]>
Commit | Line | Data |
---|---|---|
6aa8b732 AK |
1 | /* |
2 | * Kernel-based Virtual Machine driver for Linux | |
3 | * | |
4 | * This module enables machines with Intel VT-x extensions to run virtual | |
5 | * machines without emulation or binary translation. | |
6 | * | |
7 | * MMU support | |
8 | * | |
9 | * Copyright (C) 2006 Qumranet, Inc. | |
221d059d | 10 | * Copyright 2010 Red Hat, Inc. and/or its affilates. |
6aa8b732 AK |
11 | * |
12 | * Authors: | |
13 | * Yaniv Kamay <yaniv@qumranet.com> | |
14 | * Avi Kivity <avi@qumranet.com> | |
15 | * | |
16 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
17 | * the COPYING file in the top-level directory. | |
18 | * | |
19 | */ | |
20 | ||
21 | /* | |
22 | * We need the mmu code to access both 32-bit and 64-bit guest ptes, | |
23 | * so the code in this file is compiled twice, once per pte size. | |
24 | */ | |
25 | ||
26 | #if PTTYPE == 64 | |
27 | #define pt_element_t u64 | |
28 | #define guest_walker guest_walker64 | |
29 | #define FNAME(name) paging##64_##name | |
30 | #define PT_BASE_ADDR_MASK PT64_BASE_ADDR_MASK | |
e04da980 JR |
31 | #define PT_LVL_ADDR_MASK(lvl) PT64_LVL_ADDR_MASK(lvl) |
32 | #define PT_LVL_OFFSET_MASK(lvl) PT64_LVL_OFFSET_MASK(lvl) | |
6aa8b732 | 33 | #define PT_INDEX(addr, level) PT64_INDEX(addr, level) |
6aa8b732 | 34 | #define PT_LEVEL_MASK(level) PT64_LEVEL_MASK(level) |
c7addb90 | 35 | #define PT_LEVEL_BITS PT64_LEVEL_BITS |
cea0f0e7 AK |
36 | #ifdef CONFIG_X86_64 |
37 | #define PT_MAX_FULL_LEVELS 4 | |
b3e4e63f | 38 | #define CMPXCHG cmpxchg |
cea0f0e7 | 39 | #else |
b3e4e63f | 40 | #define CMPXCHG cmpxchg64 |
cea0f0e7 AK |
41 | #define PT_MAX_FULL_LEVELS 2 |
42 | #endif | |
6aa8b732 AK |
43 | #elif PTTYPE == 32 |
44 | #define pt_element_t u32 | |
45 | #define guest_walker guest_walker32 | |
46 | #define FNAME(name) paging##32_##name | |
47 | #define PT_BASE_ADDR_MASK PT32_BASE_ADDR_MASK | |
e04da980 JR |
48 | #define PT_LVL_ADDR_MASK(lvl) PT32_LVL_ADDR_MASK(lvl) |
49 | #define PT_LVL_OFFSET_MASK(lvl) PT32_LVL_OFFSET_MASK(lvl) | |
6aa8b732 | 50 | #define PT_INDEX(addr, level) PT32_INDEX(addr, level) |
6aa8b732 | 51 | #define PT_LEVEL_MASK(level) PT32_LEVEL_MASK(level) |
c7addb90 | 52 | #define PT_LEVEL_BITS PT32_LEVEL_BITS |
cea0f0e7 | 53 | #define PT_MAX_FULL_LEVELS 2 |
b3e4e63f | 54 | #define CMPXCHG cmpxchg |
6aa8b732 AK |
55 | #else |
56 | #error Invalid PTTYPE value | |
57 | #endif | |
58 | ||
e04da980 JR |
59 | #define gpte_to_gfn_lvl FNAME(gpte_to_gfn_lvl) |
60 | #define gpte_to_gfn(pte) gpte_to_gfn_lvl((pte), PT_PAGE_TABLE_LEVEL) | |
5fb07ddb | 61 | |
6aa8b732 AK |
62 | /* |
63 | * The guest_walker structure emulates the behavior of the hardware page | |
64 | * table walker. | |
65 | */ | |
66 | struct guest_walker { | |
67 | int level; | |
cea0f0e7 | 68 | gfn_t table_gfn[PT_MAX_FULL_LEVELS]; |
7819026e MT |
69 | pt_element_t ptes[PT_MAX_FULL_LEVELS]; |
70 | gpa_t pte_gpa[PT_MAX_FULL_LEVELS]; | |
fe135d2c AK |
71 | unsigned pt_access; |
72 | unsigned pte_access; | |
815af8d4 | 73 | gfn_t gfn; |
7993ba43 | 74 | u32 error_code; |
6aa8b732 AK |
75 | }; |
76 | ||
e04da980 | 77 | static gfn_t gpte_to_gfn_lvl(pt_element_t gpte, int lvl) |
5fb07ddb | 78 | { |
e04da980 | 79 | return (gpte & PT_LVL_ADDR_MASK(lvl)) >> PAGE_SHIFT; |
5fb07ddb AK |
80 | } |
81 | ||
b3e4e63f MT |
82 | static bool FNAME(cmpxchg_gpte)(struct kvm *kvm, |
83 | gfn_t table_gfn, unsigned index, | |
84 | pt_element_t orig_pte, pt_element_t new_pte) | |
85 | { | |
86 | pt_element_t ret; | |
87 | pt_element_t *table; | |
88 | struct page *page; | |
89 | ||
90 | page = gfn_to_page(kvm, table_gfn); | |
72dc67a6 | 91 | |
b3e4e63f | 92 | table = kmap_atomic(page, KM_USER0); |
b3e4e63f | 93 | ret = CMPXCHG(&table[index], orig_pte, new_pte); |
b3e4e63f MT |
94 | kunmap_atomic(table, KM_USER0); |
95 | ||
96 | kvm_release_page_dirty(page); | |
97 | ||
98 | return (ret != orig_pte); | |
99 | } | |
100 | ||
bedbe4ee AK |
101 | static unsigned FNAME(gpte_access)(struct kvm_vcpu *vcpu, pt_element_t gpte) |
102 | { | |
103 | unsigned access; | |
104 | ||
105 | access = (gpte & (PT_WRITABLE_MASK | PT_USER_MASK)) | ACC_EXEC_MASK; | |
106 | #if PTTYPE == 64 | |
107 | if (is_nx(vcpu)) | |
108 | access &= ~(gpte >> PT64_NX_SHIFT); | |
109 | #endif | |
110 | return access; | |
111 | } | |
112 | ||
ac79c978 AK |
113 | /* |
114 | * Fetch a guest pte for a guest virtual address | |
115 | */ | |
7993ba43 AK |
116 | static int FNAME(walk_addr)(struct guest_walker *walker, |
117 | struct kvm_vcpu *vcpu, gva_t addr, | |
73b1087e | 118 | int write_fault, int user_fault, int fetch_fault) |
6aa8b732 | 119 | { |
42bf3f0a | 120 | pt_element_t pte; |
cea0f0e7 | 121 | gfn_t table_gfn; |
f59c1d2d | 122 | unsigned index, pt_access, uninitialized_var(pte_access); |
42bf3f0a | 123 | gpa_t pte_gpa; |
f59c1d2d | 124 | bool eperm, present, rsvd_fault; |
6aa8b732 | 125 | |
07420171 AK |
126 | trace_kvm_mmu_pagetable_walk(addr, write_fault, user_fault, |
127 | fetch_fault); | |
b3e4e63f | 128 | walk: |
f59c1d2d AK |
129 | present = true; |
130 | eperm = rsvd_fault = false; | |
ad312c7c ZX |
131 | walker->level = vcpu->arch.mmu.root_level; |
132 | pte = vcpu->arch.cr3; | |
1b0973bd AK |
133 | #if PTTYPE == 64 |
134 | if (!is_long_mode(vcpu)) { | |
6de4f3ad | 135 | pte = kvm_pdptr_read(vcpu, (addr >> 30) & 3); |
07420171 | 136 | trace_kvm_mmu_paging_element(pte, walker->level); |
f59c1d2d AK |
137 | if (!is_present_gpte(pte)) { |
138 | present = false; | |
139 | goto error; | |
140 | } | |
1b0973bd AK |
141 | --walker->level; |
142 | } | |
143 | #endif | |
a9058ecd | 144 | ASSERT((!is_long_mode(vcpu) && is_pae(vcpu)) || |
24993d53 | 145 | (vcpu->arch.cr3 & CR3_NONPAE_RESERVED_BITS) == 0); |
6aa8b732 | 146 | |
fe135d2c | 147 | pt_access = ACC_ALL; |
ac79c978 AK |
148 | |
149 | for (;;) { | |
42bf3f0a | 150 | index = PT_INDEX(addr, walker->level); |
ac79c978 | 151 | |
5fb07ddb | 152 | table_gfn = gpte_to_gfn(pte); |
1755fbcc | 153 | pte_gpa = gfn_to_gpa(table_gfn); |
ec8d4eae | 154 | pte_gpa += index * sizeof(pt_element_t); |
42bf3f0a | 155 | walker->table_gfn[walker->level - 1] = table_gfn; |
7819026e | 156 | walker->pte_gpa[walker->level - 1] = pte_gpa; |
42bf3f0a | 157 | |
f59c1d2d AK |
158 | if (kvm_read_guest(vcpu->kvm, pte_gpa, &pte, sizeof(pte))) { |
159 | present = false; | |
160 | break; | |
161 | } | |
a6085fba | 162 | |
07420171 | 163 | trace_kvm_mmu_paging_element(pte, walker->level); |
42bf3f0a | 164 | |
f59c1d2d AK |
165 | if (!is_present_gpte(pte)) { |
166 | present = false; | |
167 | break; | |
168 | } | |
7993ba43 | 169 | |
f59c1d2d AK |
170 | if (is_rsvd_bits_set(vcpu, pte, walker->level)) { |
171 | rsvd_fault = true; | |
172 | break; | |
173 | } | |
82725b20 | 174 | |
8dae4445 | 175 | if (write_fault && !is_writable_pte(pte)) |
7993ba43 | 176 | if (user_fault || is_write_protection(vcpu)) |
f59c1d2d | 177 | eperm = true; |
7993ba43 | 178 | |
42bf3f0a | 179 | if (user_fault && !(pte & PT_USER_MASK)) |
f59c1d2d | 180 | eperm = true; |
7993ba43 | 181 | |
73b1087e | 182 | #if PTTYPE == 64 |
24222c2f | 183 | if (fetch_fault && (pte & PT64_NX_MASK)) |
f59c1d2d | 184 | eperm = true; |
73b1087e AK |
185 | #endif |
186 | ||
f59c1d2d | 187 | if (!eperm && !rsvd_fault && !(pte & PT_ACCESSED_MASK)) { |
07420171 AK |
188 | trace_kvm_mmu_set_accessed_bit(table_gfn, index, |
189 | sizeof(pte)); | |
b3e4e63f MT |
190 | if (FNAME(cmpxchg_gpte)(vcpu->kvm, table_gfn, |
191 | index, pte, pte|PT_ACCESSED_MASK)) | |
192 | goto walk; | |
f3b8c964 | 193 | mark_page_dirty(vcpu->kvm, table_gfn); |
42bf3f0a | 194 | pte |= PT_ACCESSED_MASK; |
bf3f8e86 | 195 | } |
815af8d4 | 196 | |
bedbe4ee | 197 | pte_access = pt_access & FNAME(gpte_access)(vcpu, pte); |
fe135d2c | 198 | |
7819026e MT |
199 | walker->ptes[walker->level - 1] = pte; |
200 | ||
e04da980 JR |
201 | if ((walker->level == PT_PAGE_TABLE_LEVEL) || |
202 | ((walker->level == PT_DIRECTORY_LEVEL) && | |
814a59d2 | 203 | is_large_pte(pte) && |
e04da980 JR |
204 | (PTTYPE == 64 || is_pse(vcpu))) || |
205 | ((walker->level == PT_PDPE_LEVEL) && | |
814a59d2 | 206 | is_large_pte(pte) && |
e04da980 JR |
207 | is_long_mode(vcpu))) { |
208 | int lvl = walker->level; | |
209 | ||
210 | walker->gfn = gpte_to_gfn_lvl(pte, lvl); | |
211 | walker->gfn += (addr & PT_LVL_OFFSET_MASK(lvl)) | |
212 | >> PAGE_SHIFT; | |
213 | ||
214 | if (PTTYPE == 32 && | |
215 | walker->level == PT_DIRECTORY_LEVEL && | |
216 | is_cpuid_PSE36()) | |
da928521 | 217 | walker->gfn += pse36_gfn_delta(pte); |
e04da980 | 218 | |
ac79c978 | 219 | break; |
815af8d4 | 220 | } |
ac79c978 | 221 | |
fe135d2c | 222 | pt_access = pte_access; |
ac79c978 AK |
223 | --walker->level; |
224 | } | |
42bf3f0a | 225 | |
f59c1d2d AK |
226 | if (!present || eperm || rsvd_fault) |
227 | goto error; | |
228 | ||
43a3795a | 229 | if (write_fault && !is_dirty_gpte(pte)) { |
b3e4e63f MT |
230 | bool ret; |
231 | ||
07420171 | 232 | trace_kvm_mmu_set_dirty_bit(table_gfn, index, sizeof(pte)); |
b3e4e63f MT |
233 | ret = FNAME(cmpxchg_gpte)(vcpu->kvm, table_gfn, index, pte, |
234 | pte|PT_DIRTY_MASK); | |
235 | if (ret) | |
236 | goto walk; | |
f3b8c964 | 237 | mark_page_dirty(vcpu->kvm, table_gfn); |
42bf3f0a | 238 | pte |= PT_DIRTY_MASK; |
7819026e | 239 | walker->ptes[walker->level - 1] = pte; |
42bf3f0a AK |
240 | } |
241 | ||
fe135d2c AK |
242 | walker->pt_access = pt_access; |
243 | walker->pte_access = pte_access; | |
244 | pgprintk("%s: pte %llx pte_access %x pt_access %x\n", | |
518c5a05 | 245 | __func__, (u64)pte, pte_access, pt_access); |
7993ba43 AK |
246 | return 1; |
247 | ||
f59c1d2d | 248 | error: |
7993ba43 | 249 | walker->error_code = 0; |
f59c1d2d AK |
250 | if (present) |
251 | walker->error_code |= PFERR_PRESENT_MASK; | |
7993ba43 AK |
252 | if (write_fault) |
253 | walker->error_code |= PFERR_WRITE_MASK; | |
254 | if (user_fault) | |
255 | walker->error_code |= PFERR_USER_MASK; | |
b0eeec29 | 256 | if (fetch_fault && is_nx(vcpu)) |
73b1087e | 257 | walker->error_code |= PFERR_FETCH_MASK; |
82725b20 DE |
258 | if (rsvd_fault) |
259 | walker->error_code |= PFERR_RSVD_MASK; | |
07420171 | 260 | trace_kvm_mmu_walker_error(walker->error_code); |
fe551881 | 261 | return 0; |
6aa8b732 AK |
262 | } |
263 | ||
ac3cd03c | 264 | static void FNAME(update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp, |
489f1d65 | 265 | u64 *spte, const void *pte) |
0028425f AK |
266 | { |
267 | pt_element_t gpte; | |
41074d07 | 268 | unsigned pte_access; |
35149e21 | 269 | pfn_t pfn; |
fbc5d139 | 270 | u64 new_spte; |
0028425f | 271 | |
0028425f | 272 | gpte = *(const pt_element_t *)pte; |
c7addb90 | 273 | if (~gpte & (PT_PRESENT_MASK | PT_ACCESSED_MASK)) { |
fbc5d139 | 274 | if (!is_present_gpte(gpte)) { |
ac3cd03c | 275 | if (sp->unsync) |
fbc5d139 AK |
276 | new_spte = shadow_trap_nonpresent_pte; |
277 | else | |
278 | new_spte = shadow_notrap_nonpresent_pte; | |
279 | __set_spte(spte, new_spte); | |
280 | } | |
c7addb90 AK |
281 | return; |
282 | } | |
b8688d51 | 283 | pgprintk("%s: gpte %llx spte %p\n", __func__, (u64)gpte, spte); |
ac3cd03c | 284 | pte_access = sp->role.access & FNAME(gpte_access)(vcpu, gpte); |
d7824fff AK |
285 | if (gpte_to_gfn(gpte) != vcpu->arch.update_pte.gfn) |
286 | return; | |
35149e21 AL |
287 | pfn = vcpu->arch.update_pte.pfn; |
288 | if (is_error_pfn(pfn)) | |
d7824fff | 289 | return; |
e930bffe AA |
290 | if (mmu_notifier_retry(vcpu, vcpu->arch.update_pte.mmu_seq)) |
291 | return; | |
35149e21 | 292 | kvm_get_pfn(pfn); |
1403283a IE |
293 | /* |
294 | * we call mmu_set_spte() with reset_host_protection = true beacuse that | |
295 | * vcpu->arch.update_pte.pfn was fetched from get_user_pages(write = 1). | |
296 | */ | |
ac3cd03c | 297 | mmu_set_spte(vcpu, spte, sp->role.access, pte_access, 0, 0, |
cb83cad2 | 298 | is_dirty_gpte(gpte), NULL, PT_PAGE_TABLE_LEVEL, |
1403283a | 299 | gpte_to_gfn(gpte), pfn, true, true); |
0028425f AK |
300 | } |
301 | ||
39c8c672 AK |
302 | static bool FNAME(gpte_changed)(struct kvm_vcpu *vcpu, |
303 | struct guest_walker *gw, int level) | |
304 | { | |
305 | int r; | |
306 | pt_element_t curr_pte; | |
307 | ||
308 | r = kvm_read_guest_atomic(vcpu->kvm, gw->pte_gpa[level - 1], | |
309 | &curr_pte, sizeof(curr_pte)); | |
310 | return r || curr_pte != gw->ptes[level - 1]; | |
311 | } | |
312 | ||
6aa8b732 AK |
313 | /* |
314 | * Fetch a shadow pte for a specific level in the paging hierarchy. | |
315 | */ | |
e7a04c99 AK |
316 | static u64 *FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr, |
317 | struct guest_walker *gw, | |
7e4e4056 | 318 | int user_fault, int write_fault, int hlevel, |
e7a04c99 | 319 | int *ptwrite, pfn_t pfn) |
6aa8b732 | 320 | { |
abb9e0b8 | 321 | unsigned access = gw->pt_access; |
5991b332 | 322 | struct kvm_mmu_page *sp = NULL; |
32ef26a3 | 323 | u64 *sptep = NULL; |
0b3c9333 | 324 | int uninitialized_var(level); |
84754cd8 | 325 | bool dirty = is_dirty_gpte(gw->ptes[gw->level - 1]); |
5991b332 | 326 | int top_level; |
84754cd8 | 327 | unsigned direct_access; |
e7a04c99 | 328 | struct kvm_shadow_walk_iterator iterator; |
abb9e0b8 | 329 | |
43a3795a | 330 | if (!is_present_gpte(gw->ptes[gw->level - 1])) |
e7a04c99 | 331 | return NULL; |
6aa8b732 | 332 | |
84754cd8 XG |
333 | direct_access = gw->pt_access & gw->pte_access; |
334 | if (!dirty) | |
335 | direct_access &= ~ACC_WRITE_MASK; | |
336 | ||
5991b332 AK |
337 | top_level = vcpu->arch.mmu.root_level; |
338 | if (top_level == PT32E_ROOT_LEVEL) | |
339 | top_level = PT32_ROOT_LEVEL; | |
340 | /* | |
341 | * Verify that the top-level gpte is still there. Since the page | |
342 | * is a root page, it is either write protected (and cannot be | |
343 | * changed from now on) or it is invalid (in which case, we don't | |
344 | * really care if it changes underneath us after this point). | |
345 | */ | |
346 | if (FNAME(gpte_changed)(vcpu, gw, top_level)) | |
347 | goto out_gpte_changed; | |
348 | ||
0b3c9333 AK |
349 | for (shadow_walk_init(&iterator, vcpu, addr); |
350 | shadow_walk_okay(&iterator) && iterator.level > gw->level; | |
351 | shadow_walk_next(&iterator)) { | |
352 | gfn_t table_gfn; | |
353 | ||
e7a04c99 AK |
354 | level = iterator.level; |
355 | sptep = iterator.sptep; | |
abb9e0b8 | 356 | |
a3aa51cf | 357 | drop_large_spte(vcpu, sptep); |
ef0197e8 | 358 | |
5991b332 AK |
359 | sp = NULL; |
360 | if (!is_shadow_present_pte(*sptep)) { | |
361 | table_gfn = gw->table_gfn[level - 2]; | |
362 | sp = kvm_mmu_get_page(vcpu, table_gfn, addr, level-1, | |
363 | false, access, sptep); | |
364 | } | |
0b3c9333 AK |
365 | |
366 | /* | |
367 | * Verify that the gpte in the page we've just write | |
368 | * protected is still there. | |
369 | */ | |
370 | if (FNAME(gpte_changed)(vcpu, gw, level - 1)) | |
371 | goto out_gpte_changed; | |
abb9e0b8 | 372 | |
5991b332 AK |
373 | if (sp) |
374 | link_shadow_page(sptep, sp); | |
e7a04c99 | 375 | } |
050e6499 | 376 | |
0b3c9333 AK |
377 | for (; |
378 | shadow_walk_okay(&iterator) && iterator.level > hlevel; | |
379 | shadow_walk_next(&iterator)) { | |
380 | gfn_t direct_gfn; | |
381 | ||
382 | level = iterator.level; | |
383 | sptep = iterator.sptep; | |
384 | ||
385 | validate_direct_spte(vcpu, sptep, direct_access); | |
386 | ||
387 | drop_large_spte(vcpu, sptep); | |
388 | ||
389 | if (is_shadow_present_pte(*sptep)) | |
390 | continue; | |
391 | ||
392 | direct_gfn = gw->gfn & ~(KVM_PAGES_PER_HPAGE(level) - 1); | |
393 | ||
394 | sp = kvm_mmu_get_page(vcpu, direct_gfn, addr, level-1, | |
395 | true, direct_access, sptep); | |
396 | link_shadow_page(sptep, sp); | |
397 | } | |
398 | ||
399 | sptep = iterator.sptep; | |
400 | level = iterator.level; | |
401 | ||
402 | mmu_set_spte(vcpu, sptep, access, gw->pte_access & access, | |
403 | user_fault, write_fault, dirty, ptwrite, level, | |
404 | gw->gfn, pfn, false, true); | |
405 | ||
e7a04c99 | 406 | return sptep; |
0b3c9333 AK |
407 | |
408 | out_gpte_changed: | |
5991b332 AK |
409 | if (sp) |
410 | kvm_mmu_put_page(sp, sptep); | |
0b3c9333 AK |
411 | kvm_release_pfn_clean(pfn); |
412 | return NULL; | |
6aa8b732 AK |
413 | } |
414 | ||
6aa8b732 AK |
415 | /* |
416 | * Page fault handler. There are several causes for a page fault: | |
417 | * - there is no shadow pte for the guest pte | |
418 | * - write access through a shadow pte marked read only so that we can set | |
419 | * the dirty bit | |
420 | * - write access to a shadow pte marked read only so we can update the page | |
421 | * dirty bitmap, when userspace requests it | |
422 | * - mmio access; in this case we will never install a present shadow pte | |
423 | * - normal guest page fault due to the guest pte marked not present, not | |
424 | * writable, or not executable | |
425 | * | |
e2dec939 AK |
426 | * Returns: 1 if we need to emulate the instruction, 0 otherwise, or |
427 | * a negative value on error. | |
6aa8b732 AK |
428 | */ |
429 | static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr, | |
430 | u32 error_code) | |
431 | { | |
432 | int write_fault = error_code & PFERR_WRITE_MASK; | |
6aa8b732 | 433 | int user_fault = error_code & PFERR_USER_MASK; |
73b1087e | 434 | int fetch_fault = error_code & PFERR_FETCH_MASK; |
6aa8b732 | 435 | struct guest_walker walker; |
d555c333 | 436 | u64 *sptep; |
cea0f0e7 | 437 | int write_pt = 0; |
e2dec939 | 438 | int r; |
35149e21 | 439 | pfn_t pfn; |
7e4e4056 | 440 | int level = PT_PAGE_TABLE_LEVEL; |
e930bffe | 441 | unsigned long mmu_seq; |
6aa8b732 | 442 | |
b8688d51 | 443 | pgprintk("%s: addr %lx err %x\n", __func__, addr, error_code); |
37a7d8b0 | 444 | kvm_mmu_audit(vcpu, "pre page fault"); |
714b93da | 445 | |
e2dec939 AK |
446 | r = mmu_topup_memory_caches(vcpu); |
447 | if (r) | |
448 | return r; | |
714b93da | 449 | |
6aa8b732 | 450 | /* |
a8b876b1 | 451 | * Look up the guest pte for the faulting address. |
6aa8b732 | 452 | */ |
73b1087e AK |
453 | r = FNAME(walk_addr)(&walker, vcpu, addr, write_fault, user_fault, |
454 | fetch_fault); | |
6aa8b732 AK |
455 | |
456 | /* | |
457 | * The page is not mapped by the guest. Let the guest handle it. | |
458 | */ | |
7993ba43 | 459 | if (!r) { |
b8688d51 | 460 | pgprintk("%s: guest page fault\n", __func__); |
7993ba43 | 461 | inject_page_fault(vcpu, addr, walker.error_code); |
ad312c7c | 462 | vcpu->arch.last_pt_write_count = 0; /* reset fork detector */ |
6aa8b732 AK |
463 | return 0; |
464 | } | |
465 | ||
7e4e4056 JR |
466 | if (walker.level >= PT_DIRECTORY_LEVEL) { |
467 | level = min(walker.level, mapping_level(vcpu, walker.gfn)); | |
468 | walker.gfn = walker.gfn & ~(KVM_PAGES_PER_HPAGE(level) - 1); | |
05da4558 | 469 | } |
7e4e4056 | 470 | |
e930bffe | 471 | mmu_seq = vcpu->kvm->mmu_notifier_seq; |
4c2155ce | 472 | smp_rmb(); |
35149e21 | 473 | pfn = gfn_to_pfn(vcpu->kvm, walker.gfn); |
d7824fff | 474 | |
d196e343 | 475 | /* mmio */ |
bf998156 HY |
476 | if (is_error_pfn(pfn)) |
477 | return kvm_handle_bad_page(vcpu->kvm, walker.gfn, pfn); | |
d196e343 | 478 | |
aaee2c94 | 479 | spin_lock(&vcpu->kvm->mmu_lock); |
e930bffe AA |
480 | if (mmu_notifier_retry(vcpu, mmu_seq)) |
481 | goto out_unlock; | |
eb787d10 | 482 | kvm_mmu_free_some_pages(vcpu); |
d555c333 | 483 | sptep = FNAME(fetch)(vcpu, addr, &walker, user_fault, write_fault, |
7e4e4056 | 484 | level, &write_pt, pfn); |
a24e8099 | 485 | (void)sptep; |
b8688d51 | 486 | pgprintk("%s: shadow pte %p %llx ptwrite %d\n", __func__, |
d555c333 | 487 | sptep, *sptep, write_pt); |
cea0f0e7 | 488 | |
a25f7e1f | 489 | if (!write_pt) |
ad312c7c | 490 | vcpu->arch.last_pt_write_count = 0; /* reset fork detector */ |
a25f7e1f | 491 | |
1165f5fe | 492 | ++vcpu->stat.pf_fixed; |
37a7d8b0 | 493 | kvm_mmu_audit(vcpu, "post page fault (fixed)"); |
aaee2c94 | 494 | spin_unlock(&vcpu->kvm->mmu_lock); |
6aa8b732 | 495 | |
cea0f0e7 | 496 | return write_pt; |
e930bffe AA |
497 | |
498 | out_unlock: | |
499 | spin_unlock(&vcpu->kvm->mmu_lock); | |
500 | kvm_release_pfn_clean(pfn); | |
501 | return 0; | |
6aa8b732 AK |
502 | } |
503 | ||
a461930b | 504 | static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva) |
a7052897 | 505 | { |
a461930b | 506 | struct kvm_shadow_walk_iterator iterator; |
f78978aa | 507 | struct kvm_mmu_page *sp; |
08e850c6 | 508 | gpa_t pte_gpa = -1; |
a461930b AK |
509 | int level; |
510 | u64 *sptep; | |
4539b358 | 511 | int need_flush = 0; |
a461930b AK |
512 | |
513 | spin_lock(&vcpu->kvm->mmu_lock); | |
a7052897 | 514 | |
a461930b AK |
515 | for_each_shadow_entry(vcpu, gva, iterator) { |
516 | level = iterator.level; | |
517 | sptep = iterator.sptep; | |
ad218f85 | 518 | |
f78978aa | 519 | sp = page_header(__pa(sptep)); |
884a0ff0 | 520 | if (is_last_spte(*sptep, level)) { |
22c9b2d1 | 521 | int offset, shift; |
08e850c6 | 522 | |
f78978aa XG |
523 | if (!sp->unsync) |
524 | break; | |
525 | ||
22c9b2d1 XG |
526 | shift = PAGE_SHIFT - |
527 | (PT_LEVEL_BITS - PT64_LEVEL_BITS) * level; | |
528 | offset = sp->role.quadrant << shift; | |
529 | ||
530 | pte_gpa = (sp->gfn << PAGE_SHIFT) + offset; | |
08e850c6 | 531 | pte_gpa += (sptep - sp->spt) * sizeof(pt_element_t); |
a461930b AK |
532 | |
533 | if (is_shadow_present_pte(*sptep)) { | |
a461930b AK |
534 | if (is_large_pte(*sptep)) |
535 | --vcpu->kvm->stat.lpages; | |
be38d276 AK |
536 | drop_spte(vcpu->kvm, sptep, |
537 | shadow_trap_nonpresent_pte); | |
4539b358 | 538 | need_flush = 1; |
be38d276 AK |
539 | } else |
540 | __set_spte(sptep, shadow_trap_nonpresent_pte); | |
a461930b | 541 | break; |
87917239 | 542 | } |
a7052897 | 543 | |
f78978aa | 544 | if (!is_shadow_present_pte(*sptep) || !sp->unsync_children) |
a461930b AK |
545 | break; |
546 | } | |
a7052897 | 547 | |
4539b358 AA |
548 | if (need_flush) |
549 | kvm_flush_remote_tlbs(vcpu->kvm); | |
08e850c6 AK |
550 | |
551 | atomic_inc(&vcpu->kvm->arch.invlpg_counter); | |
552 | ||
ad218f85 | 553 | spin_unlock(&vcpu->kvm->mmu_lock); |
08e850c6 AK |
554 | |
555 | if (pte_gpa == -1) | |
556 | return; | |
557 | ||
558 | if (mmu_topup_memory_caches(vcpu)) | |
559 | return; | |
560 | kvm_mmu_pte_write(vcpu, pte_gpa, NULL, sizeof(pt_element_t), 0); | |
a7052897 MT |
561 | } |
562 | ||
1871c602 GN |
563 | static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t vaddr, u32 access, |
564 | u32 *error) | |
6aa8b732 AK |
565 | { |
566 | struct guest_walker walker; | |
e119d117 AK |
567 | gpa_t gpa = UNMAPPED_GVA; |
568 | int r; | |
6aa8b732 | 569 | |
1871c602 GN |
570 | r = FNAME(walk_addr)(&walker, vcpu, vaddr, |
571 | !!(access & PFERR_WRITE_MASK), | |
572 | !!(access & PFERR_USER_MASK), | |
573 | !!(access & PFERR_FETCH_MASK)); | |
6aa8b732 | 574 | |
e119d117 | 575 | if (r) { |
1755fbcc | 576 | gpa = gfn_to_gpa(walker.gfn); |
e119d117 | 577 | gpa |= vaddr & ~PAGE_MASK; |
1871c602 GN |
578 | } else if (error) |
579 | *error = walker.error_code; | |
6aa8b732 AK |
580 | |
581 | return gpa; | |
582 | } | |
583 | ||
c7addb90 AK |
584 | static void FNAME(prefetch_page)(struct kvm_vcpu *vcpu, |
585 | struct kvm_mmu_page *sp) | |
586 | { | |
eab9f71f AK |
587 | int i, j, offset, r; |
588 | pt_element_t pt[256 / sizeof(pt_element_t)]; | |
589 | gpa_t pte_gpa; | |
c7addb90 | 590 | |
f6e2c02b | 591 | if (sp->role.direct |
e5a4c8ca | 592 | || (PTTYPE == 32 && sp->role.level > PT_PAGE_TABLE_LEVEL)) { |
c7addb90 AK |
593 | nonpaging_prefetch_page(vcpu, sp); |
594 | return; | |
595 | } | |
596 | ||
eab9f71f AK |
597 | pte_gpa = gfn_to_gpa(sp->gfn); |
598 | if (PTTYPE == 32) { | |
e5a4c8ca | 599 | offset = sp->role.quadrant << PT64_LEVEL_BITS; |
eab9f71f AK |
600 | pte_gpa += offset * sizeof(pt_element_t); |
601 | } | |
7ec54588 | 602 | |
eab9f71f AK |
603 | for (i = 0; i < PT64_ENT_PER_PAGE; i += ARRAY_SIZE(pt)) { |
604 | r = kvm_read_guest_atomic(vcpu->kvm, pte_gpa, pt, sizeof pt); | |
605 | pte_gpa += ARRAY_SIZE(pt) * sizeof(pt_element_t); | |
606 | for (j = 0; j < ARRAY_SIZE(pt); ++j) | |
43a3795a | 607 | if (r || is_present_gpte(pt[j])) |
eab9f71f AK |
608 | sp->spt[i+j] = shadow_trap_nonpresent_pte; |
609 | else | |
610 | sp->spt[i+j] = shadow_notrap_nonpresent_pte; | |
7ec54588 | 611 | } |
c7addb90 AK |
612 | } |
613 | ||
e8bc217a MT |
614 | /* |
615 | * Using the cached information from sp->gfns is safe because: | |
616 | * - The spte has a reference to the struct page, so the pfn for a given gfn | |
617 | * can't change unless all sptes pointing to it are nuked first. | |
e8bc217a | 618 | */ |
be71e061 XG |
619 | static int FNAME(sync_page)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp, |
620 | bool clear_unsync) | |
e8bc217a MT |
621 | { |
622 | int i, offset, nr_present; | |
1403283a | 623 | bool reset_host_protection; |
51fb60d8 | 624 | gpa_t first_pte_gpa; |
e8bc217a MT |
625 | |
626 | offset = nr_present = 0; | |
627 | ||
2032a93d LJ |
628 | /* direct kvm_mmu_page can not be unsync. */ |
629 | BUG_ON(sp->role.direct); | |
630 | ||
e8bc217a MT |
631 | if (PTTYPE == 32) |
632 | offset = sp->role.quadrant << PT64_LEVEL_BITS; | |
633 | ||
51fb60d8 GJ |
634 | first_pte_gpa = gfn_to_gpa(sp->gfn) + offset * sizeof(pt_element_t); |
635 | ||
e8bc217a MT |
636 | for (i = 0; i < PT64_ENT_PER_PAGE; i++) { |
637 | unsigned pte_access; | |
638 | pt_element_t gpte; | |
639 | gpa_t pte_gpa; | |
f55c3f41 | 640 | gfn_t gfn; |
e8bc217a MT |
641 | |
642 | if (!is_shadow_present_pte(sp->spt[i])) | |
643 | continue; | |
644 | ||
51fb60d8 | 645 | pte_gpa = first_pte_gpa + i * sizeof(pt_element_t); |
e8bc217a MT |
646 | |
647 | if (kvm_read_guest_atomic(vcpu->kvm, pte_gpa, &gpte, | |
648 | sizeof(pt_element_t))) | |
649 | return -EINVAL; | |
650 | ||
f55c3f41 | 651 | gfn = gpte_to_gfn(gpte); |
a1f4d395 | 652 | if (gfn != sp->gfns[i] || |
f55c3f41 | 653 | !is_present_gpte(gpte) || !(gpte & PT_ACCESSED_MASK)) { |
e8bc217a MT |
654 | u64 nonpresent; |
655 | ||
be71e061 | 656 | if (is_present_gpte(gpte) || !clear_unsync) |
e8bc217a MT |
657 | nonpresent = shadow_trap_nonpresent_pte; |
658 | else | |
659 | nonpresent = shadow_notrap_nonpresent_pte; | |
be38d276 | 660 | drop_spte(vcpu->kvm, &sp->spt[i], nonpresent); |
e8bc217a MT |
661 | continue; |
662 | } | |
663 | ||
664 | nr_present++; | |
665 | pte_access = sp->role.access & FNAME(gpte_access)(vcpu, gpte); | |
1403283a IE |
666 | if (!(sp->spt[i] & SPTE_HOST_WRITEABLE)) { |
667 | pte_access &= ~ACC_WRITE_MASK; | |
668 | reset_host_protection = 0; | |
669 | } else { | |
670 | reset_host_protection = 1; | |
671 | } | |
e8bc217a | 672 | set_spte(vcpu, &sp->spt[i], pte_access, 0, 0, |
7e4e4056 | 673 | is_dirty_gpte(gpte), PT_PAGE_TABLE_LEVEL, gfn, |
1403283a IE |
674 | spte_to_pfn(sp->spt[i]), true, false, |
675 | reset_host_protection); | |
e8bc217a MT |
676 | } |
677 | ||
678 | return !nr_present; | |
679 | } | |
680 | ||
6aa8b732 AK |
681 | #undef pt_element_t |
682 | #undef guest_walker | |
683 | #undef FNAME | |
684 | #undef PT_BASE_ADDR_MASK | |
685 | #undef PT_INDEX | |
6aa8b732 | 686 | #undef PT_LEVEL_MASK |
e04da980 JR |
687 | #undef PT_LVL_ADDR_MASK |
688 | #undef PT_LVL_OFFSET_MASK | |
c7addb90 | 689 | #undef PT_LEVEL_BITS |
cea0f0e7 | 690 | #undef PT_MAX_FULL_LEVELS |
5fb07ddb | 691 | #undef gpte_to_gfn |
e04da980 | 692 | #undef gpte_to_gfn_lvl |
b3e4e63f | 693 | #undef CMPXCHG |