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" |
e495606d | 25 | |
edf88417 | 26 | #include <linux/kvm_host.h> |
6aa8b732 AK |
27 | #include <linux/types.h> |
28 | #include <linux/string.h> | |
6aa8b732 AK |
29 | #include <linux/mm.h> |
30 | #include <linux/highmem.h> | |
31 | #include <linux/module.h> | |
448353ca | 32 | #include <linux/swap.h> |
05da4558 | 33 | #include <linux/hugetlb.h> |
2f333bcb | 34 | #include <linux/compiler.h> |
bc6678a3 | 35 | #include <linux/srcu.h> |
5a0e3ad6 | 36 | #include <linux/slab.h> |
bf998156 | 37 | #include <linux/uaccess.h> |
6aa8b732 | 38 | |
e495606d AK |
39 | #include <asm/page.h> |
40 | #include <asm/cmpxchg.h> | |
4e542370 | 41 | #include <asm/io.h> |
13673a90 | 42 | #include <asm/vmx.h> |
6aa8b732 | 43 | |
18552672 JR |
44 | /* |
45 | * When setting this variable to true it enables Two-Dimensional-Paging | |
46 | * where the hardware walks 2 page tables: | |
47 | * 1. the guest-virtual to guest-physical | |
48 | * 2. while doing 1. it walks guest-physical to host-physical | |
49 | * If the hardware supports that we don't need to do shadow paging. | |
50 | */ | |
2f333bcb | 51 | bool tdp_enabled = false; |
18552672 | 52 | |
8b1fe17c XG |
53 | enum { |
54 | AUDIT_PRE_PAGE_FAULT, | |
55 | AUDIT_POST_PAGE_FAULT, | |
56 | AUDIT_PRE_PTE_WRITE, | |
6903074c XG |
57 | AUDIT_POST_PTE_WRITE, |
58 | AUDIT_PRE_SYNC, | |
59 | AUDIT_POST_SYNC | |
8b1fe17c | 60 | }; |
37a7d8b0 | 61 | |
8b1fe17c | 62 | #undef MMU_DEBUG |
37a7d8b0 AK |
63 | |
64 | #ifdef MMU_DEBUG | |
65 | ||
66 | #define pgprintk(x...) do { if (dbg) printk(x); } while (0) | |
67 | #define rmap_printk(x...) do { if (dbg) printk(x); } while (0) | |
68 | ||
69 | #else | |
70 | ||
71 | #define pgprintk(x...) do { } while (0) | |
72 | #define rmap_printk(x...) do { } while (0) | |
73 | ||
74 | #endif | |
75 | ||
8b1fe17c | 76 | #ifdef MMU_DEBUG |
476bc001 | 77 | static bool dbg = 0; |
6ada8cca | 78 | module_param(dbg, bool, 0644); |
37a7d8b0 | 79 | #endif |
6aa8b732 | 80 | |
d6c69ee9 YD |
81 | #ifndef MMU_DEBUG |
82 | #define ASSERT(x) do { } while (0) | |
83 | #else | |
6aa8b732 AK |
84 | #define ASSERT(x) \ |
85 | if (!(x)) { \ | |
86 | printk(KERN_WARNING "assertion failed %s:%d: %s\n", \ | |
87 | __FILE__, __LINE__, #x); \ | |
88 | } | |
d6c69ee9 | 89 | #endif |
6aa8b732 | 90 | |
957ed9ef XG |
91 | #define PTE_PREFETCH_NUM 8 |
92 | ||
00763e41 | 93 | #define PT_FIRST_AVAIL_BITS_SHIFT 10 |
6aa8b732 AK |
94 | #define PT64_SECOND_AVAIL_BITS_SHIFT 52 |
95 | ||
6aa8b732 AK |
96 | #define PT64_LEVEL_BITS 9 |
97 | ||
98 | #define PT64_LEVEL_SHIFT(level) \ | |
d77c26fc | 99 | (PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS) |
6aa8b732 | 100 | |
6aa8b732 AK |
101 | #define PT64_INDEX(address, level)\ |
102 | (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1)) | |
103 | ||
104 | ||
105 | #define PT32_LEVEL_BITS 10 | |
106 | ||
107 | #define PT32_LEVEL_SHIFT(level) \ | |
d77c26fc | 108 | (PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS) |
6aa8b732 | 109 | |
e04da980 JR |
110 | #define PT32_LVL_OFFSET_MASK(level) \ |
111 | (PT32_BASE_ADDR_MASK & ((1ULL << (PAGE_SHIFT + (((level) - 1) \ | |
112 | * PT32_LEVEL_BITS))) - 1)) | |
6aa8b732 AK |
113 | |
114 | #define PT32_INDEX(address, level)\ | |
115 | (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1)) | |
116 | ||
117 | ||
27aba766 | 118 | #define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1)) |
6aa8b732 AK |
119 | #define PT64_DIR_BASE_ADDR_MASK \ |
120 | (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1)) | |
e04da980 JR |
121 | #define PT64_LVL_ADDR_MASK(level) \ |
122 | (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + (((level) - 1) \ | |
123 | * PT64_LEVEL_BITS))) - 1)) | |
124 | #define PT64_LVL_OFFSET_MASK(level) \ | |
125 | (PT64_BASE_ADDR_MASK & ((1ULL << (PAGE_SHIFT + (((level) - 1) \ | |
126 | * PT64_LEVEL_BITS))) - 1)) | |
6aa8b732 AK |
127 | |
128 | #define PT32_BASE_ADDR_MASK PAGE_MASK | |
129 | #define PT32_DIR_BASE_ADDR_MASK \ | |
130 | (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1)) | |
e04da980 JR |
131 | #define PT32_LVL_ADDR_MASK(level) \ |
132 | (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + (((level) - 1) \ | |
133 | * PT32_LEVEL_BITS))) - 1)) | |
6aa8b732 | 134 | |
79539cec AK |
135 | #define PT64_PERM_MASK (PT_PRESENT_MASK | PT_WRITABLE_MASK | PT_USER_MASK \ |
136 | | PT64_NX_MASK) | |
6aa8b732 | 137 | |
fe135d2c AK |
138 | #define ACC_EXEC_MASK 1 |
139 | #define ACC_WRITE_MASK PT_WRITABLE_MASK | |
140 | #define ACC_USER_MASK PT_USER_MASK | |
141 | #define ACC_ALL (ACC_EXEC_MASK | ACC_WRITE_MASK | ACC_USER_MASK) | |
142 | ||
90bb6fc5 AK |
143 | #include <trace/events/kvm.h> |
144 | ||
07420171 AK |
145 | #define CREATE_TRACE_POINTS |
146 | #include "mmutrace.h" | |
147 | ||
1403283a IE |
148 | #define SPTE_HOST_WRITEABLE (1ULL << PT_FIRST_AVAIL_BITS_SHIFT) |
149 | ||
135f8c2b AK |
150 | #define SHADOW_PT_INDEX(addr, level) PT64_INDEX(addr, level) |
151 | ||
220f773a TY |
152 | /* make pte_list_desc fit well in cache line */ |
153 | #define PTE_LIST_EXT 3 | |
154 | ||
53c07b18 XG |
155 | struct pte_list_desc { |
156 | u64 *sptes[PTE_LIST_EXT]; | |
157 | struct pte_list_desc *more; | |
cd4a4e53 AK |
158 | }; |
159 | ||
2d11123a AK |
160 | struct kvm_shadow_walk_iterator { |
161 | u64 addr; | |
162 | hpa_t shadow_addr; | |
2d11123a | 163 | u64 *sptep; |
dd3bfd59 | 164 | int level; |
2d11123a AK |
165 | unsigned index; |
166 | }; | |
167 | ||
168 | #define for_each_shadow_entry(_vcpu, _addr, _walker) \ | |
169 | for (shadow_walk_init(&(_walker), _vcpu, _addr); \ | |
170 | shadow_walk_okay(&(_walker)); \ | |
171 | shadow_walk_next(&(_walker))) | |
172 | ||
c2a2ac2b XG |
173 | #define for_each_shadow_entry_lockless(_vcpu, _addr, _walker, spte) \ |
174 | for (shadow_walk_init(&(_walker), _vcpu, _addr); \ | |
175 | shadow_walk_okay(&(_walker)) && \ | |
176 | ({ spte = mmu_spte_get_lockless(_walker.sptep); 1; }); \ | |
177 | __shadow_walk_next(&(_walker), spte)) | |
178 | ||
53c07b18 | 179 | static struct kmem_cache *pte_list_desc_cache; |
d3d25b04 | 180 | static struct kmem_cache *mmu_page_header_cache; |
45221ab6 | 181 | static struct percpu_counter kvm_total_used_mmu_pages; |
b5a33a75 | 182 | |
7b52345e SY |
183 | static u64 __read_mostly shadow_nx_mask; |
184 | static u64 __read_mostly shadow_x_mask; /* mutual exclusive with nx_mask */ | |
185 | static u64 __read_mostly shadow_user_mask; | |
186 | static u64 __read_mostly shadow_accessed_mask; | |
187 | static u64 __read_mostly shadow_dirty_mask; | |
ce88decf XG |
188 | static u64 __read_mostly shadow_mmio_mask; |
189 | ||
190 | static void mmu_spte_set(u64 *sptep, u64 spte); | |
191 | ||
192 | void kvm_mmu_set_mmio_spte_mask(u64 mmio_mask) | |
193 | { | |
194 | shadow_mmio_mask = mmio_mask; | |
195 | } | |
196 | EXPORT_SYMBOL_GPL(kvm_mmu_set_mmio_spte_mask); | |
197 | ||
198 | static void mark_mmio_spte(u64 *sptep, u64 gfn, unsigned access) | |
199 | { | |
200 | access &= ACC_WRITE_MASK | ACC_USER_MASK; | |
201 | ||
4f022648 | 202 | trace_mark_mmio_spte(sptep, gfn, access); |
ce88decf XG |
203 | mmu_spte_set(sptep, shadow_mmio_mask | access | gfn << PAGE_SHIFT); |
204 | } | |
205 | ||
206 | static bool is_mmio_spte(u64 spte) | |
207 | { | |
208 | return (spte & shadow_mmio_mask) == shadow_mmio_mask; | |
209 | } | |
210 | ||
211 | static gfn_t get_mmio_spte_gfn(u64 spte) | |
212 | { | |
213 | return (spte & ~shadow_mmio_mask) >> PAGE_SHIFT; | |
214 | } | |
215 | ||
216 | static unsigned get_mmio_spte_access(u64 spte) | |
217 | { | |
218 | return (spte & ~shadow_mmio_mask) & ~PAGE_MASK; | |
219 | } | |
220 | ||
221 | static bool set_mmio_spte(u64 *sptep, gfn_t gfn, pfn_t pfn, unsigned access) | |
222 | { | |
223 | if (unlikely(is_noslot_pfn(pfn))) { | |
224 | mark_mmio_spte(sptep, gfn, access); | |
225 | return true; | |
226 | } | |
227 | ||
228 | return false; | |
229 | } | |
c7addb90 | 230 | |
82725b20 DE |
231 | static inline u64 rsvd_bits(int s, int e) |
232 | { | |
233 | return ((1ULL << (e - s + 1)) - 1) << s; | |
234 | } | |
235 | ||
7b52345e | 236 | void kvm_mmu_set_mask_ptes(u64 user_mask, u64 accessed_mask, |
4b12f0de | 237 | u64 dirty_mask, u64 nx_mask, u64 x_mask) |
7b52345e SY |
238 | { |
239 | shadow_user_mask = user_mask; | |
240 | shadow_accessed_mask = accessed_mask; | |
241 | shadow_dirty_mask = dirty_mask; | |
242 | shadow_nx_mask = nx_mask; | |
243 | shadow_x_mask = x_mask; | |
244 | } | |
245 | EXPORT_SYMBOL_GPL(kvm_mmu_set_mask_ptes); | |
246 | ||
6aa8b732 AK |
247 | static int is_cpuid_PSE36(void) |
248 | { | |
249 | return 1; | |
250 | } | |
251 | ||
73b1087e AK |
252 | static int is_nx(struct kvm_vcpu *vcpu) |
253 | { | |
f6801dff | 254 | return vcpu->arch.efer & EFER_NX; |
73b1087e AK |
255 | } |
256 | ||
c7addb90 AK |
257 | static int is_shadow_present_pte(u64 pte) |
258 | { | |
ce88decf | 259 | return pte & PT_PRESENT_MASK && !is_mmio_spte(pte); |
c7addb90 AK |
260 | } |
261 | ||
05da4558 MT |
262 | static int is_large_pte(u64 pte) |
263 | { | |
264 | return pte & PT_PAGE_SIZE_MASK; | |
265 | } | |
266 | ||
43a3795a | 267 | static int is_dirty_gpte(unsigned long pte) |
e3c5e7ec | 268 | { |
439e218a | 269 | return pte & PT_DIRTY_MASK; |
e3c5e7ec AK |
270 | } |
271 | ||
43a3795a | 272 | static int is_rmap_spte(u64 pte) |
cd4a4e53 | 273 | { |
4b1a80fa | 274 | return is_shadow_present_pte(pte); |
cd4a4e53 AK |
275 | } |
276 | ||
776e6633 MT |
277 | static int is_last_spte(u64 pte, int level) |
278 | { | |
279 | if (level == PT_PAGE_TABLE_LEVEL) | |
280 | return 1; | |
852e3c19 | 281 | if (is_large_pte(pte)) |
776e6633 MT |
282 | return 1; |
283 | return 0; | |
284 | } | |
285 | ||
35149e21 | 286 | static pfn_t spte_to_pfn(u64 pte) |
0b49ea86 | 287 | { |
35149e21 | 288 | return (pte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT; |
0b49ea86 AK |
289 | } |
290 | ||
da928521 AK |
291 | static gfn_t pse36_gfn_delta(u32 gpte) |
292 | { | |
293 | int shift = 32 - PT32_DIR_PSE36_SHIFT - PAGE_SHIFT; | |
294 | ||
295 | return (gpte & PT32_DIR_PSE36_MASK) << shift; | |
296 | } | |
297 | ||
603e0651 | 298 | #ifdef CONFIG_X86_64 |
d555c333 | 299 | static void __set_spte(u64 *sptep, u64 spte) |
e663ee64 | 300 | { |
603e0651 | 301 | *sptep = spte; |
e663ee64 AK |
302 | } |
303 | ||
603e0651 | 304 | static void __update_clear_spte_fast(u64 *sptep, u64 spte) |
a9221dd5 | 305 | { |
603e0651 XG |
306 | *sptep = spte; |
307 | } | |
308 | ||
309 | static u64 __update_clear_spte_slow(u64 *sptep, u64 spte) | |
310 | { | |
311 | return xchg(sptep, spte); | |
312 | } | |
c2a2ac2b XG |
313 | |
314 | static u64 __get_spte_lockless(u64 *sptep) | |
315 | { | |
316 | return ACCESS_ONCE(*sptep); | |
317 | } | |
ce88decf XG |
318 | |
319 | static bool __check_direct_spte_mmio_pf(u64 spte) | |
320 | { | |
321 | /* It is valid if the spte is zapped. */ | |
322 | return spte == 0ull; | |
323 | } | |
a9221dd5 | 324 | #else |
603e0651 XG |
325 | union split_spte { |
326 | struct { | |
327 | u32 spte_low; | |
328 | u32 spte_high; | |
329 | }; | |
330 | u64 spte; | |
331 | }; | |
a9221dd5 | 332 | |
c2a2ac2b XG |
333 | static void count_spte_clear(u64 *sptep, u64 spte) |
334 | { | |
335 | struct kvm_mmu_page *sp = page_header(__pa(sptep)); | |
336 | ||
337 | if (is_shadow_present_pte(spte)) | |
338 | return; | |
339 | ||
340 | /* Ensure the spte is completely set before we increase the count */ | |
341 | smp_wmb(); | |
342 | sp->clear_spte_count++; | |
343 | } | |
344 | ||
603e0651 XG |
345 | static void __set_spte(u64 *sptep, u64 spte) |
346 | { | |
347 | union split_spte *ssptep, sspte; | |
a9221dd5 | 348 | |
603e0651 XG |
349 | ssptep = (union split_spte *)sptep; |
350 | sspte = (union split_spte)spte; | |
351 | ||
352 | ssptep->spte_high = sspte.spte_high; | |
353 | ||
354 | /* | |
355 | * If we map the spte from nonpresent to present, We should store | |
356 | * the high bits firstly, then set present bit, so cpu can not | |
357 | * fetch this spte while we are setting the spte. | |
358 | */ | |
359 | smp_wmb(); | |
360 | ||
361 | ssptep->spte_low = sspte.spte_low; | |
a9221dd5 AK |
362 | } |
363 | ||
603e0651 XG |
364 | static void __update_clear_spte_fast(u64 *sptep, u64 spte) |
365 | { | |
366 | union split_spte *ssptep, sspte; | |
367 | ||
368 | ssptep = (union split_spte *)sptep; | |
369 | sspte = (union split_spte)spte; | |
370 | ||
371 | ssptep->spte_low = sspte.spte_low; | |
372 | ||
373 | /* | |
374 | * If we map the spte from present to nonpresent, we should clear | |
375 | * present bit firstly to avoid vcpu fetch the old high bits. | |
376 | */ | |
377 | smp_wmb(); | |
378 | ||
379 | ssptep->spte_high = sspte.spte_high; | |
c2a2ac2b | 380 | count_spte_clear(sptep, spte); |
603e0651 XG |
381 | } |
382 | ||
383 | static u64 __update_clear_spte_slow(u64 *sptep, u64 spte) | |
384 | { | |
385 | union split_spte *ssptep, sspte, orig; | |
386 | ||
387 | ssptep = (union split_spte *)sptep; | |
388 | sspte = (union split_spte)spte; | |
389 | ||
390 | /* xchg acts as a barrier before the setting of the high bits */ | |
391 | orig.spte_low = xchg(&ssptep->spte_low, sspte.spte_low); | |
41bc3186 ZJ |
392 | orig.spte_high = ssptep->spte_high; |
393 | ssptep->spte_high = sspte.spte_high; | |
c2a2ac2b | 394 | count_spte_clear(sptep, spte); |
603e0651 XG |
395 | |
396 | return orig.spte; | |
397 | } | |
c2a2ac2b XG |
398 | |
399 | /* | |
400 | * The idea using the light way get the spte on x86_32 guest is from | |
401 | * gup_get_pte(arch/x86/mm/gup.c). | |
402 | * The difference is we can not catch the spte tlb flush if we leave | |
403 | * guest mode, so we emulate it by increase clear_spte_count when spte | |
404 | * is cleared. | |
405 | */ | |
406 | static u64 __get_spte_lockless(u64 *sptep) | |
407 | { | |
408 | struct kvm_mmu_page *sp = page_header(__pa(sptep)); | |
409 | union split_spte spte, *orig = (union split_spte *)sptep; | |
410 | int count; | |
411 | ||
412 | retry: | |
413 | count = sp->clear_spte_count; | |
414 | smp_rmb(); | |
415 | ||
416 | spte.spte_low = orig->spte_low; | |
417 | smp_rmb(); | |
418 | ||
419 | spte.spte_high = orig->spte_high; | |
420 | smp_rmb(); | |
421 | ||
422 | if (unlikely(spte.spte_low != orig->spte_low || | |
423 | count != sp->clear_spte_count)) | |
424 | goto retry; | |
425 | ||
426 | return spte.spte; | |
427 | } | |
ce88decf XG |
428 | |
429 | static bool __check_direct_spte_mmio_pf(u64 spte) | |
430 | { | |
431 | union split_spte sspte = (union split_spte)spte; | |
432 | u32 high_mmio_mask = shadow_mmio_mask >> 32; | |
433 | ||
434 | /* It is valid if the spte is zapped. */ | |
435 | if (spte == 0ull) | |
436 | return true; | |
437 | ||
438 | /* It is valid if the spte is being zapped. */ | |
439 | if (sspte.spte_low == 0ull && | |
440 | (sspte.spte_high & high_mmio_mask) == high_mmio_mask) | |
441 | return true; | |
442 | ||
443 | return false; | |
444 | } | |
603e0651 XG |
445 | #endif |
446 | ||
8672b721 XG |
447 | static bool spte_has_volatile_bits(u64 spte) |
448 | { | |
449 | if (!shadow_accessed_mask) | |
450 | return false; | |
451 | ||
452 | if (!is_shadow_present_pte(spte)) | |
453 | return false; | |
454 | ||
4132779b XG |
455 | if ((spte & shadow_accessed_mask) && |
456 | (!is_writable_pte(spte) || (spte & shadow_dirty_mask))) | |
8672b721 XG |
457 | return false; |
458 | ||
459 | return true; | |
460 | } | |
461 | ||
4132779b XG |
462 | static bool spte_is_bit_cleared(u64 old_spte, u64 new_spte, u64 bit_mask) |
463 | { | |
464 | return (old_spte & bit_mask) && !(new_spte & bit_mask); | |
465 | } | |
466 | ||
1df9f2dc XG |
467 | /* Rules for using mmu_spte_set: |
468 | * Set the sptep from nonpresent to present. | |
469 | * Note: the sptep being assigned *must* be either not present | |
470 | * or in a state where the hardware will not attempt to update | |
471 | * the spte. | |
472 | */ | |
473 | static void mmu_spte_set(u64 *sptep, u64 new_spte) | |
474 | { | |
475 | WARN_ON(is_shadow_present_pte(*sptep)); | |
476 | __set_spte(sptep, new_spte); | |
477 | } | |
478 | ||
479 | /* Rules for using mmu_spte_update: | |
480 | * Update the state bits, it means the mapped pfn is not changged. | |
481 | */ | |
482 | static void mmu_spte_update(u64 *sptep, u64 new_spte) | |
b79b93f9 | 483 | { |
4132779b XG |
484 | u64 mask, old_spte = *sptep; |
485 | ||
486 | WARN_ON(!is_rmap_spte(new_spte)); | |
b79b93f9 | 487 | |
1df9f2dc XG |
488 | if (!is_shadow_present_pte(old_spte)) |
489 | return mmu_spte_set(sptep, new_spte); | |
490 | ||
4132779b XG |
491 | new_spte |= old_spte & shadow_dirty_mask; |
492 | ||
493 | mask = shadow_accessed_mask; | |
494 | if (is_writable_pte(old_spte)) | |
495 | mask |= shadow_dirty_mask; | |
496 | ||
497 | if (!spte_has_volatile_bits(old_spte) || (new_spte & mask) == mask) | |
603e0651 | 498 | __update_clear_spte_fast(sptep, new_spte); |
4132779b | 499 | else |
603e0651 | 500 | old_spte = __update_clear_spte_slow(sptep, new_spte); |
4132779b XG |
501 | |
502 | if (!shadow_accessed_mask) | |
503 | return; | |
504 | ||
505 | if (spte_is_bit_cleared(old_spte, new_spte, shadow_accessed_mask)) | |
506 | kvm_set_pfn_accessed(spte_to_pfn(old_spte)); | |
507 | if (spte_is_bit_cleared(old_spte, new_spte, shadow_dirty_mask)) | |
508 | kvm_set_pfn_dirty(spte_to_pfn(old_spte)); | |
b79b93f9 AK |
509 | } |
510 | ||
1df9f2dc XG |
511 | /* |
512 | * Rules for using mmu_spte_clear_track_bits: | |
513 | * It sets the sptep from present to nonpresent, and track the | |
514 | * state bits, it is used to clear the last level sptep. | |
515 | */ | |
516 | static int mmu_spte_clear_track_bits(u64 *sptep) | |
517 | { | |
518 | pfn_t pfn; | |
519 | u64 old_spte = *sptep; | |
520 | ||
521 | if (!spte_has_volatile_bits(old_spte)) | |
603e0651 | 522 | __update_clear_spte_fast(sptep, 0ull); |
1df9f2dc | 523 | else |
603e0651 | 524 | old_spte = __update_clear_spte_slow(sptep, 0ull); |
1df9f2dc XG |
525 | |
526 | if (!is_rmap_spte(old_spte)) | |
527 | return 0; | |
528 | ||
529 | pfn = spte_to_pfn(old_spte); | |
530 | if (!shadow_accessed_mask || old_spte & shadow_accessed_mask) | |
531 | kvm_set_pfn_accessed(pfn); | |
532 | if (!shadow_dirty_mask || (old_spte & shadow_dirty_mask)) | |
533 | kvm_set_pfn_dirty(pfn); | |
534 | return 1; | |
535 | } | |
536 | ||
537 | /* | |
538 | * Rules for using mmu_spte_clear_no_track: | |
539 | * Directly clear spte without caring the state bits of sptep, | |
540 | * it is used to set the upper level spte. | |
541 | */ | |
542 | static void mmu_spte_clear_no_track(u64 *sptep) | |
543 | { | |
603e0651 | 544 | __update_clear_spte_fast(sptep, 0ull); |
1df9f2dc XG |
545 | } |
546 | ||
c2a2ac2b XG |
547 | static u64 mmu_spte_get_lockless(u64 *sptep) |
548 | { | |
549 | return __get_spte_lockless(sptep); | |
550 | } | |
551 | ||
552 | static void walk_shadow_page_lockless_begin(struct kvm_vcpu *vcpu) | |
553 | { | |
c142786c AK |
554 | /* |
555 | * Prevent page table teardown by making any free-er wait during | |
556 | * kvm_flush_remote_tlbs() IPI to all active vcpus. | |
557 | */ | |
558 | local_irq_disable(); | |
559 | vcpu->mode = READING_SHADOW_PAGE_TABLES; | |
560 | /* | |
561 | * Make sure a following spte read is not reordered ahead of the write | |
562 | * to vcpu->mode. | |
563 | */ | |
564 | smp_mb(); | |
c2a2ac2b XG |
565 | } |
566 | ||
567 | static void walk_shadow_page_lockless_end(struct kvm_vcpu *vcpu) | |
568 | { | |
c142786c AK |
569 | /* |
570 | * Make sure the write to vcpu->mode is not reordered in front of | |
571 | * reads to sptes. If it does, kvm_commit_zap_page() can see us | |
572 | * OUTSIDE_GUEST_MODE and proceed to free the shadow page table. | |
573 | */ | |
574 | smp_mb(); | |
575 | vcpu->mode = OUTSIDE_GUEST_MODE; | |
576 | local_irq_enable(); | |
c2a2ac2b XG |
577 | } |
578 | ||
e2dec939 | 579 | static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache, |
2e3e5882 | 580 | struct kmem_cache *base_cache, int min) |
714b93da AK |
581 | { |
582 | void *obj; | |
583 | ||
584 | if (cache->nobjs >= min) | |
e2dec939 | 585 | return 0; |
714b93da | 586 | while (cache->nobjs < ARRAY_SIZE(cache->objects)) { |
2e3e5882 | 587 | obj = kmem_cache_zalloc(base_cache, GFP_KERNEL); |
714b93da | 588 | if (!obj) |
e2dec939 | 589 | return -ENOMEM; |
714b93da AK |
590 | cache->objects[cache->nobjs++] = obj; |
591 | } | |
e2dec939 | 592 | return 0; |
714b93da AK |
593 | } |
594 | ||
f759e2b4 XG |
595 | static int mmu_memory_cache_free_objects(struct kvm_mmu_memory_cache *cache) |
596 | { | |
597 | return cache->nobjs; | |
598 | } | |
599 | ||
e8ad9a70 XG |
600 | static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc, |
601 | struct kmem_cache *cache) | |
714b93da AK |
602 | { |
603 | while (mc->nobjs) | |
e8ad9a70 | 604 | kmem_cache_free(cache, mc->objects[--mc->nobjs]); |
714b93da AK |
605 | } |
606 | ||
c1158e63 | 607 | static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache *cache, |
2e3e5882 | 608 | int min) |
c1158e63 | 609 | { |
842f22ed | 610 | void *page; |
c1158e63 AK |
611 | |
612 | if (cache->nobjs >= min) | |
613 | return 0; | |
614 | while (cache->nobjs < ARRAY_SIZE(cache->objects)) { | |
842f22ed | 615 | page = (void *)__get_free_page(GFP_KERNEL); |
c1158e63 AK |
616 | if (!page) |
617 | return -ENOMEM; | |
842f22ed | 618 | cache->objects[cache->nobjs++] = page; |
c1158e63 AK |
619 | } |
620 | return 0; | |
621 | } | |
622 | ||
623 | static void mmu_free_memory_cache_page(struct kvm_mmu_memory_cache *mc) | |
624 | { | |
625 | while (mc->nobjs) | |
c4d198d5 | 626 | free_page((unsigned long)mc->objects[--mc->nobjs]); |
c1158e63 AK |
627 | } |
628 | ||
2e3e5882 | 629 | static int mmu_topup_memory_caches(struct kvm_vcpu *vcpu) |
714b93da | 630 | { |
e2dec939 AK |
631 | int r; |
632 | ||
53c07b18 | 633 | r = mmu_topup_memory_cache(&vcpu->arch.mmu_pte_list_desc_cache, |
67052b35 | 634 | pte_list_desc_cache, 8 + PTE_PREFETCH_NUM); |
d3d25b04 AK |
635 | if (r) |
636 | goto out; | |
ad312c7c | 637 | r = mmu_topup_memory_cache_page(&vcpu->arch.mmu_page_cache, 8); |
d3d25b04 AK |
638 | if (r) |
639 | goto out; | |
ad312c7c | 640 | r = mmu_topup_memory_cache(&vcpu->arch.mmu_page_header_cache, |
2e3e5882 | 641 | mmu_page_header_cache, 4); |
e2dec939 AK |
642 | out: |
643 | return r; | |
714b93da AK |
644 | } |
645 | ||
646 | static void mmu_free_memory_caches(struct kvm_vcpu *vcpu) | |
647 | { | |
53c07b18 XG |
648 | mmu_free_memory_cache(&vcpu->arch.mmu_pte_list_desc_cache, |
649 | pte_list_desc_cache); | |
ad312c7c | 650 | mmu_free_memory_cache_page(&vcpu->arch.mmu_page_cache); |
e8ad9a70 XG |
651 | mmu_free_memory_cache(&vcpu->arch.mmu_page_header_cache, |
652 | mmu_page_header_cache); | |
714b93da AK |
653 | } |
654 | ||
80feb89a | 655 | static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc) |
714b93da AK |
656 | { |
657 | void *p; | |
658 | ||
659 | BUG_ON(!mc->nobjs); | |
660 | p = mc->objects[--mc->nobjs]; | |
714b93da AK |
661 | return p; |
662 | } | |
663 | ||
53c07b18 | 664 | static struct pte_list_desc *mmu_alloc_pte_list_desc(struct kvm_vcpu *vcpu) |
714b93da | 665 | { |
80feb89a | 666 | return mmu_memory_cache_alloc(&vcpu->arch.mmu_pte_list_desc_cache); |
714b93da AK |
667 | } |
668 | ||
53c07b18 | 669 | static void mmu_free_pte_list_desc(struct pte_list_desc *pte_list_desc) |
714b93da | 670 | { |
53c07b18 | 671 | kmem_cache_free(pte_list_desc_cache, pte_list_desc); |
714b93da AK |
672 | } |
673 | ||
2032a93d LJ |
674 | static gfn_t kvm_mmu_page_get_gfn(struct kvm_mmu_page *sp, int index) |
675 | { | |
676 | if (!sp->role.direct) | |
677 | return sp->gfns[index]; | |
678 | ||
679 | return sp->gfn + (index << ((sp->role.level - 1) * PT64_LEVEL_BITS)); | |
680 | } | |
681 | ||
682 | static void kvm_mmu_page_set_gfn(struct kvm_mmu_page *sp, int index, gfn_t gfn) | |
683 | { | |
684 | if (sp->role.direct) | |
685 | BUG_ON(gfn != kvm_mmu_page_get_gfn(sp, index)); | |
686 | else | |
687 | sp->gfns[index] = gfn; | |
688 | } | |
689 | ||
05da4558 | 690 | /* |
d4dbf470 TY |
691 | * Return the pointer to the large page information for a given gfn, |
692 | * handling slots that are not large page aligned. | |
05da4558 | 693 | */ |
d4dbf470 TY |
694 | static struct kvm_lpage_info *lpage_info_slot(gfn_t gfn, |
695 | struct kvm_memory_slot *slot, | |
696 | int level) | |
05da4558 MT |
697 | { |
698 | unsigned long idx; | |
699 | ||
fb03cb6f | 700 | idx = gfn_to_index(gfn, slot->base_gfn, level); |
db3fe4eb | 701 | return &slot->arch.lpage_info[level - 2][idx]; |
05da4558 MT |
702 | } |
703 | ||
704 | static void account_shadowed(struct kvm *kvm, gfn_t gfn) | |
705 | { | |
d25797b2 | 706 | struct kvm_memory_slot *slot; |
d4dbf470 | 707 | struct kvm_lpage_info *linfo; |
d25797b2 | 708 | int i; |
05da4558 | 709 | |
a1f4d395 | 710 | slot = gfn_to_memslot(kvm, gfn); |
d25797b2 JR |
711 | for (i = PT_DIRECTORY_LEVEL; |
712 | i < PT_PAGE_TABLE_LEVEL + KVM_NR_PAGE_SIZES; ++i) { | |
d4dbf470 TY |
713 | linfo = lpage_info_slot(gfn, slot, i); |
714 | linfo->write_count += 1; | |
d25797b2 | 715 | } |
332b207d | 716 | kvm->arch.indirect_shadow_pages++; |
05da4558 MT |
717 | } |
718 | ||
719 | static void unaccount_shadowed(struct kvm *kvm, gfn_t gfn) | |
720 | { | |
d25797b2 | 721 | struct kvm_memory_slot *slot; |
d4dbf470 | 722 | struct kvm_lpage_info *linfo; |
d25797b2 | 723 | int i; |
05da4558 | 724 | |
a1f4d395 | 725 | slot = gfn_to_memslot(kvm, gfn); |
d25797b2 JR |
726 | for (i = PT_DIRECTORY_LEVEL; |
727 | i < PT_PAGE_TABLE_LEVEL + KVM_NR_PAGE_SIZES; ++i) { | |
d4dbf470 TY |
728 | linfo = lpage_info_slot(gfn, slot, i); |
729 | linfo->write_count -= 1; | |
730 | WARN_ON(linfo->write_count < 0); | |
d25797b2 | 731 | } |
332b207d | 732 | kvm->arch.indirect_shadow_pages--; |
05da4558 MT |
733 | } |
734 | ||
d25797b2 JR |
735 | static int has_wrprotected_page(struct kvm *kvm, |
736 | gfn_t gfn, | |
737 | int level) | |
05da4558 | 738 | { |
2843099f | 739 | struct kvm_memory_slot *slot; |
d4dbf470 | 740 | struct kvm_lpage_info *linfo; |
05da4558 | 741 | |
a1f4d395 | 742 | slot = gfn_to_memslot(kvm, gfn); |
05da4558 | 743 | if (slot) { |
d4dbf470 TY |
744 | linfo = lpage_info_slot(gfn, slot, level); |
745 | return linfo->write_count; | |
05da4558 MT |
746 | } |
747 | ||
748 | return 1; | |
749 | } | |
750 | ||
d25797b2 | 751 | static int host_mapping_level(struct kvm *kvm, gfn_t gfn) |
05da4558 | 752 | { |
8f0b1ab6 | 753 | unsigned long page_size; |
d25797b2 | 754 | int i, ret = 0; |
05da4558 | 755 | |
8f0b1ab6 | 756 | page_size = kvm_host_page_size(kvm, gfn); |
05da4558 | 757 | |
d25797b2 JR |
758 | for (i = PT_PAGE_TABLE_LEVEL; |
759 | i < (PT_PAGE_TABLE_LEVEL + KVM_NR_PAGE_SIZES); ++i) { | |
760 | if (page_size >= KVM_HPAGE_SIZE(i)) | |
761 | ret = i; | |
762 | else | |
763 | break; | |
764 | } | |
765 | ||
4c2155ce | 766 | return ret; |
05da4558 MT |
767 | } |
768 | ||
5d163b1c XG |
769 | static struct kvm_memory_slot * |
770 | gfn_to_memslot_dirty_bitmap(struct kvm_vcpu *vcpu, gfn_t gfn, | |
771 | bool no_dirty_log) | |
05da4558 MT |
772 | { |
773 | struct kvm_memory_slot *slot; | |
5d163b1c XG |
774 | |
775 | slot = gfn_to_memslot(vcpu->kvm, gfn); | |
776 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID || | |
777 | (no_dirty_log && slot->dirty_bitmap)) | |
778 | slot = NULL; | |
779 | ||
780 | return slot; | |
781 | } | |
782 | ||
783 | static bool mapping_level_dirty_bitmap(struct kvm_vcpu *vcpu, gfn_t large_gfn) | |
784 | { | |
a0a8eaba | 785 | return !gfn_to_memslot_dirty_bitmap(vcpu, large_gfn, true); |
936a5fe6 AA |
786 | } |
787 | ||
788 | static int mapping_level(struct kvm_vcpu *vcpu, gfn_t large_gfn) | |
789 | { | |
790 | int host_level, level, max_level; | |
05da4558 | 791 | |
d25797b2 JR |
792 | host_level = host_mapping_level(vcpu->kvm, large_gfn); |
793 | ||
794 | if (host_level == PT_PAGE_TABLE_LEVEL) | |
795 | return host_level; | |
796 | ||
878403b7 SY |
797 | max_level = kvm_x86_ops->get_lpage_level() < host_level ? |
798 | kvm_x86_ops->get_lpage_level() : host_level; | |
799 | ||
800 | for (level = PT_DIRECTORY_LEVEL; level <= max_level; ++level) | |
d25797b2 JR |
801 | if (has_wrprotected_page(vcpu->kvm, large_gfn, level)) |
802 | break; | |
d25797b2 JR |
803 | |
804 | return level - 1; | |
05da4558 MT |
805 | } |
806 | ||
290fc38d | 807 | /* |
53c07b18 | 808 | * Pte mapping structures: |
cd4a4e53 | 809 | * |
53c07b18 | 810 | * If pte_list bit zero is zero, then pte_list point to the spte. |
cd4a4e53 | 811 | * |
53c07b18 XG |
812 | * If pte_list bit zero is one, (then pte_list & ~1) points to a struct |
813 | * pte_list_desc containing more mappings. | |
53a27b39 | 814 | * |
53c07b18 | 815 | * Returns the number of pte entries before the spte was added or zero if |
53a27b39 MT |
816 | * the spte was not added. |
817 | * | |
cd4a4e53 | 818 | */ |
53c07b18 XG |
819 | static int pte_list_add(struct kvm_vcpu *vcpu, u64 *spte, |
820 | unsigned long *pte_list) | |
cd4a4e53 | 821 | { |
53c07b18 | 822 | struct pte_list_desc *desc; |
53a27b39 | 823 | int i, count = 0; |
cd4a4e53 | 824 | |
53c07b18 XG |
825 | if (!*pte_list) { |
826 | rmap_printk("pte_list_add: %p %llx 0->1\n", spte, *spte); | |
827 | *pte_list = (unsigned long)spte; | |
828 | } else if (!(*pte_list & 1)) { | |
829 | rmap_printk("pte_list_add: %p %llx 1->many\n", spte, *spte); | |
830 | desc = mmu_alloc_pte_list_desc(vcpu); | |
831 | desc->sptes[0] = (u64 *)*pte_list; | |
d555c333 | 832 | desc->sptes[1] = spte; |
53c07b18 | 833 | *pte_list = (unsigned long)desc | 1; |
cb16a7b3 | 834 | ++count; |
cd4a4e53 | 835 | } else { |
53c07b18 XG |
836 | rmap_printk("pte_list_add: %p %llx many->many\n", spte, *spte); |
837 | desc = (struct pte_list_desc *)(*pte_list & ~1ul); | |
838 | while (desc->sptes[PTE_LIST_EXT-1] && desc->more) { | |
cd4a4e53 | 839 | desc = desc->more; |
53c07b18 | 840 | count += PTE_LIST_EXT; |
53a27b39 | 841 | } |
53c07b18 XG |
842 | if (desc->sptes[PTE_LIST_EXT-1]) { |
843 | desc->more = mmu_alloc_pte_list_desc(vcpu); | |
cd4a4e53 AK |
844 | desc = desc->more; |
845 | } | |
d555c333 | 846 | for (i = 0; desc->sptes[i]; ++i) |
cb16a7b3 | 847 | ++count; |
d555c333 | 848 | desc->sptes[i] = spte; |
cd4a4e53 | 849 | } |
53a27b39 | 850 | return count; |
cd4a4e53 AK |
851 | } |
852 | ||
53c07b18 XG |
853 | static void |
854 | pte_list_desc_remove_entry(unsigned long *pte_list, struct pte_list_desc *desc, | |
855 | int i, struct pte_list_desc *prev_desc) | |
cd4a4e53 AK |
856 | { |
857 | int j; | |
858 | ||
53c07b18 | 859 | for (j = PTE_LIST_EXT - 1; !desc->sptes[j] && j > i; --j) |
cd4a4e53 | 860 | ; |
d555c333 AK |
861 | desc->sptes[i] = desc->sptes[j]; |
862 | desc->sptes[j] = NULL; | |
cd4a4e53 AK |
863 | if (j != 0) |
864 | return; | |
865 | if (!prev_desc && !desc->more) | |
53c07b18 | 866 | *pte_list = (unsigned long)desc->sptes[0]; |
cd4a4e53 AK |
867 | else |
868 | if (prev_desc) | |
869 | prev_desc->more = desc->more; | |
870 | else | |
53c07b18 XG |
871 | *pte_list = (unsigned long)desc->more | 1; |
872 | mmu_free_pte_list_desc(desc); | |
cd4a4e53 AK |
873 | } |
874 | ||
53c07b18 | 875 | static void pte_list_remove(u64 *spte, unsigned long *pte_list) |
cd4a4e53 | 876 | { |
53c07b18 XG |
877 | struct pte_list_desc *desc; |
878 | struct pte_list_desc *prev_desc; | |
cd4a4e53 AK |
879 | int i; |
880 | ||
53c07b18 XG |
881 | if (!*pte_list) { |
882 | printk(KERN_ERR "pte_list_remove: %p 0->BUG\n", spte); | |
cd4a4e53 | 883 | BUG(); |
53c07b18 XG |
884 | } else if (!(*pte_list & 1)) { |
885 | rmap_printk("pte_list_remove: %p 1->0\n", spte); | |
886 | if ((u64 *)*pte_list != spte) { | |
887 | printk(KERN_ERR "pte_list_remove: %p 1->BUG\n", spte); | |
cd4a4e53 AK |
888 | BUG(); |
889 | } | |
53c07b18 | 890 | *pte_list = 0; |
cd4a4e53 | 891 | } else { |
53c07b18 XG |
892 | rmap_printk("pte_list_remove: %p many->many\n", spte); |
893 | desc = (struct pte_list_desc *)(*pte_list & ~1ul); | |
cd4a4e53 AK |
894 | prev_desc = NULL; |
895 | while (desc) { | |
53c07b18 | 896 | for (i = 0; i < PTE_LIST_EXT && desc->sptes[i]; ++i) |
d555c333 | 897 | if (desc->sptes[i] == spte) { |
53c07b18 | 898 | pte_list_desc_remove_entry(pte_list, |
714b93da | 899 | desc, i, |
cd4a4e53 AK |
900 | prev_desc); |
901 | return; | |
902 | } | |
903 | prev_desc = desc; | |
904 | desc = desc->more; | |
905 | } | |
53c07b18 | 906 | pr_err("pte_list_remove: %p many->many\n", spte); |
cd4a4e53 AK |
907 | BUG(); |
908 | } | |
909 | } | |
910 | ||
67052b35 XG |
911 | typedef void (*pte_list_walk_fn) (u64 *spte); |
912 | static void pte_list_walk(unsigned long *pte_list, pte_list_walk_fn fn) | |
913 | { | |
914 | struct pte_list_desc *desc; | |
915 | int i; | |
916 | ||
917 | if (!*pte_list) | |
918 | return; | |
919 | ||
920 | if (!(*pte_list & 1)) | |
921 | return fn((u64 *)*pte_list); | |
922 | ||
923 | desc = (struct pte_list_desc *)(*pte_list & ~1ul); | |
924 | while (desc) { | |
925 | for (i = 0; i < PTE_LIST_EXT && desc->sptes[i]; ++i) | |
926 | fn(desc->sptes[i]); | |
927 | desc = desc->more; | |
928 | } | |
929 | } | |
930 | ||
9373e2c0 | 931 | static unsigned long *__gfn_to_rmap(gfn_t gfn, int level, |
9b9b1492 | 932 | struct kvm_memory_slot *slot) |
53c07b18 | 933 | { |
53c07b18 XG |
934 | struct kvm_lpage_info *linfo; |
935 | ||
53c07b18 XG |
936 | if (likely(level == PT_PAGE_TABLE_LEVEL)) |
937 | return &slot->rmap[gfn - slot->base_gfn]; | |
938 | ||
939 | linfo = lpage_info_slot(gfn, slot, level); | |
53c07b18 XG |
940 | return &linfo->rmap_pde; |
941 | } | |
942 | ||
9b9b1492 TY |
943 | /* |
944 | * Take gfn and return the reverse mapping to it. | |
945 | */ | |
946 | static unsigned long *gfn_to_rmap(struct kvm *kvm, gfn_t gfn, int level) | |
947 | { | |
948 | struct kvm_memory_slot *slot; | |
949 | ||
950 | slot = gfn_to_memslot(kvm, gfn); | |
9373e2c0 | 951 | return __gfn_to_rmap(gfn, level, slot); |
9b9b1492 TY |
952 | } |
953 | ||
f759e2b4 XG |
954 | static bool rmap_can_add(struct kvm_vcpu *vcpu) |
955 | { | |
956 | struct kvm_mmu_memory_cache *cache; | |
957 | ||
958 | cache = &vcpu->arch.mmu_pte_list_desc_cache; | |
959 | return mmu_memory_cache_free_objects(cache); | |
960 | } | |
961 | ||
53c07b18 XG |
962 | static int rmap_add(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn) |
963 | { | |
964 | struct kvm_mmu_page *sp; | |
965 | unsigned long *rmapp; | |
966 | ||
53c07b18 XG |
967 | sp = page_header(__pa(spte)); |
968 | kvm_mmu_page_set_gfn(sp, spte - sp->spt, gfn); | |
969 | rmapp = gfn_to_rmap(vcpu->kvm, gfn, sp->role.level); | |
970 | return pte_list_add(vcpu, spte, rmapp); | |
971 | } | |
972 | ||
53c07b18 XG |
973 | static void rmap_remove(struct kvm *kvm, u64 *spte) |
974 | { | |
975 | struct kvm_mmu_page *sp; | |
976 | gfn_t gfn; | |
977 | unsigned long *rmapp; | |
978 | ||
979 | sp = page_header(__pa(spte)); | |
980 | gfn = kvm_mmu_page_get_gfn(sp, spte - sp->spt); | |
981 | rmapp = gfn_to_rmap(kvm, gfn, sp->role.level); | |
982 | pte_list_remove(spte, rmapp); | |
983 | } | |
984 | ||
1e3f42f0 TY |
985 | /* |
986 | * Used by the following functions to iterate through the sptes linked by a | |
987 | * rmap. All fields are private and not assumed to be used outside. | |
988 | */ | |
989 | struct rmap_iterator { | |
990 | /* private fields */ | |
991 | struct pte_list_desc *desc; /* holds the sptep if not NULL */ | |
992 | int pos; /* index of the sptep */ | |
993 | }; | |
994 | ||
995 | /* | |
996 | * Iteration must be started by this function. This should also be used after | |
997 | * removing/dropping sptes from the rmap link because in such cases the | |
998 | * information in the itererator may not be valid. | |
999 | * | |
1000 | * Returns sptep if found, NULL otherwise. | |
1001 | */ | |
1002 | static u64 *rmap_get_first(unsigned long rmap, struct rmap_iterator *iter) | |
1003 | { | |
1004 | if (!rmap) | |
1005 | return NULL; | |
1006 | ||
1007 | if (!(rmap & 1)) { | |
1008 | iter->desc = NULL; | |
1009 | return (u64 *)rmap; | |
1010 | } | |
1011 | ||
1012 | iter->desc = (struct pte_list_desc *)(rmap & ~1ul); | |
1013 | iter->pos = 0; | |
1014 | return iter->desc->sptes[iter->pos]; | |
1015 | } | |
1016 | ||
1017 | /* | |
1018 | * Must be used with a valid iterator: e.g. after rmap_get_first(). | |
1019 | * | |
1020 | * Returns sptep if found, NULL otherwise. | |
1021 | */ | |
1022 | static u64 *rmap_get_next(struct rmap_iterator *iter) | |
1023 | { | |
1024 | if (iter->desc) { | |
1025 | if (iter->pos < PTE_LIST_EXT - 1) { | |
1026 | u64 *sptep; | |
1027 | ||
1028 | ++iter->pos; | |
1029 | sptep = iter->desc->sptes[iter->pos]; | |
1030 | if (sptep) | |
1031 | return sptep; | |
1032 | } | |
1033 | ||
1034 | iter->desc = iter->desc->more; | |
1035 | ||
1036 | if (iter->desc) { | |
1037 | iter->pos = 0; | |
1038 | /* desc->sptes[0] cannot be NULL */ | |
1039 | return iter->desc->sptes[iter->pos]; | |
1040 | } | |
1041 | } | |
1042 | ||
1043 | return NULL; | |
1044 | } | |
1045 | ||
c3707958 | 1046 | static void drop_spte(struct kvm *kvm, u64 *sptep) |
e4b502ea | 1047 | { |
1df9f2dc | 1048 | if (mmu_spte_clear_track_bits(sptep)) |
eb45fda4 | 1049 | rmap_remove(kvm, sptep); |
be38d276 AK |
1050 | } |
1051 | ||
a0ed4607 | 1052 | static int __rmap_write_protect(struct kvm *kvm, unsigned long *rmapp, int level) |
98348e95 | 1053 | { |
1e3f42f0 TY |
1054 | u64 *sptep; |
1055 | struct rmap_iterator iter; | |
a0ed4607 | 1056 | int write_protected = 0; |
374cbac0 | 1057 | |
1e3f42f0 TY |
1058 | for (sptep = rmap_get_first(*rmapp, &iter); sptep;) { |
1059 | BUG_ON(!(*sptep & PT_PRESENT_MASK)); | |
1060 | rmap_printk("rmap_write_protect: spte %p %llx\n", sptep, *sptep); | |
a0ed4607 | 1061 | |
1e3f42f0 TY |
1062 | if (!is_writable_pte(*sptep)) { |
1063 | sptep = rmap_get_next(&iter); | |
a0ed4607 | 1064 | continue; |
1e3f42f0 | 1065 | } |
a0ed4607 TY |
1066 | |
1067 | if (level == PT_PAGE_TABLE_LEVEL) { | |
1e3f42f0 TY |
1068 | mmu_spte_update(sptep, *sptep & ~PT_WRITABLE_MASK); |
1069 | sptep = rmap_get_next(&iter); | |
a0ed4607 | 1070 | } else { |
1e3f42f0 TY |
1071 | BUG_ON(!is_large_pte(*sptep)); |
1072 | drop_spte(kvm, sptep); | |
a0ed4607 | 1073 | --kvm->stat.lpages; |
1e3f42f0 | 1074 | sptep = rmap_get_first(*rmapp, &iter); |
caa5b8a5 | 1075 | } |
a0ed4607 TY |
1076 | |
1077 | write_protected = 1; | |
374cbac0 | 1078 | } |
855149aa | 1079 | |
a0ed4607 TY |
1080 | return write_protected; |
1081 | } | |
1082 | ||
5dc99b23 TY |
1083 | /** |
1084 | * kvm_mmu_write_protect_pt_masked - write protect selected PT level pages | |
1085 | * @kvm: kvm instance | |
1086 | * @slot: slot to protect | |
1087 | * @gfn_offset: start of the BITS_PER_LONG pages we care about | |
1088 | * @mask: indicates which pages we should protect | |
1089 | * | |
1090 | * Used when we do not need to care about huge page mappings: e.g. during dirty | |
1091 | * logging we do not have any such mappings. | |
1092 | */ | |
1093 | void kvm_mmu_write_protect_pt_masked(struct kvm *kvm, | |
1094 | struct kvm_memory_slot *slot, | |
1095 | gfn_t gfn_offset, unsigned long mask) | |
a0ed4607 TY |
1096 | { |
1097 | unsigned long *rmapp; | |
a0ed4607 | 1098 | |
5dc99b23 TY |
1099 | while (mask) { |
1100 | rmapp = &slot->rmap[gfn_offset + __ffs(mask)]; | |
1101 | __rmap_write_protect(kvm, rmapp, PT_PAGE_TABLE_LEVEL); | |
05da4558 | 1102 | |
5dc99b23 TY |
1103 | /* clear the first set bit */ |
1104 | mask &= mask - 1; | |
1105 | } | |
374cbac0 AK |
1106 | } |
1107 | ||
95d4c16c TY |
1108 | static int rmap_write_protect(struct kvm *kvm, u64 gfn) |
1109 | { | |
1110 | struct kvm_memory_slot *slot; | |
5dc99b23 TY |
1111 | unsigned long *rmapp; |
1112 | int i; | |
1113 | int write_protected = 0; | |
95d4c16c TY |
1114 | |
1115 | slot = gfn_to_memslot(kvm, gfn); | |
5dc99b23 TY |
1116 | |
1117 | for (i = PT_PAGE_TABLE_LEVEL; | |
1118 | i < PT_PAGE_TABLE_LEVEL + KVM_NR_PAGE_SIZES; ++i) { | |
1119 | rmapp = __gfn_to_rmap(gfn, i, slot); | |
1120 | write_protected |= __rmap_write_protect(kvm, rmapp, i); | |
1121 | } | |
1122 | ||
1123 | return write_protected; | |
95d4c16c TY |
1124 | } |
1125 | ||
8a8365c5 FD |
1126 | static int kvm_unmap_rmapp(struct kvm *kvm, unsigned long *rmapp, |
1127 | unsigned long data) | |
e930bffe | 1128 | { |
1e3f42f0 TY |
1129 | u64 *sptep; |
1130 | struct rmap_iterator iter; | |
e930bffe AA |
1131 | int need_tlb_flush = 0; |
1132 | ||
1e3f42f0 TY |
1133 | while ((sptep = rmap_get_first(*rmapp, &iter))) { |
1134 | BUG_ON(!(*sptep & PT_PRESENT_MASK)); | |
1135 | rmap_printk("kvm_rmap_unmap_hva: spte %p %llx\n", sptep, *sptep); | |
1136 | ||
1137 | drop_spte(kvm, sptep); | |
e930bffe AA |
1138 | need_tlb_flush = 1; |
1139 | } | |
1e3f42f0 | 1140 | |
e930bffe AA |
1141 | return need_tlb_flush; |
1142 | } | |
1143 | ||
8a8365c5 FD |
1144 | static int kvm_set_pte_rmapp(struct kvm *kvm, unsigned long *rmapp, |
1145 | unsigned long data) | |
3da0dd43 | 1146 | { |
1e3f42f0 TY |
1147 | u64 *sptep; |
1148 | struct rmap_iterator iter; | |
3da0dd43 | 1149 | int need_flush = 0; |
1e3f42f0 | 1150 | u64 new_spte; |
3da0dd43 IE |
1151 | pte_t *ptep = (pte_t *)data; |
1152 | pfn_t new_pfn; | |
1153 | ||
1154 | WARN_ON(pte_huge(*ptep)); | |
1155 | new_pfn = pte_pfn(*ptep); | |
1e3f42f0 TY |
1156 | |
1157 | for (sptep = rmap_get_first(*rmapp, &iter); sptep;) { | |
1158 | BUG_ON(!is_shadow_present_pte(*sptep)); | |
1159 | rmap_printk("kvm_set_pte_rmapp: spte %p %llx\n", sptep, *sptep); | |
1160 | ||
3da0dd43 | 1161 | need_flush = 1; |
1e3f42f0 | 1162 | |
3da0dd43 | 1163 | if (pte_write(*ptep)) { |
1e3f42f0 TY |
1164 | drop_spte(kvm, sptep); |
1165 | sptep = rmap_get_first(*rmapp, &iter); | |
3da0dd43 | 1166 | } else { |
1e3f42f0 | 1167 | new_spte = *sptep & ~PT64_BASE_ADDR_MASK; |
3da0dd43 IE |
1168 | new_spte |= (u64)new_pfn << PAGE_SHIFT; |
1169 | ||
1170 | new_spte &= ~PT_WRITABLE_MASK; | |
1171 | new_spte &= ~SPTE_HOST_WRITEABLE; | |
b79b93f9 | 1172 | new_spte &= ~shadow_accessed_mask; |
1e3f42f0 TY |
1173 | |
1174 | mmu_spte_clear_track_bits(sptep); | |
1175 | mmu_spte_set(sptep, new_spte); | |
1176 | sptep = rmap_get_next(&iter); | |
3da0dd43 IE |
1177 | } |
1178 | } | |
1e3f42f0 | 1179 | |
3da0dd43 IE |
1180 | if (need_flush) |
1181 | kvm_flush_remote_tlbs(kvm); | |
1182 | ||
1183 | return 0; | |
1184 | } | |
1185 | ||
8a8365c5 FD |
1186 | static int kvm_handle_hva(struct kvm *kvm, unsigned long hva, |
1187 | unsigned long data, | |
3da0dd43 | 1188 | int (*handler)(struct kvm *kvm, unsigned long *rmapp, |
8a8365c5 | 1189 | unsigned long data)) |
e930bffe | 1190 | { |
be6ba0f0 | 1191 | int j; |
90bb6fc5 | 1192 | int ret; |
e930bffe | 1193 | int retval = 0; |
bc6678a3 | 1194 | struct kvm_memslots *slots; |
be6ba0f0 | 1195 | struct kvm_memory_slot *memslot; |
bc6678a3 | 1196 | |
90d83dc3 | 1197 | slots = kvm_memslots(kvm); |
e930bffe | 1198 | |
be6ba0f0 | 1199 | kvm_for_each_memslot(memslot, slots) { |
e930bffe AA |
1200 | unsigned long start = memslot->userspace_addr; |
1201 | unsigned long end; | |
1202 | ||
e930bffe AA |
1203 | end = start + (memslot->npages << PAGE_SHIFT); |
1204 | if (hva >= start && hva < end) { | |
1205 | gfn_t gfn_offset = (hva - start) >> PAGE_SHIFT; | |
d4dbf470 | 1206 | gfn_t gfn = memslot->base_gfn + gfn_offset; |
852e3c19 | 1207 | |
90bb6fc5 | 1208 | ret = handler(kvm, &memslot->rmap[gfn_offset], data); |
852e3c19 JR |
1209 | |
1210 | for (j = 0; j < KVM_NR_PAGE_SIZES - 1; ++j) { | |
d4dbf470 TY |
1211 | struct kvm_lpage_info *linfo; |
1212 | ||
1213 | linfo = lpage_info_slot(gfn, memslot, | |
1214 | PT_DIRECTORY_LEVEL + j); | |
1215 | ret |= handler(kvm, &linfo->rmap_pde, data); | |
852e3c19 | 1216 | } |
90bb6fc5 AK |
1217 | trace_kvm_age_page(hva, memslot, ret); |
1218 | retval |= ret; | |
e930bffe AA |
1219 | } |
1220 | } | |
1221 | ||
1222 | return retval; | |
1223 | } | |
1224 | ||
1225 | int kvm_unmap_hva(struct kvm *kvm, unsigned long hva) | |
1226 | { | |
3da0dd43 IE |
1227 | return kvm_handle_hva(kvm, hva, 0, kvm_unmap_rmapp); |
1228 | } | |
1229 | ||
1230 | void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte) | |
1231 | { | |
8a8365c5 | 1232 | kvm_handle_hva(kvm, hva, (unsigned long)&pte, kvm_set_pte_rmapp); |
e930bffe AA |
1233 | } |
1234 | ||
8a8365c5 FD |
1235 | static int kvm_age_rmapp(struct kvm *kvm, unsigned long *rmapp, |
1236 | unsigned long data) | |
e930bffe | 1237 | { |
1e3f42f0 | 1238 | u64 *sptep; |
79f702a6 | 1239 | struct rmap_iterator uninitialized_var(iter); |
e930bffe AA |
1240 | int young = 0; |
1241 | ||
6316e1c8 | 1242 | /* |
3f6d8c8a XH |
1243 | * In case of absence of EPT Access and Dirty Bits supports, |
1244 | * emulate the accessed bit for EPT, by checking if this page has | |
6316e1c8 RR |
1245 | * an EPT mapping, and clearing it if it does. On the next access, |
1246 | * a new EPT mapping will be established. | |
1247 | * This has some overhead, but not as much as the cost of swapping | |
1248 | * out actively used pages or breaking up actively used hugepages. | |
1249 | */ | |
534e38b4 | 1250 | if (!shadow_accessed_mask) |
6316e1c8 | 1251 | return kvm_unmap_rmapp(kvm, rmapp, data); |
534e38b4 | 1252 | |
1e3f42f0 TY |
1253 | for (sptep = rmap_get_first(*rmapp, &iter); sptep; |
1254 | sptep = rmap_get_next(&iter)) { | |
3f6d8c8a | 1255 | BUG_ON(!is_shadow_present_pte(*sptep)); |
1e3f42f0 | 1256 | |
3f6d8c8a | 1257 | if (*sptep & shadow_accessed_mask) { |
e930bffe | 1258 | young = 1; |
3f6d8c8a XH |
1259 | clear_bit((ffs(shadow_accessed_mask) - 1), |
1260 | (unsigned long *)sptep); | |
e930bffe | 1261 | } |
e930bffe | 1262 | } |
1e3f42f0 | 1263 | |
e930bffe AA |
1264 | return young; |
1265 | } | |
1266 | ||
8ee53820 AA |
1267 | static int kvm_test_age_rmapp(struct kvm *kvm, unsigned long *rmapp, |
1268 | unsigned long data) | |
1269 | { | |
1e3f42f0 TY |
1270 | u64 *sptep; |
1271 | struct rmap_iterator iter; | |
8ee53820 AA |
1272 | int young = 0; |
1273 | ||
1274 | /* | |
1275 | * If there's no access bit in the secondary pte set by the | |
1276 | * hardware it's up to gup-fast/gup to set the access bit in | |
1277 | * the primary pte or in the page structure. | |
1278 | */ | |
1279 | if (!shadow_accessed_mask) | |
1280 | goto out; | |
1281 | ||
1e3f42f0 TY |
1282 | for (sptep = rmap_get_first(*rmapp, &iter); sptep; |
1283 | sptep = rmap_get_next(&iter)) { | |
3f6d8c8a | 1284 | BUG_ON(!is_shadow_present_pte(*sptep)); |
1e3f42f0 | 1285 | |
3f6d8c8a | 1286 | if (*sptep & shadow_accessed_mask) { |
8ee53820 AA |
1287 | young = 1; |
1288 | break; | |
1289 | } | |
8ee53820 AA |
1290 | } |
1291 | out: | |
1292 | return young; | |
1293 | } | |
1294 | ||
53a27b39 MT |
1295 | #define RMAP_RECYCLE_THRESHOLD 1000 |
1296 | ||
852e3c19 | 1297 | static void rmap_recycle(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn) |
53a27b39 MT |
1298 | { |
1299 | unsigned long *rmapp; | |
852e3c19 JR |
1300 | struct kvm_mmu_page *sp; |
1301 | ||
1302 | sp = page_header(__pa(spte)); | |
53a27b39 | 1303 | |
852e3c19 | 1304 | rmapp = gfn_to_rmap(vcpu->kvm, gfn, sp->role.level); |
53a27b39 | 1305 | |
3da0dd43 | 1306 | kvm_unmap_rmapp(vcpu->kvm, rmapp, 0); |
53a27b39 MT |
1307 | kvm_flush_remote_tlbs(vcpu->kvm); |
1308 | } | |
1309 | ||
e930bffe AA |
1310 | int kvm_age_hva(struct kvm *kvm, unsigned long hva) |
1311 | { | |
3da0dd43 | 1312 | return kvm_handle_hva(kvm, hva, 0, kvm_age_rmapp); |
e930bffe AA |
1313 | } |
1314 | ||
8ee53820 AA |
1315 | int kvm_test_age_hva(struct kvm *kvm, unsigned long hva) |
1316 | { | |
1317 | return kvm_handle_hva(kvm, hva, 0, kvm_test_age_rmapp); | |
1318 | } | |
1319 | ||
d6c69ee9 | 1320 | #ifdef MMU_DEBUG |
47ad8e68 | 1321 | static int is_empty_shadow_page(u64 *spt) |
6aa8b732 | 1322 | { |
139bdb2d AK |
1323 | u64 *pos; |
1324 | u64 *end; | |
1325 | ||
47ad8e68 | 1326 | for (pos = spt, end = pos + PAGE_SIZE / sizeof(u64); pos != end; pos++) |
3c915510 | 1327 | if (is_shadow_present_pte(*pos)) { |
b8688d51 | 1328 | printk(KERN_ERR "%s: %p %llx\n", __func__, |
139bdb2d | 1329 | pos, *pos); |
6aa8b732 | 1330 | return 0; |
139bdb2d | 1331 | } |
6aa8b732 AK |
1332 | return 1; |
1333 | } | |
d6c69ee9 | 1334 | #endif |
6aa8b732 | 1335 | |
45221ab6 DH |
1336 | /* |
1337 | * This value is the sum of all of the kvm instances's | |
1338 | * kvm->arch.n_used_mmu_pages values. We need a global, | |
1339 | * aggregate version in order to make the slab shrinker | |
1340 | * faster | |
1341 | */ | |
1342 | static inline void kvm_mod_used_mmu_pages(struct kvm *kvm, int nr) | |
1343 | { | |
1344 | kvm->arch.n_used_mmu_pages += nr; | |
1345 | percpu_counter_add(&kvm_total_used_mmu_pages, nr); | |
1346 | } | |
1347 | ||
bd4c86ea XG |
1348 | /* |
1349 | * Remove the sp from shadow page cache, after call it, | |
1350 | * we can not find this sp from the cache, and the shadow | |
1351 | * page table is still valid. | |
1352 | * It should be under the protection of mmu lock. | |
1353 | */ | |
1354 | static void kvm_mmu_isolate_page(struct kvm_mmu_page *sp) | |
260746c0 | 1355 | { |
4db35314 | 1356 | ASSERT(is_empty_shadow_page(sp->spt)); |
7775834a | 1357 | hlist_del(&sp->hash_link); |
2032a93d | 1358 | if (!sp->role.direct) |
842f22ed | 1359 | free_page((unsigned long)sp->gfns); |
bd4c86ea XG |
1360 | } |
1361 | ||
1362 | /* | |
1363 | * Free the shadow page table and the sp, we can do it | |
1364 | * out of the protection of mmu lock. | |
1365 | */ | |
1366 | static void kvm_mmu_free_page(struct kvm_mmu_page *sp) | |
1367 | { | |
1368 | list_del(&sp->link); | |
1369 | free_page((unsigned long)sp->spt); | |
e8ad9a70 | 1370 | kmem_cache_free(mmu_page_header_cache, sp); |
260746c0 AK |
1371 | } |
1372 | ||
cea0f0e7 AK |
1373 | static unsigned kvm_page_table_hashfn(gfn_t gfn) |
1374 | { | |
1ae0a13d | 1375 | return gfn & ((1 << KVM_MMU_HASH_SHIFT) - 1); |
cea0f0e7 AK |
1376 | } |
1377 | ||
714b93da | 1378 | static void mmu_page_add_parent_pte(struct kvm_vcpu *vcpu, |
4db35314 | 1379 | struct kvm_mmu_page *sp, u64 *parent_pte) |
cea0f0e7 | 1380 | { |
cea0f0e7 AK |
1381 | if (!parent_pte) |
1382 | return; | |
cea0f0e7 | 1383 | |
67052b35 | 1384 | pte_list_add(vcpu, parent_pte, &sp->parent_ptes); |
cea0f0e7 AK |
1385 | } |
1386 | ||
4db35314 | 1387 | static void mmu_page_remove_parent_pte(struct kvm_mmu_page *sp, |
cea0f0e7 AK |
1388 | u64 *parent_pte) |
1389 | { | |
67052b35 | 1390 | pte_list_remove(parent_pte, &sp->parent_ptes); |
cea0f0e7 AK |
1391 | } |
1392 | ||
bcdd9a93 XG |
1393 | static void drop_parent_pte(struct kvm_mmu_page *sp, |
1394 | u64 *parent_pte) | |
1395 | { | |
1396 | mmu_page_remove_parent_pte(sp, parent_pte); | |
1df9f2dc | 1397 | mmu_spte_clear_no_track(parent_pte); |
bcdd9a93 XG |
1398 | } |
1399 | ||
67052b35 XG |
1400 | static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu, |
1401 | u64 *parent_pte, int direct) | |
ad8cfbe3 | 1402 | { |
67052b35 | 1403 | struct kvm_mmu_page *sp; |
80feb89a TY |
1404 | sp = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_header_cache); |
1405 | sp->spt = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_cache); | |
67052b35 | 1406 | if (!direct) |
80feb89a | 1407 | sp->gfns = mmu_memory_cache_alloc(&vcpu->arch.mmu_page_cache); |
67052b35 XG |
1408 | set_page_private(virt_to_page(sp->spt), (unsigned long)sp); |
1409 | list_add(&sp->link, &vcpu->kvm->arch.active_mmu_pages); | |
93a5cef0 | 1410 | bitmap_zero(sp->slot_bitmap, KVM_MEM_SLOTS_NUM); |
67052b35 XG |
1411 | sp->parent_ptes = 0; |
1412 | mmu_page_add_parent_pte(vcpu, sp, parent_pte); | |
1413 | kvm_mod_used_mmu_pages(vcpu->kvm, +1); | |
1414 | return sp; | |
ad8cfbe3 MT |
1415 | } |
1416 | ||
67052b35 | 1417 | static void mark_unsync(u64 *spte); |
1047df1f | 1418 | static void kvm_mmu_mark_parents_unsync(struct kvm_mmu_page *sp) |
0074ff63 | 1419 | { |
67052b35 | 1420 | pte_list_walk(&sp->parent_ptes, mark_unsync); |
0074ff63 MT |
1421 | } |
1422 | ||
67052b35 | 1423 | static void mark_unsync(u64 *spte) |
0074ff63 | 1424 | { |
67052b35 | 1425 | struct kvm_mmu_page *sp; |
1047df1f | 1426 | unsigned int index; |
0074ff63 | 1427 | |
67052b35 | 1428 | sp = page_header(__pa(spte)); |
1047df1f XG |
1429 | index = spte - sp->spt; |
1430 | if (__test_and_set_bit(index, sp->unsync_child_bitmap)) | |
0074ff63 | 1431 | return; |
1047df1f | 1432 | if (sp->unsync_children++) |
0074ff63 | 1433 | return; |
1047df1f | 1434 | kvm_mmu_mark_parents_unsync(sp); |
0074ff63 MT |
1435 | } |
1436 | ||
e8bc217a | 1437 | static int nonpaging_sync_page(struct kvm_vcpu *vcpu, |
a4a8e6f7 | 1438 | struct kvm_mmu_page *sp) |
e8bc217a MT |
1439 | { |
1440 | return 1; | |
1441 | } | |
1442 | ||
a7052897 MT |
1443 | static void nonpaging_invlpg(struct kvm_vcpu *vcpu, gva_t gva) |
1444 | { | |
1445 | } | |
1446 | ||
0f53b5b1 XG |
1447 | static void nonpaging_update_pte(struct kvm_vcpu *vcpu, |
1448 | struct kvm_mmu_page *sp, u64 *spte, | |
7c562522 | 1449 | const void *pte) |
0f53b5b1 XG |
1450 | { |
1451 | WARN_ON(1); | |
1452 | } | |
1453 | ||
60c8aec6 MT |
1454 | #define KVM_PAGE_ARRAY_NR 16 |
1455 | ||
1456 | struct kvm_mmu_pages { | |
1457 | struct mmu_page_and_offset { | |
1458 | struct kvm_mmu_page *sp; | |
1459 | unsigned int idx; | |
1460 | } page[KVM_PAGE_ARRAY_NR]; | |
1461 | unsigned int nr; | |
1462 | }; | |
1463 | ||
cded19f3 HE |
1464 | static int mmu_pages_add(struct kvm_mmu_pages *pvec, struct kvm_mmu_page *sp, |
1465 | int idx) | |
4731d4c7 | 1466 | { |
60c8aec6 | 1467 | int i; |
4731d4c7 | 1468 | |
60c8aec6 MT |
1469 | if (sp->unsync) |
1470 | for (i=0; i < pvec->nr; i++) | |
1471 | if (pvec->page[i].sp == sp) | |
1472 | return 0; | |
1473 | ||
1474 | pvec->page[pvec->nr].sp = sp; | |
1475 | pvec->page[pvec->nr].idx = idx; | |
1476 | pvec->nr++; | |
1477 | return (pvec->nr == KVM_PAGE_ARRAY_NR); | |
1478 | } | |
1479 | ||
1480 | static int __mmu_unsync_walk(struct kvm_mmu_page *sp, | |
1481 | struct kvm_mmu_pages *pvec) | |
1482 | { | |
1483 | int i, ret, nr_unsync_leaf = 0; | |
4731d4c7 | 1484 | |
37178b8b | 1485 | for_each_set_bit(i, sp->unsync_child_bitmap, 512) { |
7a8f1a74 | 1486 | struct kvm_mmu_page *child; |
4731d4c7 MT |
1487 | u64 ent = sp->spt[i]; |
1488 | ||
7a8f1a74 XG |
1489 | if (!is_shadow_present_pte(ent) || is_large_pte(ent)) |
1490 | goto clear_child_bitmap; | |
1491 | ||
1492 | child = page_header(ent & PT64_BASE_ADDR_MASK); | |
1493 | ||
1494 | if (child->unsync_children) { | |
1495 | if (mmu_pages_add(pvec, child, i)) | |
1496 | return -ENOSPC; | |
1497 | ||
1498 | ret = __mmu_unsync_walk(child, pvec); | |
1499 | if (!ret) | |
1500 | goto clear_child_bitmap; | |
1501 | else if (ret > 0) | |
1502 | nr_unsync_leaf += ret; | |
1503 | else | |
1504 | return ret; | |
1505 | } else if (child->unsync) { | |
1506 | nr_unsync_leaf++; | |
1507 | if (mmu_pages_add(pvec, child, i)) | |
1508 | return -ENOSPC; | |
1509 | } else | |
1510 | goto clear_child_bitmap; | |
1511 | ||
1512 | continue; | |
1513 | ||
1514 | clear_child_bitmap: | |
1515 | __clear_bit(i, sp->unsync_child_bitmap); | |
1516 | sp->unsync_children--; | |
1517 | WARN_ON((int)sp->unsync_children < 0); | |
4731d4c7 MT |
1518 | } |
1519 | ||
4731d4c7 | 1520 | |
60c8aec6 MT |
1521 | return nr_unsync_leaf; |
1522 | } | |
1523 | ||
1524 | static int mmu_unsync_walk(struct kvm_mmu_page *sp, | |
1525 | struct kvm_mmu_pages *pvec) | |
1526 | { | |
1527 | if (!sp->unsync_children) | |
1528 | return 0; | |
1529 | ||
1530 | mmu_pages_add(pvec, sp, 0); | |
1531 | return __mmu_unsync_walk(sp, pvec); | |
4731d4c7 MT |
1532 | } |
1533 | ||
4731d4c7 MT |
1534 | static void kvm_unlink_unsync_page(struct kvm *kvm, struct kvm_mmu_page *sp) |
1535 | { | |
1536 | WARN_ON(!sp->unsync); | |
5e1b3ddb | 1537 | trace_kvm_mmu_sync_page(sp); |
4731d4c7 MT |
1538 | sp->unsync = 0; |
1539 | --kvm->stat.mmu_unsync; | |
1540 | } | |
1541 | ||
7775834a XG |
1542 | static int kvm_mmu_prepare_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp, |
1543 | struct list_head *invalid_list); | |
1544 | static void kvm_mmu_commit_zap_page(struct kvm *kvm, | |
1545 | struct list_head *invalid_list); | |
4731d4c7 | 1546 | |
f41d335a XG |
1547 | #define for_each_gfn_sp(kvm, sp, gfn, pos) \ |
1548 | hlist_for_each_entry(sp, pos, \ | |
7ae680eb XG |
1549 | &(kvm)->arch.mmu_page_hash[kvm_page_table_hashfn(gfn)], hash_link) \ |
1550 | if ((sp)->gfn != (gfn)) {} else | |
1551 | ||
f41d335a XG |
1552 | #define for_each_gfn_indirect_valid_sp(kvm, sp, gfn, pos) \ |
1553 | hlist_for_each_entry(sp, pos, \ | |
7ae680eb XG |
1554 | &(kvm)->arch.mmu_page_hash[kvm_page_table_hashfn(gfn)], hash_link) \ |
1555 | if ((sp)->gfn != (gfn) || (sp)->role.direct || \ | |
1556 | (sp)->role.invalid) {} else | |
1557 | ||
f918b443 | 1558 | /* @sp->gfn should be write-protected at the call site */ |
1d9dc7e0 | 1559 | static int __kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp, |
d98ba053 | 1560 | struct list_head *invalid_list, bool clear_unsync) |
4731d4c7 | 1561 | { |
5b7e0102 | 1562 | if (sp->role.cr4_pae != !!is_pae(vcpu)) { |
d98ba053 | 1563 | kvm_mmu_prepare_zap_page(vcpu->kvm, sp, invalid_list); |
4731d4c7 MT |
1564 | return 1; |
1565 | } | |
1566 | ||
f918b443 | 1567 | if (clear_unsync) |
1d9dc7e0 | 1568 | kvm_unlink_unsync_page(vcpu->kvm, sp); |
1d9dc7e0 | 1569 | |
a4a8e6f7 | 1570 | if (vcpu->arch.mmu.sync_page(vcpu, sp)) { |
d98ba053 | 1571 | kvm_mmu_prepare_zap_page(vcpu->kvm, sp, invalid_list); |
4731d4c7 MT |
1572 | return 1; |
1573 | } | |
1574 | ||
1575 | kvm_mmu_flush_tlb(vcpu); | |
4731d4c7 MT |
1576 | return 0; |
1577 | } | |
1578 | ||
1d9dc7e0 XG |
1579 | static int kvm_sync_page_transient(struct kvm_vcpu *vcpu, |
1580 | struct kvm_mmu_page *sp) | |
1581 | { | |
d98ba053 | 1582 | LIST_HEAD(invalid_list); |
1d9dc7e0 XG |
1583 | int ret; |
1584 | ||
d98ba053 | 1585 | ret = __kvm_sync_page(vcpu, sp, &invalid_list, false); |
be71e061 | 1586 | if (ret) |
d98ba053 XG |
1587 | kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list); |
1588 | ||
1d9dc7e0 XG |
1589 | return ret; |
1590 | } | |
1591 | ||
e37fa785 XG |
1592 | #ifdef CONFIG_KVM_MMU_AUDIT |
1593 | #include "mmu_audit.c" | |
1594 | #else | |
1595 | static void kvm_mmu_audit(struct kvm_vcpu *vcpu, int point) { } | |
1596 | static void mmu_audit_disable(void) { } | |
1597 | #endif | |
1598 | ||
d98ba053 XG |
1599 | static int kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp, |
1600 | struct list_head *invalid_list) | |
1d9dc7e0 | 1601 | { |
d98ba053 | 1602 | return __kvm_sync_page(vcpu, sp, invalid_list, true); |
1d9dc7e0 XG |
1603 | } |
1604 | ||
9f1a122f XG |
1605 | /* @gfn should be write-protected at the call site */ |
1606 | static void kvm_sync_pages(struct kvm_vcpu *vcpu, gfn_t gfn) | |
1607 | { | |
9f1a122f | 1608 | struct kvm_mmu_page *s; |
f41d335a | 1609 | struct hlist_node *node; |
d98ba053 | 1610 | LIST_HEAD(invalid_list); |
9f1a122f XG |
1611 | bool flush = false; |
1612 | ||
f41d335a | 1613 | for_each_gfn_indirect_valid_sp(vcpu->kvm, s, gfn, node) { |
7ae680eb | 1614 | if (!s->unsync) |
9f1a122f XG |
1615 | continue; |
1616 | ||
1617 | WARN_ON(s->role.level != PT_PAGE_TABLE_LEVEL); | |
a4a8e6f7 | 1618 | kvm_unlink_unsync_page(vcpu->kvm, s); |
9f1a122f | 1619 | if ((s->role.cr4_pae != !!is_pae(vcpu)) || |
a4a8e6f7 | 1620 | (vcpu->arch.mmu.sync_page(vcpu, s))) { |
d98ba053 | 1621 | kvm_mmu_prepare_zap_page(vcpu->kvm, s, &invalid_list); |
9f1a122f XG |
1622 | continue; |
1623 | } | |
9f1a122f XG |
1624 | flush = true; |
1625 | } | |
1626 | ||
d98ba053 | 1627 | kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list); |
9f1a122f XG |
1628 | if (flush) |
1629 | kvm_mmu_flush_tlb(vcpu); | |
1630 | } | |
1631 | ||
60c8aec6 MT |
1632 | struct mmu_page_path { |
1633 | struct kvm_mmu_page *parent[PT64_ROOT_LEVEL-1]; | |
1634 | unsigned int idx[PT64_ROOT_LEVEL-1]; | |
4731d4c7 MT |
1635 | }; |
1636 | ||
60c8aec6 MT |
1637 | #define for_each_sp(pvec, sp, parents, i) \ |
1638 | for (i = mmu_pages_next(&pvec, &parents, -1), \ | |
1639 | sp = pvec.page[i].sp; \ | |
1640 | i < pvec.nr && ({ sp = pvec.page[i].sp; 1;}); \ | |
1641 | i = mmu_pages_next(&pvec, &parents, i)) | |
1642 | ||
cded19f3 HE |
1643 | static int mmu_pages_next(struct kvm_mmu_pages *pvec, |
1644 | struct mmu_page_path *parents, | |
1645 | int i) | |
60c8aec6 MT |
1646 | { |
1647 | int n; | |
1648 | ||
1649 | for (n = i+1; n < pvec->nr; n++) { | |
1650 | struct kvm_mmu_page *sp = pvec->page[n].sp; | |
1651 | ||
1652 | if (sp->role.level == PT_PAGE_TABLE_LEVEL) { | |
1653 | parents->idx[0] = pvec->page[n].idx; | |
1654 | return n; | |
1655 | } | |
1656 | ||
1657 | parents->parent[sp->role.level-2] = sp; | |
1658 | parents->idx[sp->role.level-1] = pvec->page[n].idx; | |
1659 | } | |
1660 | ||
1661 | return n; | |
1662 | } | |
1663 | ||
cded19f3 | 1664 | static void mmu_pages_clear_parents(struct mmu_page_path *parents) |
4731d4c7 | 1665 | { |
60c8aec6 MT |
1666 | struct kvm_mmu_page *sp; |
1667 | unsigned int level = 0; | |
1668 | ||
1669 | do { | |
1670 | unsigned int idx = parents->idx[level]; | |
4731d4c7 | 1671 | |
60c8aec6 MT |
1672 | sp = parents->parent[level]; |
1673 | if (!sp) | |
1674 | return; | |
1675 | ||
1676 | --sp->unsync_children; | |
1677 | WARN_ON((int)sp->unsync_children < 0); | |
1678 | __clear_bit(idx, sp->unsync_child_bitmap); | |
1679 | level++; | |
1680 | } while (level < PT64_ROOT_LEVEL-1 && !sp->unsync_children); | |
4731d4c7 MT |
1681 | } |
1682 | ||
60c8aec6 MT |
1683 | static void kvm_mmu_pages_init(struct kvm_mmu_page *parent, |
1684 | struct mmu_page_path *parents, | |
1685 | struct kvm_mmu_pages *pvec) | |
4731d4c7 | 1686 | { |
60c8aec6 MT |
1687 | parents->parent[parent->role.level-1] = NULL; |
1688 | pvec->nr = 0; | |
1689 | } | |
4731d4c7 | 1690 | |
60c8aec6 MT |
1691 | static void mmu_sync_children(struct kvm_vcpu *vcpu, |
1692 | struct kvm_mmu_page *parent) | |
1693 | { | |
1694 | int i; | |
1695 | struct kvm_mmu_page *sp; | |
1696 | struct mmu_page_path parents; | |
1697 | struct kvm_mmu_pages pages; | |
d98ba053 | 1698 | LIST_HEAD(invalid_list); |
60c8aec6 MT |
1699 | |
1700 | kvm_mmu_pages_init(parent, &parents, &pages); | |
1701 | while (mmu_unsync_walk(parent, &pages)) { | |
b1a36821 MT |
1702 | int protected = 0; |
1703 | ||
1704 | for_each_sp(pages, sp, parents, i) | |
1705 | protected |= rmap_write_protect(vcpu->kvm, sp->gfn); | |
1706 | ||
1707 | if (protected) | |
1708 | kvm_flush_remote_tlbs(vcpu->kvm); | |
1709 | ||
60c8aec6 | 1710 | for_each_sp(pages, sp, parents, i) { |
d98ba053 | 1711 | kvm_sync_page(vcpu, sp, &invalid_list); |
60c8aec6 MT |
1712 | mmu_pages_clear_parents(&parents); |
1713 | } | |
d98ba053 | 1714 | kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list); |
4731d4c7 | 1715 | cond_resched_lock(&vcpu->kvm->mmu_lock); |
60c8aec6 MT |
1716 | kvm_mmu_pages_init(parent, &parents, &pages); |
1717 | } | |
4731d4c7 MT |
1718 | } |
1719 | ||
c3707958 XG |
1720 | static void init_shadow_page_table(struct kvm_mmu_page *sp) |
1721 | { | |
1722 | int i; | |
1723 | ||
1724 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) | |
1725 | sp->spt[i] = 0ull; | |
1726 | } | |
1727 | ||
a30f47cb XG |
1728 | static void __clear_sp_write_flooding_count(struct kvm_mmu_page *sp) |
1729 | { | |
1730 | sp->write_flooding_count = 0; | |
1731 | } | |
1732 | ||
1733 | static void clear_sp_write_flooding_count(u64 *spte) | |
1734 | { | |
1735 | struct kvm_mmu_page *sp = page_header(__pa(spte)); | |
1736 | ||
1737 | __clear_sp_write_flooding_count(sp); | |
1738 | } | |
1739 | ||
cea0f0e7 AK |
1740 | static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu, |
1741 | gfn_t gfn, | |
1742 | gva_t gaddr, | |
1743 | unsigned level, | |
f6e2c02b | 1744 | int direct, |
41074d07 | 1745 | unsigned access, |
f7d9c7b7 | 1746 | u64 *parent_pte) |
cea0f0e7 AK |
1747 | { |
1748 | union kvm_mmu_page_role role; | |
cea0f0e7 | 1749 | unsigned quadrant; |
9f1a122f | 1750 | struct kvm_mmu_page *sp; |
f41d335a | 1751 | struct hlist_node *node; |
9f1a122f | 1752 | bool need_sync = false; |
cea0f0e7 | 1753 | |
a770f6f2 | 1754 | role = vcpu->arch.mmu.base_role; |
cea0f0e7 | 1755 | role.level = level; |
f6e2c02b | 1756 | role.direct = direct; |
84b0c8c6 | 1757 | if (role.direct) |
5b7e0102 | 1758 | role.cr4_pae = 0; |
41074d07 | 1759 | role.access = access; |
c5a78f2b JR |
1760 | if (!vcpu->arch.mmu.direct_map |
1761 | && vcpu->arch.mmu.root_level <= PT32_ROOT_LEVEL) { | |
cea0f0e7 AK |
1762 | quadrant = gaddr >> (PAGE_SHIFT + (PT64_PT_BITS * level)); |
1763 | quadrant &= (1 << ((PT32_PT_BITS - PT64_PT_BITS) * level)) - 1; | |
1764 | role.quadrant = quadrant; | |
1765 | } | |
f41d335a | 1766 | for_each_gfn_sp(vcpu->kvm, sp, gfn, node) { |
7ae680eb XG |
1767 | if (!need_sync && sp->unsync) |
1768 | need_sync = true; | |
4731d4c7 | 1769 | |
7ae680eb XG |
1770 | if (sp->role.word != role.word) |
1771 | continue; | |
4731d4c7 | 1772 | |
7ae680eb XG |
1773 | if (sp->unsync && kvm_sync_page_transient(vcpu, sp)) |
1774 | break; | |
e02aa901 | 1775 | |
7ae680eb XG |
1776 | mmu_page_add_parent_pte(vcpu, sp, parent_pte); |
1777 | if (sp->unsync_children) { | |
a8eeb04a | 1778 | kvm_make_request(KVM_REQ_MMU_SYNC, vcpu); |
7ae680eb XG |
1779 | kvm_mmu_mark_parents_unsync(sp); |
1780 | } else if (sp->unsync) | |
1781 | kvm_mmu_mark_parents_unsync(sp); | |
e02aa901 | 1782 | |
a30f47cb | 1783 | __clear_sp_write_flooding_count(sp); |
7ae680eb XG |
1784 | trace_kvm_mmu_get_page(sp, false); |
1785 | return sp; | |
1786 | } | |
dfc5aa00 | 1787 | ++vcpu->kvm->stat.mmu_cache_miss; |
2032a93d | 1788 | sp = kvm_mmu_alloc_page(vcpu, parent_pte, direct); |
4db35314 AK |
1789 | if (!sp) |
1790 | return sp; | |
4db35314 AK |
1791 | sp->gfn = gfn; |
1792 | sp->role = role; | |
7ae680eb XG |
1793 | hlist_add_head(&sp->hash_link, |
1794 | &vcpu->kvm->arch.mmu_page_hash[kvm_page_table_hashfn(gfn)]); | |
f6e2c02b | 1795 | if (!direct) { |
b1a36821 MT |
1796 | if (rmap_write_protect(vcpu->kvm, gfn)) |
1797 | kvm_flush_remote_tlbs(vcpu->kvm); | |
9f1a122f XG |
1798 | if (level > PT_PAGE_TABLE_LEVEL && need_sync) |
1799 | kvm_sync_pages(vcpu, gfn); | |
1800 | ||
4731d4c7 MT |
1801 | account_shadowed(vcpu->kvm, gfn); |
1802 | } | |
c3707958 | 1803 | init_shadow_page_table(sp); |
f691fe1d | 1804 | trace_kvm_mmu_get_page(sp, true); |
4db35314 | 1805 | return sp; |
cea0f0e7 AK |
1806 | } |
1807 | ||
2d11123a AK |
1808 | static void shadow_walk_init(struct kvm_shadow_walk_iterator *iterator, |
1809 | struct kvm_vcpu *vcpu, u64 addr) | |
1810 | { | |
1811 | iterator->addr = addr; | |
1812 | iterator->shadow_addr = vcpu->arch.mmu.root_hpa; | |
1813 | iterator->level = vcpu->arch.mmu.shadow_root_level; | |
81407ca5 JR |
1814 | |
1815 | if (iterator->level == PT64_ROOT_LEVEL && | |
1816 | vcpu->arch.mmu.root_level < PT64_ROOT_LEVEL && | |
1817 | !vcpu->arch.mmu.direct_map) | |
1818 | --iterator->level; | |
1819 | ||
2d11123a AK |
1820 | if (iterator->level == PT32E_ROOT_LEVEL) { |
1821 | iterator->shadow_addr | |
1822 | = vcpu->arch.mmu.pae_root[(addr >> 30) & 3]; | |
1823 | iterator->shadow_addr &= PT64_BASE_ADDR_MASK; | |
1824 | --iterator->level; | |
1825 | if (!iterator->shadow_addr) | |
1826 | iterator->level = 0; | |
1827 | } | |
1828 | } | |
1829 | ||
1830 | static bool shadow_walk_okay(struct kvm_shadow_walk_iterator *iterator) | |
1831 | { | |
1832 | if (iterator->level < PT_PAGE_TABLE_LEVEL) | |
1833 | return false; | |
4d88954d | 1834 | |
2d11123a AK |
1835 | iterator->index = SHADOW_PT_INDEX(iterator->addr, iterator->level); |
1836 | iterator->sptep = ((u64 *)__va(iterator->shadow_addr)) + iterator->index; | |
1837 | return true; | |
1838 | } | |
1839 | ||
c2a2ac2b XG |
1840 | static void __shadow_walk_next(struct kvm_shadow_walk_iterator *iterator, |
1841 | u64 spte) | |
2d11123a | 1842 | { |
c2a2ac2b | 1843 | if (is_last_spte(spte, iterator->level)) { |
052331be XG |
1844 | iterator->level = 0; |
1845 | return; | |
1846 | } | |
1847 | ||
c2a2ac2b | 1848 | iterator->shadow_addr = spte & PT64_BASE_ADDR_MASK; |
2d11123a AK |
1849 | --iterator->level; |
1850 | } | |
1851 | ||
c2a2ac2b XG |
1852 | static void shadow_walk_next(struct kvm_shadow_walk_iterator *iterator) |
1853 | { | |
1854 | return __shadow_walk_next(iterator, *iterator->sptep); | |
1855 | } | |
1856 | ||
32ef26a3 AK |
1857 | static void link_shadow_page(u64 *sptep, struct kvm_mmu_page *sp) |
1858 | { | |
1859 | u64 spte; | |
1860 | ||
1861 | spte = __pa(sp->spt) | |
1862 | | PT_PRESENT_MASK | PT_ACCESSED_MASK | |
1863 | | PT_WRITABLE_MASK | PT_USER_MASK; | |
1df9f2dc | 1864 | mmu_spte_set(sptep, spte); |
32ef26a3 AK |
1865 | } |
1866 | ||
a3aa51cf AK |
1867 | static void drop_large_spte(struct kvm_vcpu *vcpu, u64 *sptep) |
1868 | { | |
1869 | if (is_large_pte(*sptep)) { | |
c3707958 | 1870 | drop_spte(vcpu->kvm, sptep); |
6addd1aa | 1871 | --vcpu->kvm->stat.lpages; |
a3aa51cf AK |
1872 | kvm_flush_remote_tlbs(vcpu->kvm); |
1873 | } | |
1874 | } | |
1875 | ||
a357bd22 AK |
1876 | static void validate_direct_spte(struct kvm_vcpu *vcpu, u64 *sptep, |
1877 | unsigned direct_access) | |
1878 | { | |
1879 | if (is_shadow_present_pte(*sptep) && !is_large_pte(*sptep)) { | |
1880 | struct kvm_mmu_page *child; | |
1881 | ||
1882 | /* | |
1883 | * For the direct sp, if the guest pte's dirty bit | |
1884 | * changed form clean to dirty, it will corrupt the | |
1885 | * sp's access: allow writable in the read-only sp, | |
1886 | * so we should update the spte at this point to get | |
1887 | * a new sp with the correct access. | |
1888 | */ | |
1889 | child = page_header(*sptep & PT64_BASE_ADDR_MASK); | |
1890 | if (child->role.access == direct_access) | |
1891 | return; | |
1892 | ||
bcdd9a93 | 1893 | drop_parent_pte(child, sptep); |
a357bd22 AK |
1894 | kvm_flush_remote_tlbs(vcpu->kvm); |
1895 | } | |
1896 | } | |
1897 | ||
505aef8f | 1898 | static bool mmu_page_zap_pte(struct kvm *kvm, struct kvm_mmu_page *sp, |
38e3b2b2 XG |
1899 | u64 *spte) |
1900 | { | |
1901 | u64 pte; | |
1902 | struct kvm_mmu_page *child; | |
1903 | ||
1904 | pte = *spte; | |
1905 | if (is_shadow_present_pte(pte)) { | |
505aef8f | 1906 | if (is_last_spte(pte, sp->role.level)) { |
c3707958 | 1907 | drop_spte(kvm, spte); |
505aef8f XG |
1908 | if (is_large_pte(pte)) |
1909 | --kvm->stat.lpages; | |
1910 | } else { | |
38e3b2b2 | 1911 | child = page_header(pte & PT64_BASE_ADDR_MASK); |
bcdd9a93 | 1912 | drop_parent_pte(child, spte); |
38e3b2b2 | 1913 | } |
505aef8f XG |
1914 | return true; |
1915 | } | |
1916 | ||
1917 | if (is_mmio_spte(pte)) | |
ce88decf | 1918 | mmu_spte_clear_no_track(spte); |
c3707958 | 1919 | |
505aef8f | 1920 | return false; |
38e3b2b2 XG |
1921 | } |
1922 | ||
90cb0529 | 1923 | static void kvm_mmu_page_unlink_children(struct kvm *kvm, |
4db35314 | 1924 | struct kvm_mmu_page *sp) |
a436036b | 1925 | { |
697fe2e2 | 1926 | unsigned i; |
697fe2e2 | 1927 | |
38e3b2b2 XG |
1928 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) |
1929 | mmu_page_zap_pte(kvm, sp, sp->spt + i); | |
a436036b AK |
1930 | } |
1931 | ||
4db35314 | 1932 | static void kvm_mmu_put_page(struct kvm_mmu_page *sp, u64 *parent_pte) |
cea0f0e7 | 1933 | { |
4db35314 | 1934 | mmu_page_remove_parent_pte(sp, parent_pte); |
a436036b AK |
1935 | } |
1936 | ||
31aa2b44 | 1937 | static void kvm_mmu_unlink_parents(struct kvm *kvm, struct kvm_mmu_page *sp) |
a436036b | 1938 | { |
1e3f42f0 TY |
1939 | u64 *sptep; |
1940 | struct rmap_iterator iter; | |
a436036b | 1941 | |
1e3f42f0 TY |
1942 | while ((sptep = rmap_get_first(sp->parent_ptes, &iter))) |
1943 | drop_parent_pte(sp, sptep); | |
31aa2b44 AK |
1944 | } |
1945 | ||
60c8aec6 | 1946 | static int mmu_zap_unsync_children(struct kvm *kvm, |
7775834a XG |
1947 | struct kvm_mmu_page *parent, |
1948 | struct list_head *invalid_list) | |
4731d4c7 | 1949 | { |
60c8aec6 MT |
1950 | int i, zapped = 0; |
1951 | struct mmu_page_path parents; | |
1952 | struct kvm_mmu_pages pages; | |
4731d4c7 | 1953 | |
60c8aec6 | 1954 | if (parent->role.level == PT_PAGE_TABLE_LEVEL) |
4731d4c7 | 1955 | return 0; |
60c8aec6 MT |
1956 | |
1957 | kvm_mmu_pages_init(parent, &parents, &pages); | |
1958 | while (mmu_unsync_walk(parent, &pages)) { | |
1959 | struct kvm_mmu_page *sp; | |
1960 | ||
1961 | for_each_sp(pages, sp, parents, i) { | |
7775834a | 1962 | kvm_mmu_prepare_zap_page(kvm, sp, invalid_list); |
60c8aec6 | 1963 | mmu_pages_clear_parents(&parents); |
77662e00 | 1964 | zapped++; |
60c8aec6 | 1965 | } |
60c8aec6 MT |
1966 | kvm_mmu_pages_init(parent, &parents, &pages); |
1967 | } | |
1968 | ||
1969 | return zapped; | |
4731d4c7 MT |
1970 | } |
1971 | ||
7775834a XG |
1972 | static int kvm_mmu_prepare_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp, |
1973 | struct list_head *invalid_list) | |
31aa2b44 | 1974 | { |
4731d4c7 | 1975 | int ret; |
f691fe1d | 1976 | |
7775834a | 1977 | trace_kvm_mmu_prepare_zap_page(sp); |
31aa2b44 | 1978 | ++kvm->stat.mmu_shadow_zapped; |
7775834a | 1979 | ret = mmu_zap_unsync_children(kvm, sp, invalid_list); |
4db35314 | 1980 | kvm_mmu_page_unlink_children(kvm, sp); |
31aa2b44 | 1981 | kvm_mmu_unlink_parents(kvm, sp); |
f6e2c02b | 1982 | if (!sp->role.invalid && !sp->role.direct) |
5b5c6a5a | 1983 | unaccount_shadowed(kvm, sp->gfn); |
4731d4c7 MT |
1984 | if (sp->unsync) |
1985 | kvm_unlink_unsync_page(kvm, sp); | |
4db35314 | 1986 | if (!sp->root_count) { |
54a4f023 GJ |
1987 | /* Count self */ |
1988 | ret++; | |
7775834a | 1989 | list_move(&sp->link, invalid_list); |
aa6bd187 | 1990 | kvm_mod_used_mmu_pages(kvm, -1); |
2e53d63a | 1991 | } else { |
5b5c6a5a | 1992 | list_move(&sp->link, &kvm->arch.active_mmu_pages); |
2e53d63a MT |
1993 | kvm_reload_remote_mmus(kvm); |
1994 | } | |
7775834a XG |
1995 | |
1996 | sp->role.invalid = 1; | |
4731d4c7 | 1997 | return ret; |
a436036b AK |
1998 | } |
1999 | ||
7775834a XG |
2000 | static void kvm_mmu_commit_zap_page(struct kvm *kvm, |
2001 | struct list_head *invalid_list) | |
2002 | { | |
2003 | struct kvm_mmu_page *sp; | |
2004 | ||
2005 | if (list_empty(invalid_list)) | |
2006 | return; | |
2007 | ||
c142786c AK |
2008 | /* |
2009 | * wmb: make sure everyone sees our modifications to the page tables | |
2010 | * rmb: make sure we see changes to vcpu->mode | |
2011 | */ | |
2012 | smp_mb(); | |
4f022648 | 2013 | |
c142786c AK |
2014 | /* |
2015 | * Wait for all vcpus to exit guest mode and/or lockless shadow | |
2016 | * page table walks. | |
2017 | */ | |
2018 | kvm_flush_remote_tlbs(kvm); | |
c2a2ac2b | 2019 | |
7775834a XG |
2020 | do { |
2021 | sp = list_first_entry(invalid_list, struct kvm_mmu_page, link); | |
2022 | WARN_ON(!sp->role.invalid || sp->root_count); | |
bd4c86ea | 2023 | kvm_mmu_isolate_page(sp); |
aa6bd187 | 2024 | kvm_mmu_free_page(sp); |
7775834a | 2025 | } while (!list_empty(invalid_list)); |
7775834a XG |
2026 | } |
2027 | ||
82ce2c96 IE |
2028 | /* |
2029 | * Changing the number of mmu pages allocated to the vm | |
49d5ca26 | 2030 | * Note: if goal_nr_mmu_pages is too small, you will get dead lock |
82ce2c96 | 2031 | */ |
49d5ca26 | 2032 | void kvm_mmu_change_mmu_pages(struct kvm *kvm, unsigned int goal_nr_mmu_pages) |
82ce2c96 | 2033 | { |
d98ba053 | 2034 | LIST_HEAD(invalid_list); |
82ce2c96 IE |
2035 | /* |
2036 | * If we set the number of mmu pages to be smaller be than the | |
2037 | * number of actived pages , we must to free some mmu pages before we | |
2038 | * change the value | |
2039 | */ | |
2040 | ||
49d5ca26 DH |
2041 | if (kvm->arch.n_used_mmu_pages > goal_nr_mmu_pages) { |
2042 | while (kvm->arch.n_used_mmu_pages > goal_nr_mmu_pages && | |
77662e00 | 2043 | !list_empty(&kvm->arch.active_mmu_pages)) { |
82ce2c96 IE |
2044 | struct kvm_mmu_page *page; |
2045 | ||
f05e70ac | 2046 | page = container_of(kvm->arch.active_mmu_pages.prev, |
82ce2c96 | 2047 | struct kvm_mmu_page, link); |
80b63faf | 2048 | kvm_mmu_prepare_zap_page(kvm, page, &invalid_list); |
82ce2c96 | 2049 | } |
aa6bd187 | 2050 | kvm_mmu_commit_zap_page(kvm, &invalid_list); |
49d5ca26 | 2051 | goal_nr_mmu_pages = kvm->arch.n_used_mmu_pages; |
82ce2c96 | 2052 | } |
82ce2c96 | 2053 | |
49d5ca26 | 2054 | kvm->arch.n_max_mmu_pages = goal_nr_mmu_pages; |
82ce2c96 IE |
2055 | } |
2056 | ||
1cb3f3ae | 2057 | int kvm_mmu_unprotect_page(struct kvm *kvm, gfn_t gfn) |
a436036b | 2058 | { |
4db35314 | 2059 | struct kvm_mmu_page *sp; |
f41d335a | 2060 | struct hlist_node *node; |
d98ba053 | 2061 | LIST_HEAD(invalid_list); |
a436036b AK |
2062 | int r; |
2063 | ||
9ad17b10 | 2064 | pgprintk("%s: looking for gfn %llx\n", __func__, gfn); |
a436036b | 2065 | r = 0; |
1cb3f3ae | 2066 | spin_lock(&kvm->mmu_lock); |
f41d335a | 2067 | for_each_gfn_indirect_valid_sp(kvm, sp, gfn, node) { |
9ad17b10 | 2068 | pgprintk("%s: gfn %llx role %x\n", __func__, gfn, |
7ae680eb XG |
2069 | sp->role.word); |
2070 | r = 1; | |
f41d335a | 2071 | kvm_mmu_prepare_zap_page(kvm, sp, &invalid_list); |
7ae680eb | 2072 | } |
d98ba053 | 2073 | kvm_mmu_commit_zap_page(kvm, &invalid_list); |
1cb3f3ae XG |
2074 | spin_unlock(&kvm->mmu_lock); |
2075 | ||
a436036b | 2076 | return r; |
cea0f0e7 | 2077 | } |
1cb3f3ae | 2078 | EXPORT_SYMBOL_GPL(kvm_mmu_unprotect_page); |
cea0f0e7 | 2079 | |
38c335f1 | 2080 | static void page_header_update_slot(struct kvm *kvm, void *pte, gfn_t gfn) |
6aa8b732 | 2081 | { |
bc6678a3 | 2082 | int slot = memslot_id(kvm, gfn); |
4db35314 | 2083 | struct kvm_mmu_page *sp = page_header(__pa(pte)); |
6aa8b732 | 2084 | |
291f26bc | 2085 | __set_bit(slot, sp->slot_bitmap); |
6aa8b732 AK |
2086 | } |
2087 | ||
74be52e3 SY |
2088 | /* |
2089 | * The function is based on mtrr_type_lookup() in | |
2090 | * arch/x86/kernel/cpu/mtrr/generic.c | |
2091 | */ | |
2092 | static int get_mtrr_type(struct mtrr_state_type *mtrr_state, | |
2093 | u64 start, u64 end) | |
2094 | { | |
2095 | int i; | |
2096 | u64 base, mask; | |
2097 | u8 prev_match, curr_match; | |
2098 | int num_var_ranges = KVM_NR_VAR_MTRR; | |
2099 | ||
2100 | if (!mtrr_state->enabled) | |
2101 | return 0xFF; | |
2102 | ||
2103 | /* Make end inclusive end, instead of exclusive */ | |
2104 | end--; | |
2105 | ||
2106 | /* Look in fixed ranges. Just return the type as per start */ | |
2107 | if (mtrr_state->have_fixed && (start < 0x100000)) { | |
2108 | int idx; | |
2109 | ||
2110 | if (start < 0x80000) { | |
2111 | idx = 0; | |
2112 | idx += (start >> 16); | |
2113 | return mtrr_state->fixed_ranges[idx]; | |
2114 | } else if (start < 0xC0000) { | |
2115 | idx = 1 * 8; | |
2116 | idx += ((start - 0x80000) >> 14); | |
2117 | return mtrr_state->fixed_ranges[idx]; | |
2118 | } else if (start < 0x1000000) { | |
2119 | idx = 3 * 8; | |
2120 | idx += ((start - 0xC0000) >> 12); | |
2121 | return mtrr_state->fixed_ranges[idx]; | |
2122 | } | |
2123 | } | |
2124 | ||
2125 | /* | |
2126 | * Look in variable ranges | |
2127 | * Look of multiple ranges matching this address and pick type | |
2128 | * as per MTRR precedence | |
2129 | */ | |
2130 | if (!(mtrr_state->enabled & 2)) | |
2131 | return mtrr_state->def_type; | |
2132 | ||
2133 | prev_match = 0xFF; | |
2134 | for (i = 0; i < num_var_ranges; ++i) { | |
2135 | unsigned short start_state, end_state; | |
2136 | ||
2137 | if (!(mtrr_state->var_ranges[i].mask_lo & (1 << 11))) | |
2138 | continue; | |
2139 | ||
2140 | base = (((u64)mtrr_state->var_ranges[i].base_hi) << 32) + | |
2141 | (mtrr_state->var_ranges[i].base_lo & PAGE_MASK); | |
2142 | mask = (((u64)mtrr_state->var_ranges[i].mask_hi) << 32) + | |
2143 | (mtrr_state->var_ranges[i].mask_lo & PAGE_MASK); | |
2144 | ||
2145 | start_state = ((start & mask) == (base & mask)); | |
2146 | end_state = ((end & mask) == (base & mask)); | |
2147 | if (start_state != end_state) | |
2148 | return 0xFE; | |
2149 | ||
2150 | if ((start & mask) != (base & mask)) | |
2151 | continue; | |
2152 | ||
2153 | curr_match = mtrr_state->var_ranges[i].base_lo & 0xff; | |
2154 | if (prev_match == 0xFF) { | |
2155 | prev_match = curr_match; | |
2156 | continue; | |
2157 | } | |
2158 | ||
2159 | if (prev_match == MTRR_TYPE_UNCACHABLE || | |
2160 | curr_match == MTRR_TYPE_UNCACHABLE) | |
2161 | return MTRR_TYPE_UNCACHABLE; | |
2162 | ||
2163 | if ((prev_match == MTRR_TYPE_WRBACK && | |
2164 | curr_match == MTRR_TYPE_WRTHROUGH) || | |
2165 | (prev_match == MTRR_TYPE_WRTHROUGH && | |
2166 | curr_match == MTRR_TYPE_WRBACK)) { | |
2167 | prev_match = MTRR_TYPE_WRTHROUGH; | |
2168 | curr_match = MTRR_TYPE_WRTHROUGH; | |
2169 | } | |
2170 | ||
2171 | if (prev_match != curr_match) | |
2172 | return MTRR_TYPE_UNCACHABLE; | |
2173 | } | |
2174 | ||
2175 | if (prev_match != 0xFF) | |
2176 | return prev_match; | |
2177 | ||
2178 | return mtrr_state->def_type; | |
2179 | } | |
2180 | ||
4b12f0de | 2181 | u8 kvm_get_guest_memory_type(struct kvm_vcpu *vcpu, gfn_t gfn) |
74be52e3 SY |
2182 | { |
2183 | u8 mtrr; | |
2184 | ||
2185 | mtrr = get_mtrr_type(&vcpu->arch.mtrr_state, gfn << PAGE_SHIFT, | |
2186 | (gfn << PAGE_SHIFT) + PAGE_SIZE); | |
2187 | if (mtrr == 0xfe || mtrr == 0xff) | |
2188 | mtrr = MTRR_TYPE_WRBACK; | |
2189 | return mtrr; | |
2190 | } | |
4b12f0de | 2191 | EXPORT_SYMBOL_GPL(kvm_get_guest_memory_type); |
74be52e3 | 2192 | |
9cf5cf5a XG |
2193 | static void __kvm_unsync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp) |
2194 | { | |
2195 | trace_kvm_mmu_unsync_page(sp); | |
2196 | ++vcpu->kvm->stat.mmu_unsync; | |
2197 | sp->unsync = 1; | |
2198 | ||
2199 | kvm_mmu_mark_parents_unsync(sp); | |
9cf5cf5a XG |
2200 | } |
2201 | ||
2202 | static void kvm_unsync_pages(struct kvm_vcpu *vcpu, gfn_t gfn) | |
4731d4c7 | 2203 | { |
4731d4c7 | 2204 | struct kvm_mmu_page *s; |
f41d335a | 2205 | struct hlist_node *node; |
9cf5cf5a | 2206 | |
f41d335a | 2207 | for_each_gfn_indirect_valid_sp(vcpu->kvm, s, gfn, node) { |
7ae680eb | 2208 | if (s->unsync) |
4731d4c7 | 2209 | continue; |
9cf5cf5a XG |
2210 | WARN_ON(s->role.level != PT_PAGE_TABLE_LEVEL); |
2211 | __kvm_unsync_page(vcpu, s); | |
4731d4c7 | 2212 | } |
4731d4c7 MT |
2213 | } |
2214 | ||
2215 | static int mmu_need_write_protect(struct kvm_vcpu *vcpu, gfn_t gfn, | |
2216 | bool can_unsync) | |
2217 | { | |
9cf5cf5a | 2218 | struct kvm_mmu_page *s; |
f41d335a | 2219 | struct hlist_node *node; |
9cf5cf5a XG |
2220 | bool need_unsync = false; |
2221 | ||
f41d335a | 2222 | for_each_gfn_indirect_valid_sp(vcpu->kvm, s, gfn, node) { |
36a2e677 XG |
2223 | if (!can_unsync) |
2224 | return 1; | |
2225 | ||
9cf5cf5a | 2226 | if (s->role.level != PT_PAGE_TABLE_LEVEL) |
4731d4c7 | 2227 | return 1; |
9cf5cf5a XG |
2228 | |
2229 | if (!need_unsync && !s->unsync) { | |
9cf5cf5a XG |
2230 | need_unsync = true; |
2231 | } | |
4731d4c7 | 2232 | } |
9cf5cf5a XG |
2233 | if (need_unsync) |
2234 | kvm_unsync_pages(vcpu, gfn); | |
4731d4c7 MT |
2235 | return 0; |
2236 | } | |
2237 | ||
d555c333 | 2238 | static int set_spte(struct kvm_vcpu *vcpu, u64 *sptep, |
1e73f9dd | 2239 | unsigned pte_access, int user_fault, |
640d9b0d | 2240 | int write_fault, int level, |
c2d0ee46 | 2241 | gfn_t gfn, pfn_t pfn, bool speculative, |
9bdbba13 | 2242 | bool can_unsync, bool host_writable) |
1c4f1fd6 | 2243 | { |
b330aa0c | 2244 | u64 spte, entry = *sptep; |
1e73f9dd | 2245 | int ret = 0; |
64d4d521 | 2246 | |
ce88decf XG |
2247 | if (set_mmio_spte(sptep, gfn, pfn, pte_access)) |
2248 | return 0; | |
2249 | ||
982c2565 | 2250 | spte = PT_PRESENT_MASK; |
947da538 | 2251 | if (!speculative) |
3201b5d9 | 2252 | spte |= shadow_accessed_mask; |
640d9b0d | 2253 | |
7b52345e SY |
2254 | if (pte_access & ACC_EXEC_MASK) |
2255 | spte |= shadow_x_mask; | |
2256 | else | |
2257 | spte |= shadow_nx_mask; | |
1c4f1fd6 | 2258 | if (pte_access & ACC_USER_MASK) |
7b52345e | 2259 | spte |= shadow_user_mask; |
852e3c19 | 2260 | if (level > PT_PAGE_TABLE_LEVEL) |
05da4558 | 2261 | spte |= PT_PAGE_SIZE_MASK; |
b0bc3ee2 | 2262 | if (tdp_enabled) |
4b12f0de SY |
2263 | spte |= kvm_x86_ops->get_mt_mask(vcpu, gfn, |
2264 | kvm_is_mmio_pfn(pfn)); | |
1c4f1fd6 | 2265 | |
9bdbba13 | 2266 | if (host_writable) |
1403283a | 2267 | spte |= SPTE_HOST_WRITEABLE; |
f8e453b0 XG |
2268 | else |
2269 | pte_access &= ~ACC_WRITE_MASK; | |
1403283a | 2270 | |
35149e21 | 2271 | spte |= (u64)pfn << PAGE_SHIFT; |
1c4f1fd6 AK |
2272 | |
2273 | if ((pte_access & ACC_WRITE_MASK) | |
c5a78f2b JR |
2274 | || (!vcpu->arch.mmu.direct_map && write_fault |
2275 | && !is_write_protection(vcpu) && !user_fault)) { | |
1c4f1fd6 | 2276 | |
852e3c19 JR |
2277 | if (level > PT_PAGE_TABLE_LEVEL && |
2278 | has_wrprotected_page(vcpu->kvm, gfn, level)) { | |
38187c83 | 2279 | ret = 1; |
c3707958 | 2280 | drop_spte(vcpu->kvm, sptep); |
be38d276 | 2281 | goto done; |
38187c83 MT |
2282 | } |
2283 | ||
1c4f1fd6 | 2284 | spte |= PT_WRITABLE_MASK; |
1c4f1fd6 | 2285 | |
c5a78f2b | 2286 | if (!vcpu->arch.mmu.direct_map |
411c588d | 2287 | && !(pte_access & ACC_WRITE_MASK)) { |
69325a12 | 2288 | spte &= ~PT_USER_MASK; |
411c588d AK |
2289 | /* |
2290 | * If we converted a user page to a kernel page, | |
2291 | * so that the kernel can write to it when cr0.wp=0, | |
2292 | * then we should prevent the kernel from executing it | |
2293 | * if SMEP is enabled. | |
2294 | */ | |
2295 | if (kvm_read_cr4_bits(vcpu, X86_CR4_SMEP)) | |
2296 | spte |= PT64_NX_MASK; | |
2297 | } | |
69325a12 | 2298 | |
ecc5589f MT |
2299 | /* |
2300 | * Optimization: for pte sync, if spte was writable the hash | |
2301 | * lookup is unnecessary (and expensive). Write protection | |
2302 | * is responsibility of mmu_get_page / kvm_sync_page. | |
2303 | * Same reasoning can be applied to dirty page accounting. | |
2304 | */ | |
8dae4445 | 2305 | if (!can_unsync && is_writable_pte(*sptep)) |
ecc5589f MT |
2306 | goto set_pte; |
2307 | ||
4731d4c7 | 2308 | if (mmu_need_write_protect(vcpu, gfn, can_unsync)) { |
9ad17b10 | 2309 | pgprintk("%s: found shadow page for %llx, marking ro\n", |
b8688d51 | 2310 | __func__, gfn); |
1e73f9dd | 2311 | ret = 1; |
1c4f1fd6 | 2312 | pte_access &= ~ACC_WRITE_MASK; |
8dae4445 | 2313 | if (is_writable_pte(spte)) |
1c4f1fd6 | 2314 | spte &= ~PT_WRITABLE_MASK; |
1c4f1fd6 AK |
2315 | } |
2316 | } | |
2317 | ||
1c4f1fd6 AK |
2318 | if (pte_access & ACC_WRITE_MASK) |
2319 | mark_page_dirty(vcpu->kvm, gfn); | |
2320 | ||
38187c83 | 2321 | set_pte: |
1df9f2dc | 2322 | mmu_spte_update(sptep, spte); |
b330aa0c XG |
2323 | /* |
2324 | * If we overwrite a writable spte with a read-only one we | |
2325 | * should flush remote TLBs. Otherwise rmap_write_protect | |
2326 | * will find a read-only spte, even though the writable spte | |
2327 | * might be cached on a CPU's TLB. | |
2328 | */ | |
2329 | if (is_writable_pte(entry) && !is_writable_pte(*sptep)) | |
2330 | kvm_flush_remote_tlbs(vcpu->kvm); | |
be38d276 | 2331 | done: |
1e73f9dd MT |
2332 | return ret; |
2333 | } | |
2334 | ||
d555c333 | 2335 | static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *sptep, |
1e73f9dd | 2336 | unsigned pt_access, unsigned pte_access, |
640d9b0d | 2337 | int user_fault, int write_fault, |
b90a0e6c | 2338 | int *emulate, int level, gfn_t gfn, |
1403283a | 2339 | pfn_t pfn, bool speculative, |
9bdbba13 | 2340 | bool host_writable) |
1e73f9dd MT |
2341 | { |
2342 | int was_rmapped = 0; | |
53a27b39 | 2343 | int rmap_count; |
1e73f9dd MT |
2344 | |
2345 | pgprintk("%s: spte %llx access %x write_fault %d" | |
9ad17b10 | 2346 | " user_fault %d gfn %llx\n", |
d555c333 | 2347 | __func__, *sptep, pt_access, |
1e73f9dd MT |
2348 | write_fault, user_fault, gfn); |
2349 | ||
d555c333 | 2350 | if (is_rmap_spte(*sptep)) { |
1e73f9dd MT |
2351 | /* |
2352 | * If we overwrite a PTE page pointer with a 2MB PMD, unlink | |
2353 | * the parent of the now unreachable PTE. | |
2354 | */ | |
852e3c19 JR |
2355 | if (level > PT_PAGE_TABLE_LEVEL && |
2356 | !is_large_pte(*sptep)) { | |
1e73f9dd | 2357 | struct kvm_mmu_page *child; |
d555c333 | 2358 | u64 pte = *sptep; |
1e73f9dd MT |
2359 | |
2360 | child = page_header(pte & PT64_BASE_ADDR_MASK); | |
bcdd9a93 | 2361 | drop_parent_pte(child, sptep); |
3be2264b | 2362 | kvm_flush_remote_tlbs(vcpu->kvm); |
d555c333 | 2363 | } else if (pfn != spte_to_pfn(*sptep)) { |
9ad17b10 | 2364 | pgprintk("hfn old %llx new %llx\n", |
d555c333 | 2365 | spte_to_pfn(*sptep), pfn); |
c3707958 | 2366 | drop_spte(vcpu->kvm, sptep); |
91546356 | 2367 | kvm_flush_remote_tlbs(vcpu->kvm); |
6bed6b9e JR |
2368 | } else |
2369 | was_rmapped = 1; | |
1e73f9dd | 2370 | } |
852e3c19 | 2371 | |
d555c333 | 2372 | if (set_spte(vcpu, sptep, pte_access, user_fault, write_fault, |
640d9b0d | 2373 | level, gfn, pfn, speculative, true, |
9bdbba13 | 2374 | host_writable)) { |
1e73f9dd | 2375 | if (write_fault) |
b90a0e6c | 2376 | *emulate = 1; |
5304efde | 2377 | kvm_mmu_flush_tlb(vcpu); |
a378b4e6 | 2378 | } |
1e73f9dd | 2379 | |
ce88decf XG |
2380 | if (unlikely(is_mmio_spte(*sptep) && emulate)) |
2381 | *emulate = 1; | |
2382 | ||
d555c333 | 2383 | pgprintk("%s: setting spte %llx\n", __func__, *sptep); |
9ad17b10 | 2384 | pgprintk("instantiating %s PTE (%s) at %llx (%llx) addr %p\n", |
d555c333 | 2385 | is_large_pte(*sptep)? "2MB" : "4kB", |
a205bc19 JR |
2386 | *sptep & PT_PRESENT_MASK ?"RW":"R", gfn, |
2387 | *sptep, sptep); | |
d555c333 | 2388 | if (!was_rmapped && is_large_pte(*sptep)) |
05da4558 MT |
2389 | ++vcpu->kvm->stat.lpages; |
2390 | ||
ffb61bb3 XG |
2391 | if (is_shadow_present_pte(*sptep)) { |
2392 | page_header_update_slot(vcpu->kvm, sptep, gfn); | |
2393 | if (!was_rmapped) { | |
2394 | rmap_count = rmap_add(vcpu, sptep, gfn); | |
2395 | if (rmap_count > RMAP_RECYCLE_THRESHOLD) | |
2396 | rmap_recycle(vcpu, sptep, gfn); | |
2397 | } | |
1c4f1fd6 | 2398 | } |
9ed5520d | 2399 | kvm_release_pfn_clean(pfn); |
1c4f1fd6 AK |
2400 | } |
2401 | ||
6aa8b732 AK |
2402 | static void nonpaging_new_cr3(struct kvm_vcpu *vcpu) |
2403 | { | |
2404 | } | |
2405 | ||
957ed9ef XG |
2406 | static pfn_t pte_prefetch_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn, |
2407 | bool no_dirty_log) | |
2408 | { | |
2409 | struct kvm_memory_slot *slot; | |
2410 | unsigned long hva; | |
2411 | ||
5d163b1c | 2412 | slot = gfn_to_memslot_dirty_bitmap(vcpu, gfn, no_dirty_log); |
957ed9ef | 2413 | if (!slot) { |
fce92dce XG |
2414 | get_page(fault_page); |
2415 | return page_to_pfn(fault_page); | |
957ed9ef XG |
2416 | } |
2417 | ||
2418 | hva = gfn_to_hva_memslot(slot, gfn); | |
2419 | ||
2420 | return hva_to_pfn_atomic(vcpu->kvm, hva); | |
2421 | } | |
2422 | ||
2423 | static int direct_pte_prefetch_many(struct kvm_vcpu *vcpu, | |
2424 | struct kvm_mmu_page *sp, | |
2425 | u64 *start, u64 *end) | |
2426 | { | |
2427 | struct page *pages[PTE_PREFETCH_NUM]; | |
2428 | unsigned access = sp->role.access; | |
2429 | int i, ret; | |
2430 | gfn_t gfn; | |
2431 | ||
2432 | gfn = kvm_mmu_page_get_gfn(sp, start - sp->spt); | |
5d163b1c | 2433 | if (!gfn_to_memslot_dirty_bitmap(vcpu, gfn, access & ACC_WRITE_MASK)) |
957ed9ef XG |
2434 | return -1; |
2435 | ||
2436 | ret = gfn_to_page_many_atomic(vcpu->kvm, gfn, pages, end - start); | |
2437 | if (ret <= 0) | |
2438 | return -1; | |
2439 | ||
2440 | for (i = 0; i < ret; i++, gfn++, start++) | |
2441 | mmu_set_spte(vcpu, start, ACC_ALL, | |
640d9b0d | 2442 | access, 0, 0, NULL, |
957ed9ef XG |
2443 | sp->role.level, gfn, |
2444 | page_to_pfn(pages[i]), true, true); | |
2445 | ||
2446 | return 0; | |
2447 | } | |
2448 | ||
2449 | static void __direct_pte_prefetch(struct kvm_vcpu *vcpu, | |
2450 | struct kvm_mmu_page *sp, u64 *sptep) | |
2451 | { | |
2452 | u64 *spte, *start = NULL; | |
2453 | int i; | |
2454 | ||
2455 | WARN_ON(!sp->role.direct); | |
2456 | ||
2457 | i = (sptep - sp->spt) & ~(PTE_PREFETCH_NUM - 1); | |
2458 | spte = sp->spt + i; | |
2459 | ||
2460 | for (i = 0; i < PTE_PREFETCH_NUM; i++, spte++) { | |
c3707958 | 2461 | if (is_shadow_present_pte(*spte) || spte == sptep) { |
957ed9ef XG |
2462 | if (!start) |
2463 | continue; | |
2464 | if (direct_pte_prefetch_many(vcpu, sp, start, spte) < 0) | |
2465 | break; | |
2466 | start = NULL; | |
2467 | } else if (!start) | |
2468 | start = spte; | |
2469 | } | |
2470 | } | |
2471 | ||
2472 | static void direct_pte_prefetch(struct kvm_vcpu *vcpu, u64 *sptep) | |
2473 | { | |
2474 | struct kvm_mmu_page *sp; | |
2475 | ||
2476 | /* | |
2477 | * Since it's no accessed bit on EPT, it's no way to | |
2478 | * distinguish between actually accessed translations | |
2479 | * and prefetched, so disable pte prefetch if EPT is | |
2480 | * enabled. | |
2481 | */ | |
2482 | if (!shadow_accessed_mask) | |
2483 | return; | |
2484 | ||
2485 | sp = page_header(__pa(sptep)); | |
2486 | if (sp->role.level > PT_PAGE_TABLE_LEVEL) | |
2487 | return; | |
2488 | ||
2489 | __direct_pte_prefetch(vcpu, sp, sptep); | |
2490 | } | |
2491 | ||
9f652d21 | 2492 | static int __direct_map(struct kvm_vcpu *vcpu, gpa_t v, int write, |
2ec4739d XG |
2493 | int map_writable, int level, gfn_t gfn, pfn_t pfn, |
2494 | bool prefault) | |
140754bc | 2495 | { |
9f652d21 | 2496 | struct kvm_shadow_walk_iterator iterator; |
140754bc | 2497 | struct kvm_mmu_page *sp; |
b90a0e6c | 2498 | int emulate = 0; |
140754bc | 2499 | gfn_t pseudo_gfn; |
6aa8b732 | 2500 | |
9f652d21 | 2501 | for_each_shadow_entry(vcpu, (u64)gfn << PAGE_SHIFT, iterator) { |
852e3c19 | 2502 | if (iterator.level == level) { |
612819c3 MT |
2503 | unsigned pte_access = ACC_ALL; |
2504 | ||
612819c3 | 2505 | mmu_set_spte(vcpu, iterator.sptep, ACC_ALL, pte_access, |
b90a0e6c | 2506 | 0, write, &emulate, |
2ec4739d | 2507 | level, gfn, pfn, prefault, map_writable); |
957ed9ef | 2508 | direct_pte_prefetch(vcpu, iterator.sptep); |
9f652d21 AK |
2509 | ++vcpu->stat.pf_fixed; |
2510 | break; | |
6aa8b732 AK |
2511 | } |
2512 | ||
c3707958 | 2513 | if (!is_shadow_present_pte(*iterator.sptep)) { |
c9fa0b3b LJ |
2514 | u64 base_addr = iterator.addr; |
2515 | ||
2516 | base_addr &= PT64_LVL_ADDR_MASK(iterator.level); | |
2517 | pseudo_gfn = base_addr >> PAGE_SHIFT; | |
9f652d21 AK |
2518 | sp = kvm_mmu_get_page(vcpu, pseudo_gfn, iterator.addr, |
2519 | iterator.level - 1, | |
2520 | 1, ACC_ALL, iterator.sptep); | |
2521 | if (!sp) { | |
2522 | pgprintk("nonpaging_map: ENOMEM\n"); | |
2523 | kvm_release_pfn_clean(pfn); | |
2524 | return -ENOMEM; | |
2525 | } | |
140754bc | 2526 | |
1df9f2dc XG |
2527 | mmu_spte_set(iterator.sptep, |
2528 | __pa(sp->spt) | |
2529 | | PT_PRESENT_MASK | PT_WRITABLE_MASK | |
2530 | | shadow_user_mask | shadow_x_mask | |
2531 | | shadow_accessed_mask); | |
9f652d21 AK |
2532 | } |
2533 | } | |
b90a0e6c | 2534 | return emulate; |
6aa8b732 AK |
2535 | } |
2536 | ||
77db5cbd | 2537 | static void kvm_send_hwpoison_signal(unsigned long address, struct task_struct *tsk) |
bf998156 | 2538 | { |
77db5cbd HY |
2539 | siginfo_t info; |
2540 | ||
2541 | info.si_signo = SIGBUS; | |
2542 | info.si_errno = 0; | |
2543 | info.si_code = BUS_MCEERR_AR; | |
2544 | info.si_addr = (void __user *)address; | |
2545 | info.si_addr_lsb = PAGE_SHIFT; | |
bf998156 | 2546 | |
77db5cbd | 2547 | send_sig_info(SIGBUS, &info, tsk); |
bf998156 HY |
2548 | } |
2549 | ||
d7c55201 | 2550 | static int kvm_handle_bad_page(struct kvm_vcpu *vcpu, gfn_t gfn, pfn_t pfn) |
bf998156 HY |
2551 | { |
2552 | kvm_release_pfn_clean(pfn); | |
2553 | if (is_hwpoison_pfn(pfn)) { | |
bebb106a | 2554 | kvm_send_hwpoison_signal(gfn_to_hva(vcpu->kvm, gfn), current); |
bf998156 | 2555 | return 0; |
d7c55201 | 2556 | } |
edba23e5 | 2557 | |
d7c55201 | 2558 | return -EFAULT; |
bf998156 HY |
2559 | } |
2560 | ||
936a5fe6 AA |
2561 | static void transparent_hugepage_adjust(struct kvm_vcpu *vcpu, |
2562 | gfn_t *gfnp, pfn_t *pfnp, int *levelp) | |
2563 | { | |
2564 | pfn_t pfn = *pfnp; | |
2565 | gfn_t gfn = *gfnp; | |
2566 | int level = *levelp; | |
2567 | ||
2568 | /* | |
2569 | * Check if it's a transparent hugepage. If this would be an | |
2570 | * hugetlbfs page, level wouldn't be set to | |
2571 | * PT_PAGE_TABLE_LEVEL and there would be no adjustment done | |
2572 | * here. | |
2573 | */ | |
2574 | if (!is_error_pfn(pfn) && !kvm_is_mmio_pfn(pfn) && | |
2575 | level == PT_PAGE_TABLE_LEVEL && | |
2576 | PageTransCompound(pfn_to_page(pfn)) && | |
2577 | !has_wrprotected_page(vcpu->kvm, gfn, PT_DIRECTORY_LEVEL)) { | |
2578 | unsigned long mask; | |
2579 | /* | |
2580 | * mmu_notifier_retry was successful and we hold the | |
2581 | * mmu_lock here, so the pmd can't become splitting | |
2582 | * from under us, and in turn | |
2583 | * __split_huge_page_refcount() can't run from under | |
2584 | * us and we can safely transfer the refcount from | |
2585 | * PG_tail to PG_head as we switch the pfn to tail to | |
2586 | * head. | |
2587 | */ | |
2588 | *levelp = level = PT_DIRECTORY_LEVEL; | |
2589 | mask = KVM_PAGES_PER_HPAGE(level) - 1; | |
2590 | VM_BUG_ON((gfn & mask) != (pfn & mask)); | |
2591 | if (pfn & mask) { | |
2592 | gfn &= ~mask; | |
2593 | *gfnp = gfn; | |
2594 | kvm_release_pfn_clean(pfn); | |
2595 | pfn &= ~mask; | |
c3586667 | 2596 | kvm_get_pfn(pfn); |
936a5fe6 AA |
2597 | *pfnp = pfn; |
2598 | } | |
2599 | } | |
2600 | } | |
2601 | ||
d7c55201 XG |
2602 | static bool mmu_invalid_pfn(pfn_t pfn) |
2603 | { | |
ce88decf | 2604 | return unlikely(is_invalid_pfn(pfn)); |
d7c55201 XG |
2605 | } |
2606 | ||
2607 | static bool handle_abnormal_pfn(struct kvm_vcpu *vcpu, gva_t gva, gfn_t gfn, | |
2608 | pfn_t pfn, unsigned access, int *ret_val) | |
2609 | { | |
2610 | bool ret = true; | |
2611 | ||
2612 | /* The pfn is invalid, report the error! */ | |
2613 | if (unlikely(is_invalid_pfn(pfn))) { | |
2614 | *ret_val = kvm_handle_bad_page(vcpu, gfn, pfn); | |
2615 | goto exit; | |
2616 | } | |
2617 | ||
ce88decf | 2618 | if (unlikely(is_noslot_pfn(pfn))) |
d7c55201 | 2619 | vcpu_cache_mmio_info(vcpu, gva, gfn, access); |
d7c55201 XG |
2620 | |
2621 | ret = false; | |
2622 | exit: | |
2623 | return ret; | |
2624 | } | |
2625 | ||
78b2c54a | 2626 | static bool try_async_pf(struct kvm_vcpu *vcpu, bool prefault, gfn_t gfn, |
060c2abe XG |
2627 | gva_t gva, pfn_t *pfn, bool write, bool *writable); |
2628 | ||
2629 | static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, int write, gfn_t gfn, | |
78b2c54a | 2630 | bool prefault) |
10589a46 MT |
2631 | { |
2632 | int r; | |
852e3c19 | 2633 | int level; |
936a5fe6 | 2634 | int force_pt_level; |
35149e21 | 2635 | pfn_t pfn; |
e930bffe | 2636 | unsigned long mmu_seq; |
612819c3 | 2637 | bool map_writable; |
aaee2c94 | 2638 | |
936a5fe6 AA |
2639 | force_pt_level = mapping_level_dirty_bitmap(vcpu, gfn); |
2640 | if (likely(!force_pt_level)) { | |
2641 | level = mapping_level(vcpu, gfn); | |
2642 | /* | |
2643 | * This path builds a PAE pagetable - so we can map | |
2644 | * 2mb pages at maximum. Therefore check if the level | |
2645 | * is larger than that. | |
2646 | */ | |
2647 | if (level > PT_DIRECTORY_LEVEL) | |
2648 | level = PT_DIRECTORY_LEVEL; | |
852e3c19 | 2649 | |
936a5fe6 AA |
2650 | gfn &= ~(KVM_PAGES_PER_HPAGE(level) - 1); |
2651 | } else | |
2652 | level = PT_PAGE_TABLE_LEVEL; | |
05da4558 | 2653 | |
e930bffe | 2654 | mmu_seq = vcpu->kvm->mmu_notifier_seq; |
4c2155ce | 2655 | smp_rmb(); |
060c2abe | 2656 | |
78b2c54a | 2657 | if (try_async_pf(vcpu, prefault, gfn, v, &pfn, write, &map_writable)) |
060c2abe | 2658 | return 0; |
aaee2c94 | 2659 | |
d7c55201 XG |
2660 | if (handle_abnormal_pfn(vcpu, v, gfn, pfn, ACC_ALL, &r)) |
2661 | return r; | |
d196e343 | 2662 | |
aaee2c94 | 2663 | spin_lock(&vcpu->kvm->mmu_lock); |
e930bffe AA |
2664 | if (mmu_notifier_retry(vcpu, mmu_seq)) |
2665 | goto out_unlock; | |
eb787d10 | 2666 | kvm_mmu_free_some_pages(vcpu); |
936a5fe6 AA |
2667 | if (likely(!force_pt_level)) |
2668 | transparent_hugepage_adjust(vcpu, &gfn, &pfn, &level); | |
2ec4739d XG |
2669 | r = __direct_map(vcpu, v, write, map_writable, level, gfn, pfn, |
2670 | prefault); | |
aaee2c94 MT |
2671 | spin_unlock(&vcpu->kvm->mmu_lock); |
2672 | ||
aaee2c94 | 2673 | |
10589a46 | 2674 | return r; |
e930bffe AA |
2675 | |
2676 | out_unlock: | |
2677 | spin_unlock(&vcpu->kvm->mmu_lock); | |
2678 | kvm_release_pfn_clean(pfn); | |
2679 | return 0; | |
10589a46 MT |
2680 | } |
2681 | ||
2682 | ||
17ac10ad AK |
2683 | static void mmu_free_roots(struct kvm_vcpu *vcpu) |
2684 | { | |
2685 | int i; | |
4db35314 | 2686 | struct kvm_mmu_page *sp; |
d98ba053 | 2687 | LIST_HEAD(invalid_list); |
17ac10ad | 2688 | |
ad312c7c | 2689 | if (!VALID_PAGE(vcpu->arch.mmu.root_hpa)) |
7b53aa56 | 2690 | return; |
aaee2c94 | 2691 | spin_lock(&vcpu->kvm->mmu_lock); |
81407ca5 JR |
2692 | if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL && |
2693 | (vcpu->arch.mmu.root_level == PT64_ROOT_LEVEL || | |
2694 | vcpu->arch.mmu.direct_map)) { | |
ad312c7c | 2695 | hpa_t root = vcpu->arch.mmu.root_hpa; |
17ac10ad | 2696 | |
4db35314 AK |
2697 | sp = page_header(root); |
2698 | --sp->root_count; | |
d98ba053 XG |
2699 | if (!sp->root_count && sp->role.invalid) { |
2700 | kvm_mmu_prepare_zap_page(vcpu->kvm, sp, &invalid_list); | |
2701 | kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list); | |
2702 | } | |
ad312c7c | 2703 | vcpu->arch.mmu.root_hpa = INVALID_PAGE; |
aaee2c94 | 2704 | spin_unlock(&vcpu->kvm->mmu_lock); |
17ac10ad AK |
2705 | return; |
2706 | } | |
17ac10ad | 2707 | for (i = 0; i < 4; ++i) { |
ad312c7c | 2708 | hpa_t root = vcpu->arch.mmu.pae_root[i]; |
17ac10ad | 2709 | |
417726a3 | 2710 | if (root) { |
417726a3 | 2711 | root &= PT64_BASE_ADDR_MASK; |
4db35314 AK |
2712 | sp = page_header(root); |
2713 | --sp->root_count; | |
2e53d63a | 2714 | if (!sp->root_count && sp->role.invalid) |
d98ba053 XG |
2715 | kvm_mmu_prepare_zap_page(vcpu->kvm, sp, |
2716 | &invalid_list); | |
417726a3 | 2717 | } |
ad312c7c | 2718 | vcpu->arch.mmu.pae_root[i] = INVALID_PAGE; |
17ac10ad | 2719 | } |
d98ba053 | 2720 | kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list); |
aaee2c94 | 2721 | spin_unlock(&vcpu->kvm->mmu_lock); |
ad312c7c | 2722 | vcpu->arch.mmu.root_hpa = INVALID_PAGE; |
17ac10ad AK |
2723 | } |
2724 | ||
8986ecc0 MT |
2725 | static int mmu_check_root(struct kvm_vcpu *vcpu, gfn_t root_gfn) |
2726 | { | |
2727 | int ret = 0; | |
2728 | ||
2729 | if (!kvm_is_visible_gfn(vcpu->kvm, root_gfn)) { | |
a8eeb04a | 2730 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); |
8986ecc0 MT |
2731 | ret = 1; |
2732 | } | |
2733 | ||
2734 | return ret; | |
2735 | } | |
2736 | ||
651dd37a JR |
2737 | static int mmu_alloc_direct_roots(struct kvm_vcpu *vcpu) |
2738 | { | |
2739 | struct kvm_mmu_page *sp; | |
7ebaf15e | 2740 | unsigned i; |
651dd37a JR |
2741 | |
2742 | if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) { | |
2743 | spin_lock(&vcpu->kvm->mmu_lock); | |
2744 | kvm_mmu_free_some_pages(vcpu); | |
2745 | sp = kvm_mmu_get_page(vcpu, 0, 0, PT64_ROOT_LEVEL, | |
2746 | 1, ACC_ALL, NULL); | |
2747 | ++sp->root_count; | |
2748 | spin_unlock(&vcpu->kvm->mmu_lock); | |
2749 | vcpu->arch.mmu.root_hpa = __pa(sp->spt); | |
2750 | } else if (vcpu->arch.mmu.shadow_root_level == PT32E_ROOT_LEVEL) { | |
2751 | for (i = 0; i < 4; ++i) { | |
2752 | hpa_t root = vcpu->arch.mmu.pae_root[i]; | |
2753 | ||
2754 | ASSERT(!VALID_PAGE(root)); | |
2755 | spin_lock(&vcpu->kvm->mmu_lock); | |
2756 | kvm_mmu_free_some_pages(vcpu); | |
649497d1 AK |
2757 | sp = kvm_mmu_get_page(vcpu, i << (30 - PAGE_SHIFT), |
2758 | i << 30, | |
651dd37a JR |
2759 | PT32_ROOT_LEVEL, 1, ACC_ALL, |
2760 | NULL); | |
2761 | root = __pa(sp->spt); | |
2762 | ++sp->root_count; | |
2763 | spin_unlock(&vcpu->kvm->mmu_lock); | |
2764 | vcpu->arch.mmu.pae_root[i] = root | PT_PRESENT_MASK; | |
651dd37a | 2765 | } |
6292757f | 2766 | vcpu->arch.mmu.root_hpa = __pa(vcpu->arch.mmu.pae_root); |
651dd37a JR |
2767 | } else |
2768 | BUG(); | |
2769 | ||
2770 | return 0; | |
2771 | } | |
2772 | ||
2773 | static int mmu_alloc_shadow_roots(struct kvm_vcpu *vcpu) | |
17ac10ad | 2774 | { |
4db35314 | 2775 | struct kvm_mmu_page *sp; |
81407ca5 JR |
2776 | u64 pdptr, pm_mask; |
2777 | gfn_t root_gfn; | |
2778 | int i; | |
3bb65a22 | 2779 | |
5777ed34 | 2780 | root_gfn = vcpu->arch.mmu.get_cr3(vcpu) >> PAGE_SHIFT; |
17ac10ad | 2781 | |
651dd37a JR |
2782 | if (mmu_check_root(vcpu, root_gfn)) |
2783 | return 1; | |
2784 | ||
2785 | /* | |
2786 | * Do we shadow a long mode page table? If so we need to | |
2787 | * write-protect the guests page table root. | |
2788 | */ | |
2789 | if (vcpu->arch.mmu.root_level == PT64_ROOT_LEVEL) { | |
ad312c7c | 2790 | hpa_t root = vcpu->arch.mmu.root_hpa; |
17ac10ad AK |
2791 | |
2792 | ASSERT(!VALID_PAGE(root)); | |
651dd37a | 2793 | |
8facbbff | 2794 | spin_lock(&vcpu->kvm->mmu_lock); |
24955b6c | 2795 | kvm_mmu_free_some_pages(vcpu); |
651dd37a JR |
2796 | sp = kvm_mmu_get_page(vcpu, root_gfn, 0, PT64_ROOT_LEVEL, |
2797 | 0, ACC_ALL, NULL); | |
4db35314 AK |
2798 | root = __pa(sp->spt); |
2799 | ++sp->root_count; | |
8facbbff | 2800 | spin_unlock(&vcpu->kvm->mmu_lock); |
ad312c7c | 2801 | vcpu->arch.mmu.root_hpa = root; |
8986ecc0 | 2802 | return 0; |
17ac10ad | 2803 | } |
f87f9288 | 2804 | |
651dd37a JR |
2805 | /* |
2806 | * We shadow a 32 bit page table. This may be a legacy 2-level | |
81407ca5 JR |
2807 | * or a PAE 3-level page table. In either case we need to be aware that |
2808 | * the shadow page table may be a PAE or a long mode page table. | |
651dd37a | 2809 | */ |
81407ca5 JR |
2810 | pm_mask = PT_PRESENT_MASK; |
2811 | if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) | |
2812 | pm_mask |= PT_ACCESSED_MASK | PT_WRITABLE_MASK | PT_USER_MASK; | |
2813 | ||
17ac10ad | 2814 | for (i = 0; i < 4; ++i) { |
ad312c7c | 2815 | hpa_t root = vcpu->arch.mmu.pae_root[i]; |
17ac10ad AK |
2816 | |
2817 | ASSERT(!VALID_PAGE(root)); | |
ad312c7c | 2818 | if (vcpu->arch.mmu.root_level == PT32E_ROOT_LEVEL) { |
e4e517b4 | 2819 | pdptr = vcpu->arch.mmu.get_pdptr(vcpu, i); |
43a3795a | 2820 | if (!is_present_gpte(pdptr)) { |
ad312c7c | 2821 | vcpu->arch.mmu.pae_root[i] = 0; |
417726a3 AK |
2822 | continue; |
2823 | } | |
6de4f3ad | 2824 | root_gfn = pdptr >> PAGE_SHIFT; |
f87f9288 JR |
2825 | if (mmu_check_root(vcpu, root_gfn)) |
2826 | return 1; | |
5a7388c2 | 2827 | } |
8facbbff | 2828 | spin_lock(&vcpu->kvm->mmu_lock); |
24955b6c | 2829 | kvm_mmu_free_some_pages(vcpu); |
4db35314 | 2830 | sp = kvm_mmu_get_page(vcpu, root_gfn, i << 30, |
651dd37a | 2831 | PT32_ROOT_LEVEL, 0, |
f7d9c7b7 | 2832 | ACC_ALL, NULL); |
4db35314 AK |
2833 | root = __pa(sp->spt); |
2834 | ++sp->root_count; | |
8facbbff AK |
2835 | spin_unlock(&vcpu->kvm->mmu_lock); |
2836 | ||
81407ca5 | 2837 | vcpu->arch.mmu.pae_root[i] = root | pm_mask; |
17ac10ad | 2838 | } |
6292757f | 2839 | vcpu->arch.mmu.root_hpa = __pa(vcpu->arch.mmu.pae_root); |
81407ca5 JR |
2840 | |
2841 | /* | |
2842 | * If we shadow a 32 bit page table with a long mode page | |
2843 | * table we enter this path. | |
2844 | */ | |
2845 | if (vcpu->arch.mmu.shadow_root_level == PT64_ROOT_LEVEL) { | |
2846 | if (vcpu->arch.mmu.lm_root == NULL) { | |
2847 | /* | |
2848 | * The additional page necessary for this is only | |
2849 | * allocated on demand. | |
2850 | */ | |
2851 | ||
2852 | u64 *lm_root; | |
2853 | ||
2854 | lm_root = (void*)get_zeroed_page(GFP_KERNEL); | |
2855 | if (lm_root == NULL) | |
2856 | return 1; | |
2857 | ||
2858 | lm_root[0] = __pa(vcpu->arch.mmu.pae_root) | pm_mask; | |
2859 | ||
2860 | vcpu->arch.mmu.lm_root = lm_root; | |
2861 | } | |
2862 | ||
2863 | vcpu->arch.mmu.root_hpa = __pa(vcpu->arch.mmu.lm_root); | |
2864 | } | |
2865 | ||
8986ecc0 | 2866 | return 0; |
17ac10ad AK |
2867 | } |
2868 | ||
651dd37a JR |
2869 | static int mmu_alloc_roots(struct kvm_vcpu *vcpu) |
2870 | { | |
2871 | if (vcpu->arch.mmu.direct_map) | |
2872 | return mmu_alloc_direct_roots(vcpu); | |
2873 | else | |
2874 | return mmu_alloc_shadow_roots(vcpu); | |
2875 | } | |
2876 | ||
0ba73cda MT |
2877 | static void mmu_sync_roots(struct kvm_vcpu *vcpu) |
2878 | { | |
2879 | int i; | |
2880 | struct kvm_mmu_page *sp; | |
2881 | ||
81407ca5 JR |
2882 | if (vcpu->arch.mmu.direct_map) |
2883 | return; | |
2884 | ||
0ba73cda MT |
2885 | if (!VALID_PAGE(vcpu->arch.mmu.root_hpa)) |
2886 | return; | |
6903074c | 2887 | |
bebb106a | 2888 | vcpu_clear_mmio_info(vcpu, ~0ul); |
0375f7fa | 2889 | kvm_mmu_audit(vcpu, AUDIT_PRE_SYNC); |
81407ca5 | 2890 | if (vcpu->arch.mmu.root_level == PT64_ROOT_LEVEL) { |
0ba73cda MT |
2891 | hpa_t root = vcpu->arch.mmu.root_hpa; |
2892 | sp = page_header(root); | |
2893 | mmu_sync_children(vcpu, sp); | |
0375f7fa | 2894 | kvm_mmu_audit(vcpu, AUDIT_POST_SYNC); |
0ba73cda MT |
2895 | return; |
2896 | } | |
2897 | for (i = 0; i < 4; ++i) { | |
2898 | hpa_t root = vcpu->arch.mmu.pae_root[i]; | |
2899 | ||
8986ecc0 | 2900 | if (root && VALID_PAGE(root)) { |
0ba73cda MT |
2901 | root &= PT64_BASE_ADDR_MASK; |
2902 | sp = page_header(root); | |
2903 | mmu_sync_children(vcpu, sp); | |
2904 | } | |
2905 | } | |
0375f7fa | 2906 | kvm_mmu_audit(vcpu, AUDIT_POST_SYNC); |
0ba73cda MT |
2907 | } |
2908 | ||
2909 | void kvm_mmu_sync_roots(struct kvm_vcpu *vcpu) | |
2910 | { | |
2911 | spin_lock(&vcpu->kvm->mmu_lock); | |
2912 | mmu_sync_roots(vcpu); | |
6cffe8ca | 2913 | spin_unlock(&vcpu->kvm->mmu_lock); |
0ba73cda MT |
2914 | } |
2915 | ||
1871c602 | 2916 | static gpa_t nonpaging_gva_to_gpa(struct kvm_vcpu *vcpu, gva_t vaddr, |
ab9ae313 | 2917 | u32 access, struct x86_exception *exception) |
6aa8b732 | 2918 | { |
ab9ae313 AK |
2919 | if (exception) |
2920 | exception->error_code = 0; | |
6aa8b732 AK |
2921 | return vaddr; |
2922 | } | |
2923 | ||
6539e738 | 2924 | static gpa_t nonpaging_gva_to_gpa_nested(struct kvm_vcpu *vcpu, gva_t vaddr, |
ab9ae313 AK |
2925 | u32 access, |
2926 | struct x86_exception *exception) | |
6539e738 | 2927 | { |
ab9ae313 AK |
2928 | if (exception) |
2929 | exception->error_code = 0; | |
6539e738 JR |
2930 | return vcpu->arch.nested_mmu.translate_gpa(vcpu, vaddr, access); |
2931 | } | |
2932 | ||
ce88decf XG |
2933 | static bool quickly_check_mmio_pf(struct kvm_vcpu *vcpu, u64 addr, bool direct) |
2934 | { | |
2935 | if (direct) | |
2936 | return vcpu_match_mmio_gpa(vcpu, addr); | |
2937 | ||
2938 | return vcpu_match_mmio_gva(vcpu, addr); | |
2939 | } | |
2940 | ||
2941 | ||
2942 | /* | |
2943 | * On direct hosts, the last spte is only allows two states | |
2944 | * for mmio page fault: | |
2945 | * - It is the mmio spte | |
2946 | * - It is zapped or it is being zapped. | |
2947 | * | |
2948 | * This function completely checks the spte when the last spte | |
2949 | * is not the mmio spte. | |
2950 | */ | |
2951 | static bool check_direct_spte_mmio_pf(u64 spte) | |
2952 | { | |
2953 | return __check_direct_spte_mmio_pf(spte); | |
2954 | } | |
2955 | ||
2956 | static u64 walk_shadow_page_get_mmio_spte(struct kvm_vcpu *vcpu, u64 addr) | |
2957 | { | |
2958 | struct kvm_shadow_walk_iterator iterator; | |
2959 | u64 spte = 0ull; | |
2960 | ||
2961 | walk_shadow_page_lockless_begin(vcpu); | |
2962 | for_each_shadow_entry_lockless(vcpu, addr, iterator, spte) | |
2963 | if (!is_shadow_present_pte(spte)) | |
2964 | break; | |
2965 | walk_shadow_page_lockless_end(vcpu); | |
2966 | ||
2967 | return spte; | |
2968 | } | |
2969 | ||
2970 | /* | |
2971 | * If it is a real mmio page fault, return 1 and emulat the instruction | |
2972 | * directly, return 0 to let CPU fault again on the address, -1 is | |
2973 | * returned if bug is detected. | |
2974 | */ | |
2975 | int handle_mmio_page_fault_common(struct kvm_vcpu *vcpu, u64 addr, bool direct) | |
2976 | { | |
2977 | u64 spte; | |
2978 | ||
2979 | if (quickly_check_mmio_pf(vcpu, addr, direct)) | |
2980 | return 1; | |
2981 | ||
2982 | spte = walk_shadow_page_get_mmio_spte(vcpu, addr); | |
2983 | ||
2984 | if (is_mmio_spte(spte)) { | |
2985 | gfn_t gfn = get_mmio_spte_gfn(spte); | |
2986 | unsigned access = get_mmio_spte_access(spte); | |
2987 | ||
2988 | if (direct) | |
2989 | addr = 0; | |
4f022648 XG |
2990 | |
2991 | trace_handle_mmio_page_fault(addr, gfn, access); | |
ce88decf XG |
2992 | vcpu_cache_mmio_info(vcpu, addr, gfn, access); |
2993 | return 1; | |
2994 | } | |
2995 | ||
2996 | /* | |
2997 | * It's ok if the gva is remapped by other cpus on shadow guest, | |
2998 | * it's a BUG if the gfn is not a mmio page. | |
2999 | */ | |
3000 | if (direct && !check_direct_spte_mmio_pf(spte)) | |
3001 | return -1; | |
3002 | ||
3003 | /* | |
3004 | * If the page table is zapped by other cpus, let CPU fault again on | |
3005 | * the address. | |
3006 | */ | |
3007 | return 0; | |
3008 | } | |
3009 | EXPORT_SYMBOL_GPL(handle_mmio_page_fault_common); | |
3010 | ||
3011 | static int handle_mmio_page_fault(struct kvm_vcpu *vcpu, u64 addr, | |
3012 | u32 error_code, bool direct) | |
3013 | { | |
3014 | int ret; | |
3015 | ||
3016 | ret = handle_mmio_page_fault_common(vcpu, addr, direct); | |
3017 | WARN_ON(ret < 0); | |
3018 | return ret; | |
3019 | } | |
3020 | ||
6aa8b732 | 3021 | static int nonpaging_page_fault(struct kvm_vcpu *vcpu, gva_t gva, |
78b2c54a | 3022 | u32 error_code, bool prefault) |
6aa8b732 | 3023 | { |
e833240f | 3024 | gfn_t gfn; |
e2dec939 | 3025 | int r; |
6aa8b732 | 3026 | |
b8688d51 | 3027 | pgprintk("%s: gva %lx error %x\n", __func__, gva, error_code); |
ce88decf XG |
3028 | |
3029 | if (unlikely(error_code & PFERR_RSVD_MASK)) | |
3030 | return handle_mmio_page_fault(vcpu, gva, error_code, true); | |
3031 | ||
e2dec939 AK |
3032 | r = mmu_topup_memory_caches(vcpu); |
3033 | if (r) | |
3034 | return r; | |
714b93da | 3035 | |
6aa8b732 | 3036 | ASSERT(vcpu); |
ad312c7c | 3037 | ASSERT(VALID_PAGE(vcpu->arch.mmu.root_hpa)); |
6aa8b732 | 3038 | |
e833240f | 3039 | gfn = gva >> PAGE_SHIFT; |
6aa8b732 | 3040 | |
e833240f | 3041 | return nonpaging_map(vcpu, gva & PAGE_MASK, |
78b2c54a | 3042 | error_code & PFERR_WRITE_MASK, gfn, prefault); |
6aa8b732 AK |
3043 | } |
3044 | ||
7e1fbeac | 3045 | static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, gfn_t gfn) |
af585b92 GN |
3046 | { |
3047 | struct kvm_arch_async_pf arch; | |
fb67e14f | 3048 | |
7c90705b | 3049 | arch.token = (vcpu->arch.apf.id++ << 12) | vcpu->vcpu_id; |
af585b92 | 3050 | arch.gfn = gfn; |
c4806acd | 3051 | arch.direct_map = vcpu->arch.mmu.direct_map; |
fb67e14f | 3052 | arch.cr3 = vcpu->arch.mmu.get_cr3(vcpu); |
af585b92 GN |
3053 | |
3054 | return kvm_setup_async_pf(vcpu, gva, gfn, &arch); | |
3055 | } | |
3056 | ||
3057 | static bool can_do_async_pf(struct kvm_vcpu *vcpu) | |
3058 | { | |
3059 | if (unlikely(!irqchip_in_kernel(vcpu->kvm) || | |
3060 | kvm_event_needs_reinjection(vcpu))) | |
3061 | return false; | |
3062 | ||
3063 | return kvm_x86_ops->interrupt_allowed(vcpu); | |
3064 | } | |
3065 | ||
78b2c54a | 3066 | static bool try_async_pf(struct kvm_vcpu *vcpu, bool prefault, gfn_t gfn, |
612819c3 | 3067 | gva_t gva, pfn_t *pfn, bool write, bool *writable) |
af585b92 GN |
3068 | { |
3069 | bool async; | |
3070 | ||
612819c3 | 3071 | *pfn = gfn_to_pfn_async(vcpu->kvm, gfn, &async, write, writable); |
af585b92 GN |
3072 | |
3073 | if (!async) | |
3074 | return false; /* *pfn has correct page already */ | |
3075 | ||
3076 | put_page(pfn_to_page(*pfn)); | |
3077 | ||
78b2c54a | 3078 | if (!prefault && can_do_async_pf(vcpu)) { |
c9b263d2 | 3079 | trace_kvm_try_async_get_page(gva, gfn); |
af585b92 GN |
3080 | if (kvm_find_async_pf_gfn(vcpu, gfn)) { |
3081 | trace_kvm_async_pf_doublefault(gva, gfn); | |
3082 | kvm_make_request(KVM_REQ_APF_HALT, vcpu); | |
3083 | return true; | |
3084 | } else if (kvm_arch_setup_async_pf(vcpu, gva, gfn)) | |
3085 | return true; | |
3086 | } | |
3087 | ||
612819c3 | 3088 | *pfn = gfn_to_pfn_prot(vcpu->kvm, gfn, write, writable); |
af585b92 GN |
3089 | |
3090 | return false; | |
3091 | } | |
3092 | ||
56028d08 | 3093 | static int tdp_page_fault(struct kvm_vcpu *vcpu, gva_t gpa, u32 error_code, |
78b2c54a | 3094 | bool prefault) |
fb72d167 | 3095 | { |
35149e21 | 3096 | pfn_t pfn; |
fb72d167 | 3097 | int r; |
852e3c19 | 3098 | int level; |
936a5fe6 | 3099 | int force_pt_level; |
05da4558 | 3100 | gfn_t gfn = gpa >> PAGE_SHIFT; |
e930bffe | 3101 | unsigned long mmu_seq; |
612819c3 MT |
3102 | int write = error_code & PFERR_WRITE_MASK; |
3103 | bool map_writable; | |
fb72d167 JR |
3104 | |
3105 | ASSERT(vcpu); | |
3106 | ASSERT(VALID_PAGE(vcpu->arch.mmu.root_hpa)); | |
3107 | ||
ce88decf XG |
3108 | if (unlikely(error_code & PFERR_RSVD_MASK)) |
3109 | return handle_mmio_page_fault(vcpu, gpa, error_code, true); | |
3110 | ||
fb72d167 JR |
3111 | r = mmu_topup_memory_caches(vcpu); |
3112 | if (r) | |
3113 | return r; | |
3114 | ||
936a5fe6 AA |
3115 | force_pt_level = mapping_level_dirty_bitmap(vcpu, gfn); |
3116 | if (likely(!force_pt_level)) { | |
3117 | level = mapping_level(vcpu, gfn); | |
3118 | gfn &= ~(KVM_PAGES_PER_HPAGE(level) - 1); | |
3119 | } else | |
3120 | level = PT_PAGE_TABLE_LEVEL; | |
852e3c19 | 3121 | |
e930bffe | 3122 | mmu_seq = vcpu->kvm->mmu_notifier_seq; |
4c2155ce | 3123 | smp_rmb(); |
af585b92 | 3124 | |
78b2c54a | 3125 | if (try_async_pf(vcpu, prefault, gfn, gpa, &pfn, write, &map_writable)) |
af585b92 GN |
3126 | return 0; |
3127 | ||
d7c55201 XG |
3128 | if (handle_abnormal_pfn(vcpu, 0, gfn, pfn, ACC_ALL, &r)) |
3129 | return r; | |
3130 | ||
fb72d167 | 3131 | spin_lock(&vcpu->kvm->mmu_lock); |
e930bffe AA |
3132 | if (mmu_notifier_retry(vcpu, mmu_seq)) |
3133 | goto out_unlock; | |
fb72d167 | 3134 | kvm_mmu_free_some_pages(vcpu); |
936a5fe6 AA |
3135 | if (likely(!force_pt_level)) |
3136 | transparent_hugepage_adjust(vcpu, &gfn, &pfn, &level); | |
612819c3 | 3137 | r = __direct_map(vcpu, gpa, write, map_writable, |
2ec4739d | 3138 | level, gfn, pfn, prefault); |
fb72d167 | 3139 | spin_unlock(&vcpu->kvm->mmu_lock); |
fb72d167 JR |
3140 | |
3141 | return r; | |
e930bffe AA |
3142 | |
3143 | out_unlock: | |
3144 | spin_unlock(&vcpu->kvm->mmu_lock); | |
3145 | kvm_release_pfn_clean(pfn); | |
3146 | return 0; | |
fb72d167 JR |
3147 | } |
3148 | ||
6aa8b732 AK |
3149 | static void nonpaging_free(struct kvm_vcpu *vcpu) |
3150 | { | |
17ac10ad | 3151 | mmu_free_roots(vcpu); |
6aa8b732 AK |
3152 | } |
3153 | ||
52fde8df JR |
3154 | static int nonpaging_init_context(struct kvm_vcpu *vcpu, |
3155 | struct kvm_mmu *context) | |
6aa8b732 | 3156 | { |
6aa8b732 AK |
3157 | context->new_cr3 = nonpaging_new_cr3; |
3158 | context->page_fault = nonpaging_page_fault; | |
6aa8b732 AK |
3159 | context->gva_to_gpa = nonpaging_gva_to_gpa; |
3160 | context->free = nonpaging_free; | |
e8bc217a | 3161 | context->sync_page = nonpaging_sync_page; |
a7052897 | 3162 | context->invlpg = nonpaging_invlpg; |
0f53b5b1 | 3163 | context->update_pte = nonpaging_update_pte; |
cea0f0e7 | 3164 | context->root_level = 0; |
6aa8b732 | 3165 | context->shadow_root_level = PT32E_ROOT_LEVEL; |
17c3ba9d | 3166 | context->root_hpa = INVALID_PAGE; |
c5a78f2b | 3167 | context->direct_map = true; |
2d48a985 | 3168 | context->nx = false; |
6aa8b732 AK |
3169 | return 0; |
3170 | } | |
3171 | ||
d835dfec | 3172 | void kvm_mmu_flush_tlb(struct kvm_vcpu *vcpu) |
6aa8b732 | 3173 | { |
1165f5fe | 3174 | ++vcpu->stat.tlb_flush; |
a8eeb04a | 3175 | kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); |
6aa8b732 AK |
3176 | } |
3177 | ||
3178 | static void paging_new_cr3(struct kvm_vcpu *vcpu) | |
3179 | { | |
9f8fe504 | 3180 | pgprintk("%s: cr3 %lx\n", __func__, kvm_read_cr3(vcpu)); |
cea0f0e7 | 3181 | mmu_free_roots(vcpu); |
6aa8b732 AK |
3182 | } |
3183 | ||
5777ed34 JR |
3184 | static unsigned long get_cr3(struct kvm_vcpu *vcpu) |
3185 | { | |
9f8fe504 | 3186 | return kvm_read_cr3(vcpu); |
5777ed34 JR |
3187 | } |
3188 | ||
6389ee94 AK |
3189 | static void inject_page_fault(struct kvm_vcpu *vcpu, |
3190 | struct x86_exception *fault) | |
6aa8b732 | 3191 | { |
6389ee94 | 3192 | vcpu->arch.mmu.inject_page_fault(vcpu, fault); |
6aa8b732 AK |
3193 | } |
3194 | ||
6aa8b732 AK |
3195 | static void paging_free(struct kvm_vcpu *vcpu) |
3196 | { | |
3197 | nonpaging_free(vcpu); | |
3198 | } | |
3199 | ||
3241f22d | 3200 | static bool is_rsvd_bits_set(struct kvm_mmu *mmu, u64 gpte, int level) |
82725b20 DE |
3201 | { |
3202 | int bit7; | |
3203 | ||
3204 | bit7 = (gpte >> 7) & 1; | |
3241f22d | 3205 | return (gpte & mmu->rsvd_bits_mask[bit7][level-1]) != 0; |
82725b20 DE |
3206 | } |
3207 | ||
ce88decf XG |
3208 | static bool sync_mmio_spte(u64 *sptep, gfn_t gfn, unsigned access, |
3209 | int *nr_present) | |
3210 | { | |
3211 | if (unlikely(is_mmio_spte(*sptep))) { | |
3212 | if (gfn != get_mmio_spte_gfn(*sptep)) { | |
3213 | mmu_spte_clear_no_track(sptep); | |
3214 | return true; | |
3215 | } | |
3216 | ||
3217 | (*nr_present)++; | |
3218 | mark_mmio_spte(sptep, gfn, access); | |
3219 | return true; | |
3220 | } | |
3221 | ||
3222 | return false; | |
3223 | } | |
3224 | ||
6aa8b732 AK |
3225 | #define PTTYPE 64 |
3226 | #include "paging_tmpl.h" | |
3227 | #undef PTTYPE | |
3228 | ||
3229 | #define PTTYPE 32 | |
3230 | #include "paging_tmpl.h" | |
3231 | #undef PTTYPE | |
3232 | ||
52fde8df | 3233 | static void reset_rsvds_bits_mask(struct kvm_vcpu *vcpu, |
4d6931c3 | 3234 | struct kvm_mmu *context) |
82725b20 | 3235 | { |
82725b20 DE |
3236 | int maxphyaddr = cpuid_maxphyaddr(vcpu); |
3237 | u64 exb_bit_rsvd = 0; | |
3238 | ||
2d48a985 | 3239 | if (!context->nx) |
82725b20 | 3240 | exb_bit_rsvd = rsvd_bits(63, 63); |
4d6931c3 | 3241 | switch (context->root_level) { |
82725b20 DE |
3242 | case PT32_ROOT_LEVEL: |
3243 | /* no rsvd bits for 2 level 4K page table entries */ | |
3244 | context->rsvd_bits_mask[0][1] = 0; | |
3245 | context->rsvd_bits_mask[0][0] = 0; | |
f815bce8 XG |
3246 | context->rsvd_bits_mask[1][0] = context->rsvd_bits_mask[0][0]; |
3247 | ||
3248 | if (!is_pse(vcpu)) { | |
3249 | context->rsvd_bits_mask[1][1] = 0; | |
3250 | break; | |
3251 | } | |
3252 | ||
82725b20 DE |
3253 | if (is_cpuid_PSE36()) |
3254 | /* 36bits PSE 4MB page */ | |
3255 | context->rsvd_bits_mask[1][1] = rsvd_bits(17, 21); | |
3256 | else | |
3257 | /* 32 bits PSE 4MB page */ | |
3258 | context->rsvd_bits_mask[1][1] = rsvd_bits(13, 21); | |
82725b20 DE |
3259 | break; |
3260 | case PT32E_ROOT_LEVEL: | |
20c466b5 DE |
3261 | context->rsvd_bits_mask[0][2] = |
3262 | rsvd_bits(maxphyaddr, 63) | | |
3263 | rsvd_bits(7, 8) | rsvd_bits(1, 2); /* PDPTE */ | |
82725b20 | 3264 | context->rsvd_bits_mask[0][1] = exb_bit_rsvd | |
4c26b4cd | 3265 | rsvd_bits(maxphyaddr, 62); /* PDE */ |
82725b20 DE |
3266 | context->rsvd_bits_mask[0][0] = exb_bit_rsvd | |
3267 | rsvd_bits(maxphyaddr, 62); /* PTE */ | |
3268 | context->rsvd_bits_mask[1][1] = exb_bit_rsvd | | |
3269 | rsvd_bits(maxphyaddr, 62) | | |
3270 | rsvd_bits(13, 20); /* large page */ | |
f815bce8 | 3271 | context->rsvd_bits_mask[1][0] = context->rsvd_bits_mask[0][0]; |
82725b20 DE |
3272 | break; |
3273 | case PT64_ROOT_LEVEL: | |
3274 | context->rsvd_bits_mask[0][3] = exb_bit_rsvd | | |
3275 | rsvd_bits(maxphyaddr, 51) | rsvd_bits(7, 8); | |
3276 | context->rsvd_bits_mask[0][2] = exb_bit_rsvd | | |
3277 | rsvd_bits(maxphyaddr, 51) | rsvd_bits(7, 8); | |
3278 | context->rsvd_bits_mask[0][1] = exb_bit_rsvd | | |
4c26b4cd | 3279 | rsvd_bits(maxphyaddr, 51); |
82725b20 DE |
3280 | context->rsvd_bits_mask[0][0] = exb_bit_rsvd | |
3281 | rsvd_bits(maxphyaddr, 51); | |
3282 | context->rsvd_bits_mask[1][3] = context->rsvd_bits_mask[0][3]; | |
e04da980 JR |
3283 | context->rsvd_bits_mask[1][2] = exb_bit_rsvd | |
3284 | rsvd_bits(maxphyaddr, 51) | | |
3285 | rsvd_bits(13, 29); | |
82725b20 | 3286 | context->rsvd_bits_mask[1][1] = exb_bit_rsvd | |
4c26b4cd SY |
3287 | rsvd_bits(maxphyaddr, 51) | |
3288 | rsvd_bits(13, 20); /* large page */ | |
f815bce8 | 3289 | context->rsvd_bits_mask[1][0] = context->rsvd_bits_mask[0][0]; |
82725b20 DE |
3290 | break; |
3291 | } | |
3292 | } | |
3293 | ||
52fde8df JR |
3294 | static int paging64_init_context_common(struct kvm_vcpu *vcpu, |
3295 | struct kvm_mmu *context, | |
3296 | int level) | |
6aa8b732 | 3297 | { |
2d48a985 | 3298 | context->nx = is_nx(vcpu); |
4d6931c3 | 3299 | context->root_level = level; |
2d48a985 | 3300 | |
4d6931c3 | 3301 | reset_rsvds_bits_mask(vcpu, context); |
6aa8b732 AK |
3302 | |
3303 | ASSERT(is_pae(vcpu)); | |
3304 | context->new_cr3 = paging_new_cr3; | |
3305 | context->page_fault = paging64_page_fault; | |
6aa8b732 | 3306 | context->gva_to_gpa = paging64_gva_to_gpa; |
e8bc217a | 3307 | context->sync_page = paging64_sync_page; |
a7052897 | 3308 | context->invlpg = paging64_invlpg; |
0f53b5b1 | 3309 | context->update_pte = paging64_update_pte; |
6aa8b732 | 3310 | context->free = paging_free; |
17ac10ad | 3311 | context->shadow_root_level = level; |
17c3ba9d | 3312 | context->root_hpa = INVALID_PAGE; |
c5a78f2b | 3313 | context->direct_map = false; |
6aa8b732 AK |
3314 | return 0; |
3315 | } | |
3316 | ||
52fde8df JR |
3317 | static int paging64_init_context(struct kvm_vcpu *vcpu, |
3318 | struct kvm_mmu *context) | |
17ac10ad | 3319 | { |
52fde8df | 3320 | return paging64_init_context_common(vcpu, context, PT64_ROOT_LEVEL); |
17ac10ad AK |
3321 | } |
3322 | ||
52fde8df JR |
3323 | static int paging32_init_context(struct kvm_vcpu *vcpu, |
3324 | struct kvm_mmu *context) | |
6aa8b732 | 3325 | { |
2d48a985 | 3326 | context->nx = false; |
4d6931c3 | 3327 | context->root_level = PT32_ROOT_LEVEL; |
2d48a985 | 3328 | |
4d6931c3 | 3329 | reset_rsvds_bits_mask(vcpu, context); |
6aa8b732 AK |
3330 | |
3331 | context->new_cr3 = paging_new_cr3; | |
3332 | context->page_fault = paging32_page_fault; | |
6aa8b732 AK |
3333 | context->gva_to_gpa = paging32_gva_to_gpa; |
3334 | context->free = paging_free; | |
e8bc217a | 3335 | context->sync_page = paging32_sync_page; |
a7052897 | 3336 | context->invlpg = paging32_invlpg; |
0f53b5b1 | 3337 | context->update_pte = paging32_update_pte; |
6aa8b732 | 3338 | context->shadow_root_level = PT32E_ROOT_LEVEL; |
17c3ba9d | 3339 | context->root_hpa = INVALID_PAGE; |
c5a78f2b | 3340 | context->direct_map = false; |
6aa8b732 AK |
3341 | return 0; |
3342 | } | |
3343 | ||
52fde8df JR |
3344 | static int paging32E_init_context(struct kvm_vcpu *vcpu, |
3345 | struct kvm_mmu *context) | |
6aa8b732 | 3346 | { |
52fde8df | 3347 | return paging64_init_context_common(vcpu, context, PT32E_ROOT_LEVEL); |
6aa8b732 AK |
3348 | } |
3349 | ||
fb72d167 JR |
3350 | static int init_kvm_tdp_mmu(struct kvm_vcpu *vcpu) |
3351 | { | |
14dfe855 | 3352 | struct kvm_mmu *context = vcpu->arch.walk_mmu; |
fb72d167 | 3353 | |
c445f8ef | 3354 | context->base_role.word = 0; |
fb72d167 JR |
3355 | context->new_cr3 = nonpaging_new_cr3; |
3356 | context->page_fault = tdp_page_fault; | |
3357 | context->free = nonpaging_free; | |
e8bc217a | 3358 | context->sync_page = nonpaging_sync_page; |
a7052897 | 3359 | context->invlpg = nonpaging_invlpg; |
0f53b5b1 | 3360 | context->update_pte = nonpaging_update_pte; |
67253af5 | 3361 | context->shadow_root_level = kvm_x86_ops->get_tdp_level(); |
fb72d167 | 3362 | context->root_hpa = INVALID_PAGE; |
c5a78f2b | 3363 | context->direct_map = true; |
1c97f0a0 | 3364 | context->set_cr3 = kvm_x86_ops->set_tdp_cr3; |
5777ed34 | 3365 | context->get_cr3 = get_cr3; |
e4e517b4 | 3366 | context->get_pdptr = kvm_pdptr_read; |
cb659db8 | 3367 | context->inject_page_fault = kvm_inject_page_fault; |
fb72d167 JR |
3368 | |
3369 | if (!is_paging(vcpu)) { | |
2d48a985 | 3370 | context->nx = false; |
fb72d167 JR |
3371 | context->gva_to_gpa = nonpaging_gva_to_gpa; |
3372 | context->root_level = 0; | |
3373 | } else if (is_long_mode(vcpu)) { | |
2d48a985 | 3374 | context->nx = is_nx(vcpu); |
fb72d167 | 3375 | context->root_level = PT64_ROOT_LEVEL; |
4d6931c3 DB |
3376 | reset_rsvds_bits_mask(vcpu, context); |
3377 | context->gva_to_gpa = paging64_gva_to_gpa; | |
fb72d167 | 3378 | } else if (is_pae(vcpu)) { |
2d48a985 | 3379 | context->nx = is_nx(vcpu); |
fb72d167 | 3380 | context->root_level = PT32E_ROOT_LEVEL; |
4d6931c3 DB |
3381 | reset_rsvds_bits_mask(vcpu, context); |
3382 | context->gva_to_gpa = paging64_gva_to_gpa; | |
fb72d167 | 3383 | } else { |
2d48a985 | 3384 | context->nx = false; |
fb72d167 | 3385 | context->root_level = PT32_ROOT_LEVEL; |
4d6931c3 DB |
3386 | reset_rsvds_bits_mask(vcpu, context); |
3387 | context->gva_to_gpa = paging32_gva_to_gpa; | |
fb72d167 JR |
3388 | } |
3389 | ||
3390 | return 0; | |
3391 | } | |
3392 | ||
52fde8df | 3393 | int kvm_init_shadow_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *context) |
6aa8b732 | 3394 | { |
a770f6f2 | 3395 | int r; |
411c588d | 3396 | bool smep = kvm_read_cr4_bits(vcpu, X86_CR4_SMEP); |
6aa8b732 | 3397 | ASSERT(vcpu); |
ad312c7c | 3398 | ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa)); |
6aa8b732 AK |
3399 | |
3400 | if (!is_paging(vcpu)) | |
52fde8df | 3401 | r = nonpaging_init_context(vcpu, context); |
a9058ecd | 3402 | else if (is_long_mode(vcpu)) |
52fde8df | 3403 | r = paging64_init_context(vcpu, context); |
6aa8b732 | 3404 | else if (is_pae(vcpu)) |
52fde8df | 3405 | r = paging32E_init_context(vcpu, context); |
6aa8b732 | 3406 | else |
52fde8df | 3407 | r = paging32_init_context(vcpu, context); |
a770f6f2 | 3408 | |
5b7e0102 | 3409 | vcpu->arch.mmu.base_role.cr4_pae = !!is_pae(vcpu); |
f43addd4 | 3410 | vcpu->arch.mmu.base_role.cr0_wp = is_write_protection(vcpu); |
411c588d AK |
3411 | vcpu->arch.mmu.base_role.smep_andnot_wp |
3412 | = smep && !is_write_protection(vcpu); | |
52fde8df JR |
3413 | |
3414 | return r; | |
3415 | } | |
3416 | EXPORT_SYMBOL_GPL(kvm_init_shadow_mmu); | |
3417 | ||
3418 | static int init_kvm_softmmu(struct kvm_vcpu *vcpu) | |
3419 | { | |
14dfe855 | 3420 | int r = kvm_init_shadow_mmu(vcpu, vcpu->arch.walk_mmu); |
52fde8df | 3421 | |
14dfe855 JR |
3422 | vcpu->arch.walk_mmu->set_cr3 = kvm_x86_ops->set_cr3; |
3423 | vcpu->arch.walk_mmu->get_cr3 = get_cr3; | |
e4e517b4 | 3424 | vcpu->arch.walk_mmu->get_pdptr = kvm_pdptr_read; |
14dfe855 | 3425 | vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault; |
a770f6f2 AK |
3426 | |
3427 | return r; | |
6aa8b732 AK |
3428 | } |
3429 | ||
02f59dc9 JR |
3430 | static int init_kvm_nested_mmu(struct kvm_vcpu *vcpu) |
3431 | { | |
3432 | struct kvm_mmu *g_context = &vcpu->arch.nested_mmu; | |
3433 | ||
3434 | g_context->get_cr3 = get_cr3; | |
e4e517b4 | 3435 | g_context->get_pdptr = kvm_pdptr_read; |
02f59dc9 JR |
3436 | g_context->inject_page_fault = kvm_inject_page_fault; |
3437 | ||
3438 | /* | |
3439 | * Note that arch.mmu.gva_to_gpa translates l2_gva to l1_gpa. The | |
3440 | * translation of l2_gpa to l1_gpa addresses is done using the | |
3441 | * arch.nested_mmu.gva_to_gpa function. Basically the gva_to_gpa | |
3442 | * functions between mmu and nested_mmu are swapped. | |
3443 | */ | |
3444 | if (!is_paging(vcpu)) { | |
2d48a985 | 3445 | g_context->nx = false; |
02f59dc9 JR |
3446 | g_context->root_level = 0; |
3447 | g_context->gva_to_gpa = nonpaging_gva_to_gpa_nested; | |
3448 | } else if (is_long_mode(vcpu)) { | |
2d48a985 | 3449 | g_context->nx = is_nx(vcpu); |
02f59dc9 | 3450 | g_context->root_level = PT64_ROOT_LEVEL; |
4d6931c3 | 3451 | reset_rsvds_bits_mask(vcpu, g_context); |
02f59dc9 JR |
3452 | g_context->gva_to_gpa = paging64_gva_to_gpa_nested; |
3453 | } else if (is_pae(vcpu)) { | |
2d48a985 | 3454 | g_context->nx = is_nx(vcpu); |
02f59dc9 | 3455 | g_context->root_level = PT32E_ROOT_LEVEL; |
4d6931c3 | 3456 | reset_rsvds_bits_mask(vcpu, g_context); |
02f59dc9 JR |
3457 | g_context->gva_to_gpa = paging64_gva_to_gpa_nested; |
3458 | } else { | |
2d48a985 | 3459 | g_context->nx = false; |
02f59dc9 | 3460 | g_context->root_level = PT32_ROOT_LEVEL; |
4d6931c3 | 3461 | reset_rsvds_bits_mask(vcpu, g_context); |
02f59dc9 JR |
3462 | g_context->gva_to_gpa = paging32_gva_to_gpa_nested; |
3463 | } | |
3464 | ||
3465 | return 0; | |
3466 | } | |
3467 | ||
fb72d167 JR |
3468 | static int init_kvm_mmu(struct kvm_vcpu *vcpu) |
3469 | { | |
02f59dc9 JR |
3470 | if (mmu_is_nested(vcpu)) |
3471 | return init_kvm_nested_mmu(vcpu); | |
3472 | else if (tdp_enabled) | |
fb72d167 JR |
3473 | return init_kvm_tdp_mmu(vcpu); |
3474 | else | |
3475 | return init_kvm_softmmu(vcpu); | |
3476 | } | |
3477 | ||
6aa8b732 AK |
3478 | static void destroy_kvm_mmu(struct kvm_vcpu *vcpu) |
3479 | { | |
3480 | ASSERT(vcpu); | |
62ad0755 SY |
3481 | if (VALID_PAGE(vcpu->arch.mmu.root_hpa)) |
3482 | /* mmu.free() should set root_hpa = INVALID_PAGE */ | |
ad312c7c | 3483 | vcpu->arch.mmu.free(vcpu); |
6aa8b732 AK |
3484 | } |
3485 | ||
3486 | int kvm_mmu_reset_context(struct kvm_vcpu *vcpu) | |
17c3ba9d AK |
3487 | { |
3488 | destroy_kvm_mmu(vcpu); | |
f8f7e5ee | 3489 | return init_kvm_mmu(vcpu); |
17c3ba9d | 3490 | } |
8668a3c4 | 3491 | EXPORT_SYMBOL_GPL(kvm_mmu_reset_context); |
17c3ba9d AK |
3492 | |
3493 | int kvm_mmu_load(struct kvm_vcpu *vcpu) | |
6aa8b732 | 3494 | { |
714b93da AK |
3495 | int r; |
3496 | ||
e2dec939 | 3497 | r = mmu_topup_memory_caches(vcpu); |
17c3ba9d AK |
3498 | if (r) |
3499 | goto out; | |
8986ecc0 | 3500 | r = mmu_alloc_roots(vcpu); |
8facbbff | 3501 | spin_lock(&vcpu->kvm->mmu_lock); |
0ba73cda | 3502 | mmu_sync_roots(vcpu); |
aaee2c94 | 3503 | spin_unlock(&vcpu->kvm->mmu_lock); |
8986ecc0 MT |
3504 | if (r) |
3505 | goto out; | |
3662cb1c | 3506 | /* set_cr3() should ensure TLB has been flushed */ |
f43addd4 | 3507 | vcpu->arch.mmu.set_cr3(vcpu, vcpu->arch.mmu.root_hpa); |
714b93da AK |
3508 | out: |
3509 | return r; | |
6aa8b732 | 3510 | } |
17c3ba9d AK |
3511 | EXPORT_SYMBOL_GPL(kvm_mmu_load); |
3512 | ||
3513 | void kvm_mmu_unload(struct kvm_vcpu *vcpu) | |
3514 | { | |
3515 | mmu_free_roots(vcpu); | |
3516 | } | |
4b16184c | 3517 | EXPORT_SYMBOL_GPL(kvm_mmu_unload); |
6aa8b732 | 3518 | |
0028425f | 3519 | static void mmu_pte_write_new_pte(struct kvm_vcpu *vcpu, |
7c562522 XG |
3520 | struct kvm_mmu_page *sp, u64 *spte, |
3521 | const void *new) | |
0028425f | 3522 | { |
30945387 | 3523 | if (sp->role.level != PT_PAGE_TABLE_LEVEL) { |
7e4e4056 JR |
3524 | ++vcpu->kvm->stat.mmu_pde_zapped; |
3525 | return; | |
30945387 | 3526 | } |
0028425f | 3527 | |
4cee5764 | 3528 | ++vcpu->kvm->stat.mmu_pte_updated; |
7c562522 | 3529 | vcpu->arch.mmu.update_pte(vcpu, sp, spte, new); |
0028425f AK |
3530 | } |
3531 | ||
79539cec AK |
3532 | static bool need_remote_flush(u64 old, u64 new) |
3533 | { | |
3534 | if (!is_shadow_present_pte(old)) | |
3535 | return false; | |
3536 | if (!is_shadow_present_pte(new)) | |
3537 | return true; | |
3538 | if ((old ^ new) & PT64_BASE_ADDR_MASK) | |
3539 | return true; | |
3540 | old ^= PT64_NX_MASK; | |
3541 | new ^= PT64_NX_MASK; | |
3542 | return (old & ~new & PT64_PERM_MASK) != 0; | |
3543 | } | |
3544 | ||
0671a8e7 XG |
3545 | static void mmu_pte_write_flush_tlb(struct kvm_vcpu *vcpu, bool zap_page, |
3546 | bool remote_flush, bool local_flush) | |
79539cec | 3547 | { |
0671a8e7 XG |
3548 | if (zap_page) |
3549 | return; | |
3550 | ||
3551 | if (remote_flush) | |
79539cec | 3552 | kvm_flush_remote_tlbs(vcpu->kvm); |
0671a8e7 | 3553 | else if (local_flush) |
79539cec AK |
3554 | kvm_mmu_flush_tlb(vcpu); |
3555 | } | |
3556 | ||
889e5cbc XG |
3557 | static u64 mmu_pte_write_fetch_gpte(struct kvm_vcpu *vcpu, gpa_t *gpa, |
3558 | const u8 *new, int *bytes) | |
da4a00f0 | 3559 | { |
889e5cbc XG |
3560 | u64 gentry; |
3561 | int r; | |
72016f3a | 3562 | |
72016f3a AK |
3563 | /* |
3564 | * Assume that the pte write on a page table of the same type | |
49b26e26 XG |
3565 | * as the current vcpu paging mode since we update the sptes only |
3566 | * when they have the same mode. | |
72016f3a | 3567 | */ |
889e5cbc | 3568 | if (is_pae(vcpu) && *bytes == 4) { |
72016f3a | 3569 | /* Handle a 32-bit guest writing two halves of a 64-bit gpte */ |
889e5cbc XG |
3570 | *gpa &= ~(gpa_t)7; |
3571 | *bytes = 8; | |
3572 | r = kvm_read_guest(vcpu->kvm, *gpa, &gentry, min(*bytes, 8)); | |
72016f3a AK |
3573 | if (r) |
3574 | gentry = 0; | |
08e850c6 AK |
3575 | new = (const u8 *)&gentry; |
3576 | } | |
3577 | ||
889e5cbc | 3578 | switch (*bytes) { |
08e850c6 AK |
3579 | case 4: |
3580 | gentry = *(const u32 *)new; | |
3581 | break; | |
3582 | case 8: | |
3583 | gentry = *(const u64 *)new; | |
3584 | break; | |
3585 | default: | |
3586 | gentry = 0; | |
3587 | break; | |
72016f3a AK |
3588 | } |
3589 | ||
889e5cbc XG |
3590 | return gentry; |
3591 | } | |
3592 | ||
3593 | /* | |
3594 | * If we're seeing too many writes to a page, it may no longer be a page table, | |
3595 | * or we may be forking, in which case it is better to unmap the page. | |
3596 | */ | |
a138fe75 | 3597 | static bool detect_write_flooding(struct kvm_mmu_page *sp) |
889e5cbc | 3598 | { |
a30f47cb XG |
3599 | /* |
3600 | * Skip write-flooding detected for the sp whose level is 1, because | |
3601 | * it can become unsync, then the guest page is not write-protected. | |
3602 | */ | |
f71fa31f | 3603 | if (sp->role.level == PT_PAGE_TABLE_LEVEL) |
a30f47cb | 3604 | return false; |
3246af0e | 3605 | |
a30f47cb | 3606 | return ++sp->write_flooding_count >= 3; |
889e5cbc XG |
3607 | } |
3608 | ||
3609 | /* | |
3610 | * Misaligned accesses are too much trouble to fix up; also, they usually | |
3611 | * indicate a page is not used as a page table. | |
3612 | */ | |
3613 | static bool detect_write_misaligned(struct kvm_mmu_page *sp, gpa_t gpa, | |
3614 | int bytes) | |
3615 | { | |
3616 | unsigned offset, pte_size, misaligned; | |
3617 | ||
3618 | pgprintk("misaligned: gpa %llx bytes %d role %x\n", | |
3619 | gpa, bytes, sp->role.word); | |
3620 | ||
3621 | offset = offset_in_page(gpa); | |
3622 | pte_size = sp->role.cr4_pae ? 8 : 4; | |
5d9ca30e XG |
3623 | |
3624 | /* | |
3625 | * Sometimes, the OS only writes the last one bytes to update status | |
3626 | * bits, for example, in linux, andb instruction is used in clear_bit(). | |
3627 | */ | |
3628 | if (!(offset & (pte_size - 1)) && bytes == 1) | |
3629 | return false; | |
3630 | ||
889e5cbc XG |
3631 | misaligned = (offset ^ (offset + bytes - 1)) & ~(pte_size - 1); |
3632 | misaligned |= bytes < 4; | |
3633 | ||
3634 | return misaligned; | |
3635 | } | |
3636 | ||
3637 | static u64 *get_written_sptes(struct kvm_mmu_page *sp, gpa_t gpa, int *nspte) | |
3638 | { | |
3639 | unsigned page_offset, quadrant; | |
3640 | u64 *spte; | |
3641 | int level; | |
3642 | ||
3643 | page_offset = offset_in_page(gpa); | |
3644 | level = sp->role.level; | |
3645 | *nspte = 1; | |
3646 | if (!sp->role.cr4_pae) { | |
3647 | page_offset <<= 1; /* 32->64 */ | |
3648 | /* | |
3649 | * A 32-bit pde maps 4MB while the shadow pdes map | |
3650 | * only 2MB. So we need to double the offset again | |
3651 | * and zap two pdes instead of one. | |
3652 | */ | |
3653 | if (level == PT32_ROOT_LEVEL) { | |
3654 | page_offset &= ~7; /* kill rounding error */ | |
3655 | page_offset <<= 1; | |
3656 | *nspte = 2; | |
3657 | } | |
3658 | quadrant = page_offset >> PAGE_SHIFT; | |
3659 | page_offset &= ~PAGE_MASK; | |
3660 | if (quadrant != sp->role.quadrant) | |
3661 | return NULL; | |
3662 | } | |
3663 | ||
3664 | spte = &sp->spt[page_offset / sizeof(*spte)]; | |
3665 | return spte; | |
3666 | } | |
3667 | ||
3668 | void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa, | |
3669 | const u8 *new, int bytes) | |
3670 | { | |
3671 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
3672 | union kvm_mmu_page_role mask = { .word = 0 }; | |
3673 | struct kvm_mmu_page *sp; | |
3674 | struct hlist_node *node; | |
3675 | LIST_HEAD(invalid_list); | |
3676 | u64 entry, gentry, *spte; | |
3677 | int npte; | |
a30f47cb | 3678 | bool remote_flush, local_flush, zap_page; |
889e5cbc XG |
3679 | |
3680 | /* | |
3681 | * If we don't have indirect shadow pages, it means no page is | |
3682 | * write-protected, so we can exit simply. | |
3683 | */ | |
3684 | if (!ACCESS_ONCE(vcpu->kvm->arch.indirect_shadow_pages)) | |
3685 | return; | |
3686 | ||
3687 | zap_page = remote_flush = local_flush = false; | |
3688 | ||
3689 | pgprintk("%s: gpa %llx bytes %d\n", __func__, gpa, bytes); | |
3690 | ||
3691 | gentry = mmu_pte_write_fetch_gpte(vcpu, &gpa, new, &bytes); | |
3692 | ||
3693 | /* | |
3694 | * No need to care whether allocation memory is successful | |
3695 | * or not since pte prefetch is skiped if it does not have | |
3696 | * enough objects in the cache. | |
3697 | */ | |
3698 | mmu_topup_memory_caches(vcpu); | |
3699 | ||
3700 | spin_lock(&vcpu->kvm->mmu_lock); | |
3701 | ++vcpu->kvm->stat.mmu_pte_write; | |
0375f7fa | 3702 | kvm_mmu_audit(vcpu, AUDIT_PRE_PTE_WRITE); |
889e5cbc | 3703 | |
fa1de2bf | 3704 | mask.cr0_wp = mask.cr4_pae = mask.nxe = 1; |
f41d335a | 3705 | for_each_gfn_indirect_valid_sp(vcpu->kvm, sp, gfn, node) { |
a30f47cb | 3706 | if (detect_write_misaligned(sp, gpa, bytes) || |
a138fe75 | 3707 | detect_write_flooding(sp)) { |
0671a8e7 | 3708 | zap_page |= !!kvm_mmu_prepare_zap_page(vcpu->kvm, sp, |
f41d335a | 3709 | &invalid_list); |
4cee5764 | 3710 | ++vcpu->kvm->stat.mmu_flooded; |
0e7bc4b9 AK |
3711 | continue; |
3712 | } | |
889e5cbc XG |
3713 | |
3714 | spte = get_written_sptes(sp, gpa, &npte); | |
3715 | if (!spte) | |
3716 | continue; | |
3717 | ||
0671a8e7 | 3718 | local_flush = true; |
ac1b714e | 3719 | while (npte--) { |
79539cec | 3720 | entry = *spte; |
38e3b2b2 | 3721 | mmu_page_zap_pte(vcpu->kvm, sp, spte); |
fa1de2bf XG |
3722 | if (gentry && |
3723 | !((sp->role.word ^ vcpu->arch.mmu.base_role.word) | |
f759e2b4 | 3724 | & mask.word) && rmap_can_add(vcpu)) |
7c562522 | 3725 | mmu_pte_write_new_pte(vcpu, sp, spte, &gentry); |
0671a8e7 XG |
3726 | if (!remote_flush && need_remote_flush(entry, *spte)) |
3727 | remote_flush = true; | |
ac1b714e | 3728 | ++spte; |
9b7a0325 | 3729 | } |
9b7a0325 | 3730 | } |
0671a8e7 | 3731 | mmu_pte_write_flush_tlb(vcpu, zap_page, remote_flush, local_flush); |
d98ba053 | 3732 | kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list); |
0375f7fa | 3733 | kvm_mmu_audit(vcpu, AUDIT_POST_PTE_WRITE); |
aaee2c94 | 3734 | spin_unlock(&vcpu->kvm->mmu_lock); |
da4a00f0 AK |
3735 | } |
3736 | ||
a436036b AK |
3737 | int kvm_mmu_unprotect_page_virt(struct kvm_vcpu *vcpu, gva_t gva) |
3738 | { | |
10589a46 MT |
3739 | gpa_t gpa; |
3740 | int r; | |
a436036b | 3741 | |
c5a78f2b | 3742 | if (vcpu->arch.mmu.direct_map) |
60f24784 AK |
3743 | return 0; |
3744 | ||
1871c602 | 3745 | gpa = kvm_mmu_gva_to_gpa_read(vcpu, gva, NULL); |
10589a46 | 3746 | |
10589a46 | 3747 | r = kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT); |
1cb3f3ae | 3748 | |
10589a46 | 3749 | return r; |
a436036b | 3750 | } |
577bdc49 | 3751 | EXPORT_SYMBOL_GPL(kvm_mmu_unprotect_page_virt); |
a436036b | 3752 | |
22d95b12 | 3753 | void __kvm_mmu_free_some_pages(struct kvm_vcpu *vcpu) |
ebeace86 | 3754 | { |
d98ba053 | 3755 | LIST_HEAD(invalid_list); |
103ad25a | 3756 | |
e0df7b9f | 3757 | while (kvm_mmu_available_pages(vcpu->kvm) < KVM_REFILL_PAGES && |
3b80fffe | 3758 | !list_empty(&vcpu->kvm->arch.active_mmu_pages)) { |
4db35314 | 3759 | struct kvm_mmu_page *sp; |
ebeace86 | 3760 | |
f05e70ac | 3761 | sp = container_of(vcpu->kvm->arch.active_mmu_pages.prev, |
4db35314 | 3762 | struct kvm_mmu_page, link); |
e0df7b9f | 3763 | kvm_mmu_prepare_zap_page(vcpu->kvm, sp, &invalid_list); |
4cee5764 | 3764 | ++vcpu->kvm->stat.mmu_recycled; |
ebeace86 | 3765 | } |
aa6bd187 | 3766 | kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list); |
ebeace86 | 3767 | } |
ebeace86 | 3768 | |
1cb3f3ae XG |
3769 | static bool is_mmio_page_fault(struct kvm_vcpu *vcpu, gva_t addr) |
3770 | { | |
3771 | if (vcpu->arch.mmu.direct_map || mmu_is_nested(vcpu)) | |
3772 | return vcpu_match_mmio_gpa(vcpu, addr); | |
3773 | ||
3774 | return vcpu_match_mmio_gva(vcpu, addr); | |
3775 | } | |
3776 | ||
dc25e89e AP |
3777 | int kvm_mmu_page_fault(struct kvm_vcpu *vcpu, gva_t cr2, u32 error_code, |
3778 | void *insn, int insn_len) | |
3067714c | 3779 | { |
1cb3f3ae | 3780 | int r, emulation_type = EMULTYPE_RETRY; |
3067714c AK |
3781 | enum emulation_result er; |
3782 | ||
56028d08 | 3783 | r = vcpu->arch.mmu.page_fault(vcpu, cr2, error_code, false); |
3067714c AK |
3784 | if (r < 0) |
3785 | goto out; | |
3786 | ||
3787 | if (!r) { | |
3788 | r = 1; | |
3789 | goto out; | |
3790 | } | |
3791 | ||
1cb3f3ae XG |
3792 | if (is_mmio_page_fault(vcpu, cr2)) |
3793 | emulation_type = 0; | |
3794 | ||
3795 | er = x86_emulate_instruction(vcpu, cr2, emulation_type, insn, insn_len); | |
3067714c AK |
3796 | |
3797 | switch (er) { | |
3798 | case EMULATE_DONE: | |
3799 | return 1; | |
3800 | case EMULATE_DO_MMIO: | |
3801 | ++vcpu->stat.mmio_exits; | |
6d77dbfc | 3802 | /* fall through */ |
3067714c | 3803 | case EMULATE_FAIL: |
3f5d18a9 | 3804 | return 0; |
3067714c AK |
3805 | default: |
3806 | BUG(); | |
3807 | } | |
3808 | out: | |
3067714c AK |
3809 | return r; |
3810 | } | |
3811 | EXPORT_SYMBOL_GPL(kvm_mmu_page_fault); | |
3812 | ||
a7052897 MT |
3813 | void kvm_mmu_invlpg(struct kvm_vcpu *vcpu, gva_t gva) |
3814 | { | |
a7052897 | 3815 | vcpu->arch.mmu.invlpg(vcpu, gva); |
a7052897 MT |
3816 | kvm_mmu_flush_tlb(vcpu); |
3817 | ++vcpu->stat.invlpg; | |
3818 | } | |
3819 | EXPORT_SYMBOL_GPL(kvm_mmu_invlpg); | |
3820 | ||
18552672 JR |
3821 | void kvm_enable_tdp(void) |
3822 | { | |
3823 | tdp_enabled = true; | |
3824 | } | |
3825 | EXPORT_SYMBOL_GPL(kvm_enable_tdp); | |
3826 | ||
5f4cb662 JR |
3827 | void kvm_disable_tdp(void) |
3828 | { | |
3829 | tdp_enabled = false; | |
3830 | } | |
3831 | EXPORT_SYMBOL_GPL(kvm_disable_tdp); | |
3832 | ||
6aa8b732 AK |
3833 | static void free_mmu_pages(struct kvm_vcpu *vcpu) |
3834 | { | |
ad312c7c | 3835 | free_page((unsigned long)vcpu->arch.mmu.pae_root); |
81407ca5 JR |
3836 | if (vcpu->arch.mmu.lm_root != NULL) |
3837 | free_page((unsigned long)vcpu->arch.mmu.lm_root); | |
6aa8b732 AK |
3838 | } |
3839 | ||
3840 | static int alloc_mmu_pages(struct kvm_vcpu *vcpu) | |
3841 | { | |
17ac10ad | 3842 | struct page *page; |
6aa8b732 AK |
3843 | int i; |
3844 | ||
3845 | ASSERT(vcpu); | |
3846 | ||
17ac10ad AK |
3847 | /* |
3848 | * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64. | |
3849 | * Therefore we need to allocate shadow page tables in the first | |
3850 | * 4GB of memory, which happens to fit the DMA32 zone. | |
3851 | */ | |
3852 | page = alloc_page(GFP_KERNEL | __GFP_DMA32); | |
3853 | if (!page) | |
d7fa6ab2 WY |
3854 | return -ENOMEM; |
3855 | ||
ad312c7c | 3856 | vcpu->arch.mmu.pae_root = page_address(page); |
17ac10ad | 3857 | for (i = 0; i < 4; ++i) |
ad312c7c | 3858 | vcpu->arch.mmu.pae_root[i] = INVALID_PAGE; |
17ac10ad | 3859 | |
6aa8b732 | 3860 | return 0; |
6aa8b732 AK |
3861 | } |
3862 | ||
8018c27b | 3863 | int kvm_mmu_create(struct kvm_vcpu *vcpu) |
6aa8b732 | 3864 | { |
6aa8b732 | 3865 | ASSERT(vcpu); |
e459e322 XG |
3866 | |
3867 | vcpu->arch.walk_mmu = &vcpu->arch.mmu; | |
3868 | vcpu->arch.mmu.root_hpa = INVALID_PAGE; | |
3869 | vcpu->arch.mmu.translate_gpa = translate_gpa; | |
3870 | vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa; | |
6aa8b732 | 3871 | |
8018c27b IM |
3872 | return alloc_mmu_pages(vcpu); |
3873 | } | |
6aa8b732 | 3874 | |
8018c27b IM |
3875 | int kvm_mmu_setup(struct kvm_vcpu *vcpu) |
3876 | { | |
3877 | ASSERT(vcpu); | |
ad312c7c | 3878 | ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa)); |
2c264957 | 3879 | |
8018c27b | 3880 | return init_kvm_mmu(vcpu); |
6aa8b732 AK |
3881 | } |
3882 | ||
90cb0529 | 3883 | void kvm_mmu_slot_remove_write_access(struct kvm *kvm, int slot) |
6aa8b732 | 3884 | { |
4db35314 | 3885 | struct kvm_mmu_page *sp; |
6aa8b732 | 3886 | |
f05e70ac | 3887 | list_for_each_entry(sp, &kvm->arch.active_mmu_pages, link) { |
6aa8b732 AK |
3888 | int i; |
3889 | u64 *pt; | |
3890 | ||
291f26bc | 3891 | if (!test_bit(slot, sp->slot_bitmap)) |
6aa8b732 AK |
3892 | continue; |
3893 | ||
4db35314 | 3894 | pt = sp->spt; |
8234b22e | 3895 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) { |
da8dc75f XG |
3896 | if (!is_shadow_present_pte(pt[i]) || |
3897 | !is_last_spte(pt[i], sp->role.level)) | |
3898 | continue; | |
3899 | ||
3900 | if (is_large_pte(pt[i])) { | |
c3707958 | 3901 | drop_spte(kvm, &pt[i]); |
8234b22e | 3902 | --kvm->stat.lpages; |
da8dc75f | 3903 | continue; |
8234b22e | 3904 | } |
da8dc75f | 3905 | |
6aa8b732 | 3906 | /* avoid RMW */ |
01c168ac | 3907 | if (is_writable_pte(pt[i])) |
1df9f2dc XG |
3908 | mmu_spte_update(&pt[i], |
3909 | pt[i] & ~PT_WRITABLE_MASK); | |
8234b22e | 3910 | } |
6aa8b732 | 3911 | } |
171d595d | 3912 | kvm_flush_remote_tlbs(kvm); |
6aa8b732 | 3913 | } |
37a7d8b0 | 3914 | |
90cb0529 | 3915 | void kvm_mmu_zap_all(struct kvm *kvm) |
e0fa826f | 3916 | { |
4db35314 | 3917 | struct kvm_mmu_page *sp, *node; |
d98ba053 | 3918 | LIST_HEAD(invalid_list); |
e0fa826f | 3919 | |
aaee2c94 | 3920 | spin_lock(&kvm->mmu_lock); |
3246af0e | 3921 | restart: |
f05e70ac | 3922 | list_for_each_entry_safe(sp, node, &kvm->arch.active_mmu_pages, link) |
d98ba053 | 3923 | if (kvm_mmu_prepare_zap_page(kvm, sp, &invalid_list)) |
3246af0e XG |
3924 | goto restart; |
3925 | ||
d98ba053 | 3926 | kvm_mmu_commit_zap_page(kvm, &invalid_list); |
aaee2c94 | 3927 | spin_unlock(&kvm->mmu_lock); |
e0fa826f DL |
3928 | } |
3929 | ||
3d56cbdf JK |
3930 | static void kvm_mmu_remove_some_alloc_mmu_pages(struct kvm *kvm, |
3931 | struct list_head *invalid_list) | |
3ee16c81 IE |
3932 | { |
3933 | struct kvm_mmu_page *page; | |
3934 | ||
3935 | page = container_of(kvm->arch.active_mmu_pages.prev, | |
3936 | struct kvm_mmu_page, link); | |
3d56cbdf | 3937 | kvm_mmu_prepare_zap_page(kvm, page, invalid_list); |
3ee16c81 IE |
3938 | } |
3939 | ||
1495f230 | 3940 | static int mmu_shrink(struct shrinker *shrink, struct shrink_control *sc) |
3ee16c81 IE |
3941 | { |
3942 | struct kvm *kvm; | |
1495f230 | 3943 | int nr_to_scan = sc->nr_to_scan; |
45221ab6 DH |
3944 | |
3945 | if (nr_to_scan == 0) | |
3946 | goto out; | |
3ee16c81 | 3947 | |
e935b837 | 3948 | raw_spin_lock(&kvm_lock); |
3ee16c81 IE |
3949 | |
3950 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
3d56cbdf | 3951 | int idx; |
d98ba053 | 3952 | LIST_HEAD(invalid_list); |
3ee16c81 | 3953 | |
19526396 GN |
3954 | /* |
3955 | * n_used_mmu_pages is accessed without holding kvm->mmu_lock | |
3956 | * here. We may skip a VM instance errorneosly, but we do not | |
3957 | * want to shrink a VM that only started to populate its MMU | |
3958 | * anyway. | |
3959 | */ | |
3960 | if (kvm->arch.n_used_mmu_pages > 0) { | |
3961 | if (!nr_to_scan--) | |
3962 | break; | |
3963 | continue; | |
3964 | } | |
3965 | ||
f656ce01 | 3966 | idx = srcu_read_lock(&kvm->srcu); |
3ee16c81 | 3967 | spin_lock(&kvm->mmu_lock); |
3ee16c81 | 3968 | |
19526396 | 3969 | kvm_mmu_remove_some_alloc_mmu_pages(kvm, &invalid_list); |
d98ba053 | 3970 | kvm_mmu_commit_zap_page(kvm, &invalid_list); |
19526396 | 3971 | |
3ee16c81 | 3972 | spin_unlock(&kvm->mmu_lock); |
f656ce01 | 3973 | srcu_read_unlock(&kvm->srcu, idx); |
19526396 GN |
3974 | |
3975 | list_move_tail(&kvm->vm_list, &vm_list); | |
3976 | break; | |
3ee16c81 | 3977 | } |
3ee16c81 | 3978 | |
e935b837 | 3979 | raw_spin_unlock(&kvm_lock); |
3ee16c81 | 3980 | |
45221ab6 DH |
3981 | out: |
3982 | return percpu_counter_read_positive(&kvm_total_used_mmu_pages); | |
3ee16c81 IE |
3983 | } |
3984 | ||
3985 | static struct shrinker mmu_shrinker = { | |
3986 | .shrink = mmu_shrink, | |
3987 | .seeks = DEFAULT_SEEKS * 10, | |
3988 | }; | |
3989 | ||
2ddfd20e | 3990 | static void mmu_destroy_caches(void) |
b5a33a75 | 3991 | { |
53c07b18 XG |
3992 | if (pte_list_desc_cache) |
3993 | kmem_cache_destroy(pte_list_desc_cache); | |
d3d25b04 AK |
3994 | if (mmu_page_header_cache) |
3995 | kmem_cache_destroy(mmu_page_header_cache); | |
b5a33a75 AK |
3996 | } |
3997 | ||
3998 | int kvm_mmu_module_init(void) | |
3999 | { | |
53c07b18 XG |
4000 | pte_list_desc_cache = kmem_cache_create("pte_list_desc", |
4001 | sizeof(struct pte_list_desc), | |
20c2df83 | 4002 | 0, 0, NULL); |
53c07b18 | 4003 | if (!pte_list_desc_cache) |
b5a33a75 AK |
4004 | goto nomem; |
4005 | ||
d3d25b04 AK |
4006 | mmu_page_header_cache = kmem_cache_create("kvm_mmu_page_header", |
4007 | sizeof(struct kvm_mmu_page), | |
20c2df83 | 4008 | 0, 0, NULL); |
d3d25b04 AK |
4009 | if (!mmu_page_header_cache) |
4010 | goto nomem; | |
4011 | ||
45bf21a8 WY |
4012 | if (percpu_counter_init(&kvm_total_used_mmu_pages, 0)) |
4013 | goto nomem; | |
4014 | ||
3ee16c81 IE |
4015 | register_shrinker(&mmu_shrinker); |
4016 | ||
b5a33a75 AK |
4017 | return 0; |
4018 | ||
4019 | nomem: | |
3ee16c81 | 4020 | mmu_destroy_caches(); |
b5a33a75 AK |
4021 | return -ENOMEM; |
4022 | } | |
4023 | ||
3ad82a7e ZX |
4024 | /* |
4025 | * Caculate mmu pages needed for kvm. | |
4026 | */ | |
4027 | unsigned int kvm_mmu_calculate_mmu_pages(struct kvm *kvm) | |
4028 | { | |
3ad82a7e ZX |
4029 | unsigned int nr_mmu_pages; |
4030 | unsigned int nr_pages = 0; | |
bc6678a3 | 4031 | struct kvm_memslots *slots; |
be6ba0f0 | 4032 | struct kvm_memory_slot *memslot; |
3ad82a7e | 4033 | |
90d83dc3 LJ |
4034 | slots = kvm_memslots(kvm); |
4035 | ||
be6ba0f0 XG |
4036 | kvm_for_each_memslot(memslot, slots) |
4037 | nr_pages += memslot->npages; | |
3ad82a7e ZX |
4038 | |
4039 | nr_mmu_pages = nr_pages * KVM_PERMILLE_MMU_PAGES / 1000; | |
4040 | nr_mmu_pages = max(nr_mmu_pages, | |
4041 | (unsigned int) KVM_MIN_ALLOC_MMU_PAGES); | |
4042 | ||
4043 | return nr_mmu_pages; | |
4044 | } | |
4045 | ||
94d8b056 MT |
4046 | int kvm_mmu_get_spte_hierarchy(struct kvm_vcpu *vcpu, u64 addr, u64 sptes[4]) |
4047 | { | |
4048 | struct kvm_shadow_walk_iterator iterator; | |
c2a2ac2b | 4049 | u64 spte; |
94d8b056 MT |
4050 | int nr_sptes = 0; |
4051 | ||
c2a2ac2b XG |
4052 | walk_shadow_page_lockless_begin(vcpu); |
4053 | for_each_shadow_entry_lockless(vcpu, addr, iterator, spte) { | |
4054 | sptes[iterator.level-1] = spte; | |
94d8b056 | 4055 | nr_sptes++; |
c2a2ac2b | 4056 | if (!is_shadow_present_pte(spte)) |
94d8b056 MT |
4057 | break; |
4058 | } | |
c2a2ac2b | 4059 | walk_shadow_page_lockless_end(vcpu); |
94d8b056 MT |
4060 | |
4061 | return nr_sptes; | |
4062 | } | |
4063 | EXPORT_SYMBOL_GPL(kvm_mmu_get_spte_hierarchy); | |
4064 | ||
c42fffe3 XG |
4065 | void kvm_mmu_destroy(struct kvm_vcpu *vcpu) |
4066 | { | |
4067 | ASSERT(vcpu); | |
4068 | ||
4069 | destroy_kvm_mmu(vcpu); | |
4070 | free_mmu_pages(vcpu); | |
4071 | mmu_free_memory_caches(vcpu); | |
b034cf01 XG |
4072 | } |
4073 | ||
b034cf01 XG |
4074 | void kvm_mmu_module_exit(void) |
4075 | { | |
4076 | mmu_destroy_caches(); | |
4077 | percpu_counter_destroy(&kvm_total_used_mmu_pages); | |
4078 | unregister_shrinker(&mmu_shrinker); | |
c42fffe3 XG |
4079 | mmu_audit_disable(); |
4080 | } |