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 | |
d77c26fc MD |
93 | #define PT32_DIR_PSE36_MASK \ |
94 | (((1ULL << PT32_DIR_PSE36_SIZE) - 1) << PT32_DIR_PSE36_SHIFT) | |
6aa8b732 AK |
95 | |
96 | ||
6aa8b732 AK |
97 | #define PT_FIRST_AVAIL_BITS_SHIFT 9 |
98 | #define PT64_SECOND_AVAIL_BITS_SHIFT 52 | |
99 | ||
6aa8b732 AK |
100 | #define PT_SHADOW_IO_MARK (1ULL << PT_FIRST_AVAIL_BITS_SHIFT) |
101 | ||
6aa8b732 AK |
102 | #define VALID_PAGE(x) ((x) != INVALID_PAGE) |
103 | ||
104 | #define PT64_LEVEL_BITS 9 | |
105 | ||
106 | #define PT64_LEVEL_SHIFT(level) \ | |
d77c26fc | 107 | (PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS) |
6aa8b732 AK |
108 | |
109 | #define PT64_LEVEL_MASK(level) \ | |
110 | (((1ULL << PT64_LEVEL_BITS) - 1) << PT64_LEVEL_SHIFT(level)) | |
111 | ||
112 | #define PT64_INDEX(address, level)\ | |
113 | (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1)) | |
114 | ||
115 | ||
116 | #define PT32_LEVEL_BITS 10 | |
117 | ||
118 | #define PT32_LEVEL_SHIFT(level) \ | |
d77c26fc | 119 | (PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS) |
6aa8b732 AK |
120 | |
121 | #define PT32_LEVEL_MASK(level) \ | |
122 | (((1ULL << PT32_LEVEL_BITS) - 1) << PT32_LEVEL_SHIFT(level)) | |
123 | ||
124 | #define PT32_INDEX(address, level)\ | |
125 | (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1)) | |
126 | ||
127 | ||
27aba766 | 128 | #define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1)) |
6aa8b732 AK |
129 | #define PT64_DIR_BASE_ADDR_MASK \ |
130 | (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1)) | |
131 | ||
132 | #define PT32_BASE_ADDR_MASK PAGE_MASK | |
133 | #define PT32_DIR_BASE_ADDR_MASK \ | |
134 | (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1)) | |
135 | ||
136 | ||
137 | #define PFERR_PRESENT_MASK (1U << 0) | |
138 | #define PFERR_WRITE_MASK (1U << 1) | |
139 | #define PFERR_USER_MASK (1U << 2) | |
73b1087e | 140 | #define PFERR_FETCH_MASK (1U << 4) |
6aa8b732 AK |
141 | |
142 | #define PT64_ROOT_LEVEL 4 | |
143 | #define PT32_ROOT_LEVEL 2 | |
144 | #define PT32E_ROOT_LEVEL 3 | |
145 | ||
146 | #define PT_DIRECTORY_LEVEL 2 | |
147 | #define PT_PAGE_TABLE_LEVEL 1 | |
148 | ||
cd4a4e53 AK |
149 | #define RMAP_EXT 4 |
150 | ||
151 | struct kvm_rmap_desc { | |
152 | u64 *shadow_ptes[RMAP_EXT]; | |
153 | struct kvm_rmap_desc *more; | |
154 | }; | |
155 | ||
b5a33a75 AK |
156 | static struct kmem_cache *pte_chain_cache; |
157 | static struct kmem_cache *rmap_desc_cache; | |
d3d25b04 | 158 | static struct kmem_cache *mmu_page_header_cache; |
b5a33a75 | 159 | |
c7addb90 AK |
160 | static u64 __read_mostly shadow_trap_nonpresent_pte; |
161 | static u64 __read_mostly shadow_notrap_nonpresent_pte; | |
162 | ||
163 | void kvm_mmu_set_nonpresent_ptes(u64 trap_pte, u64 notrap_pte) | |
164 | { | |
165 | shadow_trap_nonpresent_pte = trap_pte; | |
166 | shadow_notrap_nonpresent_pte = notrap_pte; | |
167 | } | |
168 | EXPORT_SYMBOL_GPL(kvm_mmu_set_nonpresent_ptes); | |
169 | ||
6aa8b732 AK |
170 | static int is_write_protection(struct kvm_vcpu *vcpu) |
171 | { | |
707d92fa | 172 | return vcpu->cr0 & X86_CR0_WP; |
6aa8b732 AK |
173 | } |
174 | ||
175 | static int is_cpuid_PSE36(void) | |
176 | { | |
177 | return 1; | |
178 | } | |
179 | ||
73b1087e AK |
180 | static int is_nx(struct kvm_vcpu *vcpu) |
181 | { | |
182 | return vcpu->shadow_efer & EFER_NX; | |
183 | } | |
184 | ||
6aa8b732 AK |
185 | static int is_present_pte(unsigned long pte) |
186 | { | |
187 | return pte & PT_PRESENT_MASK; | |
188 | } | |
189 | ||
c7addb90 AK |
190 | static int is_shadow_present_pte(u64 pte) |
191 | { | |
192 | pte &= ~PT_SHADOW_IO_MARK; | |
193 | return pte != shadow_trap_nonpresent_pte | |
194 | && pte != shadow_notrap_nonpresent_pte; | |
195 | } | |
196 | ||
6aa8b732 AK |
197 | static int is_writeble_pte(unsigned long pte) |
198 | { | |
199 | return pte & PT_WRITABLE_MASK; | |
200 | } | |
201 | ||
e3c5e7ec AK |
202 | static int is_dirty_pte(unsigned long pte) |
203 | { | |
204 | return pte & PT_DIRTY_MASK; | |
205 | } | |
206 | ||
6aa8b732 AK |
207 | static int is_io_pte(unsigned long pte) |
208 | { | |
209 | return pte & PT_SHADOW_IO_MARK; | |
210 | } | |
211 | ||
cd4a4e53 AK |
212 | static int is_rmap_pte(u64 pte) |
213 | { | |
9647c14c IE |
214 | return pte != shadow_trap_nonpresent_pte |
215 | && pte != shadow_notrap_nonpresent_pte; | |
cd4a4e53 AK |
216 | } |
217 | ||
e663ee64 AK |
218 | static void set_shadow_pte(u64 *sptep, u64 spte) |
219 | { | |
220 | #ifdef CONFIG_X86_64 | |
221 | set_64bit((unsigned long *)sptep, spte); | |
222 | #else | |
223 | set_64bit((unsigned long long *)sptep, spte); | |
224 | #endif | |
225 | } | |
226 | ||
e2dec939 | 227 | static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache, |
2e3e5882 | 228 | struct kmem_cache *base_cache, int min) |
714b93da AK |
229 | { |
230 | void *obj; | |
231 | ||
232 | if (cache->nobjs >= min) | |
e2dec939 | 233 | return 0; |
714b93da | 234 | while (cache->nobjs < ARRAY_SIZE(cache->objects)) { |
2e3e5882 | 235 | obj = kmem_cache_zalloc(base_cache, GFP_KERNEL); |
714b93da | 236 | if (!obj) |
e2dec939 | 237 | return -ENOMEM; |
714b93da AK |
238 | cache->objects[cache->nobjs++] = obj; |
239 | } | |
e2dec939 | 240 | return 0; |
714b93da AK |
241 | } |
242 | ||
243 | static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc) | |
244 | { | |
245 | while (mc->nobjs) | |
246 | kfree(mc->objects[--mc->nobjs]); | |
247 | } | |
248 | ||
c1158e63 | 249 | static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache *cache, |
2e3e5882 | 250 | int min) |
c1158e63 AK |
251 | { |
252 | struct page *page; | |
253 | ||
254 | if (cache->nobjs >= min) | |
255 | return 0; | |
256 | while (cache->nobjs < ARRAY_SIZE(cache->objects)) { | |
2e3e5882 | 257 | page = alloc_page(GFP_KERNEL); |
c1158e63 AK |
258 | if (!page) |
259 | return -ENOMEM; | |
260 | set_page_private(page, 0); | |
261 | cache->objects[cache->nobjs++] = page_address(page); | |
262 | } | |
263 | return 0; | |
264 | } | |
265 | ||
266 | static void mmu_free_memory_cache_page(struct kvm_mmu_memory_cache *mc) | |
267 | { | |
268 | while (mc->nobjs) | |
c4d198d5 | 269 | free_page((unsigned long)mc->objects[--mc->nobjs]); |
c1158e63 AK |
270 | } |
271 | ||
2e3e5882 | 272 | static int mmu_topup_memory_caches(struct kvm_vcpu *vcpu) |
714b93da | 273 | { |
e2dec939 AK |
274 | int r; |
275 | ||
2e3e5882 | 276 | kvm_mmu_free_some_pages(vcpu); |
e2dec939 | 277 | r = mmu_topup_memory_cache(&vcpu->mmu_pte_chain_cache, |
2e3e5882 | 278 | pte_chain_cache, 4); |
e2dec939 AK |
279 | if (r) |
280 | goto out; | |
281 | r = mmu_topup_memory_cache(&vcpu->mmu_rmap_desc_cache, | |
2e3e5882 | 282 | rmap_desc_cache, 1); |
d3d25b04 AK |
283 | if (r) |
284 | goto out; | |
290fc38d | 285 | r = mmu_topup_memory_cache_page(&vcpu->mmu_page_cache, 8); |
d3d25b04 AK |
286 | if (r) |
287 | goto out; | |
288 | r = mmu_topup_memory_cache(&vcpu->mmu_page_header_cache, | |
2e3e5882 | 289 | mmu_page_header_cache, 4); |
e2dec939 AK |
290 | out: |
291 | return r; | |
714b93da AK |
292 | } |
293 | ||
294 | static void mmu_free_memory_caches(struct kvm_vcpu *vcpu) | |
295 | { | |
296 | mmu_free_memory_cache(&vcpu->mmu_pte_chain_cache); | |
297 | mmu_free_memory_cache(&vcpu->mmu_rmap_desc_cache); | |
c1158e63 | 298 | mmu_free_memory_cache_page(&vcpu->mmu_page_cache); |
d3d25b04 | 299 | mmu_free_memory_cache(&vcpu->mmu_page_header_cache); |
714b93da AK |
300 | } |
301 | ||
302 | static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc, | |
303 | size_t size) | |
304 | { | |
305 | void *p; | |
306 | ||
307 | BUG_ON(!mc->nobjs); | |
308 | p = mc->objects[--mc->nobjs]; | |
309 | memset(p, 0, size); | |
310 | return p; | |
311 | } | |
312 | ||
714b93da AK |
313 | static struct kvm_pte_chain *mmu_alloc_pte_chain(struct kvm_vcpu *vcpu) |
314 | { | |
315 | return mmu_memory_cache_alloc(&vcpu->mmu_pte_chain_cache, | |
316 | sizeof(struct kvm_pte_chain)); | |
317 | } | |
318 | ||
90cb0529 | 319 | static void mmu_free_pte_chain(struct kvm_pte_chain *pc) |
714b93da | 320 | { |
90cb0529 | 321 | kfree(pc); |
714b93da AK |
322 | } |
323 | ||
324 | static struct kvm_rmap_desc *mmu_alloc_rmap_desc(struct kvm_vcpu *vcpu) | |
325 | { | |
326 | return mmu_memory_cache_alloc(&vcpu->mmu_rmap_desc_cache, | |
327 | sizeof(struct kvm_rmap_desc)); | |
328 | } | |
329 | ||
90cb0529 | 330 | static void mmu_free_rmap_desc(struct kvm_rmap_desc *rd) |
714b93da | 331 | { |
90cb0529 | 332 | kfree(rd); |
714b93da AK |
333 | } |
334 | ||
290fc38d IE |
335 | /* |
336 | * Take gfn and return the reverse mapping to it. | |
337 | * Note: gfn must be unaliased before this function get called | |
338 | */ | |
339 | ||
340 | static unsigned long *gfn_to_rmap(struct kvm *kvm, gfn_t gfn) | |
341 | { | |
342 | struct kvm_memory_slot *slot; | |
343 | ||
344 | slot = gfn_to_memslot(kvm, gfn); | |
345 | return &slot->rmap[gfn - slot->base_gfn]; | |
346 | } | |
347 | ||
cd4a4e53 AK |
348 | /* |
349 | * Reverse mapping data structures: | |
350 | * | |
290fc38d IE |
351 | * If rmapp bit zero is zero, then rmapp point to the shadw page table entry |
352 | * that points to page_address(page). | |
cd4a4e53 | 353 | * |
290fc38d IE |
354 | * If rmapp bit zero is one, (then rmap & ~1) points to a struct kvm_rmap_desc |
355 | * containing more mappings. | |
cd4a4e53 | 356 | */ |
290fc38d | 357 | static void rmap_add(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn) |
cd4a4e53 | 358 | { |
290fc38d | 359 | struct kvm_mmu_page *page; |
cd4a4e53 | 360 | struct kvm_rmap_desc *desc; |
290fc38d | 361 | unsigned long *rmapp; |
cd4a4e53 AK |
362 | int i; |
363 | ||
364 | if (!is_rmap_pte(*spte)) | |
365 | return; | |
290fc38d IE |
366 | gfn = unalias_gfn(vcpu->kvm, gfn); |
367 | page = page_header(__pa(spte)); | |
368 | page->gfns[spte - page->spt] = gfn; | |
369 | rmapp = gfn_to_rmap(vcpu->kvm, gfn); | |
370 | if (!*rmapp) { | |
cd4a4e53 | 371 | rmap_printk("rmap_add: %p %llx 0->1\n", spte, *spte); |
290fc38d IE |
372 | *rmapp = (unsigned long)spte; |
373 | } else if (!(*rmapp & 1)) { | |
cd4a4e53 | 374 | rmap_printk("rmap_add: %p %llx 1->many\n", spte, *spte); |
714b93da | 375 | desc = mmu_alloc_rmap_desc(vcpu); |
290fc38d | 376 | desc->shadow_ptes[0] = (u64 *)*rmapp; |
cd4a4e53 | 377 | desc->shadow_ptes[1] = spte; |
290fc38d | 378 | *rmapp = (unsigned long)desc | 1; |
cd4a4e53 AK |
379 | } else { |
380 | rmap_printk("rmap_add: %p %llx many->many\n", spte, *spte); | |
290fc38d | 381 | desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul); |
cd4a4e53 AK |
382 | while (desc->shadow_ptes[RMAP_EXT-1] && desc->more) |
383 | desc = desc->more; | |
384 | if (desc->shadow_ptes[RMAP_EXT-1]) { | |
714b93da | 385 | desc->more = mmu_alloc_rmap_desc(vcpu); |
cd4a4e53 AK |
386 | desc = desc->more; |
387 | } | |
388 | for (i = 0; desc->shadow_ptes[i]; ++i) | |
389 | ; | |
390 | desc->shadow_ptes[i] = spte; | |
391 | } | |
392 | } | |
393 | ||
290fc38d | 394 | static void rmap_desc_remove_entry(unsigned long *rmapp, |
cd4a4e53 AK |
395 | struct kvm_rmap_desc *desc, |
396 | int i, | |
397 | struct kvm_rmap_desc *prev_desc) | |
398 | { | |
399 | int j; | |
400 | ||
401 | for (j = RMAP_EXT - 1; !desc->shadow_ptes[j] && j > i; --j) | |
402 | ; | |
403 | desc->shadow_ptes[i] = desc->shadow_ptes[j]; | |
11718b4d | 404 | desc->shadow_ptes[j] = NULL; |
cd4a4e53 AK |
405 | if (j != 0) |
406 | return; | |
407 | if (!prev_desc && !desc->more) | |
290fc38d | 408 | *rmapp = (unsigned long)desc->shadow_ptes[0]; |
cd4a4e53 AK |
409 | else |
410 | if (prev_desc) | |
411 | prev_desc->more = desc->more; | |
412 | else | |
290fc38d | 413 | *rmapp = (unsigned long)desc->more | 1; |
90cb0529 | 414 | mmu_free_rmap_desc(desc); |
cd4a4e53 AK |
415 | } |
416 | ||
290fc38d | 417 | static void rmap_remove(struct kvm *kvm, u64 *spte) |
cd4a4e53 | 418 | { |
cd4a4e53 AK |
419 | struct kvm_rmap_desc *desc; |
420 | struct kvm_rmap_desc *prev_desc; | |
290fc38d IE |
421 | struct kvm_mmu_page *page; |
422 | unsigned long *rmapp; | |
cd4a4e53 AK |
423 | int i; |
424 | ||
425 | if (!is_rmap_pte(*spte)) | |
426 | return; | |
290fc38d IE |
427 | page = page_header(__pa(spte)); |
428 | rmapp = gfn_to_rmap(kvm, page->gfns[spte - page->spt]); | |
429 | if (!*rmapp) { | |
cd4a4e53 AK |
430 | printk(KERN_ERR "rmap_remove: %p %llx 0->BUG\n", spte, *spte); |
431 | BUG(); | |
290fc38d | 432 | } else if (!(*rmapp & 1)) { |
cd4a4e53 | 433 | rmap_printk("rmap_remove: %p %llx 1->0\n", spte, *spte); |
290fc38d | 434 | if ((u64 *)*rmapp != spte) { |
cd4a4e53 AK |
435 | printk(KERN_ERR "rmap_remove: %p %llx 1->BUG\n", |
436 | spte, *spte); | |
437 | BUG(); | |
438 | } | |
290fc38d | 439 | *rmapp = 0; |
cd4a4e53 AK |
440 | } else { |
441 | rmap_printk("rmap_remove: %p %llx many->many\n", spte, *spte); | |
290fc38d | 442 | desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul); |
cd4a4e53 AK |
443 | prev_desc = NULL; |
444 | while (desc) { | |
445 | for (i = 0; i < RMAP_EXT && desc->shadow_ptes[i]; ++i) | |
446 | if (desc->shadow_ptes[i] == spte) { | |
290fc38d | 447 | rmap_desc_remove_entry(rmapp, |
714b93da | 448 | desc, i, |
cd4a4e53 AK |
449 | prev_desc); |
450 | return; | |
451 | } | |
452 | prev_desc = desc; | |
453 | desc = desc->more; | |
454 | } | |
455 | BUG(); | |
456 | } | |
457 | } | |
458 | ||
98348e95 | 459 | static u64 *rmap_next(struct kvm *kvm, unsigned long *rmapp, u64 *spte) |
374cbac0 | 460 | { |
374cbac0 | 461 | struct kvm_rmap_desc *desc; |
98348e95 IE |
462 | struct kvm_rmap_desc *prev_desc; |
463 | u64 *prev_spte; | |
464 | int i; | |
465 | ||
466 | if (!*rmapp) | |
467 | return NULL; | |
468 | else if (!(*rmapp & 1)) { | |
469 | if (!spte) | |
470 | return (u64 *)*rmapp; | |
471 | return NULL; | |
472 | } | |
473 | desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul); | |
474 | prev_desc = NULL; | |
475 | prev_spte = NULL; | |
476 | while (desc) { | |
477 | for (i = 0; i < RMAP_EXT && desc->shadow_ptes[i]; ++i) { | |
478 | if (prev_spte == spte) | |
479 | return desc->shadow_ptes[i]; | |
480 | prev_spte = desc->shadow_ptes[i]; | |
481 | } | |
482 | desc = desc->more; | |
483 | } | |
484 | return NULL; | |
485 | } | |
486 | ||
487 | static void rmap_write_protect(struct kvm *kvm, u64 gfn) | |
488 | { | |
290fc38d | 489 | unsigned long *rmapp; |
374cbac0 AK |
490 | u64 *spte; |
491 | ||
4a4c9924 AL |
492 | gfn = unalias_gfn(kvm, gfn); |
493 | rmapp = gfn_to_rmap(kvm, gfn); | |
374cbac0 | 494 | |
98348e95 IE |
495 | spte = rmap_next(kvm, rmapp, NULL); |
496 | while (spte) { | |
374cbac0 | 497 | BUG_ON(!spte); |
374cbac0 | 498 | BUG_ON(!(*spte & PT_PRESENT_MASK)); |
374cbac0 | 499 | rmap_printk("rmap_write_protect: spte %p %llx\n", spte, *spte); |
9647c14c IE |
500 | if (is_writeble_pte(*spte)) |
501 | set_shadow_pte(spte, *spte & ~PT_WRITABLE_MASK); | |
4a4c9924 | 502 | kvm_flush_remote_tlbs(kvm); |
9647c14c | 503 | spte = rmap_next(kvm, rmapp, spte); |
374cbac0 AK |
504 | } |
505 | } | |
506 | ||
d6c69ee9 | 507 | #ifdef MMU_DEBUG |
47ad8e68 | 508 | static int is_empty_shadow_page(u64 *spt) |
6aa8b732 | 509 | { |
139bdb2d AK |
510 | u64 *pos; |
511 | u64 *end; | |
512 | ||
47ad8e68 | 513 | for (pos = spt, end = pos + PAGE_SIZE / sizeof(u64); pos != end; pos++) |
c7addb90 | 514 | if ((*pos & ~PT_SHADOW_IO_MARK) != shadow_trap_nonpresent_pte) { |
139bdb2d AK |
515 | printk(KERN_ERR "%s: %p %llx\n", __FUNCTION__, |
516 | pos, *pos); | |
6aa8b732 | 517 | return 0; |
139bdb2d | 518 | } |
6aa8b732 AK |
519 | return 1; |
520 | } | |
d6c69ee9 | 521 | #endif |
6aa8b732 | 522 | |
90cb0529 | 523 | static void kvm_mmu_free_page(struct kvm *kvm, |
4b02d6da | 524 | struct kvm_mmu_page *page_head) |
260746c0 | 525 | { |
47ad8e68 | 526 | ASSERT(is_empty_shadow_page(page_head->spt)); |
d3d25b04 | 527 | list_del(&page_head->link); |
c1158e63 | 528 | __free_page(virt_to_page(page_head->spt)); |
290fc38d | 529 | __free_page(virt_to_page(page_head->gfns)); |
90cb0529 AK |
530 | kfree(page_head); |
531 | ++kvm->n_free_mmu_pages; | |
260746c0 AK |
532 | } |
533 | ||
cea0f0e7 AK |
534 | static unsigned kvm_page_table_hashfn(gfn_t gfn) |
535 | { | |
536 | return gfn; | |
537 | } | |
538 | ||
25c0de2c AK |
539 | static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu, |
540 | u64 *parent_pte) | |
6aa8b732 AK |
541 | { |
542 | struct kvm_mmu_page *page; | |
543 | ||
d3d25b04 | 544 | if (!vcpu->kvm->n_free_mmu_pages) |
25c0de2c | 545 | return NULL; |
6aa8b732 | 546 | |
d3d25b04 AK |
547 | page = mmu_memory_cache_alloc(&vcpu->mmu_page_header_cache, |
548 | sizeof *page); | |
549 | page->spt = mmu_memory_cache_alloc(&vcpu->mmu_page_cache, PAGE_SIZE); | |
290fc38d | 550 | page->gfns = mmu_memory_cache_alloc(&vcpu->mmu_page_cache, PAGE_SIZE); |
d3d25b04 AK |
551 | set_page_private(virt_to_page(page->spt), (unsigned long)page); |
552 | list_add(&page->link, &vcpu->kvm->active_mmu_pages); | |
47ad8e68 | 553 | ASSERT(is_empty_shadow_page(page->spt)); |
6aa8b732 | 554 | page->slot_bitmap = 0; |
cea0f0e7 | 555 | page->multimapped = 0; |
6aa8b732 | 556 | page->parent_pte = parent_pte; |
ebeace86 | 557 | --vcpu->kvm->n_free_mmu_pages; |
25c0de2c | 558 | return page; |
6aa8b732 AK |
559 | } |
560 | ||
714b93da AK |
561 | static void mmu_page_add_parent_pte(struct kvm_vcpu *vcpu, |
562 | struct kvm_mmu_page *page, u64 *parent_pte) | |
cea0f0e7 AK |
563 | { |
564 | struct kvm_pte_chain *pte_chain; | |
565 | struct hlist_node *node; | |
566 | int i; | |
567 | ||
568 | if (!parent_pte) | |
569 | return; | |
570 | if (!page->multimapped) { | |
571 | u64 *old = page->parent_pte; | |
572 | ||
573 | if (!old) { | |
574 | page->parent_pte = parent_pte; | |
575 | return; | |
576 | } | |
577 | page->multimapped = 1; | |
714b93da | 578 | pte_chain = mmu_alloc_pte_chain(vcpu); |
cea0f0e7 AK |
579 | INIT_HLIST_HEAD(&page->parent_ptes); |
580 | hlist_add_head(&pte_chain->link, &page->parent_ptes); | |
581 | pte_chain->parent_ptes[0] = old; | |
582 | } | |
583 | hlist_for_each_entry(pte_chain, node, &page->parent_ptes, link) { | |
584 | if (pte_chain->parent_ptes[NR_PTE_CHAIN_ENTRIES-1]) | |
585 | continue; | |
586 | for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) | |
587 | if (!pte_chain->parent_ptes[i]) { | |
588 | pte_chain->parent_ptes[i] = parent_pte; | |
589 | return; | |
590 | } | |
591 | } | |
714b93da | 592 | pte_chain = mmu_alloc_pte_chain(vcpu); |
cea0f0e7 AK |
593 | BUG_ON(!pte_chain); |
594 | hlist_add_head(&pte_chain->link, &page->parent_ptes); | |
595 | pte_chain->parent_ptes[0] = parent_pte; | |
596 | } | |
597 | ||
90cb0529 | 598 | static void mmu_page_remove_parent_pte(struct kvm_mmu_page *page, |
cea0f0e7 AK |
599 | u64 *parent_pte) |
600 | { | |
601 | struct kvm_pte_chain *pte_chain; | |
602 | struct hlist_node *node; | |
603 | int i; | |
604 | ||
605 | if (!page->multimapped) { | |
606 | BUG_ON(page->parent_pte != parent_pte); | |
607 | page->parent_pte = NULL; | |
608 | return; | |
609 | } | |
610 | hlist_for_each_entry(pte_chain, node, &page->parent_ptes, link) | |
611 | for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) { | |
612 | if (!pte_chain->parent_ptes[i]) | |
613 | break; | |
614 | if (pte_chain->parent_ptes[i] != parent_pte) | |
615 | continue; | |
697fe2e2 AK |
616 | while (i + 1 < NR_PTE_CHAIN_ENTRIES |
617 | && pte_chain->parent_ptes[i + 1]) { | |
cea0f0e7 AK |
618 | pte_chain->parent_ptes[i] |
619 | = pte_chain->parent_ptes[i + 1]; | |
620 | ++i; | |
621 | } | |
622 | pte_chain->parent_ptes[i] = NULL; | |
697fe2e2 AK |
623 | if (i == 0) { |
624 | hlist_del(&pte_chain->link); | |
90cb0529 | 625 | mmu_free_pte_chain(pte_chain); |
697fe2e2 AK |
626 | if (hlist_empty(&page->parent_ptes)) { |
627 | page->multimapped = 0; | |
628 | page->parent_pte = NULL; | |
629 | } | |
630 | } | |
cea0f0e7 AK |
631 | return; |
632 | } | |
633 | BUG(); | |
634 | } | |
635 | ||
f67a46f4 | 636 | static struct kvm_mmu_page *kvm_mmu_lookup_page(struct kvm *kvm, |
cea0f0e7 AK |
637 | gfn_t gfn) |
638 | { | |
639 | unsigned index; | |
640 | struct hlist_head *bucket; | |
641 | struct kvm_mmu_page *page; | |
642 | struct hlist_node *node; | |
643 | ||
644 | pgprintk("%s: looking for gfn %lx\n", __FUNCTION__, gfn); | |
645 | index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES; | |
f67a46f4 | 646 | bucket = &kvm->mmu_page_hash[index]; |
cea0f0e7 AK |
647 | hlist_for_each_entry(page, node, bucket, hash_link) |
648 | if (page->gfn == gfn && !page->role.metaphysical) { | |
649 | pgprintk("%s: found role %x\n", | |
650 | __FUNCTION__, page->role.word); | |
651 | return page; | |
652 | } | |
653 | return NULL; | |
654 | } | |
655 | ||
656 | static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu, | |
657 | gfn_t gfn, | |
658 | gva_t gaddr, | |
659 | unsigned level, | |
660 | int metaphysical, | |
d28c6cfb | 661 | unsigned hugepage_access, |
cea0f0e7 AK |
662 | u64 *parent_pte) |
663 | { | |
664 | union kvm_mmu_page_role role; | |
665 | unsigned index; | |
666 | unsigned quadrant; | |
667 | struct hlist_head *bucket; | |
668 | struct kvm_mmu_page *page; | |
669 | struct hlist_node *node; | |
670 | ||
671 | role.word = 0; | |
672 | role.glevels = vcpu->mmu.root_level; | |
673 | role.level = level; | |
674 | role.metaphysical = metaphysical; | |
d28c6cfb | 675 | role.hugepage_access = hugepage_access; |
cea0f0e7 AK |
676 | if (vcpu->mmu.root_level <= PT32_ROOT_LEVEL) { |
677 | quadrant = gaddr >> (PAGE_SHIFT + (PT64_PT_BITS * level)); | |
678 | quadrant &= (1 << ((PT32_PT_BITS - PT64_PT_BITS) * level)) - 1; | |
679 | role.quadrant = quadrant; | |
680 | } | |
681 | pgprintk("%s: looking gfn %lx role %x\n", __FUNCTION__, | |
682 | gfn, role.word); | |
683 | index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES; | |
684 | bucket = &vcpu->kvm->mmu_page_hash[index]; | |
685 | hlist_for_each_entry(page, node, bucket, hash_link) | |
686 | if (page->gfn == gfn && page->role.word == role.word) { | |
714b93da | 687 | mmu_page_add_parent_pte(vcpu, page, parent_pte); |
cea0f0e7 AK |
688 | pgprintk("%s: found\n", __FUNCTION__); |
689 | return page; | |
690 | } | |
691 | page = kvm_mmu_alloc_page(vcpu, parent_pte); | |
692 | if (!page) | |
693 | return page; | |
694 | pgprintk("%s: adding gfn %lx role %x\n", __FUNCTION__, gfn, role.word); | |
695 | page->gfn = gfn; | |
696 | page->role = role; | |
697 | hlist_add_head(&page->hash_link, bucket); | |
c7addb90 | 698 | vcpu->mmu.prefetch_page(vcpu, page); |
374cbac0 | 699 | if (!metaphysical) |
4a4c9924 | 700 | rmap_write_protect(vcpu->kvm, gfn); |
cea0f0e7 AK |
701 | return page; |
702 | } | |
703 | ||
90cb0529 | 704 | static void kvm_mmu_page_unlink_children(struct kvm *kvm, |
a436036b AK |
705 | struct kvm_mmu_page *page) |
706 | { | |
697fe2e2 AK |
707 | unsigned i; |
708 | u64 *pt; | |
709 | u64 ent; | |
710 | ||
47ad8e68 | 711 | pt = page->spt; |
697fe2e2 AK |
712 | |
713 | if (page->role.level == PT_PAGE_TABLE_LEVEL) { | |
714 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) { | |
c7addb90 | 715 | if (is_shadow_present_pte(pt[i])) |
290fc38d | 716 | rmap_remove(kvm, &pt[i]); |
c7addb90 | 717 | pt[i] = shadow_trap_nonpresent_pte; |
697fe2e2 | 718 | } |
90cb0529 | 719 | kvm_flush_remote_tlbs(kvm); |
697fe2e2 AK |
720 | return; |
721 | } | |
722 | ||
723 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) { | |
724 | ent = pt[i]; | |
725 | ||
c7addb90 AK |
726 | pt[i] = shadow_trap_nonpresent_pte; |
727 | if (!is_shadow_present_pte(ent)) | |
697fe2e2 AK |
728 | continue; |
729 | ent &= PT64_BASE_ADDR_MASK; | |
90cb0529 | 730 | mmu_page_remove_parent_pte(page_header(ent), &pt[i]); |
697fe2e2 | 731 | } |
90cb0529 | 732 | kvm_flush_remote_tlbs(kvm); |
a436036b AK |
733 | } |
734 | ||
90cb0529 | 735 | static void kvm_mmu_put_page(struct kvm_mmu_page *page, |
cea0f0e7 AK |
736 | u64 *parent_pte) |
737 | { | |
90cb0529 | 738 | mmu_page_remove_parent_pte(page, parent_pte); |
a436036b AK |
739 | } |
740 | ||
12b7d28f AK |
741 | static void kvm_mmu_reset_last_pte_updated(struct kvm *kvm) |
742 | { | |
743 | int i; | |
744 | ||
745 | for (i = 0; i < KVM_MAX_VCPUS; ++i) | |
746 | if (kvm->vcpus[i]) | |
747 | kvm->vcpus[i]->last_pte_updated = NULL; | |
748 | } | |
749 | ||
90cb0529 | 750 | static void kvm_mmu_zap_page(struct kvm *kvm, |
a436036b AK |
751 | struct kvm_mmu_page *page) |
752 | { | |
753 | u64 *parent_pte; | |
754 | ||
755 | while (page->multimapped || page->parent_pte) { | |
756 | if (!page->multimapped) | |
757 | parent_pte = page->parent_pte; | |
758 | else { | |
759 | struct kvm_pte_chain *chain; | |
760 | ||
761 | chain = container_of(page->parent_ptes.first, | |
762 | struct kvm_pte_chain, link); | |
763 | parent_pte = chain->parent_ptes[0]; | |
764 | } | |
697fe2e2 | 765 | BUG_ON(!parent_pte); |
90cb0529 | 766 | kvm_mmu_put_page(page, parent_pte); |
c7addb90 | 767 | set_shadow_pte(parent_pte, shadow_trap_nonpresent_pte); |
a436036b | 768 | } |
90cb0529 | 769 | kvm_mmu_page_unlink_children(kvm, page); |
3bb65a22 AK |
770 | if (!page->root_count) { |
771 | hlist_del(&page->hash_link); | |
90cb0529 | 772 | kvm_mmu_free_page(kvm, page); |
36868f7b | 773 | } else |
90cb0529 | 774 | list_move(&page->link, &kvm->active_mmu_pages); |
12b7d28f | 775 | kvm_mmu_reset_last_pte_updated(kvm); |
a436036b AK |
776 | } |
777 | ||
82ce2c96 IE |
778 | /* |
779 | * Changing the number of mmu pages allocated to the vm | |
780 | * Note: if kvm_nr_mmu_pages is too small, you will get dead lock | |
781 | */ | |
782 | void kvm_mmu_change_mmu_pages(struct kvm *kvm, unsigned int kvm_nr_mmu_pages) | |
783 | { | |
784 | /* | |
785 | * If we set the number of mmu pages to be smaller be than the | |
786 | * number of actived pages , we must to free some mmu pages before we | |
787 | * change the value | |
788 | */ | |
789 | ||
790 | if ((kvm->n_alloc_mmu_pages - kvm->n_free_mmu_pages) > | |
791 | kvm_nr_mmu_pages) { | |
792 | int n_used_mmu_pages = kvm->n_alloc_mmu_pages | |
793 | - kvm->n_free_mmu_pages; | |
794 | ||
795 | while (n_used_mmu_pages > kvm_nr_mmu_pages) { | |
796 | struct kvm_mmu_page *page; | |
797 | ||
798 | page = container_of(kvm->active_mmu_pages.prev, | |
799 | struct kvm_mmu_page, link); | |
800 | kvm_mmu_zap_page(kvm, page); | |
801 | n_used_mmu_pages--; | |
802 | } | |
803 | kvm->n_free_mmu_pages = 0; | |
804 | } | |
805 | else | |
806 | kvm->n_free_mmu_pages += kvm_nr_mmu_pages | |
807 | - kvm->n_alloc_mmu_pages; | |
808 | ||
809 | kvm->n_alloc_mmu_pages = kvm_nr_mmu_pages; | |
810 | } | |
811 | ||
f67a46f4 | 812 | static int kvm_mmu_unprotect_page(struct kvm *kvm, gfn_t gfn) |
a436036b AK |
813 | { |
814 | unsigned index; | |
815 | struct hlist_head *bucket; | |
816 | struct kvm_mmu_page *page; | |
817 | struct hlist_node *node, *n; | |
818 | int r; | |
819 | ||
820 | pgprintk("%s: looking for gfn %lx\n", __FUNCTION__, gfn); | |
821 | r = 0; | |
822 | index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES; | |
f67a46f4 | 823 | bucket = &kvm->mmu_page_hash[index]; |
a436036b AK |
824 | hlist_for_each_entry_safe(page, node, n, bucket, hash_link) |
825 | if (page->gfn == gfn && !page->role.metaphysical) { | |
697fe2e2 AK |
826 | pgprintk("%s: gfn %lx role %x\n", __FUNCTION__, gfn, |
827 | page->role.word); | |
f67a46f4 | 828 | kvm_mmu_zap_page(kvm, page); |
a436036b AK |
829 | r = 1; |
830 | } | |
831 | return r; | |
cea0f0e7 AK |
832 | } |
833 | ||
f67a46f4 | 834 | static void mmu_unshadow(struct kvm *kvm, gfn_t gfn) |
97a0a01e AK |
835 | { |
836 | struct kvm_mmu_page *page; | |
837 | ||
f67a46f4 | 838 | while ((page = kvm_mmu_lookup_page(kvm, gfn)) != NULL) { |
97a0a01e AK |
839 | pgprintk("%s: zap %lx %x\n", |
840 | __FUNCTION__, gfn, page->role.word); | |
f67a46f4 | 841 | kvm_mmu_zap_page(kvm, page); |
97a0a01e AK |
842 | } |
843 | } | |
844 | ||
6aa8b732 AK |
845 | static void page_header_update_slot(struct kvm *kvm, void *pte, gpa_t gpa) |
846 | { | |
847 | int slot = memslot_id(kvm, gfn_to_memslot(kvm, gpa >> PAGE_SHIFT)); | |
848 | struct kvm_mmu_page *page_head = page_header(__pa(pte)); | |
849 | ||
850 | __set_bit(slot, &page_head->slot_bitmap); | |
851 | } | |
852 | ||
4a4c9924 | 853 | hpa_t safe_gpa_to_hpa(struct kvm *kvm, gpa_t gpa) |
6aa8b732 | 854 | { |
4a4c9924 | 855 | hpa_t hpa = gpa_to_hpa(kvm, gpa); |
6aa8b732 AK |
856 | |
857 | return is_error_hpa(hpa) ? bad_page_address | (gpa & ~PAGE_MASK): hpa; | |
858 | } | |
859 | ||
4a4c9924 | 860 | hpa_t gpa_to_hpa(struct kvm *kvm, gpa_t gpa) |
6aa8b732 | 861 | { |
6aa8b732 AK |
862 | struct page *page; |
863 | ||
864 | ASSERT((gpa & HPA_ERR_MASK) == 0); | |
4a4c9924 | 865 | page = gfn_to_page(kvm, gpa >> PAGE_SHIFT); |
954bbbc2 | 866 | if (!page) |
6aa8b732 | 867 | return gpa | HPA_ERR_MASK; |
6aa8b732 AK |
868 | return ((hpa_t)page_to_pfn(page) << PAGE_SHIFT) |
869 | | (gpa & (PAGE_SIZE-1)); | |
870 | } | |
871 | ||
872 | hpa_t gva_to_hpa(struct kvm_vcpu *vcpu, gva_t gva) | |
873 | { | |
874 | gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, gva); | |
875 | ||
876 | if (gpa == UNMAPPED_GVA) | |
877 | return UNMAPPED_GVA; | |
4a4c9924 | 878 | return gpa_to_hpa(vcpu->kvm, gpa); |
6aa8b732 AK |
879 | } |
880 | ||
039576c0 AK |
881 | struct page *gva_to_page(struct kvm_vcpu *vcpu, gva_t gva) |
882 | { | |
883 | gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, gva); | |
884 | ||
885 | if (gpa == UNMAPPED_GVA) | |
886 | return NULL; | |
4a4c9924 | 887 | return pfn_to_page(gpa_to_hpa(vcpu->kvm, gpa) >> PAGE_SHIFT); |
039576c0 AK |
888 | } |
889 | ||
6aa8b732 AK |
890 | static void nonpaging_new_cr3(struct kvm_vcpu *vcpu) |
891 | { | |
892 | } | |
893 | ||
894 | static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, hpa_t p) | |
895 | { | |
896 | int level = PT32E_ROOT_LEVEL; | |
897 | hpa_t table_addr = vcpu->mmu.root_hpa; | |
898 | ||
899 | for (; ; level--) { | |
900 | u32 index = PT64_INDEX(v, level); | |
901 | u64 *table; | |
cea0f0e7 | 902 | u64 pte; |
6aa8b732 AK |
903 | |
904 | ASSERT(VALID_PAGE(table_addr)); | |
905 | table = __va(table_addr); | |
906 | ||
907 | if (level == 1) { | |
9647c14c IE |
908 | int was_rmapped; |
909 | ||
cea0f0e7 | 910 | pte = table[index]; |
9647c14c | 911 | was_rmapped = is_rmap_pte(pte); |
c7addb90 | 912 | if (is_shadow_present_pte(pte) && is_writeble_pte(pte)) |
cea0f0e7 | 913 | return 0; |
6aa8b732 AK |
914 | mark_page_dirty(vcpu->kvm, v >> PAGE_SHIFT); |
915 | page_header_update_slot(vcpu->kvm, table, v); | |
916 | table[index] = p | PT_PRESENT_MASK | PT_WRITABLE_MASK | | |
917 | PT_USER_MASK; | |
9647c14c IE |
918 | if (!was_rmapped) |
919 | rmap_add(vcpu, &table[index], v >> PAGE_SHIFT); | |
6aa8b732 AK |
920 | return 0; |
921 | } | |
922 | ||
c7addb90 | 923 | if (table[index] == shadow_trap_nonpresent_pte) { |
25c0de2c | 924 | struct kvm_mmu_page *new_table; |
cea0f0e7 | 925 | gfn_t pseudo_gfn; |
6aa8b732 | 926 | |
cea0f0e7 AK |
927 | pseudo_gfn = (v & PT64_DIR_BASE_ADDR_MASK) |
928 | >> PAGE_SHIFT; | |
929 | new_table = kvm_mmu_get_page(vcpu, pseudo_gfn, | |
930 | v, level - 1, | |
6bfccdc9 | 931 | 1, 3, &table[index]); |
25c0de2c | 932 | if (!new_table) { |
6aa8b732 AK |
933 | pgprintk("nonpaging_map: ENOMEM\n"); |
934 | return -ENOMEM; | |
935 | } | |
936 | ||
47ad8e68 | 937 | table[index] = __pa(new_table->spt) | PT_PRESENT_MASK |
25c0de2c | 938 | | PT_WRITABLE_MASK | PT_USER_MASK; |
6aa8b732 AK |
939 | } |
940 | table_addr = table[index] & PT64_BASE_ADDR_MASK; | |
941 | } | |
942 | } | |
943 | ||
c7addb90 AK |
944 | static void nonpaging_prefetch_page(struct kvm_vcpu *vcpu, |
945 | struct kvm_mmu_page *sp) | |
946 | { | |
947 | int i; | |
948 | ||
949 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) | |
950 | sp->spt[i] = shadow_trap_nonpresent_pte; | |
951 | } | |
952 | ||
17ac10ad AK |
953 | static void mmu_free_roots(struct kvm_vcpu *vcpu) |
954 | { | |
955 | int i; | |
3bb65a22 | 956 | struct kvm_mmu_page *page; |
17ac10ad | 957 | |
7b53aa56 AK |
958 | if (!VALID_PAGE(vcpu->mmu.root_hpa)) |
959 | return; | |
17ac10ad AK |
960 | #ifdef CONFIG_X86_64 |
961 | if (vcpu->mmu.shadow_root_level == PT64_ROOT_LEVEL) { | |
962 | hpa_t root = vcpu->mmu.root_hpa; | |
963 | ||
3bb65a22 AK |
964 | page = page_header(root); |
965 | --page->root_count; | |
17ac10ad AK |
966 | vcpu->mmu.root_hpa = INVALID_PAGE; |
967 | return; | |
968 | } | |
969 | #endif | |
970 | for (i = 0; i < 4; ++i) { | |
971 | hpa_t root = vcpu->mmu.pae_root[i]; | |
972 | ||
417726a3 | 973 | if (root) { |
417726a3 AK |
974 | root &= PT64_BASE_ADDR_MASK; |
975 | page = page_header(root); | |
976 | --page->root_count; | |
977 | } | |
17ac10ad AK |
978 | vcpu->mmu.pae_root[i] = INVALID_PAGE; |
979 | } | |
980 | vcpu->mmu.root_hpa = INVALID_PAGE; | |
981 | } | |
982 | ||
983 | static void mmu_alloc_roots(struct kvm_vcpu *vcpu) | |
984 | { | |
985 | int i; | |
cea0f0e7 | 986 | gfn_t root_gfn; |
3bb65a22 AK |
987 | struct kvm_mmu_page *page; |
988 | ||
cea0f0e7 | 989 | root_gfn = vcpu->cr3 >> PAGE_SHIFT; |
17ac10ad AK |
990 | |
991 | #ifdef CONFIG_X86_64 | |
992 | if (vcpu->mmu.shadow_root_level == PT64_ROOT_LEVEL) { | |
993 | hpa_t root = vcpu->mmu.root_hpa; | |
994 | ||
995 | ASSERT(!VALID_PAGE(root)); | |
68a99f6d | 996 | page = kvm_mmu_get_page(vcpu, root_gfn, 0, |
d28c6cfb | 997 | PT64_ROOT_LEVEL, 0, 0, NULL); |
47ad8e68 | 998 | root = __pa(page->spt); |
3bb65a22 | 999 | ++page->root_count; |
17ac10ad AK |
1000 | vcpu->mmu.root_hpa = root; |
1001 | return; | |
1002 | } | |
1003 | #endif | |
1004 | for (i = 0; i < 4; ++i) { | |
1005 | hpa_t root = vcpu->mmu.pae_root[i]; | |
1006 | ||
1007 | ASSERT(!VALID_PAGE(root)); | |
417726a3 AK |
1008 | if (vcpu->mmu.root_level == PT32E_ROOT_LEVEL) { |
1009 | if (!is_present_pte(vcpu->pdptrs[i])) { | |
1010 | vcpu->mmu.pae_root[i] = 0; | |
1011 | continue; | |
1012 | } | |
cea0f0e7 | 1013 | root_gfn = vcpu->pdptrs[i] >> PAGE_SHIFT; |
417726a3 | 1014 | } else if (vcpu->mmu.root_level == 0) |
cea0f0e7 | 1015 | root_gfn = 0; |
68a99f6d | 1016 | page = kvm_mmu_get_page(vcpu, root_gfn, i << 30, |
cea0f0e7 | 1017 | PT32_ROOT_LEVEL, !is_paging(vcpu), |
d28c6cfb | 1018 | 0, NULL); |
47ad8e68 | 1019 | root = __pa(page->spt); |
3bb65a22 | 1020 | ++page->root_count; |
17ac10ad AK |
1021 | vcpu->mmu.pae_root[i] = root | PT_PRESENT_MASK; |
1022 | } | |
1023 | vcpu->mmu.root_hpa = __pa(vcpu->mmu.pae_root); | |
1024 | } | |
1025 | ||
6aa8b732 AK |
1026 | static gpa_t nonpaging_gva_to_gpa(struct kvm_vcpu *vcpu, gva_t vaddr) |
1027 | { | |
1028 | return vaddr; | |
1029 | } | |
1030 | ||
1031 | static int nonpaging_page_fault(struct kvm_vcpu *vcpu, gva_t gva, | |
1032 | u32 error_code) | |
1033 | { | |
6aa8b732 | 1034 | gpa_t addr = gva; |
ebeace86 | 1035 | hpa_t paddr; |
e2dec939 | 1036 | int r; |
6aa8b732 | 1037 | |
e2dec939 AK |
1038 | r = mmu_topup_memory_caches(vcpu); |
1039 | if (r) | |
1040 | return r; | |
714b93da | 1041 | |
6aa8b732 AK |
1042 | ASSERT(vcpu); |
1043 | ASSERT(VALID_PAGE(vcpu->mmu.root_hpa)); | |
1044 | ||
6aa8b732 | 1045 | |
4a4c9924 | 1046 | paddr = gpa_to_hpa(vcpu->kvm, addr & PT64_BASE_ADDR_MASK); |
6aa8b732 | 1047 | |
ebeace86 AK |
1048 | if (is_error_hpa(paddr)) |
1049 | return 1; | |
6aa8b732 | 1050 | |
ebeace86 | 1051 | return nonpaging_map(vcpu, addr & PAGE_MASK, paddr); |
6aa8b732 AK |
1052 | } |
1053 | ||
6aa8b732 AK |
1054 | static void nonpaging_free(struct kvm_vcpu *vcpu) |
1055 | { | |
17ac10ad | 1056 | mmu_free_roots(vcpu); |
6aa8b732 AK |
1057 | } |
1058 | ||
1059 | static int nonpaging_init_context(struct kvm_vcpu *vcpu) | |
1060 | { | |
1061 | struct kvm_mmu *context = &vcpu->mmu; | |
1062 | ||
1063 | context->new_cr3 = nonpaging_new_cr3; | |
1064 | context->page_fault = nonpaging_page_fault; | |
6aa8b732 AK |
1065 | context->gva_to_gpa = nonpaging_gva_to_gpa; |
1066 | context->free = nonpaging_free; | |
c7addb90 | 1067 | context->prefetch_page = nonpaging_prefetch_page; |
cea0f0e7 | 1068 | context->root_level = 0; |
6aa8b732 | 1069 | context->shadow_root_level = PT32E_ROOT_LEVEL; |
17c3ba9d | 1070 | context->root_hpa = INVALID_PAGE; |
6aa8b732 AK |
1071 | return 0; |
1072 | } | |
1073 | ||
6aa8b732 AK |
1074 | static void kvm_mmu_flush_tlb(struct kvm_vcpu *vcpu) |
1075 | { | |
1165f5fe | 1076 | ++vcpu->stat.tlb_flush; |
cbdd1bea | 1077 | kvm_x86_ops->tlb_flush(vcpu); |
6aa8b732 AK |
1078 | } |
1079 | ||
1080 | static void paging_new_cr3(struct kvm_vcpu *vcpu) | |
1081 | { | |
374cbac0 | 1082 | pgprintk("%s: cr3 %lx\n", __FUNCTION__, vcpu->cr3); |
cea0f0e7 | 1083 | mmu_free_roots(vcpu); |
6aa8b732 AK |
1084 | } |
1085 | ||
6aa8b732 AK |
1086 | static void inject_page_fault(struct kvm_vcpu *vcpu, |
1087 | u64 addr, | |
1088 | u32 err_code) | |
1089 | { | |
cbdd1bea | 1090 | kvm_x86_ops->inject_page_fault(vcpu, addr, err_code); |
6aa8b732 AK |
1091 | } |
1092 | ||
6aa8b732 AK |
1093 | static void paging_free(struct kvm_vcpu *vcpu) |
1094 | { | |
1095 | nonpaging_free(vcpu); | |
1096 | } | |
1097 | ||
1098 | #define PTTYPE 64 | |
1099 | #include "paging_tmpl.h" | |
1100 | #undef PTTYPE | |
1101 | ||
1102 | #define PTTYPE 32 | |
1103 | #include "paging_tmpl.h" | |
1104 | #undef PTTYPE | |
1105 | ||
17ac10ad | 1106 | static int paging64_init_context_common(struct kvm_vcpu *vcpu, int level) |
6aa8b732 AK |
1107 | { |
1108 | struct kvm_mmu *context = &vcpu->mmu; | |
1109 | ||
1110 | ASSERT(is_pae(vcpu)); | |
1111 | context->new_cr3 = paging_new_cr3; | |
1112 | context->page_fault = paging64_page_fault; | |
6aa8b732 | 1113 | context->gva_to_gpa = paging64_gva_to_gpa; |
c7addb90 | 1114 | context->prefetch_page = paging64_prefetch_page; |
6aa8b732 | 1115 | context->free = paging_free; |
17ac10ad AK |
1116 | context->root_level = level; |
1117 | context->shadow_root_level = level; | |
17c3ba9d | 1118 | context->root_hpa = INVALID_PAGE; |
6aa8b732 AK |
1119 | return 0; |
1120 | } | |
1121 | ||
17ac10ad AK |
1122 | static int paging64_init_context(struct kvm_vcpu *vcpu) |
1123 | { | |
1124 | return paging64_init_context_common(vcpu, PT64_ROOT_LEVEL); | |
1125 | } | |
1126 | ||
6aa8b732 AK |
1127 | static int paging32_init_context(struct kvm_vcpu *vcpu) |
1128 | { | |
1129 | struct kvm_mmu *context = &vcpu->mmu; | |
1130 | ||
1131 | context->new_cr3 = paging_new_cr3; | |
1132 | context->page_fault = paging32_page_fault; | |
6aa8b732 AK |
1133 | context->gva_to_gpa = paging32_gva_to_gpa; |
1134 | context->free = paging_free; | |
c7addb90 | 1135 | context->prefetch_page = paging32_prefetch_page; |
6aa8b732 AK |
1136 | context->root_level = PT32_ROOT_LEVEL; |
1137 | context->shadow_root_level = PT32E_ROOT_LEVEL; | |
17c3ba9d | 1138 | context->root_hpa = INVALID_PAGE; |
6aa8b732 AK |
1139 | return 0; |
1140 | } | |
1141 | ||
1142 | static int paging32E_init_context(struct kvm_vcpu *vcpu) | |
1143 | { | |
17ac10ad | 1144 | return paging64_init_context_common(vcpu, PT32E_ROOT_LEVEL); |
6aa8b732 AK |
1145 | } |
1146 | ||
1147 | static int init_kvm_mmu(struct kvm_vcpu *vcpu) | |
1148 | { | |
1149 | ASSERT(vcpu); | |
1150 | ASSERT(!VALID_PAGE(vcpu->mmu.root_hpa)); | |
1151 | ||
1152 | if (!is_paging(vcpu)) | |
1153 | return nonpaging_init_context(vcpu); | |
a9058ecd | 1154 | else if (is_long_mode(vcpu)) |
6aa8b732 AK |
1155 | return paging64_init_context(vcpu); |
1156 | else if (is_pae(vcpu)) | |
1157 | return paging32E_init_context(vcpu); | |
1158 | else | |
1159 | return paging32_init_context(vcpu); | |
1160 | } | |
1161 | ||
1162 | static void destroy_kvm_mmu(struct kvm_vcpu *vcpu) | |
1163 | { | |
1164 | ASSERT(vcpu); | |
1165 | if (VALID_PAGE(vcpu->mmu.root_hpa)) { | |
1166 | vcpu->mmu.free(vcpu); | |
1167 | vcpu->mmu.root_hpa = INVALID_PAGE; | |
1168 | } | |
1169 | } | |
1170 | ||
1171 | int kvm_mmu_reset_context(struct kvm_vcpu *vcpu) | |
17c3ba9d AK |
1172 | { |
1173 | destroy_kvm_mmu(vcpu); | |
1174 | return init_kvm_mmu(vcpu); | |
1175 | } | |
8668a3c4 | 1176 | EXPORT_SYMBOL_GPL(kvm_mmu_reset_context); |
17c3ba9d AK |
1177 | |
1178 | int kvm_mmu_load(struct kvm_vcpu *vcpu) | |
6aa8b732 | 1179 | { |
714b93da AK |
1180 | int r; |
1181 | ||
11ec2804 | 1182 | mutex_lock(&vcpu->kvm->lock); |
e2dec939 | 1183 | r = mmu_topup_memory_caches(vcpu); |
17c3ba9d AK |
1184 | if (r) |
1185 | goto out; | |
1186 | mmu_alloc_roots(vcpu); | |
cbdd1bea | 1187 | kvm_x86_ops->set_cr3(vcpu, vcpu->mmu.root_hpa); |
17c3ba9d | 1188 | kvm_mmu_flush_tlb(vcpu); |
714b93da | 1189 | out: |
11ec2804 | 1190 | mutex_unlock(&vcpu->kvm->lock); |
714b93da | 1191 | return r; |
6aa8b732 | 1192 | } |
17c3ba9d AK |
1193 | EXPORT_SYMBOL_GPL(kvm_mmu_load); |
1194 | ||
1195 | void kvm_mmu_unload(struct kvm_vcpu *vcpu) | |
1196 | { | |
1197 | mmu_free_roots(vcpu); | |
1198 | } | |
6aa8b732 | 1199 | |
09072daf | 1200 | static void mmu_pte_write_zap_pte(struct kvm_vcpu *vcpu, |
ac1b714e AK |
1201 | struct kvm_mmu_page *page, |
1202 | u64 *spte) | |
1203 | { | |
1204 | u64 pte; | |
1205 | struct kvm_mmu_page *child; | |
1206 | ||
1207 | pte = *spte; | |
c7addb90 | 1208 | if (is_shadow_present_pte(pte)) { |
ac1b714e | 1209 | if (page->role.level == PT_PAGE_TABLE_LEVEL) |
290fc38d | 1210 | rmap_remove(vcpu->kvm, spte); |
ac1b714e AK |
1211 | else { |
1212 | child = page_header(pte & PT64_BASE_ADDR_MASK); | |
90cb0529 | 1213 | mmu_page_remove_parent_pte(child, spte); |
ac1b714e AK |
1214 | } |
1215 | } | |
c7addb90 | 1216 | set_shadow_pte(spte, shadow_trap_nonpresent_pte); |
d9e368d6 | 1217 | kvm_flush_remote_tlbs(vcpu->kvm); |
ac1b714e AK |
1218 | } |
1219 | ||
0028425f AK |
1220 | static void mmu_pte_write_new_pte(struct kvm_vcpu *vcpu, |
1221 | struct kvm_mmu_page *page, | |
1222 | u64 *spte, | |
c7addb90 AK |
1223 | const void *new, int bytes, |
1224 | int offset_in_pte) | |
0028425f AK |
1225 | { |
1226 | if (page->role.level != PT_PAGE_TABLE_LEVEL) | |
1227 | return; | |
1228 | ||
1229 | if (page->role.glevels == PT32_ROOT_LEVEL) | |
c7addb90 AK |
1230 | paging32_update_pte(vcpu, page, spte, new, bytes, |
1231 | offset_in_pte); | |
0028425f | 1232 | else |
c7addb90 AK |
1233 | paging64_update_pte(vcpu, page, spte, new, bytes, |
1234 | offset_in_pte); | |
0028425f AK |
1235 | } |
1236 | ||
12b7d28f AK |
1237 | static bool last_updated_pte_accessed(struct kvm_vcpu *vcpu) |
1238 | { | |
1239 | u64 *spte = vcpu->last_pte_updated; | |
1240 | ||
1241 | return !!(spte && (*spte & PT_ACCESSED_MASK)); | |
1242 | } | |
1243 | ||
09072daf | 1244 | void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa, |
fe551881 | 1245 | const u8 *new, int bytes) |
da4a00f0 | 1246 | { |
9b7a0325 AK |
1247 | gfn_t gfn = gpa >> PAGE_SHIFT; |
1248 | struct kvm_mmu_page *page; | |
0e7bc4b9 | 1249 | struct hlist_node *node, *n; |
9b7a0325 AK |
1250 | struct hlist_head *bucket; |
1251 | unsigned index; | |
1252 | u64 *spte; | |
9b7a0325 | 1253 | unsigned offset = offset_in_page(gpa); |
0e7bc4b9 | 1254 | unsigned pte_size; |
9b7a0325 | 1255 | unsigned page_offset; |
0e7bc4b9 | 1256 | unsigned misaligned; |
fce0657f | 1257 | unsigned quadrant; |
9b7a0325 | 1258 | int level; |
86a5ba02 | 1259 | int flooded = 0; |
ac1b714e | 1260 | int npte; |
9b7a0325 | 1261 | |
da4a00f0 | 1262 | pgprintk("%s: gpa %llx bytes %d\n", __FUNCTION__, gpa, bytes); |
c7addb90 | 1263 | kvm_mmu_audit(vcpu, "pre pte write"); |
12b7d28f AK |
1264 | if (gfn == vcpu->last_pt_write_gfn |
1265 | && !last_updated_pte_accessed(vcpu)) { | |
86a5ba02 AK |
1266 | ++vcpu->last_pt_write_count; |
1267 | if (vcpu->last_pt_write_count >= 3) | |
1268 | flooded = 1; | |
1269 | } else { | |
1270 | vcpu->last_pt_write_gfn = gfn; | |
1271 | vcpu->last_pt_write_count = 1; | |
12b7d28f | 1272 | vcpu->last_pte_updated = NULL; |
86a5ba02 | 1273 | } |
9b7a0325 AK |
1274 | index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES; |
1275 | bucket = &vcpu->kvm->mmu_page_hash[index]; | |
0e7bc4b9 | 1276 | hlist_for_each_entry_safe(page, node, n, bucket, hash_link) { |
9b7a0325 AK |
1277 | if (page->gfn != gfn || page->role.metaphysical) |
1278 | continue; | |
0e7bc4b9 AK |
1279 | pte_size = page->role.glevels == PT32_ROOT_LEVEL ? 4 : 8; |
1280 | misaligned = (offset ^ (offset + bytes - 1)) & ~(pte_size - 1); | |
e925c5ba | 1281 | misaligned |= bytes < 4; |
86a5ba02 | 1282 | if (misaligned || flooded) { |
0e7bc4b9 AK |
1283 | /* |
1284 | * Misaligned accesses are too much trouble to fix | |
1285 | * up; also, they usually indicate a page is not used | |
1286 | * as a page table. | |
86a5ba02 AK |
1287 | * |
1288 | * If we're seeing too many writes to a page, | |
1289 | * it may no longer be a page table, or we may be | |
1290 | * forking, in which case it is better to unmap the | |
1291 | * page. | |
0e7bc4b9 AK |
1292 | */ |
1293 | pgprintk("misaligned: gpa %llx bytes %d role %x\n", | |
1294 | gpa, bytes, page->role.word); | |
90cb0529 | 1295 | kvm_mmu_zap_page(vcpu->kvm, page); |
0e7bc4b9 AK |
1296 | continue; |
1297 | } | |
9b7a0325 AK |
1298 | page_offset = offset; |
1299 | level = page->role.level; | |
ac1b714e | 1300 | npte = 1; |
9b7a0325 | 1301 | if (page->role.glevels == PT32_ROOT_LEVEL) { |
ac1b714e AK |
1302 | page_offset <<= 1; /* 32->64 */ |
1303 | /* | |
1304 | * A 32-bit pde maps 4MB while the shadow pdes map | |
1305 | * only 2MB. So we need to double the offset again | |
1306 | * and zap two pdes instead of one. | |
1307 | */ | |
1308 | if (level == PT32_ROOT_LEVEL) { | |
6b8d0f9b | 1309 | page_offset &= ~7; /* kill rounding error */ |
ac1b714e AK |
1310 | page_offset <<= 1; |
1311 | npte = 2; | |
1312 | } | |
fce0657f | 1313 | quadrant = page_offset >> PAGE_SHIFT; |
9b7a0325 | 1314 | page_offset &= ~PAGE_MASK; |
fce0657f AK |
1315 | if (quadrant != page->role.quadrant) |
1316 | continue; | |
9b7a0325 | 1317 | } |
47ad8e68 | 1318 | spte = &page->spt[page_offset / sizeof(*spte)]; |
ac1b714e | 1319 | while (npte--) { |
09072daf | 1320 | mmu_pte_write_zap_pte(vcpu, page, spte); |
c7addb90 AK |
1321 | mmu_pte_write_new_pte(vcpu, page, spte, new, bytes, |
1322 | page_offset & (pte_size - 1)); | |
ac1b714e | 1323 | ++spte; |
9b7a0325 | 1324 | } |
9b7a0325 | 1325 | } |
c7addb90 | 1326 | kvm_mmu_audit(vcpu, "post pte write"); |
da4a00f0 AK |
1327 | } |
1328 | ||
a436036b AK |
1329 | int kvm_mmu_unprotect_page_virt(struct kvm_vcpu *vcpu, gva_t gva) |
1330 | { | |
1331 | gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, gva); | |
1332 | ||
f67a46f4 | 1333 | return kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT); |
a436036b AK |
1334 | } |
1335 | ||
22d95b12 | 1336 | void __kvm_mmu_free_some_pages(struct kvm_vcpu *vcpu) |
ebeace86 AK |
1337 | { |
1338 | while (vcpu->kvm->n_free_mmu_pages < KVM_REFILL_PAGES) { | |
1339 | struct kvm_mmu_page *page; | |
1340 | ||
1341 | page = container_of(vcpu->kvm->active_mmu_pages.prev, | |
1342 | struct kvm_mmu_page, link); | |
90cb0529 | 1343 | kvm_mmu_zap_page(vcpu->kvm, page); |
ebeace86 AK |
1344 | } |
1345 | } | |
ebeace86 | 1346 | |
6aa8b732 AK |
1347 | static void free_mmu_pages(struct kvm_vcpu *vcpu) |
1348 | { | |
f51234c2 | 1349 | struct kvm_mmu_page *page; |
6aa8b732 | 1350 | |
f51234c2 AK |
1351 | while (!list_empty(&vcpu->kvm->active_mmu_pages)) { |
1352 | page = container_of(vcpu->kvm->active_mmu_pages.next, | |
1353 | struct kvm_mmu_page, link); | |
90cb0529 | 1354 | kvm_mmu_zap_page(vcpu->kvm, page); |
f51234c2 | 1355 | } |
17ac10ad | 1356 | free_page((unsigned long)vcpu->mmu.pae_root); |
6aa8b732 AK |
1357 | } |
1358 | ||
1359 | static int alloc_mmu_pages(struct kvm_vcpu *vcpu) | |
1360 | { | |
17ac10ad | 1361 | struct page *page; |
6aa8b732 AK |
1362 | int i; |
1363 | ||
1364 | ASSERT(vcpu); | |
1365 | ||
82ce2c96 IE |
1366 | if (vcpu->kvm->n_requested_mmu_pages) |
1367 | vcpu->kvm->n_free_mmu_pages = vcpu->kvm->n_requested_mmu_pages; | |
1368 | else | |
1369 | vcpu->kvm->n_free_mmu_pages = vcpu->kvm->n_alloc_mmu_pages; | |
17ac10ad AK |
1370 | /* |
1371 | * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64. | |
1372 | * Therefore we need to allocate shadow page tables in the first | |
1373 | * 4GB of memory, which happens to fit the DMA32 zone. | |
1374 | */ | |
1375 | page = alloc_page(GFP_KERNEL | __GFP_DMA32); | |
1376 | if (!page) | |
1377 | goto error_1; | |
1378 | vcpu->mmu.pae_root = page_address(page); | |
1379 | for (i = 0; i < 4; ++i) | |
1380 | vcpu->mmu.pae_root[i] = INVALID_PAGE; | |
1381 | ||
6aa8b732 AK |
1382 | return 0; |
1383 | ||
1384 | error_1: | |
1385 | free_mmu_pages(vcpu); | |
1386 | return -ENOMEM; | |
1387 | } | |
1388 | ||
8018c27b | 1389 | int kvm_mmu_create(struct kvm_vcpu *vcpu) |
6aa8b732 | 1390 | { |
6aa8b732 AK |
1391 | ASSERT(vcpu); |
1392 | ASSERT(!VALID_PAGE(vcpu->mmu.root_hpa)); | |
6aa8b732 | 1393 | |
8018c27b IM |
1394 | return alloc_mmu_pages(vcpu); |
1395 | } | |
6aa8b732 | 1396 | |
8018c27b IM |
1397 | int kvm_mmu_setup(struct kvm_vcpu *vcpu) |
1398 | { | |
1399 | ASSERT(vcpu); | |
1400 | ASSERT(!VALID_PAGE(vcpu->mmu.root_hpa)); | |
2c264957 | 1401 | |
8018c27b | 1402 | return init_kvm_mmu(vcpu); |
6aa8b732 AK |
1403 | } |
1404 | ||
1405 | void kvm_mmu_destroy(struct kvm_vcpu *vcpu) | |
1406 | { | |
1407 | ASSERT(vcpu); | |
1408 | ||
1409 | destroy_kvm_mmu(vcpu); | |
1410 | free_mmu_pages(vcpu); | |
714b93da | 1411 | mmu_free_memory_caches(vcpu); |
6aa8b732 AK |
1412 | } |
1413 | ||
90cb0529 | 1414 | void kvm_mmu_slot_remove_write_access(struct kvm *kvm, int slot) |
6aa8b732 AK |
1415 | { |
1416 | struct kvm_mmu_page *page; | |
1417 | ||
1418 | list_for_each_entry(page, &kvm->active_mmu_pages, link) { | |
1419 | int i; | |
1420 | u64 *pt; | |
1421 | ||
1422 | if (!test_bit(slot, &page->slot_bitmap)) | |
1423 | continue; | |
1424 | ||
47ad8e68 | 1425 | pt = page->spt; |
6aa8b732 AK |
1426 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) |
1427 | /* avoid RMW */ | |
9647c14c | 1428 | if (pt[i] & PT_WRITABLE_MASK) |
6aa8b732 | 1429 | pt[i] &= ~PT_WRITABLE_MASK; |
6aa8b732 AK |
1430 | } |
1431 | } | |
37a7d8b0 | 1432 | |
90cb0529 | 1433 | void kvm_mmu_zap_all(struct kvm *kvm) |
e0fa826f | 1434 | { |
90cb0529 | 1435 | struct kvm_mmu_page *page, *node; |
e0fa826f | 1436 | |
90cb0529 AK |
1437 | list_for_each_entry_safe(page, node, &kvm->active_mmu_pages, link) |
1438 | kvm_mmu_zap_page(kvm, page); | |
e0fa826f | 1439 | |
90cb0529 | 1440 | kvm_flush_remote_tlbs(kvm); |
e0fa826f DL |
1441 | } |
1442 | ||
b5a33a75 AK |
1443 | void kvm_mmu_module_exit(void) |
1444 | { | |
1445 | if (pte_chain_cache) | |
1446 | kmem_cache_destroy(pte_chain_cache); | |
1447 | if (rmap_desc_cache) | |
1448 | kmem_cache_destroy(rmap_desc_cache); | |
d3d25b04 AK |
1449 | if (mmu_page_header_cache) |
1450 | kmem_cache_destroy(mmu_page_header_cache); | |
b5a33a75 AK |
1451 | } |
1452 | ||
1453 | int kvm_mmu_module_init(void) | |
1454 | { | |
1455 | pte_chain_cache = kmem_cache_create("kvm_pte_chain", | |
1456 | sizeof(struct kvm_pte_chain), | |
20c2df83 | 1457 | 0, 0, NULL); |
b5a33a75 AK |
1458 | if (!pte_chain_cache) |
1459 | goto nomem; | |
1460 | rmap_desc_cache = kmem_cache_create("kvm_rmap_desc", | |
1461 | sizeof(struct kvm_rmap_desc), | |
20c2df83 | 1462 | 0, 0, NULL); |
b5a33a75 AK |
1463 | if (!rmap_desc_cache) |
1464 | goto nomem; | |
1465 | ||
d3d25b04 AK |
1466 | mmu_page_header_cache = kmem_cache_create("kvm_mmu_page_header", |
1467 | sizeof(struct kvm_mmu_page), | |
20c2df83 | 1468 | 0, 0, NULL); |
d3d25b04 AK |
1469 | if (!mmu_page_header_cache) |
1470 | goto nomem; | |
1471 | ||
b5a33a75 AK |
1472 | return 0; |
1473 | ||
1474 | nomem: | |
1475 | kvm_mmu_module_exit(); | |
1476 | return -ENOMEM; | |
1477 | } | |
1478 | ||
37a7d8b0 AK |
1479 | #ifdef AUDIT |
1480 | ||
1481 | static const char *audit_msg; | |
1482 | ||
1483 | static gva_t canonicalize(gva_t gva) | |
1484 | { | |
1485 | #ifdef CONFIG_X86_64 | |
1486 | gva = (long long)(gva << 16) >> 16; | |
1487 | #endif | |
1488 | return gva; | |
1489 | } | |
1490 | ||
1491 | static void audit_mappings_page(struct kvm_vcpu *vcpu, u64 page_pte, | |
1492 | gva_t va, int level) | |
1493 | { | |
1494 | u64 *pt = __va(page_pte & PT64_BASE_ADDR_MASK); | |
1495 | int i; | |
1496 | gva_t va_delta = 1ul << (PAGE_SHIFT + 9 * (level - 1)); | |
1497 | ||
1498 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i, va += va_delta) { | |
1499 | u64 ent = pt[i]; | |
1500 | ||
c7addb90 | 1501 | if (ent == shadow_trap_nonpresent_pte) |
37a7d8b0 AK |
1502 | continue; |
1503 | ||
1504 | va = canonicalize(va); | |
c7addb90 AK |
1505 | if (level > 1) { |
1506 | if (ent == shadow_notrap_nonpresent_pte) | |
1507 | printk(KERN_ERR "audit: (%s) nontrapping pte" | |
1508 | " in nonleaf level: levels %d gva %lx" | |
1509 | " level %d pte %llx\n", audit_msg, | |
1510 | vcpu->mmu.root_level, va, level, ent); | |
1511 | ||
37a7d8b0 | 1512 | audit_mappings_page(vcpu, ent, va, level - 1); |
c7addb90 | 1513 | } else { |
37a7d8b0 AK |
1514 | gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, va); |
1515 | hpa_t hpa = gpa_to_hpa(vcpu, gpa); | |
1516 | ||
c7addb90 | 1517 | if (is_shadow_present_pte(ent) |
37a7d8b0 | 1518 | && (ent & PT64_BASE_ADDR_MASK) != hpa) |
c7addb90 AK |
1519 | printk(KERN_ERR "xx audit error: (%s) levels %d" |
1520 | " gva %lx gpa %llx hpa %llx ent %llx %d\n", | |
37a7d8b0 | 1521 | audit_msg, vcpu->mmu.root_level, |
d77c26fc MD |
1522 | va, gpa, hpa, ent, |
1523 | is_shadow_present_pte(ent)); | |
c7addb90 AK |
1524 | else if (ent == shadow_notrap_nonpresent_pte |
1525 | && !is_error_hpa(hpa)) | |
1526 | printk(KERN_ERR "audit: (%s) notrap shadow," | |
1527 | " valid guest gva %lx\n", audit_msg, va); | |
1528 | ||
37a7d8b0 AK |
1529 | } |
1530 | } | |
1531 | } | |
1532 | ||
1533 | static void audit_mappings(struct kvm_vcpu *vcpu) | |
1534 | { | |
1ea252af | 1535 | unsigned i; |
37a7d8b0 AK |
1536 | |
1537 | if (vcpu->mmu.root_level == 4) | |
1538 | audit_mappings_page(vcpu, vcpu->mmu.root_hpa, 0, 4); | |
1539 | else | |
1540 | for (i = 0; i < 4; ++i) | |
1541 | if (vcpu->mmu.pae_root[i] & PT_PRESENT_MASK) | |
1542 | audit_mappings_page(vcpu, | |
1543 | vcpu->mmu.pae_root[i], | |
1544 | i << 30, | |
1545 | 2); | |
1546 | } | |
1547 | ||
1548 | static int count_rmaps(struct kvm_vcpu *vcpu) | |
1549 | { | |
1550 | int nmaps = 0; | |
1551 | int i, j, k; | |
1552 | ||
1553 | for (i = 0; i < KVM_MEMORY_SLOTS; ++i) { | |
1554 | struct kvm_memory_slot *m = &vcpu->kvm->memslots[i]; | |
1555 | struct kvm_rmap_desc *d; | |
1556 | ||
1557 | for (j = 0; j < m->npages; ++j) { | |
290fc38d | 1558 | unsigned long *rmapp = &m->rmap[j]; |
37a7d8b0 | 1559 | |
290fc38d | 1560 | if (!*rmapp) |
37a7d8b0 | 1561 | continue; |
290fc38d | 1562 | if (!(*rmapp & 1)) { |
37a7d8b0 AK |
1563 | ++nmaps; |
1564 | continue; | |
1565 | } | |
290fc38d | 1566 | d = (struct kvm_rmap_desc *)(*rmapp & ~1ul); |
37a7d8b0 AK |
1567 | while (d) { |
1568 | for (k = 0; k < RMAP_EXT; ++k) | |
1569 | if (d->shadow_ptes[k]) | |
1570 | ++nmaps; | |
1571 | else | |
1572 | break; | |
1573 | d = d->more; | |
1574 | } | |
1575 | } | |
1576 | } | |
1577 | return nmaps; | |
1578 | } | |
1579 | ||
1580 | static int count_writable_mappings(struct kvm_vcpu *vcpu) | |
1581 | { | |
1582 | int nmaps = 0; | |
1583 | struct kvm_mmu_page *page; | |
1584 | int i; | |
1585 | ||
1586 | list_for_each_entry(page, &vcpu->kvm->active_mmu_pages, link) { | |
47ad8e68 | 1587 | u64 *pt = page->spt; |
37a7d8b0 AK |
1588 | |
1589 | if (page->role.level != PT_PAGE_TABLE_LEVEL) | |
1590 | continue; | |
1591 | ||
1592 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) { | |
1593 | u64 ent = pt[i]; | |
1594 | ||
1595 | if (!(ent & PT_PRESENT_MASK)) | |
1596 | continue; | |
1597 | if (!(ent & PT_WRITABLE_MASK)) | |
1598 | continue; | |
1599 | ++nmaps; | |
1600 | } | |
1601 | } | |
1602 | return nmaps; | |
1603 | } | |
1604 | ||
1605 | static void audit_rmap(struct kvm_vcpu *vcpu) | |
1606 | { | |
1607 | int n_rmap = count_rmaps(vcpu); | |
1608 | int n_actual = count_writable_mappings(vcpu); | |
1609 | ||
1610 | if (n_rmap != n_actual) | |
1611 | printk(KERN_ERR "%s: (%s) rmap %d actual %d\n", | |
1612 | __FUNCTION__, audit_msg, n_rmap, n_actual); | |
1613 | } | |
1614 | ||
1615 | static void audit_write_protection(struct kvm_vcpu *vcpu) | |
1616 | { | |
1617 | struct kvm_mmu_page *page; | |
290fc38d IE |
1618 | struct kvm_memory_slot *slot; |
1619 | unsigned long *rmapp; | |
1620 | gfn_t gfn; | |
37a7d8b0 AK |
1621 | |
1622 | list_for_each_entry(page, &vcpu->kvm->active_mmu_pages, link) { | |
37a7d8b0 AK |
1623 | if (page->role.metaphysical) |
1624 | continue; | |
1625 | ||
290fc38d IE |
1626 | slot = gfn_to_memslot(vcpu->kvm, page->gfn); |
1627 | gfn = unalias_gfn(vcpu->kvm, page->gfn); | |
1628 | rmapp = &slot->rmap[gfn - slot->base_gfn]; | |
1629 | if (*rmapp) | |
37a7d8b0 AK |
1630 | printk(KERN_ERR "%s: (%s) shadow page has writable" |
1631 | " mappings: gfn %lx role %x\n", | |
1632 | __FUNCTION__, audit_msg, page->gfn, | |
1633 | page->role.word); | |
1634 | } | |
1635 | } | |
1636 | ||
1637 | static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg) | |
1638 | { | |
1639 | int olddbg = dbg; | |
1640 | ||
1641 | dbg = 0; | |
1642 | audit_msg = msg; | |
1643 | audit_rmap(vcpu); | |
1644 | audit_write_protection(vcpu); | |
1645 | audit_mappings(vcpu); | |
1646 | dbg = olddbg; | |
1647 | } | |
1648 | ||
1649 | #endif |