]>
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. | |
10 | * | |
11 | * Authors: | |
12 | * Yaniv Kamay <yaniv@qumranet.com> | |
13 | * Avi Kivity <avi@qumranet.com> | |
14 | * | |
15 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
16 | * the COPYING file in the top-level directory. | |
17 | * | |
18 | */ | |
19 | ||
20 | /* | |
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. | |
23 | */ | |
24 | ||
25 | #if PTTYPE == 64 | |
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 | |
e04da980 JR |
30 | #define PT_LVL_ADDR_MASK(lvl) PT64_LVL_ADDR_MASK(lvl) |
31 | #define PT_LVL_OFFSET_MASK(lvl) PT64_LVL_OFFSET_MASK(lvl) | |
6aa8b732 | 32 | #define PT_INDEX(addr, level) PT64_INDEX(addr, level) |
6aa8b732 | 33 | #define PT_LEVEL_MASK(level) PT64_LEVEL_MASK(level) |
c7addb90 | 34 | #define PT_LEVEL_BITS PT64_LEVEL_BITS |
cea0f0e7 AK |
35 | #ifdef CONFIG_X86_64 |
36 | #define PT_MAX_FULL_LEVELS 4 | |
b3e4e63f | 37 | #define CMPXCHG cmpxchg |
cea0f0e7 | 38 | #else |
b3e4e63f | 39 | #define CMPXCHG cmpxchg64 |
cea0f0e7 AK |
40 | #define PT_MAX_FULL_LEVELS 2 |
41 | #endif | |
6aa8b732 AK |
42 | #elif PTTYPE == 32 |
43 | #define pt_element_t u32 | |
44 | #define guest_walker guest_walker32 | |
45 | #define FNAME(name) paging##32_##name | |
46 | #define PT_BASE_ADDR_MASK PT32_BASE_ADDR_MASK | |
e04da980 JR |
47 | #define PT_LVL_ADDR_MASK(lvl) PT32_LVL_ADDR_MASK(lvl) |
48 | #define PT_LVL_OFFSET_MASK(lvl) PT32_LVL_OFFSET_MASK(lvl) | |
6aa8b732 | 49 | #define PT_INDEX(addr, level) PT32_INDEX(addr, level) |
6aa8b732 | 50 | #define PT_LEVEL_MASK(level) PT32_LEVEL_MASK(level) |
c7addb90 | 51 | #define PT_LEVEL_BITS PT32_LEVEL_BITS |
cea0f0e7 | 52 | #define PT_MAX_FULL_LEVELS 2 |
b3e4e63f | 53 | #define CMPXCHG cmpxchg |
6aa8b732 AK |
54 | #else |
55 | #error Invalid PTTYPE value | |
56 | #endif | |
57 | ||
e04da980 JR |
58 | #define gpte_to_gfn_lvl FNAME(gpte_to_gfn_lvl) |
59 | #define gpte_to_gfn(pte) gpte_to_gfn_lvl((pte), PT_PAGE_TABLE_LEVEL) | |
5fb07ddb | 60 | |
6aa8b732 AK |
61 | /* |
62 | * The guest_walker structure emulates the behavior of the hardware page | |
63 | * table walker. | |
64 | */ | |
65 | struct guest_walker { | |
66 | int level; | |
cea0f0e7 | 67 | gfn_t table_gfn[PT_MAX_FULL_LEVELS]; |
7819026e MT |
68 | pt_element_t ptes[PT_MAX_FULL_LEVELS]; |
69 | gpa_t pte_gpa[PT_MAX_FULL_LEVELS]; | |
fe135d2c AK |
70 | unsigned pt_access; |
71 | unsigned pte_access; | |
815af8d4 | 72 | gfn_t gfn; |
7993ba43 | 73 | u32 error_code; |
6aa8b732 AK |
74 | }; |
75 | ||
e04da980 | 76 | static gfn_t gpte_to_gfn_lvl(pt_element_t gpte, int lvl) |
5fb07ddb | 77 | { |
e04da980 | 78 | return (gpte & PT_LVL_ADDR_MASK(lvl)) >> PAGE_SHIFT; |
5fb07ddb AK |
79 | } |
80 | ||
b3e4e63f MT |
81 | static bool FNAME(cmpxchg_gpte)(struct kvm *kvm, |
82 | gfn_t table_gfn, unsigned index, | |
83 | pt_element_t orig_pte, pt_element_t new_pte) | |
84 | { | |
85 | pt_element_t ret; | |
86 | pt_element_t *table; | |
87 | struct page *page; | |
88 | ||
89 | page = gfn_to_page(kvm, table_gfn); | |
72dc67a6 | 90 | |
b3e4e63f | 91 | table = kmap_atomic(page, KM_USER0); |
b3e4e63f | 92 | ret = CMPXCHG(&table[index], orig_pte, new_pte); |
b3e4e63f MT |
93 | kunmap_atomic(table, KM_USER0); |
94 | ||
95 | kvm_release_page_dirty(page); | |
96 | ||
97 | return (ret != orig_pte); | |
98 | } | |
99 | ||
bedbe4ee AK |
100 | static unsigned FNAME(gpte_access)(struct kvm_vcpu *vcpu, pt_element_t gpte) |
101 | { | |
102 | unsigned access; | |
103 | ||
104 | access = (gpte & (PT_WRITABLE_MASK | PT_USER_MASK)) | ACC_EXEC_MASK; | |
105 | #if PTTYPE == 64 | |
106 | if (is_nx(vcpu)) | |
107 | access &= ~(gpte >> PT64_NX_SHIFT); | |
108 | #endif | |
109 | return access; | |
110 | } | |
111 | ||
ac79c978 AK |
112 | /* |
113 | * Fetch a guest pte for a guest virtual address | |
114 | */ | |
7993ba43 AK |
115 | static int FNAME(walk_addr)(struct guest_walker *walker, |
116 | struct kvm_vcpu *vcpu, gva_t addr, | |
73b1087e | 117 | int write_fault, int user_fault, int fetch_fault) |
6aa8b732 | 118 | { |
42bf3f0a | 119 | pt_element_t pte; |
cea0f0e7 | 120 | gfn_t table_gfn; |
fe135d2c | 121 | unsigned index, pt_access, pte_access; |
42bf3f0a | 122 | gpa_t pte_gpa; |
82725b20 | 123 | int rsvd_fault = 0; |
6aa8b732 | 124 | |
07420171 AK |
125 | trace_kvm_mmu_pagetable_walk(addr, write_fault, user_fault, |
126 | fetch_fault); | |
b3e4e63f | 127 | walk: |
ad312c7c ZX |
128 | walker->level = vcpu->arch.mmu.root_level; |
129 | pte = vcpu->arch.cr3; | |
1b0973bd AK |
130 | #if PTTYPE == 64 |
131 | if (!is_long_mode(vcpu)) { | |
6de4f3ad | 132 | pte = kvm_pdptr_read(vcpu, (addr >> 30) & 3); |
07420171 | 133 | trace_kvm_mmu_paging_element(pte, walker->level); |
43a3795a | 134 | if (!is_present_gpte(pte)) |
7993ba43 | 135 | goto not_present; |
1b0973bd AK |
136 | --walker->level; |
137 | } | |
138 | #endif | |
a9058ecd | 139 | ASSERT((!is_long_mode(vcpu) && is_pae(vcpu)) || |
24993d53 | 140 | (vcpu->arch.cr3 & CR3_NONPAE_RESERVED_BITS) == 0); |
6aa8b732 | 141 | |
fe135d2c | 142 | pt_access = ACC_ALL; |
ac79c978 AK |
143 | |
144 | for (;;) { | |
42bf3f0a | 145 | index = PT_INDEX(addr, walker->level); |
ac79c978 | 146 | |
5fb07ddb | 147 | table_gfn = gpte_to_gfn(pte); |
1755fbcc | 148 | pte_gpa = gfn_to_gpa(table_gfn); |
ec8d4eae | 149 | pte_gpa += index * sizeof(pt_element_t); |
42bf3f0a | 150 | walker->table_gfn[walker->level - 1] = table_gfn; |
7819026e | 151 | walker->pte_gpa[walker->level - 1] = pte_gpa; |
42bf3f0a | 152 | |
a6085fba MT |
153 | if (kvm_read_guest(vcpu->kvm, pte_gpa, &pte, sizeof(pte))) |
154 | goto not_present; | |
155 | ||
07420171 | 156 | trace_kvm_mmu_paging_element(pte, walker->level); |
42bf3f0a | 157 | |
43a3795a | 158 | if (!is_present_gpte(pte)) |
7993ba43 AK |
159 | goto not_present; |
160 | ||
82725b20 DE |
161 | rsvd_fault = is_rsvd_bits_set(vcpu, pte, walker->level); |
162 | if (rsvd_fault) | |
163 | goto access_error; | |
164 | ||
8dae4445 | 165 | if (write_fault && !is_writable_pte(pte)) |
7993ba43 AK |
166 | if (user_fault || is_write_protection(vcpu)) |
167 | goto access_error; | |
168 | ||
42bf3f0a | 169 | if (user_fault && !(pte & PT_USER_MASK)) |
7993ba43 AK |
170 | goto access_error; |
171 | ||
73b1087e | 172 | #if PTTYPE == 64 |
24222c2f | 173 | if (fetch_fault && (pte & PT64_NX_MASK)) |
73b1087e AK |
174 | goto access_error; |
175 | #endif | |
176 | ||
42bf3f0a | 177 | if (!(pte & PT_ACCESSED_MASK)) { |
07420171 AK |
178 | trace_kvm_mmu_set_accessed_bit(table_gfn, index, |
179 | sizeof(pte)); | |
b3e4e63f MT |
180 | if (FNAME(cmpxchg_gpte)(vcpu->kvm, table_gfn, |
181 | index, pte, pte|PT_ACCESSED_MASK)) | |
182 | goto walk; | |
f3b8c964 | 183 | mark_page_dirty(vcpu->kvm, table_gfn); |
42bf3f0a | 184 | pte |= PT_ACCESSED_MASK; |
bf3f8e86 | 185 | } |
815af8d4 | 186 | |
bedbe4ee | 187 | pte_access = pt_access & FNAME(gpte_access)(vcpu, pte); |
fe135d2c | 188 | |
7819026e MT |
189 | walker->ptes[walker->level - 1] = pte; |
190 | ||
e04da980 JR |
191 | if ((walker->level == PT_PAGE_TABLE_LEVEL) || |
192 | ((walker->level == PT_DIRECTORY_LEVEL) && | |
814a59d2 | 193 | is_large_pte(pte) && |
e04da980 JR |
194 | (PTTYPE == 64 || is_pse(vcpu))) || |
195 | ((walker->level == PT_PDPE_LEVEL) && | |
814a59d2 | 196 | is_large_pte(pte) && |
e04da980 JR |
197 | is_long_mode(vcpu))) { |
198 | int lvl = walker->level; | |
199 | ||
200 | walker->gfn = gpte_to_gfn_lvl(pte, lvl); | |
201 | walker->gfn += (addr & PT_LVL_OFFSET_MASK(lvl)) | |
202 | >> PAGE_SHIFT; | |
203 | ||
204 | if (PTTYPE == 32 && | |
205 | walker->level == PT_DIRECTORY_LEVEL && | |
206 | is_cpuid_PSE36()) | |
da928521 | 207 | walker->gfn += pse36_gfn_delta(pte); |
e04da980 | 208 | |
ac79c978 | 209 | break; |
815af8d4 | 210 | } |
ac79c978 | 211 | |
fe135d2c | 212 | pt_access = pte_access; |
ac79c978 AK |
213 | --walker->level; |
214 | } | |
42bf3f0a | 215 | |
43a3795a | 216 | if (write_fault && !is_dirty_gpte(pte)) { |
b3e4e63f MT |
217 | bool ret; |
218 | ||
07420171 | 219 | trace_kvm_mmu_set_dirty_bit(table_gfn, index, sizeof(pte)); |
b3e4e63f MT |
220 | ret = FNAME(cmpxchg_gpte)(vcpu->kvm, table_gfn, index, pte, |
221 | pte|PT_DIRTY_MASK); | |
222 | if (ret) | |
223 | goto walk; | |
f3b8c964 | 224 | mark_page_dirty(vcpu->kvm, table_gfn); |
42bf3f0a | 225 | pte |= PT_DIRTY_MASK; |
7819026e | 226 | walker->ptes[walker->level - 1] = pte; |
42bf3f0a AK |
227 | } |
228 | ||
fe135d2c AK |
229 | walker->pt_access = pt_access; |
230 | walker->pte_access = pte_access; | |
231 | pgprintk("%s: pte %llx pte_access %x pt_access %x\n", | |
b8688d51 | 232 | __func__, (u64)pte, pt_access, pte_access); |
7993ba43 AK |
233 | return 1; |
234 | ||
235 | not_present: | |
236 | walker->error_code = 0; | |
237 | goto err; | |
238 | ||
239 | access_error: | |
240 | walker->error_code = PFERR_PRESENT_MASK; | |
241 | ||
242 | err: | |
243 | if (write_fault) | |
244 | walker->error_code |= PFERR_WRITE_MASK; | |
245 | if (user_fault) | |
246 | walker->error_code |= PFERR_USER_MASK; | |
73b1087e AK |
247 | if (fetch_fault) |
248 | walker->error_code |= PFERR_FETCH_MASK; | |
82725b20 DE |
249 | if (rsvd_fault) |
250 | walker->error_code |= PFERR_RSVD_MASK; | |
07420171 | 251 | trace_kvm_mmu_walker_error(walker->error_code); |
fe551881 | 252 | return 0; |
6aa8b732 AK |
253 | } |
254 | ||
0028425f | 255 | static void FNAME(update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *page, |
489f1d65 | 256 | u64 *spte, const void *pte) |
0028425f AK |
257 | { |
258 | pt_element_t gpte; | |
41074d07 | 259 | unsigned pte_access; |
35149e21 | 260 | pfn_t pfn; |
fbc5d139 | 261 | u64 new_spte; |
0028425f | 262 | |
0028425f | 263 | gpte = *(const pt_element_t *)pte; |
c7addb90 | 264 | if (~gpte & (PT_PRESENT_MASK | PT_ACCESSED_MASK)) { |
fbc5d139 AK |
265 | if (!is_present_gpte(gpte)) { |
266 | if (page->unsync) | |
267 | new_spte = shadow_trap_nonpresent_pte; | |
268 | else | |
269 | new_spte = shadow_notrap_nonpresent_pte; | |
270 | __set_spte(spte, new_spte); | |
271 | } | |
c7addb90 AK |
272 | return; |
273 | } | |
b8688d51 | 274 | pgprintk("%s: gpte %llx spte %p\n", __func__, (u64)gpte, spte); |
41074d07 | 275 | pte_access = page->role.access & FNAME(gpte_access)(vcpu, gpte); |
d7824fff AK |
276 | if (gpte_to_gfn(gpte) != vcpu->arch.update_pte.gfn) |
277 | return; | |
35149e21 AL |
278 | pfn = vcpu->arch.update_pte.pfn; |
279 | if (is_error_pfn(pfn)) | |
d7824fff | 280 | return; |
e930bffe AA |
281 | if (mmu_notifier_retry(vcpu, vcpu->arch.update_pte.mmu_seq)) |
282 | return; | |
35149e21 | 283 | kvm_get_pfn(pfn); |
1403283a IE |
284 | /* |
285 | * we call mmu_set_spte() with reset_host_protection = true beacuse that | |
286 | * vcpu->arch.update_pte.pfn was fetched from get_user_pages(write = 1). | |
287 | */ | |
1c4f1fd6 | 288 | mmu_set_spte(vcpu, spte, page->role.access, pte_access, 0, 0, |
7e4e4056 | 289 | gpte & PT_DIRTY_MASK, NULL, PT_PAGE_TABLE_LEVEL, |
1403283a | 290 | gpte_to_gfn(gpte), pfn, true, true); |
0028425f AK |
291 | } |
292 | ||
6aa8b732 AK |
293 | /* |
294 | * Fetch a shadow pte for a specific level in the paging hierarchy. | |
295 | */ | |
e7a04c99 AK |
296 | static u64 *FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr, |
297 | struct guest_walker *gw, | |
7e4e4056 | 298 | int user_fault, int write_fault, int hlevel, |
e7a04c99 | 299 | int *ptwrite, pfn_t pfn) |
6aa8b732 | 300 | { |
abb9e0b8 AK |
301 | unsigned access = gw->pt_access; |
302 | struct kvm_mmu_page *shadow_page; | |
bde89223 | 303 | u64 spte, *sptep = NULL; |
f6e2c02b | 304 | int direct; |
abb9e0b8 AK |
305 | gfn_t table_gfn; |
306 | int r; | |
e7a04c99 | 307 | int level; |
abb9e0b8 | 308 | pt_element_t curr_pte; |
e7a04c99 | 309 | struct kvm_shadow_walk_iterator iterator; |
abb9e0b8 | 310 | |
43a3795a | 311 | if (!is_present_gpte(gw->ptes[gw->level - 1])) |
e7a04c99 | 312 | return NULL; |
6aa8b732 | 313 | |
e7a04c99 AK |
314 | for_each_shadow_entry(vcpu, addr, iterator) { |
315 | level = iterator.level; | |
316 | sptep = iterator.sptep; | |
7e4e4056 | 317 | if (iterator.level == hlevel) { |
e7a04c99 AK |
318 | mmu_set_spte(vcpu, sptep, access, |
319 | gw->pte_access & access, | |
320 | user_fault, write_fault, | |
321 | gw->ptes[gw->level-1] & PT_DIRTY_MASK, | |
852e3c19 | 322 | ptwrite, level, |
1403283a | 323 | gw->gfn, pfn, false, true); |
e7a04c99 AK |
324 | break; |
325 | } | |
6aa8b732 | 326 | |
e7a04c99 AK |
327 | if (is_shadow_present_pte(*sptep) && !is_large_pte(*sptep)) |
328 | continue; | |
abb9e0b8 | 329 | |
e7a04c99 | 330 | if (is_large_pte(*sptep)) { |
c5bc2242 | 331 | rmap_remove(vcpu->kvm, sptep); |
d555c333 | 332 | __set_spte(sptep, shadow_trap_nonpresent_pte); |
e7a04c99 | 333 | kvm_flush_remote_tlbs(vcpu->kvm); |
7819026e | 334 | } |
ef0197e8 | 335 | |
7e4e4056 JR |
336 | if (level <= gw->level) { |
337 | int delta = level - gw->level + 1; | |
f6e2c02b | 338 | direct = 1; |
7e4e4056 | 339 | if (!is_dirty_gpte(gw->ptes[level - delta])) |
e7a04c99 | 340 | access &= ~ACC_WRITE_MASK; |
7e4e4056 JR |
341 | table_gfn = gpte_to_gfn(gw->ptes[level - delta]); |
342 | /* advance table_gfn when emulating 1gb pages with 4k */ | |
343 | if (delta == 0) | |
344 | table_gfn += PT_INDEX(addr, level); | |
6aa0b9de | 345 | access &= gw->pte_access; |
e7a04c99 | 346 | } else { |
f6e2c02b | 347 | direct = 0; |
e7a04c99 AK |
348 | table_gfn = gw->table_gfn[level - 2]; |
349 | } | |
350 | shadow_page = kvm_mmu_get_page(vcpu, table_gfn, addr, level-1, | |
f6e2c02b AK |
351 | direct, access, sptep); |
352 | if (!direct) { | |
e7a04c99 AK |
353 | r = kvm_read_guest_atomic(vcpu->kvm, |
354 | gw->pte_gpa[level - 2], | |
355 | &curr_pte, sizeof(curr_pte)); | |
356 | if (r || curr_pte != gw->ptes[level - 2]) { | |
357 | kvm_mmu_put_page(shadow_page, sptep); | |
358 | kvm_release_pfn_clean(pfn); | |
359 | sptep = NULL; | |
360 | break; | |
361 | } | |
362 | } | |
abb9e0b8 | 363 | |
e7a04c99 AK |
364 | spte = __pa(shadow_page->spt) |
365 | | PT_PRESENT_MASK | PT_ACCESSED_MASK | |
366 | | PT_WRITABLE_MASK | PT_USER_MASK; | |
367 | *sptep = spte; | |
368 | } | |
050e6499 | 369 | |
e7a04c99 | 370 | return sptep; |
6aa8b732 AK |
371 | } |
372 | ||
6aa8b732 AK |
373 | /* |
374 | * Page fault handler. There are several causes for a page fault: | |
375 | * - there is no shadow pte for the guest pte | |
376 | * - write access through a shadow pte marked read only so that we can set | |
377 | * the dirty bit | |
378 | * - write access to a shadow pte marked read only so we can update the page | |
379 | * dirty bitmap, when userspace requests it | |
380 | * - mmio access; in this case we will never install a present shadow pte | |
381 | * - normal guest page fault due to the guest pte marked not present, not | |
382 | * writable, or not executable | |
383 | * | |
e2dec939 AK |
384 | * Returns: 1 if we need to emulate the instruction, 0 otherwise, or |
385 | * a negative value on error. | |
6aa8b732 AK |
386 | */ |
387 | static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr, | |
388 | u32 error_code) | |
389 | { | |
390 | int write_fault = error_code & PFERR_WRITE_MASK; | |
6aa8b732 | 391 | int user_fault = error_code & PFERR_USER_MASK; |
73b1087e | 392 | int fetch_fault = error_code & PFERR_FETCH_MASK; |
6aa8b732 | 393 | struct guest_walker walker; |
d555c333 | 394 | u64 *sptep; |
cea0f0e7 | 395 | int write_pt = 0; |
e2dec939 | 396 | int r; |
35149e21 | 397 | pfn_t pfn; |
7e4e4056 | 398 | int level = PT_PAGE_TABLE_LEVEL; |
e930bffe | 399 | unsigned long mmu_seq; |
6aa8b732 | 400 | |
b8688d51 | 401 | pgprintk("%s: addr %lx err %x\n", __func__, addr, error_code); |
37a7d8b0 | 402 | kvm_mmu_audit(vcpu, "pre page fault"); |
714b93da | 403 | |
e2dec939 AK |
404 | r = mmu_topup_memory_caches(vcpu); |
405 | if (r) | |
406 | return r; | |
714b93da | 407 | |
6aa8b732 | 408 | /* |
a8b876b1 | 409 | * Look up the guest pte for the faulting address. |
6aa8b732 | 410 | */ |
73b1087e AK |
411 | r = FNAME(walk_addr)(&walker, vcpu, addr, write_fault, user_fault, |
412 | fetch_fault); | |
6aa8b732 AK |
413 | |
414 | /* | |
415 | * The page is not mapped by the guest. Let the guest handle it. | |
416 | */ | |
7993ba43 | 417 | if (!r) { |
b8688d51 | 418 | pgprintk("%s: guest page fault\n", __func__); |
7993ba43 | 419 | inject_page_fault(vcpu, addr, walker.error_code); |
ad312c7c | 420 | vcpu->arch.last_pt_write_count = 0; /* reset fork detector */ |
6aa8b732 AK |
421 | return 0; |
422 | } | |
423 | ||
7e4e4056 JR |
424 | if (walker.level >= PT_DIRECTORY_LEVEL) { |
425 | level = min(walker.level, mapping_level(vcpu, walker.gfn)); | |
426 | walker.gfn = walker.gfn & ~(KVM_PAGES_PER_HPAGE(level) - 1); | |
05da4558 | 427 | } |
7e4e4056 | 428 | |
e930bffe | 429 | mmu_seq = vcpu->kvm->mmu_notifier_seq; |
4c2155ce | 430 | smp_rmb(); |
35149e21 | 431 | pfn = gfn_to_pfn(vcpu->kvm, walker.gfn); |
d7824fff | 432 | |
d196e343 | 433 | /* mmio */ |
bf998156 HY |
434 | if (is_error_pfn(pfn)) |
435 | return kvm_handle_bad_page(vcpu->kvm, walker.gfn, pfn); | |
d196e343 | 436 | |
aaee2c94 | 437 | spin_lock(&vcpu->kvm->mmu_lock); |
e930bffe AA |
438 | if (mmu_notifier_retry(vcpu, mmu_seq)) |
439 | goto out_unlock; | |
eb787d10 | 440 | kvm_mmu_free_some_pages(vcpu); |
d555c333 | 441 | sptep = FNAME(fetch)(vcpu, addr, &walker, user_fault, write_fault, |
7e4e4056 | 442 | level, &write_pt, pfn); |
b8688d51 | 443 | pgprintk("%s: shadow pte %p %llx ptwrite %d\n", __func__, |
d555c333 | 444 | sptep, *sptep, write_pt); |
cea0f0e7 | 445 | |
a25f7e1f | 446 | if (!write_pt) |
ad312c7c | 447 | vcpu->arch.last_pt_write_count = 0; /* reset fork detector */ |
a25f7e1f | 448 | |
1165f5fe | 449 | ++vcpu->stat.pf_fixed; |
37a7d8b0 | 450 | kvm_mmu_audit(vcpu, "post page fault (fixed)"); |
aaee2c94 | 451 | spin_unlock(&vcpu->kvm->mmu_lock); |
6aa8b732 | 452 | |
cea0f0e7 | 453 | return write_pt; |
e930bffe AA |
454 | |
455 | out_unlock: | |
456 | spin_unlock(&vcpu->kvm->mmu_lock); | |
457 | kvm_release_pfn_clean(pfn); | |
458 | return 0; | |
6aa8b732 AK |
459 | } |
460 | ||
a461930b | 461 | static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva) |
a7052897 | 462 | { |
a461930b | 463 | struct kvm_shadow_walk_iterator iterator; |
08e850c6 | 464 | gpa_t pte_gpa = -1; |
a461930b AK |
465 | int level; |
466 | u64 *sptep; | |
4539b358 | 467 | int need_flush = 0; |
a461930b AK |
468 | |
469 | spin_lock(&vcpu->kvm->mmu_lock); | |
a7052897 | 470 | |
a461930b AK |
471 | for_each_shadow_entry(vcpu, gva, iterator) { |
472 | level = iterator.level; | |
473 | sptep = iterator.sptep; | |
ad218f85 | 474 | |
884a0ff0 | 475 | if (is_last_spte(*sptep, level)) { |
08e850c6 | 476 | struct kvm_mmu_page *sp = page_header(__pa(sptep)); |
22c9b2d1 | 477 | int offset, shift; |
08e850c6 | 478 | |
22c9b2d1 XG |
479 | shift = PAGE_SHIFT - |
480 | (PT_LEVEL_BITS - PT64_LEVEL_BITS) * level; | |
481 | offset = sp->role.quadrant << shift; | |
482 | ||
483 | pte_gpa = (sp->gfn << PAGE_SHIFT) + offset; | |
08e850c6 | 484 | pte_gpa += (sptep - sp->spt) * sizeof(pt_element_t); |
a461930b AK |
485 | |
486 | if (is_shadow_present_pte(*sptep)) { | |
487 | rmap_remove(vcpu->kvm, sptep); | |
488 | if (is_large_pte(*sptep)) | |
489 | --vcpu->kvm->stat.lpages; | |
4539b358 | 490 | need_flush = 1; |
a461930b | 491 | } |
d555c333 | 492 | __set_spte(sptep, shadow_trap_nonpresent_pte); |
a461930b | 493 | break; |
87917239 | 494 | } |
a7052897 | 495 | |
a461930b AK |
496 | if (!is_shadow_present_pte(*sptep)) |
497 | break; | |
498 | } | |
a7052897 | 499 | |
4539b358 AA |
500 | if (need_flush) |
501 | kvm_flush_remote_tlbs(vcpu->kvm); | |
08e850c6 AK |
502 | |
503 | atomic_inc(&vcpu->kvm->arch.invlpg_counter); | |
504 | ||
ad218f85 | 505 | spin_unlock(&vcpu->kvm->mmu_lock); |
08e850c6 AK |
506 | |
507 | if (pte_gpa == -1) | |
508 | return; | |
509 | ||
510 | if (mmu_topup_memory_caches(vcpu)) | |
511 | return; | |
512 | kvm_mmu_pte_write(vcpu, pte_gpa, NULL, sizeof(pt_element_t), 0); | |
a7052897 MT |
513 | } |
514 | ||
1871c602 GN |
515 | static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t vaddr, u32 access, |
516 | u32 *error) | |
6aa8b732 AK |
517 | { |
518 | struct guest_walker walker; | |
e119d117 AK |
519 | gpa_t gpa = UNMAPPED_GVA; |
520 | int r; | |
6aa8b732 | 521 | |
1871c602 GN |
522 | r = FNAME(walk_addr)(&walker, vcpu, vaddr, |
523 | !!(access & PFERR_WRITE_MASK), | |
524 | !!(access & PFERR_USER_MASK), | |
525 | !!(access & PFERR_FETCH_MASK)); | |
6aa8b732 | 526 | |
e119d117 | 527 | if (r) { |
1755fbcc | 528 | gpa = gfn_to_gpa(walker.gfn); |
e119d117 | 529 | gpa |= vaddr & ~PAGE_MASK; |
1871c602 GN |
530 | } else if (error) |
531 | *error = walker.error_code; | |
6aa8b732 AK |
532 | |
533 | return gpa; | |
534 | } | |
535 | ||
c7addb90 AK |
536 | static void FNAME(prefetch_page)(struct kvm_vcpu *vcpu, |
537 | struct kvm_mmu_page *sp) | |
538 | { | |
eab9f71f AK |
539 | int i, j, offset, r; |
540 | pt_element_t pt[256 / sizeof(pt_element_t)]; | |
541 | gpa_t pte_gpa; | |
c7addb90 | 542 | |
f6e2c02b | 543 | if (sp->role.direct |
e5a4c8ca | 544 | || (PTTYPE == 32 && sp->role.level > PT_PAGE_TABLE_LEVEL)) { |
c7addb90 AK |
545 | nonpaging_prefetch_page(vcpu, sp); |
546 | return; | |
547 | } | |
548 | ||
eab9f71f AK |
549 | pte_gpa = gfn_to_gpa(sp->gfn); |
550 | if (PTTYPE == 32) { | |
e5a4c8ca | 551 | offset = sp->role.quadrant << PT64_LEVEL_BITS; |
eab9f71f AK |
552 | pte_gpa += offset * sizeof(pt_element_t); |
553 | } | |
7ec54588 | 554 | |
eab9f71f AK |
555 | for (i = 0; i < PT64_ENT_PER_PAGE; i += ARRAY_SIZE(pt)) { |
556 | r = kvm_read_guest_atomic(vcpu->kvm, pte_gpa, pt, sizeof pt); | |
557 | pte_gpa += ARRAY_SIZE(pt) * sizeof(pt_element_t); | |
558 | for (j = 0; j < ARRAY_SIZE(pt); ++j) | |
43a3795a | 559 | if (r || is_present_gpte(pt[j])) |
eab9f71f AK |
560 | sp->spt[i+j] = shadow_trap_nonpresent_pte; |
561 | else | |
562 | sp->spt[i+j] = shadow_notrap_nonpresent_pte; | |
7ec54588 | 563 | } |
c7addb90 AK |
564 | } |
565 | ||
e8bc217a MT |
566 | /* |
567 | * Using the cached information from sp->gfns is safe because: | |
568 | * - The spte has a reference to the struct page, so the pfn for a given gfn | |
569 | * can't change unless all sptes pointing to it are nuked first. | |
570 | * - Alias changes zap the entire shadow cache. | |
571 | */ | |
572 | static int FNAME(sync_page)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp) | |
573 | { | |
574 | int i, offset, nr_present; | |
1403283a | 575 | bool reset_host_protection; |
51fb60d8 | 576 | gpa_t first_pte_gpa; |
e8bc217a MT |
577 | |
578 | offset = nr_present = 0; | |
579 | ||
580 | if (PTTYPE == 32) | |
581 | offset = sp->role.quadrant << PT64_LEVEL_BITS; | |
582 | ||
51fb60d8 GJ |
583 | first_pte_gpa = gfn_to_gpa(sp->gfn) + offset * sizeof(pt_element_t); |
584 | ||
e8bc217a MT |
585 | for (i = 0; i < PT64_ENT_PER_PAGE; i++) { |
586 | unsigned pte_access; | |
587 | pt_element_t gpte; | |
588 | gpa_t pte_gpa; | |
589 | gfn_t gfn = sp->gfns[i]; | |
590 | ||
591 | if (!is_shadow_present_pte(sp->spt[i])) | |
592 | continue; | |
593 | ||
51fb60d8 | 594 | pte_gpa = first_pte_gpa + i * sizeof(pt_element_t); |
e8bc217a MT |
595 | |
596 | if (kvm_read_guest_atomic(vcpu->kvm, pte_gpa, &gpte, | |
597 | sizeof(pt_element_t))) | |
598 | return -EINVAL; | |
599 | ||
43a3795a | 600 | if (gpte_to_gfn(gpte) != gfn || !is_present_gpte(gpte) || |
e8bc217a MT |
601 | !(gpte & PT_ACCESSED_MASK)) { |
602 | u64 nonpresent; | |
603 | ||
604 | rmap_remove(vcpu->kvm, &sp->spt[i]); | |
43a3795a | 605 | if (is_present_gpte(gpte)) |
e8bc217a MT |
606 | nonpresent = shadow_trap_nonpresent_pte; |
607 | else | |
608 | nonpresent = shadow_notrap_nonpresent_pte; | |
d555c333 | 609 | __set_spte(&sp->spt[i], nonpresent); |
e8bc217a MT |
610 | continue; |
611 | } | |
612 | ||
613 | nr_present++; | |
614 | pte_access = sp->role.access & FNAME(gpte_access)(vcpu, gpte); | |
1403283a IE |
615 | if (!(sp->spt[i] & SPTE_HOST_WRITEABLE)) { |
616 | pte_access &= ~ACC_WRITE_MASK; | |
617 | reset_host_protection = 0; | |
618 | } else { | |
619 | reset_host_protection = 1; | |
620 | } | |
e8bc217a | 621 | set_spte(vcpu, &sp->spt[i], pte_access, 0, 0, |
7e4e4056 | 622 | is_dirty_gpte(gpte), PT_PAGE_TABLE_LEVEL, gfn, |
1403283a IE |
623 | spte_to_pfn(sp->spt[i]), true, false, |
624 | reset_host_protection); | |
e8bc217a MT |
625 | } |
626 | ||
627 | return !nr_present; | |
628 | } | |
629 | ||
6aa8b732 AK |
630 | #undef pt_element_t |
631 | #undef guest_walker | |
632 | #undef FNAME | |
633 | #undef PT_BASE_ADDR_MASK | |
634 | #undef PT_INDEX | |
6aa8b732 | 635 | #undef PT_LEVEL_MASK |
e04da980 JR |
636 | #undef PT_LVL_ADDR_MASK |
637 | #undef PT_LVL_OFFSET_MASK | |
c7addb90 | 638 | #undef PT_LEVEL_BITS |
cea0f0e7 | 639 | #undef PT_MAX_FULL_LEVELS |
5fb07ddb | 640 | #undef gpte_to_gfn |
e04da980 | 641 | #undef gpte_to_gfn_lvl |
b3e4e63f | 642 | #undef CMPXCHG |