2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
9 * Copyright (C) 2006 Qumranet, Inc.
12 * Yaniv Kamay <yaniv@qumranet.com>
13 * Avi Kivity <avi@qumranet.com>
15 * This work is licensed under the terms of the GNU GPL, version 2. See
16 * the COPYING file in the top-level directory.
21 * We need the mmu code to access both 32-bit and 64-bit guest ptes,
22 * so the code in this file is compiled twice, once per pte size.
26 #define pt_element_t u64
27 #define guest_walker guest_walker64
28 #define FNAME(name) paging##64_##name
29 #define PT_BASE_ADDR_MASK PT64_BASE_ADDR_MASK
30 #define PT_DIR_BASE_ADDR_MASK PT64_DIR_BASE_ADDR_MASK
31 #define PT_INDEX(addr, level) PT64_INDEX(addr, level)
32 #define SHADOW_PT_INDEX(addr, level) PT64_INDEX(addr, level)
33 #define PT_LEVEL_MASK(level) PT64_LEVEL_MASK(level)
34 #define PT_LEVEL_BITS PT64_LEVEL_BITS
36 #define PT_MAX_FULL_LEVELS 4
38 #define PT_MAX_FULL_LEVELS 2
41 #define pt_element_t u32
42 #define guest_walker guest_walker32
43 #define FNAME(name) paging##32_##name
44 #define PT_BASE_ADDR_MASK PT32_BASE_ADDR_MASK
45 #define PT_DIR_BASE_ADDR_MASK PT32_DIR_BASE_ADDR_MASK
46 #define PT_INDEX(addr, level) PT32_INDEX(addr, level)
47 #define SHADOW_PT_INDEX(addr, level) PT64_INDEX(addr, level)
48 #define PT_LEVEL_MASK(level) PT32_LEVEL_MASK(level)
49 #define PT_LEVEL_BITS PT32_LEVEL_BITS
50 #define PT_MAX_FULL_LEVELS 2
52 #error Invalid PTTYPE value
56 * The guest_walker structure emulates the behavior of the hardware page
61 gfn_t table_gfn
[PT_MAX_FULL_LEVELS
];
63 pt_element_t inherited_ar
;
69 * Fetch a guest pte for a guest virtual address
71 static int FNAME(walk_addr
)(struct guest_walker
*walker
,
72 struct kvm_vcpu
*vcpu
, gva_t addr
,
73 int write_fault
, int user_fault
, int fetch_fault
)
76 struct kvm_memory_slot
*slot
;
84 pgprintk("%s: addr %lx\n", __FUNCTION__
, addr
);
85 walker
->level
= vcpu
->mmu
.root_level
;
88 if (!is_long_mode(vcpu
)) {
89 pte
= vcpu
->pdptrs
[(addr
>> 30) & 3];
90 if (!is_present_pte(pte
))
95 ASSERT((!is_long_mode(vcpu
) && is_pae(vcpu
)) ||
96 (vcpu
->cr3
& CR3_NONPAE_RESERVED_BITS
) == 0);
98 walker
->inherited_ar
= PT_USER_MASK
| PT_WRITABLE_MASK
;
101 index
= PT_INDEX(addr
, walker
->level
);
103 table_gfn
= (pte
& PT64_BASE_ADDR_MASK
) >> PAGE_SHIFT
;
104 walker
->table_gfn
[walker
->level
- 1] = table_gfn
;
105 pgprintk("%s: table_gfn[%d] %lx\n", __FUNCTION__
,
106 walker
->level
- 1, table_gfn
);
108 slot
= gfn_to_memslot(vcpu
->kvm
, table_gfn
);
109 hpa
= safe_gpa_to_hpa(vcpu
->kvm
, pte
& PT64_BASE_ADDR_MASK
);
110 page
= pfn_to_page(hpa
>> PAGE_SHIFT
);
112 table
= kmap_atomic(page
, KM_USER0
);
114 kunmap_atomic(table
, KM_USER0
);
116 if (!is_present_pte(pte
))
119 if (write_fault
&& !is_writeble_pte(pte
))
120 if (user_fault
|| is_write_protection(vcpu
))
123 if (user_fault
&& !(pte
& PT_USER_MASK
))
127 if (fetch_fault
&& is_nx(vcpu
) && (pte
& PT64_NX_MASK
))
131 if (!(pte
& PT_ACCESSED_MASK
)) {
132 mark_page_dirty(vcpu
->kvm
, table_gfn
);
133 pte
|= PT_ACCESSED_MASK
;
134 table
= kmap_atomic(page
, KM_USER0
);
136 kunmap_atomic(table
, KM_USER0
);
139 if (walker
->level
== PT_PAGE_TABLE_LEVEL
) {
140 walker
->gfn
= (pte
& PT_BASE_ADDR_MASK
) >> PAGE_SHIFT
;
144 if (walker
->level
== PT_DIRECTORY_LEVEL
145 && (pte
& PT_PAGE_SIZE_MASK
)
146 && (PTTYPE
== 64 || is_pse(vcpu
))) {
147 walker
->gfn
= (pte
& PT_DIR_BASE_ADDR_MASK
)
149 walker
->gfn
+= PT_INDEX(addr
, PT_PAGE_TABLE_LEVEL
);
153 walker
->inherited_ar
&= pte
;
157 if (write_fault
&& !is_dirty_pte(pte
)) {
158 mark_page_dirty(vcpu
->kvm
, table_gfn
);
159 pte
|= PT_DIRTY_MASK
;
160 table
= kmap_atomic(page
, KM_USER0
);
162 kunmap_atomic(table
, KM_USER0
);
163 pte_gpa
= table_gfn
<< PAGE_SHIFT
;
164 pte_gpa
+= index
* sizeof(pt_element_t
);
165 kvm_mmu_pte_write(vcpu
, pte_gpa
, (u8
*)&pte
, sizeof(pte
));
169 pgprintk("%s: pte %llx\n", __FUNCTION__
, (u64
)pte
);
173 walker
->error_code
= 0;
177 walker
->error_code
= PFERR_PRESENT_MASK
;
181 walker
->error_code
|= PFERR_WRITE_MASK
;
183 walker
->error_code
|= PFERR_USER_MASK
;
185 walker
->error_code
|= PFERR_FETCH_MASK
;
189 static void FNAME(set_pte_common
)(struct kvm_vcpu
*vcpu
,
197 struct guest_walker
*walker
,
201 int dirty
= gpte
& PT_DIRTY_MASK
;
203 int was_rmapped
= is_rmap_pte(*shadow_pte
);
205 pgprintk("%s: spte %llx gpte %llx access %llx write_fault %d"
206 " user_fault %d gfn %lx\n",
207 __FUNCTION__
, *shadow_pte
, (u64
)gpte
, access_bits
,
208 write_fault
, user_fault
, gfn
);
211 * We don't set the accessed bit, since we sometimes want to see
212 * whether the guest actually used the pte (in order to detect
215 spte
= PT_PRESENT_MASK
| PT_DIRTY_MASK
;
216 spte
|= gpte
& PT64_NX_MASK
;
218 access_bits
&= ~PT_WRITABLE_MASK
;
220 paddr
= gpa_to_hpa(vcpu
->kvm
, gaddr
& PT64_BASE_ADDR_MASK
);
222 spte
|= PT_PRESENT_MASK
;
223 if (access_bits
& PT_USER_MASK
)
224 spte
|= PT_USER_MASK
;
226 if (is_error_hpa(paddr
)) {
227 set_shadow_pte(shadow_pte
,
228 shadow_trap_nonpresent_pte
| PT_SHADOW_IO_MARK
);
234 if ((access_bits
& PT_WRITABLE_MASK
)
235 || (write_fault
&& !is_write_protection(vcpu
) && !user_fault
)) {
236 struct kvm_mmu_page
*shadow
;
238 spte
|= PT_WRITABLE_MASK
;
240 mmu_unshadow(vcpu
->kvm
, gfn
);
244 shadow
= kvm_mmu_lookup_page(vcpu
->kvm
, gfn
);
246 pgprintk("%s: found shadow page for %lx, marking ro\n",
248 access_bits
&= ~PT_WRITABLE_MASK
;
249 if (is_writeble_pte(spte
)) {
250 spte
&= ~PT_WRITABLE_MASK
;
251 kvm_x86_ops
->tlb_flush(vcpu
);
260 if (access_bits
& PT_WRITABLE_MASK
)
261 mark_page_dirty(vcpu
->kvm
, gaddr
>> PAGE_SHIFT
);
263 pgprintk("%s: setting spte %llx\n", __FUNCTION__
, spte
);
264 set_shadow_pte(shadow_pte
, spte
);
265 page_header_update_slot(vcpu
->kvm
, shadow_pte
, gaddr
);
267 rmap_add(vcpu
, shadow_pte
, (gaddr
& PT64_BASE_ADDR_MASK
)
269 if (!ptwrite
|| !*ptwrite
)
270 vcpu
->last_pte_updated
= shadow_pte
;
273 static void FNAME(set_pte
)(struct kvm_vcpu
*vcpu
, pt_element_t gpte
,
274 u64
*shadow_pte
, u64 access_bits
,
275 int user_fault
, int write_fault
, int *ptwrite
,
276 struct guest_walker
*walker
, gfn_t gfn
)
279 FNAME(set_pte_common
)(vcpu
, shadow_pte
, gpte
& PT_BASE_ADDR_MASK
,
280 gpte
, access_bits
, user_fault
, write_fault
,
281 ptwrite
, walker
, gfn
);
284 static void FNAME(update_pte
)(struct kvm_vcpu
*vcpu
, struct kvm_mmu_page
*page
,
285 u64
*spte
, const void *pte
, int bytes
,
290 gpte
= *(const pt_element_t
*)pte
;
291 if (~gpte
& (PT_PRESENT_MASK
| PT_ACCESSED_MASK
)) {
292 if (!offset_in_pte
&& !is_present_pte(gpte
))
293 set_shadow_pte(spte
, shadow_notrap_nonpresent_pte
);
296 if (bytes
< sizeof(pt_element_t
))
298 pgprintk("%s: gpte %llx spte %p\n", __FUNCTION__
, (u64
)gpte
, spte
);
299 FNAME(set_pte
)(vcpu
, gpte
, spte
, PT_USER_MASK
| PT_WRITABLE_MASK
, 0,
301 (gpte
& PT_BASE_ADDR_MASK
) >> PAGE_SHIFT
);
304 static void FNAME(set_pde
)(struct kvm_vcpu
*vcpu
, pt_element_t gpde
,
305 u64
*shadow_pte
, u64 access_bits
,
306 int user_fault
, int write_fault
, int *ptwrite
,
307 struct guest_walker
*walker
, gfn_t gfn
)
312 gaddr
= (gpa_t
)gfn
<< PAGE_SHIFT
;
313 if (PTTYPE
== 32 && is_cpuid_PSE36())
314 gaddr
|= (gpde
& PT32_DIR_PSE36_MASK
) <<
315 (32 - PT32_DIR_PSE36_SHIFT
);
316 FNAME(set_pte_common
)(vcpu
, shadow_pte
, gaddr
,
317 gpde
, access_bits
, user_fault
, write_fault
,
318 ptwrite
, walker
, gfn
);
322 * Fetch a shadow pte for a specific level in the paging hierarchy.
324 static u64
*FNAME(fetch
)(struct kvm_vcpu
*vcpu
, gva_t addr
,
325 struct guest_walker
*walker
,
326 int user_fault
, int write_fault
, int *ptwrite
)
331 u64
*prev_shadow_ent
= NULL
;
333 if (!is_present_pte(walker
->pte
))
336 shadow_addr
= vcpu
->mmu
.root_hpa
;
337 level
= vcpu
->mmu
.shadow_root_level
;
338 if (level
== PT32E_ROOT_LEVEL
) {
339 shadow_addr
= vcpu
->mmu
.pae_root
[(addr
>> 30) & 3];
340 shadow_addr
&= PT64_BASE_ADDR_MASK
;
345 u32 index
= SHADOW_PT_INDEX(addr
, level
);
346 struct kvm_mmu_page
*shadow_page
;
350 unsigned hugepage_access
= 0;
352 shadow_ent
= ((u64
*)__va(shadow_addr
)) + index
;
353 if (is_shadow_present_pte(*shadow_ent
)) {
354 if (level
== PT_PAGE_TABLE_LEVEL
)
356 shadow_addr
= *shadow_ent
& PT64_BASE_ADDR_MASK
;
357 prev_shadow_ent
= shadow_ent
;
361 if (level
== PT_PAGE_TABLE_LEVEL
)
364 if (level
- 1 == PT_PAGE_TABLE_LEVEL
365 && walker
->level
== PT_DIRECTORY_LEVEL
) {
367 hugepage_access
= walker
->pte
;
368 hugepage_access
&= PT_USER_MASK
| PT_WRITABLE_MASK
;
369 if (!is_dirty_pte(walker
->pte
))
370 hugepage_access
&= ~PT_WRITABLE_MASK
;
371 hugepage_access
>>= PT_WRITABLE_SHIFT
;
372 if (walker
->pte
& PT64_NX_MASK
)
373 hugepage_access
|= (1 << 2);
374 table_gfn
= (walker
->pte
& PT_BASE_ADDR_MASK
)
378 table_gfn
= walker
->table_gfn
[level
- 2];
380 shadow_page
= kvm_mmu_get_page(vcpu
, table_gfn
, addr
, level
-1,
381 metaphysical
, hugepage_access
,
383 shadow_addr
= __pa(shadow_page
->spt
);
384 shadow_pte
= shadow_addr
| PT_PRESENT_MASK
| PT_ACCESSED_MASK
385 | PT_WRITABLE_MASK
| PT_USER_MASK
;
386 *shadow_ent
= shadow_pte
;
387 prev_shadow_ent
= shadow_ent
;
390 if (walker
->level
== PT_DIRECTORY_LEVEL
) {
391 FNAME(set_pde
)(vcpu
, walker
->pte
, shadow_ent
,
392 walker
->inherited_ar
, user_fault
, write_fault
,
393 ptwrite
, walker
, walker
->gfn
);
395 ASSERT(walker
->level
== PT_PAGE_TABLE_LEVEL
);
396 FNAME(set_pte
)(vcpu
, walker
->pte
, shadow_ent
,
397 walker
->inherited_ar
, user_fault
, write_fault
,
398 ptwrite
, walker
, walker
->gfn
);
404 * Page fault handler. There are several causes for a page fault:
405 * - there is no shadow pte for the guest pte
406 * - write access through a shadow pte marked read only so that we can set
408 * - write access to a shadow pte marked read only so we can update the page
409 * dirty bitmap, when userspace requests it
410 * - mmio access; in this case we will never install a present shadow pte
411 * - normal guest page fault due to the guest pte marked not present, not
412 * writable, or not executable
414 * Returns: 1 if we need to emulate the instruction, 0 otherwise, or
415 * a negative value on error.
417 static int FNAME(page_fault
)(struct kvm_vcpu
*vcpu
, gva_t addr
,
420 int write_fault
= error_code
& PFERR_WRITE_MASK
;
421 int user_fault
= error_code
& PFERR_USER_MASK
;
422 int fetch_fault
= error_code
& PFERR_FETCH_MASK
;
423 struct guest_walker walker
;
428 pgprintk("%s: addr %lx err %x\n", __FUNCTION__
, addr
, error_code
);
429 kvm_mmu_audit(vcpu
, "pre page fault");
431 r
= mmu_topup_memory_caches(vcpu
);
436 * Look up the shadow pte for the faulting address.
438 r
= FNAME(walk_addr
)(&walker
, vcpu
, addr
, write_fault
, user_fault
,
442 * The page is not mapped by the guest. Let the guest handle it.
445 pgprintk("%s: guest page fault\n", __FUNCTION__
);
446 inject_page_fault(vcpu
, addr
, walker
.error_code
);
447 vcpu
->last_pt_write_count
= 0; /* reset fork detector */
451 shadow_pte
= FNAME(fetch
)(vcpu
, addr
, &walker
, user_fault
, write_fault
,
453 pgprintk("%s: shadow pte %p %llx ptwrite %d\n", __FUNCTION__
,
454 shadow_pte
, *shadow_pte
, write_pt
);
457 vcpu
->last_pt_write_count
= 0; /* reset fork detector */
460 * mmio: emulate if accessible, otherwise its a guest fault.
462 if (is_io_pte(*shadow_pte
))
465 ++vcpu
->stat
.pf_fixed
;
466 kvm_mmu_audit(vcpu
, "post page fault (fixed)");
471 static gpa_t
FNAME(gva_to_gpa
)(struct kvm_vcpu
*vcpu
, gva_t vaddr
)
473 struct guest_walker walker
;
474 gpa_t gpa
= UNMAPPED_GVA
;
477 r
= FNAME(walk_addr
)(&walker
, vcpu
, vaddr
, 0, 0, 0);
480 gpa
= (gpa_t
)walker
.gfn
<< PAGE_SHIFT
;
481 gpa
|= vaddr
& ~PAGE_MASK
;
487 static void FNAME(prefetch_page
)(struct kvm_vcpu
*vcpu
,
488 struct kvm_mmu_page
*sp
)
493 if (sp
->role
.metaphysical
|| PTTYPE
== 32) {
494 nonpaging_prefetch_page(vcpu
, sp
);
498 gpt
= kmap_atomic(gfn_to_page(vcpu
->kvm
, sp
->gfn
), KM_USER0
);
499 for (i
= 0; i
< PT64_ENT_PER_PAGE
; ++i
)
500 if (is_present_pte(gpt
[i
]))
501 sp
->spt
[i
] = shadow_trap_nonpresent_pte
;
503 sp
->spt
[i
] = shadow_notrap_nonpresent_pte
;
504 kunmap_atomic(gpt
, KM_USER0
);
510 #undef PT_BASE_ADDR_MASK
512 #undef SHADOW_PT_INDEX
514 #undef PT_DIR_BASE_ADDR_MASK
516 #undef PT_MAX_FULL_LEVELS