]>
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. | |
221d059d | 10 | * Copyright 2010 Red Hat, Inc. and/or its affilates. |
6aa8b732 AK |
11 | * |
12 | * Authors: | |
13 | * Yaniv Kamay <yaniv@qumranet.com> | |
14 | * Avi Kivity <avi@qumranet.com> | |
15 | * | |
16 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
17 | * the COPYING file in the top-level directory. | |
18 | * | |
19 | */ | |
e495606d | 20 | |
1d737c8a | 21 | #include "mmu.h" |
836a1b3c | 22 | #include "x86.h" |
6de4f3ad | 23 | #include "kvm_cache_regs.h" |
e495606d | 24 | |
edf88417 | 25 | #include <linux/kvm_host.h> |
6aa8b732 AK |
26 | #include <linux/types.h> |
27 | #include <linux/string.h> | |
6aa8b732 AK |
28 | #include <linux/mm.h> |
29 | #include <linux/highmem.h> | |
30 | #include <linux/module.h> | |
448353ca | 31 | #include <linux/swap.h> |
05da4558 | 32 | #include <linux/hugetlb.h> |
2f333bcb | 33 | #include <linux/compiler.h> |
bc6678a3 | 34 | #include <linux/srcu.h> |
5a0e3ad6 | 35 | #include <linux/slab.h> |
bf998156 | 36 | #include <linux/uaccess.h> |
6aa8b732 | 37 | |
e495606d AK |
38 | #include <asm/page.h> |
39 | #include <asm/cmpxchg.h> | |
4e542370 | 40 | #include <asm/io.h> |
13673a90 | 41 | #include <asm/vmx.h> |
6aa8b732 | 42 | |
18552672 JR |
43 | /* |
44 | * When setting this variable to true it enables Two-Dimensional-Paging | |
45 | * where the hardware walks 2 page tables: | |
46 | * 1. the guest-virtual to guest-physical | |
47 | * 2. while doing 1. it walks guest-physical to host-physical | |
48 | * If the hardware supports that we don't need to do shadow paging. | |
49 | */ | |
2f333bcb | 50 | bool tdp_enabled = false; |
18552672 | 51 | |
37a7d8b0 AK |
52 | #undef MMU_DEBUG |
53 | ||
54 | #undef AUDIT | |
55 | ||
56 | #ifdef AUDIT | |
57 | static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg); | |
58 | #else | |
59 | static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg) {} | |
60 | #endif | |
61 | ||
62 | #ifdef MMU_DEBUG | |
63 | ||
64 | #define pgprintk(x...) do { if (dbg) printk(x); } while (0) | |
65 | #define rmap_printk(x...) do { if (dbg) printk(x); } while (0) | |
66 | ||
67 | #else | |
68 | ||
69 | #define pgprintk(x...) do { } while (0) | |
70 | #define rmap_printk(x...) do { } while (0) | |
71 | ||
72 | #endif | |
73 | ||
74 | #if defined(MMU_DEBUG) || defined(AUDIT) | |
6ada8cca AK |
75 | static int dbg = 0; |
76 | module_param(dbg, bool, 0644); | |
37a7d8b0 | 77 | #endif |
6aa8b732 | 78 | |
582801a9 MT |
79 | static int oos_shadow = 1; |
80 | module_param(oos_shadow, bool, 0644); | |
81 | ||
d6c69ee9 YD |
82 | #ifndef MMU_DEBUG |
83 | #define ASSERT(x) do { } while (0) | |
84 | #else | |
6aa8b732 AK |
85 | #define ASSERT(x) \ |
86 | if (!(x)) { \ | |
87 | printk(KERN_WARNING "assertion failed %s:%d: %s\n", \ | |
88 | __FILE__, __LINE__, #x); \ | |
89 | } | |
d6c69ee9 | 90 | #endif |
6aa8b732 | 91 | |
6aa8b732 AK |
92 | #define PT_FIRST_AVAIL_BITS_SHIFT 9 |
93 | #define PT64_SECOND_AVAIL_BITS_SHIFT 52 | |
94 | ||
6aa8b732 AK |
95 | #define VALID_PAGE(x) ((x) != INVALID_PAGE) |
96 | ||
97 | #define PT64_LEVEL_BITS 9 | |
98 | ||
99 | #define PT64_LEVEL_SHIFT(level) \ | |
d77c26fc | 100 | (PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS) |
6aa8b732 AK |
101 | |
102 | #define PT64_LEVEL_MASK(level) \ | |
103 | (((1ULL << PT64_LEVEL_BITS) - 1) << PT64_LEVEL_SHIFT(level)) | |
104 | ||
105 | #define PT64_INDEX(address, level)\ | |
106 | (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1)) | |
107 | ||
108 | ||
109 | #define PT32_LEVEL_BITS 10 | |
110 | ||
111 | #define PT32_LEVEL_SHIFT(level) \ | |
d77c26fc | 112 | (PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS) |
6aa8b732 AK |
113 | |
114 | #define PT32_LEVEL_MASK(level) \ | |
115 | (((1ULL << PT32_LEVEL_BITS) - 1) << PT32_LEVEL_SHIFT(level)) | |
e04da980 JR |
116 | #define PT32_LVL_OFFSET_MASK(level) \ |
117 | (PT32_BASE_ADDR_MASK & ((1ULL << (PAGE_SHIFT + (((level) - 1) \ | |
118 | * PT32_LEVEL_BITS))) - 1)) | |
6aa8b732 AK |
119 | |
120 | #define PT32_INDEX(address, level)\ | |
121 | (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1)) | |
122 | ||
123 | ||
27aba766 | 124 | #define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1)) |
6aa8b732 AK |
125 | #define PT64_DIR_BASE_ADDR_MASK \ |
126 | (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1)) | |
e04da980 JR |
127 | #define PT64_LVL_ADDR_MASK(level) \ |
128 | (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + (((level) - 1) \ | |
129 | * PT64_LEVEL_BITS))) - 1)) | |
130 | #define PT64_LVL_OFFSET_MASK(level) \ | |
131 | (PT64_BASE_ADDR_MASK & ((1ULL << (PAGE_SHIFT + (((level) - 1) \ | |
132 | * PT64_LEVEL_BITS))) - 1)) | |
6aa8b732 AK |
133 | |
134 | #define PT32_BASE_ADDR_MASK PAGE_MASK | |
135 | #define PT32_DIR_BASE_ADDR_MASK \ | |
136 | (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1)) | |
e04da980 JR |
137 | #define PT32_LVL_ADDR_MASK(level) \ |
138 | (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + (((level) - 1) \ | |
139 | * PT32_LEVEL_BITS))) - 1)) | |
6aa8b732 | 140 | |
79539cec AK |
141 | #define PT64_PERM_MASK (PT_PRESENT_MASK | PT_WRITABLE_MASK | PT_USER_MASK \ |
142 | | PT64_NX_MASK) | |
6aa8b732 | 143 | |
cd4a4e53 AK |
144 | #define RMAP_EXT 4 |
145 | ||
fe135d2c AK |
146 | #define ACC_EXEC_MASK 1 |
147 | #define ACC_WRITE_MASK PT_WRITABLE_MASK | |
148 | #define ACC_USER_MASK PT_USER_MASK | |
149 | #define ACC_ALL (ACC_EXEC_MASK | ACC_WRITE_MASK | ACC_USER_MASK) | |
150 | ||
90bb6fc5 AK |
151 | #include <trace/events/kvm.h> |
152 | ||
07420171 AK |
153 | #define CREATE_TRACE_POINTS |
154 | #include "mmutrace.h" | |
155 | ||
1403283a IE |
156 | #define SPTE_HOST_WRITEABLE (1ULL << PT_FIRST_AVAIL_BITS_SHIFT) |
157 | ||
135f8c2b AK |
158 | #define SHADOW_PT_INDEX(addr, level) PT64_INDEX(addr, level) |
159 | ||
cd4a4e53 | 160 | struct kvm_rmap_desc { |
d555c333 | 161 | u64 *sptes[RMAP_EXT]; |
cd4a4e53 AK |
162 | struct kvm_rmap_desc *more; |
163 | }; | |
164 | ||
2d11123a AK |
165 | struct kvm_shadow_walk_iterator { |
166 | u64 addr; | |
167 | hpa_t shadow_addr; | |
168 | int level; | |
169 | u64 *sptep; | |
170 | unsigned index; | |
171 | }; | |
172 | ||
173 | #define for_each_shadow_entry(_vcpu, _addr, _walker) \ | |
174 | for (shadow_walk_init(&(_walker), _vcpu, _addr); \ | |
175 | shadow_walk_okay(&(_walker)); \ | |
176 | shadow_walk_next(&(_walker))) | |
177 | ||
6b18493d | 178 | typedef int (*mmu_parent_walk_fn) (struct kvm_mmu_page *sp); |
ad8cfbe3 | 179 | |
b5a33a75 AK |
180 | static struct kmem_cache *pte_chain_cache; |
181 | static struct kmem_cache *rmap_desc_cache; | |
d3d25b04 | 182 | static struct kmem_cache *mmu_page_header_cache; |
b5a33a75 | 183 | |
c7addb90 AK |
184 | static u64 __read_mostly shadow_trap_nonpresent_pte; |
185 | static u64 __read_mostly shadow_notrap_nonpresent_pte; | |
7b52345e SY |
186 | static u64 __read_mostly shadow_base_present_pte; |
187 | static u64 __read_mostly shadow_nx_mask; | |
188 | static u64 __read_mostly shadow_x_mask; /* mutual exclusive with nx_mask */ | |
189 | static u64 __read_mostly shadow_user_mask; | |
190 | static u64 __read_mostly shadow_accessed_mask; | |
191 | static u64 __read_mostly shadow_dirty_mask; | |
c7addb90 | 192 | |
82725b20 DE |
193 | static inline u64 rsvd_bits(int s, int e) |
194 | { | |
195 | return ((1ULL << (e - s + 1)) - 1) << s; | |
196 | } | |
197 | ||
c7addb90 AK |
198 | void kvm_mmu_set_nonpresent_ptes(u64 trap_pte, u64 notrap_pte) |
199 | { | |
200 | shadow_trap_nonpresent_pte = trap_pte; | |
201 | shadow_notrap_nonpresent_pte = notrap_pte; | |
202 | } | |
203 | EXPORT_SYMBOL_GPL(kvm_mmu_set_nonpresent_ptes); | |
204 | ||
7b52345e SY |
205 | void kvm_mmu_set_base_ptes(u64 base_pte) |
206 | { | |
207 | shadow_base_present_pte = base_pte; | |
208 | } | |
209 | EXPORT_SYMBOL_GPL(kvm_mmu_set_base_ptes); | |
210 | ||
211 | void kvm_mmu_set_mask_ptes(u64 user_mask, u64 accessed_mask, | |
4b12f0de | 212 | u64 dirty_mask, u64 nx_mask, u64 x_mask) |
7b52345e SY |
213 | { |
214 | shadow_user_mask = user_mask; | |
215 | shadow_accessed_mask = accessed_mask; | |
216 | shadow_dirty_mask = dirty_mask; | |
217 | shadow_nx_mask = nx_mask; | |
218 | shadow_x_mask = x_mask; | |
219 | } | |
220 | EXPORT_SYMBOL_GPL(kvm_mmu_set_mask_ptes); | |
221 | ||
3dbe1415 | 222 | static bool is_write_protection(struct kvm_vcpu *vcpu) |
6aa8b732 | 223 | { |
4d4ec087 | 224 | return kvm_read_cr0_bits(vcpu, X86_CR0_WP); |
6aa8b732 AK |
225 | } |
226 | ||
227 | static int is_cpuid_PSE36(void) | |
228 | { | |
229 | return 1; | |
230 | } | |
231 | ||
73b1087e AK |
232 | static int is_nx(struct kvm_vcpu *vcpu) |
233 | { | |
f6801dff | 234 | return vcpu->arch.efer & EFER_NX; |
73b1087e AK |
235 | } |
236 | ||
c7addb90 AK |
237 | static int is_shadow_present_pte(u64 pte) |
238 | { | |
c7addb90 AK |
239 | return pte != shadow_trap_nonpresent_pte |
240 | && pte != shadow_notrap_nonpresent_pte; | |
241 | } | |
242 | ||
05da4558 MT |
243 | static int is_large_pte(u64 pte) |
244 | { | |
245 | return pte & PT_PAGE_SIZE_MASK; | |
246 | } | |
247 | ||
8dae4445 | 248 | static int is_writable_pte(unsigned long pte) |
6aa8b732 AK |
249 | { |
250 | return pte & PT_WRITABLE_MASK; | |
251 | } | |
252 | ||
43a3795a | 253 | static int is_dirty_gpte(unsigned long pte) |
e3c5e7ec | 254 | { |
439e218a | 255 | return pte & PT_DIRTY_MASK; |
e3c5e7ec AK |
256 | } |
257 | ||
43a3795a | 258 | static int is_rmap_spte(u64 pte) |
cd4a4e53 | 259 | { |
4b1a80fa | 260 | return is_shadow_present_pte(pte); |
cd4a4e53 AK |
261 | } |
262 | ||
776e6633 MT |
263 | static int is_last_spte(u64 pte, int level) |
264 | { | |
265 | if (level == PT_PAGE_TABLE_LEVEL) | |
266 | return 1; | |
852e3c19 | 267 | if (is_large_pte(pte)) |
776e6633 MT |
268 | return 1; |
269 | return 0; | |
270 | } | |
271 | ||
35149e21 | 272 | static pfn_t spte_to_pfn(u64 pte) |
0b49ea86 | 273 | { |
35149e21 | 274 | return (pte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT; |
0b49ea86 AK |
275 | } |
276 | ||
da928521 AK |
277 | static gfn_t pse36_gfn_delta(u32 gpte) |
278 | { | |
279 | int shift = 32 - PT32_DIR_PSE36_SHIFT - PAGE_SHIFT; | |
280 | ||
281 | return (gpte & PT32_DIR_PSE36_MASK) << shift; | |
282 | } | |
283 | ||
d555c333 | 284 | static void __set_spte(u64 *sptep, u64 spte) |
e663ee64 AK |
285 | { |
286 | #ifdef CONFIG_X86_64 | |
287 | set_64bit((unsigned long *)sptep, spte); | |
288 | #else | |
289 | set_64bit((unsigned long long *)sptep, spte); | |
290 | #endif | |
291 | } | |
292 | ||
e2dec939 | 293 | static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache, |
2e3e5882 | 294 | struct kmem_cache *base_cache, int min) |
714b93da AK |
295 | { |
296 | void *obj; | |
297 | ||
298 | if (cache->nobjs >= min) | |
e2dec939 | 299 | return 0; |
714b93da | 300 | while (cache->nobjs < ARRAY_SIZE(cache->objects)) { |
2e3e5882 | 301 | obj = kmem_cache_zalloc(base_cache, GFP_KERNEL); |
714b93da | 302 | if (!obj) |
e2dec939 | 303 | return -ENOMEM; |
714b93da AK |
304 | cache->objects[cache->nobjs++] = obj; |
305 | } | |
e2dec939 | 306 | return 0; |
714b93da AK |
307 | } |
308 | ||
e8ad9a70 XG |
309 | static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc, |
310 | struct kmem_cache *cache) | |
714b93da AK |
311 | { |
312 | while (mc->nobjs) | |
e8ad9a70 | 313 | kmem_cache_free(cache, mc->objects[--mc->nobjs]); |
714b93da AK |
314 | } |
315 | ||
c1158e63 | 316 | static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache *cache, |
2e3e5882 | 317 | int min) |
c1158e63 AK |
318 | { |
319 | struct page *page; | |
320 | ||
321 | if (cache->nobjs >= min) | |
322 | return 0; | |
323 | while (cache->nobjs < ARRAY_SIZE(cache->objects)) { | |
2e3e5882 | 324 | page = alloc_page(GFP_KERNEL); |
c1158e63 AK |
325 | if (!page) |
326 | return -ENOMEM; | |
c1158e63 AK |
327 | cache->objects[cache->nobjs++] = page_address(page); |
328 | } | |
329 | return 0; | |
330 | } | |
331 | ||
332 | static void mmu_free_memory_cache_page(struct kvm_mmu_memory_cache *mc) | |
333 | { | |
334 | while (mc->nobjs) | |
c4d198d5 | 335 | free_page((unsigned long)mc->objects[--mc->nobjs]); |
c1158e63 AK |
336 | } |
337 | ||
2e3e5882 | 338 | static int mmu_topup_memory_caches(struct kvm_vcpu *vcpu) |
714b93da | 339 | { |
e2dec939 AK |
340 | int r; |
341 | ||
ad312c7c | 342 | r = mmu_topup_memory_cache(&vcpu->arch.mmu_pte_chain_cache, |
2e3e5882 | 343 | pte_chain_cache, 4); |
e2dec939 AK |
344 | if (r) |
345 | goto out; | |
ad312c7c | 346 | r = mmu_topup_memory_cache(&vcpu->arch.mmu_rmap_desc_cache, |
c41ef344 | 347 | rmap_desc_cache, 4); |
d3d25b04 AK |
348 | if (r) |
349 | goto out; | |
ad312c7c | 350 | r = mmu_topup_memory_cache_page(&vcpu->arch.mmu_page_cache, 8); |
d3d25b04 AK |
351 | if (r) |
352 | goto out; | |
ad312c7c | 353 | r = mmu_topup_memory_cache(&vcpu->arch.mmu_page_header_cache, |
2e3e5882 | 354 | mmu_page_header_cache, 4); |
e2dec939 AK |
355 | out: |
356 | return r; | |
714b93da AK |
357 | } |
358 | ||
359 | static void mmu_free_memory_caches(struct kvm_vcpu *vcpu) | |
360 | { | |
e8ad9a70 XG |
361 | mmu_free_memory_cache(&vcpu->arch.mmu_pte_chain_cache, pte_chain_cache); |
362 | mmu_free_memory_cache(&vcpu->arch.mmu_rmap_desc_cache, rmap_desc_cache); | |
ad312c7c | 363 | mmu_free_memory_cache_page(&vcpu->arch.mmu_page_cache); |
e8ad9a70 XG |
364 | mmu_free_memory_cache(&vcpu->arch.mmu_page_header_cache, |
365 | mmu_page_header_cache); | |
714b93da AK |
366 | } |
367 | ||
368 | static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc, | |
369 | size_t size) | |
370 | { | |
371 | void *p; | |
372 | ||
373 | BUG_ON(!mc->nobjs); | |
374 | p = mc->objects[--mc->nobjs]; | |
714b93da AK |
375 | return p; |
376 | } | |
377 | ||
714b93da AK |
378 | static struct kvm_pte_chain *mmu_alloc_pte_chain(struct kvm_vcpu *vcpu) |
379 | { | |
ad312c7c | 380 | return mmu_memory_cache_alloc(&vcpu->arch.mmu_pte_chain_cache, |
714b93da AK |
381 | sizeof(struct kvm_pte_chain)); |
382 | } | |
383 | ||
90cb0529 | 384 | static void mmu_free_pte_chain(struct kvm_pte_chain *pc) |
714b93da | 385 | { |
e8ad9a70 | 386 | kmem_cache_free(pte_chain_cache, pc); |
714b93da AK |
387 | } |
388 | ||
389 | static struct kvm_rmap_desc *mmu_alloc_rmap_desc(struct kvm_vcpu *vcpu) | |
390 | { | |
ad312c7c | 391 | return mmu_memory_cache_alloc(&vcpu->arch.mmu_rmap_desc_cache, |
714b93da AK |
392 | sizeof(struct kvm_rmap_desc)); |
393 | } | |
394 | ||
90cb0529 | 395 | static void mmu_free_rmap_desc(struct kvm_rmap_desc *rd) |
714b93da | 396 | { |
e8ad9a70 | 397 | kmem_cache_free(rmap_desc_cache, rd); |
714b93da AK |
398 | } |
399 | ||
05da4558 MT |
400 | /* |
401 | * Return the pointer to the largepage write count for a given | |
402 | * gfn, handling slots that are not large page aligned. | |
403 | */ | |
d25797b2 JR |
404 | static int *slot_largepage_idx(gfn_t gfn, |
405 | struct kvm_memory_slot *slot, | |
406 | int level) | |
05da4558 MT |
407 | { |
408 | unsigned long idx; | |
409 | ||
d25797b2 JR |
410 | idx = (gfn / KVM_PAGES_PER_HPAGE(level)) - |
411 | (slot->base_gfn / KVM_PAGES_PER_HPAGE(level)); | |
412 | return &slot->lpage_info[level - 2][idx].write_count; | |
05da4558 MT |
413 | } |
414 | ||
415 | static void account_shadowed(struct kvm *kvm, gfn_t gfn) | |
416 | { | |
d25797b2 | 417 | struct kvm_memory_slot *slot; |
05da4558 | 418 | int *write_count; |
d25797b2 | 419 | int i; |
05da4558 | 420 | |
2843099f | 421 | gfn = unalias_gfn(kvm, gfn); |
d25797b2 JR |
422 | |
423 | slot = gfn_to_memslot_unaliased(kvm, gfn); | |
424 | for (i = PT_DIRECTORY_LEVEL; | |
425 | i < PT_PAGE_TABLE_LEVEL + KVM_NR_PAGE_SIZES; ++i) { | |
426 | write_count = slot_largepage_idx(gfn, slot, i); | |
427 | *write_count += 1; | |
428 | } | |
05da4558 MT |
429 | } |
430 | ||
431 | static void unaccount_shadowed(struct kvm *kvm, gfn_t gfn) | |
432 | { | |
d25797b2 | 433 | struct kvm_memory_slot *slot; |
05da4558 | 434 | int *write_count; |
d25797b2 | 435 | int i; |
05da4558 | 436 | |
2843099f | 437 | gfn = unalias_gfn(kvm, gfn); |
77a1a715 | 438 | slot = gfn_to_memslot_unaliased(kvm, gfn); |
d25797b2 JR |
439 | for (i = PT_DIRECTORY_LEVEL; |
440 | i < PT_PAGE_TABLE_LEVEL + KVM_NR_PAGE_SIZES; ++i) { | |
d25797b2 JR |
441 | write_count = slot_largepage_idx(gfn, slot, i); |
442 | *write_count -= 1; | |
443 | WARN_ON(*write_count < 0); | |
444 | } | |
05da4558 MT |
445 | } |
446 | ||
d25797b2 JR |
447 | static int has_wrprotected_page(struct kvm *kvm, |
448 | gfn_t gfn, | |
449 | int level) | |
05da4558 | 450 | { |
2843099f | 451 | struct kvm_memory_slot *slot; |
05da4558 MT |
452 | int *largepage_idx; |
453 | ||
2843099f IE |
454 | gfn = unalias_gfn(kvm, gfn); |
455 | slot = gfn_to_memslot_unaliased(kvm, gfn); | |
05da4558 | 456 | if (slot) { |
d25797b2 | 457 | largepage_idx = slot_largepage_idx(gfn, slot, level); |
05da4558 MT |
458 | return *largepage_idx; |
459 | } | |
460 | ||
461 | return 1; | |
462 | } | |
463 | ||
d25797b2 | 464 | static int host_mapping_level(struct kvm *kvm, gfn_t gfn) |
05da4558 | 465 | { |
8f0b1ab6 | 466 | unsigned long page_size; |
d25797b2 | 467 | int i, ret = 0; |
05da4558 | 468 | |
8f0b1ab6 | 469 | page_size = kvm_host_page_size(kvm, gfn); |
05da4558 | 470 | |
d25797b2 JR |
471 | for (i = PT_PAGE_TABLE_LEVEL; |
472 | i < (PT_PAGE_TABLE_LEVEL + KVM_NR_PAGE_SIZES); ++i) { | |
473 | if (page_size >= KVM_HPAGE_SIZE(i)) | |
474 | ret = i; | |
475 | else | |
476 | break; | |
477 | } | |
478 | ||
4c2155ce | 479 | return ret; |
05da4558 MT |
480 | } |
481 | ||
d25797b2 | 482 | static int mapping_level(struct kvm_vcpu *vcpu, gfn_t large_gfn) |
05da4558 MT |
483 | { |
484 | struct kvm_memory_slot *slot; | |
878403b7 | 485 | int host_level, level, max_level; |
05da4558 MT |
486 | |
487 | slot = gfn_to_memslot(vcpu->kvm, large_gfn); | |
488 | if (slot && slot->dirty_bitmap) | |
d25797b2 | 489 | return PT_PAGE_TABLE_LEVEL; |
05da4558 | 490 | |
d25797b2 JR |
491 | host_level = host_mapping_level(vcpu->kvm, large_gfn); |
492 | ||
493 | if (host_level == PT_PAGE_TABLE_LEVEL) | |
494 | return host_level; | |
495 | ||
878403b7 SY |
496 | max_level = kvm_x86_ops->get_lpage_level() < host_level ? |
497 | kvm_x86_ops->get_lpage_level() : host_level; | |
498 | ||
499 | for (level = PT_DIRECTORY_LEVEL; level <= max_level; ++level) | |
d25797b2 JR |
500 | if (has_wrprotected_page(vcpu->kvm, large_gfn, level)) |
501 | break; | |
d25797b2 JR |
502 | |
503 | return level - 1; | |
05da4558 MT |
504 | } |
505 | ||
290fc38d IE |
506 | /* |
507 | * Take gfn and return the reverse mapping to it. | |
508 | * Note: gfn must be unaliased before this function get called | |
509 | */ | |
510 | ||
44ad9944 | 511 | static unsigned long *gfn_to_rmap(struct kvm *kvm, gfn_t gfn, int level) |
290fc38d IE |
512 | { |
513 | struct kvm_memory_slot *slot; | |
05da4558 | 514 | unsigned long idx; |
290fc38d IE |
515 | |
516 | slot = gfn_to_memslot(kvm, gfn); | |
44ad9944 | 517 | if (likely(level == PT_PAGE_TABLE_LEVEL)) |
05da4558 MT |
518 | return &slot->rmap[gfn - slot->base_gfn]; |
519 | ||
44ad9944 JR |
520 | idx = (gfn / KVM_PAGES_PER_HPAGE(level)) - |
521 | (slot->base_gfn / KVM_PAGES_PER_HPAGE(level)); | |
05da4558 | 522 | |
44ad9944 | 523 | return &slot->lpage_info[level - 2][idx].rmap_pde; |
290fc38d IE |
524 | } |
525 | ||
cd4a4e53 AK |
526 | /* |
527 | * Reverse mapping data structures: | |
528 | * | |
290fc38d IE |
529 | * If rmapp bit zero is zero, then rmapp point to the shadw page table entry |
530 | * that points to page_address(page). | |
cd4a4e53 | 531 | * |
290fc38d IE |
532 | * If rmapp bit zero is one, (then rmap & ~1) points to a struct kvm_rmap_desc |
533 | * containing more mappings. | |
53a27b39 MT |
534 | * |
535 | * Returns the number of rmap entries before the spte was added or zero if | |
536 | * the spte was not added. | |
537 | * | |
cd4a4e53 | 538 | */ |
44ad9944 | 539 | static int rmap_add(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn) |
cd4a4e53 | 540 | { |
4db35314 | 541 | struct kvm_mmu_page *sp; |
cd4a4e53 | 542 | struct kvm_rmap_desc *desc; |
290fc38d | 543 | unsigned long *rmapp; |
53a27b39 | 544 | int i, count = 0; |
cd4a4e53 | 545 | |
43a3795a | 546 | if (!is_rmap_spte(*spte)) |
53a27b39 | 547 | return count; |
290fc38d | 548 | gfn = unalias_gfn(vcpu->kvm, gfn); |
4db35314 AK |
549 | sp = page_header(__pa(spte)); |
550 | sp->gfns[spte - sp->spt] = gfn; | |
44ad9944 | 551 | rmapp = gfn_to_rmap(vcpu->kvm, gfn, sp->role.level); |
290fc38d | 552 | if (!*rmapp) { |
cd4a4e53 | 553 | rmap_printk("rmap_add: %p %llx 0->1\n", spte, *spte); |
290fc38d IE |
554 | *rmapp = (unsigned long)spte; |
555 | } else if (!(*rmapp & 1)) { | |
cd4a4e53 | 556 | rmap_printk("rmap_add: %p %llx 1->many\n", spte, *spte); |
714b93da | 557 | desc = mmu_alloc_rmap_desc(vcpu); |
d555c333 AK |
558 | desc->sptes[0] = (u64 *)*rmapp; |
559 | desc->sptes[1] = spte; | |
290fc38d | 560 | *rmapp = (unsigned long)desc | 1; |
cd4a4e53 AK |
561 | } else { |
562 | rmap_printk("rmap_add: %p %llx many->many\n", spte, *spte); | |
290fc38d | 563 | desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul); |
d555c333 | 564 | while (desc->sptes[RMAP_EXT-1] && desc->more) { |
cd4a4e53 | 565 | desc = desc->more; |
53a27b39 MT |
566 | count += RMAP_EXT; |
567 | } | |
d555c333 | 568 | if (desc->sptes[RMAP_EXT-1]) { |
714b93da | 569 | desc->more = mmu_alloc_rmap_desc(vcpu); |
cd4a4e53 AK |
570 | desc = desc->more; |
571 | } | |
d555c333 | 572 | for (i = 0; desc->sptes[i]; ++i) |
cd4a4e53 | 573 | ; |
d555c333 | 574 | desc->sptes[i] = spte; |
cd4a4e53 | 575 | } |
53a27b39 | 576 | return count; |
cd4a4e53 AK |
577 | } |
578 | ||
290fc38d | 579 | static void rmap_desc_remove_entry(unsigned long *rmapp, |
cd4a4e53 AK |
580 | struct kvm_rmap_desc *desc, |
581 | int i, | |
582 | struct kvm_rmap_desc *prev_desc) | |
583 | { | |
584 | int j; | |
585 | ||
d555c333 | 586 | for (j = RMAP_EXT - 1; !desc->sptes[j] && j > i; --j) |
cd4a4e53 | 587 | ; |
d555c333 AK |
588 | desc->sptes[i] = desc->sptes[j]; |
589 | desc->sptes[j] = NULL; | |
cd4a4e53 AK |
590 | if (j != 0) |
591 | return; | |
592 | if (!prev_desc && !desc->more) | |
d555c333 | 593 | *rmapp = (unsigned long)desc->sptes[0]; |
cd4a4e53 AK |
594 | else |
595 | if (prev_desc) | |
596 | prev_desc->more = desc->more; | |
597 | else | |
290fc38d | 598 | *rmapp = (unsigned long)desc->more | 1; |
90cb0529 | 599 | mmu_free_rmap_desc(desc); |
cd4a4e53 AK |
600 | } |
601 | ||
290fc38d | 602 | static void rmap_remove(struct kvm *kvm, u64 *spte) |
cd4a4e53 | 603 | { |
cd4a4e53 AK |
604 | struct kvm_rmap_desc *desc; |
605 | struct kvm_rmap_desc *prev_desc; | |
4db35314 | 606 | struct kvm_mmu_page *sp; |
35149e21 | 607 | pfn_t pfn; |
290fc38d | 608 | unsigned long *rmapp; |
cd4a4e53 AK |
609 | int i; |
610 | ||
43a3795a | 611 | if (!is_rmap_spte(*spte)) |
cd4a4e53 | 612 | return; |
4db35314 | 613 | sp = page_header(__pa(spte)); |
35149e21 | 614 | pfn = spte_to_pfn(*spte); |
7b52345e | 615 | if (*spte & shadow_accessed_mask) |
35149e21 | 616 | kvm_set_pfn_accessed(pfn); |
8dae4445 | 617 | if (is_writable_pte(*spte)) |
acb66dd0 | 618 | kvm_set_pfn_dirty(pfn); |
44ad9944 | 619 | rmapp = gfn_to_rmap(kvm, sp->gfns[spte - sp->spt], sp->role.level); |
290fc38d | 620 | if (!*rmapp) { |
cd4a4e53 AK |
621 | printk(KERN_ERR "rmap_remove: %p %llx 0->BUG\n", spte, *spte); |
622 | BUG(); | |
290fc38d | 623 | } else if (!(*rmapp & 1)) { |
cd4a4e53 | 624 | rmap_printk("rmap_remove: %p %llx 1->0\n", spte, *spte); |
290fc38d | 625 | if ((u64 *)*rmapp != spte) { |
cd4a4e53 AK |
626 | printk(KERN_ERR "rmap_remove: %p %llx 1->BUG\n", |
627 | spte, *spte); | |
628 | BUG(); | |
629 | } | |
290fc38d | 630 | *rmapp = 0; |
cd4a4e53 AK |
631 | } else { |
632 | rmap_printk("rmap_remove: %p %llx many->many\n", spte, *spte); | |
290fc38d | 633 | desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul); |
cd4a4e53 AK |
634 | prev_desc = NULL; |
635 | while (desc) { | |
d555c333 AK |
636 | for (i = 0; i < RMAP_EXT && desc->sptes[i]; ++i) |
637 | if (desc->sptes[i] == spte) { | |
290fc38d | 638 | rmap_desc_remove_entry(rmapp, |
714b93da | 639 | desc, i, |
cd4a4e53 AK |
640 | prev_desc); |
641 | return; | |
642 | } | |
643 | prev_desc = desc; | |
644 | desc = desc->more; | |
645 | } | |
186a3e52 | 646 | pr_err("rmap_remove: %p %llx many->many\n", spte, *spte); |
cd4a4e53 AK |
647 | BUG(); |
648 | } | |
649 | } | |
650 | ||
98348e95 | 651 | static u64 *rmap_next(struct kvm *kvm, unsigned long *rmapp, u64 *spte) |
374cbac0 | 652 | { |
374cbac0 | 653 | struct kvm_rmap_desc *desc; |
98348e95 IE |
654 | u64 *prev_spte; |
655 | int i; | |
656 | ||
657 | if (!*rmapp) | |
658 | return NULL; | |
659 | else if (!(*rmapp & 1)) { | |
660 | if (!spte) | |
661 | return (u64 *)*rmapp; | |
662 | return NULL; | |
663 | } | |
664 | desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul); | |
98348e95 IE |
665 | prev_spte = NULL; |
666 | while (desc) { | |
d555c333 | 667 | for (i = 0; i < RMAP_EXT && desc->sptes[i]; ++i) { |
98348e95 | 668 | if (prev_spte == spte) |
d555c333 AK |
669 | return desc->sptes[i]; |
670 | prev_spte = desc->sptes[i]; | |
98348e95 IE |
671 | } |
672 | desc = desc->more; | |
673 | } | |
674 | return NULL; | |
675 | } | |
676 | ||
b1a36821 | 677 | static int rmap_write_protect(struct kvm *kvm, u64 gfn) |
98348e95 | 678 | { |
290fc38d | 679 | unsigned long *rmapp; |
374cbac0 | 680 | u64 *spte; |
44ad9944 | 681 | int i, write_protected = 0; |
374cbac0 | 682 | |
4a4c9924 | 683 | gfn = unalias_gfn(kvm, gfn); |
44ad9944 | 684 | rmapp = gfn_to_rmap(kvm, gfn, PT_PAGE_TABLE_LEVEL); |
374cbac0 | 685 | |
98348e95 IE |
686 | spte = rmap_next(kvm, rmapp, NULL); |
687 | while (spte) { | |
374cbac0 | 688 | BUG_ON(!spte); |
374cbac0 | 689 | BUG_ON(!(*spte & PT_PRESENT_MASK)); |
374cbac0 | 690 | rmap_printk("rmap_write_protect: spte %p %llx\n", spte, *spte); |
8dae4445 | 691 | if (is_writable_pte(*spte)) { |
d555c333 | 692 | __set_spte(spte, *spte & ~PT_WRITABLE_MASK); |
caa5b8a5 ED |
693 | write_protected = 1; |
694 | } | |
9647c14c | 695 | spte = rmap_next(kvm, rmapp, spte); |
374cbac0 | 696 | } |
855149aa | 697 | if (write_protected) { |
35149e21 | 698 | pfn_t pfn; |
855149aa IE |
699 | |
700 | spte = rmap_next(kvm, rmapp, NULL); | |
35149e21 AL |
701 | pfn = spte_to_pfn(*spte); |
702 | kvm_set_pfn_dirty(pfn); | |
855149aa IE |
703 | } |
704 | ||
05da4558 | 705 | /* check for huge page mappings */ |
44ad9944 JR |
706 | for (i = PT_DIRECTORY_LEVEL; |
707 | i < PT_PAGE_TABLE_LEVEL + KVM_NR_PAGE_SIZES; ++i) { | |
708 | rmapp = gfn_to_rmap(kvm, gfn, i); | |
709 | spte = rmap_next(kvm, rmapp, NULL); | |
710 | while (spte) { | |
711 | BUG_ON(!spte); | |
712 | BUG_ON(!(*spte & PT_PRESENT_MASK)); | |
713 | BUG_ON((*spte & (PT_PAGE_SIZE_MASK|PT_PRESENT_MASK)) != (PT_PAGE_SIZE_MASK|PT_PRESENT_MASK)); | |
714 | pgprintk("rmap_write_protect(large): spte %p %llx %lld\n", spte, *spte, gfn); | |
8dae4445 | 715 | if (is_writable_pte(*spte)) { |
44ad9944 JR |
716 | rmap_remove(kvm, spte); |
717 | --kvm->stat.lpages; | |
718 | __set_spte(spte, shadow_trap_nonpresent_pte); | |
719 | spte = NULL; | |
720 | write_protected = 1; | |
721 | } | |
722 | spte = rmap_next(kvm, rmapp, spte); | |
05da4558 | 723 | } |
05da4558 MT |
724 | } |
725 | ||
b1a36821 | 726 | return write_protected; |
374cbac0 AK |
727 | } |
728 | ||
8a8365c5 FD |
729 | static int kvm_unmap_rmapp(struct kvm *kvm, unsigned long *rmapp, |
730 | unsigned long data) | |
e930bffe AA |
731 | { |
732 | u64 *spte; | |
733 | int need_tlb_flush = 0; | |
734 | ||
735 | while ((spte = rmap_next(kvm, rmapp, NULL))) { | |
736 | BUG_ON(!(*spte & PT_PRESENT_MASK)); | |
737 | rmap_printk("kvm_rmap_unmap_hva: spte %p %llx\n", spte, *spte); | |
738 | rmap_remove(kvm, spte); | |
d555c333 | 739 | __set_spte(spte, shadow_trap_nonpresent_pte); |
e930bffe AA |
740 | need_tlb_flush = 1; |
741 | } | |
742 | return need_tlb_flush; | |
743 | } | |
744 | ||
8a8365c5 FD |
745 | static int kvm_set_pte_rmapp(struct kvm *kvm, unsigned long *rmapp, |
746 | unsigned long data) | |
3da0dd43 IE |
747 | { |
748 | int need_flush = 0; | |
749 | u64 *spte, new_spte; | |
750 | pte_t *ptep = (pte_t *)data; | |
751 | pfn_t new_pfn; | |
752 | ||
753 | WARN_ON(pte_huge(*ptep)); | |
754 | new_pfn = pte_pfn(*ptep); | |
755 | spte = rmap_next(kvm, rmapp, NULL); | |
756 | while (spte) { | |
757 | BUG_ON(!is_shadow_present_pte(*spte)); | |
758 | rmap_printk("kvm_set_pte_rmapp: spte %p %llx\n", spte, *spte); | |
759 | need_flush = 1; | |
760 | if (pte_write(*ptep)) { | |
761 | rmap_remove(kvm, spte); | |
762 | __set_spte(spte, shadow_trap_nonpresent_pte); | |
763 | spte = rmap_next(kvm, rmapp, NULL); | |
764 | } else { | |
765 | new_spte = *spte &~ (PT64_BASE_ADDR_MASK); | |
766 | new_spte |= (u64)new_pfn << PAGE_SHIFT; | |
767 | ||
768 | new_spte &= ~PT_WRITABLE_MASK; | |
769 | new_spte &= ~SPTE_HOST_WRITEABLE; | |
8dae4445 | 770 | if (is_writable_pte(*spte)) |
3da0dd43 IE |
771 | kvm_set_pfn_dirty(spte_to_pfn(*spte)); |
772 | __set_spte(spte, new_spte); | |
773 | spte = rmap_next(kvm, rmapp, spte); | |
774 | } | |
775 | } | |
776 | if (need_flush) | |
777 | kvm_flush_remote_tlbs(kvm); | |
778 | ||
779 | return 0; | |
780 | } | |
781 | ||
8a8365c5 FD |
782 | static int kvm_handle_hva(struct kvm *kvm, unsigned long hva, |
783 | unsigned long data, | |
3da0dd43 | 784 | int (*handler)(struct kvm *kvm, unsigned long *rmapp, |
8a8365c5 | 785 | unsigned long data)) |
e930bffe | 786 | { |
852e3c19 | 787 | int i, j; |
90bb6fc5 | 788 | int ret; |
e930bffe | 789 | int retval = 0; |
bc6678a3 MT |
790 | struct kvm_memslots *slots; |
791 | ||
90d83dc3 | 792 | slots = kvm_memslots(kvm); |
e930bffe | 793 | |
46a26bf5 MT |
794 | for (i = 0; i < slots->nmemslots; i++) { |
795 | struct kvm_memory_slot *memslot = &slots->memslots[i]; | |
e930bffe AA |
796 | unsigned long start = memslot->userspace_addr; |
797 | unsigned long end; | |
798 | ||
e930bffe AA |
799 | end = start + (memslot->npages << PAGE_SHIFT); |
800 | if (hva >= start && hva < end) { | |
801 | gfn_t gfn_offset = (hva - start) >> PAGE_SHIFT; | |
852e3c19 | 802 | |
90bb6fc5 | 803 | ret = handler(kvm, &memslot->rmap[gfn_offset], data); |
852e3c19 JR |
804 | |
805 | for (j = 0; j < KVM_NR_PAGE_SIZES - 1; ++j) { | |
806 | int idx = gfn_offset; | |
807 | idx /= KVM_PAGES_PER_HPAGE(PT_DIRECTORY_LEVEL + j); | |
90bb6fc5 | 808 | ret |= handler(kvm, |
3da0dd43 IE |
809 | &memslot->lpage_info[j][idx].rmap_pde, |
810 | data); | |
852e3c19 | 811 | } |
90bb6fc5 AK |
812 | trace_kvm_age_page(hva, memslot, ret); |
813 | retval |= ret; | |
e930bffe AA |
814 | } |
815 | } | |
816 | ||
817 | return retval; | |
818 | } | |
819 | ||
820 | int kvm_unmap_hva(struct kvm *kvm, unsigned long hva) | |
821 | { | |
3da0dd43 IE |
822 | return kvm_handle_hva(kvm, hva, 0, kvm_unmap_rmapp); |
823 | } | |
824 | ||
825 | void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte) | |
826 | { | |
8a8365c5 | 827 | kvm_handle_hva(kvm, hva, (unsigned long)&pte, kvm_set_pte_rmapp); |
e930bffe AA |
828 | } |
829 | ||
8a8365c5 FD |
830 | static int kvm_age_rmapp(struct kvm *kvm, unsigned long *rmapp, |
831 | unsigned long data) | |
e930bffe AA |
832 | { |
833 | u64 *spte; | |
834 | int young = 0; | |
835 | ||
6316e1c8 RR |
836 | /* |
837 | * Emulate the accessed bit for EPT, by checking if this page has | |
838 | * an EPT mapping, and clearing it if it does. On the next access, | |
839 | * a new EPT mapping will be established. | |
840 | * This has some overhead, but not as much as the cost of swapping | |
841 | * out actively used pages or breaking up actively used hugepages. | |
842 | */ | |
534e38b4 | 843 | if (!shadow_accessed_mask) |
6316e1c8 | 844 | return kvm_unmap_rmapp(kvm, rmapp, data); |
534e38b4 | 845 | |
e930bffe AA |
846 | spte = rmap_next(kvm, rmapp, NULL); |
847 | while (spte) { | |
848 | int _young; | |
849 | u64 _spte = *spte; | |
850 | BUG_ON(!(_spte & PT_PRESENT_MASK)); | |
851 | _young = _spte & PT_ACCESSED_MASK; | |
852 | if (_young) { | |
853 | young = 1; | |
854 | clear_bit(PT_ACCESSED_SHIFT, (unsigned long *)spte); | |
855 | } | |
856 | spte = rmap_next(kvm, rmapp, spte); | |
857 | } | |
858 | return young; | |
859 | } | |
860 | ||
53a27b39 MT |
861 | #define RMAP_RECYCLE_THRESHOLD 1000 |
862 | ||
852e3c19 | 863 | static void rmap_recycle(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn) |
53a27b39 MT |
864 | { |
865 | unsigned long *rmapp; | |
852e3c19 JR |
866 | struct kvm_mmu_page *sp; |
867 | ||
868 | sp = page_header(__pa(spte)); | |
53a27b39 MT |
869 | |
870 | gfn = unalias_gfn(vcpu->kvm, gfn); | |
852e3c19 | 871 | rmapp = gfn_to_rmap(vcpu->kvm, gfn, sp->role.level); |
53a27b39 | 872 | |
3da0dd43 | 873 | kvm_unmap_rmapp(vcpu->kvm, rmapp, 0); |
53a27b39 MT |
874 | kvm_flush_remote_tlbs(vcpu->kvm); |
875 | } | |
876 | ||
e930bffe AA |
877 | int kvm_age_hva(struct kvm *kvm, unsigned long hva) |
878 | { | |
3da0dd43 | 879 | return kvm_handle_hva(kvm, hva, 0, kvm_age_rmapp); |
e930bffe AA |
880 | } |
881 | ||
d6c69ee9 | 882 | #ifdef MMU_DEBUG |
47ad8e68 | 883 | static int is_empty_shadow_page(u64 *spt) |
6aa8b732 | 884 | { |
139bdb2d AK |
885 | u64 *pos; |
886 | u64 *end; | |
887 | ||
47ad8e68 | 888 | for (pos = spt, end = pos + PAGE_SIZE / sizeof(u64); pos != end; pos++) |
3c915510 | 889 | if (is_shadow_present_pte(*pos)) { |
b8688d51 | 890 | printk(KERN_ERR "%s: %p %llx\n", __func__, |
139bdb2d | 891 | pos, *pos); |
6aa8b732 | 892 | return 0; |
139bdb2d | 893 | } |
6aa8b732 AK |
894 | return 1; |
895 | } | |
d6c69ee9 | 896 | #endif |
6aa8b732 | 897 | |
4db35314 | 898 | static void kvm_mmu_free_page(struct kvm *kvm, struct kvm_mmu_page *sp) |
260746c0 | 899 | { |
4db35314 AK |
900 | ASSERT(is_empty_shadow_page(sp->spt)); |
901 | list_del(&sp->link); | |
902 | __free_page(virt_to_page(sp->spt)); | |
903 | __free_page(virt_to_page(sp->gfns)); | |
e8ad9a70 | 904 | kmem_cache_free(mmu_page_header_cache, sp); |
f05e70ac | 905 | ++kvm->arch.n_free_mmu_pages; |
260746c0 AK |
906 | } |
907 | ||
cea0f0e7 AK |
908 | static unsigned kvm_page_table_hashfn(gfn_t gfn) |
909 | { | |
1ae0a13d | 910 | return gfn & ((1 << KVM_MMU_HASH_SHIFT) - 1); |
cea0f0e7 AK |
911 | } |
912 | ||
25c0de2c AK |
913 | static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu, |
914 | u64 *parent_pte) | |
6aa8b732 | 915 | { |
4db35314 | 916 | struct kvm_mmu_page *sp; |
6aa8b732 | 917 | |
ad312c7c ZX |
918 | sp = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_header_cache, sizeof *sp); |
919 | sp->spt = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_cache, PAGE_SIZE); | |
920 | sp->gfns = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_cache, PAGE_SIZE); | |
4db35314 | 921 | set_page_private(virt_to_page(sp->spt), (unsigned long)sp); |
f05e70ac | 922 | list_add(&sp->link, &vcpu->kvm->arch.active_mmu_pages); |
291f26bc | 923 | bitmap_zero(sp->slot_bitmap, KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS); |
4db35314 AK |
924 | sp->multimapped = 0; |
925 | sp->parent_pte = parent_pte; | |
f05e70ac | 926 | --vcpu->kvm->arch.n_free_mmu_pages; |
4db35314 | 927 | return sp; |
6aa8b732 AK |
928 | } |
929 | ||
714b93da | 930 | static void mmu_page_add_parent_pte(struct kvm_vcpu *vcpu, |
4db35314 | 931 | struct kvm_mmu_page *sp, u64 *parent_pte) |
cea0f0e7 AK |
932 | { |
933 | struct kvm_pte_chain *pte_chain; | |
934 | struct hlist_node *node; | |
935 | int i; | |
936 | ||
937 | if (!parent_pte) | |
938 | return; | |
4db35314 AK |
939 | if (!sp->multimapped) { |
940 | u64 *old = sp->parent_pte; | |
cea0f0e7 AK |
941 | |
942 | if (!old) { | |
4db35314 | 943 | sp->parent_pte = parent_pte; |
cea0f0e7 AK |
944 | return; |
945 | } | |
4db35314 | 946 | sp->multimapped = 1; |
714b93da | 947 | pte_chain = mmu_alloc_pte_chain(vcpu); |
4db35314 AK |
948 | INIT_HLIST_HEAD(&sp->parent_ptes); |
949 | hlist_add_head(&pte_chain->link, &sp->parent_ptes); | |
cea0f0e7 AK |
950 | pte_chain->parent_ptes[0] = old; |
951 | } | |
4db35314 | 952 | hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link) { |
cea0f0e7 AK |
953 | if (pte_chain->parent_ptes[NR_PTE_CHAIN_ENTRIES-1]) |
954 | continue; | |
955 | for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) | |
956 | if (!pte_chain->parent_ptes[i]) { | |
957 | pte_chain->parent_ptes[i] = parent_pte; | |
958 | return; | |
959 | } | |
960 | } | |
714b93da | 961 | pte_chain = mmu_alloc_pte_chain(vcpu); |
cea0f0e7 | 962 | BUG_ON(!pte_chain); |
4db35314 | 963 | hlist_add_head(&pte_chain->link, &sp->parent_ptes); |
cea0f0e7 AK |
964 | pte_chain->parent_ptes[0] = parent_pte; |
965 | } | |
966 | ||
4db35314 | 967 | static void mmu_page_remove_parent_pte(struct kvm_mmu_page *sp, |
cea0f0e7 AK |
968 | u64 *parent_pte) |
969 | { | |
970 | struct kvm_pte_chain *pte_chain; | |
971 | struct hlist_node *node; | |
972 | int i; | |
973 | ||
4db35314 AK |
974 | if (!sp->multimapped) { |
975 | BUG_ON(sp->parent_pte != parent_pte); | |
976 | sp->parent_pte = NULL; | |
cea0f0e7 AK |
977 | return; |
978 | } | |
4db35314 | 979 | hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link) |
cea0f0e7 AK |
980 | for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) { |
981 | if (!pte_chain->parent_ptes[i]) | |
982 | break; | |
983 | if (pte_chain->parent_ptes[i] != parent_pte) | |
984 | continue; | |
697fe2e2 AK |
985 | while (i + 1 < NR_PTE_CHAIN_ENTRIES |
986 | && pte_chain->parent_ptes[i + 1]) { | |
cea0f0e7 AK |
987 | pte_chain->parent_ptes[i] |
988 | = pte_chain->parent_ptes[i + 1]; | |
989 | ++i; | |
990 | } | |
991 | pte_chain->parent_ptes[i] = NULL; | |
697fe2e2 AK |
992 | if (i == 0) { |
993 | hlist_del(&pte_chain->link); | |
90cb0529 | 994 | mmu_free_pte_chain(pte_chain); |
4db35314 AK |
995 | if (hlist_empty(&sp->parent_ptes)) { |
996 | sp->multimapped = 0; | |
997 | sp->parent_pte = NULL; | |
697fe2e2 AK |
998 | } |
999 | } | |
cea0f0e7 AK |
1000 | return; |
1001 | } | |
1002 | BUG(); | |
1003 | } | |
1004 | ||
ad8cfbe3 | 1005 | |
6b18493d | 1006 | static void mmu_parent_walk(struct kvm_mmu_page *sp, mmu_parent_walk_fn fn) |
ad8cfbe3 MT |
1007 | { |
1008 | struct kvm_pte_chain *pte_chain; | |
1009 | struct hlist_node *node; | |
1010 | struct kvm_mmu_page *parent_sp; | |
1011 | int i; | |
1012 | ||
1013 | if (!sp->multimapped && sp->parent_pte) { | |
1014 | parent_sp = page_header(__pa(sp->parent_pte)); | |
6b18493d XG |
1015 | fn(parent_sp); |
1016 | mmu_parent_walk(parent_sp, fn); | |
ad8cfbe3 MT |
1017 | return; |
1018 | } | |
1019 | hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link) | |
1020 | for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) { | |
1021 | if (!pte_chain->parent_ptes[i]) | |
1022 | break; | |
1023 | parent_sp = page_header(__pa(pte_chain->parent_ptes[i])); | |
6b18493d XG |
1024 | fn(parent_sp); |
1025 | mmu_parent_walk(parent_sp, fn); | |
ad8cfbe3 MT |
1026 | } |
1027 | } | |
1028 | ||
0074ff63 MT |
1029 | static void kvm_mmu_update_unsync_bitmap(u64 *spte) |
1030 | { | |
1031 | unsigned int index; | |
1032 | struct kvm_mmu_page *sp = page_header(__pa(spte)); | |
1033 | ||
1034 | index = spte - sp->spt; | |
60c8aec6 MT |
1035 | if (!__test_and_set_bit(index, sp->unsync_child_bitmap)) |
1036 | sp->unsync_children++; | |
1037 | WARN_ON(!sp->unsync_children); | |
0074ff63 MT |
1038 | } |
1039 | ||
1040 | static void kvm_mmu_update_parents_unsync(struct kvm_mmu_page *sp) | |
1041 | { | |
1042 | struct kvm_pte_chain *pte_chain; | |
1043 | struct hlist_node *node; | |
1044 | int i; | |
1045 | ||
1046 | if (!sp->parent_pte) | |
1047 | return; | |
1048 | ||
1049 | if (!sp->multimapped) { | |
1050 | kvm_mmu_update_unsync_bitmap(sp->parent_pte); | |
1051 | return; | |
1052 | } | |
1053 | ||
1054 | hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link) | |
1055 | for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) { | |
1056 | if (!pte_chain->parent_ptes[i]) | |
1057 | break; | |
1058 | kvm_mmu_update_unsync_bitmap(pte_chain->parent_ptes[i]); | |
1059 | } | |
1060 | } | |
1061 | ||
6b18493d | 1062 | static int unsync_walk_fn(struct kvm_mmu_page *sp) |
0074ff63 | 1063 | { |
0074ff63 MT |
1064 | kvm_mmu_update_parents_unsync(sp); |
1065 | return 1; | |
1066 | } | |
1067 | ||
6b18493d | 1068 | static void kvm_mmu_mark_parents_unsync(struct kvm_mmu_page *sp) |
0074ff63 | 1069 | { |
6b18493d | 1070 | mmu_parent_walk(sp, unsync_walk_fn); |
0074ff63 MT |
1071 | kvm_mmu_update_parents_unsync(sp); |
1072 | } | |
1073 | ||
d761a501 AK |
1074 | static void nonpaging_prefetch_page(struct kvm_vcpu *vcpu, |
1075 | struct kvm_mmu_page *sp) | |
1076 | { | |
1077 | int i; | |
1078 | ||
1079 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) | |
1080 | sp->spt[i] = shadow_trap_nonpresent_pte; | |
1081 | } | |
1082 | ||
e8bc217a MT |
1083 | static int nonpaging_sync_page(struct kvm_vcpu *vcpu, |
1084 | struct kvm_mmu_page *sp) | |
1085 | { | |
1086 | return 1; | |
1087 | } | |
1088 | ||
a7052897 MT |
1089 | static void nonpaging_invlpg(struct kvm_vcpu *vcpu, gva_t gva) |
1090 | { | |
1091 | } | |
1092 | ||
60c8aec6 MT |
1093 | #define KVM_PAGE_ARRAY_NR 16 |
1094 | ||
1095 | struct kvm_mmu_pages { | |
1096 | struct mmu_page_and_offset { | |
1097 | struct kvm_mmu_page *sp; | |
1098 | unsigned int idx; | |
1099 | } page[KVM_PAGE_ARRAY_NR]; | |
1100 | unsigned int nr; | |
1101 | }; | |
1102 | ||
0074ff63 MT |
1103 | #define for_each_unsync_children(bitmap, idx) \ |
1104 | for (idx = find_first_bit(bitmap, 512); \ | |
1105 | idx < 512; \ | |
1106 | idx = find_next_bit(bitmap, 512, idx+1)) | |
1107 | ||
cded19f3 HE |
1108 | static int mmu_pages_add(struct kvm_mmu_pages *pvec, struct kvm_mmu_page *sp, |
1109 | int idx) | |
4731d4c7 | 1110 | { |
60c8aec6 | 1111 | int i; |
4731d4c7 | 1112 | |
60c8aec6 MT |
1113 | if (sp->unsync) |
1114 | for (i=0; i < pvec->nr; i++) | |
1115 | if (pvec->page[i].sp == sp) | |
1116 | return 0; | |
1117 | ||
1118 | pvec->page[pvec->nr].sp = sp; | |
1119 | pvec->page[pvec->nr].idx = idx; | |
1120 | pvec->nr++; | |
1121 | return (pvec->nr == KVM_PAGE_ARRAY_NR); | |
1122 | } | |
1123 | ||
1124 | static int __mmu_unsync_walk(struct kvm_mmu_page *sp, | |
1125 | struct kvm_mmu_pages *pvec) | |
1126 | { | |
1127 | int i, ret, nr_unsync_leaf = 0; | |
4731d4c7 | 1128 | |
0074ff63 | 1129 | for_each_unsync_children(sp->unsync_child_bitmap, i) { |
4731d4c7 MT |
1130 | u64 ent = sp->spt[i]; |
1131 | ||
87917239 | 1132 | if (is_shadow_present_pte(ent) && !is_large_pte(ent)) { |
4731d4c7 MT |
1133 | struct kvm_mmu_page *child; |
1134 | child = page_header(ent & PT64_BASE_ADDR_MASK); | |
1135 | ||
1136 | if (child->unsync_children) { | |
60c8aec6 MT |
1137 | if (mmu_pages_add(pvec, child, i)) |
1138 | return -ENOSPC; | |
1139 | ||
1140 | ret = __mmu_unsync_walk(child, pvec); | |
1141 | if (!ret) | |
1142 | __clear_bit(i, sp->unsync_child_bitmap); | |
1143 | else if (ret > 0) | |
1144 | nr_unsync_leaf += ret; | |
1145 | else | |
4731d4c7 MT |
1146 | return ret; |
1147 | } | |
1148 | ||
1149 | if (child->unsync) { | |
60c8aec6 MT |
1150 | nr_unsync_leaf++; |
1151 | if (mmu_pages_add(pvec, child, i)) | |
1152 | return -ENOSPC; | |
4731d4c7 MT |
1153 | } |
1154 | } | |
1155 | } | |
1156 | ||
0074ff63 | 1157 | if (find_first_bit(sp->unsync_child_bitmap, 512) == 512) |
4731d4c7 MT |
1158 | sp->unsync_children = 0; |
1159 | ||
60c8aec6 MT |
1160 | return nr_unsync_leaf; |
1161 | } | |
1162 | ||
1163 | static int mmu_unsync_walk(struct kvm_mmu_page *sp, | |
1164 | struct kvm_mmu_pages *pvec) | |
1165 | { | |
1166 | if (!sp->unsync_children) | |
1167 | return 0; | |
1168 | ||
1169 | mmu_pages_add(pvec, sp, 0); | |
1170 | return __mmu_unsync_walk(sp, pvec); | |
4731d4c7 MT |
1171 | } |
1172 | ||
4731d4c7 MT |
1173 | static void kvm_unlink_unsync_page(struct kvm *kvm, struct kvm_mmu_page *sp) |
1174 | { | |
1175 | WARN_ON(!sp->unsync); | |
5e1b3ddb | 1176 | trace_kvm_mmu_sync_page(sp); |
4731d4c7 MT |
1177 | sp->unsync = 0; |
1178 | --kvm->stat.mmu_unsync; | |
1179 | } | |
1180 | ||
1181 | static int kvm_mmu_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp); | |
1182 | ||
1d9dc7e0 XG |
1183 | static int __kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp, |
1184 | bool clear_unsync) | |
4731d4c7 | 1185 | { |
5b7e0102 | 1186 | if (sp->role.cr4_pae != !!is_pae(vcpu)) { |
4731d4c7 MT |
1187 | kvm_mmu_zap_page(vcpu->kvm, sp); |
1188 | return 1; | |
1189 | } | |
1190 | ||
1d9dc7e0 XG |
1191 | if (clear_unsync) { |
1192 | if (rmap_write_protect(vcpu->kvm, sp->gfn)) | |
1193 | kvm_flush_remote_tlbs(vcpu->kvm); | |
1194 | kvm_unlink_unsync_page(vcpu->kvm, sp); | |
1195 | } | |
1196 | ||
4731d4c7 MT |
1197 | if (vcpu->arch.mmu.sync_page(vcpu, sp)) { |
1198 | kvm_mmu_zap_page(vcpu->kvm, sp); | |
1199 | return 1; | |
1200 | } | |
1201 | ||
1202 | kvm_mmu_flush_tlb(vcpu); | |
4731d4c7 MT |
1203 | return 0; |
1204 | } | |
1205 | ||
1d9dc7e0 XG |
1206 | static void mmu_convert_notrap(struct kvm_mmu_page *sp); |
1207 | static int kvm_sync_page_transient(struct kvm_vcpu *vcpu, | |
1208 | struct kvm_mmu_page *sp) | |
1209 | { | |
1210 | int ret; | |
1211 | ||
1212 | ret = __kvm_sync_page(vcpu, sp, false); | |
1213 | if (!ret) | |
1214 | mmu_convert_notrap(sp); | |
1215 | return ret; | |
1216 | } | |
1217 | ||
1218 | static int kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp) | |
1219 | { | |
1220 | return __kvm_sync_page(vcpu, sp, true); | |
1221 | } | |
1222 | ||
60c8aec6 MT |
1223 | struct mmu_page_path { |
1224 | struct kvm_mmu_page *parent[PT64_ROOT_LEVEL-1]; | |
1225 | unsigned int idx[PT64_ROOT_LEVEL-1]; | |
4731d4c7 MT |
1226 | }; |
1227 | ||
60c8aec6 MT |
1228 | #define for_each_sp(pvec, sp, parents, i) \ |
1229 | for (i = mmu_pages_next(&pvec, &parents, -1), \ | |
1230 | sp = pvec.page[i].sp; \ | |
1231 | i < pvec.nr && ({ sp = pvec.page[i].sp; 1;}); \ | |
1232 | i = mmu_pages_next(&pvec, &parents, i)) | |
1233 | ||
cded19f3 HE |
1234 | static int mmu_pages_next(struct kvm_mmu_pages *pvec, |
1235 | struct mmu_page_path *parents, | |
1236 | int i) | |
60c8aec6 MT |
1237 | { |
1238 | int n; | |
1239 | ||
1240 | for (n = i+1; n < pvec->nr; n++) { | |
1241 | struct kvm_mmu_page *sp = pvec->page[n].sp; | |
1242 | ||
1243 | if (sp->role.level == PT_PAGE_TABLE_LEVEL) { | |
1244 | parents->idx[0] = pvec->page[n].idx; | |
1245 | return n; | |
1246 | } | |
1247 | ||
1248 | parents->parent[sp->role.level-2] = sp; | |
1249 | parents->idx[sp->role.level-1] = pvec->page[n].idx; | |
1250 | } | |
1251 | ||
1252 | return n; | |
1253 | } | |
1254 | ||
cded19f3 | 1255 | static void mmu_pages_clear_parents(struct mmu_page_path *parents) |
4731d4c7 | 1256 | { |
60c8aec6 MT |
1257 | struct kvm_mmu_page *sp; |
1258 | unsigned int level = 0; | |
1259 | ||
1260 | do { | |
1261 | unsigned int idx = parents->idx[level]; | |
4731d4c7 | 1262 | |
60c8aec6 MT |
1263 | sp = parents->parent[level]; |
1264 | if (!sp) | |
1265 | return; | |
1266 | ||
1267 | --sp->unsync_children; | |
1268 | WARN_ON((int)sp->unsync_children < 0); | |
1269 | __clear_bit(idx, sp->unsync_child_bitmap); | |
1270 | level++; | |
1271 | } while (level < PT64_ROOT_LEVEL-1 && !sp->unsync_children); | |
4731d4c7 MT |
1272 | } |
1273 | ||
60c8aec6 MT |
1274 | static void kvm_mmu_pages_init(struct kvm_mmu_page *parent, |
1275 | struct mmu_page_path *parents, | |
1276 | struct kvm_mmu_pages *pvec) | |
4731d4c7 | 1277 | { |
60c8aec6 MT |
1278 | parents->parent[parent->role.level-1] = NULL; |
1279 | pvec->nr = 0; | |
1280 | } | |
4731d4c7 | 1281 | |
60c8aec6 MT |
1282 | static void mmu_sync_children(struct kvm_vcpu *vcpu, |
1283 | struct kvm_mmu_page *parent) | |
1284 | { | |
1285 | int i; | |
1286 | struct kvm_mmu_page *sp; | |
1287 | struct mmu_page_path parents; | |
1288 | struct kvm_mmu_pages pages; | |
1289 | ||
1290 | kvm_mmu_pages_init(parent, &parents, &pages); | |
1291 | while (mmu_unsync_walk(parent, &pages)) { | |
b1a36821 MT |
1292 | int protected = 0; |
1293 | ||
1294 | for_each_sp(pages, sp, parents, i) | |
1295 | protected |= rmap_write_protect(vcpu->kvm, sp->gfn); | |
1296 | ||
1297 | if (protected) | |
1298 | kvm_flush_remote_tlbs(vcpu->kvm); | |
1299 | ||
60c8aec6 MT |
1300 | for_each_sp(pages, sp, parents, i) { |
1301 | kvm_sync_page(vcpu, sp); | |
1302 | mmu_pages_clear_parents(&parents); | |
1303 | } | |
4731d4c7 | 1304 | cond_resched_lock(&vcpu->kvm->mmu_lock); |
60c8aec6 MT |
1305 | kvm_mmu_pages_init(parent, &parents, &pages); |
1306 | } | |
4731d4c7 MT |
1307 | } |
1308 | ||
cea0f0e7 AK |
1309 | static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu, |
1310 | gfn_t gfn, | |
1311 | gva_t gaddr, | |
1312 | unsigned level, | |
f6e2c02b | 1313 | int direct, |
41074d07 | 1314 | unsigned access, |
f7d9c7b7 | 1315 | u64 *parent_pte) |
cea0f0e7 AK |
1316 | { |
1317 | union kvm_mmu_page_role role; | |
1318 | unsigned index; | |
1319 | unsigned quadrant; | |
1320 | struct hlist_head *bucket; | |
e02aa901 | 1321 | struct kvm_mmu_page *sp, *unsync_sp = NULL; |
4731d4c7 | 1322 | struct hlist_node *node, *tmp; |
cea0f0e7 | 1323 | |
a770f6f2 | 1324 | role = vcpu->arch.mmu.base_role; |
cea0f0e7 | 1325 | role.level = level; |
f6e2c02b | 1326 | role.direct = direct; |
84b0c8c6 | 1327 | if (role.direct) |
5b7e0102 | 1328 | role.cr4_pae = 0; |
41074d07 | 1329 | role.access = access; |
ad312c7c | 1330 | if (vcpu->arch.mmu.root_level <= PT32_ROOT_LEVEL) { |
cea0f0e7 AK |
1331 | quadrant = gaddr >> (PAGE_SHIFT + (PT64_PT_BITS * level)); |
1332 | quadrant &= (1 << ((PT32_PT_BITS - PT64_PT_BITS) * level)) - 1; | |
1333 | role.quadrant = quadrant; | |
1334 | } | |
1ae0a13d | 1335 | index = kvm_page_table_hashfn(gfn); |
f05e70ac | 1336 | bucket = &vcpu->kvm->arch.mmu_page_hash[index]; |
4731d4c7 MT |
1337 | hlist_for_each_entry_safe(sp, node, tmp, bucket, hash_link) |
1338 | if (sp->gfn == gfn) { | |
1339 | if (sp->unsync) | |
e02aa901 | 1340 | unsync_sp = sp; |
4731d4c7 MT |
1341 | |
1342 | if (sp->role.word != role.word) | |
1343 | continue; | |
1344 | ||
e02aa901 XG |
1345 | if (!direct && unsync_sp && |
1346 | kvm_sync_page_transient(vcpu, unsync_sp)) { | |
1347 | unsync_sp = NULL; | |
1348 | break; | |
1349 | } | |
1350 | ||
4db35314 | 1351 | mmu_page_add_parent_pte(vcpu, sp, parent_pte); |
0074ff63 MT |
1352 | if (sp->unsync_children) { |
1353 | set_bit(KVM_REQ_MMU_SYNC, &vcpu->requests); | |
6b18493d | 1354 | kvm_mmu_mark_parents_unsync(sp); |
e02aa901 XG |
1355 | } else if (sp->unsync) |
1356 | kvm_mmu_mark_parents_unsync(sp); | |
1357 | ||
f691fe1d | 1358 | trace_kvm_mmu_get_page(sp, false); |
4db35314 | 1359 | return sp; |
cea0f0e7 | 1360 | } |
e02aa901 XG |
1361 | if (!direct && unsync_sp) |
1362 | kvm_sync_page(vcpu, unsync_sp); | |
1363 | ||
dfc5aa00 | 1364 | ++vcpu->kvm->stat.mmu_cache_miss; |
4db35314 AK |
1365 | sp = kvm_mmu_alloc_page(vcpu, parent_pte); |
1366 | if (!sp) | |
1367 | return sp; | |
4db35314 AK |
1368 | sp->gfn = gfn; |
1369 | sp->role = role; | |
1370 | hlist_add_head(&sp->hash_link, bucket); | |
f6e2c02b | 1371 | if (!direct) { |
b1a36821 MT |
1372 | if (rmap_write_protect(vcpu->kvm, gfn)) |
1373 | kvm_flush_remote_tlbs(vcpu->kvm); | |
4731d4c7 MT |
1374 | account_shadowed(vcpu->kvm, gfn); |
1375 | } | |
131d8279 AK |
1376 | if (shadow_trap_nonpresent_pte != shadow_notrap_nonpresent_pte) |
1377 | vcpu->arch.mmu.prefetch_page(vcpu, sp); | |
1378 | else | |
1379 | nonpaging_prefetch_page(vcpu, sp); | |
f691fe1d | 1380 | trace_kvm_mmu_get_page(sp, true); |
4db35314 | 1381 | return sp; |
cea0f0e7 AK |
1382 | } |
1383 | ||
2d11123a AK |
1384 | static void shadow_walk_init(struct kvm_shadow_walk_iterator *iterator, |
1385 | struct kvm_vcpu *vcpu, u64 addr) | |
1386 | { | |
1387 | iterator->addr = addr; | |
1388 | iterator->shadow_addr = vcpu->arch.mmu.root_hpa; | |
1389 | iterator->level = vcpu->arch.mmu.shadow_root_level; | |
1390 | if (iterator->level == PT32E_ROOT_LEVEL) { | |
1391 | iterator->shadow_addr | |
1392 | = vcpu->arch.mmu.pae_root[(addr >> 30) & 3]; | |
1393 | iterator->shadow_addr &= PT64_BASE_ADDR_MASK; | |
1394 | --iterator->level; | |
1395 | if (!iterator->shadow_addr) | |
1396 | iterator->level = 0; | |
1397 | } | |
1398 | } | |
1399 | ||
1400 | static bool shadow_walk_okay(struct kvm_shadow_walk_iterator *iterator) | |
1401 | { | |
1402 | if (iterator->level < PT_PAGE_TABLE_LEVEL) | |
1403 | return false; | |
4d88954d MT |
1404 | |
1405 | if (iterator->level == PT_PAGE_TABLE_LEVEL) | |
1406 | if (is_large_pte(*iterator->sptep)) | |
1407 | return false; | |
1408 | ||
2d11123a AK |
1409 | iterator->index = SHADOW_PT_INDEX(iterator->addr, iterator->level); |
1410 | iterator->sptep = ((u64 *)__va(iterator->shadow_addr)) + iterator->index; | |
1411 | return true; | |
1412 | } | |
1413 | ||
1414 | static void shadow_walk_next(struct kvm_shadow_walk_iterator *iterator) | |
1415 | { | |
1416 | iterator->shadow_addr = *iterator->sptep & PT64_BASE_ADDR_MASK; | |
1417 | --iterator->level; | |
1418 | } | |
1419 | ||
90cb0529 | 1420 | static void kvm_mmu_page_unlink_children(struct kvm *kvm, |
4db35314 | 1421 | struct kvm_mmu_page *sp) |
a436036b | 1422 | { |
697fe2e2 AK |
1423 | unsigned i; |
1424 | u64 *pt; | |
1425 | u64 ent; | |
1426 | ||
4db35314 | 1427 | pt = sp->spt; |
697fe2e2 | 1428 | |
697fe2e2 AK |
1429 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) { |
1430 | ent = pt[i]; | |
1431 | ||
05da4558 | 1432 | if (is_shadow_present_pte(ent)) { |
776e6633 | 1433 | if (!is_last_spte(ent, sp->role.level)) { |
05da4558 MT |
1434 | ent &= PT64_BASE_ADDR_MASK; |
1435 | mmu_page_remove_parent_pte(page_header(ent), | |
1436 | &pt[i]); | |
1437 | } else { | |
776e6633 MT |
1438 | if (is_large_pte(ent)) |
1439 | --kvm->stat.lpages; | |
05da4558 MT |
1440 | rmap_remove(kvm, &pt[i]); |
1441 | } | |
1442 | } | |
c7addb90 | 1443 | pt[i] = shadow_trap_nonpresent_pte; |
697fe2e2 | 1444 | } |
a436036b AK |
1445 | } |
1446 | ||
4db35314 | 1447 | static void kvm_mmu_put_page(struct kvm_mmu_page *sp, u64 *parent_pte) |
cea0f0e7 | 1448 | { |
4db35314 | 1449 | mmu_page_remove_parent_pte(sp, parent_pte); |
a436036b AK |
1450 | } |
1451 | ||
12b7d28f AK |
1452 | static void kvm_mmu_reset_last_pte_updated(struct kvm *kvm) |
1453 | { | |
1454 | int i; | |
988a2cae | 1455 | struct kvm_vcpu *vcpu; |
12b7d28f | 1456 | |
988a2cae GN |
1457 | kvm_for_each_vcpu(i, vcpu, kvm) |
1458 | vcpu->arch.last_pte_updated = NULL; | |
12b7d28f AK |
1459 | } |
1460 | ||
31aa2b44 | 1461 | static void kvm_mmu_unlink_parents(struct kvm *kvm, struct kvm_mmu_page *sp) |
a436036b AK |
1462 | { |
1463 | u64 *parent_pte; | |
1464 | ||
4db35314 AK |
1465 | while (sp->multimapped || sp->parent_pte) { |
1466 | if (!sp->multimapped) | |
1467 | parent_pte = sp->parent_pte; | |
a436036b AK |
1468 | else { |
1469 | struct kvm_pte_chain *chain; | |
1470 | ||
4db35314 | 1471 | chain = container_of(sp->parent_ptes.first, |
a436036b AK |
1472 | struct kvm_pte_chain, link); |
1473 | parent_pte = chain->parent_ptes[0]; | |
1474 | } | |
697fe2e2 | 1475 | BUG_ON(!parent_pte); |
4db35314 | 1476 | kvm_mmu_put_page(sp, parent_pte); |
d555c333 | 1477 | __set_spte(parent_pte, shadow_trap_nonpresent_pte); |
a436036b | 1478 | } |
31aa2b44 AK |
1479 | } |
1480 | ||
60c8aec6 MT |
1481 | static int mmu_zap_unsync_children(struct kvm *kvm, |
1482 | struct kvm_mmu_page *parent) | |
4731d4c7 | 1483 | { |
60c8aec6 MT |
1484 | int i, zapped = 0; |
1485 | struct mmu_page_path parents; | |
1486 | struct kvm_mmu_pages pages; | |
4731d4c7 | 1487 | |
60c8aec6 | 1488 | if (parent->role.level == PT_PAGE_TABLE_LEVEL) |
4731d4c7 | 1489 | return 0; |
60c8aec6 MT |
1490 | |
1491 | kvm_mmu_pages_init(parent, &parents, &pages); | |
1492 | while (mmu_unsync_walk(parent, &pages)) { | |
1493 | struct kvm_mmu_page *sp; | |
1494 | ||
1495 | for_each_sp(pages, sp, parents, i) { | |
1496 | kvm_mmu_zap_page(kvm, sp); | |
1497 | mmu_pages_clear_parents(&parents); | |
77662e00 | 1498 | zapped++; |
60c8aec6 | 1499 | } |
60c8aec6 MT |
1500 | kvm_mmu_pages_init(parent, &parents, &pages); |
1501 | } | |
1502 | ||
1503 | return zapped; | |
4731d4c7 MT |
1504 | } |
1505 | ||
07385413 | 1506 | static int kvm_mmu_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp) |
31aa2b44 | 1507 | { |
4731d4c7 | 1508 | int ret; |
f691fe1d AK |
1509 | |
1510 | trace_kvm_mmu_zap_page(sp); | |
31aa2b44 | 1511 | ++kvm->stat.mmu_shadow_zapped; |
4731d4c7 | 1512 | ret = mmu_zap_unsync_children(kvm, sp); |
4db35314 | 1513 | kvm_mmu_page_unlink_children(kvm, sp); |
31aa2b44 | 1514 | kvm_mmu_unlink_parents(kvm, sp); |
5b5c6a5a | 1515 | kvm_flush_remote_tlbs(kvm); |
f6e2c02b | 1516 | if (!sp->role.invalid && !sp->role.direct) |
5b5c6a5a | 1517 | unaccount_shadowed(kvm, sp->gfn); |
4731d4c7 MT |
1518 | if (sp->unsync) |
1519 | kvm_unlink_unsync_page(kvm, sp); | |
4db35314 | 1520 | if (!sp->root_count) { |
54a4f023 GJ |
1521 | /* Count self */ |
1522 | ret++; | |
4db35314 AK |
1523 | hlist_del(&sp->hash_link); |
1524 | kvm_mmu_free_page(kvm, sp); | |
2e53d63a | 1525 | } else { |
2e53d63a | 1526 | sp->role.invalid = 1; |
5b5c6a5a | 1527 | list_move(&sp->link, &kvm->arch.active_mmu_pages); |
2e53d63a MT |
1528 | kvm_reload_remote_mmus(kvm); |
1529 | } | |
12b7d28f | 1530 | kvm_mmu_reset_last_pte_updated(kvm); |
4731d4c7 | 1531 | return ret; |
a436036b AK |
1532 | } |
1533 | ||
82ce2c96 IE |
1534 | /* |
1535 | * Changing the number of mmu pages allocated to the vm | |
1536 | * Note: if kvm_nr_mmu_pages is too small, you will get dead lock | |
1537 | */ | |
1538 | void kvm_mmu_change_mmu_pages(struct kvm *kvm, unsigned int kvm_nr_mmu_pages) | |
1539 | { | |
025dbbf3 MT |
1540 | int used_pages; |
1541 | ||
1542 | used_pages = kvm->arch.n_alloc_mmu_pages - kvm->arch.n_free_mmu_pages; | |
1543 | used_pages = max(0, used_pages); | |
1544 | ||
82ce2c96 IE |
1545 | /* |
1546 | * If we set the number of mmu pages to be smaller be than the | |
1547 | * number of actived pages , we must to free some mmu pages before we | |
1548 | * change the value | |
1549 | */ | |
1550 | ||
025dbbf3 | 1551 | if (used_pages > kvm_nr_mmu_pages) { |
77662e00 XG |
1552 | while (used_pages > kvm_nr_mmu_pages && |
1553 | !list_empty(&kvm->arch.active_mmu_pages)) { | |
82ce2c96 IE |
1554 | struct kvm_mmu_page *page; |
1555 | ||
f05e70ac | 1556 | page = container_of(kvm->arch.active_mmu_pages.prev, |
82ce2c96 | 1557 | struct kvm_mmu_page, link); |
77662e00 | 1558 | used_pages -= kvm_mmu_zap_page(kvm, page); |
82ce2c96 | 1559 | } |
77662e00 | 1560 | kvm_nr_mmu_pages = used_pages; |
f05e70ac | 1561 | kvm->arch.n_free_mmu_pages = 0; |
82ce2c96 IE |
1562 | } |
1563 | else | |
f05e70ac ZX |
1564 | kvm->arch.n_free_mmu_pages += kvm_nr_mmu_pages |
1565 | - kvm->arch.n_alloc_mmu_pages; | |
82ce2c96 | 1566 | |
f05e70ac | 1567 | kvm->arch.n_alloc_mmu_pages = kvm_nr_mmu_pages; |
82ce2c96 IE |
1568 | } |
1569 | ||
f67a46f4 | 1570 | static int kvm_mmu_unprotect_page(struct kvm *kvm, gfn_t gfn) |
a436036b AK |
1571 | { |
1572 | unsigned index; | |
1573 | struct hlist_head *bucket; | |
4db35314 | 1574 | struct kvm_mmu_page *sp; |
a436036b AK |
1575 | struct hlist_node *node, *n; |
1576 | int r; | |
1577 | ||
b8688d51 | 1578 | pgprintk("%s: looking for gfn %lx\n", __func__, gfn); |
a436036b | 1579 | r = 0; |
1ae0a13d | 1580 | index = kvm_page_table_hashfn(gfn); |
f05e70ac | 1581 | bucket = &kvm->arch.mmu_page_hash[index]; |
3246af0e | 1582 | restart: |
4db35314 | 1583 | hlist_for_each_entry_safe(sp, node, n, bucket, hash_link) |
f6e2c02b | 1584 | if (sp->gfn == gfn && !sp->role.direct) { |
b8688d51 | 1585 | pgprintk("%s: gfn %lx role %x\n", __func__, gfn, |
4db35314 | 1586 | sp->role.word); |
a436036b | 1587 | r = 1; |
07385413 | 1588 | if (kvm_mmu_zap_page(kvm, sp)) |
3246af0e | 1589 | goto restart; |
a436036b AK |
1590 | } |
1591 | return r; | |
cea0f0e7 AK |
1592 | } |
1593 | ||
f67a46f4 | 1594 | static void mmu_unshadow(struct kvm *kvm, gfn_t gfn) |
97a0a01e | 1595 | { |
4677a3b6 AK |
1596 | unsigned index; |
1597 | struct hlist_head *bucket; | |
4db35314 | 1598 | struct kvm_mmu_page *sp; |
4677a3b6 | 1599 | struct hlist_node *node, *nn; |
97a0a01e | 1600 | |
4677a3b6 AK |
1601 | index = kvm_page_table_hashfn(gfn); |
1602 | bucket = &kvm->arch.mmu_page_hash[index]; | |
3246af0e | 1603 | restart: |
4677a3b6 | 1604 | hlist_for_each_entry_safe(sp, node, nn, bucket, hash_link) { |
f6e2c02b | 1605 | if (sp->gfn == gfn && !sp->role.direct |
4677a3b6 AK |
1606 | && !sp->role.invalid) { |
1607 | pgprintk("%s: zap %lx %x\n", | |
1608 | __func__, gfn, sp->role.word); | |
77662e00 | 1609 | if (kvm_mmu_zap_page(kvm, sp)) |
3246af0e | 1610 | goto restart; |
4677a3b6 | 1611 | } |
97a0a01e AK |
1612 | } |
1613 | } | |
1614 | ||
38c335f1 | 1615 | static void page_header_update_slot(struct kvm *kvm, void *pte, gfn_t gfn) |
6aa8b732 | 1616 | { |
bc6678a3 | 1617 | int slot = memslot_id(kvm, gfn); |
4db35314 | 1618 | struct kvm_mmu_page *sp = page_header(__pa(pte)); |
6aa8b732 | 1619 | |
291f26bc | 1620 | __set_bit(slot, sp->slot_bitmap); |
6aa8b732 AK |
1621 | } |
1622 | ||
6844dec6 MT |
1623 | static void mmu_convert_notrap(struct kvm_mmu_page *sp) |
1624 | { | |
1625 | int i; | |
1626 | u64 *pt = sp->spt; | |
1627 | ||
1628 | if (shadow_trap_nonpresent_pte == shadow_notrap_nonpresent_pte) | |
1629 | return; | |
1630 | ||
1631 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) { | |
1632 | if (pt[i] == shadow_notrap_nonpresent_pte) | |
d555c333 | 1633 | __set_spte(&pt[i], shadow_trap_nonpresent_pte); |
6844dec6 MT |
1634 | } |
1635 | } | |
1636 | ||
74be52e3 SY |
1637 | /* |
1638 | * The function is based on mtrr_type_lookup() in | |
1639 | * arch/x86/kernel/cpu/mtrr/generic.c | |
1640 | */ | |
1641 | static int get_mtrr_type(struct mtrr_state_type *mtrr_state, | |
1642 | u64 start, u64 end) | |
1643 | { | |
1644 | int i; | |
1645 | u64 base, mask; | |
1646 | u8 prev_match, curr_match; | |
1647 | int num_var_ranges = KVM_NR_VAR_MTRR; | |
1648 | ||
1649 | if (!mtrr_state->enabled) | |
1650 | return 0xFF; | |
1651 | ||
1652 | /* Make end inclusive end, instead of exclusive */ | |
1653 | end--; | |
1654 | ||
1655 | /* Look in fixed ranges. Just return the type as per start */ | |
1656 | if (mtrr_state->have_fixed && (start < 0x100000)) { | |
1657 | int idx; | |
1658 | ||
1659 | if (start < 0x80000) { | |
1660 | idx = 0; | |
1661 | idx += (start >> 16); | |
1662 | return mtrr_state->fixed_ranges[idx]; | |
1663 | } else if (start < 0xC0000) { | |
1664 | idx = 1 * 8; | |
1665 | idx += ((start - 0x80000) >> 14); | |
1666 | return mtrr_state->fixed_ranges[idx]; | |
1667 | } else if (start < 0x1000000) { | |
1668 | idx = 3 * 8; | |
1669 | idx += ((start - 0xC0000) >> 12); | |
1670 | return mtrr_state->fixed_ranges[idx]; | |
1671 | } | |
1672 | } | |
1673 | ||
1674 | /* | |
1675 | * Look in variable ranges | |
1676 | * Look of multiple ranges matching this address and pick type | |
1677 | * as per MTRR precedence | |
1678 | */ | |
1679 | if (!(mtrr_state->enabled & 2)) | |
1680 | return mtrr_state->def_type; | |
1681 | ||
1682 | prev_match = 0xFF; | |
1683 | for (i = 0; i < num_var_ranges; ++i) { | |
1684 | unsigned short start_state, end_state; | |
1685 | ||
1686 | if (!(mtrr_state->var_ranges[i].mask_lo & (1 << 11))) | |
1687 | continue; | |
1688 | ||
1689 | base = (((u64)mtrr_state->var_ranges[i].base_hi) << 32) + | |
1690 | (mtrr_state->var_ranges[i].base_lo & PAGE_MASK); | |
1691 | mask = (((u64)mtrr_state->var_ranges[i].mask_hi) << 32) + | |
1692 | (mtrr_state->var_ranges[i].mask_lo & PAGE_MASK); | |
1693 | ||
1694 | start_state = ((start & mask) == (base & mask)); | |
1695 | end_state = ((end & mask) == (base & mask)); | |
1696 | if (start_state != end_state) | |
1697 | return 0xFE; | |
1698 | ||
1699 | if ((start & mask) != (base & mask)) | |
1700 | continue; | |
1701 | ||
1702 | curr_match = mtrr_state->var_ranges[i].base_lo & 0xff; | |
1703 | if (prev_match == 0xFF) { | |
1704 | prev_match = curr_match; | |
1705 | continue; | |
1706 | } | |
1707 | ||
1708 | if (prev_match == MTRR_TYPE_UNCACHABLE || | |
1709 | curr_match == MTRR_TYPE_UNCACHABLE) | |
1710 | return MTRR_TYPE_UNCACHABLE; | |
1711 | ||
1712 | if ((prev_match == MTRR_TYPE_WRBACK && | |
1713 | curr_match == MTRR_TYPE_WRTHROUGH) || | |
1714 | (prev_match == MTRR_TYPE_WRTHROUGH && | |
1715 | curr_match == MTRR_TYPE_WRBACK)) { | |
1716 | prev_match = MTRR_TYPE_WRTHROUGH; | |
1717 | curr_match = MTRR_TYPE_WRTHROUGH; | |
1718 | } | |
1719 | ||
1720 | if (prev_match != curr_match) | |
1721 | return MTRR_TYPE_UNCACHABLE; | |
1722 | } | |
1723 | ||
1724 | if (prev_match != 0xFF) | |
1725 | return prev_match; | |
1726 | ||
1727 | return mtrr_state->def_type; | |
1728 | } | |
1729 | ||
4b12f0de | 1730 | u8 kvm_get_guest_memory_type(struct kvm_vcpu *vcpu, gfn_t gfn) |
74be52e3 SY |
1731 | { |
1732 | u8 mtrr; | |
1733 | ||
1734 | mtrr = get_mtrr_type(&vcpu->arch.mtrr_state, gfn << PAGE_SHIFT, | |
1735 | (gfn << PAGE_SHIFT) + PAGE_SIZE); | |
1736 | if (mtrr == 0xfe || mtrr == 0xff) | |
1737 | mtrr = MTRR_TYPE_WRBACK; | |
1738 | return mtrr; | |
1739 | } | |
4b12f0de | 1740 | EXPORT_SYMBOL_GPL(kvm_get_guest_memory_type); |
74be52e3 | 1741 | |
9cf5cf5a XG |
1742 | static void __kvm_unsync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp) |
1743 | { | |
1744 | trace_kvm_mmu_unsync_page(sp); | |
1745 | ++vcpu->kvm->stat.mmu_unsync; | |
1746 | sp->unsync = 1; | |
1747 | ||
1748 | kvm_mmu_mark_parents_unsync(sp); | |
1749 | mmu_convert_notrap(sp); | |
1750 | } | |
1751 | ||
1752 | static void kvm_unsync_pages(struct kvm_vcpu *vcpu, gfn_t gfn) | |
4731d4c7 | 1753 | { |
4731d4c7 MT |
1754 | struct hlist_head *bucket; |
1755 | struct kvm_mmu_page *s; | |
1756 | struct hlist_node *node, *n; | |
9cf5cf5a | 1757 | unsigned index; |
4731d4c7 | 1758 | |
9cf5cf5a | 1759 | index = kvm_page_table_hashfn(gfn); |
4731d4c7 | 1760 | bucket = &vcpu->kvm->arch.mmu_page_hash[index]; |
9cf5cf5a | 1761 | |
4731d4c7 | 1762 | hlist_for_each_entry_safe(s, node, n, bucket, hash_link) { |
9cf5cf5a XG |
1763 | if (s->gfn != gfn || s->role.direct || s->unsync || |
1764 | s->role.invalid) | |
4731d4c7 | 1765 | continue; |
9cf5cf5a XG |
1766 | WARN_ON(s->role.level != PT_PAGE_TABLE_LEVEL); |
1767 | __kvm_unsync_page(vcpu, s); | |
4731d4c7 | 1768 | } |
4731d4c7 MT |
1769 | } |
1770 | ||
1771 | static int mmu_need_write_protect(struct kvm_vcpu *vcpu, gfn_t gfn, | |
1772 | bool can_unsync) | |
1773 | { | |
9cf5cf5a XG |
1774 | unsigned index; |
1775 | struct hlist_head *bucket; | |
1776 | struct kvm_mmu_page *s; | |
1777 | struct hlist_node *node, *n; | |
1778 | bool need_unsync = false; | |
1779 | ||
1780 | index = kvm_page_table_hashfn(gfn); | |
1781 | bucket = &vcpu->kvm->arch.mmu_page_hash[index]; | |
1782 | hlist_for_each_entry_safe(s, node, n, bucket, hash_link) { | |
1783 | if (s->gfn != gfn || s->role.direct || s->role.invalid) | |
1784 | continue; | |
4731d4c7 | 1785 | |
9cf5cf5a | 1786 | if (s->role.level != PT_PAGE_TABLE_LEVEL) |
4731d4c7 | 1787 | return 1; |
9cf5cf5a XG |
1788 | |
1789 | if (!need_unsync && !s->unsync) { | |
1790 | if (!can_unsync || !oos_shadow) | |
1791 | return 1; | |
1792 | need_unsync = true; | |
1793 | } | |
4731d4c7 | 1794 | } |
9cf5cf5a XG |
1795 | if (need_unsync) |
1796 | kvm_unsync_pages(vcpu, gfn); | |
4731d4c7 MT |
1797 | return 0; |
1798 | } | |
1799 | ||
d555c333 | 1800 | static int set_spte(struct kvm_vcpu *vcpu, u64 *sptep, |
1e73f9dd | 1801 | unsigned pte_access, int user_fault, |
852e3c19 | 1802 | int write_fault, int dirty, int level, |
c2d0ee46 | 1803 | gfn_t gfn, pfn_t pfn, bool speculative, |
1403283a | 1804 | bool can_unsync, bool reset_host_protection) |
1c4f1fd6 AK |
1805 | { |
1806 | u64 spte; | |
1e73f9dd | 1807 | int ret = 0; |
64d4d521 | 1808 | |
1c4f1fd6 AK |
1809 | /* |
1810 | * We don't set the accessed bit, since we sometimes want to see | |
1811 | * whether the guest actually used the pte (in order to detect | |
1812 | * demand paging). | |
1813 | */ | |
7b52345e | 1814 | spte = shadow_base_present_pte | shadow_dirty_mask; |
947da538 | 1815 | if (!speculative) |
3201b5d9 | 1816 | spte |= shadow_accessed_mask; |
1c4f1fd6 AK |
1817 | if (!dirty) |
1818 | pte_access &= ~ACC_WRITE_MASK; | |
7b52345e SY |
1819 | if (pte_access & ACC_EXEC_MASK) |
1820 | spte |= shadow_x_mask; | |
1821 | else | |
1822 | spte |= shadow_nx_mask; | |
1c4f1fd6 | 1823 | if (pte_access & ACC_USER_MASK) |
7b52345e | 1824 | spte |= shadow_user_mask; |
852e3c19 | 1825 | if (level > PT_PAGE_TABLE_LEVEL) |
05da4558 | 1826 | spte |= PT_PAGE_SIZE_MASK; |
4b12f0de SY |
1827 | if (tdp_enabled) |
1828 | spte |= kvm_x86_ops->get_mt_mask(vcpu, gfn, | |
1829 | kvm_is_mmio_pfn(pfn)); | |
1c4f1fd6 | 1830 | |
1403283a IE |
1831 | if (reset_host_protection) |
1832 | spte |= SPTE_HOST_WRITEABLE; | |
1833 | ||
35149e21 | 1834 | spte |= (u64)pfn << PAGE_SHIFT; |
1c4f1fd6 AK |
1835 | |
1836 | if ((pte_access & ACC_WRITE_MASK) | |
1837 | || (write_fault && !is_write_protection(vcpu) && !user_fault)) { | |
1c4f1fd6 | 1838 | |
852e3c19 JR |
1839 | if (level > PT_PAGE_TABLE_LEVEL && |
1840 | has_wrprotected_page(vcpu->kvm, gfn, level)) { | |
38187c83 | 1841 | ret = 1; |
6d74229f | 1842 | rmap_remove(vcpu->kvm, sptep); |
38187c83 MT |
1843 | spte = shadow_trap_nonpresent_pte; |
1844 | goto set_pte; | |
1845 | } | |
1846 | ||
1c4f1fd6 | 1847 | spte |= PT_WRITABLE_MASK; |
1c4f1fd6 | 1848 | |
69325a12 AK |
1849 | if (!tdp_enabled && !(pte_access & ACC_WRITE_MASK)) |
1850 | spte &= ~PT_USER_MASK; | |
1851 | ||
ecc5589f MT |
1852 | /* |
1853 | * Optimization: for pte sync, if spte was writable the hash | |
1854 | * lookup is unnecessary (and expensive). Write protection | |
1855 | * is responsibility of mmu_get_page / kvm_sync_page. | |
1856 | * Same reasoning can be applied to dirty page accounting. | |
1857 | */ | |
8dae4445 | 1858 | if (!can_unsync && is_writable_pte(*sptep)) |
ecc5589f MT |
1859 | goto set_pte; |
1860 | ||
4731d4c7 | 1861 | if (mmu_need_write_protect(vcpu, gfn, can_unsync)) { |
1c4f1fd6 | 1862 | pgprintk("%s: found shadow page for %lx, marking ro\n", |
b8688d51 | 1863 | __func__, gfn); |
1e73f9dd | 1864 | ret = 1; |
1c4f1fd6 | 1865 | pte_access &= ~ACC_WRITE_MASK; |
8dae4445 | 1866 | if (is_writable_pte(spte)) |
1c4f1fd6 | 1867 | spte &= ~PT_WRITABLE_MASK; |
1c4f1fd6 AK |
1868 | } |
1869 | } | |
1870 | ||
1c4f1fd6 AK |
1871 | if (pte_access & ACC_WRITE_MASK) |
1872 | mark_page_dirty(vcpu->kvm, gfn); | |
1873 | ||
38187c83 | 1874 | set_pte: |
d555c333 | 1875 | __set_spte(sptep, spte); |
1e73f9dd MT |
1876 | return ret; |
1877 | } | |
1878 | ||
d555c333 | 1879 | static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *sptep, |
1e73f9dd MT |
1880 | unsigned pt_access, unsigned pte_access, |
1881 | int user_fault, int write_fault, int dirty, | |
852e3c19 | 1882 | int *ptwrite, int level, gfn_t gfn, |
1403283a IE |
1883 | pfn_t pfn, bool speculative, |
1884 | bool reset_host_protection) | |
1e73f9dd MT |
1885 | { |
1886 | int was_rmapped = 0; | |
8dae4445 | 1887 | int was_writable = is_writable_pte(*sptep); |
53a27b39 | 1888 | int rmap_count; |
1e73f9dd MT |
1889 | |
1890 | pgprintk("%s: spte %llx access %x write_fault %d" | |
1891 | " user_fault %d gfn %lx\n", | |
d555c333 | 1892 | __func__, *sptep, pt_access, |
1e73f9dd MT |
1893 | write_fault, user_fault, gfn); |
1894 | ||
d555c333 | 1895 | if (is_rmap_spte(*sptep)) { |
1e73f9dd MT |
1896 | /* |
1897 | * If we overwrite a PTE page pointer with a 2MB PMD, unlink | |
1898 | * the parent of the now unreachable PTE. | |
1899 | */ | |
852e3c19 JR |
1900 | if (level > PT_PAGE_TABLE_LEVEL && |
1901 | !is_large_pte(*sptep)) { | |
1e73f9dd | 1902 | struct kvm_mmu_page *child; |
d555c333 | 1903 | u64 pte = *sptep; |
1e73f9dd MT |
1904 | |
1905 | child = page_header(pte & PT64_BASE_ADDR_MASK); | |
d555c333 | 1906 | mmu_page_remove_parent_pte(child, sptep); |
3be2264b MT |
1907 | __set_spte(sptep, shadow_trap_nonpresent_pte); |
1908 | kvm_flush_remote_tlbs(vcpu->kvm); | |
d555c333 | 1909 | } else if (pfn != spte_to_pfn(*sptep)) { |
1e73f9dd | 1910 | pgprintk("hfn old %lx new %lx\n", |
d555c333 AK |
1911 | spte_to_pfn(*sptep), pfn); |
1912 | rmap_remove(vcpu->kvm, sptep); | |
91546356 XG |
1913 | __set_spte(sptep, shadow_trap_nonpresent_pte); |
1914 | kvm_flush_remote_tlbs(vcpu->kvm); | |
6bed6b9e JR |
1915 | } else |
1916 | was_rmapped = 1; | |
1e73f9dd | 1917 | } |
852e3c19 | 1918 | |
d555c333 | 1919 | if (set_spte(vcpu, sptep, pte_access, user_fault, write_fault, |
1403283a IE |
1920 | dirty, level, gfn, pfn, speculative, true, |
1921 | reset_host_protection)) { | |
1e73f9dd MT |
1922 | if (write_fault) |
1923 | *ptwrite = 1; | |
a378b4e6 MT |
1924 | kvm_x86_ops->tlb_flush(vcpu); |
1925 | } | |
1e73f9dd | 1926 | |
d555c333 | 1927 | pgprintk("%s: setting spte %llx\n", __func__, *sptep); |
1e73f9dd | 1928 | pgprintk("instantiating %s PTE (%s) at %ld (%llx) addr %p\n", |
d555c333 | 1929 | is_large_pte(*sptep)? "2MB" : "4kB", |
a205bc19 JR |
1930 | *sptep & PT_PRESENT_MASK ?"RW":"R", gfn, |
1931 | *sptep, sptep); | |
d555c333 | 1932 | if (!was_rmapped && is_large_pte(*sptep)) |
05da4558 MT |
1933 | ++vcpu->kvm->stat.lpages; |
1934 | ||
d555c333 | 1935 | page_header_update_slot(vcpu->kvm, sptep, gfn); |
1c4f1fd6 | 1936 | if (!was_rmapped) { |
44ad9944 | 1937 | rmap_count = rmap_add(vcpu, sptep, gfn); |
acb66dd0 | 1938 | kvm_release_pfn_clean(pfn); |
53a27b39 | 1939 | if (rmap_count > RMAP_RECYCLE_THRESHOLD) |
852e3c19 | 1940 | rmap_recycle(vcpu, sptep, gfn); |
75e68e60 | 1941 | } else { |
8dae4445 | 1942 | if (was_writable) |
35149e21 | 1943 | kvm_release_pfn_dirty(pfn); |
75e68e60 | 1944 | else |
35149e21 | 1945 | kvm_release_pfn_clean(pfn); |
1c4f1fd6 | 1946 | } |
1b7fcd32 | 1947 | if (speculative) { |
d555c333 | 1948 | vcpu->arch.last_pte_updated = sptep; |
1b7fcd32 AK |
1949 | vcpu->arch.last_pte_gfn = gfn; |
1950 | } | |
1c4f1fd6 AK |
1951 | } |
1952 | ||
6aa8b732 AK |
1953 | static void nonpaging_new_cr3(struct kvm_vcpu *vcpu) |
1954 | { | |
1955 | } | |
1956 | ||
9f652d21 | 1957 | static int __direct_map(struct kvm_vcpu *vcpu, gpa_t v, int write, |
852e3c19 | 1958 | int level, gfn_t gfn, pfn_t pfn) |
140754bc | 1959 | { |
9f652d21 | 1960 | struct kvm_shadow_walk_iterator iterator; |
140754bc | 1961 | struct kvm_mmu_page *sp; |
9f652d21 | 1962 | int pt_write = 0; |
140754bc | 1963 | gfn_t pseudo_gfn; |
6aa8b732 | 1964 | |
9f652d21 | 1965 | for_each_shadow_entry(vcpu, (u64)gfn << PAGE_SHIFT, iterator) { |
852e3c19 | 1966 | if (iterator.level == level) { |
9f652d21 AK |
1967 | mmu_set_spte(vcpu, iterator.sptep, ACC_ALL, ACC_ALL, |
1968 | 0, write, 1, &pt_write, | |
1403283a | 1969 | level, gfn, pfn, false, true); |
9f652d21 AK |
1970 | ++vcpu->stat.pf_fixed; |
1971 | break; | |
6aa8b732 AK |
1972 | } |
1973 | ||
9f652d21 AK |
1974 | if (*iterator.sptep == shadow_trap_nonpresent_pte) { |
1975 | pseudo_gfn = (iterator.addr & PT64_DIR_BASE_ADDR_MASK) >> PAGE_SHIFT; | |
1976 | sp = kvm_mmu_get_page(vcpu, pseudo_gfn, iterator.addr, | |
1977 | iterator.level - 1, | |
1978 | 1, ACC_ALL, iterator.sptep); | |
1979 | if (!sp) { | |
1980 | pgprintk("nonpaging_map: ENOMEM\n"); | |
1981 | kvm_release_pfn_clean(pfn); | |
1982 | return -ENOMEM; | |
1983 | } | |
140754bc | 1984 | |
d555c333 AK |
1985 | __set_spte(iterator.sptep, |
1986 | __pa(sp->spt) | |
1987 | | PT_PRESENT_MASK | PT_WRITABLE_MASK | |
1988 | | shadow_user_mask | shadow_x_mask); | |
9f652d21 AK |
1989 | } |
1990 | } | |
1991 | return pt_write; | |
6aa8b732 AK |
1992 | } |
1993 | ||
bf998156 HY |
1994 | static void kvm_send_hwpoison_signal(struct kvm *kvm, gfn_t gfn) |
1995 | { | |
1996 | char buf[1]; | |
1997 | void __user *hva; | |
1998 | int r; | |
1999 | ||
2000 | /* Touch the page, so send SIGBUS */ | |
2001 | hva = (void __user *)gfn_to_hva(kvm, gfn); | |
2002 | r = copy_from_user(buf, hva, 1); | |
2003 | } | |
2004 | ||
2005 | static int kvm_handle_bad_page(struct kvm *kvm, gfn_t gfn, pfn_t pfn) | |
2006 | { | |
2007 | kvm_release_pfn_clean(pfn); | |
2008 | if (is_hwpoison_pfn(pfn)) { | |
2009 | kvm_send_hwpoison_signal(kvm, gfn); | |
2010 | return 0; | |
2011 | } | |
2012 | return 1; | |
2013 | } | |
2014 | ||
10589a46 MT |
2015 | static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, int write, gfn_t gfn) |
2016 | { | |
2017 | int r; | |
852e3c19 | 2018 | int level; |
35149e21 | 2019 | pfn_t pfn; |
e930bffe | 2020 | unsigned long mmu_seq; |
aaee2c94 | 2021 | |
852e3c19 JR |
2022 | level = mapping_level(vcpu, gfn); |
2023 | ||
2024 | /* | |
2025 | * This path builds a PAE pagetable - so we can map 2mb pages at | |
2026 | * maximum. Therefore check if the level is larger than that. | |
2027 | */ | |
2028 | if (level > PT_DIRECTORY_LEVEL) | |
2029 | level = PT_DIRECTORY_LEVEL; | |
2030 | ||
2031 | gfn &= ~(KVM_PAGES_PER_HPAGE(level) - 1); | |
05da4558 | 2032 | |
e930bffe | 2033 | mmu_seq = vcpu->kvm->mmu_notifier_seq; |
4c2155ce | 2034 | smp_rmb(); |
35149e21 | 2035 | pfn = gfn_to_pfn(vcpu->kvm, gfn); |
aaee2c94 | 2036 | |
d196e343 | 2037 | /* mmio */ |
bf998156 HY |
2038 | if (is_error_pfn(pfn)) |
2039 | return kvm_handle_bad_page(vcpu->kvm, gfn, pfn); | |
d196e343 | 2040 | |
aaee2c94 | 2041 | spin_lock(&vcpu->kvm->mmu_lock); |
e930bffe AA |
2042 | if (mmu_notifier_retry(vcpu, mmu_seq)) |
2043 | goto out_unlock; | |
eb787d10 | 2044 | kvm_mmu_free_some_pages(vcpu); |
852e3c19 | 2045 | r = __direct_map(vcpu, v, write, level, gfn, pfn); |
aaee2c94 MT |
2046 | spin_unlock(&vcpu->kvm->mmu_lock); |
2047 | ||
aaee2c94 | 2048 | |
10589a46 | 2049 | return r; |
e930bffe AA |
2050 | |
2051 | out_unlock: | |
2052 | spin_unlock(&vcpu->kvm->mmu_lock); | |
2053 | kvm_release_pfn_clean(pfn); | |
2054 | return 0; | |
10589a46 MT |
2055 | } |
2056 | ||
2057 | ||
17ac10ad AK |
2058 | static void mmu_free_roots(struct kvm_vcpu *vcpu) |
2059 | { | |
2060 | int i; | |
4db35314 | 2061 | struct kvm_mmu_page *sp; |
17ac10ad | 2062 | |
ad312c7c | 2063 | if (!VALID_PAGE(vcpu->arch.mmu.root_hpa)) |
7b53aa56 | 2064 | return; |
aaee2c94 | 2065 | spin_lock(&vcpu->kvm->mmu_lock); |
ad312c7c ZX |
2066 | if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) { |
2067 | hpa_t root = vcpu->arch.mmu.root_hpa; | |
17ac10ad | 2068 | |
4db35314 AK |
2069 | sp = page_header(root); |
2070 | --sp->root_count; | |
2e53d63a MT |
2071 | if (!sp->root_count && sp->role.invalid) |
2072 | kvm_mmu_zap_page(vcpu->kvm, sp); | |
ad312c7c | 2073 | vcpu->arch.mmu.root_hpa = INVALID_PAGE; |
aaee2c94 | 2074 | spin_unlock(&vcpu->kvm->mmu_lock); |
17ac10ad AK |
2075 | return; |
2076 | } | |
17ac10ad | 2077 | for (i = 0; i < 4; ++i) { |
ad312c7c | 2078 | hpa_t root = vcpu->arch.mmu.pae_root[i]; |
17ac10ad | 2079 | |
417726a3 | 2080 | if (root) { |
417726a3 | 2081 | root &= PT64_BASE_ADDR_MASK; |
4db35314 AK |
2082 | sp = page_header(root); |
2083 | --sp->root_count; | |
2e53d63a MT |
2084 | if (!sp->root_count && sp->role.invalid) |
2085 | kvm_mmu_zap_page(vcpu->kvm, sp); | |
417726a3 | 2086 | } |
ad312c7c | 2087 | vcpu->arch.mmu.pae_root[i] = INVALID_PAGE; |
17ac10ad | 2088 | } |
aaee2c94 | 2089 | spin_unlock(&vcpu->kvm->mmu_lock); |
ad312c7c | 2090 | vcpu->arch.mmu.root_hpa = INVALID_PAGE; |
17ac10ad AK |
2091 | } |
2092 | ||
8986ecc0 MT |
2093 | static int mmu_check_root(struct kvm_vcpu *vcpu, gfn_t root_gfn) |
2094 | { | |
2095 | int ret = 0; | |
2096 | ||
2097 | if (!kvm_is_visible_gfn(vcpu->kvm, root_gfn)) { | |
2098 | set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests); | |
2099 | ret = 1; | |
2100 | } | |
2101 | ||
2102 | return ret; | |
2103 | } | |
2104 | ||
2105 | static int mmu_alloc_roots(struct kvm_vcpu *vcpu) | |
17ac10ad AK |
2106 | { |
2107 | int i; | |
cea0f0e7 | 2108 | gfn_t root_gfn; |
4db35314 | 2109 | struct kvm_mmu_page *sp; |
f6e2c02b | 2110 | int direct = 0; |
6de4f3ad | 2111 | u64 pdptr; |
3bb65a22 | 2112 | |
ad312c7c | 2113 | root_gfn = vcpu->arch.cr3 >> PAGE_SHIFT; |
17ac10ad | 2114 | |
ad312c7c ZX |
2115 | if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) { |
2116 | hpa_t root = vcpu->arch.mmu.root_hpa; | |
17ac10ad AK |
2117 | |
2118 | ASSERT(!VALID_PAGE(root)); | |
8986ecc0 MT |
2119 | if (mmu_check_root(vcpu, root_gfn)) |
2120 | return 1; | |
5a7388c2 EN |
2121 | if (tdp_enabled) { |
2122 | direct = 1; | |
2123 | root_gfn = 0; | |
2124 | } | |
8facbbff | 2125 | spin_lock(&vcpu->kvm->mmu_lock); |
24955b6c | 2126 | kvm_mmu_free_some_pages(vcpu); |
4db35314 | 2127 | sp = kvm_mmu_get_page(vcpu, root_gfn, 0, |
f6e2c02b | 2128 | PT64_ROOT_LEVEL, direct, |
fb72d167 | 2129 | ACC_ALL, NULL); |
4db35314 AK |
2130 | root = __pa(sp->spt); |
2131 | ++sp->root_count; | |
8facbbff | 2132 | spin_unlock(&vcpu->kvm->mmu_lock); |
ad312c7c | 2133 | vcpu->arch.mmu.root_hpa = root; |
8986ecc0 | 2134 | return 0; |
17ac10ad | 2135 | } |
f6e2c02b | 2136 | direct = !is_paging(vcpu); |
17ac10ad | 2137 | for (i = 0; i < 4; ++i) { |
ad312c7c | 2138 | hpa_t root = vcpu->arch.mmu.pae_root[i]; |
17ac10ad AK |
2139 | |
2140 | ASSERT(!VALID_PAGE(root)); | |
ad312c7c | 2141 | if (vcpu->arch.mmu.root_level == PT32E_ROOT_LEVEL) { |
6de4f3ad | 2142 | pdptr = kvm_pdptr_read(vcpu, i); |
43a3795a | 2143 | if (!is_present_gpte(pdptr)) { |
ad312c7c | 2144 | vcpu->arch.mmu.pae_root[i] = 0; |
417726a3 AK |
2145 | continue; |
2146 | } | |
6de4f3ad | 2147 | root_gfn = pdptr >> PAGE_SHIFT; |
ad312c7c | 2148 | } else if (vcpu->arch.mmu.root_level == 0) |
cea0f0e7 | 2149 | root_gfn = 0; |
8986ecc0 MT |
2150 | if (mmu_check_root(vcpu, root_gfn)) |
2151 | return 1; | |
5a7388c2 EN |
2152 | if (tdp_enabled) { |
2153 | direct = 1; | |
2154 | root_gfn = i << 30; | |
2155 | } | |
8facbbff | 2156 | spin_lock(&vcpu->kvm->mmu_lock); |
24955b6c | 2157 | kvm_mmu_free_some_pages(vcpu); |
4db35314 | 2158 | sp = kvm_mmu_get_page(vcpu, root_gfn, i << 30, |
f6e2c02b | 2159 | PT32_ROOT_LEVEL, direct, |
f7d9c7b7 | 2160 | ACC_ALL, NULL); |
4db35314 AK |
2161 | root = __pa(sp->spt); |
2162 | ++sp->root_count; | |
8facbbff AK |
2163 | spin_unlock(&vcpu->kvm->mmu_lock); |
2164 | ||
ad312c7c | 2165 | vcpu->arch.mmu.pae_root[i] = root | PT_PRESENT_MASK; |
17ac10ad | 2166 | } |
ad312c7c | 2167 | vcpu->arch.mmu.root_hpa = __pa(vcpu->arch.mmu.pae_root); |
8986ecc0 | 2168 | return 0; |
17ac10ad AK |
2169 | } |
2170 | ||
0ba73cda MT |
2171 | static void mmu_sync_roots(struct kvm_vcpu *vcpu) |
2172 | { | |
2173 | int i; | |
2174 | struct kvm_mmu_page *sp; | |
2175 | ||
2176 | if (!VALID_PAGE(vcpu->arch.mmu.root_hpa)) | |
2177 | return; | |
2178 | if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) { | |
2179 | hpa_t root = vcpu->arch.mmu.root_hpa; | |
2180 | sp = page_header(root); | |
2181 | mmu_sync_children(vcpu, sp); | |
2182 | return; | |
2183 | } | |
2184 | for (i = 0; i < 4; ++i) { | |
2185 | hpa_t root = vcpu->arch.mmu.pae_root[i]; | |
2186 | ||
8986ecc0 | 2187 | if (root && VALID_PAGE(root)) { |
0ba73cda MT |
2188 | root &= PT64_BASE_ADDR_MASK; |
2189 | sp = page_header(root); | |
2190 | mmu_sync_children(vcpu, sp); | |
2191 | } | |
2192 | } | |
2193 | } | |
2194 | ||
2195 | void kvm_mmu_sync_roots(struct kvm_vcpu *vcpu) | |
2196 | { | |
2197 | spin_lock(&vcpu->kvm->mmu_lock); | |
2198 | mmu_sync_roots(vcpu); | |
6cffe8ca | 2199 | spin_unlock(&vcpu->kvm->mmu_lock); |
0ba73cda MT |
2200 | } |
2201 | ||
1871c602 GN |
2202 | static gpa_t nonpaging_gva_to_gpa(struct kvm_vcpu *vcpu, gva_t vaddr, |
2203 | u32 access, u32 *error) | |
6aa8b732 | 2204 | { |
1871c602 GN |
2205 | if (error) |
2206 | *error = 0; | |
6aa8b732 AK |
2207 | return vaddr; |
2208 | } | |
2209 | ||
2210 | static int nonpaging_page_fault(struct kvm_vcpu *vcpu, gva_t gva, | |
3f3e7124 | 2211 | u32 error_code) |
6aa8b732 | 2212 | { |
e833240f | 2213 | gfn_t gfn; |
e2dec939 | 2214 | int r; |
6aa8b732 | 2215 | |
b8688d51 | 2216 | pgprintk("%s: gva %lx error %x\n", __func__, gva, error_code); |
e2dec939 AK |
2217 | r = mmu_topup_memory_caches(vcpu); |
2218 | if (r) | |
2219 | return r; | |
714b93da | 2220 | |
6aa8b732 | 2221 | ASSERT(vcpu); |
ad312c7c | 2222 | ASSERT(VALID_PAGE(vcpu->arch.mmu.root_hpa)); |
6aa8b732 | 2223 | |
e833240f | 2224 | gfn = gva >> PAGE_SHIFT; |
6aa8b732 | 2225 | |
e833240f AK |
2226 | return nonpaging_map(vcpu, gva & PAGE_MASK, |
2227 | error_code & PFERR_WRITE_MASK, gfn); | |
6aa8b732 AK |
2228 | } |
2229 | ||
fb72d167 JR |
2230 | static int tdp_page_fault(struct kvm_vcpu *vcpu, gva_t gpa, |
2231 | u32 error_code) | |
2232 | { | |
35149e21 | 2233 | pfn_t pfn; |
fb72d167 | 2234 | int r; |
852e3c19 | 2235 | int level; |
05da4558 | 2236 | gfn_t gfn = gpa >> PAGE_SHIFT; |
e930bffe | 2237 | unsigned long mmu_seq; |
fb72d167 JR |
2238 | |
2239 | ASSERT(vcpu); | |
2240 | ASSERT(VALID_PAGE(vcpu->arch.mmu.root_hpa)); | |
2241 | ||
2242 | r = mmu_topup_memory_caches(vcpu); | |
2243 | if (r) | |
2244 | return r; | |
2245 | ||
852e3c19 JR |
2246 | level = mapping_level(vcpu, gfn); |
2247 | ||
2248 | gfn &= ~(KVM_PAGES_PER_HPAGE(level) - 1); | |
2249 | ||
e930bffe | 2250 | mmu_seq = vcpu->kvm->mmu_notifier_seq; |
4c2155ce | 2251 | smp_rmb(); |
35149e21 | 2252 | pfn = gfn_to_pfn(vcpu->kvm, gfn); |
bf998156 HY |
2253 | if (is_error_pfn(pfn)) |
2254 | return kvm_handle_bad_page(vcpu->kvm, gfn, pfn); | |
fb72d167 | 2255 | spin_lock(&vcpu->kvm->mmu_lock); |
e930bffe AA |
2256 | if (mmu_notifier_retry(vcpu, mmu_seq)) |
2257 | goto out_unlock; | |
fb72d167 JR |
2258 | kvm_mmu_free_some_pages(vcpu); |
2259 | r = __direct_map(vcpu, gpa, error_code & PFERR_WRITE_MASK, | |
852e3c19 | 2260 | level, gfn, pfn); |
fb72d167 | 2261 | spin_unlock(&vcpu->kvm->mmu_lock); |
fb72d167 JR |
2262 | |
2263 | return r; | |
e930bffe AA |
2264 | |
2265 | out_unlock: | |
2266 | spin_unlock(&vcpu->kvm->mmu_lock); | |
2267 | kvm_release_pfn_clean(pfn); | |
2268 | return 0; | |
fb72d167 JR |
2269 | } |
2270 | ||
6aa8b732 AK |
2271 | static void nonpaging_free(struct kvm_vcpu *vcpu) |
2272 | { | |
17ac10ad | 2273 | mmu_free_roots(vcpu); |
6aa8b732 AK |
2274 | } |
2275 | ||
2276 | static int nonpaging_init_context(struct kvm_vcpu *vcpu) | |
2277 | { | |
ad312c7c | 2278 | struct kvm_mmu *context = &vcpu->arch.mmu; |
6aa8b732 AK |
2279 | |
2280 | context->new_cr3 = nonpaging_new_cr3; | |
2281 | context->page_fault = nonpaging_page_fault; | |
6aa8b732 AK |
2282 | context->gva_to_gpa = nonpaging_gva_to_gpa; |
2283 | context->free = nonpaging_free; | |
c7addb90 | 2284 | context->prefetch_page = nonpaging_prefetch_page; |
e8bc217a | 2285 | context->sync_page = nonpaging_sync_page; |
a7052897 | 2286 | context->invlpg = nonpaging_invlpg; |
cea0f0e7 | 2287 | context->root_level = 0; |
6aa8b732 | 2288 | context->shadow_root_level = PT32E_ROOT_LEVEL; |
17c3ba9d | 2289 | context->root_hpa = INVALID_PAGE; |
6aa8b732 AK |
2290 | return 0; |
2291 | } | |
2292 | ||
d835dfec | 2293 | void kvm_mmu_flush_tlb(struct kvm_vcpu *vcpu) |
6aa8b732 | 2294 | { |
1165f5fe | 2295 | ++vcpu->stat.tlb_flush; |
cbdd1bea | 2296 | kvm_x86_ops->tlb_flush(vcpu); |
6aa8b732 AK |
2297 | } |
2298 | ||
2299 | static void paging_new_cr3(struct kvm_vcpu *vcpu) | |
2300 | { | |
b8688d51 | 2301 | pgprintk("%s: cr3 %lx\n", __func__, vcpu->arch.cr3); |
cea0f0e7 | 2302 | mmu_free_roots(vcpu); |
6aa8b732 AK |
2303 | } |
2304 | ||
6aa8b732 AK |
2305 | static void inject_page_fault(struct kvm_vcpu *vcpu, |
2306 | u64 addr, | |
2307 | u32 err_code) | |
2308 | { | |
c3c91fee | 2309 | kvm_inject_page_fault(vcpu, addr, err_code); |
6aa8b732 AK |
2310 | } |
2311 | ||
6aa8b732 AK |
2312 | static void paging_free(struct kvm_vcpu *vcpu) |
2313 | { | |
2314 | nonpaging_free(vcpu); | |
2315 | } | |
2316 | ||
82725b20 DE |
2317 | static bool is_rsvd_bits_set(struct kvm_vcpu *vcpu, u64 gpte, int level) |
2318 | { | |
2319 | int bit7; | |
2320 | ||
2321 | bit7 = (gpte >> 7) & 1; | |
2322 | return (gpte & vcpu->arch.mmu.rsvd_bits_mask[bit7][level-1]) != 0; | |
2323 | } | |
2324 | ||
6aa8b732 AK |
2325 | #define PTTYPE 64 |
2326 | #include "paging_tmpl.h" | |
2327 | #undef PTTYPE | |
2328 | ||
2329 | #define PTTYPE 32 | |
2330 | #include "paging_tmpl.h" | |
2331 | #undef PTTYPE | |
2332 | ||
82725b20 DE |
2333 | static void reset_rsvds_bits_mask(struct kvm_vcpu *vcpu, int level) |
2334 | { | |
2335 | struct kvm_mmu *context = &vcpu->arch.mmu; | |
2336 | int maxphyaddr = cpuid_maxphyaddr(vcpu); | |
2337 | u64 exb_bit_rsvd = 0; | |
2338 | ||
2339 | if (!is_nx(vcpu)) | |
2340 | exb_bit_rsvd = rsvd_bits(63, 63); | |
2341 | switch (level) { | |
2342 | case PT32_ROOT_LEVEL: | |
2343 | /* no rsvd bits for 2 level 4K page table entries */ | |
2344 | context->rsvd_bits_mask[0][1] = 0; | |
2345 | context->rsvd_bits_mask[0][0] = 0; | |
f815bce8 XG |
2346 | context->rsvd_bits_mask[1][0] = context->rsvd_bits_mask[0][0]; |
2347 | ||
2348 | if (!is_pse(vcpu)) { | |
2349 | context->rsvd_bits_mask[1][1] = 0; | |
2350 | break; | |
2351 | } | |
2352 | ||
82725b20 DE |
2353 | if (is_cpuid_PSE36()) |
2354 | /* 36bits PSE 4MB page */ | |
2355 | context->rsvd_bits_mask[1][1] = rsvd_bits(17, 21); | |
2356 | else | |
2357 | /* 32 bits PSE 4MB page */ | |
2358 | context->rsvd_bits_mask[1][1] = rsvd_bits(13, 21); | |
82725b20 DE |
2359 | break; |
2360 | case PT32E_ROOT_LEVEL: | |
20c466b5 DE |
2361 | context->rsvd_bits_mask[0][2] = |
2362 | rsvd_bits(maxphyaddr, 63) | | |
2363 | rsvd_bits(7, 8) | rsvd_bits(1, 2); /* PDPTE */ | |
82725b20 | 2364 | context->rsvd_bits_mask[0][1] = exb_bit_rsvd | |
4c26b4cd | 2365 | rsvd_bits(maxphyaddr, 62); /* PDE */ |
82725b20 DE |
2366 | context->rsvd_bits_mask[0][0] = exb_bit_rsvd | |
2367 | rsvd_bits(maxphyaddr, 62); /* PTE */ | |
2368 | context->rsvd_bits_mask[1][1] = exb_bit_rsvd | | |
2369 | rsvd_bits(maxphyaddr, 62) | | |
2370 | rsvd_bits(13, 20); /* large page */ | |
f815bce8 | 2371 | context->rsvd_bits_mask[1][0] = context->rsvd_bits_mask[0][0]; |
82725b20 DE |
2372 | break; |
2373 | case PT64_ROOT_LEVEL: | |
2374 | context->rsvd_bits_mask[0][3] = exb_bit_rsvd | | |
2375 | rsvd_bits(maxphyaddr, 51) | rsvd_bits(7, 8); | |
2376 | context->rsvd_bits_mask[0][2] = exb_bit_rsvd | | |
2377 | rsvd_bits(maxphyaddr, 51) | rsvd_bits(7, 8); | |
2378 | context->rsvd_bits_mask[0][1] = exb_bit_rsvd | | |
4c26b4cd | 2379 | rsvd_bits(maxphyaddr, 51); |
82725b20 DE |
2380 | context->rsvd_bits_mask[0][0] = exb_bit_rsvd | |
2381 | rsvd_bits(maxphyaddr, 51); | |
2382 | context->rsvd_bits_mask[1][3] = context->rsvd_bits_mask[0][3]; | |
e04da980 JR |
2383 | context->rsvd_bits_mask[1][2] = exb_bit_rsvd | |
2384 | rsvd_bits(maxphyaddr, 51) | | |
2385 | rsvd_bits(13, 29); | |
82725b20 | 2386 | context->rsvd_bits_mask[1][1] = exb_bit_rsvd | |
4c26b4cd SY |
2387 | rsvd_bits(maxphyaddr, 51) | |
2388 | rsvd_bits(13, 20); /* large page */ | |
f815bce8 | 2389 | context->rsvd_bits_mask[1][0] = context->rsvd_bits_mask[0][0]; |
82725b20 DE |
2390 | break; |
2391 | } | |
2392 | } | |
2393 | ||
17ac10ad | 2394 | static int paging64_init_context_common(struct kvm_vcpu *vcpu, int level) |
6aa8b732 | 2395 | { |
ad312c7c | 2396 | struct kvm_mmu *context = &vcpu->arch.mmu; |
6aa8b732 AK |
2397 | |
2398 | ASSERT(is_pae(vcpu)); | |
2399 | context->new_cr3 = paging_new_cr3; | |
2400 | context->page_fault = paging64_page_fault; | |
6aa8b732 | 2401 | context->gva_to_gpa = paging64_gva_to_gpa; |
c7addb90 | 2402 | context->prefetch_page = paging64_prefetch_page; |
e8bc217a | 2403 | context->sync_page = paging64_sync_page; |
a7052897 | 2404 | context->invlpg = paging64_invlpg; |
6aa8b732 | 2405 | context->free = paging_free; |
17ac10ad AK |
2406 | context->root_level = level; |
2407 | context->shadow_root_level = level; | |
17c3ba9d | 2408 | context->root_hpa = INVALID_PAGE; |
6aa8b732 AK |
2409 | return 0; |
2410 | } | |
2411 | ||
17ac10ad AK |
2412 | static int paging64_init_context(struct kvm_vcpu *vcpu) |
2413 | { | |
82725b20 | 2414 | reset_rsvds_bits_mask(vcpu, PT64_ROOT_LEVEL); |
17ac10ad AK |
2415 | return paging64_init_context_common(vcpu, PT64_ROOT_LEVEL); |
2416 | } | |
2417 | ||
6aa8b732 AK |
2418 | static int paging32_init_context(struct kvm_vcpu *vcpu) |
2419 | { | |
ad312c7c | 2420 | struct kvm_mmu *context = &vcpu->arch.mmu; |
6aa8b732 | 2421 | |
82725b20 | 2422 | reset_rsvds_bits_mask(vcpu, PT32_ROOT_LEVEL); |
6aa8b732 AK |
2423 | context->new_cr3 = paging_new_cr3; |
2424 | context->page_fault = paging32_page_fault; | |
6aa8b732 AK |
2425 | context->gva_to_gpa = paging32_gva_to_gpa; |
2426 | context->free = paging_free; | |
c7addb90 | 2427 | context->prefetch_page = paging32_prefetch_page; |
e8bc217a | 2428 | context->sync_page = paging32_sync_page; |
a7052897 | 2429 | context->invlpg = paging32_invlpg; |
6aa8b732 AK |
2430 | context->root_level = PT32_ROOT_LEVEL; |
2431 | context->shadow_root_level = PT32E_ROOT_LEVEL; | |
17c3ba9d | 2432 | context->root_hpa = INVALID_PAGE; |
6aa8b732 AK |
2433 | return 0; |
2434 | } | |
2435 | ||
2436 | static int paging32E_init_context(struct kvm_vcpu *vcpu) | |
2437 | { | |
82725b20 | 2438 | reset_rsvds_bits_mask(vcpu, PT32E_ROOT_LEVEL); |
17ac10ad | 2439 | return paging64_init_context_common(vcpu, PT32E_ROOT_LEVEL); |
6aa8b732 AK |
2440 | } |
2441 | ||
fb72d167 JR |
2442 | static int init_kvm_tdp_mmu(struct kvm_vcpu *vcpu) |
2443 | { | |
2444 | struct kvm_mmu *context = &vcpu->arch.mmu; | |
2445 | ||
2446 | context->new_cr3 = nonpaging_new_cr3; | |
2447 | context->page_fault = tdp_page_fault; | |
2448 | context->free = nonpaging_free; | |
2449 | context->prefetch_page = nonpaging_prefetch_page; | |
e8bc217a | 2450 | context->sync_page = nonpaging_sync_page; |
a7052897 | 2451 | context->invlpg = nonpaging_invlpg; |
67253af5 | 2452 | context->shadow_root_level = kvm_x86_ops->get_tdp_level(); |
fb72d167 JR |
2453 | context->root_hpa = INVALID_PAGE; |
2454 | ||
2455 | if (!is_paging(vcpu)) { | |
2456 | context->gva_to_gpa = nonpaging_gva_to_gpa; | |
2457 | context->root_level = 0; | |
2458 | } else if (is_long_mode(vcpu)) { | |
82725b20 | 2459 | reset_rsvds_bits_mask(vcpu, PT64_ROOT_LEVEL); |
fb72d167 JR |
2460 | context->gva_to_gpa = paging64_gva_to_gpa; |
2461 | context->root_level = PT64_ROOT_LEVEL; | |
2462 | } else if (is_pae(vcpu)) { | |
82725b20 | 2463 | reset_rsvds_bits_mask(vcpu, PT32E_ROOT_LEVEL); |
fb72d167 JR |
2464 | context->gva_to_gpa = paging64_gva_to_gpa; |
2465 | context->root_level = PT32E_ROOT_LEVEL; | |
2466 | } else { | |
82725b20 | 2467 | reset_rsvds_bits_mask(vcpu, PT32_ROOT_LEVEL); |
fb72d167 JR |
2468 | context->gva_to_gpa = paging32_gva_to_gpa; |
2469 | context->root_level = PT32_ROOT_LEVEL; | |
2470 | } | |
2471 | ||
2472 | return 0; | |
2473 | } | |
2474 | ||
2475 | static int init_kvm_softmmu(struct kvm_vcpu *vcpu) | |
6aa8b732 | 2476 | { |
a770f6f2 AK |
2477 | int r; |
2478 | ||
6aa8b732 | 2479 | ASSERT(vcpu); |
ad312c7c | 2480 | ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa)); |
6aa8b732 AK |
2481 | |
2482 | if (!is_paging(vcpu)) | |
a770f6f2 | 2483 | r = nonpaging_init_context(vcpu); |
a9058ecd | 2484 | else if (is_long_mode(vcpu)) |
a770f6f2 | 2485 | r = paging64_init_context(vcpu); |
6aa8b732 | 2486 | else if (is_pae(vcpu)) |
a770f6f2 | 2487 | r = paging32E_init_context(vcpu); |
6aa8b732 | 2488 | else |
a770f6f2 AK |
2489 | r = paging32_init_context(vcpu); |
2490 | ||
5b7e0102 | 2491 | vcpu->arch.mmu.base_role.cr4_pae = !!is_pae(vcpu); |
3dbe1415 | 2492 | vcpu->arch.mmu.base_role.cr0_wp = is_write_protection(vcpu); |
a770f6f2 AK |
2493 | |
2494 | return r; | |
6aa8b732 AK |
2495 | } |
2496 | ||
fb72d167 JR |
2497 | static int init_kvm_mmu(struct kvm_vcpu *vcpu) |
2498 | { | |
35149e21 AL |
2499 | vcpu->arch.update_pte.pfn = bad_pfn; |
2500 | ||
fb72d167 JR |
2501 | if (tdp_enabled) |
2502 | return init_kvm_tdp_mmu(vcpu); | |
2503 | else | |
2504 | return init_kvm_softmmu(vcpu); | |
2505 | } | |
2506 | ||
6aa8b732 AK |
2507 | static void destroy_kvm_mmu(struct kvm_vcpu *vcpu) |
2508 | { | |
2509 | ASSERT(vcpu); | |
62ad0755 SY |
2510 | if (VALID_PAGE(vcpu->arch.mmu.root_hpa)) |
2511 | /* mmu.free() should set root_hpa = INVALID_PAGE */ | |
ad312c7c | 2512 | vcpu->arch.mmu.free(vcpu); |
6aa8b732 AK |
2513 | } |
2514 | ||
2515 | int kvm_mmu_reset_context(struct kvm_vcpu *vcpu) | |
17c3ba9d AK |
2516 | { |
2517 | destroy_kvm_mmu(vcpu); | |
2518 | return init_kvm_mmu(vcpu); | |
2519 | } | |
8668a3c4 | 2520 | EXPORT_SYMBOL_GPL(kvm_mmu_reset_context); |
17c3ba9d AK |
2521 | |
2522 | int kvm_mmu_load(struct kvm_vcpu *vcpu) | |
6aa8b732 | 2523 | { |
714b93da AK |
2524 | int r; |
2525 | ||
e2dec939 | 2526 | r = mmu_topup_memory_caches(vcpu); |
17c3ba9d AK |
2527 | if (r) |
2528 | goto out; | |
8986ecc0 | 2529 | r = mmu_alloc_roots(vcpu); |
8facbbff | 2530 | spin_lock(&vcpu->kvm->mmu_lock); |
0ba73cda | 2531 | mmu_sync_roots(vcpu); |
aaee2c94 | 2532 | spin_unlock(&vcpu->kvm->mmu_lock); |
8986ecc0 MT |
2533 | if (r) |
2534 | goto out; | |
3662cb1c | 2535 | /* set_cr3() should ensure TLB has been flushed */ |
ad312c7c | 2536 | kvm_x86_ops->set_cr3(vcpu, vcpu->arch.mmu.root_hpa); |
714b93da AK |
2537 | out: |
2538 | return r; | |
6aa8b732 | 2539 | } |
17c3ba9d AK |
2540 | EXPORT_SYMBOL_GPL(kvm_mmu_load); |
2541 | ||
2542 | void kvm_mmu_unload(struct kvm_vcpu *vcpu) | |
2543 | { | |
2544 | mmu_free_roots(vcpu); | |
2545 | } | |
6aa8b732 | 2546 | |
09072daf | 2547 | static void mmu_pte_write_zap_pte(struct kvm_vcpu *vcpu, |
4db35314 | 2548 | struct kvm_mmu_page *sp, |
ac1b714e AK |
2549 | u64 *spte) |
2550 | { | |
2551 | u64 pte; | |
2552 | struct kvm_mmu_page *child; | |
2553 | ||
2554 | pte = *spte; | |
c7addb90 | 2555 | if (is_shadow_present_pte(pte)) { |
776e6633 | 2556 | if (is_last_spte(pte, sp->role.level)) |
290fc38d | 2557 | rmap_remove(vcpu->kvm, spte); |
ac1b714e AK |
2558 | else { |
2559 | child = page_header(pte & PT64_BASE_ADDR_MASK); | |
90cb0529 | 2560 | mmu_page_remove_parent_pte(child, spte); |
ac1b714e AK |
2561 | } |
2562 | } | |
d555c333 | 2563 | __set_spte(spte, shadow_trap_nonpresent_pte); |
05da4558 MT |
2564 | if (is_large_pte(pte)) |
2565 | --vcpu->kvm->stat.lpages; | |
ac1b714e AK |
2566 | } |
2567 | ||
0028425f | 2568 | static void mmu_pte_write_new_pte(struct kvm_vcpu *vcpu, |
4db35314 | 2569 | struct kvm_mmu_page *sp, |
0028425f | 2570 | u64 *spte, |
489f1d65 | 2571 | const void *new) |
0028425f | 2572 | { |
30945387 | 2573 | if (sp->role.level != PT_PAGE_TABLE_LEVEL) { |
7e4e4056 JR |
2574 | ++vcpu->kvm->stat.mmu_pde_zapped; |
2575 | return; | |
30945387 | 2576 | } |
0028425f | 2577 | |
4cee5764 | 2578 | ++vcpu->kvm->stat.mmu_pte_updated; |
5b7e0102 | 2579 | if (!sp->role.cr4_pae) |
489f1d65 | 2580 | paging32_update_pte(vcpu, sp, spte, new); |
0028425f | 2581 | else |
489f1d65 | 2582 | paging64_update_pte(vcpu, sp, spte, new); |
0028425f AK |
2583 | } |
2584 | ||
79539cec AK |
2585 | static bool need_remote_flush(u64 old, u64 new) |
2586 | { | |
2587 | if (!is_shadow_present_pte(old)) | |
2588 | return false; | |
2589 | if (!is_shadow_present_pte(new)) | |
2590 | return true; | |
2591 | if ((old ^ new) & PT64_BASE_ADDR_MASK) | |
2592 | return true; | |
2593 | old ^= PT64_NX_MASK; | |
2594 | new ^= PT64_NX_MASK; | |
2595 | return (old & ~new & PT64_PERM_MASK) != 0; | |
2596 | } | |
2597 | ||
2598 | static void mmu_pte_write_flush_tlb(struct kvm_vcpu *vcpu, u64 old, u64 new) | |
2599 | { | |
2600 | if (need_remote_flush(old, new)) | |
2601 | kvm_flush_remote_tlbs(vcpu->kvm); | |
2602 | else | |
2603 | kvm_mmu_flush_tlb(vcpu); | |
2604 | } | |
2605 | ||
12b7d28f AK |
2606 | static bool last_updated_pte_accessed(struct kvm_vcpu *vcpu) |
2607 | { | |
ad312c7c | 2608 | u64 *spte = vcpu->arch.last_pte_updated; |
12b7d28f | 2609 | |
7b52345e | 2610 | return !!(spte && (*spte & shadow_accessed_mask)); |
12b7d28f AK |
2611 | } |
2612 | ||
d7824fff | 2613 | static void mmu_guess_page_from_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa, |
72016f3a | 2614 | u64 gpte) |
d7824fff AK |
2615 | { |
2616 | gfn_t gfn; | |
35149e21 | 2617 | pfn_t pfn; |
d7824fff | 2618 | |
43a3795a | 2619 | if (!is_present_gpte(gpte)) |
d7824fff AK |
2620 | return; |
2621 | gfn = (gpte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT; | |
72dc67a6 | 2622 | |
e930bffe | 2623 | vcpu->arch.update_pte.mmu_seq = vcpu->kvm->mmu_notifier_seq; |
4c2155ce | 2624 | smp_rmb(); |
35149e21 | 2625 | pfn = gfn_to_pfn(vcpu->kvm, gfn); |
72dc67a6 | 2626 | |
35149e21 AL |
2627 | if (is_error_pfn(pfn)) { |
2628 | kvm_release_pfn_clean(pfn); | |
d196e343 AK |
2629 | return; |
2630 | } | |
d7824fff | 2631 | vcpu->arch.update_pte.gfn = gfn; |
35149e21 | 2632 | vcpu->arch.update_pte.pfn = pfn; |
d7824fff AK |
2633 | } |
2634 | ||
1b7fcd32 AK |
2635 | static void kvm_mmu_access_page(struct kvm_vcpu *vcpu, gfn_t gfn) |
2636 | { | |
2637 | u64 *spte = vcpu->arch.last_pte_updated; | |
2638 | ||
2639 | if (spte | |
2640 | && vcpu->arch.last_pte_gfn == gfn | |
2641 | && shadow_accessed_mask | |
2642 | && !(*spte & shadow_accessed_mask) | |
2643 | && is_shadow_present_pte(*spte)) | |
2644 | set_bit(PT_ACCESSED_SHIFT, (unsigned long *)spte); | |
2645 | } | |
2646 | ||
09072daf | 2647 | void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa, |
ad218f85 MT |
2648 | const u8 *new, int bytes, |
2649 | bool guest_initiated) | |
da4a00f0 | 2650 | { |
9b7a0325 | 2651 | gfn_t gfn = gpa >> PAGE_SHIFT; |
4db35314 | 2652 | struct kvm_mmu_page *sp; |
0e7bc4b9 | 2653 | struct hlist_node *node, *n; |
9b7a0325 AK |
2654 | struct hlist_head *bucket; |
2655 | unsigned index; | |
489f1d65 | 2656 | u64 entry, gentry; |
9b7a0325 | 2657 | u64 *spte; |
9b7a0325 | 2658 | unsigned offset = offset_in_page(gpa); |
0e7bc4b9 | 2659 | unsigned pte_size; |
9b7a0325 | 2660 | unsigned page_offset; |
0e7bc4b9 | 2661 | unsigned misaligned; |
fce0657f | 2662 | unsigned quadrant; |
9b7a0325 | 2663 | int level; |
86a5ba02 | 2664 | int flooded = 0; |
ac1b714e | 2665 | int npte; |
489f1d65 | 2666 | int r; |
08e850c6 | 2667 | int invlpg_counter; |
9b7a0325 | 2668 | |
b8688d51 | 2669 | pgprintk("%s: gpa %llx bytes %d\n", __func__, gpa, bytes); |
72016f3a | 2670 | |
08e850c6 | 2671 | invlpg_counter = atomic_read(&vcpu->kvm->arch.invlpg_counter); |
72016f3a AK |
2672 | |
2673 | /* | |
2674 | * Assume that the pte write on a page table of the same type | |
2675 | * as the current vcpu paging mode. This is nearly always true | |
2676 | * (might be false while changing modes). Note it is verified later | |
2677 | * by update_pte(). | |
2678 | */ | |
08e850c6 | 2679 | if ((is_pae(vcpu) && bytes == 4) || !new) { |
72016f3a | 2680 | /* Handle a 32-bit guest writing two halves of a 64-bit gpte */ |
08e850c6 AK |
2681 | if (is_pae(vcpu)) { |
2682 | gpa &= ~(gpa_t)7; | |
2683 | bytes = 8; | |
2684 | } | |
2685 | r = kvm_read_guest(vcpu->kvm, gpa, &gentry, min(bytes, 8)); | |
72016f3a AK |
2686 | if (r) |
2687 | gentry = 0; | |
08e850c6 AK |
2688 | new = (const u8 *)&gentry; |
2689 | } | |
2690 | ||
2691 | switch (bytes) { | |
2692 | case 4: | |
2693 | gentry = *(const u32 *)new; | |
2694 | break; | |
2695 | case 8: | |
2696 | gentry = *(const u64 *)new; | |
2697 | break; | |
2698 | default: | |
2699 | gentry = 0; | |
2700 | break; | |
72016f3a AK |
2701 | } |
2702 | ||
2703 | mmu_guess_page_from_pte_write(vcpu, gpa, gentry); | |
aaee2c94 | 2704 | spin_lock(&vcpu->kvm->mmu_lock); |
08e850c6 AK |
2705 | if (atomic_read(&vcpu->kvm->arch.invlpg_counter) != invlpg_counter) |
2706 | gentry = 0; | |
1b7fcd32 | 2707 | kvm_mmu_access_page(vcpu, gfn); |
eb787d10 | 2708 | kvm_mmu_free_some_pages(vcpu); |
4cee5764 | 2709 | ++vcpu->kvm->stat.mmu_pte_write; |
c7addb90 | 2710 | kvm_mmu_audit(vcpu, "pre pte write"); |
ad218f85 MT |
2711 | if (guest_initiated) { |
2712 | if (gfn == vcpu->arch.last_pt_write_gfn | |
2713 | && !last_updated_pte_accessed(vcpu)) { | |
2714 | ++vcpu->arch.last_pt_write_count; | |
2715 | if (vcpu->arch.last_pt_write_count >= 3) | |
2716 | flooded = 1; | |
2717 | } else { | |
2718 | vcpu->arch.last_pt_write_gfn = gfn; | |
2719 | vcpu->arch.last_pt_write_count = 1; | |
2720 | vcpu->arch.last_pte_updated = NULL; | |
2721 | } | |
86a5ba02 | 2722 | } |
1ae0a13d | 2723 | index = kvm_page_table_hashfn(gfn); |
f05e70ac | 2724 | bucket = &vcpu->kvm->arch.mmu_page_hash[index]; |
3246af0e XG |
2725 | |
2726 | restart: | |
4db35314 | 2727 | hlist_for_each_entry_safe(sp, node, n, bucket, hash_link) { |
f6e2c02b | 2728 | if (sp->gfn != gfn || sp->role.direct || sp->role.invalid) |
9b7a0325 | 2729 | continue; |
5b7e0102 | 2730 | pte_size = sp->role.cr4_pae ? 8 : 4; |
0e7bc4b9 | 2731 | misaligned = (offset ^ (offset + bytes - 1)) & ~(pte_size - 1); |
e925c5ba | 2732 | misaligned |= bytes < 4; |
86a5ba02 | 2733 | if (misaligned || flooded) { |
0e7bc4b9 AK |
2734 | /* |
2735 | * Misaligned accesses are too much trouble to fix | |
2736 | * up; also, they usually indicate a page is not used | |
2737 | * as a page table. | |
86a5ba02 AK |
2738 | * |
2739 | * If we're seeing too many writes to a page, | |
2740 | * it may no longer be a page table, or we may be | |
2741 | * forking, in which case it is better to unmap the | |
2742 | * page. | |
0e7bc4b9 AK |
2743 | */ |
2744 | pgprintk("misaligned: gpa %llx bytes %d role %x\n", | |
4db35314 | 2745 | gpa, bytes, sp->role.word); |
07385413 | 2746 | if (kvm_mmu_zap_page(vcpu->kvm, sp)) |
3246af0e | 2747 | goto restart; |
4cee5764 | 2748 | ++vcpu->kvm->stat.mmu_flooded; |
0e7bc4b9 AK |
2749 | continue; |
2750 | } | |
9b7a0325 | 2751 | page_offset = offset; |
4db35314 | 2752 | level = sp->role.level; |
ac1b714e | 2753 | npte = 1; |
5b7e0102 | 2754 | if (!sp->role.cr4_pae) { |
ac1b714e AK |
2755 | page_offset <<= 1; /* 32->64 */ |
2756 | /* | |
2757 | * A 32-bit pde maps 4MB while the shadow pdes map | |
2758 | * only 2MB. So we need to double the offset again | |
2759 | * and zap two pdes instead of one. | |
2760 | */ | |
2761 | if (level == PT32_ROOT_LEVEL) { | |
6b8d0f9b | 2762 | page_offset &= ~7; /* kill rounding error */ |
ac1b714e AK |
2763 | page_offset <<= 1; |
2764 | npte = 2; | |
2765 | } | |
fce0657f | 2766 | quadrant = page_offset >> PAGE_SHIFT; |
9b7a0325 | 2767 | page_offset &= ~PAGE_MASK; |
4db35314 | 2768 | if (quadrant != sp->role.quadrant) |
fce0657f | 2769 | continue; |
9b7a0325 | 2770 | } |
4db35314 | 2771 | spte = &sp->spt[page_offset / sizeof(*spte)]; |
ac1b714e | 2772 | while (npte--) { |
79539cec | 2773 | entry = *spte; |
4db35314 | 2774 | mmu_pte_write_zap_pte(vcpu, sp, spte); |
72016f3a AK |
2775 | if (gentry) |
2776 | mmu_pte_write_new_pte(vcpu, sp, spte, &gentry); | |
79539cec | 2777 | mmu_pte_write_flush_tlb(vcpu, entry, *spte); |
ac1b714e | 2778 | ++spte; |
9b7a0325 | 2779 | } |
9b7a0325 | 2780 | } |
c7addb90 | 2781 | kvm_mmu_audit(vcpu, "post pte write"); |
aaee2c94 | 2782 | spin_unlock(&vcpu->kvm->mmu_lock); |
35149e21 AL |
2783 | if (!is_error_pfn(vcpu->arch.update_pte.pfn)) { |
2784 | kvm_release_pfn_clean(vcpu->arch.update_pte.pfn); | |
2785 | vcpu->arch.update_pte.pfn = bad_pfn; | |
d7824fff | 2786 | } |
da4a00f0 AK |
2787 | } |
2788 | ||
a436036b AK |
2789 | int kvm_mmu_unprotect_page_virt(struct kvm_vcpu *vcpu, gva_t gva) |
2790 | { | |
10589a46 MT |
2791 | gpa_t gpa; |
2792 | int r; | |
a436036b | 2793 | |
60f24784 AK |
2794 | if (tdp_enabled) |
2795 | return 0; | |
2796 | ||
1871c602 | 2797 | gpa = kvm_mmu_gva_to_gpa_read(vcpu, gva, NULL); |
10589a46 | 2798 | |
aaee2c94 | 2799 | spin_lock(&vcpu->kvm->mmu_lock); |
10589a46 | 2800 | r = kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT); |
aaee2c94 | 2801 | spin_unlock(&vcpu->kvm->mmu_lock); |
10589a46 | 2802 | return r; |
a436036b | 2803 | } |
577bdc49 | 2804 | EXPORT_SYMBOL_GPL(kvm_mmu_unprotect_page_virt); |
a436036b | 2805 | |
22d95b12 | 2806 | void __kvm_mmu_free_some_pages(struct kvm_vcpu *vcpu) |
ebeace86 | 2807 | { |
3b80fffe IE |
2808 | while (vcpu->kvm->arch.n_free_mmu_pages < KVM_REFILL_PAGES && |
2809 | !list_empty(&vcpu->kvm->arch.active_mmu_pages)) { | |
4db35314 | 2810 | struct kvm_mmu_page *sp; |
ebeace86 | 2811 | |
f05e70ac | 2812 | sp = container_of(vcpu->kvm->arch.active_mmu_pages.prev, |
4db35314 AK |
2813 | struct kvm_mmu_page, link); |
2814 | kvm_mmu_zap_page(vcpu->kvm, sp); | |
4cee5764 | 2815 | ++vcpu->kvm->stat.mmu_recycled; |
ebeace86 AK |
2816 | } |
2817 | } | |
ebeace86 | 2818 | |
3067714c AK |
2819 | int kvm_mmu_page_fault(struct kvm_vcpu *vcpu, gva_t cr2, u32 error_code) |
2820 | { | |
2821 | int r; | |
2822 | enum emulation_result er; | |
2823 | ||
ad312c7c | 2824 | r = vcpu->arch.mmu.page_fault(vcpu, cr2, error_code); |
3067714c AK |
2825 | if (r < 0) |
2826 | goto out; | |
2827 | ||
2828 | if (!r) { | |
2829 | r = 1; | |
2830 | goto out; | |
2831 | } | |
2832 | ||
b733bfb5 AK |
2833 | r = mmu_topup_memory_caches(vcpu); |
2834 | if (r) | |
2835 | goto out; | |
2836 | ||
851ba692 | 2837 | er = emulate_instruction(vcpu, cr2, error_code, 0); |
3067714c AK |
2838 | |
2839 | switch (er) { | |
2840 | case EMULATE_DONE: | |
2841 | return 1; | |
2842 | case EMULATE_DO_MMIO: | |
2843 | ++vcpu->stat.mmio_exits; | |
6d77dbfc | 2844 | /* fall through */ |
3067714c | 2845 | case EMULATE_FAIL: |
3f5d18a9 | 2846 | return 0; |
3067714c AK |
2847 | default: |
2848 | BUG(); | |
2849 | } | |
2850 | out: | |
3067714c AK |
2851 | return r; |
2852 | } | |
2853 | EXPORT_SYMBOL_GPL(kvm_mmu_page_fault); | |
2854 | ||
a7052897 MT |
2855 | void kvm_mmu_invlpg(struct kvm_vcpu *vcpu, gva_t gva) |
2856 | { | |
a7052897 | 2857 | vcpu->arch.mmu.invlpg(vcpu, gva); |
a7052897 MT |
2858 | kvm_mmu_flush_tlb(vcpu); |
2859 | ++vcpu->stat.invlpg; | |
2860 | } | |
2861 | EXPORT_SYMBOL_GPL(kvm_mmu_invlpg); | |
2862 | ||
18552672 JR |
2863 | void kvm_enable_tdp(void) |
2864 | { | |
2865 | tdp_enabled = true; | |
2866 | } | |
2867 | EXPORT_SYMBOL_GPL(kvm_enable_tdp); | |
2868 | ||
5f4cb662 JR |
2869 | void kvm_disable_tdp(void) |
2870 | { | |
2871 | tdp_enabled = false; | |
2872 | } | |
2873 | EXPORT_SYMBOL_GPL(kvm_disable_tdp); | |
2874 | ||
6aa8b732 AK |
2875 | static void free_mmu_pages(struct kvm_vcpu *vcpu) |
2876 | { | |
ad312c7c | 2877 | free_page((unsigned long)vcpu->arch.mmu.pae_root); |
6aa8b732 AK |
2878 | } |
2879 | ||
2880 | static int alloc_mmu_pages(struct kvm_vcpu *vcpu) | |
2881 | { | |
17ac10ad | 2882 | struct page *page; |
6aa8b732 AK |
2883 | int i; |
2884 | ||
2885 | ASSERT(vcpu); | |
2886 | ||
17ac10ad AK |
2887 | /* |
2888 | * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64. | |
2889 | * Therefore we need to allocate shadow page tables in the first | |
2890 | * 4GB of memory, which happens to fit the DMA32 zone. | |
2891 | */ | |
2892 | page = alloc_page(GFP_KERNEL | __GFP_DMA32); | |
2893 | if (!page) | |
d7fa6ab2 WY |
2894 | return -ENOMEM; |
2895 | ||
ad312c7c | 2896 | vcpu->arch.mmu.pae_root = page_address(page); |
17ac10ad | 2897 | for (i = 0; i < 4; ++i) |
ad312c7c | 2898 | vcpu->arch.mmu.pae_root[i] = INVALID_PAGE; |
17ac10ad | 2899 | |
6aa8b732 | 2900 | return 0; |
6aa8b732 AK |
2901 | } |
2902 | ||
8018c27b | 2903 | int kvm_mmu_create(struct kvm_vcpu *vcpu) |
6aa8b732 | 2904 | { |
6aa8b732 | 2905 | ASSERT(vcpu); |
ad312c7c | 2906 | ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa)); |
6aa8b732 | 2907 | |
8018c27b IM |
2908 | return alloc_mmu_pages(vcpu); |
2909 | } | |
6aa8b732 | 2910 | |
8018c27b IM |
2911 | int kvm_mmu_setup(struct kvm_vcpu *vcpu) |
2912 | { | |
2913 | ASSERT(vcpu); | |
ad312c7c | 2914 | ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa)); |
2c264957 | 2915 | |
8018c27b | 2916 | return init_kvm_mmu(vcpu); |
6aa8b732 AK |
2917 | } |
2918 | ||
2919 | void kvm_mmu_destroy(struct kvm_vcpu *vcpu) | |
2920 | { | |
2921 | ASSERT(vcpu); | |
2922 | ||
2923 | destroy_kvm_mmu(vcpu); | |
2924 | free_mmu_pages(vcpu); | |
714b93da | 2925 | mmu_free_memory_caches(vcpu); |
6aa8b732 AK |
2926 | } |
2927 | ||
90cb0529 | 2928 | void kvm_mmu_slot_remove_write_access(struct kvm *kvm, int slot) |
6aa8b732 | 2929 | { |
4db35314 | 2930 | struct kvm_mmu_page *sp; |
6aa8b732 | 2931 | |
f05e70ac | 2932 | list_for_each_entry(sp, &kvm->arch.active_mmu_pages, link) { |
6aa8b732 AK |
2933 | int i; |
2934 | u64 *pt; | |
2935 | ||
291f26bc | 2936 | if (!test_bit(slot, sp->slot_bitmap)) |
6aa8b732 AK |
2937 | continue; |
2938 | ||
4db35314 | 2939 | pt = sp->spt; |
6aa8b732 AK |
2940 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) |
2941 | /* avoid RMW */ | |
9647c14c | 2942 | if (pt[i] & PT_WRITABLE_MASK) |
6aa8b732 | 2943 | pt[i] &= ~PT_WRITABLE_MASK; |
6aa8b732 | 2944 | } |
171d595d | 2945 | kvm_flush_remote_tlbs(kvm); |
6aa8b732 | 2946 | } |
37a7d8b0 | 2947 | |
90cb0529 | 2948 | void kvm_mmu_zap_all(struct kvm *kvm) |
e0fa826f | 2949 | { |
4db35314 | 2950 | struct kvm_mmu_page *sp, *node; |
e0fa826f | 2951 | |
aaee2c94 | 2952 | spin_lock(&kvm->mmu_lock); |
3246af0e | 2953 | restart: |
f05e70ac | 2954 | list_for_each_entry_safe(sp, node, &kvm->arch.active_mmu_pages, link) |
07385413 | 2955 | if (kvm_mmu_zap_page(kvm, sp)) |
3246af0e XG |
2956 | goto restart; |
2957 | ||
aaee2c94 | 2958 | spin_unlock(&kvm->mmu_lock); |
e0fa826f | 2959 | |
90cb0529 | 2960 | kvm_flush_remote_tlbs(kvm); |
e0fa826f DL |
2961 | } |
2962 | ||
d35b8dd9 | 2963 | static int kvm_mmu_remove_some_alloc_mmu_pages(struct kvm *kvm) |
3ee16c81 IE |
2964 | { |
2965 | struct kvm_mmu_page *page; | |
2966 | ||
2967 | page = container_of(kvm->arch.active_mmu_pages.prev, | |
2968 | struct kvm_mmu_page, link); | |
54a4f023 | 2969 | return kvm_mmu_zap_page(kvm, page); |
3ee16c81 IE |
2970 | } |
2971 | ||
7f8275d0 | 2972 | static int mmu_shrink(struct shrinker *shrink, int nr_to_scan, gfp_t gfp_mask) |
3ee16c81 IE |
2973 | { |
2974 | struct kvm *kvm; | |
2975 | struct kvm *kvm_freed = NULL; | |
2976 | int cache_count = 0; | |
2977 | ||
2978 | spin_lock(&kvm_lock); | |
2979 | ||
2980 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
d35b8dd9 | 2981 | int npages, idx, freed_pages; |
3ee16c81 | 2982 | |
f656ce01 | 2983 | idx = srcu_read_lock(&kvm->srcu); |
3ee16c81 IE |
2984 | spin_lock(&kvm->mmu_lock); |
2985 | npages = kvm->arch.n_alloc_mmu_pages - | |
2986 | kvm->arch.n_free_mmu_pages; | |
2987 | cache_count += npages; | |
2988 | if (!kvm_freed && nr_to_scan > 0 && npages > 0) { | |
d35b8dd9 GJ |
2989 | freed_pages = kvm_mmu_remove_some_alloc_mmu_pages(kvm); |
2990 | cache_count -= freed_pages; | |
3ee16c81 IE |
2991 | kvm_freed = kvm; |
2992 | } | |
2993 | nr_to_scan--; | |
2994 | ||
2995 | spin_unlock(&kvm->mmu_lock); | |
f656ce01 | 2996 | srcu_read_unlock(&kvm->srcu, idx); |
3ee16c81 IE |
2997 | } |
2998 | if (kvm_freed) | |
2999 | list_move_tail(&kvm_freed->vm_list, &vm_list); | |
3000 | ||
3001 | spin_unlock(&kvm_lock); | |
3002 | ||
3003 | return cache_count; | |
3004 | } | |
3005 | ||
3006 | static struct shrinker mmu_shrinker = { | |
3007 | .shrink = mmu_shrink, | |
3008 | .seeks = DEFAULT_SEEKS * 10, | |
3009 | }; | |
3010 | ||
2ddfd20e | 3011 | static void mmu_destroy_caches(void) |
b5a33a75 AK |
3012 | { |
3013 | if (pte_chain_cache) | |
3014 | kmem_cache_destroy(pte_chain_cache); | |
3015 | if (rmap_desc_cache) | |
3016 | kmem_cache_destroy(rmap_desc_cache); | |
d3d25b04 AK |
3017 | if (mmu_page_header_cache) |
3018 | kmem_cache_destroy(mmu_page_header_cache); | |
b5a33a75 AK |
3019 | } |
3020 | ||
3ee16c81 IE |
3021 | void kvm_mmu_module_exit(void) |
3022 | { | |
3023 | mmu_destroy_caches(); | |
3024 | unregister_shrinker(&mmu_shrinker); | |
3025 | } | |
3026 | ||
b5a33a75 AK |
3027 | int kvm_mmu_module_init(void) |
3028 | { | |
3029 | pte_chain_cache = kmem_cache_create("kvm_pte_chain", | |
3030 | sizeof(struct kvm_pte_chain), | |
20c2df83 | 3031 | 0, 0, NULL); |
b5a33a75 AK |
3032 | if (!pte_chain_cache) |
3033 | goto nomem; | |
3034 | rmap_desc_cache = kmem_cache_create("kvm_rmap_desc", | |
3035 | sizeof(struct kvm_rmap_desc), | |
20c2df83 | 3036 | 0, 0, NULL); |
b5a33a75 AK |
3037 | if (!rmap_desc_cache) |
3038 | goto nomem; | |
3039 | ||
d3d25b04 AK |
3040 | mmu_page_header_cache = kmem_cache_create("kvm_mmu_page_header", |
3041 | sizeof(struct kvm_mmu_page), | |
20c2df83 | 3042 | 0, 0, NULL); |
d3d25b04 AK |
3043 | if (!mmu_page_header_cache) |
3044 | goto nomem; | |
3045 | ||
3ee16c81 IE |
3046 | register_shrinker(&mmu_shrinker); |
3047 | ||
b5a33a75 AK |
3048 | return 0; |
3049 | ||
3050 | nomem: | |
3ee16c81 | 3051 | mmu_destroy_caches(); |
b5a33a75 AK |
3052 | return -ENOMEM; |
3053 | } | |
3054 | ||
3ad82a7e ZX |
3055 | /* |
3056 | * Caculate mmu pages needed for kvm. | |
3057 | */ | |
3058 | unsigned int kvm_mmu_calculate_mmu_pages(struct kvm *kvm) | |
3059 | { | |
3060 | int i; | |
3061 | unsigned int nr_mmu_pages; | |
3062 | unsigned int nr_pages = 0; | |
bc6678a3 | 3063 | struct kvm_memslots *slots; |
3ad82a7e | 3064 | |
90d83dc3 LJ |
3065 | slots = kvm_memslots(kvm); |
3066 | ||
bc6678a3 MT |
3067 | for (i = 0; i < slots->nmemslots; i++) |
3068 | nr_pages += slots->memslots[i].npages; | |
3ad82a7e ZX |
3069 | |
3070 | nr_mmu_pages = nr_pages * KVM_PERMILLE_MMU_PAGES / 1000; | |
3071 | nr_mmu_pages = max(nr_mmu_pages, | |
3072 | (unsigned int) KVM_MIN_ALLOC_MMU_PAGES); | |
3073 | ||
3074 | return nr_mmu_pages; | |
3075 | } | |
3076 | ||
2f333bcb MT |
3077 | static void *pv_mmu_peek_buffer(struct kvm_pv_mmu_op_buffer *buffer, |
3078 | unsigned len) | |
3079 | { | |
3080 | if (len > buffer->len) | |
3081 | return NULL; | |
3082 | return buffer->ptr; | |
3083 | } | |
3084 | ||
3085 | static void *pv_mmu_read_buffer(struct kvm_pv_mmu_op_buffer *buffer, | |
3086 | unsigned len) | |
3087 | { | |
3088 | void *ret; | |
3089 | ||
3090 | ret = pv_mmu_peek_buffer(buffer, len); | |
3091 | if (!ret) | |
3092 | return ret; | |
3093 | buffer->ptr += len; | |
3094 | buffer->len -= len; | |
3095 | buffer->processed += len; | |
3096 | return ret; | |
3097 | } | |
3098 | ||
3099 | static int kvm_pv_mmu_write(struct kvm_vcpu *vcpu, | |
3100 | gpa_t addr, gpa_t value) | |
3101 | { | |
3102 | int bytes = 8; | |
3103 | int r; | |
3104 | ||
3105 | if (!is_long_mode(vcpu) && !is_pae(vcpu)) | |
3106 | bytes = 4; | |
3107 | ||
3108 | r = mmu_topup_memory_caches(vcpu); | |
3109 | if (r) | |
3110 | return r; | |
3111 | ||
3200f405 | 3112 | if (!emulator_write_phys(vcpu, addr, &value, bytes)) |
2f333bcb MT |
3113 | return -EFAULT; |
3114 | ||
3115 | return 1; | |
3116 | } | |
3117 | ||
3118 | static int kvm_pv_mmu_flush_tlb(struct kvm_vcpu *vcpu) | |
3119 | { | |
a8cd0244 | 3120 | kvm_set_cr3(vcpu, vcpu->arch.cr3); |
2f333bcb MT |
3121 | return 1; |
3122 | } | |
3123 | ||
3124 | static int kvm_pv_mmu_release_pt(struct kvm_vcpu *vcpu, gpa_t addr) | |
3125 | { | |
3126 | spin_lock(&vcpu->kvm->mmu_lock); | |
3127 | mmu_unshadow(vcpu->kvm, addr >> PAGE_SHIFT); | |
3128 | spin_unlock(&vcpu->kvm->mmu_lock); | |
3129 | return 1; | |
3130 | } | |
3131 | ||
3132 | static int kvm_pv_mmu_op_one(struct kvm_vcpu *vcpu, | |
3133 | struct kvm_pv_mmu_op_buffer *buffer) | |
3134 | { | |
3135 | struct kvm_mmu_op_header *header; | |
3136 | ||
3137 | header = pv_mmu_peek_buffer(buffer, sizeof *header); | |
3138 | if (!header) | |
3139 | return 0; | |
3140 | switch (header->op) { | |
3141 | case KVM_MMU_OP_WRITE_PTE: { | |
3142 | struct kvm_mmu_op_write_pte *wpte; | |
3143 | ||
3144 | wpte = pv_mmu_read_buffer(buffer, sizeof *wpte); | |
3145 | if (!wpte) | |
3146 | return 0; | |
3147 | return kvm_pv_mmu_write(vcpu, wpte->pte_phys, | |
3148 | wpte->pte_val); | |
3149 | } | |
3150 | case KVM_MMU_OP_FLUSH_TLB: { | |
3151 | struct kvm_mmu_op_flush_tlb *ftlb; | |
3152 | ||
3153 | ftlb = pv_mmu_read_buffer(buffer, sizeof *ftlb); | |
3154 | if (!ftlb) | |
3155 | return 0; | |
3156 | return kvm_pv_mmu_flush_tlb(vcpu); | |
3157 | } | |
3158 | case KVM_MMU_OP_RELEASE_PT: { | |
3159 | struct kvm_mmu_op_release_pt *rpt; | |
3160 | ||
3161 | rpt = pv_mmu_read_buffer(buffer, sizeof *rpt); | |
3162 | if (!rpt) | |
3163 | return 0; | |
3164 | return kvm_pv_mmu_release_pt(vcpu, rpt->pt_phys); | |
3165 | } | |
3166 | default: return 0; | |
3167 | } | |
3168 | } | |
3169 | ||
3170 | int kvm_pv_mmu_op(struct kvm_vcpu *vcpu, unsigned long bytes, | |
3171 | gpa_t addr, unsigned long *ret) | |
3172 | { | |
3173 | int r; | |
6ad18fba | 3174 | struct kvm_pv_mmu_op_buffer *buffer = &vcpu->arch.mmu_op_buffer; |
2f333bcb | 3175 | |
6ad18fba DH |
3176 | buffer->ptr = buffer->buf; |
3177 | buffer->len = min_t(unsigned long, bytes, sizeof buffer->buf); | |
3178 | buffer->processed = 0; | |
2f333bcb | 3179 | |
6ad18fba | 3180 | r = kvm_read_guest(vcpu->kvm, addr, buffer->buf, buffer->len); |
2f333bcb MT |
3181 | if (r) |
3182 | goto out; | |
3183 | ||
6ad18fba DH |
3184 | while (buffer->len) { |
3185 | r = kvm_pv_mmu_op_one(vcpu, buffer); | |
2f333bcb MT |
3186 | if (r < 0) |
3187 | goto out; | |
3188 | if (r == 0) | |
3189 | break; | |
3190 | } | |
3191 | ||
3192 | r = 1; | |
3193 | out: | |
6ad18fba | 3194 | *ret = buffer->processed; |
2f333bcb MT |
3195 | return r; |
3196 | } | |
3197 | ||
94d8b056 MT |
3198 | int kvm_mmu_get_spte_hierarchy(struct kvm_vcpu *vcpu, u64 addr, u64 sptes[4]) |
3199 | { | |
3200 | struct kvm_shadow_walk_iterator iterator; | |
3201 | int nr_sptes = 0; | |
3202 | ||
3203 | spin_lock(&vcpu->kvm->mmu_lock); | |
3204 | for_each_shadow_entry(vcpu, addr, iterator) { | |
3205 | sptes[iterator.level-1] = *iterator.sptep; | |
3206 | nr_sptes++; | |
3207 | if (!is_shadow_present_pte(*iterator.sptep)) | |
3208 | break; | |
3209 | } | |
3210 | spin_unlock(&vcpu->kvm->mmu_lock); | |
3211 | ||
3212 | return nr_sptes; | |
3213 | } | |
3214 | EXPORT_SYMBOL_GPL(kvm_mmu_get_spte_hierarchy); | |
3215 | ||
37a7d8b0 AK |
3216 | #ifdef AUDIT |
3217 | ||
3218 | static const char *audit_msg; | |
3219 | ||
3220 | static gva_t canonicalize(gva_t gva) | |
3221 | { | |
3222 | #ifdef CONFIG_X86_64 | |
3223 | gva = (long long)(gva << 16) >> 16; | |
3224 | #endif | |
3225 | return gva; | |
3226 | } | |
3227 | ||
08a3732b | 3228 | |
805d32de | 3229 | typedef void (*inspect_spte_fn) (struct kvm *kvm, u64 *sptep); |
08a3732b MT |
3230 | |
3231 | static void __mmu_spte_walk(struct kvm *kvm, struct kvm_mmu_page *sp, | |
3232 | inspect_spte_fn fn) | |
3233 | { | |
3234 | int i; | |
3235 | ||
3236 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) { | |
3237 | u64 ent = sp->spt[i]; | |
3238 | ||
3239 | if (is_shadow_present_pte(ent)) { | |
2920d728 | 3240 | if (!is_last_spte(ent, sp->role.level)) { |
08a3732b MT |
3241 | struct kvm_mmu_page *child; |
3242 | child = page_header(ent & PT64_BASE_ADDR_MASK); | |
3243 | __mmu_spte_walk(kvm, child, fn); | |
2920d728 | 3244 | } else |
805d32de | 3245 | fn(kvm, &sp->spt[i]); |
08a3732b MT |
3246 | } |
3247 | } | |
3248 | } | |
3249 | ||
3250 | static void mmu_spte_walk(struct kvm_vcpu *vcpu, inspect_spte_fn fn) | |
3251 | { | |
3252 | int i; | |
3253 | struct kvm_mmu_page *sp; | |
3254 | ||
3255 | if (!VALID_PAGE(vcpu->arch.mmu.root_hpa)) | |
3256 | return; | |
3257 | if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) { | |
3258 | hpa_t root = vcpu->arch.mmu.root_hpa; | |
3259 | sp = page_header(root); | |
3260 | __mmu_spte_walk(vcpu->kvm, sp, fn); | |
3261 | return; | |
3262 | } | |
3263 | for (i = 0; i < 4; ++i) { | |
3264 | hpa_t root = vcpu->arch.mmu.pae_root[i]; | |
3265 | ||
3266 | if (root && VALID_PAGE(root)) { | |
3267 | root &= PT64_BASE_ADDR_MASK; | |
3268 | sp = page_header(root); | |
3269 | __mmu_spte_walk(vcpu->kvm, sp, fn); | |
3270 | } | |
3271 | } | |
3272 | return; | |
3273 | } | |
3274 | ||
37a7d8b0 AK |
3275 | static void audit_mappings_page(struct kvm_vcpu *vcpu, u64 page_pte, |
3276 | gva_t va, int level) | |
3277 | { | |
3278 | u64 *pt = __va(page_pte & PT64_BASE_ADDR_MASK); | |
3279 | int i; | |
3280 | gva_t va_delta = 1ul << (PAGE_SHIFT + 9 * (level - 1)); | |
3281 | ||
3282 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i, va += va_delta) { | |
3283 | u64 ent = pt[i]; | |
3284 | ||
c7addb90 | 3285 | if (ent == shadow_trap_nonpresent_pte) |
37a7d8b0 AK |
3286 | continue; |
3287 | ||
3288 | va = canonicalize(va); | |
2920d728 MT |
3289 | if (is_shadow_present_pte(ent) && !is_last_spte(ent, level)) |
3290 | audit_mappings_page(vcpu, ent, va, level - 1); | |
3291 | else { | |
1871c602 | 3292 | gpa_t gpa = kvm_mmu_gva_to_gpa_read(vcpu, va, NULL); |
34382539 JK |
3293 | gfn_t gfn = gpa >> PAGE_SHIFT; |
3294 | pfn_t pfn = gfn_to_pfn(vcpu->kvm, gfn); | |
3295 | hpa_t hpa = (hpa_t)pfn << PAGE_SHIFT; | |
37a7d8b0 | 3296 | |
2aaf65e8 MT |
3297 | if (is_error_pfn(pfn)) { |
3298 | kvm_release_pfn_clean(pfn); | |
3299 | continue; | |
3300 | } | |
3301 | ||
c7addb90 | 3302 | if (is_shadow_present_pte(ent) |
37a7d8b0 | 3303 | && (ent & PT64_BASE_ADDR_MASK) != hpa) |
c7addb90 AK |
3304 | printk(KERN_ERR "xx audit error: (%s) levels %d" |
3305 | " gva %lx gpa %llx hpa %llx ent %llx %d\n", | |
ad312c7c | 3306 | audit_msg, vcpu->arch.mmu.root_level, |
d77c26fc MD |
3307 | va, gpa, hpa, ent, |
3308 | is_shadow_present_pte(ent)); | |
c7addb90 AK |
3309 | else if (ent == shadow_notrap_nonpresent_pte |
3310 | && !is_error_hpa(hpa)) | |
3311 | printk(KERN_ERR "audit: (%s) notrap shadow," | |
3312 | " valid guest gva %lx\n", audit_msg, va); | |
35149e21 | 3313 | kvm_release_pfn_clean(pfn); |
c7addb90 | 3314 | |
37a7d8b0 AK |
3315 | } |
3316 | } | |
3317 | } | |
3318 | ||
3319 | static void audit_mappings(struct kvm_vcpu *vcpu) | |
3320 | { | |
1ea252af | 3321 | unsigned i; |
37a7d8b0 | 3322 | |
ad312c7c ZX |
3323 | if (vcpu->arch.mmu.root_level == 4) |
3324 | audit_mappings_page(vcpu, vcpu->arch.mmu.root_hpa, 0, 4); | |
37a7d8b0 AK |
3325 | else |
3326 | for (i = 0; i < 4; ++i) | |
ad312c7c | 3327 | if (vcpu->arch.mmu.pae_root[i] & PT_PRESENT_MASK) |
37a7d8b0 | 3328 | audit_mappings_page(vcpu, |
ad312c7c | 3329 | vcpu->arch.mmu.pae_root[i], |
37a7d8b0 AK |
3330 | i << 30, |
3331 | 2); | |
3332 | } | |
3333 | ||
3334 | static int count_rmaps(struct kvm_vcpu *vcpu) | |
3335 | { | |
805d32de XG |
3336 | struct kvm *kvm = vcpu->kvm; |
3337 | struct kvm_memslots *slots; | |
37a7d8b0 | 3338 | int nmaps = 0; |
bc6678a3 | 3339 | int i, j, k, idx; |
37a7d8b0 | 3340 | |
bc6678a3 | 3341 | idx = srcu_read_lock(&kvm->srcu); |
90d83dc3 | 3342 | slots = kvm_memslots(kvm); |
37a7d8b0 | 3343 | for (i = 0; i < KVM_MEMORY_SLOTS; ++i) { |
bc6678a3 | 3344 | struct kvm_memory_slot *m = &slots->memslots[i]; |
37a7d8b0 AK |
3345 | struct kvm_rmap_desc *d; |
3346 | ||
3347 | for (j = 0; j < m->npages; ++j) { | |
290fc38d | 3348 | unsigned long *rmapp = &m->rmap[j]; |
37a7d8b0 | 3349 | |
290fc38d | 3350 | if (!*rmapp) |
37a7d8b0 | 3351 | continue; |
290fc38d | 3352 | if (!(*rmapp & 1)) { |
37a7d8b0 AK |
3353 | ++nmaps; |
3354 | continue; | |
3355 | } | |
290fc38d | 3356 | d = (struct kvm_rmap_desc *)(*rmapp & ~1ul); |
37a7d8b0 AK |
3357 | while (d) { |
3358 | for (k = 0; k < RMAP_EXT; ++k) | |
d555c333 | 3359 | if (d->sptes[k]) |
37a7d8b0 AK |
3360 | ++nmaps; |
3361 | else | |
3362 | break; | |
3363 | d = d->more; | |
3364 | } | |
3365 | } | |
3366 | } | |
bc6678a3 | 3367 | srcu_read_unlock(&kvm->srcu, idx); |
37a7d8b0 AK |
3368 | return nmaps; |
3369 | } | |
3370 | ||
805d32de | 3371 | void inspect_spte_has_rmap(struct kvm *kvm, u64 *sptep) |
08a3732b MT |
3372 | { |
3373 | unsigned long *rmapp; | |
3374 | struct kvm_mmu_page *rev_sp; | |
3375 | gfn_t gfn; | |
3376 | ||
3377 | if (*sptep & PT_WRITABLE_MASK) { | |
3378 | rev_sp = page_header(__pa(sptep)); | |
3379 | gfn = rev_sp->gfns[sptep - rev_sp->spt]; | |
3380 | ||
3381 | if (!gfn_to_memslot(kvm, gfn)) { | |
3382 | if (!printk_ratelimit()) | |
3383 | return; | |
3384 | printk(KERN_ERR "%s: no memslot for gfn %ld\n", | |
3385 | audit_msg, gfn); | |
3386 | printk(KERN_ERR "%s: index %ld of sp (gfn=%lx)\n", | |
805d32de | 3387 | audit_msg, (long int)(sptep - rev_sp->spt), |
08a3732b MT |
3388 | rev_sp->gfn); |
3389 | dump_stack(); | |
3390 | return; | |
3391 | } | |
3392 | ||
2920d728 | 3393 | rmapp = gfn_to_rmap(kvm, rev_sp->gfns[sptep - rev_sp->spt], |
805d32de | 3394 | rev_sp->role.level); |
08a3732b MT |
3395 | if (!*rmapp) { |
3396 | if (!printk_ratelimit()) | |
3397 | return; | |
3398 | printk(KERN_ERR "%s: no rmap for writable spte %llx\n", | |
3399 | audit_msg, *sptep); | |
3400 | dump_stack(); | |
3401 | } | |
3402 | } | |
3403 | ||
3404 | } | |
3405 | ||
3406 | void audit_writable_sptes_have_rmaps(struct kvm_vcpu *vcpu) | |
3407 | { | |
3408 | mmu_spte_walk(vcpu, inspect_spte_has_rmap); | |
3409 | } | |
3410 | ||
3411 | static void check_writable_mappings_rmap(struct kvm_vcpu *vcpu) | |
37a7d8b0 | 3412 | { |
4db35314 | 3413 | struct kvm_mmu_page *sp; |
37a7d8b0 AK |
3414 | int i; |
3415 | ||
f05e70ac | 3416 | list_for_each_entry(sp, &vcpu->kvm->arch.active_mmu_pages, link) { |
4db35314 | 3417 | u64 *pt = sp->spt; |
37a7d8b0 | 3418 | |
4db35314 | 3419 | if (sp->role.level != PT_PAGE_TABLE_LEVEL) |
37a7d8b0 AK |
3420 | continue; |
3421 | ||
3422 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) { | |
3423 | u64 ent = pt[i]; | |
3424 | ||
3425 | if (!(ent & PT_PRESENT_MASK)) | |
3426 | continue; | |
3427 | if (!(ent & PT_WRITABLE_MASK)) | |
3428 | continue; | |
805d32de | 3429 | inspect_spte_has_rmap(vcpu->kvm, &pt[i]); |
37a7d8b0 AK |
3430 | } |
3431 | } | |
08a3732b | 3432 | return; |
37a7d8b0 AK |
3433 | } |
3434 | ||
3435 | static void audit_rmap(struct kvm_vcpu *vcpu) | |
3436 | { | |
08a3732b MT |
3437 | check_writable_mappings_rmap(vcpu); |
3438 | count_rmaps(vcpu); | |
37a7d8b0 AK |
3439 | } |
3440 | ||
3441 | static void audit_write_protection(struct kvm_vcpu *vcpu) | |
3442 | { | |
4db35314 | 3443 | struct kvm_mmu_page *sp; |
290fc38d IE |
3444 | struct kvm_memory_slot *slot; |
3445 | unsigned long *rmapp; | |
e58b0f9e | 3446 | u64 *spte; |
290fc38d | 3447 | gfn_t gfn; |
37a7d8b0 | 3448 | |
f05e70ac | 3449 | list_for_each_entry(sp, &vcpu->kvm->arch.active_mmu_pages, link) { |
f6e2c02b | 3450 | if (sp->role.direct) |
37a7d8b0 | 3451 | continue; |
e58b0f9e MT |
3452 | if (sp->unsync) |
3453 | continue; | |
37a7d8b0 | 3454 | |
4db35314 | 3455 | gfn = unalias_gfn(vcpu->kvm, sp->gfn); |
2843099f | 3456 | slot = gfn_to_memslot_unaliased(vcpu->kvm, sp->gfn); |
290fc38d | 3457 | rmapp = &slot->rmap[gfn - slot->base_gfn]; |
e58b0f9e MT |
3458 | |
3459 | spte = rmap_next(vcpu->kvm, rmapp, NULL); | |
3460 | while (spte) { | |
3461 | if (*spte & PT_WRITABLE_MASK) | |
3462 | printk(KERN_ERR "%s: (%s) shadow page has " | |
3463 | "writable mappings: gfn %lx role %x\n", | |
b8688d51 | 3464 | __func__, audit_msg, sp->gfn, |
4db35314 | 3465 | sp->role.word); |
e58b0f9e MT |
3466 | spte = rmap_next(vcpu->kvm, rmapp, spte); |
3467 | } | |
37a7d8b0 AK |
3468 | } |
3469 | } | |
3470 | ||
3471 | static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg) | |
3472 | { | |
3473 | int olddbg = dbg; | |
3474 | ||
3475 | dbg = 0; | |
3476 | audit_msg = msg; | |
3477 | audit_rmap(vcpu); | |
3478 | audit_write_protection(vcpu); | |
2aaf65e8 MT |
3479 | if (strcmp("pre pte write", audit_msg) != 0) |
3480 | audit_mappings(vcpu); | |
08a3732b | 3481 | audit_writable_sptes_have_rmaps(vcpu); |
37a7d8b0 AK |
3482 | dbg = olddbg; |
3483 | } | |
3484 | ||
3485 | #endif |