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 | 19 | |
1d737c8a | 20 | #include "mmu.h" |
e495606d | 21 | |
edf88417 | 22 | #include <linux/kvm_host.h> |
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> | |
448353ca | 28 | #include <linux/swap.h> |
05da4558 | 29 | #include <linux/hugetlb.h> |
2f333bcb | 30 | #include <linux/compiler.h> |
6aa8b732 | 31 | |
e495606d AK |
32 | #include <asm/page.h> |
33 | #include <asm/cmpxchg.h> | |
4e542370 | 34 | #include <asm/io.h> |
13673a90 | 35 | #include <asm/vmx.h> |
6aa8b732 | 36 | |
18552672 JR |
37 | /* |
38 | * When setting this variable to true it enables Two-Dimensional-Paging | |
39 | * where the hardware walks 2 page tables: | |
40 | * 1. the guest-virtual to guest-physical | |
41 | * 2. while doing 1. it walks guest-physical to host-physical | |
42 | * If the hardware supports that we don't need to do shadow paging. | |
43 | */ | |
2f333bcb | 44 | bool tdp_enabled = false; |
18552672 | 45 | |
37a7d8b0 AK |
46 | #undef MMU_DEBUG |
47 | ||
48 | #undef AUDIT | |
49 | ||
50 | #ifdef AUDIT | |
51 | static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg); | |
52 | #else | |
53 | static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg) {} | |
54 | #endif | |
55 | ||
56 | #ifdef MMU_DEBUG | |
57 | ||
58 | #define pgprintk(x...) do { if (dbg) printk(x); } while (0) | |
59 | #define rmap_printk(x...) do { if (dbg) printk(x); } while (0) | |
60 | ||
61 | #else | |
62 | ||
63 | #define pgprintk(x...) do { } while (0) | |
64 | #define rmap_printk(x...) do { } while (0) | |
65 | ||
66 | #endif | |
67 | ||
68 | #if defined(MMU_DEBUG) || defined(AUDIT) | |
6ada8cca AK |
69 | static int dbg = 0; |
70 | module_param(dbg, bool, 0644); | |
37a7d8b0 | 71 | #endif |
6aa8b732 | 72 | |
582801a9 MT |
73 | static int oos_shadow = 1; |
74 | module_param(oos_shadow, bool, 0644); | |
75 | ||
d6c69ee9 YD |
76 | #ifndef MMU_DEBUG |
77 | #define ASSERT(x) do { } while (0) | |
78 | #else | |
6aa8b732 AK |
79 | #define ASSERT(x) \ |
80 | if (!(x)) { \ | |
81 | printk(KERN_WARNING "assertion failed %s:%d: %s\n", \ | |
82 | __FILE__, __LINE__, #x); \ | |
83 | } | |
d6c69ee9 | 84 | #endif |
6aa8b732 | 85 | |
6aa8b732 AK |
86 | #define PT_FIRST_AVAIL_BITS_SHIFT 9 |
87 | #define PT64_SECOND_AVAIL_BITS_SHIFT 52 | |
88 | ||
6aa8b732 AK |
89 | #define VALID_PAGE(x) ((x) != INVALID_PAGE) |
90 | ||
91 | #define PT64_LEVEL_BITS 9 | |
92 | ||
93 | #define PT64_LEVEL_SHIFT(level) \ | |
d77c26fc | 94 | (PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS) |
6aa8b732 AK |
95 | |
96 | #define PT64_LEVEL_MASK(level) \ | |
97 | (((1ULL << PT64_LEVEL_BITS) - 1) << PT64_LEVEL_SHIFT(level)) | |
98 | ||
99 | #define PT64_INDEX(address, level)\ | |
100 | (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1)) | |
101 | ||
102 | ||
103 | #define PT32_LEVEL_BITS 10 | |
104 | ||
105 | #define PT32_LEVEL_SHIFT(level) \ | |
d77c26fc | 106 | (PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS) |
6aa8b732 AK |
107 | |
108 | #define PT32_LEVEL_MASK(level) \ | |
109 | (((1ULL << PT32_LEVEL_BITS) - 1) << PT32_LEVEL_SHIFT(level)) | |
110 | ||
111 | #define PT32_INDEX(address, level)\ | |
112 | (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1)) | |
113 | ||
114 | ||
27aba766 | 115 | #define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1)) |
6aa8b732 AK |
116 | #define PT64_DIR_BASE_ADDR_MASK \ |
117 | (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1)) | |
118 | ||
119 | #define PT32_BASE_ADDR_MASK PAGE_MASK | |
120 | #define PT32_DIR_BASE_ADDR_MASK \ | |
121 | (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1)) | |
122 | ||
79539cec AK |
123 | #define PT64_PERM_MASK (PT_PRESENT_MASK | PT_WRITABLE_MASK | PT_USER_MASK \ |
124 | | PT64_NX_MASK) | |
6aa8b732 AK |
125 | |
126 | #define PFERR_PRESENT_MASK (1U << 0) | |
127 | #define PFERR_WRITE_MASK (1U << 1) | |
128 | #define PFERR_USER_MASK (1U << 2) | |
73b1087e | 129 | #define PFERR_FETCH_MASK (1U << 4) |
6aa8b732 | 130 | |
6aa8b732 AK |
131 | #define PT_DIRECTORY_LEVEL 2 |
132 | #define PT_PAGE_TABLE_LEVEL 1 | |
133 | ||
cd4a4e53 AK |
134 | #define RMAP_EXT 4 |
135 | ||
fe135d2c AK |
136 | #define ACC_EXEC_MASK 1 |
137 | #define ACC_WRITE_MASK PT_WRITABLE_MASK | |
138 | #define ACC_USER_MASK PT_USER_MASK | |
139 | #define ACC_ALL (ACC_EXEC_MASK | ACC_WRITE_MASK | ACC_USER_MASK) | |
140 | ||
135f8c2b AK |
141 | #define SHADOW_PT_INDEX(addr, level) PT64_INDEX(addr, level) |
142 | ||
cd4a4e53 AK |
143 | struct kvm_rmap_desc { |
144 | u64 *shadow_ptes[RMAP_EXT]; | |
145 | struct kvm_rmap_desc *more; | |
146 | }; | |
147 | ||
3d000db5 AK |
148 | struct kvm_shadow_walk { |
149 | int (*entry)(struct kvm_shadow_walk *walk, struct kvm_vcpu *vcpu, | |
d40a1ee4 | 150 | u64 addr, u64 *spte, int level); |
3d000db5 AK |
151 | }; |
152 | ||
4731d4c7 MT |
153 | struct kvm_unsync_walk { |
154 | int (*entry) (struct kvm_mmu_page *sp, struct kvm_unsync_walk *walk); | |
155 | }; | |
156 | ||
ad8cfbe3 MT |
157 | typedef int (*mmu_parent_walk_fn) (struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp); |
158 | ||
b5a33a75 AK |
159 | static struct kmem_cache *pte_chain_cache; |
160 | static struct kmem_cache *rmap_desc_cache; | |
d3d25b04 | 161 | static struct kmem_cache *mmu_page_header_cache; |
b5a33a75 | 162 | |
c7addb90 AK |
163 | static u64 __read_mostly shadow_trap_nonpresent_pte; |
164 | static u64 __read_mostly shadow_notrap_nonpresent_pte; | |
7b52345e SY |
165 | static u64 __read_mostly shadow_base_present_pte; |
166 | static u64 __read_mostly shadow_nx_mask; | |
167 | static u64 __read_mostly shadow_x_mask; /* mutual exclusive with nx_mask */ | |
168 | static u64 __read_mostly shadow_user_mask; | |
169 | static u64 __read_mostly shadow_accessed_mask; | |
170 | static u64 __read_mostly shadow_dirty_mask; | |
64d4d521 | 171 | static u64 __read_mostly shadow_mt_mask; |
c7addb90 AK |
172 | |
173 | void kvm_mmu_set_nonpresent_ptes(u64 trap_pte, u64 notrap_pte) | |
174 | { | |
175 | shadow_trap_nonpresent_pte = trap_pte; | |
176 | shadow_notrap_nonpresent_pte = notrap_pte; | |
177 | } | |
178 | EXPORT_SYMBOL_GPL(kvm_mmu_set_nonpresent_ptes); | |
179 | ||
7b52345e SY |
180 | void kvm_mmu_set_base_ptes(u64 base_pte) |
181 | { | |
182 | shadow_base_present_pte = base_pte; | |
183 | } | |
184 | EXPORT_SYMBOL_GPL(kvm_mmu_set_base_ptes); | |
185 | ||
186 | void kvm_mmu_set_mask_ptes(u64 user_mask, u64 accessed_mask, | |
64d4d521 | 187 | u64 dirty_mask, u64 nx_mask, u64 x_mask, u64 mt_mask) |
7b52345e SY |
188 | { |
189 | shadow_user_mask = user_mask; | |
190 | shadow_accessed_mask = accessed_mask; | |
191 | shadow_dirty_mask = dirty_mask; | |
192 | shadow_nx_mask = nx_mask; | |
193 | shadow_x_mask = x_mask; | |
64d4d521 | 194 | shadow_mt_mask = mt_mask; |
7b52345e SY |
195 | } |
196 | EXPORT_SYMBOL_GPL(kvm_mmu_set_mask_ptes); | |
197 | ||
6aa8b732 AK |
198 | static int is_write_protection(struct kvm_vcpu *vcpu) |
199 | { | |
ad312c7c | 200 | return vcpu->arch.cr0 & X86_CR0_WP; |
6aa8b732 AK |
201 | } |
202 | ||
203 | static int is_cpuid_PSE36(void) | |
204 | { | |
205 | return 1; | |
206 | } | |
207 | ||
73b1087e AK |
208 | static int is_nx(struct kvm_vcpu *vcpu) |
209 | { | |
ad312c7c | 210 | return vcpu->arch.shadow_efer & EFER_NX; |
73b1087e AK |
211 | } |
212 | ||
6aa8b732 AK |
213 | static int is_present_pte(unsigned long pte) |
214 | { | |
215 | return pte & PT_PRESENT_MASK; | |
216 | } | |
217 | ||
c7addb90 AK |
218 | static int is_shadow_present_pte(u64 pte) |
219 | { | |
c7addb90 AK |
220 | return pte != shadow_trap_nonpresent_pte |
221 | && pte != shadow_notrap_nonpresent_pte; | |
222 | } | |
223 | ||
05da4558 MT |
224 | static int is_large_pte(u64 pte) |
225 | { | |
226 | return pte & PT_PAGE_SIZE_MASK; | |
227 | } | |
228 | ||
6aa8b732 AK |
229 | static int is_writeble_pte(unsigned long pte) |
230 | { | |
231 | return pte & PT_WRITABLE_MASK; | |
232 | } | |
233 | ||
e3c5e7ec AK |
234 | static int is_dirty_pte(unsigned long pte) |
235 | { | |
7b52345e | 236 | return pte & shadow_dirty_mask; |
e3c5e7ec AK |
237 | } |
238 | ||
cd4a4e53 AK |
239 | static int is_rmap_pte(u64 pte) |
240 | { | |
4b1a80fa | 241 | return is_shadow_present_pte(pte); |
cd4a4e53 AK |
242 | } |
243 | ||
35149e21 | 244 | static pfn_t spte_to_pfn(u64 pte) |
0b49ea86 | 245 | { |
35149e21 | 246 | return (pte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT; |
0b49ea86 AK |
247 | } |
248 | ||
da928521 AK |
249 | static gfn_t pse36_gfn_delta(u32 gpte) |
250 | { | |
251 | int shift = 32 - PT32_DIR_PSE36_SHIFT - PAGE_SHIFT; | |
252 | ||
253 | return (gpte & PT32_DIR_PSE36_MASK) << shift; | |
254 | } | |
255 | ||
e663ee64 AK |
256 | static void set_shadow_pte(u64 *sptep, u64 spte) |
257 | { | |
258 | #ifdef CONFIG_X86_64 | |
259 | set_64bit((unsigned long *)sptep, spte); | |
260 | #else | |
261 | set_64bit((unsigned long long *)sptep, spte); | |
262 | #endif | |
263 | } | |
264 | ||
e2dec939 | 265 | static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache, |
2e3e5882 | 266 | struct kmem_cache *base_cache, int min) |
714b93da AK |
267 | { |
268 | void *obj; | |
269 | ||
270 | if (cache->nobjs >= min) | |
e2dec939 | 271 | return 0; |
714b93da | 272 | while (cache->nobjs < ARRAY_SIZE(cache->objects)) { |
2e3e5882 | 273 | obj = kmem_cache_zalloc(base_cache, GFP_KERNEL); |
714b93da | 274 | if (!obj) |
e2dec939 | 275 | return -ENOMEM; |
714b93da AK |
276 | cache->objects[cache->nobjs++] = obj; |
277 | } | |
e2dec939 | 278 | return 0; |
714b93da AK |
279 | } |
280 | ||
281 | static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc) | |
282 | { | |
283 | while (mc->nobjs) | |
284 | kfree(mc->objects[--mc->nobjs]); | |
285 | } | |
286 | ||
c1158e63 | 287 | static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache *cache, |
2e3e5882 | 288 | int min) |
c1158e63 AK |
289 | { |
290 | struct page *page; | |
291 | ||
292 | if (cache->nobjs >= min) | |
293 | return 0; | |
294 | while (cache->nobjs < ARRAY_SIZE(cache->objects)) { | |
2e3e5882 | 295 | page = alloc_page(GFP_KERNEL); |
c1158e63 AK |
296 | if (!page) |
297 | return -ENOMEM; | |
298 | set_page_private(page, 0); | |
299 | cache->objects[cache->nobjs++] = page_address(page); | |
300 | } | |
301 | return 0; | |
302 | } | |
303 | ||
304 | static void mmu_free_memory_cache_page(struct kvm_mmu_memory_cache *mc) | |
305 | { | |
306 | while (mc->nobjs) | |
c4d198d5 | 307 | free_page((unsigned long)mc->objects[--mc->nobjs]); |
c1158e63 AK |
308 | } |
309 | ||
2e3e5882 | 310 | static int mmu_topup_memory_caches(struct kvm_vcpu *vcpu) |
714b93da | 311 | { |
e2dec939 AK |
312 | int r; |
313 | ||
ad312c7c | 314 | r = mmu_topup_memory_cache(&vcpu->arch.mmu_pte_chain_cache, |
2e3e5882 | 315 | pte_chain_cache, 4); |
e2dec939 AK |
316 | if (r) |
317 | goto out; | |
ad312c7c | 318 | r = mmu_topup_memory_cache(&vcpu->arch.mmu_rmap_desc_cache, |
c41ef344 | 319 | rmap_desc_cache, 4); |
d3d25b04 AK |
320 | if (r) |
321 | goto out; | |
ad312c7c | 322 | r = mmu_topup_memory_cache_page(&vcpu->arch.mmu_page_cache, 8); |
d3d25b04 AK |
323 | if (r) |
324 | goto out; | |
ad312c7c | 325 | r = mmu_topup_memory_cache(&vcpu->arch.mmu_page_header_cache, |
2e3e5882 | 326 | mmu_page_header_cache, 4); |
e2dec939 AK |
327 | out: |
328 | return r; | |
714b93da AK |
329 | } |
330 | ||
331 | static void mmu_free_memory_caches(struct kvm_vcpu *vcpu) | |
332 | { | |
ad312c7c ZX |
333 | mmu_free_memory_cache(&vcpu->arch.mmu_pte_chain_cache); |
334 | mmu_free_memory_cache(&vcpu->arch.mmu_rmap_desc_cache); | |
335 | mmu_free_memory_cache_page(&vcpu->arch.mmu_page_cache); | |
336 | mmu_free_memory_cache(&vcpu->arch.mmu_page_header_cache); | |
714b93da AK |
337 | } |
338 | ||
339 | static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc, | |
340 | size_t size) | |
341 | { | |
342 | void *p; | |
343 | ||
344 | BUG_ON(!mc->nobjs); | |
345 | p = mc->objects[--mc->nobjs]; | |
346 | memset(p, 0, size); | |
347 | return p; | |
348 | } | |
349 | ||
714b93da AK |
350 | static struct kvm_pte_chain *mmu_alloc_pte_chain(struct kvm_vcpu *vcpu) |
351 | { | |
ad312c7c | 352 | return mmu_memory_cache_alloc(&vcpu->arch.mmu_pte_chain_cache, |
714b93da AK |
353 | sizeof(struct kvm_pte_chain)); |
354 | } | |
355 | ||
90cb0529 | 356 | static void mmu_free_pte_chain(struct kvm_pte_chain *pc) |
714b93da | 357 | { |
90cb0529 | 358 | kfree(pc); |
714b93da AK |
359 | } |
360 | ||
361 | static struct kvm_rmap_desc *mmu_alloc_rmap_desc(struct kvm_vcpu *vcpu) | |
362 | { | |
ad312c7c | 363 | return mmu_memory_cache_alloc(&vcpu->arch.mmu_rmap_desc_cache, |
714b93da AK |
364 | sizeof(struct kvm_rmap_desc)); |
365 | } | |
366 | ||
90cb0529 | 367 | static void mmu_free_rmap_desc(struct kvm_rmap_desc *rd) |
714b93da | 368 | { |
90cb0529 | 369 | kfree(rd); |
714b93da AK |
370 | } |
371 | ||
05da4558 MT |
372 | /* |
373 | * Return the pointer to the largepage write count for a given | |
374 | * gfn, handling slots that are not large page aligned. | |
375 | */ | |
376 | static int *slot_largepage_idx(gfn_t gfn, struct kvm_memory_slot *slot) | |
377 | { | |
378 | unsigned long idx; | |
379 | ||
380 | idx = (gfn / KVM_PAGES_PER_HPAGE) - | |
381 | (slot->base_gfn / KVM_PAGES_PER_HPAGE); | |
382 | return &slot->lpage_info[idx].write_count; | |
383 | } | |
384 | ||
385 | static void account_shadowed(struct kvm *kvm, gfn_t gfn) | |
386 | { | |
387 | int *write_count; | |
388 | ||
2843099f IE |
389 | gfn = unalias_gfn(kvm, gfn); |
390 | write_count = slot_largepage_idx(gfn, | |
391 | gfn_to_memslot_unaliased(kvm, gfn)); | |
05da4558 | 392 | *write_count += 1; |
05da4558 MT |
393 | } |
394 | ||
395 | static void unaccount_shadowed(struct kvm *kvm, gfn_t gfn) | |
396 | { | |
397 | int *write_count; | |
398 | ||
2843099f IE |
399 | gfn = unalias_gfn(kvm, gfn); |
400 | write_count = slot_largepage_idx(gfn, | |
401 | gfn_to_memslot_unaliased(kvm, gfn)); | |
05da4558 MT |
402 | *write_count -= 1; |
403 | WARN_ON(*write_count < 0); | |
404 | } | |
405 | ||
406 | static int has_wrprotected_page(struct kvm *kvm, gfn_t gfn) | |
407 | { | |
2843099f | 408 | struct kvm_memory_slot *slot; |
05da4558 MT |
409 | int *largepage_idx; |
410 | ||
2843099f IE |
411 | gfn = unalias_gfn(kvm, gfn); |
412 | slot = gfn_to_memslot_unaliased(kvm, gfn); | |
05da4558 MT |
413 | if (slot) { |
414 | largepage_idx = slot_largepage_idx(gfn, slot); | |
415 | return *largepage_idx; | |
416 | } | |
417 | ||
418 | return 1; | |
419 | } | |
420 | ||
421 | static int host_largepage_backed(struct kvm *kvm, gfn_t gfn) | |
422 | { | |
423 | struct vm_area_struct *vma; | |
424 | unsigned long addr; | |
4c2155ce | 425 | int ret = 0; |
05da4558 MT |
426 | |
427 | addr = gfn_to_hva(kvm, gfn); | |
428 | if (kvm_is_error_hva(addr)) | |
4c2155ce | 429 | return ret; |
05da4558 | 430 | |
4c2155ce | 431 | down_read(¤t->mm->mmap_sem); |
05da4558 MT |
432 | vma = find_vma(current->mm, addr); |
433 | if (vma && is_vm_hugetlb_page(vma)) | |
4c2155ce MT |
434 | ret = 1; |
435 | up_read(¤t->mm->mmap_sem); | |
05da4558 | 436 | |
4c2155ce | 437 | return ret; |
05da4558 MT |
438 | } |
439 | ||
440 | static int is_largepage_backed(struct kvm_vcpu *vcpu, gfn_t large_gfn) | |
441 | { | |
442 | struct kvm_memory_slot *slot; | |
443 | ||
444 | if (has_wrprotected_page(vcpu->kvm, large_gfn)) | |
445 | return 0; | |
446 | ||
447 | if (!host_largepage_backed(vcpu->kvm, large_gfn)) | |
448 | return 0; | |
449 | ||
450 | slot = gfn_to_memslot(vcpu->kvm, large_gfn); | |
451 | if (slot && slot->dirty_bitmap) | |
452 | return 0; | |
453 | ||
454 | return 1; | |
455 | } | |
456 | ||
290fc38d IE |
457 | /* |
458 | * Take gfn and return the reverse mapping to it. | |
459 | * Note: gfn must be unaliased before this function get called | |
460 | */ | |
461 | ||
05da4558 | 462 | static unsigned long *gfn_to_rmap(struct kvm *kvm, gfn_t gfn, int lpage) |
290fc38d IE |
463 | { |
464 | struct kvm_memory_slot *slot; | |
05da4558 | 465 | unsigned long idx; |
290fc38d IE |
466 | |
467 | slot = gfn_to_memslot(kvm, gfn); | |
05da4558 MT |
468 | if (!lpage) |
469 | return &slot->rmap[gfn - slot->base_gfn]; | |
470 | ||
471 | idx = (gfn / KVM_PAGES_PER_HPAGE) - | |
472 | (slot->base_gfn / KVM_PAGES_PER_HPAGE); | |
473 | ||
474 | return &slot->lpage_info[idx].rmap_pde; | |
290fc38d IE |
475 | } |
476 | ||
cd4a4e53 AK |
477 | /* |
478 | * Reverse mapping data structures: | |
479 | * | |
290fc38d IE |
480 | * If rmapp bit zero is zero, then rmapp point to the shadw page table entry |
481 | * that points to page_address(page). | |
cd4a4e53 | 482 | * |
290fc38d IE |
483 | * If rmapp bit zero is one, (then rmap & ~1) points to a struct kvm_rmap_desc |
484 | * containing more mappings. | |
cd4a4e53 | 485 | */ |
05da4558 | 486 | static void rmap_add(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn, int lpage) |
cd4a4e53 | 487 | { |
4db35314 | 488 | struct kvm_mmu_page *sp; |
cd4a4e53 | 489 | struct kvm_rmap_desc *desc; |
290fc38d | 490 | unsigned long *rmapp; |
cd4a4e53 AK |
491 | int i; |
492 | ||
493 | if (!is_rmap_pte(*spte)) | |
494 | return; | |
290fc38d | 495 | gfn = unalias_gfn(vcpu->kvm, gfn); |
4db35314 AK |
496 | sp = page_header(__pa(spte)); |
497 | sp->gfns[spte - sp->spt] = gfn; | |
05da4558 | 498 | rmapp = gfn_to_rmap(vcpu->kvm, gfn, lpage); |
290fc38d | 499 | if (!*rmapp) { |
cd4a4e53 | 500 | rmap_printk("rmap_add: %p %llx 0->1\n", spte, *spte); |
290fc38d IE |
501 | *rmapp = (unsigned long)spte; |
502 | } else if (!(*rmapp & 1)) { | |
cd4a4e53 | 503 | rmap_printk("rmap_add: %p %llx 1->many\n", spte, *spte); |
714b93da | 504 | desc = mmu_alloc_rmap_desc(vcpu); |
290fc38d | 505 | desc->shadow_ptes[0] = (u64 *)*rmapp; |
cd4a4e53 | 506 | desc->shadow_ptes[1] = spte; |
290fc38d | 507 | *rmapp = (unsigned long)desc | 1; |
cd4a4e53 AK |
508 | } else { |
509 | rmap_printk("rmap_add: %p %llx many->many\n", spte, *spte); | |
290fc38d | 510 | desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul); |
cd4a4e53 AK |
511 | while (desc->shadow_ptes[RMAP_EXT-1] && desc->more) |
512 | desc = desc->more; | |
513 | if (desc->shadow_ptes[RMAP_EXT-1]) { | |
714b93da | 514 | desc->more = mmu_alloc_rmap_desc(vcpu); |
cd4a4e53 AK |
515 | desc = desc->more; |
516 | } | |
517 | for (i = 0; desc->shadow_ptes[i]; ++i) | |
518 | ; | |
519 | desc->shadow_ptes[i] = spte; | |
520 | } | |
521 | } | |
522 | ||
290fc38d | 523 | static void rmap_desc_remove_entry(unsigned long *rmapp, |
cd4a4e53 AK |
524 | struct kvm_rmap_desc *desc, |
525 | int i, | |
526 | struct kvm_rmap_desc *prev_desc) | |
527 | { | |
528 | int j; | |
529 | ||
530 | for (j = RMAP_EXT - 1; !desc->shadow_ptes[j] && j > i; --j) | |
531 | ; | |
532 | desc->shadow_ptes[i] = desc->shadow_ptes[j]; | |
11718b4d | 533 | desc->shadow_ptes[j] = NULL; |
cd4a4e53 AK |
534 | if (j != 0) |
535 | return; | |
536 | if (!prev_desc && !desc->more) | |
290fc38d | 537 | *rmapp = (unsigned long)desc->shadow_ptes[0]; |
cd4a4e53 AK |
538 | else |
539 | if (prev_desc) | |
540 | prev_desc->more = desc->more; | |
541 | else | |
290fc38d | 542 | *rmapp = (unsigned long)desc->more | 1; |
90cb0529 | 543 | mmu_free_rmap_desc(desc); |
cd4a4e53 AK |
544 | } |
545 | ||
290fc38d | 546 | static void rmap_remove(struct kvm *kvm, u64 *spte) |
cd4a4e53 | 547 | { |
cd4a4e53 AK |
548 | struct kvm_rmap_desc *desc; |
549 | struct kvm_rmap_desc *prev_desc; | |
4db35314 | 550 | struct kvm_mmu_page *sp; |
35149e21 | 551 | pfn_t pfn; |
290fc38d | 552 | unsigned long *rmapp; |
cd4a4e53 AK |
553 | int i; |
554 | ||
555 | if (!is_rmap_pte(*spte)) | |
556 | return; | |
4db35314 | 557 | sp = page_header(__pa(spte)); |
35149e21 | 558 | pfn = spte_to_pfn(*spte); |
7b52345e | 559 | if (*spte & shadow_accessed_mask) |
35149e21 | 560 | kvm_set_pfn_accessed(pfn); |
b4231d61 | 561 | if (is_writeble_pte(*spte)) |
35149e21 | 562 | kvm_release_pfn_dirty(pfn); |
b4231d61 | 563 | else |
35149e21 | 564 | kvm_release_pfn_clean(pfn); |
05da4558 | 565 | rmapp = gfn_to_rmap(kvm, sp->gfns[spte - sp->spt], is_large_pte(*spte)); |
290fc38d | 566 | if (!*rmapp) { |
cd4a4e53 AK |
567 | printk(KERN_ERR "rmap_remove: %p %llx 0->BUG\n", spte, *spte); |
568 | BUG(); | |
290fc38d | 569 | } else if (!(*rmapp & 1)) { |
cd4a4e53 | 570 | rmap_printk("rmap_remove: %p %llx 1->0\n", spte, *spte); |
290fc38d | 571 | if ((u64 *)*rmapp != spte) { |
cd4a4e53 AK |
572 | printk(KERN_ERR "rmap_remove: %p %llx 1->BUG\n", |
573 | spte, *spte); | |
574 | BUG(); | |
575 | } | |
290fc38d | 576 | *rmapp = 0; |
cd4a4e53 AK |
577 | } else { |
578 | rmap_printk("rmap_remove: %p %llx many->many\n", spte, *spte); | |
290fc38d | 579 | desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul); |
cd4a4e53 AK |
580 | prev_desc = NULL; |
581 | while (desc) { | |
582 | for (i = 0; i < RMAP_EXT && desc->shadow_ptes[i]; ++i) | |
583 | if (desc->shadow_ptes[i] == spte) { | |
290fc38d | 584 | rmap_desc_remove_entry(rmapp, |
714b93da | 585 | desc, i, |
cd4a4e53 AK |
586 | prev_desc); |
587 | return; | |
588 | } | |
589 | prev_desc = desc; | |
590 | desc = desc->more; | |
591 | } | |
592 | BUG(); | |
593 | } | |
594 | } | |
595 | ||
98348e95 | 596 | static u64 *rmap_next(struct kvm *kvm, unsigned long *rmapp, u64 *spte) |
374cbac0 | 597 | { |
374cbac0 | 598 | struct kvm_rmap_desc *desc; |
98348e95 IE |
599 | struct kvm_rmap_desc *prev_desc; |
600 | u64 *prev_spte; | |
601 | int i; | |
602 | ||
603 | if (!*rmapp) | |
604 | return NULL; | |
605 | else if (!(*rmapp & 1)) { | |
606 | if (!spte) | |
607 | return (u64 *)*rmapp; | |
608 | return NULL; | |
609 | } | |
610 | desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul); | |
611 | prev_desc = NULL; | |
612 | prev_spte = NULL; | |
613 | while (desc) { | |
614 | for (i = 0; i < RMAP_EXT && desc->shadow_ptes[i]; ++i) { | |
615 | if (prev_spte == spte) | |
616 | return desc->shadow_ptes[i]; | |
617 | prev_spte = desc->shadow_ptes[i]; | |
618 | } | |
619 | desc = desc->more; | |
620 | } | |
621 | return NULL; | |
622 | } | |
623 | ||
b1a36821 | 624 | static int rmap_write_protect(struct kvm *kvm, u64 gfn) |
98348e95 | 625 | { |
290fc38d | 626 | unsigned long *rmapp; |
374cbac0 | 627 | u64 *spte; |
caa5b8a5 | 628 | int write_protected = 0; |
374cbac0 | 629 | |
4a4c9924 | 630 | gfn = unalias_gfn(kvm, gfn); |
05da4558 | 631 | rmapp = gfn_to_rmap(kvm, gfn, 0); |
374cbac0 | 632 | |
98348e95 IE |
633 | spte = rmap_next(kvm, rmapp, NULL); |
634 | while (spte) { | |
374cbac0 | 635 | BUG_ON(!spte); |
374cbac0 | 636 | BUG_ON(!(*spte & PT_PRESENT_MASK)); |
374cbac0 | 637 | rmap_printk("rmap_write_protect: spte %p %llx\n", spte, *spte); |
caa5b8a5 | 638 | if (is_writeble_pte(*spte)) { |
9647c14c | 639 | set_shadow_pte(spte, *spte & ~PT_WRITABLE_MASK); |
caa5b8a5 ED |
640 | write_protected = 1; |
641 | } | |
9647c14c | 642 | spte = rmap_next(kvm, rmapp, spte); |
374cbac0 | 643 | } |
855149aa | 644 | if (write_protected) { |
35149e21 | 645 | pfn_t pfn; |
855149aa IE |
646 | |
647 | spte = rmap_next(kvm, rmapp, NULL); | |
35149e21 AL |
648 | pfn = spte_to_pfn(*spte); |
649 | kvm_set_pfn_dirty(pfn); | |
855149aa IE |
650 | } |
651 | ||
05da4558 MT |
652 | /* check for huge page mappings */ |
653 | rmapp = gfn_to_rmap(kvm, gfn, 1); | |
654 | spte = rmap_next(kvm, rmapp, NULL); | |
655 | while (spte) { | |
656 | BUG_ON(!spte); | |
657 | BUG_ON(!(*spte & PT_PRESENT_MASK)); | |
658 | BUG_ON((*spte & (PT_PAGE_SIZE_MASK|PT_PRESENT_MASK)) != (PT_PAGE_SIZE_MASK|PT_PRESENT_MASK)); | |
659 | pgprintk("rmap_write_protect(large): spte %p %llx %lld\n", spte, *spte, gfn); | |
660 | if (is_writeble_pte(*spte)) { | |
661 | rmap_remove(kvm, spte); | |
662 | --kvm->stat.lpages; | |
663 | set_shadow_pte(spte, shadow_trap_nonpresent_pte); | |
6597ca09 | 664 | spte = NULL; |
05da4558 MT |
665 | write_protected = 1; |
666 | } | |
667 | spte = rmap_next(kvm, rmapp, spte); | |
668 | } | |
669 | ||
b1a36821 | 670 | return write_protected; |
374cbac0 AK |
671 | } |
672 | ||
e930bffe AA |
673 | static int kvm_unmap_rmapp(struct kvm *kvm, unsigned long *rmapp) |
674 | { | |
675 | u64 *spte; | |
676 | int need_tlb_flush = 0; | |
677 | ||
678 | while ((spte = rmap_next(kvm, rmapp, NULL))) { | |
679 | BUG_ON(!(*spte & PT_PRESENT_MASK)); | |
680 | rmap_printk("kvm_rmap_unmap_hva: spte %p %llx\n", spte, *spte); | |
681 | rmap_remove(kvm, spte); | |
682 | set_shadow_pte(spte, shadow_trap_nonpresent_pte); | |
683 | need_tlb_flush = 1; | |
684 | } | |
685 | return need_tlb_flush; | |
686 | } | |
687 | ||
688 | static int kvm_handle_hva(struct kvm *kvm, unsigned long hva, | |
689 | int (*handler)(struct kvm *kvm, unsigned long *rmapp)) | |
690 | { | |
691 | int i; | |
692 | int retval = 0; | |
693 | ||
694 | /* | |
695 | * If mmap_sem isn't taken, we can look the memslots with only | |
696 | * the mmu_lock by skipping over the slots with userspace_addr == 0. | |
697 | */ | |
698 | for (i = 0; i < kvm->nmemslots; i++) { | |
699 | struct kvm_memory_slot *memslot = &kvm->memslots[i]; | |
700 | unsigned long start = memslot->userspace_addr; | |
701 | unsigned long end; | |
702 | ||
703 | /* mmu_lock protects userspace_addr */ | |
704 | if (!start) | |
705 | continue; | |
706 | ||
707 | end = start + (memslot->npages << PAGE_SHIFT); | |
708 | if (hva >= start && hva < end) { | |
709 | gfn_t gfn_offset = (hva - start) >> PAGE_SHIFT; | |
710 | retval |= handler(kvm, &memslot->rmap[gfn_offset]); | |
711 | retval |= handler(kvm, | |
712 | &memslot->lpage_info[ | |
713 | gfn_offset / | |
714 | KVM_PAGES_PER_HPAGE].rmap_pde); | |
715 | } | |
716 | } | |
717 | ||
718 | return retval; | |
719 | } | |
720 | ||
721 | int kvm_unmap_hva(struct kvm *kvm, unsigned long hva) | |
722 | { | |
723 | return kvm_handle_hva(kvm, hva, kvm_unmap_rmapp); | |
724 | } | |
725 | ||
726 | static int kvm_age_rmapp(struct kvm *kvm, unsigned long *rmapp) | |
727 | { | |
728 | u64 *spte; | |
729 | int young = 0; | |
730 | ||
534e38b4 SY |
731 | /* always return old for EPT */ |
732 | if (!shadow_accessed_mask) | |
733 | return 0; | |
734 | ||
e930bffe AA |
735 | spte = rmap_next(kvm, rmapp, NULL); |
736 | while (spte) { | |
737 | int _young; | |
738 | u64 _spte = *spte; | |
739 | BUG_ON(!(_spte & PT_PRESENT_MASK)); | |
740 | _young = _spte & PT_ACCESSED_MASK; | |
741 | if (_young) { | |
742 | young = 1; | |
743 | clear_bit(PT_ACCESSED_SHIFT, (unsigned long *)spte); | |
744 | } | |
745 | spte = rmap_next(kvm, rmapp, spte); | |
746 | } | |
747 | return young; | |
748 | } | |
749 | ||
750 | int kvm_age_hva(struct kvm *kvm, unsigned long hva) | |
751 | { | |
752 | return kvm_handle_hva(kvm, hva, kvm_age_rmapp); | |
753 | } | |
754 | ||
d6c69ee9 | 755 | #ifdef MMU_DEBUG |
47ad8e68 | 756 | static int is_empty_shadow_page(u64 *spt) |
6aa8b732 | 757 | { |
139bdb2d AK |
758 | u64 *pos; |
759 | u64 *end; | |
760 | ||
47ad8e68 | 761 | for (pos = spt, end = pos + PAGE_SIZE / sizeof(u64); pos != end; pos++) |
3c915510 | 762 | if (is_shadow_present_pte(*pos)) { |
b8688d51 | 763 | printk(KERN_ERR "%s: %p %llx\n", __func__, |
139bdb2d | 764 | pos, *pos); |
6aa8b732 | 765 | return 0; |
139bdb2d | 766 | } |
6aa8b732 AK |
767 | return 1; |
768 | } | |
d6c69ee9 | 769 | #endif |
6aa8b732 | 770 | |
4db35314 | 771 | static void kvm_mmu_free_page(struct kvm *kvm, struct kvm_mmu_page *sp) |
260746c0 | 772 | { |
4db35314 AK |
773 | ASSERT(is_empty_shadow_page(sp->spt)); |
774 | list_del(&sp->link); | |
775 | __free_page(virt_to_page(sp->spt)); | |
776 | __free_page(virt_to_page(sp->gfns)); | |
777 | kfree(sp); | |
f05e70ac | 778 | ++kvm->arch.n_free_mmu_pages; |
260746c0 AK |
779 | } |
780 | ||
cea0f0e7 AK |
781 | static unsigned kvm_page_table_hashfn(gfn_t gfn) |
782 | { | |
1ae0a13d | 783 | return gfn & ((1 << KVM_MMU_HASH_SHIFT) - 1); |
cea0f0e7 AK |
784 | } |
785 | ||
25c0de2c AK |
786 | static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu, |
787 | u64 *parent_pte) | |
6aa8b732 | 788 | { |
4db35314 | 789 | struct kvm_mmu_page *sp; |
6aa8b732 | 790 | |
ad312c7c ZX |
791 | sp = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_header_cache, sizeof *sp); |
792 | sp->spt = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_cache, PAGE_SIZE); | |
793 | sp->gfns = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_cache, PAGE_SIZE); | |
4db35314 | 794 | set_page_private(virt_to_page(sp->spt), (unsigned long)sp); |
f05e70ac | 795 | list_add(&sp->link, &vcpu->kvm->arch.active_mmu_pages); |
6cffe8ca | 796 | INIT_LIST_HEAD(&sp->oos_link); |
4db35314 | 797 | ASSERT(is_empty_shadow_page(sp->spt)); |
291f26bc | 798 | bitmap_zero(sp->slot_bitmap, KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS); |
4db35314 AK |
799 | sp->multimapped = 0; |
800 | sp->parent_pte = parent_pte; | |
f05e70ac | 801 | --vcpu->kvm->arch.n_free_mmu_pages; |
4db35314 | 802 | return sp; |
6aa8b732 AK |
803 | } |
804 | ||
714b93da | 805 | static void mmu_page_add_parent_pte(struct kvm_vcpu *vcpu, |
4db35314 | 806 | struct kvm_mmu_page *sp, u64 *parent_pte) |
cea0f0e7 AK |
807 | { |
808 | struct kvm_pte_chain *pte_chain; | |
809 | struct hlist_node *node; | |
810 | int i; | |
811 | ||
812 | if (!parent_pte) | |
813 | return; | |
4db35314 AK |
814 | if (!sp->multimapped) { |
815 | u64 *old = sp->parent_pte; | |
cea0f0e7 AK |
816 | |
817 | if (!old) { | |
4db35314 | 818 | sp->parent_pte = parent_pte; |
cea0f0e7 AK |
819 | return; |
820 | } | |
4db35314 | 821 | sp->multimapped = 1; |
714b93da | 822 | pte_chain = mmu_alloc_pte_chain(vcpu); |
4db35314 AK |
823 | INIT_HLIST_HEAD(&sp->parent_ptes); |
824 | hlist_add_head(&pte_chain->link, &sp->parent_ptes); | |
cea0f0e7 AK |
825 | pte_chain->parent_ptes[0] = old; |
826 | } | |
4db35314 | 827 | hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link) { |
cea0f0e7 AK |
828 | if (pte_chain->parent_ptes[NR_PTE_CHAIN_ENTRIES-1]) |
829 | continue; | |
830 | for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) | |
831 | if (!pte_chain->parent_ptes[i]) { | |
832 | pte_chain->parent_ptes[i] = parent_pte; | |
833 | return; | |
834 | } | |
835 | } | |
714b93da | 836 | pte_chain = mmu_alloc_pte_chain(vcpu); |
cea0f0e7 | 837 | BUG_ON(!pte_chain); |
4db35314 | 838 | hlist_add_head(&pte_chain->link, &sp->parent_ptes); |
cea0f0e7 AK |
839 | pte_chain->parent_ptes[0] = parent_pte; |
840 | } | |
841 | ||
4db35314 | 842 | static void mmu_page_remove_parent_pte(struct kvm_mmu_page *sp, |
cea0f0e7 AK |
843 | u64 *parent_pte) |
844 | { | |
845 | struct kvm_pte_chain *pte_chain; | |
846 | struct hlist_node *node; | |
847 | int i; | |
848 | ||
4db35314 AK |
849 | if (!sp->multimapped) { |
850 | BUG_ON(sp->parent_pte != parent_pte); | |
851 | sp->parent_pte = NULL; | |
cea0f0e7 AK |
852 | return; |
853 | } | |
4db35314 | 854 | hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link) |
cea0f0e7 AK |
855 | for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) { |
856 | if (!pte_chain->parent_ptes[i]) | |
857 | break; | |
858 | if (pte_chain->parent_ptes[i] != parent_pte) | |
859 | continue; | |
697fe2e2 AK |
860 | while (i + 1 < NR_PTE_CHAIN_ENTRIES |
861 | && pte_chain->parent_ptes[i + 1]) { | |
cea0f0e7 AK |
862 | pte_chain->parent_ptes[i] |
863 | = pte_chain->parent_ptes[i + 1]; | |
864 | ++i; | |
865 | } | |
866 | pte_chain->parent_ptes[i] = NULL; | |
697fe2e2 AK |
867 | if (i == 0) { |
868 | hlist_del(&pte_chain->link); | |
90cb0529 | 869 | mmu_free_pte_chain(pte_chain); |
4db35314 AK |
870 | if (hlist_empty(&sp->parent_ptes)) { |
871 | sp->multimapped = 0; | |
872 | sp->parent_pte = NULL; | |
697fe2e2 AK |
873 | } |
874 | } | |
cea0f0e7 AK |
875 | return; |
876 | } | |
877 | BUG(); | |
878 | } | |
879 | ||
ad8cfbe3 MT |
880 | |
881 | static void mmu_parent_walk(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp, | |
882 | mmu_parent_walk_fn fn) | |
883 | { | |
884 | struct kvm_pte_chain *pte_chain; | |
885 | struct hlist_node *node; | |
886 | struct kvm_mmu_page *parent_sp; | |
887 | int i; | |
888 | ||
889 | if (!sp->multimapped && sp->parent_pte) { | |
890 | parent_sp = page_header(__pa(sp->parent_pte)); | |
891 | fn(vcpu, parent_sp); | |
892 | mmu_parent_walk(vcpu, parent_sp, fn); | |
893 | return; | |
894 | } | |
895 | hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link) | |
896 | for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) { | |
897 | if (!pte_chain->parent_ptes[i]) | |
898 | break; | |
899 | parent_sp = page_header(__pa(pte_chain->parent_ptes[i])); | |
900 | fn(vcpu, parent_sp); | |
901 | mmu_parent_walk(vcpu, parent_sp, fn); | |
902 | } | |
903 | } | |
904 | ||
0074ff63 MT |
905 | static void kvm_mmu_update_unsync_bitmap(u64 *spte) |
906 | { | |
907 | unsigned int index; | |
908 | struct kvm_mmu_page *sp = page_header(__pa(spte)); | |
909 | ||
910 | index = spte - sp->spt; | |
60c8aec6 MT |
911 | if (!__test_and_set_bit(index, sp->unsync_child_bitmap)) |
912 | sp->unsync_children++; | |
913 | WARN_ON(!sp->unsync_children); | |
0074ff63 MT |
914 | } |
915 | ||
916 | static void kvm_mmu_update_parents_unsync(struct kvm_mmu_page *sp) | |
917 | { | |
918 | struct kvm_pte_chain *pte_chain; | |
919 | struct hlist_node *node; | |
920 | int i; | |
921 | ||
922 | if (!sp->parent_pte) | |
923 | return; | |
924 | ||
925 | if (!sp->multimapped) { | |
926 | kvm_mmu_update_unsync_bitmap(sp->parent_pte); | |
927 | return; | |
928 | } | |
929 | ||
930 | hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link) | |
931 | for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) { | |
932 | if (!pte_chain->parent_ptes[i]) | |
933 | break; | |
934 | kvm_mmu_update_unsync_bitmap(pte_chain->parent_ptes[i]); | |
935 | } | |
936 | } | |
937 | ||
938 | static int unsync_walk_fn(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp) | |
939 | { | |
0074ff63 MT |
940 | kvm_mmu_update_parents_unsync(sp); |
941 | return 1; | |
942 | } | |
943 | ||
944 | static void kvm_mmu_mark_parents_unsync(struct kvm_vcpu *vcpu, | |
945 | struct kvm_mmu_page *sp) | |
946 | { | |
947 | mmu_parent_walk(vcpu, sp, unsync_walk_fn); | |
948 | kvm_mmu_update_parents_unsync(sp); | |
949 | } | |
950 | ||
d761a501 AK |
951 | static void nonpaging_prefetch_page(struct kvm_vcpu *vcpu, |
952 | struct kvm_mmu_page *sp) | |
953 | { | |
954 | int i; | |
955 | ||
956 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) | |
957 | sp->spt[i] = shadow_trap_nonpresent_pte; | |
958 | } | |
959 | ||
e8bc217a MT |
960 | static int nonpaging_sync_page(struct kvm_vcpu *vcpu, |
961 | struct kvm_mmu_page *sp) | |
962 | { | |
963 | return 1; | |
964 | } | |
965 | ||
a7052897 MT |
966 | static void nonpaging_invlpg(struct kvm_vcpu *vcpu, gva_t gva) |
967 | { | |
968 | } | |
969 | ||
60c8aec6 MT |
970 | #define KVM_PAGE_ARRAY_NR 16 |
971 | ||
972 | struct kvm_mmu_pages { | |
973 | struct mmu_page_and_offset { | |
974 | struct kvm_mmu_page *sp; | |
975 | unsigned int idx; | |
976 | } page[KVM_PAGE_ARRAY_NR]; | |
977 | unsigned int nr; | |
978 | }; | |
979 | ||
0074ff63 MT |
980 | #define for_each_unsync_children(bitmap, idx) \ |
981 | for (idx = find_first_bit(bitmap, 512); \ | |
982 | idx < 512; \ | |
983 | idx = find_next_bit(bitmap, 512, idx+1)) | |
984 | ||
60c8aec6 MT |
985 | int mmu_pages_add(struct kvm_mmu_pages *pvec, struct kvm_mmu_page *sp, |
986 | int idx) | |
4731d4c7 | 987 | { |
60c8aec6 | 988 | int i; |
4731d4c7 | 989 | |
60c8aec6 MT |
990 | if (sp->unsync) |
991 | for (i=0; i < pvec->nr; i++) | |
992 | if (pvec->page[i].sp == sp) | |
993 | return 0; | |
994 | ||
995 | pvec->page[pvec->nr].sp = sp; | |
996 | pvec->page[pvec->nr].idx = idx; | |
997 | pvec->nr++; | |
998 | return (pvec->nr == KVM_PAGE_ARRAY_NR); | |
999 | } | |
1000 | ||
1001 | static int __mmu_unsync_walk(struct kvm_mmu_page *sp, | |
1002 | struct kvm_mmu_pages *pvec) | |
1003 | { | |
1004 | int i, ret, nr_unsync_leaf = 0; | |
4731d4c7 | 1005 | |
0074ff63 | 1006 | for_each_unsync_children(sp->unsync_child_bitmap, i) { |
4731d4c7 MT |
1007 | u64 ent = sp->spt[i]; |
1008 | ||
87917239 | 1009 | if (is_shadow_present_pte(ent) && !is_large_pte(ent)) { |
4731d4c7 MT |
1010 | struct kvm_mmu_page *child; |
1011 | child = page_header(ent & PT64_BASE_ADDR_MASK); | |
1012 | ||
1013 | if (child->unsync_children) { | |
60c8aec6 MT |
1014 | if (mmu_pages_add(pvec, child, i)) |
1015 | return -ENOSPC; | |
1016 | ||
1017 | ret = __mmu_unsync_walk(child, pvec); | |
1018 | if (!ret) | |
1019 | __clear_bit(i, sp->unsync_child_bitmap); | |
1020 | else if (ret > 0) | |
1021 | nr_unsync_leaf += ret; | |
1022 | else | |
4731d4c7 MT |
1023 | return ret; |
1024 | } | |
1025 | ||
1026 | if (child->unsync) { | |
60c8aec6 MT |
1027 | nr_unsync_leaf++; |
1028 | if (mmu_pages_add(pvec, child, i)) | |
1029 | return -ENOSPC; | |
4731d4c7 MT |
1030 | } |
1031 | } | |
1032 | } | |
1033 | ||
0074ff63 | 1034 | if (find_first_bit(sp->unsync_child_bitmap, 512) == 512) |
4731d4c7 MT |
1035 | sp->unsync_children = 0; |
1036 | ||
60c8aec6 MT |
1037 | return nr_unsync_leaf; |
1038 | } | |
1039 | ||
1040 | static int mmu_unsync_walk(struct kvm_mmu_page *sp, | |
1041 | struct kvm_mmu_pages *pvec) | |
1042 | { | |
1043 | if (!sp->unsync_children) | |
1044 | return 0; | |
1045 | ||
1046 | mmu_pages_add(pvec, sp, 0); | |
1047 | return __mmu_unsync_walk(sp, pvec); | |
4731d4c7 MT |
1048 | } |
1049 | ||
4db35314 | 1050 | static struct kvm_mmu_page *kvm_mmu_lookup_page(struct kvm *kvm, gfn_t gfn) |
cea0f0e7 AK |
1051 | { |
1052 | unsigned index; | |
1053 | struct hlist_head *bucket; | |
4db35314 | 1054 | struct kvm_mmu_page *sp; |
cea0f0e7 AK |
1055 | struct hlist_node *node; |
1056 | ||
b8688d51 | 1057 | pgprintk("%s: looking for gfn %lx\n", __func__, gfn); |
1ae0a13d | 1058 | index = kvm_page_table_hashfn(gfn); |
f05e70ac | 1059 | bucket = &kvm->arch.mmu_page_hash[index]; |
4db35314 | 1060 | hlist_for_each_entry(sp, node, bucket, hash_link) |
2e53d63a MT |
1061 | if (sp->gfn == gfn && !sp->role.metaphysical |
1062 | && !sp->role.invalid) { | |
cea0f0e7 | 1063 | pgprintk("%s: found role %x\n", |
b8688d51 | 1064 | __func__, sp->role.word); |
4db35314 | 1065 | return sp; |
cea0f0e7 AK |
1066 | } |
1067 | return NULL; | |
1068 | } | |
1069 | ||
6cffe8ca MT |
1070 | static void kvm_unlink_unsync_global(struct kvm *kvm, struct kvm_mmu_page *sp) |
1071 | { | |
1072 | list_del(&sp->oos_link); | |
1073 | --kvm->stat.mmu_unsync_global; | |
1074 | } | |
1075 | ||
4731d4c7 MT |
1076 | static void kvm_unlink_unsync_page(struct kvm *kvm, struct kvm_mmu_page *sp) |
1077 | { | |
1078 | WARN_ON(!sp->unsync); | |
1079 | sp->unsync = 0; | |
6cffe8ca MT |
1080 | if (sp->global) |
1081 | kvm_unlink_unsync_global(kvm, sp); | |
4731d4c7 MT |
1082 | --kvm->stat.mmu_unsync; |
1083 | } | |
1084 | ||
1085 | static int kvm_mmu_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp); | |
1086 | ||
1087 | static int kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp) | |
1088 | { | |
1089 | if (sp->role.glevels != vcpu->arch.mmu.root_level) { | |
1090 | kvm_mmu_zap_page(vcpu->kvm, sp); | |
1091 | return 1; | |
1092 | } | |
1093 | ||
b1a36821 MT |
1094 | if (rmap_write_protect(vcpu->kvm, sp->gfn)) |
1095 | kvm_flush_remote_tlbs(vcpu->kvm); | |
0c0f40bd | 1096 | kvm_unlink_unsync_page(vcpu->kvm, sp); |
4731d4c7 MT |
1097 | if (vcpu->arch.mmu.sync_page(vcpu, sp)) { |
1098 | kvm_mmu_zap_page(vcpu->kvm, sp); | |
1099 | return 1; | |
1100 | } | |
1101 | ||
1102 | kvm_mmu_flush_tlb(vcpu); | |
4731d4c7 MT |
1103 | return 0; |
1104 | } | |
1105 | ||
60c8aec6 MT |
1106 | struct mmu_page_path { |
1107 | struct kvm_mmu_page *parent[PT64_ROOT_LEVEL-1]; | |
1108 | unsigned int idx[PT64_ROOT_LEVEL-1]; | |
4731d4c7 MT |
1109 | }; |
1110 | ||
60c8aec6 MT |
1111 | #define for_each_sp(pvec, sp, parents, i) \ |
1112 | for (i = mmu_pages_next(&pvec, &parents, -1), \ | |
1113 | sp = pvec.page[i].sp; \ | |
1114 | i < pvec.nr && ({ sp = pvec.page[i].sp; 1;}); \ | |
1115 | i = mmu_pages_next(&pvec, &parents, i)) | |
1116 | ||
1117 | int mmu_pages_next(struct kvm_mmu_pages *pvec, struct mmu_page_path *parents, | |
1118 | int i) | |
1119 | { | |
1120 | int n; | |
1121 | ||
1122 | for (n = i+1; n < pvec->nr; n++) { | |
1123 | struct kvm_mmu_page *sp = pvec->page[n].sp; | |
1124 | ||
1125 | if (sp->role.level == PT_PAGE_TABLE_LEVEL) { | |
1126 | parents->idx[0] = pvec->page[n].idx; | |
1127 | return n; | |
1128 | } | |
1129 | ||
1130 | parents->parent[sp->role.level-2] = sp; | |
1131 | parents->idx[sp->role.level-1] = pvec->page[n].idx; | |
1132 | } | |
1133 | ||
1134 | return n; | |
1135 | } | |
1136 | ||
1137 | void mmu_pages_clear_parents(struct mmu_page_path *parents) | |
4731d4c7 | 1138 | { |
60c8aec6 MT |
1139 | struct kvm_mmu_page *sp; |
1140 | unsigned int level = 0; | |
1141 | ||
1142 | do { | |
1143 | unsigned int idx = parents->idx[level]; | |
4731d4c7 | 1144 | |
60c8aec6 MT |
1145 | sp = parents->parent[level]; |
1146 | if (!sp) | |
1147 | return; | |
1148 | ||
1149 | --sp->unsync_children; | |
1150 | WARN_ON((int)sp->unsync_children < 0); | |
1151 | __clear_bit(idx, sp->unsync_child_bitmap); | |
1152 | level++; | |
1153 | } while (level < PT64_ROOT_LEVEL-1 && !sp->unsync_children); | |
4731d4c7 MT |
1154 | } |
1155 | ||
60c8aec6 MT |
1156 | static void kvm_mmu_pages_init(struct kvm_mmu_page *parent, |
1157 | struct mmu_page_path *parents, | |
1158 | struct kvm_mmu_pages *pvec) | |
4731d4c7 | 1159 | { |
60c8aec6 MT |
1160 | parents->parent[parent->role.level-1] = NULL; |
1161 | pvec->nr = 0; | |
1162 | } | |
4731d4c7 | 1163 | |
60c8aec6 MT |
1164 | static void mmu_sync_children(struct kvm_vcpu *vcpu, |
1165 | struct kvm_mmu_page *parent) | |
1166 | { | |
1167 | int i; | |
1168 | struct kvm_mmu_page *sp; | |
1169 | struct mmu_page_path parents; | |
1170 | struct kvm_mmu_pages pages; | |
1171 | ||
1172 | kvm_mmu_pages_init(parent, &parents, &pages); | |
1173 | while (mmu_unsync_walk(parent, &pages)) { | |
b1a36821 MT |
1174 | int protected = 0; |
1175 | ||
1176 | for_each_sp(pages, sp, parents, i) | |
1177 | protected |= rmap_write_protect(vcpu->kvm, sp->gfn); | |
1178 | ||
1179 | if (protected) | |
1180 | kvm_flush_remote_tlbs(vcpu->kvm); | |
1181 | ||
60c8aec6 MT |
1182 | for_each_sp(pages, sp, parents, i) { |
1183 | kvm_sync_page(vcpu, sp); | |
1184 | mmu_pages_clear_parents(&parents); | |
1185 | } | |
4731d4c7 | 1186 | cond_resched_lock(&vcpu->kvm->mmu_lock); |
60c8aec6 MT |
1187 | kvm_mmu_pages_init(parent, &parents, &pages); |
1188 | } | |
4731d4c7 MT |
1189 | } |
1190 | ||
cea0f0e7 AK |
1191 | static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu, |
1192 | gfn_t gfn, | |
1193 | gva_t gaddr, | |
1194 | unsigned level, | |
1195 | int metaphysical, | |
41074d07 | 1196 | unsigned access, |
f7d9c7b7 | 1197 | u64 *parent_pte) |
cea0f0e7 AK |
1198 | { |
1199 | union kvm_mmu_page_role role; | |
1200 | unsigned index; | |
1201 | unsigned quadrant; | |
1202 | struct hlist_head *bucket; | |
4db35314 | 1203 | struct kvm_mmu_page *sp; |
4731d4c7 | 1204 | struct hlist_node *node, *tmp; |
cea0f0e7 | 1205 | |
a770f6f2 | 1206 | role = vcpu->arch.mmu.base_role; |
cea0f0e7 AK |
1207 | role.level = level; |
1208 | role.metaphysical = metaphysical; | |
41074d07 | 1209 | role.access = access; |
ad312c7c | 1210 | if (vcpu->arch.mmu.root_level <= PT32_ROOT_LEVEL) { |
cea0f0e7 AK |
1211 | quadrant = gaddr >> (PAGE_SHIFT + (PT64_PT_BITS * level)); |
1212 | quadrant &= (1 << ((PT32_PT_BITS - PT64_PT_BITS) * level)) - 1; | |
1213 | role.quadrant = quadrant; | |
1214 | } | |
b8688d51 | 1215 | pgprintk("%s: looking gfn %lx role %x\n", __func__, |
cea0f0e7 | 1216 | gfn, role.word); |
1ae0a13d | 1217 | index = kvm_page_table_hashfn(gfn); |
f05e70ac | 1218 | bucket = &vcpu->kvm->arch.mmu_page_hash[index]; |
4731d4c7 MT |
1219 | hlist_for_each_entry_safe(sp, node, tmp, bucket, hash_link) |
1220 | if (sp->gfn == gfn) { | |
1221 | if (sp->unsync) | |
1222 | if (kvm_sync_page(vcpu, sp)) | |
1223 | continue; | |
1224 | ||
1225 | if (sp->role.word != role.word) | |
1226 | continue; | |
1227 | ||
4db35314 | 1228 | mmu_page_add_parent_pte(vcpu, sp, parent_pte); |
0074ff63 MT |
1229 | if (sp->unsync_children) { |
1230 | set_bit(KVM_REQ_MMU_SYNC, &vcpu->requests); | |
1231 | kvm_mmu_mark_parents_unsync(vcpu, sp); | |
1232 | } | |
b8688d51 | 1233 | pgprintk("%s: found\n", __func__); |
4db35314 | 1234 | return sp; |
cea0f0e7 | 1235 | } |
dfc5aa00 | 1236 | ++vcpu->kvm->stat.mmu_cache_miss; |
4db35314 AK |
1237 | sp = kvm_mmu_alloc_page(vcpu, parent_pte); |
1238 | if (!sp) | |
1239 | return sp; | |
b8688d51 | 1240 | pgprintk("%s: adding gfn %lx role %x\n", __func__, gfn, role.word); |
4db35314 AK |
1241 | sp->gfn = gfn; |
1242 | sp->role = role; | |
e2078318 | 1243 | sp->global = role.cr4_pge; |
4db35314 | 1244 | hlist_add_head(&sp->hash_link, bucket); |
4731d4c7 | 1245 | if (!metaphysical) { |
b1a36821 MT |
1246 | if (rmap_write_protect(vcpu->kvm, gfn)) |
1247 | kvm_flush_remote_tlbs(vcpu->kvm); | |
4731d4c7 MT |
1248 | account_shadowed(vcpu->kvm, gfn); |
1249 | } | |
131d8279 AK |
1250 | if (shadow_trap_nonpresent_pte != shadow_notrap_nonpresent_pte) |
1251 | vcpu->arch.mmu.prefetch_page(vcpu, sp); | |
1252 | else | |
1253 | nonpaging_prefetch_page(vcpu, sp); | |
4db35314 | 1254 | return sp; |
cea0f0e7 AK |
1255 | } |
1256 | ||
3d000db5 | 1257 | static int walk_shadow(struct kvm_shadow_walk *walker, |
d40a1ee4 | 1258 | struct kvm_vcpu *vcpu, u64 addr) |
3d000db5 AK |
1259 | { |
1260 | hpa_t shadow_addr; | |
1261 | int level; | |
1262 | int r; | |
1263 | u64 *sptep; | |
1264 | unsigned index; | |
1265 | ||
1266 | shadow_addr = vcpu->arch.mmu.root_hpa; | |
1267 | level = vcpu->arch.mmu.shadow_root_level; | |
1268 | if (level == PT32E_ROOT_LEVEL) { | |
1269 | shadow_addr = vcpu->arch.mmu.pae_root[(addr >> 30) & 3]; | |
1270 | shadow_addr &= PT64_BASE_ADDR_MASK; | |
eb64f1e8 MT |
1271 | if (!shadow_addr) |
1272 | return 1; | |
3d000db5 AK |
1273 | --level; |
1274 | } | |
1275 | ||
1276 | while (level >= PT_PAGE_TABLE_LEVEL) { | |
1277 | index = SHADOW_PT_INDEX(addr, level); | |
1278 | sptep = ((u64 *)__va(shadow_addr)) + index; | |
1279 | r = walker->entry(walker, vcpu, addr, sptep, level); | |
1280 | if (r) | |
1281 | return r; | |
1282 | shadow_addr = *sptep & PT64_BASE_ADDR_MASK; | |
1283 | --level; | |
1284 | } | |
1285 | return 0; | |
1286 | } | |
1287 | ||
90cb0529 | 1288 | static void kvm_mmu_page_unlink_children(struct kvm *kvm, |
4db35314 | 1289 | struct kvm_mmu_page *sp) |
a436036b | 1290 | { |
697fe2e2 AK |
1291 | unsigned i; |
1292 | u64 *pt; | |
1293 | u64 ent; | |
1294 | ||
4db35314 | 1295 | pt = sp->spt; |
697fe2e2 | 1296 | |
4db35314 | 1297 | if (sp->role.level == PT_PAGE_TABLE_LEVEL) { |
697fe2e2 | 1298 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) { |
c7addb90 | 1299 | if (is_shadow_present_pte(pt[i])) |
290fc38d | 1300 | rmap_remove(kvm, &pt[i]); |
c7addb90 | 1301 | pt[i] = shadow_trap_nonpresent_pte; |
697fe2e2 AK |
1302 | } |
1303 | return; | |
1304 | } | |
1305 | ||
1306 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) { | |
1307 | ent = pt[i]; | |
1308 | ||
05da4558 MT |
1309 | if (is_shadow_present_pte(ent)) { |
1310 | if (!is_large_pte(ent)) { | |
1311 | ent &= PT64_BASE_ADDR_MASK; | |
1312 | mmu_page_remove_parent_pte(page_header(ent), | |
1313 | &pt[i]); | |
1314 | } else { | |
1315 | --kvm->stat.lpages; | |
1316 | rmap_remove(kvm, &pt[i]); | |
1317 | } | |
1318 | } | |
c7addb90 | 1319 | pt[i] = shadow_trap_nonpresent_pte; |
697fe2e2 | 1320 | } |
a436036b AK |
1321 | } |
1322 | ||
4db35314 | 1323 | static void kvm_mmu_put_page(struct kvm_mmu_page *sp, u64 *parent_pte) |
cea0f0e7 | 1324 | { |
4db35314 | 1325 | mmu_page_remove_parent_pte(sp, parent_pte); |
a436036b AK |
1326 | } |
1327 | ||
12b7d28f AK |
1328 | static void kvm_mmu_reset_last_pte_updated(struct kvm *kvm) |
1329 | { | |
1330 | int i; | |
1331 | ||
1332 | for (i = 0; i < KVM_MAX_VCPUS; ++i) | |
1333 | if (kvm->vcpus[i]) | |
ad312c7c | 1334 | kvm->vcpus[i]->arch.last_pte_updated = NULL; |
12b7d28f AK |
1335 | } |
1336 | ||
31aa2b44 | 1337 | static void kvm_mmu_unlink_parents(struct kvm *kvm, struct kvm_mmu_page *sp) |
a436036b AK |
1338 | { |
1339 | u64 *parent_pte; | |
1340 | ||
4db35314 AK |
1341 | while (sp->multimapped || sp->parent_pte) { |
1342 | if (!sp->multimapped) | |
1343 | parent_pte = sp->parent_pte; | |
a436036b AK |
1344 | else { |
1345 | struct kvm_pte_chain *chain; | |
1346 | ||
4db35314 | 1347 | chain = container_of(sp->parent_ptes.first, |
a436036b AK |
1348 | struct kvm_pte_chain, link); |
1349 | parent_pte = chain->parent_ptes[0]; | |
1350 | } | |
697fe2e2 | 1351 | BUG_ON(!parent_pte); |
4db35314 | 1352 | kvm_mmu_put_page(sp, parent_pte); |
c7addb90 | 1353 | set_shadow_pte(parent_pte, shadow_trap_nonpresent_pte); |
a436036b | 1354 | } |
31aa2b44 AK |
1355 | } |
1356 | ||
60c8aec6 MT |
1357 | static int mmu_zap_unsync_children(struct kvm *kvm, |
1358 | struct kvm_mmu_page *parent) | |
4731d4c7 | 1359 | { |
60c8aec6 MT |
1360 | int i, zapped = 0; |
1361 | struct mmu_page_path parents; | |
1362 | struct kvm_mmu_pages pages; | |
4731d4c7 | 1363 | |
60c8aec6 | 1364 | if (parent->role.level == PT_PAGE_TABLE_LEVEL) |
4731d4c7 | 1365 | return 0; |
60c8aec6 MT |
1366 | |
1367 | kvm_mmu_pages_init(parent, &parents, &pages); | |
1368 | while (mmu_unsync_walk(parent, &pages)) { | |
1369 | struct kvm_mmu_page *sp; | |
1370 | ||
1371 | for_each_sp(pages, sp, parents, i) { | |
1372 | kvm_mmu_zap_page(kvm, sp); | |
1373 | mmu_pages_clear_parents(&parents); | |
1374 | } | |
1375 | zapped += pages.nr; | |
1376 | kvm_mmu_pages_init(parent, &parents, &pages); | |
1377 | } | |
1378 | ||
1379 | return zapped; | |
4731d4c7 MT |
1380 | } |
1381 | ||
07385413 | 1382 | static int kvm_mmu_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp) |
31aa2b44 | 1383 | { |
4731d4c7 | 1384 | int ret; |
31aa2b44 | 1385 | ++kvm->stat.mmu_shadow_zapped; |
4731d4c7 | 1386 | ret = mmu_zap_unsync_children(kvm, sp); |
4db35314 | 1387 | kvm_mmu_page_unlink_children(kvm, sp); |
31aa2b44 | 1388 | kvm_mmu_unlink_parents(kvm, sp); |
5b5c6a5a AK |
1389 | kvm_flush_remote_tlbs(kvm); |
1390 | if (!sp->role.invalid && !sp->role.metaphysical) | |
1391 | unaccount_shadowed(kvm, sp->gfn); | |
4731d4c7 MT |
1392 | if (sp->unsync) |
1393 | kvm_unlink_unsync_page(kvm, sp); | |
4db35314 AK |
1394 | if (!sp->root_count) { |
1395 | hlist_del(&sp->hash_link); | |
1396 | kvm_mmu_free_page(kvm, sp); | |
2e53d63a | 1397 | } else { |
2e53d63a | 1398 | sp->role.invalid = 1; |
5b5c6a5a | 1399 | list_move(&sp->link, &kvm->arch.active_mmu_pages); |
2e53d63a MT |
1400 | kvm_reload_remote_mmus(kvm); |
1401 | } | |
12b7d28f | 1402 | kvm_mmu_reset_last_pte_updated(kvm); |
4731d4c7 | 1403 | return ret; |
a436036b AK |
1404 | } |
1405 | ||
82ce2c96 IE |
1406 | /* |
1407 | * Changing the number of mmu pages allocated to the vm | |
1408 | * Note: if kvm_nr_mmu_pages is too small, you will get dead lock | |
1409 | */ | |
1410 | void kvm_mmu_change_mmu_pages(struct kvm *kvm, unsigned int kvm_nr_mmu_pages) | |
1411 | { | |
1412 | /* | |
1413 | * If we set the number of mmu pages to be smaller be than the | |
1414 | * number of actived pages , we must to free some mmu pages before we | |
1415 | * change the value | |
1416 | */ | |
1417 | ||
f05e70ac | 1418 | if ((kvm->arch.n_alloc_mmu_pages - kvm->arch.n_free_mmu_pages) > |
82ce2c96 | 1419 | kvm_nr_mmu_pages) { |
f05e70ac ZX |
1420 | int n_used_mmu_pages = kvm->arch.n_alloc_mmu_pages |
1421 | - kvm->arch.n_free_mmu_pages; | |
82ce2c96 IE |
1422 | |
1423 | while (n_used_mmu_pages > kvm_nr_mmu_pages) { | |
1424 | struct kvm_mmu_page *page; | |
1425 | ||
f05e70ac | 1426 | page = container_of(kvm->arch.active_mmu_pages.prev, |
82ce2c96 IE |
1427 | struct kvm_mmu_page, link); |
1428 | kvm_mmu_zap_page(kvm, page); | |
1429 | n_used_mmu_pages--; | |
1430 | } | |
f05e70ac | 1431 | kvm->arch.n_free_mmu_pages = 0; |
82ce2c96 IE |
1432 | } |
1433 | else | |
f05e70ac ZX |
1434 | kvm->arch.n_free_mmu_pages += kvm_nr_mmu_pages |
1435 | - kvm->arch.n_alloc_mmu_pages; | |
82ce2c96 | 1436 | |
f05e70ac | 1437 | kvm->arch.n_alloc_mmu_pages = kvm_nr_mmu_pages; |
82ce2c96 IE |
1438 | } |
1439 | ||
f67a46f4 | 1440 | static int kvm_mmu_unprotect_page(struct kvm *kvm, gfn_t gfn) |
a436036b AK |
1441 | { |
1442 | unsigned index; | |
1443 | struct hlist_head *bucket; | |
4db35314 | 1444 | struct kvm_mmu_page *sp; |
a436036b AK |
1445 | struct hlist_node *node, *n; |
1446 | int r; | |
1447 | ||
b8688d51 | 1448 | pgprintk("%s: looking for gfn %lx\n", __func__, gfn); |
a436036b | 1449 | r = 0; |
1ae0a13d | 1450 | index = kvm_page_table_hashfn(gfn); |
f05e70ac | 1451 | bucket = &kvm->arch.mmu_page_hash[index]; |
4db35314 AK |
1452 | hlist_for_each_entry_safe(sp, node, n, bucket, hash_link) |
1453 | if (sp->gfn == gfn && !sp->role.metaphysical) { | |
b8688d51 | 1454 | pgprintk("%s: gfn %lx role %x\n", __func__, gfn, |
4db35314 | 1455 | sp->role.word); |
a436036b | 1456 | r = 1; |
07385413 MT |
1457 | if (kvm_mmu_zap_page(kvm, sp)) |
1458 | n = bucket->first; | |
a436036b AK |
1459 | } |
1460 | return r; | |
cea0f0e7 AK |
1461 | } |
1462 | ||
f67a46f4 | 1463 | static void mmu_unshadow(struct kvm *kvm, gfn_t gfn) |
97a0a01e | 1464 | { |
4db35314 | 1465 | struct kvm_mmu_page *sp; |
97a0a01e | 1466 | |
4db35314 | 1467 | while ((sp = kvm_mmu_lookup_page(kvm, gfn)) != NULL) { |
b8688d51 | 1468 | pgprintk("%s: zap %lx %x\n", __func__, gfn, sp->role.word); |
4db35314 | 1469 | kvm_mmu_zap_page(kvm, sp); |
97a0a01e AK |
1470 | } |
1471 | } | |
1472 | ||
38c335f1 | 1473 | static void page_header_update_slot(struct kvm *kvm, void *pte, gfn_t gfn) |
6aa8b732 | 1474 | { |
38c335f1 | 1475 | int slot = memslot_id(kvm, gfn_to_memslot(kvm, gfn)); |
4db35314 | 1476 | struct kvm_mmu_page *sp = page_header(__pa(pte)); |
6aa8b732 | 1477 | |
291f26bc | 1478 | __set_bit(slot, sp->slot_bitmap); |
6aa8b732 AK |
1479 | } |
1480 | ||
6844dec6 MT |
1481 | static void mmu_convert_notrap(struct kvm_mmu_page *sp) |
1482 | { | |
1483 | int i; | |
1484 | u64 *pt = sp->spt; | |
1485 | ||
1486 | if (shadow_trap_nonpresent_pte == shadow_notrap_nonpresent_pte) | |
1487 | return; | |
1488 | ||
1489 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) { | |
1490 | if (pt[i] == shadow_notrap_nonpresent_pte) | |
1491 | set_shadow_pte(&pt[i], shadow_trap_nonpresent_pte); | |
1492 | } | |
1493 | } | |
1494 | ||
039576c0 AK |
1495 | struct page *gva_to_page(struct kvm_vcpu *vcpu, gva_t gva) |
1496 | { | |
72dc67a6 IE |
1497 | struct page *page; |
1498 | ||
ad312c7c | 1499 | gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gva); |
039576c0 AK |
1500 | |
1501 | if (gpa == UNMAPPED_GVA) | |
1502 | return NULL; | |
72dc67a6 | 1503 | |
72dc67a6 | 1504 | page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT); |
72dc67a6 IE |
1505 | |
1506 | return page; | |
039576c0 AK |
1507 | } |
1508 | ||
74be52e3 SY |
1509 | /* |
1510 | * The function is based on mtrr_type_lookup() in | |
1511 | * arch/x86/kernel/cpu/mtrr/generic.c | |
1512 | */ | |
1513 | static int get_mtrr_type(struct mtrr_state_type *mtrr_state, | |
1514 | u64 start, u64 end) | |
1515 | { | |
1516 | int i; | |
1517 | u64 base, mask; | |
1518 | u8 prev_match, curr_match; | |
1519 | int num_var_ranges = KVM_NR_VAR_MTRR; | |
1520 | ||
1521 | if (!mtrr_state->enabled) | |
1522 | return 0xFF; | |
1523 | ||
1524 | /* Make end inclusive end, instead of exclusive */ | |
1525 | end--; | |
1526 | ||
1527 | /* Look in fixed ranges. Just return the type as per start */ | |
1528 | if (mtrr_state->have_fixed && (start < 0x100000)) { | |
1529 | int idx; | |
1530 | ||
1531 | if (start < 0x80000) { | |
1532 | idx = 0; | |
1533 | idx += (start >> 16); | |
1534 | return mtrr_state->fixed_ranges[idx]; | |
1535 | } else if (start < 0xC0000) { | |
1536 | idx = 1 * 8; | |
1537 | idx += ((start - 0x80000) >> 14); | |
1538 | return mtrr_state->fixed_ranges[idx]; | |
1539 | } else if (start < 0x1000000) { | |
1540 | idx = 3 * 8; | |
1541 | idx += ((start - 0xC0000) >> 12); | |
1542 | return mtrr_state->fixed_ranges[idx]; | |
1543 | } | |
1544 | } | |
1545 | ||
1546 | /* | |
1547 | * Look in variable ranges | |
1548 | * Look of multiple ranges matching this address and pick type | |
1549 | * as per MTRR precedence | |
1550 | */ | |
1551 | if (!(mtrr_state->enabled & 2)) | |
1552 | return mtrr_state->def_type; | |
1553 | ||
1554 | prev_match = 0xFF; | |
1555 | for (i = 0; i < num_var_ranges; ++i) { | |
1556 | unsigned short start_state, end_state; | |
1557 | ||
1558 | if (!(mtrr_state->var_ranges[i].mask_lo & (1 << 11))) | |
1559 | continue; | |
1560 | ||
1561 | base = (((u64)mtrr_state->var_ranges[i].base_hi) << 32) + | |
1562 | (mtrr_state->var_ranges[i].base_lo & PAGE_MASK); | |
1563 | mask = (((u64)mtrr_state->var_ranges[i].mask_hi) << 32) + | |
1564 | (mtrr_state->var_ranges[i].mask_lo & PAGE_MASK); | |
1565 | ||
1566 | start_state = ((start & mask) == (base & mask)); | |
1567 | end_state = ((end & mask) == (base & mask)); | |
1568 | if (start_state != end_state) | |
1569 | return 0xFE; | |
1570 | ||
1571 | if ((start & mask) != (base & mask)) | |
1572 | continue; | |
1573 | ||
1574 | curr_match = mtrr_state->var_ranges[i].base_lo & 0xff; | |
1575 | if (prev_match == 0xFF) { | |
1576 | prev_match = curr_match; | |
1577 | continue; | |
1578 | } | |
1579 | ||
1580 | if (prev_match == MTRR_TYPE_UNCACHABLE || | |
1581 | curr_match == MTRR_TYPE_UNCACHABLE) | |
1582 | return MTRR_TYPE_UNCACHABLE; | |
1583 | ||
1584 | if ((prev_match == MTRR_TYPE_WRBACK && | |
1585 | curr_match == MTRR_TYPE_WRTHROUGH) || | |
1586 | (prev_match == MTRR_TYPE_WRTHROUGH && | |
1587 | curr_match == MTRR_TYPE_WRBACK)) { | |
1588 | prev_match = MTRR_TYPE_WRTHROUGH; | |
1589 | curr_match = MTRR_TYPE_WRTHROUGH; | |
1590 | } | |
1591 | ||
1592 | if (prev_match != curr_match) | |
1593 | return MTRR_TYPE_UNCACHABLE; | |
1594 | } | |
1595 | ||
1596 | if (prev_match != 0xFF) | |
1597 | return prev_match; | |
1598 | ||
1599 | return mtrr_state->def_type; | |
1600 | } | |
1601 | ||
1602 | static u8 get_memory_type(struct kvm_vcpu *vcpu, gfn_t gfn) | |
1603 | { | |
1604 | u8 mtrr; | |
1605 | ||
1606 | mtrr = get_mtrr_type(&vcpu->arch.mtrr_state, gfn << PAGE_SHIFT, | |
1607 | (gfn << PAGE_SHIFT) + PAGE_SIZE); | |
1608 | if (mtrr == 0xfe || mtrr == 0xff) | |
1609 | mtrr = MTRR_TYPE_WRBACK; | |
1610 | return mtrr; | |
1611 | } | |
1612 | ||
4731d4c7 MT |
1613 | static int kvm_unsync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp) |
1614 | { | |
1615 | unsigned index; | |
1616 | struct hlist_head *bucket; | |
1617 | struct kvm_mmu_page *s; | |
1618 | struct hlist_node *node, *n; | |
1619 | ||
1620 | index = kvm_page_table_hashfn(sp->gfn); | |
1621 | bucket = &vcpu->kvm->arch.mmu_page_hash[index]; | |
1622 | /* don't unsync if pagetable is shadowed with multiple roles */ | |
1623 | hlist_for_each_entry_safe(s, node, n, bucket, hash_link) { | |
1624 | if (s->gfn != sp->gfn || s->role.metaphysical) | |
1625 | continue; | |
1626 | if (s->role.word != sp->role.word) | |
1627 | return 1; | |
1628 | } | |
4731d4c7 MT |
1629 | ++vcpu->kvm->stat.mmu_unsync; |
1630 | sp->unsync = 1; | |
6cffe8ca MT |
1631 | |
1632 | if (sp->global) { | |
1633 | list_add(&sp->oos_link, &vcpu->kvm->arch.oos_global_pages); | |
1634 | ++vcpu->kvm->stat.mmu_unsync_global; | |
1635 | } else | |
1636 | kvm_mmu_mark_parents_unsync(vcpu, sp); | |
1637 | ||
4731d4c7 MT |
1638 | mmu_convert_notrap(sp); |
1639 | return 0; | |
1640 | } | |
1641 | ||
1642 | static int mmu_need_write_protect(struct kvm_vcpu *vcpu, gfn_t gfn, | |
1643 | bool can_unsync) | |
1644 | { | |
1645 | struct kvm_mmu_page *shadow; | |
1646 | ||
1647 | shadow = kvm_mmu_lookup_page(vcpu->kvm, gfn); | |
1648 | if (shadow) { | |
1649 | if (shadow->role.level != PT_PAGE_TABLE_LEVEL) | |
1650 | return 1; | |
1651 | if (shadow->unsync) | |
1652 | return 0; | |
582801a9 | 1653 | if (can_unsync && oos_shadow) |
4731d4c7 MT |
1654 | return kvm_unsync_page(vcpu, shadow); |
1655 | return 1; | |
1656 | } | |
1657 | return 0; | |
1658 | } | |
1659 | ||
1e73f9dd MT |
1660 | static int set_spte(struct kvm_vcpu *vcpu, u64 *shadow_pte, |
1661 | unsigned pte_access, int user_fault, | |
1662 | int write_fault, int dirty, int largepage, | |
6cffe8ca | 1663 | int global, gfn_t gfn, pfn_t pfn, bool speculative, |
4731d4c7 | 1664 | bool can_unsync) |
1c4f1fd6 AK |
1665 | { |
1666 | u64 spte; | |
1e73f9dd | 1667 | int ret = 0; |
64d4d521 | 1668 | u64 mt_mask = shadow_mt_mask; |
6cffe8ca MT |
1669 | struct kvm_mmu_page *sp = page_header(__pa(shadow_pte)); |
1670 | ||
1671 | if (!global && sp->global) { | |
1672 | sp->global = 0; | |
1673 | if (sp->unsync) { | |
1674 | kvm_unlink_unsync_global(vcpu->kvm, sp); | |
1675 | kvm_mmu_mark_parents_unsync(vcpu, sp); | |
1676 | } | |
1677 | } | |
64d4d521 | 1678 | |
1c4f1fd6 AK |
1679 | /* |
1680 | * We don't set the accessed bit, since we sometimes want to see | |
1681 | * whether the guest actually used the pte (in order to detect | |
1682 | * demand paging). | |
1683 | */ | |
7b52345e | 1684 | spte = shadow_base_present_pte | shadow_dirty_mask; |
947da538 | 1685 | if (!speculative) |
3201b5d9 | 1686 | spte |= shadow_accessed_mask; |
1c4f1fd6 AK |
1687 | if (!dirty) |
1688 | pte_access &= ~ACC_WRITE_MASK; | |
7b52345e SY |
1689 | if (pte_access & ACC_EXEC_MASK) |
1690 | spte |= shadow_x_mask; | |
1691 | else | |
1692 | spte |= shadow_nx_mask; | |
1c4f1fd6 | 1693 | if (pte_access & ACC_USER_MASK) |
7b52345e | 1694 | spte |= shadow_user_mask; |
05da4558 MT |
1695 | if (largepage) |
1696 | spte |= PT_PAGE_SIZE_MASK; | |
64d4d521 | 1697 | if (mt_mask) { |
2aaf69dc SY |
1698 | if (!kvm_is_mmio_pfn(pfn)) { |
1699 | mt_mask = get_memory_type(vcpu, gfn) << | |
1700 | kvm_x86_ops->get_mt_mask_shift(); | |
1701 | mt_mask |= VMX_EPT_IGMT_BIT; | |
1702 | } else | |
1703 | mt_mask = MTRR_TYPE_UNCACHABLE << | |
1704 | kvm_x86_ops->get_mt_mask_shift(); | |
64d4d521 SY |
1705 | spte |= mt_mask; |
1706 | } | |
1c4f1fd6 | 1707 | |
35149e21 | 1708 | spte |= (u64)pfn << PAGE_SHIFT; |
1c4f1fd6 AK |
1709 | |
1710 | if ((pte_access & ACC_WRITE_MASK) | |
1711 | || (write_fault && !is_write_protection(vcpu) && !user_fault)) { | |
1c4f1fd6 | 1712 | |
38187c83 MT |
1713 | if (largepage && has_wrprotected_page(vcpu->kvm, gfn)) { |
1714 | ret = 1; | |
1715 | spte = shadow_trap_nonpresent_pte; | |
1716 | goto set_pte; | |
1717 | } | |
1718 | ||
1c4f1fd6 | 1719 | spte |= PT_WRITABLE_MASK; |
1c4f1fd6 | 1720 | |
ecc5589f MT |
1721 | /* |
1722 | * Optimization: for pte sync, if spte was writable the hash | |
1723 | * lookup is unnecessary (and expensive). Write protection | |
1724 | * is responsibility of mmu_get_page / kvm_sync_page. | |
1725 | * Same reasoning can be applied to dirty page accounting. | |
1726 | */ | |
1727 | if (!can_unsync && is_writeble_pte(*shadow_pte)) | |
1728 | goto set_pte; | |
1729 | ||
4731d4c7 | 1730 | if (mmu_need_write_protect(vcpu, gfn, can_unsync)) { |
1c4f1fd6 | 1731 | pgprintk("%s: found shadow page for %lx, marking ro\n", |
b8688d51 | 1732 | __func__, gfn); |
1e73f9dd | 1733 | ret = 1; |
1c4f1fd6 | 1734 | pte_access &= ~ACC_WRITE_MASK; |
a378b4e6 | 1735 | if (is_writeble_pte(spte)) |
1c4f1fd6 | 1736 | spte &= ~PT_WRITABLE_MASK; |
1c4f1fd6 AK |
1737 | } |
1738 | } | |
1739 | ||
1c4f1fd6 AK |
1740 | if (pte_access & ACC_WRITE_MASK) |
1741 | mark_page_dirty(vcpu->kvm, gfn); | |
1742 | ||
38187c83 | 1743 | set_pte: |
1c4f1fd6 | 1744 | set_shadow_pte(shadow_pte, spte); |
1e73f9dd MT |
1745 | return ret; |
1746 | } | |
1747 | ||
1e73f9dd MT |
1748 | static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *shadow_pte, |
1749 | unsigned pt_access, unsigned pte_access, | |
1750 | int user_fault, int write_fault, int dirty, | |
6cffe8ca MT |
1751 | int *ptwrite, int largepage, int global, |
1752 | gfn_t gfn, pfn_t pfn, bool speculative) | |
1e73f9dd MT |
1753 | { |
1754 | int was_rmapped = 0; | |
1755 | int was_writeble = is_writeble_pte(*shadow_pte); | |
1756 | ||
1757 | pgprintk("%s: spte %llx access %x write_fault %d" | |
1758 | " user_fault %d gfn %lx\n", | |
1759 | __func__, *shadow_pte, pt_access, | |
1760 | write_fault, user_fault, gfn); | |
1761 | ||
1762 | if (is_rmap_pte(*shadow_pte)) { | |
1763 | /* | |
1764 | * If we overwrite a PTE page pointer with a 2MB PMD, unlink | |
1765 | * the parent of the now unreachable PTE. | |
1766 | */ | |
1767 | if (largepage && !is_large_pte(*shadow_pte)) { | |
1768 | struct kvm_mmu_page *child; | |
1769 | u64 pte = *shadow_pte; | |
1770 | ||
1771 | child = page_header(pte & PT64_BASE_ADDR_MASK); | |
1772 | mmu_page_remove_parent_pte(child, shadow_pte); | |
1773 | } else if (pfn != spte_to_pfn(*shadow_pte)) { | |
1774 | pgprintk("hfn old %lx new %lx\n", | |
1775 | spte_to_pfn(*shadow_pte), pfn); | |
1776 | rmap_remove(vcpu->kvm, shadow_pte); | |
1777 | } else { | |
1778 | if (largepage) | |
1779 | was_rmapped = is_large_pte(*shadow_pte); | |
1780 | else | |
1781 | was_rmapped = 1; | |
1782 | } | |
1783 | } | |
1784 | if (set_spte(vcpu, shadow_pte, pte_access, user_fault, write_fault, | |
6cffe8ca | 1785 | dirty, largepage, global, gfn, pfn, speculative, true)) { |
1e73f9dd MT |
1786 | if (write_fault) |
1787 | *ptwrite = 1; | |
a378b4e6 MT |
1788 | kvm_x86_ops->tlb_flush(vcpu); |
1789 | } | |
1e73f9dd MT |
1790 | |
1791 | pgprintk("%s: setting spte %llx\n", __func__, *shadow_pte); | |
1792 | pgprintk("instantiating %s PTE (%s) at %ld (%llx) addr %p\n", | |
1793 | is_large_pte(*shadow_pte)? "2MB" : "4kB", | |
1794 | is_present_pte(*shadow_pte)?"RW":"R", gfn, | |
1795 | *shadow_pte, shadow_pte); | |
1796 | if (!was_rmapped && is_large_pte(*shadow_pte)) | |
05da4558 MT |
1797 | ++vcpu->kvm->stat.lpages; |
1798 | ||
1c4f1fd6 AK |
1799 | page_header_update_slot(vcpu->kvm, shadow_pte, gfn); |
1800 | if (!was_rmapped) { | |
05da4558 | 1801 | rmap_add(vcpu, shadow_pte, gfn, largepage); |
1c4f1fd6 | 1802 | if (!is_rmap_pte(*shadow_pte)) |
35149e21 | 1803 | kvm_release_pfn_clean(pfn); |
75e68e60 IE |
1804 | } else { |
1805 | if (was_writeble) | |
35149e21 | 1806 | kvm_release_pfn_dirty(pfn); |
75e68e60 | 1807 | else |
35149e21 | 1808 | kvm_release_pfn_clean(pfn); |
1c4f1fd6 | 1809 | } |
1b7fcd32 | 1810 | if (speculative) { |
ad312c7c | 1811 | vcpu->arch.last_pte_updated = shadow_pte; |
1b7fcd32 AK |
1812 | vcpu->arch.last_pte_gfn = gfn; |
1813 | } | |
1c4f1fd6 AK |
1814 | } |
1815 | ||
6aa8b732 AK |
1816 | static void nonpaging_new_cr3(struct kvm_vcpu *vcpu) |
1817 | { | |
1818 | } | |
1819 | ||
140754bc AK |
1820 | struct direct_shadow_walk { |
1821 | struct kvm_shadow_walk walker; | |
1822 | pfn_t pfn; | |
1823 | int write; | |
1824 | int largepage; | |
1825 | int pt_write; | |
1826 | }; | |
6aa8b732 | 1827 | |
140754bc AK |
1828 | static int direct_map_entry(struct kvm_shadow_walk *_walk, |
1829 | struct kvm_vcpu *vcpu, | |
d40a1ee4 | 1830 | u64 addr, u64 *sptep, int level) |
140754bc AK |
1831 | { |
1832 | struct direct_shadow_walk *walk = | |
1833 | container_of(_walk, struct direct_shadow_walk, walker); | |
1834 | struct kvm_mmu_page *sp; | |
1835 | gfn_t pseudo_gfn; | |
1836 | gfn_t gfn = addr >> PAGE_SHIFT; | |
1837 | ||
1838 | if (level == PT_PAGE_TABLE_LEVEL | |
1839 | || (walk->largepage && level == PT_DIRECTORY_LEVEL)) { | |
1840 | mmu_set_spte(vcpu, sptep, ACC_ALL, ACC_ALL, | |
1841 | 0, walk->write, 1, &walk->pt_write, | |
6cffe8ca | 1842 | walk->largepage, 0, gfn, walk->pfn, false); |
bc2d4299 | 1843 | ++vcpu->stat.pf_fixed; |
140754bc AK |
1844 | return 1; |
1845 | } | |
6aa8b732 | 1846 | |
140754bc AK |
1847 | if (*sptep == shadow_trap_nonpresent_pte) { |
1848 | pseudo_gfn = (addr & PT64_DIR_BASE_ADDR_MASK) >> PAGE_SHIFT; | |
d40a1ee4 | 1849 | sp = kvm_mmu_get_page(vcpu, pseudo_gfn, (gva_t)addr, level - 1, |
140754bc AK |
1850 | 1, ACC_ALL, sptep); |
1851 | if (!sp) { | |
1852 | pgprintk("nonpaging_map: ENOMEM\n"); | |
1853 | kvm_release_pfn_clean(walk->pfn); | |
1854 | return -ENOMEM; | |
6aa8b732 AK |
1855 | } |
1856 | ||
140754bc AK |
1857 | set_shadow_pte(sptep, |
1858 | __pa(sp->spt) | |
1859 | | PT_PRESENT_MASK | PT_WRITABLE_MASK | |
1860 | | shadow_user_mask | shadow_x_mask); | |
6aa8b732 | 1861 | } |
140754bc AK |
1862 | return 0; |
1863 | } | |
1864 | ||
1865 | static int __direct_map(struct kvm_vcpu *vcpu, gpa_t v, int write, | |
1866 | int largepage, gfn_t gfn, pfn_t pfn) | |
1867 | { | |
1868 | int r; | |
1869 | struct direct_shadow_walk walker = { | |
1870 | .walker = { .entry = direct_map_entry, }, | |
1871 | .pfn = pfn, | |
1872 | .largepage = largepage, | |
1873 | .write = write, | |
1874 | .pt_write = 0, | |
1875 | }; | |
1876 | ||
d40a1ee4 | 1877 | r = walk_shadow(&walker.walker, vcpu, gfn << PAGE_SHIFT); |
140754bc AK |
1878 | if (r < 0) |
1879 | return r; | |
1880 | return walker.pt_write; | |
6aa8b732 AK |
1881 | } |
1882 | ||
10589a46 MT |
1883 | static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, int write, gfn_t gfn) |
1884 | { | |
1885 | int r; | |
05da4558 | 1886 | int largepage = 0; |
35149e21 | 1887 | pfn_t pfn; |
e930bffe | 1888 | unsigned long mmu_seq; |
aaee2c94 | 1889 | |
05da4558 MT |
1890 | if (is_largepage_backed(vcpu, gfn & ~(KVM_PAGES_PER_HPAGE-1))) { |
1891 | gfn &= ~(KVM_PAGES_PER_HPAGE-1); | |
1892 | largepage = 1; | |
1893 | } | |
1894 | ||
e930bffe | 1895 | mmu_seq = vcpu->kvm->mmu_notifier_seq; |
4c2155ce | 1896 | smp_rmb(); |
35149e21 | 1897 | pfn = gfn_to_pfn(vcpu->kvm, gfn); |
aaee2c94 | 1898 | |
d196e343 | 1899 | /* mmio */ |
35149e21 AL |
1900 | if (is_error_pfn(pfn)) { |
1901 | kvm_release_pfn_clean(pfn); | |
d196e343 AK |
1902 | return 1; |
1903 | } | |
1904 | ||
aaee2c94 | 1905 | spin_lock(&vcpu->kvm->mmu_lock); |
e930bffe AA |
1906 | if (mmu_notifier_retry(vcpu, mmu_seq)) |
1907 | goto out_unlock; | |
eb787d10 | 1908 | kvm_mmu_free_some_pages(vcpu); |
6c41f428 | 1909 | r = __direct_map(vcpu, v, write, largepage, gfn, pfn); |
aaee2c94 MT |
1910 | spin_unlock(&vcpu->kvm->mmu_lock); |
1911 | ||
aaee2c94 | 1912 | |
10589a46 | 1913 | return r; |
e930bffe AA |
1914 | |
1915 | out_unlock: | |
1916 | spin_unlock(&vcpu->kvm->mmu_lock); | |
1917 | kvm_release_pfn_clean(pfn); | |
1918 | return 0; | |
10589a46 MT |
1919 | } |
1920 | ||
1921 | ||
17ac10ad AK |
1922 | static void mmu_free_roots(struct kvm_vcpu *vcpu) |
1923 | { | |
1924 | int i; | |
4db35314 | 1925 | struct kvm_mmu_page *sp; |
17ac10ad | 1926 | |
ad312c7c | 1927 | if (!VALID_PAGE(vcpu->arch.mmu.root_hpa)) |
7b53aa56 | 1928 | return; |
aaee2c94 | 1929 | spin_lock(&vcpu->kvm->mmu_lock); |
ad312c7c ZX |
1930 | if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) { |
1931 | hpa_t root = vcpu->arch.mmu.root_hpa; | |
17ac10ad | 1932 | |
4db35314 AK |
1933 | sp = page_header(root); |
1934 | --sp->root_count; | |
2e53d63a MT |
1935 | if (!sp->root_count && sp->role.invalid) |
1936 | kvm_mmu_zap_page(vcpu->kvm, sp); | |
ad312c7c | 1937 | vcpu->arch.mmu.root_hpa = INVALID_PAGE; |
aaee2c94 | 1938 | spin_unlock(&vcpu->kvm->mmu_lock); |
17ac10ad AK |
1939 | return; |
1940 | } | |
17ac10ad | 1941 | for (i = 0; i < 4; ++i) { |
ad312c7c | 1942 | hpa_t root = vcpu->arch.mmu.pae_root[i]; |
17ac10ad | 1943 | |
417726a3 | 1944 | if (root) { |
417726a3 | 1945 | root &= PT64_BASE_ADDR_MASK; |
4db35314 AK |
1946 | sp = page_header(root); |
1947 | --sp->root_count; | |
2e53d63a MT |
1948 | if (!sp->root_count && sp->role.invalid) |
1949 | kvm_mmu_zap_page(vcpu->kvm, sp); | |
417726a3 | 1950 | } |
ad312c7c | 1951 | vcpu->arch.mmu.pae_root[i] = INVALID_PAGE; |
17ac10ad | 1952 | } |
aaee2c94 | 1953 | spin_unlock(&vcpu->kvm->mmu_lock); |
ad312c7c | 1954 | vcpu->arch.mmu.root_hpa = INVALID_PAGE; |
17ac10ad AK |
1955 | } |
1956 | ||
1957 | static void mmu_alloc_roots(struct kvm_vcpu *vcpu) | |
1958 | { | |
1959 | int i; | |
cea0f0e7 | 1960 | gfn_t root_gfn; |
4db35314 | 1961 | struct kvm_mmu_page *sp; |
fb72d167 | 1962 | int metaphysical = 0; |
3bb65a22 | 1963 | |
ad312c7c | 1964 | root_gfn = vcpu->arch.cr3 >> PAGE_SHIFT; |
17ac10ad | 1965 | |
ad312c7c ZX |
1966 | if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) { |
1967 | hpa_t root = vcpu->arch.mmu.root_hpa; | |
17ac10ad AK |
1968 | |
1969 | ASSERT(!VALID_PAGE(root)); | |
fb72d167 JR |
1970 | if (tdp_enabled) |
1971 | metaphysical = 1; | |
4db35314 | 1972 | sp = kvm_mmu_get_page(vcpu, root_gfn, 0, |
fb72d167 JR |
1973 | PT64_ROOT_LEVEL, metaphysical, |
1974 | ACC_ALL, NULL); | |
4db35314 AK |
1975 | root = __pa(sp->spt); |
1976 | ++sp->root_count; | |
ad312c7c | 1977 | vcpu->arch.mmu.root_hpa = root; |
17ac10ad AK |
1978 | return; |
1979 | } | |
fb72d167 JR |
1980 | metaphysical = !is_paging(vcpu); |
1981 | if (tdp_enabled) | |
1982 | metaphysical = 1; | |
17ac10ad | 1983 | for (i = 0; i < 4; ++i) { |
ad312c7c | 1984 | hpa_t root = vcpu->arch.mmu.pae_root[i]; |
17ac10ad AK |
1985 | |
1986 | ASSERT(!VALID_PAGE(root)); | |
ad312c7c ZX |
1987 | if (vcpu->arch.mmu.root_level == PT32E_ROOT_LEVEL) { |
1988 | if (!is_present_pte(vcpu->arch.pdptrs[i])) { | |
1989 | vcpu->arch.mmu.pae_root[i] = 0; | |
417726a3 AK |
1990 | continue; |
1991 | } | |
ad312c7c ZX |
1992 | root_gfn = vcpu->arch.pdptrs[i] >> PAGE_SHIFT; |
1993 | } else if (vcpu->arch.mmu.root_level == 0) | |
cea0f0e7 | 1994 | root_gfn = 0; |
4db35314 | 1995 | sp = kvm_mmu_get_page(vcpu, root_gfn, i << 30, |
fb72d167 | 1996 | PT32_ROOT_LEVEL, metaphysical, |
f7d9c7b7 | 1997 | ACC_ALL, NULL); |
4db35314 AK |
1998 | root = __pa(sp->spt); |
1999 | ++sp->root_count; | |
ad312c7c | 2000 | vcpu->arch.mmu.pae_root[i] = root | PT_PRESENT_MASK; |
17ac10ad | 2001 | } |
ad312c7c | 2002 | vcpu->arch.mmu.root_hpa = __pa(vcpu->arch.mmu.pae_root); |
17ac10ad AK |
2003 | } |
2004 | ||
0ba73cda MT |
2005 | static void mmu_sync_roots(struct kvm_vcpu *vcpu) |
2006 | { | |
2007 | int i; | |
2008 | struct kvm_mmu_page *sp; | |
2009 | ||
2010 | if (!VALID_PAGE(vcpu->arch.mmu.root_hpa)) | |
2011 | return; | |
2012 | if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) { | |
2013 | hpa_t root = vcpu->arch.mmu.root_hpa; | |
2014 | sp = page_header(root); | |
2015 | mmu_sync_children(vcpu, sp); | |
2016 | return; | |
2017 | } | |
2018 | for (i = 0; i < 4; ++i) { | |
2019 | hpa_t root = vcpu->arch.mmu.pae_root[i]; | |
2020 | ||
2021 | if (root) { | |
2022 | root &= PT64_BASE_ADDR_MASK; | |
2023 | sp = page_header(root); | |
2024 | mmu_sync_children(vcpu, sp); | |
2025 | } | |
2026 | } | |
2027 | } | |
2028 | ||
6cffe8ca MT |
2029 | static void mmu_sync_global(struct kvm_vcpu *vcpu) |
2030 | { | |
2031 | struct kvm *kvm = vcpu->kvm; | |
2032 | struct kvm_mmu_page *sp, *n; | |
2033 | ||
2034 | list_for_each_entry_safe(sp, n, &kvm->arch.oos_global_pages, oos_link) | |
2035 | kvm_sync_page(vcpu, sp); | |
2036 | } | |
2037 | ||
0ba73cda MT |
2038 | void kvm_mmu_sync_roots(struct kvm_vcpu *vcpu) |
2039 | { | |
2040 | spin_lock(&vcpu->kvm->mmu_lock); | |
2041 | mmu_sync_roots(vcpu); | |
6cffe8ca MT |
2042 | spin_unlock(&vcpu->kvm->mmu_lock); |
2043 | } | |
2044 | ||
2045 | void kvm_mmu_sync_global(struct kvm_vcpu *vcpu) | |
2046 | { | |
2047 | spin_lock(&vcpu->kvm->mmu_lock); | |
2048 | mmu_sync_global(vcpu); | |
0ba73cda MT |
2049 | spin_unlock(&vcpu->kvm->mmu_lock); |
2050 | } | |
2051 | ||
6aa8b732 AK |
2052 | static gpa_t nonpaging_gva_to_gpa(struct kvm_vcpu *vcpu, gva_t vaddr) |
2053 | { | |
2054 | return vaddr; | |
2055 | } | |
2056 | ||
2057 | static int nonpaging_page_fault(struct kvm_vcpu *vcpu, gva_t gva, | |
3f3e7124 | 2058 | u32 error_code) |
6aa8b732 | 2059 | { |
e833240f | 2060 | gfn_t gfn; |
e2dec939 | 2061 | int r; |
6aa8b732 | 2062 | |
b8688d51 | 2063 | pgprintk("%s: gva %lx error %x\n", __func__, gva, error_code); |
e2dec939 AK |
2064 | r = mmu_topup_memory_caches(vcpu); |
2065 | if (r) | |
2066 | return r; | |
714b93da | 2067 | |
6aa8b732 | 2068 | ASSERT(vcpu); |
ad312c7c | 2069 | ASSERT(VALID_PAGE(vcpu->arch.mmu.root_hpa)); |
6aa8b732 | 2070 | |
e833240f | 2071 | gfn = gva >> PAGE_SHIFT; |
6aa8b732 | 2072 | |
e833240f AK |
2073 | return nonpaging_map(vcpu, gva & PAGE_MASK, |
2074 | error_code & PFERR_WRITE_MASK, gfn); | |
6aa8b732 AK |
2075 | } |
2076 | ||
fb72d167 JR |
2077 | static int tdp_page_fault(struct kvm_vcpu *vcpu, gva_t gpa, |
2078 | u32 error_code) | |
2079 | { | |
35149e21 | 2080 | pfn_t pfn; |
fb72d167 | 2081 | int r; |
05da4558 MT |
2082 | int largepage = 0; |
2083 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
e930bffe | 2084 | unsigned long mmu_seq; |
fb72d167 JR |
2085 | |
2086 | ASSERT(vcpu); | |
2087 | ASSERT(VALID_PAGE(vcpu->arch.mmu.root_hpa)); | |
2088 | ||
2089 | r = mmu_topup_memory_caches(vcpu); | |
2090 | if (r) | |
2091 | return r; | |
2092 | ||
05da4558 MT |
2093 | if (is_largepage_backed(vcpu, gfn & ~(KVM_PAGES_PER_HPAGE-1))) { |
2094 | gfn &= ~(KVM_PAGES_PER_HPAGE-1); | |
2095 | largepage = 1; | |
2096 | } | |
e930bffe | 2097 | mmu_seq = vcpu->kvm->mmu_notifier_seq; |
4c2155ce | 2098 | smp_rmb(); |
35149e21 | 2099 | pfn = gfn_to_pfn(vcpu->kvm, gfn); |
35149e21 AL |
2100 | if (is_error_pfn(pfn)) { |
2101 | kvm_release_pfn_clean(pfn); | |
fb72d167 JR |
2102 | return 1; |
2103 | } | |
2104 | spin_lock(&vcpu->kvm->mmu_lock); | |
e930bffe AA |
2105 | if (mmu_notifier_retry(vcpu, mmu_seq)) |
2106 | goto out_unlock; | |
fb72d167 JR |
2107 | kvm_mmu_free_some_pages(vcpu); |
2108 | r = __direct_map(vcpu, gpa, error_code & PFERR_WRITE_MASK, | |
6c41f428 | 2109 | largepage, gfn, pfn); |
fb72d167 | 2110 | spin_unlock(&vcpu->kvm->mmu_lock); |
fb72d167 JR |
2111 | |
2112 | return r; | |
e930bffe AA |
2113 | |
2114 | out_unlock: | |
2115 | spin_unlock(&vcpu->kvm->mmu_lock); | |
2116 | kvm_release_pfn_clean(pfn); | |
2117 | return 0; | |
fb72d167 JR |
2118 | } |
2119 | ||
6aa8b732 AK |
2120 | static void nonpaging_free(struct kvm_vcpu *vcpu) |
2121 | { | |
17ac10ad | 2122 | mmu_free_roots(vcpu); |
6aa8b732 AK |
2123 | } |
2124 | ||
2125 | static int nonpaging_init_context(struct kvm_vcpu *vcpu) | |
2126 | { | |
ad312c7c | 2127 | struct kvm_mmu *context = &vcpu->arch.mmu; |
6aa8b732 AK |
2128 | |
2129 | context->new_cr3 = nonpaging_new_cr3; | |
2130 | context->page_fault = nonpaging_page_fault; | |
6aa8b732 AK |
2131 | context->gva_to_gpa = nonpaging_gva_to_gpa; |
2132 | context->free = nonpaging_free; | |
c7addb90 | 2133 | context->prefetch_page = nonpaging_prefetch_page; |
e8bc217a | 2134 | context->sync_page = nonpaging_sync_page; |
a7052897 | 2135 | context->invlpg = nonpaging_invlpg; |
cea0f0e7 | 2136 | context->root_level = 0; |
6aa8b732 | 2137 | context->shadow_root_level = PT32E_ROOT_LEVEL; |
17c3ba9d | 2138 | context->root_hpa = INVALID_PAGE; |
6aa8b732 AK |
2139 | return 0; |
2140 | } | |
2141 | ||
d835dfec | 2142 | void kvm_mmu_flush_tlb(struct kvm_vcpu *vcpu) |
6aa8b732 | 2143 | { |
1165f5fe | 2144 | ++vcpu->stat.tlb_flush; |
cbdd1bea | 2145 | kvm_x86_ops->tlb_flush(vcpu); |
6aa8b732 AK |
2146 | } |
2147 | ||
2148 | static void paging_new_cr3(struct kvm_vcpu *vcpu) | |
2149 | { | |
b8688d51 | 2150 | pgprintk("%s: cr3 %lx\n", __func__, vcpu->arch.cr3); |
cea0f0e7 | 2151 | mmu_free_roots(vcpu); |
6aa8b732 AK |
2152 | } |
2153 | ||
6aa8b732 AK |
2154 | static void inject_page_fault(struct kvm_vcpu *vcpu, |
2155 | u64 addr, | |
2156 | u32 err_code) | |
2157 | { | |
c3c91fee | 2158 | kvm_inject_page_fault(vcpu, addr, err_code); |
6aa8b732 AK |
2159 | } |
2160 | ||
6aa8b732 AK |
2161 | static void paging_free(struct kvm_vcpu *vcpu) |
2162 | { | |
2163 | nonpaging_free(vcpu); | |
2164 | } | |
2165 | ||
2166 | #define PTTYPE 64 | |
2167 | #include "paging_tmpl.h" | |
2168 | #undef PTTYPE | |
2169 | ||
2170 | #define PTTYPE 32 | |
2171 | #include "paging_tmpl.h" | |
2172 | #undef PTTYPE | |
2173 | ||
17ac10ad | 2174 | static int paging64_init_context_common(struct kvm_vcpu *vcpu, int level) |
6aa8b732 | 2175 | { |
ad312c7c | 2176 | struct kvm_mmu *context = &vcpu->arch.mmu; |
6aa8b732 AK |
2177 | |
2178 | ASSERT(is_pae(vcpu)); | |
2179 | context->new_cr3 = paging_new_cr3; | |
2180 | context->page_fault = paging64_page_fault; | |
6aa8b732 | 2181 | context->gva_to_gpa = paging64_gva_to_gpa; |
c7addb90 | 2182 | context->prefetch_page = paging64_prefetch_page; |
e8bc217a | 2183 | context->sync_page = paging64_sync_page; |
a7052897 | 2184 | context->invlpg = paging64_invlpg; |
6aa8b732 | 2185 | context->free = paging_free; |
17ac10ad AK |
2186 | context->root_level = level; |
2187 | context->shadow_root_level = level; | |
17c3ba9d | 2188 | context->root_hpa = INVALID_PAGE; |
6aa8b732 AK |
2189 | return 0; |
2190 | } | |
2191 | ||
17ac10ad AK |
2192 | static int paging64_init_context(struct kvm_vcpu *vcpu) |
2193 | { | |
2194 | return paging64_init_context_common(vcpu, PT64_ROOT_LEVEL); | |
2195 | } | |
2196 | ||
6aa8b732 AK |
2197 | static int paging32_init_context(struct kvm_vcpu *vcpu) |
2198 | { | |
ad312c7c | 2199 | struct kvm_mmu *context = &vcpu->arch.mmu; |
6aa8b732 AK |
2200 | |
2201 | context->new_cr3 = paging_new_cr3; | |
2202 | context->page_fault = paging32_page_fault; | |
6aa8b732 AK |
2203 | context->gva_to_gpa = paging32_gva_to_gpa; |
2204 | context->free = paging_free; | |
c7addb90 | 2205 | context->prefetch_page = paging32_prefetch_page; |
e8bc217a | 2206 | context->sync_page = paging32_sync_page; |
a7052897 | 2207 | context->invlpg = paging32_invlpg; |
6aa8b732 AK |
2208 | context->root_level = PT32_ROOT_LEVEL; |
2209 | context->shadow_root_level = PT32E_ROOT_LEVEL; | |
17c3ba9d | 2210 | context->root_hpa = INVALID_PAGE; |
6aa8b732 AK |
2211 | return 0; |
2212 | } | |
2213 | ||
2214 | static int paging32E_init_context(struct kvm_vcpu *vcpu) | |
2215 | { | |
17ac10ad | 2216 | return paging64_init_context_common(vcpu, PT32E_ROOT_LEVEL); |
6aa8b732 AK |
2217 | } |
2218 | ||
fb72d167 JR |
2219 | static int init_kvm_tdp_mmu(struct kvm_vcpu *vcpu) |
2220 | { | |
2221 | struct kvm_mmu *context = &vcpu->arch.mmu; | |
2222 | ||
2223 | context->new_cr3 = nonpaging_new_cr3; | |
2224 | context->page_fault = tdp_page_fault; | |
2225 | context->free = nonpaging_free; | |
2226 | context->prefetch_page = nonpaging_prefetch_page; | |
e8bc217a | 2227 | context->sync_page = nonpaging_sync_page; |
a7052897 | 2228 | context->invlpg = nonpaging_invlpg; |
67253af5 | 2229 | context->shadow_root_level = kvm_x86_ops->get_tdp_level(); |
fb72d167 JR |
2230 | context->root_hpa = INVALID_PAGE; |
2231 | ||
2232 | if (!is_paging(vcpu)) { | |
2233 | context->gva_to_gpa = nonpaging_gva_to_gpa; | |
2234 | context->root_level = 0; | |
2235 | } else if (is_long_mode(vcpu)) { | |
2236 | context->gva_to_gpa = paging64_gva_to_gpa; | |
2237 | context->root_level = PT64_ROOT_LEVEL; | |
2238 | } else if (is_pae(vcpu)) { | |
2239 | context->gva_to_gpa = paging64_gva_to_gpa; | |
2240 | context->root_level = PT32E_ROOT_LEVEL; | |
2241 | } else { | |
2242 | context->gva_to_gpa = paging32_gva_to_gpa; | |
2243 | context->root_level = PT32_ROOT_LEVEL; | |
2244 | } | |
2245 | ||
2246 | return 0; | |
2247 | } | |
2248 | ||
2249 | static int init_kvm_softmmu(struct kvm_vcpu *vcpu) | |
6aa8b732 | 2250 | { |
a770f6f2 AK |
2251 | int r; |
2252 | ||
6aa8b732 | 2253 | ASSERT(vcpu); |
ad312c7c | 2254 | ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa)); |
6aa8b732 AK |
2255 | |
2256 | if (!is_paging(vcpu)) | |
a770f6f2 | 2257 | r = nonpaging_init_context(vcpu); |
a9058ecd | 2258 | else if (is_long_mode(vcpu)) |
a770f6f2 | 2259 | r = paging64_init_context(vcpu); |
6aa8b732 | 2260 | else if (is_pae(vcpu)) |
a770f6f2 | 2261 | r = paging32E_init_context(vcpu); |
6aa8b732 | 2262 | else |
a770f6f2 AK |
2263 | r = paging32_init_context(vcpu); |
2264 | ||
2265 | vcpu->arch.mmu.base_role.glevels = vcpu->arch.mmu.root_level; | |
2266 | ||
2267 | return r; | |
6aa8b732 AK |
2268 | } |
2269 | ||
fb72d167 JR |
2270 | static int init_kvm_mmu(struct kvm_vcpu *vcpu) |
2271 | { | |
35149e21 AL |
2272 | vcpu->arch.update_pte.pfn = bad_pfn; |
2273 | ||
fb72d167 JR |
2274 | if (tdp_enabled) |
2275 | return init_kvm_tdp_mmu(vcpu); | |
2276 | else | |
2277 | return init_kvm_softmmu(vcpu); | |
2278 | } | |
2279 | ||
6aa8b732 AK |
2280 | static void destroy_kvm_mmu(struct kvm_vcpu *vcpu) |
2281 | { | |
2282 | ASSERT(vcpu); | |
ad312c7c ZX |
2283 | if (VALID_PAGE(vcpu->arch.mmu.root_hpa)) { |
2284 | vcpu->arch.mmu.free(vcpu); | |
2285 | vcpu->arch.mmu.root_hpa = INVALID_PAGE; | |
6aa8b732 AK |
2286 | } |
2287 | } | |
2288 | ||
2289 | int kvm_mmu_reset_context(struct kvm_vcpu *vcpu) | |
17c3ba9d AK |
2290 | { |
2291 | destroy_kvm_mmu(vcpu); | |
2292 | return init_kvm_mmu(vcpu); | |
2293 | } | |
8668a3c4 | 2294 | EXPORT_SYMBOL_GPL(kvm_mmu_reset_context); |
17c3ba9d AK |
2295 | |
2296 | int kvm_mmu_load(struct kvm_vcpu *vcpu) | |
6aa8b732 | 2297 | { |
714b93da AK |
2298 | int r; |
2299 | ||
e2dec939 | 2300 | r = mmu_topup_memory_caches(vcpu); |
17c3ba9d AK |
2301 | if (r) |
2302 | goto out; | |
aaee2c94 | 2303 | spin_lock(&vcpu->kvm->mmu_lock); |
eb787d10 | 2304 | kvm_mmu_free_some_pages(vcpu); |
17c3ba9d | 2305 | mmu_alloc_roots(vcpu); |
0ba73cda | 2306 | mmu_sync_roots(vcpu); |
aaee2c94 | 2307 | spin_unlock(&vcpu->kvm->mmu_lock); |
ad312c7c | 2308 | kvm_x86_ops->set_cr3(vcpu, vcpu->arch.mmu.root_hpa); |
17c3ba9d | 2309 | kvm_mmu_flush_tlb(vcpu); |
714b93da AK |
2310 | out: |
2311 | return r; | |
6aa8b732 | 2312 | } |
17c3ba9d AK |
2313 | EXPORT_SYMBOL_GPL(kvm_mmu_load); |
2314 | ||
2315 | void kvm_mmu_unload(struct kvm_vcpu *vcpu) | |
2316 | { | |
2317 | mmu_free_roots(vcpu); | |
2318 | } | |
6aa8b732 | 2319 | |
09072daf | 2320 | static void mmu_pte_write_zap_pte(struct kvm_vcpu *vcpu, |
4db35314 | 2321 | struct kvm_mmu_page *sp, |
ac1b714e AK |
2322 | u64 *spte) |
2323 | { | |
2324 | u64 pte; | |
2325 | struct kvm_mmu_page *child; | |
2326 | ||
2327 | pte = *spte; | |
c7addb90 | 2328 | if (is_shadow_present_pte(pte)) { |
05da4558 MT |
2329 | if (sp->role.level == PT_PAGE_TABLE_LEVEL || |
2330 | is_large_pte(pte)) | |
290fc38d | 2331 | rmap_remove(vcpu->kvm, spte); |
ac1b714e AK |
2332 | else { |
2333 | child = page_header(pte & PT64_BASE_ADDR_MASK); | |
90cb0529 | 2334 | mmu_page_remove_parent_pte(child, spte); |
ac1b714e AK |
2335 | } |
2336 | } | |
c7addb90 | 2337 | set_shadow_pte(spte, shadow_trap_nonpresent_pte); |
05da4558 MT |
2338 | if (is_large_pte(pte)) |
2339 | --vcpu->kvm->stat.lpages; | |
ac1b714e AK |
2340 | } |
2341 | ||
0028425f | 2342 | static void mmu_pte_write_new_pte(struct kvm_vcpu *vcpu, |
4db35314 | 2343 | struct kvm_mmu_page *sp, |
0028425f | 2344 | u64 *spte, |
489f1d65 | 2345 | const void *new) |
0028425f | 2346 | { |
30945387 MT |
2347 | if (sp->role.level != PT_PAGE_TABLE_LEVEL) { |
2348 | if (!vcpu->arch.update_pte.largepage || | |
2349 | sp->role.glevels == PT32_ROOT_LEVEL) { | |
2350 | ++vcpu->kvm->stat.mmu_pde_zapped; | |
2351 | return; | |
2352 | } | |
2353 | } | |
0028425f | 2354 | |
4cee5764 | 2355 | ++vcpu->kvm->stat.mmu_pte_updated; |
4db35314 | 2356 | if (sp->role.glevels == PT32_ROOT_LEVEL) |
489f1d65 | 2357 | paging32_update_pte(vcpu, sp, spte, new); |
0028425f | 2358 | else |
489f1d65 | 2359 | paging64_update_pte(vcpu, sp, spte, new); |
0028425f AK |
2360 | } |
2361 | ||
79539cec AK |
2362 | static bool need_remote_flush(u64 old, u64 new) |
2363 | { | |
2364 | if (!is_shadow_present_pte(old)) | |
2365 | return false; | |
2366 | if (!is_shadow_present_pte(new)) | |
2367 | return true; | |
2368 | if ((old ^ new) & PT64_BASE_ADDR_MASK) | |
2369 | return true; | |
2370 | old ^= PT64_NX_MASK; | |
2371 | new ^= PT64_NX_MASK; | |
2372 | return (old & ~new & PT64_PERM_MASK) != 0; | |
2373 | } | |
2374 | ||
2375 | static void mmu_pte_write_flush_tlb(struct kvm_vcpu *vcpu, u64 old, u64 new) | |
2376 | { | |
2377 | if (need_remote_flush(old, new)) | |
2378 | kvm_flush_remote_tlbs(vcpu->kvm); | |
2379 | else | |
2380 | kvm_mmu_flush_tlb(vcpu); | |
2381 | } | |
2382 | ||
12b7d28f AK |
2383 | static bool last_updated_pte_accessed(struct kvm_vcpu *vcpu) |
2384 | { | |
ad312c7c | 2385 | u64 *spte = vcpu->arch.last_pte_updated; |
12b7d28f | 2386 | |
7b52345e | 2387 | return !!(spte && (*spte & shadow_accessed_mask)); |
12b7d28f AK |
2388 | } |
2389 | ||
d7824fff AK |
2390 | static void mmu_guess_page_from_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa, |
2391 | const u8 *new, int bytes) | |
2392 | { | |
2393 | gfn_t gfn; | |
2394 | int r; | |
2395 | u64 gpte = 0; | |
35149e21 | 2396 | pfn_t pfn; |
d7824fff | 2397 | |
05da4558 MT |
2398 | vcpu->arch.update_pte.largepage = 0; |
2399 | ||
d7824fff AK |
2400 | if (bytes != 4 && bytes != 8) |
2401 | return; | |
2402 | ||
2403 | /* | |
2404 | * Assume that the pte write on a page table of the same type | |
2405 | * as the current vcpu paging mode. This is nearly always true | |
2406 | * (might be false while changing modes). Note it is verified later | |
2407 | * by update_pte(). | |
2408 | */ | |
2409 | if (is_pae(vcpu)) { | |
2410 | /* Handle a 32-bit guest writing two halves of a 64-bit gpte */ | |
2411 | if ((bytes == 4) && (gpa % 4 == 0)) { | |
2412 | r = kvm_read_guest(vcpu->kvm, gpa & ~(u64)7, &gpte, 8); | |
2413 | if (r) | |
2414 | return; | |
2415 | memcpy((void *)&gpte + (gpa % 8), new, 4); | |
2416 | } else if ((bytes == 8) && (gpa % 8 == 0)) { | |
2417 | memcpy((void *)&gpte, new, 8); | |
2418 | } | |
2419 | } else { | |
2420 | if ((bytes == 4) && (gpa % 4 == 0)) | |
2421 | memcpy((void *)&gpte, new, 4); | |
2422 | } | |
2423 | if (!is_present_pte(gpte)) | |
2424 | return; | |
2425 | gfn = (gpte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT; | |
72dc67a6 | 2426 | |
05da4558 MT |
2427 | if (is_large_pte(gpte) && is_largepage_backed(vcpu, gfn)) { |
2428 | gfn &= ~(KVM_PAGES_PER_HPAGE-1); | |
2429 | vcpu->arch.update_pte.largepage = 1; | |
2430 | } | |
e930bffe | 2431 | vcpu->arch.update_pte.mmu_seq = vcpu->kvm->mmu_notifier_seq; |
4c2155ce | 2432 | smp_rmb(); |
35149e21 | 2433 | pfn = gfn_to_pfn(vcpu->kvm, gfn); |
72dc67a6 | 2434 | |
35149e21 AL |
2435 | if (is_error_pfn(pfn)) { |
2436 | kvm_release_pfn_clean(pfn); | |
d196e343 AK |
2437 | return; |
2438 | } | |
d7824fff | 2439 | vcpu->arch.update_pte.gfn = gfn; |
35149e21 | 2440 | vcpu->arch.update_pte.pfn = pfn; |
d7824fff AK |
2441 | } |
2442 | ||
1b7fcd32 AK |
2443 | static void kvm_mmu_access_page(struct kvm_vcpu *vcpu, gfn_t gfn) |
2444 | { | |
2445 | u64 *spte = vcpu->arch.last_pte_updated; | |
2446 | ||
2447 | if (spte | |
2448 | && vcpu->arch.last_pte_gfn == gfn | |
2449 | && shadow_accessed_mask | |
2450 | && !(*spte & shadow_accessed_mask) | |
2451 | && is_shadow_present_pte(*spte)) | |
2452 | set_bit(PT_ACCESSED_SHIFT, (unsigned long *)spte); | |
2453 | } | |
2454 | ||
09072daf | 2455 | void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa, |
ad218f85 MT |
2456 | const u8 *new, int bytes, |
2457 | bool guest_initiated) | |
da4a00f0 | 2458 | { |
9b7a0325 | 2459 | gfn_t gfn = gpa >> PAGE_SHIFT; |
4db35314 | 2460 | struct kvm_mmu_page *sp; |
0e7bc4b9 | 2461 | struct hlist_node *node, *n; |
9b7a0325 AK |
2462 | struct hlist_head *bucket; |
2463 | unsigned index; | |
489f1d65 | 2464 | u64 entry, gentry; |
9b7a0325 | 2465 | u64 *spte; |
9b7a0325 | 2466 | unsigned offset = offset_in_page(gpa); |
0e7bc4b9 | 2467 | unsigned pte_size; |
9b7a0325 | 2468 | unsigned page_offset; |
0e7bc4b9 | 2469 | unsigned misaligned; |
fce0657f | 2470 | unsigned quadrant; |
9b7a0325 | 2471 | int level; |
86a5ba02 | 2472 | int flooded = 0; |
ac1b714e | 2473 | int npte; |
489f1d65 | 2474 | int r; |
9b7a0325 | 2475 | |
b8688d51 | 2476 | pgprintk("%s: gpa %llx bytes %d\n", __func__, gpa, bytes); |
d7824fff | 2477 | mmu_guess_page_from_pte_write(vcpu, gpa, new, bytes); |
aaee2c94 | 2478 | spin_lock(&vcpu->kvm->mmu_lock); |
1b7fcd32 | 2479 | kvm_mmu_access_page(vcpu, gfn); |
eb787d10 | 2480 | kvm_mmu_free_some_pages(vcpu); |
4cee5764 | 2481 | ++vcpu->kvm->stat.mmu_pte_write; |
c7addb90 | 2482 | kvm_mmu_audit(vcpu, "pre pte write"); |
ad218f85 MT |
2483 | if (guest_initiated) { |
2484 | if (gfn == vcpu->arch.last_pt_write_gfn | |
2485 | && !last_updated_pte_accessed(vcpu)) { | |
2486 | ++vcpu->arch.last_pt_write_count; | |
2487 | if (vcpu->arch.last_pt_write_count >= 3) | |
2488 | flooded = 1; | |
2489 | } else { | |
2490 | vcpu->arch.last_pt_write_gfn = gfn; | |
2491 | vcpu->arch.last_pt_write_count = 1; | |
2492 | vcpu->arch.last_pte_updated = NULL; | |
2493 | } | |
86a5ba02 | 2494 | } |
1ae0a13d | 2495 | index = kvm_page_table_hashfn(gfn); |
f05e70ac | 2496 | bucket = &vcpu->kvm->arch.mmu_page_hash[index]; |
4db35314 | 2497 | hlist_for_each_entry_safe(sp, node, n, bucket, hash_link) { |
5b5c6a5a | 2498 | if (sp->gfn != gfn || sp->role.metaphysical || sp->role.invalid) |
9b7a0325 | 2499 | continue; |
4db35314 | 2500 | pte_size = sp->role.glevels == PT32_ROOT_LEVEL ? 4 : 8; |
0e7bc4b9 | 2501 | misaligned = (offset ^ (offset + bytes - 1)) & ~(pte_size - 1); |
e925c5ba | 2502 | misaligned |= bytes < 4; |
86a5ba02 | 2503 | if (misaligned || flooded) { |
0e7bc4b9 AK |
2504 | /* |
2505 | * Misaligned accesses are too much trouble to fix | |
2506 | * up; also, they usually indicate a page is not used | |
2507 | * as a page table. | |
86a5ba02 AK |
2508 | * |
2509 | * If we're seeing too many writes to a page, | |
2510 | * it may no longer be a page table, or we may be | |
2511 | * forking, in which case it is better to unmap the | |
2512 | * page. | |
0e7bc4b9 AK |
2513 | */ |
2514 | pgprintk("misaligned: gpa %llx bytes %d role %x\n", | |
4db35314 | 2515 | gpa, bytes, sp->role.word); |
07385413 MT |
2516 | if (kvm_mmu_zap_page(vcpu->kvm, sp)) |
2517 | n = bucket->first; | |
4cee5764 | 2518 | ++vcpu->kvm->stat.mmu_flooded; |
0e7bc4b9 AK |
2519 | continue; |
2520 | } | |
9b7a0325 | 2521 | page_offset = offset; |
4db35314 | 2522 | level = sp->role.level; |
ac1b714e | 2523 | npte = 1; |
4db35314 | 2524 | if (sp->role.glevels == PT32_ROOT_LEVEL) { |
ac1b714e AK |
2525 | page_offset <<= 1; /* 32->64 */ |
2526 | /* | |
2527 | * A 32-bit pde maps 4MB while the shadow pdes map | |
2528 | * only 2MB. So we need to double the offset again | |
2529 | * and zap two pdes instead of one. | |
2530 | */ | |
2531 | if (level == PT32_ROOT_LEVEL) { | |
6b8d0f9b | 2532 | page_offset &= ~7; /* kill rounding error */ |
ac1b714e AK |
2533 | page_offset <<= 1; |
2534 | npte = 2; | |
2535 | } | |
fce0657f | 2536 | quadrant = page_offset >> PAGE_SHIFT; |
9b7a0325 | 2537 | page_offset &= ~PAGE_MASK; |
4db35314 | 2538 | if (quadrant != sp->role.quadrant) |
fce0657f | 2539 | continue; |
9b7a0325 | 2540 | } |
4db35314 | 2541 | spte = &sp->spt[page_offset / sizeof(*spte)]; |
489f1d65 DE |
2542 | if ((gpa & (pte_size - 1)) || (bytes < pte_size)) { |
2543 | gentry = 0; | |
2544 | r = kvm_read_guest_atomic(vcpu->kvm, | |
2545 | gpa & ~(u64)(pte_size - 1), | |
2546 | &gentry, pte_size); | |
2547 | new = (const void *)&gentry; | |
2548 | if (r < 0) | |
2549 | new = NULL; | |
2550 | } | |
ac1b714e | 2551 | while (npte--) { |
79539cec | 2552 | entry = *spte; |
4db35314 | 2553 | mmu_pte_write_zap_pte(vcpu, sp, spte); |
489f1d65 DE |
2554 | if (new) |
2555 | mmu_pte_write_new_pte(vcpu, sp, spte, new); | |
79539cec | 2556 | mmu_pte_write_flush_tlb(vcpu, entry, *spte); |
ac1b714e | 2557 | ++spte; |
9b7a0325 | 2558 | } |
9b7a0325 | 2559 | } |
c7addb90 | 2560 | kvm_mmu_audit(vcpu, "post pte write"); |
aaee2c94 | 2561 | spin_unlock(&vcpu->kvm->mmu_lock); |
35149e21 AL |
2562 | if (!is_error_pfn(vcpu->arch.update_pte.pfn)) { |
2563 | kvm_release_pfn_clean(vcpu->arch.update_pte.pfn); | |
2564 | vcpu->arch.update_pte.pfn = bad_pfn; | |
d7824fff | 2565 | } |
da4a00f0 AK |
2566 | } |
2567 | ||
a436036b AK |
2568 | int kvm_mmu_unprotect_page_virt(struct kvm_vcpu *vcpu, gva_t gva) |
2569 | { | |
10589a46 MT |
2570 | gpa_t gpa; |
2571 | int r; | |
a436036b | 2572 | |
10589a46 | 2573 | gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gva); |
10589a46 | 2574 | |
aaee2c94 | 2575 | spin_lock(&vcpu->kvm->mmu_lock); |
10589a46 | 2576 | r = kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT); |
aaee2c94 | 2577 | spin_unlock(&vcpu->kvm->mmu_lock); |
10589a46 | 2578 | return r; |
a436036b | 2579 | } |
577bdc49 | 2580 | EXPORT_SYMBOL_GPL(kvm_mmu_unprotect_page_virt); |
a436036b | 2581 | |
22d95b12 | 2582 | void __kvm_mmu_free_some_pages(struct kvm_vcpu *vcpu) |
ebeace86 | 2583 | { |
f05e70ac | 2584 | while (vcpu->kvm->arch.n_free_mmu_pages < KVM_REFILL_PAGES) { |
4db35314 | 2585 | struct kvm_mmu_page *sp; |
ebeace86 | 2586 | |
f05e70ac | 2587 | sp = container_of(vcpu->kvm->arch.active_mmu_pages.prev, |
4db35314 AK |
2588 | struct kvm_mmu_page, link); |
2589 | kvm_mmu_zap_page(vcpu->kvm, sp); | |
4cee5764 | 2590 | ++vcpu->kvm->stat.mmu_recycled; |
ebeace86 AK |
2591 | } |
2592 | } | |
ebeace86 | 2593 | |
3067714c AK |
2594 | int kvm_mmu_page_fault(struct kvm_vcpu *vcpu, gva_t cr2, u32 error_code) |
2595 | { | |
2596 | int r; | |
2597 | enum emulation_result er; | |
2598 | ||
ad312c7c | 2599 | r = vcpu->arch.mmu.page_fault(vcpu, cr2, error_code); |
3067714c AK |
2600 | if (r < 0) |
2601 | goto out; | |
2602 | ||
2603 | if (!r) { | |
2604 | r = 1; | |
2605 | goto out; | |
2606 | } | |
2607 | ||
b733bfb5 AK |
2608 | r = mmu_topup_memory_caches(vcpu); |
2609 | if (r) | |
2610 | goto out; | |
2611 | ||
3067714c | 2612 | er = emulate_instruction(vcpu, vcpu->run, cr2, error_code, 0); |
3067714c AK |
2613 | |
2614 | switch (er) { | |
2615 | case EMULATE_DONE: | |
2616 | return 1; | |
2617 | case EMULATE_DO_MMIO: | |
2618 | ++vcpu->stat.mmio_exits; | |
2619 | return 0; | |
2620 | case EMULATE_FAIL: | |
2621 | kvm_report_emulation_failure(vcpu, "pagetable"); | |
2622 | return 1; | |
2623 | default: | |
2624 | BUG(); | |
2625 | } | |
2626 | out: | |
3067714c AK |
2627 | return r; |
2628 | } | |
2629 | EXPORT_SYMBOL_GPL(kvm_mmu_page_fault); | |
2630 | ||
a7052897 MT |
2631 | void kvm_mmu_invlpg(struct kvm_vcpu *vcpu, gva_t gva) |
2632 | { | |
a7052897 | 2633 | vcpu->arch.mmu.invlpg(vcpu, gva); |
a7052897 MT |
2634 | kvm_mmu_flush_tlb(vcpu); |
2635 | ++vcpu->stat.invlpg; | |
2636 | } | |
2637 | EXPORT_SYMBOL_GPL(kvm_mmu_invlpg); | |
2638 | ||
18552672 JR |
2639 | void kvm_enable_tdp(void) |
2640 | { | |
2641 | tdp_enabled = true; | |
2642 | } | |
2643 | EXPORT_SYMBOL_GPL(kvm_enable_tdp); | |
2644 | ||
5f4cb662 JR |
2645 | void kvm_disable_tdp(void) |
2646 | { | |
2647 | tdp_enabled = false; | |
2648 | } | |
2649 | EXPORT_SYMBOL_GPL(kvm_disable_tdp); | |
2650 | ||
6aa8b732 AK |
2651 | static void free_mmu_pages(struct kvm_vcpu *vcpu) |
2652 | { | |
4db35314 | 2653 | struct kvm_mmu_page *sp; |
6aa8b732 | 2654 | |
f05e70ac ZX |
2655 | while (!list_empty(&vcpu->kvm->arch.active_mmu_pages)) { |
2656 | sp = container_of(vcpu->kvm->arch.active_mmu_pages.next, | |
4db35314 AK |
2657 | struct kvm_mmu_page, link); |
2658 | kvm_mmu_zap_page(vcpu->kvm, sp); | |
8d2d73b9 | 2659 | cond_resched(); |
f51234c2 | 2660 | } |
ad312c7c | 2661 | free_page((unsigned long)vcpu->arch.mmu.pae_root); |
6aa8b732 AK |
2662 | } |
2663 | ||
2664 | static int alloc_mmu_pages(struct kvm_vcpu *vcpu) | |
2665 | { | |
17ac10ad | 2666 | struct page *page; |
6aa8b732 AK |
2667 | int i; |
2668 | ||
2669 | ASSERT(vcpu); | |
2670 | ||
f05e70ac ZX |
2671 | if (vcpu->kvm->arch.n_requested_mmu_pages) |
2672 | vcpu->kvm->arch.n_free_mmu_pages = | |
2673 | vcpu->kvm->arch.n_requested_mmu_pages; | |
82ce2c96 | 2674 | else |
f05e70ac ZX |
2675 | vcpu->kvm->arch.n_free_mmu_pages = |
2676 | vcpu->kvm->arch.n_alloc_mmu_pages; | |
17ac10ad AK |
2677 | /* |
2678 | * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64. | |
2679 | * Therefore we need to allocate shadow page tables in the first | |
2680 | * 4GB of memory, which happens to fit the DMA32 zone. | |
2681 | */ | |
2682 | page = alloc_page(GFP_KERNEL | __GFP_DMA32); | |
2683 | if (!page) | |
2684 | goto error_1; | |
ad312c7c | 2685 | vcpu->arch.mmu.pae_root = page_address(page); |
17ac10ad | 2686 | for (i = 0; i < 4; ++i) |
ad312c7c | 2687 | vcpu->arch.mmu.pae_root[i] = INVALID_PAGE; |
17ac10ad | 2688 | |
6aa8b732 AK |
2689 | return 0; |
2690 | ||
2691 | error_1: | |
2692 | free_mmu_pages(vcpu); | |
2693 | return -ENOMEM; | |
2694 | } | |
2695 | ||
8018c27b | 2696 | int kvm_mmu_create(struct kvm_vcpu *vcpu) |
6aa8b732 | 2697 | { |
6aa8b732 | 2698 | ASSERT(vcpu); |
ad312c7c | 2699 | ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa)); |
6aa8b732 | 2700 | |
8018c27b IM |
2701 | return alloc_mmu_pages(vcpu); |
2702 | } | |
6aa8b732 | 2703 | |
8018c27b IM |
2704 | int kvm_mmu_setup(struct kvm_vcpu *vcpu) |
2705 | { | |
2706 | ASSERT(vcpu); | |
ad312c7c | 2707 | ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa)); |
2c264957 | 2708 | |
8018c27b | 2709 | return init_kvm_mmu(vcpu); |
6aa8b732 AK |
2710 | } |
2711 | ||
2712 | void kvm_mmu_destroy(struct kvm_vcpu *vcpu) | |
2713 | { | |
2714 | ASSERT(vcpu); | |
2715 | ||
2716 | destroy_kvm_mmu(vcpu); | |
2717 | free_mmu_pages(vcpu); | |
714b93da | 2718 | mmu_free_memory_caches(vcpu); |
6aa8b732 AK |
2719 | } |
2720 | ||
90cb0529 | 2721 | void kvm_mmu_slot_remove_write_access(struct kvm *kvm, int slot) |
6aa8b732 | 2722 | { |
4db35314 | 2723 | struct kvm_mmu_page *sp; |
6aa8b732 | 2724 | |
2245a28f | 2725 | spin_lock(&kvm->mmu_lock); |
f05e70ac | 2726 | list_for_each_entry(sp, &kvm->arch.active_mmu_pages, link) { |
6aa8b732 AK |
2727 | int i; |
2728 | u64 *pt; | |
2729 | ||
291f26bc | 2730 | if (!test_bit(slot, sp->slot_bitmap)) |
6aa8b732 AK |
2731 | continue; |
2732 | ||
4db35314 | 2733 | pt = sp->spt; |
6aa8b732 AK |
2734 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) |
2735 | /* avoid RMW */ | |
9647c14c | 2736 | if (pt[i] & PT_WRITABLE_MASK) |
6aa8b732 | 2737 | pt[i] &= ~PT_WRITABLE_MASK; |
6aa8b732 | 2738 | } |
171d595d | 2739 | kvm_flush_remote_tlbs(kvm); |
2245a28f | 2740 | spin_unlock(&kvm->mmu_lock); |
6aa8b732 | 2741 | } |
37a7d8b0 | 2742 | |
90cb0529 | 2743 | void kvm_mmu_zap_all(struct kvm *kvm) |
e0fa826f | 2744 | { |
4db35314 | 2745 | struct kvm_mmu_page *sp, *node; |
e0fa826f | 2746 | |
aaee2c94 | 2747 | spin_lock(&kvm->mmu_lock); |
f05e70ac | 2748 | list_for_each_entry_safe(sp, node, &kvm->arch.active_mmu_pages, link) |
07385413 MT |
2749 | if (kvm_mmu_zap_page(kvm, sp)) |
2750 | node = container_of(kvm->arch.active_mmu_pages.next, | |
2751 | struct kvm_mmu_page, link); | |
aaee2c94 | 2752 | spin_unlock(&kvm->mmu_lock); |
e0fa826f | 2753 | |
90cb0529 | 2754 | kvm_flush_remote_tlbs(kvm); |
e0fa826f DL |
2755 | } |
2756 | ||
8b2cf73c | 2757 | static void kvm_mmu_remove_one_alloc_mmu_page(struct kvm *kvm) |
3ee16c81 IE |
2758 | { |
2759 | struct kvm_mmu_page *page; | |
2760 | ||
2761 | page = container_of(kvm->arch.active_mmu_pages.prev, | |
2762 | struct kvm_mmu_page, link); | |
2763 | kvm_mmu_zap_page(kvm, page); | |
2764 | } | |
2765 | ||
2766 | static int mmu_shrink(int nr_to_scan, gfp_t gfp_mask) | |
2767 | { | |
2768 | struct kvm *kvm; | |
2769 | struct kvm *kvm_freed = NULL; | |
2770 | int cache_count = 0; | |
2771 | ||
2772 | spin_lock(&kvm_lock); | |
2773 | ||
2774 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
2775 | int npages; | |
2776 | ||
5a4c9288 MT |
2777 | if (!down_read_trylock(&kvm->slots_lock)) |
2778 | continue; | |
3ee16c81 IE |
2779 | spin_lock(&kvm->mmu_lock); |
2780 | npages = kvm->arch.n_alloc_mmu_pages - | |
2781 | kvm->arch.n_free_mmu_pages; | |
2782 | cache_count += npages; | |
2783 | if (!kvm_freed && nr_to_scan > 0 && npages > 0) { | |
2784 | kvm_mmu_remove_one_alloc_mmu_page(kvm); | |
2785 | cache_count--; | |
2786 | kvm_freed = kvm; | |
2787 | } | |
2788 | nr_to_scan--; | |
2789 | ||
2790 | spin_unlock(&kvm->mmu_lock); | |
5a4c9288 | 2791 | up_read(&kvm->slots_lock); |
3ee16c81 IE |
2792 | } |
2793 | if (kvm_freed) | |
2794 | list_move_tail(&kvm_freed->vm_list, &vm_list); | |
2795 | ||
2796 | spin_unlock(&kvm_lock); | |
2797 | ||
2798 | return cache_count; | |
2799 | } | |
2800 | ||
2801 | static struct shrinker mmu_shrinker = { | |
2802 | .shrink = mmu_shrink, | |
2803 | .seeks = DEFAULT_SEEKS * 10, | |
2804 | }; | |
2805 | ||
2ddfd20e | 2806 | static void mmu_destroy_caches(void) |
b5a33a75 AK |
2807 | { |
2808 | if (pte_chain_cache) | |
2809 | kmem_cache_destroy(pte_chain_cache); | |
2810 | if (rmap_desc_cache) | |
2811 | kmem_cache_destroy(rmap_desc_cache); | |
d3d25b04 AK |
2812 | if (mmu_page_header_cache) |
2813 | kmem_cache_destroy(mmu_page_header_cache); | |
b5a33a75 AK |
2814 | } |
2815 | ||
3ee16c81 IE |
2816 | void kvm_mmu_module_exit(void) |
2817 | { | |
2818 | mmu_destroy_caches(); | |
2819 | unregister_shrinker(&mmu_shrinker); | |
2820 | } | |
2821 | ||
b5a33a75 AK |
2822 | int kvm_mmu_module_init(void) |
2823 | { | |
2824 | pte_chain_cache = kmem_cache_create("kvm_pte_chain", | |
2825 | sizeof(struct kvm_pte_chain), | |
20c2df83 | 2826 | 0, 0, NULL); |
b5a33a75 AK |
2827 | if (!pte_chain_cache) |
2828 | goto nomem; | |
2829 | rmap_desc_cache = kmem_cache_create("kvm_rmap_desc", | |
2830 | sizeof(struct kvm_rmap_desc), | |
20c2df83 | 2831 | 0, 0, NULL); |
b5a33a75 AK |
2832 | if (!rmap_desc_cache) |
2833 | goto nomem; | |
2834 | ||
d3d25b04 AK |
2835 | mmu_page_header_cache = kmem_cache_create("kvm_mmu_page_header", |
2836 | sizeof(struct kvm_mmu_page), | |
20c2df83 | 2837 | 0, 0, NULL); |
d3d25b04 AK |
2838 | if (!mmu_page_header_cache) |
2839 | goto nomem; | |
2840 | ||
3ee16c81 IE |
2841 | register_shrinker(&mmu_shrinker); |
2842 | ||
b5a33a75 AK |
2843 | return 0; |
2844 | ||
2845 | nomem: | |
3ee16c81 | 2846 | mmu_destroy_caches(); |
b5a33a75 AK |
2847 | return -ENOMEM; |
2848 | } | |
2849 | ||
3ad82a7e ZX |
2850 | /* |
2851 | * Caculate mmu pages needed for kvm. | |
2852 | */ | |
2853 | unsigned int kvm_mmu_calculate_mmu_pages(struct kvm *kvm) | |
2854 | { | |
2855 | int i; | |
2856 | unsigned int nr_mmu_pages; | |
2857 | unsigned int nr_pages = 0; | |
2858 | ||
2859 | for (i = 0; i < kvm->nmemslots; i++) | |
2860 | nr_pages += kvm->memslots[i].npages; | |
2861 | ||
2862 | nr_mmu_pages = nr_pages * KVM_PERMILLE_MMU_PAGES / 1000; | |
2863 | nr_mmu_pages = max(nr_mmu_pages, | |
2864 | (unsigned int) KVM_MIN_ALLOC_MMU_PAGES); | |
2865 | ||
2866 | return nr_mmu_pages; | |
2867 | } | |
2868 | ||
2f333bcb MT |
2869 | static void *pv_mmu_peek_buffer(struct kvm_pv_mmu_op_buffer *buffer, |
2870 | unsigned len) | |
2871 | { | |
2872 | if (len > buffer->len) | |
2873 | return NULL; | |
2874 | return buffer->ptr; | |
2875 | } | |
2876 | ||
2877 | static void *pv_mmu_read_buffer(struct kvm_pv_mmu_op_buffer *buffer, | |
2878 | unsigned len) | |
2879 | { | |
2880 | void *ret; | |
2881 | ||
2882 | ret = pv_mmu_peek_buffer(buffer, len); | |
2883 | if (!ret) | |
2884 | return ret; | |
2885 | buffer->ptr += len; | |
2886 | buffer->len -= len; | |
2887 | buffer->processed += len; | |
2888 | return ret; | |
2889 | } | |
2890 | ||
2891 | static int kvm_pv_mmu_write(struct kvm_vcpu *vcpu, | |
2892 | gpa_t addr, gpa_t value) | |
2893 | { | |
2894 | int bytes = 8; | |
2895 | int r; | |
2896 | ||
2897 | if (!is_long_mode(vcpu) && !is_pae(vcpu)) | |
2898 | bytes = 4; | |
2899 | ||
2900 | r = mmu_topup_memory_caches(vcpu); | |
2901 | if (r) | |
2902 | return r; | |
2903 | ||
3200f405 | 2904 | if (!emulator_write_phys(vcpu, addr, &value, bytes)) |
2f333bcb MT |
2905 | return -EFAULT; |
2906 | ||
2907 | return 1; | |
2908 | } | |
2909 | ||
2910 | static int kvm_pv_mmu_flush_tlb(struct kvm_vcpu *vcpu) | |
2911 | { | |
2912 | kvm_x86_ops->tlb_flush(vcpu); | |
6ad9f15c | 2913 | set_bit(KVM_REQ_MMU_SYNC, &vcpu->requests); |
2f333bcb MT |
2914 | return 1; |
2915 | } | |
2916 | ||
2917 | static int kvm_pv_mmu_release_pt(struct kvm_vcpu *vcpu, gpa_t addr) | |
2918 | { | |
2919 | spin_lock(&vcpu->kvm->mmu_lock); | |
2920 | mmu_unshadow(vcpu->kvm, addr >> PAGE_SHIFT); | |
2921 | spin_unlock(&vcpu->kvm->mmu_lock); | |
2922 | return 1; | |
2923 | } | |
2924 | ||
2925 | static int kvm_pv_mmu_op_one(struct kvm_vcpu *vcpu, | |
2926 | struct kvm_pv_mmu_op_buffer *buffer) | |
2927 | { | |
2928 | struct kvm_mmu_op_header *header; | |
2929 | ||
2930 | header = pv_mmu_peek_buffer(buffer, sizeof *header); | |
2931 | if (!header) | |
2932 | return 0; | |
2933 | switch (header->op) { | |
2934 | case KVM_MMU_OP_WRITE_PTE: { | |
2935 | struct kvm_mmu_op_write_pte *wpte; | |
2936 | ||
2937 | wpte = pv_mmu_read_buffer(buffer, sizeof *wpte); | |
2938 | if (!wpte) | |
2939 | return 0; | |
2940 | return kvm_pv_mmu_write(vcpu, wpte->pte_phys, | |
2941 | wpte->pte_val); | |
2942 | } | |
2943 | case KVM_MMU_OP_FLUSH_TLB: { | |
2944 | struct kvm_mmu_op_flush_tlb *ftlb; | |
2945 | ||
2946 | ftlb = pv_mmu_read_buffer(buffer, sizeof *ftlb); | |
2947 | if (!ftlb) | |
2948 | return 0; | |
2949 | return kvm_pv_mmu_flush_tlb(vcpu); | |
2950 | } | |
2951 | case KVM_MMU_OP_RELEASE_PT: { | |
2952 | struct kvm_mmu_op_release_pt *rpt; | |
2953 | ||
2954 | rpt = pv_mmu_read_buffer(buffer, sizeof *rpt); | |
2955 | if (!rpt) | |
2956 | return 0; | |
2957 | return kvm_pv_mmu_release_pt(vcpu, rpt->pt_phys); | |
2958 | } | |
2959 | default: return 0; | |
2960 | } | |
2961 | } | |
2962 | ||
2963 | int kvm_pv_mmu_op(struct kvm_vcpu *vcpu, unsigned long bytes, | |
2964 | gpa_t addr, unsigned long *ret) | |
2965 | { | |
2966 | int r; | |
6ad18fba | 2967 | struct kvm_pv_mmu_op_buffer *buffer = &vcpu->arch.mmu_op_buffer; |
2f333bcb | 2968 | |
6ad18fba DH |
2969 | buffer->ptr = buffer->buf; |
2970 | buffer->len = min_t(unsigned long, bytes, sizeof buffer->buf); | |
2971 | buffer->processed = 0; | |
2f333bcb | 2972 | |
6ad18fba | 2973 | r = kvm_read_guest(vcpu->kvm, addr, buffer->buf, buffer->len); |
2f333bcb MT |
2974 | if (r) |
2975 | goto out; | |
2976 | ||
6ad18fba DH |
2977 | while (buffer->len) { |
2978 | r = kvm_pv_mmu_op_one(vcpu, buffer); | |
2f333bcb MT |
2979 | if (r < 0) |
2980 | goto out; | |
2981 | if (r == 0) | |
2982 | break; | |
2983 | } | |
2984 | ||
2985 | r = 1; | |
2986 | out: | |
6ad18fba | 2987 | *ret = buffer->processed; |
2f333bcb MT |
2988 | return r; |
2989 | } | |
2990 | ||
37a7d8b0 AK |
2991 | #ifdef AUDIT |
2992 | ||
2993 | static const char *audit_msg; | |
2994 | ||
2995 | static gva_t canonicalize(gva_t gva) | |
2996 | { | |
2997 | #ifdef CONFIG_X86_64 | |
2998 | gva = (long long)(gva << 16) >> 16; | |
2999 | #endif | |
3000 | return gva; | |
3001 | } | |
3002 | ||
3003 | static void audit_mappings_page(struct kvm_vcpu *vcpu, u64 page_pte, | |
3004 | gva_t va, int level) | |
3005 | { | |
3006 | u64 *pt = __va(page_pte & PT64_BASE_ADDR_MASK); | |
3007 | int i; | |
3008 | gva_t va_delta = 1ul << (PAGE_SHIFT + 9 * (level - 1)); | |
3009 | ||
3010 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i, va += va_delta) { | |
3011 | u64 ent = pt[i]; | |
3012 | ||
c7addb90 | 3013 | if (ent == shadow_trap_nonpresent_pte) |
37a7d8b0 AK |
3014 | continue; |
3015 | ||
3016 | va = canonicalize(va); | |
c7addb90 AK |
3017 | if (level > 1) { |
3018 | if (ent == shadow_notrap_nonpresent_pte) | |
3019 | printk(KERN_ERR "audit: (%s) nontrapping pte" | |
3020 | " in nonleaf level: levels %d gva %lx" | |
3021 | " level %d pte %llx\n", audit_msg, | |
ad312c7c | 3022 | vcpu->arch.mmu.root_level, va, level, ent); |
c7addb90 | 3023 | |
37a7d8b0 | 3024 | audit_mappings_page(vcpu, ent, va, level - 1); |
c7addb90 | 3025 | } else { |
ad312c7c | 3026 | gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, va); |
35149e21 | 3027 | hpa_t hpa = (hpa_t)gpa_to_pfn(vcpu, gpa) << PAGE_SHIFT; |
37a7d8b0 | 3028 | |
c7addb90 | 3029 | if (is_shadow_present_pte(ent) |
37a7d8b0 | 3030 | && (ent & PT64_BASE_ADDR_MASK) != hpa) |
c7addb90 AK |
3031 | printk(KERN_ERR "xx audit error: (%s) levels %d" |
3032 | " gva %lx gpa %llx hpa %llx ent %llx %d\n", | |
ad312c7c | 3033 | audit_msg, vcpu->arch.mmu.root_level, |
d77c26fc MD |
3034 | va, gpa, hpa, ent, |
3035 | is_shadow_present_pte(ent)); | |
c7addb90 AK |
3036 | else if (ent == shadow_notrap_nonpresent_pte |
3037 | && !is_error_hpa(hpa)) | |
3038 | printk(KERN_ERR "audit: (%s) notrap shadow," | |
3039 | " valid guest gva %lx\n", audit_msg, va); | |
35149e21 | 3040 | kvm_release_pfn_clean(pfn); |
c7addb90 | 3041 | |
37a7d8b0 AK |
3042 | } |
3043 | } | |
3044 | } | |
3045 | ||
3046 | static void audit_mappings(struct kvm_vcpu *vcpu) | |
3047 | { | |
1ea252af | 3048 | unsigned i; |
37a7d8b0 | 3049 | |
ad312c7c ZX |
3050 | if (vcpu->arch.mmu.root_level == 4) |
3051 | audit_mappings_page(vcpu, vcpu->arch.mmu.root_hpa, 0, 4); | |
37a7d8b0 AK |
3052 | else |
3053 | for (i = 0; i < 4; ++i) | |
ad312c7c | 3054 | if (vcpu->arch.mmu.pae_root[i] & PT_PRESENT_MASK) |
37a7d8b0 | 3055 | audit_mappings_page(vcpu, |
ad312c7c | 3056 | vcpu->arch.mmu.pae_root[i], |
37a7d8b0 AK |
3057 | i << 30, |
3058 | 2); | |
3059 | } | |
3060 | ||
3061 | static int count_rmaps(struct kvm_vcpu *vcpu) | |
3062 | { | |
3063 | int nmaps = 0; | |
3064 | int i, j, k; | |
3065 | ||
3066 | for (i = 0; i < KVM_MEMORY_SLOTS; ++i) { | |
3067 | struct kvm_memory_slot *m = &vcpu->kvm->memslots[i]; | |
3068 | struct kvm_rmap_desc *d; | |
3069 | ||
3070 | for (j = 0; j < m->npages; ++j) { | |
290fc38d | 3071 | unsigned long *rmapp = &m->rmap[j]; |
37a7d8b0 | 3072 | |
290fc38d | 3073 | if (!*rmapp) |
37a7d8b0 | 3074 | continue; |
290fc38d | 3075 | if (!(*rmapp & 1)) { |
37a7d8b0 AK |
3076 | ++nmaps; |
3077 | continue; | |
3078 | } | |
290fc38d | 3079 | d = (struct kvm_rmap_desc *)(*rmapp & ~1ul); |
37a7d8b0 AK |
3080 | while (d) { |
3081 | for (k = 0; k < RMAP_EXT; ++k) | |
3082 | if (d->shadow_ptes[k]) | |
3083 | ++nmaps; | |
3084 | else | |
3085 | break; | |
3086 | d = d->more; | |
3087 | } | |
3088 | } | |
3089 | } | |
3090 | return nmaps; | |
3091 | } | |
3092 | ||
3093 | static int count_writable_mappings(struct kvm_vcpu *vcpu) | |
3094 | { | |
3095 | int nmaps = 0; | |
4db35314 | 3096 | struct kvm_mmu_page *sp; |
37a7d8b0 AK |
3097 | int i; |
3098 | ||
f05e70ac | 3099 | list_for_each_entry(sp, &vcpu->kvm->arch.active_mmu_pages, link) { |
4db35314 | 3100 | u64 *pt = sp->spt; |
37a7d8b0 | 3101 | |
4db35314 | 3102 | if (sp->role.level != PT_PAGE_TABLE_LEVEL) |
37a7d8b0 AK |
3103 | continue; |
3104 | ||
3105 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) { | |
3106 | u64 ent = pt[i]; | |
3107 | ||
3108 | if (!(ent & PT_PRESENT_MASK)) | |
3109 | continue; | |
3110 | if (!(ent & PT_WRITABLE_MASK)) | |
3111 | continue; | |
3112 | ++nmaps; | |
3113 | } | |
3114 | } | |
3115 | return nmaps; | |
3116 | } | |
3117 | ||
3118 | static void audit_rmap(struct kvm_vcpu *vcpu) | |
3119 | { | |
3120 | int n_rmap = count_rmaps(vcpu); | |
3121 | int n_actual = count_writable_mappings(vcpu); | |
3122 | ||
3123 | if (n_rmap != n_actual) | |
3124 | printk(KERN_ERR "%s: (%s) rmap %d actual %d\n", | |
b8688d51 | 3125 | __func__, audit_msg, n_rmap, n_actual); |
37a7d8b0 AK |
3126 | } |
3127 | ||
3128 | static void audit_write_protection(struct kvm_vcpu *vcpu) | |
3129 | { | |
4db35314 | 3130 | struct kvm_mmu_page *sp; |
290fc38d IE |
3131 | struct kvm_memory_slot *slot; |
3132 | unsigned long *rmapp; | |
3133 | gfn_t gfn; | |
37a7d8b0 | 3134 | |
f05e70ac | 3135 | list_for_each_entry(sp, &vcpu->kvm->arch.active_mmu_pages, link) { |
4db35314 | 3136 | if (sp->role.metaphysical) |
37a7d8b0 AK |
3137 | continue; |
3138 | ||
4db35314 | 3139 | gfn = unalias_gfn(vcpu->kvm, sp->gfn); |
2843099f | 3140 | slot = gfn_to_memslot_unaliased(vcpu->kvm, sp->gfn); |
290fc38d IE |
3141 | rmapp = &slot->rmap[gfn - slot->base_gfn]; |
3142 | if (*rmapp) | |
37a7d8b0 AK |
3143 | printk(KERN_ERR "%s: (%s) shadow page has writable" |
3144 | " mappings: gfn %lx role %x\n", | |
b8688d51 | 3145 | __func__, audit_msg, sp->gfn, |
4db35314 | 3146 | sp->role.word); |
37a7d8b0 AK |
3147 | } |
3148 | } | |
3149 | ||
3150 | static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg) | |
3151 | { | |
3152 | int olddbg = dbg; | |
3153 | ||
3154 | dbg = 0; | |
3155 | audit_msg = msg; | |
3156 | audit_rmap(vcpu); | |
3157 | audit_write_protection(vcpu); | |
3158 | audit_mappings(vcpu); | |
3159 | dbg = olddbg; | |
3160 | } | |
3161 | ||
3162 | #endif |