]>
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 | */ | |
e495606d AK |
19 | |
20 | #include "vmx.h" | |
21 | #include "kvm.h" | |
22 | ||
6aa8b732 AK |
23 | #include <linux/types.h> |
24 | #include <linux/string.h> | |
6aa8b732 AK |
25 | #include <linux/mm.h> |
26 | #include <linux/highmem.h> | |
27 | #include <linux/module.h> | |
28 | ||
e495606d AK |
29 | #include <asm/page.h> |
30 | #include <asm/cmpxchg.h> | |
6aa8b732 | 31 | |
37a7d8b0 AK |
32 | #undef MMU_DEBUG |
33 | ||
34 | #undef AUDIT | |
35 | ||
36 | #ifdef AUDIT | |
37 | static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg); | |
38 | #else | |
39 | static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg) {} | |
40 | #endif | |
41 | ||
42 | #ifdef MMU_DEBUG | |
43 | ||
44 | #define pgprintk(x...) do { if (dbg) printk(x); } while (0) | |
45 | #define rmap_printk(x...) do { if (dbg) printk(x); } while (0) | |
46 | ||
47 | #else | |
48 | ||
49 | #define pgprintk(x...) do { } while (0) | |
50 | #define rmap_printk(x...) do { } while (0) | |
51 | ||
52 | #endif | |
53 | ||
54 | #if defined(MMU_DEBUG) || defined(AUDIT) | |
55 | static int dbg = 1; | |
56 | #endif | |
6aa8b732 | 57 | |
d6c69ee9 YD |
58 | #ifndef MMU_DEBUG |
59 | #define ASSERT(x) do { } while (0) | |
60 | #else | |
6aa8b732 AK |
61 | #define ASSERT(x) \ |
62 | if (!(x)) { \ | |
63 | printk(KERN_WARNING "assertion failed %s:%d: %s\n", \ | |
64 | __FILE__, __LINE__, #x); \ | |
65 | } | |
d6c69ee9 | 66 | #endif |
6aa8b732 | 67 | |
cea0f0e7 AK |
68 | #define PT64_PT_BITS 9 |
69 | #define PT64_ENT_PER_PAGE (1 << PT64_PT_BITS) | |
70 | #define PT32_PT_BITS 10 | |
71 | #define PT32_ENT_PER_PAGE (1 << PT32_PT_BITS) | |
6aa8b732 AK |
72 | |
73 | #define PT_WRITABLE_SHIFT 1 | |
74 | ||
75 | #define PT_PRESENT_MASK (1ULL << 0) | |
76 | #define PT_WRITABLE_MASK (1ULL << PT_WRITABLE_SHIFT) | |
77 | #define PT_USER_MASK (1ULL << 2) | |
78 | #define PT_PWT_MASK (1ULL << 3) | |
79 | #define PT_PCD_MASK (1ULL << 4) | |
80 | #define PT_ACCESSED_MASK (1ULL << 5) | |
81 | #define PT_DIRTY_MASK (1ULL << 6) | |
82 | #define PT_PAGE_SIZE_MASK (1ULL << 7) | |
83 | #define PT_PAT_MASK (1ULL << 7) | |
84 | #define PT_GLOBAL_MASK (1ULL << 8) | |
85 | #define PT64_NX_MASK (1ULL << 63) | |
86 | ||
87 | #define PT_PAT_SHIFT 7 | |
88 | #define PT_DIR_PAT_SHIFT 12 | |
89 | #define PT_DIR_PAT_MASK (1ULL << PT_DIR_PAT_SHIFT) | |
90 | ||
91 | #define PT32_DIR_PSE36_SIZE 4 | |
92 | #define PT32_DIR_PSE36_SHIFT 13 | |
93 | #define PT32_DIR_PSE36_MASK (((1ULL << PT32_DIR_PSE36_SIZE) - 1) << PT32_DIR_PSE36_SHIFT) | |
94 | ||
95 | ||
6aa8b732 AK |
96 | #define PT_FIRST_AVAIL_BITS_SHIFT 9 |
97 | #define PT64_SECOND_AVAIL_BITS_SHIFT 52 | |
98 | ||
6aa8b732 AK |
99 | #define PT_SHADOW_IO_MARK (1ULL << PT_FIRST_AVAIL_BITS_SHIFT) |
100 | ||
6aa8b732 AK |
101 | #define VALID_PAGE(x) ((x) != INVALID_PAGE) |
102 | ||
103 | #define PT64_LEVEL_BITS 9 | |
104 | ||
105 | #define PT64_LEVEL_SHIFT(level) \ | |
106 | ( PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS ) | |
107 | ||
108 | #define PT64_LEVEL_MASK(level) \ | |
109 | (((1ULL << PT64_LEVEL_BITS) - 1) << PT64_LEVEL_SHIFT(level)) | |
110 | ||
111 | #define PT64_INDEX(address, level)\ | |
112 | (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1)) | |
113 | ||
114 | ||
115 | #define PT32_LEVEL_BITS 10 | |
116 | ||
117 | #define PT32_LEVEL_SHIFT(level) \ | |
118 | ( PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS ) | |
119 | ||
120 | #define PT32_LEVEL_MASK(level) \ | |
121 | (((1ULL << PT32_LEVEL_BITS) - 1) << PT32_LEVEL_SHIFT(level)) | |
122 | ||
123 | #define PT32_INDEX(address, level)\ | |
124 | (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1)) | |
125 | ||
126 | ||
27aba766 | 127 | #define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1)) |
6aa8b732 AK |
128 | #define PT64_DIR_BASE_ADDR_MASK \ |
129 | (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1)) | |
130 | ||
131 | #define PT32_BASE_ADDR_MASK PAGE_MASK | |
132 | #define PT32_DIR_BASE_ADDR_MASK \ | |
133 | (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1)) | |
134 | ||
135 | ||
136 | #define PFERR_PRESENT_MASK (1U << 0) | |
137 | #define PFERR_WRITE_MASK (1U << 1) | |
138 | #define PFERR_USER_MASK (1U << 2) | |
73b1087e | 139 | #define PFERR_FETCH_MASK (1U << 4) |
6aa8b732 AK |
140 | |
141 | #define PT64_ROOT_LEVEL 4 | |
142 | #define PT32_ROOT_LEVEL 2 | |
143 | #define PT32E_ROOT_LEVEL 3 | |
144 | ||
145 | #define PT_DIRECTORY_LEVEL 2 | |
146 | #define PT_PAGE_TABLE_LEVEL 1 | |
147 | ||
cd4a4e53 AK |
148 | #define RMAP_EXT 4 |
149 | ||
150 | struct kvm_rmap_desc { | |
151 | u64 *shadow_ptes[RMAP_EXT]; | |
152 | struct kvm_rmap_desc *more; | |
153 | }; | |
154 | ||
b5a33a75 AK |
155 | static struct kmem_cache *pte_chain_cache; |
156 | static struct kmem_cache *rmap_desc_cache; | |
d3d25b04 | 157 | static struct kmem_cache *mmu_page_header_cache; |
b5a33a75 | 158 | |
c7addb90 AK |
159 | static u64 __read_mostly shadow_trap_nonpresent_pte; |
160 | static u64 __read_mostly shadow_notrap_nonpresent_pte; | |
161 | ||
162 | void kvm_mmu_set_nonpresent_ptes(u64 trap_pte, u64 notrap_pte) | |
163 | { | |
164 | shadow_trap_nonpresent_pte = trap_pte; | |
165 | shadow_notrap_nonpresent_pte = notrap_pte; | |
166 | } | |
167 | EXPORT_SYMBOL_GPL(kvm_mmu_set_nonpresent_ptes); | |
168 | ||
6aa8b732 AK |
169 | static int is_write_protection(struct kvm_vcpu *vcpu) |
170 | { | |
707d92fa | 171 | return vcpu->cr0 & X86_CR0_WP; |
6aa8b732 AK |
172 | } |
173 | ||
174 | static int is_cpuid_PSE36(void) | |
175 | { | |
176 | return 1; | |
177 | } | |
178 | ||
73b1087e AK |
179 | static int is_nx(struct kvm_vcpu *vcpu) |
180 | { | |
181 | return vcpu->shadow_efer & EFER_NX; | |
182 | } | |
183 | ||
6aa8b732 AK |
184 | static int is_present_pte(unsigned long pte) |
185 | { | |
186 | return pte & PT_PRESENT_MASK; | |
187 | } | |
188 | ||
c7addb90 AK |
189 | static int is_shadow_present_pte(u64 pte) |
190 | { | |
191 | pte &= ~PT_SHADOW_IO_MARK; | |
192 | return pte != shadow_trap_nonpresent_pte | |
193 | && pte != shadow_notrap_nonpresent_pte; | |
194 | } | |
195 | ||
6aa8b732 AK |
196 | static int is_writeble_pte(unsigned long pte) |
197 | { | |
198 | return pte & PT_WRITABLE_MASK; | |
199 | } | |
200 | ||
201 | static int is_io_pte(unsigned long pte) | |
202 | { | |
203 | return pte & PT_SHADOW_IO_MARK; | |
204 | } | |
205 | ||
cd4a4e53 AK |
206 | static int is_rmap_pte(u64 pte) |
207 | { | |
208 | return (pte & (PT_WRITABLE_MASK | PT_PRESENT_MASK)) | |
209 | == (PT_WRITABLE_MASK | PT_PRESENT_MASK); | |
210 | } | |
211 | ||
e663ee64 AK |
212 | static void set_shadow_pte(u64 *sptep, u64 spte) |
213 | { | |
214 | #ifdef CONFIG_X86_64 | |
215 | set_64bit((unsigned long *)sptep, spte); | |
216 | #else | |
217 | set_64bit((unsigned long long *)sptep, spte); | |
218 | #endif | |
219 | } | |
220 | ||
e2dec939 | 221 | static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache, |
2e3e5882 | 222 | struct kmem_cache *base_cache, int min) |
714b93da AK |
223 | { |
224 | void *obj; | |
225 | ||
226 | if (cache->nobjs >= min) | |
e2dec939 | 227 | return 0; |
714b93da | 228 | while (cache->nobjs < ARRAY_SIZE(cache->objects)) { |
2e3e5882 | 229 | obj = kmem_cache_zalloc(base_cache, GFP_KERNEL); |
714b93da | 230 | if (!obj) |
e2dec939 | 231 | return -ENOMEM; |
714b93da AK |
232 | cache->objects[cache->nobjs++] = obj; |
233 | } | |
e2dec939 | 234 | return 0; |
714b93da AK |
235 | } |
236 | ||
237 | static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc) | |
238 | { | |
239 | while (mc->nobjs) | |
240 | kfree(mc->objects[--mc->nobjs]); | |
241 | } | |
242 | ||
c1158e63 | 243 | static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache *cache, |
2e3e5882 | 244 | int min) |
c1158e63 AK |
245 | { |
246 | struct page *page; | |
247 | ||
248 | if (cache->nobjs >= min) | |
249 | return 0; | |
250 | while (cache->nobjs < ARRAY_SIZE(cache->objects)) { | |
2e3e5882 | 251 | page = alloc_page(GFP_KERNEL); |
c1158e63 AK |
252 | if (!page) |
253 | return -ENOMEM; | |
254 | set_page_private(page, 0); | |
255 | cache->objects[cache->nobjs++] = page_address(page); | |
256 | } | |
257 | return 0; | |
258 | } | |
259 | ||
260 | static void mmu_free_memory_cache_page(struct kvm_mmu_memory_cache *mc) | |
261 | { | |
262 | while (mc->nobjs) | |
c4d198d5 | 263 | free_page((unsigned long)mc->objects[--mc->nobjs]); |
c1158e63 AK |
264 | } |
265 | ||
2e3e5882 | 266 | static int mmu_topup_memory_caches(struct kvm_vcpu *vcpu) |
714b93da | 267 | { |
e2dec939 AK |
268 | int r; |
269 | ||
2e3e5882 | 270 | kvm_mmu_free_some_pages(vcpu); |
e2dec939 | 271 | r = mmu_topup_memory_cache(&vcpu->mmu_pte_chain_cache, |
2e3e5882 | 272 | pte_chain_cache, 4); |
e2dec939 AK |
273 | if (r) |
274 | goto out; | |
275 | r = mmu_topup_memory_cache(&vcpu->mmu_rmap_desc_cache, | |
2e3e5882 | 276 | rmap_desc_cache, 1); |
d3d25b04 AK |
277 | if (r) |
278 | goto out; | |
290fc38d | 279 | r = mmu_topup_memory_cache_page(&vcpu->mmu_page_cache, 8); |
d3d25b04 AK |
280 | if (r) |
281 | goto out; | |
282 | r = mmu_topup_memory_cache(&vcpu->mmu_page_header_cache, | |
2e3e5882 | 283 | mmu_page_header_cache, 4); |
e2dec939 AK |
284 | out: |
285 | return r; | |
714b93da AK |
286 | } |
287 | ||
288 | static void mmu_free_memory_caches(struct kvm_vcpu *vcpu) | |
289 | { | |
290 | mmu_free_memory_cache(&vcpu->mmu_pte_chain_cache); | |
291 | mmu_free_memory_cache(&vcpu->mmu_rmap_desc_cache); | |
c1158e63 | 292 | mmu_free_memory_cache_page(&vcpu->mmu_page_cache); |
d3d25b04 | 293 | mmu_free_memory_cache(&vcpu->mmu_page_header_cache); |
714b93da AK |
294 | } |
295 | ||
296 | static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc, | |
297 | size_t size) | |
298 | { | |
299 | void *p; | |
300 | ||
301 | BUG_ON(!mc->nobjs); | |
302 | p = mc->objects[--mc->nobjs]; | |
303 | memset(p, 0, size); | |
304 | return p; | |
305 | } | |
306 | ||
714b93da AK |
307 | static struct kvm_pte_chain *mmu_alloc_pte_chain(struct kvm_vcpu *vcpu) |
308 | { | |
309 | return mmu_memory_cache_alloc(&vcpu->mmu_pte_chain_cache, | |
310 | sizeof(struct kvm_pte_chain)); | |
311 | } | |
312 | ||
90cb0529 | 313 | static void mmu_free_pte_chain(struct kvm_pte_chain *pc) |
714b93da | 314 | { |
90cb0529 | 315 | kfree(pc); |
714b93da AK |
316 | } |
317 | ||
318 | static struct kvm_rmap_desc *mmu_alloc_rmap_desc(struct kvm_vcpu *vcpu) | |
319 | { | |
320 | return mmu_memory_cache_alloc(&vcpu->mmu_rmap_desc_cache, | |
321 | sizeof(struct kvm_rmap_desc)); | |
322 | } | |
323 | ||
90cb0529 | 324 | static void mmu_free_rmap_desc(struct kvm_rmap_desc *rd) |
714b93da | 325 | { |
90cb0529 | 326 | kfree(rd); |
714b93da AK |
327 | } |
328 | ||
290fc38d IE |
329 | /* |
330 | * Take gfn and return the reverse mapping to it. | |
331 | * Note: gfn must be unaliased before this function get called | |
332 | */ | |
333 | ||
334 | static unsigned long *gfn_to_rmap(struct kvm *kvm, gfn_t gfn) | |
335 | { | |
336 | struct kvm_memory_slot *slot; | |
337 | ||
338 | slot = gfn_to_memslot(kvm, gfn); | |
339 | return &slot->rmap[gfn - slot->base_gfn]; | |
340 | } | |
341 | ||
cd4a4e53 AK |
342 | /* |
343 | * Reverse mapping data structures: | |
344 | * | |
290fc38d IE |
345 | * If rmapp bit zero is zero, then rmapp point to the shadw page table entry |
346 | * that points to page_address(page). | |
cd4a4e53 | 347 | * |
290fc38d IE |
348 | * If rmapp bit zero is one, (then rmap & ~1) points to a struct kvm_rmap_desc |
349 | * containing more mappings. | |
cd4a4e53 | 350 | */ |
290fc38d | 351 | static void rmap_add(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn) |
cd4a4e53 | 352 | { |
290fc38d | 353 | struct kvm_mmu_page *page; |
cd4a4e53 | 354 | struct kvm_rmap_desc *desc; |
290fc38d | 355 | unsigned long *rmapp; |
cd4a4e53 AK |
356 | int i; |
357 | ||
358 | if (!is_rmap_pte(*spte)) | |
359 | return; | |
290fc38d IE |
360 | gfn = unalias_gfn(vcpu->kvm, gfn); |
361 | page = page_header(__pa(spte)); | |
362 | page->gfns[spte - page->spt] = gfn; | |
363 | rmapp = gfn_to_rmap(vcpu->kvm, gfn); | |
364 | if (!*rmapp) { | |
cd4a4e53 | 365 | rmap_printk("rmap_add: %p %llx 0->1\n", spte, *spte); |
290fc38d IE |
366 | *rmapp = (unsigned long)spte; |
367 | } else if (!(*rmapp & 1)) { | |
cd4a4e53 | 368 | rmap_printk("rmap_add: %p %llx 1->many\n", spte, *spte); |
714b93da | 369 | desc = mmu_alloc_rmap_desc(vcpu); |
290fc38d | 370 | desc->shadow_ptes[0] = (u64 *)*rmapp; |
cd4a4e53 | 371 | desc->shadow_ptes[1] = spte; |
290fc38d | 372 | *rmapp = (unsigned long)desc | 1; |
cd4a4e53 AK |
373 | } else { |
374 | rmap_printk("rmap_add: %p %llx many->many\n", spte, *spte); | |
290fc38d | 375 | desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul); |
cd4a4e53 AK |
376 | while (desc->shadow_ptes[RMAP_EXT-1] && desc->more) |
377 | desc = desc->more; | |
378 | if (desc->shadow_ptes[RMAP_EXT-1]) { | |
714b93da | 379 | desc->more = mmu_alloc_rmap_desc(vcpu); |
cd4a4e53 AK |
380 | desc = desc->more; |
381 | } | |
382 | for (i = 0; desc->shadow_ptes[i]; ++i) | |
383 | ; | |
384 | desc->shadow_ptes[i] = spte; | |
385 | } | |
386 | } | |
387 | ||
290fc38d | 388 | static void rmap_desc_remove_entry(unsigned long *rmapp, |
cd4a4e53 AK |
389 | struct kvm_rmap_desc *desc, |
390 | int i, | |
391 | struct kvm_rmap_desc *prev_desc) | |
392 | { | |
393 | int j; | |
394 | ||
395 | for (j = RMAP_EXT - 1; !desc->shadow_ptes[j] && j > i; --j) | |
396 | ; | |
397 | desc->shadow_ptes[i] = desc->shadow_ptes[j]; | |
11718b4d | 398 | desc->shadow_ptes[j] = NULL; |
cd4a4e53 AK |
399 | if (j != 0) |
400 | return; | |
401 | if (!prev_desc && !desc->more) | |
290fc38d | 402 | *rmapp = (unsigned long)desc->shadow_ptes[0]; |
cd4a4e53 AK |
403 | else |
404 | if (prev_desc) | |
405 | prev_desc->more = desc->more; | |
406 | else | |
290fc38d | 407 | *rmapp = (unsigned long)desc->more | 1; |
90cb0529 | 408 | mmu_free_rmap_desc(desc); |
cd4a4e53 AK |
409 | } |
410 | ||
290fc38d | 411 | static void rmap_remove(struct kvm *kvm, u64 *spte) |
cd4a4e53 | 412 | { |
cd4a4e53 AK |
413 | struct kvm_rmap_desc *desc; |
414 | struct kvm_rmap_desc *prev_desc; | |
290fc38d IE |
415 | struct kvm_mmu_page *page; |
416 | unsigned long *rmapp; | |
cd4a4e53 AK |
417 | int i; |
418 | ||
419 | if (!is_rmap_pte(*spte)) | |
420 | return; | |
290fc38d IE |
421 | page = page_header(__pa(spte)); |
422 | rmapp = gfn_to_rmap(kvm, page->gfns[spte - page->spt]); | |
423 | if (!*rmapp) { | |
cd4a4e53 AK |
424 | printk(KERN_ERR "rmap_remove: %p %llx 0->BUG\n", spte, *spte); |
425 | BUG(); | |
290fc38d | 426 | } else if (!(*rmapp & 1)) { |
cd4a4e53 | 427 | rmap_printk("rmap_remove: %p %llx 1->0\n", spte, *spte); |
290fc38d | 428 | if ((u64 *)*rmapp != spte) { |
cd4a4e53 AK |
429 | printk(KERN_ERR "rmap_remove: %p %llx 1->BUG\n", |
430 | spte, *spte); | |
431 | BUG(); | |
432 | } | |
290fc38d | 433 | *rmapp = 0; |
cd4a4e53 AK |
434 | } else { |
435 | rmap_printk("rmap_remove: %p %llx many->many\n", spte, *spte); | |
290fc38d | 436 | desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul); |
cd4a4e53 AK |
437 | prev_desc = NULL; |
438 | while (desc) { | |
439 | for (i = 0; i < RMAP_EXT && desc->shadow_ptes[i]; ++i) | |
440 | if (desc->shadow_ptes[i] == spte) { | |
290fc38d | 441 | rmap_desc_remove_entry(rmapp, |
714b93da | 442 | desc, i, |
cd4a4e53 AK |
443 | prev_desc); |
444 | return; | |
445 | } | |
446 | prev_desc = desc; | |
447 | desc = desc->more; | |
448 | } | |
449 | BUG(); | |
450 | } | |
451 | } | |
452 | ||
714b93da | 453 | static void rmap_write_protect(struct kvm_vcpu *vcpu, u64 gfn) |
374cbac0 | 454 | { |
374cbac0 | 455 | struct kvm_rmap_desc *desc; |
290fc38d | 456 | unsigned long *rmapp; |
374cbac0 AK |
457 | u64 *spte; |
458 | ||
290fc38d IE |
459 | gfn = unalias_gfn(vcpu->kvm, gfn); |
460 | rmapp = gfn_to_rmap(vcpu->kvm, gfn); | |
374cbac0 | 461 | |
290fc38d IE |
462 | while (*rmapp) { |
463 | if (!(*rmapp & 1)) | |
464 | spte = (u64 *)*rmapp; | |
374cbac0 | 465 | else { |
290fc38d | 466 | desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul); |
374cbac0 AK |
467 | spte = desc->shadow_ptes[0]; |
468 | } | |
469 | BUG_ON(!spte); | |
374cbac0 AK |
470 | BUG_ON(!(*spte & PT_PRESENT_MASK)); |
471 | BUG_ON(!(*spte & PT_WRITABLE_MASK)); | |
472 | rmap_printk("rmap_write_protect: spte %p %llx\n", spte, *spte); | |
290fc38d | 473 | rmap_remove(vcpu->kvm, spte); |
e663ee64 | 474 | set_shadow_pte(spte, *spte & ~PT_WRITABLE_MASK); |
88a97f0b | 475 | kvm_flush_remote_tlbs(vcpu->kvm); |
374cbac0 AK |
476 | } |
477 | } | |
478 | ||
d6c69ee9 | 479 | #ifdef MMU_DEBUG |
47ad8e68 | 480 | static int is_empty_shadow_page(u64 *spt) |
6aa8b732 | 481 | { |
139bdb2d AK |
482 | u64 *pos; |
483 | u64 *end; | |
484 | ||
47ad8e68 | 485 | for (pos = spt, end = pos + PAGE_SIZE / sizeof(u64); pos != end; pos++) |
c7addb90 | 486 | if ((*pos & ~PT_SHADOW_IO_MARK) != shadow_trap_nonpresent_pte) { |
139bdb2d AK |
487 | printk(KERN_ERR "%s: %p %llx\n", __FUNCTION__, |
488 | pos, *pos); | |
6aa8b732 | 489 | return 0; |
139bdb2d | 490 | } |
6aa8b732 AK |
491 | return 1; |
492 | } | |
d6c69ee9 | 493 | #endif |
6aa8b732 | 494 | |
90cb0529 | 495 | static void kvm_mmu_free_page(struct kvm *kvm, |
4b02d6da | 496 | struct kvm_mmu_page *page_head) |
260746c0 | 497 | { |
47ad8e68 | 498 | ASSERT(is_empty_shadow_page(page_head->spt)); |
d3d25b04 | 499 | list_del(&page_head->link); |
c1158e63 | 500 | __free_page(virt_to_page(page_head->spt)); |
290fc38d | 501 | __free_page(virt_to_page(page_head->gfns)); |
90cb0529 AK |
502 | kfree(page_head); |
503 | ++kvm->n_free_mmu_pages; | |
260746c0 AK |
504 | } |
505 | ||
cea0f0e7 AK |
506 | static unsigned kvm_page_table_hashfn(gfn_t gfn) |
507 | { | |
508 | return gfn; | |
509 | } | |
510 | ||
25c0de2c AK |
511 | static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu, |
512 | u64 *parent_pte) | |
6aa8b732 AK |
513 | { |
514 | struct kvm_mmu_page *page; | |
515 | ||
d3d25b04 | 516 | if (!vcpu->kvm->n_free_mmu_pages) |
25c0de2c | 517 | return NULL; |
6aa8b732 | 518 | |
d3d25b04 AK |
519 | page = mmu_memory_cache_alloc(&vcpu->mmu_page_header_cache, |
520 | sizeof *page); | |
521 | page->spt = mmu_memory_cache_alloc(&vcpu->mmu_page_cache, PAGE_SIZE); | |
290fc38d | 522 | page->gfns = mmu_memory_cache_alloc(&vcpu->mmu_page_cache, PAGE_SIZE); |
d3d25b04 AK |
523 | set_page_private(virt_to_page(page->spt), (unsigned long)page); |
524 | list_add(&page->link, &vcpu->kvm->active_mmu_pages); | |
47ad8e68 | 525 | ASSERT(is_empty_shadow_page(page->spt)); |
6aa8b732 | 526 | page->slot_bitmap = 0; |
cea0f0e7 | 527 | page->multimapped = 0; |
6aa8b732 | 528 | page->parent_pte = parent_pte; |
ebeace86 | 529 | --vcpu->kvm->n_free_mmu_pages; |
25c0de2c | 530 | return page; |
6aa8b732 AK |
531 | } |
532 | ||
714b93da AK |
533 | static void mmu_page_add_parent_pte(struct kvm_vcpu *vcpu, |
534 | struct kvm_mmu_page *page, u64 *parent_pte) | |
cea0f0e7 AK |
535 | { |
536 | struct kvm_pte_chain *pte_chain; | |
537 | struct hlist_node *node; | |
538 | int i; | |
539 | ||
540 | if (!parent_pte) | |
541 | return; | |
542 | if (!page->multimapped) { | |
543 | u64 *old = page->parent_pte; | |
544 | ||
545 | if (!old) { | |
546 | page->parent_pte = parent_pte; | |
547 | return; | |
548 | } | |
549 | page->multimapped = 1; | |
714b93da | 550 | pte_chain = mmu_alloc_pte_chain(vcpu); |
cea0f0e7 AK |
551 | INIT_HLIST_HEAD(&page->parent_ptes); |
552 | hlist_add_head(&pte_chain->link, &page->parent_ptes); | |
553 | pte_chain->parent_ptes[0] = old; | |
554 | } | |
555 | hlist_for_each_entry(pte_chain, node, &page->parent_ptes, link) { | |
556 | if (pte_chain->parent_ptes[NR_PTE_CHAIN_ENTRIES-1]) | |
557 | continue; | |
558 | for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) | |
559 | if (!pte_chain->parent_ptes[i]) { | |
560 | pte_chain->parent_ptes[i] = parent_pte; | |
561 | return; | |
562 | } | |
563 | } | |
714b93da | 564 | pte_chain = mmu_alloc_pte_chain(vcpu); |
cea0f0e7 AK |
565 | BUG_ON(!pte_chain); |
566 | hlist_add_head(&pte_chain->link, &page->parent_ptes); | |
567 | pte_chain->parent_ptes[0] = parent_pte; | |
568 | } | |
569 | ||
90cb0529 | 570 | static void mmu_page_remove_parent_pte(struct kvm_mmu_page *page, |
cea0f0e7 AK |
571 | u64 *parent_pte) |
572 | { | |
573 | struct kvm_pte_chain *pte_chain; | |
574 | struct hlist_node *node; | |
575 | int i; | |
576 | ||
577 | if (!page->multimapped) { | |
578 | BUG_ON(page->parent_pte != parent_pte); | |
579 | page->parent_pte = NULL; | |
580 | return; | |
581 | } | |
582 | hlist_for_each_entry(pte_chain, node, &page->parent_ptes, link) | |
583 | for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) { | |
584 | if (!pte_chain->parent_ptes[i]) | |
585 | break; | |
586 | if (pte_chain->parent_ptes[i] != parent_pte) | |
587 | continue; | |
697fe2e2 AK |
588 | while (i + 1 < NR_PTE_CHAIN_ENTRIES |
589 | && pte_chain->parent_ptes[i + 1]) { | |
cea0f0e7 AK |
590 | pte_chain->parent_ptes[i] |
591 | = pte_chain->parent_ptes[i + 1]; | |
592 | ++i; | |
593 | } | |
594 | pte_chain->parent_ptes[i] = NULL; | |
697fe2e2 AK |
595 | if (i == 0) { |
596 | hlist_del(&pte_chain->link); | |
90cb0529 | 597 | mmu_free_pte_chain(pte_chain); |
697fe2e2 AK |
598 | if (hlist_empty(&page->parent_ptes)) { |
599 | page->multimapped = 0; | |
600 | page->parent_pte = NULL; | |
601 | } | |
602 | } | |
cea0f0e7 AK |
603 | return; |
604 | } | |
605 | BUG(); | |
606 | } | |
607 | ||
608 | static struct kvm_mmu_page *kvm_mmu_lookup_page(struct kvm_vcpu *vcpu, | |
609 | gfn_t gfn) | |
610 | { | |
611 | unsigned index; | |
612 | struct hlist_head *bucket; | |
613 | struct kvm_mmu_page *page; | |
614 | struct hlist_node *node; | |
615 | ||
616 | pgprintk("%s: looking for gfn %lx\n", __FUNCTION__, gfn); | |
617 | index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES; | |
618 | bucket = &vcpu->kvm->mmu_page_hash[index]; | |
619 | hlist_for_each_entry(page, node, bucket, hash_link) | |
620 | if (page->gfn == gfn && !page->role.metaphysical) { | |
621 | pgprintk("%s: found role %x\n", | |
622 | __FUNCTION__, page->role.word); | |
623 | return page; | |
624 | } | |
625 | return NULL; | |
626 | } | |
627 | ||
628 | static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu, | |
629 | gfn_t gfn, | |
630 | gva_t gaddr, | |
631 | unsigned level, | |
632 | int metaphysical, | |
d28c6cfb | 633 | unsigned hugepage_access, |
cea0f0e7 AK |
634 | u64 *parent_pte) |
635 | { | |
636 | union kvm_mmu_page_role role; | |
637 | unsigned index; | |
638 | unsigned quadrant; | |
639 | struct hlist_head *bucket; | |
640 | struct kvm_mmu_page *page; | |
641 | struct hlist_node *node; | |
642 | ||
643 | role.word = 0; | |
644 | role.glevels = vcpu->mmu.root_level; | |
645 | role.level = level; | |
646 | role.metaphysical = metaphysical; | |
d28c6cfb | 647 | role.hugepage_access = hugepage_access; |
cea0f0e7 AK |
648 | if (vcpu->mmu.root_level <= PT32_ROOT_LEVEL) { |
649 | quadrant = gaddr >> (PAGE_SHIFT + (PT64_PT_BITS * level)); | |
650 | quadrant &= (1 << ((PT32_PT_BITS - PT64_PT_BITS) * level)) - 1; | |
651 | role.quadrant = quadrant; | |
652 | } | |
653 | pgprintk("%s: looking gfn %lx role %x\n", __FUNCTION__, | |
654 | gfn, role.word); | |
655 | index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES; | |
656 | bucket = &vcpu->kvm->mmu_page_hash[index]; | |
657 | hlist_for_each_entry(page, node, bucket, hash_link) | |
658 | if (page->gfn == gfn && page->role.word == role.word) { | |
714b93da | 659 | mmu_page_add_parent_pte(vcpu, page, parent_pte); |
cea0f0e7 AK |
660 | pgprintk("%s: found\n", __FUNCTION__); |
661 | return page; | |
662 | } | |
663 | page = kvm_mmu_alloc_page(vcpu, parent_pte); | |
664 | if (!page) | |
665 | return page; | |
666 | pgprintk("%s: adding gfn %lx role %x\n", __FUNCTION__, gfn, role.word); | |
667 | page->gfn = gfn; | |
668 | page->role = role; | |
669 | hlist_add_head(&page->hash_link, bucket); | |
c7addb90 | 670 | vcpu->mmu.prefetch_page(vcpu, page); |
374cbac0 | 671 | if (!metaphysical) |
714b93da | 672 | rmap_write_protect(vcpu, gfn); |
cea0f0e7 AK |
673 | return page; |
674 | } | |
675 | ||
90cb0529 | 676 | static void kvm_mmu_page_unlink_children(struct kvm *kvm, |
a436036b AK |
677 | struct kvm_mmu_page *page) |
678 | { | |
697fe2e2 AK |
679 | unsigned i; |
680 | u64 *pt; | |
681 | u64 ent; | |
682 | ||
47ad8e68 | 683 | pt = page->spt; |
697fe2e2 AK |
684 | |
685 | if (page->role.level == PT_PAGE_TABLE_LEVEL) { | |
686 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) { | |
c7addb90 | 687 | if (is_shadow_present_pte(pt[i])) |
290fc38d | 688 | rmap_remove(kvm, &pt[i]); |
c7addb90 | 689 | pt[i] = shadow_trap_nonpresent_pte; |
697fe2e2 | 690 | } |
90cb0529 | 691 | kvm_flush_remote_tlbs(kvm); |
697fe2e2 AK |
692 | return; |
693 | } | |
694 | ||
695 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) { | |
696 | ent = pt[i]; | |
697 | ||
c7addb90 AK |
698 | pt[i] = shadow_trap_nonpresent_pte; |
699 | if (!is_shadow_present_pte(ent)) | |
697fe2e2 AK |
700 | continue; |
701 | ent &= PT64_BASE_ADDR_MASK; | |
90cb0529 | 702 | mmu_page_remove_parent_pte(page_header(ent), &pt[i]); |
697fe2e2 | 703 | } |
90cb0529 | 704 | kvm_flush_remote_tlbs(kvm); |
a436036b AK |
705 | } |
706 | ||
90cb0529 | 707 | static void kvm_mmu_put_page(struct kvm_mmu_page *page, |
cea0f0e7 AK |
708 | u64 *parent_pte) |
709 | { | |
90cb0529 | 710 | mmu_page_remove_parent_pte(page, parent_pte); |
a436036b AK |
711 | } |
712 | ||
12b7d28f AK |
713 | static void kvm_mmu_reset_last_pte_updated(struct kvm *kvm) |
714 | { | |
715 | int i; | |
716 | ||
717 | for (i = 0; i < KVM_MAX_VCPUS; ++i) | |
718 | if (kvm->vcpus[i]) | |
719 | kvm->vcpus[i]->last_pte_updated = NULL; | |
720 | } | |
721 | ||
90cb0529 | 722 | static void kvm_mmu_zap_page(struct kvm *kvm, |
a436036b AK |
723 | struct kvm_mmu_page *page) |
724 | { | |
725 | u64 *parent_pte; | |
726 | ||
727 | while (page->multimapped || page->parent_pte) { | |
728 | if (!page->multimapped) | |
729 | parent_pte = page->parent_pte; | |
730 | else { | |
731 | struct kvm_pte_chain *chain; | |
732 | ||
733 | chain = container_of(page->parent_ptes.first, | |
734 | struct kvm_pte_chain, link); | |
735 | parent_pte = chain->parent_ptes[0]; | |
736 | } | |
697fe2e2 | 737 | BUG_ON(!parent_pte); |
90cb0529 | 738 | kvm_mmu_put_page(page, parent_pte); |
c7addb90 | 739 | set_shadow_pte(parent_pte, shadow_trap_nonpresent_pte); |
a436036b | 740 | } |
90cb0529 | 741 | kvm_mmu_page_unlink_children(kvm, page); |
3bb65a22 AK |
742 | if (!page->root_count) { |
743 | hlist_del(&page->hash_link); | |
90cb0529 | 744 | kvm_mmu_free_page(kvm, page); |
36868f7b | 745 | } else |
90cb0529 | 746 | list_move(&page->link, &kvm->active_mmu_pages); |
12b7d28f | 747 | kvm_mmu_reset_last_pte_updated(kvm); |
a436036b AK |
748 | } |
749 | ||
82ce2c96 IE |
750 | /* |
751 | * Changing the number of mmu pages allocated to the vm | |
752 | * Note: if kvm_nr_mmu_pages is too small, you will get dead lock | |
753 | */ | |
754 | void kvm_mmu_change_mmu_pages(struct kvm *kvm, unsigned int kvm_nr_mmu_pages) | |
755 | { | |
756 | /* | |
757 | * If we set the number of mmu pages to be smaller be than the | |
758 | * number of actived pages , we must to free some mmu pages before we | |
759 | * change the value | |
760 | */ | |
761 | ||
762 | if ((kvm->n_alloc_mmu_pages - kvm->n_free_mmu_pages) > | |
763 | kvm_nr_mmu_pages) { | |
764 | int n_used_mmu_pages = kvm->n_alloc_mmu_pages | |
765 | - kvm->n_free_mmu_pages; | |
766 | ||
767 | while (n_used_mmu_pages > kvm_nr_mmu_pages) { | |
768 | struct kvm_mmu_page *page; | |
769 | ||
770 | page = container_of(kvm->active_mmu_pages.prev, | |
771 | struct kvm_mmu_page, link); | |
772 | kvm_mmu_zap_page(kvm, page); | |
773 | n_used_mmu_pages--; | |
774 | } | |
775 | kvm->n_free_mmu_pages = 0; | |
776 | } | |
777 | else | |
778 | kvm->n_free_mmu_pages += kvm_nr_mmu_pages | |
779 | - kvm->n_alloc_mmu_pages; | |
780 | ||
781 | kvm->n_alloc_mmu_pages = kvm_nr_mmu_pages; | |
782 | } | |
783 | ||
a436036b AK |
784 | static int kvm_mmu_unprotect_page(struct kvm_vcpu *vcpu, gfn_t gfn) |
785 | { | |
786 | unsigned index; | |
787 | struct hlist_head *bucket; | |
788 | struct kvm_mmu_page *page; | |
789 | struct hlist_node *node, *n; | |
790 | int r; | |
791 | ||
792 | pgprintk("%s: looking for gfn %lx\n", __FUNCTION__, gfn); | |
793 | r = 0; | |
794 | index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES; | |
795 | bucket = &vcpu->kvm->mmu_page_hash[index]; | |
796 | hlist_for_each_entry_safe(page, node, n, bucket, hash_link) | |
797 | if (page->gfn == gfn && !page->role.metaphysical) { | |
697fe2e2 AK |
798 | pgprintk("%s: gfn %lx role %x\n", __FUNCTION__, gfn, |
799 | page->role.word); | |
90cb0529 | 800 | kvm_mmu_zap_page(vcpu->kvm, page); |
a436036b AK |
801 | r = 1; |
802 | } | |
803 | return r; | |
cea0f0e7 AK |
804 | } |
805 | ||
97a0a01e AK |
806 | static void mmu_unshadow(struct kvm_vcpu *vcpu, gfn_t gfn) |
807 | { | |
808 | struct kvm_mmu_page *page; | |
809 | ||
810 | while ((page = kvm_mmu_lookup_page(vcpu, gfn)) != NULL) { | |
811 | pgprintk("%s: zap %lx %x\n", | |
812 | __FUNCTION__, gfn, page->role.word); | |
90cb0529 | 813 | kvm_mmu_zap_page(vcpu->kvm, page); |
97a0a01e AK |
814 | } |
815 | } | |
816 | ||
6aa8b732 AK |
817 | static void page_header_update_slot(struct kvm *kvm, void *pte, gpa_t gpa) |
818 | { | |
819 | int slot = memslot_id(kvm, gfn_to_memslot(kvm, gpa >> PAGE_SHIFT)); | |
820 | struct kvm_mmu_page *page_head = page_header(__pa(pte)); | |
821 | ||
822 | __set_bit(slot, &page_head->slot_bitmap); | |
823 | } | |
824 | ||
825 | hpa_t safe_gpa_to_hpa(struct kvm_vcpu *vcpu, gpa_t gpa) | |
826 | { | |
827 | hpa_t hpa = gpa_to_hpa(vcpu, gpa); | |
828 | ||
829 | return is_error_hpa(hpa) ? bad_page_address | (gpa & ~PAGE_MASK): hpa; | |
830 | } | |
831 | ||
832 | hpa_t gpa_to_hpa(struct kvm_vcpu *vcpu, gpa_t gpa) | |
833 | { | |
6aa8b732 AK |
834 | struct page *page; |
835 | ||
836 | ASSERT((gpa & HPA_ERR_MASK) == 0); | |
954bbbc2 AK |
837 | page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT); |
838 | if (!page) | |
6aa8b732 | 839 | return gpa | HPA_ERR_MASK; |
6aa8b732 AK |
840 | return ((hpa_t)page_to_pfn(page) << PAGE_SHIFT) |
841 | | (gpa & (PAGE_SIZE-1)); | |
842 | } | |
843 | ||
844 | hpa_t gva_to_hpa(struct kvm_vcpu *vcpu, gva_t gva) | |
845 | { | |
846 | gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, gva); | |
847 | ||
848 | if (gpa == UNMAPPED_GVA) | |
849 | return UNMAPPED_GVA; | |
850 | return gpa_to_hpa(vcpu, gpa); | |
851 | } | |
852 | ||
039576c0 AK |
853 | struct page *gva_to_page(struct kvm_vcpu *vcpu, gva_t gva) |
854 | { | |
855 | gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, gva); | |
856 | ||
857 | if (gpa == UNMAPPED_GVA) | |
858 | return NULL; | |
859 | return pfn_to_page(gpa_to_hpa(vcpu, gpa) >> PAGE_SHIFT); | |
860 | } | |
861 | ||
6aa8b732 AK |
862 | static void nonpaging_new_cr3(struct kvm_vcpu *vcpu) |
863 | { | |
864 | } | |
865 | ||
866 | static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, hpa_t p) | |
867 | { | |
868 | int level = PT32E_ROOT_LEVEL; | |
869 | hpa_t table_addr = vcpu->mmu.root_hpa; | |
870 | ||
871 | for (; ; level--) { | |
872 | u32 index = PT64_INDEX(v, level); | |
873 | u64 *table; | |
cea0f0e7 | 874 | u64 pte; |
6aa8b732 AK |
875 | |
876 | ASSERT(VALID_PAGE(table_addr)); | |
877 | table = __va(table_addr); | |
878 | ||
879 | if (level == 1) { | |
cea0f0e7 | 880 | pte = table[index]; |
c7addb90 | 881 | if (is_shadow_present_pte(pte) && is_writeble_pte(pte)) |
cea0f0e7 | 882 | return 0; |
6aa8b732 AK |
883 | mark_page_dirty(vcpu->kvm, v >> PAGE_SHIFT); |
884 | page_header_update_slot(vcpu->kvm, table, v); | |
885 | table[index] = p | PT_PRESENT_MASK | PT_WRITABLE_MASK | | |
886 | PT_USER_MASK; | |
290fc38d | 887 | rmap_add(vcpu, &table[index], v >> PAGE_SHIFT); |
6aa8b732 AK |
888 | return 0; |
889 | } | |
890 | ||
c7addb90 | 891 | if (table[index] == shadow_trap_nonpresent_pte) { |
25c0de2c | 892 | struct kvm_mmu_page *new_table; |
cea0f0e7 | 893 | gfn_t pseudo_gfn; |
6aa8b732 | 894 | |
cea0f0e7 AK |
895 | pseudo_gfn = (v & PT64_DIR_BASE_ADDR_MASK) |
896 | >> PAGE_SHIFT; | |
897 | new_table = kvm_mmu_get_page(vcpu, pseudo_gfn, | |
898 | v, level - 1, | |
d28c6cfb | 899 | 1, 0, &table[index]); |
25c0de2c | 900 | if (!new_table) { |
6aa8b732 AK |
901 | pgprintk("nonpaging_map: ENOMEM\n"); |
902 | return -ENOMEM; | |
903 | } | |
904 | ||
47ad8e68 | 905 | table[index] = __pa(new_table->spt) | PT_PRESENT_MASK |
25c0de2c | 906 | | PT_WRITABLE_MASK | PT_USER_MASK; |
6aa8b732 AK |
907 | } |
908 | table_addr = table[index] & PT64_BASE_ADDR_MASK; | |
909 | } | |
910 | } | |
911 | ||
c7addb90 AK |
912 | static void nonpaging_prefetch_page(struct kvm_vcpu *vcpu, |
913 | struct kvm_mmu_page *sp) | |
914 | { | |
915 | int i; | |
916 | ||
917 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) | |
918 | sp->spt[i] = shadow_trap_nonpresent_pte; | |
919 | } | |
920 | ||
17ac10ad AK |
921 | static void mmu_free_roots(struct kvm_vcpu *vcpu) |
922 | { | |
923 | int i; | |
3bb65a22 | 924 | struct kvm_mmu_page *page; |
17ac10ad | 925 | |
7b53aa56 AK |
926 | if (!VALID_PAGE(vcpu->mmu.root_hpa)) |
927 | return; | |
17ac10ad AK |
928 | #ifdef CONFIG_X86_64 |
929 | if (vcpu->mmu.shadow_root_level == PT64_ROOT_LEVEL) { | |
930 | hpa_t root = vcpu->mmu.root_hpa; | |
931 | ||
3bb65a22 AK |
932 | page = page_header(root); |
933 | --page->root_count; | |
17ac10ad AK |
934 | vcpu->mmu.root_hpa = INVALID_PAGE; |
935 | return; | |
936 | } | |
937 | #endif | |
938 | for (i = 0; i < 4; ++i) { | |
939 | hpa_t root = vcpu->mmu.pae_root[i]; | |
940 | ||
417726a3 | 941 | if (root) { |
417726a3 AK |
942 | root &= PT64_BASE_ADDR_MASK; |
943 | page = page_header(root); | |
944 | --page->root_count; | |
945 | } | |
17ac10ad AK |
946 | vcpu->mmu.pae_root[i] = INVALID_PAGE; |
947 | } | |
948 | vcpu->mmu.root_hpa = INVALID_PAGE; | |
949 | } | |
950 | ||
951 | static void mmu_alloc_roots(struct kvm_vcpu *vcpu) | |
952 | { | |
953 | int i; | |
cea0f0e7 | 954 | gfn_t root_gfn; |
3bb65a22 AK |
955 | struct kvm_mmu_page *page; |
956 | ||
cea0f0e7 | 957 | root_gfn = vcpu->cr3 >> PAGE_SHIFT; |
17ac10ad AK |
958 | |
959 | #ifdef CONFIG_X86_64 | |
960 | if (vcpu->mmu.shadow_root_level == PT64_ROOT_LEVEL) { | |
961 | hpa_t root = vcpu->mmu.root_hpa; | |
962 | ||
963 | ASSERT(!VALID_PAGE(root)); | |
68a99f6d | 964 | page = kvm_mmu_get_page(vcpu, root_gfn, 0, |
d28c6cfb | 965 | PT64_ROOT_LEVEL, 0, 0, NULL); |
47ad8e68 | 966 | root = __pa(page->spt); |
3bb65a22 | 967 | ++page->root_count; |
17ac10ad AK |
968 | vcpu->mmu.root_hpa = root; |
969 | return; | |
970 | } | |
971 | #endif | |
972 | for (i = 0; i < 4; ++i) { | |
973 | hpa_t root = vcpu->mmu.pae_root[i]; | |
974 | ||
975 | ASSERT(!VALID_PAGE(root)); | |
417726a3 AK |
976 | if (vcpu->mmu.root_level == PT32E_ROOT_LEVEL) { |
977 | if (!is_present_pte(vcpu->pdptrs[i])) { | |
978 | vcpu->mmu.pae_root[i] = 0; | |
979 | continue; | |
980 | } | |
cea0f0e7 | 981 | root_gfn = vcpu->pdptrs[i] >> PAGE_SHIFT; |
417726a3 | 982 | } else if (vcpu->mmu.root_level == 0) |
cea0f0e7 | 983 | root_gfn = 0; |
68a99f6d | 984 | page = kvm_mmu_get_page(vcpu, root_gfn, i << 30, |
cea0f0e7 | 985 | PT32_ROOT_LEVEL, !is_paging(vcpu), |
d28c6cfb | 986 | 0, NULL); |
47ad8e68 | 987 | root = __pa(page->spt); |
3bb65a22 | 988 | ++page->root_count; |
17ac10ad AK |
989 | vcpu->mmu.pae_root[i] = root | PT_PRESENT_MASK; |
990 | } | |
991 | vcpu->mmu.root_hpa = __pa(vcpu->mmu.pae_root); | |
992 | } | |
993 | ||
6aa8b732 AK |
994 | static gpa_t nonpaging_gva_to_gpa(struct kvm_vcpu *vcpu, gva_t vaddr) |
995 | { | |
996 | return vaddr; | |
997 | } | |
998 | ||
999 | static int nonpaging_page_fault(struct kvm_vcpu *vcpu, gva_t gva, | |
1000 | u32 error_code) | |
1001 | { | |
6aa8b732 | 1002 | gpa_t addr = gva; |
ebeace86 | 1003 | hpa_t paddr; |
e2dec939 | 1004 | int r; |
6aa8b732 | 1005 | |
e2dec939 AK |
1006 | r = mmu_topup_memory_caches(vcpu); |
1007 | if (r) | |
1008 | return r; | |
714b93da | 1009 | |
6aa8b732 AK |
1010 | ASSERT(vcpu); |
1011 | ASSERT(VALID_PAGE(vcpu->mmu.root_hpa)); | |
1012 | ||
6aa8b732 | 1013 | |
ebeace86 | 1014 | paddr = gpa_to_hpa(vcpu , addr & PT64_BASE_ADDR_MASK); |
6aa8b732 | 1015 | |
ebeace86 AK |
1016 | if (is_error_hpa(paddr)) |
1017 | return 1; | |
6aa8b732 | 1018 | |
ebeace86 | 1019 | return nonpaging_map(vcpu, addr & PAGE_MASK, paddr); |
6aa8b732 AK |
1020 | } |
1021 | ||
6aa8b732 AK |
1022 | static void nonpaging_free(struct kvm_vcpu *vcpu) |
1023 | { | |
17ac10ad | 1024 | mmu_free_roots(vcpu); |
6aa8b732 AK |
1025 | } |
1026 | ||
1027 | static int nonpaging_init_context(struct kvm_vcpu *vcpu) | |
1028 | { | |
1029 | struct kvm_mmu *context = &vcpu->mmu; | |
1030 | ||
1031 | context->new_cr3 = nonpaging_new_cr3; | |
1032 | context->page_fault = nonpaging_page_fault; | |
6aa8b732 AK |
1033 | context->gva_to_gpa = nonpaging_gva_to_gpa; |
1034 | context->free = nonpaging_free; | |
c7addb90 | 1035 | context->prefetch_page = nonpaging_prefetch_page; |
cea0f0e7 | 1036 | context->root_level = 0; |
6aa8b732 | 1037 | context->shadow_root_level = PT32E_ROOT_LEVEL; |
17c3ba9d | 1038 | context->root_hpa = INVALID_PAGE; |
6aa8b732 AK |
1039 | return 0; |
1040 | } | |
1041 | ||
6aa8b732 AK |
1042 | static void kvm_mmu_flush_tlb(struct kvm_vcpu *vcpu) |
1043 | { | |
1165f5fe | 1044 | ++vcpu->stat.tlb_flush; |
cbdd1bea | 1045 | kvm_x86_ops->tlb_flush(vcpu); |
6aa8b732 AK |
1046 | } |
1047 | ||
1048 | static void paging_new_cr3(struct kvm_vcpu *vcpu) | |
1049 | { | |
374cbac0 | 1050 | pgprintk("%s: cr3 %lx\n", __FUNCTION__, vcpu->cr3); |
cea0f0e7 | 1051 | mmu_free_roots(vcpu); |
6aa8b732 AK |
1052 | } |
1053 | ||
6aa8b732 AK |
1054 | static void inject_page_fault(struct kvm_vcpu *vcpu, |
1055 | u64 addr, | |
1056 | u32 err_code) | |
1057 | { | |
cbdd1bea | 1058 | kvm_x86_ops->inject_page_fault(vcpu, addr, err_code); |
6aa8b732 AK |
1059 | } |
1060 | ||
6aa8b732 AK |
1061 | static void paging_free(struct kvm_vcpu *vcpu) |
1062 | { | |
1063 | nonpaging_free(vcpu); | |
1064 | } | |
1065 | ||
1066 | #define PTTYPE 64 | |
1067 | #include "paging_tmpl.h" | |
1068 | #undef PTTYPE | |
1069 | ||
1070 | #define PTTYPE 32 | |
1071 | #include "paging_tmpl.h" | |
1072 | #undef PTTYPE | |
1073 | ||
17ac10ad | 1074 | static int paging64_init_context_common(struct kvm_vcpu *vcpu, int level) |
6aa8b732 AK |
1075 | { |
1076 | struct kvm_mmu *context = &vcpu->mmu; | |
1077 | ||
1078 | ASSERT(is_pae(vcpu)); | |
1079 | context->new_cr3 = paging_new_cr3; | |
1080 | context->page_fault = paging64_page_fault; | |
6aa8b732 | 1081 | context->gva_to_gpa = paging64_gva_to_gpa; |
c7addb90 | 1082 | context->prefetch_page = paging64_prefetch_page; |
6aa8b732 | 1083 | context->free = paging_free; |
17ac10ad AK |
1084 | context->root_level = level; |
1085 | context->shadow_root_level = level; | |
17c3ba9d | 1086 | context->root_hpa = INVALID_PAGE; |
6aa8b732 AK |
1087 | return 0; |
1088 | } | |
1089 | ||
17ac10ad AK |
1090 | static int paging64_init_context(struct kvm_vcpu *vcpu) |
1091 | { | |
1092 | return paging64_init_context_common(vcpu, PT64_ROOT_LEVEL); | |
1093 | } | |
1094 | ||
6aa8b732 AK |
1095 | static int paging32_init_context(struct kvm_vcpu *vcpu) |
1096 | { | |
1097 | struct kvm_mmu *context = &vcpu->mmu; | |
1098 | ||
1099 | context->new_cr3 = paging_new_cr3; | |
1100 | context->page_fault = paging32_page_fault; | |
6aa8b732 AK |
1101 | context->gva_to_gpa = paging32_gva_to_gpa; |
1102 | context->free = paging_free; | |
c7addb90 | 1103 | context->prefetch_page = paging32_prefetch_page; |
6aa8b732 AK |
1104 | context->root_level = PT32_ROOT_LEVEL; |
1105 | context->shadow_root_level = PT32E_ROOT_LEVEL; | |
17c3ba9d | 1106 | context->root_hpa = INVALID_PAGE; |
6aa8b732 AK |
1107 | return 0; |
1108 | } | |
1109 | ||
1110 | static int paging32E_init_context(struct kvm_vcpu *vcpu) | |
1111 | { | |
17ac10ad | 1112 | return paging64_init_context_common(vcpu, PT32E_ROOT_LEVEL); |
6aa8b732 AK |
1113 | } |
1114 | ||
1115 | static int init_kvm_mmu(struct kvm_vcpu *vcpu) | |
1116 | { | |
1117 | ASSERT(vcpu); | |
1118 | ASSERT(!VALID_PAGE(vcpu->mmu.root_hpa)); | |
1119 | ||
1120 | if (!is_paging(vcpu)) | |
1121 | return nonpaging_init_context(vcpu); | |
a9058ecd | 1122 | else if (is_long_mode(vcpu)) |
6aa8b732 AK |
1123 | return paging64_init_context(vcpu); |
1124 | else if (is_pae(vcpu)) | |
1125 | return paging32E_init_context(vcpu); | |
1126 | else | |
1127 | return paging32_init_context(vcpu); | |
1128 | } | |
1129 | ||
1130 | static void destroy_kvm_mmu(struct kvm_vcpu *vcpu) | |
1131 | { | |
1132 | ASSERT(vcpu); | |
1133 | if (VALID_PAGE(vcpu->mmu.root_hpa)) { | |
1134 | vcpu->mmu.free(vcpu); | |
1135 | vcpu->mmu.root_hpa = INVALID_PAGE; | |
1136 | } | |
1137 | } | |
1138 | ||
1139 | int kvm_mmu_reset_context(struct kvm_vcpu *vcpu) | |
17c3ba9d AK |
1140 | { |
1141 | destroy_kvm_mmu(vcpu); | |
1142 | return init_kvm_mmu(vcpu); | |
1143 | } | |
8668a3c4 | 1144 | EXPORT_SYMBOL_GPL(kvm_mmu_reset_context); |
17c3ba9d AK |
1145 | |
1146 | int kvm_mmu_load(struct kvm_vcpu *vcpu) | |
6aa8b732 | 1147 | { |
714b93da AK |
1148 | int r; |
1149 | ||
11ec2804 | 1150 | mutex_lock(&vcpu->kvm->lock); |
e2dec939 | 1151 | r = mmu_topup_memory_caches(vcpu); |
17c3ba9d AK |
1152 | if (r) |
1153 | goto out; | |
1154 | mmu_alloc_roots(vcpu); | |
cbdd1bea | 1155 | kvm_x86_ops->set_cr3(vcpu, vcpu->mmu.root_hpa); |
17c3ba9d | 1156 | kvm_mmu_flush_tlb(vcpu); |
714b93da | 1157 | out: |
11ec2804 | 1158 | mutex_unlock(&vcpu->kvm->lock); |
714b93da | 1159 | return r; |
6aa8b732 | 1160 | } |
17c3ba9d AK |
1161 | EXPORT_SYMBOL_GPL(kvm_mmu_load); |
1162 | ||
1163 | void kvm_mmu_unload(struct kvm_vcpu *vcpu) | |
1164 | { | |
1165 | mmu_free_roots(vcpu); | |
1166 | } | |
6aa8b732 | 1167 | |
09072daf | 1168 | static void mmu_pte_write_zap_pte(struct kvm_vcpu *vcpu, |
ac1b714e AK |
1169 | struct kvm_mmu_page *page, |
1170 | u64 *spte) | |
1171 | { | |
1172 | u64 pte; | |
1173 | struct kvm_mmu_page *child; | |
1174 | ||
1175 | pte = *spte; | |
c7addb90 | 1176 | if (is_shadow_present_pte(pte)) { |
ac1b714e | 1177 | if (page->role.level == PT_PAGE_TABLE_LEVEL) |
290fc38d | 1178 | rmap_remove(vcpu->kvm, spte); |
ac1b714e AK |
1179 | else { |
1180 | child = page_header(pte & PT64_BASE_ADDR_MASK); | |
90cb0529 | 1181 | mmu_page_remove_parent_pte(child, spte); |
ac1b714e AK |
1182 | } |
1183 | } | |
c7addb90 | 1184 | set_shadow_pte(spte, shadow_trap_nonpresent_pte); |
d9e368d6 | 1185 | kvm_flush_remote_tlbs(vcpu->kvm); |
ac1b714e AK |
1186 | } |
1187 | ||
0028425f AK |
1188 | static void mmu_pte_write_new_pte(struct kvm_vcpu *vcpu, |
1189 | struct kvm_mmu_page *page, | |
1190 | u64 *spte, | |
c7addb90 AK |
1191 | const void *new, int bytes, |
1192 | int offset_in_pte) | |
0028425f AK |
1193 | { |
1194 | if (page->role.level != PT_PAGE_TABLE_LEVEL) | |
1195 | return; | |
1196 | ||
1197 | if (page->role.glevels == PT32_ROOT_LEVEL) | |
c7addb90 AK |
1198 | paging32_update_pte(vcpu, page, spte, new, bytes, |
1199 | offset_in_pte); | |
0028425f | 1200 | else |
c7addb90 AK |
1201 | paging64_update_pte(vcpu, page, spte, new, bytes, |
1202 | offset_in_pte); | |
0028425f AK |
1203 | } |
1204 | ||
12b7d28f AK |
1205 | static bool last_updated_pte_accessed(struct kvm_vcpu *vcpu) |
1206 | { | |
1207 | u64 *spte = vcpu->last_pte_updated; | |
1208 | ||
1209 | return !!(spte && (*spte & PT_ACCESSED_MASK)); | |
1210 | } | |
1211 | ||
09072daf | 1212 | void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa, |
fe551881 | 1213 | const u8 *new, int bytes) |
da4a00f0 | 1214 | { |
9b7a0325 AK |
1215 | gfn_t gfn = gpa >> PAGE_SHIFT; |
1216 | struct kvm_mmu_page *page; | |
0e7bc4b9 | 1217 | struct hlist_node *node, *n; |
9b7a0325 AK |
1218 | struct hlist_head *bucket; |
1219 | unsigned index; | |
1220 | u64 *spte; | |
9b7a0325 | 1221 | unsigned offset = offset_in_page(gpa); |
0e7bc4b9 | 1222 | unsigned pte_size; |
9b7a0325 | 1223 | unsigned page_offset; |
0e7bc4b9 | 1224 | unsigned misaligned; |
fce0657f | 1225 | unsigned quadrant; |
9b7a0325 | 1226 | int level; |
86a5ba02 | 1227 | int flooded = 0; |
ac1b714e | 1228 | int npte; |
9b7a0325 | 1229 | |
da4a00f0 | 1230 | pgprintk("%s: gpa %llx bytes %d\n", __FUNCTION__, gpa, bytes); |
c7addb90 | 1231 | kvm_mmu_audit(vcpu, "pre pte write"); |
12b7d28f AK |
1232 | if (gfn == vcpu->last_pt_write_gfn |
1233 | && !last_updated_pte_accessed(vcpu)) { | |
86a5ba02 AK |
1234 | ++vcpu->last_pt_write_count; |
1235 | if (vcpu->last_pt_write_count >= 3) | |
1236 | flooded = 1; | |
1237 | } else { | |
1238 | vcpu->last_pt_write_gfn = gfn; | |
1239 | vcpu->last_pt_write_count = 1; | |
12b7d28f | 1240 | vcpu->last_pte_updated = NULL; |
86a5ba02 | 1241 | } |
9b7a0325 AK |
1242 | index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES; |
1243 | bucket = &vcpu->kvm->mmu_page_hash[index]; | |
0e7bc4b9 | 1244 | hlist_for_each_entry_safe(page, node, n, bucket, hash_link) { |
9b7a0325 AK |
1245 | if (page->gfn != gfn || page->role.metaphysical) |
1246 | continue; | |
0e7bc4b9 AK |
1247 | pte_size = page->role.glevels == PT32_ROOT_LEVEL ? 4 : 8; |
1248 | misaligned = (offset ^ (offset + bytes - 1)) & ~(pte_size - 1); | |
e925c5ba | 1249 | misaligned |= bytes < 4; |
86a5ba02 | 1250 | if (misaligned || flooded) { |
0e7bc4b9 AK |
1251 | /* |
1252 | * Misaligned accesses are too much trouble to fix | |
1253 | * up; also, they usually indicate a page is not used | |
1254 | * as a page table. | |
86a5ba02 AK |
1255 | * |
1256 | * If we're seeing too many writes to a page, | |
1257 | * it may no longer be a page table, or we may be | |
1258 | * forking, in which case it is better to unmap the | |
1259 | * page. | |
0e7bc4b9 AK |
1260 | */ |
1261 | pgprintk("misaligned: gpa %llx bytes %d role %x\n", | |
1262 | gpa, bytes, page->role.word); | |
90cb0529 | 1263 | kvm_mmu_zap_page(vcpu->kvm, page); |
0e7bc4b9 AK |
1264 | continue; |
1265 | } | |
9b7a0325 AK |
1266 | page_offset = offset; |
1267 | level = page->role.level; | |
ac1b714e | 1268 | npte = 1; |
9b7a0325 | 1269 | if (page->role.glevels == PT32_ROOT_LEVEL) { |
ac1b714e AK |
1270 | page_offset <<= 1; /* 32->64 */ |
1271 | /* | |
1272 | * A 32-bit pde maps 4MB while the shadow pdes map | |
1273 | * only 2MB. So we need to double the offset again | |
1274 | * and zap two pdes instead of one. | |
1275 | */ | |
1276 | if (level == PT32_ROOT_LEVEL) { | |
6b8d0f9b | 1277 | page_offset &= ~7; /* kill rounding error */ |
ac1b714e AK |
1278 | page_offset <<= 1; |
1279 | npte = 2; | |
1280 | } | |
fce0657f | 1281 | quadrant = page_offset >> PAGE_SHIFT; |
9b7a0325 | 1282 | page_offset &= ~PAGE_MASK; |
fce0657f AK |
1283 | if (quadrant != page->role.quadrant) |
1284 | continue; | |
9b7a0325 | 1285 | } |
47ad8e68 | 1286 | spte = &page->spt[page_offset / sizeof(*spte)]; |
ac1b714e | 1287 | while (npte--) { |
09072daf | 1288 | mmu_pte_write_zap_pte(vcpu, page, spte); |
c7addb90 AK |
1289 | mmu_pte_write_new_pte(vcpu, page, spte, new, bytes, |
1290 | page_offset & (pte_size - 1)); | |
ac1b714e | 1291 | ++spte; |
9b7a0325 | 1292 | } |
9b7a0325 | 1293 | } |
c7addb90 | 1294 | kvm_mmu_audit(vcpu, "post pte write"); |
da4a00f0 AK |
1295 | } |
1296 | ||
a436036b AK |
1297 | int kvm_mmu_unprotect_page_virt(struct kvm_vcpu *vcpu, gva_t gva) |
1298 | { | |
1299 | gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, gva); | |
1300 | ||
1301 | return kvm_mmu_unprotect_page(vcpu, gpa >> PAGE_SHIFT); | |
1302 | } | |
1303 | ||
22d95b12 | 1304 | void __kvm_mmu_free_some_pages(struct kvm_vcpu *vcpu) |
ebeace86 AK |
1305 | { |
1306 | while (vcpu->kvm->n_free_mmu_pages < KVM_REFILL_PAGES) { | |
1307 | struct kvm_mmu_page *page; | |
1308 | ||
1309 | page = container_of(vcpu->kvm->active_mmu_pages.prev, | |
1310 | struct kvm_mmu_page, link); | |
90cb0529 | 1311 | kvm_mmu_zap_page(vcpu->kvm, page); |
ebeace86 AK |
1312 | } |
1313 | } | |
ebeace86 | 1314 | |
6aa8b732 AK |
1315 | static void free_mmu_pages(struct kvm_vcpu *vcpu) |
1316 | { | |
f51234c2 | 1317 | struct kvm_mmu_page *page; |
6aa8b732 | 1318 | |
f51234c2 AK |
1319 | while (!list_empty(&vcpu->kvm->active_mmu_pages)) { |
1320 | page = container_of(vcpu->kvm->active_mmu_pages.next, | |
1321 | struct kvm_mmu_page, link); | |
90cb0529 | 1322 | kvm_mmu_zap_page(vcpu->kvm, page); |
f51234c2 | 1323 | } |
17ac10ad | 1324 | free_page((unsigned long)vcpu->mmu.pae_root); |
6aa8b732 AK |
1325 | } |
1326 | ||
1327 | static int alloc_mmu_pages(struct kvm_vcpu *vcpu) | |
1328 | { | |
17ac10ad | 1329 | struct page *page; |
6aa8b732 AK |
1330 | int i; |
1331 | ||
1332 | ASSERT(vcpu); | |
1333 | ||
82ce2c96 IE |
1334 | if (vcpu->kvm->n_requested_mmu_pages) |
1335 | vcpu->kvm->n_free_mmu_pages = vcpu->kvm->n_requested_mmu_pages; | |
1336 | else | |
1337 | vcpu->kvm->n_free_mmu_pages = vcpu->kvm->n_alloc_mmu_pages; | |
17ac10ad AK |
1338 | /* |
1339 | * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64. | |
1340 | * Therefore we need to allocate shadow page tables in the first | |
1341 | * 4GB of memory, which happens to fit the DMA32 zone. | |
1342 | */ | |
1343 | page = alloc_page(GFP_KERNEL | __GFP_DMA32); | |
1344 | if (!page) | |
1345 | goto error_1; | |
1346 | vcpu->mmu.pae_root = page_address(page); | |
1347 | for (i = 0; i < 4; ++i) | |
1348 | vcpu->mmu.pae_root[i] = INVALID_PAGE; | |
1349 | ||
6aa8b732 AK |
1350 | return 0; |
1351 | ||
1352 | error_1: | |
1353 | free_mmu_pages(vcpu); | |
1354 | return -ENOMEM; | |
1355 | } | |
1356 | ||
8018c27b | 1357 | int kvm_mmu_create(struct kvm_vcpu *vcpu) |
6aa8b732 | 1358 | { |
6aa8b732 AK |
1359 | ASSERT(vcpu); |
1360 | ASSERT(!VALID_PAGE(vcpu->mmu.root_hpa)); | |
6aa8b732 | 1361 | |
8018c27b IM |
1362 | return alloc_mmu_pages(vcpu); |
1363 | } | |
6aa8b732 | 1364 | |
8018c27b IM |
1365 | int kvm_mmu_setup(struct kvm_vcpu *vcpu) |
1366 | { | |
1367 | ASSERT(vcpu); | |
1368 | ASSERT(!VALID_PAGE(vcpu->mmu.root_hpa)); | |
2c264957 | 1369 | |
8018c27b | 1370 | return init_kvm_mmu(vcpu); |
6aa8b732 AK |
1371 | } |
1372 | ||
1373 | void kvm_mmu_destroy(struct kvm_vcpu *vcpu) | |
1374 | { | |
1375 | ASSERT(vcpu); | |
1376 | ||
1377 | destroy_kvm_mmu(vcpu); | |
1378 | free_mmu_pages(vcpu); | |
714b93da | 1379 | mmu_free_memory_caches(vcpu); |
6aa8b732 AK |
1380 | } |
1381 | ||
90cb0529 | 1382 | void kvm_mmu_slot_remove_write_access(struct kvm *kvm, int slot) |
6aa8b732 AK |
1383 | { |
1384 | struct kvm_mmu_page *page; | |
1385 | ||
1386 | list_for_each_entry(page, &kvm->active_mmu_pages, link) { | |
1387 | int i; | |
1388 | u64 *pt; | |
1389 | ||
1390 | if (!test_bit(slot, &page->slot_bitmap)) | |
1391 | continue; | |
1392 | ||
47ad8e68 | 1393 | pt = page->spt; |
6aa8b732 AK |
1394 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) |
1395 | /* avoid RMW */ | |
cd4a4e53 | 1396 | if (pt[i] & PT_WRITABLE_MASK) { |
290fc38d | 1397 | rmap_remove(kvm, &pt[i]); |
6aa8b732 | 1398 | pt[i] &= ~PT_WRITABLE_MASK; |
cd4a4e53 | 1399 | } |
6aa8b732 AK |
1400 | } |
1401 | } | |
37a7d8b0 | 1402 | |
90cb0529 | 1403 | void kvm_mmu_zap_all(struct kvm *kvm) |
e0fa826f | 1404 | { |
90cb0529 | 1405 | struct kvm_mmu_page *page, *node; |
e0fa826f | 1406 | |
90cb0529 AK |
1407 | list_for_each_entry_safe(page, node, &kvm->active_mmu_pages, link) |
1408 | kvm_mmu_zap_page(kvm, page); | |
e0fa826f | 1409 | |
90cb0529 | 1410 | kvm_flush_remote_tlbs(kvm); |
e0fa826f DL |
1411 | } |
1412 | ||
b5a33a75 AK |
1413 | void kvm_mmu_module_exit(void) |
1414 | { | |
1415 | if (pte_chain_cache) | |
1416 | kmem_cache_destroy(pte_chain_cache); | |
1417 | if (rmap_desc_cache) | |
1418 | kmem_cache_destroy(rmap_desc_cache); | |
d3d25b04 AK |
1419 | if (mmu_page_header_cache) |
1420 | kmem_cache_destroy(mmu_page_header_cache); | |
b5a33a75 AK |
1421 | } |
1422 | ||
1423 | int kvm_mmu_module_init(void) | |
1424 | { | |
1425 | pte_chain_cache = kmem_cache_create("kvm_pte_chain", | |
1426 | sizeof(struct kvm_pte_chain), | |
20c2df83 | 1427 | 0, 0, NULL); |
b5a33a75 AK |
1428 | if (!pte_chain_cache) |
1429 | goto nomem; | |
1430 | rmap_desc_cache = kmem_cache_create("kvm_rmap_desc", | |
1431 | sizeof(struct kvm_rmap_desc), | |
20c2df83 | 1432 | 0, 0, NULL); |
b5a33a75 AK |
1433 | if (!rmap_desc_cache) |
1434 | goto nomem; | |
1435 | ||
d3d25b04 AK |
1436 | mmu_page_header_cache = kmem_cache_create("kvm_mmu_page_header", |
1437 | sizeof(struct kvm_mmu_page), | |
20c2df83 | 1438 | 0, 0, NULL); |
d3d25b04 AK |
1439 | if (!mmu_page_header_cache) |
1440 | goto nomem; | |
1441 | ||
b5a33a75 AK |
1442 | return 0; |
1443 | ||
1444 | nomem: | |
1445 | kvm_mmu_module_exit(); | |
1446 | return -ENOMEM; | |
1447 | } | |
1448 | ||
37a7d8b0 AK |
1449 | #ifdef AUDIT |
1450 | ||
1451 | static const char *audit_msg; | |
1452 | ||
1453 | static gva_t canonicalize(gva_t gva) | |
1454 | { | |
1455 | #ifdef CONFIG_X86_64 | |
1456 | gva = (long long)(gva << 16) >> 16; | |
1457 | #endif | |
1458 | return gva; | |
1459 | } | |
1460 | ||
1461 | static void audit_mappings_page(struct kvm_vcpu *vcpu, u64 page_pte, | |
1462 | gva_t va, int level) | |
1463 | { | |
1464 | u64 *pt = __va(page_pte & PT64_BASE_ADDR_MASK); | |
1465 | int i; | |
1466 | gva_t va_delta = 1ul << (PAGE_SHIFT + 9 * (level - 1)); | |
1467 | ||
1468 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i, va += va_delta) { | |
1469 | u64 ent = pt[i]; | |
1470 | ||
c7addb90 | 1471 | if (ent == shadow_trap_nonpresent_pte) |
37a7d8b0 AK |
1472 | continue; |
1473 | ||
1474 | va = canonicalize(va); | |
c7addb90 AK |
1475 | if (level > 1) { |
1476 | if (ent == shadow_notrap_nonpresent_pte) | |
1477 | printk(KERN_ERR "audit: (%s) nontrapping pte" | |
1478 | " in nonleaf level: levels %d gva %lx" | |
1479 | " level %d pte %llx\n", audit_msg, | |
1480 | vcpu->mmu.root_level, va, level, ent); | |
1481 | ||
37a7d8b0 | 1482 | audit_mappings_page(vcpu, ent, va, level - 1); |
c7addb90 | 1483 | } else { |
37a7d8b0 AK |
1484 | gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, va); |
1485 | hpa_t hpa = gpa_to_hpa(vcpu, gpa); | |
1486 | ||
c7addb90 | 1487 | if (is_shadow_present_pte(ent) |
37a7d8b0 | 1488 | && (ent & PT64_BASE_ADDR_MASK) != hpa) |
c7addb90 AK |
1489 | printk(KERN_ERR "xx audit error: (%s) levels %d" |
1490 | " gva %lx gpa %llx hpa %llx ent %llx %d\n", | |
37a7d8b0 | 1491 | audit_msg, vcpu->mmu.root_level, |
c7addb90 AK |
1492 | va, gpa, hpa, ent, is_shadow_present_pte(ent)); |
1493 | else if (ent == shadow_notrap_nonpresent_pte | |
1494 | && !is_error_hpa(hpa)) | |
1495 | printk(KERN_ERR "audit: (%s) notrap shadow," | |
1496 | " valid guest gva %lx\n", audit_msg, va); | |
1497 | ||
37a7d8b0 AK |
1498 | } |
1499 | } | |
1500 | } | |
1501 | ||
1502 | static void audit_mappings(struct kvm_vcpu *vcpu) | |
1503 | { | |
1ea252af | 1504 | unsigned i; |
37a7d8b0 AK |
1505 | |
1506 | if (vcpu->mmu.root_level == 4) | |
1507 | audit_mappings_page(vcpu, vcpu->mmu.root_hpa, 0, 4); | |
1508 | else | |
1509 | for (i = 0; i < 4; ++i) | |
1510 | if (vcpu->mmu.pae_root[i] & PT_PRESENT_MASK) | |
1511 | audit_mappings_page(vcpu, | |
1512 | vcpu->mmu.pae_root[i], | |
1513 | i << 30, | |
1514 | 2); | |
1515 | } | |
1516 | ||
1517 | static int count_rmaps(struct kvm_vcpu *vcpu) | |
1518 | { | |
1519 | int nmaps = 0; | |
1520 | int i, j, k; | |
1521 | ||
1522 | for (i = 0; i < KVM_MEMORY_SLOTS; ++i) { | |
1523 | struct kvm_memory_slot *m = &vcpu->kvm->memslots[i]; | |
1524 | struct kvm_rmap_desc *d; | |
1525 | ||
1526 | for (j = 0; j < m->npages; ++j) { | |
290fc38d | 1527 | unsigned long *rmapp = &m->rmap[j]; |
37a7d8b0 | 1528 | |
290fc38d | 1529 | if (!*rmapp) |
37a7d8b0 | 1530 | continue; |
290fc38d | 1531 | if (!(*rmapp & 1)) { |
37a7d8b0 AK |
1532 | ++nmaps; |
1533 | continue; | |
1534 | } | |
290fc38d | 1535 | d = (struct kvm_rmap_desc *)(*rmapp & ~1ul); |
37a7d8b0 AK |
1536 | while (d) { |
1537 | for (k = 0; k < RMAP_EXT; ++k) | |
1538 | if (d->shadow_ptes[k]) | |
1539 | ++nmaps; | |
1540 | else | |
1541 | break; | |
1542 | d = d->more; | |
1543 | } | |
1544 | } | |
1545 | } | |
1546 | return nmaps; | |
1547 | } | |
1548 | ||
1549 | static int count_writable_mappings(struct kvm_vcpu *vcpu) | |
1550 | { | |
1551 | int nmaps = 0; | |
1552 | struct kvm_mmu_page *page; | |
1553 | int i; | |
1554 | ||
1555 | list_for_each_entry(page, &vcpu->kvm->active_mmu_pages, link) { | |
47ad8e68 | 1556 | u64 *pt = page->spt; |
37a7d8b0 AK |
1557 | |
1558 | if (page->role.level != PT_PAGE_TABLE_LEVEL) | |
1559 | continue; | |
1560 | ||
1561 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) { | |
1562 | u64 ent = pt[i]; | |
1563 | ||
1564 | if (!(ent & PT_PRESENT_MASK)) | |
1565 | continue; | |
1566 | if (!(ent & PT_WRITABLE_MASK)) | |
1567 | continue; | |
1568 | ++nmaps; | |
1569 | } | |
1570 | } | |
1571 | return nmaps; | |
1572 | } | |
1573 | ||
1574 | static void audit_rmap(struct kvm_vcpu *vcpu) | |
1575 | { | |
1576 | int n_rmap = count_rmaps(vcpu); | |
1577 | int n_actual = count_writable_mappings(vcpu); | |
1578 | ||
1579 | if (n_rmap != n_actual) | |
1580 | printk(KERN_ERR "%s: (%s) rmap %d actual %d\n", | |
1581 | __FUNCTION__, audit_msg, n_rmap, n_actual); | |
1582 | } | |
1583 | ||
1584 | static void audit_write_protection(struct kvm_vcpu *vcpu) | |
1585 | { | |
1586 | struct kvm_mmu_page *page; | |
290fc38d IE |
1587 | struct kvm_memory_slot *slot; |
1588 | unsigned long *rmapp; | |
1589 | gfn_t gfn; | |
37a7d8b0 AK |
1590 | |
1591 | list_for_each_entry(page, &vcpu->kvm->active_mmu_pages, link) { | |
37a7d8b0 AK |
1592 | if (page->role.metaphysical) |
1593 | continue; | |
1594 | ||
290fc38d IE |
1595 | slot = gfn_to_memslot(vcpu->kvm, page->gfn); |
1596 | gfn = unalias_gfn(vcpu->kvm, page->gfn); | |
1597 | rmapp = &slot->rmap[gfn - slot->base_gfn]; | |
1598 | if (*rmapp) | |
37a7d8b0 AK |
1599 | printk(KERN_ERR "%s: (%s) shadow page has writable" |
1600 | " mappings: gfn %lx role %x\n", | |
1601 | __FUNCTION__, audit_msg, page->gfn, | |
1602 | page->role.word); | |
1603 | } | |
1604 | } | |
1605 | ||
1606 | static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg) | |
1607 | { | |
1608 | int olddbg = dbg; | |
1609 | ||
1610 | dbg = 0; | |
1611 | audit_msg = msg; | |
1612 | audit_rmap(vcpu); | |
1613 | audit_write_protection(vcpu); | |
1614 | audit_mappings(vcpu); | |
1615 | dbg = olddbg; | |
1616 | } | |
1617 | ||
1618 | #endif |