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" | |
34c16eec | 22 | #include "x86.h" |
e495606d | 23 | |
6aa8b732 AK |
24 | #include <linux/types.h> |
25 | #include <linux/string.h> | |
6aa8b732 AK |
26 | #include <linux/mm.h> |
27 | #include <linux/highmem.h> | |
28 | #include <linux/module.h> | |
29 | ||
e495606d AK |
30 | #include <asm/page.h> |
31 | #include <asm/cmpxchg.h> | |
6aa8b732 | 32 | |
37a7d8b0 AK |
33 | #undef MMU_DEBUG |
34 | ||
35 | #undef AUDIT | |
36 | ||
37 | #ifdef AUDIT | |
38 | static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg); | |
39 | #else | |
40 | static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg) {} | |
41 | #endif | |
42 | ||
43 | #ifdef MMU_DEBUG | |
44 | ||
45 | #define pgprintk(x...) do { if (dbg) printk(x); } while (0) | |
46 | #define rmap_printk(x...) do { if (dbg) printk(x); } while (0) | |
47 | ||
48 | #else | |
49 | ||
50 | #define pgprintk(x...) do { } while (0) | |
51 | #define rmap_printk(x...) do { } while (0) | |
52 | ||
53 | #endif | |
54 | ||
55 | #if defined(MMU_DEBUG) || defined(AUDIT) | |
56 | static int dbg = 1; | |
57 | #endif | |
6aa8b732 | 58 | |
d6c69ee9 YD |
59 | #ifndef MMU_DEBUG |
60 | #define ASSERT(x) do { } while (0) | |
61 | #else | |
6aa8b732 AK |
62 | #define ASSERT(x) \ |
63 | if (!(x)) { \ | |
64 | printk(KERN_WARNING "assertion failed %s:%d: %s\n", \ | |
65 | __FILE__, __LINE__, #x); \ | |
66 | } | |
d6c69ee9 | 67 | #endif |
6aa8b732 | 68 | |
cea0f0e7 AK |
69 | #define PT64_PT_BITS 9 |
70 | #define PT64_ENT_PER_PAGE (1 << PT64_PT_BITS) | |
71 | #define PT32_PT_BITS 10 | |
72 | #define PT32_ENT_PER_PAGE (1 << PT32_PT_BITS) | |
6aa8b732 AK |
73 | |
74 | #define PT_WRITABLE_SHIFT 1 | |
75 | ||
76 | #define PT_PRESENT_MASK (1ULL << 0) | |
77 | #define PT_WRITABLE_MASK (1ULL << PT_WRITABLE_SHIFT) | |
78 | #define PT_USER_MASK (1ULL << 2) | |
79 | #define PT_PWT_MASK (1ULL << 3) | |
80 | #define PT_PCD_MASK (1ULL << 4) | |
81 | #define PT_ACCESSED_MASK (1ULL << 5) | |
82 | #define PT_DIRTY_MASK (1ULL << 6) | |
83 | #define PT_PAGE_SIZE_MASK (1ULL << 7) | |
84 | #define PT_PAT_MASK (1ULL << 7) | |
85 | #define PT_GLOBAL_MASK (1ULL << 8) | |
86 | #define PT64_NX_MASK (1ULL << 63) | |
87 | ||
88 | #define PT_PAT_SHIFT 7 | |
89 | #define PT_DIR_PAT_SHIFT 12 | |
90 | #define PT_DIR_PAT_MASK (1ULL << PT_DIR_PAT_SHIFT) | |
91 | ||
92 | #define PT32_DIR_PSE36_SIZE 4 | |
93 | #define PT32_DIR_PSE36_SHIFT 13 | |
d77c26fc MD |
94 | #define PT32_DIR_PSE36_MASK \ |
95 | (((1ULL << PT32_DIR_PSE36_SIZE) - 1) << PT32_DIR_PSE36_SHIFT) | |
6aa8b732 AK |
96 | |
97 | ||
6aa8b732 AK |
98 | #define PT_FIRST_AVAIL_BITS_SHIFT 9 |
99 | #define PT64_SECOND_AVAIL_BITS_SHIFT 52 | |
100 | ||
6aa8b732 AK |
101 | #define PT_SHADOW_IO_MARK (1ULL << PT_FIRST_AVAIL_BITS_SHIFT) |
102 | ||
6aa8b732 AK |
103 | #define VALID_PAGE(x) ((x) != INVALID_PAGE) |
104 | ||
105 | #define PT64_LEVEL_BITS 9 | |
106 | ||
107 | #define PT64_LEVEL_SHIFT(level) \ | |
d77c26fc | 108 | (PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS) |
6aa8b732 AK |
109 | |
110 | #define PT64_LEVEL_MASK(level) \ | |
111 | (((1ULL << PT64_LEVEL_BITS) - 1) << PT64_LEVEL_SHIFT(level)) | |
112 | ||
113 | #define PT64_INDEX(address, level)\ | |
114 | (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1)) | |
115 | ||
116 | ||
117 | #define PT32_LEVEL_BITS 10 | |
118 | ||
119 | #define PT32_LEVEL_SHIFT(level) \ | |
d77c26fc | 120 | (PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS) |
6aa8b732 AK |
121 | |
122 | #define PT32_LEVEL_MASK(level) \ | |
123 | (((1ULL << PT32_LEVEL_BITS) - 1) << PT32_LEVEL_SHIFT(level)) | |
124 | ||
125 | #define PT32_INDEX(address, level)\ | |
126 | (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1)) | |
127 | ||
128 | ||
27aba766 | 129 | #define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1)) |
6aa8b732 AK |
130 | #define PT64_DIR_BASE_ADDR_MASK \ |
131 | (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1)) | |
132 | ||
133 | #define PT32_BASE_ADDR_MASK PAGE_MASK | |
134 | #define PT32_DIR_BASE_ADDR_MASK \ | |
135 | (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1)) | |
136 | ||
137 | ||
138 | #define PFERR_PRESENT_MASK (1U << 0) | |
139 | #define PFERR_WRITE_MASK (1U << 1) | |
140 | #define PFERR_USER_MASK (1U << 2) | |
73b1087e | 141 | #define PFERR_FETCH_MASK (1U << 4) |
6aa8b732 AK |
142 | |
143 | #define PT64_ROOT_LEVEL 4 | |
144 | #define PT32_ROOT_LEVEL 2 | |
145 | #define PT32E_ROOT_LEVEL 3 | |
146 | ||
147 | #define PT_DIRECTORY_LEVEL 2 | |
148 | #define PT_PAGE_TABLE_LEVEL 1 | |
149 | ||
cd4a4e53 AK |
150 | #define RMAP_EXT 4 |
151 | ||
152 | struct kvm_rmap_desc { | |
153 | u64 *shadow_ptes[RMAP_EXT]; | |
154 | struct kvm_rmap_desc *more; | |
155 | }; | |
156 | ||
b5a33a75 AK |
157 | static struct kmem_cache *pte_chain_cache; |
158 | static struct kmem_cache *rmap_desc_cache; | |
d3d25b04 | 159 | static struct kmem_cache *mmu_page_header_cache; |
b5a33a75 | 160 | |
c7addb90 AK |
161 | static u64 __read_mostly shadow_trap_nonpresent_pte; |
162 | static u64 __read_mostly shadow_notrap_nonpresent_pte; | |
163 | ||
164 | void kvm_mmu_set_nonpresent_ptes(u64 trap_pte, u64 notrap_pte) | |
165 | { | |
166 | shadow_trap_nonpresent_pte = trap_pte; | |
167 | shadow_notrap_nonpresent_pte = notrap_pte; | |
168 | } | |
169 | EXPORT_SYMBOL_GPL(kvm_mmu_set_nonpresent_ptes); | |
170 | ||
6aa8b732 AK |
171 | static int is_write_protection(struct kvm_vcpu *vcpu) |
172 | { | |
707d92fa | 173 | return vcpu->cr0 & X86_CR0_WP; |
6aa8b732 AK |
174 | } |
175 | ||
176 | static int is_cpuid_PSE36(void) | |
177 | { | |
178 | return 1; | |
179 | } | |
180 | ||
73b1087e AK |
181 | static int is_nx(struct kvm_vcpu *vcpu) |
182 | { | |
183 | return vcpu->shadow_efer & EFER_NX; | |
184 | } | |
185 | ||
6aa8b732 AK |
186 | static int is_present_pte(unsigned long pte) |
187 | { | |
188 | return pte & PT_PRESENT_MASK; | |
189 | } | |
190 | ||
c7addb90 AK |
191 | static int is_shadow_present_pte(u64 pte) |
192 | { | |
193 | pte &= ~PT_SHADOW_IO_MARK; | |
194 | return pte != shadow_trap_nonpresent_pte | |
195 | && pte != shadow_notrap_nonpresent_pte; | |
196 | } | |
197 | ||
6aa8b732 AK |
198 | static int is_writeble_pte(unsigned long pte) |
199 | { | |
200 | return pte & PT_WRITABLE_MASK; | |
201 | } | |
202 | ||
e3c5e7ec AK |
203 | static int is_dirty_pte(unsigned long pte) |
204 | { | |
205 | return pte & PT_DIRTY_MASK; | |
206 | } | |
207 | ||
6aa8b732 AK |
208 | static int is_io_pte(unsigned long pte) |
209 | { | |
210 | return pte & PT_SHADOW_IO_MARK; | |
211 | } | |
212 | ||
cd4a4e53 AK |
213 | static int is_rmap_pte(u64 pte) |
214 | { | |
9647c14c IE |
215 | return pte != shadow_trap_nonpresent_pte |
216 | && pte != shadow_notrap_nonpresent_pte; | |
cd4a4e53 AK |
217 | } |
218 | ||
e663ee64 AK |
219 | static void set_shadow_pte(u64 *sptep, u64 spte) |
220 | { | |
221 | #ifdef CONFIG_X86_64 | |
222 | set_64bit((unsigned long *)sptep, spte); | |
223 | #else | |
224 | set_64bit((unsigned long long *)sptep, spte); | |
225 | #endif | |
226 | } | |
227 | ||
e2dec939 | 228 | static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache, |
2e3e5882 | 229 | struct kmem_cache *base_cache, int min) |
714b93da AK |
230 | { |
231 | void *obj; | |
232 | ||
233 | if (cache->nobjs >= min) | |
e2dec939 | 234 | return 0; |
714b93da | 235 | while (cache->nobjs < ARRAY_SIZE(cache->objects)) { |
2e3e5882 | 236 | obj = kmem_cache_zalloc(base_cache, GFP_KERNEL); |
714b93da | 237 | if (!obj) |
e2dec939 | 238 | return -ENOMEM; |
714b93da AK |
239 | cache->objects[cache->nobjs++] = obj; |
240 | } | |
e2dec939 | 241 | return 0; |
714b93da AK |
242 | } |
243 | ||
244 | static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc) | |
245 | { | |
246 | while (mc->nobjs) | |
247 | kfree(mc->objects[--mc->nobjs]); | |
248 | } | |
249 | ||
c1158e63 | 250 | static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache *cache, |
2e3e5882 | 251 | int min) |
c1158e63 AK |
252 | { |
253 | struct page *page; | |
254 | ||
255 | if (cache->nobjs >= min) | |
256 | return 0; | |
257 | while (cache->nobjs < ARRAY_SIZE(cache->objects)) { | |
2e3e5882 | 258 | page = alloc_page(GFP_KERNEL); |
c1158e63 AK |
259 | if (!page) |
260 | return -ENOMEM; | |
261 | set_page_private(page, 0); | |
262 | cache->objects[cache->nobjs++] = page_address(page); | |
263 | } | |
264 | return 0; | |
265 | } | |
266 | ||
267 | static void mmu_free_memory_cache_page(struct kvm_mmu_memory_cache *mc) | |
268 | { | |
269 | while (mc->nobjs) | |
c4d198d5 | 270 | free_page((unsigned long)mc->objects[--mc->nobjs]); |
c1158e63 AK |
271 | } |
272 | ||
2e3e5882 | 273 | static int mmu_topup_memory_caches(struct kvm_vcpu *vcpu) |
714b93da | 274 | { |
e2dec939 AK |
275 | int r; |
276 | ||
2e3e5882 | 277 | kvm_mmu_free_some_pages(vcpu); |
e2dec939 | 278 | r = mmu_topup_memory_cache(&vcpu->mmu_pte_chain_cache, |
2e3e5882 | 279 | pte_chain_cache, 4); |
e2dec939 AK |
280 | if (r) |
281 | goto out; | |
282 | r = mmu_topup_memory_cache(&vcpu->mmu_rmap_desc_cache, | |
2e3e5882 | 283 | rmap_desc_cache, 1); |
d3d25b04 AK |
284 | if (r) |
285 | goto out; | |
290fc38d | 286 | r = mmu_topup_memory_cache_page(&vcpu->mmu_page_cache, 8); |
d3d25b04 AK |
287 | if (r) |
288 | goto out; | |
289 | r = mmu_topup_memory_cache(&vcpu->mmu_page_header_cache, | |
2e3e5882 | 290 | mmu_page_header_cache, 4); |
e2dec939 AK |
291 | out: |
292 | return r; | |
714b93da AK |
293 | } |
294 | ||
295 | static void mmu_free_memory_caches(struct kvm_vcpu *vcpu) | |
296 | { | |
297 | mmu_free_memory_cache(&vcpu->mmu_pte_chain_cache); | |
298 | mmu_free_memory_cache(&vcpu->mmu_rmap_desc_cache); | |
c1158e63 | 299 | mmu_free_memory_cache_page(&vcpu->mmu_page_cache); |
d3d25b04 | 300 | mmu_free_memory_cache(&vcpu->mmu_page_header_cache); |
714b93da AK |
301 | } |
302 | ||
303 | static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc, | |
304 | size_t size) | |
305 | { | |
306 | void *p; | |
307 | ||
308 | BUG_ON(!mc->nobjs); | |
309 | p = mc->objects[--mc->nobjs]; | |
310 | memset(p, 0, size); | |
311 | return p; | |
312 | } | |
313 | ||
714b93da AK |
314 | static struct kvm_pte_chain *mmu_alloc_pte_chain(struct kvm_vcpu *vcpu) |
315 | { | |
316 | return mmu_memory_cache_alloc(&vcpu->mmu_pte_chain_cache, | |
317 | sizeof(struct kvm_pte_chain)); | |
318 | } | |
319 | ||
90cb0529 | 320 | static void mmu_free_pte_chain(struct kvm_pte_chain *pc) |
714b93da | 321 | { |
90cb0529 | 322 | kfree(pc); |
714b93da AK |
323 | } |
324 | ||
325 | static struct kvm_rmap_desc *mmu_alloc_rmap_desc(struct kvm_vcpu *vcpu) | |
326 | { | |
327 | return mmu_memory_cache_alloc(&vcpu->mmu_rmap_desc_cache, | |
328 | sizeof(struct kvm_rmap_desc)); | |
329 | } | |
330 | ||
90cb0529 | 331 | static void mmu_free_rmap_desc(struct kvm_rmap_desc *rd) |
714b93da | 332 | { |
90cb0529 | 333 | kfree(rd); |
714b93da AK |
334 | } |
335 | ||
290fc38d IE |
336 | /* |
337 | * Take gfn and return the reverse mapping to it. | |
338 | * Note: gfn must be unaliased before this function get called | |
339 | */ | |
340 | ||
341 | static unsigned long *gfn_to_rmap(struct kvm *kvm, gfn_t gfn) | |
342 | { | |
343 | struct kvm_memory_slot *slot; | |
344 | ||
345 | slot = gfn_to_memslot(kvm, gfn); | |
346 | return &slot->rmap[gfn - slot->base_gfn]; | |
347 | } | |
348 | ||
cd4a4e53 AK |
349 | /* |
350 | * Reverse mapping data structures: | |
351 | * | |
290fc38d IE |
352 | * If rmapp bit zero is zero, then rmapp point to the shadw page table entry |
353 | * that points to page_address(page). | |
cd4a4e53 | 354 | * |
290fc38d IE |
355 | * If rmapp bit zero is one, (then rmap & ~1) points to a struct kvm_rmap_desc |
356 | * containing more mappings. | |
cd4a4e53 | 357 | */ |
290fc38d | 358 | static void rmap_add(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn) |
cd4a4e53 | 359 | { |
290fc38d | 360 | struct kvm_mmu_page *page; |
cd4a4e53 | 361 | struct kvm_rmap_desc *desc; |
290fc38d | 362 | unsigned long *rmapp; |
cd4a4e53 AK |
363 | int i; |
364 | ||
365 | if (!is_rmap_pte(*spte)) | |
366 | return; | |
290fc38d IE |
367 | gfn = unalias_gfn(vcpu->kvm, gfn); |
368 | page = page_header(__pa(spte)); | |
369 | page->gfns[spte - page->spt] = gfn; | |
370 | rmapp = gfn_to_rmap(vcpu->kvm, gfn); | |
371 | if (!*rmapp) { | |
cd4a4e53 | 372 | rmap_printk("rmap_add: %p %llx 0->1\n", spte, *spte); |
290fc38d IE |
373 | *rmapp = (unsigned long)spte; |
374 | } else if (!(*rmapp & 1)) { | |
cd4a4e53 | 375 | rmap_printk("rmap_add: %p %llx 1->many\n", spte, *spte); |
714b93da | 376 | desc = mmu_alloc_rmap_desc(vcpu); |
290fc38d | 377 | desc->shadow_ptes[0] = (u64 *)*rmapp; |
cd4a4e53 | 378 | desc->shadow_ptes[1] = spte; |
290fc38d | 379 | *rmapp = (unsigned long)desc | 1; |
cd4a4e53 AK |
380 | } else { |
381 | rmap_printk("rmap_add: %p %llx many->many\n", spte, *spte); | |
290fc38d | 382 | desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul); |
cd4a4e53 AK |
383 | while (desc->shadow_ptes[RMAP_EXT-1] && desc->more) |
384 | desc = desc->more; | |
385 | if (desc->shadow_ptes[RMAP_EXT-1]) { | |
714b93da | 386 | desc->more = mmu_alloc_rmap_desc(vcpu); |
cd4a4e53 AK |
387 | desc = desc->more; |
388 | } | |
389 | for (i = 0; desc->shadow_ptes[i]; ++i) | |
390 | ; | |
391 | desc->shadow_ptes[i] = spte; | |
392 | } | |
393 | } | |
394 | ||
290fc38d | 395 | static void rmap_desc_remove_entry(unsigned long *rmapp, |
cd4a4e53 AK |
396 | struct kvm_rmap_desc *desc, |
397 | int i, | |
398 | struct kvm_rmap_desc *prev_desc) | |
399 | { | |
400 | int j; | |
401 | ||
402 | for (j = RMAP_EXT - 1; !desc->shadow_ptes[j] && j > i; --j) | |
403 | ; | |
404 | desc->shadow_ptes[i] = desc->shadow_ptes[j]; | |
11718b4d | 405 | desc->shadow_ptes[j] = NULL; |
cd4a4e53 AK |
406 | if (j != 0) |
407 | return; | |
408 | if (!prev_desc && !desc->more) | |
290fc38d | 409 | *rmapp = (unsigned long)desc->shadow_ptes[0]; |
cd4a4e53 AK |
410 | else |
411 | if (prev_desc) | |
412 | prev_desc->more = desc->more; | |
413 | else | |
290fc38d | 414 | *rmapp = (unsigned long)desc->more | 1; |
90cb0529 | 415 | mmu_free_rmap_desc(desc); |
cd4a4e53 AK |
416 | } |
417 | ||
290fc38d | 418 | static void rmap_remove(struct kvm *kvm, u64 *spte) |
cd4a4e53 | 419 | { |
cd4a4e53 AK |
420 | struct kvm_rmap_desc *desc; |
421 | struct kvm_rmap_desc *prev_desc; | |
290fc38d | 422 | struct kvm_mmu_page *page; |
b4231d61 | 423 | struct page *release_page; |
290fc38d | 424 | unsigned long *rmapp; |
cd4a4e53 AK |
425 | int i; |
426 | ||
427 | if (!is_rmap_pte(*spte)) | |
428 | return; | |
290fc38d | 429 | page = page_header(__pa(spte)); |
b4231d61 IE |
430 | release_page = pfn_to_page((*spte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT); |
431 | if (is_writeble_pte(*spte)) | |
432 | kvm_release_page_dirty(release_page); | |
433 | else | |
434 | kvm_release_page_clean(release_page); | |
290fc38d IE |
435 | rmapp = gfn_to_rmap(kvm, page->gfns[spte - page->spt]); |
436 | if (!*rmapp) { | |
cd4a4e53 AK |
437 | printk(KERN_ERR "rmap_remove: %p %llx 0->BUG\n", spte, *spte); |
438 | BUG(); | |
290fc38d | 439 | } else if (!(*rmapp & 1)) { |
cd4a4e53 | 440 | rmap_printk("rmap_remove: %p %llx 1->0\n", spte, *spte); |
290fc38d | 441 | if ((u64 *)*rmapp != spte) { |
cd4a4e53 AK |
442 | printk(KERN_ERR "rmap_remove: %p %llx 1->BUG\n", |
443 | spte, *spte); | |
444 | BUG(); | |
445 | } | |
290fc38d | 446 | *rmapp = 0; |
cd4a4e53 AK |
447 | } else { |
448 | rmap_printk("rmap_remove: %p %llx many->many\n", spte, *spte); | |
290fc38d | 449 | desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul); |
cd4a4e53 AK |
450 | prev_desc = NULL; |
451 | while (desc) { | |
452 | for (i = 0; i < RMAP_EXT && desc->shadow_ptes[i]; ++i) | |
453 | if (desc->shadow_ptes[i] == spte) { | |
290fc38d | 454 | rmap_desc_remove_entry(rmapp, |
714b93da | 455 | desc, i, |
cd4a4e53 AK |
456 | prev_desc); |
457 | return; | |
458 | } | |
459 | prev_desc = desc; | |
460 | desc = desc->more; | |
461 | } | |
462 | BUG(); | |
463 | } | |
464 | } | |
465 | ||
98348e95 | 466 | static u64 *rmap_next(struct kvm *kvm, unsigned long *rmapp, u64 *spte) |
374cbac0 | 467 | { |
374cbac0 | 468 | struct kvm_rmap_desc *desc; |
98348e95 IE |
469 | struct kvm_rmap_desc *prev_desc; |
470 | u64 *prev_spte; | |
471 | int i; | |
472 | ||
473 | if (!*rmapp) | |
474 | return NULL; | |
475 | else if (!(*rmapp & 1)) { | |
476 | if (!spte) | |
477 | return (u64 *)*rmapp; | |
478 | return NULL; | |
479 | } | |
480 | desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul); | |
481 | prev_desc = NULL; | |
482 | prev_spte = NULL; | |
483 | while (desc) { | |
484 | for (i = 0; i < RMAP_EXT && desc->shadow_ptes[i]; ++i) { | |
485 | if (prev_spte == spte) | |
486 | return desc->shadow_ptes[i]; | |
487 | prev_spte = desc->shadow_ptes[i]; | |
488 | } | |
489 | desc = desc->more; | |
490 | } | |
491 | return NULL; | |
492 | } | |
493 | ||
494 | static void rmap_write_protect(struct kvm *kvm, u64 gfn) | |
495 | { | |
290fc38d | 496 | unsigned long *rmapp; |
374cbac0 AK |
497 | u64 *spte; |
498 | ||
4a4c9924 AL |
499 | gfn = unalias_gfn(kvm, gfn); |
500 | rmapp = gfn_to_rmap(kvm, gfn); | |
374cbac0 | 501 | |
98348e95 IE |
502 | spte = rmap_next(kvm, rmapp, NULL); |
503 | while (spte) { | |
374cbac0 | 504 | BUG_ON(!spte); |
374cbac0 | 505 | BUG_ON(!(*spte & PT_PRESENT_MASK)); |
374cbac0 | 506 | rmap_printk("rmap_write_protect: spte %p %llx\n", spte, *spte); |
9647c14c IE |
507 | if (is_writeble_pte(*spte)) |
508 | set_shadow_pte(spte, *spte & ~PT_WRITABLE_MASK); | |
4a4c9924 | 509 | kvm_flush_remote_tlbs(kvm); |
9647c14c | 510 | spte = rmap_next(kvm, rmapp, spte); |
374cbac0 AK |
511 | } |
512 | } | |
513 | ||
d6c69ee9 | 514 | #ifdef MMU_DEBUG |
47ad8e68 | 515 | static int is_empty_shadow_page(u64 *spt) |
6aa8b732 | 516 | { |
139bdb2d AK |
517 | u64 *pos; |
518 | u64 *end; | |
519 | ||
47ad8e68 | 520 | for (pos = spt, end = pos + PAGE_SIZE / sizeof(u64); pos != end; pos++) |
c7addb90 | 521 | if ((*pos & ~PT_SHADOW_IO_MARK) != shadow_trap_nonpresent_pte) { |
139bdb2d AK |
522 | printk(KERN_ERR "%s: %p %llx\n", __FUNCTION__, |
523 | pos, *pos); | |
6aa8b732 | 524 | return 0; |
139bdb2d | 525 | } |
6aa8b732 AK |
526 | return 1; |
527 | } | |
d6c69ee9 | 528 | #endif |
6aa8b732 | 529 | |
90cb0529 | 530 | static void kvm_mmu_free_page(struct kvm *kvm, |
4b02d6da | 531 | struct kvm_mmu_page *page_head) |
260746c0 | 532 | { |
47ad8e68 | 533 | ASSERT(is_empty_shadow_page(page_head->spt)); |
d3d25b04 | 534 | list_del(&page_head->link); |
c1158e63 | 535 | __free_page(virt_to_page(page_head->spt)); |
290fc38d | 536 | __free_page(virt_to_page(page_head->gfns)); |
90cb0529 AK |
537 | kfree(page_head); |
538 | ++kvm->n_free_mmu_pages; | |
260746c0 AK |
539 | } |
540 | ||
cea0f0e7 AK |
541 | static unsigned kvm_page_table_hashfn(gfn_t gfn) |
542 | { | |
543 | return gfn; | |
544 | } | |
545 | ||
25c0de2c AK |
546 | static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu, |
547 | u64 *parent_pte) | |
6aa8b732 AK |
548 | { |
549 | struct kvm_mmu_page *page; | |
550 | ||
d3d25b04 | 551 | if (!vcpu->kvm->n_free_mmu_pages) |
25c0de2c | 552 | return NULL; |
6aa8b732 | 553 | |
d3d25b04 AK |
554 | page = mmu_memory_cache_alloc(&vcpu->mmu_page_header_cache, |
555 | sizeof *page); | |
556 | page->spt = mmu_memory_cache_alloc(&vcpu->mmu_page_cache, PAGE_SIZE); | |
290fc38d | 557 | page->gfns = mmu_memory_cache_alloc(&vcpu->mmu_page_cache, PAGE_SIZE); |
d3d25b04 AK |
558 | set_page_private(virt_to_page(page->spt), (unsigned long)page); |
559 | list_add(&page->link, &vcpu->kvm->active_mmu_pages); | |
47ad8e68 | 560 | ASSERT(is_empty_shadow_page(page->spt)); |
6aa8b732 | 561 | page->slot_bitmap = 0; |
cea0f0e7 | 562 | page->multimapped = 0; |
6aa8b732 | 563 | page->parent_pte = parent_pte; |
ebeace86 | 564 | --vcpu->kvm->n_free_mmu_pages; |
25c0de2c | 565 | return page; |
6aa8b732 AK |
566 | } |
567 | ||
714b93da AK |
568 | static void mmu_page_add_parent_pte(struct kvm_vcpu *vcpu, |
569 | struct kvm_mmu_page *page, u64 *parent_pte) | |
cea0f0e7 AK |
570 | { |
571 | struct kvm_pte_chain *pte_chain; | |
572 | struct hlist_node *node; | |
573 | int i; | |
574 | ||
575 | if (!parent_pte) | |
576 | return; | |
577 | if (!page->multimapped) { | |
578 | u64 *old = page->parent_pte; | |
579 | ||
580 | if (!old) { | |
581 | page->parent_pte = parent_pte; | |
582 | return; | |
583 | } | |
584 | page->multimapped = 1; | |
714b93da | 585 | pte_chain = mmu_alloc_pte_chain(vcpu); |
cea0f0e7 AK |
586 | INIT_HLIST_HEAD(&page->parent_ptes); |
587 | hlist_add_head(&pte_chain->link, &page->parent_ptes); | |
588 | pte_chain->parent_ptes[0] = old; | |
589 | } | |
590 | hlist_for_each_entry(pte_chain, node, &page->parent_ptes, link) { | |
591 | if (pte_chain->parent_ptes[NR_PTE_CHAIN_ENTRIES-1]) | |
592 | continue; | |
593 | for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) | |
594 | if (!pte_chain->parent_ptes[i]) { | |
595 | pte_chain->parent_ptes[i] = parent_pte; | |
596 | return; | |
597 | } | |
598 | } | |
714b93da | 599 | pte_chain = mmu_alloc_pte_chain(vcpu); |
cea0f0e7 AK |
600 | BUG_ON(!pte_chain); |
601 | hlist_add_head(&pte_chain->link, &page->parent_ptes); | |
602 | pte_chain->parent_ptes[0] = parent_pte; | |
603 | } | |
604 | ||
90cb0529 | 605 | static void mmu_page_remove_parent_pte(struct kvm_mmu_page *page, |
cea0f0e7 AK |
606 | u64 *parent_pte) |
607 | { | |
608 | struct kvm_pte_chain *pte_chain; | |
609 | struct hlist_node *node; | |
610 | int i; | |
611 | ||
612 | if (!page->multimapped) { | |
613 | BUG_ON(page->parent_pte != parent_pte); | |
614 | page->parent_pte = NULL; | |
615 | return; | |
616 | } | |
617 | hlist_for_each_entry(pte_chain, node, &page->parent_ptes, link) | |
618 | for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) { | |
619 | if (!pte_chain->parent_ptes[i]) | |
620 | break; | |
621 | if (pte_chain->parent_ptes[i] != parent_pte) | |
622 | continue; | |
697fe2e2 AK |
623 | while (i + 1 < NR_PTE_CHAIN_ENTRIES |
624 | && pte_chain->parent_ptes[i + 1]) { | |
cea0f0e7 AK |
625 | pte_chain->parent_ptes[i] |
626 | = pte_chain->parent_ptes[i + 1]; | |
627 | ++i; | |
628 | } | |
629 | pte_chain->parent_ptes[i] = NULL; | |
697fe2e2 AK |
630 | if (i == 0) { |
631 | hlist_del(&pte_chain->link); | |
90cb0529 | 632 | mmu_free_pte_chain(pte_chain); |
697fe2e2 AK |
633 | if (hlist_empty(&page->parent_ptes)) { |
634 | page->multimapped = 0; | |
635 | page->parent_pte = NULL; | |
636 | } | |
637 | } | |
cea0f0e7 AK |
638 | return; |
639 | } | |
640 | BUG(); | |
641 | } | |
642 | ||
f67a46f4 | 643 | static struct kvm_mmu_page *kvm_mmu_lookup_page(struct kvm *kvm, |
cea0f0e7 AK |
644 | gfn_t gfn) |
645 | { | |
646 | unsigned index; | |
647 | struct hlist_head *bucket; | |
648 | struct kvm_mmu_page *page; | |
649 | struct hlist_node *node; | |
650 | ||
651 | pgprintk("%s: looking for gfn %lx\n", __FUNCTION__, gfn); | |
652 | index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES; | |
f67a46f4 | 653 | bucket = &kvm->mmu_page_hash[index]; |
cea0f0e7 AK |
654 | hlist_for_each_entry(page, node, bucket, hash_link) |
655 | if (page->gfn == gfn && !page->role.metaphysical) { | |
656 | pgprintk("%s: found role %x\n", | |
657 | __FUNCTION__, page->role.word); | |
658 | return page; | |
659 | } | |
660 | return NULL; | |
661 | } | |
662 | ||
663 | static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu, | |
664 | gfn_t gfn, | |
665 | gva_t gaddr, | |
666 | unsigned level, | |
667 | int metaphysical, | |
d28c6cfb | 668 | unsigned hugepage_access, |
cea0f0e7 AK |
669 | u64 *parent_pte) |
670 | { | |
671 | union kvm_mmu_page_role role; | |
672 | unsigned index; | |
673 | unsigned quadrant; | |
674 | struct hlist_head *bucket; | |
675 | struct kvm_mmu_page *page; | |
676 | struct hlist_node *node; | |
677 | ||
678 | role.word = 0; | |
679 | role.glevels = vcpu->mmu.root_level; | |
680 | role.level = level; | |
681 | role.metaphysical = metaphysical; | |
d28c6cfb | 682 | role.hugepage_access = hugepage_access; |
cea0f0e7 AK |
683 | if (vcpu->mmu.root_level <= PT32_ROOT_LEVEL) { |
684 | quadrant = gaddr >> (PAGE_SHIFT + (PT64_PT_BITS * level)); | |
685 | quadrant &= (1 << ((PT32_PT_BITS - PT64_PT_BITS) * level)) - 1; | |
686 | role.quadrant = quadrant; | |
687 | } | |
688 | pgprintk("%s: looking gfn %lx role %x\n", __FUNCTION__, | |
689 | gfn, role.word); | |
690 | index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES; | |
691 | bucket = &vcpu->kvm->mmu_page_hash[index]; | |
692 | hlist_for_each_entry(page, node, bucket, hash_link) | |
693 | if (page->gfn == gfn && page->role.word == role.word) { | |
714b93da | 694 | mmu_page_add_parent_pte(vcpu, page, parent_pte); |
cea0f0e7 AK |
695 | pgprintk("%s: found\n", __FUNCTION__); |
696 | return page; | |
697 | } | |
698 | page = kvm_mmu_alloc_page(vcpu, parent_pte); | |
699 | if (!page) | |
700 | return page; | |
701 | pgprintk("%s: adding gfn %lx role %x\n", __FUNCTION__, gfn, role.word); | |
702 | page->gfn = gfn; | |
703 | page->role = role; | |
704 | hlist_add_head(&page->hash_link, bucket); | |
c7addb90 | 705 | vcpu->mmu.prefetch_page(vcpu, page); |
374cbac0 | 706 | if (!metaphysical) |
4a4c9924 | 707 | rmap_write_protect(vcpu->kvm, gfn); |
cea0f0e7 AK |
708 | return page; |
709 | } | |
710 | ||
90cb0529 | 711 | static void kvm_mmu_page_unlink_children(struct kvm *kvm, |
a436036b AK |
712 | struct kvm_mmu_page *page) |
713 | { | |
697fe2e2 AK |
714 | unsigned i; |
715 | u64 *pt; | |
716 | u64 ent; | |
717 | ||
47ad8e68 | 718 | pt = page->spt; |
697fe2e2 AK |
719 | |
720 | if (page->role.level == PT_PAGE_TABLE_LEVEL) { | |
721 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) { | |
c7addb90 | 722 | if (is_shadow_present_pte(pt[i])) |
290fc38d | 723 | rmap_remove(kvm, &pt[i]); |
c7addb90 | 724 | pt[i] = shadow_trap_nonpresent_pte; |
697fe2e2 | 725 | } |
90cb0529 | 726 | kvm_flush_remote_tlbs(kvm); |
697fe2e2 AK |
727 | return; |
728 | } | |
729 | ||
730 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) { | |
731 | ent = pt[i]; | |
732 | ||
c7addb90 AK |
733 | pt[i] = shadow_trap_nonpresent_pte; |
734 | if (!is_shadow_present_pte(ent)) | |
697fe2e2 AK |
735 | continue; |
736 | ent &= PT64_BASE_ADDR_MASK; | |
90cb0529 | 737 | mmu_page_remove_parent_pte(page_header(ent), &pt[i]); |
697fe2e2 | 738 | } |
90cb0529 | 739 | kvm_flush_remote_tlbs(kvm); |
a436036b AK |
740 | } |
741 | ||
90cb0529 | 742 | static void kvm_mmu_put_page(struct kvm_mmu_page *page, |
cea0f0e7 AK |
743 | u64 *parent_pte) |
744 | { | |
90cb0529 | 745 | mmu_page_remove_parent_pte(page, parent_pte); |
a436036b AK |
746 | } |
747 | ||
12b7d28f AK |
748 | static void kvm_mmu_reset_last_pte_updated(struct kvm *kvm) |
749 | { | |
750 | int i; | |
751 | ||
752 | for (i = 0; i < KVM_MAX_VCPUS; ++i) | |
753 | if (kvm->vcpus[i]) | |
754 | kvm->vcpus[i]->last_pte_updated = NULL; | |
755 | } | |
756 | ||
90cb0529 | 757 | static void kvm_mmu_zap_page(struct kvm *kvm, |
a436036b AK |
758 | struct kvm_mmu_page *page) |
759 | { | |
760 | u64 *parent_pte; | |
761 | ||
4cee5764 | 762 | ++kvm->stat.mmu_shadow_zapped; |
a436036b AK |
763 | while (page->multimapped || page->parent_pte) { |
764 | if (!page->multimapped) | |
765 | parent_pte = page->parent_pte; | |
766 | else { | |
767 | struct kvm_pte_chain *chain; | |
768 | ||
769 | chain = container_of(page->parent_ptes.first, | |
770 | struct kvm_pte_chain, link); | |
771 | parent_pte = chain->parent_ptes[0]; | |
772 | } | |
697fe2e2 | 773 | BUG_ON(!parent_pte); |
90cb0529 | 774 | kvm_mmu_put_page(page, parent_pte); |
c7addb90 | 775 | set_shadow_pte(parent_pte, shadow_trap_nonpresent_pte); |
a436036b | 776 | } |
90cb0529 | 777 | kvm_mmu_page_unlink_children(kvm, page); |
3bb65a22 AK |
778 | if (!page->root_count) { |
779 | hlist_del(&page->hash_link); | |
90cb0529 | 780 | kvm_mmu_free_page(kvm, page); |
36868f7b | 781 | } else |
90cb0529 | 782 | list_move(&page->link, &kvm->active_mmu_pages); |
12b7d28f | 783 | kvm_mmu_reset_last_pte_updated(kvm); |
a436036b AK |
784 | } |
785 | ||
82ce2c96 IE |
786 | /* |
787 | * Changing the number of mmu pages allocated to the vm | |
788 | * Note: if kvm_nr_mmu_pages is too small, you will get dead lock | |
789 | */ | |
790 | void kvm_mmu_change_mmu_pages(struct kvm *kvm, unsigned int kvm_nr_mmu_pages) | |
791 | { | |
792 | /* | |
793 | * If we set the number of mmu pages to be smaller be than the | |
794 | * number of actived pages , we must to free some mmu pages before we | |
795 | * change the value | |
796 | */ | |
797 | ||
798 | if ((kvm->n_alloc_mmu_pages - kvm->n_free_mmu_pages) > | |
799 | kvm_nr_mmu_pages) { | |
800 | int n_used_mmu_pages = kvm->n_alloc_mmu_pages | |
801 | - kvm->n_free_mmu_pages; | |
802 | ||
803 | while (n_used_mmu_pages > kvm_nr_mmu_pages) { | |
804 | struct kvm_mmu_page *page; | |
805 | ||
806 | page = container_of(kvm->active_mmu_pages.prev, | |
807 | struct kvm_mmu_page, link); | |
808 | kvm_mmu_zap_page(kvm, page); | |
809 | n_used_mmu_pages--; | |
810 | } | |
811 | kvm->n_free_mmu_pages = 0; | |
812 | } | |
813 | else | |
814 | kvm->n_free_mmu_pages += kvm_nr_mmu_pages | |
815 | - kvm->n_alloc_mmu_pages; | |
816 | ||
817 | kvm->n_alloc_mmu_pages = kvm_nr_mmu_pages; | |
818 | } | |
819 | ||
f67a46f4 | 820 | static int kvm_mmu_unprotect_page(struct kvm *kvm, gfn_t gfn) |
a436036b AK |
821 | { |
822 | unsigned index; | |
823 | struct hlist_head *bucket; | |
824 | struct kvm_mmu_page *page; | |
825 | struct hlist_node *node, *n; | |
826 | int r; | |
827 | ||
828 | pgprintk("%s: looking for gfn %lx\n", __FUNCTION__, gfn); | |
829 | r = 0; | |
830 | index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES; | |
f67a46f4 | 831 | bucket = &kvm->mmu_page_hash[index]; |
a436036b AK |
832 | hlist_for_each_entry_safe(page, node, n, bucket, hash_link) |
833 | if (page->gfn == gfn && !page->role.metaphysical) { | |
697fe2e2 AK |
834 | pgprintk("%s: gfn %lx role %x\n", __FUNCTION__, gfn, |
835 | page->role.word); | |
f67a46f4 | 836 | kvm_mmu_zap_page(kvm, page); |
a436036b AK |
837 | r = 1; |
838 | } | |
839 | return r; | |
cea0f0e7 AK |
840 | } |
841 | ||
f67a46f4 | 842 | static void mmu_unshadow(struct kvm *kvm, gfn_t gfn) |
97a0a01e AK |
843 | { |
844 | struct kvm_mmu_page *page; | |
845 | ||
f67a46f4 | 846 | while ((page = kvm_mmu_lookup_page(kvm, gfn)) != NULL) { |
97a0a01e AK |
847 | pgprintk("%s: zap %lx %x\n", |
848 | __FUNCTION__, gfn, page->role.word); | |
f67a46f4 | 849 | kvm_mmu_zap_page(kvm, page); |
97a0a01e AK |
850 | } |
851 | } | |
852 | ||
6aa8b732 AK |
853 | static void page_header_update_slot(struct kvm *kvm, void *pte, gpa_t gpa) |
854 | { | |
855 | int slot = memslot_id(kvm, gfn_to_memslot(kvm, gpa >> PAGE_SHIFT)); | |
856 | struct kvm_mmu_page *page_head = page_header(__pa(pte)); | |
857 | ||
858 | __set_bit(slot, &page_head->slot_bitmap); | |
859 | } | |
860 | ||
4a4c9924 | 861 | hpa_t gpa_to_hpa(struct kvm *kvm, gpa_t gpa) |
6aa8b732 | 862 | { |
6aa8b732 | 863 | struct page *page; |
cea7bb21 | 864 | hpa_t hpa; |
6aa8b732 AK |
865 | |
866 | ASSERT((gpa & HPA_ERR_MASK) == 0); | |
4a4c9924 | 867 | page = gfn_to_page(kvm, gpa >> PAGE_SHIFT); |
cea7bb21 IE |
868 | hpa = ((hpa_t)page_to_pfn(page) << PAGE_SHIFT) | (gpa & (PAGE_SIZE-1)); |
869 | if (is_error_page(page)) | |
870 | return hpa | HPA_ERR_MASK; | |
871 | return hpa; | |
6aa8b732 AK |
872 | } |
873 | ||
874 | hpa_t gva_to_hpa(struct kvm_vcpu *vcpu, gva_t gva) | |
875 | { | |
876 | gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, gva); | |
877 | ||
878 | if (gpa == UNMAPPED_GVA) | |
879 | return UNMAPPED_GVA; | |
4a4c9924 | 880 | return gpa_to_hpa(vcpu->kvm, gpa); |
6aa8b732 AK |
881 | } |
882 | ||
039576c0 AK |
883 | struct page *gva_to_page(struct kvm_vcpu *vcpu, gva_t gva) |
884 | { | |
885 | gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, gva); | |
886 | ||
887 | if (gpa == UNMAPPED_GVA) | |
888 | return NULL; | |
4a4c9924 | 889 | return pfn_to_page(gpa_to_hpa(vcpu->kvm, gpa) >> PAGE_SHIFT); |
039576c0 AK |
890 | } |
891 | ||
6aa8b732 AK |
892 | static void nonpaging_new_cr3(struct kvm_vcpu *vcpu) |
893 | { | |
894 | } | |
895 | ||
896 | static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, hpa_t p) | |
897 | { | |
898 | int level = PT32E_ROOT_LEVEL; | |
899 | hpa_t table_addr = vcpu->mmu.root_hpa; | |
b4231d61 | 900 | struct page *page; |
6aa8b732 | 901 | |
b4231d61 | 902 | page = pfn_to_page(p >> PAGE_SHIFT); |
6aa8b732 AK |
903 | for (; ; level--) { |
904 | u32 index = PT64_INDEX(v, level); | |
905 | u64 *table; | |
cea0f0e7 | 906 | u64 pte; |
6aa8b732 AK |
907 | |
908 | ASSERT(VALID_PAGE(table_addr)); | |
909 | table = __va(table_addr); | |
910 | ||
911 | if (level == 1) { | |
9647c14c IE |
912 | int was_rmapped; |
913 | ||
cea0f0e7 | 914 | pte = table[index]; |
9647c14c | 915 | was_rmapped = is_rmap_pte(pte); |
2065b372 | 916 | if (is_shadow_present_pte(pte) && is_writeble_pte(pte)) { |
b4231d61 | 917 | kvm_release_page_clean(page); |
cea0f0e7 | 918 | return 0; |
2065b372 | 919 | } |
6aa8b732 AK |
920 | mark_page_dirty(vcpu->kvm, v >> PAGE_SHIFT); |
921 | page_header_update_slot(vcpu->kvm, table, v); | |
922 | table[index] = p | PT_PRESENT_MASK | PT_WRITABLE_MASK | | |
923 | PT_USER_MASK; | |
9647c14c IE |
924 | if (!was_rmapped) |
925 | rmap_add(vcpu, &table[index], v >> PAGE_SHIFT); | |
8a7ae055 | 926 | else |
b4231d61 IE |
927 | kvm_release_page_clean(page); |
928 | ||
6aa8b732 AK |
929 | return 0; |
930 | } | |
931 | ||
c7addb90 | 932 | if (table[index] == shadow_trap_nonpresent_pte) { |
25c0de2c | 933 | struct kvm_mmu_page *new_table; |
cea0f0e7 | 934 | gfn_t pseudo_gfn; |
6aa8b732 | 935 | |
cea0f0e7 AK |
936 | pseudo_gfn = (v & PT64_DIR_BASE_ADDR_MASK) |
937 | >> PAGE_SHIFT; | |
938 | new_table = kvm_mmu_get_page(vcpu, pseudo_gfn, | |
939 | v, level - 1, | |
6bfccdc9 | 940 | 1, 3, &table[index]); |
25c0de2c | 941 | if (!new_table) { |
6aa8b732 | 942 | pgprintk("nonpaging_map: ENOMEM\n"); |
b4231d61 | 943 | kvm_release_page_clean(page); |
6aa8b732 AK |
944 | return -ENOMEM; |
945 | } | |
946 | ||
47ad8e68 | 947 | table[index] = __pa(new_table->spt) | PT_PRESENT_MASK |
25c0de2c | 948 | | PT_WRITABLE_MASK | PT_USER_MASK; |
6aa8b732 AK |
949 | } |
950 | table_addr = table[index] & PT64_BASE_ADDR_MASK; | |
951 | } | |
952 | } | |
953 | ||
c7addb90 AK |
954 | static void nonpaging_prefetch_page(struct kvm_vcpu *vcpu, |
955 | struct kvm_mmu_page *sp) | |
956 | { | |
957 | int i; | |
958 | ||
959 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) | |
960 | sp->spt[i] = shadow_trap_nonpresent_pte; | |
961 | } | |
962 | ||
17ac10ad AK |
963 | static void mmu_free_roots(struct kvm_vcpu *vcpu) |
964 | { | |
965 | int i; | |
3bb65a22 | 966 | struct kvm_mmu_page *page; |
17ac10ad | 967 | |
7b53aa56 AK |
968 | if (!VALID_PAGE(vcpu->mmu.root_hpa)) |
969 | return; | |
17ac10ad AK |
970 | #ifdef CONFIG_X86_64 |
971 | if (vcpu->mmu.shadow_root_level == PT64_ROOT_LEVEL) { | |
972 | hpa_t root = vcpu->mmu.root_hpa; | |
973 | ||
3bb65a22 AK |
974 | page = page_header(root); |
975 | --page->root_count; | |
17ac10ad AK |
976 | vcpu->mmu.root_hpa = INVALID_PAGE; |
977 | return; | |
978 | } | |
979 | #endif | |
980 | for (i = 0; i < 4; ++i) { | |
981 | hpa_t root = vcpu->mmu.pae_root[i]; | |
982 | ||
417726a3 | 983 | if (root) { |
417726a3 AK |
984 | root &= PT64_BASE_ADDR_MASK; |
985 | page = page_header(root); | |
986 | --page->root_count; | |
987 | } | |
17ac10ad AK |
988 | vcpu->mmu.pae_root[i] = INVALID_PAGE; |
989 | } | |
990 | vcpu->mmu.root_hpa = INVALID_PAGE; | |
991 | } | |
992 | ||
993 | static void mmu_alloc_roots(struct kvm_vcpu *vcpu) | |
994 | { | |
995 | int i; | |
cea0f0e7 | 996 | gfn_t root_gfn; |
3bb65a22 AK |
997 | struct kvm_mmu_page *page; |
998 | ||
cea0f0e7 | 999 | root_gfn = vcpu->cr3 >> PAGE_SHIFT; |
17ac10ad AK |
1000 | |
1001 | #ifdef CONFIG_X86_64 | |
1002 | if (vcpu->mmu.shadow_root_level == PT64_ROOT_LEVEL) { | |
1003 | hpa_t root = vcpu->mmu.root_hpa; | |
1004 | ||
1005 | ASSERT(!VALID_PAGE(root)); | |
68a99f6d | 1006 | page = kvm_mmu_get_page(vcpu, root_gfn, 0, |
d28c6cfb | 1007 | PT64_ROOT_LEVEL, 0, 0, NULL); |
47ad8e68 | 1008 | root = __pa(page->spt); |
3bb65a22 | 1009 | ++page->root_count; |
17ac10ad AK |
1010 | vcpu->mmu.root_hpa = root; |
1011 | return; | |
1012 | } | |
1013 | #endif | |
1014 | for (i = 0; i < 4; ++i) { | |
1015 | hpa_t root = vcpu->mmu.pae_root[i]; | |
1016 | ||
1017 | ASSERT(!VALID_PAGE(root)); | |
417726a3 AK |
1018 | if (vcpu->mmu.root_level == PT32E_ROOT_LEVEL) { |
1019 | if (!is_present_pte(vcpu->pdptrs[i])) { | |
1020 | vcpu->mmu.pae_root[i] = 0; | |
1021 | continue; | |
1022 | } | |
cea0f0e7 | 1023 | root_gfn = vcpu->pdptrs[i] >> PAGE_SHIFT; |
417726a3 | 1024 | } else if (vcpu->mmu.root_level == 0) |
cea0f0e7 | 1025 | root_gfn = 0; |
68a99f6d | 1026 | page = kvm_mmu_get_page(vcpu, root_gfn, i << 30, |
cea0f0e7 | 1027 | PT32_ROOT_LEVEL, !is_paging(vcpu), |
d28c6cfb | 1028 | 0, NULL); |
47ad8e68 | 1029 | root = __pa(page->spt); |
3bb65a22 | 1030 | ++page->root_count; |
17ac10ad AK |
1031 | vcpu->mmu.pae_root[i] = root | PT_PRESENT_MASK; |
1032 | } | |
1033 | vcpu->mmu.root_hpa = __pa(vcpu->mmu.pae_root); | |
1034 | } | |
1035 | ||
6aa8b732 AK |
1036 | static gpa_t nonpaging_gva_to_gpa(struct kvm_vcpu *vcpu, gva_t vaddr) |
1037 | { | |
1038 | return vaddr; | |
1039 | } | |
1040 | ||
1041 | static int nonpaging_page_fault(struct kvm_vcpu *vcpu, gva_t gva, | |
1042 | u32 error_code) | |
1043 | { | |
6aa8b732 | 1044 | gpa_t addr = gva; |
ebeace86 | 1045 | hpa_t paddr; |
e2dec939 | 1046 | int r; |
6aa8b732 | 1047 | |
e2dec939 AK |
1048 | r = mmu_topup_memory_caches(vcpu); |
1049 | if (r) | |
1050 | return r; | |
714b93da | 1051 | |
6aa8b732 AK |
1052 | ASSERT(vcpu); |
1053 | ASSERT(VALID_PAGE(vcpu->mmu.root_hpa)); | |
1054 | ||
6aa8b732 | 1055 | |
4a4c9924 | 1056 | paddr = gpa_to_hpa(vcpu->kvm, addr & PT64_BASE_ADDR_MASK); |
6aa8b732 | 1057 | |
8a7ae055 | 1058 | if (is_error_hpa(paddr)) { |
b4231d61 IE |
1059 | kvm_release_page_clean(pfn_to_page((paddr & PT64_BASE_ADDR_MASK) |
1060 | >> PAGE_SHIFT)); | |
ebeace86 | 1061 | return 1; |
8a7ae055 | 1062 | } |
6aa8b732 | 1063 | |
ebeace86 | 1064 | return nonpaging_map(vcpu, addr & PAGE_MASK, paddr); |
6aa8b732 AK |
1065 | } |
1066 | ||
6aa8b732 AK |
1067 | static void nonpaging_free(struct kvm_vcpu *vcpu) |
1068 | { | |
17ac10ad | 1069 | mmu_free_roots(vcpu); |
6aa8b732 AK |
1070 | } |
1071 | ||
1072 | static int nonpaging_init_context(struct kvm_vcpu *vcpu) | |
1073 | { | |
1074 | struct kvm_mmu *context = &vcpu->mmu; | |
1075 | ||
1076 | context->new_cr3 = nonpaging_new_cr3; | |
1077 | context->page_fault = nonpaging_page_fault; | |
6aa8b732 AK |
1078 | context->gva_to_gpa = nonpaging_gva_to_gpa; |
1079 | context->free = nonpaging_free; | |
c7addb90 | 1080 | context->prefetch_page = nonpaging_prefetch_page; |
cea0f0e7 | 1081 | context->root_level = 0; |
6aa8b732 | 1082 | context->shadow_root_level = PT32E_ROOT_LEVEL; |
17c3ba9d | 1083 | context->root_hpa = INVALID_PAGE; |
6aa8b732 AK |
1084 | return 0; |
1085 | } | |
1086 | ||
6aa8b732 AK |
1087 | static void kvm_mmu_flush_tlb(struct kvm_vcpu *vcpu) |
1088 | { | |
1165f5fe | 1089 | ++vcpu->stat.tlb_flush; |
cbdd1bea | 1090 | kvm_x86_ops->tlb_flush(vcpu); |
6aa8b732 AK |
1091 | } |
1092 | ||
1093 | static void paging_new_cr3(struct kvm_vcpu *vcpu) | |
1094 | { | |
374cbac0 | 1095 | pgprintk("%s: cr3 %lx\n", __FUNCTION__, vcpu->cr3); |
cea0f0e7 | 1096 | mmu_free_roots(vcpu); |
6aa8b732 AK |
1097 | } |
1098 | ||
6aa8b732 AK |
1099 | static void inject_page_fault(struct kvm_vcpu *vcpu, |
1100 | u64 addr, | |
1101 | u32 err_code) | |
1102 | { | |
cbdd1bea | 1103 | kvm_x86_ops->inject_page_fault(vcpu, addr, err_code); |
6aa8b732 AK |
1104 | } |
1105 | ||
6aa8b732 AK |
1106 | static void paging_free(struct kvm_vcpu *vcpu) |
1107 | { | |
1108 | nonpaging_free(vcpu); | |
1109 | } | |
1110 | ||
1111 | #define PTTYPE 64 | |
1112 | #include "paging_tmpl.h" | |
1113 | #undef PTTYPE | |
1114 | ||
1115 | #define PTTYPE 32 | |
1116 | #include "paging_tmpl.h" | |
1117 | #undef PTTYPE | |
1118 | ||
17ac10ad | 1119 | static int paging64_init_context_common(struct kvm_vcpu *vcpu, int level) |
6aa8b732 AK |
1120 | { |
1121 | struct kvm_mmu *context = &vcpu->mmu; | |
1122 | ||
1123 | ASSERT(is_pae(vcpu)); | |
1124 | context->new_cr3 = paging_new_cr3; | |
1125 | context->page_fault = paging64_page_fault; | |
6aa8b732 | 1126 | context->gva_to_gpa = paging64_gva_to_gpa; |
c7addb90 | 1127 | context->prefetch_page = paging64_prefetch_page; |
6aa8b732 | 1128 | context->free = paging_free; |
17ac10ad AK |
1129 | context->root_level = level; |
1130 | context->shadow_root_level = level; | |
17c3ba9d | 1131 | context->root_hpa = INVALID_PAGE; |
6aa8b732 AK |
1132 | return 0; |
1133 | } | |
1134 | ||
17ac10ad AK |
1135 | static int paging64_init_context(struct kvm_vcpu *vcpu) |
1136 | { | |
1137 | return paging64_init_context_common(vcpu, PT64_ROOT_LEVEL); | |
1138 | } | |
1139 | ||
6aa8b732 AK |
1140 | static int paging32_init_context(struct kvm_vcpu *vcpu) |
1141 | { | |
1142 | struct kvm_mmu *context = &vcpu->mmu; | |
1143 | ||
1144 | context->new_cr3 = paging_new_cr3; | |
1145 | context->page_fault = paging32_page_fault; | |
6aa8b732 AK |
1146 | context->gva_to_gpa = paging32_gva_to_gpa; |
1147 | context->free = paging_free; | |
c7addb90 | 1148 | context->prefetch_page = paging32_prefetch_page; |
6aa8b732 AK |
1149 | context->root_level = PT32_ROOT_LEVEL; |
1150 | context->shadow_root_level = PT32E_ROOT_LEVEL; | |
17c3ba9d | 1151 | context->root_hpa = INVALID_PAGE; |
6aa8b732 AK |
1152 | return 0; |
1153 | } | |
1154 | ||
1155 | static int paging32E_init_context(struct kvm_vcpu *vcpu) | |
1156 | { | |
17ac10ad | 1157 | return paging64_init_context_common(vcpu, PT32E_ROOT_LEVEL); |
6aa8b732 AK |
1158 | } |
1159 | ||
1160 | static int init_kvm_mmu(struct kvm_vcpu *vcpu) | |
1161 | { | |
1162 | ASSERT(vcpu); | |
1163 | ASSERT(!VALID_PAGE(vcpu->mmu.root_hpa)); | |
1164 | ||
1165 | if (!is_paging(vcpu)) | |
1166 | return nonpaging_init_context(vcpu); | |
a9058ecd | 1167 | else if (is_long_mode(vcpu)) |
6aa8b732 AK |
1168 | return paging64_init_context(vcpu); |
1169 | else if (is_pae(vcpu)) | |
1170 | return paging32E_init_context(vcpu); | |
1171 | else | |
1172 | return paging32_init_context(vcpu); | |
1173 | } | |
1174 | ||
1175 | static void destroy_kvm_mmu(struct kvm_vcpu *vcpu) | |
1176 | { | |
1177 | ASSERT(vcpu); | |
1178 | if (VALID_PAGE(vcpu->mmu.root_hpa)) { | |
1179 | vcpu->mmu.free(vcpu); | |
1180 | vcpu->mmu.root_hpa = INVALID_PAGE; | |
1181 | } | |
1182 | } | |
1183 | ||
1184 | int kvm_mmu_reset_context(struct kvm_vcpu *vcpu) | |
17c3ba9d AK |
1185 | { |
1186 | destroy_kvm_mmu(vcpu); | |
1187 | return init_kvm_mmu(vcpu); | |
1188 | } | |
8668a3c4 | 1189 | EXPORT_SYMBOL_GPL(kvm_mmu_reset_context); |
17c3ba9d AK |
1190 | |
1191 | int kvm_mmu_load(struct kvm_vcpu *vcpu) | |
6aa8b732 | 1192 | { |
714b93da AK |
1193 | int r; |
1194 | ||
11ec2804 | 1195 | mutex_lock(&vcpu->kvm->lock); |
e2dec939 | 1196 | r = mmu_topup_memory_caches(vcpu); |
17c3ba9d AK |
1197 | if (r) |
1198 | goto out; | |
1199 | mmu_alloc_roots(vcpu); | |
cbdd1bea | 1200 | kvm_x86_ops->set_cr3(vcpu, vcpu->mmu.root_hpa); |
17c3ba9d | 1201 | kvm_mmu_flush_tlb(vcpu); |
714b93da | 1202 | out: |
11ec2804 | 1203 | mutex_unlock(&vcpu->kvm->lock); |
714b93da | 1204 | return r; |
6aa8b732 | 1205 | } |
17c3ba9d AK |
1206 | EXPORT_SYMBOL_GPL(kvm_mmu_load); |
1207 | ||
1208 | void kvm_mmu_unload(struct kvm_vcpu *vcpu) | |
1209 | { | |
1210 | mmu_free_roots(vcpu); | |
1211 | } | |
6aa8b732 | 1212 | |
09072daf | 1213 | static void mmu_pte_write_zap_pte(struct kvm_vcpu *vcpu, |
ac1b714e AK |
1214 | struct kvm_mmu_page *page, |
1215 | u64 *spte) | |
1216 | { | |
1217 | u64 pte; | |
1218 | struct kvm_mmu_page *child; | |
1219 | ||
1220 | pte = *spte; | |
c7addb90 | 1221 | if (is_shadow_present_pte(pte)) { |
ac1b714e | 1222 | if (page->role.level == PT_PAGE_TABLE_LEVEL) |
290fc38d | 1223 | rmap_remove(vcpu->kvm, spte); |
ac1b714e AK |
1224 | else { |
1225 | child = page_header(pte & PT64_BASE_ADDR_MASK); | |
90cb0529 | 1226 | mmu_page_remove_parent_pte(child, spte); |
ac1b714e AK |
1227 | } |
1228 | } | |
c7addb90 | 1229 | set_shadow_pte(spte, shadow_trap_nonpresent_pte); |
d9e368d6 | 1230 | kvm_flush_remote_tlbs(vcpu->kvm); |
ac1b714e AK |
1231 | } |
1232 | ||
0028425f AK |
1233 | static void mmu_pte_write_new_pte(struct kvm_vcpu *vcpu, |
1234 | struct kvm_mmu_page *page, | |
1235 | u64 *spte, | |
c7addb90 AK |
1236 | const void *new, int bytes, |
1237 | int offset_in_pte) | |
0028425f | 1238 | { |
4cee5764 AK |
1239 | if (page->role.level != PT_PAGE_TABLE_LEVEL) { |
1240 | ++vcpu->kvm->stat.mmu_pde_zapped; | |
0028425f | 1241 | return; |
4cee5764 | 1242 | } |
0028425f | 1243 | |
4cee5764 | 1244 | ++vcpu->kvm->stat.mmu_pte_updated; |
0028425f | 1245 | if (page->role.glevels == PT32_ROOT_LEVEL) |
c7addb90 AK |
1246 | paging32_update_pte(vcpu, page, spte, new, bytes, |
1247 | offset_in_pte); | |
0028425f | 1248 | else |
c7addb90 AK |
1249 | paging64_update_pte(vcpu, page, spte, new, bytes, |
1250 | offset_in_pte); | |
0028425f AK |
1251 | } |
1252 | ||
12b7d28f AK |
1253 | static bool last_updated_pte_accessed(struct kvm_vcpu *vcpu) |
1254 | { | |
1255 | u64 *spte = vcpu->last_pte_updated; | |
1256 | ||
1257 | return !!(spte && (*spte & PT_ACCESSED_MASK)); | |
1258 | } | |
1259 | ||
09072daf | 1260 | void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa, |
fe551881 | 1261 | const u8 *new, int bytes) |
da4a00f0 | 1262 | { |
9b7a0325 AK |
1263 | gfn_t gfn = gpa >> PAGE_SHIFT; |
1264 | struct kvm_mmu_page *page; | |
0e7bc4b9 | 1265 | struct hlist_node *node, *n; |
9b7a0325 AK |
1266 | struct hlist_head *bucket; |
1267 | unsigned index; | |
1268 | u64 *spte; | |
9b7a0325 | 1269 | unsigned offset = offset_in_page(gpa); |
0e7bc4b9 | 1270 | unsigned pte_size; |
9b7a0325 | 1271 | unsigned page_offset; |
0e7bc4b9 | 1272 | unsigned misaligned; |
fce0657f | 1273 | unsigned quadrant; |
9b7a0325 | 1274 | int level; |
86a5ba02 | 1275 | int flooded = 0; |
ac1b714e | 1276 | int npte; |
9b7a0325 | 1277 | |
da4a00f0 | 1278 | pgprintk("%s: gpa %llx bytes %d\n", __FUNCTION__, gpa, bytes); |
4cee5764 | 1279 | ++vcpu->kvm->stat.mmu_pte_write; |
c7addb90 | 1280 | kvm_mmu_audit(vcpu, "pre pte write"); |
12b7d28f AK |
1281 | if (gfn == vcpu->last_pt_write_gfn |
1282 | && !last_updated_pte_accessed(vcpu)) { | |
86a5ba02 AK |
1283 | ++vcpu->last_pt_write_count; |
1284 | if (vcpu->last_pt_write_count >= 3) | |
1285 | flooded = 1; | |
1286 | } else { | |
1287 | vcpu->last_pt_write_gfn = gfn; | |
1288 | vcpu->last_pt_write_count = 1; | |
12b7d28f | 1289 | vcpu->last_pte_updated = NULL; |
86a5ba02 | 1290 | } |
9b7a0325 AK |
1291 | index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES; |
1292 | bucket = &vcpu->kvm->mmu_page_hash[index]; | |
0e7bc4b9 | 1293 | hlist_for_each_entry_safe(page, node, n, bucket, hash_link) { |
9b7a0325 AK |
1294 | if (page->gfn != gfn || page->role.metaphysical) |
1295 | continue; | |
0e7bc4b9 AK |
1296 | pte_size = page->role.glevels == PT32_ROOT_LEVEL ? 4 : 8; |
1297 | misaligned = (offset ^ (offset + bytes - 1)) & ~(pte_size - 1); | |
e925c5ba | 1298 | misaligned |= bytes < 4; |
86a5ba02 | 1299 | if (misaligned || flooded) { |
0e7bc4b9 AK |
1300 | /* |
1301 | * Misaligned accesses are too much trouble to fix | |
1302 | * up; also, they usually indicate a page is not used | |
1303 | * as a page table. | |
86a5ba02 AK |
1304 | * |
1305 | * If we're seeing too many writes to a page, | |
1306 | * it may no longer be a page table, or we may be | |
1307 | * forking, in which case it is better to unmap the | |
1308 | * page. | |
0e7bc4b9 AK |
1309 | */ |
1310 | pgprintk("misaligned: gpa %llx bytes %d role %x\n", | |
1311 | gpa, bytes, page->role.word); | |
90cb0529 | 1312 | kvm_mmu_zap_page(vcpu->kvm, page); |
4cee5764 | 1313 | ++vcpu->kvm->stat.mmu_flooded; |
0e7bc4b9 AK |
1314 | continue; |
1315 | } | |
9b7a0325 AK |
1316 | page_offset = offset; |
1317 | level = page->role.level; | |
ac1b714e | 1318 | npte = 1; |
9b7a0325 | 1319 | if (page->role.glevels == PT32_ROOT_LEVEL) { |
ac1b714e AK |
1320 | page_offset <<= 1; /* 32->64 */ |
1321 | /* | |
1322 | * A 32-bit pde maps 4MB while the shadow pdes map | |
1323 | * only 2MB. So we need to double the offset again | |
1324 | * and zap two pdes instead of one. | |
1325 | */ | |
1326 | if (level == PT32_ROOT_LEVEL) { | |
6b8d0f9b | 1327 | page_offset &= ~7; /* kill rounding error */ |
ac1b714e AK |
1328 | page_offset <<= 1; |
1329 | npte = 2; | |
1330 | } | |
fce0657f | 1331 | quadrant = page_offset >> PAGE_SHIFT; |
9b7a0325 | 1332 | page_offset &= ~PAGE_MASK; |
fce0657f AK |
1333 | if (quadrant != page->role.quadrant) |
1334 | continue; | |
9b7a0325 | 1335 | } |
47ad8e68 | 1336 | spte = &page->spt[page_offset / sizeof(*spte)]; |
ac1b714e | 1337 | while (npte--) { |
09072daf | 1338 | mmu_pte_write_zap_pte(vcpu, page, spte); |
c7addb90 AK |
1339 | mmu_pte_write_new_pte(vcpu, page, spte, new, bytes, |
1340 | page_offset & (pte_size - 1)); | |
ac1b714e | 1341 | ++spte; |
9b7a0325 | 1342 | } |
9b7a0325 | 1343 | } |
c7addb90 | 1344 | kvm_mmu_audit(vcpu, "post pte write"); |
da4a00f0 AK |
1345 | } |
1346 | ||
a436036b AK |
1347 | int kvm_mmu_unprotect_page_virt(struct kvm_vcpu *vcpu, gva_t gva) |
1348 | { | |
1349 | gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, gva); | |
1350 | ||
f67a46f4 | 1351 | return kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT); |
a436036b AK |
1352 | } |
1353 | ||
22d95b12 | 1354 | void __kvm_mmu_free_some_pages(struct kvm_vcpu *vcpu) |
ebeace86 AK |
1355 | { |
1356 | while (vcpu->kvm->n_free_mmu_pages < KVM_REFILL_PAGES) { | |
1357 | struct kvm_mmu_page *page; | |
1358 | ||
1359 | page = container_of(vcpu->kvm->active_mmu_pages.prev, | |
1360 | struct kvm_mmu_page, link); | |
90cb0529 | 1361 | kvm_mmu_zap_page(vcpu->kvm, page); |
4cee5764 | 1362 | ++vcpu->kvm->stat.mmu_recycled; |
ebeace86 AK |
1363 | } |
1364 | } | |
ebeace86 | 1365 | |
3067714c AK |
1366 | int kvm_mmu_page_fault(struct kvm_vcpu *vcpu, gva_t cr2, u32 error_code) |
1367 | { | |
1368 | int r; | |
1369 | enum emulation_result er; | |
1370 | ||
1371 | mutex_lock(&vcpu->kvm->lock); | |
1372 | r = vcpu->mmu.page_fault(vcpu, cr2, error_code); | |
1373 | if (r < 0) | |
1374 | goto out; | |
1375 | ||
1376 | if (!r) { | |
1377 | r = 1; | |
1378 | goto out; | |
1379 | } | |
1380 | ||
b733bfb5 AK |
1381 | r = mmu_topup_memory_caches(vcpu); |
1382 | if (r) | |
1383 | goto out; | |
1384 | ||
3067714c AK |
1385 | er = emulate_instruction(vcpu, vcpu->run, cr2, error_code, 0); |
1386 | mutex_unlock(&vcpu->kvm->lock); | |
1387 | ||
1388 | switch (er) { | |
1389 | case EMULATE_DONE: | |
1390 | return 1; | |
1391 | case EMULATE_DO_MMIO: | |
1392 | ++vcpu->stat.mmio_exits; | |
1393 | return 0; | |
1394 | case EMULATE_FAIL: | |
1395 | kvm_report_emulation_failure(vcpu, "pagetable"); | |
1396 | return 1; | |
1397 | default: | |
1398 | BUG(); | |
1399 | } | |
1400 | out: | |
1401 | mutex_unlock(&vcpu->kvm->lock); | |
1402 | return r; | |
1403 | } | |
1404 | EXPORT_SYMBOL_GPL(kvm_mmu_page_fault); | |
1405 | ||
6aa8b732 AK |
1406 | static void free_mmu_pages(struct kvm_vcpu *vcpu) |
1407 | { | |
f51234c2 | 1408 | struct kvm_mmu_page *page; |
6aa8b732 | 1409 | |
f51234c2 AK |
1410 | while (!list_empty(&vcpu->kvm->active_mmu_pages)) { |
1411 | page = container_of(vcpu->kvm->active_mmu_pages.next, | |
1412 | struct kvm_mmu_page, link); | |
90cb0529 | 1413 | kvm_mmu_zap_page(vcpu->kvm, page); |
f51234c2 | 1414 | } |
17ac10ad | 1415 | free_page((unsigned long)vcpu->mmu.pae_root); |
6aa8b732 AK |
1416 | } |
1417 | ||
1418 | static int alloc_mmu_pages(struct kvm_vcpu *vcpu) | |
1419 | { | |
17ac10ad | 1420 | struct page *page; |
6aa8b732 AK |
1421 | int i; |
1422 | ||
1423 | ASSERT(vcpu); | |
1424 | ||
82ce2c96 IE |
1425 | if (vcpu->kvm->n_requested_mmu_pages) |
1426 | vcpu->kvm->n_free_mmu_pages = vcpu->kvm->n_requested_mmu_pages; | |
1427 | else | |
1428 | vcpu->kvm->n_free_mmu_pages = vcpu->kvm->n_alloc_mmu_pages; | |
17ac10ad AK |
1429 | /* |
1430 | * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64. | |
1431 | * Therefore we need to allocate shadow page tables in the first | |
1432 | * 4GB of memory, which happens to fit the DMA32 zone. | |
1433 | */ | |
1434 | page = alloc_page(GFP_KERNEL | __GFP_DMA32); | |
1435 | if (!page) | |
1436 | goto error_1; | |
1437 | vcpu->mmu.pae_root = page_address(page); | |
1438 | for (i = 0; i < 4; ++i) | |
1439 | vcpu->mmu.pae_root[i] = INVALID_PAGE; | |
1440 | ||
6aa8b732 AK |
1441 | return 0; |
1442 | ||
1443 | error_1: | |
1444 | free_mmu_pages(vcpu); | |
1445 | return -ENOMEM; | |
1446 | } | |
1447 | ||
8018c27b | 1448 | int kvm_mmu_create(struct kvm_vcpu *vcpu) |
6aa8b732 | 1449 | { |
6aa8b732 AK |
1450 | ASSERT(vcpu); |
1451 | ASSERT(!VALID_PAGE(vcpu->mmu.root_hpa)); | |
6aa8b732 | 1452 | |
8018c27b IM |
1453 | return alloc_mmu_pages(vcpu); |
1454 | } | |
6aa8b732 | 1455 | |
8018c27b IM |
1456 | int kvm_mmu_setup(struct kvm_vcpu *vcpu) |
1457 | { | |
1458 | ASSERT(vcpu); | |
1459 | ASSERT(!VALID_PAGE(vcpu->mmu.root_hpa)); | |
2c264957 | 1460 | |
8018c27b | 1461 | return init_kvm_mmu(vcpu); |
6aa8b732 AK |
1462 | } |
1463 | ||
1464 | void kvm_mmu_destroy(struct kvm_vcpu *vcpu) | |
1465 | { | |
1466 | ASSERT(vcpu); | |
1467 | ||
1468 | destroy_kvm_mmu(vcpu); | |
1469 | free_mmu_pages(vcpu); | |
714b93da | 1470 | mmu_free_memory_caches(vcpu); |
6aa8b732 AK |
1471 | } |
1472 | ||
90cb0529 | 1473 | void kvm_mmu_slot_remove_write_access(struct kvm *kvm, int slot) |
6aa8b732 AK |
1474 | { |
1475 | struct kvm_mmu_page *page; | |
1476 | ||
1477 | list_for_each_entry(page, &kvm->active_mmu_pages, link) { | |
1478 | int i; | |
1479 | u64 *pt; | |
1480 | ||
1481 | if (!test_bit(slot, &page->slot_bitmap)) | |
1482 | continue; | |
1483 | ||
47ad8e68 | 1484 | pt = page->spt; |
6aa8b732 AK |
1485 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) |
1486 | /* avoid RMW */ | |
9647c14c | 1487 | if (pt[i] & PT_WRITABLE_MASK) |
6aa8b732 | 1488 | pt[i] &= ~PT_WRITABLE_MASK; |
6aa8b732 AK |
1489 | } |
1490 | } | |
37a7d8b0 | 1491 | |
90cb0529 | 1492 | void kvm_mmu_zap_all(struct kvm *kvm) |
e0fa826f | 1493 | { |
90cb0529 | 1494 | struct kvm_mmu_page *page, *node; |
e0fa826f | 1495 | |
90cb0529 AK |
1496 | list_for_each_entry_safe(page, node, &kvm->active_mmu_pages, link) |
1497 | kvm_mmu_zap_page(kvm, page); | |
e0fa826f | 1498 | |
90cb0529 | 1499 | kvm_flush_remote_tlbs(kvm); |
e0fa826f DL |
1500 | } |
1501 | ||
b5a33a75 AK |
1502 | void kvm_mmu_module_exit(void) |
1503 | { | |
1504 | if (pte_chain_cache) | |
1505 | kmem_cache_destroy(pte_chain_cache); | |
1506 | if (rmap_desc_cache) | |
1507 | kmem_cache_destroy(rmap_desc_cache); | |
d3d25b04 AK |
1508 | if (mmu_page_header_cache) |
1509 | kmem_cache_destroy(mmu_page_header_cache); | |
b5a33a75 AK |
1510 | } |
1511 | ||
1512 | int kvm_mmu_module_init(void) | |
1513 | { | |
1514 | pte_chain_cache = kmem_cache_create("kvm_pte_chain", | |
1515 | sizeof(struct kvm_pte_chain), | |
20c2df83 | 1516 | 0, 0, NULL); |
b5a33a75 AK |
1517 | if (!pte_chain_cache) |
1518 | goto nomem; | |
1519 | rmap_desc_cache = kmem_cache_create("kvm_rmap_desc", | |
1520 | sizeof(struct kvm_rmap_desc), | |
20c2df83 | 1521 | 0, 0, NULL); |
b5a33a75 AK |
1522 | if (!rmap_desc_cache) |
1523 | goto nomem; | |
1524 | ||
d3d25b04 AK |
1525 | mmu_page_header_cache = kmem_cache_create("kvm_mmu_page_header", |
1526 | sizeof(struct kvm_mmu_page), | |
20c2df83 | 1527 | 0, 0, NULL); |
d3d25b04 AK |
1528 | if (!mmu_page_header_cache) |
1529 | goto nomem; | |
1530 | ||
b5a33a75 AK |
1531 | return 0; |
1532 | ||
1533 | nomem: | |
1534 | kvm_mmu_module_exit(); | |
1535 | return -ENOMEM; | |
1536 | } | |
1537 | ||
3ad82a7e ZX |
1538 | /* |
1539 | * Caculate mmu pages needed for kvm. | |
1540 | */ | |
1541 | unsigned int kvm_mmu_calculate_mmu_pages(struct kvm *kvm) | |
1542 | { | |
1543 | int i; | |
1544 | unsigned int nr_mmu_pages; | |
1545 | unsigned int nr_pages = 0; | |
1546 | ||
1547 | for (i = 0; i < kvm->nmemslots; i++) | |
1548 | nr_pages += kvm->memslots[i].npages; | |
1549 | ||
1550 | nr_mmu_pages = nr_pages * KVM_PERMILLE_MMU_PAGES / 1000; | |
1551 | nr_mmu_pages = max(nr_mmu_pages, | |
1552 | (unsigned int) KVM_MIN_ALLOC_MMU_PAGES); | |
1553 | ||
1554 | return nr_mmu_pages; | |
1555 | } | |
1556 | ||
37a7d8b0 AK |
1557 | #ifdef AUDIT |
1558 | ||
1559 | static const char *audit_msg; | |
1560 | ||
1561 | static gva_t canonicalize(gva_t gva) | |
1562 | { | |
1563 | #ifdef CONFIG_X86_64 | |
1564 | gva = (long long)(gva << 16) >> 16; | |
1565 | #endif | |
1566 | return gva; | |
1567 | } | |
1568 | ||
1569 | static void audit_mappings_page(struct kvm_vcpu *vcpu, u64 page_pte, | |
1570 | gva_t va, int level) | |
1571 | { | |
1572 | u64 *pt = __va(page_pte & PT64_BASE_ADDR_MASK); | |
1573 | int i; | |
1574 | gva_t va_delta = 1ul << (PAGE_SHIFT + 9 * (level - 1)); | |
1575 | ||
1576 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i, va += va_delta) { | |
1577 | u64 ent = pt[i]; | |
1578 | ||
c7addb90 | 1579 | if (ent == shadow_trap_nonpresent_pte) |
37a7d8b0 AK |
1580 | continue; |
1581 | ||
1582 | va = canonicalize(va); | |
c7addb90 AK |
1583 | if (level > 1) { |
1584 | if (ent == shadow_notrap_nonpresent_pte) | |
1585 | printk(KERN_ERR "audit: (%s) nontrapping pte" | |
1586 | " in nonleaf level: levels %d gva %lx" | |
1587 | " level %d pte %llx\n", audit_msg, | |
1588 | vcpu->mmu.root_level, va, level, ent); | |
1589 | ||
37a7d8b0 | 1590 | audit_mappings_page(vcpu, ent, va, level - 1); |
c7addb90 | 1591 | } else { |
37a7d8b0 AK |
1592 | gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, va); |
1593 | hpa_t hpa = gpa_to_hpa(vcpu, gpa); | |
8a7ae055 | 1594 | struct page *page; |
37a7d8b0 | 1595 | |
c7addb90 | 1596 | if (is_shadow_present_pte(ent) |
37a7d8b0 | 1597 | && (ent & PT64_BASE_ADDR_MASK) != hpa) |
c7addb90 AK |
1598 | printk(KERN_ERR "xx audit error: (%s) levels %d" |
1599 | " gva %lx gpa %llx hpa %llx ent %llx %d\n", | |
37a7d8b0 | 1600 | audit_msg, vcpu->mmu.root_level, |
d77c26fc MD |
1601 | va, gpa, hpa, ent, |
1602 | is_shadow_present_pte(ent)); | |
c7addb90 AK |
1603 | else if (ent == shadow_notrap_nonpresent_pte |
1604 | && !is_error_hpa(hpa)) | |
1605 | printk(KERN_ERR "audit: (%s) notrap shadow," | |
1606 | " valid guest gva %lx\n", audit_msg, va); | |
8a7ae055 IE |
1607 | page = pfn_to_page((gpa & PT64_BASE_ADDR_MASK) |
1608 | >> PAGE_SHIFT); | |
b4231d61 | 1609 | kvm_release_page_clean(page); |
c7addb90 | 1610 | |
37a7d8b0 AK |
1611 | } |
1612 | } | |
1613 | } | |
1614 | ||
1615 | static void audit_mappings(struct kvm_vcpu *vcpu) | |
1616 | { | |
1ea252af | 1617 | unsigned i; |
37a7d8b0 AK |
1618 | |
1619 | if (vcpu->mmu.root_level == 4) | |
1620 | audit_mappings_page(vcpu, vcpu->mmu.root_hpa, 0, 4); | |
1621 | else | |
1622 | for (i = 0; i < 4; ++i) | |
1623 | if (vcpu->mmu.pae_root[i] & PT_PRESENT_MASK) | |
1624 | audit_mappings_page(vcpu, | |
1625 | vcpu->mmu.pae_root[i], | |
1626 | i << 30, | |
1627 | 2); | |
1628 | } | |
1629 | ||
1630 | static int count_rmaps(struct kvm_vcpu *vcpu) | |
1631 | { | |
1632 | int nmaps = 0; | |
1633 | int i, j, k; | |
1634 | ||
1635 | for (i = 0; i < KVM_MEMORY_SLOTS; ++i) { | |
1636 | struct kvm_memory_slot *m = &vcpu->kvm->memslots[i]; | |
1637 | struct kvm_rmap_desc *d; | |
1638 | ||
1639 | for (j = 0; j < m->npages; ++j) { | |
290fc38d | 1640 | unsigned long *rmapp = &m->rmap[j]; |
37a7d8b0 | 1641 | |
290fc38d | 1642 | if (!*rmapp) |
37a7d8b0 | 1643 | continue; |
290fc38d | 1644 | if (!(*rmapp & 1)) { |
37a7d8b0 AK |
1645 | ++nmaps; |
1646 | continue; | |
1647 | } | |
290fc38d | 1648 | d = (struct kvm_rmap_desc *)(*rmapp & ~1ul); |
37a7d8b0 AK |
1649 | while (d) { |
1650 | for (k = 0; k < RMAP_EXT; ++k) | |
1651 | if (d->shadow_ptes[k]) | |
1652 | ++nmaps; | |
1653 | else | |
1654 | break; | |
1655 | d = d->more; | |
1656 | } | |
1657 | } | |
1658 | } | |
1659 | return nmaps; | |
1660 | } | |
1661 | ||
1662 | static int count_writable_mappings(struct kvm_vcpu *vcpu) | |
1663 | { | |
1664 | int nmaps = 0; | |
1665 | struct kvm_mmu_page *page; | |
1666 | int i; | |
1667 | ||
1668 | list_for_each_entry(page, &vcpu->kvm->active_mmu_pages, link) { | |
47ad8e68 | 1669 | u64 *pt = page->spt; |
37a7d8b0 AK |
1670 | |
1671 | if (page->role.level != PT_PAGE_TABLE_LEVEL) | |
1672 | continue; | |
1673 | ||
1674 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) { | |
1675 | u64 ent = pt[i]; | |
1676 | ||
1677 | if (!(ent & PT_PRESENT_MASK)) | |
1678 | continue; | |
1679 | if (!(ent & PT_WRITABLE_MASK)) | |
1680 | continue; | |
1681 | ++nmaps; | |
1682 | } | |
1683 | } | |
1684 | return nmaps; | |
1685 | } | |
1686 | ||
1687 | static void audit_rmap(struct kvm_vcpu *vcpu) | |
1688 | { | |
1689 | int n_rmap = count_rmaps(vcpu); | |
1690 | int n_actual = count_writable_mappings(vcpu); | |
1691 | ||
1692 | if (n_rmap != n_actual) | |
1693 | printk(KERN_ERR "%s: (%s) rmap %d actual %d\n", | |
1694 | __FUNCTION__, audit_msg, n_rmap, n_actual); | |
1695 | } | |
1696 | ||
1697 | static void audit_write_protection(struct kvm_vcpu *vcpu) | |
1698 | { | |
1699 | struct kvm_mmu_page *page; | |
290fc38d IE |
1700 | struct kvm_memory_slot *slot; |
1701 | unsigned long *rmapp; | |
1702 | gfn_t gfn; | |
37a7d8b0 AK |
1703 | |
1704 | list_for_each_entry(page, &vcpu->kvm->active_mmu_pages, link) { | |
37a7d8b0 AK |
1705 | if (page->role.metaphysical) |
1706 | continue; | |
1707 | ||
290fc38d IE |
1708 | slot = gfn_to_memslot(vcpu->kvm, page->gfn); |
1709 | gfn = unalias_gfn(vcpu->kvm, page->gfn); | |
1710 | rmapp = &slot->rmap[gfn - slot->base_gfn]; | |
1711 | if (*rmapp) | |
37a7d8b0 AK |
1712 | printk(KERN_ERR "%s: (%s) shadow page has writable" |
1713 | " mappings: gfn %lx role %x\n", | |
1714 | __FUNCTION__, audit_msg, page->gfn, | |
1715 | page->role.word); | |
1716 | } | |
1717 | } | |
1718 | ||
1719 | static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg) | |
1720 | { | |
1721 | int olddbg = dbg; | |
1722 | ||
1723 | dbg = 0; | |
1724 | audit_msg = msg; | |
1725 | audit_rmap(vcpu); | |
1726 | audit_write_protection(vcpu); | |
1727 | audit_mappings(vcpu); | |
1728 | dbg = olddbg; | |
1729 | } | |
1730 | ||
1731 | #endif |