]>
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. | |
9611c187 | 10 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
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 | |
af585b92 | 21 | #include "irq.h" |
1d737c8a | 22 | #include "mmu.h" |
836a1b3c | 23 | #include "x86.h" |
6de4f3ad | 24 | #include "kvm_cache_regs.h" |
5f7dde7b | 25 | #include "cpuid.h" |
e495606d | 26 | |
edf88417 | 27 | #include <linux/kvm_host.h> |
6aa8b732 AK |
28 | #include <linux/types.h> |
29 | #include <linux/string.h> | |
6aa8b732 AK |
30 | #include <linux/mm.h> |
31 | #include <linux/highmem.h> | |
32 | #include <linux/module.h> | |
448353ca | 33 | #include <linux/swap.h> |
05da4558 | 34 | #include <linux/hugetlb.h> |
2f333bcb | 35 | #include <linux/compiler.h> |
bc6678a3 | 36 | #include <linux/srcu.h> |
5a0e3ad6 | 37 | #include <linux/slab.h> |
bf998156 | 38 | #include <linux/uaccess.h> |
6aa8b732 | 39 | |
e495606d AK |
40 | #include <asm/page.h> |
41 | #include <asm/cmpxchg.h> | |
4e542370 | 42 | #include <asm/io.h> |
13673a90 | 43 | #include <asm/vmx.h> |
6aa8b732 | 44 | |
18552672 JR |
45 | /* |
46 | * When setting this variable to true it enables Two-Dimensional-Paging | |
47 | * where the hardware walks 2 page tables: | |
48 | * 1. the guest-virtual to guest-physical | |
49 | * 2. while doing 1. it walks guest-physical to host-physical | |
50 | * If the hardware supports that we don't need to do shadow paging. | |
51 | */ | |
2f333bcb | 52 | bool tdp_enabled = false; |
18552672 | 53 | |
8b1fe17c XG |
54 | enum { |
55 | AUDIT_PRE_PAGE_FAULT, | |
56 | AUDIT_POST_PAGE_FAULT, | |
57 | AUDIT_PRE_PTE_WRITE, | |
6903074c XG |
58 | AUDIT_POST_PTE_WRITE, |
59 | AUDIT_PRE_SYNC, | |
60 | AUDIT_POST_SYNC | |
8b1fe17c | 61 | }; |
37a7d8b0 | 62 | |
8b1fe17c | 63 | #undef MMU_DEBUG |
37a7d8b0 AK |
64 | |
65 | #ifdef MMU_DEBUG | |
fa4a2c08 PB |
66 | static bool dbg = 0; |
67 | module_param(dbg, bool, 0644); | |
37a7d8b0 AK |
68 | |
69 | #define pgprintk(x...) do { if (dbg) printk(x); } while (0) | |
70 | #define rmap_printk(x...) do { if (dbg) printk(x); } while (0) | |
fa4a2c08 | 71 | #define MMU_WARN_ON(x) WARN_ON(x) |
37a7d8b0 | 72 | #else |
37a7d8b0 AK |
73 | #define pgprintk(x...) do { } while (0) |
74 | #define rmap_printk(x...) do { } while (0) | |
fa4a2c08 | 75 | #define MMU_WARN_ON(x) do { } while (0) |
d6c69ee9 | 76 | #endif |
6aa8b732 | 77 | |
957ed9ef XG |
78 | #define PTE_PREFETCH_NUM 8 |
79 | ||
00763e41 | 80 | #define PT_FIRST_AVAIL_BITS_SHIFT 10 |
6aa8b732 AK |
81 | #define PT64_SECOND_AVAIL_BITS_SHIFT 52 |
82 | ||
6aa8b732 AK |
83 | #define PT64_LEVEL_BITS 9 |
84 | ||
85 | #define PT64_LEVEL_SHIFT(level) \ | |
d77c26fc | 86 | (PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS) |
6aa8b732 | 87 | |
6aa8b732 AK |
88 | #define PT64_INDEX(address, level)\ |
89 | (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1)) | |
90 | ||
91 | ||
92 | #define PT32_LEVEL_BITS 10 | |
93 | ||
94 | #define PT32_LEVEL_SHIFT(level) \ | |
d77c26fc | 95 | (PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS) |
6aa8b732 | 96 | |
e04da980 JR |
97 | #define PT32_LVL_OFFSET_MASK(level) \ |
98 | (PT32_BASE_ADDR_MASK & ((1ULL << (PAGE_SHIFT + (((level) - 1) \ | |
99 | * PT32_LEVEL_BITS))) - 1)) | |
6aa8b732 AK |
100 | |
101 | #define PT32_INDEX(address, level)\ | |
102 | (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1)) | |
103 | ||
104 | ||
27aba766 | 105 | #define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1)) |
6aa8b732 AK |
106 | #define PT64_DIR_BASE_ADDR_MASK \ |
107 | (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1)) | |
e04da980 JR |
108 | #define PT64_LVL_ADDR_MASK(level) \ |
109 | (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + (((level) - 1) \ | |
110 | * PT64_LEVEL_BITS))) - 1)) | |
111 | #define PT64_LVL_OFFSET_MASK(level) \ | |
112 | (PT64_BASE_ADDR_MASK & ((1ULL << (PAGE_SHIFT + (((level) - 1) \ | |
113 | * PT64_LEVEL_BITS))) - 1)) | |
6aa8b732 AK |
114 | |
115 | #define PT32_BASE_ADDR_MASK PAGE_MASK | |
116 | #define PT32_DIR_BASE_ADDR_MASK \ | |
117 | (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1)) | |
e04da980 JR |
118 | #define PT32_LVL_ADDR_MASK(level) \ |
119 | (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + (((level) - 1) \ | |
120 | * PT32_LEVEL_BITS))) - 1)) | |
6aa8b732 | 121 | |
53166229 GN |
122 | #define PT64_PERM_MASK (PT_PRESENT_MASK | PT_WRITABLE_MASK | shadow_user_mask \ |
123 | | shadow_x_mask | shadow_nx_mask) | |
6aa8b732 | 124 | |
fe135d2c AK |
125 | #define ACC_EXEC_MASK 1 |
126 | #define ACC_WRITE_MASK PT_WRITABLE_MASK | |
127 | #define ACC_USER_MASK PT_USER_MASK | |
128 | #define ACC_ALL (ACC_EXEC_MASK | ACC_WRITE_MASK | ACC_USER_MASK) | |
129 | ||
90bb6fc5 AK |
130 | #include <trace/events/kvm.h> |
131 | ||
07420171 AK |
132 | #define CREATE_TRACE_POINTS |
133 | #include "mmutrace.h" | |
134 | ||
49fde340 XG |
135 | #define SPTE_HOST_WRITEABLE (1ULL << PT_FIRST_AVAIL_BITS_SHIFT) |
136 | #define SPTE_MMU_WRITEABLE (1ULL << (PT_FIRST_AVAIL_BITS_SHIFT + 1)) | |
1403283a | 137 | |
135f8c2b AK |
138 | #define SHADOW_PT_INDEX(addr, level) PT64_INDEX(addr, level) |
139 | ||
220f773a TY |
140 | /* make pte_list_desc fit well in cache line */ |
141 | #define PTE_LIST_EXT 3 | |
142 | ||
53c07b18 XG |
143 | struct pte_list_desc { |
144 | u64 *sptes[PTE_LIST_EXT]; | |
145 | struct pte_list_desc *more; | |
cd4a4e53 AK |
146 | }; |
147 | ||
2d11123a AK |
148 | struct kvm_shadow_walk_iterator { |
149 | u64 addr; | |
150 | hpa_t shadow_addr; | |
2d11123a | 151 | u64 *sptep; |
dd3bfd59 | 152 | int level; |
2d11123a AK |
153 | unsigned index; |
154 | }; | |
155 | ||
156 | #define for_each_shadow_entry(_vcpu, _addr, _walker) \ | |
157 | for (shadow_walk_init(&(_walker), _vcpu, _addr); \ | |
158 | shadow_walk_okay(&(_walker)); \ | |
159 | shadow_walk_next(&(_walker))) | |
160 | ||
c2a2ac2b XG |
161 | #define for_each_shadow_entry_lockless(_vcpu, _addr, _walker, spte) \ |
162 | for (shadow_walk_init(&(_walker), _vcpu, _addr); \ | |
163 | shadow_walk_okay(&(_walker)) && \ | |
164 | ({ spte = mmu_spte_get_lockless(_walker.sptep); 1; }); \ | |
165 | __shadow_walk_next(&(_walker), spte)) | |
166 | ||
53c07b18 | 167 | static struct kmem_cache *pte_list_desc_cache; |
d3d25b04 | 168 | static struct kmem_cache *mmu_page_header_cache; |
45221ab6 | 169 | static struct percpu_counter kvm_total_used_mmu_pages; |
b5a33a75 | 170 | |
7b52345e SY |
171 | static u64 __read_mostly shadow_nx_mask; |
172 | static u64 __read_mostly shadow_x_mask; /* mutual exclusive with nx_mask */ | |
173 | static u64 __read_mostly shadow_user_mask; | |
174 | static u64 __read_mostly shadow_accessed_mask; | |
175 | static u64 __read_mostly shadow_dirty_mask; | |
ce88decf XG |
176 | static u64 __read_mostly shadow_mmio_mask; |
177 | ||
178 | static void mmu_spte_set(u64 *sptep, u64 spte); | |
e676505a | 179 | static void mmu_free_roots(struct kvm_vcpu *vcpu); |
ce88decf XG |
180 | |
181 | void kvm_mmu_set_mmio_spte_mask(u64 mmio_mask) | |
182 | { | |
183 | shadow_mmio_mask = mmio_mask; | |
184 | } | |
185 | EXPORT_SYMBOL_GPL(kvm_mmu_set_mmio_spte_mask); | |
186 | ||
f2fd125d | 187 | /* |
ee3d1570 DM |
188 | * the low bit of the generation number is always presumed to be zero. |
189 | * This disables mmio caching during memslot updates. The concept is | |
190 | * similar to a seqcount but instead of retrying the access we just punt | |
191 | * and ignore the cache. | |
192 | * | |
193 | * spte bits 3-11 are used as bits 1-9 of the generation number, | |
194 | * the bits 52-61 are used as bits 10-19 of the generation number. | |
f2fd125d | 195 | */ |
ee3d1570 | 196 | #define MMIO_SPTE_GEN_LOW_SHIFT 2 |
f2fd125d XG |
197 | #define MMIO_SPTE_GEN_HIGH_SHIFT 52 |
198 | ||
ee3d1570 DM |
199 | #define MMIO_GEN_SHIFT 20 |
200 | #define MMIO_GEN_LOW_SHIFT 10 | |
201 | #define MMIO_GEN_LOW_MASK ((1 << MMIO_GEN_LOW_SHIFT) - 2) | |
f8f55942 | 202 | #define MMIO_GEN_MASK ((1 << MMIO_GEN_SHIFT) - 1) |
f2fd125d XG |
203 | |
204 | static u64 generation_mmio_spte_mask(unsigned int gen) | |
205 | { | |
206 | u64 mask; | |
207 | ||
842bb26a | 208 | WARN_ON(gen & ~MMIO_GEN_MASK); |
f2fd125d XG |
209 | |
210 | mask = (gen & MMIO_GEN_LOW_MASK) << MMIO_SPTE_GEN_LOW_SHIFT; | |
211 | mask |= ((u64)gen >> MMIO_GEN_LOW_SHIFT) << MMIO_SPTE_GEN_HIGH_SHIFT; | |
212 | return mask; | |
213 | } | |
214 | ||
215 | static unsigned int get_mmio_spte_generation(u64 spte) | |
216 | { | |
217 | unsigned int gen; | |
218 | ||
219 | spte &= ~shadow_mmio_mask; | |
220 | ||
221 | gen = (spte >> MMIO_SPTE_GEN_LOW_SHIFT) & MMIO_GEN_LOW_MASK; | |
222 | gen |= (spte >> MMIO_SPTE_GEN_HIGH_SHIFT) << MMIO_GEN_LOW_SHIFT; | |
223 | return gen; | |
224 | } | |
225 | ||
54bf36aa | 226 | static unsigned int kvm_current_mmio_generation(struct kvm_vcpu *vcpu) |
f8f55942 | 227 | { |
54bf36aa | 228 | return kvm_vcpu_memslots(vcpu)->generation & MMIO_GEN_MASK; |
f8f55942 XG |
229 | } |
230 | ||
54bf36aa | 231 | static void mark_mmio_spte(struct kvm_vcpu *vcpu, u64 *sptep, u64 gfn, |
f2fd125d | 232 | unsigned access) |
ce88decf | 233 | { |
54bf36aa | 234 | unsigned int gen = kvm_current_mmio_generation(vcpu); |
f8f55942 | 235 | u64 mask = generation_mmio_spte_mask(gen); |
95b0430d | 236 | |
ce88decf | 237 | access &= ACC_WRITE_MASK | ACC_USER_MASK; |
f2fd125d | 238 | mask |= shadow_mmio_mask | access | gfn << PAGE_SHIFT; |
f2fd125d | 239 | |
f8f55942 | 240 | trace_mark_mmio_spte(sptep, gfn, access, gen); |
f2fd125d | 241 | mmu_spte_set(sptep, mask); |
ce88decf XG |
242 | } |
243 | ||
244 | static bool is_mmio_spte(u64 spte) | |
245 | { | |
246 | return (spte & shadow_mmio_mask) == shadow_mmio_mask; | |
247 | } | |
248 | ||
249 | static gfn_t get_mmio_spte_gfn(u64 spte) | |
250 | { | |
842bb26a | 251 | u64 mask = generation_mmio_spte_mask(MMIO_GEN_MASK) | shadow_mmio_mask; |
f2fd125d | 252 | return (spte & ~mask) >> PAGE_SHIFT; |
ce88decf XG |
253 | } |
254 | ||
255 | static unsigned get_mmio_spte_access(u64 spte) | |
256 | { | |
842bb26a | 257 | u64 mask = generation_mmio_spte_mask(MMIO_GEN_MASK) | shadow_mmio_mask; |
f2fd125d | 258 | return (spte & ~mask) & ~PAGE_MASK; |
ce88decf XG |
259 | } |
260 | ||
54bf36aa | 261 | static bool set_mmio_spte(struct kvm_vcpu *vcpu, u64 *sptep, gfn_t gfn, |
f2fd125d | 262 | pfn_t pfn, unsigned access) |
ce88decf XG |
263 | { |
264 | if (unlikely(is_noslot_pfn(pfn))) { | |
54bf36aa | 265 | mark_mmio_spte(vcpu, sptep, gfn, access); |
ce88decf XG |
266 | return true; |
267 | } | |
268 | ||
269 | return false; | |
270 | } | |
c7addb90 | 271 | |
54bf36aa | 272 | static bool check_mmio_spte(struct kvm_vcpu *vcpu, u64 spte) |
f8f55942 | 273 | { |
089504c0 XG |
274 | unsigned int kvm_gen, spte_gen; |
275 | ||
54bf36aa | 276 | kvm_gen = kvm_current_mmio_generation(vcpu); |
089504c0 XG |
277 | spte_gen = get_mmio_spte_generation(spte); |
278 | ||
279 | trace_check_mmio_spte(spte, kvm_gen, spte_gen); | |
280 | return likely(kvm_gen == spte_gen); | |
f8f55942 XG |
281 | } |
282 | ||
7b52345e | 283 | void kvm_mmu_set_mask_ptes(u64 user_mask, u64 accessed_mask, |
4b12f0de | 284 | u64 dirty_mask, u64 nx_mask, u64 x_mask) |
7b52345e SY |
285 | { |
286 | shadow_user_mask = user_mask; | |
287 | shadow_accessed_mask = accessed_mask; | |
288 | shadow_dirty_mask = dirty_mask; | |
289 | shadow_nx_mask = nx_mask; | |
290 | shadow_x_mask = x_mask; | |
291 | } | |
292 | EXPORT_SYMBOL_GPL(kvm_mmu_set_mask_ptes); | |
293 | ||
6aa8b732 AK |
294 | static int is_cpuid_PSE36(void) |
295 | { | |
296 | return 1; | |
297 | } | |
298 | ||
73b1087e AK |
299 | static int is_nx(struct kvm_vcpu *vcpu) |
300 | { | |
f6801dff | 301 | return vcpu->arch.efer & EFER_NX; |
73b1087e AK |
302 | } |
303 | ||
c7addb90 AK |
304 | static int is_shadow_present_pte(u64 pte) |
305 | { | |
ce88decf | 306 | return pte & PT_PRESENT_MASK && !is_mmio_spte(pte); |
c7addb90 AK |
307 | } |
308 | ||
05da4558 MT |
309 | static int is_large_pte(u64 pte) |
310 | { | |
311 | return pte & PT_PAGE_SIZE_MASK; | |
312 | } | |
313 | ||
43a3795a | 314 | static int is_rmap_spte(u64 pte) |
cd4a4e53 | 315 | { |
4b1a80fa | 316 | return is_shadow_present_pte(pte); |
cd4a4e53 AK |
317 | } |
318 | ||
776e6633 MT |
319 | static int is_last_spte(u64 pte, int level) |
320 | { | |
321 | if (level == PT_PAGE_TABLE_LEVEL) | |
322 | return 1; | |
852e3c19 | 323 | if (is_large_pte(pte)) |
776e6633 MT |
324 | return 1; |
325 | return 0; | |
326 | } | |
327 | ||
35149e21 | 328 | static pfn_t spte_to_pfn(u64 pte) |
0b49ea86 | 329 | { |
35149e21 | 330 | return (pte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT; |
0b49ea86 AK |
331 | } |
332 | ||
da928521 AK |
333 | static gfn_t pse36_gfn_delta(u32 gpte) |
334 | { | |
335 | int shift = 32 - PT32_DIR_PSE36_SHIFT - PAGE_SHIFT; | |
336 | ||
337 | return (gpte & PT32_DIR_PSE36_MASK) << shift; | |
338 | } | |
339 | ||
603e0651 | 340 | #ifdef CONFIG_X86_64 |
d555c333 | 341 | static void __set_spte(u64 *sptep, u64 spte) |
e663ee64 | 342 | { |
603e0651 | 343 | *sptep = spte; |
e663ee64 AK |
344 | } |
345 | ||
603e0651 | 346 | static void __update_clear_spte_fast(u64 *sptep, u64 spte) |
a9221dd5 | 347 | { |
603e0651 XG |
348 | *sptep = spte; |
349 | } | |
350 | ||
351 | static u64 __update_clear_spte_slow(u64 *sptep, u64 spte) | |
352 | { | |
353 | return xchg(sptep, spte); | |
354 | } | |
c2a2ac2b XG |
355 | |
356 | static u64 __get_spte_lockless(u64 *sptep) | |
357 | { | |
358 | return ACCESS_ONCE(*sptep); | |
359 | } | |
ce88decf XG |
360 | |
361 | static bool __check_direct_spte_mmio_pf(u64 spte) | |
362 | { | |
363 | /* It is valid if the spte is zapped. */ | |
364 | return spte == 0ull; | |
365 | } | |
a9221dd5 | 366 | #else |
603e0651 XG |
367 | union split_spte { |
368 | struct { | |
369 | u32 spte_low; | |
370 | u32 spte_high; | |
371 | }; | |
372 | u64 spte; | |
373 | }; | |
a9221dd5 | 374 | |
c2a2ac2b XG |
375 | static void count_spte_clear(u64 *sptep, u64 spte) |
376 | { | |
377 | struct kvm_mmu_page *sp = page_header(__pa(sptep)); | |
378 | ||
379 | if (is_shadow_present_pte(spte)) | |
380 | return; | |
381 | ||
382 | /* Ensure the spte is completely set before we increase the count */ | |
383 | smp_wmb(); | |
384 | sp->clear_spte_count++; | |
385 | } | |
386 | ||
603e0651 XG |
387 | static void __set_spte(u64 *sptep, u64 spte) |
388 | { | |
389 | union split_spte *ssptep, sspte; | |
a9221dd5 | 390 | |
603e0651 XG |
391 | ssptep = (union split_spte *)sptep; |
392 | sspte = (union split_spte)spte; | |
393 | ||
394 | ssptep->spte_high = sspte.spte_high; | |
395 | ||
396 | /* | |
397 | * If we map the spte from nonpresent to present, We should store | |
398 | * the high bits firstly, then set present bit, so cpu can not | |
399 | * fetch this spte while we are setting the spte. | |
400 | */ | |
401 | smp_wmb(); | |
402 | ||
403 | ssptep->spte_low = sspte.spte_low; | |
a9221dd5 AK |
404 | } |
405 | ||
603e0651 XG |
406 | static void __update_clear_spte_fast(u64 *sptep, u64 spte) |
407 | { | |
408 | union split_spte *ssptep, sspte; | |
409 | ||
410 | ssptep = (union split_spte *)sptep; | |
411 | sspte = (union split_spte)spte; | |
412 | ||
413 | ssptep->spte_low = sspte.spte_low; | |
414 | ||
415 | /* | |
416 | * If we map the spte from present to nonpresent, we should clear | |
417 | * present bit firstly to avoid vcpu fetch the old high bits. | |
418 | */ | |
419 | smp_wmb(); | |
420 | ||
421 | ssptep->spte_high = sspte.spte_high; | |
c2a2ac2b | 422 | count_spte_clear(sptep, spte); |
603e0651 XG |
423 | } |
424 | ||
425 | static u64 __update_clear_spte_slow(u64 *sptep, u64 spte) | |
426 | { | |
427 | union split_spte *ssptep, sspte, orig; | |
428 | ||
429 | ssptep = (union split_spte *)sptep; | |
430 | sspte = (union split_spte)spte; | |
431 | ||
432 | /* xchg acts as a barrier before the setting of the high bits */ | |
433 | orig.spte_low = xchg(&ssptep->spte_low, sspte.spte_low); | |
41bc3186 ZJ |
434 | orig.spte_high = ssptep->spte_high; |
435 | ssptep->spte_high = sspte.spte_high; | |
c2a2ac2b | 436 | count_spte_clear(sptep, spte); |
603e0651 XG |
437 | |
438 | return orig.spte; | |
439 | } | |
c2a2ac2b XG |
440 | |
441 | /* | |
442 | * The idea using the light way get the spte on x86_32 guest is from | |
443 | * gup_get_pte(arch/x86/mm/gup.c). | |
accaefe0 XG |
444 | * |
445 | * An spte tlb flush may be pending, because kvm_set_pte_rmapp | |
446 | * coalesces them and we are running out of the MMU lock. Therefore | |
447 | * we need to protect against in-progress updates of the spte. | |
448 | * | |
449 | * Reading the spte while an update is in progress may get the old value | |
450 | * for the high part of the spte. The race is fine for a present->non-present | |
451 | * change (because the high part of the spte is ignored for non-present spte), | |
452 | * but for a present->present change we must reread the spte. | |
453 | * | |
454 | * All such changes are done in two steps (present->non-present and | |
455 | * non-present->present), hence it is enough to count the number of | |
456 | * present->non-present updates: if it changed while reading the spte, | |
457 | * we might have hit the race. This is done using clear_spte_count. | |
c2a2ac2b XG |
458 | */ |
459 | static u64 __get_spte_lockless(u64 *sptep) | |
460 | { | |
461 | struct kvm_mmu_page *sp = page_header(__pa(sptep)); | |
462 | union split_spte spte, *orig = (union split_spte *)sptep; | |
463 | int count; | |
464 | ||
465 | retry: | |
466 | count = sp->clear_spte_count; | |
467 | smp_rmb(); | |
468 | ||
469 | spte.spte_low = orig->spte_low; | |
470 | smp_rmb(); | |
471 | ||
472 | spte.spte_high = orig->spte_high; | |
473 | smp_rmb(); | |
474 | ||
475 | if (unlikely(spte.spte_low != orig->spte_low || | |
476 | count != sp->clear_spte_count)) | |
477 | goto retry; | |
478 | ||
479 | return spte.spte; | |
480 | } | |
ce88decf XG |
481 | |
482 | static bool __check_direct_spte_mmio_pf(u64 spte) | |
483 | { | |
484 | union split_spte sspte = (union split_spte)spte; | |
485 | u32 high_mmio_mask = shadow_mmio_mask >> 32; | |
486 | ||
487 | /* It is valid if the spte is zapped. */ | |
488 | if (spte == 0ull) | |
489 | return true; | |
490 | ||
491 | /* It is valid if the spte is being zapped. */ | |
492 | if (sspte.spte_low == 0ull && | |
493 | (sspte.spte_high & high_mmio_mask) == high_mmio_mask) | |
494 | return true; | |
495 | ||
496 | return false; | |
497 | } | |
603e0651 XG |
498 | #endif |
499 | ||
c7ba5b48 XG |
500 | static bool spte_is_locklessly_modifiable(u64 spte) |
501 | { | |
feb3eb70 GN |
502 | return (spte & (SPTE_HOST_WRITEABLE | SPTE_MMU_WRITEABLE)) == |
503 | (SPTE_HOST_WRITEABLE | SPTE_MMU_WRITEABLE); | |
c7ba5b48 XG |
504 | } |
505 | ||
8672b721 XG |
506 | static bool spte_has_volatile_bits(u64 spte) |
507 | { | |
c7ba5b48 XG |
508 | /* |
509 | * Always atomicly update spte if it can be updated | |
510 | * out of mmu-lock, it can ensure dirty bit is not lost, | |
511 | * also, it can help us to get a stable is_writable_pte() | |
512 | * to ensure tlb flush is not missed. | |
513 | */ | |
514 | if (spte_is_locklessly_modifiable(spte)) | |
515 | return true; | |
516 | ||
8672b721 XG |
517 | if (!shadow_accessed_mask) |
518 | return false; | |
519 | ||
520 | if (!is_shadow_present_pte(spte)) | |
521 | return false; | |
522 | ||
4132779b XG |
523 | if ((spte & shadow_accessed_mask) && |
524 | (!is_writable_pte(spte) || (spte & shadow_dirty_mask))) | |
8672b721 XG |
525 | return false; |
526 | ||
527 | return true; | |
528 | } | |
529 | ||
4132779b XG |
530 | static bool spte_is_bit_cleared(u64 old_spte, u64 new_spte, u64 bit_mask) |
531 | { | |
532 | return (old_spte & bit_mask) && !(new_spte & bit_mask); | |
533 | } | |
534 | ||
7e71a59b KH |
535 | static bool spte_is_bit_changed(u64 old_spte, u64 new_spte, u64 bit_mask) |
536 | { | |
537 | return (old_spte & bit_mask) != (new_spte & bit_mask); | |
538 | } | |
539 | ||
1df9f2dc XG |
540 | /* Rules for using mmu_spte_set: |
541 | * Set the sptep from nonpresent to present. | |
542 | * Note: the sptep being assigned *must* be either not present | |
543 | * or in a state where the hardware will not attempt to update | |
544 | * the spte. | |
545 | */ | |
546 | static void mmu_spte_set(u64 *sptep, u64 new_spte) | |
547 | { | |
548 | WARN_ON(is_shadow_present_pte(*sptep)); | |
549 | __set_spte(sptep, new_spte); | |
550 | } | |
551 | ||
552 | /* Rules for using mmu_spte_update: | |
553 | * Update the state bits, it means the mapped pfn is not changged. | |
6e7d0354 XG |
554 | * |
555 | * Whenever we overwrite a writable spte with a read-only one we | |
556 | * should flush remote TLBs. Otherwise rmap_write_protect | |
557 | * will find a read-only spte, even though the writable spte | |
558 | * might be cached on a CPU's TLB, the return value indicates this | |
559 | * case. | |
1df9f2dc | 560 | */ |
6e7d0354 | 561 | static bool mmu_spte_update(u64 *sptep, u64 new_spte) |
b79b93f9 | 562 | { |
c7ba5b48 | 563 | u64 old_spte = *sptep; |
6e7d0354 | 564 | bool ret = false; |
4132779b XG |
565 | |
566 | WARN_ON(!is_rmap_spte(new_spte)); | |
b79b93f9 | 567 | |
6e7d0354 XG |
568 | if (!is_shadow_present_pte(old_spte)) { |
569 | mmu_spte_set(sptep, new_spte); | |
570 | return ret; | |
571 | } | |
4132779b | 572 | |
c7ba5b48 | 573 | if (!spte_has_volatile_bits(old_spte)) |
603e0651 | 574 | __update_clear_spte_fast(sptep, new_spte); |
4132779b | 575 | else |
603e0651 | 576 | old_spte = __update_clear_spte_slow(sptep, new_spte); |
4132779b | 577 | |
c7ba5b48 XG |
578 | /* |
579 | * For the spte updated out of mmu-lock is safe, since | |
580 | * we always atomicly update it, see the comments in | |
581 | * spte_has_volatile_bits(). | |
582 | */ | |
7f31c959 XG |
583 | if (spte_is_locklessly_modifiable(old_spte) && |
584 | !is_writable_pte(new_spte)) | |
6e7d0354 XG |
585 | ret = true; |
586 | ||
4132779b | 587 | if (!shadow_accessed_mask) |
6e7d0354 | 588 | return ret; |
4132779b | 589 | |
7e71a59b KH |
590 | /* |
591 | * Flush TLB when accessed/dirty bits are changed in the page tables, | |
592 | * to guarantee consistency between TLB and page tables. | |
593 | */ | |
594 | if (spte_is_bit_changed(old_spte, new_spte, | |
595 | shadow_accessed_mask | shadow_dirty_mask)) | |
596 | ret = true; | |
597 | ||
4132779b XG |
598 | if (spte_is_bit_cleared(old_spte, new_spte, shadow_accessed_mask)) |
599 | kvm_set_pfn_accessed(spte_to_pfn(old_spte)); | |
600 | if (spte_is_bit_cleared(old_spte, new_spte, shadow_dirty_mask)) | |
601 | kvm_set_pfn_dirty(spte_to_pfn(old_spte)); | |
6e7d0354 XG |
602 | |
603 | return ret; | |
b79b93f9 AK |
604 | } |
605 | ||
1df9f2dc XG |
606 | /* |
607 | * Rules for using mmu_spte_clear_track_bits: | |
608 | * It sets the sptep from present to nonpresent, and track the | |
609 | * state bits, it is used to clear the last level sptep. | |
610 | */ | |
611 | static int mmu_spte_clear_track_bits(u64 *sptep) | |
612 | { | |
613 | pfn_t pfn; | |
614 | u64 old_spte = *sptep; | |
615 | ||
616 | if (!spte_has_volatile_bits(old_spte)) | |
603e0651 | 617 | __update_clear_spte_fast(sptep, 0ull); |
1df9f2dc | 618 | else |
603e0651 | 619 | old_spte = __update_clear_spte_slow(sptep, 0ull); |
1df9f2dc XG |
620 | |
621 | if (!is_rmap_spte(old_spte)) | |
622 | return 0; | |
623 | ||
624 | pfn = spte_to_pfn(old_spte); | |
86fde74c XG |
625 | |
626 | /* | |
627 | * KVM does not hold the refcount of the page used by | |
628 | * kvm mmu, before reclaiming the page, we should | |
629 | * unmap it from mmu first. | |
630 | */ | |
bf4bea8e | 631 | WARN_ON(!kvm_is_reserved_pfn(pfn) && !page_count(pfn_to_page(pfn))); |
86fde74c | 632 | |
1df9f2dc XG |
633 | if (!shadow_accessed_mask || old_spte & shadow_accessed_mask) |
634 | kvm_set_pfn_accessed(pfn); | |
635 | if (!shadow_dirty_mask || (old_spte & shadow_dirty_mask)) | |
636 | kvm_set_pfn_dirty(pfn); | |
637 | return 1; | |
638 | } | |
639 | ||
640 | /* | |
641 | * Rules for using mmu_spte_clear_no_track: | |
642 | * Directly clear spte without caring the state bits of sptep, | |
643 | * it is used to set the upper level spte. | |
644 | */ | |
645 | static void mmu_spte_clear_no_track(u64 *sptep) | |
646 | { | |
603e0651 | 647 | __update_clear_spte_fast(sptep, 0ull); |
1df9f2dc XG |
648 | } |
649 | ||
c2a2ac2b XG |
650 | static u64 mmu_spte_get_lockless(u64 *sptep) |
651 | { | |
652 | return __get_spte_lockless(sptep); | |
653 | } | |
654 | ||
655 | static void walk_shadow_page_lockless_begin(struct kvm_vcpu *vcpu) | |
656 | { | |
c142786c AK |
657 | /* |
658 | * Prevent page table teardown by making any free-er wait during | |
659 | * kvm_flush_remote_tlbs() IPI to all active vcpus. | |
660 | */ | |
661 | local_irq_disable(); | |
662 | vcpu->mode = READING_SHADOW_PAGE_TABLES; | |
663 | /* | |
664 | * Make sure a following spte read is not reordered ahead of the write | |
665 | * to vcpu->mode. | |
666 | */ | |
667 | smp_mb(); | |
c2a2ac2b XG |
668 | } |
669 | ||
670 | static void walk_shadow_page_lockless_end(struct kvm_vcpu *vcpu) | |
671 | { | |
c142786c AK |
672 | /* |
673 | * Make sure the write to vcpu->mode is not reordered in front of | |
674 | * reads to sptes. If it does, kvm_commit_zap_page() can see us | |
675 | * OUTSIDE_GUEST_MODE and proceed to free the shadow page table. | |
676 | */ | |
677 | smp_mb(); | |
678 | vcpu->mode = OUTSIDE_GUEST_MODE; | |
679 | local_irq_enable(); | |
c2a2ac2b XG |
680 | } |
681 | ||
e2dec939 | 682 | static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache, |
2e3e5882 | 683 | struct kmem_cache *base_cache, int min) |
714b93da AK |
684 | { |
685 | void *obj; | |
686 | ||
687 | if (cache->nobjs >= min) | |
e2dec939 | 688 | return 0; |
714b93da | 689 | while (cache->nobjs < ARRAY_SIZE(cache->objects)) { |
2e3e5882 | 690 | obj = kmem_cache_zalloc(base_cache, GFP_KERNEL); |
714b93da | 691 | if (!obj) |
e2dec939 | 692 | return -ENOMEM; |
714b93da AK |
693 | cache->objects[cache->nobjs++] = obj; |
694 | } | |
e2dec939 | 695 | return 0; |
714b93da AK |
696 | } |
697 | ||
f759e2b4 XG |
698 | static int mmu_memory_cache_free_objects(struct kvm_mmu_memory_cache *cache) |
699 | { | |
700 | return cache->nobjs; | |
701 | } | |
702 | ||
e8ad9a70 XG |
703 | static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc, |
704 | struct kmem_cache *cache) | |
714b93da AK |
705 | { |
706 | while (mc->nobjs) | |
e8ad9a70 | 707 | kmem_cache_free(cache, mc->objects[--mc->nobjs]); |
714b93da AK |
708 | } |
709 | ||
c1158e63 | 710 | static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache *cache, |
2e3e5882 | 711 | int min) |
c1158e63 | 712 | { |
842f22ed | 713 | void *page; |
c1158e63 AK |
714 | |
715 | if (cache->nobjs >= min) | |
716 | return 0; | |
717 | while (cache->nobjs < ARRAY_SIZE(cache->objects)) { | |
842f22ed | 718 | page = (void *)__get_free_page(GFP_KERNEL); |
c1158e63 AK |
719 | if (!page) |
720 | return -ENOMEM; | |
842f22ed | 721 | cache->objects[cache->nobjs++] = page; |
c1158e63 AK |
722 | } |
723 | return 0; | |
724 | } | |
725 | ||
726 | static void mmu_free_memory_cache_page(struct kvm_mmu_memory_cache *mc) | |
727 | { | |
728 | while (mc->nobjs) | |
c4d198d5 | 729 | free_page((unsigned long)mc->objects[--mc->nobjs]); |
c1158e63 AK |
730 | } |
731 | ||
2e3e5882 | 732 | static int mmu_topup_memory_caches(struct kvm_vcpu *vcpu) |
714b93da | 733 | { |
e2dec939 AK |
734 | int r; |
735 | ||
53c07b18 | 736 | r = mmu_topup_memory_cache(&vcpu->arch.mmu_pte_list_desc_cache, |
67052b35 | 737 | pte_list_desc_cache, 8 + PTE_PREFETCH_NUM); |
d3d25b04 AK |
738 | if (r) |
739 | goto out; | |
ad312c7c | 740 | r = mmu_topup_memory_cache_page(&vcpu->arch.mmu_page_cache, 8); |
d3d25b04 AK |
741 | if (r) |
742 | goto out; | |
ad312c7c | 743 | r = mmu_topup_memory_cache(&vcpu->arch.mmu_page_header_cache, |
2e3e5882 | 744 | mmu_page_header_cache, 4); |
e2dec939 AK |
745 | out: |
746 | return r; | |
714b93da AK |
747 | } |
748 | ||
749 | static void mmu_free_memory_caches(struct kvm_vcpu *vcpu) | |
750 | { | |
53c07b18 XG |
751 | mmu_free_memory_cache(&vcpu->arch.mmu_pte_list_desc_cache, |
752 | pte_list_desc_cache); | |
ad312c7c | 753 | mmu_free_memory_cache_page(&vcpu->arch.mmu_page_cache); |
e8ad9a70 XG |
754 | mmu_free_memory_cache(&vcpu->arch.mmu_page_header_cache, |
755 | mmu_page_header_cache); | |
714b93da AK |
756 | } |
757 | ||
80feb89a | 758 | static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc) |
714b93da AK |
759 | { |
760 | void *p; | |
761 | ||
762 | BUG_ON(!mc->nobjs); | |
763 | p = mc->objects[--mc->nobjs]; | |
714b93da AK |
764 | return p; |
765 | } | |
766 | ||
53c07b18 | 767 | static struct pte_list_desc *mmu_alloc_pte_list_desc(struct kvm_vcpu *vcpu) |
714b93da | 768 | { |
80feb89a | 769 | return mmu_memory_cache_alloc(&vcpu->arch.mmu_pte_list_desc_cache); |
714b93da AK |
770 | } |
771 | ||
53c07b18 | 772 | static void mmu_free_pte_list_desc(struct pte_list_desc *pte_list_desc) |
714b93da | 773 | { |
53c07b18 | 774 | kmem_cache_free(pte_list_desc_cache, pte_list_desc); |
714b93da AK |
775 | } |
776 | ||
2032a93d LJ |
777 | static gfn_t kvm_mmu_page_get_gfn(struct kvm_mmu_page *sp, int index) |
778 | { | |
779 | if (!sp->role.direct) | |
780 | return sp->gfns[index]; | |
781 | ||
782 | return sp->gfn + (index << ((sp->role.level - 1) * PT64_LEVEL_BITS)); | |
783 | } | |
784 | ||
785 | static void kvm_mmu_page_set_gfn(struct kvm_mmu_page *sp, int index, gfn_t gfn) | |
786 | { | |
787 | if (sp->role.direct) | |
788 | BUG_ON(gfn != kvm_mmu_page_get_gfn(sp, index)); | |
789 | else | |
790 | sp->gfns[index] = gfn; | |
791 | } | |
792 | ||
05da4558 | 793 | /* |
d4dbf470 TY |
794 | * Return the pointer to the large page information for a given gfn, |
795 | * handling slots that are not large page aligned. | |
05da4558 | 796 | */ |
d4dbf470 TY |
797 | static struct kvm_lpage_info *lpage_info_slot(gfn_t gfn, |
798 | struct kvm_memory_slot *slot, | |
799 | int level) | |
05da4558 MT |
800 | { |
801 | unsigned long idx; | |
802 | ||
fb03cb6f | 803 | idx = gfn_to_index(gfn, slot->base_gfn, level); |
db3fe4eb | 804 | return &slot->arch.lpage_info[level - 2][idx]; |
05da4558 MT |
805 | } |
806 | ||
3ed1a478 | 807 | static void account_shadowed(struct kvm *kvm, struct kvm_mmu_page *sp) |
05da4558 | 808 | { |
699023e2 | 809 | struct kvm_memslots *slots; |
d25797b2 | 810 | struct kvm_memory_slot *slot; |
d4dbf470 | 811 | struct kvm_lpage_info *linfo; |
3ed1a478 | 812 | gfn_t gfn; |
d25797b2 | 813 | int i; |
05da4558 | 814 | |
3ed1a478 | 815 | gfn = sp->gfn; |
699023e2 PB |
816 | slots = kvm_memslots_for_spte_role(kvm, sp->role); |
817 | slot = __gfn_to_memslot(slots, gfn); | |
8a3d08f1 | 818 | for (i = PT_DIRECTORY_LEVEL; i <= PT_MAX_HUGEPAGE_LEVEL; ++i) { |
d4dbf470 TY |
819 | linfo = lpage_info_slot(gfn, slot, i); |
820 | linfo->write_count += 1; | |
d25797b2 | 821 | } |
332b207d | 822 | kvm->arch.indirect_shadow_pages++; |
05da4558 MT |
823 | } |
824 | ||
3ed1a478 | 825 | static void unaccount_shadowed(struct kvm *kvm, struct kvm_mmu_page *sp) |
05da4558 | 826 | { |
699023e2 | 827 | struct kvm_memslots *slots; |
d25797b2 | 828 | struct kvm_memory_slot *slot; |
d4dbf470 | 829 | struct kvm_lpage_info *linfo; |
3ed1a478 | 830 | gfn_t gfn; |
d25797b2 | 831 | int i; |
05da4558 | 832 | |
3ed1a478 | 833 | gfn = sp->gfn; |
699023e2 PB |
834 | slots = kvm_memslots_for_spte_role(kvm, sp->role); |
835 | slot = __gfn_to_memslot(slots, gfn); | |
8a3d08f1 | 836 | for (i = PT_DIRECTORY_LEVEL; i <= PT_MAX_HUGEPAGE_LEVEL; ++i) { |
d4dbf470 TY |
837 | linfo = lpage_info_slot(gfn, slot, i); |
838 | linfo->write_count -= 1; | |
839 | WARN_ON(linfo->write_count < 0); | |
d25797b2 | 840 | } |
332b207d | 841 | kvm->arch.indirect_shadow_pages--; |
05da4558 MT |
842 | } |
843 | ||
54bf36aa | 844 | static int has_wrprotected_page(struct kvm_vcpu *vcpu, |
d25797b2 JR |
845 | gfn_t gfn, |
846 | int level) | |
05da4558 | 847 | { |
2843099f | 848 | struct kvm_memory_slot *slot; |
d4dbf470 | 849 | struct kvm_lpage_info *linfo; |
05da4558 | 850 | |
54bf36aa | 851 | slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); |
05da4558 | 852 | if (slot) { |
d4dbf470 TY |
853 | linfo = lpage_info_slot(gfn, slot, level); |
854 | return linfo->write_count; | |
05da4558 MT |
855 | } |
856 | ||
857 | return 1; | |
858 | } | |
859 | ||
d25797b2 | 860 | static int host_mapping_level(struct kvm *kvm, gfn_t gfn) |
05da4558 | 861 | { |
8f0b1ab6 | 862 | unsigned long page_size; |
d25797b2 | 863 | int i, ret = 0; |
05da4558 | 864 | |
8f0b1ab6 | 865 | page_size = kvm_host_page_size(kvm, gfn); |
05da4558 | 866 | |
8a3d08f1 | 867 | for (i = PT_PAGE_TABLE_LEVEL; i <= PT_MAX_HUGEPAGE_LEVEL; ++i) { |
d25797b2 JR |
868 | if (page_size >= KVM_HPAGE_SIZE(i)) |
869 | ret = i; | |
870 | else | |
871 | break; | |
872 | } | |
873 | ||
4c2155ce | 874 | return ret; |
05da4558 MT |
875 | } |
876 | ||
5d163b1c XG |
877 | static struct kvm_memory_slot * |
878 | gfn_to_memslot_dirty_bitmap(struct kvm_vcpu *vcpu, gfn_t gfn, | |
879 | bool no_dirty_log) | |
05da4558 MT |
880 | { |
881 | struct kvm_memory_slot *slot; | |
5d163b1c | 882 | |
54bf36aa | 883 | slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); |
5d163b1c XG |
884 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID || |
885 | (no_dirty_log && slot->dirty_bitmap)) | |
886 | slot = NULL; | |
887 | ||
888 | return slot; | |
889 | } | |
890 | ||
891 | static bool mapping_level_dirty_bitmap(struct kvm_vcpu *vcpu, gfn_t large_gfn) | |
892 | { | |
a0a8eaba | 893 | return !gfn_to_memslot_dirty_bitmap(vcpu, large_gfn, true); |
936a5fe6 AA |
894 | } |
895 | ||
896 | static int mapping_level(struct kvm_vcpu *vcpu, gfn_t large_gfn) | |
897 | { | |
898 | int host_level, level, max_level; | |
05da4558 | 899 | |
d25797b2 JR |
900 | host_level = host_mapping_level(vcpu->kvm, large_gfn); |
901 | ||
902 | if (host_level == PT_PAGE_TABLE_LEVEL) | |
903 | return host_level; | |
904 | ||
55dd98c3 | 905 | max_level = min(kvm_x86_ops->get_lpage_level(), host_level); |
878403b7 SY |
906 | |
907 | for (level = PT_DIRECTORY_LEVEL; level <= max_level; ++level) | |
54bf36aa | 908 | if (has_wrprotected_page(vcpu, large_gfn, level)) |
d25797b2 | 909 | break; |
d25797b2 JR |
910 | |
911 | return level - 1; | |
05da4558 MT |
912 | } |
913 | ||
290fc38d | 914 | /* |
53c07b18 | 915 | * Pte mapping structures: |
cd4a4e53 | 916 | * |
53c07b18 | 917 | * If pte_list bit zero is zero, then pte_list point to the spte. |
cd4a4e53 | 918 | * |
53c07b18 XG |
919 | * If pte_list bit zero is one, (then pte_list & ~1) points to a struct |
920 | * pte_list_desc containing more mappings. | |
53a27b39 | 921 | * |
53c07b18 | 922 | * Returns the number of pte entries before the spte was added or zero if |
53a27b39 MT |
923 | * the spte was not added. |
924 | * | |
cd4a4e53 | 925 | */ |
53c07b18 XG |
926 | static int pte_list_add(struct kvm_vcpu *vcpu, u64 *spte, |
927 | unsigned long *pte_list) | |
cd4a4e53 | 928 | { |
53c07b18 | 929 | struct pte_list_desc *desc; |
53a27b39 | 930 | int i, count = 0; |
cd4a4e53 | 931 | |
53c07b18 XG |
932 | if (!*pte_list) { |
933 | rmap_printk("pte_list_add: %p %llx 0->1\n", spte, *spte); | |
934 | *pte_list = (unsigned long)spte; | |
935 | } else if (!(*pte_list & 1)) { | |
936 | rmap_printk("pte_list_add: %p %llx 1->many\n", spte, *spte); | |
937 | desc = mmu_alloc_pte_list_desc(vcpu); | |
938 | desc->sptes[0] = (u64 *)*pte_list; | |
d555c333 | 939 | desc->sptes[1] = spte; |
53c07b18 | 940 | *pte_list = (unsigned long)desc | 1; |
cb16a7b3 | 941 | ++count; |
cd4a4e53 | 942 | } else { |
53c07b18 XG |
943 | rmap_printk("pte_list_add: %p %llx many->many\n", spte, *spte); |
944 | desc = (struct pte_list_desc *)(*pte_list & ~1ul); | |
945 | while (desc->sptes[PTE_LIST_EXT-1] && desc->more) { | |
cd4a4e53 | 946 | desc = desc->more; |
53c07b18 | 947 | count += PTE_LIST_EXT; |
53a27b39 | 948 | } |
53c07b18 XG |
949 | if (desc->sptes[PTE_LIST_EXT-1]) { |
950 | desc->more = mmu_alloc_pte_list_desc(vcpu); | |
cd4a4e53 AK |
951 | desc = desc->more; |
952 | } | |
d555c333 | 953 | for (i = 0; desc->sptes[i]; ++i) |
cb16a7b3 | 954 | ++count; |
d555c333 | 955 | desc->sptes[i] = spte; |
cd4a4e53 | 956 | } |
53a27b39 | 957 | return count; |
cd4a4e53 AK |
958 | } |
959 | ||
53c07b18 XG |
960 | static void |
961 | pte_list_desc_remove_entry(unsigned long *pte_list, struct pte_list_desc *desc, | |
962 | int i, struct pte_list_desc *prev_desc) | |
cd4a4e53 AK |
963 | { |
964 | int j; | |
965 | ||
53c07b18 | 966 | for (j = PTE_LIST_EXT - 1; !desc->sptes[j] && j > i; --j) |
cd4a4e53 | 967 | ; |
d555c333 AK |
968 | desc->sptes[i] = desc->sptes[j]; |
969 | desc->sptes[j] = NULL; | |
cd4a4e53 AK |
970 | if (j != 0) |
971 | return; | |
972 | if (!prev_desc && !desc->more) | |
53c07b18 | 973 | *pte_list = (unsigned long)desc->sptes[0]; |
cd4a4e53 AK |
974 | else |
975 | if (prev_desc) | |
976 | prev_desc->more = desc->more; | |
977 | else | |
53c07b18 XG |
978 | *pte_list = (unsigned long)desc->more | 1; |
979 | mmu_free_pte_list_desc(desc); | |
cd4a4e53 AK |
980 | } |
981 | ||
53c07b18 | 982 | static void pte_list_remove(u64 *spte, unsigned long *pte_list) |
cd4a4e53 | 983 | { |
53c07b18 XG |
984 | struct pte_list_desc *desc; |
985 | struct pte_list_desc *prev_desc; | |
cd4a4e53 AK |
986 | int i; |
987 | ||
53c07b18 XG |
988 | if (!*pte_list) { |
989 | printk(KERN_ERR "pte_list_remove: %p 0->BUG\n", spte); | |
cd4a4e53 | 990 | BUG(); |
53c07b18 XG |
991 | } else if (!(*pte_list & 1)) { |
992 | rmap_printk("pte_list_remove: %p 1->0\n", spte); | |
993 | if ((u64 *)*pte_list != spte) { | |
994 | printk(KERN_ERR "pte_list_remove: %p 1->BUG\n", spte); | |
cd4a4e53 AK |
995 | BUG(); |
996 | } | |
53c07b18 | 997 | *pte_list = 0; |
cd4a4e53 | 998 | } else { |
53c07b18 XG |
999 | rmap_printk("pte_list_remove: %p many->many\n", spte); |
1000 | desc = (struct pte_list_desc *)(*pte_list & ~1ul); | |
cd4a4e53 AK |
1001 | prev_desc = NULL; |
1002 | while (desc) { | |
53c07b18 | 1003 | for (i = 0; i < PTE_LIST_EXT && desc->sptes[i]; ++i) |
d555c333 | 1004 | if (desc->sptes[i] == spte) { |
53c07b18 | 1005 | pte_list_desc_remove_entry(pte_list, |
714b93da | 1006 | desc, i, |
cd4a4e53 AK |
1007 | prev_desc); |
1008 | return; | |
1009 | } | |
1010 | prev_desc = desc; | |
1011 | desc = desc->more; | |
1012 | } | |
53c07b18 | 1013 | pr_err("pte_list_remove: %p many->many\n", spte); |
cd4a4e53 AK |
1014 | BUG(); |
1015 | } | |
1016 | } | |
1017 | ||
67052b35 XG |
1018 | typedef void (*pte_list_walk_fn) (u64 *spte); |
1019 | static void pte_list_walk(unsigned long *pte_list, pte_list_walk_fn fn) | |
1020 | { | |
1021 | struct pte_list_desc *desc; | |
1022 | int i; | |
1023 | ||
1024 | if (!*pte_list) | |
1025 | return; | |
1026 | ||
1027 | if (!(*pte_list & 1)) | |
1028 | return fn((u64 *)*pte_list); | |
1029 | ||
1030 | desc = (struct pte_list_desc *)(*pte_list & ~1ul); | |
1031 | while (desc) { | |
1032 | for (i = 0; i < PTE_LIST_EXT && desc->sptes[i]; ++i) | |
1033 | fn(desc->sptes[i]); | |
1034 | desc = desc->more; | |
1035 | } | |
1036 | } | |
1037 | ||
9373e2c0 | 1038 | static unsigned long *__gfn_to_rmap(gfn_t gfn, int level, |
9b9b1492 | 1039 | struct kvm_memory_slot *slot) |
53c07b18 | 1040 | { |
77d11309 | 1041 | unsigned long idx; |
53c07b18 | 1042 | |
77d11309 | 1043 | idx = gfn_to_index(gfn, slot->base_gfn, level); |
d89cc617 | 1044 | return &slot->arch.rmap[level - PT_PAGE_TABLE_LEVEL][idx]; |
53c07b18 XG |
1045 | } |
1046 | ||
9b9b1492 TY |
1047 | /* |
1048 | * Take gfn and return the reverse mapping to it. | |
1049 | */ | |
e4cd1da9 | 1050 | static unsigned long *gfn_to_rmap(struct kvm *kvm, gfn_t gfn, struct kvm_mmu_page *sp) |
9b9b1492 | 1051 | { |
699023e2 | 1052 | struct kvm_memslots *slots; |
9b9b1492 TY |
1053 | struct kvm_memory_slot *slot; |
1054 | ||
699023e2 PB |
1055 | slots = kvm_memslots_for_spte_role(kvm, sp->role); |
1056 | slot = __gfn_to_memslot(slots, gfn); | |
e4cd1da9 | 1057 | return __gfn_to_rmap(gfn, sp->role.level, slot); |
9b9b1492 TY |
1058 | } |
1059 | ||
f759e2b4 XG |
1060 | static bool rmap_can_add(struct kvm_vcpu *vcpu) |
1061 | { | |
1062 | struct kvm_mmu_memory_cache *cache; | |
1063 | ||
1064 | cache = &vcpu->arch.mmu_pte_list_desc_cache; | |
1065 | return mmu_memory_cache_free_objects(cache); | |
1066 | } | |
1067 | ||
53c07b18 XG |
1068 | static int rmap_add(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn) |
1069 | { | |
1070 | struct kvm_mmu_page *sp; | |
1071 | unsigned long *rmapp; | |
1072 | ||
53c07b18 XG |
1073 | sp = page_header(__pa(spte)); |
1074 | kvm_mmu_page_set_gfn(sp, spte - sp->spt, gfn); | |
e4cd1da9 | 1075 | rmapp = gfn_to_rmap(vcpu->kvm, gfn, sp); |
53c07b18 XG |
1076 | return pte_list_add(vcpu, spte, rmapp); |
1077 | } | |
1078 | ||
53c07b18 XG |
1079 | static void rmap_remove(struct kvm *kvm, u64 *spte) |
1080 | { | |
1081 | struct kvm_mmu_page *sp; | |
1082 | gfn_t gfn; | |
1083 | unsigned long *rmapp; | |
1084 | ||
1085 | sp = page_header(__pa(spte)); | |
1086 | gfn = kvm_mmu_page_get_gfn(sp, spte - sp->spt); | |
e4cd1da9 | 1087 | rmapp = gfn_to_rmap(kvm, gfn, sp); |
53c07b18 XG |
1088 | pte_list_remove(spte, rmapp); |
1089 | } | |
1090 | ||
1e3f42f0 TY |
1091 | /* |
1092 | * Used by the following functions to iterate through the sptes linked by a | |
1093 | * rmap. All fields are private and not assumed to be used outside. | |
1094 | */ | |
1095 | struct rmap_iterator { | |
1096 | /* private fields */ | |
1097 | struct pte_list_desc *desc; /* holds the sptep if not NULL */ | |
1098 | int pos; /* index of the sptep */ | |
1099 | }; | |
1100 | ||
1101 | /* | |
1102 | * Iteration must be started by this function. This should also be used after | |
1103 | * removing/dropping sptes from the rmap link because in such cases the | |
1104 | * information in the itererator may not be valid. | |
1105 | * | |
1106 | * Returns sptep if found, NULL otherwise. | |
1107 | */ | |
1108 | static u64 *rmap_get_first(unsigned long rmap, struct rmap_iterator *iter) | |
1109 | { | |
1110 | if (!rmap) | |
1111 | return NULL; | |
1112 | ||
1113 | if (!(rmap & 1)) { | |
1114 | iter->desc = NULL; | |
1115 | return (u64 *)rmap; | |
1116 | } | |
1117 | ||
1118 | iter->desc = (struct pte_list_desc *)(rmap & ~1ul); | |
1119 | iter->pos = 0; | |
1120 | return iter->desc->sptes[iter->pos]; | |
1121 | } | |
1122 | ||
1123 | /* | |
1124 | * Must be used with a valid iterator: e.g. after rmap_get_first(). | |
1125 | * | |
1126 | * Returns sptep if found, NULL otherwise. | |
1127 | */ | |
1128 | static u64 *rmap_get_next(struct rmap_iterator *iter) | |
1129 | { | |
1130 | if (iter->desc) { | |
1131 | if (iter->pos < PTE_LIST_EXT - 1) { | |
1132 | u64 *sptep; | |
1133 | ||
1134 | ++iter->pos; | |
1135 | sptep = iter->desc->sptes[iter->pos]; | |
1136 | if (sptep) | |
1137 | return sptep; | |
1138 | } | |
1139 | ||
1140 | iter->desc = iter->desc->more; | |
1141 | ||
1142 | if (iter->desc) { | |
1143 | iter->pos = 0; | |
1144 | /* desc->sptes[0] cannot be NULL */ | |
1145 | return iter->desc->sptes[iter->pos]; | |
1146 | } | |
1147 | } | |
1148 | ||
1149 | return NULL; | |
1150 | } | |
1151 | ||
0d536790 XG |
1152 | #define for_each_rmap_spte(_rmap_, _iter_, _spte_) \ |
1153 | for (_spte_ = rmap_get_first(*_rmap_, _iter_); \ | |
1154 | _spte_ && ({BUG_ON(!is_shadow_present_pte(*_spte_)); 1;}); \ | |
1155 | _spte_ = rmap_get_next(_iter_)) | |
1156 | ||
c3707958 | 1157 | static void drop_spte(struct kvm *kvm, u64 *sptep) |
e4b502ea | 1158 | { |
1df9f2dc | 1159 | if (mmu_spte_clear_track_bits(sptep)) |
eb45fda4 | 1160 | rmap_remove(kvm, sptep); |
be38d276 AK |
1161 | } |
1162 | ||
8e22f955 XG |
1163 | |
1164 | static bool __drop_large_spte(struct kvm *kvm, u64 *sptep) | |
1165 | { | |
1166 | if (is_large_pte(*sptep)) { | |
1167 | WARN_ON(page_header(__pa(sptep))->role.level == | |
1168 | PT_PAGE_TABLE_LEVEL); | |
1169 | drop_spte(kvm, sptep); | |
1170 | --kvm->stat.lpages; | |
1171 | return true; | |
1172 | } | |
1173 | ||
1174 | return false; | |
1175 | } | |
1176 | ||
1177 | static void drop_large_spte(struct kvm_vcpu *vcpu, u64 *sptep) | |
1178 | { | |
1179 | if (__drop_large_spte(vcpu->kvm, sptep)) | |
1180 | kvm_flush_remote_tlbs(vcpu->kvm); | |
1181 | } | |
1182 | ||
1183 | /* | |
49fde340 | 1184 | * Write-protect on the specified @sptep, @pt_protect indicates whether |
c126d94f | 1185 | * spte write-protection is caused by protecting shadow page table. |
49fde340 | 1186 | * |
b4619660 | 1187 | * Note: write protection is difference between dirty logging and spte |
49fde340 XG |
1188 | * protection: |
1189 | * - for dirty logging, the spte can be set to writable at anytime if | |
1190 | * its dirty bitmap is properly set. | |
1191 | * - for spte protection, the spte can be writable only after unsync-ing | |
1192 | * shadow page. | |
8e22f955 | 1193 | * |
c126d94f | 1194 | * Return true if tlb need be flushed. |
8e22f955 | 1195 | */ |
c126d94f | 1196 | static bool spte_write_protect(struct kvm *kvm, u64 *sptep, bool pt_protect) |
d13bc5b5 XG |
1197 | { |
1198 | u64 spte = *sptep; | |
1199 | ||
49fde340 XG |
1200 | if (!is_writable_pte(spte) && |
1201 | !(pt_protect && spte_is_locklessly_modifiable(spte))) | |
d13bc5b5 XG |
1202 | return false; |
1203 | ||
1204 | rmap_printk("rmap_write_protect: spte %p %llx\n", sptep, *sptep); | |
1205 | ||
49fde340 XG |
1206 | if (pt_protect) |
1207 | spte &= ~SPTE_MMU_WRITEABLE; | |
d13bc5b5 | 1208 | spte = spte & ~PT_WRITABLE_MASK; |
49fde340 | 1209 | |
c126d94f | 1210 | return mmu_spte_update(sptep, spte); |
d13bc5b5 XG |
1211 | } |
1212 | ||
49fde340 | 1213 | static bool __rmap_write_protect(struct kvm *kvm, unsigned long *rmapp, |
245c3912 | 1214 | bool pt_protect) |
98348e95 | 1215 | { |
1e3f42f0 TY |
1216 | u64 *sptep; |
1217 | struct rmap_iterator iter; | |
d13bc5b5 | 1218 | bool flush = false; |
374cbac0 | 1219 | |
0d536790 | 1220 | for_each_rmap_spte(rmapp, &iter, sptep) |
c126d94f | 1221 | flush |= spte_write_protect(kvm, sptep, pt_protect); |
855149aa | 1222 | |
d13bc5b5 | 1223 | return flush; |
a0ed4607 TY |
1224 | } |
1225 | ||
f4b4b180 KH |
1226 | static bool spte_clear_dirty(struct kvm *kvm, u64 *sptep) |
1227 | { | |
1228 | u64 spte = *sptep; | |
1229 | ||
1230 | rmap_printk("rmap_clear_dirty: spte %p %llx\n", sptep, *sptep); | |
1231 | ||
1232 | spte &= ~shadow_dirty_mask; | |
1233 | ||
1234 | return mmu_spte_update(sptep, spte); | |
1235 | } | |
1236 | ||
1237 | static bool __rmap_clear_dirty(struct kvm *kvm, unsigned long *rmapp) | |
1238 | { | |
1239 | u64 *sptep; | |
1240 | struct rmap_iterator iter; | |
1241 | bool flush = false; | |
1242 | ||
0d536790 | 1243 | for_each_rmap_spte(rmapp, &iter, sptep) |
f4b4b180 | 1244 | flush |= spte_clear_dirty(kvm, sptep); |
f4b4b180 KH |
1245 | |
1246 | return flush; | |
1247 | } | |
1248 | ||
1249 | static bool spte_set_dirty(struct kvm *kvm, u64 *sptep) | |
1250 | { | |
1251 | u64 spte = *sptep; | |
1252 | ||
1253 | rmap_printk("rmap_set_dirty: spte %p %llx\n", sptep, *sptep); | |
1254 | ||
1255 | spte |= shadow_dirty_mask; | |
1256 | ||
1257 | return mmu_spte_update(sptep, spte); | |
1258 | } | |
1259 | ||
1260 | static bool __rmap_set_dirty(struct kvm *kvm, unsigned long *rmapp) | |
1261 | { | |
1262 | u64 *sptep; | |
1263 | struct rmap_iterator iter; | |
1264 | bool flush = false; | |
1265 | ||
0d536790 | 1266 | for_each_rmap_spte(rmapp, &iter, sptep) |
f4b4b180 | 1267 | flush |= spte_set_dirty(kvm, sptep); |
f4b4b180 KH |
1268 | |
1269 | return flush; | |
1270 | } | |
1271 | ||
5dc99b23 | 1272 | /** |
3b0f1d01 | 1273 | * kvm_mmu_write_protect_pt_masked - write protect selected PT level pages |
5dc99b23 TY |
1274 | * @kvm: kvm instance |
1275 | * @slot: slot to protect | |
1276 | * @gfn_offset: start of the BITS_PER_LONG pages we care about | |
1277 | * @mask: indicates which pages we should protect | |
1278 | * | |
1279 | * Used when we do not need to care about huge page mappings: e.g. during dirty | |
1280 | * logging we do not have any such mappings. | |
1281 | */ | |
3b0f1d01 | 1282 | static void kvm_mmu_write_protect_pt_masked(struct kvm *kvm, |
5dc99b23 TY |
1283 | struct kvm_memory_slot *slot, |
1284 | gfn_t gfn_offset, unsigned long mask) | |
a0ed4607 TY |
1285 | { |
1286 | unsigned long *rmapp; | |
a0ed4607 | 1287 | |
5dc99b23 | 1288 | while (mask) { |
65fbe37c TY |
1289 | rmapp = __gfn_to_rmap(slot->base_gfn + gfn_offset + __ffs(mask), |
1290 | PT_PAGE_TABLE_LEVEL, slot); | |
245c3912 | 1291 | __rmap_write_protect(kvm, rmapp, false); |
05da4558 | 1292 | |
5dc99b23 TY |
1293 | /* clear the first set bit */ |
1294 | mask &= mask - 1; | |
1295 | } | |
374cbac0 AK |
1296 | } |
1297 | ||
f4b4b180 KH |
1298 | /** |
1299 | * kvm_mmu_clear_dirty_pt_masked - clear MMU D-bit for PT level pages | |
1300 | * @kvm: kvm instance | |
1301 | * @slot: slot to clear D-bit | |
1302 | * @gfn_offset: start of the BITS_PER_LONG pages we care about | |
1303 | * @mask: indicates which pages we should clear D-bit | |
1304 | * | |
1305 | * Used for PML to re-log the dirty GPAs after userspace querying dirty_bitmap. | |
1306 | */ | |
1307 | void kvm_mmu_clear_dirty_pt_masked(struct kvm *kvm, | |
1308 | struct kvm_memory_slot *slot, | |
1309 | gfn_t gfn_offset, unsigned long mask) | |
1310 | { | |
1311 | unsigned long *rmapp; | |
1312 | ||
1313 | while (mask) { | |
1314 | rmapp = __gfn_to_rmap(slot->base_gfn + gfn_offset + __ffs(mask), | |
1315 | PT_PAGE_TABLE_LEVEL, slot); | |
1316 | __rmap_clear_dirty(kvm, rmapp); | |
1317 | ||
1318 | /* clear the first set bit */ | |
1319 | mask &= mask - 1; | |
1320 | } | |
1321 | } | |
1322 | EXPORT_SYMBOL_GPL(kvm_mmu_clear_dirty_pt_masked); | |
1323 | ||
3b0f1d01 KH |
1324 | /** |
1325 | * kvm_arch_mmu_enable_log_dirty_pt_masked - enable dirty logging for selected | |
1326 | * PT level pages. | |
1327 | * | |
1328 | * It calls kvm_mmu_write_protect_pt_masked to write protect selected pages to | |
1329 | * enable dirty logging for them. | |
1330 | * | |
1331 | * Used when we do not need to care about huge page mappings: e.g. during dirty | |
1332 | * logging we do not have any such mappings. | |
1333 | */ | |
1334 | void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm, | |
1335 | struct kvm_memory_slot *slot, | |
1336 | gfn_t gfn_offset, unsigned long mask) | |
1337 | { | |
88178fd4 KH |
1338 | if (kvm_x86_ops->enable_log_dirty_pt_masked) |
1339 | kvm_x86_ops->enable_log_dirty_pt_masked(kvm, slot, gfn_offset, | |
1340 | mask); | |
1341 | else | |
1342 | kvm_mmu_write_protect_pt_masked(kvm, slot, gfn_offset, mask); | |
3b0f1d01 KH |
1343 | } |
1344 | ||
54bf36aa | 1345 | static bool rmap_write_protect(struct kvm_vcpu *vcpu, u64 gfn) |
95d4c16c TY |
1346 | { |
1347 | struct kvm_memory_slot *slot; | |
5dc99b23 TY |
1348 | unsigned long *rmapp; |
1349 | int i; | |
2f84569f | 1350 | bool write_protected = false; |
95d4c16c | 1351 | |
54bf36aa | 1352 | slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); |
5dc99b23 | 1353 | |
8a3d08f1 | 1354 | for (i = PT_PAGE_TABLE_LEVEL; i <= PT_MAX_HUGEPAGE_LEVEL; ++i) { |
5dc99b23 | 1355 | rmapp = __gfn_to_rmap(gfn, i, slot); |
54bf36aa | 1356 | write_protected |= __rmap_write_protect(vcpu->kvm, rmapp, true); |
5dc99b23 TY |
1357 | } |
1358 | ||
1359 | return write_protected; | |
95d4c16c TY |
1360 | } |
1361 | ||
6a49f85c | 1362 | static bool kvm_zap_rmapp(struct kvm *kvm, unsigned long *rmapp) |
e930bffe | 1363 | { |
1e3f42f0 TY |
1364 | u64 *sptep; |
1365 | struct rmap_iterator iter; | |
6a49f85c | 1366 | bool flush = false; |
e930bffe | 1367 | |
1e3f42f0 TY |
1368 | while ((sptep = rmap_get_first(*rmapp, &iter))) { |
1369 | BUG_ON(!(*sptep & PT_PRESENT_MASK)); | |
6a49f85c | 1370 | rmap_printk("%s: spte %p %llx.\n", __func__, sptep, *sptep); |
1e3f42f0 TY |
1371 | |
1372 | drop_spte(kvm, sptep); | |
6a49f85c | 1373 | flush = true; |
e930bffe | 1374 | } |
1e3f42f0 | 1375 | |
6a49f85c XG |
1376 | return flush; |
1377 | } | |
1378 | ||
1379 | static int kvm_unmap_rmapp(struct kvm *kvm, unsigned long *rmapp, | |
1380 | struct kvm_memory_slot *slot, gfn_t gfn, int level, | |
1381 | unsigned long data) | |
1382 | { | |
1383 | return kvm_zap_rmapp(kvm, rmapp); | |
e930bffe AA |
1384 | } |
1385 | ||
8a8365c5 | 1386 | static int kvm_set_pte_rmapp(struct kvm *kvm, unsigned long *rmapp, |
8a9522d2 ALC |
1387 | struct kvm_memory_slot *slot, gfn_t gfn, int level, |
1388 | unsigned long data) | |
3da0dd43 | 1389 | { |
1e3f42f0 TY |
1390 | u64 *sptep; |
1391 | struct rmap_iterator iter; | |
3da0dd43 | 1392 | int need_flush = 0; |
1e3f42f0 | 1393 | u64 new_spte; |
3da0dd43 IE |
1394 | pte_t *ptep = (pte_t *)data; |
1395 | pfn_t new_pfn; | |
1396 | ||
1397 | WARN_ON(pte_huge(*ptep)); | |
1398 | new_pfn = pte_pfn(*ptep); | |
1e3f42f0 | 1399 | |
0d536790 XG |
1400 | restart: |
1401 | for_each_rmap_spte(rmapp, &iter, sptep) { | |
8a9522d2 ALC |
1402 | rmap_printk("kvm_set_pte_rmapp: spte %p %llx gfn %llx (%d)\n", |
1403 | sptep, *sptep, gfn, level); | |
1e3f42f0 | 1404 | |
3da0dd43 | 1405 | need_flush = 1; |
1e3f42f0 | 1406 | |
3da0dd43 | 1407 | if (pte_write(*ptep)) { |
1e3f42f0 | 1408 | drop_spte(kvm, sptep); |
0d536790 | 1409 | goto restart; |
3da0dd43 | 1410 | } else { |
1e3f42f0 | 1411 | new_spte = *sptep & ~PT64_BASE_ADDR_MASK; |
3da0dd43 IE |
1412 | new_spte |= (u64)new_pfn << PAGE_SHIFT; |
1413 | ||
1414 | new_spte &= ~PT_WRITABLE_MASK; | |
1415 | new_spte &= ~SPTE_HOST_WRITEABLE; | |
b79b93f9 | 1416 | new_spte &= ~shadow_accessed_mask; |
1e3f42f0 TY |
1417 | |
1418 | mmu_spte_clear_track_bits(sptep); | |
1419 | mmu_spte_set(sptep, new_spte); | |
3da0dd43 IE |
1420 | } |
1421 | } | |
1e3f42f0 | 1422 | |
3da0dd43 IE |
1423 | if (need_flush) |
1424 | kvm_flush_remote_tlbs(kvm); | |
1425 | ||
1426 | return 0; | |
1427 | } | |
1428 | ||
6ce1f4e2 XG |
1429 | struct slot_rmap_walk_iterator { |
1430 | /* input fields. */ | |
1431 | struct kvm_memory_slot *slot; | |
1432 | gfn_t start_gfn; | |
1433 | gfn_t end_gfn; | |
1434 | int start_level; | |
1435 | int end_level; | |
1436 | ||
1437 | /* output fields. */ | |
1438 | gfn_t gfn; | |
1439 | unsigned long *rmap; | |
1440 | int level; | |
1441 | ||
1442 | /* private field. */ | |
1443 | unsigned long *end_rmap; | |
1444 | }; | |
1445 | ||
1446 | static void | |
1447 | rmap_walk_init_level(struct slot_rmap_walk_iterator *iterator, int level) | |
1448 | { | |
1449 | iterator->level = level; | |
1450 | iterator->gfn = iterator->start_gfn; | |
1451 | iterator->rmap = __gfn_to_rmap(iterator->gfn, level, iterator->slot); | |
1452 | iterator->end_rmap = __gfn_to_rmap(iterator->end_gfn, level, | |
1453 | iterator->slot); | |
1454 | } | |
1455 | ||
1456 | static void | |
1457 | slot_rmap_walk_init(struct slot_rmap_walk_iterator *iterator, | |
1458 | struct kvm_memory_slot *slot, int start_level, | |
1459 | int end_level, gfn_t start_gfn, gfn_t end_gfn) | |
1460 | { | |
1461 | iterator->slot = slot; | |
1462 | iterator->start_level = start_level; | |
1463 | iterator->end_level = end_level; | |
1464 | iterator->start_gfn = start_gfn; | |
1465 | iterator->end_gfn = end_gfn; | |
1466 | ||
1467 | rmap_walk_init_level(iterator, iterator->start_level); | |
1468 | } | |
1469 | ||
1470 | static bool slot_rmap_walk_okay(struct slot_rmap_walk_iterator *iterator) | |
1471 | { | |
1472 | return !!iterator->rmap; | |
1473 | } | |
1474 | ||
1475 | static void slot_rmap_walk_next(struct slot_rmap_walk_iterator *iterator) | |
1476 | { | |
1477 | if (++iterator->rmap <= iterator->end_rmap) { | |
1478 | iterator->gfn += (1UL << KVM_HPAGE_GFN_SHIFT(iterator->level)); | |
1479 | return; | |
1480 | } | |
1481 | ||
1482 | if (++iterator->level > iterator->end_level) { | |
1483 | iterator->rmap = NULL; | |
1484 | return; | |
1485 | } | |
1486 | ||
1487 | rmap_walk_init_level(iterator, iterator->level); | |
1488 | } | |
1489 | ||
1490 | #define for_each_slot_rmap_range(_slot_, _start_level_, _end_level_, \ | |
1491 | _start_gfn, _end_gfn, _iter_) \ | |
1492 | for (slot_rmap_walk_init(_iter_, _slot_, _start_level_, \ | |
1493 | _end_level_, _start_gfn, _end_gfn); \ | |
1494 | slot_rmap_walk_okay(_iter_); \ | |
1495 | slot_rmap_walk_next(_iter_)) | |
1496 | ||
84504ef3 TY |
1497 | static int kvm_handle_hva_range(struct kvm *kvm, |
1498 | unsigned long start, | |
1499 | unsigned long end, | |
1500 | unsigned long data, | |
1501 | int (*handler)(struct kvm *kvm, | |
1502 | unsigned long *rmapp, | |
048212d0 | 1503 | struct kvm_memory_slot *slot, |
8a9522d2 ALC |
1504 | gfn_t gfn, |
1505 | int level, | |
84504ef3 | 1506 | unsigned long data)) |
e930bffe | 1507 | { |
bc6678a3 | 1508 | struct kvm_memslots *slots; |
be6ba0f0 | 1509 | struct kvm_memory_slot *memslot; |
6ce1f4e2 XG |
1510 | struct slot_rmap_walk_iterator iterator; |
1511 | int ret = 0; | |
9da0e4d5 | 1512 | int i; |
bc6678a3 | 1513 | |
9da0e4d5 PB |
1514 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
1515 | slots = __kvm_memslots(kvm, i); | |
1516 | kvm_for_each_memslot(memslot, slots) { | |
1517 | unsigned long hva_start, hva_end; | |
1518 | gfn_t gfn_start, gfn_end; | |
e930bffe | 1519 | |
9da0e4d5 PB |
1520 | hva_start = max(start, memslot->userspace_addr); |
1521 | hva_end = min(end, memslot->userspace_addr + | |
1522 | (memslot->npages << PAGE_SHIFT)); | |
1523 | if (hva_start >= hva_end) | |
1524 | continue; | |
1525 | /* | |
1526 | * {gfn(page) | page intersects with [hva_start, hva_end)} = | |
1527 | * {gfn_start, gfn_start+1, ..., gfn_end-1}. | |
1528 | */ | |
1529 | gfn_start = hva_to_gfn_memslot(hva_start, memslot); | |
1530 | gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot); | |
1531 | ||
1532 | for_each_slot_rmap_range(memslot, PT_PAGE_TABLE_LEVEL, | |
1533 | PT_MAX_HUGEPAGE_LEVEL, | |
1534 | gfn_start, gfn_end - 1, | |
1535 | &iterator) | |
1536 | ret |= handler(kvm, iterator.rmap, memslot, | |
1537 | iterator.gfn, iterator.level, data); | |
1538 | } | |
e930bffe AA |
1539 | } |
1540 | ||
f395302e | 1541 | return ret; |
e930bffe AA |
1542 | } |
1543 | ||
84504ef3 TY |
1544 | static int kvm_handle_hva(struct kvm *kvm, unsigned long hva, |
1545 | unsigned long data, | |
1546 | int (*handler)(struct kvm *kvm, unsigned long *rmapp, | |
048212d0 | 1547 | struct kvm_memory_slot *slot, |
8a9522d2 | 1548 | gfn_t gfn, int level, |
84504ef3 TY |
1549 | unsigned long data)) |
1550 | { | |
1551 | return kvm_handle_hva_range(kvm, hva, hva + 1, data, handler); | |
e930bffe AA |
1552 | } |
1553 | ||
1554 | int kvm_unmap_hva(struct kvm *kvm, unsigned long hva) | |
1555 | { | |
3da0dd43 IE |
1556 | return kvm_handle_hva(kvm, hva, 0, kvm_unmap_rmapp); |
1557 | } | |
1558 | ||
b3ae2096 TY |
1559 | int kvm_unmap_hva_range(struct kvm *kvm, unsigned long start, unsigned long end) |
1560 | { | |
1561 | return kvm_handle_hva_range(kvm, start, end, 0, kvm_unmap_rmapp); | |
1562 | } | |
1563 | ||
3da0dd43 IE |
1564 | void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte) |
1565 | { | |
8a8365c5 | 1566 | kvm_handle_hva(kvm, hva, (unsigned long)&pte, kvm_set_pte_rmapp); |
e930bffe AA |
1567 | } |
1568 | ||
8a8365c5 | 1569 | static int kvm_age_rmapp(struct kvm *kvm, unsigned long *rmapp, |
8a9522d2 ALC |
1570 | struct kvm_memory_slot *slot, gfn_t gfn, int level, |
1571 | unsigned long data) | |
e930bffe | 1572 | { |
1e3f42f0 | 1573 | u64 *sptep; |
79f702a6 | 1574 | struct rmap_iterator uninitialized_var(iter); |
e930bffe AA |
1575 | int young = 0; |
1576 | ||
57128468 | 1577 | BUG_ON(!shadow_accessed_mask); |
534e38b4 | 1578 | |
0d536790 | 1579 | for_each_rmap_spte(rmapp, &iter, sptep) |
3f6d8c8a | 1580 | if (*sptep & shadow_accessed_mask) { |
e930bffe | 1581 | young = 1; |
3f6d8c8a XH |
1582 | clear_bit((ffs(shadow_accessed_mask) - 1), |
1583 | (unsigned long *)sptep); | |
e930bffe | 1584 | } |
0d536790 | 1585 | |
8a9522d2 | 1586 | trace_kvm_age_page(gfn, level, slot, young); |
e930bffe AA |
1587 | return young; |
1588 | } | |
1589 | ||
8ee53820 | 1590 | static int kvm_test_age_rmapp(struct kvm *kvm, unsigned long *rmapp, |
8a9522d2 ALC |
1591 | struct kvm_memory_slot *slot, gfn_t gfn, |
1592 | int level, unsigned long data) | |
8ee53820 | 1593 | { |
1e3f42f0 TY |
1594 | u64 *sptep; |
1595 | struct rmap_iterator iter; | |
8ee53820 AA |
1596 | int young = 0; |
1597 | ||
1598 | /* | |
1599 | * If there's no access bit in the secondary pte set by the | |
1600 | * hardware it's up to gup-fast/gup to set the access bit in | |
1601 | * the primary pte or in the page structure. | |
1602 | */ | |
1603 | if (!shadow_accessed_mask) | |
1604 | goto out; | |
1605 | ||
0d536790 | 1606 | for_each_rmap_spte(rmapp, &iter, sptep) |
3f6d8c8a | 1607 | if (*sptep & shadow_accessed_mask) { |
8ee53820 AA |
1608 | young = 1; |
1609 | break; | |
1610 | } | |
8ee53820 AA |
1611 | out: |
1612 | return young; | |
1613 | } | |
1614 | ||
53a27b39 MT |
1615 | #define RMAP_RECYCLE_THRESHOLD 1000 |
1616 | ||
852e3c19 | 1617 | static void rmap_recycle(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn) |
53a27b39 MT |
1618 | { |
1619 | unsigned long *rmapp; | |
852e3c19 JR |
1620 | struct kvm_mmu_page *sp; |
1621 | ||
1622 | sp = page_header(__pa(spte)); | |
53a27b39 | 1623 | |
e4cd1da9 | 1624 | rmapp = gfn_to_rmap(vcpu->kvm, gfn, sp); |
53a27b39 | 1625 | |
8a9522d2 | 1626 | kvm_unmap_rmapp(vcpu->kvm, rmapp, NULL, gfn, sp->role.level, 0); |
53a27b39 MT |
1627 | kvm_flush_remote_tlbs(vcpu->kvm); |
1628 | } | |
1629 | ||
57128468 | 1630 | int kvm_age_hva(struct kvm *kvm, unsigned long start, unsigned long end) |
e930bffe | 1631 | { |
57128468 ALC |
1632 | /* |
1633 | * In case of absence of EPT Access and Dirty Bits supports, | |
1634 | * emulate the accessed bit for EPT, by checking if this page has | |
1635 | * an EPT mapping, and clearing it if it does. On the next access, | |
1636 | * a new EPT mapping will be established. | |
1637 | * This has some overhead, but not as much as the cost of swapping | |
1638 | * out actively used pages or breaking up actively used hugepages. | |
1639 | */ | |
1640 | if (!shadow_accessed_mask) { | |
1641 | /* | |
1642 | * We are holding the kvm->mmu_lock, and we are blowing up | |
1643 | * shadow PTEs. MMU notifier consumers need to be kept at bay. | |
1644 | * This is correct as long as we don't decouple the mmu_lock | |
1645 | * protected regions (like invalidate_range_start|end does). | |
1646 | */ | |
1647 | kvm->mmu_notifier_seq++; | |
1648 | return kvm_handle_hva_range(kvm, start, end, 0, | |
1649 | kvm_unmap_rmapp); | |
1650 | } | |
1651 | ||
1652 | return kvm_handle_hva_range(kvm, start, end, 0, kvm_age_rmapp); | |
e930bffe AA |
1653 | } |
1654 | ||
8ee53820 AA |
1655 | int kvm_test_age_hva(struct kvm *kvm, unsigned long hva) |
1656 | { | |
1657 | return kvm_handle_hva(kvm, hva, 0, kvm_test_age_rmapp); | |
1658 | } | |
1659 | ||
d6c69ee9 | 1660 | #ifdef MMU_DEBUG |
47ad8e68 | 1661 | static int is_empty_shadow_page(u64 *spt) |
6aa8b732 | 1662 | { |
139bdb2d AK |
1663 | u64 *pos; |
1664 | u64 *end; | |
1665 | ||
47ad8e68 | 1666 | for (pos = spt, end = pos + PAGE_SIZE / sizeof(u64); pos != end; pos++) |
3c915510 | 1667 | if (is_shadow_present_pte(*pos)) { |
b8688d51 | 1668 | printk(KERN_ERR "%s: %p %llx\n", __func__, |
139bdb2d | 1669 | pos, *pos); |
6aa8b732 | 1670 | return 0; |
139bdb2d | 1671 | } |
6aa8b732 AK |
1672 | return 1; |
1673 | } | |
d6c69ee9 | 1674 | #endif |
6aa8b732 | 1675 | |
45221ab6 DH |
1676 | /* |
1677 | * This value is the sum of all of the kvm instances's | |
1678 | * kvm->arch.n_used_mmu_pages values. We need a global, | |
1679 | * aggregate version in order to make the slab shrinker | |
1680 | * faster | |
1681 | */ | |
1682 | static inline void kvm_mod_used_mmu_pages(struct kvm *kvm, int nr) | |
1683 | { | |
1684 | kvm->arch.n_used_mmu_pages += nr; | |
1685 | percpu_counter_add(&kvm_total_used_mmu_pages, nr); | |
1686 | } | |
1687 | ||
834be0d8 | 1688 | static void kvm_mmu_free_page(struct kvm_mmu_page *sp) |
260746c0 | 1689 | { |
fa4a2c08 | 1690 | MMU_WARN_ON(!is_empty_shadow_page(sp->spt)); |
7775834a | 1691 | hlist_del(&sp->hash_link); |
bd4c86ea XG |
1692 | list_del(&sp->link); |
1693 | free_page((unsigned long)sp->spt); | |
834be0d8 GN |
1694 | if (!sp->role.direct) |
1695 | free_page((unsigned long)sp->gfns); | |
e8ad9a70 | 1696 | kmem_cache_free(mmu_page_header_cache, sp); |
260746c0 AK |
1697 | } |
1698 | ||
cea0f0e7 AK |
1699 | static unsigned kvm_page_table_hashfn(gfn_t gfn) |
1700 | { | |
1ae0a13d | 1701 | return gfn & ((1 << KVM_MMU_HASH_SHIFT) - 1); |
cea0f0e7 AK |
1702 | } |
1703 | ||
714b93da | 1704 | static void mmu_page_add_parent_pte(struct kvm_vcpu *vcpu, |
4db35314 | 1705 | struct kvm_mmu_page *sp, u64 *parent_pte) |
cea0f0e7 | 1706 | { |
cea0f0e7 AK |
1707 | if (!parent_pte) |
1708 | return; | |
cea0f0e7 | 1709 | |
67052b35 | 1710 | pte_list_add(vcpu, parent_pte, &sp->parent_ptes); |
cea0f0e7 AK |
1711 | } |
1712 | ||
4db35314 | 1713 | static void mmu_page_remove_parent_pte(struct kvm_mmu_page *sp, |
cea0f0e7 AK |
1714 | u64 *parent_pte) |
1715 | { | |
67052b35 | 1716 | pte_list_remove(parent_pte, &sp->parent_ptes); |
cea0f0e7 AK |
1717 | } |
1718 | ||
bcdd9a93 XG |
1719 | static void drop_parent_pte(struct kvm_mmu_page *sp, |
1720 | u64 *parent_pte) | |
1721 | { | |
1722 | mmu_page_remove_parent_pte(sp, parent_pte); | |
1df9f2dc | 1723 | mmu_spte_clear_no_track(parent_pte); |
bcdd9a93 XG |
1724 | } |
1725 | ||
67052b35 XG |
1726 | static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu, |
1727 | u64 *parent_pte, int direct) | |
ad8cfbe3 | 1728 | { |
67052b35 | 1729 | struct kvm_mmu_page *sp; |
7ddca7e4 | 1730 | |
80feb89a TY |
1731 | sp = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_header_cache); |
1732 | sp->spt = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_cache); | |
67052b35 | 1733 | if (!direct) |
80feb89a | 1734 | sp->gfns = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_cache); |
67052b35 | 1735 | set_page_private(virt_to_page(sp->spt), (unsigned long)sp); |
5304b8d3 XG |
1736 | |
1737 | /* | |
1738 | * The active_mmu_pages list is the FIFO list, do not move the | |
1739 | * page until it is zapped. kvm_zap_obsolete_pages depends on | |
1740 | * this feature. See the comments in kvm_zap_obsolete_pages(). | |
1741 | */ | |
67052b35 | 1742 | list_add(&sp->link, &vcpu->kvm->arch.active_mmu_pages); |
67052b35 XG |
1743 | sp->parent_ptes = 0; |
1744 | mmu_page_add_parent_pte(vcpu, sp, parent_pte); | |
1745 | kvm_mod_used_mmu_pages(vcpu->kvm, +1); | |
1746 | return sp; | |
ad8cfbe3 MT |
1747 | } |
1748 | ||
67052b35 | 1749 | static void mark_unsync(u64 *spte); |
1047df1f | 1750 | static void kvm_mmu_mark_parents_unsync(struct kvm_mmu_page *sp) |
0074ff63 | 1751 | { |
67052b35 | 1752 | pte_list_walk(&sp->parent_ptes, mark_unsync); |
0074ff63 MT |
1753 | } |
1754 | ||
67052b35 | 1755 | static void mark_unsync(u64 *spte) |
0074ff63 | 1756 | { |
67052b35 | 1757 | struct kvm_mmu_page *sp; |
1047df1f | 1758 | unsigned int index; |
0074ff63 | 1759 | |
67052b35 | 1760 | sp = page_header(__pa(spte)); |
1047df1f XG |
1761 | index = spte - sp->spt; |
1762 | if (__test_and_set_bit(index, sp->unsync_child_bitmap)) | |
0074ff63 | 1763 | return; |
1047df1f | 1764 | if (sp->unsync_children++) |
0074ff63 | 1765 | return; |
1047df1f | 1766 | kvm_mmu_mark_parents_unsync(sp); |
0074ff63 MT |
1767 | } |
1768 | ||
e8bc217a | 1769 | static int nonpaging_sync_page(struct kvm_vcpu *vcpu, |
a4a8e6f7 | 1770 | struct kvm_mmu_page *sp) |
e8bc217a MT |
1771 | { |
1772 | return 1; | |
1773 | } | |
1774 | ||
a7052897 MT |
1775 | static void nonpaging_invlpg(struct kvm_vcpu *vcpu, gva_t gva) |
1776 | { | |
1777 | } | |
1778 | ||
0f53b5b1 XG |
1779 | static void nonpaging_update_pte(struct kvm_vcpu *vcpu, |
1780 | struct kvm_mmu_page *sp, u64 *spte, | |
7c562522 | 1781 | const void *pte) |
0f53b5b1 XG |
1782 | { |
1783 | WARN_ON(1); | |
1784 | } | |
1785 | ||
60c8aec6 MT |
1786 | #define KVM_PAGE_ARRAY_NR 16 |
1787 | ||
1788 | struct kvm_mmu_pages { | |
1789 | struct mmu_page_and_offset { | |
1790 | struct kvm_mmu_page *sp; | |
1791 | unsigned int idx; | |
1792 | } page[KVM_PAGE_ARRAY_NR]; | |
1793 | unsigned int nr; | |
1794 | }; | |
1795 | ||
cded19f3 HE |
1796 | static int mmu_pages_add(struct kvm_mmu_pages *pvec, struct kvm_mmu_page *sp, |
1797 | int idx) | |
4731d4c7 | 1798 | { |
60c8aec6 | 1799 | int i; |
4731d4c7 | 1800 | |
60c8aec6 MT |
1801 | if (sp->unsync) |
1802 | for (i=0; i < pvec->nr; i++) | |
1803 | if (pvec->page[i].sp == sp) | |
1804 | return 0; | |
1805 | ||
1806 | pvec->page[pvec->nr].sp = sp; | |
1807 | pvec->page[pvec->nr].idx = idx; | |
1808 | pvec->nr++; | |
1809 | return (pvec->nr == KVM_PAGE_ARRAY_NR); | |
1810 | } | |
1811 | ||
1812 | static int __mmu_unsync_walk(struct kvm_mmu_page *sp, | |
1813 | struct kvm_mmu_pages *pvec) | |
1814 | { | |
1815 | int i, ret, nr_unsync_leaf = 0; | |
4731d4c7 | 1816 | |
37178b8b | 1817 | for_each_set_bit(i, sp->unsync_child_bitmap, 512) { |
7a8f1a74 | 1818 | struct kvm_mmu_page *child; |
4731d4c7 MT |
1819 | u64 ent = sp->spt[i]; |
1820 | ||
7a8f1a74 XG |
1821 | if (!is_shadow_present_pte(ent) || is_large_pte(ent)) |
1822 | goto clear_child_bitmap; | |
1823 | ||
1824 | child = page_header(ent & PT64_BASE_ADDR_MASK); | |
1825 | ||
1826 | if (child->unsync_children) { | |
1827 | if (mmu_pages_add(pvec, child, i)) | |
1828 | return -ENOSPC; | |
1829 | ||
1830 | ret = __mmu_unsync_walk(child, pvec); | |
1831 | if (!ret) | |
1832 | goto clear_child_bitmap; | |
1833 | else if (ret > 0) | |
1834 | nr_unsync_leaf += ret; | |
1835 | else | |
1836 | return ret; | |
1837 | } else if (child->unsync) { | |
1838 | nr_unsync_leaf++; | |
1839 | if (mmu_pages_add(pvec, child, i)) | |
1840 | return -ENOSPC; | |
1841 | } else | |
1842 | goto clear_child_bitmap; | |
1843 | ||
1844 | continue; | |
1845 | ||
1846 | clear_child_bitmap: | |
1847 | __clear_bit(i, sp->unsync_child_bitmap); | |
1848 | sp->unsync_children--; | |
1849 | WARN_ON((int)sp->unsync_children < 0); | |
4731d4c7 MT |
1850 | } |
1851 | ||
4731d4c7 | 1852 | |
60c8aec6 MT |
1853 | return nr_unsync_leaf; |
1854 | } | |
1855 | ||
1856 | static int mmu_unsync_walk(struct kvm_mmu_page *sp, | |
1857 | struct kvm_mmu_pages *pvec) | |
1858 | { | |
1859 | if (!sp->unsync_children) | |
1860 | return 0; | |
1861 | ||
1862 | mmu_pages_add(pvec, sp, 0); | |
1863 | return __mmu_unsync_walk(sp, pvec); | |
4731d4c7 MT |
1864 | } |
1865 | ||
4731d4c7 MT |
1866 | static void kvm_unlink_unsync_page(struct kvm *kvm, struct kvm_mmu_page *sp) |
1867 | { | |
1868 | WARN_ON(!sp->unsync); | |
5e1b3ddb | 1869 | trace_kvm_mmu_sync_page(sp); |
4731d4c7 MT |
1870 | sp->unsync = 0; |
1871 | --kvm->stat.mmu_unsync; | |
1872 | } | |
1873 | ||
7775834a XG |
1874 | static int kvm_mmu_prepare_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp, |
1875 | struct list_head *invalid_list); | |
1876 | static void kvm_mmu_commit_zap_page(struct kvm *kvm, | |
1877 | struct list_head *invalid_list); | |
4731d4c7 | 1878 | |
f34d251d XG |
1879 | /* |
1880 | * NOTE: we should pay more attention on the zapped-obsolete page | |
1881 | * (is_obsolete_sp(sp) && sp->role.invalid) when you do hash list walk | |
1882 | * since it has been deleted from active_mmu_pages but still can be found | |
1883 | * at hast list. | |
1884 | * | |
1885 | * for_each_gfn_indirect_valid_sp has skipped that kind of page and | |
1886 | * kvm_mmu_get_page(), the only user of for_each_gfn_sp(), has skipped | |
1887 | * all the obsolete pages. | |
1888 | */ | |
1044b030 TY |
1889 | #define for_each_gfn_sp(_kvm, _sp, _gfn) \ |
1890 | hlist_for_each_entry(_sp, \ | |
1891 | &(_kvm)->arch.mmu_page_hash[kvm_page_table_hashfn(_gfn)], hash_link) \ | |
1892 | if ((_sp)->gfn != (_gfn)) {} else | |
1893 | ||
1894 | #define for_each_gfn_indirect_valid_sp(_kvm, _sp, _gfn) \ | |
1895 | for_each_gfn_sp(_kvm, _sp, _gfn) \ | |
1896 | if ((_sp)->role.direct || (_sp)->role.invalid) {} else | |
7ae680eb | 1897 | |
f918b443 | 1898 | /* @sp->gfn should be write-protected at the call site */ |
1d9dc7e0 | 1899 | static int __kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp, |
d98ba053 | 1900 | struct list_head *invalid_list, bool clear_unsync) |
4731d4c7 | 1901 | { |
5b7e0102 | 1902 | if (sp->role.cr4_pae != !!is_pae(vcpu)) { |
d98ba053 | 1903 | kvm_mmu_prepare_zap_page(vcpu->kvm, sp, invalid_list); |
4731d4c7 MT |
1904 | return 1; |
1905 | } | |
1906 | ||
f918b443 | 1907 | if (clear_unsync) |
1d9dc7e0 | 1908 | kvm_unlink_unsync_page(vcpu->kvm, sp); |
1d9dc7e0 | 1909 | |
a4a8e6f7 | 1910 | if (vcpu->arch.mmu.sync_page(vcpu, sp)) { |
d98ba053 | 1911 | kvm_mmu_prepare_zap_page(vcpu->kvm, sp, invalid_list); |
4731d4c7 MT |
1912 | return 1; |
1913 | } | |
1914 | ||
77c3913b | 1915 | kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); |
4731d4c7 MT |
1916 | return 0; |
1917 | } | |
1918 | ||
1d9dc7e0 XG |
1919 | static int kvm_sync_page_transient(struct kvm_vcpu *vcpu, |
1920 | struct kvm_mmu_page *sp) | |
1921 | { | |
d98ba053 | 1922 | LIST_HEAD(invalid_list); |
1d9dc7e0 XG |
1923 | int ret; |
1924 | ||
d98ba053 | 1925 | ret = __kvm_sync_page(vcpu, sp, &invalid_list, false); |
be71e061 | 1926 | if (ret) |
d98ba053 XG |
1927 | kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list); |
1928 | ||
1d9dc7e0 XG |
1929 | return ret; |
1930 | } | |
1931 | ||
e37fa785 XG |
1932 | #ifdef CONFIG_KVM_MMU_AUDIT |
1933 | #include "mmu_audit.c" | |
1934 | #else | |
1935 | static void kvm_mmu_audit(struct kvm_vcpu *vcpu, int point) { } | |
1936 | static void mmu_audit_disable(void) { } | |
1937 | #endif | |
1938 | ||
d98ba053 XG |
1939 | static int kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp, |
1940 | struct list_head *invalid_list) | |
1d9dc7e0 | 1941 | { |
d98ba053 | 1942 | return __kvm_sync_page(vcpu, sp, invalid_list, true); |
1d9dc7e0 XG |
1943 | } |
1944 | ||
9f1a122f XG |
1945 | /* @gfn should be write-protected at the call site */ |
1946 | static void kvm_sync_pages(struct kvm_vcpu *vcpu, gfn_t gfn) | |
1947 | { | |
9f1a122f | 1948 | struct kvm_mmu_page *s; |
d98ba053 | 1949 | LIST_HEAD(invalid_list); |
9f1a122f XG |
1950 | bool flush = false; |
1951 | ||
b67bfe0d | 1952 | for_each_gfn_indirect_valid_sp(vcpu->kvm, s, gfn) { |
7ae680eb | 1953 | if (!s->unsync) |
9f1a122f XG |
1954 | continue; |
1955 | ||
1956 | WARN_ON(s->role.level != PT_PAGE_TABLE_LEVEL); | |
a4a8e6f7 | 1957 | kvm_unlink_unsync_page(vcpu->kvm, s); |
9f1a122f | 1958 | if ((s->role.cr4_pae != !!is_pae(vcpu)) || |
a4a8e6f7 | 1959 | (vcpu->arch.mmu.sync_page(vcpu, s))) { |
d98ba053 | 1960 | kvm_mmu_prepare_zap_page(vcpu->kvm, s, &invalid_list); |
9f1a122f XG |
1961 | continue; |
1962 | } | |
9f1a122f XG |
1963 | flush = true; |
1964 | } | |
1965 | ||
d98ba053 | 1966 | kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list); |
9f1a122f | 1967 | if (flush) |
77c3913b | 1968 | kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); |
9f1a122f XG |
1969 | } |
1970 | ||
60c8aec6 MT |
1971 | struct mmu_page_path { |
1972 | struct kvm_mmu_page *parent[PT64_ROOT_LEVEL-1]; | |
1973 | unsigned int idx[PT64_ROOT_LEVEL-1]; | |
4731d4c7 MT |
1974 | }; |
1975 | ||
60c8aec6 MT |
1976 | #define for_each_sp(pvec, sp, parents, i) \ |
1977 | for (i = mmu_pages_next(&pvec, &parents, -1), \ | |
1978 | sp = pvec.page[i].sp; \ | |
1979 | i < pvec.nr && ({ sp = pvec.page[i].sp; 1;}); \ | |
1980 | i = mmu_pages_next(&pvec, &parents, i)) | |
1981 | ||
cded19f3 HE |
1982 | static int mmu_pages_next(struct kvm_mmu_pages *pvec, |
1983 | struct mmu_page_path *parents, | |
1984 | int i) | |
60c8aec6 MT |
1985 | { |
1986 | int n; | |
1987 | ||
1988 | for (n = i+1; n < pvec->nr; n++) { | |
1989 | struct kvm_mmu_page *sp = pvec->page[n].sp; | |
1990 | ||
1991 | if (sp->role.level == PT_PAGE_TABLE_LEVEL) { | |
1992 | parents->idx[0] = pvec->page[n].idx; | |
1993 | return n; | |
1994 | } | |
1995 | ||
1996 | parents->parent[sp->role.level-2] = sp; | |
1997 | parents->idx[sp->role.level-1] = pvec->page[n].idx; | |
1998 | } | |
1999 | ||
2000 | return n; | |
2001 | } | |
2002 | ||
cded19f3 | 2003 | static void mmu_pages_clear_parents(struct mmu_page_path *parents) |
4731d4c7 | 2004 | { |
60c8aec6 MT |
2005 | struct kvm_mmu_page *sp; |
2006 | unsigned int level = 0; | |
2007 | ||
2008 | do { | |
2009 | unsigned int idx = parents->idx[level]; | |
4731d4c7 | 2010 | |
60c8aec6 MT |
2011 | sp = parents->parent[level]; |
2012 | if (!sp) | |
2013 | return; | |
2014 | ||
2015 | --sp->unsync_children; | |
2016 | WARN_ON((int)sp->unsync_children < 0); | |
2017 | __clear_bit(idx, sp->unsync_child_bitmap); | |
2018 | level++; | |
2019 | } while (level < PT64_ROOT_LEVEL-1 && !sp->unsync_children); | |
4731d4c7 MT |
2020 | } |
2021 | ||
60c8aec6 MT |
2022 | static void kvm_mmu_pages_init(struct kvm_mmu_page *parent, |
2023 | struct mmu_page_path *parents, | |
2024 | struct kvm_mmu_pages *pvec) | |
4731d4c7 | 2025 | { |
60c8aec6 MT |
2026 | parents->parent[parent->role.level-1] = NULL; |
2027 | pvec->nr = 0; | |
2028 | } | |
4731d4c7 | 2029 | |
60c8aec6 MT |
2030 | static void mmu_sync_children(struct kvm_vcpu *vcpu, |
2031 | struct kvm_mmu_page *parent) | |
2032 | { | |
2033 | int i; | |
2034 | struct kvm_mmu_page *sp; | |
2035 | struct mmu_page_path parents; | |
2036 | struct kvm_mmu_pages pages; | |
d98ba053 | 2037 | LIST_HEAD(invalid_list); |
60c8aec6 MT |
2038 | |
2039 | kvm_mmu_pages_init(parent, &parents, &pages); | |
2040 | while (mmu_unsync_walk(parent, &pages)) { | |
2f84569f | 2041 | bool protected = false; |
b1a36821 MT |
2042 | |
2043 | for_each_sp(pages, sp, parents, i) | |
54bf36aa | 2044 | protected |= rmap_write_protect(vcpu, sp->gfn); |
b1a36821 MT |
2045 | |
2046 | if (protected) | |
2047 | kvm_flush_remote_tlbs(vcpu->kvm); | |
2048 | ||
60c8aec6 | 2049 | for_each_sp(pages, sp, parents, i) { |
d98ba053 | 2050 | kvm_sync_page(vcpu, sp, &invalid_list); |
60c8aec6 MT |
2051 | mmu_pages_clear_parents(&parents); |
2052 | } | |
d98ba053 | 2053 | kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list); |
4731d4c7 | 2054 | cond_resched_lock(&vcpu->kvm->mmu_lock); |
60c8aec6 MT |
2055 | kvm_mmu_pages_init(parent, &parents, &pages); |
2056 | } | |
4731d4c7 MT |
2057 | } |
2058 | ||
c3707958 XG |
2059 | static void init_shadow_page_table(struct kvm_mmu_page *sp) |
2060 | { | |
2061 | int i; | |
2062 | ||
2063 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) | |
2064 | sp->spt[i] = 0ull; | |
2065 | } | |
2066 | ||
a30f47cb XG |
2067 | static void __clear_sp_write_flooding_count(struct kvm_mmu_page *sp) |
2068 | { | |
2069 | sp->write_flooding_count = 0; | |
2070 | } | |
2071 | ||
2072 | static void clear_sp_write_flooding_count(u64 *spte) | |
2073 | { | |
2074 | struct kvm_mmu_page *sp = page_header(__pa(spte)); | |
2075 | ||
2076 | __clear_sp_write_flooding_count(sp); | |
2077 | } | |
2078 | ||
5304b8d3 XG |
2079 | static bool is_obsolete_sp(struct kvm *kvm, struct kvm_mmu_page *sp) |
2080 | { | |
2081 | return unlikely(sp->mmu_valid_gen != kvm->arch.mmu_valid_gen); | |
2082 | } | |
2083 | ||
cea0f0e7 AK |
2084 | static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu, |
2085 | gfn_t gfn, | |
2086 | gva_t gaddr, | |
2087 | unsigned level, | |
f6e2c02b | 2088 | int direct, |
41074d07 | 2089 | unsigned access, |
f7d9c7b7 | 2090 | u64 *parent_pte) |
cea0f0e7 AK |
2091 | { |
2092 | union kvm_mmu_page_role role; | |
cea0f0e7 | 2093 | unsigned quadrant; |
9f1a122f | 2094 | struct kvm_mmu_page *sp; |
9f1a122f | 2095 | bool need_sync = false; |
cea0f0e7 | 2096 | |
a770f6f2 | 2097 | role = vcpu->arch.mmu.base_role; |
cea0f0e7 | 2098 | role.level = level; |
f6e2c02b | 2099 | role.direct = direct; |
84b0c8c6 | 2100 | if (role.direct) |
5b7e0102 | 2101 | role.cr4_pae = 0; |
41074d07 | 2102 | role.access = access; |
c5a78f2b JR |
2103 | if (!vcpu->arch.mmu.direct_map |
2104 | && vcpu->arch.mmu.root_level <= PT32_ROOT_LEVEL) { | |
cea0f0e7 AK |
2105 | quadrant = gaddr >> (PAGE_SHIFT + (PT64_PT_BITS * level)); |
2106 | quadrant &= (1 << ((PT32_PT_BITS - PT64_PT_BITS) * level)) - 1; | |
2107 | role.quadrant = quadrant; | |
2108 | } | |
b67bfe0d | 2109 | for_each_gfn_sp(vcpu->kvm, sp, gfn) { |
7f52af74 XG |
2110 | if (is_obsolete_sp(vcpu->kvm, sp)) |
2111 | continue; | |
2112 | ||
7ae680eb XG |
2113 | if (!need_sync && sp->unsync) |
2114 | need_sync = true; | |
4731d4c7 | 2115 | |
7ae680eb XG |
2116 | if (sp->role.word != role.word) |
2117 | continue; | |
4731d4c7 | 2118 | |
7ae680eb XG |
2119 | if (sp->unsync && kvm_sync_page_transient(vcpu, sp)) |
2120 | break; | |
e02aa901 | 2121 | |
7ae680eb XG |
2122 | mmu_page_add_parent_pte(vcpu, sp, parent_pte); |
2123 | if (sp->unsync_children) { | |
a8eeb04a | 2124 | kvm_make_request(KVM_REQ_MMU_SYNC, vcpu); |
7ae680eb XG |
2125 | kvm_mmu_mark_parents_unsync(sp); |
2126 | } else if (sp->unsync) | |
2127 | kvm_mmu_mark_parents_unsync(sp); | |
e02aa901 | 2128 | |
a30f47cb | 2129 | __clear_sp_write_flooding_count(sp); |
7ae680eb XG |
2130 | trace_kvm_mmu_get_page(sp, false); |
2131 | return sp; | |
2132 | } | |
dfc5aa00 | 2133 | ++vcpu->kvm->stat.mmu_cache_miss; |
2032a93d | 2134 | sp = kvm_mmu_alloc_page(vcpu, parent_pte, direct); |
4db35314 AK |
2135 | if (!sp) |
2136 | return sp; | |
4db35314 AK |
2137 | sp->gfn = gfn; |
2138 | sp->role = role; | |
7ae680eb XG |
2139 | hlist_add_head(&sp->hash_link, |
2140 | &vcpu->kvm->arch.mmu_page_hash[kvm_page_table_hashfn(gfn)]); | |
f6e2c02b | 2141 | if (!direct) { |
54bf36aa | 2142 | if (rmap_write_protect(vcpu, gfn)) |
b1a36821 | 2143 | kvm_flush_remote_tlbs(vcpu->kvm); |
9f1a122f XG |
2144 | if (level > PT_PAGE_TABLE_LEVEL && need_sync) |
2145 | kvm_sync_pages(vcpu, gfn); | |
2146 | ||
3ed1a478 | 2147 | account_shadowed(vcpu->kvm, sp); |
4731d4c7 | 2148 | } |
5304b8d3 | 2149 | sp->mmu_valid_gen = vcpu->kvm->arch.mmu_valid_gen; |
c3707958 | 2150 | init_shadow_page_table(sp); |
f691fe1d | 2151 | trace_kvm_mmu_get_page(sp, true); |
4db35314 | 2152 | return sp; |
cea0f0e7 AK |
2153 | } |
2154 | ||
2d11123a AK |
2155 | static void shadow_walk_init(struct kvm_shadow_walk_iterator *iterator, |
2156 | struct kvm_vcpu *vcpu, u64 addr) | |
2157 | { | |
2158 | iterator->addr = addr; | |
2159 | iterator->shadow_addr = vcpu->arch.mmu.root_hpa; | |
2160 | iterator->level = vcpu->arch.mmu.shadow_root_level; | |
81407ca5 JR |
2161 | |
2162 | if (iterator->level == PT64_ROOT_LEVEL && | |
2163 | vcpu->arch.mmu.root_level < PT64_ROOT_LEVEL && | |
2164 | !vcpu->arch.mmu.direct_map) | |
2165 | --iterator->level; | |
2166 | ||
2d11123a AK |
2167 | if (iterator->level == PT32E_ROOT_LEVEL) { |
2168 | iterator->shadow_addr | |
2169 | = vcpu->arch.mmu.pae_root[(addr >> 30) & 3]; | |
2170 | iterator->shadow_addr &= PT64_BASE_ADDR_MASK; | |
2171 | --iterator->level; | |
2172 | if (!iterator->shadow_addr) | |
2173 | iterator->level = 0; | |
2174 | } | |
2175 | } | |
2176 | ||
2177 | static bool shadow_walk_okay(struct kvm_shadow_walk_iterator *iterator) | |
2178 | { | |
2179 | if (iterator->level < PT_PAGE_TABLE_LEVEL) | |
2180 | return false; | |
4d88954d | 2181 | |
2d11123a AK |
2182 | iterator->index = SHADOW_PT_INDEX(iterator->addr, iterator->level); |
2183 | iterator->sptep = ((u64 *)__va(iterator->shadow_addr)) + iterator->index; | |
2184 | return true; | |
2185 | } | |
2186 | ||
c2a2ac2b XG |
2187 | static void __shadow_walk_next(struct kvm_shadow_walk_iterator *iterator, |
2188 | u64 spte) | |
2d11123a | 2189 | { |
c2a2ac2b | 2190 | if (is_last_spte(spte, iterator->level)) { |
052331be XG |
2191 | iterator->level = 0; |
2192 | return; | |
2193 | } | |
2194 | ||
c2a2ac2b | 2195 | iterator->shadow_addr = spte & PT64_BASE_ADDR_MASK; |
2d11123a AK |
2196 | --iterator->level; |
2197 | } | |
2198 | ||
c2a2ac2b XG |
2199 | static void shadow_walk_next(struct kvm_shadow_walk_iterator *iterator) |
2200 | { | |
2201 | return __shadow_walk_next(iterator, *iterator->sptep); | |
2202 | } | |
2203 | ||
7a1638ce | 2204 | static void link_shadow_page(u64 *sptep, struct kvm_mmu_page *sp, bool accessed) |
32ef26a3 AK |
2205 | { |
2206 | u64 spte; | |
2207 | ||
7a1638ce YZ |
2208 | BUILD_BUG_ON(VMX_EPT_READABLE_MASK != PT_PRESENT_MASK || |
2209 | VMX_EPT_WRITABLE_MASK != PT_WRITABLE_MASK); | |
2210 | ||
24db2734 | 2211 | spte = __pa(sp->spt) | PT_PRESENT_MASK | PT_WRITABLE_MASK | |
7a1638ce YZ |
2212 | shadow_user_mask | shadow_x_mask; |
2213 | ||
2214 | if (accessed) | |
2215 | spte |= shadow_accessed_mask; | |
24db2734 | 2216 | |
1df9f2dc | 2217 | mmu_spte_set(sptep, spte); |
32ef26a3 AK |
2218 | } |
2219 | ||
a357bd22 AK |
2220 | static void validate_direct_spte(struct kvm_vcpu *vcpu, u64 *sptep, |
2221 | unsigned direct_access) | |
2222 | { | |
2223 | if (is_shadow_present_pte(*sptep) && !is_large_pte(*sptep)) { | |
2224 | struct kvm_mmu_page *child; | |
2225 | ||
2226 | /* | |
2227 | * For the direct sp, if the guest pte's dirty bit | |
2228 | * changed form clean to dirty, it will corrupt the | |
2229 | * sp's access: allow writable in the read-only sp, | |
2230 | * so we should update the spte at this point to get | |
2231 | * a new sp with the correct access. | |
2232 | */ | |
2233 | child = page_header(*sptep & PT64_BASE_ADDR_MASK); | |
2234 | if (child->role.access == direct_access) | |
2235 | return; | |
2236 | ||
bcdd9a93 | 2237 | drop_parent_pte(child, sptep); |
a357bd22 AK |
2238 | kvm_flush_remote_tlbs(vcpu->kvm); |
2239 | } | |
2240 | } | |
2241 | ||
505aef8f | 2242 | static bool mmu_page_zap_pte(struct kvm *kvm, struct kvm_mmu_page *sp, |
38e3b2b2 XG |
2243 | u64 *spte) |
2244 | { | |
2245 | u64 pte; | |
2246 | struct kvm_mmu_page *child; | |
2247 | ||
2248 | pte = *spte; | |
2249 | if (is_shadow_present_pte(pte)) { | |
505aef8f | 2250 | if (is_last_spte(pte, sp->role.level)) { |
c3707958 | 2251 | drop_spte(kvm, spte); |
505aef8f XG |
2252 | if (is_large_pte(pte)) |
2253 | --kvm->stat.lpages; | |
2254 | } else { | |
38e3b2b2 | 2255 | child = page_header(pte & PT64_BASE_ADDR_MASK); |
bcdd9a93 | 2256 | drop_parent_pte(child, spte); |
38e3b2b2 | 2257 | } |
505aef8f XG |
2258 | return true; |
2259 | } | |
2260 | ||
2261 | if (is_mmio_spte(pte)) | |
ce88decf | 2262 | mmu_spte_clear_no_track(spte); |
c3707958 | 2263 | |
505aef8f | 2264 | return false; |
38e3b2b2 XG |
2265 | } |
2266 | ||
90cb0529 | 2267 | static void kvm_mmu_page_unlink_children(struct kvm *kvm, |
4db35314 | 2268 | struct kvm_mmu_page *sp) |
a436036b | 2269 | { |
697fe2e2 | 2270 | unsigned i; |
697fe2e2 | 2271 | |
38e3b2b2 XG |
2272 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) |
2273 | mmu_page_zap_pte(kvm, sp, sp->spt + i); | |
a436036b AK |
2274 | } |
2275 | ||
4db35314 | 2276 | static void kvm_mmu_put_page(struct kvm_mmu_page *sp, u64 *parent_pte) |
cea0f0e7 | 2277 | { |
4db35314 | 2278 | mmu_page_remove_parent_pte(sp, parent_pte); |
a436036b AK |
2279 | } |
2280 | ||
31aa2b44 | 2281 | static void kvm_mmu_unlink_parents(struct kvm *kvm, struct kvm_mmu_page *sp) |
a436036b | 2282 | { |
1e3f42f0 TY |
2283 | u64 *sptep; |
2284 | struct rmap_iterator iter; | |
a436036b | 2285 | |
1e3f42f0 TY |
2286 | while ((sptep = rmap_get_first(sp->parent_ptes, &iter))) |
2287 | drop_parent_pte(sp, sptep); | |
31aa2b44 AK |
2288 | } |
2289 | ||
60c8aec6 | 2290 | static int mmu_zap_unsync_children(struct kvm *kvm, |
7775834a XG |
2291 | struct kvm_mmu_page *parent, |
2292 | struct list_head *invalid_list) | |
4731d4c7 | 2293 | { |
60c8aec6 MT |
2294 | int i, zapped = 0; |
2295 | struct mmu_page_path parents; | |
2296 | struct kvm_mmu_pages pages; | |
4731d4c7 | 2297 | |
60c8aec6 | 2298 | if (parent->role.level == PT_PAGE_TABLE_LEVEL) |
4731d4c7 | 2299 | return 0; |
60c8aec6 MT |
2300 | |
2301 | kvm_mmu_pages_init(parent, &parents, &pages); | |
2302 | while (mmu_unsync_walk(parent, &pages)) { | |
2303 | struct kvm_mmu_page *sp; | |
2304 | ||
2305 | for_each_sp(pages, sp, parents, i) { | |
7775834a | 2306 | kvm_mmu_prepare_zap_page(kvm, sp, invalid_list); |
60c8aec6 | 2307 | mmu_pages_clear_parents(&parents); |
77662e00 | 2308 | zapped++; |
60c8aec6 | 2309 | } |
60c8aec6 MT |
2310 | kvm_mmu_pages_init(parent, &parents, &pages); |
2311 | } | |
2312 | ||
2313 | return zapped; | |
4731d4c7 MT |
2314 | } |
2315 | ||
7775834a XG |
2316 | static int kvm_mmu_prepare_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp, |
2317 | struct list_head *invalid_list) | |
31aa2b44 | 2318 | { |
4731d4c7 | 2319 | int ret; |
f691fe1d | 2320 | |
7775834a | 2321 | trace_kvm_mmu_prepare_zap_page(sp); |
31aa2b44 | 2322 | ++kvm->stat.mmu_shadow_zapped; |
7775834a | 2323 | ret = mmu_zap_unsync_children(kvm, sp, invalid_list); |
4db35314 | 2324 | kvm_mmu_page_unlink_children(kvm, sp); |
31aa2b44 | 2325 | kvm_mmu_unlink_parents(kvm, sp); |
5304b8d3 | 2326 | |
f6e2c02b | 2327 | if (!sp->role.invalid && !sp->role.direct) |
3ed1a478 | 2328 | unaccount_shadowed(kvm, sp); |
5304b8d3 | 2329 | |
4731d4c7 MT |
2330 | if (sp->unsync) |
2331 | kvm_unlink_unsync_page(kvm, sp); | |
4db35314 | 2332 | if (!sp->root_count) { |
54a4f023 GJ |
2333 | /* Count self */ |
2334 | ret++; | |
7775834a | 2335 | list_move(&sp->link, invalid_list); |
aa6bd187 | 2336 | kvm_mod_used_mmu_pages(kvm, -1); |
2e53d63a | 2337 | } else { |
5b5c6a5a | 2338 | list_move(&sp->link, &kvm->arch.active_mmu_pages); |
05988d72 GN |
2339 | |
2340 | /* | |
2341 | * The obsolete pages can not be used on any vcpus. | |
2342 | * See the comments in kvm_mmu_invalidate_zap_all_pages(). | |
2343 | */ | |
2344 | if (!sp->role.invalid && !is_obsolete_sp(kvm, sp)) | |
2345 | kvm_reload_remote_mmus(kvm); | |
2e53d63a | 2346 | } |
7775834a XG |
2347 | |
2348 | sp->role.invalid = 1; | |
4731d4c7 | 2349 | return ret; |
a436036b AK |
2350 | } |
2351 | ||
7775834a XG |
2352 | static void kvm_mmu_commit_zap_page(struct kvm *kvm, |
2353 | struct list_head *invalid_list) | |
2354 | { | |
945315b9 | 2355 | struct kvm_mmu_page *sp, *nsp; |
7775834a XG |
2356 | |
2357 | if (list_empty(invalid_list)) | |
2358 | return; | |
2359 | ||
c142786c AK |
2360 | /* |
2361 | * wmb: make sure everyone sees our modifications to the page tables | |
2362 | * rmb: make sure we see changes to vcpu->mode | |
2363 | */ | |
2364 | smp_mb(); | |
4f022648 | 2365 | |
c142786c AK |
2366 | /* |
2367 | * Wait for all vcpus to exit guest mode and/or lockless shadow | |
2368 | * page table walks. | |
2369 | */ | |
2370 | kvm_flush_remote_tlbs(kvm); | |
c2a2ac2b | 2371 | |
945315b9 | 2372 | list_for_each_entry_safe(sp, nsp, invalid_list, link) { |
7775834a | 2373 | WARN_ON(!sp->role.invalid || sp->root_count); |
aa6bd187 | 2374 | kvm_mmu_free_page(sp); |
945315b9 | 2375 | } |
7775834a XG |
2376 | } |
2377 | ||
5da59607 TY |
2378 | static bool prepare_zap_oldest_mmu_page(struct kvm *kvm, |
2379 | struct list_head *invalid_list) | |
2380 | { | |
2381 | struct kvm_mmu_page *sp; | |
2382 | ||
2383 | if (list_empty(&kvm->arch.active_mmu_pages)) | |
2384 | return false; | |
2385 | ||
2386 | sp = list_entry(kvm->arch.active_mmu_pages.prev, | |
2387 | struct kvm_mmu_page, link); | |
2388 | kvm_mmu_prepare_zap_page(kvm, sp, invalid_list); | |
2389 | ||
2390 | return true; | |
2391 | } | |
2392 | ||
82ce2c96 IE |
2393 | /* |
2394 | * Changing the number of mmu pages allocated to the vm | |
49d5ca26 | 2395 | * Note: if goal_nr_mmu_pages is too small, you will get dead lock |
82ce2c96 | 2396 | */ |
49d5ca26 | 2397 | void kvm_mmu_change_mmu_pages(struct kvm *kvm, unsigned int goal_nr_mmu_pages) |
82ce2c96 | 2398 | { |
d98ba053 | 2399 | LIST_HEAD(invalid_list); |
82ce2c96 | 2400 | |
b34cb590 TY |
2401 | spin_lock(&kvm->mmu_lock); |
2402 | ||
49d5ca26 | 2403 | if (kvm->arch.n_used_mmu_pages > goal_nr_mmu_pages) { |
5da59607 TY |
2404 | /* Need to free some mmu pages to achieve the goal. */ |
2405 | while (kvm->arch.n_used_mmu_pages > goal_nr_mmu_pages) | |
2406 | if (!prepare_zap_oldest_mmu_page(kvm, &invalid_list)) | |
2407 | break; | |
82ce2c96 | 2408 | |
aa6bd187 | 2409 | kvm_mmu_commit_zap_page(kvm, &invalid_list); |
49d5ca26 | 2410 | goal_nr_mmu_pages = kvm->arch.n_used_mmu_pages; |
82ce2c96 | 2411 | } |
82ce2c96 | 2412 | |
49d5ca26 | 2413 | kvm->arch.n_max_mmu_pages = goal_nr_mmu_pages; |
b34cb590 TY |
2414 | |
2415 | spin_unlock(&kvm->mmu_lock); | |
82ce2c96 IE |
2416 | } |
2417 | ||
1cb3f3ae | 2418 | int kvm_mmu_unprotect_page(struct kvm *kvm, gfn_t gfn) |
a436036b | 2419 | { |
4db35314 | 2420 | struct kvm_mmu_page *sp; |
d98ba053 | 2421 | LIST_HEAD(invalid_list); |
a436036b AK |
2422 | int r; |
2423 | ||
9ad17b10 | 2424 | pgprintk("%s: looking for gfn %llx\n", __func__, gfn); |
a436036b | 2425 | r = 0; |
1cb3f3ae | 2426 | spin_lock(&kvm->mmu_lock); |
b67bfe0d | 2427 | for_each_gfn_indirect_valid_sp(kvm, sp, gfn) { |
9ad17b10 | 2428 | pgprintk("%s: gfn %llx role %x\n", __func__, gfn, |
7ae680eb XG |
2429 | sp->role.word); |
2430 | r = 1; | |
f41d335a | 2431 | kvm_mmu_prepare_zap_page(kvm, sp, &invalid_list); |
7ae680eb | 2432 | } |
d98ba053 | 2433 | kvm_mmu_commit_zap_page(kvm, &invalid_list); |
1cb3f3ae XG |
2434 | spin_unlock(&kvm->mmu_lock); |
2435 | ||
a436036b | 2436 | return r; |
cea0f0e7 | 2437 | } |
1cb3f3ae | 2438 | EXPORT_SYMBOL_GPL(kvm_mmu_unprotect_page); |
cea0f0e7 | 2439 | |
9cf5cf5a XG |
2440 | static void __kvm_unsync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp) |
2441 | { | |
2442 | trace_kvm_mmu_unsync_page(sp); | |
2443 | ++vcpu->kvm->stat.mmu_unsync; | |
2444 | sp->unsync = 1; | |
2445 | ||
2446 | kvm_mmu_mark_parents_unsync(sp); | |
9cf5cf5a XG |
2447 | } |
2448 | ||
2449 | static void kvm_unsync_pages(struct kvm_vcpu *vcpu, gfn_t gfn) | |
4731d4c7 | 2450 | { |
4731d4c7 | 2451 | struct kvm_mmu_page *s; |
9cf5cf5a | 2452 | |
b67bfe0d | 2453 | for_each_gfn_indirect_valid_sp(vcpu->kvm, s, gfn) { |
7ae680eb | 2454 | if (s->unsync) |
4731d4c7 | 2455 | continue; |
9cf5cf5a XG |
2456 | WARN_ON(s->role.level != PT_PAGE_TABLE_LEVEL); |
2457 | __kvm_unsync_page(vcpu, s); | |
4731d4c7 | 2458 | } |
4731d4c7 MT |
2459 | } |
2460 | ||
2461 | static int mmu_need_write_protect(struct kvm_vcpu *vcpu, gfn_t gfn, | |
2462 | bool can_unsync) | |
2463 | { | |
9cf5cf5a | 2464 | struct kvm_mmu_page *s; |
9cf5cf5a XG |
2465 | bool need_unsync = false; |
2466 | ||
b67bfe0d | 2467 | for_each_gfn_indirect_valid_sp(vcpu->kvm, s, gfn) { |
36a2e677 XG |
2468 | if (!can_unsync) |
2469 | return 1; | |
2470 | ||
9cf5cf5a | 2471 | if (s->role.level != PT_PAGE_TABLE_LEVEL) |
4731d4c7 | 2472 | return 1; |
9cf5cf5a | 2473 | |
9bb4f6b1 | 2474 | if (!s->unsync) |
9cf5cf5a | 2475 | need_unsync = true; |
4731d4c7 | 2476 | } |
9cf5cf5a XG |
2477 | if (need_unsync) |
2478 | kvm_unsync_pages(vcpu, gfn); | |
4731d4c7 MT |
2479 | return 0; |
2480 | } | |
2481 | ||
d1fe9219 PB |
2482 | static bool kvm_is_mmio_pfn(pfn_t pfn) |
2483 | { | |
2484 | if (pfn_valid(pfn)) | |
2485 | return !is_zero_pfn(pfn) && PageReserved(pfn_to_page(pfn)); | |
2486 | ||
2487 | return true; | |
2488 | } | |
2489 | ||
d555c333 | 2490 | static int set_spte(struct kvm_vcpu *vcpu, u64 *sptep, |
c2288505 | 2491 | unsigned pte_access, int level, |
c2d0ee46 | 2492 | gfn_t gfn, pfn_t pfn, bool speculative, |
9bdbba13 | 2493 | bool can_unsync, bool host_writable) |
1c4f1fd6 | 2494 | { |
6e7d0354 | 2495 | u64 spte; |
1e73f9dd | 2496 | int ret = 0; |
64d4d521 | 2497 | |
54bf36aa | 2498 | if (set_mmio_spte(vcpu, sptep, gfn, pfn, pte_access)) |
ce88decf XG |
2499 | return 0; |
2500 | ||
982c2565 | 2501 | spte = PT_PRESENT_MASK; |
947da538 | 2502 | if (!speculative) |
3201b5d9 | 2503 | spte |= shadow_accessed_mask; |
640d9b0d | 2504 | |
7b52345e SY |
2505 | if (pte_access & ACC_EXEC_MASK) |
2506 | spte |= shadow_x_mask; | |
2507 | else | |
2508 | spte |= shadow_nx_mask; | |
49fde340 | 2509 | |
1c4f1fd6 | 2510 | if (pte_access & ACC_USER_MASK) |
7b52345e | 2511 | spte |= shadow_user_mask; |
49fde340 | 2512 | |
852e3c19 | 2513 | if (level > PT_PAGE_TABLE_LEVEL) |
05da4558 | 2514 | spte |= PT_PAGE_SIZE_MASK; |
b0bc3ee2 | 2515 | if (tdp_enabled) |
4b12f0de | 2516 | spte |= kvm_x86_ops->get_mt_mask(vcpu, gfn, |
d1fe9219 | 2517 | kvm_is_mmio_pfn(pfn)); |
1c4f1fd6 | 2518 | |
9bdbba13 | 2519 | if (host_writable) |
1403283a | 2520 | spte |= SPTE_HOST_WRITEABLE; |
f8e453b0 XG |
2521 | else |
2522 | pte_access &= ~ACC_WRITE_MASK; | |
1403283a | 2523 | |
35149e21 | 2524 | spte |= (u64)pfn << PAGE_SHIFT; |
1c4f1fd6 | 2525 | |
c2288505 | 2526 | if (pte_access & ACC_WRITE_MASK) { |
1c4f1fd6 | 2527 | |
c2193463 | 2528 | /* |
7751babd XG |
2529 | * Other vcpu creates new sp in the window between |
2530 | * mapping_level() and acquiring mmu-lock. We can | |
2531 | * allow guest to retry the access, the mapping can | |
2532 | * be fixed if guest refault. | |
c2193463 | 2533 | */ |
852e3c19 | 2534 | if (level > PT_PAGE_TABLE_LEVEL && |
54bf36aa | 2535 | has_wrprotected_page(vcpu, gfn, level)) |
be38d276 | 2536 | goto done; |
38187c83 | 2537 | |
49fde340 | 2538 | spte |= PT_WRITABLE_MASK | SPTE_MMU_WRITEABLE; |
1c4f1fd6 | 2539 | |
ecc5589f MT |
2540 | /* |
2541 | * Optimization: for pte sync, if spte was writable the hash | |
2542 | * lookup is unnecessary (and expensive). Write protection | |
2543 | * is responsibility of mmu_get_page / kvm_sync_page. | |
2544 | * Same reasoning can be applied to dirty page accounting. | |
2545 | */ | |
8dae4445 | 2546 | if (!can_unsync && is_writable_pte(*sptep)) |
ecc5589f MT |
2547 | goto set_pte; |
2548 | ||
4731d4c7 | 2549 | if (mmu_need_write_protect(vcpu, gfn, can_unsync)) { |
9ad17b10 | 2550 | pgprintk("%s: found shadow page for %llx, marking ro\n", |
b8688d51 | 2551 | __func__, gfn); |
1e73f9dd | 2552 | ret = 1; |
1c4f1fd6 | 2553 | pte_access &= ~ACC_WRITE_MASK; |
49fde340 | 2554 | spte &= ~(PT_WRITABLE_MASK | SPTE_MMU_WRITEABLE); |
1c4f1fd6 AK |
2555 | } |
2556 | } | |
2557 | ||
9b51a630 | 2558 | if (pte_access & ACC_WRITE_MASK) { |
54bf36aa | 2559 | kvm_vcpu_mark_page_dirty(vcpu, gfn); |
9b51a630 KH |
2560 | spte |= shadow_dirty_mask; |
2561 | } | |
1c4f1fd6 | 2562 | |
38187c83 | 2563 | set_pte: |
6e7d0354 | 2564 | if (mmu_spte_update(sptep, spte)) |
b330aa0c | 2565 | kvm_flush_remote_tlbs(vcpu->kvm); |
be38d276 | 2566 | done: |
1e73f9dd MT |
2567 | return ret; |
2568 | } | |
2569 | ||
d555c333 | 2570 | static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *sptep, |
f7616203 XG |
2571 | unsigned pte_access, int write_fault, int *emulate, |
2572 | int level, gfn_t gfn, pfn_t pfn, bool speculative, | |
2573 | bool host_writable) | |
1e73f9dd MT |
2574 | { |
2575 | int was_rmapped = 0; | |
53a27b39 | 2576 | int rmap_count; |
1e73f9dd | 2577 | |
f7616203 XG |
2578 | pgprintk("%s: spte %llx write_fault %d gfn %llx\n", __func__, |
2579 | *sptep, write_fault, gfn); | |
1e73f9dd | 2580 | |
d555c333 | 2581 | if (is_rmap_spte(*sptep)) { |
1e73f9dd MT |
2582 | /* |
2583 | * If we overwrite a PTE page pointer with a 2MB PMD, unlink | |
2584 | * the parent of the now unreachable PTE. | |
2585 | */ | |
852e3c19 JR |
2586 | if (level > PT_PAGE_TABLE_LEVEL && |
2587 | !is_large_pte(*sptep)) { | |
1e73f9dd | 2588 | struct kvm_mmu_page *child; |
d555c333 | 2589 | u64 pte = *sptep; |
1e73f9dd MT |
2590 | |
2591 | child = page_header(pte & PT64_BASE_ADDR_MASK); | |
bcdd9a93 | 2592 | drop_parent_pte(child, sptep); |
3be2264b | 2593 | kvm_flush_remote_tlbs(vcpu->kvm); |
d555c333 | 2594 | } else if (pfn != spte_to_pfn(*sptep)) { |
9ad17b10 | 2595 | pgprintk("hfn old %llx new %llx\n", |
d555c333 | 2596 | spte_to_pfn(*sptep), pfn); |
c3707958 | 2597 | drop_spte(vcpu->kvm, sptep); |
91546356 | 2598 | kvm_flush_remote_tlbs(vcpu->kvm); |
6bed6b9e JR |
2599 | } else |
2600 | was_rmapped = 1; | |
1e73f9dd | 2601 | } |
852e3c19 | 2602 | |
c2288505 XG |
2603 | if (set_spte(vcpu, sptep, pte_access, level, gfn, pfn, speculative, |
2604 | true, host_writable)) { | |
1e73f9dd | 2605 | if (write_fault) |
b90a0e6c | 2606 | *emulate = 1; |
77c3913b | 2607 | kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); |
a378b4e6 | 2608 | } |
1e73f9dd | 2609 | |
ce88decf XG |
2610 | if (unlikely(is_mmio_spte(*sptep) && emulate)) |
2611 | *emulate = 1; | |
2612 | ||
d555c333 | 2613 | pgprintk("%s: setting spte %llx\n", __func__, *sptep); |
9ad17b10 | 2614 | pgprintk("instantiating %s PTE (%s) at %llx (%llx) addr %p\n", |
d555c333 | 2615 | is_large_pte(*sptep)? "2MB" : "4kB", |
a205bc19 JR |
2616 | *sptep & PT_PRESENT_MASK ?"RW":"R", gfn, |
2617 | *sptep, sptep); | |
d555c333 | 2618 | if (!was_rmapped && is_large_pte(*sptep)) |
05da4558 MT |
2619 | ++vcpu->kvm->stat.lpages; |
2620 | ||
ffb61bb3 | 2621 | if (is_shadow_present_pte(*sptep)) { |
ffb61bb3 XG |
2622 | if (!was_rmapped) { |
2623 | rmap_count = rmap_add(vcpu, sptep, gfn); | |
2624 | if (rmap_count > RMAP_RECYCLE_THRESHOLD) | |
2625 | rmap_recycle(vcpu, sptep, gfn); | |
2626 | } | |
1c4f1fd6 | 2627 | } |
cb9aaa30 | 2628 | |
f3ac1a4b | 2629 | kvm_release_pfn_clean(pfn); |
1c4f1fd6 AK |
2630 | } |
2631 | ||
957ed9ef XG |
2632 | static pfn_t pte_prefetch_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn, |
2633 | bool no_dirty_log) | |
2634 | { | |
2635 | struct kvm_memory_slot *slot; | |
957ed9ef | 2636 | |
5d163b1c | 2637 | slot = gfn_to_memslot_dirty_bitmap(vcpu, gfn, no_dirty_log); |
903816fa | 2638 | if (!slot) |
6c8ee57b | 2639 | return KVM_PFN_ERR_FAULT; |
957ed9ef | 2640 | |
037d92dc | 2641 | return gfn_to_pfn_memslot_atomic(slot, gfn); |
957ed9ef XG |
2642 | } |
2643 | ||
2644 | static int direct_pte_prefetch_many(struct kvm_vcpu *vcpu, | |
2645 | struct kvm_mmu_page *sp, | |
2646 | u64 *start, u64 *end) | |
2647 | { | |
2648 | struct page *pages[PTE_PREFETCH_NUM]; | |
d9ef13c2 | 2649 | struct kvm_memory_slot *slot; |
957ed9ef XG |
2650 | unsigned access = sp->role.access; |
2651 | int i, ret; | |
2652 | gfn_t gfn; | |
2653 | ||
2654 | gfn = kvm_mmu_page_get_gfn(sp, start - sp->spt); | |
d9ef13c2 PB |
2655 | slot = gfn_to_memslot_dirty_bitmap(vcpu, gfn, access & ACC_WRITE_MASK); |
2656 | if (!slot) | |
957ed9ef XG |
2657 | return -1; |
2658 | ||
d9ef13c2 | 2659 | ret = gfn_to_page_many_atomic(slot, gfn, pages, end - start); |
957ed9ef XG |
2660 | if (ret <= 0) |
2661 | return -1; | |
2662 | ||
2663 | for (i = 0; i < ret; i++, gfn++, start++) | |
f7616203 | 2664 | mmu_set_spte(vcpu, start, access, 0, NULL, |
c2288505 XG |
2665 | sp->role.level, gfn, page_to_pfn(pages[i]), |
2666 | true, true); | |
957ed9ef XG |
2667 | |
2668 | return 0; | |
2669 | } | |
2670 | ||
2671 | static void __direct_pte_prefetch(struct kvm_vcpu *vcpu, | |
2672 | struct kvm_mmu_page *sp, u64 *sptep) | |
2673 | { | |
2674 | u64 *spte, *start = NULL; | |
2675 | int i; | |
2676 | ||
2677 | WARN_ON(!sp->role.direct); | |
2678 | ||
2679 | i = (sptep - sp->spt) & ~(PTE_PREFETCH_NUM - 1); | |
2680 | spte = sp->spt + i; | |
2681 | ||
2682 | for (i = 0; i < PTE_PREFETCH_NUM; i++, spte++) { | |
c3707958 | 2683 | if (is_shadow_present_pte(*spte) || spte == sptep) { |
957ed9ef XG |
2684 | if (!start) |
2685 | continue; | |
2686 | if (direct_pte_prefetch_many(vcpu, sp, start, spte) < 0) | |
2687 | break; | |
2688 | start = NULL; | |
2689 | } else if (!start) | |
2690 | start = spte; | |
2691 | } | |
2692 | } | |
2693 | ||
2694 | static void direct_pte_prefetch(struct kvm_vcpu *vcpu, u64 *sptep) | |
2695 | { | |
2696 | struct kvm_mmu_page *sp; | |
2697 | ||
2698 | /* | |
2699 | * Since it's no accessed bit on EPT, it's no way to | |
2700 | * distinguish between actually accessed translations | |
2701 | * and prefetched, so disable pte prefetch if EPT is | |
2702 | * enabled. | |
2703 | */ | |
2704 | if (!shadow_accessed_mask) | |
2705 | return; | |
2706 | ||
2707 | sp = page_header(__pa(sptep)); | |
2708 | if (sp->role.level > PT_PAGE_TABLE_LEVEL) | |
2709 | return; | |
2710 | ||
2711 | __direct_pte_prefetch(vcpu, sp, sptep); | |
2712 | } | |
2713 | ||
9f652d21 | 2714 | static int __direct_map(struct kvm_vcpu *vcpu, gpa_t v, int write, |
2ec4739d XG |
2715 | int map_writable, int level, gfn_t gfn, pfn_t pfn, |
2716 | bool prefault) | |
140754bc | 2717 | { |
9f652d21 | 2718 | struct kvm_shadow_walk_iterator iterator; |
140754bc | 2719 | struct kvm_mmu_page *sp; |
b90a0e6c | 2720 | int emulate = 0; |
140754bc | 2721 | gfn_t pseudo_gfn; |
6aa8b732 | 2722 | |
989c6b34 MT |
2723 | if (!VALID_PAGE(vcpu->arch.mmu.root_hpa)) |
2724 | return 0; | |
2725 | ||
9f652d21 | 2726 | for_each_shadow_entry(vcpu, (u64)gfn << PAGE_SHIFT, iterator) { |
852e3c19 | 2727 | if (iterator.level == level) { |
f7616203 | 2728 | mmu_set_spte(vcpu, iterator.sptep, ACC_ALL, |
c2288505 XG |
2729 | write, &emulate, level, gfn, pfn, |
2730 | prefault, map_writable); | |
957ed9ef | 2731 | direct_pte_prefetch(vcpu, iterator.sptep); |
9f652d21 AK |
2732 | ++vcpu->stat.pf_fixed; |
2733 | break; | |
6aa8b732 AK |
2734 | } |
2735 | ||
404381c5 | 2736 | drop_large_spte(vcpu, iterator.sptep); |
c3707958 | 2737 | if (!is_shadow_present_pte(*iterator.sptep)) { |
c9fa0b3b LJ |
2738 | u64 base_addr = iterator.addr; |
2739 | ||
2740 | base_addr &= PT64_LVL_ADDR_MASK(iterator.level); | |
2741 | pseudo_gfn = base_addr >> PAGE_SHIFT; | |
9f652d21 AK |
2742 | sp = kvm_mmu_get_page(vcpu, pseudo_gfn, iterator.addr, |
2743 | iterator.level - 1, | |
2744 | 1, ACC_ALL, iterator.sptep); | |
140754bc | 2745 | |
7a1638ce | 2746 | link_shadow_page(iterator.sptep, sp, true); |
9f652d21 AK |
2747 | } |
2748 | } | |
b90a0e6c | 2749 | return emulate; |
6aa8b732 AK |
2750 | } |
2751 | ||
77db5cbd | 2752 | static void kvm_send_hwpoison_signal(unsigned long address, struct task_struct *tsk) |
bf998156 | 2753 | { |
77db5cbd HY |
2754 | siginfo_t info; |
2755 | ||
2756 | info.si_signo = SIGBUS; | |
2757 | info.si_errno = 0; | |
2758 | info.si_code = BUS_MCEERR_AR; | |
2759 | info.si_addr = (void __user *)address; | |
2760 | info.si_addr_lsb = PAGE_SHIFT; | |
bf998156 | 2761 | |
77db5cbd | 2762 | send_sig_info(SIGBUS, &info, tsk); |
bf998156 HY |
2763 | } |
2764 | ||
d7c55201 | 2765 | static int kvm_handle_bad_page(struct kvm_vcpu *vcpu, gfn_t gfn, pfn_t pfn) |
bf998156 | 2766 | { |
4d8b81ab XG |
2767 | /* |
2768 | * Do not cache the mmio info caused by writing the readonly gfn | |
2769 | * into the spte otherwise read access on readonly gfn also can | |
2770 | * caused mmio page fault and treat it as mmio access. | |
2771 | * Return 1 to tell kvm to emulate it. | |
2772 | */ | |
2773 | if (pfn == KVM_PFN_ERR_RO_FAULT) | |
2774 | return 1; | |
2775 | ||
e6c1502b | 2776 | if (pfn == KVM_PFN_ERR_HWPOISON) { |
54bf36aa | 2777 | kvm_send_hwpoison_signal(kvm_vcpu_gfn_to_hva(vcpu, gfn), current); |
bf998156 | 2778 | return 0; |
d7c55201 | 2779 | } |
edba23e5 | 2780 | |
d7c55201 | 2781 | return -EFAULT; |
bf998156 HY |
2782 | } |
2783 | ||
936a5fe6 AA |
2784 | static void transparent_hugepage_adjust(struct kvm_vcpu *vcpu, |
2785 | gfn_t *gfnp, pfn_t *pfnp, int *levelp) | |
2786 | { | |
2787 | pfn_t pfn = *pfnp; | |
2788 | gfn_t gfn = *gfnp; | |
2789 | int level = *levelp; | |
2790 | ||
2791 | /* | |
2792 | * Check if it's a transparent hugepage. If this would be an | |
2793 | * hugetlbfs page, level wouldn't be set to | |
2794 | * PT_PAGE_TABLE_LEVEL and there would be no adjustment done | |
2795 | * here. | |
2796 | */ | |
bf4bea8e | 2797 | if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn) && |
936a5fe6 AA |
2798 | level == PT_PAGE_TABLE_LEVEL && |
2799 | PageTransCompound(pfn_to_page(pfn)) && | |
54bf36aa | 2800 | !has_wrprotected_page(vcpu, gfn, PT_DIRECTORY_LEVEL)) { |
936a5fe6 AA |
2801 | unsigned long mask; |
2802 | /* | |
2803 | * mmu_notifier_retry was successful and we hold the | |
2804 | * mmu_lock here, so the pmd can't become splitting | |
2805 | * from under us, and in turn | |
2806 | * __split_huge_page_refcount() can't run from under | |
2807 | * us and we can safely transfer the refcount from | |
2808 | * PG_tail to PG_head as we switch the pfn to tail to | |
2809 | * head. | |
2810 | */ | |
2811 | *levelp = level = PT_DIRECTORY_LEVEL; | |
2812 | mask = KVM_PAGES_PER_HPAGE(level) - 1; | |
2813 | VM_BUG_ON((gfn & mask) != (pfn & mask)); | |
2814 | if (pfn & mask) { | |
2815 | gfn &= ~mask; | |
2816 | *gfnp = gfn; | |
2817 | kvm_release_pfn_clean(pfn); | |
2818 | pfn &= ~mask; | |
c3586667 | 2819 | kvm_get_pfn(pfn); |
936a5fe6 AA |
2820 | *pfnp = pfn; |
2821 | } | |
2822 | } | |
2823 | } | |
2824 | ||
d7c55201 XG |
2825 | static bool handle_abnormal_pfn(struct kvm_vcpu *vcpu, gva_t gva, gfn_t gfn, |
2826 | pfn_t pfn, unsigned access, int *ret_val) | |
2827 | { | |
2828 | bool ret = true; | |
2829 | ||
2830 | /* The pfn is invalid, report the error! */ | |
81c52c56 | 2831 | if (unlikely(is_error_pfn(pfn))) { |
d7c55201 XG |
2832 | *ret_val = kvm_handle_bad_page(vcpu, gfn, pfn); |
2833 | goto exit; | |
2834 | } | |
2835 | ||
ce88decf | 2836 | if (unlikely(is_noslot_pfn(pfn))) |
d7c55201 | 2837 | vcpu_cache_mmio_info(vcpu, gva, gfn, access); |
d7c55201 XG |
2838 | |
2839 | ret = false; | |
2840 | exit: | |
2841 | return ret; | |
2842 | } | |
2843 | ||
e5552fd2 | 2844 | static bool page_fault_can_be_fast(u32 error_code) |
c7ba5b48 | 2845 | { |
1c118b82 XG |
2846 | /* |
2847 | * Do not fix the mmio spte with invalid generation number which | |
2848 | * need to be updated by slow page fault path. | |
2849 | */ | |
2850 | if (unlikely(error_code & PFERR_RSVD_MASK)) | |
2851 | return false; | |
2852 | ||
c7ba5b48 XG |
2853 | /* |
2854 | * #PF can be fast only if the shadow page table is present and it | |
2855 | * is caused by write-protect, that means we just need change the | |
2856 | * W bit of the spte which can be done out of mmu-lock. | |
2857 | */ | |
2858 | if (!(error_code & PFERR_PRESENT_MASK) || | |
2859 | !(error_code & PFERR_WRITE_MASK)) | |
2860 | return false; | |
2861 | ||
2862 | return true; | |
2863 | } | |
2864 | ||
2865 | static bool | |
92a476cb XG |
2866 | fast_pf_fix_direct_spte(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp, |
2867 | u64 *sptep, u64 spte) | |
c7ba5b48 | 2868 | { |
c7ba5b48 XG |
2869 | gfn_t gfn; |
2870 | ||
2871 | WARN_ON(!sp->role.direct); | |
2872 | ||
2873 | /* | |
2874 | * The gfn of direct spte is stable since it is calculated | |
2875 | * by sp->gfn. | |
2876 | */ | |
2877 | gfn = kvm_mmu_page_get_gfn(sp, sptep - sp->spt); | |
2878 | ||
9b51a630 KH |
2879 | /* |
2880 | * Theoretically we could also set dirty bit (and flush TLB) here in | |
2881 | * order to eliminate unnecessary PML logging. See comments in | |
2882 | * set_spte. But fast_page_fault is very unlikely to happen with PML | |
2883 | * enabled, so we do not do this. This might result in the same GPA | |
2884 | * to be logged in PML buffer again when the write really happens, and | |
2885 | * eventually to be called by mark_page_dirty twice. But it's also no | |
2886 | * harm. This also avoids the TLB flush needed after setting dirty bit | |
2887 | * so non-PML cases won't be impacted. | |
2888 | * | |
2889 | * Compare with set_spte where instead shadow_dirty_mask is set. | |
2890 | */ | |
c7ba5b48 | 2891 | if (cmpxchg64(sptep, spte, spte | PT_WRITABLE_MASK) == spte) |
54bf36aa | 2892 | kvm_vcpu_mark_page_dirty(vcpu, gfn); |
c7ba5b48 XG |
2893 | |
2894 | return true; | |
2895 | } | |
2896 | ||
2897 | /* | |
2898 | * Return value: | |
2899 | * - true: let the vcpu to access on the same address again. | |
2900 | * - false: let the real page fault path to fix it. | |
2901 | */ | |
2902 | static bool fast_page_fault(struct kvm_vcpu *vcpu, gva_t gva, int level, | |
2903 | u32 error_code) | |
2904 | { | |
2905 | struct kvm_shadow_walk_iterator iterator; | |
92a476cb | 2906 | struct kvm_mmu_page *sp; |
c7ba5b48 XG |
2907 | bool ret = false; |
2908 | u64 spte = 0ull; | |
2909 | ||
37f6a4e2 MT |
2910 | if (!VALID_PAGE(vcpu->arch.mmu.root_hpa)) |
2911 | return false; | |
2912 | ||
e5552fd2 | 2913 | if (!page_fault_can_be_fast(error_code)) |
c7ba5b48 XG |
2914 | return false; |
2915 | ||
2916 | walk_shadow_page_lockless_begin(vcpu); | |
2917 | for_each_shadow_entry_lockless(vcpu, gva, iterator, spte) | |
2918 | if (!is_shadow_present_pte(spte) || iterator.level < level) | |
2919 | break; | |
2920 | ||
2921 | /* | |
2922 | * If the mapping has been changed, let the vcpu fault on the | |
2923 | * same address again. | |
2924 | */ | |
2925 | if (!is_rmap_spte(spte)) { | |
2926 | ret = true; | |
2927 | goto exit; | |
2928 | } | |
2929 | ||
92a476cb XG |
2930 | sp = page_header(__pa(iterator.sptep)); |
2931 | if (!is_last_spte(spte, sp->role.level)) | |
c7ba5b48 XG |
2932 | goto exit; |
2933 | ||
2934 | /* | |
2935 | * Check if it is a spurious fault caused by TLB lazily flushed. | |
2936 | * | |
2937 | * Need not check the access of upper level table entries since | |
2938 | * they are always ACC_ALL. | |
2939 | */ | |
2940 | if (is_writable_pte(spte)) { | |
2941 | ret = true; | |
2942 | goto exit; | |
2943 | } | |
2944 | ||
2945 | /* | |
2946 | * Currently, to simplify the code, only the spte write-protected | |
2947 | * by dirty-log can be fast fixed. | |
2948 | */ | |
2949 | if (!spte_is_locklessly_modifiable(spte)) | |
2950 | goto exit; | |
2951 | ||
c126d94f XG |
2952 | /* |
2953 | * Do not fix write-permission on the large spte since we only dirty | |
2954 | * the first page into the dirty-bitmap in fast_pf_fix_direct_spte() | |
2955 | * that means other pages are missed if its slot is dirty-logged. | |
2956 | * | |
2957 | * Instead, we let the slow page fault path create a normal spte to | |
2958 | * fix the access. | |
2959 | * | |
2960 | * See the comments in kvm_arch_commit_memory_region(). | |
2961 | */ | |
2962 | if (sp->role.level > PT_PAGE_TABLE_LEVEL) | |
2963 | goto exit; | |
2964 | ||
c7ba5b48 XG |
2965 | /* |
2966 | * Currently, fast page fault only works for direct mapping since | |
2967 | * the gfn is not stable for indirect shadow page. | |
2968 | * See Documentation/virtual/kvm/locking.txt to get more detail. | |
2969 | */ | |
92a476cb | 2970 | ret = fast_pf_fix_direct_spte(vcpu, sp, iterator.sptep, spte); |
c7ba5b48 | 2971 | exit: |
a72faf25 XG |
2972 | trace_fast_page_fault(vcpu, gva, error_code, iterator.sptep, |
2973 | spte, ret); | |
c7ba5b48 XG |
2974 | walk_shadow_page_lockless_end(vcpu); |
2975 | ||
2976 | return ret; | |
2977 | } | |
2978 | ||
78b2c54a | 2979 | static bool try_async_pf(struct kvm_vcpu *vcpu, bool prefault, gfn_t gfn, |
060c2abe | 2980 | gva_t gva, pfn_t *pfn, bool write, bool *writable); |
450e0b41 | 2981 | static void make_mmu_pages_available(struct kvm_vcpu *vcpu); |
060c2abe | 2982 | |
c7ba5b48 XG |
2983 | static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, u32 error_code, |
2984 | gfn_t gfn, bool prefault) | |
10589a46 MT |
2985 | { |
2986 | int r; | |
852e3c19 | 2987 | int level; |
936a5fe6 | 2988 | int force_pt_level; |
35149e21 | 2989 | pfn_t pfn; |
e930bffe | 2990 | unsigned long mmu_seq; |
c7ba5b48 | 2991 | bool map_writable, write = error_code & PFERR_WRITE_MASK; |
aaee2c94 | 2992 | |
936a5fe6 AA |
2993 | force_pt_level = mapping_level_dirty_bitmap(vcpu, gfn); |
2994 | if (likely(!force_pt_level)) { | |
2995 | level = mapping_level(vcpu, gfn); | |
2996 | /* | |
2997 | * This path builds a PAE pagetable - so we can map | |
2998 | * 2mb pages at maximum. Therefore check if the level | |
2999 | * is larger than that. | |
3000 | */ | |
3001 | if (level > PT_DIRECTORY_LEVEL) | |
3002 | level = PT_DIRECTORY_LEVEL; | |
852e3c19 | 3003 | |
936a5fe6 AA |
3004 | gfn &= ~(KVM_PAGES_PER_HPAGE(level) - 1); |
3005 | } else | |
3006 | level = PT_PAGE_TABLE_LEVEL; | |
05da4558 | 3007 | |
c7ba5b48 XG |
3008 | if (fast_page_fault(vcpu, v, level, error_code)) |
3009 | return 0; | |
3010 | ||
e930bffe | 3011 | mmu_seq = vcpu->kvm->mmu_notifier_seq; |
4c2155ce | 3012 | smp_rmb(); |
060c2abe | 3013 | |
78b2c54a | 3014 | if (try_async_pf(vcpu, prefault, gfn, v, &pfn, write, &map_writable)) |
060c2abe | 3015 | return 0; |
aaee2c94 | 3016 | |
d7c55201 XG |
3017 | if (handle_abnormal_pfn(vcpu, v, gfn, pfn, ACC_ALL, &r)) |
3018 | return r; | |
d196e343 | 3019 | |
aaee2c94 | 3020 | spin_lock(&vcpu->kvm->mmu_lock); |
8ca40a70 | 3021 | if (mmu_notifier_retry(vcpu->kvm, mmu_seq)) |
e930bffe | 3022 | goto out_unlock; |
450e0b41 | 3023 | make_mmu_pages_available(vcpu); |
936a5fe6 AA |
3024 | if (likely(!force_pt_level)) |
3025 | transparent_hugepage_adjust(vcpu, &gfn, &pfn, &level); | |
2ec4739d XG |
3026 | r = __direct_map(vcpu, v, write, map_writable, level, gfn, pfn, |
3027 | prefault); | |
aaee2c94 MT |
3028 | spin_unlock(&vcpu->kvm->mmu_lock); |
3029 | ||
aaee2c94 | 3030 | |
10589a46 | 3031 | return r; |
e930bffe AA |
3032 | |
3033 | out_unlock: | |
3034 | spin_unlock(&vcpu->kvm->mmu_lock); | |
3035 | kvm_release_pfn_clean(pfn); | |
3036 | return 0; | |
10589a46 MT |
3037 | } |
3038 | ||
3039 | ||
17ac10ad AK |
3040 | static void mmu_free_roots(struct kvm_vcpu *vcpu) |
3041 | { | |
3042 | int i; | |
4db35314 | 3043 | struct kvm_mmu_page *sp; |
d98ba053 | 3044 | LIST_HEAD(invalid_list); |
17ac10ad | 3045 | |
ad312c7c | 3046 | if (!VALID_PAGE(vcpu->arch.mmu.root_hpa)) |
7b53aa56 | 3047 | return; |
35af577a | 3048 | |
81407ca5 JR |
3049 | if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL && |
3050 | (vcpu->arch.mmu.root_level == PT64_ROOT_LEVEL || | |
3051 | vcpu->arch.mmu.direct_map)) { | |
ad312c7c | 3052 | hpa_t root = vcpu->arch.mmu.root_hpa; |
17ac10ad | 3053 | |
35af577a | 3054 | spin_lock(&vcpu->kvm->mmu_lock); |
4db35314 AK |
3055 | sp = page_header(root); |
3056 | --sp->root_count; | |
d98ba053 XG |
3057 | if (!sp->root_count && sp->role.invalid) { |
3058 | kvm_mmu_prepare_zap_page(vcpu->kvm, sp, &invalid_list); | |
3059 | kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list); | |
3060 | } | |
aaee2c94 | 3061 | spin_unlock(&vcpu->kvm->mmu_lock); |
35af577a | 3062 | vcpu->arch.mmu.root_hpa = INVALID_PAGE; |
17ac10ad AK |
3063 | return; |
3064 | } | |
35af577a GN |
3065 | |
3066 | spin_lock(&vcpu->kvm->mmu_lock); | |
17ac10ad | 3067 | for (i = 0; i < 4; ++i) { |
ad312c7c | 3068 | hpa_t root = vcpu->arch.mmu.pae_root[i]; |
17ac10ad | 3069 | |
417726a3 | 3070 | if (root) { |
417726a3 | 3071 | root &= PT64_BASE_ADDR_MASK; |
4db35314 AK |
3072 | sp = page_header(root); |
3073 | --sp->root_count; | |
2e53d63a | 3074 | if (!sp->root_count && sp->role.invalid) |
d98ba053 XG |
3075 | kvm_mmu_prepare_zap_page(vcpu->kvm, sp, |
3076 | &invalid_list); | |
417726a3 | 3077 | } |
ad312c7c | 3078 | vcpu->arch.mmu.pae_root[i] = INVALID_PAGE; |
17ac10ad | 3079 | } |
d98ba053 | 3080 | kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list); |
aaee2c94 | 3081 | spin_unlock(&vcpu->kvm->mmu_lock); |
ad312c7c | 3082 | vcpu->arch.mmu.root_hpa = INVALID_PAGE; |
17ac10ad AK |
3083 | } |
3084 | ||
8986ecc0 MT |
3085 | static int mmu_check_root(struct kvm_vcpu *vcpu, gfn_t root_gfn) |
3086 | { | |
3087 | int ret = 0; | |
3088 | ||
3089 | if (!kvm_is_visible_gfn(vcpu->kvm, root_gfn)) { | |
a8eeb04a | 3090 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); |
8986ecc0 MT |
3091 | ret = 1; |
3092 | } | |
3093 | ||
3094 | return ret; | |
3095 | } | |
3096 | ||
651dd37a JR |
3097 | static int mmu_alloc_direct_roots(struct kvm_vcpu *vcpu) |
3098 | { | |
3099 | struct kvm_mmu_page *sp; | |
7ebaf15e | 3100 | unsigned i; |
651dd37a JR |
3101 | |
3102 | if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) { | |
3103 | spin_lock(&vcpu->kvm->mmu_lock); | |
450e0b41 | 3104 | make_mmu_pages_available(vcpu); |
651dd37a JR |
3105 | sp = kvm_mmu_get_page(vcpu, 0, 0, PT64_ROOT_LEVEL, |
3106 | 1, ACC_ALL, NULL); | |
3107 | ++sp->root_count; | |
3108 | spin_unlock(&vcpu->kvm->mmu_lock); | |
3109 | vcpu->arch.mmu.root_hpa = __pa(sp->spt); | |
3110 | } else if (vcpu->arch.mmu.shadow_root_level == PT32E_ROOT_LEVEL) { | |
3111 | for (i = 0; i < 4; ++i) { | |
3112 | hpa_t root = vcpu->arch.mmu.pae_root[i]; | |
3113 | ||
fa4a2c08 | 3114 | MMU_WARN_ON(VALID_PAGE(root)); |
651dd37a | 3115 | spin_lock(&vcpu->kvm->mmu_lock); |
450e0b41 | 3116 | make_mmu_pages_available(vcpu); |
649497d1 AK |
3117 | sp = kvm_mmu_get_page(vcpu, i << (30 - PAGE_SHIFT), |
3118 | i << 30, | |
651dd37a JR |
3119 | PT32_ROOT_LEVEL, 1, ACC_ALL, |
3120 | NULL); | |
3121 | root = __pa(sp->spt); | |
3122 | ++sp->root_count; | |
3123 | spin_unlock(&vcpu->kvm->mmu_lock); | |
3124 | vcpu->arch.mmu.pae_root[i] = root | PT_PRESENT_MASK; | |
651dd37a | 3125 | } |
6292757f | 3126 | vcpu->arch.mmu.root_hpa = __pa(vcpu->arch.mmu.pae_root); |
651dd37a JR |
3127 | } else |
3128 | BUG(); | |
3129 | ||
3130 | return 0; | |
3131 | } | |
3132 | ||
3133 | static int mmu_alloc_shadow_roots(struct kvm_vcpu *vcpu) | |
17ac10ad | 3134 | { |
4db35314 | 3135 | struct kvm_mmu_page *sp; |
81407ca5 JR |
3136 | u64 pdptr, pm_mask; |
3137 | gfn_t root_gfn; | |
3138 | int i; | |
3bb65a22 | 3139 | |
5777ed34 | 3140 | root_gfn = vcpu->arch.mmu.get_cr3(vcpu) >> PAGE_SHIFT; |
17ac10ad | 3141 | |
651dd37a JR |
3142 | if (mmu_check_root(vcpu, root_gfn)) |
3143 | return 1; | |
3144 | ||
3145 | /* | |
3146 | * Do we shadow a long mode page table? If so we need to | |
3147 | * write-protect the guests page table root. | |
3148 | */ | |
3149 | if (vcpu->arch.mmu.root_level == PT64_ROOT_LEVEL) { | |
ad312c7c | 3150 | hpa_t root = vcpu->arch.mmu.root_hpa; |
17ac10ad | 3151 | |
fa4a2c08 | 3152 | MMU_WARN_ON(VALID_PAGE(root)); |
651dd37a | 3153 | |
8facbbff | 3154 | spin_lock(&vcpu->kvm->mmu_lock); |
450e0b41 | 3155 | make_mmu_pages_available(vcpu); |
651dd37a JR |
3156 | sp = kvm_mmu_get_page(vcpu, root_gfn, 0, PT64_ROOT_LEVEL, |
3157 | 0, ACC_ALL, NULL); | |
4db35314 AK |
3158 | root = __pa(sp->spt); |
3159 | ++sp->root_count; | |
8facbbff | 3160 | spin_unlock(&vcpu->kvm->mmu_lock); |
ad312c7c | 3161 | vcpu->arch.mmu.root_hpa = root; |
8986ecc0 | 3162 | return 0; |
17ac10ad | 3163 | } |
f87f9288 | 3164 | |
651dd37a JR |
3165 | /* |
3166 | * We shadow a 32 bit page table. This may be a legacy 2-level | |
81407ca5 JR |
3167 | * or a PAE 3-level page table. In either case we need to be aware that |
3168 | * the shadow page table may be a PAE or a long mode page table. | |
651dd37a | 3169 | */ |
81407ca5 JR |
3170 | pm_mask = PT_PRESENT_MASK; |
3171 | if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) | |
3172 | pm_mask |= PT_ACCESSED_MASK | PT_WRITABLE_MASK | PT_USER_MASK; | |
3173 | ||
17ac10ad | 3174 | for (i = 0; i < 4; ++i) { |
ad312c7c | 3175 | hpa_t root = vcpu->arch.mmu.pae_root[i]; |
17ac10ad | 3176 | |
fa4a2c08 | 3177 | MMU_WARN_ON(VALID_PAGE(root)); |
ad312c7c | 3178 | if (vcpu->arch.mmu.root_level == PT32E_ROOT_LEVEL) { |
e4e517b4 | 3179 | pdptr = vcpu->arch.mmu.get_pdptr(vcpu, i); |
43a3795a | 3180 | if (!is_present_gpte(pdptr)) { |
ad312c7c | 3181 | vcpu->arch.mmu.pae_root[i] = 0; |
417726a3 AK |
3182 | continue; |
3183 | } | |
6de4f3ad | 3184 | root_gfn = pdptr >> PAGE_SHIFT; |
f87f9288 JR |
3185 | if (mmu_check_root(vcpu, root_gfn)) |
3186 | return 1; | |
5a7388c2 | 3187 | } |
8facbbff | 3188 | spin_lock(&vcpu->kvm->mmu_lock); |
450e0b41 | 3189 | make_mmu_pages_available(vcpu); |
4db35314 | 3190 | sp = kvm_mmu_get_page(vcpu, root_gfn, i << 30, |
651dd37a | 3191 | PT32_ROOT_LEVEL, 0, |
f7d9c7b7 | 3192 | ACC_ALL, NULL); |
4db35314 AK |
3193 | root = __pa(sp->spt); |
3194 | ++sp->root_count; | |
8facbbff AK |
3195 | spin_unlock(&vcpu->kvm->mmu_lock); |
3196 | ||
81407ca5 | 3197 | vcpu->arch.mmu.pae_root[i] = root | pm_mask; |
17ac10ad | 3198 | } |
6292757f | 3199 | vcpu->arch.mmu.root_hpa = __pa(vcpu->arch.mmu.pae_root); |
81407ca5 JR |
3200 | |
3201 | /* | |
3202 | * If we shadow a 32 bit page table with a long mode page | |
3203 | * table we enter this path. | |
3204 | */ | |
3205 | if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) { | |
3206 | if (vcpu->arch.mmu.lm_root == NULL) { | |
3207 | /* | |
3208 | * The additional page necessary for this is only | |
3209 | * allocated on demand. | |
3210 | */ | |
3211 | ||
3212 | u64 *lm_root; | |
3213 | ||
3214 | lm_root = (void*)get_zeroed_page(GFP_KERNEL); | |
3215 | if (lm_root == NULL) | |
3216 | return 1; | |
3217 | ||
3218 | lm_root[0] = __pa(vcpu->arch.mmu.pae_root) | pm_mask; | |
3219 | ||
3220 | vcpu->arch.mmu.lm_root = lm_root; | |
3221 | } | |
3222 | ||
3223 | vcpu->arch.mmu.root_hpa = __pa(vcpu->arch.mmu.lm_root); | |
3224 | } | |
3225 | ||
8986ecc0 | 3226 | return 0; |
17ac10ad AK |
3227 | } |
3228 | ||
651dd37a JR |
3229 | static int mmu_alloc_roots(struct kvm_vcpu *vcpu) |
3230 | { | |
3231 | if (vcpu->arch.mmu.direct_map) | |
3232 | return mmu_alloc_direct_roots(vcpu); | |
3233 | else | |
3234 | return mmu_alloc_shadow_roots(vcpu); | |
3235 | } | |
3236 | ||
0ba73cda MT |
3237 | static void mmu_sync_roots(struct kvm_vcpu *vcpu) |
3238 | { | |
3239 | int i; | |
3240 | struct kvm_mmu_page *sp; | |
3241 | ||
81407ca5 JR |
3242 | if (vcpu->arch.mmu.direct_map) |
3243 | return; | |
3244 | ||
0ba73cda MT |
3245 | if (!VALID_PAGE(vcpu->arch.mmu.root_hpa)) |
3246 | return; | |
6903074c | 3247 | |
56f17dd3 | 3248 | vcpu_clear_mmio_info(vcpu, MMIO_GVA_ANY); |
0375f7fa | 3249 | kvm_mmu_audit(vcpu, AUDIT_PRE_SYNC); |
81407ca5 | 3250 | if (vcpu->arch.mmu.root_level == PT64_ROOT_LEVEL) { |
0ba73cda MT |
3251 | hpa_t root = vcpu->arch.mmu.root_hpa; |
3252 | sp = page_header(root); | |
3253 | mmu_sync_children(vcpu, sp); | |
0375f7fa | 3254 | kvm_mmu_audit(vcpu, AUDIT_POST_SYNC); |
0ba73cda MT |
3255 | return; |
3256 | } | |
3257 | for (i = 0; i < 4; ++i) { | |
3258 | hpa_t root = vcpu->arch.mmu.pae_root[i]; | |
3259 | ||
8986ecc0 | 3260 | if (root && VALID_PAGE(root)) { |
0ba73cda MT |
3261 | root &= PT64_BASE_ADDR_MASK; |
3262 | sp = page_header(root); | |
3263 | mmu_sync_children(vcpu, sp); | |
3264 | } | |
3265 | } | |
0375f7fa | 3266 | kvm_mmu_audit(vcpu, AUDIT_POST_SYNC); |
0ba73cda MT |
3267 | } |
3268 | ||
3269 | void kvm_mmu_sync_roots(struct kvm_vcpu *vcpu) | |
3270 | { | |
3271 | spin_lock(&vcpu->kvm->mmu_lock); | |
3272 | mmu_sync_roots(vcpu); | |
6cffe8ca | 3273 | spin_unlock(&vcpu->kvm->mmu_lock); |
0ba73cda | 3274 | } |
bfd0a56b | 3275 | EXPORT_SYMBOL_GPL(kvm_mmu_sync_roots); |
0ba73cda | 3276 | |
1871c602 | 3277 | static gpa_t nonpaging_gva_to_gpa(struct kvm_vcpu *vcpu, gva_t vaddr, |
ab9ae313 | 3278 | u32 access, struct x86_exception *exception) |
6aa8b732 | 3279 | { |
ab9ae313 AK |
3280 | if (exception) |
3281 | exception->error_code = 0; | |
6aa8b732 AK |
3282 | return vaddr; |
3283 | } | |
3284 | ||
6539e738 | 3285 | static gpa_t nonpaging_gva_to_gpa_nested(struct kvm_vcpu *vcpu, gva_t vaddr, |
ab9ae313 AK |
3286 | u32 access, |
3287 | struct x86_exception *exception) | |
6539e738 | 3288 | { |
ab9ae313 AK |
3289 | if (exception) |
3290 | exception->error_code = 0; | |
54987b7a | 3291 | return vcpu->arch.nested_mmu.translate_gpa(vcpu, vaddr, access, exception); |
6539e738 JR |
3292 | } |
3293 | ||
ce88decf XG |
3294 | static bool quickly_check_mmio_pf(struct kvm_vcpu *vcpu, u64 addr, bool direct) |
3295 | { | |
3296 | if (direct) | |
3297 | return vcpu_match_mmio_gpa(vcpu, addr); | |
3298 | ||
3299 | return vcpu_match_mmio_gva(vcpu, addr); | |
3300 | } | |
3301 | ||
3302 | ||
3303 | /* | |
3304 | * On direct hosts, the last spte is only allows two states | |
3305 | * for mmio page fault: | |
3306 | * - It is the mmio spte | |
3307 | * - It is zapped or it is being zapped. | |
3308 | * | |
3309 | * This function completely checks the spte when the last spte | |
3310 | * is not the mmio spte. | |
3311 | */ | |
3312 | static bool check_direct_spte_mmio_pf(u64 spte) | |
3313 | { | |
3314 | return __check_direct_spte_mmio_pf(spte); | |
3315 | } | |
3316 | ||
3317 | static u64 walk_shadow_page_get_mmio_spte(struct kvm_vcpu *vcpu, u64 addr) | |
3318 | { | |
3319 | struct kvm_shadow_walk_iterator iterator; | |
3320 | u64 spte = 0ull; | |
3321 | ||
37f6a4e2 MT |
3322 | if (!VALID_PAGE(vcpu->arch.mmu.root_hpa)) |
3323 | return spte; | |
3324 | ||
ce88decf XG |
3325 | walk_shadow_page_lockless_begin(vcpu); |
3326 | for_each_shadow_entry_lockless(vcpu, addr, iterator, spte) | |
3327 | if (!is_shadow_present_pte(spte)) | |
3328 | break; | |
3329 | walk_shadow_page_lockless_end(vcpu); | |
3330 | ||
3331 | return spte; | |
3332 | } | |
3333 | ||
ce88decf XG |
3334 | int handle_mmio_page_fault_common(struct kvm_vcpu *vcpu, u64 addr, bool direct) |
3335 | { | |
3336 | u64 spte; | |
3337 | ||
3338 | if (quickly_check_mmio_pf(vcpu, addr, direct)) | |
b37fbea6 | 3339 | return RET_MMIO_PF_EMULATE; |
ce88decf XG |
3340 | |
3341 | spte = walk_shadow_page_get_mmio_spte(vcpu, addr); | |
3342 | ||
3343 | if (is_mmio_spte(spte)) { | |
3344 | gfn_t gfn = get_mmio_spte_gfn(spte); | |
3345 | unsigned access = get_mmio_spte_access(spte); | |
3346 | ||
54bf36aa | 3347 | if (!check_mmio_spte(vcpu, spte)) |
f8f55942 XG |
3348 | return RET_MMIO_PF_INVALID; |
3349 | ||
ce88decf XG |
3350 | if (direct) |
3351 | addr = 0; | |
4f022648 XG |
3352 | |
3353 | trace_handle_mmio_page_fault(addr, gfn, access); | |
ce88decf | 3354 | vcpu_cache_mmio_info(vcpu, addr, gfn, access); |
b37fbea6 | 3355 | return RET_MMIO_PF_EMULATE; |
ce88decf XG |
3356 | } |
3357 | ||
3358 | /* | |
3359 | * It's ok if the gva is remapped by other cpus on shadow guest, | |
3360 | * it's a BUG if the gfn is not a mmio page. | |
3361 | */ | |
3362 | if (direct && !check_direct_spte_mmio_pf(spte)) | |
b37fbea6 | 3363 | return RET_MMIO_PF_BUG; |
ce88decf XG |
3364 | |
3365 | /* | |
3366 | * If the page table is zapped by other cpus, let CPU fault again on | |
3367 | * the address. | |
3368 | */ | |
b37fbea6 | 3369 | return RET_MMIO_PF_RETRY; |
ce88decf XG |
3370 | } |
3371 | EXPORT_SYMBOL_GPL(handle_mmio_page_fault_common); | |
3372 | ||
3373 | static int handle_mmio_page_fault(struct kvm_vcpu *vcpu, u64 addr, | |
3374 | u32 error_code, bool direct) | |
3375 | { | |
3376 | int ret; | |
3377 | ||
3378 | ret = handle_mmio_page_fault_common(vcpu, addr, direct); | |
b37fbea6 | 3379 | WARN_ON(ret == RET_MMIO_PF_BUG); |
ce88decf XG |
3380 | return ret; |
3381 | } | |
3382 | ||
6aa8b732 | 3383 | static int nonpaging_page_fault(struct kvm_vcpu *vcpu, gva_t gva, |
78b2c54a | 3384 | u32 error_code, bool prefault) |
6aa8b732 | 3385 | { |
e833240f | 3386 | gfn_t gfn; |
e2dec939 | 3387 | int r; |
6aa8b732 | 3388 | |
b8688d51 | 3389 | pgprintk("%s: gva %lx error %x\n", __func__, gva, error_code); |
ce88decf | 3390 | |
f8f55942 XG |
3391 | if (unlikely(error_code & PFERR_RSVD_MASK)) { |
3392 | r = handle_mmio_page_fault(vcpu, gva, error_code, true); | |
3393 | ||
3394 | if (likely(r != RET_MMIO_PF_INVALID)) | |
3395 | return r; | |
3396 | } | |
ce88decf | 3397 | |
e2dec939 AK |
3398 | r = mmu_topup_memory_caches(vcpu); |
3399 | if (r) | |
3400 | return r; | |
714b93da | 3401 | |
fa4a2c08 | 3402 | MMU_WARN_ON(!VALID_PAGE(vcpu->arch.mmu.root_hpa)); |
6aa8b732 | 3403 | |
e833240f | 3404 | gfn = gva >> PAGE_SHIFT; |
6aa8b732 | 3405 | |
e833240f | 3406 | return nonpaging_map(vcpu, gva & PAGE_MASK, |
c7ba5b48 | 3407 | error_code, gfn, prefault); |
6aa8b732 AK |
3408 | } |
3409 | ||
7e1fbeac | 3410 | static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, gfn_t gfn) |
af585b92 GN |
3411 | { |
3412 | struct kvm_arch_async_pf arch; | |
fb67e14f | 3413 | |
7c90705b | 3414 | arch.token = (vcpu->arch.apf.id++ << 12) | vcpu->vcpu_id; |
af585b92 | 3415 | arch.gfn = gfn; |
c4806acd | 3416 | arch.direct_map = vcpu->arch.mmu.direct_map; |
fb67e14f | 3417 | arch.cr3 = vcpu->arch.mmu.get_cr3(vcpu); |
af585b92 | 3418 | |
54bf36aa | 3419 | return kvm_setup_async_pf(vcpu, gva, kvm_vcpu_gfn_to_hva(vcpu, gfn), &arch); |
af585b92 GN |
3420 | } |
3421 | ||
3422 | static bool can_do_async_pf(struct kvm_vcpu *vcpu) | |
3423 | { | |
3424 | if (unlikely(!irqchip_in_kernel(vcpu->kvm) || | |
3425 | kvm_event_needs_reinjection(vcpu))) | |
3426 | return false; | |
3427 | ||
3428 | return kvm_x86_ops->interrupt_allowed(vcpu); | |
3429 | } | |
3430 | ||
78b2c54a | 3431 | static bool try_async_pf(struct kvm_vcpu *vcpu, bool prefault, gfn_t gfn, |
612819c3 | 3432 | gva_t gva, pfn_t *pfn, bool write, bool *writable) |
af585b92 | 3433 | { |
3520469d | 3434 | struct kvm_memory_slot *slot; |
af585b92 GN |
3435 | bool async; |
3436 | ||
54bf36aa | 3437 | slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); |
3520469d PB |
3438 | async = false; |
3439 | *pfn = __gfn_to_pfn_memslot(slot, gfn, false, &async, write, writable); | |
af585b92 GN |
3440 | if (!async) |
3441 | return false; /* *pfn has correct page already */ | |
3442 | ||
78b2c54a | 3443 | if (!prefault && can_do_async_pf(vcpu)) { |
c9b263d2 | 3444 | trace_kvm_try_async_get_page(gva, gfn); |
af585b92 GN |
3445 | if (kvm_find_async_pf_gfn(vcpu, gfn)) { |
3446 | trace_kvm_async_pf_doublefault(gva, gfn); | |
3447 | kvm_make_request(KVM_REQ_APF_HALT, vcpu); | |
3448 | return true; | |
3449 | } else if (kvm_arch_setup_async_pf(vcpu, gva, gfn)) | |
3450 | return true; | |
3451 | } | |
3452 | ||
3520469d | 3453 | *pfn = __gfn_to_pfn_memslot(slot, gfn, false, NULL, write, writable); |
af585b92 GN |
3454 | return false; |
3455 | } | |
3456 | ||
6a39bbc5 XG |
3457 | static bool |
3458 | check_hugepage_cache_consistency(struct kvm_vcpu *vcpu, gfn_t gfn, int level) | |
3459 | { | |
3460 | int page_num = KVM_PAGES_PER_HPAGE(level); | |
3461 | ||
3462 | gfn &= ~(page_num - 1); | |
3463 | ||
3464 | return kvm_mtrr_check_gfn_range_consistency(vcpu, gfn, page_num); | |
3465 | } | |
3466 | ||
56028d08 | 3467 | static int tdp_page_fault(struct kvm_vcpu *vcpu, gva_t gpa, u32 error_code, |
78b2c54a | 3468 | bool prefault) |
fb72d167 | 3469 | { |
35149e21 | 3470 | pfn_t pfn; |
fb72d167 | 3471 | int r; |
852e3c19 | 3472 | int level; |
936a5fe6 | 3473 | int force_pt_level; |
05da4558 | 3474 | gfn_t gfn = gpa >> PAGE_SHIFT; |
e930bffe | 3475 | unsigned long mmu_seq; |
612819c3 MT |
3476 | int write = error_code & PFERR_WRITE_MASK; |
3477 | bool map_writable; | |
fb72d167 | 3478 | |
fa4a2c08 | 3479 | MMU_WARN_ON(!VALID_PAGE(vcpu->arch.mmu.root_hpa)); |
fb72d167 | 3480 | |
f8f55942 XG |
3481 | if (unlikely(error_code & PFERR_RSVD_MASK)) { |
3482 | r = handle_mmio_page_fault(vcpu, gpa, error_code, true); | |
3483 | ||
3484 | if (likely(r != RET_MMIO_PF_INVALID)) | |
3485 | return r; | |
3486 | } | |
ce88decf | 3487 | |
fb72d167 JR |
3488 | r = mmu_topup_memory_caches(vcpu); |
3489 | if (r) | |
3490 | return r; | |
3491 | ||
6a39bbc5 XG |
3492 | if (mapping_level_dirty_bitmap(vcpu, gfn) || |
3493 | !check_hugepage_cache_consistency(vcpu, gfn, PT_DIRECTORY_LEVEL)) | |
3494 | force_pt_level = 1; | |
3495 | else | |
3496 | force_pt_level = 0; | |
3497 | ||
936a5fe6 AA |
3498 | if (likely(!force_pt_level)) { |
3499 | level = mapping_level(vcpu, gfn); | |
6a39bbc5 XG |
3500 | if (level > PT_DIRECTORY_LEVEL && |
3501 | !check_hugepage_cache_consistency(vcpu, gfn, level)) | |
3502 | level = PT_DIRECTORY_LEVEL; | |
936a5fe6 AA |
3503 | gfn &= ~(KVM_PAGES_PER_HPAGE(level) - 1); |
3504 | } else | |
3505 | level = PT_PAGE_TABLE_LEVEL; | |
852e3c19 | 3506 | |
c7ba5b48 XG |
3507 | if (fast_page_fault(vcpu, gpa, level, error_code)) |
3508 | return 0; | |
3509 | ||
e930bffe | 3510 | mmu_seq = vcpu->kvm->mmu_notifier_seq; |
4c2155ce | 3511 | smp_rmb(); |
af585b92 | 3512 | |
78b2c54a | 3513 | if (try_async_pf(vcpu, prefault, gfn, gpa, &pfn, write, &map_writable)) |
af585b92 GN |
3514 | return 0; |
3515 | ||
d7c55201 XG |
3516 | if (handle_abnormal_pfn(vcpu, 0, gfn, pfn, ACC_ALL, &r)) |
3517 | return r; | |
3518 | ||
fb72d167 | 3519 | spin_lock(&vcpu->kvm->mmu_lock); |
8ca40a70 | 3520 | if (mmu_notifier_retry(vcpu->kvm, mmu_seq)) |
e930bffe | 3521 | goto out_unlock; |
450e0b41 | 3522 | make_mmu_pages_available(vcpu); |
936a5fe6 AA |
3523 | if (likely(!force_pt_level)) |
3524 | transparent_hugepage_adjust(vcpu, &gfn, &pfn, &level); | |
612819c3 | 3525 | r = __direct_map(vcpu, gpa, write, map_writable, |
2ec4739d | 3526 | level, gfn, pfn, prefault); |
fb72d167 | 3527 | spin_unlock(&vcpu->kvm->mmu_lock); |
fb72d167 JR |
3528 | |
3529 | return r; | |
e930bffe AA |
3530 | |
3531 | out_unlock: | |
3532 | spin_unlock(&vcpu->kvm->mmu_lock); | |
3533 | kvm_release_pfn_clean(pfn); | |
3534 | return 0; | |
fb72d167 JR |
3535 | } |
3536 | ||
8a3c1a33 PB |
3537 | static void nonpaging_init_context(struct kvm_vcpu *vcpu, |
3538 | struct kvm_mmu *context) | |
6aa8b732 | 3539 | { |
6aa8b732 | 3540 | context->page_fault = nonpaging_page_fault; |
6aa8b732 | 3541 | context->gva_to_gpa = nonpaging_gva_to_gpa; |
e8bc217a | 3542 | context->sync_page = nonpaging_sync_page; |
a7052897 | 3543 | context->invlpg = nonpaging_invlpg; |
0f53b5b1 | 3544 | context->update_pte = nonpaging_update_pte; |
cea0f0e7 | 3545 | context->root_level = 0; |
6aa8b732 | 3546 | context->shadow_root_level = PT32E_ROOT_LEVEL; |
17c3ba9d | 3547 | context->root_hpa = INVALID_PAGE; |
c5a78f2b | 3548 | context->direct_map = true; |
2d48a985 | 3549 | context->nx = false; |
6aa8b732 AK |
3550 | } |
3551 | ||
d8d173da | 3552 | void kvm_mmu_new_cr3(struct kvm_vcpu *vcpu) |
6aa8b732 | 3553 | { |
cea0f0e7 | 3554 | mmu_free_roots(vcpu); |
6aa8b732 AK |
3555 | } |
3556 | ||
5777ed34 JR |
3557 | static unsigned long get_cr3(struct kvm_vcpu *vcpu) |
3558 | { | |
9f8fe504 | 3559 | return kvm_read_cr3(vcpu); |
5777ed34 JR |
3560 | } |
3561 | ||
6389ee94 AK |
3562 | static void inject_page_fault(struct kvm_vcpu *vcpu, |
3563 | struct x86_exception *fault) | |
6aa8b732 | 3564 | { |
6389ee94 | 3565 | vcpu->arch.mmu.inject_page_fault(vcpu, fault); |
6aa8b732 AK |
3566 | } |
3567 | ||
54bf36aa | 3568 | static bool sync_mmio_spte(struct kvm_vcpu *vcpu, u64 *sptep, gfn_t gfn, |
f2fd125d | 3569 | unsigned access, int *nr_present) |
ce88decf XG |
3570 | { |
3571 | if (unlikely(is_mmio_spte(*sptep))) { | |
3572 | if (gfn != get_mmio_spte_gfn(*sptep)) { | |
3573 | mmu_spte_clear_no_track(sptep); | |
3574 | return true; | |
3575 | } | |
3576 | ||
3577 | (*nr_present)++; | |
54bf36aa | 3578 | mark_mmio_spte(vcpu, sptep, gfn, access); |
ce88decf XG |
3579 | return true; |
3580 | } | |
3581 | ||
3582 | return false; | |
3583 | } | |
3584 | ||
6fd01b71 AK |
3585 | static inline bool is_last_gpte(struct kvm_mmu *mmu, unsigned level, unsigned gpte) |
3586 | { | |
3587 | unsigned index; | |
3588 | ||
3589 | index = level - 1; | |
3590 | index |= (gpte & PT_PAGE_SIZE_MASK) >> (PT_PAGE_SIZE_SHIFT - 2); | |
3591 | return mmu->last_pte_bitmap & (1 << index); | |
3592 | } | |
3593 | ||
37406aaa NHE |
3594 | #define PTTYPE_EPT 18 /* arbitrary */ |
3595 | #define PTTYPE PTTYPE_EPT | |
3596 | #include "paging_tmpl.h" | |
3597 | #undef PTTYPE | |
3598 | ||
6aa8b732 AK |
3599 | #define PTTYPE 64 |
3600 | #include "paging_tmpl.h" | |
3601 | #undef PTTYPE | |
3602 | ||
3603 | #define PTTYPE 32 | |
3604 | #include "paging_tmpl.h" | |
3605 | #undef PTTYPE | |
3606 | ||
52fde8df | 3607 | static void reset_rsvds_bits_mask(struct kvm_vcpu *vcpu, |
4d6931c3 | 3608 | struct kvm_mmu *context) |
82725b20 | 3609 | { |
82725b20 DE |
3610 | int maxphyaddr = cpuid_maxphyaddr(vcpu); |
3611 | u64 exb_bit_rsvd = 0; | |
5f7dde7b | 3612 | u64 gbpages_bit_rsvd = 0; |
a0c0feb5 | 3613 | u64 nonleaf_bit8_rsvd = 0; |
82725b20 | 3614 | |
25d92081 YZ |
3615 | context->bad_mt_xwr = 0; |
3616 | ||
2d48a985 | 3617 | if (!context->nx) |
82725b20 | 3618 | exb_bit_rsvd = rsvd_bits(63, 63); |
5f7dde7b NA |
3619 | if (!guest_cpuid_has_gbpages(vcpu)) |
3620 | gbpages_bit_rsvd = rsvd_bits(7, 7); | |
a0c0feb5 PB |
3621 | |
3622 | /* | |
3623 | * Non-leaf PML4Es and PDPEs reserve bit 8 (which would be the G bit for | |
3624 | * leaf entries) on AMD CPUs only. | |
3625 | */ | |
3626 | if (guest_cpuid_is_amd(vcpu)) | |
3627 | nonleaf_bit8_rsvd = rsvd_bits(8, 8); | |
3628 | ||
4d6931c3 | 3629 | switch (context->root_level) { |
82725b20 DE |
3630 | case PT32_ROOT_LEVEL: |
3631 | /* no rsvd bits for 2 level 4K page table entries */ | |
3632 | context->rsvd_bits_mask[0][1] = 0; | |
3633 | context->rsvd_bits_mask[0][0] = 0; | |
f815bce8 XG |
3634 | context->rsvd_bits_mask[1][0] = context->rsvd_bits_mask[0][0]; |
3635 | ||
3636 | if (!is_pse(vcpu)) { | |
3637 | context->rsvd_bits_mask[1][1] = 0; | |
3638 | break; | |
3639 | } | |
3640 | ||
82725b20 DE |
3641 | if (is_cpuid_PSE36()) |
3642 | /* 36bits PSE 4MB page */ | |
3643 | context->rsvd_bits_mask[1][1] = rsvd_bits(17, 21); | |
3644 | else | |
3645 | /* 32 bits PSE 4MB page */ | |
3646 | context->rsvd_bits_mask[1][1] = rsvd_bits(13, 21); | |
82725b20 DE |
3647 | break; |
3648 | case PT32E_ROOT_LEVEL: | |
20c466b5 DE |
3649 | context->rsvd_bits_mask[0][2] = |
3650 | rsvd_bits(maxphyaddr, 63) | | |
cd9ae5fe | 3651 | rsvd_bits(5, 8) | rsvd_bits(1, 2); /* PDPTE */ |
82725b20 | 3652 | context->rsvd_bits_mask[0][1] = exb_bit_rsvd | |
4c26b4cd | 3653 | rsvd_bits(maxphyaddr, 62); /* PDE */ |
82725b20 DE |
3654 | context->rsvd_bits_mask[0][0] = exb_bit_rsvd | |
3655 | rsvd_bits(maxphyaddr, 62); /* PTE */ | |
3656 | context->rsvd_bits_mask[1][1] = exb_bit_rsvd | | |
3657 | rsvd_bits(maxphyaddr, 62) | | |
3658 | rsvd_bits(13, 20); /* large page */ | |
f815bce8 | 3659 | context->rsvd_bits_mask[1][0] = context->rsvd_bits_mask[0][0]; |
82725b20 DE |
3660 | break; |
3661 | case PT64_ROOT_LEVEL: | |
3662 | context->rsvd_bits_mask[0][3] = exb_bit_rsvd | | |
a0c0feb5 | 3663 | nonleaf_bit8_rsvd | rsvd_bits(7, 7) | rsvd_bits(maxphyaddr, 51); |
82725b20 | 3664 | context->rsvd_bits_mask[0][2] = exb_bit_rsvd | |
a0c0feb5 | 3665 | nonleaf_bit8_rsvd | gbpages_bit_rsvd | rsvd_bits(maxphyaddr, 51); |
82725b20 | 3666 | context->rsvd_bits_mask[0][1] = exb_bit_rsvd | |
4c26b4cd | 3667 | rsvd_bits(maxphyaddr, 51); |
82725b20 DE |
3668 | context->rsvd_bits_mask[0][0] = exb_bit_rsvd | |
3669 | rsvd_bits(maxphyaddr, 51); | |
3670 | context->rsvd_bits_mask[1][3] = context->rsvd_bits_mask[0][3]; | |
e04da980 | 3671 | context->rsvd_bits_mask[1][2] = exb_bit_rsvd | |
5f7dde7b | 3672 | gbpages_bit_rsvd | rsvd_bits(maxphyaddr, 51) | |
e04da980 | 3673 | rsvd_bits(13, 29); |
82725b20 | 3674 | context->rsvd_bits_mask[1][1] = exb_bit_rsvd | |
4c26b4cd SY |
3675 | rsvd_bits(maxphyaddr, 51) | |
3676 | rsvd_bits(13, 20); /* large page */ | |
f815bce8 | 3677 | context->rsvd_bits_mask[1][0] = context->rsvd_bits_mask[0][0]; |
82725b20 DE |
3678 | break; |
3679 | } | |
3680 | } | |
3681 | ||
25d92081 YZ |
3682 | static void reset_rsvds_bits_mask_ept(struct kvm_vcpu *vcpu, |
3683 | struct kvm_mmu *context, bool execonly) | |
3684 | { | |
3685 | int maxphyaddr = cpuid_maxphyaddr(vcpu); | |
3686 | int pte; | |
3687 | ||
3688 | context->rsvd_bits_mask[0][3] = | |
3689 | rsvd_bits(maxphyaddr, 51) | rsvd_bits(3, 7); | |
3690 | context->rsvd_bits_mask[0][2] = | |
3691 | rsvd_bits(maxphyaddr, 51) | rsvd_bits(3, 6); | |
3692 | context->rsvd_bits_mask[0][1] = | |
3693 | rsvd_bits(maxphyaddr, 51) | rsvd_bits(3, 6); | |
3694 | context->rsvd_bits_mask[0][0] = rsvd_bits(maxphyaddr, 51); | |
3695 | ||
3696 | /* large page */ | |
3697 | context->rsvd_bits_mask[1][3] = context->rsvd_bits_mask[0][3]; | |
3698 | context->rsvd_bits_mask[1][2] = | |
3699 | rsvd_bits(maxphyaddr, 51) | rsvd_bits(12, 29); | |
3700 | context->rsvd_bits_mask[1][1] = | |
3701 | rsvd_bits(maxphyaddr, 51) | rsvd_bits(12, 20); | |
3702 | context->rsvd_bits_mask[1][0] = context->rsvd_bits_mask[0][0]; | |
3703 | ||
3704 | for (pte = 0; pte < 64; pte++) { | |
3705 | int rwx_bits = pte & 7; | |
3706 | int mt = pte >> 3; | |
3707 | if (mt == 0x2 || mt == 0x3 || mt == 0x7 || | |
3708 | rwx_bits == 0x2 || rwx_bits == 0x6 || | |
3709 | (rwx_bits == 0x4 && !execonly)) | |
3710 | context->bad_mt_xwr |= (1ull << pte); | |
3711 | } | |
3712 | } | |
3713 | ||
edc90b7d XG |
3714 | static void update_permission_bitmask(struct kvm_vcpu *vcpu, |
3715 | struct kvm_mmu *mmu, bool ept) | |
97d64b78 AK |
3716 | { |
3717 | unsigned bit, byte, pfec; | |
3718 | u8 map; | |
66386ade | 3719 | bool fault, x, w, u, wf, uf, ff, smapf, cr4_smap, cr4_smep, smap = 0; |
97d64b78 | 3720 | |
66386ade | 3721 | cr4_smep = kvm_read_cr4_bits(vcpu, X86_CR4_SMEP); |
97ec8c06 | 3722 | cr4_smap = kvm_read_cr4_bits(vcpu, X86_CR4_SMAP); |
97d64b78 AK |
3723 | for (byte = 0; byte < ARRAY_SIZE(mmu->permissions); ++byte) { |
3724 | pfec = byte << 1; | |
3725 | map = 0; | |
3726 | wf = pfec & PFERR_WRITE_MASK; | |
3727 | uf = pfec & PFERR_USER_MASK; | |
3728 | ff = pfec & PFERR_FETCH_MASK; | |
97ec8c06 FW |
3729 | /* |
3730 | * PFERR_RSVD_MASK bit is set in PFEC if the access is not | |
3731 | * subject to SMAP restrictions, and cleared otherwise. The | |
3732 | * bit is only meaningful if the SMAP bit is set in CR4. | |
3733 | */ | |
3734 | smapf = !(pfec & PFERR_RSVD_MASK); | |
97d64b78 AK |
3735 | for (bit = 0; bit < 8; ++bit) { |
3736 | x = bit & ACC_EXEC_MASK; | |
3737 | w = bit & ACC_WRITE_MASK; | |
3738 | u = bit & ACC_USER_MASK; | |
3739 | ||
25d92081 YZ |
3740 | if (!ept) { |
3741 | /* Not really needed: !nx will cause pte.nx to fault */ | |
3742 | x |= !mmu->nx; | |
3743 | /* Allow supervisor writes if !cr0.wp */ | |
3744 | w |= !is_write_protection(vcpu) && !uf; | |
3745 | /* Disallow supervisor fetches of user code if cr4.smep */ | |
66386ade | 3746 | x &= !(cr4_smep && u && !uf); |
97ec8c06 FW |
3747 | |
3748 | /* | |
3749 | * SMAP:kernel-mode data accesses from user-mode | |
3750 | * mappings should fault. A fault is considered | |
3751 | * as a SMAP violation if all of the following | |
3752 | * conditions are ture: | |
3753 | * - X86_CR4_SMAP is set in CR4 | |
3754 | * - An user page is accessed | |
3755 | * - Page fault in kernel mode | |
3756 | * - if CPL = 3 or X86_EFLAGS_AC is clear | |
3757 | * | |
3758 | * Here, we cover the first three conditions. | |
3759 | * The fourth is computed dynamically in | |
3760 | * permission_fault() and is in smapf. | |
3761 | * | |
3762 | * Also, SMAP does not affect instruction | |
3763 | * fetches, add the !ff check here to make it | |
3764 | * clearer. | |
3765 | */ | |
3766 | smap = cr4_smap && u && !uf && !ff; | |
25d92081 YZ |
3767 | } else |
3768 | /* Not really needed: no U/S accesses on ept */ | |
3769 | u = 1; | |
97d64b78 | 3770 | |
97ec8c06 FW |
3771 | fault = (ff && !x) || (uf && !u) || (wf && !w) || |
3772 | (smapf && smap); | |
97d64b78 AK |
3773 | map |= fault << bit; |
3774 | } | |
3775 | mmu->permissions[byte] = map; | |
3776 | } | |
3777 | } | |
3778 | ||
6fd01b71 AK |
3779 | static void update_last_pte_bitmap(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu) |
3780 | { | |
3781 | u8 map; | |
3782 | unsigned level, root_level = mmu->root_level; | |
3783 | const unsigned ps_set_index = 1 << 2; /* bit 2 of index: ps */ | |
3784 | ||
3785 | if (root_level == PT32E_ROOT_LEVEL) | |
3786 | --root_level; | |
3787 | /* PT_PAGE_TABLE_LEVEL always terminates */ | |
3788 | map = 1 | (1 << ps_set_index); | |
3789 | for (level = PT_DIRECTORY_LEVEL; level <= root_level; ++level) { | |
3790 | if (level <= PT_PDPE_LEVEL | |
3791 | && (mmu->root_level >= PT32E_ROOT_LEVEL || is_pse(vcpu))) | |
3792 | map |= 1 << (ps_set_index | (level - 1)); | |
3793 | } | |
3794 | mmu->last_pte_bitmap = map; | |
3795 | } | |
3796 | ||
8a3c1a33 PB |
3797 | static void paging64_init_context_common(struct kvm_vcpu *vcpu, |
3798 | struct kvm_mmu *context, | |
3799 | int level) | |
6aa8b732 | 3800 | { |
2d48a985 | 3801 | context->nx = is_nx(vcpu); |
4d6931c3 | 3802 | context->root_level = level; |
2d48a985 | 3803 | |
4d6931c3 | 3804 | reset_rsvds_bits_mask(vcpu, context); |
25d92081 | 3805 | update_permission_bitmask(vcpu, context, false); |
6fd01b71 | 3806 | update_last_pte_bitmap(vcpu, context); |
6aa8b732 | 3807 | |
fa4a2c08 | 3808 | MMU_WARN_ON(!is_pae(vcpu)); |
6aa8b732 | 3809 | context->page_fault = paging64_page_fault; |
6aa8b732 | 3810 | context->gva_to_gpa = paging64_gva_to_gpa; |
e8bc217a | 3811 | context->sync_page = paging64_sync_page; |
a7052897 | 3812 | context->invlpg = paging64_invlpg; |
0f53b5b1 | 3813 | context->update_pte = paging64_update_pte; |
17ac10ad | 3814 | context->shadow_root_level = level; |
17c3ba9d | 3815 | context->root_hpa = INVALID_PAGE; |
c5a78f2b | 3816 | context->direct_map = false; |
6aa8b732 AK |
3817 | } |
3818 | ||
8a3c1a33 PB |
3819 | static void paging64_init_context(struct kvm_vcpu *vcpu, |
3820 | struct kvm_mmu *context) | |
17ac10ad | 3821 | { |
8a3c1a33 | 3822 | paging64_init_context_common(vcpu, context, PT64_ROOT_LEVEL); |
17ac10ad AK |
3823 | } |
3824 | ||
8a3c1a33 PB |
3825 | static void paging32_init_context(struct kvm_vcpu *vcpu, |
3826 | struct kvm_mmu *context) | |
6aa8b732 | 3827 | { |
2d48a985 | 3828 | context->nx = false; |
4d6931c3 | 3829 | context->root_level = PT32_ROOT_LEVEL; |
2d48a985 | 3830 | |
4d6931c3 | 3831 | reset_rsvds_bits_mask(vcpu, context); |
25d92081 | 3832 | update_permission_bitmask(vcpu, context, false); |
6fd01b71 | 3833 | update_last_pte_bitmap(vcpu, context); |
6aa8b732 | 3834 | |
6aa8b732 | 3835 | context->page_fault = paging32_page_fault; |
6aa8b732 | 3836 | context->gva_to_gpa = paging32_gva_to_gpa; |
e8bc217a | 3837 | context->sync_page = paging32_sync_page; |
a7052897 | 3838 | context->invlpg = paging32_invlpg; |
0f53b5b1 | 3839 | context->update_pte = paging32_update_pte; |
6aa8b732 | 3840 | context->shadow_root_level = PT32E_ROOT_LEVEL; |
17c3ba9d | 3841 | context->root_hpa = INVALID_PAGE; |
c5a78f2b | 3842 | context->direct_map = false; |
6aa8b732 AK |
3843 | } |
3844 | ||
8a3c1a33 PB |
3845 | static void paging32E_init_context(struct kvm_vcpu *vcpu, |
3846 | struct kvm_mmu *context) | |
6aa8b732 | 3847 | { |
8a3c1a33 | 3848 | paging64_init_context_common(vcpu, context, PT32E_ROOT_LEVEL); |
6aa8b732 AK |
3849 | } |
3850 | ||
8a3c1a33 | 3851 | static void init_kvm_tdp_mmu(struct kvm_vcpu *vcpu) |
fb72d167 | 3852 | { |
ad896af0 | 3853 | struct kvm_mmu *context = &vcpu->arch.mmu; |
fb72d167 | 3854 | |
c445f8ef | 3855 | context->base_role.word = 0; |
699023e2 | 3856 | context->base_role.smm = is_smm(vcpu); |
fb72d167 | 3857 | context->page_fault = tdp_page_fault; |
e8bc217a | 3858 | context->sync_page = nonpaging_sync_page; |
a7052897 | 3859 | context->invlpg = nonpaging_invlpg; |
0f53b5b1 | 3860 | context->update_pte = nonpaging_update_pte; |
67253af5 | 3861 | context->shadow_root_level = kvm_x86_ops->get_tdp_level(); |
fb72d167 | 3862 | context->root_hpa = INVALID_PAGE; |
c5a78f2b | 3863 | context->direct_map = true; |
1c97f0a0 | 3864 | context->set_cr3 = kvm_x86_ops->set_tdp_cr3; |
5777ed34 | 3865 | context->get_cr3 = get_cr3; |
e4e517b4 | 3866 | context->get_pdptr = kvm_pdptr_read; |
cb659db8 | 3867 | context->inject_page_fault = kvm_inject_page_fault; |
fb72d167 JR |
3868 | |
3869 | if (!is_paging(vcpu)) { | |
2d48a985 | 3870 | context->nx = false; |
fb72d167 JR |
3871 | context->gva_to_gpa = nonpaging_gva_to_gpa; |
3872 | context->root_level = 0; | |
3873 | } else if (is_long_mode(vcpu)) { | |
2d48a985 | 3874 | context->nx = is_nx(vcpu); |
fb72d167 | 3875 | context->root_level = PT64_ROOT_LEVEL; |
4d6931c3 DB |
3876 | reset_rsvds_bits_mask(vcpu, context); |
3877 | context->gva_to_gpa = paging64_gva_to_gpa; | |
fb72d167 | 3878 | } else if (is_pae(vcpu)) { |
2d48a985 | 3879 | context->nx = is_nx(vcpu); |
fb72d167 | 3880 | context->root_level = PT32E_ROOT_LEVEL; |
4d6931c3 DB |
3881 | reset_rsvds_bits_mask(vcpu, context); |
3882 | context->gva_to_gpa = paging64_gva_to_gpa; | |
fb72d167 | 3883 | } else { |
2d48a985 | 3884 | context->nx = false; |
fb72d167 | 3885 | context->root_level = PT32_ROOT_LEVEL; |
4d6931c3 DB |
3886 | reset_rsvds_bits_mask(vcpu, context); |
3887 | context->gva_to_gpa = paging32_gva_to_gpa; | |
fb72d167 JR |
3888 | } |
3889 | ||
25d92081 | 3890 | update_permission_bitmask(vcpu, context, false); |
6fd01b71 | 3891 | update_last_pte_bitmap(vcpu, context); |
fb72d167 JR |
3892 | } |
3893 | ||
ad896af0 | 3894 | void kvm_init_shadow_mmu(struct kvm_vcpu *vcpu) |
6aa8b732 | 3895 | { |
411c588d | 3896 | bool smep = kvm_read_cr4_bits(vcpu, X86_CR4_SMEP); |
edc90b7d | 3897 | bool smap = kvm_read_cr4_bits(vcpu, X86_CR4_SMAP); |
ad896af0 PB |
3898 | struct kvm_mmu *context = &vcpu->arch.mmu; |
3899 | ||
fa4a2c08 | 3900 | MMU_WARN_ON(VALID_PAGE(context->root_hpa)); |
6aa8b732 AK |
3901 | |
3902 | if (!is_paging(vcpu)) | |
8a3c1a33 | 3903 | nonpaging_init_context(vcpu, context); |
a9058ecd | 3904 | else if (is_long_mode(vcpu)) |
8a3c1a33 | 3905 | paging64_init_context(vcpu, context); |
6aa8b732 | 3906 | else if (is_pae(vcpu)) |
8a3c1a33 | 3907 | paging32E_init_context(vcpu, context); |
6aa8b732 | 3908 | else |
8a3c1a33 | 3909 | paging32_init_context(vcpu, context); |
a770f6f2 | 3910 | |
ad896af0 PB |
3911 | context->base_role.nxe = is_nx(vcpu); |
3912 | context->base_role.cr4_pae = !!is_pae(vcpu); | |
3913 | context->base_role.cr0_wp = is_write_protection(vcpu); | |
3914 | context->base_role.smep_andnot_wp | |
411c588d | 3915 | = smep && !is_write_protection(vcpu); |
edc90b7d XG |
3916 | context->base_role.smap_andnot_wp |
3917 | = smap && !is_write_protection(vcpu); | |
699023e2 | 3918 | context->base_role.smm = is_smm(vcpu); |
52fde8df JR |
3919 | } |
3920 | EXPORT_SYMBOL_GPL(kvm_init_shadow_mmu); | |
3921 | ||
ad896af0 | 3922 | void kvm_init_shadow_ept_mmu(struct kvm_vcpu *vcpu, bool execonly) |
155a97a3 | 3923 | { |
ad896af0 PB |
3924 | struct kvm_mmu *context = &vcpu->arch.mmu; |
3925 | ||
fa4a2c08 | 3926 | MMU_WARN_ON(VALID_PAGE(context->root_hpa)); |
155a97a3 NHE |
3927 | |
3928 | context->shadow_root_level = kvm_x86_ops->get_tdp_level(); | |
3929 | ||
3930 | context->nx = true; | |
155a97a3 NHE |
3931 | context->page_fault = ept_page_fault; |
3932 | context->gva_to_gpa = ept_gva_to_gpa; | |
3933 | context->sync_page = ept_sync_page; | |
3934 | context->invlpg = ept_invlpg; | |
3935 | context->update_pte = ept_update_pte; | |
155a97a3 NHE |
3936 | context->root_level = context->shadow_root_level; |
3937 | context->root_hpa = INVALID_PAGE; | |
3938 | context->direct_map = false; | |
3939 | ||
3940 | update_permission_bitmask(vcpu, context, true); | |
3941 | reset_rsvds_bits_mask_ept(vcpu, context, execonly); | |
155a97a3 NHE |
3942 | } |
3943 | EXPORT_SYMBOL_GPL(kvm_init_shadow_ept_mmu); | |
3944 | ||
8a3c1a33 | 3945 | static void init_kvm_softmmu(struct kvm_vcpu *vcpu) |
52fde8df | 3946 | { |
ad896af0 PB |
3947 | struct kvm_mmu *context = &vcpu->arch.mmu; |
3948 | ||
3949 | kvm_init_shadow_mmu(vcpu); | |
3950 | context->set_cr3 = kvm_x86_ops->set_cr3; | |
3951 | context->get_cr3 = get_cr3; | |
3952 | context->get_pdptr = kvm_pdptr_read; | |
3953 | context->inject_page_fault = kvm_inject_page_fault; | |
6aa8b732 AK |
3954 | } |
3955 | ||
8a3c1a33 | 3956 | static void init_kvm_nested_mmu(struct kvm_vcpu *vcpu) |
02f59dc9 JR |
3957 | { |
3958 | struct kvm_mmu *g_context = &vcpu->arch.nested_mmu; | |
3959 | ||
3960 | g_context->get_cr3 = get_cr3; | |
e4e517b4 | 3961 | g_context->get_pdptr = kvm_pdptr_read; |
02f59dc9 JR |
3962 | g_context->inject_page_fault = kvm_inject_page_fault; |
3963 | ||
3964 | /* | |
3965 | * Note that arch.mmu.gva_to_gpa translates l2_gva to l1_gpa. The | |
3966 | * translation of l2_gpa to l1_gpa addresses is done using the | |
3967 | * arch.nested_mmu.gva_to_gpa function. Basically the gva_to_gpa | |
3968 | * functions between mmu and nested_mmu are swapped. | |
3969 | */ | |
3970 | if (!is_paging(vcpu)) { | |
2d48a985 | 3971 | g_context->nx = false; |
02f59dc9 JR |
3972 | g_context->root_level = 0; |
3973 | g_context->gva_to_gpa = nonpaging_gva_to_gpa_nested; | |
3974 | } else if (is_long_mode(vcpu)) { | |
2d48a985 | 3975 | g_context->nx = is_nx(vcpu); |
02f59dc9 | 3976 | g_context->root_level = PT64_ROOT_LEVEL; |
4d6931c3 | 3977 | reset_rsvds_bits_mask(vcpu, g_context); |
02f59dc9 JR |
3978 | g_context->gva_to_gpa = paging64_gva_to_gpa_nested; |
3979 | } else if (is_pae(vcpu)) { | |
2d48a985 | 3980 | g_context->nx = is_nx(vcpu); |
02f59dc9 | 3981 | g_context->root_level = PT32E_ROOT_LEVEL; |
4d6931c3 | 3982 | reset_rsvds_bits_mask(vcpu, g_context); |
02f59dc9 JR |
3983 | g_context->gva_to_gpa = paging64_gva_to_gpa_nested; |
3984 | } else { | |
2d48a985 | 3985 | g_context->nx = false; |
02f59dc9 | 3986 | g_context->root_level = PT32_ROOT_LEVEL; |
4d6931c3 | 3987 | reset_rsvds_bits_mask(vcpu, g_context); |
02f59dc9 JR |
3988 | g_context->gva_to_gpa = paging32_gva_to_gpa_nested; |
3989 | } | |
3990 | ||
25d92081 | 3991 | update_permission_bitmask(vcpu, g_context, false); |
6fd01b71 | 3992 | update_last_pte_bitmap(vcpu, g_context); |
02f59dc9 JR |
3993 | } |
3994 | ||
8a3c1a33 | 3995 | static void init_kvm_mmu(struct kvm_vcpu *vcpu) |
fb72d167 | 3996 | { |
02f59dc9 | 3997 | if (mmu_is_nested(vcpu)) |
e0c6db3e | 3998 | init_kvm_nested_mmu(vcpu); |
02f59dc9 | 3999 | else if (tdp_enabled) |
e0c6db3e | 4000 | init_kvm_tdp_mmu(vcpu); |
fb72d167 | 4001 | else |
e0c6db3e | 4002 | init_kvm_softmmu(vcpu); |
fb72d167 JR |
4003 | } |
4004 | ||
8a3c1a33 | 4005 | void kvm_mmu_reset_context(struct kvm_vcpu *vcpu) |
6aa8b732 | 4006 | { |
95f93af4 | 4007 | kvm_mmu_unload(vcpu); |
8a3c1a33 | 4008 | init_kvm_mmu(vcpu); |
17c3ba9d | 4009 | } |
8668a3c4 | 4010 | EXPORT_SYMBOL_GPL(kvm_mmu_reset_context); |
17c3ba9d AK |
4011 | |
4012 | int kvm_mmu_load(struct kvm_vcpu *vcpu) | |
6aa8b732 | 4013 | { |
714b93da AK |
4014 | int r; |
4015 | ||
e2dec939 | 4016 | r = mmu_topup_memory_caches(vcpu); |
17c3ba9d AK |
4017 | if (r) |
4018 | goto out; | |
8986ecc0 | 4019 | r = mmu_alloc_roots(vcpu); |
e2858b4a | 4020 | kvm_mmu_sync_roots(vcpu); |
8986ecc0 MT |
4021 | if (r) |
4022 | goto out; | |
3662cb1c | 4023 | /* set_cr3() should ensure TLB has been flushed */ |
f43addd4 | 4024 | vcpu->arch.mmu.set_cr3(vcpu, vcpu->arch.mmu.root_hpa); |
714b93da AK |
4025 | out: |
4026 | return r; | |
6aa8b732 | 4027 | } |
17c3ba9d AK |
4028 | EXPORT_SYMBOL_GPL(kvm_mmu_load); |
4029 | ||
4030 | void kvm_mmu_unload(struct kvm_vcpu *vcpu) | |
4031 | { | |
4032 | mmu_free_roots(vcpu); | |
95f93af4 | 4033 | WARN_ON(VALID_PAGE(vcpu->arch.mmu.root_hpa)); |
17c3ba9d | 4034 | } |
4b16184c | 4035 | EXPORT_SYMBOL_GPL(kvm_mmu_unload); |
6aa8b732 | 4036 | |
0028425f | 4037 | static void mmu_pte_write_new_pte(struct kvm_vcpu *vcpu, |
7c562522 XG |
4038 | struct kvm_mmu_page *sp, u64 *spte, |
4039 | const void *new) | |
0028425f | 4040 | { |
30945387 | 4041 | if (sp->role.level != PT_PAGE_TABLE_LEVEL) { |
7e4e4056 JR |
4042 | ++vcpu->kvm->stat.mmu_pde_zapped; |
4043 | return; | |
30945387 | 4044 | } |
0028425f | 4045 | |
4cee5764 | 4046 | ++vcpu->kvm->stat.mmu_pte_updated; |
7c562522 | 4047 | vcpu->arch.mmu.update_pte(vcpu, sp, spte, new); |
0028425f AK |
4048 | } |
4049 | ||
79539cec AK |
4050 | static bool need_remote_flush(u64 old, u64 new) |
4051 | { | |
4052 | if (!is_shadow_present_pte(old)) | |
4053 | return false; | |
4054 | if (!is_shadow_present_pte(new)) | |
4055 | return true; | |
4056 | if ((old ^ new) & PT64_BASE_ADDR_MASK) | |
4057 | return true; | |
53166229 GN |
4058 | old ^= shadow_nx_mask; |
4059 | new ^= shadow_nx_mask; | |
79539cec AK |
4060 | return (old & ~new & PT64_PERM_MASK) != 0; |
4061 | } | |
4062 | ||
0671a8e7 XG |
4063 | static void mmu_pte_write_flush_tlb(struct kvm_vcpu *vcpu, bool zap_page, |
4064 | bool remote_flush, bool local_flush) | |
79539cec | 4065 | { |
0671a8e7 XG |
4066 | if (zap_page) |
4067 | return; | |
4068 | ||
4069 | if (remote_flush) | |
79539cec | 4070 | kvm_flush_remote_tlbs(vcpu->kvm); |
0671a8e7 | 4071 | else if (local_flush) |
77c3913b | 4072 | kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); |
79539cec AK |
4073 | } |
4074 | ||
889e5cbc XG |
4075 | static u64 mmu_pte_write_fetch_gpte(struct kvm_vcpu *vcpu, gpa_t *gpa, |
4076 | const u8 *new, int *bytes) | |
da4a00f0 | 4077 | { |
889e5cbc XG |
4078 | u64 gentry; |
4079 | int r; | |
72016f3a | 4080 | |
72016f3a AK |
4081 | /* |
4082 | * Assume that the pte write on a page table of the same type | |
49b26e26 XG |
4083 | * as the current vcpu paging mode since we update the sptes only |
4084 | * when they have the same mode. | |
72016f3a | 4085 | */ |
889e5cbc | 4086 | if (is_pae(vcpu) && *bytes == 4) { |
72016f3a | 4087 | /* Handle a 32-bit guest writing two halves of a 64-bit gpte */ |
889e5cbc XG |
4088 | *gpa &= ~(gpa_t)7; |
4089 | *bytes = 8; | |
54bf36aa | 4090 | r = kvm_vcpu_read_guest(vcpu, *gpa, &gentry, 8); |
72016f3a AK |
4091 | if (r) |
4092 | gentry = 0; | |
08e850c6 AK |
4093 | new = (const u8 *)&gentry; |
4094 | } | |
4095 | ||
889e5cbc | 4096 | switch (*bytes) { |
08e850c6 AK |
4097 | case 4: |
4098 | gentry = *(const u32 *)new; | |
4099 | break; | |
4100 | case 8: | |
4101 | gentry = *(const u64 *)new; | |
4102 | break; | |
4103 | default: | |
4104 | gentry = 0; | |
4105 | break; | |
72016f3a AK |
4106 | } |
4107 | ||
889e5cbc XG |
4108 | return gentry; |
4109 | } | |
4110 | ||
4111 | /* | |
4112 | * If we're seeing too many writes to a page, it may no longer be a page table, | |
4113 | * or we may be forking, in which case it is better to unmap the page. | |
4114 | */ | |
a138fe75 | 4115 | static bool detect_write_flooding(struct kvm_mmu_page *sp) |
889e5cbc | 4116 | { |
a30f47cb XG |
4117 | /* |
4118 | * Skip write-flooding detected for the sp whose level is 1, because | |
4119 | * it can become unsync, then the guest page is not write-protected. | |
4120 | */ | |
f71fa31f | 4121 | if (sp->role.level == PT_PAGE_TABLE_LEVEL) |
a30f47cb | 4122 | return false; |
3246af0e | 4123 | |
a30f47cb | 4124 | return ++sp->write_flooding_count >= 3; |
889e5cbc XG |
4125 | } |
4126 | ||
4127 | /* | |
4128 | * Misaligned accesses are too much trouble to fix up; also, they usually | |
4129 | * indicate a page is not used as a page table. | |
4130 | */ | |
4131 | static bool detect_write_misaligned(struct kvm_mmu_page *sp, gpa_t gpa, | |
4132 | int bytes) | |
4133 | { | |
4134 | unsigned offset, pte_size, misaligned; | |
4135 | ||
4136 | pgprintk("misaligned: gpa %llx bytes %d role %x\n", | |
4137 | gpa, bytes, sp->role.word); | |
4138 | ||
4139 | offset = offset_in_page(gpa); | |
4140 | pte_size = sp->role.cr4_pae ? 8 : 4; | |
5d9ca30e XG |
4141 | |
4142 | /* | |
4143 | * Sometimes, the OS only writes the last one bytes to update status | |
4144 | * bits, for example, in linux, andb instruction is used in clear_bit(). | |
4145 | */ | |
4146 | if (!(offset & (pte_size - 1)) && bytes == 1) | |
4147 | return false; | |
4148 | ||
889e5cbc XG |
4149 | misaligned = (offset ^ (offset + bytes - 1)) & ~(pte_size - 1); |
4150 | misaligned |= bytes < 4; | |
4151 | ||
4152 | return misaligned; | |
4153 | } | |
4154 | ||
4155 | static u64 *get_written_sptes(struct kvm_mmu_page *sp, gpa_t gpa, int *nspte) | |
4156 | { | |
4157 | unsigned page_offset, quadrant; | |
4158 | u64 *spte; | |
4159 | int level; | |
4160 | ||
4161 | page_offset = offset_in_page(gpa); | |
4162 | level = sp->role.level; | |
4163 | *nspte = 1; | |
4164 | if (!sp->role.cr4_pae) { | |
4165 | page_offset <<= 1; /* 32->64 */ | |
4166 | /* | |
4167 | * A 32-bit pde maps 4MB while the shadow pdes map | |
4168 | * only 2MB. So we need to double the offset again | |
4169 | * and zap two pdes instead of one. | |
4170 | */ | |
4171 | if (level == PT32_ROOT_LEVEL) { | |
4172 | page_offset &= ~7; /* kill rounding error */ | |
4173 | page_offset <<= 1; | |
4174 | *nspte = 2; | |
4175 | } | |
4176 | quadrant = page_offset >> PAGE_SHIFT; | |
4177 | page_offset &= ~PAGE_MASK; | |
4178 | if (quadrant != sp->role.quadrant) | |
4179 | return NULL; | |
4180 | } | |
4181 | ||
4182 | spte = &sp->spt[page_offset / sizeof(*spte)]; | |
4183 | return spte; | |
4184 | } | |
4185 | ||
4186 | void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa, | |
4187 | const u8 *new, int bytes) | |
4188 | { | |
4189 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
889e5cbc | 4190 | struct kvm_mmu_page *sp; |
889e5cbc XG |
4191 | LIST_HEAD(invalid_list); |
4192 | u64 entry, gentry, *spte; | |
4193 | int npte; | |
a30f47cb | 4194 | bool remote_flush, local_flush, zap_page; |
4141259b AM |
4195 | union kvm_mmu_page_role mask = { }; |
4196 | ||
4197 | mask.cr0_wp = 1; | |
4198 | mask.cr4_pae = 1; | |
4199 | mask.nxe = 1; | |
4200 | mask.smep_andnot_wp = 1; | |
4201 | mask.smap_andnot_wp = 1; | |
699023e2 | 4202 | mask.smm = 1; |
889e5cbc XG |
4203 | |
4204 | /* | |
4205 | * If we don't have indirect shadow pages, it means no page is | |
4206 | * write-protected, so we can exit simply. | |
4207 | */ | |
4208 | if (!ACCESS_ONCE(vcpu->kvm->arch.indirect_shadow_pages)) | |
4209 | return; | |
4210 | ||
4211 | zap_page = remote_flush = local_flush = false; | |
4212 | ||
4213 | pgprintk("%s: gpa %llx bytes %d\n", __func__, gpa, bytes); | |
4214 | ||
4215 | gentry = mmu_pte_write_fetch_gpte(vcpu, &gpa, new, &bytes); | |
4216 | ||
4217 | /* | |
4218 | * No need to care whether allocation memory is successful | |
4219 | * or not since pte prefetch is skiped if it does not have | |
4220 | * enough objects in the cache. | |
4221 | */ | |
4222 | mmu_topup_memory_caches(vcpu); | |
4223 | ||
4224 | spin_lock(&vcpu->kvm->mmu_lock); | |
4225 | ++vcpu->kvm->stat.mmu_pte_write; | |
0375f7fa | 4226 | kvm_mmu_audit(vcpu, AUDIT_PRE_PTE_WRITE); |
889e5cbc | 4227 | |
b67bfe0d | 4228 | for_each_gfn_indirect_valid_sp(vcpu->kvm, sp, gfn) { |
a30f47cb | 4229 | if (detect_write_misaligned(sp, gpa, bytes) || |
a138fe75 | 4230 | detect_write_flooding(sp)) { |
0671a8e7 | 4231 | zap_page |= !!kvm_mmu_prepare_zap_page(vcpu->kvm, sp, |
f41d335a | 4232 | &invalid_list); |
4cee5764 | 4233 | ++vcpu->kvm->stat.mmu_flooded; |
0e7bc4b9 AK |
4234 | continue; |
4235 | } | |
889e5cbc XG |
4236 | |
4237 | spte = get_written_sptes(sp, gpa, &npte); | |
4238 | if (!spte) | |
4239 | continue; | |
4240 | ||
0671a8e7 | 4241 | local_flush = true; |
ac1b714e | 4242 | while (npte--) { |
79539cec | 4243 | entry = *spte; |
38e3b2b2 | 4244 | mmu_page_zap_pte(vcpu->kvm, sp, spte); |
fa1de2bf XG |
4245 | if (gentry && |
4246 | !((sp->role.word ^ vcpu->arch.mmu.base_role.word) | |
f759e2b4 | 4247 | & mask.word) && rmap_can_add(vcpu)) |
7c562522 | 4248 | mmu_pte_write_new_pte(vcpu, sp, spte, &gentry); |
9bb4f6b1 | 4249 | if (need_remote_flush(entry, *spte)) |
0671a8e7 | 4250 | remote_flush = true; |
ac1b714e | 4251 | ++spte; |
9b7a0325 | 4252 | } |
9b7a0325 | 4253 | } |
0671a8e7 | 4254 | mmu_pte_write_flush_tlb(vcpu, zap_page, remote_flush, local_flush); |
d98ba053 | 4255 | kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list); |
0375f7fa | 4256 | kvm_mmu_audit(vcpu, AUDIT_POST_PTE_WRITE); |
aaee2c94 | 4257 | spin_unlock(&vcpu->kvm->mmu_lock); |
da4a00f0 AK |
4258 | } |
4259 | ||
a436036b AK |
4260 | int kvm_mmu_unprotect_page_virt(struct kvm_vcpu *vcpu, gva_t gva) |
4261 | { | |
10589a46 MT |
4262 | gpa_t gpa; |
4263 | int r; | |
a436036b | 4264 | |
c5a78f2b | 4265 | if (vcpu->arch.mmu.direct_map) |
60f24784 AK |
4266 | return 0; |
4267 | ||
1871c602 | 4268 | gpa = kvm_mmu_gva_to_gpa_read(vcpu, gva, NULL); |
10589a46 | 4269 | |
10589a46 | 4270 | r = kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT); |
1cb3f3ae | 4271 | |
10589a46 | 4272 | return r; |
a436036b | 4273 | } |
577bdc49 | 4274 | EXPORT_SYMBOL_GPL(kvm_mmu_unprotect_page_virt); |
a436036b | 4275 | |
81f4f76b | 4276 | static void make_mmu_pages_available(struct kvm_vcpu *vcpu) |
ebeace86 | 4277 | { |
d98ba053 | 4278 | LIST_HEAD(invalid_list); |
103ad25a | 4279 | |
81f4f76b TY |
4280 | if (likely(kvm_mmu_available_pages(vcpu->kvm) >= KVM_MIN_FREE_MMU_PAGES)) |
4281 | return; | |
4282 | ||
5da59607 TY |
4283 | while (kvm_mmu_available_pages(vcpu->kvm) < KVM_REFILL_PAGES) { |
4284 | if (!prepare_zap_oldest_mmu_page(vcpu->kvm, &invalid_list)) | |
4285 | break; | |
ebeace86 | 4286 | |
4cee5764 | 4287 | ++vcpu->kvm->stat.mmu_recycled; |
ebeace86 | 4288 | } |
aa6bd187 | 4289 | kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list); |
ebeace86 | 4290 | } |
ebeace86 | 4291 | |
1cb3f3ae XG |
4292 | static bool is_mmio_page_fault(struct kvm_vcpu *vcpu, gva_t addr) |
4293 | { | |
4294 | if (vcpu->arch.mmu.direct_map || mmu_is_nested(vcpu)) | |
4295 | return vcpu_match_mmio_gpa(vcpu, addr); | |
4296 | ||
4297 | return vcpu_match_mmio_gva(vcpu, addr); | |
4298 | } | |
4299 | ||
dc25e89e AP |
4300 | int kvm_mmu_page_fault(struct kvm_vcpu *vcpu, gva_t cr2, u32 error_code, |
4301 | void *insn, int insn_len) | |
3067714c | 4302 | { |
1cb3f3ae | 4303 | int r, emulation_type = EMULTYPE_RETRY; |
3067714c AK |
4304 | enum emulation_result er; |
4305 | ||
56028d08 | 4306 | r = vcpu->arch.mmu.page_fault(vcpu, cr2, error_code, false); |
3067714c AK |
4307 | if (r < 0) |
4308 | goto out; | |
4309 | ||
4310 | if (!r) { | |
4311 | r = 1; | |
4312 | goto out; | |
4313 | } | |
4314 | ||
1cb3f3ae XG |
4315 | if (is_mmio_page_fault(vcpu, cr2)) |
4316 | emulation_type = 0; | |
4317 | ||
4318 | er = x86_emulate_instruction(vcpu, cr2, emulation_type, insn, insn_len); | |
3067714c AK |
4319 | |
4320 | switch (er) { | |
4321 | case EMULATE_DONE: | |
4322 | return 1; | |
ac0a48c3 | 4323 | case EMULATE_USER_EXIT: |
3067714c | 4324 | ++vcpu->stat.mmio_exits; |
6d77dbfc | 4325 | /* fall through */ |
3067714c | 4326 | case EMULATE_FAIL: |
3f5d18a9 | 4327 | return 0; |
3067714c AK |
4328 | default: |
4329 | BUG(); | |
4330 | } | |
4331 | out: | |
3067714c AK |
4332 | return r; |
4333 | } | |
4334 | EXPORT_SYMBOL_GPL(kvm_mmu_page_fault); | |
4335 | ||
a7052897 MT |
4336 | void kvm_mmu_invlpg(struct kvm_vcpu *vcpu, gva_t gva) |
4337 | { | |
a7052897 | 4338 | vcpu->arch.mmu.invlpg(vcpu, gva); |
77c3913b | 4339 | kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); |
a7052897 MT |
4340 | ++vcpu->stat.invlpg; |
4341 | } | |
4342 | EXPORT_SYMBOL_GPL(kvm_mmu_invlpg); | |
4343 | ||
18552672 JR |
4344 | void kvm_enable_tdp(void) |
4345 | { | |
4346 | tdp_enabled = true; | |
4347 | } | |
4348 | EXPORT_SYMBOL_GPL(kvm_enable_tdp); | |
4349 | ||
5f4cb662 JR |
4350 | void kvm_disable_tdp(void) |
4351 | { | |
4352 | tdp_enabled = false; | |
4353 | } | |
4354 | EXPORT_SYMBOL_GPL(kvm_disable_tdp); | |
4355 | ||
6aa8b732 AK |
4356 | static void free_mmu_pages(struct kvm_vcpu *vcpu) |
4357 | { | |
ad312c7c | 4358 | free_page((unsigned long)vcpu->arch.mmu.pae_root); |
81407ca5 JR |
4359 | if (vcpu->arch.mmu.lm_root != NULL) |
4360 | free_page((unsigned long)vcpu->arch.mmu.lm_root); | |
6aa8b732 AK |
4361 | } |
4362 | ||
4363 | static int alloc_mmu_pages(struct kvm_vcpu *vcpu) | |
4364 | { | |
17ac10ad | 4365 | struct page *page; |
6aa8b732 AK |
4366 | int i; |
4367 | ||
17ac10ad AK |
4368 | /* |
4369 | * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64. | |
4370 | * Therefore we need to allocate shadow page tables in the first | |
4371 | * 4GB of memory, which happens to fit the DMA32 zone. | |
4372 | */ | |
4373 | page = alloc_page(GFP_KERNEL | __GFP_DMA32); | |
4374 | if (!page) | |
d7fa6ab2 WY |
4375 | return -ENOMEM; |
4376 | ||
ad312c7c | 4377 | vcpu->arch.mmu.pae_root = page_address(page); |
17ac10ad | 4378 | for (i = 0; i < 4; ++i) |
ad312c7c | 4379 | vcpu->arch.mmu.pae_root[i] = INVALID_PAGE; |
17ac10ad | 4380 | |
6aa8b732 | 4381 | return 0; |
6aa8b732 AK |
4382 | } |
4383 | ||
8018c27b | 4384 | int kvm_mmu_create(struct kvm_vcpu *vcpu) |
6aa8b732 | 4385 | { |
e459e322 XG |
4386 | vcpu->arch.walk_mmu = &vcpu->arch.mmu; |
4387 | vcpu->arch.mmu.root_hpa = INVALID_PAGE; | |
4388 | vcpu->arch.mmu.translate_gpa = translate_gpa; | |
4389 | vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa; | |
6aa8b732 | 4390 | |
8018c27b IM |
4391 | return alloc_mmu_pages(vcpu); |
4392 | } | |
6aa8b732 | 4393 | |
8a3c1a33 | 4394 | void kvm_mmu_setup(struct kvm_vcpu *vcpu) |
8018c27b | 4395 | { |
fa4a2c08 | 4396 | MMU_WARN_ON(VALID_PAGE(vcpu->arch.mmu.root_hpa)); |
2c264957 | 4397 | |
8a3c1a33 | 4398 | init_kvm_mmu(vcpu); |
6aa8b732 AK |
4399 | } |
4400 | ||
1bad2b2a XG |
4401 | /* The return value indicates if tlb flush on all vcpus is needed. */ |
4402 | typedef bool (*slot_level_handler) (struct kvm *kvm, unsigned long *rmap); | |
4403 | ||
4404 | /* The caller should hold mmu-lock before calling this function. */ | |
4405 | static bool | |
4406 | slot_handle_level_range(struct kvm *kvm, struct kvm_memory_slot *memslot, | |
4407 | slot_level_handler fn, int start_level, int end_level, | |
4408 | gfn_t start_gfn, gfn_t end_gfn, bool lock_flush_tlb) | |
4409 | { | |
4410 | struct slot_rmap_walk_iterator iterator; | |
4411 | bool flush = false; | |
4412 | ||
4413 | for_each_slot_rmap_range(memslot, start_level, end_level, start_gfn, | |
4414 | end_gfn, &iterator) { | |
4415 | if (iterator.rmap) | |
4416 | flush |= fn(kvm, iterator.rmap); | |
4417 | ||
4418 | if (need_resched() || spin_needbreak(&kvm->mmu_lock)) { | |
4419 | if (flush && lock_flush_tlb) { | |
4420 | kvm_flush_remote_tlbs(kvm); | |
4421 | flush = false; | |
4422 | } | |
4423 | cond_resched_lock(&kvm->mmu_lock); | |
4424 | } | |
4425 | } | |
4426 | ||
4427 | if (flush && lock_flush_tlb) { | |
4428 | kvm_flush_remote_tlbs(kvm); | |
4429 | flush = false; | |
4430 | } | |
4431 | ||
4432 | return flush; | |
4433 | } | |
4434 | ||
4435 | static bool | |
4436 | slot_handle_level(struct kvm *kvm, struct kvm_memory_slot *memslot, | |
4437 | slot_level_handler fn, int start_level, int end_level, | |
4438 | bool lock_flush_tlb) | |
4439 | { | |
4440 | return slot_handle_level_range(kvm, memslot, fn, start_level, | |
4441 | end_level, memslot->base_gfn, | |
4442 | memslot->base_gfn + memslot->npages - 1, | |
4443 | lock_flush_tlb); | |
4444 | } | |
4445 | ||
4446 | static bool | |
4447 | slot_handle_all_level(struct kvm *kvm, struct kvm_memory_slot *memslot, | |
4448 | slot_level_handler fn, bool lock_flush_tlb) | |
4449 | { | |
4450 | return slot_handle_level(kvm, memslot, fn, PT_PAGE_TABLE_LEVEL, | |
4451 | PT_MAX_HUGEPAGE_LEVEL, lock_flush_tlb); | |
4452 | } | |
4453 | ||
4454 | static bool | |
4455 | slot_handle_large_level(struct kvm *kvm, struct kvm_memory_slot *memslot, | |
4456 | slot_level_handler fn, bool lock_flush_tlb) | |
4457 | { | |
4458 | return slot_handle_level(kvm, memslot, fn, PT_PAGE_TABLE_LEVEL + 1, | |
4459 | PT_MAX_HUGEPAGE_LEVEL, lock_flush_tlb); | |
4460 | } | |
4461 | ||
4462 | static bool | |
4463 | slot_handle_leaf(struct kvm *kvm, struct kvm_memory_slot *memslot, | |
4464 | slot_level_handler fn, bool lock_flush_tlb) | |
4465 | { | |
4466 | return slot_handle_level(kvm, memslot, fn, PT_PAGE_TABLE_LEVEL, | |
4467 | PT_PAGE_TABLE_LEVEL, lock_flush_tlb); | |
4468 | } | |
4469 | ||
efdfe536 XG |
4470 | void kvm_zap_gfn_range(struct kvm *kvm, gfn_t gfn_start, gfn_t gfn_end) |
4471 | { | |
4472 | struct kvm_memslots *slots; | |
4473 | struct kvm_memory_slot *memslot; | |
9da0e4d5 | 4474 | int i; |
efdfe536 XG |
4475 | |
4476 | spin_lock(&kvm->mmu_lock); | |
9da0e4d5 PB |
4477 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
4478 | slots = __kvm_memslots(kvm, i); | |
4479 | kvm_for_each_memslot(memslot, slots) { | |
4480 | gfn_t start, end; | |
4481 | ||
4482 | start = max(gfn_start, memslot->base_gfn); | |
4483 | end = min(gfn_end, memslot->base_gfn + memslot->npages); | |
4484 | if (start >= end) | |
4485 | continue; | |
efdfe536 | 4486 | |
9da0e4d5 PB |
4487 | slot_handle_level_range(kvm, memslot, kvm_zap_rmapp, |
4488 | PT_PAGE_TABLE_LEVEL, PT_MAX_HUGEPAGE_LEVEL, | |
4489 | start, end - 1, true); | |
4490 | } | |
efdfe536 XG |
4491 | } |
4492 | ||
4493 | spin_unlock(&kvm->mmu_lock); | |
4494 | } | |
4495 | ||
d77aa73c XG |
4496 | static bool slot_rmap_write_protect(struct kvm *kvm, unsigned long *rmapp) |
4497 | { | |
4498 | return __rmap_write_protect(kvm, rmapp, false); | |
4499 | } | |
4500 | ||
1c91cad4 KH |
4501 | void kvm_mmu_slot_remove_write_access(struct kvm *kvm, |
4502 | struct kvm_memory_slot *memslot) | |
6aa8b732 | 4503 | { |
d77aa73c | 4504 | bool flush; |
6aa8b732 | 4505 | |
9d1beefb | 4506 | spin_lock(&kvm->mmu_lock); |
d77aa73c XG |
4507 | flush = slot_handle_all_level(kvm, memslot, slot_rmap_write_protect, |
4508 | false); | |
9d1beefb | 4509 | spin_unlock(&kvm->mmu_lock); |
198c74f4 XG |
4510 | |
4511 | /* | |
4512 | * kvm_mmu_slot_remove_write_access() and kvm_vm_ioctl_get_dirty_log() | |
4513 | * which do tlb flush out of mmu-lock should be serialized by | |
4514 | * kvm->slots_lock otherwise tlb flush would be missed. | |
4515 | */ | |
4516 | lockdep_assert_held(&kvm->slots_lock); | |
4517 | ||
4518 | /* | |
4519 | * We can flush all the TLBs out of the mmu lock without TLB | |
4520 | * corruption since we just change the spte from writable to | |
4521 | * readonly so that we only need to care the case of changing | |
4522 | * spte from present to present (changing the spte from present | |
4523 | * to nonpresent will flush all the TLBs immediately), in other | |
4524 | * words, the only case we care is mmu_spte_update() where we | |
4525 | * haved checked SPTE_HOST_WRITEABLE | SPTE_MMU_WRITEABLE | |
4526 | * instead of PT_WRITABLE_MASK, that means it does not depend | |
4527 | * on PT_WRITABLE_MASK anymore. | |
4528 | */ | |
d91ffee9 KH |
4529 | if (flush) |
4530 | kvm_flush_remote_tlbs(kvm); | |
6aa8b732 | 4531 | } |
37a7d8b0 | 4532 | |
3ea3b7fa WL |
4533 | static bool kvm_mmu_zap_collapsible_spte(struct kvm *kvm, |
4534 | unsigned long *rmapp) | |
4535 | { | |
4536 | u64 *sptep; | |
4537 | struct rmap_iterator iter; | |
4538 | int need_tlb_flush = 0; | |
4539 | pfn_t pfn; | |
4540 | struct kvm_mmu_page *sp; | |
4541 | ||
0d536790 XG |
4542 | restart: |
4543 | for_each_rmap_spte(rmapp, &iter, sptep) { | |
3ea3b7fa WL |
4544 | sp = page_header(__pa(sptep)); |
4545 | pfn = spte_to_pfn(*sptep); | |
4546 | ||
4547 | /* | |
decf6333 XG |
4548 | * We cannot do huge page mapping for indirect shadow pages, |
4549 | * which are found on the last rmap (level = 1) when not using | |
4550 | * tdp; such shadow pages are synced with the page table in | |
4551 | * the guest, and the guest page table is using 4K page size | |
4552 | * mapping if the indirect sp has level = 1. | |
3ea3b7fa WL |
4553 | */ |
4554 | if (sp->role.direct && | |
4555 | !kvm_is_reserved_pfn(pfn) && | |
4556 | PageTransCompound(pfn_to_page(pfn))) { | |
4557 | drop_spte(kvm, sptep); | |
3ea3b7fa | 4558 | need_tlb_flush = 1; |
0d536790 XG |
4559 | goto restart; |
4560 | } | |
3ea3b7fa WL |
4561 | } |
4562 | ||
4563 | return need_tlb_flush; | |
4564 | } | |
4565 | ||
4566 | void kvm_mmu_zap_collapsible_sptes(struct kvm *kvm, | |
f36f3f28 | 4567 | const struct kvm_memory_slot *memslot) |
3ea3b7fa | 4568 | { |
f36f3f28 | 4569 | /* FIXME: const-ify all uses of struct kvm_memory_slot. */ |
3ea3b7fa | 4570 | spin_lock(&kvm->mmu_lock); |
f36f3f28 PB |
4571 | slot_handle_leaf(kvm, (struct kvm_memory_slot *)memslot, |
4572 | kvm_mmu_zap_collapsible_spte, true); | |
3ea3b7fa WL |
4573 | spin_unlock(&kvm->mmu_lock); |
4574 | } | |
4575 | ||
f4b4b180 KH |
4576 | void kvm_mmu_slot_leaf_clear_dirty(struct kvm *kvm, |
4577 | struct kvm_memory_slot *memslot) | |
4578 | { | |
d77aa73c | 4579 | bool flush; |
f4b4b180 KH |
4580 | |
4581 | spin_lock(&kvm->mmu_lock); | |
d77aa73c | 4582 | flush = slot_handle_leaf(kvm, memslot, __rmap_clear_dirty, false); |
f4b4b180 KH |
4583 | spin_unlock(&kvm->mmu_lock); |
4584 | ||
4585 | lockdep_assert_held(&kvm->slots_lock); | |
4586 | ||
4587 | /* | |
4588 | * It's also safe to flush TLBs out of mmu lock here as currently this | |
4589 | * function is only used for dirty logging, in which case flushing TLB | |
4590 | * out of mmu lock also guarantees no dirty pages will be lost in | |
4591 | * dirty_bitmap. | |
4592 | */ | |
4593 | if (flush) | |
4594 | kvm_flush_remote_tlbs(kvm); | |
4595 | } | |
4596 | EXPORT_SYMBOL_GPL(kvm_mmu_slot_leaf_clear_dirty); | |
4597 | ||
4598 | void kvm_mmu_slot_largepage_remove_write_access(struct kvm *kvm, | |
4599 | struct kvm_memory_slot *memslot) | |
4600 | { | |
d77aa73c | 4601 | bool flush; |
f4b4b180 KH |
4602 | |
4603 | spin_lock(&kvm->mmu_lock); | |
d77aa73c XG |
4604 | flush = slot_handle_large_level(kvm, memslot, slot_rmap_write_protect, |
4605 | false); | |
f4b4b180 KH |
4606 | spin_unlock(&kvm->mmu_lock); |
4607 | ||
4608 | /* see kvm_mmu_slot_remove_write_access */ | |
4609 | lockdep_assert_held(&kvm->slots_lock); | |
4610 | ||
4611 | if (flush) | |
4612 | kvm_flush_remote_tlbs(kvm); | |
4613 | } | |
4614 | EXPORT_SYMBOL_GPL(kvm_mmu_slot_largepage_remove_write_access); | |
4615 | ||
4616 | void kvm_mmu_slot_set_dirty(struct kvm *kvm, | |
4617 | struct kvm_memory_slot *memslot) | |
4618 | { | |
d77aa73c | 4619 | bool flush; |
f4b4b180 KH |
4620 | |
4621 | spin_lock(&kvm->mmu_lock); | |
d77aa73c | 4622 | flush = slot_handle_all_level(kvm, memslot, __rmap_set_dirty, false); |
f4b4b180 KH |
4623 | spin_unlock(&kvm->mmu_lock); |
4624 | ||
4625 | lockdep_assert_held(&kvm->slots_lock); | |
4626 | ||
4627 | /* see kvm_mmu_slot_leaf_clear_dirty */ | |
4628 | if (flush) | |
4629 | kvm_flush_remote_tlbs(kvm); | |
4630 | } | |
4631 | EXPORT_SYMBOL_GPL(kvm_mmu_slot_set_dirty); | |
4632 | ||
e7d11c7a | 4633 | #define BATCH_ZAP_PAGES 10 |
5304b8d3 XG |
4634 | static void kvm_zap_obsolete_pages(struct kvm *kvm) |
4635 | { | |
4636 | struct kvm_mmu_page *sp, *node; | |
e7d11c7a | 4637 | int batch = 0; |
5304b8d3 XG |
4638 | |
4639 | restart: | |
4640 | list_for_each_entry_safe_reverse(sp, node, | |
4641 | &kvm->arch.active_mmu_pages, link) { | |
e7d11c7a XG |
4642 | int ret; |
4643 | ||
5304b8d3 XG |
4644 | /* |
4645 | * No obsolete page exists before new created page since | |
4646 | * active_mmu_pages is the FIFO list. | |
4647 | */ | |
4648 | if (!is_obsolete_sp(kvm, sp)) | |
4649 | break; | |
4650 | ||
4651 | /* | |
5304b8d3 XG |
4652 | * Since we are reversely walking the list and the invalid |
4653 | * list will be moved to the head, skip the invalid page | |
4654 | * can help us to avoid the infinity list walking. | |
4655 | */ | |
4656 | if (sp->role.invalid) | |
4657 | continue; | |
4658 | ||
f34d251d XG |
4659 | /* |
4660 | * Need not flush tlb since we only zap the sp with invalid | |
4661 | * generation number. | |
4662 | */ | |
e7d11c7a | 4663 | if (batch >= BATCH_ZAP_PAGES && |
f34d251d | 4664 | cond_resched_lock(&kvm->mmu_lock)) { |
e7d11c7a | 4665 | batch = 0; |
5304b8d3 XG |
4666 | goto restart; |
4667 | } | |
4668 | ||
365c8868 XG |
4669 | ret = kvm_mmu_prepare_zap_page(kvm, sp, |
4670 | &kvm->arch.zapped_obsolete_pages); | |
e7d11c7a XG |
4671 | batch += ret; |
4672 | ||
4673 | if (ret) | |
5304b8d3 XG |
4674 | goto restart; |
4675 | } | |
4676 | ||
f34d251d XG |
4677 | /* |
4678 | * Should flush tlb before free page tables since lockless-walking | |
4679 | * may use the pages. | |
4680 | */ | |
365c8868 | 4681 | kvm_mmu_commit_zap_page(kvm, &kvm->arch.zapped_obsolete_pages); |
5304b8d3 XG |
4682 | } |
4683 | ||
4684 | /* | |
4685 | * Fast invalidate all shadow pages and use lock-break technique | |
4686 | * to zap obsolete pages. | |
4687 | * | |
4688 | * It's required when memslot is being deleted or VM is being | |
4689 | * destroyed, in these cases, we should ensure that KVM MMU does | |
4690 | * not use any resource of the being-deleted slot or all slots | |
4691 | * after calling the function. | |
4692 | */ | |
4693 | void kvm_mmu_invalidate_zap_all_pages(struct kvm *kvm) | |
4694 | { | |
4695 | spin_lock(&kvm->mmu_lock); | |
35006126 | 4696 | trace_kvm_mmu_invalidate_zap_all_pages(kvm); |
5304b8d3 XG |
4697 | kvm->arch.mmu_valid_gen++; |
4698 | ||
f34d251d XG |
4699 | /* |
4700 | * Notify all vcpus to reload its shadow page table | |
4701 | * and flush TLB. Then all vcpus will switch to new | |
4702 | * shadow page table with the new mmu_valid_gen. | |
4703 | * | |
4704 | * Note: we should do this under the protection of | |
4705 | * mmu-lock, otherwise, vcpu would purge shadow page | |
4706 | * but miss tlb flush. | |
4707 | */ | |
4708 | kvm_reload_remote_mmus(kvm); | |
4709 | ||
5304b8d3 XG |
4710 | kvm_zap_obsolete_pages(kvm); |
4711 | spin_unlock(&kvm->mmu_lock); | |
4712 | } | |
4713 | ||
365c8868 XG |
4714 | static bool kvm_has_zapped_obsolete_pages(struct kvm *kvm) |
4715 | { | |
4716 | return unlikely(!list_empty_careful(&kvm->arch.zapped_obsolete_pages)); | |
4717 | } | |
4718 | ||
54bf36aa | 4719 | void kvm_mmu_invalidate_mmio_sptes(struct kvm *kvm, struct kvm_memslots *slots) |
f8f55942 XG |
4720 | { |
4721 | /* | |
4722 | * The very rare case: if the generation-number is round, | |
4723 | * zap all shadow pages. | |
f8f55942 | 4724 | */ |
54bf36aa | 4725 | if (unlikely((slots->generation & MMIO_GEN_MASK) == 0)) { |
a629df7e | 4726 | printk_ratelimited(KERN_DEBUG "kvm: zapping shadow pages for mmio generation wraparound\n"); |
a8eca9dc | 4727 | kvm_mmu_invalidate_zap_all_pages(kvm); |
7a2e8aaf | 4728 | } |
f8f55942 XG |
4729 | } |
4730 | ||
70534a73 DC |
4731 | static unsigned long |
4732 | mmu_shrink_scan(struct shrinker *shrink, struct shrink_control *sc) | |
3ee16c81 IE |
4733 | { |
4734 | struct kvm *kvm; | |
1495f230 | 4735 | int nr_to_scan = sc->nr_to_scan; |
70534a73 | 4736 | unsigned long freed = 0; |
3ee16c81 | 4737 | |
2f303b74 | 4738 | spin_lock(&kvm_lock); |
3ee16c81 IE |
4739 | |
4740 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
3d56cbdf | 4741 | int idx; |
d98ba053 | 4742 | LIST_HEAD(invalid_list); |
3ee16c81 | 4743 | |
35f2d16b TY |
4744 | /* |
4745 | * Never scan more than sc->nr_to_scan VM instances. | |
4746 | * Will not hit this condition practically since we do not try | |
4747 | * to shrink more than one VM and it is very unlikely to see | |
4748 | * !n_used_mmu_pages so many times. | |
4749 | */ | |
4750 | if (!nr_to_scan--) | |
4751 | break; | |
19526396 GN |
4752 | /* |
4753 | * n_used_mmu_pages is accessed without holding kvm->mmu_lock | |
4754 | * here. We may skip a VM instance errorneosly, but we do not | |
4755 | * want to shrink a VM that only started to populate its MMU | |
4756 | * anyway. | |
4757 | */ | |
365c8868 XG |
4758 | if (!kvm->arch.n_used_mmu_pages && |
4759 | !kvm_has_zapped_obsolete_pages(kvm)) | |
19526396 | 4760 | continue; |
19526396 | 4761 | |
f656ce01 | 4762 | idx = srcu_read_lock(&kvm->srcu); |
3ee16c81 | 4763 | spin_lock(&kvm->mmu_lock); |
3ee16c81 | 4764 | |
365c8868 XG |
4765 | if (kvm_has_zapped_obsolete_pages(kvm)) { |
4766 | kvm_mmu_commit_zap_page(kvm, | |
4767 | &kvm->arch.zapped_obsolete_pages); | |
4768 | goto unlock; | |
4769 | } | |
4770 | ||
70534a73 DC |
4771 | if (prepare_zap_oldest_mmu_page(kvm, &invalid_list)) |
4772 | freed++; | |
d98ba053 | 4773 | kvm_mmu_commit_zap_page(kvm, &invalid_list); |
19526396 | 4774 | |
365c8868 | 4775 | unlock: |
3ee16c81 | 4776 | spin_unlock(&kvm->mmu_lock); |
f656ce01 | 4777 | srcu_read_unlock(&kvm->srcu, idx); |
19526396 | 4778 | |
70534a73 DC |
4779 | /* |
4780 | * unfair on small ones | |
4781 | * per-vm shrinkers cry out | |
4782 | * sadness comes quickly | |
4783 | */ | |
19526396 GN |
4784 | list_move_tail(&kvm->vm_list, &vm_list); |
4785 | break; | |
3ee16c81 | 4786 | } |
3ee16c81 | 4787 | |
2f303b74 | 4788 | spin_unlock(&kvm_lock); |
70534a73 | 4789 | return freed; |
70534a73 DC |
4790 | } |
4791 | ||
4792 | static unsigned long | |
4793 | mmu_shrink_count(struct shrinker *shrink, struct shrink_control *sc) | |
4794 | { | |
45221ab6 | 4795 | return percpu_counter_read_positive(&kvm_total_used_mmu_pages); |
3ee16c81 IE |
4796 | } |
4797 | ||
4798 | static struct shrinker mmu_shrinker = { | |
70534a73 DC |
4799 | .count_objects = mmu_shrink_count, |
4800 | .scan_objects = mmu_shrink_scan, | |
3ee16c81 IE |
4801 | .seeks = DEFAULT_SEEKS * 10, |
4802 | }; | |
4803 | ||
2ddfd20e | 4804 | static void mmu_destroy_caches(void) |
b5a33a75 | 4805 | { |
53c07b18 XG |
4806 | if (pte_list_desc_cache) |
4807 | kmem_cache_destroy(pte_list_desc_cache); | |
d3d25b04 AK |
4808 | if (mmu_page_header_cache) |
4809 | kmem_cache_destroy(mmu_page_header_cache); | |
b5a33a75 AK |
4810 | } |
4811 | ||
4812 | int kvm_mmu_module_init(void) | |
4813 | { | |
53c07b18 XG |
4814 | pte_list_desc_cache = kmem_cache_create("pte_list_desc", |
4815 | sizeof(struct pte_list_desc), | |
20c2df83 | 4816 | 0, 0, NULL); |
53c07b18 | 4817 | if (!pte_list_desc_cache) |
b5a33a75 AK |
4818 | goto nomem; |
4819 | ||
d3d25b04 AK |
4820 | mmu_page_header_cache = kmem_cache_create("kvm_mmu_page_header", |
4821 | sizeof(struct kvm_mmu_page), | |
20c2df83 | 4822 | 0, 0, NULL); |
d3d25b04 AK |
4823 | if (!mmu_page_header_cache) |
4824 | goto nomem; | |
4825 | ||
908c7f19 | 4826 | if (percpu_counter_init(&kvm_total_used_mmu_pages, 0, GFP_KERNEL)) |
45bf21a8 WY |
4827 | goto nomem; |
4828 | ||
3ee16c81 IE |
4829 | register_shrinker(&mmu_shrinker); |
4830 | ||
b5a33a75 AK |
4831 | return 0; |
4832 | ||
4833 | nomem: | |
3ee16c81 | 4834 | mmu_destroy_caches(); |
b5a33a75 AK |
4835 | return -ENOMEM; |
4836 | } | |
4837 | ||
3ad82a7e ZX |
4838 | /* |
4839 | * Caculate mmu pages needed for kvm. | |
4840 | */ | |
4841 | unsigned int kvm_mmu_calculate_mmu_pages(struct kvm *kvm) | |
4842 | { | |
3ad82a7e ZX |
4843 | unsigned int nr_mmu_pages; |
4844 | unsigned int nr_pages = 0; | |
bc6678a3 | 4845 | struct kvm_memslots *slots; |
be6ba0f0 | 4846 | struct kvm_memory_slot *memslot; |
9da0e4d5 | 4847 | int i; |
3ad82a7e | 4848 | |
9da0e4d5 PB |
4849 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
4850 | slots = __kvm_memslots(kvm, i); | |
90d83dc3 | 4851 | |
9da0e4d5 PB |
4852 | kvm_for_each_memslot(memslot, slots) |
4853 | nr_pages += memslot->npages; | |
4854 | } | |
3ad82a7e ZX |
4855 | |
4856 | nr_mmu_pages = nr_pages * KVM_PERMILLE_MMU_PAGES / 1000; | |
4857 | nr_mmu_pages = max(nr_mmu_pages, | |
9da0e4d5 | 4858 | (unsigned int) KVM_MIN_ALLOC_MMU_PAGES); |
3ad82a7e ZX |
4859 | |
4860 | return nr_mmu_pages; | |
4861 | } | |
4862 | ||
94d8b056 MT |
4863 | int kvm_mmu_get_spte_hierarchy(struct kvm_vcpu *vcpu, u64 addr, u64 sptes[4]) |
4864 | { | |
4865 | struct kvm_shadow_walk_iterator iterator; | |
c2a2ac2b | 4866 | u64 spte; |
94d8b056 MT |
4867 | int nr_sptes = 0; |
4868 | ||
37f6a4e2 MT |
4869 | if (!VALID_PAGE(vcpu->arch.mmu.root_hpa)) |
4870 | return nr_sptes; | |
4871 | ||
c2a2ac2b XG |
4872 | walk_shadow_page_lockless_begin(vcpu); |
4873 | for_each_shadow_entry_lockless(vcpu, addr, iterator, spte) { | |
4874 | sptes[iterator.level-1] = spte; | |
94d8b056 | 4875 | nr_sptes++; |
c2a2ac2b | 4876 | if (!is_shadow_present_pte(spte)) |
94d8b056 MT |
4877 | break; |
4878 | } | |
c2a2ac2b | 4879 | walk_shadow_page_lockless_end(vcpu); |
94d8b056 MT |
4880 | |
4881 | return nr_sptes; | |
4882 | } | |
4883 | EXPORT_SYMBOL_GPL(kvm_mmu_get_spte_hierarchy); | |
4884 | ||
c42fffe3 XG |
4885 | void kvm_mmu_destroy(struct kvm_vcpu *vcpu) |
4886 | { | |
95f93af4 | 4887 | kvm_mmu_unload(vcpu); |
c42fffe3 XG |
4888 | free_mmu_pages(vcpu); |
4889 | mmu_free_memory_caches(vcpu); | |
b034cf01 XG |
4890 | } |
4891 | ||
b034cf01 XG |
4892 | void kvm_mmu_module_exit(void) |
4893 | { | |
4894 | mmu_destroy_caches(); | |
4895 | percpu_counter_destroy(&kvm_total_used_mmu_pages); | |
4896 | unregister_shrinker(&mmu_shrinker); | |
c42fffe3 XG |
4897 | mmu_audit_disable(); |
4898 | } |