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 | |
ec8d4eae | 153 | kvm_read_guest(vcpu->kvm, pte_gpa, &pte, sizeof(pte)); |
07420171 | 154 | trace_kvm_mmu_paging_element(pte, walker->level); |
42bf3f0a | 155 | |
43a3795a | 156 | if (!is_present_gpte(pte)) |
7993ba43 AK |
157 | goto not_present; |
158 | ||
82725b20 DE |
159 | rsvd_fault = is_rsvd_bits_set(vcpu, pte, walker->level); |
160 | if (rsvd_fault) | |
161 | goto access_error; | |
162 | ||
42bf3f0a | 163 | if (write_fault && !is_writeble_pte(pte)) |
7993ba43 AK |
164 | if (user_fault || is_write_protection(vcpu)) |
165 | goto access_error; | |
166 | ||
42bf3f0a | 167 | if (user_fault && !(pte & PT_USER_MASK)) |
7993ba43 AK |
168 | goto access_error; |
169 | ||
73b1087e | 170 | #if PTTYPE == 64 |
42bf3f0a | 171 | if (fetch_fault && is_nx(vcpu) && (pte & PT64_NX_MASK)) |
73b1087e AK |
172 | goto access_error; |
173 | #endif | |
174 | ||
42bf3f0a | 175 | if (!(pte & PT_ACCESSED_MASK)) { |
07420171 AK |
176 | trace_kvm_mmu_set_accessed_bit(table_gfn, index, |
177 | sizeof(pte)); | |
bf3f8e86 | 178 | mark_page_dirty(vcpu->kvm, table_gfn); |
b3e4e63f MT |
179 | if (FNAME(cmpxchg_gpte)(vcpu->kvm, table_gfn, |
180 | index, pte, pte|PT_ACCESSED_MASK)) | |
181 | goto walk; | |
42bf3f0a | 182 | pte |= PT_ACCESSED_MASK; |
bf3f8e86 | 183 | } |
815af8d4 | 184 | |
bedbe4ee | 185 | pte_access = pt_access & FNAME(gpte_access)(vcpu, pte); |
fe135d2c | 186 | |
7819026e MT |
187 | walker->ptes[walker->level - 1] = pte; |
188 | ||
e04da980 JR |
189 | if ((walker->level == PT_PAGE_TABLE_LEVEL) || |
190 | ((walker->level == PT_DIRECTORY_LEVEL) && | |
191 | (pte & PT_PAGE_SIZE_MASK) && | |
192 | (PTTYPE == 64 || is_pse(vcpu))) || | |
193 | ((walker->level == PT_PDPE_LEVEL) && | |
194 | (pte & PT_PAGE_SIZE_MASK) && | |
195 | is_long_mode(vcpu))) { | |
196 | int lvl = walker->level; | |
197 | ||
198 | walker->gfn = gpte_to_gfn_lvl(pte, lvl); | |
199 | walker->gfn += (addr & PT_LVL_OFFSET_MASK(lvl)) | |
200 | >> PAGE_SHIFT; | |
201 | ||
202 | if (PTTYPE == 32 && | |
203 | walker->level == PT_DIRECTORY_LEVEL && | |
204 | is_cpuid_PSE36()) | |
da928521 | 205 | walker->gfn += pse36_gfn_delta(pte); |
e04da980 | 206 | |
ac79c978 | 207 | break; |
815af8d4 | 208 | } |
ac79c978 | 209 | |
fe135d2c | 210 | pt_access = pte_access; |
ac79c978 AK |
211 | --walker->level; |
212 | } | |
42bf3f0a | 213 | |
43a3795a | 214 | if (write_fault && !is_dirty_gpte(pte)) { |
b3e4e63f MT |
215 | bool ret; |
216 | ||
07420171 | 217 | trace_kvm_mmu_set_dirty_bit(table_gfn, index, sizeof(pte)); |
42bf3f0a | 218 | mark_page_dirty(vcpu->kvm, table_gfn); |
b3e4e63f MT |
219 | ret = FNAME(cmpxchg_gpte)(vcpu->kvm, table_gfn, index, pte, |
220 | pte|PT_DIRTY_MASK); | |
221 | if (ret) | |
222 | goto walk; | |
42bf3f0a | 223 | pte |= PT_DIRTY_MASK; |
7819026e | 224 | walker->ptes[walker->level - 1] = pte; |
42bf3f0a AK |
225 | } |
226 | ||
fe135d2c AK |
227 | walker->pt_access = pt_access; |
228 | walker->pte_access = pte_access; | |
229 | pgprintk("%s: pte %llx pte_access %x pt_access %x\n", | |
b8688d51 | 230 | __func__, (u64)pte, pt_access, pte_access); |
7993ba43 AK |
231 | return 1; |
232 | ||
233 | not_present: | |
234 | walker->error_code = 0; | |
235 | goto err; | |
236 | ||
237 | access_error: | |
238 | walker->error_code = PFERR_PRESENT_MASK; | |
239 | ||
240 | err: | |
241 | if (write_fault) | |
242 | walker->error_code |= PFERR_WRITE_MASK; | |
243 | if (user_fault) | |
244 | walker->error_code |= PFERR_USER_MASK; | |
73b1087e AK |
245 | if (fetch_fault) |
246 | walker->error_code |= PFERR_FETCH_MASK; | |
82725b20 DE |
247 | if (rsvd_fault) |
248 | walker->error_code |= PFERR_RSVD_MASK; | |
07420171 | 249 | trace_kvm_mmu_walker_error(walker->error_code); |
fe551881 | 250 | return 0; |
6aa8b732 AK |
251 | } |
252 | ||
0028425f | 253 | static void FNAME(update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *page, |
489f1d65 | 254 | u64 *spte, const void *pte) |
0028425f AK |
255 | { |
256 | pt_element_t gpte; | |
41074d07 | 257 | unsigned pte_access; |
35149e21 | 258 | pfn_t pfn; |
0028425f | 259 | |
0028425f | 260 | gpte = *(const pt_element_t *)pte; |
c7addb90 | 261 | if (~gpte & (PT_PRESENT_MASK | PT_ACCESSED_MASK)) { |
43a3795a | 262 | if (!is_present_gpte(gpte)) |
d555c333 | 263 | __set_spte(spte, shadow_notrap_nonpresent_pte); |
c7addb90 AK |
264 | return; |
265 | } | |
b8688d51 | 266 | pgprintk("%s: gpte %llx spte %p\n", __func__, (u64)gpte, spte); |
41074d07 | 267 | pte_access = page->role.access & FNAME(gpte_access)(vcpu, gpte); |
d7824fff AK |
268 | if (gpte_to_gfn(gpte) != vcpu->arch.update_pte.gfn) |
269 | return; | |
35149e21 AL |
270 | pfn = vcpu->arch.update_pte.pfn; |
271 | if (is_error_pfn(pfn)) | |
d7824fff | 272 | return; |
e930bffe AA |
273 | if (mmu_notifier_retry(vcpu, vcpu->arch.update_pte.mmu_seq)) |
274 | return; | |
35149e21 | 275 | kvm_get_pfn(pfn); |
1c4f1fd6 | 276 | mmu_set_spte(vcpu, spte, page->role.access, pte_access, 0, 0, |
7e4e4056 | 277 | gpte & PT_DIRTY_MASK, NULL, PT_PAGE_TABLE_LEVEL, |
c2d0ee46 | 278 | gpte_to_gfn(gpte), pfn, true); |
0028425f AK |
279 | } |
280 | ||
6aa8b732 AK |
281 | /* |
282 | * Fetch a shadow pte for a specific level in the paging hierarchy. | |
283 | */ | |
e7a04c99 AK |
284 | static u64 *FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr, |
285 | struct guest_walker *gw, | |
7e4e4056 | 286 | int user_fault, int write_fault, int hlevel, |
e7a04c99 | 287 | int *ptwrite, pfn_t pfn) |
6aa8b732 | 288 | { |
abb9e0b8 AK |
289 | unsigned access = gw->pt_access; |
290 | struct kvm_mmu_page *shadow_page; | |
bde89223 | 291 | u64 spte, *sptep = NULL; |
f6e2c02b | 292 | int direct; |
abb9e0b8 AK |
293 | gfn_t table_gfn; |
294 | int r; | |
e7a04c99 | 295 | int level; |
abb9e0b8 | 296 | pt_element_t curr_pte; |
e7a04c99 | 297 | struct kvm_shadow_walk_iterator iterator; |
abb9e0b8 | 298 | |
43a3795a | 299 | if (!is_present_gpte(gw->ptes[gw->level - 1])) |
e7a04c99 | 300 | return NULL; |
6aa8b732 | 301 | |
e7a04c99 AK |
302 | for_each_shadow_entry(vcpu, addr, iterator) { |
303 | level = iterator.level; | |
304 | sptep = iterator.sptep; | |
7e4e4056 | 305 | if (iterator.level == hlevel) { |
e7a04c99 AK |
306 | mmu_set_spte(vcpu, sptep, access, |
307 | gw->pte_access & access, | |
308 | user_fault, write_fault, | |
309 | gw->ptes[gw->level-1] & PT_DIRTY_MASK, | |
852e3c19 | 310 | ptwrite, level, |
e7a04c99 AK |
311 | gw->gfn, pfn, false); |
312 | break; | |
313 | } | |
6aa8b732 | 314 | |
e7a04c99 AK |
315 | if (is_shadow_present_pte(*sptep) && !is_large_pte(*sptep)) |
316 | continue; | |
abb9e0b8 | 317 | |
e7a04c99 | 318 | if (is_large_pte(*sptep)) { |
c5bc2242 | 319 | rmap_remove(vcpu->kvm, sptep); |
d555c333 | 320 | __set_spte(sptep, shadow_trap_nonpresent_pte); |
e7a04c99 | 321 | kvm_flush_remote_tlbs(vcpu->kvm); |
7819026e | 322 | } |
ef0197e8 | 323 | |
7e4e4056 JR |
324 | if (level <= gw->level) { |
325 | int delta = level - gw->level + 1; | |
f6e2c02b | 326 | direct = 1; |
7e4e4056 | 327 | if (!is_dirty_gpte(gw->ptes[level - delta])) |
e7a04c99 | 328 | access &= ~ACC_WRITE_MASK; |
7e4e4056 JR |
329 | table_gfn = gpte_to_gfn(gw->ptes[level - delta]); |
330 | /* advance table_gfn when emulating 1gb pages with 4k */ | |
331 | if (delta == 0) | |
332 | table_gfn += PT_INDEX(addr, level); | |
e7a04c99 | 333 | } else { |
f6e2c02b | 334 | direct = 0; |
e7a04c99 AK |
335 | table_gfn = gw->table_gfn[level - 2]; |
336 | } | |
337 | shadow_page = kvm_mmu_get_page(vcpu, table_gfn, addr, level-1, | |
f6e2c02b AK |
338 | direct, access, sptep); |
339 | if (!direct) { | |
e7a04c99 AK |
340 | r = kvm_read_guest_atomic(vcpu->kvm, |
341 | gw->pte_gpa[level - 2], | |
342 | &curr_pte, sizeof(curr_pte)); | |
343 | if (r || curr_pte != gw->ptes[level - 2]) { | |
344 | kvm_mmu_put_page(shadow_page, sptep); | |
345 | kvm_release_pfn_clean(pfn); | |
346 | sptep = NULL; | |
347 | break; | |
348 | } | |
349 | } | |
abb9e0b8 | 350 | |
e7a04c99 AK |
351 | spte = __pa(shadow_page->spt) |
352 | | PT_PRESENT_MASK | PT_ACCESSED_MASK | |
353 | | PT_WRITABLE_MASK | PT_USER_MASK; | |
354 | *sptep = spte; | |
355 | } | |
050e6499 | 356 | |
e7a04c99 | 357 | return sptep; |
6aa8b732 AK |
358 | } |
359 | ||
6aa8b732 AK |
360 | /* |
361 | * Page fault handler. There are several causes for a page fault: | |
362 | * - there is no shadow pte for the guest pte | |
363 | * - write access through a shadow pte marked read only so that we can set | |
364 | * the dirty bit | |
365 | * - write access to a shadow pte marked read only so we can update the page | |
366 | * dirty bitmap, when userspace requests it | |
367 | * - mmio access; in this case we will never install a present shadow pte | |
368 | * - normal guest page fault due to the guest pte marked not present, not | |
369 | * writable, or not executable | |
370 | * | |
e2dec939 AK |
371 | * Returns: 1 if we need to emulate the instruction, 0 otherwise, or |
372 | * a negative value on error. | |
6aa8b732 AK |
373 | */ |
374 | static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr, | |
375 | u32 error_code) | |
376 | { | |
377 | int write_fault = error_code & PFERR_WRITE_MASK; | |
6aa8b732 | 378 | int user_fault = error_code & PFERR_USER_MASK; |
73b1087e | 379 | int fetch_fault = error_code & PFERR_FETCH_MASK; |
6aa8b732 | 380 | struct guest_walker walker; |
d555c333 | 381 | u64 *sptep; |
cea0f0e7 | 382 | int write_pt = 0; |
e2dec939 | 383 | int r; |
35149e21 | 384 | pfn_t pfn; |
7e4e4056 | 385 | int level = PT_PAGE_TABLE_LEVEL; |
e930bffe | 386 | unsigned long mmu_seq; |
6aa8b732 | 387 | |
b8688d51 | 388 | pgprintk("%s: addr %lx err %x\n", __func__, addr, error_code); |
37a7d8b0 | 389 | kvm_mmu_audit(vcpu, "pre page fault"); |
714b93da | 390 | |
e2dec939 AK |
391 | r = mmu_topup_memory_caches(vcpu); |
392 | if (r) | |
393 | return r; | |
714b93da | 394 | |
6aa8b732 | 395 | /* |
a8b876b1 | 396 | * Look up the guest pte for the faulting address. |
6aa8b732 | 397 | */ |
73b1087e AK |
398 | r = FNAME(walk_addr)(&walker, vcpu, addr, write_fault, user_fault, |
399 | fetch_fault); | |
6aa8b732 AK |
400 | |
401 | /* | |
402 | * The page is not mapped by the guest. Let the guest handle it. | |
403 | */ | |
7993ba43 | 404 | if (!r) { |
b8688d51 | 405 | pgprintk("%s: guest page fault\n", __func__); |
7993ba43 | 406 | inject_page_fault(vcpu, addr, walker.error_code); |
ad312c7c | 407 | vcpu->arch.last_pt_write_count = 0; /* reset fork detector */ |
6aa8b732 AK |
408 | return 0; |
409 | } | |
410 | ||
7e4e4056 JR |
411 | if (walker.level >= PT_DIRECTORY_LEVEL) { |
412 | level = min(walker.level, mapping_level(vcpu, walker.gfn)); | |
413 | walker.gfn = walker.gfn & ~(KVM_PAGES_PER_HPAGE(level) - 1); | |
05da4558 | 414 | } |
7e4e4056 | 415 | |
e930bffe | 416 | mmu_seq = vcpu->kvm->mmu_notifier_seq; |
4c2155ce | 417 | smp_rmb(); |
35149e21 | 418 | pfn = gfn_to_pfn(vcpu->kvm, walker.gfn); |
d7824fff | 419 | |
d196e343 | 420 | /* mmio */ |
35149e21 | 421 | if (is_error_pfn(pfn)) { |
ebb0e626 | 422 | pgprintk("gfn %lx is mmio\n", walker.gfn); |
35149e21 | 423 | kvm_release_pfn_clean(pfn); |
d196e343 AK |
424 | return 1; |
425 | } | |
426 | ||
aaee2c94 | 427 | spin_lock(&vcpu->kvm->mmu_lock); |
e930bffe AA |
428 | if (mmu_notifier_retry(vcpu, mmu_seq)) |
429 | goto out_unlock; | |
eb787d10 | 430 | kvm_mmu_free_some_pages(vcpu); |
d555c333 | 431 | sptep = FNAME(fetch)(vcpu, addr, &walker, user_fault, write_fault, |
7e4e4056 | 432 | level, &write_pt, pfn); |
b8688d51 | 433 | pgprintk("%s: shadow pte %p %llx ptwrite %d\n", __func__, |
d555c333 | 434 | sptep, *sptep, write_pt); |
cea0f0e7 | 435 | |
a25f7e1f | 436 | if (!write_pt) |
ad312c7c | 437 | vcpu->arch.last_pt_write_count = 0; /* reset fork detector */ |
a25f7e1f | 438 | |
1165f5fe | 439 | ++vcpu->stat.pf_fixed; |
37a7d8b0 | 440 | kvm_mmu_audit(vcpu, "post page fault (fixed)"); |
aaee2c94 | 441 | spin_unlock(&vcpu->kvm->mmu_lock); |
6aa8b732 | 442 | |
cea0f0e7 | 443 | return write_pt; |
e930bffe AA |
444 | |
445 | out_unlock: | |
446 | spin_unlock(&vcpu->kvm->mmu_lock); | |
447 | kvm_release_pfn_clean(pfn); | |
448 | return 0; | |
6aa8b732 AK |
449 | } |
450 | ||
a461930b | 451 | static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva) |
a7052897 | 452 | { |
a461930b AK |
453 | struct kvm_shadow_walk_iterator iterator; |
454 | pt_element_t gpte; | |
455 | gpa_t pte_gpa = -1; | |
456 | int level; | |
457 | u64 *sptep; | |
4539b358 | 458 | int need_flush = 0; |
a461930b AK |
459 | |
460 | spin_lock(&vcpu->kvm->mmu_lock); | |
a7052897 | 461 | |
a461930b AK |
462 | for_each_shadow_entry(vcpu, gva, iterator) { |
463 | level = iterator.level; | |
464 | sptep = iterator.sptep; | |
ad218f85 | 465 | |
a461930b | 466 | /* FIXME: properly handle invlpg on large guest pages */ |
7e4e4056 JR |
467 | if (level == PT_PAGE_TABLE_LEVEL || |
468 | ((level == PT_DIRECTORY_LEVEL && is_large_pte(*sptep))) || | |
469 | ((level == PT_PDPE_LEVEL && is_large_pte(*sptep)))) { | |
a461930b | 470 | struct kvm_mmu_page *sp = page_header(__pa(sptep)); |
ad218f85 | 471 | |
a461930b AK |
472 | pte_gpa = (sp->gfn << PAGE_SHIFT); |
473 | pte_gpa += (sptep - sp->spt) * sizeof(pt_element_t); | |
474 | ||
475 | if (is_shadow_present_pte(*sptep)) { | |
476 | rmap_remove(vcpu->kvm, sptep); | |
477 | if (is_large_pte(*sptep)) | |
478 | --vcpu->kvm->stat.lpages; | |
4539b358 | 479 | need_flush = 1; |
a461930b | 480 | } |
d555c333 | 481 | __set_spte(sptep, shadow_trap_nonpresent_pte); |
a461930b | 482 | break; |
87917239 | 483 | } |
a7052897 | 484 | |
a461930b AK |
485 | if (!is_shadow_present_pte(*sptep)) |
486 | break; | |
487 | } | |
a7052897 | 488 | |
4539b358 AA |
489 | if (need_flush) |
490 | kvm_flush_remote_tlbs(vcpu->kvm); | |
ad218f85 | 491 | spin_unlock(&vcpu->kvm->mmu_lock); |
a461930b AK |
492 | |
493 | if (pte_gpa == -1) | |
ad218f85 | 494 | return; |
a461930b | 495 | if (kvm_read_guest_atomic(vcpu->kvm, pte_gpa, &gpte, |
ad218f85 MT |
496 | sizeof(pt_element_t))) |
497 | return; | |
43a3795a | 498 | if (is_present_gpte(gpte) && (gpte & PT_ACCESSED_MASK)) { |
ad218f85 MT |
499 | if (mmu_topup_memory_caches(vcpu)) |
500 | return; | |
a461930b | 501 | kvm_mmu_pte_write(vcpu, pte_gpa, (const u8 *)&gpte, |
ad218f85 MT |
502 | sizeof(pt_element_t), 0); |
503 | } | |
a7052897 MT |
504 | } |
505 | ||
6aa8b732 AK |
506 | static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t vaddr) |
507 | { | |
508 | struct guest_walker walker; | |
e119d117 AK |
509 | gpa_t gpa = UNMAPPED_GVA; |
510 | int r; | |
6aa8b732 | 511 | |
e119d117 | 512 | r = FNAME(walk_addr)(&walker, vcpu, vaddr, 0, 0, 0); |
6aa8b732 | 513 | |
e119d117 | 514 | if (r) { |
1755fbcc | 515 | gpa = gfn_to_gpa(walker.gfn); |
e119d117 | 516 | gpa |= vaddr & ~PAGE_MASK; |
6aa8b732 AK |
517 | } |
518 | ||
519 | return gpa; | |
520 | } | |
521 | ||
c7addb90 AK |
522 | static void FNAME(prefetch_page)(struct kvm_vcpu *vcpu, |
523 | struct kvm_mmu_page *sp) | |
524 | { | |
eab9f71f AK |
525 | int i, j, offset, r; |
526 | pt_element_t pt[256 / sizeof(pt_element_t)]; | |
527 | gpa_t pte_gpa; | |
c7addb90 | 528 | |
f6e2c02b | 529 | if (sp->role.direct |
e5a4c8ca | 530 | || (PTTYPE == 32 && sp->role.level > PT_PAGE_TABLE_LEVEL)) { |
c7addb90 AK |
531 | nonpaging_prefetch_page(vcpu, sp); |
532 | return; | |
533 | } | |
534 | ||
eab9f71f AK |
535 | pte_gpa = gfn_to_gpa(sp->gfn); |
536 | if (PTTYPE == 32) { | |
e5a4c8ca | 537 | offset = sp->role.quadrant << PT64_LEVEL_BITS; |
eab9f71f AK |
538 | pte_gpa += offset * sizeof(pt_element_t); |
539 | } | |
7ec54588 | 540 | |
eab9f71f AK |
541 | for (i = 0; i < PT64_ENT_PER_PAGE; i += ARRAY_SIZE(pt)) { |
542 | r = kvm_read_guest_atomic(vcpu->kvm, pte_gpa, pt, sizeof pt); | |
543 | pte_gpa += ARRAY_SIZE(pt) * sizeof(pt_element_t); | |
544 | for (j = 0; j < ARRAY_SIZE(pt); ++j) | |
43a3795a | 545 | if (r || is_present_gpte(pt[j])) |
eab9f71f AK |
546 | sp->spt[i+j] = shadow_trap_nonpresent_pte; |
547 | else | |
548 | sp->spt[i+j] = shadow_notrap_nonpresent_pte; | |
7ec54588 | 549 | } |
c7addb90 AK |
550 | } |
551 | ||
e8bc217a MT |
552 | /* |
553 | * Using the cached information from sp->gfns is safe because: | |
554 | * - The spte has a reference to the struct page, so the pfn for a given gfn | |
555 | * can't change unless all sptes pointing to it are nuked first. | |
556 | * - Alias changes zap the entire shadow cache. | |
557 | */ | |
558 | static int FNAME(sync_page)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp) | |
559 | { | |
560 | int i, offset, nr_present; | |
561 | ||
562 | offset = nr_present = 0; | |
563 | ||
564 | if (PTTYPE == 32) | |
565 | offset = sp->role.quadrant << PT64_LEVEL_BITS; | |
566 | ||
567 | for (i = 0; i < PT64_ENT_PER_PAGE; i++) { | |
568 | unsigned pte_access; | |
569 | pt_element_t gpte; | |
570 | gpa_t pte_gpa; | |
571 | gfn_t gfn = sp->gfns[i]; | |
572 | ||
573 | if (!is_shadow_present_pte(sp->spt[i])) | |
574 | continue; | |
575 | ||
576 | pte_gpa = gfn_to_gpa(sp->gfn); | |
577 | pte_gpa += (i+offset) * sizeof(pt_element_t); | |
578 | ||
579 | if (kvm_read_guest_atomic(vcpu->kvm, pte_gpa, &gpte, | |
580 | sizeof(pt_element_t))) | |
581 | return -EINVAL; | |
582 | ||
43a3795a | 583 | if (gpte_to_gfn(gpte) != gfn || !is_present_gpte(gpte) || |
e8bc217a MT |
584 | !(gpte & PT_ACCESSED_MASK)) { |
585 | u64 nonpresent; | |
586 | ||
587 | rmap_remove(vcpu->kvm, &sp->spt[i]); | |
43a3795a | 588 | if (is_present_gpte(gpte)) |
e8bc217a MT |
589 | nonpresent = shadow_trap_nonpresent_pte; |
590 | else | |
591 | nonpresent = shadow_notrap_nonpresent_pte; | |
d555c333 | 592 | __set_spte(&sp->spt[i], nonpresent); |
e8bc217a MT |
593 | continue; |
594 | } | |
595 | ||
596 | nr_present++; | |
597 | pte_access = sp->role.access & FNAME(gpte_access)(vcpu, gpte); | |
598 | set_spte(vcpu, &sp->spt[i], pte_access, 0, 0, | |
7e4e4056 | 599 | is_dirty_gpte(gpte), PT_PAGE_TABLE_LEVEL, gfn, |
4731d4c7 | 600 | spte_to_pfn(sp->spt[i]), true, false); |
e8bc217a MT |
601 | } |
602 | ||
603 | return !nr_present; | |
604 | } | |
605 | ||
6aa8b732 AK |
606 | #undef pt_element_t |
607 | #undef guest_walker | |
608 | #undef FNAME | |
609 | #undef PT_BASE_ADDR_MASK | |
610 | #undef PT_INDEX | |
6aa8b732 | 611 | #undef PT_LEVEL_MASK |
e04da980 JR |
612 | #undef PT_LVL_ADDR_MASK |
613 | #undef PT_LVL_OFFSET_MASK | |
c7addb90 | 614 | #undef PT_LEVEL_BITS |
cea0f0e7 | 615 | #undef PT_MAX_FULL_LEVELS |
5fb07ddb | 616 | #undef gpte_to_gfn |
e04da980 | 617 | #undef gpte_to_gfn_lvl |
b3e4e63f | 618 | #undef CMPXCHG |