT_WRITABLE_MASK
2230 | shadow_user_mask
| shadow_x_mask
2231 | shadow_accessed_mask
);
2237 static void kvm_send_hwpoison_signal(unsigned long address
, struct task_struct
*tsk
)
2241 info
.si_signo
= SIGBUS
;
2243 info
.si_code
= BUS_MCEERR_AR
;
2244 info
.si_addr
= (void __user
*)address
;
2245 info
.si_addr_lsb
= PAGE_SHIFT
;
2247 send_sig_info(SIGBUS
, &info
, tsk
);
2250 static int kvm_handle_bad_page(struct kvm
*kvm
, gfn_t gfn
, pfn_t pfn
)
2252 kvm_release_pfn_clean(pfn
);
2253 if (is_hwpoison_pfn(pfn
)) {
2254 kvm_send_hwpoison_signal(gfn_to_hva(kvm
, gfn
), current
);
2256 } else if (is_fault_pfn(pfn
))
2262 static void transparent_hugepage_adjust(struct kvm_vcpu
*vcpu
,
2263 gfn_t
*gfnp
, pfn_t
*pfnp
, int *levelp
)
2267 int level
= *levelp
;
2270 * Check if it's a transparent hugepage. If this would be an
2271 * hugetlbfs page, level wouldn't be set to
2272 * PT_PAGE_TABLE_LEVEL and there would be no adjustment done
2275 if (!is_error_pfn(pfn
) && !kvm_is_mmio_pfn(pfn
) &&
2276 level
== PT_PAGE_TABLE_LEVEL
&&
2277 PageTransCompound(pfn_to_page(pfn
)) &&
2278 !has_wrprotected_page(vcpu
->kvm
, gfn
, PT_DIRECTORY_LEVEL
)) {
2281 * mmu_notifier_retry was successful and we hold the
2282 * mmu_lock here, so the pmd can't become splitting
2283 * from under us, and in turn
2284 * __split_huge_page_refcount() can't run from under
2285 * us and we can safely transfer the refcount from
2286 * PG_tail to PG_head as we switch the pfn to tail to
2289 *levelp
= level
= PT_DIRECTORY_LEVEL
;
2290 mask
= KVM_PAGES_PER_HPAGE(level
) - 1;
2291 VM_BUG_ON((gfn
& mask
) != (pfn
& mask
));
2295 kvm_release_pfn_clean(pfn
);
2297 if (!get_page_unless_zero(pfn_to_page(pfn
)))
2304 static bool try_async_pf(struct kvm_vcpu
*vcpu
, bool prefault
, gfn_t gfn
,
2305 gva_t gva
, pfn_t
*pfn
, bool write
, bool *writable
);
2307 static int nonpaging_map(struct kvm_vcpu
*vcpu
, gva_t v
, int write
, gfn_t gfn
,
2314 unsigned long mmu_seq
;
2317 force_pt_level
= mapping_level_dirty_bitmap(vcpu
, gfn
);
2318 if (likely(!force_pt_level
)) {
2319 level
= mapping_level(vcpu
, gfn
);
2321 * This path builds a PAE pagetable - so we can map
2322 * 2mb pages at maximum. Therefore check if the level
2323 * is larger than that.
2325 if (level
> PT_DIRECTORY_LEVEL
)
2326 level
= PT_DIRECTORY_LEVEL
;
2328 gfn
&= ~(KVM_PAGES_PER_HPAGE(level
) - 1);
2330 level
= PT_PAGE_TABLE_LEVEL
;
2332 mmu_seq
= vcpu
->kvm
->mmu_notifier_seq
;
2335 if (try_async_pf(vcpu
, prefault
, gfn
, v
, &pfn
, write
, &map_writable
))
2339 if (is_error_pfn(pfn
))
2340 return kvm_handle_bad_page(vcpu
->kvm
, gfn
, pfn
);
2342 spin_lock(&vcpu
->kvm
->mmu_lock
);
2343 if (mmu_notifier_retry(vcpu
, mmu_seq
))
2345 kvm_mmu_free_some_pages(vcpu
);
2346 if (likely(!force_pt_level
))
2347 transparent_hugepage_adjust(vcpu
, &gfn
, &pfn
, &level
);
2348 r
= __direct_map(vcpu
, v
, write
, map_writable
, level
, gfn
, pfn
,
2350 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2356 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2357 kvm_release_pfn_clean(pfn
);
2362 static void mmu_free_roots(struct kvm_vcpu
*vcpu
)
2365 struct kvm_mmu_page
*sp
;
2366 LIST_HEAD(invalid_list
);
2368 if (!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
))
2370 spin_lock(&vcpu
->kvm
->mmu_lock
);
2371 if (vcpu
->arch
.mmu
.shadow_root_level
== PT64_ROOT_LEVEL
&&
2372 (vcpu
->arch
.mmu
.root_level
== PT64_ROOT_LEVEL
||
2373 vcpu
->arch
.mmu
.direct_map
)) {
2374 hpa_t root
= vcpu
->arch
.mmu
.root_hpa
;
2376 sp
= page_header(root
);
2378 if (!sp
->root_count
&& sp
->role
.invalid
) {
2379 kvm_mmu_prepare_zap_page(vcpu
->kvm
, sp
, &invalid_list
);
2380 kvm_mmu_commit_zap_page(vcpu
->kvm
, &invalid_list
);
2382 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
2383 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2386 for (i
= 0; i
< 4; ++i
) {
2387 hpa_t root
= vcpu
->arch
.mmu
.pae_root
[i
];
2390 root
&= PT64_BASE_ADDR_MASK
;
2391 sp
= page_header(root
);
2393 if (!sp
->root_count
&& sp
->role
.invalid
)
2394 kvm_mmu_prepare_zap_page(vcpu
->kvm
, sp
,
2397 vcpu
->arch
.mmu
.pae_root
[i
] = INVALID_PAGE
;
2399 kvm_mmu_commit_zap_page(vcpu
->kvm
, &invalid_list
);
2400 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2401 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
2404 static int mmu_check_root(struct kvm_vcpu
*vcpu
, gfn_t root_gfn
)
2408 if (!kvm_is_visible_gfn(vcpu
->kvm
, root_gfn
)) {
2409 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
2416 static int mmu_alloc_direct_roots(struct kvm_vcpu
*vcpu
)
2418 struct kvm_mmu_page
*sp
;
2421 if (vcpu
->arch
.mmu
.shadow_root_level
== PT64_ROOT_LEVEL
) {
2422 spin_lock(&vcpu
->kvm
->mmu_lock
);
2423 kvm_mmu_free_some_pages(vcpu
);
2424 sp
= kvm_mmu_get_page(vcpu
, 0, 0, PT64_ROOT_LEVEL
,
2427 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2428 vcpu
->arch
.mmu
.root_hpa
= __pa(sp
->spt
);
2429 } else if (vcpu
->arch
.mmu
.shadow_root_level
== PT32E_ROOT_LEVEL
) {
2430 for (i
= 0; i
< 4; ++i
) {
2431 hpa_t root
= vcpu
->arch
.mmu
.pae_root
[i
];
2433 ASSERT(!VALID_PAGE(root
));
2434 spin_lock(&vcpu
->kvm
->mmu_lock
);
2435 kvm_mmu_free_some_pages(vcpu
);
2436 sp
= kvm_mmu_get_page(vcpu
, i
<< (30 - PAGE_SHIFT
),
2438 PT32_ROOT_LEVEL
, 1, ACC_ALL
,
2440 root
= __pa(sp
->spt
);
2442 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2443 vcpu
->arch
.mmu
.pae_root
[i
] = root
| PT_PRESENT_MASK
;
2445 vcpu
->arch
.mmu
.root_hpa
= __pa(vcpu
->arch
.mmu
.pae_root
);
2452 static int mmu_alloc_shadow_roots(struct kvm_vcpu
*vcpu
)
2454 struct kvm_mmu_page
*sp
;
2459 root_gfn
= vcpu
->arch
.mmu
.get_cr3(vcpu
) >> PAGE_SHIFT
;
2461 if (mmu_check_root(vcpu
, root_gfn
))
2465 * Do we shadow a long mode page table? If so we need to
2466 * write-protect the guests page table root.
2468 if (vcpu
->arch
.mmu
.root_level
== PT64_ROOT_LEVEL
) {
2469 hpa_t root
= vcpu
->arch
.mmu
.root_hpa
;
2471 ASSERT(!VALID_PAGE(root
));
2473 spin_lock(&vcpu
->kvm
->mmu_lock
);
2474 kvm_mmu_free_some_pages(vcpu
);
2475 sp
= kvm_mmu_get_page(vcpu
, root_gfn
, 0, PT64_ROOT_LEVEL
,
2477 root
= __pa(sp
->spt
);
2479 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2480 vcpu
->arch
.mmu
.root_hpa
= root
;
2485 * We shadow a 32 bit page table. This may be a legacy 2-level
2486 * or a PAE 3-level page table. In either case we need to be aware that
2487 * the shadow page table may be a PAE or a long mode page table.
2489 pm_mask
= PT_PRESENT_MASK
;
2490 if (vcpu
->arch
.mmu
.shadow_root_level
== PT64_ROOT_LEVEL
)
2491 pm_mask
|= PT_ACCESSED_MASK
| PT_WRITABLE_MASK
| PT_USER_MASK
;
2493 for (i
= 0; i
< 4; ++i
) {
2494 hpa_t root
= vcpu
->arch
.mmu
.pae_root
[i
];
2496 ASSERT(!VALID_PAGE(root
));
2497 if (vcpu
->arch
.mmu
.root_level
== PT32E_ROOT_LEVEL
) {
2498 pdptr
= kvm_pdptr_read_mmu(vcpu
, &vcpu
->arch
.mmu
, i
);
2499 if (!is_present_gpte(pdptr
)) {
2500 vcpu
->arch
.mmu
.pae_root
[i
] = 0;
2503 root_gfn
= pdptr
>> PAGE_SHIFT
;
2504 if (mmu_check_root(vcpu
, root_gfn
))
2507 spin_lock(&vcpu
->kvm
->mmu_lock
);
2508 kvm_mmu_free_some_pages(vcpu
);
2509 sp
= kvm_mmu_get_page(vcpu
, root_gfn
, i
<< 30,
2512 root
= __pa(sp
->spt
);
2514 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2516 vcpu
->arch
.mmu
.pae_root
[i
] = root
| pm_mask
;
2518 vcpu
->arch
.mmu
.root_hpa
= __pa(vcpu
->arch
.mmu
.pae_root
);
2521 * If we shadow a 32 bit page table with a long mode page
2522 * table we enter this path.
2524 if (vcpu
->arch
.mmu
.shadow_root_level
== PT64_ROOT_LEVEL
) {
2525 if (vcpu
->arch
.mmu
.lm_root
== NULL
) {
2527 * The additional page necessary for this is only
2528 * allocated on demand.
2533 lm_root
= (void*)get_zeroed_page(GFP_KERNEL
);
2534 if (lm_root
== NULL
)
2537 lm_root
[0] = __pa(vcpu
->arch
.mmu
.pae_root
) | pm_mask
;
2539 vcpu
->arch
.mmu
.lm_root
= lm_root
;
2542 vcpu
->arch
.mmu
.root_hpa
= __pa(vcpu
->arch
.mmu
.lm_root
);
2548 static int mmu_alloc_roots(struct kvm_vcpu
*vcpu
)
2550 if (vcpu
->arch
.mmu
.direct_map
)
2551 return mmu_alloc_direct_roots(vcpu
);
2553 return mmu_alloc_shadow_roots(vcpu
);
2556 static void mmu_sync_roots(struct kvm_vcpu
*vcpu
)
2559 struct kvm_mmu_page
*sp
;
2561 if (vcpu
->arch
.mmu
.direct_map
)
2564 if (!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
))
2567 trace_kvm_mmu_audit(vcpu
, AUDIT_PRE_SYNC
);
2568 if (vcpu
->arch
.mmu
.root_level
== PT64_ROOT_LEVEL
) {
2569 hpa_t root
= vcpu
->arch
.mmu
.root_hpa
;
2570 sp
= page_header(root
);
2571 mmu_sync_children(vcpu
, sp
);
2572 trace_kvm_mmu_audit(vcpu
, AUDIT_POST_SYNC
);
2575 for (i
= 0; i
< 4; ++i
) {
2576 hpa_t root
= vcpu
->arch
.mmu
.pae_root
[i
];
2578 if (root
&& VALID_PAGE(root
)) {
2579 root
&= PT64_BASE_ADDR_MASK
;
2580 sp
= page_header(root
);
2581 mmu_sync_children(vcpu
, sp
);
2584 trace_kvm_mmu_audit(vcpu
, AUDIT_POST_SYNC
);
2587 void kvm_mmu_sync_roots(struct kvm_vcpu
*vcpu
)
2589 spin_lock(&vcpu
->kvm
->mmu_lock
);
2590 mmu_sync_roots(vcpu
);
2591 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2594 static gpa_t
nonpaging_gva_to_gpa(struct kvm_vcpu
*vcpu
, gva_t vaddr
,
2595 u32 access
, struct x86_exception
*exception
)
2598 exception
->error_code
= 0;
2602 static gpa_t
nonpaging_gva_to_gpa_nested(struct kvm_vcpu
*vcpu
, gva_t vaddr
,
2604 struct x86_exception
*exception
)
2607 exception
->error_code
= 0;
2608 return vcpu
->arch
.nested_mmu
.translate_gpa(vcpu
, vaddr
, access
);
2611 static int nonpaging_page_fault(struct kvm_vcpu
*vcpu
, gva_t gva
,
2612 u32 error_code
, bool prefault
)
2617 pgprintk("%s: gva %lx error %x\n", __func__
, gva
, error_code
);
2618 r
= mmu_topup_memory_caches(vcpu
);
2623 ASSERT(VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
2625 gfn
= gva
>> PAGE_SHIFT
;
2627 return nonpaging_map(vcpu
, gva
& PAGE_MASK
,
2628 error_code
& PFERR_WRITE_MASK
, gfn
, prefault
);
2631 static int kvm_arch_setup_async_pf(struct kvm_vcpu
*vcpu
, gva_t gva
, gfn_t gfn
)
2633 struct kvm_arch_async_pf arch
;
2635 arch
.token
= (vcpu
->arch
.apf
.id
++ << 12) | vcpu
->vcpu_id
;
2637 arch
.direct_map
= vcpu
->arch
.mmu
.direct_map
;
2638 arch
.cr3
= vcpu
->arch
.mmu
.get_cr3(vcpu
);
2640 return kvm_setup_async_pf(vcpu
, gva
, gfn
, &arch
);
2643 static bool can_do_async_pf(struct kvm_vcpu
*vcpu
)
2645 if (unlikely(!irqchip_in_kernel(vcpu
->kvm
) ||
2646 kvm_event_needs_reinjection(vcpu
)))
2649 return kvm_x86_ops
->interrupt_allowed(vcpu
);
2652 static bool try_async_pf(struct kvm_vcpu
*vcpu
, bool prefault
, gfn_t gfn
,
2653 gva_t gva
, pfn_t
*pfn
, bool write
, bool *writable
)
2657 *pfn
= gfn_to_pfn_async(vcpu
->kvm
, gfn
, &async
, write
, writable
);
2660 return false; /* *pfn has correct page already */
2662 put_page(pfn_to_page(*pfn
));
2664 if (!prefault
&& can_do_async_pf(vcpu
)) {
2665 trace_kvm_try_async_get_page(gva
, gfn
);
2666 if (kvm_find_async_pf_gfn(vcpu
, gfn
)) {
2667 trace_kvm_async_pf_doublefault(gva
, gfn
);
2668 kvm_make_request(KVM_REQ_APF_HALT
, vcpu
);
2670 } else if (kvm_arch_setup_async_pf(vcpu
, gva
, gfn
))
2674 *pfn
= gfn_to_pfn_prot(vcpu
->kvm
, gfn
, write
, writable
);
2679 static int tdp_page_fault(struct kvm_vcpu
*vcpu
, gva_t gpa
, u32 error_code
,
2686 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2687 unsigned long mmu_seq
;
2688 int write
= error_code
& PFERR_WRITE_MASK
;
2692 ASSERT(VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
2694 r
= mmu_topup_memory_caches(vcpu
);
2698 force_pt_level
= mapping_level_dirty_bitmap(vcpu
, gfn
);
2699 if (likely(!force_pt_level
)) {
2700 level
= mapping_level(vcpu
, gfn
);
2701 gfn
&= ~(KVM_PAGES_PER_HPAGE(level
) - 1);
2703 level
= PT_PAGE_TABLE_LEVEL
;
2705 mmu_seq
= vcpu
->kvm
->mmu_notifier_seq
;
2708 if (try_async_pf(vcpu
, prefault
, gfn
, gpa
, &pfn
, write
, &map_writable
))
2712 if (is_error_pfn(pfn
))
2713 return kvm_handle_bad_page(vcpu
->kvm
, gfn
, pfn
);
2714 spin_lock(&vcpu
->kvm
->mmu_lock
);
2715 if (mmu_notifier_retry(vcpu
, mmu_seq
))
2717 kvm_mmu_free_some_pages(vcpu
);
2718 if (likely(!force_pt_level
))
2719 transparent_hugepage_adjust(vcpu
, &gfn
, &pfn
, &level
);
2720 r
= __direct_map(vcpu
, gpa
, write
, map_writable
,
2721 level
, gfn
, pfn
, prefault
);
2722 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2727 spin_unlock(&vcpu
->kvm
->mmu_lock
);
2728 kvm_release_pfn_clean(pfn
);
2732 static void nonpaging_free(struct kvm_vcpu
*vcpu
)
2734 mmu_free_roots(vcpu
);
2737 static int nonpaging_init_context(struct kvm_vcpu
*vcpu
,
2738 struct kvm_mmu
*context
)
2740 context
->new_cr3
= nonpaging_new_cr3
;
2741 context
->page_fault
= nonpaging_page_fault
;
2742 context
->gva_to_gpa
= nonpaging_gva_to_gpa
;
2743 context
->free
= nonpaging_free
;
2744 context
->prefetch_page
= nonpaging_prefetch_page
;
2745 context
->sync_page
= nonpaging_sync_page
;
2746 context
->invlpg
= nonpaging_invlpg
;
2747 context
->update_pte
= nonpaging_update_pte
;
2748 context
->root_level
= 0;
2749 context
->shadow_root_level
= PT32E_ROOT_LEVEL
;
2750 context
->root_hpa
= INVALID_PAGE
;
2751 context
->direct_map
= true;
2752 context
->nx
= false;
2756 void kvm_mmu_flush_tlb(struct kvm_vcpu
*vcpu
)
2758 ++vcpu
->stat
.tlb_flush
;
2759 kvm_make_request(KVM_REQ_TLB_FLUSH
, vcpu
);
2762 static void paging_new_cr3(struct kvm_vcpu
*vcpu
)
2764 pgprintk("%s: cr3 %lx\n", __func__
, kvm_read_cr3(vcpu
));
2765 mmu_free_roots(vcpu
);
2768 static unsigned long get_cr3(struct kvm_vcpu
*vcpu
)
2770 return kvm_read_cr3(vcpu
);
2773 static void inject_page_fault(struct kvm_vcpu
*vcpu
,
2774 struct x86_exception
*fault
)
2776 vcpu
->arch
.mmu
.inject_page_fault(vcpu
, fault
);
2779 static void paging_free(struct kvm_vcpu
*vcpu
)
2781 nonpaging_free(vcpu
);
2784 static bool is_rsvd_bits_set(struct kvm_mmu
*mmu
, u64 gpte
, int level
)
2788 bit7
= (gpte
>> 7) & 1;
2789 return (gpte
& mmu
->rsvd_bits_mask
[bit7
][level
-1]) != 0;
2793 #include "paging_tmpl.h"
2797 #include "paging_tmpl.h"
2800 static void reset_rsvds_bits_mask(struct kvm_vcpu
*vcpu
,
2801 struct kvm_mmu
*context
,
2804 int maxphyaddr
= cpuid_maxphyaddr(vcpu
);
2805 u64 exb_bit_rsvd
= 0;
2808 exb_bit_rsvd
= rsvd_bits(63, 63);
2810 case PT32_ROOT_LEVEL
:
2811 /* no rsvd bits for 2 level 4K page table entries */
2812 context
->rsvd_bits_mask
[0][1] = 0;
2813 context
->rsvd_bits_mask
[0][0] = 0;
2814 context
->rsvd_bits_mask
[1][0] = context
->rsvd_bits_mask
[0][0];
2816 if (!is_pse(vcpu
)) {
2817 context
->rsvd_bits_mask
[1][1] = 0;
2821 if (is_cpuid_PSE36())
2822 /* 36bits PSE 4MB page */
2823 context
->rsvd_bits_mask
[1][1] = rsvd_bits(17, 21);
2825 /* 32 bits PSE 4MB page */
2826 context
->rsvd_bits_mask
[1][1] = rsvd_bits(13, 21);
2828 case PT32E_ROOT_LEVEL
:
2829 context
->rsvd_bits_mask
[0][2] =
2830 rsvd_bits(maxphyaddr
, 63) |
2831 rsvd_bits(7, 8) | rsvd_bits(1, 2); /* PDPTE */
2832 context
->rsvd_bits_mask
[0][1] = exb_bit_rsvd
|
2833 rsvd_bits(maxphyaddr
, 62); /* PDE */
2834 context
->rsvd_bits_mask
[0][0] = exb_bit_rsvd
|
2835 rsvd_bits(maxphyaddr
, 62); /* PTE */
2836 context
->rsvd_bits_mask
[1][1] = exb_bit_rsvd
|
2837 rsvd_bits(maxphyaddr
, 62) |
2838 rsvd_bits(13, 20); /* large page */
2839 context
->rsvd_bits_mask
[1][0] = context
->rsvd_bits_mask
[0][0];
2841 case PT64_ROOT_LEVEL
:
2842 context
->rsvd_bits_mask
[0][3] = exb_bit_rsvd
|
2843 rsvd_bits(maxphyaddr
, 51) | rsvd_bits(7, 8);
2844 context
->rsvd_bits_mask
[0][2] = exb_bit_rsvd
|
2845 rsvd_bits(maxphyaddr
, 51) | rsvd_bits(7, 8);
2846 context
->rsvd_bits_mask
[0][1] = exb_bit_rsvd
|
2847 rsvd_bits(maxphyaddr
, 51);
2848 context
->rsvd_bits_mask
[0][0] = exb_bit_rsvd
|
2849 rsvd_bits(maxphyaddr
, 51);
2850 context
->rsvd_bits_mask
[1][3] = context
->rsvd_bits_mask
[0][3];
2851 context
->rsvd_bits_mask
[1][2] = exb_bit_rsvd
|
2852 rsvd_bits(maxphyaddr
, 51) |
2854 context
->rsvd_bits_mask
[1][1] = exb_bit_rsvd
|
2855 rsvd_bits(maxphyaddr
, 51) |
2856 rsvd_bits(13, 20); /* large page */
2857 context
->rsvd_bits_mask
[1][0] = context
->rsvd_bits_mask
[0][0];
2862 static int paging64_init_context_common(struct kvm_vcpu
*vcpu
,
2863 struct kvm_mmu
*context
,
2866 context
->nx
= is_nx(vcpu
);
2868 reset_rsvds_bits_mask(vcpu
, context
, level
);
2870 ASSERT(is_pae(vcpu
));
2871 context
->new_cr3
= paging_new_cr3
;
2872 context
->page_fault
= paging64_page_fault
;
2873 context
->gva_to_gpa
= paging64_gva_to_gpa
;
2874 context
->prefetch_page
= paging64_prefetch_page
;
2875 context
->sync_page
= paging64_sync_page
;
2876 context
->invlpg
= paging64_invlpg
;
2877 context
->update_pte
= paging64_update_pte
;
2878 context
->free
= paging_free
;
2879 context
->root_level
= level
;
2880 context
->shadow_root_level
= level
;
2881 context
->root_hpa
= INVALID_PAGE
;
2882 context
->direct_map
= false;
2886 static int paging64_init_context(struct kvm_vcpu
*vcpu
,
2887 struct kvm_mmu
*context
)
2889 return paging64_init_context_common(vcpu
, context
, PT64_ROOT_LEVEL
);
2892 static int paging32_init_context(struct kvm_vcpu
*vcpu
,
2893 struct kvm_mmu
*context
)
2895 context
->nx
= false;
2897 reset_rsvds_bits_mask(vcpu
, context
, PT32_ROOT_LEVEL
);
2899 context
->new_cr3
= paging_new_cr3
;
2900 context
->page_fault
= paging32_page_fault
;
2901 context
->gva_to_gpa
= paging32_gva_to_gpa
;
2902 context
->free
= paging_free
;
2903 context
->prefetch_page
= paging32_prefetch_page
;
2904 context
->sync_page
= paging32_sync_page
;
2905 context
->invlpg
= paging32_invlpg
;
2906 context
->update_pte
= paging32_update_pte
;
2907 context
->root_level
= PT32_ROOT_LEVEL
;
2908 context
->shadow_root_level
= PT32E_ROOT_LEVEL
;
2909 context
->root_hpa
= INVALID_PAGE
;
2910 context
->direct_map
= false;
2914 static int paging32E_init_context(struct kvm_vcpu
*vcpu
,
2915 struct kvm_mmu
*context
)
2917 return paging64_init_context_common(vcpu
, context
, PT32E_ROOT_LEVEL
);
2920 static int init_kvm_tdp_mmu(struct kvm_vcpu
*vcpu
)
2922 struct kvm_mmu
*context
= vcpu
->arch
.walk_mmu
;
2924 context
->base_role
.word
= 0;
2925 context
->new_cr3
= nonpaging_new_cr3
;
2926 context
->page_fault
= tdp_page_fault
;
2927 context
->free
= nonpaging_free
;
2928 context
->prefetch_page
= nonpaging_prefetch_page
;
2929 context
->sync_page
= nonpaging_sync_page
;
2930 context
->invlpg
= nonpaging_invlpg
;
2931 context
->update_pte
= nonpaging_update_pte
;
2932 context
->shadow_root_level
= kvm_x86_ops
->get_tdp_level();
2933 context
->root_hpa
= INVALID_PAGE
;
2934 context
->direct_map
= true;
2935 context
->set_cr3
= kvm_x86_ops
->set_tdp_cr3
;
2936 context
->get_cr3
= get_cr3
;
2937 context
->inject_page_fault
= kvm_inject_page_fault
;
2938 context
->nx
= is_nx(vcpu
);
2940 if (!is_paging(vcpu
)) {
2941 context
->nx
= false;
2942 context
->gva_to_gpa
= nonpaging_gva_to_gpa
;
2943 context
->root_level
= 0;
2944 } else if (is_long_mode(vcpu
)) {
2945 context
->nx
= is_nx(vcpu
);
2946 reset_rsvds_bits_mask(vcpu
, context
, PT64_ROOT_LEVEL
);
2947 context
->gva_to_gpa
= paging64_gva_to_gpa
;
2948 context
->root_level
= PT64_ROOT_LEVEL
;
2949 } else if (is_pae(vcpu
)) {
2950 context
->nx
= is_nx(vcpu
);
2951 reset_rsvds_bits_mask(vcpu
, context
, PT32E_ROOT_LEVEL
);
2952 context
->gva_to_gpa
= paging64_gva_to_gpa
;
2953 context
->root_level
= PT32E_ROOT_LEVEL
;
2955 context
->nx
= false;
2956 reset_rsvds_bits_mask(vcpu
, context
, PT32_ROOT_LEVEL
);
2957 context
->gva_to_gpa
= paging32_gva_to_gpa
;
2958 context
->root_level
= PT32_ROOT_LEVEL
;
2964 int kvm_init_shadow_mmu(struct kvm_vcpu
*vcpu
, struct kvm_mmu
*context
)
2967 bool smep
= kvm_read_cr4_bits(vcpu
, X86_CR4_SMEP
);
2969 ASSERT(!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
2971 if (!is_paging(vcpu
))
2972 r
= nonpaging_init_context(vcpu
, context
);
2973 else if (is_long_mode(vcpu
))
2974 r
= paging64_init_context(vcpu
, context
);
2975 else if (is_pae(vcpu
))
2976 r
= paging32E_init_context(vcpu
, context
);
2978 r
= paging32_init_context(vcpu
, context
);
2980 vcpu
->arch
.mmu
.base_role
.cr4_pae
= !!is_pae(vcpu
);
2981 vcpu
->arch
.mmu
.base_role
.cr0_wp
= is_write_protection(vcpu
);
2982 vcpu
->arch
.mmu
.base_role
.smep_andnot_wp
2983 = smep
&& !is_write_protection(vcpu
);
2987 EXPORT_SYMBOL_GPL(kvm_init_shadow_mmu
);
2989 static int init_kvm_softmmu(struct kvm_vcpu
*vcpu
)
2991 int r
= kvm_init_shadow_mmu(vcpu
, vcpu
->arch
.walk_mmu
);
2993 vcpu
->arch
.walk_mmu
->set_cr3
= kvm_x86_ops
->set_cr3
;
2994 vcpu
->arch
.walk_mmu
->get_cr3
= get_cr3
;
2995 vcpu
->arch
.walk_mmu
->inject_page_fault
= kvm_inject_page_fault
;
3000 static int init_kvm_nested_mmu(struct kvm_vcpu
*vcpu
)
3002 struct kvm_mmu
*g_context
= &vcpu
->arch
.nested_mmu
;
3004 g_context
->get_cr3
= get_cr3
;
3005 g_context
->inject_page_fault
= kvm_inject_page_fault
;
3008 * Note that arch.mmu.gva_to_gpa translates l2_gva to l1_gpa. The
3009 * translation of l2_gpa to l1_gpa addresses is done using the
3010 * arch.nested_mmu.gva_to_gpa function. Basically the gva_to_gpa
3011 * functions between mmu and nested_mmu are swapped.
3013 if (!is_paging(vcpu
)) {
3014 g_context
->nx
= false;
3015 g_context
->root_level
= 0;
3016 g_context
->gva_to_gpa
= nonpaging_gva_to_gpa_nested
;
3017 } else if (is_long_mode(vcpu
)) {
3018 g_context
->nx
= is_nx(vcpu
);
3019 reset_rsvds_bits_mask(vcpu
, g_context
, PT64_ROOT_LEVEL
);
3020 g_context
->root_level
= PT64_ROOT_LEVEL
;
3021 g_context
->gva_to_gpa
= paging64_gva_to_gpa_nested
;
3022 } else if (is_pae(vcpu
)) {
3023 g_context
->nx
= is_nx(vcpu
);
3024 reset_rsvds_bits_mask(vcpu
, g_context
, PT32E_ROOT_LEVEL
);
3025 g_context
->root_level
= PT32E_ROOT_LEVEL
;
3026 g_context
->gva_to_gpa
= paging64_gva_to_gpa_nested
;
3028 g_context
->nx
= false;
3029 reset_rsvds_bits_mask(vcpu
, g_context
, PT32_ROOT_LEVEL
);
3030 g_context
->root_level
= PT32_ROOT_LEVEL
;
3031 g_context
->gva_to_gpa
= paging32_gva_to_gpa_nested
;
3037 static int init_kvm_mmu(struct kvm_vcpu
*vcpu
)
3039 if (mmu_is_nested(vcpu
))
3040 return init_kvm_nested_mmu(vcpu
);
3041 else if (tdp_enabled
)
3042 return init_kvm_tdp_mmu(vcpu
);
3044 return init_kvm_softmmu(vcpu
);
3047 static void destroy_kvm_mmu(struct kvm_vcpu
*vcpu
)
3050 if (VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
))
3051 /* mmu.free() should set root_hpa = INVALID_PAGE */
3052 vcpu
->arch
.mmu
.free(vcpu
);
3055 int kvm_mmu_reset_context(struct kvm_vcpu
*vcpu
)
3059 destroy_kvm_mmu(vcpu
);
3060 r
= init_kvm_mmu(vcpu
);
3065 kvm_mmu_sync_roots(vcpu
);
3066 kvm_mmu_flush_tlb(vcpu
);
3070 EXPORT_SYMBOL_GPL(kvm_mmu_reset_context
);
3072 int kvm_mmu_load(struct kvm_vcpu
*vcpu
)
3076 r
= mmu_topup_memory_caches(vcpu
);
3079 r
= mmu_alloc_roots(vcpu
);
3080 spin_lock(&vcpu
->kvm
->mmu_lock
);
3081 mmu_sync_roots(vcpu
);
3082 spin_unlock(&vcpu
->kvm
->mmu_lock
);
3085 /* set_cr3() should ensure TLB has been flushed */
3086 vcpu
->arch
.mmu
.set_cr3(vcpu
, vcpu
->arch
.mmu
.root_hpa
);
3090 EXPORT_SYMBOL_GPL(kvm_mmu_load
);
3092 void kvm_mmu_unload(struct kvm_vcpu
*vcpu
)
3094 mmu_free_roots(vcpu
);
3096 EXPORT_SYMBOL_GPL(kvm_mmu_unload
);
3098 static void mmu_pte_write_new_pte(struct kvm_vcpu
*vcpu
,
3099 struct kvm_mmu_page
*sp
, u64
*spte
,
3102 if (sp
->role
.level
!= PT_PAGE_TABLE_LEVEL
) {
3103 ++vcpu
->kvm
->stat
.mmu_pde_zapped
;
3107 ++vcpu
->kvm
->stat
.mmu_pte_updated
;
3108 vcpu
->arch
.mmu
.update_pte(vcpu
, sp
, spte
, new);
3111 static bool need_remote_flush(u64 old
, u64
new)
3113 if (!is_shadow_present_pte(old
))
3115 if (!is_shadow_present_pte(new))
3117 if ((old
^ new) & PT64_BASE_ADDR_MASK
)
3119 old
^= PT64_NX_MASK
;
3120 new ^= PT64_NX_MASK
;
3121 return (old
& ~new & PT64_PERM_MASK
) != 0;
3124 static void mmu_pte_write_flush_tlb(struct kvm_vcpu
*vcpu
, bool zap_page
,
3125 bool remote_flush
, bool local_flush
)
3131 kvm_flush_remote_tlbs(vcpu
->kvm
);
3132 else if (local_flush
)
3133 kvm_mmu_flush_tlb(vcpu
);
3136 static bool last_updated_pte_accessed(struct kvm_vcpu
*vcpu
)
3138 u64
*spte
= vcpu
->arch
.last_pte_updated
;
3140 return !!(spte
&& (*spte
& shadow_accessed_mask
));
3143 static void kvm_mmu_access_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
3145 u64
*spte
= vcpu
->arch
.last_pte_updated
;
3148 && vcpu
->arch
.last_pte_gfn
== gfn
3149 && shadow_accessed_mask
3150 && !(*spte
& shadow_accessed_mask
)
3151 && is_shadow_present_pte(*spte
))
3152 set_bit(PT_ACCESSED_SHIFT
, (unsigned long *)spte
);
3155 void kvm_mmu_pte_write(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
3156 const u8
*new, int bytes
,
3157 bool guest_initiated
)
3159 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
3160 union kvm_mmu_page_role mask
= { .word
= 0 };
3161 struct kvm_mmu_page
*sp
;
3162 struct hlist_node
*node
;
3163 LIST_HEAD(invalid_list
);
3164 u64 entry
, gentry
, *spte
;
3165 unsigned pte_size
, page_offset
, misaligned
, quadrant
, offset
;
3166 int level
, npte
, invlpg_counter
, r
, flooded
= 0;
3167 bool remote_flush
, local_flush
, zap_page
;
3170 * If we don't have indirect shadow pages, it means no page is
3171 * write-protected, so we can exit simply.
3173 if (!ACCESS_ONCE(vcpu
->kvm
->arch
.indirect_shadow_pages
))
3176 zap_page
= remote_flush
= local_flush
= false;
3177 offset
= offset_in_page(gpa
);
3179 pgprintk("%s: gpa %llx bytes %d\n", __func__
, gpa
, bytes
);
3181 invlpg_counter
= atomic_read(&vcpu
->kvm
->arch
.invlpg_counter
);
3184 * Assume that the pte write on a page table of the same type
3185 * as the current vcpu paging mode since we update the sptes only
3186 * when they have the same mode.
3188 if ((is_pae(vcpu
) && bytes
== 4) || !new) {
3189 /* Handle a 32-bit guest writing two halves of a 64-bit gpte */
3194 r
= kvm_read_guest(vcpu
->kvm
, gpa
, &gentry
, min(bytes
, 8));
3197 new = (const u8
*)&gentry
;
3202 gentry
= *(const u32
*)new;
3205 gentry
= *(const u64
*)new;
3212 spin_lock(&vcpu
->kvm
->mmu_lock
);
3213 if (atomic_read(&vcpu
->kvm
->arch
.invlpg_counter
) != invlpg_counter
)
3215 kvm_mmu_free_some_pages(vcpu
);
3216 ++vcpu
->kvm
->stat
.mmu_pte_write
;
3217 trace_kvm_mmu_audit(vcpu
, AUDIT_PRE_PTE_WRITE
);
3218 if (guest_initiated
) {
3219 kvm_mmu_access_page(vcpu
, gfn
);
3220 if (gfn
== vcpu
->arch
.last_pt_write_gfn
3221 && !last_updated_pte_accessed(vcpu
)) {
3222 ++vcpu
->arch
.last_pt_write_count
;
3223 if (vcpu
->arch
.last_pt_write_count
>= 3)
3226 vcpu
->arch
.last_pt_write_gfn
= gfn
;
3227 vcpu
->arch
.last_pt_write_count
= 1;
3228 vcpu
->arch
.last_pte_updated
= NULL
;
3232 mask
.cr0_wp
= mask
.cr4_pae
= mask
.nxe
= 1;
3233 for_each_gfn_indirect_valid_sp(vcpu
->kvm
, sp
, gfn
, node
) {
3234 pte_size
= sp
->role
.cr4_pae
? 8 : 4;
3235 misaligned
= (offset
^ (offset
+ bytes
- 1)) & ~(pte_size
- 1);
3236 misaligned
|= bytes
< 4;
3237 if (misaligned
|| flooded
) {
3239 * Misaligned accesses are too much trouble to fix
3240 * up; also, they usually indicate a page is not used
3243 * If we're seeing too many writes to a page,
3244 * it may no longer be a page table, or we may be
3245 * forking, in which case it is better to unmap the
3248 pgprintk("misaligned: gpa %llx bytes %d role %x\n",
3249 gpa
, bytes
, sp
->role
.word
);
3250 zap_page
|= !!kvm_mmu_prepare_zap_page(vcpu
->kvm
, sp
,
3252 ++vcpu
->kvm
->stat
.mmu_flooded
;
3255 page_offset
= offset
;
3256 level
= sp
->role
.level
;
3258 if (!sp
->role
.cr4_pae
) {
3259 page_offset
<<= 1; /* 32->64 */
3261 * A 32-bit pde maps 4MB while the shadow pdes map
3262 * only 2MB. So we need to double the offset again
3263 * and zap two pdes instead of one.
3265 if (level
== PT32_ROOT_LEVEL
) {
3266 page_offset
&= ~7; /* kill rounding error */
3270 quadrant
= page_offset
>> PAGE_SHIFT
;
3271 page_offset
&= ~PAGE_MASK
;
3272 if (quadrant
!= sp
->role
.quadrant
)
3276 spte
= &sp
->spt
[page_offset
/ sizeof(*spte
)];
3279 mmu_page_zap_pte(vcpu
->kvm
, sp
, spte
);
3281 !((sp
->role
.word
^ vcpu
->arch
.mmu
.base_role
.word
)
3283 mmu_pte_write_new_pte(vcpu
, sp
, spte
, &gentry
);
3284 if (!remote_flush
&& need_remote_flush(entry
, *spte
))
3285 remote_flush
= true;
3289 mmu_pte_write_flush_tlb(vcpu
, zap_page
, remote_flush
, local_flush
);
3290 kvm_mmu_commit_zap_page(vcpu
->kvm
, &invalid_list
);
3291 trace_kvm_mmu_audit(vcpu
, AUDIT_POST_PTE_WRITE
);
3292 spin_unlock(&vcpu
->kvm
->mmu_lock
);
3295 int kvm_mmu_unprotect_page_virt(struct kvm_vcpu
*vcpu
, gva_t gva
)
3300 if (vcpu
->arch
.mmu
.direct_map
)
3303 gpa
= kvm_mmu_gva_to_gpa_read(vcpu
, gva
, NULL
);
3305 spin_lock(&vcpu
->kvm
->mmu_lock
);
3306 r
= kvm_mmu_unprotect_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
3307 spin_unlock(&vcpu
->kvm
->mmu_lock
);
3310 EXPORT_SYMBOL_GPL(kvm_mmu_unprotect_page_virt
);
3312 void __kvm_mmu_free_some_pages(struct kvm_vcpu
*vcpu
)
3314 LIST_HEAD(invalid_list
);
3316 while (kvm_mmu_available_pages(vcpu
->kvm
) < KVM_REFILL_PAGES
&&
3317 !list_empty(&vcpu
->kvm
->arch
.active_mmu_pages
)) {
3318 struct kvm_mmu_page
*sp
;
3320 sp
= container_of(vcpu
->kvm
->arch
.active_mmu_pages
.prev
,
3321 struct kvm_mmu_page
, link
);
3322 kvm_mmu_prepare_zap_page(vcpu
->kvm
, sp
, &invalid_list
);
3323 kvm_mmu_commit_zap_page(vcpu
->kvm
, &invalid_list
);
3324 ++vcpu
->kvm
->stat
.mmu_recycled
;
3328 int kvm_mmu_page_fault(struct kvm_vcpu
*vcpu
, gva_t cr2
, u32 error_code
,
3329 void *insn
, int insn_len
)
3332 enum emulation_result er
;
3334 r
= vcpu
->arch
.mmu
.page_fault(vcpu
, cr2
, error_code
, false);
3343 r
= mmu_topup_memory_caches(vcpu
);
3347 er
= x86_emulate_instruction(vcpu
, cr2
, 0, insn
, insn_len
);
3352 case EMULATE_DO_MMIO
:
3353 ++vcpu
->stat
.mmio_exits
;
3363 EXPORT_SYMBOL_GPL(kvm_mmu_page_fault
);
3365 void kvm_mmu_invlpg(struct kvm_vcpu
*vcpu
, gva_t gva
)
3367 vcpu
->arch
.mmu
.invlpg(vcpu
, gva
);
3368 kvm_mmu_flush_tlb(vcpu
);
3369 ++vcpu
->stat
.invlpg
;
3371 EXPORT_SYMBOL_GPL(kvm_mmu_invlpg
);
3373 void kvm_enable_tdp(void)
3377 EXPORT_SYMBOL_GPL(kvm_enable_tdp
);
3379 void kvm_disable_tdp(void)
3381 tdp_enabled
= false;
3383 EXPORT_SYMBOL_GPL(kvm_disable_tdp
);
3385 static void free_mmu_pages(struct kvm_vcpu
*vcpu
)
3387 free_page((unsigned long)vcpu
->arch
.mmu
.pae_root
);
3388 if (vcpu
->arch
.mmu
.lm_root
!= NULL
)
3389 free_page((unsigned long)vcpu
->arch
.mmu
.lm_root
);
3392 static int alloc_mmu_pages(struct kvm_vcpu
*vcpu
)
3400 * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64.
3401 * Therefore we need to allocate shadow page tables in the first
3402 * 4GB of memory, which happens to fit the DMA32 zone.
3404 page
= alloc_page(GFP_KERNEL
| __GFP_DMA32
);
3408 vcpu
->arch
.mmu
.pae_root
= page_address(page
);
3409 for (i
= 0; i
< 4; ++i
)
3410 vcpu
->arch
.mmu
.pae_root
[i
] = INVALID_PAGE
;
3415 int kvm_mmu_create(struct kvm_vcpu
*vcpu
)
3418 ASSERT(!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
3420 return alloc_mmu_pages(vcpu
);
3423 int kvm_mmu_setup(struct kvm_vcpu
*vcpu
)
3426 ASSERT(!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
));
3428 return init_kvm_mmu(vcpu
);
3431 void kvm_mmu_slot_remove_write_access(struct kvm
*kvm
, int slot
)
3433 struct kvm_mmu_page
*sp
;
3435 list_for_each_entry(sp
, &kvm
->arch
.active_mmu_pages
, link
) {
3439 if (!test_bit(slot
, sp
->slot_bitmap
))
3443 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
) {
3444 if (!is_shadow_present_pte(pt
[i
]) ||
3445 !is_last_spte(pt
[i
], sp
->role
.level
))
3448 if (is_large_pte(pt
[i
])) {
3449 drop_spte(kvm
, &pt
[i
],
3450 shadow_trap_nonpresent_pte
);
3456 if (is_writable_pte(pt
[i
]))
3457 update_spte(&pt
[i
], pt
[i
] & ~PT_WRITABLE_MASK
);
3460 kvm_flush_remote_tlbs(kvm
);
3463 void kvm_mmu_zap_all(struct kvm
*kvm
)
3465 struct kvm_mmu_page
*sp
, *node
;
3466 LIST_HEAD(invalid_list
);
3468 spin_lock(&kvm
->mmu_lock
);
3470 list_for_each_entry_safe(sp
, node
, &kvm
->arch
.active_mmu_pages
, link
)
3471 if (kvm_mmu_prepare_zap_page(kvm
, sp
, &invalid_list
))
3474 kvm_mmu_commit_zap_page(kvm
, &invalid_list
);
3475 spin_unlock(&kvm
->mmu_lock
);
3478 static int kvm_mmu_remove_some_alloc_mmu_pages(struct kvm
*kvm
,
3479 struct list_head
*invalid_list
)
3481 struct kvm_mmu_page
*page
;
3483 page
= container_of(kvm
->arch
.active_mmu_pages
.prev
,
3484 struct kvm_mmu_page
, link
);
3485 return kvm_mmu_prepare_zap_page(kvm
, page
, invalid_list
);
3488 static int mmu_shrink(struct shrinker
*shrink
, struct shrink_control
*sc
)
3491 struct kvm
*kvm_freed
= NULL
;
3492 int nr_to_scan
= sc
->nr_to_scan
;
3494 if (nr_to_scan
== 0)
3497 raw_spin_lock(&kvm_lock
);
3499 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3500 int idx
, freed_pages
;
3501 LIST_HEAD(invalid_list
);
3503 idx
= srcu_read_lock(&kvm
->srcu
);
3504 spin_lock(&kvm
->mmu_lock
);
3505 if (!kvm_freed
&& nr_to_scan
> 0 &&
3506 kvm
->arch
.n_used_mmu_pages
> 0) {
3507 freed_pages
= kvm_mmu_remove_some_alloc_mmu_pages(kvm
,
3513 kvm_mmu_commit_zap_page(kvm
, &invalid_list
);
3514 spin_unlock(&kvm
->mmu_lock
);
3515 srcu_read_unlock(&kvm
->srcu
, idx
);
3518 list_move_tail(&kvm_freed
->vm_list
, &vm_list
);
3520 raw_spin_unlock(&kvm_lock
);
3523 return percpu_counter_read_positive(&kvm_total_used_mmu_pages
);
3526 static struct shrinker mmu_shrinker
= {
3527 .shrink
= mmu_shrink
,
3528 .seeks
= DEFAULT_SEEKS
* 10,
3531 static void mmu_destroy_caches(void)
3533 if (pte_list_desc_cache
)
3534 kmem_cache_destroy(pte_list_desc_cache
);
3535 if (mmu_page_header_cache
)
3536 kmem_cache_destroy(mmu_page_header_cache
);
3539 int kvm_mmu_module_init(void)
3541 pte_list_desc_cache
= kmem_cache_create("pte_list_desc",
3542 sizeof(struct pte_list_desc
),
3544 if (!pte_list_desc_cache
)
3547 mmu_page_header_cache
= kmem_cache_create("kvm_mmu_page_header",
3548 sizeof(struct kvm_mmu_page
),
3550 if (!mmu_page_header_cache
)
3553 if (percpu_counter_init(&kvm_total_used_mmu_pages
, 0))
3556 register_shrinker(&mmu_shrinker
);
3561 mmu_destroy_caches();
3566 * Caculate mmu pages needed for kvm.
3568 unsigned int kvm_mmu_calculate_mmu_pages(struct kvm
*kvm
)
3571 unsigned int nr_mmu_pages
;
3572 unsigned int nr_pages
= 0;
3573 struct kvm_memslots
*slots
;
3575 slots
= kvm_memslots(kvm
);
3577 for (i
= 0; i
< slots
->nmemslots
; i
++)
3578 nr_pages
+= slots
->memslots
[i
].npages
;
3580 nr_mmu_pages
= nr_pages
* KVM_PERMILLE_MMU_PAGES
/ 1000;
3581 nr_mmu_pages
= max(nr_mmu_pages
,
3582 (unsigned int) KVM_MIN_ALLOC_MMU_PAGES
);
3584 return nr_mmu_pages
;
3587 static void *pv_mmu_peek_buffer(struct kvm_pv_mmu_op_buffer
*buffer
,
3590 if (len
> buffer
->len
)
3595 static void *pv_mmu_read_buffer(struct kvm_pv_mmu_op_buffer
*buffer
,
3600 ret
= pv_mmu_peek_buffer(buffer
, len
);
3605 buffer
->processed
+= len
;
3609 static int kvm_pv_mmu_write(struct kvm_vcpu
*vcpu
,
3610 gpa_t addr
, gpa_t value
)
3615 if (!is_long_mode(vcpu
) && !is_pae(vcpu
))
3618 r
= mmu_topup_memory_caches(vcpu
);
3622 if (!emulator_write_phys(vcpu
, addr
, &value
, bytes
))
3628 static int kvm_pv_mmu_flush_tlb(struct kvm_vcpu
*vcpu
)
3630 (void)kvm_set_cr3(vcpu
, kvm_read_cr3(vcpu
));
3634 static int kvm_pv_mmu_release_pt(struct kvm_vcpu
*vcpu
, gpa_t addr
)
3636 spin_lock(&vcpu
->kvm
->mmu_lock
);
3637 mmu_unshadow(vcpu
->kvm
, addr
>> PAGE_SHIFT
);
3638 spin_unlock(&vcpu
->kvm
->mmu_lock
);
3642 static int kvm_pv_mmu_op_one(struct kvm_vcpu
*vcpu
,
3643 struct kvm_pv_mmu_op_buffer
*buffer
)
3645 struct kvm_mmu_op_header
*header
;
3647 header
= pv_mmu_peek_buffer(buffer
, sizeof *header
);
3650 switch (header
->op
) {
3651 case KVM_MMU_OP_WRITE_PTE
: {
3652 struct kvm_mmu_op_write_pte
*wpte
;
3654 wpte
= pv_mmu_read_buffer(buffer
, sizeof *wpte
);
3657 return kvm_pv_mmu_write(vcpu
, wpte
->pte_phys
,
3660 case KVM_MMU_OP_FLUSH_TLB
: {
3661 struct kvm_mmu_op_flush_tlb
*ftlb
;
3663 ftlb
= pv_mmu_read_buffer(buffer
, sizeof *ftlb
);
3666 return kvm_pv_mmu_flush_tlb(vcpu
);
3668 case KVM_MMU_OP_RELEASE_PT
: {
3669 struct kvm_mmu_op_release_pt
*rpt
;
3671 rpt
= pv_mmu_read_buffer(buffer
, sizeof *rpt
);
3674 return kvm_pv_mmu_release_pt(vcpu
, rpt
->pt_phys
);
3680 int kvm_pv_mmu_op(struct kvm_vcpu
*vcpu
, unsigned long bytes
,
3681 gpa_t addr
, unsigned long *ret
)
3684 struct kvm_pv_mmu_op_buffer
*buffer
= &vcpu
->arch
.mmu_op_buffer
;
3686 buffer
->ptr
= buffer
->buf
;
3687 buffer
->len
= min_t(unsigned long, bytes
, sizeof buffer
->buf
);
3688 buffer
->processed
= 0;
3690 r
= kvm_read_guest(vcpu
->kvm
, addr
, buffer
->buf
, buffer
->len
);
3694 while (buffer
->len
) {
3695 r
= kvm_pv_mmu_op_one(vcpu
, buffer
);
3704 *ret
= buffer
->processed
;
3708 int kvm_mmu_get_spte_hierarchy(struct kvm_vcpu
*vcpu
, u64 addr
, u64 sptes
[4])
3710 struct kvm_shadow_walk_iterator iterator
;
3713 spin_lock(&vcpu
->kvm
->mmu_lock
);
3714 for_each_shadow_entry(vcpu
, addr
, iterator
) {
3715 sptes
[iterator
.level
-1] = *iterator
.sptep
;
3717 if (!is_shadow_present_pte(*iterator
.sptep
))
3720 spin_unlock(&vcpu
->kvm
->mmu_lock
);
3724 EXPORT_SYMBOL_GPL(kvm_mmu_get_spte_hierarchy
);
3726 void kvm_mmu_destroy(struct kvm_vcpu
*vcpu
)
3730 destroy_kvm_mmu(vcpu
);
3731 free_mmu_pages(vcpu
);
3732 mmu_free_memory_caches(vcpu
);
3735 #ifdef CONFIG_KVM_MMU_AUDIT
3736 #include "mmu_audit.c"
3738 static void mmu_audit_disable(void) { }
3741 void kvm_mmu_module_exit(void)
3743 mmu_destroy_caches();
3744 percpu_counter_destroy(&kvm_total_used_mmu_pages
);
3745 unregister_shrinker(&mmu_shrinker
);
3746 mmu_audit_disable();