]>
Commit | Line | Data |
---|---|---|
20c8ccb1 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
6aa8b732 AK |
2 | /* |
3 | * Kernel-based Virtual Machine driver for Linux | |
4 | * | |
5 | * This module enables machines with Intel VT-x extensions to run virtual | |
6 | * machines without emulation or binary translation. | |
7 | * | |
8 | * MMU support | |
9 | * | |
10 | * Copyright (C) 2006 Qumranet, Inc. | |
9611c187 | 11 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
6aa8b732 AK |
12 | * |
13 | * Authors: | |
14 | * Yaniv Kamay <yaniv@qumranet.com> | |
15 | * Avi Kivity <avi@qumranet.com> | |
6aa8b732 | 16 | */ |
8d20bd63 | 17 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
e495606d | 18 | |
af585b92 | 19 | #include "irq.h" |
88197e6a | 20 | #include "ioapic.h" |
1d737c8a | 21 | #include "mmu.h" |
6ca9a6f3 | 22 | #include "mmu_internal.h" |
fe5db27d | 23 | #include "tdp_mmu.h" |
836a1b3c | 24 | #include "x86.h" |
6de4f3ad | 25 | #include "kvm_cache_regs.h" |
b0b42197 | 26 | #include "smm.h" |
2f728d66 | 27 | #include "kvm_emulate.h" |
58ea7cf7 | 28 | #include "page_track.h" |
5f7dde7b | 29 | #include "cpuid.h" |
5a9624af | 30 | #include "spte.h" |
e495606d | 31 | |
edf88417 | 32 | #include <linux/kvm_host.h> |
6aa8b732 AK |
33 | #include <linux/types.h> |
34 | #include <linux/string.h> | |
6aa8b732 AK |
35 | #include <linux/mm.h> |
36 | #include <linux/highmem.h> | |
1767e931 PG |
37 | #include <linux/moduleparam.h> |
38 | #include <linux/export.h> | |
448353ca | 39 | #include <linux/swap.h> |
05da4558 | 40 | #include <linux/hugetlb.h> |
2f333bcb | 41 | #include <linux/compiler.h> |
bc6678a3 | 42 | #include <linux/srcu.h> |
5a0e3ad6 | 43 | #include <linux/slab.h> |
3f07c014 | 44 | #include <linux/sched/signal.h> |
bf998156 | 45 | #include <linux/uaccess.h> |
114df303 | 46 | #include <linux/hash.h> |
f160c7b7 | 47 | #include <linux/kern_levels.h> |
11b36fe7 | 48 | #include <linux/kstrtox.h> |
1aa9b957 | 49 | #include <linux/kthread.h> |
6aa8b732 | 50 | |
e495606d | 51 | #include <asm/page.h> |
eb243d1d | 52 | #include <asm/memtype.h> |
e495606d | 53 | #include <asm/cmpxchg.h> |
4e542370 | 54 | #include <asm/io.h> |
4a98623d | 55 | #include <asm/set_memory.h> |
13673a90 | 56 | #include <asm/vmx.h> |
58ea7cf7 | 57 | |
1261bfa3 | 58 | #include "trace.h" |
6aa8b732 | 59 | |
b8e8c830 PB |
60 | extern bool itlb_multihit_kvm_mitigation; |
61 | ||
0b210faf SC |
62 | static bool nx_hugepage_mitigation_hard_disabled; |
63 | ||
a9d6496d | 64 | int __read_mostly nx_huge_pages = -1; |
4dfe4f40 | 65 | static uint __read_mostly nx_huge_pages_recovery_period_ms; |
13fb5927 PB |
66 | #ifdef CONFIG_PREEMPT_RT |
67 | /* Recovery can cause latency spikes, disable it for PREEMPT_RT. */ | |
68 | static uint __read_mostly nx_huge_pages_recovery_ratio = 0; | |
69 | #else | |
1aa9b957 | 70 | static uint __read_mostly nx_huge_pages_recovery_ratio = 60; |
13fb5927 | 71 | #endif |
b8e8c830 | 72 | |
0b210faf | 73 | static int get_nx_huge_pages(char *buffer, const struct kernel_param *kp); |
b8e8c830 | 74 | static int set_nx_huge_pages(const char *val, const struct kernel_param *kp); |
4dfe4f40 | 75 | static int set_nx_huge_pages_recovery_param(const char *val, const struct kernel_param *kp); |
b8e8c830 | 76 | |
d5d6c18d | 77 | static const struct kernel_param_ops nx_huge_pages_ops = { |
b8e8c830 | 78 | .set = set_nx_huge_pages, |
0b210faf | 79 | .get = get_nx_huge_pages, |
b8e8c830 PB |
80 | }; |
81 | ||
4dfe4f40 JS |
82 | static const struct kernel_param_ops nx_huge_pages_recovery_param_ops = { |
83 | .set = set_nx_huge_pages_recovery_param, | |
1aa9b957 JS |
84 | .get = param_get_uint, |
85 | }; | |
86 | ||
b8e8c830 PB |
87 | module_param_cb(nx_huge_pages, &nx_huge_pages_ops, &nx_huge_pages, 0644); |
88 | __MODULE_PARM_TYPE(nx_huge_pages, "bool"); | |
4dfe4f40 | 89 | module_param_cb(nx_huge_pages_recovery_ratio, &nx_huge_pages_recovery_param_ops, |
1aa9b957 JS |
90 | &nx_huge_pages_recovery_ratio, 0644); |
91 | __MODULE_PARM_TYPE(nx_huge_pages_recovery_ratio, "uint"); | |
4dfe4f40 JS |
92 | module_param_cb(nx_huge_pages_recovery_period_ms, &nx_huge_pages_recovery_param_ops, |
93 | &nx_huge_pages_recovery_period_ms, 0644); | |
94 | __MODULE_PARM_TYPE(nx_huge_pages_recovery_period_ms, "uint"); | |
b8e8c830 | 95 | |
71fe7013 SC |
96 | static bool __read_mostly force_flush_and_sync_on_reuse; |
97 | module_param_named(flush_on_reuse, force_flush_and_sync_on_reuse, bool, 0644); | |
98 | ||
18552672 JR |
99 | /* |
100 | * When setting this variable to true it enables Two-Dimensional-Paging | |
101 | * where the hardware walks 2 page tables: | |
102 | * 1. the guest-virtual to guest-physical | |
103 | * 2. while doing 1. it walks guest-physical to host-physical | |
104 | * If the hardware supports that we don't need to do shadow paging. | |
105 | */ | |
2f333bcb | 106 | bool tdp_enabled = false; |
18552672 | 107 | |
7f604e92 | 108 | static bool __ro_after_init tdp_mmu_allowed; |
1f98f2bd DM |
109 | |
110 | #ifdef CONFIG_X86_64 | |
111 | bool __read_mostly tdp_mmu_enabled = true; | |
112 | module_param_named(tdp_mmu, tdp_mmu_enabled, bool, 0444); | |
113 | #endif | |
114 | ||
1d92d2e8 | 115 | static int max_huge_page_level __read_mostly; |
746700d2 | 116 | static int tdp_root_level __read_mostly; |
83013059 | 117 | static int max_tdp_level __read_mostly; |
703c335d | 118 | |
957ed9ef XG |
119 | #define PTE_PREFETCH_NUM 8 |
120 | ||
90bb6fc5 AK |
121 | #include <trace/events/kvm.h> |
122 | ||
dc1cff96 | 123 | /* make pte_list_desc fit well in cache lines */ |
13236e25 | 124 | #define PTE_LIST_EXT 14 |
220f773a | 125 | |
13236e25 | 126 | /* |
141705b7 LJ |
127 | * struct pte_list_desc is the core data structure used to implement a custom |
128 | * list for tracking a set of related SPTEs, e.g. all the SPTEs that map a | |
129 | * given GFN when used in the context of rmaps. Using a custom list allows KVM | |
130 | * to optimize for the common case where many GFNs will have at most a handful | |
131 | * of SPTEs pointing at them, i.e. allows packing multiple SPTEs into a small | |
132 | * memory footprint, which in turn improves runtime performance by exploiting | |
133 | * cache locality. | |
134 | * | |
135 | * A list is comprised of one or more pte_list_desc objects (descriptors). | |
136 | * Each individual descriptor stores up to PTE_LIST_EXT SPTEs. If a descriptor | |
137 | * is full and a new SPTEs needs to be added, a new descriptor is allocated and | |
138 | * becomes the head of the list. This means that by definitions, all tail | |
139 | * descriptors are full. | |
140 | * | |
141 | * Note, the meta data fields are deliberately placed at the start of the | |
142 | * structure to optimize the cacheline layout; accessing the descriptor will | |
143 | * touch only a single cacheline so long as @spte_count<=6 (or if only the | |
144 | * descriptors metadata is accessed). | |
13236e25 | 145 | */ |
53c07b18 | 146 | struct pte_list_desc { |
53c07b18 | 147 | struct pte_list_desc *more; |
141705b7 LJ |
148 | /* The number of PTEs stored in _this_ descriptor. */ |
149 | u32 spte_count; | |
150 | /* The number of PTEs stored in all tails of this descriptor. */ | |
151 | u32 tail_count; | |
13236e25 | 152 | u64 *sptes[PTE_LIST_EXT]; |
cd4a4e53 AK |
153 | }; |
154 | ||
2d11123a AK |
155 | struct kvm_shadow_walk_iterator { |
156 | u64 addr; | |
157 | hpa_t shadow_addr; | |
2d11123a | 158 | u64 *sptep; |
dd3bfd59 | 159 | int level; |
2d11123a AK |
160 | unsigned index; |
161 | }; | |
162 | ||
7eb77e9f JS |
163 | #define for_each_shadow_entry_using_root(_vcpu, _root, _addr, _walker) \ |
164 | for (shadow_walk_init_using_root(&(_walker), (_vcpu), \ | |
165 | (_root), (_addr)); \ | |
166 | shadow_walk_okay(&(_walker)); \ | |
167 | shadow_walk_next(&(_walker))) | |
168 | ||
169 | #define for_each_shadow_entry(_vcpu, _addr, _walker) \ | |
2d11123a AK |
170 | for (shadow_walk_init(&(_walker), _vcpu, _addr); \ |
171 | shadow_walk_okay(&(_walker)); \ | |
172 | shadow_walk_next(&(_walker))) | |
173 | ||
c2a2ac2b XG |
174 | #define for_each_shadow_entry_lockless(_vcpu, _addr, _walker, spte) \ |
175 | for (shadow_walk_init(&(_walker), _vcpu, _addr); \ | |
176 | shadow_walk_okay(&(_walker)) && \ | |
177 | ({ spte = mmu_spte_get_lockless(_walker.sptep); 1; }); \ | |
178 | __shadow_walk_next(&(_walker), spte)) | |
179 | ||
53c07b18 | 180 | static struct kmem_cache *pte_list_desc_cache; |
02c00b3a | 181 | struct kmem_cache *mmu_page_header_cache; |
45221ab6 | 182 | static struct percpu_counter kvm_total_used_mmu_pages; |
b5a33a75 | 183 | |
ce88decf XG |
184 | static void mmu_spte_set(u64 *sptep, u64 spte); |
185 | ||
594e91a1 SC |
186 | struct kvm_mmu_role_regs { |
187 | const unsigned long cr0; | |
188 | const unsigned long cr4; | |
189 | const u64 efer; | |
190 | }; | |
191 | ||
335e192a PB |
192 | #define CREATE_TRACE_POINTS |
193 | #include "mmutrace.h" | |
194 | ||
594e91a1 SC |
195 | /* |
196 | * Yes, lot's of underscores. They're a hint that you probably shouldn't be | |
7a458f0e | 197 | * reading from the role_regs. Once the root_role is constructed, it becomes |
594e91a1 SC |
198 | * the single source of truth for the MMU's state. |
199 | */ | |
200 | #define BUILD_MMU_ROLE_REGS_ACCESSOR(reg, name, flag) \ | |
82ffa13f PB |
201 | static inline bool __maybe_unused \ |
202 | ____is_##reg##_##name(const struct kvm_mmu_role_regs *regs) \ | |
594e91a1 SC |
203 | { \ |
204 | return !!(regs->reg & flag); \ | |
205 | } | |
206 | BUILD_MMU_ROLE_REGS_ACCESSOR(cr0, pg, X86_CR0_PG); | |
207 | BUILD_MMU_ROLE_REGS_ACCESSOR(cr0, wp, X86_CR0_WP); | |
208 | BUILD_MMU_ROLE_REGS_ACCESSOR(cr4, pse, X86_CR4_PSE); | |
209 | BUILD_MMU_ROLE_REGS_ACCESSOR(cr4, pae, X86_CR4_PAE); | |
210 | BUILD_MMU_ROLE_REGS_ACCESSOR(cr4, smep, X86_CR4_SMEP); | |
211 | BUILD_MMU_ROLE_REGS_ACCESSOR(cr4, smap, X86_CR4_SMAP); | |
212 | BUILD_MMU_ROLE_REGS_ACCESSOR(cr4, pke, X86_CR4_PKE); | |
213 | BUILD_MMU_ROLE_REGS_ACCESSOR(cr4, la57, X86_CR4_LA57); | |
214 | BUILD_MMU_ROLE_REGS_ACCESSOR(efer, nx, EFER_NX); | |
215 | BUILD_MMU_ROLE_REGS_ACCESSOR(efer, lma, EFER_LMA); | |
216 | ||
60667724 SC |
217 | /* |
218 | * The MMU itself (with a valid role) is the single source of truth for the | |
219 | * MMU. Do not use the regs used to build the MMU/role, nor the vCPU. The | |
220 | * regs don't account for dependencies, e.g. clearing CR4 bits if CR0.PG=1, | |
221 | * and the vCPU may be incorrect/irrelevant. | |
222 | */ | |
223 | #define BUILD_MMU_ROLE_ACCESSOR(base_or_ext, reg, name) \ | |
4ac21457 | 224 | static inline bool __maybe_unused is_##reg##_##name(struct kvm_mmu *mmu) \ |
60667724 | 225 | { \ |
e5ed0fb0 | 226 | return !!(mmu->cpu_role. base_or_ext . reg##_##name); \ |
60667724 | 227 | } |
60667724 SC |
228 | BUILD_MMU_ROLE_ACCESSOR(base, cr0, wp); |
229 | BUILD_MMU_ROLE_ACCESSOR(ext, cr4, pse); | |
60667724 SC |
230 | BUILD_MMU_ROLE_ACCESSOR(ext, cr4, smep); |
231 | BUILD_MMU_ROLE_ACCESSOR(ext, cr4, smap); | |
232 | BUILD_MMU_ROLE_ACCESSOR(ext, cr4, pke); | |
233 | BUILD_MMU_ROLE_ACCESSOR(ext, cr4, la57); | |
234 | BUILD_MMU_ROLE_ACCESSOR(base, efer, nx); | |
56b321f9 | 235 | BUILD_MMU_ROLE_ACCESSOR(ext, efer, lma); |
60667724 | 236 | |
faf72962 PB |
237 | static inline bool is_cr0_pg(struct kvm_mmu *mmu) |
238 | { | |
239 | return mmu->cpu_role.base.level > 0; | |
240 | } | |
241 | ||
242 | static inline bool is_cr4_pae(struct kvm_mmu *mmu) | |
243 | { | |
244 | return !mmu->cpu_role.base.has_4_byte_gpte; | |
245 | } | |
246 | ||
594e91a1 SC |
247 | static struct kvm_mmu_role_regs vcpu_to_role_regs(struct kvm_vcpu *vcpu) |
248 | { | |
249 | struct kvm_mmu_role_regs regs = { | |
250 | .cr0 = kvm_read_cr0_bits(vcpu, KVM_MMU_CR0_ROLE_BITS), | |
251 | .cr4 = kvm_read_cr4_bits(vcpu, KVM_MMU_CR4_ROLE_BITS), | |
252 | .efer = vcpu->arch.efer, | |
253 | }; | |
254 | ||
255 | return regs; | |
256 | } | |
40ef75a7 | 257 | |
2fdcc1b3 | 258 | static unsigned long get_guest_cr3(struct kvm_vcpu *vcpu) |
40ef75a7 | 259 | { |
2fdcc1b3 | 260 | return kvm_read_cr3(vcpu); |
40ef75a7 LT |
261 | } |
262 | ||
2fdcc1b3 PB |
263 | static inline unsigned long kvm_mmu_get_guest_pgd(struct kvm_vcpu *vcpu, |
264 | struct kvm_mmu *mmu) | |
40ef75a7 | 265 | { |
aefb2f2e | 266 | if (IS_ENABLED(CONFIG_MITIGATION_RETPOLINE) && mmu->get_guest_pgd == get_guest_cr3) |
2fdcc1b3 | 267 | return kvm_read_cr3(vcpu); |
40ef75a7 | 268 | |
2fdcc1b3 | 269 | return mmu->get_guest_pgd(vcpu); |
40ef75a7 LT |
270 | } |
271 | ||
8a1300ff | 272 | static inline bool kvm_available_flush_remote_tlbs_range(void) |
40ef75a7 | 273 | { |
8a1300ff | 274 | return kvm_x86_ops.flush_remote_tlbs_range; |
40ef75a7 | 275 | } |
40ef75a7 | 276 | |
d4788996 | 277 | int kvm_arch_flush_remote_tlbs_range(struct kvm *kvm, gfn_t gfn, u64 nr_pages) |
40ef75a7 | 278 | { |
d4788996 DM |
279 | if (!kvm_x86_ops.flush_remote_tlbs_range) |
280 | return -EOPNOTSUPP; | |
40ef75a7 | 281 | |
d4788996 | 282 | return static_call(kvm_x86_flush_remote_tlbs_range)(kvm, gfn, nr_pages); |
40ef75a7 LT |
283 | } |
284 | ||
1b2dc736 HW |
285 | static gfn_t kvm_mmu_page_get_gfn(struct kvm_mmu_page *sp, int index); |
286 | ||
287 | /* Flush the range of guest memory mapped by the given SPTE. */ | |
288 | static void kvm_flush_remote_tlbs_sptep(struct kvm *kvm, u64 *sptep) | |
289 | { | |
290 | struct kvm_mmu_page *sp = sptep_to_sp(sptep); | |
291 | gfn_t gfn = kvm_mmu_page_get_gfn(sp, spte_index(sptep)); | |
292 | ||
293 | kvm_flush_remote_tlbs_gfn(kvm, gfn, sp->role.level); | |
294 | } | |
295 | ||
8f79b064 BG |
296 | static void mark_mmio_spte(struct kvm_vcpu *vcpu, u64 *sptep, u64 gfn, |
297 | unsigned int access) | |
298 | { | |
c236d962 | 299 | u64 spte = make_mmio_spte(vcpu, gfn, access); |
8f79b064 | 300 | |
c236d962 SC |
301 | trace_mark_mmio_spte(sptep, gfn, spte); |
302 | mmu_spte_set(sptep, spte); | |
ce88decf XG |
303 | } |
304 | ||
ce88decf XG |
305 | static gfn_t get_mmio_spte_gfn(u64 spte) |
306 | { | |
daa07cbc | 307 | u64 gpa = spte & shadow_nonpresent_or_rsvd_lower_gfn_mask; |
28a1f3ac | 308 | |
8a967d65 | 309 | gpa |= (spte >> SHADOW_NONPRESENT_OR_RSVD_MASK_LEN) |
28a1f3ac JS |
310 | & shadow_nonpresent_or_rsvd_mask; |
311 | ||
312 | return gpa >> PAGE_SHIFT; | |
ce88decf XG |
313 | } |
314 | ||
315 | static unsigned get_mmio_spte_access(u64 spte) | |
316 | { | |
4af77151 | 317 | return spte & shadow_mmio_access_mask; |
ce88decf XG |
318 | } |
319 | ||
54bf36aa | 320 | static bool check_mmio_spte(struct kvm_vcpu *vcpu, u64 spte) |
f8f55942 | 321 | { |
cae7ed3c | 322 | u64 kvm_gen, spte_gen, gen; |
089504c0 | 323 | |
cae7ed3c SC |
324 | gen = kvm_vcpu_memslots(vcpu)->generation; |
325 | if (unlikely(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS)) | |
326 | return false; | |
089504c0 | 327 | |
cae7ed3c | 328 | kvm_gen = gen & MMIO_SPTE_GEN_MASK; |
089504c0 XG |
329 | spte_gen = get_mmio_spte_generation(spte); |
330 | ||
331 | trace_check_mmio_spte(spte, kvm_gen, spte_gen); | |
332 | return likely(kvm_gen == spte_gen); | |
f8f55942 XG |
333 | } |
334 | ||
6aa8b732 AK |
335 | static int is_cpuid_PSE36(void) |
336 | { | |
337 | return 1; | |
338 | } | |
339 | ||
603e0651 | 340 | #ifdef CONFIG_X86_64 |
d555c333 | 341 | static void __set_spte(u64 *sptep, u64 spte) |
e663ee64 | 342 | { |
b19ee2ff | 343 | WRITE_ONCE(*sptep, spte); |
e663ee64 AK |
344 | } |
345 | ||
603e0651 | 346 | static void __update_clear_spte_fast(u64 *sptep, u64 spte) |
a9221dd5 | 347 | { |
b19ee2ff | 348 | WRITE_ONCE(*sptep, spte); |
603e0651 XG |
349 | } |
350 | ||
351 | static u64 __update_clear_spte_slow(u64 *sptep, u64 spte) | |
352 | { | |
353 | return xchg(sptep, spte); | |
354 | } | |
c2a2ac2b XG |
355 | |
356 | static u64 __get_spte_lockless(u64 *sptep) | |
357 | { | |
6aa7de05 | 358 | return READ_ONCE(*sptep); |
c2a2ac2b | 359 | } |
a9221dd5 | 360 | #else |
603e0651 XG |
361 | union split_spte { |
362 | struct { | |
363 | u32 spte_low; | |
364 | u32 spte_high; | |
365 | }; | |
366 | u64 spte; | |
367 | }; | |
a9221dd5 | 368 | |
c2a2ac2b XG |
369 | static void count_spte_clear(u64 *sptep, u64 spte) |
370 | { | |
57354682 | 371 | struct kvm_mmu_page *sp = sptep_to_sp(sptep); |
c2a2ac2b XG |
372 | |
373 | if (is_shadow_present_pte(spte)) | |
374 | return; | |
375 | ||
376 | /* Ensure the spte is completely set before we increase the count */ | |
377 | smp_wmb(); | |
378 | sp->clear_spte_count++; | |
379 | } | |
380 | ||
603e0651 XG |
381 | static void __set_spte(u64 *sptep, u64 spte) |
382 | { | |
383 | union split_spte *ssptep, sspte; | |
a9221dd5 | 384 | |
603e0651 XG |
385 | ssptep = (union split_spte *)sptep; |
386 | sspte = (union split_spte)spte; | |
387 | ||
388 | ssptep->spte_high = sspte.spte_high; | |
389 | ||
390 | /* | |
391 | * If we map the spte from nonpresent to present, We should store | |
392 | * the high bits firstly, then set present bit, so cpu can not | |
393 | * fetch this spte while we are setting the spte. | |
394 | */ | |
395 | smp_wmb(); | |
396 | ||
b19ee2ff | 397 | WRITE_ONCE(ssptep->spte_low, sspte.spte_low); |
a9221dd5 AK |
398 | } |
399 | ||
603e0651 XG |
400 | static void __update_clear_spte_fast(u64 *sptep, u64 spte) |
401 | { | |
402 | union split_spte *ssptep, sspte; | |
403 | ||
404 | ssptep = (union split_spte *)sptep; | |
405 | sspte = (union split_spte)spte; | |
406 | ||
b19ee2ff | 407 | WRITE_ONCE(ssptep->spte_low, sspte.spte_low); |
603e0651 XG |
408 | |
409 | /* | |
410 | * If we map the spte from present to nonpresent, we should clear | |
411 | * present bit firstly to avoid vcpu fetch the old high bits. | |
412 | */ | |
413 | smp_wmb(); | |
414 | ||
415 | ssptep->spte_high = sspte.spte_high; | |
c2a2ac2b | 416 | count_spte_clear(sptep, spte); |
603e0651 XG |
417 | } |
418 | ||
419 | static u64 __update_clear_spte_slow(u64 *sptep, u64 spte) | |
420 | { | |
421 | union split_spte *ssptep, sspte, orig; | |
422 | ||
423 | ssptep = (union split_spte *)sptep; | |
424 | sspte = (union split_spte)spte; | |
425 | ||
426 | /* xchg acts as a barrier before the setting of the high bits */ | |
427 | orig.spte_low = xchg(&ssptep->spte_low, sspte.spte_low); | |
41bc3186 ZJ |
428 | orig.spte_high = ssptep->spte_high; |
429 | ssptep->spte_high = sspte.spte_high; | |
c2a2ac2b | 430 | count_spte_clear(sptep, spte); |
603e0651 XG |
431 | |
432 | return orig.spte; | |
433 | } | |
c2a2ac2b XG |
434 | |
435 | /* | |
436 | * The idea using the light way get the spte on x86_32 guest is from | |
39656e83 | 437 | * gup_get_pte (mm/gup.c). |
accaefe0 | 438 | * |
aed02fe3 | 439 | * An spte tlb flush may be pending, because kvm_set_pte_rmap |
accaefe0 XG |
440 | * coalesces them and we are running out of the MMU lock. Therefore |
441 | * we need to protect against in-progress updates of the spte. | |
442 | * | |
443 | * Reading the spte while an update is in progress may get the old value | |
444 | * for the high part of the spte. The race is fine for a present->non-present | |
445 | * change (because the high part of the spte is ignored for non-present spte), | |
446 | * but for a present->present change we must reread the spte. | |
447 | * | |
448 | * All such changes are done in two steps (present->non-present and | |
449 | * non-present->present), hence it is enough to count the number of | |
450 | * present->non-present updates: if it changed while reading the spte, | |
451 | * we might have hit the race. This is done using clear_spte_count. | |
c2a2ac2b XG |
452 | */ |
453 | static u64 __get_spte_lockless(u64 *sptep) | |
454 | { | |
57354682 | 455 | struct kvm_mmu_page *sp = sptep_to_sp(sptep); |
c2a2ac2b XG |
456 | union split_spte spte, *orig = (union split_spte *)sptep; |
457 | int count; | |
458 | ||
459 | retry: | |
460 | count = sp->clear_spte_count; | |
461 | smp_rmb(); | |
462 | ||
463 | spte.spte_low = orig->spte_low; | |
464 | smp_rmb(); | |
465 | ||
466 | spte.spte_high = orig->spte_high; | |
467 | smp_rmb(); | |
468 | ||
469 | if (unlikely(spte.spte_low != orig->spte_low || | |
470 | count != sp->clear_spte_count)) | |
471 | goto retry; | |
472 | ||
473 | return spte.spte; | |
474 | } | |
603e0651 XG |
475 | #endif |
476 | ||
1df9f2dc XG |
477 | /* Rules for using mmu_spte_set: |
478 | * Set the sptep from nonpresent to present. | |
479 | * Note: the sptep being assigned *must* be either not present | |
480 | * or in a state where the hardware will not attempt to update | |
481 | * the spte. | |
482 | */ | |
483 | static void mmu_spte_set(u64 *sptep, u64 new_spte) | |
484 | { | |
20ba462d | 485 | WARN_ON_ONCE(is_shadow_present_pte(*sptep)); |
1df9f2dc XG |
486 | __set_spte(sptep, new_spte); |
487 | } | |
488 | ||
f39a058d JS |
489 | /* |
490 | * Update the SPTE (excluding the PFN), but do not track changes in its | |
491 | * accessed/dirty status. | |
1df9f2dc | 492 | */ |
f39a058d | 493 | static u64 mmu_spte_update_no_track(u64 *sptep, u64 new_spte) |
b79b93f9 | 494 | { |
c7ba5b48 | 495 | u64 old_spte = *sptep; |
4132779b | 496 | |
20ba462d | 497 | WARN_ON_ONCE(!is_shadow_present_pte(new_spte)); |
115111ef | 498 | check_spte_writable_invariants(new_spte); |
b79b93f9 | 499 | |
6e7d0354 XG |
500 | if (!is_shadow_present_pte(old_spte)) { |
501 | mmu_spte_set(sptep, new_spte); | |
f39a058d | 502 | return old_spte; |
6e7d0354 | 503 | } |
4132779b | 504 | |
c7ba5b48 | 505 | if (!spte_has_volatile_bits(old_spte)) |
603e0651 | 506 | __update_clear_spte_fast(sptep, new_spte); |
4132779b | 507 | else |
603e0651 | 508 | old_spte = __update_clear_spte_slow(sptep, new_spte); |
4132779b | 509 | |
20ba462d | 510 | WARN_ON_ONCE(spte_to_pfn(old_spte) != spte_to_pfn(new_spte)); |
83ef6c81 | 511 | |
f39a058d JS |
512 | return old_spte; |
513 | } | |
514 | ||
515 | /* Rules for using mmu_spte_update: | |
516 | * Update the state bits, it means the mapped pfn is not changed. | |
517 | * | |
02844ac1 DM |
518 | * Whenever an MMU-writable SPTE is overwritten with a read-only SPTE, remote |
519 | * TLBs must be flushed. Otherwise rmap_write_protect will find a read-only | |
520 | * spte, even though the writable spte might be cached on a CPU's TLB. | |
f39a058d JS |
521 | * |
522 | * Returns true if the TLB needs to be flushed | |
523 | */ | |
524 | static bool mmu_spte_update(u64 *sptep, u64 new_spte) | |
525 | { | |
526 | bool flush = false; | |
527 | u64 old_spte = mmu_spte_update_no_track(sptep, new_spte); | |
528 | ||
529 | if (!is_shadow_present_pte(old_spte)) | |
530 | return false; | |
531 | ||
c7ba5b48 XG |
532 | /* |
533 | * For the spte updated out of mmu-lock is safe, since | |
6a6256f9 | 534 | * we always atomically update it, see the comments in |
c7ba5b48 XG |
535 | * spte_has_volatile_bits(). |
536 | */ | |
706c9c55 | 537 | if (is_mmu_writable_spte(old_spte) && |
7f31c959 | 538 | !is_writable_pte(new_spte)) |
83ef6c81 | 539 | flush = true; |
4132779b | 540 | |
7e71a59b | 541 | /* |
83ef6c81 | 542 | * Flush TLB when accessed/dirty states are changed in the page tables, |
7e71a59b KH |
543 | * to guarantee consistency between TLB and page tables. |
544 | */ | |
7e71a59b | 545 | |
83ef6c81 JS |
546 | if (is_accessed_spte(old_spte) && !is_accessed_spte(new_spte)) { |
547 | flush = true; | |
4132779b | 548 | kvm_set_pfn_accessed(spte_to_pfn(old_spte)); |
83ef6c81 JS |
549 | } |
550 | ||
551 | if (is_dirty_spte(old_spte) && !is_dirty_spte(new_spte)) { | |
552 | flush = true; | |
4132779b | 553 | kvm_set_pfn_dirty(spte_to_pfn(old_spte)); |
83ef6c81 | 554 | } |
6e7d0354 | 555 | |
83ef6c81 | 556 | return flush; |
b79b93f9 AK |
557 | } |
558 | ||
1df9f2dc XG |
559 | /* |
560 | * Rules for using mmu_spte_clear_track_bits: | |
561 | * It sets the sptep from present to nonpresent, and track the | |
562 | * state bits, it is used to clear the last level sptep. | |
7fa2a347 | 563 | * Returns the old PTE. |
1df9f2dc | 564 | */ |
35d539c3 | 565 | static u64 mmu_spte_clear_track_bits(struct kvm *kvm, u64 *sptep) |
1df9f2dc | 566 | { |
ba049e93 | 567 | kvm_pfn_t pfn; |
1df9f2dc | 568 | u64 old_spte = *sptep; |
71f51d2c | 569 | int level = sptep_to_sp(sptep)->role.level; |
b14b2690 | 570 | struct page *page; |
1df9f2dc | 571 | |
54eb3ef5 SC |
572 | if (!is_shadow_present_pte(old_spte) || |
573 | !spte_has_volatile_bits(old_spte)) | |
603e0651 | 574 | __update_clear_spte_fast(sptep, 0ull); |
1df9f2dc | 575 | else |
603e0651 | 576 | old_spte = __update_clear_spte_slow(sptep, 0ull); |
1df9f2dc | 577 | |
afd28fe1 | 578 | if (!is_shadow_present_pte(old_spte)) |
7fa2a347 | 579 | return old_spte; |
1df9f2dc | 580 | |
71f51d2c MZ |
581 | kvm_update_page_stats(kvm, level, -1); |
582 | ||
1df9f2dc | 583 | pfn = spte_to_pfn(old_spte); |
86fde74c XG |
584 | |
585 | /* | |
b14b2690 SC |
586 | * KVM doesn't hold a reference to any pages mapped into the guest, and |
587 | * instead uses the mmu_notifier to ensure that KVM unmaps any pages | |
588 | * before they are reclaimed. Sanity check that, if the pfn is backed | |
589 | * by a refcounted page, the refcount is elevated. | |
86fde74c | 590 | */ |
b14b2690 | 591 | page = kvm_pfn_to_refcounted_page(pfn); |
20ba462d | 592 | WARN_ON_ONCE(page && !page_count(page)); |
86fde74c | 593 | |
83ef6c81 | 594 | if (is_accessed_spte(old_spte)) |
1df9f2dc | 595 | kvm_set_pfn_accessed(pfn); |
83ef6c81 JS |
596 | |
597 | if (is_dirty_spte(old_spte)) | |
1df9f2dc | 598 | kvm_set_pfn_dirty(pfn); |
83ef6c81 | 599 | |
7fa2a347 | 600 | return old_spte; |
1df9f2dc XG |
601 | } |
602 | ||
603 | /* | |
604 | * Rules for using mmu_spte_clear_no_track: | |
605 | * Directly clear spte without caring the state bits of sptep, | |
606 | * it is used to set the upper level spte. | |
607 | */ | |
608 | static void mmu_spte_clear_no_track(u64 *sptep) | |
609 | { | |
603e0651 | 610 | __update_clear_spte_fast(sptep, 0ull); |
1df9f2dc XG |
611 | } |
612 | ||
c2a2ac2b XG |
613 | static u64 mmu_spte_get_lockless(u64 *sptep) |
614 | { | |
615 | return __get_spte_lockless(sptep); | |
616 | } | |
617 | ||
f160c7b7 JS |
618 | /* Returns the Accessed status of the PTE and resets it at the same time. */ |
619 | static bool mmu_spte_age(u64 *sptep) | |
620 | { | |
621 | u64 spte = mmu_spte_get_lockless(sptep); | |
622 | ||
623 | if (!is_accessed_spte(spte)) | |
624 | return false; | |
625 | ||
ac8d57e5 | 626 | if (spte_ad_enabled(spte)) { |
f160c7b7 JS |
627 | clear_bit((ffs(shadow_accessed_mask) - 1), |
628 | (unsigned long *)sptep); | |
629 | } else { | |
630 | /* | |
631 | * Capture the dirty status of the page, so that it doesn't get | |
632 | * lost when the SPTE is marked for access tracking. | |
633 | */ | |
634 | if (is_writable_pte(spte)) | |
635 | kvm_set_pfn_dirty(spte_to_pfn(spte)); | |
636 | ||
637 | spte = mark_spte_for_access_track(spte); | |
638 | mmu_spte_update_no_track(sptep, spte); | |
639 | } | |
640 | ||
641 | return true; | |
642 | } | |
643 | ||
78fdd2f0 SC |
644 | static inline bool is_tdp_mmu_active(struct kvm_vcpu *vcpu) |
645 | { | |
646 | return tdp_mmu_enabled && vcpu->arch.mmu->root_role.direct; | |
647 | } | |
648 | ||
c2a2ac2b XG |
649 | static void walk_shadow_page_lockless_begin(struct kvm_vcpu *vcpu) |
650 | { | |
78fdd2f0 | 651 | if (is_tdp_mmu_active(vcpu)) { |
c5c8c7c5 DM |
652 | kvm_tdp_mmu_walk_lockless_begin(); |
653 | } else { | |
654 | /* | |
655 | * Prevent page table teardown by making any free-er wait during | |
656 | * kvm_flush_remote_tlbs() IPI to all active vcpus. | |
657 | */ | |
658 | local_irq_disable(); | |
36ca7e0a | 659 | |
c5c8c7c5 DM |
660 | /* |
661 | * Make sure a following spte read is not reordered ahead of the write | |
662 | * to vcpu->mode. | |
663 | */ | |
664 | smp_store_mb(vcpu->mode, READING_SHADOW_PAGE_TABLES); | |
665 | } | |
c2a2ac2b XG |
666 | } |
667 | ||
668 | static void walk_shadow_page_lockless_end(struct kvm_vcpu *vcpu) | |
669 | { | |
78fdd2f0 | 670 | if (is_tdp_mmu_active(vcpu)) { |
c5c8c7c5 DM |
671 | kvm_tdp_mmu_walk_lockless_end(); |
672 | } else { | |
673 | /* | |
674 | * Make sure the write to vcpu->mode is not reordered in front of | |
675 | * reads to sptes. If it does, kvm_mmu_commit_zap_page() can see us | |
676 | * OUTSIDE_GUEST_MODE and proceed to free the shadow page table. | |
677 | */ | |
678 | smp_store_release(&vcpu->mode, OUTSIDE_GUEST_MODE); | |
679 | local_irq_enable(); | |
680 | } | |
c2a2ac2b XG |
681 | } |
682 | ||
378f5cd6 | 683 | static int mmu_topup_memory_caches(struct kvm_vcpu *vcpu, bool maybe_indirect) |
714b93da | 684 | { |
e2dec939 AK |
685 | int r; |
686 | ||
531281ad | 687 | /* 1 rmap, 1 parent PTE per level, and the prefetched rmaps. */ |
94ce87ef SC |
688 | r = kvm_mmu_topup_memory_cache(&vcpu->arch.mmu_pte_list_desc_cache, |
689 | 1 + PT64_ROOT_MAX_LEVEL + PTE_PREFETCH_NUM); | |
d3d25b04 | 690 | if (r) |
284aa868 | 691 | return r; |
94ce87ef SC |
692 | r = kvm_mmu_topup_memory_cache(&vcpu->arch.mmu_shadow_page_cache, |
693 | PT64_ROOT_MAX_LEVEL); | |
d3d25b04 | 694 | if (r) |
171a90d7 | 695 | return r; |
378f5cd6 | 696 | if (maybe_indirect) { |
6a97575d | 697 | r = kvm_mmu_topup_memory_cache(&vcpu->arch.mmu_shadowed_info_cache, |
94ce87ef | 698 | PT64_ROOT_MAX_LEVEL); |
378f5cd6 SC |
699 | if (r) |
700 | return r; | |
701 | } | |
94ce87ef SC |
702 | return kvm_mmu_topup_memory_cache(&vcpu->arch.mmu_page_header_cache, |
703 | PT64_ROOT_MAX_LEVEL); | |
714b93da AK |
704 | } |
705 | ||
706 | static void mmu_free_memory_caches(struct kvm_vcpu *vcpu) | |
707 | { | |
94ce87ef SC |
708 | kvm_mmu_free_memory_cache(&vcpu->arch.mmu_pte_list_desc_cache); |
709 | kvm_mmu_free_memory_cache(&vcpu->arch.mmu_shadow_page_cache); | |
6a97575d | 710 | kvm_mmu_free_memory_cache(&vcpu->arch.mmu_shadowed_info_cache); |
94ce87ef | 711 | kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_header_cache); |
714b93da AK |
712 | } |
713 | ||
53c07b18 | 714 | static void mmu_free_pte_list_desc(struct pte_list_desc *pte_list_desc) |
714b93da | 715 | { |
53c07b18 | 716 | kmem_cache_free(pte_list_desc_cache, pte_list_desc); |
714b93da AK |
717 | } |
718 | ||
6a97575d DM |
719 | static bool sp_has_gptes(struct kvm_mmu_page *sp); |
720 | ||
2032a93d LJ |
721 | static gfn_t kvm_mmu_page_get_gfn(struct kvm_mmu_page *sp, int index) |
722 | { | |
84e5ffd0 LJ |
723 | if (sp->role.passthrough) |
724 | return sp->gfn; | |
725 | ||
2032a93d | 726 | if (!sp->role.direct) |
6a97575d | 727 | return sp->shadowed_translation[index] >> PAGE_SHIFT; |
2032a93d | 728 | |
2ca3129e | 729 | return sp->gfn + (index << ((sp->role.level - 1) * SPTE_LEVEL_BITS)); |
2032a93d LJ |
730 | } |
731 | ||
6a97575d DM |
732 | /* |
733 | * For leaf SPTEs, fetch the *guest* access permissions being shadowed. Note | |
734 | * that the SPTE itself may have a more constrained access permissions that | |
735 | * what the guest enforces. For example, a guest may create an executable | |
736 | * huge PTE but KVM may disallow execution to mitigate iTLB multihit. | |
737 | */ | |
738 | static u32 kvm_mmu_page_get_access(struct kvm_mmu_page *sp, int index) | |
2032a93d | 739 | { |
6a97575d DM |
740 | if (sp_has_gptes(sp)) |
741 | return sp->shadowed_translation[index] & ACC_ALL; | |
84e5ffd0 | 742 | |
6a97575d DM |
743 | /* |
744 | * For direct MMUs (e.g. TDP or non-paging guests) or passthrough SPs, | |
745 | * KVM is not shadowing any guest page tables, so the "guest access | |
746 | * permissions" are just ACC_ALL. | |
747 | * | |
748 | * For direct SPs in indirect MMUs (shadow paging), i.e. when KVM | |
749 | * is shadowing a guest huge page with small pages, the guest access | |
750 | * permissions being shadowed are the access permissions of the huge | |
751 | * page. | |
752 | * | |
753 | * In both cases, sp->role.access contains the correct access bits. | |
754 | */ | |
755 | return sp->role.access; | |
756 | } | |
757 | ||
72ae5822 SC |
758 | static void kvm_mmu_page_set_translation(struct kvm_mmu_page *sp, int index, |
759 | gfn_t gfn, unsigned int access) | |
6a97575d DM |
760 | { |
761 | if (sp_has_gptes(sp)) { | |
762 | sp->shadowed_translation[index] = (gfn << PAGE_SHIFT) | access; | |
e9f2a760 PB |
763 | return; |
764 | } | |
765 | ||
6a97575d DM |
766 | WARN_ONCE(access != kvm_mmu_page_get_access(sp, index), |
767 | "access mismatch under %s page %llx (expected %u, got %u)\n", | |
768 | sp->role.passthrough ? "passthrough" : "direct", | |
769 | sp->gfn, kvm_mmu_page_get_access(sp, index), access); | |
770 | ||
771 | WARN_ONCE(gfn != kvm_mmu_page_get_gfn(sp, index), | |
772 | "gfn mismatch under %s page %llx (expected %llx, got %llx)\n", | |
773 | sp->role.passthrough ? "passthrough" : "direct", | |
774 | sp->gfn, kvm_mmu_page_get_gfn(sp, index), gfn); | |
775 | } | |
776 | ||
72ae5822 SC |
777 | static void kvm_mmu_page_set_access(struct kvm_mmu_page *sp, int index, |
778 | unsigned int access) | |
6a97575d DM |
779 | { |
780 | gfn_t gfn = kvm_mmu_page_get_gfn(sp, index); | |
781 | ||
782 | kvm_mmu_page_set_translation(sp, index, gfn, access); | |
2032a93d LJ |
783 | } |
784 | ||
05da4558 | 785 | /* |
d4dbf470 TY |
786 | * Return the pointer to the large page information for a given gfn, |
787 | * handling slots that are not large page aligned. | |
05da4558 | 788 | */ |
d4dbf470 | 789 | static struct kvm_lpage_info *lpage_info_slot(gfn_t gfn, |
8ca6f063 | 790 | const struct kvm_memory_slot *slot, int level) |
05da4558 MT |
791 | { |
792 | unsigned long idx; | |
793 | ||
fb03cb6f | 794 | idx = gfn_to_index(gfn, slot->base_gfn, level); |
db3fe4eb | 795 | return &slot->arch.lpage_info[level - 2][idx]; |
05da4558 MT |
796 | } |
797 | ||
269e9552 | 798 | static void update_gfn_disallow_lpage_count(const struct kvm_memory_slot *slot, |
547ffaed XG |
799 | gfn_t gfn, int count) |
800 | { | |
801 | struct kvm_lpage_info *linfo; | |
802 | int i; | |
803 | ||
3bae0459 | 804 | for (i = PG_LEVEL_2M; i <= KVM_MAX_HUGEPAGE_LEVEL; ++i) { |
547ffaed XG |
805 | linfo = lpage_info_slot(gfn, slot, i); |
806 | linfo->disallow_lpage += count; | |
20ba462d | 807 | WARN_ON_ONCE(linfo->disallow_lpage < 0); |
547ffaed XG |
808 | } |
809 | } | |
810 | ||
269e9552 | 811 | void kvm_mmu_gfn_disallow_lpage(const struct kvm_memory_slot *slot, gfn_t gfn) |
547ffaed XG |
812 | { |
813 | update_gfn_disallow_lpage_count(slot, gfn, 1); | |
814 | } | |
815 | ||
269e9552 | 816 | void kvm_mmu_gfn_allow_lpage(const struct kvm_memory_slot *slot, gfn_t gfn) |
547ffaed XG |
817 | { |
818 | update_gfn_disallow_lpage_count(slot, gfn, -1); | |
819 | } | |
820 | ||
3ed1a478 | 821 | static void account_shadowed(struct kvm *kvm, struct kvm_mmu_page *sp) |
05da4558 | 822 | { |
699023e2 | 823 | struct kvm_memslots *slots; |
d25797b2 | 824 | struct kvm_memory_slot *slot; |
3ed1a478 | 825 | gfn_t gfn; |
05da4558 | 826 | |
56ca57f9 | 827 | kvm->arch.indirect_shadow_pages++; |
3ed1a478 | 828 | gfn = sp->gfn; |
699023e2 PB |
829 | slots = kvm_memslots_for_spte_role(kvm, sp->role); |
830 | slot = __gfn_to_memslot(slots, gfn); | |
56ca57f9 XG |
831 | |
832 | /* the non-leaf shadow pages are keeping readonly. */ | |
3bae0459 | 833 | if (sp->role.level > PG_LEVEL_4K) |
96316a06 | 834 | return __kvm_write_track_add_gfn(kvm, slot, gfn); |
56ca57f9 | 835 | |
547ffaed | 836 | kvm_mmu_gfn_disallow_lpage(slot, gfn); |
be911771 DM |
837 | |
838 | if (kvm_mmu_slot_gfn_write_protect(kvm, slot, gfn, PG_LEVEL_4K)) | |
4ad980ae | 839 | kvm_flush_remote_tlbs_gfn(kvm, gfn, PG_LEVEL_4K); |
05da4558 MT |
840 | } |
841 | ||
61f94478 | 842 | void track_possible_nx_huge_page(struct kvm *kvm, struct kvm_mmu_page *sp) |
b8e8c830 | 843 | { |
428e9216 SC |
844 | /* |
845 | * If it's possible to replace the shadow page with an NX huge page, | |
846 | * i.e. if the shadow page is the only thing currently preventing KVM | |
847 | * from using a huge page, add the shadow page to the list of "to be | |
848 | * zapped for NX recovery" pages. Note, the shadow page can already be | |
849 | * on the list if KVM is reusing an existing shadow page, i.e. if KVM | |
850 | * links a shadow page at multiple points. | |
851 | */ | |
61f94478 | 852 | if (!list_empty(&sp->possible_nx_huge_page_link)) |
b8e8c830 PB |
853 | return; |
854 | ||
855 | ++kvm->stat.nx_lpage_splits; | |
55c510e2 SC |
856 | list_add_tail(&sp->possible_nx_huge_page_link, |
857 | &kvm->arch.possible_nx_huge_pages); | |
b8e8c830 PB |
858 | } |
859 | ||
61f94478 SC |
860 | static void account_nx_huge_page(struct kvm *kvm, struct kvm_mmu_page *sp, |
861 | bool nx_huge_page_possible) | |
862 | { | |
863 | sp->nx_huge_page_disallowed = true; | |
864 | ||
865 | if (nx_huge_page_possible) | |
866 | track_possible_nx_huge_page(kvm, sp); | |
b8e8c830 PB |
867 | } |
868 | ||
3ed1a478 | 869 | static void unaccount_shadowed(struct kvm *kvm, struct kvm_mmu_page *sp) |
05da4558 | 870 | { |
699023e2 | 871 | struct kvm_memslots *slots; |
d25797b2 | 872 | struct kvm_memory_slot *slot; |
3ed1a478 | 873 | gfn_t gfn; |
05da4558 | 874 | |
56ca57f9 | 875 | kvm->arch.indirect_shadow_pages--; |
3ed1a478 | 876 | gfn = sp->gfn; |
699023e2 PB |
877 | slots = kvm_memslots_for_spte_role(kvm, sp->role); |
878 | slot = __gfn_to_memslot(slots, gfn); | |
3bae0459 | 879 | if (sp->role.level > PG_LEVEL_4K) |
96316a06 | 880 | return __kvm_write_track_remove_gfn(kvm, slot, gfn); |
56ca57f9 | 881 | |
547ffaed | 882 | kvm_mmu_gfn_allow_lpage(slot, gfn); |
05da4558 MT |
883 | } |
884 | ||
61f94478 | 885 | void untrack_possible_nx_huge_page(struct kvm *kvm, struct kvm_mmu_page *sp) |
b8e8c830 | 886 | { |
55c510e2 | 887 | if (list_empty(&sp->possible_nx_huge_page_link)) |
428e9216 SC |
888 | return; |
889 | ||
b8e8c830 | 890 | --kvm->stat.nx_lpage_splits; |
55c510e2 | 891 | list_del_init(&sp->possible_nx_huge_page_link); |
b8e8c830 PB |
892 | } |
893 | ||
61f94478 SC |
894 | static void unaccount_nx_huge_page(struct kvm *kvm, struct kvm_mmu_page *sp) |
895 | { | |
896 | sp->nx_huge_page_disallowed = false; | |
897 | ||
898 | untrack_possible_nx_huge_page(kvm, sp); | |
b8e8c830 PB |
899 | } |
900 | ||
f3d90f90 SC |
901 | static struct kvm_memory_slot *gfn_to_memslot_dirty_bitmap(struct kvm_vcpu *vcpu, |
902 | gfn_t gfn, | |
903 | bool no_dirty_log) | |
05da4558 MT |
904 | { |
905 | struct kvm_memory_slot *slot; | |
5d163b1c | 906 | |
54bf36aa | 907 | slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); |
91b0d268 PB |
908 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
909 | return NULL; | |
044c59c4 | 910 | if (no_dirty_log && kvm_slot_dirty_track_enabled(slot)) |
91b0d268 | 911 | return NULL; |
5d163b1c XG |
912 | |
913 | return slot; | |
914 | } | |
915 | ||
290fc38d | 916 | /* |
018aabb5 | 917 | * About rmap_head encoding: |
cd4a4e53 | 918 | * |
018aabb5 TY |
919 | * If the bit zero of rmap_head->val is clear, then it points to the only spte |
920 | * in this rmap chain. Otherwise, (rmap_head->val & ~1) points to a struct | |
53c07b18 | 921 | * pte_list_desc containing more mappings. |
018aabb5 TY |
922 | */ |
923 | ||
924 | /* | |
925 | * Returns the number of pointers in the rmap chain, not counting the new one. | |
cd4a4e53 | 926 | */ |
2ff9039a | 927 | static int pte_list_add(struct kvm_mmu_memory_cache *cache, u64 *spte, |
018aabb5 | 928 | struct kvm_rmap_head *rmap_head) |
cd4a4e53 | 929 | { |
53c07b18 | 930 | struct pte_list_desc *desc; |
13236e25 | 931 | int count = 0; |
cd4a4e53 | 932 | |
018aabb5 | 933 | if (!rmap_head->val) { |
018aabb5 TY |
934 | rmap_head->val = (unsigned long)spte; |
935 | } else if (!(rmap_head->val & 1)) { | |
2ff9039a | 936 | desc = kvm_mmu_memory_cache_alloc(cache); |
018aabb5 | 937 | desc->sptes[0] = (u64 *)rmap_head->val; |
d555c333 | 938 | desc->sptes[1] = spte; |
13236e25 | 939 | desc->spte_count = 2; |
141705b7 | 940 | desc->tail_count = 0; |
018aabb5 | 941 | rmap_head->val = (unsigned long)desc | 1; |
cb16a7b3 | 942 | ++count; |
cd4a4e53 | 943 | } else { |
018aabb5 | 944 | desc = (struct pte_list_desc *)(rmap_head->val & ~1ul); |
141705b7 LJ |
945 | count = desc->tail_count + desc->spte_count; |
946 | ||
947 | /* | |
948 | * If the previous head is full, allocate a new head descriptor | |
949 | * as tail descriptors are always kept full. | |
950 | */ | |
951 | if (desc->spte_count == PTE_LIST_EXT) { | |
952 | desc = kvm_mmu_memory_cache_alloc(cache); | |
953 | desc->more = (struct pte_list_desc *)(rmap_head->val & ~1ul); | |
954 | desc->spte_count = 0; | |
955 | desc->tail_count = count; | |
956 | rmap_head->val = (unsigned long)desc | 1; | |
cd4a4e53 | 957 | } |
13236e25 | 958 | desc->sptes[desc->spte_count++] = spte; |
cd4a4e53 | 959 | } |
53a27b39 | 960 | return count; |
cd4a4e53 AK |
961 | } |
962 | ||
069f30c6 MZ |
963 | static void pte_list_desc_remove_entry(struct kvm *kvm, |
964 | struct kvm_rmap_head *rmap_head, | |
f3d90f90 | 965 | struct pte_list_desc *desc, int i) |
cd4a4e53 | 966 | { |
141705b7 LJ |
967 | struct pte_list_desc *head_desc = (struct pte_list_desc *)(rmap_head->val & ~1ul); |
968 | int j = head_desc->spte_count - 1; | |
cd4a4e53 | 969 | |
141705b7 LJ |
970 | /* |
971 | * The head descriptor should never be empty. A new head is added only | |
972 | * when adding an entry and the previous head is full, and heads are | |
973 | * removed (this flow) when they become empty. | |
974 | */ | |
52e322ed | 975 | KVM_BUG_ON_DATA_CORRUPTION(j < 0, kvm); |
141705b7 LJ |
976 | |
977 | /* | |
978 | * Replace the to-be-freed SPTE with the last valid entry from the head | |
979 | * descriptor to ensure that tail descriptors are full at all times. | |
980 | * Note, this also means that tail_count is stable for each descriptor. | |
981 | */ | |
982 | desc->sptes[i] = head_desc->sptes[j]; | |
983 | head_desc->sptes[j] = NULL; | |
984 | head_desc->spte_count--; | |
985 | if (head_desc->spte_count) | |
cd4a4e53 | 986 | return; |
141705b7 LJ |
987 | |
988 | /* | |
989 | * The head descriptor is empty. If there are no tail descriptors, | |
54aa699e | 990 | * nullify the rmap head to mark the list as empty, else point the rmap |
141705b7 LJ |
991 | * head at the next descriptor, i.e. the new head. |
992 | */ | |
993 | if (!head_desc->more) | |
fe3c2b4c | 994 | rmap_head->val = 0; |
cd4a4e53 | 995 | else |
141705b7 LJ |
996 | rmap_head->val = (unsigned long)head_desc->more | 1; |
997 | mmu_free_pte_list_desc(head_desc); | |
cd4a4e53 AK |
998 | } |
999 | ||
069f30c6 MZ |
1000 | static void pte_list_remove(struct kvm *kvm, u64 *spte, |
1001 | struct kvm_rmap_head *rmap_head) | |
cd4a4e53 | 1002 | { |
53c07b18 | 1003 | struct pte_list_desc *desc; |
cd4a4e53 AK |
1004 | int i; |
1005 | ||
52e322ed SC |
1006 | if (KVM_BUG_ON_DATA_CORRUPTION(!rmap_head->val, kvm)) |
1007 | return; | |
1008 | ||
1009 | if (!(rmap_head->val & 1)) { | |
1010 | if (KVM_BUG_ON_DATA_CORRUPTION((u64 *)rmap_head->val != spte, kvm)) | |
1011 | return; | |
1012 | ||
018aabb5 | 1013 | rmap_head->val = 0; |
cd4a4e53 | 1014 | } else { |
018aabb5 | 1015 | desc = (struct pte_list_desc *)(rmap_head->val & ~1ul); |
cd4a4e53 | 1016 | while (desc) { |
13236e25 | 1017 | for (i = 0; i < desc->spte_count; ++i) { |
d555c333 | 1018 | if (desc->sptes[i] == spte) { |
069f30c6 MZ |
1019 | pte_list_desc_remove_entry(kvm, rmap_head, |
1020 | desc, i); | |
cd4a4e53 AK |
1021 | return; |
1022 | } | |
018aabb5 | 1023 | } |
cd4a4e53 AK |
1024 | desc = desc->more; |
1025 | } | |
52e322ed SC |
1026 | |
1027 | KVM_BUG_ON_DATA_CORRUPTION(true, kvm); | |
cd4a4e53 AK |
1028 | } |
1029 | } | |
1030 | ||
9202aee8 SC |
1031 | static void kvm_zap_one_rmap_spte(struct kvm *kvm, |
1032 | struct kvm_rmap_head *rmap_head, u64 *sptep) | |
e7912386 | 1033 | { |
71f51d2c | 1034 | mmu_spte_clear_track_bits(kvm, sptep); |
069f30c6 | 1035 | pte_list_remove(kvm, sptep, rmap_head); |
e7912386 WY |
1036 | } |
1037 | ||
9202aee8 SC |
1038 | /* Return true if at least one SPTE was zapped, false otherwise */ |
1039 | static bool kvm_zap_all_rmap_sptes(struct kvm *kvm, | |
1040 | struct kvm_rmap_head *rmap_head) | |
a75b5404 PX |
1041 | { |
1042 | struct pte_list_desc *desc, *next; | |
1043 | int i; | |
1044 | ||
1045 | if (!rmap_head->val) | |
1046 | return false; | |
1047 | ||
1048 | if (!(rmap_head->val & 1)) { | |
71f51d2c | 1049 | mmu_spte_clear_track_bits(kvm, (u64 *)rmap_head->val); |
a75b5404 PX |
1050 | goto out; |
1051 | } | |
1052 | ||
1053 | desc = (struct pte_list_desc *)(rmap_head->val & ~1ul); | |
1054 | ||
1055 | for (; desc; desc = next) { | |
1056 | for (i = 0; i < desc->spte_count; i++) | |
71f51d2c | 1057 | mmu_spte_clear_track_bits(kvm, desc->sptes[i]); |
a75b5404 PX |
1058 | next = desc->more; |
1059 | mmu_free_pte_list_desc(desc); | |
1060 | } | |
1061 | out: | |
1062 | /* rmap_head is meaningless now, remember to reset it */ | |
1063 | rmap_head->val = 0; | |
1064 | return true; | |
1065 | } | |
1066 | ||
3bcd0662 PX |
1067 | unsigned int pte_list_count(struct kvm_rmap_head *rmap_head) |
1068 | { | |
1069 | struct pte_list_desc *desc; | |
3bcd0662 PX |
1070 | |
1071 | if (!rmap_head->val) | |
1072 | return 0; | |
1073 | else if (!(rmap_head->val & 1)) | |
1074 | return 1; | |
1075 | ||
1076 | desc = (struct pte_list_desc *)(rmap_head->val & ~1ul); | |
141705b7 | 1077 | return desc->tail_count + desc->spte_count; |
3bcd0662 PX |
1078 | } |
1079 | ||
93e083d4 DM |
1080 | static struct kvm_rmap_head *gfn_to_rmap(gfn_t gfn, int level, |
1081 | const struct kvm_memory_slot *slot) | |
53c07b18 | 1082 | { |
77d11309 | 1083 | unsigned long idx; |
53c07b18 | 1084 | |
77d11309 | 1085 | idx = gfn_to_index(gfn, slot->base_gfn, level); |
3bae0459 | 1086 | return &slot->arch.rmap[level - PG_LEVEL_4K][idx]; |
53c07b18 XG |
1087 | } |
1088 | ||
53c07b18 XG |
1089 | static void rmap_remove(struct kvm *kvm, u64 *spte) |
1090 | { | |
601f8af0 DM |
1091 | struct kvm_memslots *slots; |
1092 | struct kvm_memory_slot *slot; | |
53c07b18 XG |
1093 | struct kvm_mmu_page *sp; |
1094 | gfn_t gfn; | |
018aabb5 | 1095 | struct kvm_rmap_head *rmap_head; |
53c07b18 | 1096 | |
57354682 | 1097 | sp = sptep_to_sp(spte); |
79e48cec | 1098 | gfn = kvm_mmu_page_get_gfn(sp, spte_index(spte)); |
601f8af0 DM |
1099 | |
1100 | /* | |
68be1306 DM |
1101 | * Unlike rmap_add, rmap_remove does not run in the context of a vCPU |
1102 | * so we have to determine which memslots to use based on context | |
1103 | * information in sp->role. | |
601f8af0 DM |
1104 | */ |
1105 | slots = kvm_memslots_for_spte_role(kvm, sp->role); | |
1106 | ||
1107 | slot = __gfn_to_memslot(slots, gfn); | |
93e083d4 | 1108 | rmap_head = gfn_to_rmap(gfn, sp->role.level, slot); |
601f8af0 | 1109 | |
069f30c6 | 1110 | pte_list_remove(kvm, spte, rmap_head); |
53c07b18 XG |
1111 | } |
1112 | ||
1e3f42f0 TY |
1113 | /* |
1114 | * Used by the following functions to iterate through the sptes linked by a | |
1115 | * rmap. All fields are private and not assumed to be used outside. | |
1116 | */ | |
1117 | struct rmap_iterator { | |
1118 | /* private fields */ | |
1119 | struct pte_list_desc *desc; /* holds the sptep if not NULL */ | |
1120 | int pos; /* index of the sptep */ | |
1121 | }; | |
1122 | ||
1123 | /* | |
1124 | * Iteration must be started by this function. This should also be used after | |
1125 | * removing/dropping sptes from the rmap link because in such cases the | |
0a03cbda | 1126 | * information in the iterator may not be valid. |
1e3f42f0 TY |
1127 | * |
1128 | * Returns sptep if found, NULL otherwise. | |
1129 | */ | |
018aabb5 TY |
1130 | static u64 *rmap_get_first(struct kvm_rmap_head *rmap_head, |
1131 | struct rmap_iterator *iter) | |
1e3f42f0 | 1132 | { |
77fbbbd2 TY |
1133 | u64 *sptep; |
1134 | ||
018aabb5 | 1135 | if (!rmap_head->val) |
1e3f42f0 TY |
1136 | return NULL; |
1137 | ||
018aabb5 | 1138 | if (!(rmap_head->val & 1)) { |
1e3f42f0 | 1139 | iter->desc = NULL; |
77fbbbd2 TY |
1140 | sptep = (u64 *)rmap_head->val; |
1141 | goto out; | |
1e3f42f0 TY |
1142 | } |
1143 | ||
018aabb5 | 1144 | iter->desc = (struct pte_list_desc *)(rmap_head->val & ~1ul); |
1e3f42f0 | 1145 | iter->pos = 0; |
77fbbbd2 TY |
1146 | sptep = iter->desc->sptes[iter->pos]; |
1147 | out: | |
1148 | BUG_ON(!is_shadow_present_pte(*sptep)); | |
1149 | return sptep; | |
1e3f42f0 TY |
1150 | } |
1151 | ||
1152 | /* | |
1153 | * Must be used with a valid iterator: e.g. after rmap_get_first(). | |
1154 | * | |
1155 | * Returns sptep if found, NULL otherwise. | |
1156 | */ | |
1157 | static u64 *rmap_get_next(struct rmap_iterator *iter) | |
1158 | { | |
77fbbbd2 TY |
1159 | u64 *sptep; |
1160 | ||
1e3f42f0 TY |
1161 | if (iter->desc) { |
1162 | if (iter->pos < PTE_LIST_EXT - 1) { | |
1e3f42f0 TY |
1163 | ++iter->pos; |
1164 | sptep = iter->desc->sptes[iter->pos]; | |
1165 | if (sptep) | |
77fbbbd2 | 1166 | goto out; |
1e3f42f0 TY |
1167 | } |
1168 | ||
1169 | iter->desc = iter->desc->more; | |
1170 | ||
1171 | if (iter->desc) { | |
1172 | iter->pos = 0; | |
1173 | /* desc->sptes[0] cannot be NULL */ | |
77fbbbd2 TY |
1174 | sptep = iter->desc->sptes[iter->pos]; |
1175 | goto out; | |
1e3f42f0 TY |
1176 | } |
1177 | } | |
1178 | ||
1179 | return NULL; | |
77fbbbd2 TY |
1180 | out: |
1181 | BUG_ON(!is_shadow_present_pte(*sptep)); | |
1182 | return sptep; | |
1e3f42f0 TY |
1183 | } |
1184 | ||
018aabb5 TY |
1185 | #define for_each_rmap_spte(_rmap_head_, _iter_, _spte_) \ |
1186 | for (_spte_ = rmap_get_first(_rmap_head_, _iter_); \ | |
77fbbbd2 | 1187 | _spte_; _spte_ = rmap_get_next(_iter_)) |
0d536790 | 1188 | |
c3707958 | 1189 | static void drop_spte(struct kvm *kvm, u64 *sptep) |
e4b502ea | 1190 | { |
71f51d2c | 1191 | u64 old_spte = mmu_spte_clear_track_bits(kvm, sptep); |
7fa2a347 SC |
1192 | |
1193 | if (is_shadow_present_pte(old_spte)) | |
eb45fda4 | 1194 | rmap_remove(kvm, sptep); |
be38d276 AK |
1195 | } |
1196 | ||
03787394 | 1197 | static void drop_large_spte(struct kvm *kvm, u64 *sptep, bool flush) |
8e22f955 | 1198 | { |
0cd8dc73 | 1199 | struct kvm_mmu_page *sp; |
8e22f955 | 1200 | |
0cd8dc73 | 1201 | sp = sptep_to_sp(sptep); |
20ba462d | 1202 | WARN_ON_ONCE(sp->role.level == PG_LEVEL_4K); |
c3134ce2 | 1203 | |
0cd8dc73 | 1204 | drop_spte(kvm, sptep); |
03787394 PB |
1205 | |
1206 | if (flush) | |
1b2dc736 | 1207 | kvm_flush_remote_tlbs_sptep(kvm, sptep); |
8e22f955 XG |
1208 | } |
1209 | ||
1210 | /* | |
49fde340 | 1211 | * Write-protect on the specified @sptep, @pt_protect indicates whether |
c126d94f | 1212 | * spte write-protection is caused by protecting shadow page table. |
49fde340 | 1213 | * |
b4619660 | 1214 | * Note: write protection is difference between dirty logging and spte |
49fde340 XG |
1215 | * protection: |
1216 | * - for dirty logging, the spte can be set to writable at anytime if | |
1217 | * its dirty bitmap is properly set. | |
1218 | * - for spte protection, the spte can be writable only after unsync-ing | |
1219 | * shadow page. | |
8e22f955 | 1220 | * |
c126d94f | 1221 | * Return true if tlb need be flushed. |
8e22f955 | 1222 | */ |
c4f138b4 | 1223 | static bool spte_write_protect(u64 *sptep, bool pt_protect) |
d13bc5b5 XG |
1224 | { |
1225 | u64 spte = *sptep; | |
1226 | ||
49fde340 | 1227 | if (!is_writable_pte(spte) && |
706c9c55 | 1228 | !(pt_protect && is_mmu_writable_spte(spte))) |
d13bc5b5 XG |
1229 | return false; |
1230 | ||
49fde340 | 1231 | if (pt_protect) |
5fc3424f | 1232 | spte &= ~shadow_mmu_writable_mask; |
d13bc5b5 | 1233 | spte = spte & ~PT_WRITABLE_MASK; |
49fde340 | 1234 | |
c126d94f | 1235 | return mmu_spte_update(sptep, spte); |
d13bc5b5 XG |
1236 | } |
1237 | ||
1346bbb6 DM |
1238 | static bool rmap_write_protect(struct kvm_rmap_head *rmap_head, |
1239 | bool pt_protect) | |
98348e95 | 1240 | { |
1e3f42f0 TY |
1241 | u64 *sptep; |
1242 | struct rmap_iterator iter; | |
d13bc5b5 | 1243 | bool flush = false; |
374cbac0 | 1244 | |
018aabb5 | 1245 | for_each_rmap_spte(rmap_head, &iter, sptep) |
c4f138b4 | 1246 | flush |= spte_write_protect(sptep, pt_protect); |
855149aa | 1247 | |
d13bc5b5 | 1248 | return flush; |
a0ed4607 TY |
1249 | } |
1250 | ||
c4f138b4 | 1251 | static bool spte_clear_dirty(u64 *sptep) |
f4b4b180 KH |
1252 | { |
1253 | u64 spte = *sptep; | |
1254 | ||
0fe6370e | 1255 | KVM_MMU_WARN_ON(!spte_ad_enabled(spte)); |
f4b4b180 | 1256 | spte &= ~shadow_dirty_mask; |
f4b4b180 KH |
1257 | return mmu_spte_update(sptep, spte); |
1258 | } | |
1259 | ||
1f4e5fc8 | 1260 | static bool spte_wrprot_for_clear_dirty(u64 *sptep) |
ac8d57e5 PF |
1261 | { |
1262 | bool was_writable = test_and_clear_bit(PT_WRITABLE_SHIFT, | |
1263 | (unsigned long *)sptep); | |
1f4e5fc8 | 1264 | if (was_writable && !spte_ad_enabled(*sptep)) |
ac8d57e5 PF |
1265 | kvm_set_pfn_dirty(spte_to_pfn(*sptep)); |
1266 | ||
1267 | return was_writable; | |
1268 | } | |
1269 | ||
1270 | /* | |
1271 | * Gets the GFN ready for another round of dirty logging by clearing the | |
1272 | * - D bit on ad-enabled SPTEs, and | |
1273 | * - W bit on ad-disabled SPTEs. | |
1274 | * Returns true iff any D or W bits were cleared. | |
1275 | */ | |
0a234f5d | 1276 | static bool __rmap_clear_dirty(struct kvm *kvm, struct kvm_rmap_head *rmap_head, |
269e9552 | 1277 | const struct kvm_memory_slot *slot) |
f4b4b180 KH |
1278 | { |
1279 | u64 *sptep; | |
1280 | struct rmap_iterator iter; | |
1281 | bool flush = false; | |
1282 | ||
018aabb5 | 1283 | for_each_rmap_spte(rmap_head, &iter, sptep) |
1f4e5fc8 PB |
1284 | if (spte_ad_need_write_protect(*sptep)) |
1285 | flush |= spte_wrprot_for_clear_dirty(sptep); | |
ac8d57e5 | 1286 | else |
1f4e5fc8 | 1287 | flush |= spte_clear_dirty(sptep); |
f4b4b180 KH |
1288 | |
1289 | return flush; | |
1290 | } | |
1291 | ||
5dc99b23 | 1292 | /** |
3b0f1d01 | 1293 | * kvm_mmu_write_protect_pt_masked - write protect selected PT level pages |
5dc99b23 TY |
1294 | * @kvm: kvm instance |
1295 | * @slot: slot to protect | |
1296 | * @gfn_offset: start of the BITS_PER_LONG pages we care about | |
1297 | * @mask: indicates which pages we should protect | |
1298 | * | |
89212919 | 1299 | * Used when we do not need to care about huge page mappings. |
5dc99b23 | 1300 | */ |
3b0f1d01 | 1301 | static void kvm_mmu_write_protect_pt_masked(struct kvm *kvm, |
5dc99b23 TY |
1302 | struct kvm_memory_slot *slot, |
1303 | gfn_t gfn_offset, unsigned long mask) | |
a0ed4607 | 1304 | { |
018aabb5 | 1305 | struct kvm_rmap_head *rmap_head; |
a0ed4607 | 1306 | |
1f98f2bd | 1307 | if (tdp_mmu_enabled) |
a6a0b05d BG |
1308 | kvm_tdp_mmu_clear_dirty_pt_masked(kvm, slot, |
1309 | slot->base_gfn + gfn_offset, mask, true); | |
e2209710 BG |
1310 | |
1311 | if (!kvm_memslots_have_rmaps(kvm)) | |
1312 | return; | |
1313 | ||
5dc99b23 | 1314 | while (mask) { |
93e083d4 DM |
1315 | rmap_head = gfn_to_rmap(slot->base_gfn + gfn_offset + __ffs(mask), |
1316 | PG_LEVEL_4K, slot); | |
1346bbb6 | 1317 | rmap_write_protect(rmap_head, false); |
05da4558 | 1318 | |
5dc99b23 TY |
1319 | /* clear the first set bit */ |
1320 | mask &= mask - 1; | |
1321 | } | |
374cbac0 AK |
1322 | } |
1323 | ||
f4b4b180 | 1324 | /** |
ac8d57e5 PF |
1325 | * kvm_mmu_clear_dirty_pt_masked - clear MMU D-bit for PT level pages, or write |
1326 | * protect the page if the D-bit isn't supported. | |
f4b4b180 KH |
1327 | * @kvm: kvm instance |
1328 | * @slot: slot to clear D-bit | |
1329 | * @gfn_offset: start of the BITS_PER_LONG pages we care about | |
1330 | * @mask: indicates which pages we should clear D-bit | |
1331 | * | |
1332 | * Used for PML to re-log the dirty GPAs after userspace querying dirty_bitmap. | |
1333 | */ | |
a018eba5 SC |
1334 | static void kvm_mmu_clear_dirty_pt_masked(struct kvm *kvm, |
1335 | struct kvm_memory_slot *slot, | |
1336 | gfn_t gfn_offset, unsigned long mask) | |
f4b4b180 | 1337 | { |
018aabb5 | 1338 | struct kvm_rmap_head *rmap_head; |
f4b4b180 | 1339 | |
1f98f2bd | 1340 | if (tdp_mmu_enabled) |
a6a0b05d BG |
1341 | kvm_tdp_mmu_clear_dirty_pt_masked(kvm, slot, |
1342 | slot->base_gfn + gfn_offset, mask, false); | |
e2209710 BG |
1343 | |
1344 | if (!kvm_memslots_have_rmaps(kvm)) | |
1345 | return; | |
1346 | ||
f4b4b180 | 1347 | while (mask) { |
93e083d4 DM |
1348 | rmap_head = gfn_to_rmap(slot->base_gfn + gfn_offset + __ffs(mask), |
1349 | PG_LEVEL_4K, slot); | |
0a234f5d | 1350 | __rmap_clear_dirty(kvm, rmap_head, slot); |
f4b4b180 KH |
1351 | |
1352 | /* clear the first set bit */ | |
1353 | mask &= mask - 1; | |
1354 | } | |
1355 | } | |
f4b4b180 | 1356 | |
3b0f1d01 KH |
1357 | /** |
1358 | * kvm_arch_mmu_enable_log_dirty_pt_masked - enable dirty logging for selected | |
1359 | * PT level pages. | |
1360 | * | |
1361 | * It calls kvm_mmu_write_protect_pt_masked to write protect selected pages to | |
1362 | * enable dirty logging for them. | |
1363 | * | |
89212919 KZ |
1364 | * We need to care about huge page mappings: e.g. during dirty logging we may |
1365 | * have such mappings. | |
3b0f1d01 KH |
1366 | */ |
1367 | void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm, | |
1368 | struct kvm_memory_slot *slot, | |
1369 | gfn_t gfn_offset, unsigned long mask) | |
1370 | { | |
89212919 KZ |
1371 | /* |
1372 | * Huge pages are NOT write protected when we start dirty logging in | |
1373 | * initially-all-set mode; must write protect them here so that they | |
1374 | * are split to 4K on the first write. | |
1375 | * | |
1376 | * The gfn_offset is guaranteed to be aligned to 64, but the base_gfn | |
1377 | * of memslot has no such restriction, so the range can cross two large | |
1378 | * pages. | |
1379 | */ | |
1380 | if (kvm_dirty_log_manual_protect_and_init_set(kvm)) { | |
1381 | gfn_t start = slot->base_gfn + gfn_offset + __ffs(mask); | |
1382 | gfn_t end = slot->base_gfn + gfn_offset + __fls(mask); | |
1383 | ||
cb00a70b DM |
1384 | if (READ_ONCE(eager_page_split)) |
1385 | kvm_mmu_try_split_huge_pages(kvm, slot, start, end, PG_LEVEL_4K); | |
1386 | ||
89212919 KZ |
1387 | kvm_mmu_slot_gfn_write_protect(kvm, slot, start, PG_LEVEL_2M); |
1388 | ||
1389 | /* Cross two large pages? */ | |
1390 | if (ALIGN(start << PAGE_SHIFT, PMD_SIZE) != | |
1391 | ALIGN(end << PAGE_SHIFT, PMD_SIZE)) | |
1392 | kvm_mmu_slot_gfn_write_protect(kvm, slot, end, | |
1393 | PG_LEVEL_2M); | |
1394 | } | |
1395 | ||
1396 | /* Now handle 4K PTEs. */ | |
a018eba5 SC |
1397 | if (kvm_x86_ops.cpu_dirty_log_size) |
1398 | kvm_mmu_clear_dirty_pt_masked(kvm, slot, gfn_offset, mask); | |
88178fd4 KH |
1399 | else |
1400 | kvm_mmu_write_protect_pt_masked(kvm, slot, gfn_offset, mask); | |
3b0f1d01 KH |
1401 | } |
1402 | ||
fb04a1ed PX |
1403 | int kvm_cpu_dirty_log_size(void) |
1404 | { | |
6dd03800 | 1405 | return kvm_x86_ops.cpu_dirty_log_size; |
fb04a1ed PX |
1406 | } |
1407 | ||
aeecee2e | 1408 | bool kvm_mmu_slot_gfn_write_protect(struct kvm *kvm, |
3ad93562 KZ |
1409 | struct kvm_memory_slot *slot, u64 gfn, |
1410 | int min_level) | |
95d4c16c | 1411 | { |
018aabb5 | 1412 | struct kvm_rmap_head *rmap_head; |
5dc99b23 | 1413 | int i; |
2f84569f | 1414 | bool write_protected = false; |
95d4c16c | 1415 | |
e2209710 BG |
1416 | if (kvm_memslots_have_rmaps(kvm)) { |
1417 | for (i = min_level; i <= KVM_MAX_HUGEPAGE_LEVEL; ++i) { | |
93e083d4 | 1418 | rmap_head = gfn_to_rmap(gfn, i, slot); |
1346bbb6 | 1419 | write_protected |= rmap_write_protect(rmap_head, true); |
e2209710 | 1420 | } |
5dc99b23 TY |
1421 | } |
1422 | ||
1f98f2bd | 1423 | if (tdp_mmu_enabled) |
46044f72 | 1424 | write_protected |= |
3ad93562 | 1425 | kvm_tdp_mmu_write_protect_gfn(kvm, slot, gfn, min_level); |
46044f72 | 1426 | |
5dc99b23 | 1427 | return write_protected; |
95d4c16c TY |
1428 | } |
1429 | ||
cf48f9e2 | 1430 | static bool kvm_vcpu_write_protect_gfn(struct kvm_vcpu *vcpu, u64 gfn) |
aeecee2e XG |
1431 | { |
1432 | struct kvm_memory_slot *slot; | |
1433 | ||
1434 | slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
3ad93562 | 1435 | return kvm_mmu_slot_gfn_write_protect(vcpu->kvm, slot, gfn, PG_LEVEL_4K); |
aeecee2e XG |
1436 | } |
1437 | ||
f8480721 SC |
1438 | static bool __kvm_zap_rmap(struct kvm *kvm, struct kvm_rmap_head *rmap_head, |
1439 | const struct kvm_memory_slot *slot) | |
e930bffe | 1440 | { |
9202aee8 | 1441 | return kvm_zap_all_rmap_sptes(kvm, rmap_head); |
6a49f85c XG |
1442 | } |
1443 | ||
f8480721 SC |
1444 | static bool kvm_zap_rmap(struct kvm *kvm, struct kvm_rmap_head *rmap_head, |
1445 | struct kvm_memory_slot *slot, gfn_t gfn, int level, | |
1446 | pte_t unused) | |
6a49f85c | 1447 | { |
f8480721 | 1448 | return __kvm_zap_rmap(kvm, rmap_head, slot); |
e930bffe AA |
1449 | } |
1450 | ||
aed02fe3 SC |
1451 | static bool kvm_set_pte_rmap(struct kvm *kvm, struct kvm_rmap_head *rmap_head, |
1452 | struct kvm_memory_slot *slot, gfn_t gfn, int level, | |
1453 | pte_t pte) | |
3da0dd43 | 1454 | { |
1e3f42f0 TY |
1455 | u64 *sptep; |
1456 | struct rmap_iterator iter; | |
98a26b69 | 1457 | bool need_flush = false; |
1e3f42f0 | 1458 | u64 new_spte; |
ba049e93 | 1459 | kvm_pfn_t new_pfn; |
3da0dd43 | 1460 | |
20ba462d | 1461 | WARN_ON_ONCE(pte_huge(pte)); |
3039bcc7 | 1462 | new_pfn = pte_pfn(pte); |
1e3f42f0 | 1463 | |
0d536790 | 1464 | restart: |
018aabb5 | 1465 | for_each_rmap_spte(rmap_head, &iter, sptep) { |
98a26b69 | 1466 | need_flush = true; |
1e3f42f0 | 1467 | |
3039bcc7 | 1468 | if (pte_write(pte)) { |
9202aee8 | 1469 | kvm_zap_one_rmap_spte(kvm, rmap_head, sptep); |
0d536790 | 1470 | goto restart; |
3da0dd43 | 1471 | } else { |
cb3eedab PB |
1472 | new_spte = kvm_mmu_changed_pte_notifier_make_spte( |
1473 | *sptep, new_pfn); | |
1e3f42f0 | 1474 | |
71f51d2c | 1475 | mmu_spte_clear_track_bits(kvm, sptep); |
1e3f42f0 | 1476 | mmu_spte_set(sptep, new_spte); |
3da0dd43 IE |
1477 | } |
1478 | } | |
1e3f42f0 | 1479 | |
8a1300ff | 1480 | if (need_flush && kvm_available_flush_remote_tlbs_range()) { |
9ffe9265 | 1481 | kvm_flush_remote_tlbs_gfn(kvm, gfn, level); |
98a26b69 | 1482 | return false; |
3cc5ea94 LT |
1483 | } |
1484 | ||
0cf853c5 | 1485 | return need_flush; |
3da0dd43 IE |
1486 | } |
1487 | ||
6ce1f4e2 XG |
1488 | struct slot_rmap_walk_iterator { |
1489 | /* input fields. */ | |
269e9552 | 1490 | const struct kvm_memory_slot *slot; |
6ce1f4e2 XG |
1491 | gfn_t start_gfn; |
1492 | gfn_t end_gfn; | |
1493 | int start_level; | |
1494 | int end_level; | |
1495 | ||
1496 | /* output fields. */ | |
1497 | gfn_t gfn; | |
018aabb5 | 1498 | struct kvm_rmap_head *rmap; |
6ce1f4e2 XG |
1499 | int level; |
1500 | ||
1501 | /* private field. */ | |
018aabb5 | 1502 | struct kvm_rmap_head *end_rmap; |
6ce1f4e2 XG |
1503 | }; |
1504 | ||
f3d90f90 SC |
1505 | static void rmap_walk_init_level(struct slot_rmap_walk_iterator *iterator, |
1506 | int level) | |
6ce1f4e2 XG |
1507 | { |
1508 | iterator->level = level; | |
1509 | iterator->gfn = iterator->start_gfn; | |
93e083d4 DM |
1510 | iterator->rmap = gfn_to_rmap(iterator->gfn, level, iterator->slot); |
1511 | iterator->end_rmap = gfn_to_rmap(iterator->end_gfn, level, iterator->slot); | |
6ce1f4e2 XG |
1512 | } |
1513 | ||
f3d90f90 SC |
1514 | static void slot_rmap_walk_init(struct slot_rmap_walk_iterator *iterator, |
1515 | const struct kvm_memory_slot *slot, | |
1516 | int start_level, int end_level, | |
1517 | gfn_t start_gfn, gfn_t end_gfn) | |
6ce1f4e2 XG |
1518 | { |
1519 | iterator->slot = slot; | |
1520 | iterator->start_level = start_level; | |
1521 | iterator->end_level = end_level; | |
1522 | iterator->start_gfn = start_gfn; | |
1523 | iterator->end_gfn = end_gfn; | |
1524 | ||
1525 | rmap_walk_init_level(iterator, iterator->start_level); | |
1526 | } | |
1527 | ||
1528 | static bool slot_rmap_walk_okay(struct slot_rmap_walk_iterator *iterator) | |
1529 | { | |
1530 | return !!iterator->rmap; | |
1531 | } | |
1532 | ||
1533 | static void slot_rmap_walk_next(struct slot_rmap_walk_iterator *iterator) | |
1534 | { | |
6ba1e04f | 1535 | while (++iterator->rmap <= iterator->end_rmap) { |
6ce1f4e2 | 1536 | iterator->gfn += (1UL << KVM_HPAGE_GFN_SHIFT(iterator->level)); |
6ba1e04f VS |
1537 | |
1538 | if (iterator->rmap->val) | |
1539 | return; | |
6ce1f4e2 XG |
1540 | } |
1541 | ||
1542 | if (++iterator->level > iterator->end_level) { | |
1543 | iterator->rmap = NULL; | |
1544 | return; | |
1545 | } | |
1546 | ||
1547 | rmap_walk_init_level(iterator, iterator->level); | |
1548 | } | |
1549 | ||
1550 | #define for_each_slot_rmap_range(_slot_, _start_level_, _end_level_, \ | |
1551 | _start_gfn, _end_gfn, _iter_) \ | |
1552 | for (slot_rmap_walk_init(_iter_, _slot_, _start_level_, \ | |
1553 | _end_level_, _start_gfn, _end_gfn); \ | |
1554 | slot_rmap_walk_okay(_iter_); \ | |
1555 | slot_rmap_walk_next(_iter_)) | |
1556 | ||
3039bcc7 SC |
1557 | typedef bool (*rmap_handler_t)(struct kvm *kvm, struct kvm_rmap_head *rmap_head, |
1558 | struct kvm_memory_slot *slot, gfn_t gfn, | |
1559 | int level, pte_t pte); | |
c1b91493 | 1560 | |
3039bcc7 SC |
1561 | static __always_inline bool kvm_handle_gfn_range(struct kvm *kvm, |
1562 | struct kvm_gfn_range *range, | |
1563 | rmap_handler_t handler) | |
e930bffe | 1564 | { |
6ce1f4e2 | 1565 | struct slot_rmap_walk_iterator iterator; |
3039bcc7 | 1566 | bool ret = false; |
e930bffe | 1567 | |
3039bcc7 SC |
1568 | for_each_slot_rmap_range(range->slot, PG_LEVEL_4K, KVM_MAX_HUGEPAGE_LEVEL, |
1569 | range->start, range->end - 1, &iterator) | |
1570 | ret |= handler(kvm, iterator.rmap, range->slot, iterator.gfn, | |
3e1efe2b | 1571 | iterator.level, range->arg.pte); |
e930bffe | 1572 | |
f395302e | 1573 | return ret; |
e930bffe AA |
1574 | } |
1575 | ||
3039bcc7 | 1576 | bool kvm_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range) |
84504ef3 | 1577 | { |
e2209710 | 1578 | bool flush = false; |
063afacd | 1579 | |
e2209710 | 1580 | if (kvm_memslots_have_rmaps(kvm)) |
f8480721 | 1581 | flush = kvm_handle_gfn_range(kvm, range, kvm_zap_rmap); |
063afacd | 1582 | |
1f98f2bd | 1583 | if (tdp_mmu_enabled) |
c7785d85 | 1584 | flush = kvm_tdp_mmu_unmap_gfn_range(kvm, range, flush); |
063afacd | 1585 | |
0a3869e1 SC |
1586 | if (kvm_x86_ops.set_apic_access_page_addr && |
1587 | range->slot->id == APIC_ACCESS_PAGE_PRIVATE_MEMSLOT) | |
0a8a5f2c SC |
1588 | kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD); |
1589 | ||
3039bcc7 | 1590 | return flush; |
b3ae2096 TY |
1591 | } |
1592 | ||
3039bcc7 | 1593 | bool kvm_set_spte_gfn(struct kvm *kvm, struct kvm_gfn_range *range) |
3da0dd43 | 1594 | { |
e2209710 | 1595 | bool flush = false; |
1d8dd6b3 | 1596 | |
e2209710 | 1597 | if (kvm_memslots_have_rmaps(kvm)) |
aed02fe3 | 1598 | flush = kvm_handle_gfn_range(kvm, range, kvm_set_pte_rmap); |
1d8dd6b3 | 1599 | |
1f98f2bd | 1600 | if (tdp_mmu_enabled) |
3039bcc7 | 1601 | flush |= kvm_tdp_mmu_set_spte_gfn(kvm, range); |
1d8dd6b3 | 1602 | |
3039bcc7 | 1603 | return flush; |
e930bffe AA |
1604 | } |
1605 | ||
aed02fe3 SC |
1606 | static bool kvm_age_rmap(struct kvm *kvm, struct kvm_rmap_head *rmap_head, |
1607 | struct kvm_memory_slot *slot, gfn_t gfn, int level, | |
1608 | pte_t unused) | |
e930bffe | 1609 | { |
1e3f42f0 | 1610 | u64 *sptep; |
3f649ab7 | 1611 | struct rmap_iterator iter; |
e930bffe AA |
1612 | int young = 0; |
1613 | ||
f160c7b7 JS |
1614 | for_each_rmap_spte(rmap_head, &iter, sptep) |
1615 | young |= mmu_spte_age(sptep); | |
0d536790 | 1616 | |
e930bffe AA |
1617 | return young; |
1618 | } | |
1619 | ||
aed02fe3 SC |
1620 | static bool kvm_test_age_rmap(struct kvm *kvm, struct kvm_rmap_head *rmap_head, |
1621 | struct kvm_memory_slot *slot, gfn_t gfn, | |
1622 | int level, pte_t unused) | |
8ee53820 | 1623 | { |
1e3f42f0 TY |
1624 | u64 *sptep; |
1625 | struct rmap_iterator iter; | |
8ee53820 | 1626 | |
83ef6c81 JS |
1627 | for_each_rmap_spte(rmap_head, &iter, sptep) |
1628 | if (is_accessed_spte(*sptep)) | |
98a26b69 VM |
1629 | return true; |
1630 | return false; | |
8ee53820 AA |
1631 | } |
1632 | ||
53a27b39 MT |
1633 | #define RMAP_RECYCLE_THRESHOLD 1000 |
1634 | ||
2ff9039a DM |
1635 | static void __rmap_add(struct kvm *kvm, |
1636 | struct kvm_mmu_memory_cache *cache, | |
1637 | const struct kvm_memory_slot *slot, | |
72ae5822 | 1638 | u64 *spte, gfn_t gfn, unsigned int access) |
53a27b39 | 1639 | { |
852e3c19 | 1640 | struct kvm_mmu_page *sp; |
68be1306 DM |
1641 | struct kvm_rmap_head *rmap_head; |
1642 | int rmap_count; | |
852e3c19 | 1643 | |
57354682 | 1644 | sp = sptep_to_sp(spte); |
79e48cec | 1645 | kvm_mmu_page_set_translation(sp, spte_index(spte), gfn, access); |
81cb4657 DM |
1646 | kvm_update_page_stats(kvm, sp->role.level, 1); |
1647 | ||
93e083d4 | 1648 | rmap_head = gfn_to_rmap(gfn, sp->role.level, slot); |
2ff9039a | 1649 | rmap_count = pte_list_add(cache, spte, rmap_head); |
53a27b39 | 1650 | |
604f5332 ML |
1651 | if (rmap_count > kvm->stat.max_mmu_rmap_size) |
1652 | kvm->stat.max_mmu_rmap_size = rmap_count; | |
68be1306 | 1653 | if (rmap_count > RMAP_RECYCLE_THRESHOLD) { |
9202aee8 | 1654 | kvm_zap_all_rmap_sptes(kvm, rmap_head); |
1b2dc736 | 1655 | kvm_flush_remote_tlbs_gfn(kvm, gfn, sp->role.level); |
68be1306 | 1656 | } |
53a27b39 MT |
1657 | } |
1658 | ||
2ff9039a | 1659 | static void rmap_add(struct kvm_vcpu *vcpu, const struct kvm_memory_slot *slot, |
72ae5822 | 1660 | u64 *spte, gfn_t gfn, unsigned int access) |
2ff9039a DM |
1661 | { |
1662 | struct kvm_mmu_memory_cache *cache = &vcpu->arch.mmu_pte_list_desc_cache; | |
1663 | ||
6a97575d | 1664 | __rmap_add(vcpu->kvm, cache, slot, spte, gfn, access); |
2ff9039a DM |
1665 | } |
1666 | ||
3039bcc7 | 1667 | bool kvm_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range) |
e930bffe | 1668 | { |
e2209710 | 1669 | bool young = false; |
3039bcc7 | 1670 | |
e2209710 | 1671 | if (kvm_memslots_have_rmaps(kvm)) |
aed02fe3 | 1672 | young = kvm_handle_gfn_range(kvm, range, kvm_age_rmap); |
f8e14497 | 1673 | |
1f98f2bd | 1674 | if (tdp_mmu_enabled) |
3039bcc7 | 1675 | young |= kvm_tdp_mmu_age_gfn_range(kvm, range); |
f8e14497 BG |
1676 | |
1677 | return young; | |
e930bffe AA |
1678 | } |
1679 | ||
3039bcc7 | 1680 | bool kvm_test_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range) |
8ee53820 | 1681 | { |
e2209710 | 1682 | bool young = false; |
3039bcc7 | 1683 | |
e2209710 | 1684 | if (kvm_memslots_have_rmaps(kvm)) |
aed02fe3 | 1685 | young = kvm_handle_gfn_range(kvm, range, kvm_test_age_rmap); |
f8e14497 | 1686 | |
1f98f2bd | 1687 | if (tdp_mmu_enabled) |
3039bcc7 | 1688 | young |= kvm_tdp_mmu_test_age_gfn(kvm, range); |
f8e14497 BG |
1689 | |
1690 | return young; | |
8ee53820 AA |
1691 | } |
1692 | ||
58da926c | 1693 | static void kvm_mmu_check_sptes_at_free(struct kvm_mmu_page *sp) |
6aa8b732 | 1694 | { |
870d4d4e | 1695 | #ifdef CONFIG_KVM_PROVE_MMU |
242a6dd8 | 1696 | int i; |
139bdb2d | 1697 | |
242a6dd8 | 1698 | for (i = 0; i < SPTE_ENT_PER_PAGE; i++) { |
0fe6370e | 1699 | if (KVM_MMU_WARN_ON(is_shadow_present_pte(sp->spt[i]))) |
58da926c SC |
1700 | pr_err_ratelimited("SPTE %llx (@ %p) for gfn %llx shadow-present at free", |
1701 | sp->spt[i], &sp->spt[i], | |
1702 | kvm_mmu_page_get_gfn(sp, i)); | |
242a6dd8 | 1703 | } |
d6c69ee9 | 1704 | #endif |
58da926c | 1705 | } |
6aa8b732 | 1706 | |
45221ab6 DH |
1707 | /* |
1708 | * This value is the sum of all of the kvm instances's | |
1709 | * kvm->arch.n_used_mmu_pages values. We need a global, | |
1710 | * aggregate version in order to make the slab shrinker | |
1711 | * faster | |
1712 | */ | |
d5aaad6f | 1713 | static inline void kvm_mod_used_mmu_pages(struct kvm *kvm, long nr) |
45221ab6 DH |
1714 | { |
1715 | kvm->arch.n_used_mmu_pages += nr; | |
1716 | percpu_counter_add(&kvm_total_used_mmu_pages, nr); | |
1717 | } | |
1718 | ||
43a063ca YA |
1719 | static void kvm_account_mmu_page(struct kvm *kvm, struct kvm_mmu_page *sp) |
1720 | { | |
1721 | kvm_mod_used_mmu_pages(kvm, +1); | |
1722 | kvm_account_pgtable_pages((void *)sp->spt, +1); | |
1723 | } | |
1724 | ||
1725 | static void kvm_unaccount_mmu_page(struct kvm *kvm, struct kvm_mmu_page *sp) | |
1726 | { | |
1727 | kvm_mod_used_mmu_pages(kvm, -1); | |
1728 | kvm_account_pgtable_pages((void *)sp->spt, -1); | |
1729 | } | |
1730 | ||
87654643 | 1731 | static void kvm_mmu_free_shadow_page(struct kvm_mmu_page *sp) |
260746c0 | 1732 | { |
58da926c SC |
1733 | kvm_mmu_check_sptes_at_free(sp); |
1734 | ||
7775834a | 1735 | hlist_del(&sp->hash_link); |
bd4c86ea XG |
1736 | list_del(&sp->link); |
1737 | free_page((unsigned long)sp->spt); | |
834be0d8 | 1738 | if (!sp->role.direct) |
6a97575d | 1739 | free_page((unsigned long)sp->shadowed_translation); |
e8ad9a70 | 1740 | kmem_cache_free(mmu_page_header_cache, sp); |
260746c0 AK |
1741 | } |
1742 | ||
cea0f0e7 AK |
1743 | static unsigned kvm_page_table_hashfn(gfn_t gfn) |
1744 | { | |
114df303 | 1745 | return hash_64(gfn, KVM_MMU_HASH_SHIFT); |
cea0f0e7 AK |
1746 | } |
1747 | ||
2ff9039a | 1748 | static void mmu_page_add_parent_pte(struct kvm_mmu_memory_cache *cache, |
4db35314 | 1749 | struct kvm_mmu_page *sp, u64 *parent_pte) |
cea0f0e7 | 1750 | { |
cea0f0e7 AK |
1751 | if (!parent_pte) |
1752 | return; | |
cea0f0e7 | 1753 | |
2ff9039a | 1754 | pte_list_add(cache, parent_pte, &sp->parent_ptes); |
cea0f0e7 AK |
1755 | } |
1756 | ||
069f30c6 | 1757 | static void mmu_page_remove_parent_pte(struct kvm *kvm, struct kvm_mmu_page *sp, |
cea0f0e7 AK |
1758 | u64 *parent_pte) |
1759 | { | |
069f30c6 | 1760 | pte_list_remove(kvm, parent_pte, &sp->parent_ptes); |
cea0f0e7 AK |
1761 | } |
1762 | ||
069f30c6 | 1763 | static void drop_parent_pte(struct kvm *kvm, struct kvm_mmu_page *sp, |
bcdd9a93 XG |
1764 | u64 *parent_pte) |
1765 | { | |
069f30c6 | 1766 | mmu_page_remove_parent_pte(kvm, sp, parent_pte); |
1df9f2dc | 1767 | mmu_spte_clear_no_track(parent_pte); |
bcdd9a93 XG |
1768 | } |
1769 | ||
67052b35 | 1770 | static void mark_unsync(u64 *spte); |
1047df1f | 1771 | static void kvm_mmu_mark_parents_unsync(struct kvm_mmu_page *sp) |
0074ff63 | 1772 | { |
74c4e63a TY |
1773 | u64 *sptep; |
1774 | struct rmap_iterator iter; | |
1775 | ||
1776 | for_each_rmap_spte(&sp->parent_ptes, &iter, sptep) { | |
1777 | mark_unsync(sptep); | |
1778 | } | |
0074ff63 MT |
1779 | } |
1780 | ||
67052b35 | 1781 | static void mark_unsync(u64 *spte) |
0074ff63 | 1782 | { |
67052b35 | 1783 | struct kvm_mmu_page *sp; |
0074ff63 | 1784 | |
57354682 | 1785 | sp = sptep_to_sp(spte); |
79e48cec | 1786 | if (__test_and_set_bit(spte_index(spte), sp->unsync_child_bitmap)) |
0074ff63 | 1787 | return; |
1047df1f | 1788 | if (sp->unsync_children++) |
0074ff63 | 1789 | return; |
1047df1f | 1790 | kvm_mmu_mark_parents_unsync(sp); |
0074ff63 MT |
1791 | } |
1792 | ||
60c8aec6 MT |
1793 | #define KVM_PAGE_ARRAY_NR 16 |
1794 | ||
1795 | struct kvm_mmu_pages { | |
1796 | struct mmu_page_and_offset { | |
1797 | struct kvm_mmu_page *sp; | |
1798 | unsigned int idx; | |
1799 | } page[KVM_PAGE_ARRAY_NR]; | |
1800 | unsigned int nr; | |
1801 | }; | |
1802 | ||
cded19f3 HE |
1803 | static int mmu_pages_add(struct kvm_mmu_pages *pvec, struct kvm_mmu_page *sp, |
1804 | int idx) | |
4731d4c7 | 1805 | { |
60c8aec6 | 1806 | int i; |
4731d4c7 | 1807 | |
60c8aec6 MT |
1808 | if (sp->unsync) |
1809 | for (i=0; i < pvec->nr; i++) | |
1810 | if (pvec->page[i].sp == sp) | |
1811 | return 0; | |
1812 | ||
1813 | pvec->page[pvec->nr].sp = sp; | |
1814 | pvec->page[pvec->nr].idx = idx; | |
1815 | pvec->nr++; | |
1816 | return (pvec->nr == KVM_PAGE_ARRAY_NR); | |
1817 | } | |
1818 | ||
fd951457 TY |
1819 | static inline void clear_unsync_child_bit(struct kvm_mmu_page *sp, int idx) |
1820 | { | |
1821 | --sp->unsync_children; | |
20ba462d | 1822 | WARN_ON_ONCE((int)sp->unsync_children < 0); |
fd951457 TY |
1823 | __clear_bit(idx, sp->unsync_child_bitmap); |
1824 | } | |
1825 | ||
60c8aec6 MT |
1826 | static int __mmu_unsync_walk(struct kvm_mmu_page *sp, |
1827 | struct kvm_mmu_pages *pvec) | |
1828 | { | |
1829 | int i, ret, nr_unsync_leaf = 0; | |
4731d4c7 | 1830 | |
37178b8b | 1831 | for_each_set_bit(i, sp->unsync_child_bitmap, 512) { |
7a8f1a74 | 1832 | struct kvm_mmu_page *child; |
4731d4c7 MT |
1833 | u64 ent = sp->spt[i]; |
1834 | ||
fd951457 TY |
1835 | if (!is_shadow_present_pte(ent) || is_large_pte(ent)) { |
1836 | clear_unsync_child_bit(sp, i); | |
1837 | continue; | |
1838 | } | |
7a8f1a74 | 1839 | |
5e3edd7e | 1840 | child = spte_to_child_sp(ent); |
7a8f1a74 XG |
1841 | |
1842 | if (child->unsync_children) { | |
1843 | if (mmu_pages_add(pvec, child, i)) | |
1844 | return -ENOSPC; | |
1845 | ||
1846 | ret = __mmu_unsync_walk(child, pvec); | |
fd951457 TY |
1847 | if (!ret) { |
1848 | clear_unsync_child_bit(sp, i); | |
1849 | continue; | |
1850 | } else if (ret > 0) { | |
7a8f1a74 | 1851 | nr_unsync_leaf += ret; |
fd951457 | 1852 | } else |
7a8f1a74 XG |
1853 | return ret; |
1854 | } else if (child->unsync) { | |
1855 | nr_unsync_leaf++; | |
1856 | if (mmu_pages_add(pvec, child, i)) | |
1857 | return -ENOSPC; | |
1858 | } else | |
fd951457 | 1859 | clear_unsync_child_bit(sp, i); |
4731d4c7 MT |
1860 | } |
1861 | ||
60c8aec6 MT |
1862 | return nr_unsync_leaf; |
1863 | } | |
1864 | ||
e23d3fef XG |
1865 | #define INVALID_INDEX (-1) |
1866 | ||
60c8aec6 MT |
1867 | static int mmu_unsync_walk(struct kvm_mmu_page *sp, |
1868 | struct kvm_mmu_pages *pvec) | |
1869 | { | |
0a47cd85 | 1870 | pvec->nr = 0; |
60c8aec6 MT |
1871 | if (!sp->unsync_children) |
1872 | return 0; | |
1873 | ||
e23d3fef | 1874 | mmu_pages_add(pvec, sp, INVALID_INDEX); |
60c8aec6 | 1875 | return __mmu_unsync_walk(sp, pvec); |
4731d4c7 MT |
1876 | } |
1877 | ||
4731d4c7 MT |
1878 | static void kvm_unlink_unsync_page(struct kvm *kvm, struct kvm_mmu_page *sp) |
1879 | { | |
20ba462d | 1880 | WARN_ON_ONCE(!sp->unsync); |
5e1b3ddb | 1881 | trace_kvm_mmu_sync_page(sp); |
4731d4c7 MT |
1882 | sp->unsync = 0; |
1883 | --kvm->stat.mmu_unsync; | |
1884 | } | |
1885 | ||
83cdb568 SC |
1886 | static bool kvm_mmu_prepare_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp, |
1887 | struct list_head *invalid_list); | |
7775834a XG |
1888 | static void kvm_mmu_commit_zap_page(struct kvm *kvm, |
1889 | struct list_head *invalid_list); | |
4731d4c7 | 1890 | |
767d8d8d LJ |
1891 | static bool sp_has_gptes(struct kvm_mmu_page *sp) |
1892 | { | |
1893 | if (sp->role.direct) | |
1894 | return false; | |
1895 | ||
84e5ffd0 LJ |
1896 | if (sp->role.passthrough) |
1897 | return false; | |
1898 | ||
767d8d8d LJ |
1899 | return true; |
1900 | } | |
1901 | ||
ac101b7c SC |
1902 | #define for_each_valid_sp(_kvm, _sp, _list) \ |
1903 | hlist_for_each_entry(_sp, _list, hash_link) \ | |
fac026da | 1904 | if (is_obsolete_sp((_kvm), (_sp))) { \ |
f3414bc7 | 1905 | } else |
1044b030 | 1906 | |
767d8d8d | 1907 | #define for_each_gfn_valid_sp_with_gptes(_kvm, _sp, _gfn) \ |
ac101b7c SC |
1908 | for_each_valid_sp(_kvm, _sp, \ |
1909 | &(_kvm)->arch.mmu_page_hash[kvm_page_table_hashfn(_gfn)]) \ | |
767d8d8d | 1910 | if ((_sp)->gfn != (_gfn) || !sp_has_gptes(_sp)) {} else |
7ae680eb | 1911 | |
90e44470 LJ |
1912 | static bool kvm_sync_page_check(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp) |
1913 | { | |
1914 | union kvm_mmu_page_role root_role = vcpu->arch.mmu->root_role; | |
1915 | ||
1916 | /* | |
1917 | * Ignore various flags when verifying that it's safe to sync a shadow | |
1918 | * page using the current MMU context. | |
1919 | * | |
1920 | * - level: not part of the overall MMU role and will never match as the MMU's | |
1921 | * level tracks the root level | |
1922 | * - access: updated based on the new guest PTE | |
1923 | * - quadrant: not part of the overall MMU role (similar to level) | |
1924 | */ | |
1925 | const union kvm_mmu_page_role sync_role_ign = { | |
1926 | .level = 0xf, | |
1927 | .access = 0x7, | |
1928 | .quadrant = 0x3, | |
1929 | .passthrough = 0x1, | |
1930 | }; | |
1931 | ||
1932 | /* | |
1933 | * Direct pages can never be unsync, and KVM should never attempt to | |
1934 | * sync a shadow page for a different MMU context, e.g. if the role | |
1935 | * differs then the memslot lookup (SMM vs. non-SMM) will be bogus, the | |
1936 | * reserved bits checks will be wrong, etc... | |
1937 | */ | |
c3c6c9fc | 1938 | if (WARN_ON_ONCE(sp->role.direct || !vcpu->arch.mmu->sync_spte || |
90e44470 LJ |
1939 | (sp->role.word ^ root_role.word) & ~sync_role_ign.word)) |
1940 | return false; | |
1941 | ||
1942 | return true; | |
1943 | } | |
1944 | ||
19ace7d6 LJ |
1945 | static int kvm_sync_spte(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp, int i) |
1946 | { | |
1947 | if (!sp->spt[i]) | |
1948 | return 0; | |
1949 | ||
1950 | return vcpu->arch.mmu->sync_spte(vcpu, sp, i); | |
1951 | } | |
1952 | ||
90e44470 LJ |
1953 | static int __kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp) |
1954 | { | |
c3c6c9fc LJ |
1955 | int flush = 0; |
1956 | int i; | |
1957 | ||
90e44470 LJ |
1958 | if (!kvm_sync_page_check(vcpu, sp)) |
1959 | return -1; | |
1960 | ||
c3c6c9fc | 1961 | for (i = 0; i < SPTE_ENT_PER_PAGE; i++) { |
19ace7d6 | 1962 | int ret = kvm_sync_spte(vcpu, sp, i); |
c3c6c9fc LJ |
1963 | |
1964 | if (ret < -1) | |
1965 | return -1; | |
1966 | flush |= ret; | |
1967 | } | |
1968 | ||
1969 | /* | |
1970 | * Note, any flush is purely for KVM's correctness, e.g. when dropping | |
1971 | * an existing SPTE or clearing W/A/D bits to ensure an mmu_notifier | |
1972 | * unmap or dirty logging event doesn't fail to flush. The guest is | |
1973 | * responsible for flushing the TLB to ensure any changes in protection | |
1974 | * bits are recognized, i.e. until the guest flushes or page faults on | |
1975 | * a relevant address, KVM is architecturally allowed to let vCPUs use | |
1976 | * cached translations with the old protection bits. | |
1977 | */ | |
1978 | return flush; | |
90e44470 LJ |
1979 | } |
1980 | ||
8d5678a7 | 1981 | static int kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp, |
479a1efc | 1982 | struct list_head *invalid_list) |
4731d4c7 | 1983 | { |
90e44470 | 1984 | int ret = __kvm_sync_page(vcpu, sp); |
c3e5e415 | 1985 | |
8d5678a7 | 1986 | if (ret < 0) |
d98ba053 | 1987 | kvm_mmu_prepare_zap_page(vcpu->kvm, sp, invalid_list); |
8d5678a7 | 1988 | return ret; |
4731d4c7 MT |
1989 | } |
1990 | ||
a2113634 SC |
1991 | static bool kvm_mmu_remote_flush_or_zap(struct kvm *kvm, |
1992 | struct list_head *invalid_list, | |
1993 | bool remote_flush) | |
1994 | { | |
cfd32acf | 1995 | if (!remote_flush && list_empty(invalid_list)) |
a2113634 SC |
1996 | return false; |
1997 | ||
1998 | if (!list_empty(invalid_list)) | |
1999 | kvm_mmu_commit_zap_page(kvm, invalid_list); | |
2000 | else | |
2001 | kvm_flush_remote_tlbs(kvm); | |
2002 | return true; | |
2003 | } | |
2004 | ||
002c5f73 SC |
2005 | static bool is_obsolete_sp(struct kvm *kvm, struct kvm_mmu_page *sp) |
2006 | { | |
a955cad8 SC |
2007 | if (sp->role.invalid) |
2008 | return true; | |
2009 | ||
fa3e4203 | 2010 | /* TDP MMU pages do not use the MMU generation. */ |
de0322f5 | 2011 | return !is_tdp_mmu_page(sp) && |
fac026da | 2012 | unlikely(sp->mmu_valid_gen != kvm->arch.mmu_valid_gen); |
002c5f73 SC |
2013 | } |
2014 | ||
60c8aec6 | 2015 | struct mmu_page_path { |
2a7266a8 YZ |
2016 | struct kvm_mmu_page *parent[PT64_ROOT_MAX_LEVEL]; |
2017 | unsigned int idx[PT64_ROOT_MAX_LEVEL]; | |
4731d4c7 MT |
2018 | }; |
2019 | ||
60c8aec6 | 2020 | #define for_each_sp(pvec, sp, parents, i) \ |
0a47cd85 | 2021 | for (i = mmu_pages_first(&pvec, &parents); \ |
60c8aec6 MT |
2022 | i < pvec.nr && ({ sp = pvec.page[i].sp; 1;}); \ |
2023 | i = mmu_pages_next(&pvec, &parents, i)) | |
2024 | ||
cded19f3 HE |
2025 | static int mmu_pages_next(struct kvm_mmu_pages *pvec, |
2026 | struct mmu_page_path *parents, | |
2027 | int i) | |
60c8aec6 MT |
2028 | { |
2029 | int n; | |
2030 | ||
2031 | for (n = i+1; n < pvec->nr; n++) { | |
2032 | struct kvm_mmu_page *sp = pvec->page[n].sp; | |
0a47cd85 PB |
2033 | unsigned idx = pvec->page[n].idx; |
2034 | int level = sp->role.level; | |
60c8aec6 | 2035 | |
0a47cd85 | 2036 | parents->idx[level-1] = idx; |
3bae0459 | 2037 | if (level == PG_LEVEL_4K) |
0a47cd85 | 2038 | break; |
60c8aec6 | 2039 | |
0a47cd85 | 2040 | parents->parent[level-2] = sp; |
60c8aec6 MT |
2041 | } |
2042 | ||
2043 | return n; | |
2044 | } | |
2045 | ||
0a47cd85 PB |
2046 | static int mmu_pages_first(struct kvm_mmu_pages *pvec, |
2047 | struct mmu_page_path *parents) | |
2048 | { | |
2049 | struct kvm_mmu_page *sp; | |
2050 | int level; | |
2051 | ||
2052 | if (pvec->nr == 0) | |
2053 | return 0; | |
2054 | ||
20ba462d | 2055 | WARN_ON_ONCE(pvec->page[0].idx != INVALID_INDEX); |
e23d3fef | 2056 | |
0a47cd85 PB |
2057 | sp = pvec->page[0].sp; |
2058 | level = sp->role.level; | |
20ba462d | 2059 | WARN_ON_ONCE(level == PG_LEVEL_4K); |
0a47cd85 PB |
2060 | |
2061 | parents->parent[level-2] = sp; | |
2062 | ||
2063 | /* Also set up a sentinel. Further entries in pvec are all | |
2064 | * children of sp, so this element is never overwritten. | |
2065 | */ | |
2066 | parents->parent[level-1] = NULL; | |
2067 | return mmu_pages_next(pvec, parents, 0); | |
2068 | } | |
2069 | ||
cded19f3 | 2070 | static void mmu_pages_clear_parents(struct mmu_page_path *parents) |
4731d4c7 | 2071 | { |
60c8aec6 MT |
2072 | struct kvm_mmu_page *sp; |
2073 | unsigned int level = 0; | |
2074 | ||
2075 | do { | |
2076 | unsigned int idx = parents->idx[level]; | |
60c8aec6 MT |
2077 | sp = parents->parent[level]; |
2078 | if (!sp) | |
2079 | return; | |
2080 | ||
20ba462d | 2081 | WARN_ON_ONCE(idx == INVALID_INDEX); |
fd951457 | 2082 | clear_unsync_child_bit(sp, idx); |
60c8aec6 | 2083 | level++; |
0a47cd85 | 2084 | } while (!sp->unsync_children); |
60c8aec6 | 2085 | } |
4731d4c7 | 2086 | |
65855ed8 LJ |
2087 | static int mmu_sync_children(struct kvm_vcpu *vcpu, |
2088 | struct kvm_mmu_page *parent, bool can_yield) | |
60c8aec6 MT |
2089 | { |
2090 | int i; | |
2091 | struct kvm_mmu_page *sp; | |
2092 | struct mmu_page_path parents; | |
2093 | struct kvm_mmu_pages pages; | |
d98ba053 | 2094 | LIST_HEAD(invalid_list); |
50c9e6f3 | 2095 | bool flush = false; |
60c8aec6 | 2096 | |
60c8aec6 | 2097 | while (mmu_unsync_walk(parent, &pages)) { |
2f84569f | 2098 | bool protected = false; |
b1a36821 MT |
2099 | |
2100 | for_each_sp(pages, sp, parents, i) | |
cf48f9e2 | 2101 | protected |= kvm_vcpu_write_protect_gfn(vcpu, sp->gfn); |
b1a36821 | 2102 | |
50c9e6f3 | 2103 | if (protected) { |
5591c069 | 2104 | kvm_mmu_remote_flush_or_zap(vcpu->kvm, &invalid_list, true); |
50c9e6f3 PB |
2105 | flush = false; |
2106 | } | |
b1a36821 | 2107 | |
60c8aec6 | 2108 | for_each_sp(pages, sp, parents, i) { |
479a1efc | 2109 | kvm_unlink_unsync_page(vcpu->kvm, sp); |
8d5678a7 | 2110 | flush |= kvm_sync_page(vcpu, sp, &invalid_list) > 0; |
60c8aec6 MT |
2111 | mmu_pages_clear_parents(&parents); |
2112 | } | |
531810ca | 2113 | if (need_resched() || rwlock_needbreak(&vcpu->kvm->mmu_lock)) { |
c3e5e415 | 2114 | kvm_mmu_remote_flush_or_zap(vcpu->kvm, &invalid_list, flush); |
65855ed8 LJ |
2115 | if (!can_yield) { |
2116 | kvm_make_request(KVM_REQ_MMU_SYNC, vcpu); | |
2117 | return -EINTR; | |
2118 | } | |
2119 | ||
531810ca | 2120 | cond_resched_rwlock_write(&vcpu->kvm->mmu_lock); |
50c9e6f3 PB |
2121 | flush = false; |
2122 | } | |
60c8aec6 | 2123 | } |
50c9e6f3 | 2124 | |
c3e5e415 | 2125 | kvm_mmu_remote_flush_or_zap(vcpu->kvm, &invalid_list, flush); |
65855ed8 | 2126 | return 0; |
4731d4c7 MT |
2127 | } |
2128 | ||
a30f47cb XG |
2129 | static void __clear_sp_write_flooding_count(struct kvm_mmu_page *sp) |
2130 | { | |
e5691a81 | 2131 | atomic_set(&sp->write_flooding_count, 0); |
a30f47cb XG |
2132 | } |
2133 | ||
2134 | static void clear_sp_write_flooding_count(u64 *spte) | |
2135 | { | |
57354682 | 2136 | __clear_sp_write_flooding_count(sptep_to_sp(spte)); |
a30f47cb XG |
2137 | } |
2138 | ||
cbd858b1 DM |
2139 | /* |
2140 | * The vCPU is required when finding indirect shadow pages; the shadow | |
2141 | * page may already exist and syncing it needs the vCPU pointer in | |
2142 | * order to read guest page tables. Direct shadow pages are never | |
2143 | * unsync, thus @vcpu can be NULL if @role.direct is true. | |
2144 | */ | |
3cc736b3 DM |
2145 | static struct kvm_mmu_page *kvm_mmu_find_shadow_page(struct kvm *kvm, |
2146 | struct kvm_vcpu *vcpu, | |
94c81364 DM |
2147 | gfn_t gfn, |
2148 | struct hlist_head *sp_list, | |
2149 | union kvm_mmu_page_role role) | |
cea0f0e7 | 2150 | { |
9f1a122f | 2151 | struct kvm_mmu_page *sp; |
8d5678a7 | 2152 | int ret; |
f3414bc7 | 2153 | int collisions = 0; |
2a74003a | 2154 | LIST_HEAD(invalid_list); |
cea0f0e7 | 2155 | |
3cc736b3 | 2156 | for_each_valid_sp(kvm, sp, sp_list) { |
f3414bc7 DM |
2157 | if (sp->gfn != gfn) { |
2158 | collisions++; | |
2159 | continue; | |
2160 | } | |
2161 | ||
ddc16abb SC |
2162 | if (sp->role.word != role.word) { |
2163 | /* | |
2164 | * If the guest is creating an upper-level page, zap | |
2165 | * unsync pages for the same gfn. While it's possible | |
2166 | * the guest is using recursive page tables, in all | |
2167 | * likelihood the guest has stopped using the unsync | |
2168 | * page and is installing a completely unrelated page. | |
2169 | * Unsync pages must not be left as is, because the new | |
2170 | * upper-level page will be write-protected. | |
2171 | */ | |
2e65e842 | 2172 | if (role.level > PG_LEVEL_4K && sp->unsync) |
3cc736b3 | 2173 | kvm_mmu_prepare_zap_page(kvm, sp, |
ddc16abb | 2174 | &invalid_list); |
7ae680eb | 2175 | continue; |
ddc16abb | 2176 | } |
4731d4c7 | 2177 | |
bb924ca6 DM |
2178 | /* unsync and write-flooding only apply to indirect SPs. */ |
2179 | if (sp->role.direct) | |
94c81364 | 2180 | goto out; |
fb58a9c3 | 2181 | |
2a74003a | 2182 | if (sp->unsync) { |
cbd858b1 DM |
2183 | if (KVM_BUG_ON(!vcpu, kvm)) |
2184 | break; | |
2185 | ||
07dc4f35 | 2186 | /* |
479a1efc | 2187 | * The page is good, but is stale. kvm_sync_page does |
07dc4f35 SC |
2188 | * get the latest guest state, but (unlike mmu_unsync_children) |
2189 | * it doesn't write-protect the page or mark it synchronized! | |
2190 | * This way the validity of the mapping is ensured, but the | |
2191 | * overhead of write protection is not incurred until the | |
2192 | * guest invalidates the TLB mapping. This allows multiple | |
2193 | * SPs for a single gfn to be unsync. | |
2194 | * | |
2195 | * If the sync fails, the page is zapped. If so, break | |
2196 | * in order to rebuild it. | |
2a74003a | 2197 | */ |
8d5678a7 HW |
2198 | ret = kvm_sync_page(vcpu, sp, &invalid_list); |
2199 | if (ret < 0) | |
2a74003a PB |
2200 | break; |
2201 | ||
20ba462d | 2202 | WARN_ON_ONCE(!list_empty(&invalid_list)); |
8d5678a7 | 2203 | if (ret > 0) |
3cc736b3 | 2204 | kvm_flush_remote_tlbs(kvm); |
2a74003a | 2205 | } |
e02aa901 | 2206 | |
a30f47cb | 2207 | __clear_sp_write_flooding_count(sp); |
fb58a9c3 | 2208 | |
f3414bc7 | 2209 | goto out; |
7ae680eb | 2210 | } |
47005792 | 2211 | |
94c81364 | 2212 | sp = NULL; |
3cc736b3 | 2213 | ++kvm->stat.mmu_cache_miss; |
47005792 | 2214 | |
94c81364 | 2215 | out: |
3cc736b3 | 2216 | kvm_mmu_commit_zap_page(kvm, &invalid_list); |
94c81364 | 2217 | |
3cc736b3 DM |
2218 | if (collisions > kvm->stat.max_mmu_page_hash_collisions) |
2219 | kvm->stat.max_mmu_page_hash_collisions = collisions; | |
94c81364 DM |
2220 | return sp; |
2221 | } | |
2222 | ||
2f8b1b53 DM |
2223 | /* Caches used when allocating a new shadow page. */ |
2224 | struct shadow_page_caches { | |
2225 | struct kvm_mmu_memory_cache *page_header_cache; | |
2226 | struct kvm_mmu_memory_cache *shadow_page_cache; | |
6a97575d | 2227 | struct kvm_mmu_memory_cache *shadowed_info_cache; |
2f8b1b53 DM |
2228 | }; |
2229 | ||
336081fb | 2230 | static struct kvm_mmu_page *kvm_mmu_alloc_shadow_page(struct kvm *kvm, |
2f8b1b53 | 2231 | struct shadow_page_caches *caches, |
94c81364 DM |
2232 | gfn_t gfn, |
2233 | struct hlist_head *sp_list, | |
2234 | union kvm_mmu_page_role role) | |
2235 | { | |
c306aec8 DM |
2236 | struct kvm_mmu_page *sp; |
2237 | ||
2f8b1b53 DM |
2238 | sp = kvm_mmu_memory_cache_alloc(caches->page_header_cache); |
2239 | sp->spt = kvm_mmu_memory_cache_alloc(caches->shadow_page_cache); | |
c306aec8 | 2240 | if (!role.direct) |
6a97575d | 2241 | sp->shadowed_translation = kvm_mmu_memory_cache_alloc(caches->shadowed_info_cache); |
c306aec8 DM |
2242 | |
2243 | set_page_private(virt_to_page(sp->spt), (unsigned long)sp); | |
2244 | ||
55c510e2 | 2245 | INIT_LIST_HEAD(&sp->possible_nx_huge_page_link); |
428e9216 | 2246 | |
c306aec8 DM |
2247 | /* |
2248 | * active_mmu_pages must be a FIFO list, as kvm_zap_obsolete_pages() | |
2249 | * depends on valid pages being added to the head of the list. See | |
2250 | * comments in kvm_zap_obsolete_pages(). | |
2251 | */ | |
336081fb DM |
2252 | sp->mmu_valid_gen = kvm->arch.mmu_valid_gen; |
2253 | list_add(&sp->link, &kvm->arch.active_mmu_pages); | |
43a063ca | 2254 | kvm_account_mmu_page(kvm, sp); |
47005792 | 2255 | |
4db35314 AK |
2256 | sp->gfn = gfn; |
2257 | sp->role = role; | |
ac101b7c | 2258 | hlist_add_head(&sp->hash_link, sp_list); |
be911771 | 2259 | if (sp_has_gptes(sp)) |
336081fb | 2260 | account_shadowed(kvm, sp); |
ddc16abb | 2261 | |
94c81364 DM |
2262 | return sp; |
2263 | } | |
2264 | ||
cbd858b1 | 2265 | /* Note, @vcpu may be NULL if @role.direct is true; see kvm_mmu_find_shadow_page. */ |
3cc736b3 DM |
2266 | static struct kvm_mmu_page *__kvm_mmu_get_shadow_page(struct kvm *kvm, |
2267 | struct kvm_vcpu *vcpu, | |
2f8b1b53 DM |
2268 | struct shadow_page_caches *caches, |
2269 | gfn_t gfn, | |
2270 | union kvm_mmu_page_role role) | |
94c81364 DM |
2271 | { |
2272 | struct hlist_head *sp_list; | |
2273 | struct kvm_mmu_page *sp; | |
2274 | bool created = false; | |
2275 | ||
3cc736b3 | 2276 | sp_list = &kvm->arch.mmu_page_hash[kvm_page_table_hashfn(gfn)]; |
94c81364 | 2277 | |
3cc736b3 | 2278 | sp = kvm_mmu_find_shadow_page(kvm, vcpu, gfn, sp_list, role); |
94c81364 DM |
2279 | if (!sp) { |
2280 | created = true; | |
3cc736b3 | 2281 | sp = kvm_mmu_alloc_shadow_page(kvm, caches, gfn, sp_list, role); |
94c81364 DM |
2282 | } |
2283 | ||
2284 | trace_kvm_mmu_get_page(sp, created); | |
4db35314 | 2285 | return sp; |
cea0f0e7 AK |
2286 | } |
2287 | ||
2f8b1b53 DM |
2288 | static struct kvm_mmu_page *kvm_mmu_get_shadow_page(struct kvm_vcpu *vcpu, |
2289 | gfn_t gfn, | |
2290 | union kvm_mmu_page_role role) | |
2291 | { | |
2292 | struct shadow_page_caches caches = { | |
2293 | .page_header_cache = &vcpu->arch.mmu_page_header_cache, | |
2294 | .shadow_page_cache = &vcpu->arch.mmu_shadow_page_cache, | |
6a97575d | 2295 | .shadowed_info_cache = &vcpu->arch.mmu_shadowed_info_cache, |
2f8b1b53 DM |
2296 | }; |
2297 | ||
3cc736b3 | 2298 | return __kvm_mmu_get_shadow_page(vcpu->kvm, vcpu, &caches, gfn, role); |
2f8b1b53 DM |
2299 | } |
2300 | ||
39944ab9 SC |
2301 | static union kvm_mmu_page_role kvm_mmu_child_role(u64 *sptep, bool direct, |
2302 | unsigned int access) | |
2e65e842 DM |
2303 | { |
2304 | struct kvm_mmu_page *parent_sp = sptep_to_sp(sptep); | |
2305 | union kvm_mmu_page_role role; | |
2306 | ||
2307 | role = parent_sp->role; | |
2308 | role.level--; | |
2309 | role.access = access; | |
2310 | role.direct = direct; | |
2311 | role.passthrough = 0; | |
2312 | ||
2313 | /* | |
2314 | * If the guest has 4-byte PTEs then that means it's using 32-bit, | |
2315 | * 2-level, non-PAE paging. KVM shadows such guests with PAE paging | |
2316 | * (i.e. 8-byte PTEs). The difference in PTE size means that KVM must | |
2317 | * shadow each guest page table with multiple shadow page tables, which | |
2318 | * requires extra bookkeeping in the role. | |
2319 | * | |
2320 | * Specifically, to shadow the guest's page directory (which covers a | |
2321 | * 4GiB address space), KVM uses 4 PAE page directories, each mapping | |
2322 | * 1GiB of the address space. @role.quadrant encodes which quarter of | |
2323 | * the address space each maps. | |
2324 | * | |
2325 | * To shadow the guest's page tables (which each map a 4MiB region), KVM | |
2326 | * uses 2 PAE page tables, each mapping a 2MiB region. For these, | |
2327 | * @role.quadrant encodes which half of the region they map. | |
2328 | * | |
39944ab9 SC |
2329 | * Concretely, a 4-byte PDE consumes bits 31:22, while an 8-byte PDE |
2330 | * consumes bits 29:21. To consume bits 31:30, KVM's uses 4 shadow | |
2331 | * PDPTEs; those 4 PAE page directories are pre-allocated and their | |
2332 | * quadrant is assigned in mmu_alloc_root(). A 4-byte PTE consumes | |
2333 | * bits 21:12, while an 8-byte PTE consumes bits 20:12. To consume | |
2334 | * bit 21 in the PTE (the child here), KVM propagates that bit to the | |
2335 | * quadrant, i.e. sets quadrant to '0' or '1'. The parent 8-byte PDE | |
2336 | * covers bit 21 (see above), thus the quadrant is calculated from the | |
2337 | * _least_ significant bit of the PDE index. | |
2e65e842 DM |
2338 | */ |
2339 | if (role.has_4_byte_gpte) { | |
2340 | WARN_ON_ONCE(role.level != PG_LEVEL_4K); | |
79e48cec | 2341 | role.quadrant = spte_index(sptep) & 1; |
2e65e842 DM |
2342 | } |
2343 | ||
2344 | return role; | |
2345 | } | |
2346 | ||
2347 | static struct kvm_mmu_page *kvm_mmu_get_child_sp(struct kvm_vcpu *vcpu, | |
2348 | u64 *sptep, gfn_t gfn, | |
2349 | bool direct, unsigned int access) | |
2350 | { | |
2351 | union kvm_mmu_page_role role; | |
2352 | ||
0cd8dc73 PB |
2353 | if (is_shadow_present_pte(*sptep) && !is_large_pte(*sptep)) |
2354 | return ERR_PTR(-EEXIST); | |
2355 | ||
2e65e842 | 2356 | role = kvm_mmu_child_role(sptep, direct, access); |
87654643 | 2357 | return kvm_mmu_get_shadow_page(vcpu, gfn, role); |
2e65e842 DM |
2358 | } |
2359 | ||
7eb77e9f JS |
2360 | static void shadow_walk_init_using_root(struct kvm_shadow_walk_iterator *iterator, |
2361 | struct kvm_vcpu *vcpu, hpa_t root, | |
2362 | u64 addr) | |
2d11123a AK |
2363 | { |
2364 | iterator->addr = addr; | |
7eb77e9f | 2365 | iterator->shadow_addr = root; |
a972e29c | 2366 | iterator->level = vcpu->arch.mmu->root_role.level; |
81407ca5 | 2367 | |
12ec33a7 | 2368 | if (iterator->level >= PT64_ROOT_4LEVEL && |
4d25502a | 2369 | vcpu->arch.mmu->cpu_role.base.level < PT64_ROOT_4LEVEL && |
347a0d0d | 2370 | !vcpu->arch.mmu->root_role.direct) |
12ec33a7 | 2371 | iterator->level = PT32E_ROOT_LEVEL; |
81407ca5 | 2372 | |
2d11123a | 2373 | if (iterator->level == PT32E_ROOT_LEVEL) { |
7eb77e9f JS |
2374 | /* |
2375 | * prev_root is currently only used for 64-bit hosts. So only | |
2376 | * the active root_hpa is valid here. | |
2377 | */ | |
b9e5603c | 2378 | BUG_ON(root != vcpu->arch.mmu->root.hpa); |
7eb77e9f | 2379 | |
2d11123a | 2380 | iterator->shadow_addr |
44dd3ffa | 2381 | = vcpu->arch.mmu->pae_root[(addr >> 30) & 3]; |
2ca3129e | 2382 | iterator->shadow_addr &= SPTE_BASE_ADDR_MASK; |
2d11123a AK |
2383 | --iterator->level; |
2384 | if (!iterator->shadow_addr) | |
2385 | iterator->level = 0; | |
2386 | } | |
2387 | } | |
2388 | ||
7eb77e9f JS |
2389 | static void shadow_walk_init(struct kvm_shadow_walk_iterator *iterator, |
2390 | struct kvm_vcpu *vcpu, u64 addr) | |
2391 | { | |
b9e5603c | 2392 | shadow_walk_init_using_root(iterator, vcpu, vcpu->arch.mmu->root.hpa, |
7eb77e9f JS |
2393 | addr); |
2394 | } | |
2395 | ||
2d11123a AK |
2396 | static bool shadow_walk_okay(struct kvm_shadow_walk_iterator *iterator) |
2397 | { | |
3bae0459 | 2398 | if (iterator->level < PG_LEVEL_4K) |
2d11123a | 2399 | return false; |
4d88954d | 2400 | |
2ca3129e | 2401 | iterator->index = SPTE_INDEX(iterator->addr, iterator->level); |
2d11123a AK |
2402 | iterator->sptep = ((u64 *)__va(iterator->shadow_addr)) + iterator->index; |
2403 | return true; | |
2404 | } | |
2405 | ||
c2a2ac2b XG |
2406 | static void __shadow_walk_next(struct kvm_shadow_walk_iterator *iterator, |
2407 | u64 spte) | |
2d11123a | 2408 | { |
3e44dce4 | 2409 | if (!is_shadow_present_pte(spte) || is_last_spte(spte, iterator->level)) { |
052331be XG |
2410 | iterator->level = 0; |
2411 | return; | |
2412 | } | |
2413 | ||
2ca3129e | 2414 | iterator->shadow_addr = spte & SPTE_BASE_ADDR_MASK; |
2d11123a AK |
2415 | --iterator->level; |
2416 | } | |
2417 | ||
c2a2ac2b XG |
2418 | static void shadow_walk_next(struct kvm_shadow_walk_iterator *iterator) |
2419 | { | |
bb606a9b | 2420 | __shadow_walk_next(iterator, *iterator->sptep); |
c2a2ac2b XG |
2421 | } |
2422 | ||
0cd8dc73 PB |
2423 | static void __link_shadow_page(struct kvm *kvm, |
2424 | struct kvm_mmu_memory_cache *cache, u64 *sptep, | |
03787394 | 2425 | struct kvm_mmu_page *sp, bool flush) |
cc4674d0 BG |
2426 | { |
2427 | u64 spte; | |
2428 | ||
2429 | BUILD_BUG_ON(VMX_EPT_WRITABLE_MASK != PT_WRITABLE_MASK); | |
2430 | ||
0cd8dc73 PB |
2431 | /* |
2432 | * If an SPTE is present already, it must be a leaf and therefore | |
03787394 PB |
2433 | * a large one. Drop it, and flush the TLB if needed, before |
2434 | * installing sp. | |
0cd8dc73 PB |
2435 | */ |
2436 | if (is_shadow_present_pte(*sptep)) | |
03787394 | 2437 | drop_large_spte(kvm, sptep, flush); |
0cd8dc73 | 2438 | |
cc4674d0 BG |
2439 | spte = make_nonleaf_spte(sp->spt, sp_ad_disabled(sp)); |
2440 | ||
1df9f2dc | 2441 | mmu_spte_set(sptep, spte); |
98bba238 | 2442 | |
2ff9039a | 2443 | mmu_page_add_parent_pte(cache, sp, sptep); |
98bba238 | 2444 | |
c4a48868 LJ |
2445 | /* |
2446 | * The non-direct sub-pagetable must be updated before linking. For | |
2447 | * L1 sp, the pagetable is updated via kvm_sync_page() in | |
2448 | * kvm_mmu_find_shadow_page() without write-protecting the gfn, | |
2449 | * so sp->unsync can be true or false. For higher level non-direct | |
2450 | * sp, the pagetable is updated/synced via mmu_sync_children() in | |
2451 | * FNAME(fetch)(), so sp->unsync_children can only be false. | |
2452 | * WARN_ON_ONCE() if anything happens unexpectedly. | |
2453 | */ | |
2454 | if (WARN_ON_ONCE(sp->unsync_children) || sp->unsync) | |
98bba238 | 2455 | mark_unsync(sptep); |
32ef26a3 AK |
2456 | } |
2457 | ||
2ff9039a DM |
2458 | static void link_shadow_page(struct kvm_vcpu *vcpu, u64 *sptep, |
2459 | struct kvm_mmu_page *sp) | |
2460 | { | |
03787394 | 2461 | __link_shadow_page(vcpu->kvm, &vcpu->arch.mmu_pte_list_desc_cache, sptep, sp, true); |
2ff9039a DM |
2462 | } |
2463 | ||
a357bd22 AK |
2464 | static void validate_direct_spte(struct kvm_vcpu *vcpu, u64 *sptep, |
2465 | unsigned direct_access) | |
2466 | { | |
2467 | if (is_shadow_present_pte(*sptep) && !is_large_pte(*sptep)) { | |
2468 | struct kvm_mmu_page *child; | |
2469 | ||
2470 | /* | |
2471 | * For the direct sp, if the guest pte's dirty bit | |
2472 | * changed form clean to dirty, it will corrupt the | |
2473 | * sp's access: allow writable in the read-only sp, | |
2474 | * so we should update the spte at this point to get | |
2475 | * a new sp with the correct access. | |
2476 | */ | |
5e3edd7e | 2477 | child = spte_to_child_sp(*sptep); |
a357bd22 AK |
2478 | if (child->role.access == direct_access) |
2479 | return; | |
2480 | ||
069f30c6 | 2481 | drop_parent_pte(vcpu->kvm, child, sptep); |
3cdf9374 | 2482 | kvm_flush_remote_tlbs_sptep(vcpu->kvm, sptep); |
a357bd22 AK |
2483 | } |
2484 | } | |
2485 | ||
2de4085c BG |
2486 | /* Returns the number of zapped non-leaf child shadow pages. */ |
2487 | static int mmu_page_zap_pte(struct kvm *kvm, struct kvm_mmu_page *sp, | |
2488 | u64 *spte, struct list_head *invalid_list) | |
38e3b2b2 XG |
2489 | { |
2490 | u64 pte; | |
2491 | struct kvm_mmu_page *child; | |
2492 | ||
2493 | pte = *spte; | |
2494 | if (is_shadow_present_pte(pte)) { | |
505aef8f | 2495 | if (is_last_spte(pte, sp->role.level)) { |
c3707958 | 2496 | drop_spte(kvm, spte); |
505aef8f | 2497 | } else { |
5e3edd7e | 2498 | child = spte_to_child_sp(pte); |
069f30c6 | 2499 | drop_parent_pte(kvm, child, spte); |
2de4085c BG |
2500 | |
2501 | /* | |
2502 | * Recursively zap nested TDP SPs, parentless SPs are | |
2503 | * unlikely to be used again in the near future. This | |
2504 | * avoids retaining a large number of stale nested SPs. | |
2505 | */ | |
2506 | if (tdp_enabled && invalid_list && | |
2507 | child->role.guest_mode && !child->parent_ptes.val) | |
2508 | return kvm_mmu_prepare_zap_page(kvm, child, | |
2509 | invalid_list); | |
38e3b2b2 | 2510 | } |
ace569e0 | 2511 | } else if (is_mmio_spte(pte)) { |
ce88decf | 2512 | mmu_spte_clear_no_track(spte); |
ace569e0 | 2513 | } |
2de4085c | 2514 | return 0; |
38e3b2b2 XG |
2515 | } |
2516 | ||
2de4085c BG |
2517 | static int kvm_mmu_page_unlink_children(struct kvm *kvm, |
2518 | struct kvm_mmu_page *sp, | |
2519 | struct list_head *invalid_list) | |
a436036b | 2520 | { |
2de4085c | 2521 | int zapped = 0; |
697fe2e2 | 2522 | unsigned i; |
697fe2e2 | 2523 | |
2ca3129e | 2524 | for (i = 0; i < SPTE_ENT_PER_PAGE; ++i) |
2de4085c BG |
2525 | zapped += mmu_page_zap_pte(kvm, sp, sp->spt + i, invalid_list); |
2526 | ||
2527 | return zapped; | |
a436036b AK |
2528 | } |
2529 | ||
069f30c6 | 2530 | static void kvm_mmu_unlink_parents(struct kvm *kvm, struct kvm_mmu_page *sp) |
a436036b | 2531 | { |
1e3f42f0 TY |
2532 | u64 *sptep; |
2533 | struct rmap_iterator iter; | |
a436036b | 2534 | |
018aabb5 | 2535 | while ((sptep = rmap_get_first(&sp->parent_ptes, &iter))) |
069f30c6 | 2536 | drop_parent_pte(kvm, sp, sptep); |
31aa2b44 AK |
2537 | } |
2538 | ||
60c8aec6 | 2539 | static int mmu_zap_unsync_children(struct kvm *kvm, |
7775834a XG |
2540 | struct kvm_mmu_page *parent, |
2541 | struct list_head *invalid_list) | |
4731d4c7 | 2542 | { |
60c8aec6 MT |
2543 | int i, zapped = 0; |
2544 | struct mmu_page_path parents; | |
2545 | struct kvm_mmu_pages pages; | |
4731d4c7 | 2546 | |
3bae0459 | 2547 | if (parent->role.level == PG_LEVEL_4K) |
4731d4c7 | 2548 | return 0; |
60c8aec6 | 2549 | |
60c8aec6 MT |
2550 | while (mmu_unsync_walk(parent, &pages)) { |
2551 | struct kvm_mmu_page *sp; | |
2552 | ||
2553 | for_each_sp(pages, sp, parents, i) { | |
7775834a | 2554 | kvm_mmu_prepare_zap_page(kvm, sp, invalid_list); |
60c8aec6 | 2555 | mmu_pages_clear_parents(&parents); |
77662e00 | 2556 | zapped++; |
60c8aec6 | 2557 | } |
60c8aec6 MT |
2558 | } |
2559 | ||
2560 | return zapped; | |
4731d4c7 MT |
2561 | } |
2562 | ||
83cdb568 SC |
2563 | static bool __kvm_mmu_prepare_zap_page(struct kvm *kvm, |
2564 | struct kvm_mmu_page *sp, | |
2565 | struct list_head *invalid_list, | |
2566 | int *nr_zapped) | |
31aa2b44 | 2567 | { |
527d5cd7 | 2568 | bool list_unstable, zapped_root = false; |
f691fe1d | 2569 | |
47b0c2e4 | 2570 | lockdep_assert_held_write(&kvm->mmu_lock); |
7775834a | 2571 | trace_kvm_mmu_prepare_zap_page(sp); |
31aa2b44 | 2572 | ++kvm->stat.mmu_shadow_zapped; |
83cdb568 | 2573 | *nr_zapped = mmu_zap_unsync_children(kvm, sp, invalid_list); |
2de4085c | 2574 | *nr_zapped += kvm_mmu_page_unlink_children(kvm, sp, invalid_list); |
069f30c6 | 2575 | kvm_mmu_unlink_parents(kvm, sp); |
5304b8d3 | 2576 | |
83cdb568 SC |
2577 | /* Zapping children means active_mmu_pages has become unstable. */ |
2578 | list_unstable = *nr_zapped; | |
2579 | ||
767d8d8d | 2580 | if (!sp->role.invalid && sp_has_gptes(sp)) |
3ed1a478 | 2581 | unaccount_shadowed(kvm, sp); |
5304b8d3 | 2582 | |
4731d4c7 MT |
2583 | if (sp->unsync) |
2584 | kvm_unlink_unsync_page(kvm, sp); | |
4db35314 | 2585 | if (!sp->root_count) { |
54a4f023 | 2586 | /* Count self */ |
83cdb568 | 2587 | (*nr_zapped)++; |
f95eec9b SC |
2588 | |
2589 | /* | |
2590 | * Already invalid pages (previously active roots) are not on | |
2591 | * the active page list. See list_del() in the "else" case of | |
2592 | * !sp->root_count. | |
2593 | */ | |
2594 | if (sp->role.invalid) | |
2595 | list_add(&sp->link, invalid_list); | |
2596 | else | |
2597 | list_move(&sp->link, invalid_list); | |
43a063ca | 2598 | kvm_unaccount_mmu_page(kvm, sp); |
2e53d63a | 2599 | } else { |
f95eec9b SC |
2600 | /* |
2601 | * Remove the active root from the active page list, the root | |
2602 | * will be explicitly freed when the root_count hits zero. | |
2603 | */ | |
2604 | list_del(&sp->link); | |
05988d72 | 2605 | |
10605204 SC |
2606 | /* |
2607 | * Obsolete pages cannot be used on any vCPUs, see the comment | |
2608 | * in kvm_mmu_zap_all_fast(). Note, is_obsolete_sp() also | |
2609 | * treats invalid shadow pages as being obsolete. | |
2610 | */ | |
527d5cd7 | 2611 | zapped_root = !is_obsolete_sp(kvm, sp); |
2e53d63a | 2612 | } |
7775834a | 2613 | |
55c510e2 SC |
2614 | if (sp->nx_huge_page_disallowed) |
2615 | unaccount_nx_huge_page(kvm, sp); | |
b8e8c830 | 2616 | |
7775834a | 2617 | sp->role.invalid = 1; |
527d5cd7 SC |
2618 | |
2619 | /* | |
2620 | * Make the request to free obsolete roots after marking the root | |
2621 | * invalid, otherwise other vCPUs may not see it as invalid. | |
2622 | */ | |
2623 | if (zapped_root) | |
2624 | kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_FREE_OBSOLETE_ROOTS); | |
83cdb568 SC |
2625 | return list_unstable; |
2626 | } | |
2627 | ||
2628 | static bool kvm_mmu_prepare_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp, | |
2629 | struct list_head *invalid_list) | |
2630 | { | |
2631 | int nr_zapped; | |
2632 | ||
2633 | __kvm_mmu_prepare_zap_page(kvm, sp, invalid_list, &nr_zapped); | |
2634 | return nr_zapped; | |
a436036b AK |
2635 | } |
2636 | ||
7775834a XG |
2637 | static void kvm_mmu_commit_zap_page(struct kvm *kvm, |
2638 | struct list_head *invalid_list) | |
2639 | { | |
945315b9 | 2640 | struct kvm_mmu_page *sp, *nsp; |
7775834a XG |
2641 | |
2642 | if (list_empty(invalid_list)) | |
2643 | return; | |
2644 | ||
c142786c | 2645 | /* |
9753f529 LT |
2646 | * We need to make sure everyone sees our modifications to |
2647 | * the page tables and see changes to vcpu->mode here. The barrier | |
2648 | * in the kvm_flush_remote_tlbs() achieves this. This pairs | |
2649 | * with vcpu_enter_guest and walk_shadow_page_lockless_begin/end. | |
2650 | * | |
2651 | * In addition, kvm_flush_remote_tlbs waits for all vcpus to exit | |
2652 | * guest mode and/or lockless shadow page table walks. | |
c142786c AK |
2653 | */ |
2654 | kvm_flush_remote_tlbs(kvm); | |
c2a2ac2b | 2655 | |
945315b9 | 2656 | list_for_each_entry_safe(sp, nsp, invalid_list, link) { |
20ba462d | 2657 | WARN_ON_ONCE(!sp->role.invalid || sp->root_count); |
87654643 | 2658 | kvm_mmu_free_shadow_page(sp); |
945315b9 | 2659 | } |
7775834a XG |
2660 | } |
2661 | ||
6b82ef2c SC |
2662 | static unsigned long kvm_mmu_zap_oldest_mmu_pages(struct kvm *kvm, |
2663 | unsigned long nr_to_zap) | |
5da59607 | 2664 | { |
6b82ef2c SC |
2665 | unsigned long total_zapped = 0; |
2666 | struct kvm_mmu_page *sp, *tmp; | |
ba7888dd | 2667 | LIST_HEAD(invalid_list); |
6b82ef2c SC |
2668 | bool unstable; |
2669 | int nr_zapped; | |
5da59607 TY |
2670 | |
2671 | if (list_empty(&kvm->arch.active_mmu_pages)) | |
ba7888dd SC |
2672 | return 0; |
2673 | ||
6b82ef2c | 2674 | restart: |
8fc51726 | 2675 | list_for_each_entry_safe_reverse(sp, tmp, &kvm->arch.active_mmu_pages, link) { |
6b82ef2c SC |
2676 | /* |
2677 | * Don't zap active root pages, the page itself can't be freed | |
2678 | * and zapping it will just force vCPUs to realloc and reload. | |
2679 | */ | |
2680 | if (sp->root_count) | |
2681 | continue; | |
2682 | ||
2683 | unstable = __kvm_mmu_prepare_zap_page(kvm, sp, &invalid_list, | |
2684 | &nr_zapped); | |
2685 | total_zapped += nr_zapped; | |
2686 | if (total_zapped >= nr_to_zap) | |
ba7888dd SC |
2687 | break; |
2688 | ||
6b82ef2c SC |
2689 | if (unstable) |
2690 | goto restart; | |
ba7888dd | 2691 | } |
5da59607 | 2692 | |
6b82ef2c SC |
2693 | kvm_mmu_commit_zap_page(kvm, &invalid_list); |
2694 | ||
2695 | kvm->stat.mmu_recycled += total_zapped; | |
2696 | return total_zapped; | |
2697 | } | |
2698 | ||
afe8d7e6 SC |
2699 | static inline unsigned long kvm_mmu_available_pages(struct kvm *kvm) |
2700 | { | |
2701 | if (kvm->arch.n_max_mmu_pages > kvm->arch.n_used_mmu_pages) | |
2702 | return kvm->arch.n_max_mmu_pages - | |
2703 | kvm->arch.n_used_mmu_pages; | |
2704 | ||
2705 | return 0; | |
5da59607 TY |
2706 | } |
2707 | ||
ba7888dd SC |
2708 | static int make_mmu_pages_available(struct kvm_vcpu *vcpu) |
2709 | { | |
6b82ef2c | 2710 | unsigned long avail = kvm_mmu_available_pages(vcpu->kvm); |
ba7888dd | 2711 | |
6b82ef2c | 2712 | if (likely(avail >= KVM_MIN_FREE_MMU_PAGES)) |
ba7888dd SC |
2713 | return 0; |
2714 | ||
6b82ef2c | 2715 | kvm_mmu_zap_oldest_mmu_pages(vcpu->kvm, KVM_REFILL_PAGES - avail); |
ba7888dd | 2716 | |
6e6ec584 SC |
2717 | /* |
2718 | * Note, this check is intentionally soft, it only guarantees that one | |
2719 | * page is available, while the caller may end up allocating as many as | |
2720 | * four pages, e.g. for PAE roots or for 5-level paging. Temporarily | |
2721 | * exceeding the (arbitrary by default) limit will not harm the host, | |
c4342633 | 2722 | * being too aggressive may unnecessarily kill the guest, and getting an |
6e6ec584 SC |
2723 | * exact count is far more trouble than it's worth, especially in the |
2724 | * page fault paths. | |
2725 | */ | |
ba7888dd SC |
2726 | if (!kvm_mmu_available_pages(vcpu->kvm)) |
2727 | return -ENOSPC; | |
2728 | return 0; | |
2729 | } | |
2730 | ||
82ce2c96 IE |
2731 | /* |
2732 | * Changing the number of mmu pages allocated to the vm | |
49d5ca26 | 2733 | * Note: if goal_nr_mmu_pages is too small, you will get dead lock |
82ce2c96 | 2734 | */ |
bc8a3d89 | 2735 | void kvm_mmu_change_mmu_pages(struct kvm *kvm, unsigned long goal_nr_mmu_pages) |
82ce2c96 | 2736 | { |
531810ca | 2737 | write_lock(&kvm->mmu_lock); |
b34cb590 | 2738 | |
49d5ca26 | 2739 | if (kvm->arch.n_used_mmu_pages > goal_nr_mmu_pages) { |
6b82ef2c SC |
2740 | kvm_mmu_zap_oldest_mmu_pages(kvm, kvm->arch.n_used_mmu_pages - |
2741 | goal_nr_mmu_pages); | |
82ce2c96 | 2742 | |
49d5ca26 | 2743 | goal_nr_mmu_pages = kvm->arch.n_used_mmu_pages; |
82ce2c96 | 2744 | } |
82ce2c96 | 2745 | |
49d5ca26 | 2746 | kvm->arch.n_max_mmu_pages = goal_nr_mmu_pages; |
b34cb590 | 2747 | |
531810ca | 2748 | write_unlock(&kvm->mmu_lock); |
82ce2c96 IE |
2749 | } |
2750 | ||
1cb3f3ae | 2751 | int kvm_mmu_unprotect_page(struct kvm *kvm, gfn_t gfn) |
a436036b | 2752 | { |
4db35314 | 2753 | struct kvm_mmu_page *sp; |
d98ba053 | 2754 | LIST_HEAD(invalid_list); |
a436036b AK |
2755 | int r; |
2756 | ||
a436036b | 2757 | r = 0; |
531810ca | 2758 | write_lock(&kvm->mmu_lock); |
767d8d8d | 2759 | for_each_gfn_valid_sp_with_gptes(kvm, sp, gfn) { |
7ae680eb | 2760 | r = 1; |
f41d335a | 2761 | kvm_mmu_prepare_zap_page(kvm, sp, &invalid_list); |
7ae680eb | 2762 | } |
d98ba053 | 2763 | kvm_mmu_commit_zap_page(kvm, &invalid_list); |
531810ca | 2764 | write_unlock(&kvm->mmu_lock); |
1cb3f3ae | 2765 | |
a436036b | 2766 | return r; |
cea0f0e7 | 2767 | } |
96ad91ae SC |
2768 | |
2769 | static int kvm_mmu_unprotect_page_virt(struct kvm_vcpu *vcpu, gva_t gva) | |
2770 | { | |
2771 | gpa_t gpa; | |
2772 | int r; | |
2773 | ||
347a0d0d | 2774 | if (vcpu->arch.mmu->root_role.direct) |
96ad91ae SC |
2775 | return 0; |
2776 | ||
2777 | gpa = kvm_mmu_gva_to_gpa_read(vcpu, gva, NULL); | |
2778 | ||
2779 | r = kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT); | |
2780 | ||
2781 | return r; | |
2782 | } | |
cea0f0e7 | 2783 | |
4d78d0b3 | 2784 | static void kvm_unsync_page(struct kvm *kvm, struct kvm_mmu_page *sp) |
9cf5cf5a XG |
2785 | { |
2786 | trace_kvm_mmu_unsync_page(sp); | |
4d78d0b3 | 2787 | ++kvm->stat.mmu_unsync; |
9cf5cf5a XG |
2788 | sp->unsync = 1; |
2789 | ||
2790 | kvm_mmu_mark_parents_unsync(sp); | |
9cf5cf5a XG |
2791 | } |
2792 | ||
0337f585 SC |
2793 | /* |
2794 | * Attempt to unsync any shadow pages that can be reached by the specified gfn, | |
2795 | * KVM is creating a writable mapping for said gfn. Returns 0 if all pages | |
2796 | * were marked unsync (or if there is no shadow page), -EPERM if the SPTE must | |
2797 | * be write-protected. | |
2798 | */ | |
8283e36a | 2799 | int mmu_try_to_unsync_pages(struct kvm *kvm, const struct kvm_memory_slot *slot, |
2839180c | 2800 | gfn_t gfn, bool can_unsync, bool prefetch) |
4731d4c7 | 2801 | { |
5c520e90 | 2802 | struct kvm_mmu_page *sp; |
ce25681d | 2803 | bool locked = false; |
4731d4c7 | 2804 | |
0337f585 SC |
2805 | /* |
2806 | * Force write-protection if the page is being tracked. Note, the page | |
2807 | * track machinery is used to write-protect upper-level shadow pages, | |
2808 | * i.e. this guards the role.level == 4K assertion below! | |
2809 | */ | |
7b574863 | 2810 | if (kvm_gfn_is_write_tracked(kvm, slot, gfn)) |
0337f585 | 2811 | return -EPERM; |
9cf5cf5a | 2812 | |
0337f585 SC |
2813 | /* |
2814 | * The page is not write-tracked, mark existing shadow pages unsync | |
2815 | * unless KVM is synchronizing an unsync SP (can_unsync = false). In | |
2816 | * that case, KVM must complete emulation of the guest TLB flush before | |
2817 | * allowing shadow pages to become unsync (writable by the guest). | |
2818 | */ | |
767d8d8d | 2819 | for_each_gfn_valid_sp_with_gptes(kvm, sp, gfn) { |
36a2e677 | 2820 | if (!can_unsync) |
0337f585 | 2821 | return -EPERM; |
36a2e677 | 2822 | |
5c520e90 XG |
2823 | if (sp->unsync) |
2824 | continue; | |
9cf5cf5a | 2825 | |
2839180c | 2826 | if (prefetch) |
f1c4a88c LJ |
2827 | return -EEXIST; |
2828 | ||
ce25681d SC |
2829 | /* |
2830 | * TDP MMU page faults require an additional spinlock as they | |
2831 | * run with mmu_lock held for read, not write, and the unsync | |
2832 | * logic is not thread safe. Take the spinklock regardless of | |
2833 | * the MMU type to avoid extra conditionals/parameters, there's | |
2834 | * no meaningful penalty if mmu_lock is held for write. | |
2835 | */ | |
2836 | if (!locked) { | |
2837 | locked = true; | |
4d78d0b3 | 2838 | spin_lock(&kvm->arch.mmu_unsync_pages_lock); |
ce25681d SC |
2839 | |
2840 | /* | |
2841 | * Recheck after taking the spinlock, a different vCPU | |
2842 | * may have since marked the page unsync. A false | |
2843 | * positive on the unprotected check above is not | |
2844 | * possible as clearing sp->unsync _must_ hold mmu_lock | |
2845 | * for write, i.e. unsync cannot transition from 0->1 | |
2846 | * while this CPU holds mmu_lock for read (or write). | |
2847 | */ | |
2848 | if (READ_ONCE(sp->unsync)) | |
2849 | continue; | |
2850 | } | |
2851 | ||
20ba462d | 2852 | WARN_ON_ONCE(sp->role.level != PG_LEVEL_4K); |
4d78d0b3 | 2853 | kvm_unsync_page(kvm, sp); |
4731d4c7 | 2854 | } |
ce25681d | 2855 | if (locked) |
4d78d0b3 | 2856 | spin_unlock(&kvm->arch.mmu_unsync_pages_lock); |
3d0c27ad | 2857 | |
578e1c4d JS |
2858 | /* |
2859 | * We need to ensure that the marking of unsync pages is visible | |
2860 | * before the SPTE is updated to allow writes because | |
2861 | * kvm_mmu_sync_roots() checks the unsync flags without holding | |
2862 | * the MMU lock and so can race with this. If the SPTE was updated | |
2863 | * before the page had been marked as unsync-ed, something like the | |
2864 | * following could happen: | |
2865 | * | |
2866 | * CPU 1 CPU 2 | |
2867 | * --------------------------------------------------------------------- | |
2868 | * 1.2 Host updates SPTE | |
2869 | * to be writable | |
2870 | * 2.1 Guest writes a GPTE for GVA X. | |
2871 | * (GPTE being in the guest page table shadowed | |
2872 | * by the SP from CPU 1.) | |
2873 | * This reads SPTE during the page table walk. | |
2874 | * Since SPTE.W is read as 1, there is no | |
2875 | * fault. | |
2876 | * | |
2877 | * 2.2 Guest issues TLB flush. | |
2878 | * That causes a VM Exit. | |
2879 | * | |
0337f585 SC |
2880 | * 2.3 Walking of unsync pages sees sp->unsync is |
2881 | * false and skips the page. | |
578e1c4d JS |
2882 | * |
2883 | * 2.4 Guest accesses GVA X. | |
2884 | * Since the mapping in the SP was not updated, | |
2885 | * so the old mapping for GVA X incorrectly | |
2886 | * gets used. | |
2887 | * 1.1 Host marks SP | |
2888 | * as unsync | |
2889 | * (sp->unsync = true) | |
2890 | * | |
2891 | * The write barrier below ensures that 1.1 happens before 1.2 and thus | |
264d3dc1 LJ |
2892 | * the situation in 2.4 does not arise. It pairs with the read barrier |
2893 | * in is_unsync_root(), placed between 2.1's load of SPTE.W and 2.3. | |
578e1c4d JS |
2894 | */ |
2895 | smp_wmb(); | |
2896 | ||
0337f585 | 2897 | return 0; |
4731d4c7 MT |
2898 | } |
2899 | ||
8a9f566a DM |
2900 | static int mmu_set_spte(struct kvm_vcpu *vcpu, struct kvm_memory_slot *slot, |
2901 | u64 *sptep, unsigned int pte_access, gfn_t gfn, | |
a12f4381 | 2902 | kvm_pfn_t pfn, struct kvm_page_fault *fault) |
1e73f9dd | 2903 | { |
d786c778 | 2904 | struct kvm_mmu_page *sp = sptep_to_sp(sptep); |
eb5cd7ff | 2905 | int level = sp->role.level; |
1e73f9dd | 2906 | int was_rmapped = 0; |
c4371c2a | 2907 | int ret = RET_PF_FIXED; |
c2a4eadf | 2908 | bool flush = false; |
ad67e480 | 2909 | bool wrprot; |
d786c778 | 2910 | u64 spte; |
1e73f9dd | 2911 | |
a12f4381 PB |
2912 | /* Prefetching always gets a writable pfn. */ |
2913 | bool host_writable = !fault || fault->map_writable; | |
2839180c | 2914 | bool prefetch = !fault || fault->prefetch; |
a12f4381 | 2915 | bool write_fault = fault && fault->write; |
1e73f9dd | 2916 | |
a54aa15c | 2917 | if (unlikely(is_noslot_pfn(pfn))) { |
1075d41e | 2918 | vcpu->stat.pf_mmio_spte_created++; |
a54aa15c SC |
2919 | mark_mmio_spte(vcpu, sptep, gfn, pte_access); |
2920 | return RET_PF_EMULATE; | |
2921 | } | |
2922 | ||
afd28fe1 | 2923 | if (is_shadow_present_pte(*sptep)) { |
1e73f9dd MT |
2924 | /* |
2925 | * If we overwrite a PTE page pointer with a 2MB PMD, unlink | |
2926 | * the parent of the now unreachable PTE. | |
2927 | */ | |
3bae0459 | 2928 | if (level > PG_LEVEL_4K && !is_large_pte(*sptep)) { |
1e73f9dd | 2929 | struct kvm_mmu_page *child; |
d555c333 | 2930 | u64 pte = *sptep; |
1e73f9dd | 2931 | |
5e3edd7e | 2932 | child = spte_to_child_sp(pte); |
069f30c6 | 2933 | drop_parent_pte(vcpu->kvm, child, sptep); |
c2a4eadf | 2934 | flush = true; |
d555c333 | 2935 | } else if (pfn != spte_to_pfn(*sptep)) { |
c3707958 | 2936 | drop_spte(vcpu->kvm, sptep); |
c2a4eadf | 2937 | flush = true; |
6bed6b9e JR |
2938 | } else |
2939 | was_rmapped = 1; | |
1e73f9dd | 2940 | } |
852e3c19 | 2941 | |
2839180c | 2942 | wrprot = make_spte(vcpu, sp, slot, pte_access, gfn, pfn, *sptep, prefetch, |
7158bee4 | 2943 | true, host_writable, &spte); |
d786c778 PB |
2944 | |
2945 | if (*sptep == spte) { | |
2946 | ret = RET_PF_SPURIOUS; | |
2947 | } else { | |
d786c778 | 2948 | flush |= mmu_spte_update(sptep, spte); |
5959ff4a | 2949 | trace_kvm_mmu_set_spte(level, gfn, sptep); |
d786c778 PB |
2950 | } |
2951 | ||
ad67e480 | 2952 | if (wrprot) { |
1e73f9dd | 2953 | if (write_fault) |
9b8ebbdb | 2954 | ret = RET_PF_EMULATE; |
a378b4e6 | 2955 | } |
c3134ce2 | 2956 | |
d786c778 | 2957 | if (flush) |
4ad980ae | 2958 | kvm_flush_remote_tlbs_gfn(vcpu->kvm, gfn, level); |
1e73f9dd | 2959 | |
4293ddb7 | 2960 | if (!was_rmapped) { |
d786c778 | 2961 | WARN_ON_ONCE(ret == RET_PF_SPURIOUS); |
6a97575d DM |
2962 | rmap_add(vcpu, slot, sptep, gfn, pte_access); |
2963 | } else { | |
2964 | /* Already rmapped but the pte_access bits may have changed. */ | |
79e48cec | 2965 | kvm_mmu_page_set_access(sp, spte_index(sptep), pte_access); |
1c4f1fd6 | 2966 | } |
cb9aaa30 | 2967 | |
9b8ebbdb | 2968 | return ret; |
1c4f1fd6 AK |
2969 | } |
2970 | ||
957ed9ef XG |
2971 | static int direct_pte_prefetch_many(struct kvm_vcpu *vcpu, |
2972 | struct kvm_mmu_page *sp, | |
2973 | u64 *start, u64 *end) | |
2974 | { | |
2975 | struct page *pages[PTE_PREFETCH_NUM]; | |
d9ef13c2 | 2976 | struct kvm_memory_slot *slot; |
0a2b64c5 | 2977 | unsigned int access = sp->role.access; |
957ed9ef XG |
2978 | int i, ret; |
2979 | gfn_t gfn; | |
2980 | ||
79e48cec | 2981 | gfn = kvm_mmu_page_get_gfn(sp, spte_index(start)); |
d9ef13c2 PB |
2982 | slot = gfn_to_memslot_dirty_bitmap(vcpu, gfn, access & ACC_WRITE_MASK); |
2983 | if (!slot) | |
957ed9ef XG |
2984 | return -1; |
2985 | ||
d9ef13c2 | 2986 | ret = gfn_to_page_many_atomic(slot, gfn, pages, end - start); |
957ed9ef XG |
2987 | if (ret <= 0) |
2988 | return -1; | |
2989 | ||
43fdcda9 | 2990 | for (i = 0; i < ret; i++, gfn++, start++) { |
8a9f566a | 2991 | mmu_set_spte(vcpu, slot, start, access, gfn, |
a12f4381 | 2992 | page_to_pfn(pages[i]), NULL); |
43fdcda9 JS |
2993 | put_page(pages[i]); |
2994 | } | |
957ed9ef XG |
2995 | |
2996 | return 0; | |
2997 | } | |
2998 | ||
2999 | static void __direct_pte_prefetch(struct kvm_vcpu *vcpu, | |
3000 | struct kvm_mmu_page *sp, u64 *sptep) | |
3001 | { | |
3002 | u64 *spte, *start = NULL; | |
3003 | int i; | |
3004 | ||
20ba462d | 3005 | WARN_ON_ONCE(!sp->role.direct); |
957ed9ef | 3006 | |
79e48cec | 3007 | i = spte_index(sptep) & ~(PTE_PREFETCH_NUM - 1); |
957ed9ef XG |
3008 | spte = sp->spt + i; |
3009 | ||
3010 | for (i = 0; i < PTE_PREFETCH_NUM; i++, spte++) { | |
c3707958 | 3011 | if (is_shadow_present_pte(*spte) || spte == sptep) { |
957ed9ef XG |
3012 | if (!start) |
3013 | continue; | |
3014 | if (direct_pte_prefetch_many(vcpu, sp, start, spte) < 0) | |
c6cecc4b | 3015 | return; |
957ed9ef XG |
3016 | start = NULL; |
3017 | } else if (!start) | |
3018 | start = spte; | |
3019 | } | |
c6cecc4b SC |
3020 | if (start) |
3021 | direct_pte_prefetch_many(vcpu, sp, start, spte); | |
957ed9ef XG |
3022 | } |
3023 | ||
3024 | static void direct_pte_prefetch(struct kvm_vcpu *vcpu, u64 *sptep) | |
3025 | { | |
3026 | struct kvm_mmu_page *sp; | |
3027 | ||
57354682 | 3028 | sp = sptep_to_sp(sptep); |
ac8d57e5 | 3029 | |
957ed9ef | 3030 | /* |
ac8d57e5 PF |
3031 | * Without accessed bits, there's no way to distinguish between |
3032 | * actually accessed translations and prefetched, so disable pte | |
3033 | * prefetch if accessed bits aren't available. | |
957ed9ef | 3034 | */ |
ac8d57e5 | 3035 | if (sp_ad_disabled(sp)) |
957ed9ef XG |
3036 | return; |
3037 | ||
3bae0459 | 3038 | if (sp->role.level > PG_LEVEL_4K) |
957ed9ef XG |
3039 | return; |
3040 | ||
4a42d848 DS |
3041 | /* |
3042 | * If addresses are being invalidated, skip prefetching to avoid | |
3043 | * accidentally prefetching those addresses. | |
3044 | */ | |
20ec3ebd | 3045 | if (unlikely(vcpu->kvm->mmu_invalidate_in_progress)) |
4a42d848 DS |
3046 | return; |
3047 | ||
957ed9ef XG |
3048 | __direct_pte_prefetch(vcpu, sp, sptep); |
3049 | } | |
3050 | ||
65e3b446 SC |
3051 | /* |
3052 | * Lookup the mapping level for @gfn in the current mm. | |
3053 | * | |
3054 | * WARNING! Use of host_pfn_mapping_level() requires the caller and the end | |
3055 | * consumer to be tied into KVM's handlers for MMU notifier events! | |
3056 | * | |
3057 | * There are several ways to safely use this helper: | |
3058 | * | |
20ec3ebd | 3059 | * - Check mmu_invalidate_retry_hva() after grabbing the mapping level, before |
65e3b446 SC |
3060 | * consuming it. In this case, mmu_lock doesn't need to be held during the |
3061 | * lookup, but it does need to be held while checking the MMU notifier. | |
3062 | * | |
3063 | * - Hold mmu_lock AND ensure there is no in-progress MMU notifier invalidation | |
3064 | * event for the hva. This can be done by explicit checking the MMU notifier | |
3065 | * or by ensuring that KVM already has a valid mapping that covers the hva. | |
3066 | * | |
3067 | * - Do not use the result to install new mappings, e.g. use the host mapping | |
3068 | * level only to decide whether or not to zap an entry. In this case, it's | |
3069 | * not required to hold mmu_lock (though it's highly likely the caller will | |
3070 | * want to hold mmu_lock anyways, e.g. to modify SPTEs). | |
3071 | * | |
3072 | * Note! The lookup can still race with modifications to host page tables, but | |
3073 | * the above "rules" ensure KVM will not _consume_ the result of the walk if a | |
3074 | * race with the primary MMU occurs. | |
3075 | */ | |
a8ac499b | 3076 | static int host_pfn_mapping_level(struct kvm *kvm, gfn_t gfn, |
8ca6f063 | 3077 | const struct kvm_memory_slot *slot) |
db543216 | 3078 | { |
284dc493 | 3079 | int level = PG_LEVEL_4K; |
db543216 | 3080 | unsigned long hva; |
44187235 | 3081 | unsigned long flags; |
44187235 MZ |
3082 | pgd_t pgd; |
3083 | p4d_t p4d; | |
3084 | pud_t pud; | |
3085 | pmd_t pmd; | |
db543216 | 3086 | |
293e306e SC |
3087 | /* |
3088 | * Note, using the already-retrieved memslot and __gfn_to_hva_memslot() | |
3089 | * is not solely for performance, it's also necessary to avoid the | |
3090 | * "writable" check in __gfn_to_hva_many(), which will always fail on | |
3091 | * read-only memslots due to gfn_to_hva() assuming writes. Earlier | |
3092 | * page fault steps have already verified the guest isn't writing a | |
3093 | * read-only memslot. | |
3094 | */ | |
db543216 SC |
3095 | hva = __gfn_to_hva_memslot(slot, gfn); |
3096 | ||
44187235 | 3097 | /* |
65e3b446 SC |
3098 | * Disable IRQs to prevent concurrent tear down of host page tables, |
3099 | * e.g. if the primary MMU promotes a P*D to a huge page and then frees | |
3100 | * the original page table. | |
44187235 MZ |
3101 | */ |
3102 | local_irq_save(flags); | |
3103 | ||
65e3b446 SC |
3104 | /* |
3105 | * Read each entry once. As above, a non-leaf entry can be promoted to | |
3106 | * a huge page _during_ this walk. Re-reading the entry could send the | |
3107 | * walk into the weeks, e.g. p*d_large() returns false (sees the old | |
3108 | * value) and then p*d_offset() walks into the target huge page instead | |
3109 | * of the old page table (sees the new value). | |
3110 | */ | |
44187235 MZ |
3111 | pgd = READ_ONCE(*pgd_offset(kvm->mm, hva)); |
3112 | if (pgd_none(pgd)) | |
3113 | goto out; | |
3114 | ||
3115 | p4d = READ_ONCE(*p4d_offset(&pgd, hva)); | |
3116 | if (p4d_none(p4d) || !p4d_present(p4d)) | |
3117 | goto out; | |
db543216 | 3118 | |
44187235 MZ |
3119 | pud = READ_ONCE(*pud_offset(&p4d, hva)); |
3120 | if (pud_none(pud) || !pud_present(pud)) | |
3121 | goto out; | |
3122 | ||
3123 | if (pud_large(pud)) { | |
3124 | level = PG_LEVEL_1G; | |
3125 | goto out; | |
3126 | } | |
3127 | ||
3128 | pmd = READ_ONCE(*pmd_offset(&pud, hva)); | |
3129 | if (pmd_none(pmd) || !pmd_present(pmd)) | |
3130 | goto out; | |
3131 | ||
3132 | if (pmd_large(pmd)) | |
3133 | level = PG_LEVEL_2M; | |
3134 | ||
3135 | out: | |
3136 | local_irq_restore(flags); | |
db543216 SC |
3137 | return level; |
3138 | } | |
3139 | ||
8ca6f063 BG |
3140 | int kvm_mmu_max_mapping_level(struct kvm *kvm, |
3141 | const struct kvm_memory_slot *slot, gfn_t gfn, | |
a8ac499b | 3142 | int max_level) |
1b6d9d9e SC |
3143 | { |
3144 | struct kvm_lpage_info *linfo; | |
ec607a56 | 3145 | int host_level; |
1b6d9d9e SC |
3146 | |
3147 | max_level = min(max_level, max_huge_page_level); | |
3148 | for ( ; max_level > PG_LEVEL_4K; max_level--) { | |
3149 | linfo = lpage_info_slot(gfn, slot, max_level); | |
3150 | if (!linfo->disallow_lpage) | |
3151 | break; | |
3152 | } | |
3153 | ||
3154 | if (max_level == PG_LEVEL_4K) | |
3155 | return PG_LEVEL_4K; | |
3156 | ||
a8ac499b | 3157 | host_level = host_pfn_mapping_level(kvm, gfn, slot); |
ec607a56 | 3158 | return min(host_level, max_level); |
1b6d9d9e SC |
3159 | } |
3160 | ||
73a3c659 | 3161 | void kvm_mmu_hugepage_adjust(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault) |
0885904d | 3162 | { |
e710c5f6 | 3163 | struct kvm_memory_slot *slot = fault->slot; |
17eff019 SC |
3164 | kvm_pfn_t mask; |
3165 | ||
73a3c659 | 3166 | fault->huge_page_disallowed = fault->exec && fault->nx_huge_page_workaround_enabled; |
3cf06612 | 3167 | |
73a3c659 PB |
3168 | if (unlikely(fault->max_level == PG_LEVEL_4K)) |
3169 | return; | |
17eff019 | 3170 | |
5d49f08c | 3171 | if (is_error_noslot_pfn(fault->pfn)) |
73a3c659 | 3172 | return; |
17eff019 | 3173 | |
e710c5f6 | 3174 | if (kvm_slot_dirty_track_enabled(slot)) |
73a3c659 | 3175 | return; |
293e306e | 3176 | |
3cf06612 SC |
3177 | /* |
3178 | * Enforce the iTLB multihit workaround after capturing the requested | |
3179 | * level, which will be used to do precise, accurate accounting. | |
3180 | */ | |
73a3c659 | 3181 | fault->req_level = kvm_mmu_max_mapping_level(vcpu->kvm, slot, |
a8ac499b | 3182 | fault->gfn, fault->max_level); |
73a3c659 PB |
3183 | if (fault->req_level == PG_LEVEL_4K || fault->huge_page_disallowed) |
3184 | return; | |
0885904d SC |
3185 | |
3186 | /* | |
20ec3ebd | 3187 | * mmu_invalidate_retry() was successful and mmu_lock is held, so |
17eff019 | 3188 | * the pmd can't be split from under us. |
0885904d | 3189 | */ |
73a3c659 PB |
3190 | fault->goal_level = fault->req_level; |
3191 | mask = KVM_PAGES_PER_HPAGE(fault->goal_level) - 1; | |
3192 | VM_BUG_ON((fault->gfn & mask) != (fault->pfn & mask)); | |
3193 | fault->pfn &= ~mask; | |
0885904d SC |
3194 | } |
3195 | ||
536f0e6a | 3196 | void disallowed_hugepage_adjust(struct kvm_page_fault *fault, u64 spte, int cur_level) |
b8e8c830 | 3197 | { |
536f0e6a PB |
3198 | if (cur_level > PG_LEVEL_4K && |
3199 | cur_level == fault->goal_level && | |
b8e8c830 | 3200 | is_shadow_present_pte(spte) && |
76901e56 MZ |
3201 | !is_large_pte(spte) && |
3202 | spte_to_child_sp(spte)->nx_huge_page_disallowed) { | |
b8e8c830 | 3203 | /* |
6c882ef4 DM |
3204 | * A small SPTE exists for this pfn, but FNAME(fetch), |
3205 | * direct_map(), or kvm_tdp_mmu_map() would like to create a | |
3206 | * large PTE instead: just force them to go down another level, | |
3207 | * patching back for them into pfn the next 9 bits of the | |
3208 | * address. | |
b8e8c830 | 3209 | */ |
536f0e6a PB |
3210 | u64 page_mask = KVM_PAGES_PER_HPAGE(cur_level) - |
3211 | KVM_PAGES_PER_HPAGE(cur_level - 1); | |
3212 | fault->pfn |= fault->gfn & page_mask; | |
3213 | fault->goal_level--; | |
b8e8c830 PB |
3214 | } |
3215 | } | |
3216 | ||
6c882ef4 | 3217 | static int direct_map(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault) |
140754bc | 3218 | { |
3fcf2d1b | 3219 | struct kvm_shadow_walk_iterator it; |
140754bc | 3220 | struct kvm_mmu_page *sp; |
73a3c659 | 3221 | int ret; |
43b74355 | 3222 | gfn_t base_gfn = fault->gfn; |
6aa8b732 | 3223 | |
73a3c659 | 3224 | kvm_mmu_hugepage_adjust(vcpu, fault); |
4cd071d1 | 3225 | |
f0066d94 | 3226 | trace_kvm_mmu_spte_requested(fault); |
43b74355 | 3227 | for_each_shadow_entry(vcpu, fault->addr, it) { |
b8e8c830 PB |
3228 | /* |
3229 | * We cannot overwrite existing page tables with an NX | |
3230 | * large page, as the leaf could be executable. | |
3231 | */ | |
73a3c659 | 3232 | if (fault->nx_huge_page_workaround_enabled) |
536f0e6a | 3233 | disallowed_hugepage_adjust(fault, *it.sptep, it.level); |
b8e8c830 | 3234 | |
c667a3ba | 3235 | base_gfn = gfn_round_for_level(fault->gfn, it.level); |
73a3c659 | 3236 | if (it.level == fault->goal_level) |
9f652d21 | 3237 | break; |
6aa8b732 | 3238 | |
2e65e842 | 3239 | sp = kvm_mmu_get_child_sp(vcpu, it.sptep, base_gfn, true, ACC_ALL); |
0cd8dc73 PB |
3240 | if (sp == ERR_PTR(-EEXIST)) |
3241 | continue; | |
03fffc54 SC |
3242 | |
3243 | link_shadow_page(vcpu, it.sptep, sp); | |
b5b0977f | 3244 | if (fault->huge_page_disallowed) |
55c510e2 | 3245 | account_nx_huge_page(vcpu->kvm, sp, |
428e9216 | 3246 | fault->req_level >= it.level); |
9f652d21 | 3247 | } |
3fcf2d1b | 3248 | |
b1a429fb SC |
3249 | if (WARN_ON_ONCE(it.level != fault->goal_level)) |
3250 | return -EFAULT; | |
3251 | ||
8a9f566a | 3252 | ret = mmu_set_spte(vcpu, fault->slot, it.sptep, ACC_ALL, |
a12f4381 | 3253 | base_gfn, fault->pfn, fault); |
12703759 SC |
3254 | if (ret == RET_PF_SPURIOUS) |
3255 | return ret; | |
3256 | ||
3fcf2d1b | 3257 | direct_pte_prefetch(vcpu, it.sptep); |
3fcf2d1b | 3258 | return ret; |
6aa8b732 AK |
3259 | } |
3260 | ||
cd08d178 | 3261 | static void kvm_send_hwpoison_signal(struct kvm_memory_slot *slot, gfn_t gfn) |
bf998156 | 3262 | { |
cd08d178 DM |
3263 | unsigned long hva = gfn_to_hva_memslot(slot, gfn); |
3264 | ||
3265 | send_sig_mceerr(BUS_MCEERR_AR, (void __user *)hva, PAGE_SHIFT, current); | |
bf998156 HY |
3266 | } |
3267 | ||
cd08d178 | 3268 | static int kvm_handle_error_pfn(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault) |
bf998156 | 3269 | { |
cd08d178 | 3270 | if (is_sigpending_pfn(fault->pfn)) { |
76657687 PX |
3271 | kvm_handle_signal_exit(vcpu); |
3272 | return -EINTR; | |
3273 | } | |
3274 | ||
4d8b81ab XG |
3275 | /* |
3276 | * Do not cache the mmio info caused by writing the readonly gfn | |
3277 | * into the spte otherwise read access on readonly gfn also can | |
3278 | * caused mmio page fault and treat it as mmio access. | |
4d8b81ab | 3279 | */ |
cd08d178 | 3280 | if (fault->pfn == KVM_PFN_ERR_RO_FAULT) |
9b8ebbdb | 3281 | return RET_PF_EMULATE; |
4d8b81ab | 3282 | |
cd08d178 DM |
3283 | if (fault->pfn == KVM_PFN_ERR_HWPOISON) { |
3284 | kvm_send_hwpoison_signal(fault->slot, fault->gfn); | |
9b8ebbdb | 3285 | return RET_PF_RETRY; |
d7c55201 | 3286 | } |
edba23e5 | 3287 | |
2c151b25 | 3288 | return -EFAULT; |
bf998156 HY |
3289 | } |
3290 | ||
354c908c DM |
3291 | static int kvm_handle_noslot_fault(struct kvm_vcpu *vcpu, |
3292 | struct kvm_page_fault *fault, | |
3293 | unsigned int access) | |
d7c55201 | 3294 | { |
354c908c | 3295 | gva_t gva = fault->is_tdp ? 0 : fault->addr; |
d7c55201 | 3296 | |
354c908c DM |
3297 | vcpu_cache_mmio_info(vcpu, gva, fault->gfn, |
3298 | access & shadow_mmio_access_mask); | |
3a13f4fe | 3299 | |
354c908c DM |
3300 | /* |
3301 | * If MMIO caching is disabled, emulate immediately without | |
3302 | * touching the shadow page tables as attempting to install an | |
3303 | * MMIO SPTE will just be an expensive nop. | |
3304 | */ | |
3305 | if (unlikely(!enable_mmio_caching)) | |
3306 | return RET_PF_EMULATE; | |
3307 | ||
3308 | /* | |
3309 | * Do not create an MMIO SPTE for a gfn greater than host.MAXPHYADDR, | |
3310 | * any guest that generates such gfns is running nested and is being | |
3311 | * tricked by L0 userspace (you can observe gfn > L1.MAXPHYADDR if and | |
3312 | * only if L1's MAXPHYADDR is inaccurate with respect to the | |
3313 | * hardware's). | |
3314 | */ | |
3315 | if (unlikely(fault->gfn > kvm_mmu_max_gfn())) | |
3316 | return RET_PF_EMULATE; | |
d7c55201 | 3317 | |
5276c616 | 3318 | return RET_PF_CONTINUE; |
d7c55201 XG |
3319 | } |
3320 | ||
3c8ad5a6 | 3321 | static bool page_fault_can_be_fast(struct kvm_page_fault *fault) |
c7ba5b48 | 3322 | { |
1c118b82 | 3323 | /* |
5c64aba5 SC |
3324 | * Page faults with reserved bits set, i.e. faults on MMIO SPTEs, only |
3325 | * reach the common page fault handler if the SPTE has an invalid MMIO | |
3326 | * generation number. Refreshing the MMIO generation needs to go down | |
3327 | * the slow path. Note, EPT Misconfigs do NOT set the PRESENT flag! | |
1c118b82 | 3328 | */ |
3c8ad5a6 | 3329 | if (fault->rsvd) |
1c118b82 XG |
3330 | return false; |
3331 | ||
c7ba5b48 | 3332 | /* |
f160c7b7 | 3333 | * #PF can be fast if: |
f160c7b7 | 3334 | * |
54275f74 SC |
3335 | * 1. The shadow page table entry is not present and A/D bits are |
3336 | * disabled _by KVM_, which could mean that the fault is potentially | |
3337 | * caused by access tracking (if enabled). If A/D bits are enabled | |
3338 | * by KVM, but disabled by L1 for L2, KVM is forced to disable A/D | |
3339 | * bits for L2 and employ access tracking, but the fast page fault | |
3340 | * mechanism only supports direct MMUs. | |
3341 | * 2. The shadow page table entry is present, the access is a write, | |
3342 | * and no reserved bits are set (MMIO SPTEs cannot be "fixed"), i.e. | |
3343 | * the fault was caused by a write-protection violation. If the | |
3344 | * SPTE is MMU-writable (determined later), the fault can be fixed | |
3345 | * by setting the Writable bit, which can be done out of mmu_lock. | |
c7ba5b48 | 3346 | */ |
5c64aba5 SC |
3347 | if (!fault->present) |
3348 | return !kvm_ad_enabled(); | |
3349 | ||
3350 | /* | |
3351 | * Note, instruction fetches and writes are mutually exclusive, ignore | |
3352 | * the "exec" flag. | |
3353 | */ | |
3354 | return fault->write; | |
c7ba5b48 XG |
3355 | } |
3356 | ||
97dceba2 JS |
3357 | /* |
3358 | * Returns true if the SPTE was fixed successfully. Otherwise, | |
3359 | * someone else modified the SPTE from its original value. | |
3360 | */ | |
f3d90f90 SC |
3361 | static bool fast_pf_fix_direct_spte(struct kvm_vcpu *vcpu, |
3362 | struct kvm_page_fault *fault, | |
3363 | u64 *sptep, u64 old_spte, u64 new_spte) | |
c7ba5b48 | 3364 | { |
9b51a630 KH |
3365 | /* |
3366 | * Theoretically we could also set dirty bit (and flush TLB) here in | |
3367 | * order to eliminate unnecessary PML logging. See comments in | |
3368 | * set_spte. But fast_page_fault is very unlikely to happen with PML | |
3369 | * enabled, so we do not do this. This might result in the same GPA | |
3370 | * to be logged in PML buffer again when the write really happens, and | |
3371 | * eventually to be called by mark_page_dirty twice. But it's also no | |
3372 | * harm. This also avoids the TLB flush needed after setting dirty bit | |
3373 | * so non-PML cases won't be impacted. | |
3374 | * | |
3375 | * Compare with set_spte where instead shadow_dirty_mask is set. | |
3376 | */ | |
2db2f46f | 3377 | if (!try_cmpxchg64(sptep, &old_spte, new_spte)) |
97dceba2 JS |
3378 | return false; |
3379 | ||
e710c5f6 DM |
3380 | if (is_writable_pte(new_spte) && !is_writable_pte(old_spte)) |
3381 | mark_page_dirty_in_slot(vcpu->kvm, fault->slot, fault->gfn); | |
c7ba5b48 XG |
3382 | |
3383 | return true; | |
3384 | } | |
3385 | ||
3c8ad5a6 | 3386 | static bool is_access_allowed(struct kvm_page_fault *fault, u64 spte) |
d3e328f2 | 3387 | { |
3c8ad5a6 | 3388 | if (fault->exec) |
d3e328f2 JS |
3389 | return is_executable_pte(spte); |
3390 | ||
3c8ad5a6 | 3391 | if (fault->write) |
d3e328f2 JS |
3392 | return is_writable_pte(spte); |
3393 | ||
3394 | /* Fault was on Read access */ | |
3395 | return spte & PT_PRESENT_MASK; | |
3396 | } | |
3397 | ||
6e8eb206 DM |
3398 | /* |
3399 | * Returns the last level spte pointer of the shadow page walk for the given | |
3400 | * gpa, and sets *spte to the spte value. This spte may be non-preset. If no | |
3401 | * walk could be performed, returns NULL and *spte does not contain valid data. | |
3402 | * | |
3403 | * Contract: | |
3404 | * - Must be called between walk_shadow_page_lockless_{begin,end}. | |
3405 | * - The returned sptep must not be used after walk_shadow_page_lockless_end. | |
3406 | */ | |
3407 | static u64 *fast_pf_get_last_sptep(struct kvm_vcpu *vcpu, gpa_t gpa, u64 *spte) | |
3408 | { | |
3409 | struct kvm_shadow_walk_iterator iterator; | |
3410 | u64 old_spte; | |
3411 | u64 *sptep = NULL; | |
3412 | ||
3413 | for_each_shadow_entry_lockless(vcpu, gpa, iterator, old_spte) { | |
3414 | sptep = iterator.sptep; | |
3415 | *spte = old_spte; | |
6e8eb206 DM |
3416 | } |
3417 | ||
3418 | return sptep; | |
3419 | } | |
3420 | ||
c7ba5b48 | 3421 | /* |
c4371c2a | 3422 | * Returns one of RET_PF_INVALID, RET_PF_FIXED or RET_PF_SPURIOUS. |
c7ba5b48 | 3423 | */ |
3c8ad5a6 | 3424 | static int fast_page_fault(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault) |
c7ba5b48 | 3425 | { |
92a476cb | 3426 | struct kvm_mmu_page *sp; |
c4371c2a | 3427 | int ret = RET_PF_INVALID; |
1de9992f L |
3428 | u64 spte; |
3429 | u64 *sptep; | |
97dceba2 | 3430 | uint retry_count = 0; |
c7ba5b48 | 3431 | |
3c8ad5a6 | 3432 | if (!page_fault_can_be_fast(fault)) |
c4371c2a | 3433 | return ret; |
c7ba5b48 XG |
3434 | |
3435 | walk_shadow_page_lockless_begin(vcpu); | |
c7ba5b48 | 3436 | |
97dceba2 | 3437 | do { |
d3e328f2 | 3438 | u64 new_spte; |
c7ba5b48 | 3439 | |
dfe0ecc6 | 3440 | if (tdp_mmu_enabled) |
3c8ad5a6 | 3441 | sptep = kvm_tdp_mmu_fast_pf_get_last_sptep(vcpu, fault->addr, &spte); |
6e8eb206 | 3442 | else |
3c8ad5a6 | 3443 | sptep = fast_pf_get_last_sptep(vcpu, fault->addr, &spte); |
d162f30a | 3444 | |
1de9992f L |
3445 | /* |
3446 | * It's entirely possible for the mapping to have been zapped | |
3447 | * by a different task, but the root page should always be | |
3448 | * available as the vCPU holds a reference to its root(s). | |
3449 | */ | |
3450 | if (WARN_ON_ONCE(!sptep)) | |
3451 | spte = REMOVED_SPTE; | |
3452 | ||
ec89e643 SC |
3453 | if (!is_shadow_present_pte(spte)) |
3454 | break; | |
3455 | ||
6e8eb206 | 3456 | sp = sptep_to_sp(sptep); |
97dceba2 JS |
3457 | if (!is_last_spte(spte, sp->role.level)) |
3458 | break; | |
c7ba5b48 | 3459 | |
97dceba2 | 3460 | /* |
f160c7b7 JS |
3461 | * Check whether the memory access that caused the fault would |
3462 | * still cause it if it were to be performed right now. If not, | |
3463 | * then this is a spurious fault caused by TLB lazily flushed, | |
3464 | * or some other CPU has already fixed the PTE after the | |
3465 | * current CPU took the fault. | |
97dceba2 JS |
3466 | * |
3467 | * Need not check the access of upper level table entries since | |
3468 | * they are always ACC_ALL. | |
3469 | */ | |
3c8ad5a6 | 3470 | if (is_access_allowed(fault, spte)) { |
c4371c2a | 3471 | ret = RET_PF_SPURIOUS; |
d3e328f2 JS |
3472 | break; |
3473 | } | |
f160c7b7 | 3474 | |
d3e328f2 JS |
3475 | new_spte = spte; |
3476 | ||
54275f74 SC |
3477 | /* |
3478 | * KVM only supports fixing page faults outside of MMU lock for | |
3479 | * direct MMUs, nested MMUs are always indirect, and KVM always | |
3480 | * uses A/D bits for non-nested MMUs. Thus, if A/D bits are | |
3481 | * enabled, the SPTE can't be an access-tracked SPTE. | |
3482 | */ | |
3483 | if (unlikely(!kvm_ad_enabled()) && is_access_track_spte(spte)) | |
d3e328f2 JS |
3484 | new_spte = restore_acc_track_spte(new_spte); |
3485 | ||
3486 | /* | |
54275f74 SC |
3487 | * To keep things simple, only SPTEs that are MMU-writable can |
3488 | * be made fully writable outside of mmu_lock, e.g. only SPTEs | |
3489 | * that were write-protected for dirty-logging or access | |
3490 | * tracking are handled here. Don't bother checking if the | |
3491 | * SPTE is writable to prioritize running with A/D bits enabled. | |
3492 | * The is_access_allowed() check above handles the common case | |
3493 | * of the fault being spurious, and the SPTE is known to be | |
3494 | * shadow-present, i.e. except for access tracking restoration | |
3495 | * making the new SPTE writable, the check is wasteful. | |
d3e328f2 | 3496 | */ |
706c9c55 | 3497 | if (fault->write && is_mmu_writable_spte(spte)) { |
d3e328f2 | 3498 | new_spte |= PT_WRITABLE_MASK; |
f160c7b7 JS |
3499 | |
3500 | /* | |
10c30de0 JS |
3501 | * Do not fix write-permission on the large spte when |
3502 | * dirty logging is enabled. Since we only dirty the | |
3503 | * first page into the dirty-bitmap in | |
d3e328f2 JS |
3504 | * fast_pf_fix_direct_spte(), other pages are missed |
3505 | * if its slot has dirty logging enabled. | |
3506 | * | |
3507 | * Instead, we let the slow page fault path create a | |
3508 | * normal spte to fix the access. | |
f160c7b7 | 3509 | */ |
10c30de0 JS |
3510 | if (sp->role.level > PG_LEVEL_4K && |
3511 | kvm_slot_dirty_track_enabled(fault->slot)) | |
f160c7b7 | 3512 | break; |
97dceba2 | 3513 | } |
c7ba5b48 | 3514 | |
f160c7b7 | 3515 | /* Verify that the fault can be handled in the fast path */ |
d3e328f2 | 3516 | if (new_spte == spte || |
3c8ad5a6 | 3517 | !is_access_allowed(fault, new_spte)) |
97dceba2 JS |
3518 | break; |
3519 | ||
3520 | /* | |
3521 | * Currently, fast page fault only works for direct mapping | |
3522 | * since the gfn is not stable for indirect shadow page. See | |
3ecad8c2 | 3523 | * Documentation/virt/kvm/locking.rst to get more detail. |
97dceba2 | 3524 | */ |
e710c5f6 | 3525 | if (fast_pf_fix_direct_spte(vcpu, fault, sptep, spte, new_spte)) { |
c4371c2a | 3526 | ret = RET_PF_FIXED; |
97dceba2 | 3527 | break; |
c4371c2a | 3528 | } |
97dceba2 JS |
3529 | |
3530 | if (++retry_count > 4) { | |
8d20bd63 | 3531 | pr_warn_once("Fast #PF retrying more than 4 times.\n"); |
97dceba2 JS |
3532 | break; |
3533 | } | |
3534 | ||
97dceba2 | 3535 | } while (true); |
c126d94f | 3536 | |
f0066d94 | 3537 | trace_fast_page_fault(vcpu, fault, sptep, spte, ret); |
c7ba5b48 XG |
3538 | walk_shadow_page_lockless_end(vcpu); |
3539 | ||
1075d41e SC |
3540 | if (ret != RET_PF_INVALID) |
3541 | vcpu->stat.pf_fast++; | |
3542 | ||
c4371c2a | 3543 | return ret; |
c7ba5b48 XG |
3544 | } |
3545 | ||
74b566e6 JS |
3546 | static void mmu_free_root_page(struct kvm *kvm, hpa_t *root_hpa, |
3547 | struct list_head *invalid_list) | |
17ac10ad | 3548 | { |
4db35314 | 3549 | struct kvm_mmu_page *sp; |
17ac10ad | 3550 | |
74b566e6 | 3551 | if (!VALID_PAGE(*root_hpa)) |
7b53aa56 | 3552 | return; |
35af577a | 3553 | |
c5f2d564 | 3554 | sp = root_to_sp(*root_hpa); |
20ba462d | 3555 | if (WARN_ON_ONCE(!sp)) |
9191b8f0 | 3556 | return; |
02c00b3a | 3557 | |
2bdb3d84 | 3558 | if (is_tdp_mmu_page(sp)) |
6103bc07 | 3559 | kvm_tdp_mmu_put_root(kvm, sp, false); |
76eb54e7 BG |
3560 | else if (!--sp->root_count && sp->role.invalid) |
3561 | kvm_mmu_prepare_zap_page(kvm, sp, invalid_list); | |
17ac10ad | 3562 | |
74b566e6 JS |
3563 | *root_hpa = INVALID_PAGE; |
3564 | } | |
3565 | ||
08fb59d8 | 3566 | /* roots_to_free must be some combination of the KVM_MMU_ROOT_* flags */ |
0c1c92f1 | 3567 | void kvm_mmu_free_roots(struct kvm *kvm, struct kvm_mmu *mmu, |
6a82cd1c | 3568 | ulong roots_to_free) |
74b566e6 JS |
3569 | { |
3570 | int i; | |
3571 | LIST_HEAD(invalid_list); | |
594bef79 | 3572 | bool free_active_root; |
74b566e6 | 3573 | |
f94db0c8 SC |
3574 | WARN_ON_ONCE(roots_to_free & ~KVM_MMU_ROOTS_ALL); |
3575 | ||
b94742c9 | 3576 | BUILD_BUG_ON(KVM_MMU_NUM_PREV_ROOTS >= BITS_PER_LONG); |
74b566e6 | 3577 | |
08fb59d8 | 3578 | /* Before acquiring the MMU lock, see if we need to do any real work. */ |
594bef79 PB |
3579 | free_active_root = (roots_to_free & KVM_MMU_ROOT_CURRENT) |
3580 | && VALID_PAGE(mmu->root.hpa); | |
3581 | ||
3582 | if (!free_active_root) { | |
b94742c9 JS |
3583 | for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) |
3584 | if ((roots_to_free & KVM_MMU_ROOT_PREVIOUS(i)) && | |
3585 | VALID_PAGE(mmu->prev_roots[i].hpa)) | |
3586 | break; | |
3587 | ||
3588 | if (i == KVM_MMU_NUM_PREV_ROOTS) | |
3589 | return; | |
3590 | } | |
35af577a | 3591 | |
531810ca | 3592 | write_lock(&kvm->mmu_lock); |
17ac10ad | 3593 | |
b94742c9 JS |
3594 | for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) |
3595 | if (roots_to_free & KVM_MMU_ROOT_PREVIOUS(i)) | |
4d710de9 | 3596 | mmu_free_root_page(kvm, &mmu->prev_roots[i].hpa, |
b94742c9 | 3597 | &invalid_list); |
7c390d35 | 3598 | |
08fb59d8 | 3599 | if (free_active_root) { |
0e3223d8 SC |
3600 | if (kvm_mmu_is_dummy_root(mmu->root.hpa)) { |
3601 | /* Nothing to cleanup for dummy roots. */ | |
3602 | } else if (root_to_sp(mmu->root.hpa)) { | |
b9e5603c | 3603 | mmu_free_root_page(kvm, &mmu->root.hpa, &invalid_list); |
04d45551 | 3604 | } else if (mmu->pae_root) { |
c834e5e4 SC |
3605 | for (i = 0; i < 4; ++i) { |
3606 | if (!IS_VALID_PAE_ROOT(mmu->pae_root[i])) | |
3607 | continue; | |
3608 | ||
3609 | mmu_free_root_page(kvm, &mmu->pae_root[i], | |
3610 | &invalid_list); | |
3611 | mmu->pae_root[i] = INVALID_PAE_ROOT; | |
3612 | } | |
08fb59d8 | 3613 | } |
b9e5603c PB |
3614 | mmu->root.hpa = INVALID_PAGE; |
3615 | mmu->root.pgd = 0; | |
17ac10ad | 3616 | } |
74b566e6 | 3617 | |
4d710de9 | 3618 | kvm_mmu_commit_zap_page(kvm, &invalid_list); |
531810ca | 3619 | write_unlock(&kvm->mmu_lock); |
17ac10ad | 3620 | } |
74b566e6 | 3621 | EXPORT_SYMBOL_GPL(kvm_mmu_free_roots); |
17ac10ad | 3622 | |
0c1c92f1 | 3623 | void kvm_mmu_free_guest_mode_roots(struct kvm *kvm, struct kvm_mmu *mmu) |
25b62c62 SC |
3624 | { |
3625 | unsigned long roots_to_free = 0; | |
c5f2d564 | 3626 | struct kvm_mmu_page *sp; |
25b62c62 SC |
3627 | hpa_t root_hpa; |
3628 | int i; | |
3629 | ||
3630 | /* | |
3631 | * This should not be called while L2 is active, L2 can't invalidate | |
3632 | * _only_ its own roots, e.g. INVVPID unconditionally exits. | |
3633 | */ | |
7a458f0e | 3634 | WARN_ON_ONCE(mmu->root_role.guest_mode); |
25b62c62 SC |
3635 | |
3636 | for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) { | |
3637 | root_hpa = mmu->prev_roots[i].hpa; | |
3638 | if (!VALID_PAGE(root_hpa)) | |
3639 | continue; | |
3640 | ||
c5f2d564 SC |
3641 | sp = root_to_sp(root_hpa); |
3642 | if (!sp || sp->role.guest_mode) | |
25b62c62 SC |
3643 | roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i); |
3644 | } | |
3645 | ||
0c1c92f1 | 3646 | kvm_mmu_free_roots(kvm, mmu, roots_to_free); |
25b62c62 SC |
3647 | } |
3648 | EXPORT_SYMBOL_GPL(kvm_mmu_free_guest_mode_roots); | |
3649 | ||
2e65e842 | 3650 | static hpa_t mmu_alloc_root(struct kvm_vcpu *vcpu, gfn_t gfn, int quadrant, |
86938ab6 | 3651 | u8 level) |
651dd37a | 3652 | { |
2e65e842 | 3653 | union kvm_mmu_page_role role = vcpu->arch.mmu->root_role; |
651dd37a | 3654 | struct kvm_mmu_page *sp; |
8123f265 | 3655 | |
2e65e842 | 3656 | role.level = level; |
7f497775 | 3657 | role.quadrant = quadrant; |
2e65e842 | 3658 | |
7f497775 DM |
3659 | WARN_ON_ONCE(quadrant && !role.has_4_byte_gpte); |
3660 | WARN_ON_ONCE(role.direct && role.has_4_byte_gpte); | |
2e65e842 | 3661 | |
87654643 | 3662 | sp = kvm_mmu_get_shadow_page(vcpu, gfn, role); |
8123f265 SC |
3663 | ++sp->root_count; |
3664 | ||
8123f265 SC |
3665 | return __pa(sp->spt); |
3666 | } | |
3667 | ||
3668 | static int mmu_alloc_direct_roots(struct kvm_vcpu *vcpu) | |
3669 | { | |
b37233c9 | 3670 | struct kvm_mmu *mmu = vcpu->arch.mmu; |
a972e29c | 3671 | u8 shadow_root_level = mmu->root_role.level; |
8123f265 | 3672 | hpa_t root; |
7ebaf15e | 3673 | unsigned i; |
4a38162e PB |
3674 | int r; |
3675 | ||
3676 | write_lock(&vcpu->kvm->mmu_lock); | |
3677 | r = make_mmu_pages_available(vcpu); | |
3678 | if (r < 0) | |
3679 | goto out_unlock; | |
651dd37a | 3680 | |
1f98f2bd | 3681 | if (tdp_mmu_enabled) { |
02c00b3a | 3682 | root = kvm_tdp_mmu_get_vcpu_root_hpa(vcpu); |
b9e5603c | 3683 | mmu->root.hpa = root; |
02c00b3a | 3684 | } else if (shadow_root_level >= PT64_ROOT_4LEVEL) { |
86938ab6 | 3685 | root = mmu_alloc_root(vcpu, 0, 0, shadow_root_level); |
b9e5603c | 3686 | mmu->root.hpa = root; |
8123f265 | 3687 | } else if (shadow_root_level == PT32E_ROOT_LEVEL) { |
4a38162e PB |
3688 | if (WARN_ON_ONCE(!mmu->pae_root)) { |
3689 | r = -EIO; | |
3690 | goto out_unlock; | |
3691 | } | |
73ad1606 | 3692 | |
651dd37a | 3693 | for (i = 0; i < 4; ++i) { |
c834e5e4 | 3694 | WARN_ON_ONCE(IS_VALID_PAE_ROOT(mmu->pae_root[i])); |
651dd37a | 3695 | |
7f497775 | 3696 | root = mmu_alloc_root(vcpu, i << (30 - PAGE_SHIFT), 0, |
2e65e842 | 3697 | PT32_ROOT_LEVEL); |
17e368d9 | 3698 | mmu->pae_root[i] = root | PT_PRESENT_MASK | |
d2263de1 | 3699 | shadow_me_value; |
651dd37a | 3700 | } |
b9e5603c | 3701 | mmu->root.hpa = __pa(mmu->pae_root); |
73ad1606 SC |
3702 | } else { |
3703 | WARN_ONCE(1, "Bad TDP root level = %d\n", shadow_root_level); | |
4a38162e PB |
3704 | r = -EIO; |
3705 | goto out_unlock; | |
73ad1606 | 3706 | } |
3651c7fc | 3707 | |
b9e5603c PB |
3708 | /* root.pgd is ignored for direct MMUs. */ |
3709 | mmu->root.pgd = 0; | |
4a38162e PB |
3710 | out_unlock: |
3711 | write_unlock(&vcpu->kvm->mmu_lock); | |
3712 | return r; | |
651dd37a JR |
3713 | } |
3714 | ||
1e76a3ce DS |
3715 | static int mmu_first_shadow_root_alloc(struct kvm *kvm) |
3716 | { | |
3717 | struct kvm_memslots *slots; | |
3718 | struct kvm_memory_slot *slot; | |
a54d8066 | 3719 | int r = 0, i, bkt; |
1e76a3ce DS |
3720 | |
3721 | /* | |
3722 | * Check if this is the first shadow root being allocated before | |
3723 | * taking the lock. | |
3724 | */ | |
3725 | if (kvm_shadow_root_allocated(kvm)) | |
3726 | return 0; | |
3727 | ||
3728 | mutex_lock(&kvm->slots_arch_lock); | |
3729 | ||
3730 | /* Recheck, under the lock, whether this is the first shadow root. */ | |
3731 | if (kvm_shadow_root_allocated(kvm)) | |
3732 | goto out_unlock; | |
3733 | ||
3734 | /* | |
3735 | * Check if anything actually needs to be allocated, e.g. all metadata | |
3736 | * will be allocated upfront if TDP is disabled. | |
3737 | */ | |
3738 | if (kvm_memslots_have_rmaps(kvm) && | |
3739 | kvm_page_track_write_tracking_enabled(kvm)) | |
3740 | goto out_success; | |
3741 | ||
3742 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { | |
3743 | slots = __kvm_memslots(kvm, i); | |
a54d8066 | 3744 | kvm_for_each_memslot(slot, bkt, slots) { |
1e76a3ce DS |
3745 | /* |
3746 | * Both of these functions are no-ops if the target is | |
3747 | * already allocated, so unconditionally calling both | |
3748 | * is safe. Intentionally do NOT free allocations on | |
3749 | * failure to avoid having to track which allocations | |
3750 | * were made now versus when the memslot was created. | |
3751 | * The metadata is guaranteed to be freed when the slot | |
3752 | * is freed, and will be kept/used if userspace retries | |
3753 | * KVM_RUN instead of killing the VM. | |
3754 | */ | |
3755 | r = memslot_rmap_alloc(slot, slot->npages); | |
3756 | if (r) | |
3757 | goto out_unlock; | |
3758 | r = kvm_page_track_write_tracking_alloc(slot); | |
3759 | if (r) | |
3760 | goto out_unlock; | |
3761 | } | |
3762 | } | |
3763 | ||
3764 | /* | |
3765 | * Ensure that shadow_root_allocated becomes true strictly after | |
3766 | * all the related pointers are set. | |
3767 | */ | |
3768 | out_success: | |
3769 | smp_store_release(&kvm->arch.shadow_root_allocated, true); | |
3770 | ||
3771 | out_unlock: | |
3772 | mutex_unlock(&kvm->slots_arch_lock); | |
3773 | return r; | |
3774 | } | |
3775 | ||
651dd37a | 3776 | static int mmu_alloc_shadow_roots(struct kvm_vcpu *vcpu) |
17ac10ad | 3777 | { |
b37233c9 | 3778 | struct kvm_mmu *mmu = vcpu->arch.mmu; |
6e0918ae | 3779 | u64 pdptrs[4], pm_mask; |
be01e8e2 | 3780 | gfn_t root_gfn, root_pgd; |
7f497775 | 3781 | int quadrant, i, r; |
8123f265 | 3782 | hpa_t root; |
3bb65a22 | 3783 | |
2fdcc1b3 | 3784 | root_pgd = kvm_mmu_get_guest_pgd(vcpu, mmu); |
be01e8e2 | 3785 | root_gfn = root_pgd >> PAGE_SHIFT; |
17ac10ad | 3786 | |
0e3223d8 SC |
3787 | if (!kvm_vcpu_is_visible_gfn(vcpu, root_gfn)) { |
3788 | mmu->root.hpa = kvm_mmu_get_dummy_root(); | |
3789 | return 0; | |
3790 | } | |
651dd37a | 3791 | |
4a38162e PB |
3792 | /* |
3793 | * On SVM, reading PDPTRs might access guest memory, which might fault | |
3794 | * and thus might sleep. Grab the PDPTRs before acquiring mmu_lock. | |
3795 | */ | |
4d25502a | 3796 | if (mmu->cpu_role.base.level == PT32E_ROOT_LEVEL) { |
6e0918ae SC |
3797 | for (i = 0; i < 4; ++i) { |
3798 | pdptrs[i] = mmu->get_pdptr(vcpu, i); | |
3799 | if (!(pdptrs[i] & PT_PRESENT_MASK)) | |
3800 | continue; | |
3801 | ||
0e3223d8 SC |
3802 | if (!kvm_vcpu_is_visible_gfn(vcpu, pdptrs[i] >> PAGE_SHIFT)) |
3803 | pdptrs[i] = 0; | |
6e0918ae SC |
3804 | } |
3805 | } | |
3806 | ||
1e76a3ce | 3807 | r = mmu_first_shadow_root_alloc(vcpu->kvm); |
d501f747 BG |
3808 | if (r) |
3809 | return r; | |
3810 | ||
4a38162e PB |
3811 | write_lock(&vcpu->kvm->mmu_lock); |
3812 | r = make_mmu_pages_available(vcpu); | |
3813 | if (r < 0) | |
3814 | goto out_unlock; | |
3815 | ||
651dd37a JR |
3816 | /* |
3817 | * Do we shadow a long mode page table? If so we need to | |
3818 | * write-protect the guests page table root. | |
3819 | */ | |
4d25502a | 3820 | if (mmu->cpu_role.base.level >= PT64_ROOT_4LEVEL) { |
8123f265 | 3821 | root = mmu_alloc_root(vcpu, root_gfn, 0, |
86938ab6 | 3822 | mmu->root_role.level); |
b9e5603c | 3823 | mmu->root.hpa = root; |
be01e8e2 | 3824 | goto set_root_pgd; |
17ac10ad | 3825 | } |
f87f9288 | 3826 | |
4a38162e PB |
3827 | if (WARN_ON_ONCE(!mmu->pae_root)) { |
3828 | r = -EIO; | |
3829 | goto out_unlock; | |
3830 | } | |
73ad1606 | 3831 | |
651dd37a JR |
3832 | /* |
3833 | * We shadow a 32 bit page table. This may be a legacy 2-level | |
81407ca5 JR |
3834 | * or a PAE 3-level page table. In either case we need to be aware that |
3835 | * the shadow page table may be a PAE or a long mode page table. | |
651dd37a | 3836 | */ |
e54f1ff2 | 3837 | pm_mask = PT_PRESENT_MASK | shadow_me_value; |
a972e29c | 3838 | if (mmu->root_role.level >= PT64_ROOT_4LEVEL) { |
81407ca5 JR |
3839 | pm_mask |= PT_ACCESSED_MASK | PT_WRITABLE_MASK | PT_USER_MASK; |
3840 | ||
03ca4589 | 3841 | if (WARN_ON_ONCE(!mmu->pml4_root)) { |
4a38162e PB |
3842 | r = -EIO; |
3843 | goto out_unlock; | |
3844 | } | |
03ca4589 | 3845 | mmu->pml4_root[0] = __pa(mmu->pae_root) | pm_mask; |
cb0f722a | 3846 | |
a972e29c | 3847 | if (mmu->root_role.level == PT64_ROOT_5LEVEL) { |
cb0f722a WH |
3848 | if (WARN_ON_ONCE(!mmu->pml5_root)) { |
3849 | r = -EIO; | |
3850 | goto out_unlock; | |
3851 | } | |
3852 | mmu->pml5_root[0] = __pa(mmu->pml4_root) | pm_mask; | |
3853 | } | |
04d45551 SC |
3854 | } |
3855 | ||
17ac10ad | 3856 | for (i = 0; i < 4; ++i) { |
c834e5e4 | 3857 | WARN_ON_ONCE(IS_VALID_PAE_ROOT(mmu->pae_root[i])); |
6e6ec584 | 3858 | |
4d25502a | 3859 | if (mmu->cpu_role.base.level == PT32E_ROOT_LEVEL) { |
6e0918ae | 3860 | if (!(pdptrs[i] & PT_PRESENT_MASK)) { |
c834e5e4 | 3861 | mmu->pae_root[i] = INVALID_PAE_ROOT; |
417726a3 AK |
3862 | continue; |
3863 | } | |
6e0918ae | 3864 | root_gfn = pdptrs[i] >> PAGE_SHIFT; |
5a7388c2 | 3865 | } |
8facbbff | 3866 | |
7f497775 DM |
3867 | /* |
3868 | * If shadowing 32-bit non-PAE page tables, each PAE page | |
3869 | * directory maps one quarter of the guest's non-PAE page | |
3870 | * directory. Othwerise each PAE page direct shadows one guest | |
3871 | * PAE page directory so that quadrant should be 0. | |
3872 | */ | |
3873 | quadrant = (mmu->cpu_role.base.level == PT32_ROOT_LEVEL) ? i : 0; | |
3874 | ||
3875 | root = mmu_alloc_root(vcpu, root_gfn, quadrant, PT32_ROOT_LEVEL); | |
b37233c9 | 3876 | mmu->pae_root[i] = root | pm_mask; |
17ac10ad | 3877 | } |
81407ca5 | 3878 | |
a972e29c | 3879 | if (mmu->root_role.level == PT64_ROOT_5LEVEL) |
b9e5603c | 3880 | mmu->root.hpa = __pa(mmu->pml5_root); |
a972e29c | 3881 | else if (mmu->root_role.level == PT64_ROOT_4LEVEL) |
b9e5603c | 3882 | mmu->root.hpa = __pa(mmu->pml4_root); |
ba0a194f | 3883 | else |
b9e5603c | 3884 | mmu->root.hpa = __pa(mmu->pae_root); |
81407ca5 | 3885 | |
be01e8e2 | 3886 | set_root_pgd: |
b9e5603c | 3887 | mmu->root.pgd = root_pgd; |
4a38162e PB |
3888 | out_unlock: |
3889 | write_unlock(&vcpu->kvm->mmu_lock); | |
ad7dc69a | 3890 | |
c6c937d6 | 3891 | return r; |
17ac10ad AK |
3892 | } |
3893 | ||
748e52b9 SC |
3894 | static int mmu_alloc_special_roots(struct kvm_vcpu *vcpu) |
3895 | { | |
3896 | struct kvm_mmu *mmu = vcpu->arch.mmu; | |
a972e29c | 3897 | bool need_pml5 = mmu->root_role.level > PT64_ROOT_4LEVEL; |
cb0f722a WH |
3898 | u64 *pml5_root = NULL; |
3899 | u64 *pml4_root = NULL; | |
3900 | u64 *pae_root; | |
81407ca5 JR |
3901 | |
3902 | /* | |
748e52b9 SC |
3903 | * When shadowing 32-bit or PAE NPT with 64-bit NPT, the PML4 and PDP |
3904 | * tables are allocated and initialized at root creation as there is no | |
3905 | * equivalent level in the guest's NPT to shadow. Allocate the tables | |
3906 | * on demand, as running a 32-bit L1 VMM on 64-bit KVM is very rare. | |
81407ca5 | 3907 | */ |
347a0d0d PB |
3908 | if (mmu->root_role.direct || |
3909 | mmu->cpu_role.base.level >= PT64_ROOT_4LEVEL || | |
a972e29c | 3910 | mmu->root_role.level < PT64_ROOT_4LEVEL) |
748e52b9 | 3911 | return 0; |
81407ca5 | 3912 | |
a717a780 SC |
3913 | /* |
3914 | * NPT, the only paging mode that uses this horror, uses a fixed number | |
3915 | * of levels for the shadow page tables, e.g. all MMUs are 4-level or | |
3916 | * all MMus are 5-level. Thus, this can safely require that pml5_root | |
3917 | * is allocated if the other roots are valid and pml5 is needed, as any | |
3918 | * prior MMU would also have required pml5. | |
3919 | */ | |
3920 | if (mmu->pae_root && mmu->pml4_root && (!need_pml5 || mmu->pml5_root)) | |
748e52b9 | 3921 | return 0; |
81407ca5 | 3922 | |
748e52b9 SC |
3923 | /* |
3924 | * The special roots should always be allocated in concert. Yell and | |
3925 | * bail if KVM ends up in a state where only one of the roots is valid. | |
3926 | */ | |
cb0f722a | 3927 | if (WARN_ON_ONCE(!tdp_enabled || mmu->pae_root || mmu->pml4_root || |
a717a780 | 3928 | (need_pml5 && mmu->pml5_root))) |
748e52b9 | 3929 | return -EIO; |
81407ca5 | 3930 | |
4a98623d SC |
3931 | /* |
3932 | * Unlike 32-bit NPT, the PDP table doesn't need to be in low mem, and | |
3933 | * doesn't need to be decrypted. | |
3934 | */ | |
748e52b9 SC |
3935 | pae_root = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT); |
3936 | if (!pae_root) | |
3937 | return -ENOMEM; | |
81407ca5 | 3938 | |
cb0f722a | 3939 | #ifdef CONFIG_X86_64 |
03ca4589 | 3940 | pml4_root = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT); |
cb0f722a WH |
3941 | if (!pml4_root) |
3942 | goto err_pml4; | |
3943 | ||
a717a780 | 3944 | if (need_pml5) { |
cb0f722a WH |
3945 | pml5_root = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT); |
3946 | if (!pml5_root) | |
3947 | goto err_pml5; | |
81407ca5 | 3948 | } |
cb0f722a | 3949 | #endif |
81407ca5 | 3950 | |
748e52b9 | 3951 | mmu->pae_root = pae_root; |
03ca4589 | 3952 | mmu->pml4_root = pml4_root; |
cb0f722a | 3953 | mmu->pml5_root = pml5_root; |
ad7dc69a | 3954 | |
8986ecc0 | 3955 | return 0; |
cb0f722a WH |
3956 | |
3957 | #ifdef CONFIG_X86_64 | |
3958 | err_pml5: | |
3959 | free_page((unsigned long)pml4_root); | |
3960 | err_pml4: | |
3961 | free_page((unsigned long)pae_root); | |
3962 | return -ENOMEM; | |
3963 | #endif | |
17ac10ad AK |
3964 | } |
3965 | ||
264d3dc1 LJ |
3966 | static bool is_unsync_root(hpa_t root) |
3967 | { | |
3968 | struct kvm_mmu_page *sp; | |
3969 | ||
0e3223d8 | 3970 | if (!VALID_PAGE(root) || kvm_mmu_is_dummy_root(root)) |
61b05a9f LJ |
3971 | return false; |
3972 | ||
264d3dc1 LJ |
3973 | /* |
3974 | * The read barrier orders the CPU's read of SPTE.W during the page table | |
3975 | * walk before the reads of sp->unsync/sp->unsync_children here. | |
3976 | * | |
3977 | * Even if another CPU was marking the SP as unsync-ed simultaneously, | |
3978 | * any guest page table changes are not guaranteed to be visible anyway | |
3979 | * until this VCPU issues a TLB flush strictly after those changes are | |
3980 | * made. We only need to ensure that the other CPU sets these flags | |
3981 | * before any actual changes to the page tables are made. The comments | |
3982 | * in mmu_try_to_unsync_pages() describe what could go wrong if this | |
3983 | * requirement isn't satisfied. | |
3984 | */ | |
3985 | smp_rmb(); | |
c5f2d564 | 3986 | sp = root_to_sp(root); |
5d6a3221 SC |
3987 | |
3988 | /* | |
3989 | * PAE roots (somewhat arbitrarily) aren't backed by shadow pages, the | |
3990 | * PDPTEs for a given PAE root need to be synchronized individually. | |
3991 | */ | |
3992 | if (WARN_ON_ONCE(!sp)) | |
3993 | return false; | |
3994 | ||
264d3dc1 LJ |
3995 | if (sp->unsync || sp->unsync_children) |
3996 | return true; | |
3997 | ||
3998 | return false; | |
3999 | } | |
4000 | ||
578e1c4d | 4001 | void kvm_mmu_sync_roots(struct kvm_vcpu *vcpu) |
0ba73cda MT |
4002 | { |
4003 | int i; | |
4004 | struct kvm_mmu_page *sp; | |
4005 | ||
347a0d0d | 4006 | if (vcpu->arch.mmu->root_role.direct) |
81407ca5 JR |
4007 | return; |
4008 | ||
b9e5603c | 4009 | if (!VALID_PAGE(vcpu->arch.mmu->root.hpa)) |
0ba73cda | 4010 | return; |
6903074c | 4011 | |
56f17dd3 | 4012 | vcpu_clear_mmio_info(vcpu, MMIO_GVA_ANY); |
578e1c4d | 4013 | |
4d25502a | 4014 | if (vcpu->arch.mmu->cpu_role.base.level >= PT64_ROOT_4LEVEL) { |
b9e5603c | 4015 | hpa_t root = vcpu->arch.mmu->root.hpa; |
578e1c4d | 4016 | |
264d3dc1 | 4017 | if (!is_unsync_root(root)) |
578e1c4d JS |
4018 | return; |
4019 | ||
c5f2d564 SC |
4020 | sp = root_to_sp(root); |
4021 | ||
531810ca | 4022 | write_lock(&vcpu->kvm->mmu_lock); |
65855ed8 | 4023 | mmu_sync_children(vcpu, sp, true); |
531810ca | 4024 | write_unlock(&vcpu->kvm->mmu_lock); |
0ba73cda MT |
4025 | return; |
4026 | } | |
578e1c4d | 4027 | |
531810ca | 4028 | write_lock(&vcpu->kvm->mmu_lock); |
578e1c4d | 4029 | |
0ba73cda | 4030 | for (i = 0; i < 4; ++i) { |
44dd3ffa | 4031 | hpa_t root = vcpu->arch.mmu->pae_root[i]; |
0ba73cda | 4032 | |
c834e5e4 | 4033 | if (IS_VALID_PAE_ROOT(root)) { |
5e3edd7e | 4034 | sp = spte_to_child_sp(root); |
65855ed8 | 4035 | mmu_sync_children(vcpu, sp, true); |
0ba73cda MT |
4036 | } |
4037 | } | |
0ba73cda | 4038 | |
531810ca | 4039 | write_unlock(&vcpu->kvm->mmu_lock); |
0ba73cda MT |
4040 | } |
4041 | ||
61b05a9f LJ |
4042 | void kvm_mmu_sync_prev_roots(struct kvm_vcpu *vcpu) |
4043 | { | |
4044 | unsigned long roots_to_free = 0; | |
4045 | int i; | |
4046 | ||
4047 | for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) | |
4048 | if (is_unsync_root(vcpu->arch.mmu->prev_roots[i].hpa)) | |
4049 | roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i); | |
4050 | ||
4051 | /* sync prev_roots by simply freeing them */ | |
0c1c92f1 | 4052 | kvm_mmu_free_roots(vcpu->kvm, vcpu->arch.mmu, roots_to_free); |
61b05a9f LJ |
4053 | } |
4054 | ||
1f5a21ee | 4055 | static gpa_t nonpaging_gva_to_gpa(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, |
5b22bbe7 | 4056 | gpa_t vaddr, u64 access, |
1f5a21ee | 4057 | struct x86_exception *exception) |
6aa8b732 | 4058 | { |
ab9ae313 AK |
4059 | if (exception) |
4060 | exception->error_code = 0; | |
c59a0f57 | 4061 | return kvm_translate_gpa(vcpu, mmu, vaddr, access, exception); |
6539e738 JR |
4062 | } |
4063 | ||
ded58749 | 4064 | static bool mmio_info_in_cache(struct kvm_vcpu *vcpu, u64 addr, bool direct) |
ce88decf | 4065 | { |
9034e6e8 PB |
4066 | /* |
4067 | * A nested guest cannot use the MMIO cache if it is using nested | |
4068 | * page tables, because cr2 is a nGPA while the cache stores GPAs. | |
4069 | */ | |
4070 | if (mmu_is_nested(vcpu)) | |
4071 | return false; | |
4072 | ||
ce88decf XG |
4073 | if (direct) |
4074 | return vcpu_match_mmio_gpa(vcpu, addr); | |
4075 | ||
4076 | return vcpu_match_mmio_gva(vcpu, addr); | |
4077 | } | |
4078 | ||
95fb5b02 BG |
4079 | /* |
4080 | * Return the level of the lowest level SPTE added to sptes. | |
4081 | * That SPTE may be non-present. | |
c5c8c7c5 DM |
4082 | * |
4083 | * Must be called between walk_shadow_page_lockless_{begin,end}. | |
95fb5b02 | 4084 | */ |
39b4d43e | 4085 | static int get_walk(struct kvm_vcpu *vcpu, u64 addr, u64 *sptes, int *root_level) |
ce88decf XG |
4086 | { |
4087 | struct kvm_shadow_walk_iterator iterator; | |
2aa07893 | 4088 | int leaf = -1; |
95fb5b02 | 4089 | u64 spte; |
ce88decf | 4090 | |
39b4d43e SC |
4091 | for (shadow_walk_init(&iterator, vcpu, addr), |
4092 | *root_level = iterator.level; | |
47ab8751 XG |
4093 | shadow_walk_okay(&iterator); |
4094 | __shadow_walk_next(&iterator, spte)) { | |
95fb5b02 | 4095 | leaf = iterator.level; |
47ab8751 XG |
4096 | spte = mmu_spte_get_lockless(iterator.sptep); |
4097 | ||
dde81f94 | 4098 | sptes[leaf] = spte; |
95fb5b02 BG |
4099 | } |
4100 | ||
95fb5b02 BG |
4101 | return leaf; |
4102 | } | |
4103 | ||
9aa41879 | 4104 | /* return true if reserved bit(s) are detected on a valid, non-MMIO SPTE. */ |
95fb5b02 BG |
4105 | static bool get_mmio_spte(struct kvm_vcpu *vcpu, u64 addr, u64 *sptep) |
4106 | { | |
dde81f94 | 4107 | u64 sptes[PT64_ROOT_MAX_LEVEL + 1]; |
95fb5b02 | 4108 | struct rsvd_bits_validate *rsvd_check; |
39b4d43e | 4109 | int root, leaf, level; |
95fb5b02 BG |
4110 | bool reserved = false; |
4111 | ||
c5c8c7c5 DM |
4112 | walk_shadow_page_lockless_begin(vcpu); |
4113 | ||
78fdd2f0 | 4114 | if (is_tdp_mmu_active(vcpu)) |
39b4d43e | 4115 | leaf = kvm_tdp_mmu_get_walk(vcpu, addr, sptes, &root); |
95fb5b02 | 4116 | else |
39b4d43e | 4117 | leaf = get_walk(vcpu, addr, sptes, &root); |
95fb5b02 | 4118 | |
c5c8c7c5 DM |
4119 | walk_shadow_page_lockless_end(vcpu); |
4120 | ||
2aa07893 SC |
4121 | if (unlikely(leaf < 0)) { |
4122 | *sptep = 0ull; | |
4123 | return reserved; | |
4124 | } | |
4125 | ||
9aa41879 SC |
4126 | *sptep = sptes[leaf]; |
4127 | ||
4128 | /* | |
4129 | * Skip reserved bits checks on the terminal leaf if it's not a valid | |
4130 | * SPTE. Note, this also (intentionally) skips MMIO SPTEs, which, by | |
4131 | * design, always have reserved bits set. The purpose of the checks is | |
4132 | * to detect reserved bits on non-MMIO SPTEs. i.e. buggy SPTEs. | |
4133 | */ | |
4134 | if (!is_shadow_present_pte(sptes[leaf])) | |
4135 | leaf++; | |
95fb5b02 BG |
4136 | |
4137 | rsvd_check = &vcpu->arch.mmu->shadow_zero_check; | |
4138 | ||
9aa41879 | 4139 | for (level = root; level >= leaf; level--) |
961f8445 | 4140 | reserved |= is_rsvd_spte(rsvd_check, sptes[level], level); |
47ab8751 | 4141 | |
47ab8751 | 4142 | if (reserved) { |
bb4cdf3a | 4143 | pr_err("%s: reserved bits set on MMU-present spte, addr 0x%llx, hierarchy:\n", |
47ab8751 | 4144 | __func__, addr); |
95fb5b02 | 4145 | for (level = root; level >= leaf; level--) |
bb4cdf3a SC |
4146 | pr_err("------ spte = 0x%llx level = %d, rsvd bits = 0x%llx", |
4147 | sptes[level], level, | |
961f8445 | 4148 | get_rsvd_bits(rsvd_check, sptes[level], level)); |
47ab8751 | 4149 | } |
ddce6208 | 4150 | |
47ab8751 | 4151 | return reserved; |
ce88decf XG |
4152 | } |
4153 | ||
e08d26f0 | 4154 | static int handle_mmio_page_fault(struct kvm_vcpu *vcpu, u64 addr, bool direct) |
ce88decf XG |
4155 | { |
4156 | u64 spte; | |
47ab8751 | 4157 | bool reserved; |
ce88decf | 4158 | |
ded58749 | 4159 | if (mmio_info_in_cache(vcpu, addr, direct)) |
9b8ebbdb | 4160 | return RET_PF_EMULATE; |
ce88decf | 4161 | |
95fb5b02 | 4162 | reserved = get_mmio_spte(vcpu, addr, &spte); |
20ba462d | 4163 | if (WARN_ON_ONCE(reserved)) |
9b8ebbdb | 4164 | return -EINVAL; |
ce88decf XG |
4165 | |
4166 | if (is_mmio_spte(spte)) { | |
4167 | gfn_t gfn = get_mmio_spte_gfn(spte); | |
0a2b64c5 | 4168 | unsigned int access = get_mmio_spte_access(spte); |
ce88decf | 4169 | |
54bf36aa | 4170 | if (!check_mmio_spte(vcpu, spte)) |
9b8ebbdb | 4171 | return RET_PF_INVALID; |
f8f55942 | 4172 | |
ce88decf XG |
4173 | if (direct) |
4174 | addr = 0; | |
4f022648 XG |
4175 | |
4176 | trace_handle_mmio_page_fault(addr, gfn, access); | |
ce88decf | 4177 | vcpu_cache_mmio_info(vcpu, addr, gfn, access); |
9b8ebbdb | 4178 | return RET_PF_EMULATE; |
ce88decf XG |
4179 | } |
4180 | ||
ce88decf XG |
4181 | /* |
4182 | * If the page table is zapped by other cpus, let CPU fault again on | |
4183 | * the address. | |
4184 | */ | |
9b8ebbdb | 4185 | return RET_PF_RETRY; |
ce88decf | 4186 | } |
ce88decf | 4187 | |
3d0c27ad | 4188 | static bool page_fault_handle_page_track(struct kvm_vcpu *vcpu, |
b8a5d551 | 4189 | struct kvm_page_fault *fault) |
3d0c27ad | 4190 | { |
b8a5d551 | 4191 | if (unlikely(fault->rsvd)) |
3d0c27ad XG |
4192 | return false; |
4193 | ||
b8a5d551 | 4194 | if (!fault->present || !fault->write) |
3d0c27ad XG |
4195 | return false; |
4196 | ||
4197 | /* | |
4198 | * guest is writing the page which is write tracked which can | |
4199 | * not be fixed by page fault handler. | |
4200 | */ | |
7b574863 | 4201 | if (kvm_gfn_is_write_tracked(vcpu->kvm, fault->slot, fault->gfn)) |
3d0c27ad XG |
4202 | return true; |
4203 | ||
4204 | return false; | |
4205 | } | |
4206 | ||
e5691a81 XG |
4207 | static void shadow_page_table_clear_flood(struct kvm_vcpu *vcpu, gva_t addr) |
4208 | { | |
4209 | struct kvm_shadow_walk_iterator iterator; | |
4210 | u64 spte; | |
4211 | ||
e5691a81 | 4212 | walk_shadow_page_lockless_begin(vcpu); |
3e44dce4 | 4213 | for_each_shadow_entry_lockless(vcpu, addr, iterator, spte) |
e5691a81 | 4214 | clear_sp_write_flooding_count(iterator.sptep); |
e5691a81 XG |
4215 | walk_shadow_page_lockless_end(vcpu); |
4216 | } | |
4217 | ||
6f3c1fc5 LZ |
4218 | static u32 alloc_apf_token(struct kvm_vcpu *vcpu) |
4219 | { | |
4220 | /* make sure the token value is not 0 */ | |
4221 | u32 id = vcpu->arch.apf.id; | |
4222 | ||
4223 | if (id << 12 == 0) | |
4224 | vcpu->arch.apf.id = 1; | |
4225 | ||
4226 | return (vcpu->arch.apf.id++ << 12) | vcpu->vcpu_id; | |
4227 | } | |
4228 | ||
e8c22266 VK |
4229 | static bool kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa, |
4230 | gfn_t gfn) | |
af585b92 GN |
4231 | { |
4232 | struct kvm_arch_async_pf arch; | |
fb67e14f | 4233 | |
6f3c1fc5 | 4234 | arch.token = alloc_apf_token(vcpu); |
af585b92 | 4235 | arch.gfn = gfn; |
347a0d0d | 4236 | arch.direct_map = vcpu->arch.mmu->root_role.direct; |
2fdcc1b3 | 4237 | arch.cr3 = kvm_mmu_get_guest_pgd(vcpu, vcpu->arch.mmu); |
af585b92 | 4238 | |
9f1a8526 SC |
4239 | return kvm_setup_async_pf(vcpu, cr2_or_gpa, |
4240 | kvm_vcpu_gfn_to_hva(vcpu, gfn), &arch); | |
af585b92 GN |
4241 | } |
4242 | ||
8a009d5b SC |
4243 | void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work) |
4244 | { | |
4245 | int r; | |
4246 | ||
4247 | if ((vcpu->arch.mmu->root_role.direct != work->arch.direct_map) || | |
4248 | work->wakeup_all) | |
4249 | return; | |
4250 | ||
4251 | r = kvm_mmu_reload(vcpu); | |
4252 | if (unlikely(r)) | |
4253 | return; | |
4254 | ||
4255 | if (!vcpu->arch.mmu->root_role.direct && | |
2fdcc1b3 | 4256 | work->arch.cr3 != kvm_mmu_get_guest_pgd(vcpu, vcpu->arch.mmu)) |
8a009d5b SC |
4257 | return; |
4258 | ||
258d985f | 4259 | kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true, NULL); |
8a009d5b SC |
4260 | } |
4261 | ||
ba6e3fe2 | 4262 | static int __kvm_faultin_pfn(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault) |
af585b92 | 4263 | { |
e710c5f6 | 4264 | struct kvm_memory_slot *slot = fault->slot; |
af585b92 GN |
4265 | bool async; |
4266 | ||
e0c37868 SC |
4267 | /* |
4268 | * Retry the page fault if the gfn hit a memslot that is being deleted | |
4269 | * or moved. This ensures any existing SPTEs for the old memslot will | |
4270 | * be zapped before KVM inserts a new MMIO SPTE for the gfn. | |
4271 | */ | |
4272 | if (slot && (slot->flags & KVM_MEMSLOT_INVALID)) | |
5276c616 | 4273 | return RET_PF_RETRY; |
e0c37868 | 4274 | |
9cc13d60 ML |
4275 | if (!kvm_is_visible_memslot(slot)) { |
4276 | /* Don't expose private memslots to L2. */ | |
4277 | if (is_guest_mode(vcpu)) { | |
e710c5f6 | 4278 | fault->slot = NULL; |
3647cd04 PB |
4279 | fault->pfn = KVM_PFN_NOSLOT; |
4280 | fault->map_writable = false; | |
5276c616 | 4281 | return RET_PF_CONTINUE; |
9cc13d60 ML |
4282 | } |
4283 | /* | |
4284 | * If the APIC access page exists but is disabled, go directly | |
4285 | * to emulation without caching the MMIO access or creating a | |
4286 | * MMIO SPTE. That way the cache doesn't need to be purged | |
4287 | * when the AVIC is re-enabled. | |
4288 | */ | |
4289 | if (slot && slot->id == APIC_ACCESS_PAGE_PRIVATE_MEMSLOT && | |
5276c616 SC |
4290 | !kvm_apicv_activated(vcpu->kvm)) |
4291 | return RET_PF_EMULATE; | |
3a2936de JM |
4292 | } |
4293 | ||
3520469d | 4294 | async = false; |
c8b88b33 | 4295 | fault->pfn = __gfn_to_pfn_memslot(slot, fault->gfn, false, false, &async, |
3647cd04 PB |
4296 | fault->write, &fault->map_writable, |
4297 | &fault->hva); | |
af585b92 | 4298 | if (!async) |
5276c616 | 4299 | return RET_PF_CONTINUE; /* *pfn has correct page already */ |
af585b92 | 4300 | |
2839180c | 4301 | if (!fault->prefetch && kvm_can_do_async_pf(vcpu)) { |
3647cd04 PB |
4302 | trace_kvm_try_async_get_page(fault->addr, fault->gfn); |
4303 | if (kvm_find_async_pf_gfn(vcpu, fault->gfn)) { | |
1685c0f3 | 4304 | trace_kvm_async_pf_repeated_fault(fault->addr, fault->gfn); |
af585b92 | 4305 | kvm_make_request(KVM_REQ_APF_HALT, vcpu); |
5276c616 SC |
4306 | return RET_PF_RETRY; |
4307 | } else if (kvm_arch_setup_async_pf(vcpu, fault->addr, fault->gfn)) { | |
4308 | return RET_PF_RETRY; | |
4309 | } | |
af585b92 GN |
4310 | } |
4311 | ||
76657687 PX |
4312 | /* |
4313 | * Allow gup to bail on pending non-fatal signals when it's also allowed | |
4314 | * to wait for IO. Note, gup always bails if it is unable to quickly | |
4315 | * get a page and a fatal signal, i.e. SIGKILL, is pending. | |
4316 | */ | |
4317 | fault->pfn = __gfn_to_pfn_memslot(slot, fault->gfn, false, true, NULL, | |
3647cd04 PB |
4318 | fault->write, &fault->map_writable, |
4319 | &fault->hva); | |
5276c616 | 4320 | return RET_PF_CONTINUE; |
af585b92 GN |
4321 | } |
4322 | ||
354c908c DM |
4323 | static int kvm_faultin_pfn(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault, |
4324 | unsigned int access) | |
ba6e3fe2 | 4325 | { |
56c3a4e4 DM |
4326 | int ret; |
4327 | ||
ba6e3fe2 DM |
4328 | fault->mmu_seq = vcpu->kvm->mmu_invalidate_seq; |
4329 | smp_rmb(); | |
4330 | ||
56c3a4e4 DM |
4331 | ret = __kvm_faultin_pfn(vcpu, fault); |
4332 | if (ret != RET_PF_CONTINUE) | |
4333 | return ret; | |
4334 | ||
4335 | if (unlikely(is_error_pfn(fault->pfn))) | |
cd08d178 | 4336 | return kvm_handle_error_pfn(vcpu, fault); |
56c3a4e4 | 4337 | |
354c908c DM |
4338 | if (unlikely(!fault->slot)) |
4339 | return kvm_handle_noslot_fault(vcpu, fault, access); | |
4340 | ||
56c3a4e4 | 4341 | return RET_PF_CONTINUE; |
ba6e3fe2 DM |
4342 | } |
4343 | ||
a955cad8 SC |
4344 | /* |
4345 | * Returns true if the page fault is stale and needs to be retried, i.e. if the | |
4346 | * root was invalidated by a memslot update or a relevant mmu_notifier fired. | |
4347 | */ | |
4348 | static bool is_page_fault_stale(struct kvm_vcpu *vcpu, | |
ba6e3fe2 | 4349 | struct kvm_page_fault *fault) |
a955cad8 | 4350 | { |
c5f2d564 | 4351 | struct kvm_mmu_page *sp = root_to_sp(vcpu->arch.mmu->root.hpa); |
18c841e1 SC |
4352 | |
4353 | /* Special roots, e.g. pae_root, are not backed by shadow pages. */ | |
4354 | if (sp && is_obsolete_sp(vcpu->kvm, sp)) | |
4355 | return true; | |
4356 | ||
4357 | /* | |
4358 | * Roots without an associated shadow page are considered invalid if | |
4359 | * there is a pending request to free obsolete roots. The request is | |
4360 | * only a hint that the current root _may_ be obsolete and needs to be | |
4361 | * reloaded, e.g. if the guest frees a PGD that KVM is tracking as a | |
4362 | * previous root, then __kvm_mmu_prepare_zap_page() signals all vCPUs | |
4363 | * to reload even if no vCPU is actively using the root. | |
4364 | */ | |
527d5cd7 | 4365 | if (!sp && kvm_test_request(KVM_REQ_MMU_FREE_OBSOLETE_ROOTS, vcpu)) |
a955cad8 SC |
4366 | return true; |
4367 | ||
4368 | return fault->slot && | |
ba6e3fe2 | 4369 | mmu_invalidate_retry_hva(vcpu->kvm, fault->mmu_seq, fault->hva); |
a955cad8 SC |
4370 | } |
4371 | ||
4326e57e | 4372 | static int direct_page_fault(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault) |
6aa8b732 | 4373 | { |
83f06fa7 | 4374 | int r; |
ce88decf | 4375 | |
0e3223d8 SC |
4376 | /* Dummy roots are used only for shadowing bad guest roots. */ |
4377 | if (WARN_ON_ONCE(kvm_mmu_is_dummy_root(vcpu->arch.mmu->root.hpa))) | |
4378 | return RET_PF_RETRY; | |
4379 | ||
b8a5d551 | 4380 | if (page_fault_handle_page_track(vcpu, fault)) |
9b8ebbdb | 4381 | return RET_PF_EMULATE; |
ce88decf | 4382 | |
3c8ad5a6 | 4383 | r = fast_page_fault(vcpu, fault); |
6e8eb206 DM |
4384 | if (r != RET_PF_INVALID) |
4385 | return r; | |
83291445 | 4386 | |
378f5cd6 | 4387 | r = mmu_topup_memory_caches(vcpu, false); |
e2dec939 AK |
4388 | if (r) |
4389 | return r; | |
714b93da | 4390 | |
354c908c | 4391 | r = kvm_faultin_pfn(vcpu, fault, ACC_ALL); |
5276c616 | 4392 | if (r != RET_PF_CONTINUE) |
367fd790 | 4393 | return r; |
6aa8b732 | 4394 | |
367fd790 | 4395 | r = RET_PF_RETRY; |
9aa8ab43 | 4396 | write_lock(&vcpu->kvm->mmu_lock); |
a2855afc | 4397 | |
ba6e3fe2 | 4398 | if (is_page_fault_stale(vcpu, fault)) |
367fd790 | 4399 | goto out_unlock; |
a2855afc | 4400 | |
7bd7ded6 SC |
4401 | r = make_mmu_pages_available(vcpu); |
4402 | if (r) | |
367fd790 | 4403 | goto out_unlock; |
a955cad8 | 4404 | |
6c882ef4 | 4405 | r = direct_map(vcpu, fault); |
0f90e1c1 | 4406 | |
367fd790 | 4407 | out_unlock: |
9aa8ab43 | 4408 | write_unlock(&vcpu->kvm->mmu_lock); |
3647cd04 | 4409 | kvm_release_pfn_clean(fault->pfn); |
367fd790 | 4410 | return r; |
6aa8b732 AK |
4411 | } |
4412 | ||
c501040a PB |
4413 | static int nonpaging_page_fault(struct kvm_vcpu *vcpu, |
4414 | struct kvm_page_fault *fault) | |
0f90e1c1 | 4415 | { |
0f90e1c1 | 4416 | /* This path builds a PAE pagetable, we can map 2mb pages at maximum. */ |
4326e57e PB |
4417 | fault->max_level = PG_LEVEL_2M; |
4418 | return direct_page_fault(vcpu, fault); | |
0f90e1c1 SC |
4419 | } |
4420 | ||
1261bfa3 | 4421 | int kvm_handle_page_fault(struct kvm_vcpu *vcpu, u64 error_code, |
d0006530 | 4422 | u64 fault_address, char *insn, int insn_len) |
1261bfa3 WL |
4423 | { |
4424 | int r = 1; | |
9ce372b3 | 4425 | u32 flags = vcpu->arch.apf.host_apf_flags; |
1261bfa3 | 4426 | |
736c291c SC |
4427 | #ifndef CONFIG_X86_64 |
4428 | /* A 64-bit CR2 should be impossible on 32-bit KVM. */ | |
4429 | if (WARN_ON_ONCE(fault_address >> 32)) | |
4430 | return -EFAULT; | |
4431 | #endif | |
4432 | ||
c595ceee | 4433 | vcpu->arch.l1tf_flush_l1d = true; |
9ce372b3 | 4434 | if (!flags) { |
faa03b39 | 4435 | trace_kvm_page_fault(vcpu, fault_address, error_code); |
1261bfa3 | 4436 | |
d0006530 | 4437 | if (kvm_event_needs_reinjection(vcpu)) |
1261bfa3 WL |
4438 | kvm_mmu_unprotect_page_virt(vcpu, fault_address); |
4439 | r = kvm_mmu_page_fault(vcpu, fault_address, error_code, insn, | |
4440 | insn_len); | |
9ce372b3 | 4441 | } else if (flags & KVM_PV_REASON_PAGE_NOT_PRESENT) { |
68fd66f1 | 4442 | vcpu->arch.apf.host_apf_flags = 0; |
1261bfa3 | 4443 | local_irq_disable(); |
6bca69ad | 4444 | kvm_async_pf_task_wait_schedule(fault_address); |
1261bfa3 | 4445 | local_irq_enable(); |
9ce372b3 VK |
4446 | } else { |
4447 | WARN_ONCE(1, "Unexpected host async PF flags: %x\n", flags); | |
1261bfa3 | 4448 | } |
9ce372b3 | 4449 | |
1261bfa3 WL |
4450 | return r; |
4451 | } | |
4452 | EXPORT_SYMBOL_GPL(kvm_handle_page_fault); | |
4453 | ||
9aa8ab43 DM |
4454 | #ifdef CONFIG_X86_64 |
4455 | static int kvm_tdp_mmu_page_fault(struct kvm_vcpu *vcpu, | |
4456 | struct kvm_page_fault *fault) | |
4457 | { | |
4458 | int r; | |
4459 | ||
4460 | if (page_fault_handle_page_track(vcpu, fault)) | |
4461 | return RET_PF_EMULATE; | |
4462 | ||
4463 | r = fast_page_fault(vcpu, fault); | |
4464 | if (r != RET_PF_INVALID) | |
4465 | return r; | |
4466 | ||
4467 | r = mmu_topup_memory_caches(vcpu, false); | |
4468 | if (r) | |
4469 | return r; | |
4470 | ||
4471 | r = kvm_faultin_pfn(vcpu, fault, ACC_ALL); | |
4472 | if (r != RET_PF_CONTINUE) | |
4473 | return r; | |
4474 | ||
4475 | r = RET_PF_RETRY; | |
4476 | read_lock(&vcpu->kvm->mmu_lock); | |
4477 | ||
4478 | if (is_page_fault_stale(vcpu, fault)) | |
4479 | goto out_unlock; | |
4480 | ||
9aa8ab43 DM |
4481 | r = kvm_tdp_mmu_map(vcpu, fault); |
4482 | ||
4483 | out_unlock: | |
4484 | read_unlock(&vcpu->kvm->mmu_lock); | |
4485 | kvm_release_pfn_clean(fault->pfn); | |
4486 | return r; | |
4487 | } | |
4488 | #endif | |
4489 | ||
1affe455 | 4490 | bool __kvm_mmu_honors_guest_mtrrs(bool vm_has_noncoherent_dma) |
fb72d167 | 4491 | { |
d5e90a69 | 4492 | /* |
1affe455 YZ |
4493 | * If host MTRRs are ignored (shadow_memtype_mask is non-zero), and the |
4494 | * VM has non-coherent DMA (DMA doesn't snoop CPU caches), KVM's ABI is | |
4495 | * to honor the memtype from the guest's MTRRs so that guest accesses | |
4496 | * to memory that is DMA'd aren't cached against the guest's wishes. | |
d5e90a69 SC |
4497 | * |
4498 | * Note, KVM may still ultimately ignore guest MTRRs for certain PFNs, | |
4499 | * e.g. KVM will force UC memtype for host MMIO. | |
4500 | */ | |
1affe455 YZ |
4501 | return vm_has_noncoherent_dma && shadow_memtype_mask; |
4502 | } | |
4503 | ||
4504 | int kvm_tdp_page_fault(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault) | |
4505 | { | |
4506 | /* | |
4507 | * If the guest's MTRRs may be used to compute the "real" memtype, | |
4508 | * restrict the mapping level to ensure KVM uses a consistent memtype | |
4509 | * across the entire mapping. | |
4510 | */ | |
4511 | if (kvm_mmu_honors_guest_mtrrs(vcpu->kvm)) { | |
d5e90a69 SC |
4512 | for ( ; fault->max_level > PG_LEVEL_4K; --fault->max_level) { |
4513 | int page_num = KVM_PAGES_PER_HPAGE(fault->max_level); | |
c667a3ba HW |
4514 | gfn_t base = gfn_round_for_level(fault->gfn, |
4515 | fault->max_level); | |
4326e57e | 4516 | |
d5e90a69 SC |
4517 | if (kvm_mtrr_check_gfn_range_consistency(vcpu, base, page_num)) |
4518 | break; | |
4519 | } | |
fd136902 | 4520 | } |
852e3c19 | 4521 | |
9aa8ab43 DM |
4522 | #ifdef CONFIG_X86_64 |
4523 | if (tdp_mmu_enabled) | |
4524 | return kvm_tdp_mmu_page_fault(vcpu, fault); | |
4525 | #endif | |
4526 | ||
4326e57e | 4527 | return direct_page_fault(vcpu, fault); |
fb72d167 JR |
4528 | } |
4529 | ||
84a16226 | 4530 | static void nonpaging_init_context(struct kvm_mmu *context) |
6aa8b732 | 4531 | { |
6aa8b732 | 4532 | context->page_fault = nonpaging_page_fault; |
6aa8b732 | 4533 | context->gva_to_gpa = nonpaging_gva_to_gpa; |
c3c6c9fc | 4534 | context->sync_spte = NULL; |
6aa8b732 AK |
4535 | } |
4536 | ||
be01e8e2 | 4537 | static inline bool is_root_usable(struct kvm_mmu_root_info *root, gpa_t pgd, |
0be44352 SC |
4538 | union kvm_mmu_page_role role) |
4539 | { | |
c30e000e SC |
4540 | struct kvm_mmu_page *sp; |
4541 | ||
4542 | if (!VALID_PAGE(root->hpa)) | |
4543 | return false; | |
4544 | ||
4545 | if (!role.direct && pgd != root->pgd) | |
4546 | return false; | |
4547 | ||
4548 | sp = root_to_sp(root->hpa); | |
4549 | if (WARN_ON_ONCE(!sp)) | |
4550 | return false; | |
4551 | ||
4552 | return role.word == sp->role.word; | |
0be44352 SC |
4553 | } |
4554 | ||
b94742c9 | 4555 | /* |
5499ea73 PB |
4556 | * Find out if a previously cached root matching the new pgd/role is available, |
4557 | * and insert the current root as the MRU in the cache. | |
4558 | * If a matching root is found, it is assigned to kvm_mmu->root and | |
4559 | * true is returned. | |
4560 | * If no match is found, kvm_mmu->root is left invalid, the LRU root is | |
4561 | * evicted to make room for the current root, and false is returned. | |
b94742c9 | 4562 | */ |
5499ea73 PB |
4563 | static bool cached_root_find_and_keep_current(struct kvm *kvm, struct kvm_mmu *mmu, |
4564 | gpa_t new_pgd, | |
4565 | union kvm_mmu_page_role new_role) | |
b94742c9 JS |
4566 | { |
4567 | uint i; | |
b94742c9 | 4568 | |
b9e5603c | 4569 | if (is_root_usable(&mmu->root, new_pgd, new_role)) |
0be44352 SC |
4570 | return true; |
4571 | ||
b94742c9 | 4572 | for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) { |
5499ea73 PB |
4573 | /* |
4574 | * The swaps end up rotating the cache like this: | |
4575 | * C 0 1 2 3 (on entry to the function) | |
4576 | * 0 C 1 2 3 | |
4577 | * 1 C 0 2 3 | |
4578 | * 2 C 0 1 3 | |
4579 | * 3 C 0 1 2 (on exit from the loop) | |
4580 | */ | |
b9e5603c | 4581 | swap(mmu->root, mmu->prev_roots[i]); |
b9e5603c | 4582 | if (is_root_usable(&mmu->root, new_pgd, new_role)) |
5499ea73 | 4583 | return true; |
b94742c9 JS |
4584 | } |
4585 | ||
5499ea73 PB |
4586 | kvm_mmu_free_roots(kvm, mmu, KVM_MMU_ROOT_CURRENT); |
4587 | return false; | |
b94742c9 JS |
4588 | } |
4589 | ||
5499ea73 PB |
4590 | /* |
4591 | * Find out if a previously cached root matching the new pgd/role is available. | |
4592 | * On entry, mmu->root is invalid. | |
4593 | * If a matching root is found, it is assigned to kvm_mmu->root, the LRU entry | |
4594 | * of the cache becomes invalid, and true is returned. | |
4595 | * If no match is found, kvm_mmu->root is left invalid and false is returned. | |
4596 | */ | |
4597 | static bool cached_root_find_without_current(struct kvm *kvm, struct kvm_mmu *mmu, | |
4598 | gpa_t new_pgd, | |
4599 | union kvm_mmu_page_role new_role) | |
6aa8b732 | 4600 | { |
5499ea73 PB |
4601 | uint i; |
4602 | ||
4603 | for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) | |
4604 | if (is_root_usable(&mmu->prev_roots[i], new_pgd, new_role)) | |
4605 | goto hit; | |
7c390d35 | 4606 | |
5499ea73 PB |
4607 | return false; |
4608 | ||
4609 | hit: | |
4610 | swap(mmu->root, mmu->prev_roots[i]); | |
4611 | /* Bubble up the remaining roots. */ | |
4612 | for (; i < KVM_MMU_NUM_PREV_ROOTS - 1; i++) | |
4613 | mmu->prev_roots[i] = mmu->prev_roots[i + 1]; | |
4614 | mmu->prev_roots[i].hpa = INVALID_PAGE; | |
4615 | return true; | |
4616 | } | |
4617 | ||
4618 | static bool fast_pgd_switch(struct kvm *kvm, struct kvm_mmu *mmu, | |
4619 | gpa_t new_pgd, union kvm_mmu_page_role new_role) | |
4620 | { | |
7c390d35 | 4621 | /* |
0e3223d8 SC |
4622 | * Limit reuse to 64-bit hosts+VMs without "special" roots in order to |
4623 | * avoid having to deal with PDPTEs and other complexities. | |
7c390d35 | 4624 | */ |
c5f2d564 | 4625 | if (VALID_PAGE(mmu->root.hpa) && !root_to_sp(mmu->root.hpa)) |
5499ea73 | 4626 | kvm_mmu_free_roots(kvm, mmu, KVM_MMU_ROOT_CURRENT); |
7c390d35 | 4627 | |
5499ea73 PB |
4628 | if (VALID_PAGE(mmu->root.hpa)) |
4629 | return cached_root_find_and_keep_current(kvm, mmu, new_pgd, new_role); | |
4630 | else | |
4631 | return cached_root_find_without_current(kvm, mmu, new_pgd, new_role); | |
6aa8b732 AK |
4632 | } |
4633 | ||
d2e5f333 | 4634 | void kvm_mmu_new_pgd(struct kvm_vcpu *vcpu, gpa_t new_pgd) |
6aa8b732 | 4635 | { |
0c1c92f1 | 4636 | struct kvm_mmu *mmu = vcpu->arch.mmu; |
7a458f0e | 4637 | union kvm_mmu_page_role new_role = mmu->root_role; |
0c1c92f1 | 4638 | |
a7e48ef7 WL |
4639 | /* |
4640 | * Return immediately if no usable root was found, kvm_mmu_reload() | |
4641 | * will establish a valid root prior to the next VM-Enter. | |
4642 | */ | |
4643 | if (!fast_pgd_switch(vcpu->kvm, mmu, new_pgd, new_role)) | |
b869855b | 4644 | return; |
b869855b SC |
4645 | |
4646 | /* | |
4647 | * It's possible that the cached previous root page is obsolete because | |
4648 | * of a change in the MMU generation number. However, changing the | |
527d5cd7 SC |
4649 | * generation number is accompanied by KVM_REQ_MMU_FREE_OBSOLETE_ROOTS, |
4650 | * which will free the root set here and allocate a new one. | |
b869855b SC |
4651 | */ |
4652 | kvm_make_request(KVM_REQ_LOAD_MMU_PGD, vcpu); | |
4653 | ||
b5129100 | 4654 | if (force_flush_and_sync_on_reuse) { |
b869855b SC |
4655 | kvm_make_request(KVM_REQ_MMU_SYNC, vcpu); |
4656 | kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu); | |
b5129100 | 4657 | } |
b869855b SC |
4658 | |
4659 | /* | |
4660 | * The last MMIO access's GVA and GPA are cached in the VCPU. When | |
4661 | * switching to a new CR3, that GVA->GPA mapping may no longer be | |
4662 | * valid. So clear any cached MMIO info even when we don't need to sync | |
4663 | * the shadow page tables. | |
4664 | */ | |
4665 | vcpu_clear_mmio_info(vcpu, MMIO_GVA_ANY); | |
4666 | ||
daa5b6c1 BG |
4667 | /* |
4668 | * If this is a direct root page, it doesn't have a write flooding | |
4669 | * count. Otherwise, clear the write flooding count. | |
4670 | */ | |
c30e000e SC |
4671 | if (!new_role.direct) { |
4672 | struct kvm_mmu_page *sp = root_to_sp(vcpu->arch.mmu->root.hpa); | |
4673 | ||
4674 | if (!WARN_ON_ONCE(!sp)) | |
4675 | __clear_sp_write_flooding_count(sp); | |
4676 | } | |
6aa8b732 | 4677 | } |
be01e8e2 | 4678 | EXPORT_SYMBOL_GPL(kvm_mmu_new_pgd); |
0aab33e4 | 4679 | |
54bf36aa | 4680 | static bool sync_mmio_spte(struct kvm_vcpu *vcpu, u64 *sptep, gfn_t gfn, |
c3e5e415 | 4681 | unsigned int access) |
ce88decf XG |
4682 | { |
4683 | if (unlikely(is_mmio_spte(*sptep))) { | |
4684 | if (gfn != get_mmio_spte_gfn(*sptep)) { | |
4685 | mmu_spte_clear_no_track(sptep); | |
4686 | return true; | |
4687 | } | |
4688 | ||
54bf36aa | 4689 | mark_mmio_spte(vcpu, sptep, gfn, access); |
ce88decf XG |
4690 | return true; |
4691 | } | |
4692 | ||
4693 | return false; | |
4694 | } | |
4695 | ||
37406aaa NHE |
4696 | #define PTTYPE_EPT 18 /* arbitrary */ |
4697 | #define PTTYPE PTTYPE_EPT | |
4698 | #include "paging_tmpl.h" | |
4699 | #undef PTTYPE | |
4700 | ||
6aa8b732 AK |
4701 | #define PTTYPE 64 |
4702 | #include "paging_tmpl.h" | |
4703 | #undef PTTYPE | |
4704 | ||
4705 | #define PTTYPE 32 | |
4706 | #include "paging_tmpl.h" | |
4707 | #undef PTTYPE | |
4708 | ||
f3d90f90 SC |
4709 | static void __reset_rsvds_bits_mask(struct rsvd_bits_validate *rsvd_check, |
4710 | u64 pa_bits_rsvd, int level, bool nx, | |
4711 | bool gbpages, bool pse, bool amd) | |
82725b20 | 4712 | { |
5f7dde7b | 4713 | u64 gbpages_bit_rsvd = 0; |
a0c0feb5 | 4714 | u64 nonleaf_bit8_rsvd = 0; |
5b7f575c | 4715 | u64 high_bits_rsvd; |
82725b20 | 4716 | |
a0a64f50 | 4717 | rsvd_check->bad_mt_xwr = 0; |
25d92081 | 4718 | |
6dc98b86 | 4719 | if (!gbpages) |
5f7dde7b | 4720 | gbpages_bit_rsvd = rsvd_bits(7, 7); |
a0c0feb5 | 4721 | |
5b7f575c SC |
4722 | if (level == PT32E_ROOT_LEVEL) |
4723 | high_bits_rsvd = pa_bits_rsvd & rsvd_bits(0, 62); | |
4724 | else | |
4725 | high_bits_rsvd = pa_bits_rsvd & rsvd_bits(0, 51); | |
4726 | ||
4727 | /* Note, NX doesn't exist in PDPTEs, this is handled below. */ | |
4728 | if (!nx) | |
4729 | high_bits_rsvd |= rsvd_bits(63, 63); | |
4730 | ||
a0c0feb5 PB |
4731 | /* |
4732 | * Non-leaf PML4Es and PDPEs reserve bit 8 (which would be the G bit for | |
4733 | * leaf entries) on AMD CPUs only. | |
4734 | */ | |
6fec2144 | 4735 | if (amd) |
a0c0feb5 PB |
4736 | nonleaf_bit8_rsvd = rsvd_bits(8, 8); |
4737 | ||
6dc98b86 | 4738 | switch (level) { |
82725b20 DE |
4739 | case PT32_ROOT_LEVEL: |
4740 | /* no rsvd bits for 2 level 4K page table entries */ | |
a0a64f50 XG |
4741 | rsvd_check->rsvd_bits_mask[0][1] = 0; |
4742 | rsvd_check->rsvd_bits_mask[0][0] = 0; | |
4743 | rsvd_check->rsvd_bits_mask[1][0] = | |
4744 | rsvd_check->rsvd_bits_mask[0][0]; | |
f815bce8 | 4745 | |
6dc98b86 | 4746 | if (!pse) { |
a0a64f50 | 4747 | rsvd_check->rsvd_bits_mask[1][1] = 0; |
f815bce8 XG |
4748 | break; |
4749 | } | |
4750 | ||
82725b20 DE |
4751 | if (is_cpuid_PSE36()) |
4752 | /* 36bits PSE 4MB page */ | |
a0a64f50 | 4753 | rsvd_check->rsvd_bits_mask[1][1] = rsvd_bits(17, 21); |
82725b20 DE |
4754 | else |
4755 | /* 32 bits PSE 4MB page */ | |
a0a64f50 | 4756 | rsvd_check->rsvd_bits_mask[1][1] = rsvd_bits(13, 21); |
82725b20 DE |
4757 | break; |
4758 | case PT32E_ROOT_LEVEL: | |
5b7f575c SC |
4759 | rsvd_check->rsvd_bits_mask[0][2] = rsvd_bits(63, 63) | |
4760 | high_bits_rsvd | | |
4761 | rsvd_bits(5, 8) | | |
4762 | rsvd_bits(1, 2); /* PDPTE */ | |
4763 | rsvd_check->rsvd_bits_mask[0][1] = high_bits_rsvd; /* PDE */ | |
4764 | rsvd_check->rsvd_bits_mask[0][0] = high_bits_rsvd; /* PTE */ | |
4765 | rsvd_check->rsvd_bits_mask[1][1] = high_bits_rsvd | | |
4766 | rsvd_bits(13, 20); /* large page */ | |
a0a64f50 XG |
4767 | rsvd_check->rsvd_bits_mask[1][0] = |
4768 | rsvd_check->rsvd_bits_mask[0][0]; | |
82725b20 | 4769 | break; |
855feb67 | 4770 | case PT64_ROOT_5LEVEL: |
5b7f575c SC |
4771 | rsvd_check->rsvd_bits_mask[0][4] = high_bits_rsvd | |
4772 | nonleaf_bit8_rsvd | | |
4773 | rsvd_bits(7, 7); | |
855feb67 YZ |
4774 | rsvd_check->rsvd_bits_mask[1][4] = |
4775 | rsvd_check->rsvd_bits_mask[0][4]; | |
df561f66 | 4776 | fallthrough; |
2a7266a8 | 4777 | case PT64_ROOT_4LEVEL: |
5b7f575c SC |
4778 | rsvd_check->rsvd_bits_mask[0][3] = high_bits_rsvd | |
4779 | nonleaf_bit8_rsvd | | |
4780 | rsvd_bits(7, 7); | |
4781 | rsvd_check->rsvd_bits_mask[0][2] = high_bits_rsvd | | |
4782 | gbpages_bit_rsvd; | |
4783 | rsvd_check->rsvd_bits_mask[0][1] = high_bits_rsvd; | |
4784 | rsvd_check->rsvd_bits_mask[0][0] = high_bits_rsvd; | |
a0a64f50 XG |
4785 | rsvd_check->rsvd_bits_mask[1][3] = |
4786 | rsvd_check->rsvd_bits_mask[0][3]; | |
5b7f575c SC |
4787 | rsvd_check->rsvd_bits_mask[1][2] = high_bits_rsvd | |
4788 | gbpages_bit_rsvd | | |
4789 | rsvd_bits(13, 29); | |
4790 | rsvd_check->rsvd_bits_mask[1][1] = high_bits_rsvd | | |
4791 | rsvd_bits(13, 20); /* large page */ | |
a0a64f50 XG |
4792 | rsvd_check->rsvd_bits_mask[1][0] = |
4793 | rsvd_check->rsvd_bits_mask[0][0]; | |
82725b20 DE |
4794 | break; |
4795 | } | |
4796 | } | |
4797 | ||
c919e881 KH |
4798 | static void reset_guest_rsvds_bits_mask(struct kvm_vcpu *vcpu, |
4799 | struct kvm_mmu *context) | |
6dc98b86 | 4800 | { |
b705a277 | 4801 | __reset_rsvds_bits_mask(&context->guest_rsvd_check, |
5b7f575c | 4802 | vcpu->arch.reserved_gpa_bits, |
4d25502a | 4803 | context->cpu_role.base.level, is_efer_nx(context), |
ccf31d6e | 4804 | guest_can_use(vcpu, X86_FEATURE_GBPAGES), |
4e9c0d80 | 4805 | is_cr4_pse(context), |
23493d0a | 4806 | guest_cpuid_is_amd_or_hygon(vcpu)); |
6dc98b86 XG |
4807 | } |
4808 | ||
f3d90f90 SC |
4809 | static void __reset_rsvds_bits_mask_ept(struct rsvd_bits_validate *rsvd_check, |
4810 | u64 pa_bits_rsvd, bool execonly, | |
4811 | int huge_page_level) | |
25d92081 | 4812 | { |
5b7f575c | 4813 | u64 high_bits_rsvd = pa_bits_rsvd & rsvd_bits(0, 51); |
84ea5c09 | 4814 | u64 large_1g_rsvd = 0, large_2m_rsvd = 0; |
951f9fd7 | 4815 | u64 bad_mt_xwr; |
25d92081 | 4816 | |
84ea5c09 LJ |
4817 | if (huge_page_level < PG_LEVEL_1G) |
4818 | large_1g_rsvd = rsvd_bits(7, 7); | |
4819 | if (huge_page_level < PG_LEVEL_2M) | |
4820 | large_2m_rsvd = rsvd_bits(7, 7); | |
4821 | ||
5b7f575c SC |
4822 | rsvd_check->rsvd_bits_mask[0][4] = high_bits_rsvd | rsvd_bits(3, 7); |
4823 | rsvd_check->rsvd_bits_mask[0][3] = high_bits_rsvd | rsvd_bits(3, 7); | |
84ea5c09 LJ |
4824 | rsvd_check->rsvd_bits_mask[0][2] = high_bits_rsvd | rsvd_bits(3, 6) | large_1g_rsvd; |
4825 | rsvd_check->rsvd_bits_mask[0][1] = high_bits_rsvd | rsvd_bits(3, 6) | large_2m_rsvd; | |
5b7f575c | 4826 | rsvd_check->rsvd_bits_mask[0][0] = high_bits_rsvd; |
25d92081 YZ |
4827 | |
4828 | /* large page */ | |
855feb67 | 4829 | rsvd_check->rsvd_bits_mask[1][4] = rsvd_check->rsvd_bits_mask[0][4]; |
a0a64f50 | 4830 | rsvd_check->rsvd_bits_mask[1][3] = rsvd_check->rsvd_bits_mask[0][3]; |
84ea5c09 LJ |
4831 | rsvd_check->rsvd_bits_mask[1][2] = high_bits_rsvd | rsvd_bits(12, 29) | large_1g_rsvd; |
4832 | rsvd_check->rsvd_bits_mask[1][1] = high_bits_rsvd | rsvd_bits(12, 20) | large_2m_rsvd; | |
a0a64f50 | 4833 | rsvd_check->rsvd_bits_mask[1][0] = rsvd_check->rsvd_bits_mask[0][0]; |
25d92081 | 4834 | |
951f9fd7 PB |
4835 | bad_mt_xwr = 0xFFull << (2 * 8); /* bits 3..5 must not be 2 */ |
4836 | bad_mt_xwr |= 0xFFull << (3 * 8); /* bits 3..5 must not be 3 */ | |
4837 | bad_mt_xwr |= 0xFFull << (7 * 8); /* bits 3..5 must not be 7 */ | |
4838 | bad_mt_xwr |= REPEAT_BYTE(1ull << 2); /* bits 0..2 must not be 010 */ | |
4839 | bad_mt_xwr |= REPEAT_BYTE(1ull << 6); /* bits 0..2 must not be 110 */ | |
4840 | if (!execonly) { | |
4841 | /* bits 0..2 must not be 100 unless VMX capabilities allow it */ | |
4842 | bad_mt_xwr |= REPEAT_BYTE(1ull << 4); | |
25d92081 | 4843 | } |
951f9fd7 | 4844 | rsvd_check->bad_mt_xwr = bad_mt_xwr; |
25d92081 YZ |
4845 | } |
4846 | ||
81b8eebb | 4847 | static void reset_rsvds_bits_mask_ept(struct kvm_vcpu *vcpu, |
84ea5c09 | 4848 | struct kvm_mmu *context, bool execonly, int huge_page_level) |
81b8eebb XG |
4849 | { |
4850 | __reset_rsvds_bits_mask_ept(&context->guest_rsvd_check, | |
84ea5c09 LJ |
4851 | vcpu->arch.reserved_gpa_bits, execonly, |
4852 | huge_page_level); | |
81b8eebb XG |
4853 | } |
4854 | ||
6f8e65a6 SC |
4855 | static inline u64 reserved_hpa_bits(void) |
4856 | { | |
4857 | return rsvd_bits(shadow_phys_bits, 63); | |
4858 | } | |
4859 | ||
c258b62b XG |
4860 | /* |
4861 | * the page table on host is the shadow page table for the page | |
4862 | * table in guest or amd nested guest, its mmu features completely | |
4863 | * follow the features in guest. | |
4864 | */ | |
16be1d12 SC |
4865 | static void reset_shadow_zero_bits_mask(struct kvm_vcpu *vcpu, |
4866 | struct kvm_mmu *context) | |
c258b62b | 4867 | { |
8c985b2d SC |
4868 | /* @amd adds a check on bit of SPTEs, which KVM shouldn't use anyways. */ |
4869 | bool is_amd = true; | |
4870 | /* KVM doesn't use 2-level page tables for the shadow MMU. */ | |
4871 | bool is_pse = false; | |
ea2800dd BS |
4872 | struct rsvd_bits_validate *shadow_zero_check; |
4873 | int i; | |
5f0b8199 | 4874 | |
a972e29c | 4875 | WARN_ON_ONCE(context->root_role.level < PT32E_ROOT_LEVEL); |
8c985b2d | 4876 | |
ea2800dd | 4877 | shadow_zero_check = &context->shadow_zero_check; |
b705a277 | 4878 | __reset_rsvds_bits_mask(shadow_zero_check, reserved_hpa_bits(), |
a972e29c | 4879 | context->root_role.level, |
7a458f0e | 4880 | context->root_role.efer_nx, |
ccf31d6e SC |
4881 | guest_can_use(vcpu, X86_FEATURE_GBPAGES), |
4882 | is_pse, is_amd); | |
ea2800dd BS |
4883 | |
4884 | if (!shadow_me_mask) | |
4885 | return; | |
4886 | ||
a972e29c | 4887 | for (i = context->root_role.level; --i >= 0;) { |
e54f1ff2 KH |
4888 | /* |
4889 | * So far shadow_me_value is a constant during KVM's life | |
4890 | * time. Bits in shadow_me_value are allowed to be set. | |
4891 | * Bits in shadow_me_mask but not in shadow_me_value are | |
4892 | * not allowed to be set. | |
4893 | */ | |
4894 | shadow_zero_check->rsvd_bits_mask[0][i] |= shadow_me_mask; | |
4895 | shadow_zero_check->rsvd_bits_mask[1][i] |= shadow_me_mask; | |
4896 | shadow_zero_check->rsvd_bits_mask[0][i] &= ~shadow_me_value; | |
4897 | shadow_zero_check->rsvd_bits_mask[1][i] &= ~shadow_me_value; | |
ea2800dd BS |
4898 | } |
4899 | ||
c258b62b | 4900 | } |
c258b62b | 4901 | |
6fec2144 PB |
4902 | static inline bool boot_cpu_is_amd(void) |
4903 | { | |
4904 | WARN_ON_ONCE(!tdp_enabled); | |
4905 | return shadow_x_mask == 0; | |
4906 | } | |
4907 | ||
c258b62b XG |
4908 | /* |
4909 | * the direct page table on host, use as much mmu features as | |
4910 | * possible, however, kvm currently does not do execution-protection. | |
4911 | */ | |
f3d90f90 | 4912 | static void reset_tdp_shadow_zero_bits_mask(struct kvm_mmu *context) |
c258b62b | 4913 | { |
ea2800dd BS |
4914 | struct rsvd_bits_validate *shadow_zero_check; |
4915 | int i; | |
4916 | ||
4917 | shadow_zero_check = &context->shadow_zero_check; | |
4918 | ||
6fec2144 | 4919 | if (boot_cpu_is_amd()) |
b705a277 | 4920 | __reset_rsvds_bits_mask(shadow_zero_check, reserved_hpa_bits(), |
6c6ab524 | 4921 | context->root_role.level, true, |
b8291adc | 4922 | boot_cpu_has(X86_FEATURE_GBPAGES), |
8c985b2d | 4923 | false, true); |
c258b62b | 4924 | else |
ea2800dd | 4925 | __reset_rsvds_bits_mask_ept(shadow_zero_check, |
84ea5c09 LJ |
4926 | reserved_hpa_bits(), false, |
4927 | max_huge_page_level); | |
c258b62b | 4928 | |
ea2800dd BS |
4929 | if (!shadow_me_mask) |
4930 | return; | |
4931 | ||
a972e29c | 4932 | for (i = context->root_role.level; --i >= 0;) { |
ea2800dd BS |
4933 | shadow_zero_check->rsvd_bits_mask[0][i] &= ~shadow_me_mask; |
4934 | shadow_zero_check->rsvd_bits_mask[1][i] &= ~shadow_me_mask; | |
4935 | } | |
c258b62b XG |
4936 | } |
4937 | ||
4938 | /* | |
4939 | * as the comments in reset_shadow_zero_bits_mask() except it | |
4940 | * is the shadow page table for intel nested guest. | |
4941 | */ | |
4942 | static void | |
e8f6e738 | 4943 | reset_ept_shadow_zero_bits_mask(struct kvm_mmu *context, bool execonly) |
c258b62b XG |
4944 | { |
4945 | __reset_rsvds_bits_mask_ept(&context->shadow_zero_check, | |
84ea5c09 LJ |
4946 | reserved_hpa_bits(), execonly, |
4947 | max_huge_page_level); | |
c258b62b XG |
4948 | } |
4949 | ||
09f037aa PB |
4950 | #define BYTE_MASK(access) \ |
4951 | ((1 & (access) ? 2 : 0) | \ | |
4952 | (2 & (access) ? 4 : 0) | \ | |
4953 | (3 & (access) ? 8 : 0) | \ | |
4954 | (4 & (access) ? 16 : 0) | \ | |
4955 | (5 & (access) ? 32 : 0) | \ | |
4956 | (6 & (access) ? 64 : 0) | \ | |
4957 | (7 & (access) ? 128 : 0)) | |
4958 | ||
4959 | ||
c596f147 | 4960 | static void update_permission_bitmask(struct kvm_mmu *mmu, bool ept) |
97d64b78 | 4961 | { |
09f037aa PB |
4962 | unsigned byte; |
4963 | ||
4964 | const u8 x = BYTE_MASK(ACC_EXEC_MASK); | |
4965 | const u8 w = BYTE_MASK(ACC_WRITE_MASK); | |
4966 | const u8 u = BYTE_MASK(ACC_USER_MASK); | |
4967 | ||
c596f147 SC |
4968 | bool cr4_smep = is_cr4_smep(mmu); |
4969 | bool cr4_smap = is_cr4_smap(mmu); | |
4970 | bool cr0_wp = is_cr0_wp(mmu); | |
90599c28 | 4971 | bool efer_nx = is_efer_nx(mmu); |
97d64b78 | 4972 | |
97d64b78 | 4973 | for (byte = 0; byte < ARRAY_SIZE(mmu->permissions); ++byte) { |
09f037aa PB |
4974 | unsigned pfec = byte << 1; |
4975 | ||
97ec8c06 | 4976 | /* |
09f037aa PB |
4977 | * Each "*f" variable has a 1 bit for each UWX value |
4978 | * that causes a fault with the given PFEC. | |
97ec8c06 | 4979 | */ |
97d64b78 | 4980 | |
09f037aa | 4981 | /* Faults from writes to non-writable pages */ |
a6a6d3b1 | 4982 | u8 wf = (pfec & PFERR_WRITE_MASK) ? (u8)~w : 0; |
09f037aa | 4983 | /* Faults from user mode accesses to supervisor pages */ |
a6a6d3b1 | 4984 | u8 uf = (pfec & PFERR_USER_MASK) ? (u8)~u : 0; |
09f037aa | 4985 | /* Faults from fetches of non-executable pages*/ |
a6a6d3b1 | 4986 | u8 ff = (pfec & PFERR_FETCH_MASK) ? (u8)~x : 0; |
09f037aa PB |
4987 | /* Faults from kernel mode fetches of user pages */ |
4988 | u8 smepf = 0; | |
4989 | /* Faults from kernel mode accesses of user pages */ | |
4990 | u8 smapf = 0; | |
4991 | ||
4992 | if (!ept) { | |
4993 | /* Faults from kernel mode accesses to user pages */ | |
4994 | u8 kf = (pfec & PFERR_USER_MASK) ? 0 : u; | |
4995 | ||
4996 | /* Not really needed: !nx will cause pte.nx to fault */ | |
90599c28 | 4997 | if (!efer_nx) |
09f037aa PB |
4998 | ff = 0; |
4999 | ||
5000 | /* Allow supervisor writes if !cr0.wp */ | |
5001 | if (!cr0_wp) | |
5002 | wf = (pfec & PFERR_USER_MASK) ? wf : 0; | |
5003 | ||
5004 | /* Disallow supervisor fetches of user code if cr4.smep */ | |
5005 | if (cr4_smep) | |
5006 | smepf = (pfec & PFERR_FETCH_MASK) ? kf : 0; | |
5007 | ||
5008 | /* | |
5009 | * SMAP:kernel-mode data accesses from user-mode | |
5010 | * mappings should fault. A fault is considered | |
5011 | * as a SMAP violation if all of the following | |
39337ad1 | 5012 | * conditions are true: |
09f037aa PB |
5013 | * - X86_CR4_SMAP is set in CR4 |
5014 | * - A user page is accessed | |
5015 | * - The access is not a fetch | |
4f4aa80e LJ |
5016 | * - The access is supervisor mode |
5017 | * - If implicit supervisor access or X86_EFLAGS_AC is clear | |
09f037aa | 5018 | * |
94b4a2f1 LJ |
5019 | * Here, we cover the first four conditions. |
5020 | * The fifth is computed dynamically in permission_fault(); | |
09f037aa PB |
5021 | * PFERR_RSVD_MASK bit will be set in PFEC if the access is |
5022 | * *not* subject to SMAP restrictions. | |
5023 | */ | |
5024 | if (cr4_smap) | |
5025 | smapf = (pfec & (PFERR_RSVD_MASK|PFERR_FETCH_MASK)) ? 0 : kf; | |
97d64b78 | 5026 | } |
09f037aa PB |
5027 | |
5028 | mmu->permissions[byte] = ff | uf | wf | smepf | smapf; | |
97d64b78 AK |
5029 | } |
5030 | } | |
5031 | ||
2d344105 HH |
5032 | /* |
5033 | * PKU is an additional mechanism by which the paging controls access to | |
5034 | * user-mode addresses based on the value in the PKRU register. Protection | |
5035 | * key violations are reported through a bit in the page fault error code. | |
5036 | * Unlike other bits of the error code, the PK bit is not known at the | |
5037 | * call site of e.g. gva_to_gpa; it must be computed directly in | |
5038 | * permission_fault based on two bits of PKRU, on some machine state (CR4, | |
5039 | * CR0, EFER, CPL), and on other bits of the error code and the page tables. | |
5040 | * | |
5041 | * In particular the following conditions come from the error code, the | |
5042 | * page tables and the machine state: | |
5043 | * - PK is always zero unless CR4.PKE=1 and EFER.LMA=1 | |
5044 | * - PK is always zero if RSVD=1 (reserved bit set) or F=1 (instruction fetch) | |
5045 | * - PK is always zero if U=0 in the page tables | |
5046 | * - PKRU.WD is ignored if CR0.WP=0 and the access is a supervisor access. | |
5047 | * | |
5048 | * The PKRU bitmask caches the result of these four conditions. The error | |
5049 | * code (minus the P bit) and the page table's U bit form an index into the | |
5050 | * PKRU bitmask. Two bits of the PKRU bitmask are then extracted and ANDed | |
5051 | * with the two bits of the PKRU register corresponding to the protection key. | |
5052 | * For the first three conditions above the bits will be 00, thus masking | |
5053 | * away both AD and WD. For all reads or if the last condition holds, WD | |
5054 | * only will be masked away. | |
5055 | */ | |
2e4c0661 | 5056 | static void update_pkru_bitmask(struct kvm_mmu *mmu) |
2d344105 HH |
5057 | { |
5058 | unsigned bit; | |
5059 | bool wp; | |
5060 | ||
a3ca5281 CQ |
5061 | mmu->pkru_mask = 0; |
5062 | ||
5063 | if (!is_cr4_pke(mmu)) | |
2d344105 | 5064 | return; |
2d344105 | 5065 | |
2e4c0661 | 5066 | wp = is_cr0_wp(mmu); |
2d344105 HH |
5067 | |
5068 | for (bit = 0; bit < ARRAY_SIZE(mmu->permissions); ++bit) { | |
5069 | unsigned pfec, pkey_bits; | |
5070 | bool check_pkey, check_write, ff, uf, wf, pte_user; | |
5071 | ||
5072 | pfec = bit << 1; | |
5073 | ff = pfec & PFERR_FETCH_MASK; | |
5074 | uf = pfec & PFERR_USER_MASK; | |
5075 | wf = pfec & PFERR_WRITE_MASK; | |
5076 | ||
5077 | /* PFEC.RSVD is replaced by ACC_USER_MASK. */ | |
5078 | pte_user = pfec & PFERR_RSVD_MASK; | |
5079 | ||
5080 | /* | |
5081 | * Only need to check the access which is not an | |
5082 | * instruction fetch and is to a user page. | |
5083 | */ | |
5084 | check_pkey = (!ff && pte_user); | |
5085 | /* | |
5086 | * write access is controlled by PKRU if it is a | |
5087 | * user access or CR0.WP = 1. | |
5088 | */ | |
5089 | check_write = check_pkey && wf && (uf || wp); | |
5090 | ||
5091 | /* PKRU.AD stops both read and write access. */ | |
5092 | pkey_bits = !!check_pkey; | |
5093 | /* PKRU.WD stops write access. */ | |
5094 | pkey_bits |= (!!check_write) << 1; | |
5095 | ||
5096 | mmu->pkru_mask |= (pkey_bits & 3) << pfec; | |
5097 | } | |
5098 | } | |
5099 | ||
533f9a4b SC |
5100 | static void reset_guest_paging_metadata(struct kvm_vcpu *vcpu, |
5101 | struct kvm_mmu *mmu) | |
6fd01b71 | 5102 | { |
533f9a4b SC |
5103 | if (!is_cr0_pg(mmu)) |
5104 | return; | |
6bb69c9b | 5105 | |
c919e881 | 5106 | reset_guest_rsvds_bits_mask(vcpu, mmu); |
533f9a4b SC |
5107 | update_permission_bitmask(mmu, false); |
5108 | update_pkru_bitmask(mmu); | |
6fd01b71 AK |
5109 | } |
5110 | ||
fe660f72 | 5111 | static void paging64_init_context(struct kvm_mmu *context) |
6aa8b732 | 5112 | { |
6aa8b732 | 5113 | context->page_fault = paging64_page_fault; |
6aa8b732 | 5114 | context->gva_to_gpa = paging64_gva_to_gpa; |
c3c6c9fc | 5115 | context->sync_spte = paging64_sync_spte; |
6aa8b732 AK |
5116 | } |
5117 | ||
84a16226 | 5118 | static void paging32_init_context(struct kvm_mmu *context) |
6aa8b732 | 5119 | { |
6aa8b732 | 5120 | context->page_fault = paging32_page_fault; |
6aa8b732 | 5121 | context->gva_to_gpa = paging32_gva_to_gpa; |
c3c6c9fc | 5122 | context->sync_spte = paging32_sync_spte; |
6aa8b732 AK |
5123 | } |
5124 | ||
f3d90f90 SC |
5125 | static union kvm_cpu_role kvm_calc_cpu_role(struct kvm_vcpu *vcpu, |
5126 | const struct kvm_mmu_role_regs *regs) | |
e5ed0fb0 | 5127 | { |
7a7ae829 | 5128 | union kvm_cpu_role role = {0}; |
e5ed0fb0 PB |
5129 | |
5130 | role.base.access = ACC_ALL; | |
5131 | role.base.smm = is_smm(vcpu); | |
5132 | role.base.guest_mode = is_guest_mode(vcpu); | |
5133 | role.ext.valid = 1; | |
5134 | ||
5135 | if (!____is_cr0_pg(regs)) { | |
5136 | role.base.direct = 1; | |
5137 | return role; | |
5138 | } | |
5139 | ||
5140 | role.base.efer_nx = ____is_efer_nx(regs); | |
5141 | role.base.cr0_wp = ____is_cr0_wp(regs); | |
5142 | role.base.smep_andnot_wp = ____is_cr4_smep(regs) && !____is_cr0_wp(regs); | |
5143 | role.base.smap_andnot_wp = ____is_cr4_smap(regs) && !____is_cr0_wp(regs); | |
5144 | role.base.has_4_byte_gpte = !____is_cr4_pae(regs); | |
60f3cb60 PB |
5145 | |
5146 | if (____is_efer_lma(regs)) | |
5147 | role.base.level = ____is_cr4_la57(regs) ? PT64_ROOT_5LEVEL | |
5148 | : PT64_ROOT_4LEVEL; | |
5149 | else if (____is_cr4_pae(regs)) | |
5150 | role.base.level = PT32E_ROOT_LEVEL; | |
5151 | else | |
5152 | role.base.level = PT32_ROOT_LEVEL; | |
e5ed0fb0 | 5153 | |
e5ed0fb0 PB |
5154 | role.ext.cr4_smep = ____is_cr4_smep(regs); |
5155 | role.ext.cr4_smap = ____is_cr4_smap(regs); | |
5156 | role.ext.cr4_pse = ____is_cr4_pse(regs); | |
5157 | ||
5158 | /* PKEY and LA57 are active iff long mode is active. */ | |
5159 | role.ext.cr4_pke = ____is_efer_lma(regs) && ____is_cr4_pke(regs); | |
5160 | role.ext.cr4_la57 = ____is_efer_lma(regs) && ____is_cr4_la57(regs); | |
5161 | role.ext.efer_lma = ____is_efer_lma(regs); | |
5162 | return role; | |
5163 | } | |
5164 | ||
cf9f4c0e SC |
5165 | void __kvm_mmu_refresh_passthrough_bits(struct kvm_vcpu *vcpu, |
5166 | struct kvm_mmu *mmu) | |
5167 | { | |
5168 | const bool cr0_wp = kvm_is_cr0_bit_set(vcpu, X86_CR0_WP); | |
5169 | ||
5170 | BUILD_BUG_ON((KVM_MMU_CR0_ROLE_BITS & KVM_POSSIBLE_CR0_GUEST_BITS) != X86_CR0_WP); | |
5171 | BUILD_BUG_ON((KVM_MMU_CR4_ROLE_BITS & KVM_POSSIBLE_CR4_GUEST_BITS)); | |
5172 | ||
5173 | if (is_cr0_wp(mmu) == cr0_wp) | |
5174 | return; | |
5175 | ||
5176 | mmu->cpu_role.base.cr0_wp = cr0_wp; | |
5177 | reset_guest_paging_metadata(vcpu, mmu); | |
5178 | } | |
5179 | ||
d468d94b SC |
5180 | static inline int kvm_mmu_get_tdp_level(struct kvm_vcpu *vcpu) |
5181 | { | |
746700d2 WH |
5182 | /* tdp_root_level is architecture forced level, use it if nonzero */ |
5183 | if (tdp_root_level) | |
5184 | return tdp_root_level; | |
5185 | ||
d468d94b | 5186 | /* Use 5-level TDP if and only if it's useful/necessary. */ |
83013059 | 5187 | if (max_tdp_level == 5 && cpuid_maxphyaddr(vcpu) <= 48) |
d468d94b SC |
5188 | return 4; |
5189 | ||
83013059 | 5190 | return max_tdp_level; |
d468d94b SC |
5191 | } |
5192 | ||
7a458f0e | 5193 | static union kvm_mmu_page_role |
8626c120 | 5194 | kvm_calc_tdp_mmu_root_page_role(struct kvm_vcpu *vcpu, |
7a7ae829 | 5195 | union kvm_cpu_role cpu_role) |
9fa72119 | 5196 | { |
7a458f0e | 5197 | union kvm_mmu_page_role role = {0}; |
9fa72119 | 5198 | |
7a458f0e PB |
5199 | role.access = ACC_ALL; |
5200 | role.cr0_wp = true; | |
5201 | role.efer_nx = true; | |
5202 | role.smm = cpu_role.base.smm; | |
5203 | role.guest_mode = cpu_role.base.guest_mode; | |
54275f74 | 5204 | role.ad_disabled = !kvm_ad_enabled(); |
7a458f0e PB |
5205 | role.level = kvm_mmu_get_tdp_level(vcpu); |
5206 | role.direct = true; | |
5207 | role.has_4_byte_gpte = false; | |
9fa72119 JS |
5208 | |
5209 | return role; | |
5210 | } | |
5211 | ||
39e7e2bf | 5212 | static void init_kvm_tdp_mmu(struct kvm_vcpu *vcpu, |
a7f1de9b | 5213 | union kvm_cpu_role cpu_role) |
fb72d167 | 5214 | { |
8c008659 | 5215 | struct kvm_mmu *context = &vcpu->arch.root_mmu; |
7a458f0e | 5216 | union kvm_mmu_page_role root_role = kvm_calc_tdp_mmu_root_page_role(vcpu, cpu_role); |
fb72d167 | 5217 | |
e5ed0fb0 | 5218 | if (cpu_role.as_u64 == context->cpu_role.as_u64 && |
7a458f0e | 5219 | root_role.word == context->root_role.word) |
7dcd5755 VK |
5220 | return; |
5221 | ||
e5ed0fb0 | 5222 | context->cpu_role.as_u64 = cpu_role.as_u64; |
7a458f0e | 5223 | context->root_role.word = root_role.word; |
7a02674d | 5224 | context->page_fault = kvm_tdp_page_fault; |
c3c6c9fc | 5225 | context->sync_spte = NULL; |
2fdcc1b3 | 5226 | context->get_guest_pgd = get_guest_cr3; |
e4e517b4 | 5227 | context->get_pdptr = kvm_pdptr_read; |
cb659db8 | 5228 | context->inject_page_fault = kvm_inject_page_fault; |
fb72d167 | 5229 | |
36f26787 | 5230 | if (!is_cr0_pg(context)) |
fb72d167 | 5231 | context->gva_to_gpa = nonpaging_gva_to_gpa; |
36f26787 | 5232 | else if (is_cr4_pae(context)) |
4d6931c3 | 5233 | context->gva_to_gpa = paging64_gva_to_gpa; |
f4bd6f73 | 5234 | else |
4d6931c3 | 5235 | context->gva_to_gpa = paging32_gva_to_gpa; |
fb72d167 | 5236 | |
533f9a4b | 5237 | reset_guest_paging_metadata(vcpu, context); |
e8f6e738 | 5238 | reset_tdp_shadow_zero_bits_mask(context); |
fb72d167 JR |
5239 | } |
5240 | ||
8c008659 | 5241 | static void shadow_mmu_init_context(struct kvm_vcpu *vcpu, struct kvm_mmu *context, |
7a7ae829 | 5242 | union kvm_cpu_role cpu_role, |
7a458f0e | 5243 | union kvm_mmu_page_role root_role) |
9fa72119 | 5244 | { |
e5ed0fb0 | 5245 | if (cpu_role.as_u64 == context->cpu_role.as_u64 && |
7a458f0e | 5246 | root_role.word == context->root_role.word) |
18db1b17 | 5247 | return; |
a770f6f2 | 5248 | |
e5ed0fb0 | 5249 | context->cpu_role.as_u64 = cpu_role.as_u64; |
7a458f0e | 5250 | context->root_role.word = root_role.word; |
18db1b17 | 5251 | |
36f26787 | 5252 | if (!is_cr0_pg(context)) |
84a16226 | 5253 | nonpaging_init_context(context); |
36f26787 | 5254 | else if (is_cr4_pae(context)) |
fe660f72 | 5255 | paging64_init_context(context); |
6aa8b732 | 5256 | else |
84a16226 | 5257 | paging32_init_context(context); |
a770f6f2 | 5258 | |
533f9a4b | 5259 | reset_guest_paging_metadata(vcpu, context); |
c258b62b | 5260 | reset_shadow_zero_bits_mask(vcpu, context); |
52fde8df | 5261 | } |
0f04a2ac | 5262 | |
594e91a1 | 5263 | static void kvm_init_shadow_mmu(struct kvm_vcpu *vcpu, |
a7f1de9b | 5264 | union kvm_cpu_role cpu_role) |
0f04a2ac | 5265 | { |
8c008659 | 5266 | struct kvm_mmu *context = &vcpu->arch.root_mmu; |
56b321f9 | 5267 | union kvm_mmu_page_role root_role; |
0f04a2ac | 5268 | |
56b321f9 | 5269 | root_role = cpu_role.base; |
0f04a2ac | 5270 | |
56b321f9 PB |
5271 | /* KVM uses PAE paging whenever the guest isn't using 64-bit paging. */ |
5272 | root_role.level = max_t(u32, root_role.level, PT32E_ROOT_LEVEL); | |
59505b55 | 5273 | |
56b321f9 PB |
5274 | /* |
5275 | * KVM forces EFER.NX=1 when TDP is disabled, reflect it in the MMU role. | |
5276 | * KVM uses NX when TDP is disabled to handle a variety of scenarios, | |
5277 | * notably for huge SPTEs if iTLB multi-hit mitigation is enabled and | |
5278 | * to generate correct permissions for CR0.WP=0/CR4.SMEP=1/EFER.NX=0. | |
5279 | * The iTLB multi-hit workaround can be toggled at any time, so assume | |
5280 | * NX can be used by any non-nested shadow MMU to avoid having to reset | |
5281 | * MMU contexts. | |
5282 | */ | |
5283 | root_role.efer_nx = true; | |
5284 | ||
5285 | shadow_mmu_init_context(vcpu, context, cpu_role, root_role); | |
59505b55 SC |
5286 | } |
5287 | ||
dbc4739b SC |
5288 | void kvm_init_shadow_npt_mmu(struct kvm_vcpu *vcpu, unsigned long cr0, |
5289 | unsigned long cr4, u64 efer, gpa_t nested_cr3) | |
0f04a2ac | 5290 | { |
8c008659 | 5291 | struct kvm_mmu *context = &vcpu->arch.guest_mmu; |
594e91a1 SC |
5292 | struct kvm_mmu_role_regs regs = { |
5293 | .cr0 = cr0, | |
28f091bc | 5294 | .cr4 = cr4 & ~X86_CR4_PKE, |
594e91a1 SC |
5295 | .efer = efer, |
5296 | }; | |
7a7ae829 | 5297 | union kvm_cpu_role cpu_role = kvm_calc_cpu_role(vcpu, ®s); |
56b321f9 PB |
5298 | union kvm_mmu_page_role root_role; |
5299 | ||
5300 | /* NPT requires CR0.PG=1. */ | |
5301 | WARN_ON_ONCE(cpu_role.base.direct); | |
5302 | ||
5303 | root_role = cpu_role.base; | |
5304 | root_role.level = kvm_mmu_get_tdp_level(vcpu); | |
84e5ffd0 LJ |
5305 | if (root_role.level == PT64_ROOT_5LEVEL && |
5306 | cpu_role.base.level == PT64_ROOT_4LEVEL) | |
5307 | root_role.passthrough = 1; | |
a506fdd2 | 5308 | |
7a458f0e | 5309 | shadow_mmu_init_context(vcpu, context, cpu_role, root_role); |
d2e5f333 | 5310 | kvm_mmu_new_pgd(vcpu, nested_cr3); |
0f04a2ac VK |
5311 | } |
5312 | EXPORT_SYMBOL_GPL(kvm_init_shadow_npt_mmu); | |
52fde8df | 5313 | |
7a7ae829 | 5314 | static union kvm_cpu_role |
a336282d | 5315 | kvm_calc_shadow_ept_root_page_role(struct kvm_vcpu *vcpu, bool accessed_dirty, |
bb1fcc70 | 5316 | bool execonly, u8 level) |
9fa72119 | 5317 | { |
7a7ae829 | 5318 | union kvm_cpu_role role = {0}; |
14c07ad8 | 5319 | |
daed87b8 PB |
5320 | /* |
5321 | * KVM does not support SMM transfer monitors, and consequently does not | |
5322 | * support the "entry to SMM" control either. role.base.smm is always 0. | |
5323 | */ | |
5324 | WARN_ON_ONCE(is_smm(vcpu)); | |
bb1fcc70 | 5325 | role.base.level = level; |
bb3b394d | 5326 | role.base.has_4_byte_gpte = false; |
a336282d VK |
5327 | role.base.direct = false; |
5328 | role.base.ad_disabled = !accessed_dirty; | |
5329 | role.base.guest_mode = true; | |
5330 | role.base.access = ACC_ALL; | |
9fa72119 | 5331 | |
cd6767c3 | 5332 | role.ext.word = 0; |
a336282d | 5333 | role.ext.execonly = execonly; |
cd6767c3 | 5334 | role.ext.valid = 1; |
9fa72119 JS |
5335 | |
5336 | return role; | |
5337 | } | |
5338 | ||
ae1e2d10 | 5339 | void kvm_init_shadow_ept_mmu(struct kvm_vcpu *vcpu, bool execonly, |
cc022ae1 LJ |
5340 | int huge_page_level, bool accessed_dirty, |
5341 | gpa_t new_eptp) | |
155a97a3 | 5342 | { |
8c008659 | 5343 | struct kvm_mmu *context = &vcpu->arch.guest_mmu; |
bb1fcc70 | 5344 | u8 level = vmx_eptp_page_walk_level(new_eptp); |
7a7ae829 | 5345 | union kvm_cpu_role new_mode = |
a336282d | 5346 | kvm_calc_shadow_ept_root_page_role(vcpu, accessed_dirty, |
bb1fcc70 | 5347 | execonly, level); |
a336282d | 5348 | |
e5ed0fb0 PB |
5349 | if (new_mode.as_u64 != context->cpu_role.as_u64) { |
5350 | /* EPT, and thus nested EPT, does not consume CR0, CR4, nor EFER. */ | |
5351 | context->cpu_role.as_u64 = new_mode.as_u64; | |
7a458f0e | 5352 | context->root_role.word = new_mode.base.word; |
3cffc89d | 5353 | |
3cffc89d PB |
5354 | context->page_fault = ept_page_fault; |
5355 | context->gva_to_gpa = ept_gva_to_gpa; | |
c3c6c9fc | 5356 | context->sync_spte = ept_sync_spte; |
347a0d0d | 5357 | |
3cffc89d PB |
5358 | update_permission_bitmask(context, true); |
5359 | context->pkru_mask = 0; | |
5360 | reset_rsvds_bits_mask_ept(vcpu, context, execonly, huge_page_level); | |
5361 | reset_ept_shadow_zero_bits_mask(context, execonly); | |
5362 | } | |
3dc773e7 | 5363 | |
d2e5f333 | 5364 | kvm_mmu_new_pgd(vcpu, new_eptp); |
155a97a3 NHE |
5365 | } |
5366 | EXPORT_SYMBOL_GPL(kvm_init_shadow_ept_mmu); | |
5367 | ||
39e7e2bf | 5368 | static void init_kvm_softmmu(struct kvm_vcpu *vcpu, |
a7f1de9b | 5369 | union kvm_cpu_role cpu_role) |
52fde8df | 5370 | { |
8c008659 | 5371 | struct kvm_mmu *context = &vcpu->arch.root_mmu; |
ad896af0 | 5372 | |
a7f1de9b | 5373 | kvm_init_shadow_mmu(vcpu, cpu_role); |
929d1cfa | 5374 | |
2fdcc1b3 | 5375 | context->get_guest_pgd = get_guest_cr3; |
ad896af0 PB |
5376 | context->get_pdptr = kvm_pdptr_read; |
5377 | context->inject_page_fault = kvm_inject_page_fault; | |
6aa8b732 AK |
5378 | } |
5379 | ||
39e7e2bf | 5380 | static void init_kvm_nested_mmu(struct kvm_vcpu *vcpu, |
a7f1de9b | 5381 | union kvm_cpu_role new_mode) |
02f59dc9 JR |
5382 | { |
5383 | struct kvm_mmu *g_context = &vcpu->arch.nested_mmu; | |
5384 | ||
e5ed0fb0 | 5385 | if (new_mode.as_u64 == g_context->cpu_role.as_u64) |
bf627a92 VK |
5386 | return; |
5387 | ||
e5ed0fb0 | 5388 | g_context->cpu_role.as_u64 = new_mode.as_u64; |
2fdcc1b3 | 5389 | g_context->get_guest_pgd = get_guest_cr3; |
e4e517b4 | 5390 | g_context->get_pdptr = kvm_pdptr_read; |
02f59dc9 JR |
5391 | g_context->inject_page_fault = kvm_inject_page_fault; |
5392 | ||
5efac074 PB |
5393 | /* |
5394 | * L2 page tables are never shadowed, so there is no need to sync | |
5395 | * SPTEs. | |
5396 | */ | |
9fd4a4e3 | 5397 | g_context->sync_spte = NULL; |
5efac074 | 5398 | |
02f59dc9 | 5399 | /* |
44dd3ffa | 5400 | * Note that arch.mmu->gva_to_gpa translates l2_gpa to l1_gpa using |
0af2593b DM |
5401 | * L1's nested page tables (e.g. EPT12). The nested translation |
5402 | * of l2_gva to l1_gpa is done by arch.nested_mmu.gva_to_gpa using | |
5403 | * L2's page tables as the first level of translation and L1's | |
5404 | * nested page tables as the second level of translation. Basically | |
5405 | * the gva_to_gpa functions between mmu and nested_mmu are swapped. | |
02f59dc9 | 5406 | */ |
fa4b5588 | 5407 | if (!is_paging(vcpu)) |
1f5a21ee | 5408 | g_context->gva_to_gpa = nonpaging_gva_to_gpa; |
fa4b5588 | 5409 | else if (is_long_mode(vcpu)) |
1f5a21ee | 5410 | g_context->gva_to_gpa = paging64_gva_to_gpa; |
fa4b5588 | 5411 | else if (is_pae(vcpu)) |
1f5a21ee | 5412 | g_context->gva_to_gpa = paging64_gva_to_gpa; |
fa4b5588 | 5413 | else |
1f5a21ee | 5414 | g_context->gva_to_gpa = paging32_gva_to_gpa; |
02f59dc9 | 5415 | |
533f9a4b | 5416 | reset_guest_paging_metadata(vcpu, g_context); |
02f59dc9 JR |
5417 | } |
5418 | ||
c9060662 | 5419 | void kvm_init_mmu(struct kvm_vcpu *vcpu) |
fb72d167 | 5420 | { |
39e7e2bf | 5421 | struct kvm_mmu_role_regs regs = vcpu_to_role_regs(vcpu); |
a7f1de9b | 5422 | union kvm_cpu_role cpu_role = kvm_calc_cpu_role(vcpu, ®s); |
39e7e2bf | 5423 | |
02f59dc9 | 5424 | if (mmu_is_nested(vcpu)) |
a7f1de9b | 5425 | init_kvm_nested_mmu(vcpu, cpu_role); |
02f59dc9 | 5426 | else if (tdp_enabled) |
a7f1de9b | 5427 | init_kvm_tdp_mmu(vcpu, cpu_role); |
fb72d167 | 5428 | else |
a7f1de9b | 5429 | init_kvm_softmmu(vcpu, cpu_role); |
fb72d167 | 5430 | } |
1c53da3f | 5431 | EXPORT_SYMBOL_GPL(kvm_init_mmu); |
fb72d167 | 5432 | |
49c6f875 SC |
5433 | void kvm_mmu_after_set_cpuid(struct kvm_vcpu *vcpu) |
5434 | { | |
5435 | /* | |
5436 | * Invalidate all MMU roles to force them to reinitialize as CPUID | |
5437 | * information is factored into reserved bit calculations. | |
feb627e8 VK |
5438 | * |
5439 | * Correctly handling multiple vCPU models with respect to paging and | |
5440 | * physical address properties) in a single VM would require tracking | |
5441 | * all relevant CPUID information in kvm_mmu_page_role. That is very | |
5442 | * undesirable as it would increase the memory requirements for | |
338068b5 SC |
5443 | * gfn_write_track (see struct kvm_mmu_page_role comments). For now |
5444 | * that problem is swept under the rug; KVM's CPUID API is horrific and | |
feb627e8 | 5445 | * it's all but impossible to solve it without introducing a new API. |
49c6f875 | 5446 | */ |
7a458f0e PB |
5447 | vcpu->arch.root_mmu.root_role.word = 0; |
5448 | vcpu->arch.guest_mmu.root_role.word = 0; | |
5449 | vcpu->arch.nested_mmu.root_role.word = 0; | |
e5ed0fb0 PB |
5450 | vcpu->arch.root_mmu.cpu_role.ext.valid = 0; |
5451 | vcpu->arch.guest_mmu.cpu_role.ext.valid = 0; | |
5452 | vcpu->arch.nested_mmu.cpu_role.ext.valid = 0; | |
49c6f875 | 5453 | kvm_mmu_reset_context(vcpu); |
63f5a190 SC |
5454 | |
5455 | /* | |
feb627e8 VK |
5456 | * Changing guest CPUID after KVM_RUN is forbidden, see the comment in |
5457 | * kvm_arch_vcpu_ioctl(). | |
63f5a190 | 5458 | */ |
fb3146b4 | 5459 | KVM_BUG_ON(kvm_vcpu_has_run(vcpu), vcpu->kvm); |
49c6f875 SC |
5460 | } |
5461 | ||
8a3c1a33 | 5462 | void kvm_mmu_reset_context(struct kvm_vcpu *vcpu) |
6aa8b732 | 5463 | { |
95f93af4 | 5464 | kvm_mmu_unload(vcpu); |
c9060662 | 5465 | kvm_init_mmu(vcpu); |
17c3ba9d | 5466 | } |
8668a3c4 | 5467 | EXPORT_SYMBOL_GPL(kvm_mmu_reset_context); |
17c3ba9d AK |
5468 | |
5469 | int kvm_mmu_load(struct kvm_vcpu *vcpu) | |
6aa8b732 | 5470 | { |
714b93da AK |
5471 | int r; |
5472 | ||
347a0d0d | 5473 | r = mmu_topup_memory_caches(vcpu, !vcpu->arch.mmu->root_role.direct); |
17c3ba9d AK |
5474 | if (r) |
5475 | goto out; | |
748e52b9 | 5476 | r = mmu_alloc_special_roots(vcpu); |
17c3ba9d AK |
5477 | if (r) |
5478 | goto out; | |
347a0d0d | 5479 | if (vcpu->arch.mmu->root_role.direct) |
6e6ec584 SC |
5480 | r = mmu_alloc_direct_roots(vcpu); |
5481 | else | |
5482 | r = mmu_alloc_shadow_roots(vcpu); | |
8986ecc0 MT |
5483 | if (r) |
5484 | goto out; | |
a91f387b SC |
5485 | |
5486 | kvm_mmu_sync_roots(vcpu); | |
5487 | ||
727a7e27 | 5488 | kvm_mmu_load_pgd(vcpu); |
db01416b SC |
5489 | |
5490 | /* | |
5491 | * Flush any TLB entries for the new root, the provenance of the root | |
5492 | * is unknown. Even if KVM ensures there are no stale TLB entries | |
5493 | * for a freed root, in theory another hypervisor could have left | |
5494 | * stale entries. Flushing on alloc also allows KVM to skip the TLB | |
5495 | * flush when freeing a root (see kvm_tdp_mmu_put_root()). | |
5496 | */ | |
e27bc044 | 5497 | static_call(kvm_x86_flush_tlb_current)(vcpu); |
714b93da AK |
5498 | out: |
5499 | return r; | |
6aa8b732 | 5500 | } |
17c3ba9d AK |
5501 | |
5502 | void kvm_mmu_unload(struct kvm_vcpu *vcpu) | |
5503 | { | |
0c1c92f1 PB |
5504 | struct kvm *kvm = vcpu->kvm; |
5505 | ||
5506 | kvm_mmu_free_roots(kvm, &vcpu->arch.root_mmu, KVM_MMU_ROOTS_ALL); | |
20ba462d | 5507 | WARN_ON_ONCE(VALID_PAGE(vcpu->arch.root_mmu.root.hpa)); |
0c1c92f1 | 5508 | kvm_mmu_free_roots(kvm, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL); |
20ba462d | 5509 | WARN_ON_ONCE(VALID_PAGE(vcpu->arch.guest_mmu.root.hpa)); |
6d58f275 | 5510 | vcpu_clear_mmio_info(vcpu, MMIO_GVA_ANY); |
17c3ba9d | 5511 | } |
6aa8b732 | 5512 | |
527d5cd7 SC |
5513 | static bool is_obsolete_root(struct kvm *kvm, hpa_t root_hpa) |
5514 | { | |
5515 | struct kvm_mmu_page *sp; | |
5516 | ||
5517 | if (!VALID_PAGE(root_hpa)) | |
5518 | return false; | |
5519 | ||
5520 | /* | |
5521 | * When freeing obsolete roots, treat roots as obsolete if they don't | |
0e3223d8 SC |
5522 | * have an associated shadow page, as it's impossible to determine if |
5523 | * such roots are fresh or stale. This does mean KVM will get false | |
527d5cd7 SC |
5524 | * positives and free roots that don't strictly need to be freed, but |
5525 | * such false positives are relatively rare: | |
5526 | * | |
0e3223d8 SC |
5527 | * (a) only PAE paging and nested NPT have roots without shadow pages |
5528 | * (or any shadow paging flavor with a dummy root, see note below) | |
527d5cd7 SC |
5529 | * (b) remote reloads due to a memslot update obsoletes _all_ roots |
5530 | * (c) KVM doesn't track previous roots for PAE paging, and the guest | |
5531 | * is unlikely to zap an in-use PGD. | |
0e3223d8 SC |
5532 | * |
5533 | * Note! Dummy roots are unique in that they are obsoleted by memslot | |
5534 | * _creation_! See also FNAME(fetch). | |
527d5cd7 | 5535 | */ |
c5f2d564 | 5536 | sp = root_to_sp(root_hpa); |
527d5cd7 SC |
5537 | return !sp || is_obsolete_sp(kvm, sp); |
5538 | } | |
5539 | ||
5540 | static void __kvm_mmu_free_obsolete_roots(struct kvm *kvm, struct kvm_mmu *mmu) | |
5541 | { | |
5542 | unsigned long roots_to_free = 0; | |
5543 | int i; | |
5544 | ||
5545 | if (is_obsolete_root(kvm, mmu->root.hpa)) | |
5546 | roots_to_free |= KVM_MMU_ROOT_CURRENT; | |
5547 | ||
5548 | for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) { | |
cf4a8693 | 5549 | if (is_obsolete_root(kvm, mmu->prev_roots[i].hpa)) |
527d5cd7 SC |
5550 | roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i); |
5551 | } | |
5552 | ||
5553 | if (roots_to_free) | |
5554 | kvm_mmu_free_roots(kvm, mmu, roots_to_free); | |
5555 | } | |
5556 | ||
5557 | void kvm_mmu_free_obsolete_roots(struct kvm_vcpu *vcpu) | |
5558 | { | |
5559 | __kvm_mmu_free_obsolete_roots(vcpu->kvm, &vcpu->arch.root_mmu); | |
5560 | __kvm_mmu_free_obsolete_roots(vcpu->kvm, &vcpu->arch.guest_mmu); | |
5561 | } | |
5562 | ||
889e5cbc | 5563 | static u64 mmu_pte_write_fetch_gpte(struct kvm_vcpu *vcpu, gpa_t *gpa, |
0e0fee5c | 5564 | int *bytes) |
da4a00f0 | 5565 | { |
0e0fee5c | 5566 | u64 gentry = 0; |
889e5cbc | 5567 | int r; |
72016f3a | 5568 | |
72016f3a AK |
5569 | /* |
5570 | * Assume that the pte write on a page table of the same type | |
49b26e26 XG |
5571 | * as the current vcpu paging mode since we update the sptes only |
5572 | * when they have the same mode. | |
72016f3a | 5573 | */ |
889e5cbc | 5574 | if (is_pae(vcpu) && *bytes == 4) { |
72016f3a | 5575 | /* Handle a 32-bit guest writing two halves of a 64-bit gpte */ |
889e5cbc XG |
5576 | *gpa &= ~(gpa_t)7; |
5577 | *bytes = 8; | |
08e850c6 AK |
5578 | } |
5579 | ||
0e0fee5c JS |
5580 | if (*bytes == 4 || *bytes == 8) { |
5581 | r = kvm_vcpu_read_guest_atomic(vcpu, *gpa, &gentry, *bytes); | |
5582 | if (r) | |
5583 | gentry = 0; | |
72016f3a AK |
5584 | } |
5585 | ||
889e5cbc XG |
5586 | return gentry; |
5587 | } | |
5588 | ||
5589 | /* | |
5590 | * If we're seeing too many writes to a page, it may no longer be a page table, | |
5591 | * or we may be forking, in which case it is better to unmap the page. | |
5592 | */ | |
a138fe75 | 5593 | static bool detect_write_flooding(struct kvm_mmu_page *sp) |
889e5cbc | 5594 | { |
a30f47cb XG |
5595 | /* |
5596 | * Skip write-flooding detected for the sp whose level is 1, because | |
5597 | * it can become unsync, then the guest page is not write-protected. | |
5598 | */ | |
3bae0459 | 5599 | if (sp->role.level == PG_LEVEL_4K) |
a30f47cb | 5600 | return false; |
3246af0e | 5601 | |
e5691a81 XG |
5602 | atomic_inc(&sp->write_flooding_count); |
5603 | return atomic_read(&sp->write_flooding_count) >= 3; | |
889e5cbc XG |
5604 | } |
5605 | ||
5606 | /* | |
5607 | * Misaligned accesses are too much trouble to fix up; also, they usually | |
5608 | * indicate a page is not used as a page table. | |
5609 | */ | |
5610 | static bool detect_write_misaligned(struct kvm_mmu_page *sp, gpa_t gpa, | |
5611 | int bytes) | |
5612 | { | |
5613 | unsigned offset, pte_size, misaligned; | |
5614 | ||
889e5cbc | 5615 | offset = offset_in_page(gpa); |
bb3b394d | 5616 | pte_size = sp->role.has_4_byte_gpte ? 4 : 8; |
5d9ca30e XG |
5617 | |
5618 | /* | |
5619 | * Sometimes, the OS only writes the last one bytes to update status | |
5620 | * bits, for example, in linux, andb instruction is used in clear_bit(). | |
5621 | */ | |
5622 | if (!(offset & (pte_size - 1)) && bytes == 1) | |
5623 | return false; | |
5624 | ||
889e5cbc XG |
5625 | misaligned = (offset ^ (offset + bytes - 1)) & ~(pte_size - 1); |
5626 | misaligned |= bytes < 4; | |
5627 | ||
5628 | return misaligned; | |
5629 | } | |
5630 | ||
5631 | static u64 *get_written_sptes(struct kvm_mmu_page *sp, gpa_t gpa, int *nspte) | |
5632 | { | |
5633 | unsigned page_offset, quadrant; | |
5634 | u64 *spte; | |
5635 | int level; | |
5636 | ||
5637 | page_offset = offset_in_page(gpa); | |
5638 | level = sp->role.level; | |
5639 | *nspte = 1; | |
bb3b394d | 5640 | if (sp->role.has_4_byte_gpte) { |
889e5cbc XG |
5641 | page_offset <<= 1; /* 32->64 */ |
5642 | /* | |
5643 | * A 32-bit pde maps 4MB while the shadow pdes map | |
5644 | * only 2MB. So we need to double the offset again | |
5645 | * and zap two pdes instead of one. | |
5646 | */ | |
5647 | if (level == PT32_ROOT_LEVEL) { | |
5648 | page_offset &= ~7; /* kill rounding error */ | |
5649 | page_offset <<= 1; | |
5650 | *nspte = 2; | |
5651 | } | |
5652 | quadrant = page_offset >> PAGE_SHIFT; | |
5653 | page_offset &= ~PAGE_MASK; | |
5654 | if (quadrant != sp->role.quadrant) | |
5655 | return NULL; | |
5656 | } | |
5657 | ||
5658 | spte = &sp->spt[page_offset / sizeof(*spte)]; | |
5659 | return spte; | |
5660 | } | |
5661 | ||
93284446 SC |
5662 | void kvm_mmu_track_write(struct kvm_vcpu *vcpu, gpa_t gpa, const u8 *new, |
5663 | int bytes) | |
889e5cbc XG |
5664 | { |
5665 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
889e5cbc | 5666 | struct kvm_mmu_page *sp; |
889e5cbc XG |
5667 | LIST_HEAD(invalid_list); |
5668 | u64 entry, gentry, *spte; | |
5669 | int npte; | |
06152b2d | 5670 | bool flush = false; |
889e5cbc XG |
5671 | |
5672 | /* | |
5673 | * If we don't have indirect shadow pages, it means no page is | |
5674 | * write-protected, so we can exit simply. | |
5675 | */ | |
6aa7de05 | 5676 | if (!READ_ONCE(vcpu->kvm->arch.indirect_shadow_pages)) |
889e5cbc XG |
5677 | return; |
5678 | ||
531810ca | 5679 | write_lock(&vcpu->kvm->mmu_lock); |
0e0fee5c JS |
5680 | |
5681 | gentry = mmu_pte_write_fetch_gpte(vcpu, &gpa, &bytes); | |
5682 | ||
889e5cbc | 5683 | ++vcpu->kvm->stat.mmu_pte_write; |
889e5cbc | 5684 | |
767d8d8d | 5685 | for_each_gfn_valid_sp_with_gptes(vcpu->kvm, sp, gfn) { |
a30f47cb | 5686 | if (detect_write_misaligned(sp, gpa, bytes) || |
a138fe75 | 5687 | detect_write_flooding(sp)) { |
b8c67b7a | 5688 | kvm_mmu_prepare_zap_page(vcpu->kvm, sp, &invalid_list); |
4cee5764 | 5689 | ++vcpu->kvm->stat.mmu_flooded; |
0e7bc4b9 AK |
5690 | continue; |
5691 | } | |
889e5cbc XG |
5692 | |
5693 | spte = get_written_sptes(sp, gpa, &npte); | |
5694 | if (!spte) | |
5695 | continue; | |
5696 | ||
ac1b714e | 5697 | while (npte--) { |
79539cec | 5698 | entry = *spte; |
2de4085c | 5699 | mmu_page_zap_pte(vcpu->kvm, sp, spte, NULL); |
c5e2184d SC |
5700 | if (gentry && sp->role.level != PG_LEVEL_4K) |
5701 | ++vcpu->kvm->stat.mmu_pde_zapped; | |
1441ca14 | 5702 | if (is_shadow_present_pte(entry)) |
06152b2d | 5703 | flush = true; |
ac1b714e | 5704 | ++spte; |
9b7a0325 | 5705 | } |
9b7a0325 | 5706 | } |
06152b2d | 5707 | kvm_mmu_remote_flush_or_zap(vcpu->kvm, &invalid_list, flush); |
531810ca | 5708 | write_unlock(&vcpu->kvm->mmu_lock); |
da4a00f0 AK |
5709 | } |
5710 | ||
1075d41e | 5711 | int noinline kvm_mmu_page_fault(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa, u64 error_code, |
dc25e89e | 5712 | void *insn, int insn_len) |
3067714c | 5713 | { |
92daa48b | 5714 | int r, emulation_type = EMULTYPE_PF; |
347a0d0d | 5715 | bool direct = vcpu->arch.mmu->root_role.direct; |
3067714c | 5716 | |
d09f7112 SC |
5717 | /* |
5718 | * IMPLICIT_ACCESS is a KVM-defined flag used to correctly perform SMAP | |
5719 | * checks when emulating instructions that triggers implicit access. | |
5720 | * WARN if hardware generates a fault with an error code that collides | |
5721 | * with the KVM-defined value. Clear the flag and continue on, i.e. | |
5722 | * don't terminate the VM, as KVM can't possibly be relying on a flag | |
5723 | * that KVM doesn't know about. | |
5724 | */ | |
5725 | if (WARN_ON_ONCE(error_code & PFERR_IMPLICIT_ACCESS)) | |
5726 | error_code &= ~PFERR_IMPLICIT_ACCESS; | |
5727 | ||
20ba462d | 5728 | if (WARN_ON_ONCE(!VALID_PAGE(vcpu->arch.mmu->root.hpa))) |
ddce6208 SC |
5729 | return RET_PF_RETRY; |
5730 | ||
9b8ebbdb | 5731 | r = RET_PF_INVALID; |
e9ee956e | 5732 | if (unlikely(error_code & PFERR_RSVD_MASK)) { |
736c291c | 5733 | r = handle_mmio_page_fault(vcpu, cr2_or_gpa, direct); |
472faffa | 5734 | if (r == RET_PF_EMULATE) |
e9ee956e | 5735 | goto emulate; |
e9ee956e | 5736 | } |
3067714c | 5737 | |
9b8ebbdb | 5738 | if (r == RET_PF_INVALID) { |
7a02674d | 5739 | r = kvm_mmu_do_page_fault(vcpu, cr2_or_gpa, |
258d985f SC |
5740 | lower_32_bits(error_code), false, |
5741 | &emulation_type); | |
19025e7b | 5742 | if (KVM_BUG_ON(r == RET_PF_INVALID, vcpu->kvm)) |
7b367bc9 | 5743 | return -EIO; |
9b8ebbdb PB |
5744 | } |
5745 | ||
3067714c | 5746 | if (r < 0) |
e9ee956e | 5747 | return r; |
83a2ba4c SC |
5748 | if (r != RET_PF_EMULATE) |
5749 | return 1; | |
3067714c | 5750 | |
14727754 TL |
5751 | /* |
5752 | * Before emulating the instruction, check if the error code | |
5753 | * was due to a RO violation while translating the guest page. | |
5754 | * This can occur when using nested virtualization with nested | |
5755 | * paging in both guests. If true, we simply unprotect the page | |
5756 | * and resume the guest. | |
14727754 | 5757 | */ |
347a0d0d | 5758 | if (vcpu->arch.mmu->root_role.direct && |
eebed243 | 5759 | (error_code & PFERR_NESTED_GUEST_PAGE) == PFERR_NESTED_GUEST_PAGE) { |
736c291c | 5760 | kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(cr2_or_gpa)); |
14727754 TL |
5761 | return 1; |
5762 | } | |
5763 | ||
472faffa SC |
5764 | /* |
5765 | * vcpu->arch.mmu.page_fault returned RET_PF_EMULATE, but we can still | |
5766 | * optimistically try to just unprotect the page and let the processor | |
5767 | * re-execute the instruction that caused the page fault. Do not allow | |
5768 | * retrying MMIO emulation, as it's not only pointless but could also | |
5769 | * cause us to enter an infinite loop because the processor will keep | |
6c3dfeb6 SC |
5770 | * faulting on the non-existent MMIO address. Retrying an instruction |
5771 | * from a nested guest is also pointless and dangerous as we are only | |
5772 | * explicitly shadowing L1's page tables, i.e. unprotecting something | |
5773 | * for L1 isn't going to magically fix whatever issue cause L2 to fail. | |
472faffa | 5774 | */ |
736c291c | 5775 | if (!mmio_info_in_cache(vcpu, cr2_or_gpa, direct) && !is_guest_mode(vcpu)) |
92daa48b | 5776 | emulation_type |= EMULTYPE_ALLOW_RETRY_PF; |
e9ee956e | 5777 | emulate: |
736c291c | 5778 | return x86_emulate_instruction(vcpu, cr2_or_gpa, emulation_type, insn, |
60fc3d02 | 5779 | insn_len); |
3067714c AK |
5780 | } |
5781 | EXPORT_SYMBOL_GPL(kvm_mmu_page_fault); | |
5782 | ||
9fd4a4e3 LJ |
5783 | static void __kvm_mmu_invalidate_addr(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, |
5784 | u64 addr, hpa_t root_hpa) | |
5785 | { | |
5786 | struct kvm_shadow_walk_iterator iterator; | |
5787 | ||
5788 | vcpu_clear_mmio_info(vcpu, addr); | |
5789 | ||
762b33eb LX |
5790 | /* |
5791 | * Walking and synchronizing SPTEs both assume they are operating in | |
5792 | * the context of the current MMU, and would need to be reworked if | |
5793 | * this is ever used to sync the guest_mmu, e.g. to emulate INVEPT. | |
5794 | */ | |
5795 | if (WARN_ON_ONCE(mmu != vcpu->arch.mmu)) | |
5796 | return; | |
5797 | ||
9fd4a4e3 LJ |
5798 | if (!VALID_PAGE(root_hpa)) |
5799 | return; | |
5800 | ||
5801 | write_lock(&vcpu->kvm->mmu_lock); | |
5802 | for_each_shadow_entry_using_root(vcpu, root_hpa, addr, iterator) { | |
5803 | struct kvm_mmu_page *sp = sptep_to_sp(iterator.sptep); | |
5804 | ||
5805 | if (sp->unsync) { | |
19ace7d6 | 5806 | int ret = kvm_sync_spte(vcpu, sp, iterator.index); |
9fd4a4e3 LJ |
5807 | |
5808 | if (ret < 0) | |
5809 | mmu_page_zap_pte(vcpu->kvm, sp, iterator.sptep, NULL); | |
5810 | if (ret) | |
5811 | kvm_flush_remote_tlbs_sptep(vcpu->kvm, iterator.sptep); | |
5812 | } | |
5813 | ||
5814 | if (!sp->unsync_children) | |
5815 | break; | |
5816 | } | |
5817 | write_unlock(&vcpu->kvm->mmu_lock); | |
5818 | } | |
5819 | ||
753b43c9 | 5820 | void kvm_mmu_invalidate_addr(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, |
cd42853e | 5821 | u64 addr, unsigned long roots) |
a7052897 | 5822 | { |
b94742c9 | 5823 | int i; |
7eb77e9f | 5824 | |
cd42853e LJ |
5825 | WARN_ON_ONCE(roots & ~KVM_MMU_ROOTS_ALL); |
5826 | ||
5efac074 PB |
5827 | /* It's actually a GPA for vcpu->arch.guest_mmu. */ |
5828 | if (mmu != &vcpu->arch.guest_mmu) { | |
5829 | /* INVLPG on a non-canonical address is a NOP according to the SDM. */ | |
753b43c9 | 5830 | if (is_noncanonical_address(addr, vcpu)) |
5efac074 PB |
5831 | return; |
5832 | ||
753b43c9 | 5833 | static_call(kvm_x86_flush_tlb_gva)(vcpu, addr); |
5efac074 PB |
5834 | } |
5835 | ||
9fd4a4e3 | 5836 | if (!mmu->sync_spte) |
faff8758 JS |
5837 | return; |
5838 | ||
cd42853e | 5839 | if (roots & KVM_MMU_ROOT_CURRENT) |
9fd4a4e3 | 5840 | __kvm_mmu_invalidate_addr(vcpu, mmu, addr, mmu->root.hpa); |
956bf353 | 5841 | |
cd42853e | 5842 | for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) { |
ed335278 | 5843 | if (roots & KVM_MMU_ROOT_PREVIOUS(i)) |
9fd4a4e3 | 5844 | __kvm_mmu_invalidate_addr(vcpu, mmu, addr, mmu->prev_roots[i].hpa); |
5efac074 PB |
5845 | } |
5846 | } | |
2c86c444 | 5847 | EXPORT_SYMBOL_GPL(kvm_mmu_invalidate_addr); |
956bf353 | 5848 | |
5efac074 PB |
5849 | void kvm_mmu_invlpg(struct kvm_vcpu *vcpu, gva_t gva) |
5850 | { | |
cd42853e LJ |
5851 | /* |
5852 | * INVLPG is required to invalidate any global mappings for the VA, | |
5853 | * irrespective of PCID. Blindly sync all roots as it would take | |
5854 | * roughly the same amount of work/time to determine whether any of the | |
5855 | * previous roots have a global mapping. | |
5856 | * | |
5857 | * Mappings not reachable via the current or previous cached roots will | |
5858 | * be synced when switching to that new cr3, so nothing needs to be | |
5859 | * done here for them. | |
5860 | */ | |
5861 | kvm_mmu_invalidate_addr(vcpu, vcpu->arch.walk_mmu, gva, KVM_MMU_ROOTS_ALL); | |
a7052897 MT |
5862 | ++vcpu->stat.invlpg; |
5863 | } | |
5864 | EXPORT_SYMBOL_GPL(kvm_mmu_invlpg); | |
5865 | ||
5efac074 | 5866 | |
eb4b248e JS |
5867 | void kvm_mmu_invpcid_gva(struct kvm_vcpu *vcpu, gva_t gva, unsigned long pcid) |
5868 | { | |
44dd3ffa | 5869 | struct kvm_mmu *mmu = vcpu->arch.mmu; |
9ebc3f51 | 5870 | unsigned long roots = 0; |
b94742c9 | 5871 | uint i; |
eb4b248e | 5872 | |
9ebc3f51 LJ |
5873 | if (pcid == kvm_get_active_pcid(vcpu)) |
5874 | roots |= KVM_MMU_ROOT_CURRENT; | |
eb4b248e | 5875 | |
b94742c9 JS |
5876 | for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) { |
5877 | if (VALID_PAGE(mmu->prev_roots[i].hpa) && | |
9ebc3f51 LJ |
5878 | pcid == kvm_get_pcid(vcpu, mmu->prev_roots[i].pgd)) |
5879 | roots |= KVM_MMU_ROOT_PREVIOUS(i); | |
956bf353 | 5880 | } |
ade61e28 | 5881 | |
9ebc3f51 LJ |
5882 | if (roots) |
5883 | kvm_mmu_invalidate_addr(vcpu, mmu, gva, roots); | |
eb4b248e JS |
5884 | ++vcpu->stat.invlpg; |
5885 | ||
5886 | /* | |
b94742c9 JS |
5887 | * Mappings not reachable via the current cr3 or the prev_roots will be |
5888 | * synced when switching to that cr3, so nothing needs to be done here | |
5889 | * for them. | |
eb4b248e JS |
5890 | */ |
5891 | } | |
eb4b248e | 5892 | |
746700d2 WH |
5893 | void kvm_configure_mmu(bool enable_tdp, int tdp_forced_root_level, |
5894 | int tdp_max_root_level, int tdp_huge_page_level) | |
18552672 | 5895 | { |
bde77235 | 5896 | tdp_enabled = enable_tdp; |
746700d2 | 5897 | tdp_root_level = tdp_forced_root_level; |
83013059 | 5898 | max_tdp_level = tdp_max_root_level; |
703c335d | 5899 | |
1f98f2bd DM |
5900 | #ifdef CONFIG_X86_64 |
5901 | tdp_mmu_enabled = tdp_mmu_allowed && tdp_enabled; | |
5902 | #endif | |
703c335d | 5903 | /* |
1d92d2e8 | 5904 | * max_huge_page_level reflects KVM's MMU capabilities irrespective |
703c335d SC |
5905 | * of kernel support, e.g. KVM may be capable of using 1GB pages when |
5906 | * the kernel is not. But, KVM never creates a page size greater than | |
5907 | * what is used by the kernel for any given HVA, i.e. the kernel's | |
5908 | * capabilities are ultimately consulted by kvm_mmu_hugepage_adjust(). | |
5909 | */ | |
5910 | if (tdp_enabled) | |
1d92d2e8 | 5911 | max_huge_page_level = tdp_huge_page_level; |
703c335d | 5912 | else if (boot_cpu_has(X86_FEATURE_GBPAGES)) |
1d92d2e8 | 5913 | max_huge_page_level = PG_LEVEL_1G; |
703c335d | 5914 | else |
1d92d2e8 | 5915 | max_huge_page_level = PG_LEVEL_2M; |
18552672 | 5916 | } |
bde77235 | 5917 | EXPORT_SYMBOL_GPL(kvm_configure_mmu); |
85875a13 SC |
5918 | |
5919 | /* The return value indicates if tlb flush on all vcpus is needed. */ | |
727ae377 | 5920 | typedef bool (*slot_rmaps_handler) (struct kvm *kvm, |
269e9552 HM |
5921 | struct kvm_rmap_head *rmap_head, |
5922 | const struct kvm_memory_slot *slot); | |
85875a13 | 5923 | |
727ae377 SC |
5924 | static __always_inline bool __walk_slot_rmaps(struct kvm *kvm, |
5925 | const struct kvm_memory_slot *slot, | |
5926 | slot_rmaps_handler fn, | |
5927 | int start_level, int end_level, | |
5928 | gfn_t start_gfn, gfn_t end_gfn, | |
5929 | bool flush_on_yield, bool flush) | |
85875a13 SC |
5930 | { |
5931 | struct slot_rmap_walk_iterator iterator; | |
85875a13 | 5932 | |
eddd9e83 SC |
5933 | lockdep_assert_held_write(&kvm->mmu_lock); |
5934 | ||
727ae377 | 5935 | for_each_slot_rmap_range(slot, start_level, end_level, start_gfn, |
85875a13 SC |
5936 | end_gfn, &iterator) { |
5937 | if (iterator.rmap) | |
727ae377 | 5938 | flush |= fn(kvm, iterator.rmap, slot); |
85875a13 | 5939 | |
531810ca | 5940 | if (need_resched() || rwlock_needbreak(&kvm->mmu_lock)) { |
302695a5 | 5941 | if (flush && flush_on_yield) { |
8c63e8c2 DM |
5942 | kvm_flush_remote_tlbs_range(kvm, start_gfn, |
5943 | iterator.gfn - start_gfn + 1); | |
85875a13 SC |
5944 | flush = false; |
5945 | } | |
531810ca | 5946 | cond_resched_rwlock_write(&kvm->mmu_lock); |
85875a13 SC |
5947 | } |
5948 | } | |
5949 | ||
85875a13 SC |
5950 | return flush; |
5951 | } | |
5952 | ||
727ae377 SC |
5953 | static __always_inline bool walk_slot_rmaps(struct kvm *kvm, |
5954 | const struct kvm_memory_slot *slot, | |
5955 | slot_rmaps_handler fn, | |
5956 | int start_level, int end_level, | |
5957 | bool flush_on_yield) | |
85875a13 | 5958 | { |
727ae377 SC |
5959 | return __walk_slot_rmaps(kvm, slot, fn, start_level, end_level, |
5960 | slot->base_gfn, slot->base_gfn + slot->npages - 1, | |
5961 | flush_on_yield, false); | |
85875a13 SC |
5962 | } |
5963 | ||
727ae377 SC |
5964 | static __always_inline bool walk_slot_rmaps_4k(struct kvm *kvm, |
5965 | const struct kvm_memory_slot *slot, | |
5966 | slot_rmaps_handler fn, | |
5967 | bool flush_on_yield) | |
85875a13 | 5968 | { |
727ae377 | 5969 | return walk_slot_rmaps(kvm, slot, fn, PG_LEVEL_4K, PG_LEVEL_4K, flush_on_yield); |
85875a13 SC |
5970 | } |
5971 | ||
1cfff4d9 | 5972 | static void free_mmu_pages(struct kvm_mmu *mmu) |
6aa8b732 | 5973 | { |
4a98623d SC |
5974 | if (!tdp_enabled && mmu->pae_root) |
5975 | set_memory_encrypted((unsigned long)mmu->pae_root, 1); | |
1cfff4d9 | 5976 | free_page((unsigned long)mmu->pae_root); |
03ca4589 | 5977 | free_page((unsigned long)mmu->pml4_root); |
cb0f722a | 5978 | free_page((unsigned long)mmu->pml5_root); |
6aa8b732 AK |
5979 | } |
5980 | ||
04d28e37 | 5981 | static int __kvm_mmu_create(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu) |
6aa8b732 | 5982 | { |
17ac10ad | 5983 | struct page *page; |
6aa8b732 AK |
5984 | int i; |
5985 | ||
b9e5603c PB |
5986 | mmu->root.hpa = INVALID_PAGE; |
5987 | mmu->root.pgd = 0; | |
04d28e37 SC |
5988 | for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) |
5989 | mmu->prev_roots[i] = KVM_MMU_ROOT_INFO_INVALID; | |
5990 | ||
27f4fca2 LJ |
5991 | /* vcpu->arch.guest_mmu isn't used when !tdp_enabled. */ |
5992 | if (!tdp_enabled && mmu == &vcpu->arch.guest_mmu) | |
5993 | return 0; | |
5994 | ||
17ac10ad | 5995 | /* |
b6b80c78 SC |
5996 | * When using PAE paging, the four PDPTEs are treated as 'root' pages, |
5997 | * while the PDP table is a per-vCPU construct that's allocated at MMU | |
5998 | * creation. When emulating 32-bit mode, cr3 is only 32 bits even on | |
5999 | * x86_64. Therefore we need to allocate the PDP table in the first | |
04d45551 SC |
6000 | * 4GB of memory, which happens to fit the DMA32 zone. TDP paging |
6001 | * generally doesn't use PAE paging and can skip allocating the PDP | |
6002 | * table. The main exception, handled here, is SVM's 32-bit NPT. The | |
6003 | * other exception is for shadowing L1's 32-bit or PAE NPT on 64-bit | |
84432316 | 6004 | * KVM; that horror is handled on-demand by mmu_alloc_special_roots(). |
17ac10ad | 6005 | */ |
d468d94b | 6006 | if (tdp_enabled && kvm_mmu_get_tdp_level(vcpu) > PT32E_ROOT_LEVEL) |
b6b80c78 SC |
6007 | return 0; |
6008 | ||
254272ce | 6009 | page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_DMA32); |
17ac10ad | 6010 | if (!page) |
d7fa6ab2 WY |
6011 | return -ENOMEM; |
6012 | ||
1cfff4d9 | 6013 | mmu->pae_root = page_address(page); |
4a98623d SC |
6014 | |
6015 | /* | |
6016 | * CR3 is only 32 bits when PAE paging is used, thus it's impossible to | |
6017 | * get the CPU to treat the PDPTEs as encrypted. Decrypt the page so | |
6018 | * that KVM's writes and the CPU's reads get along. Note, this is | |
6019 | * only necessary when using shadow paging, as 64-bit NPT can get at | |
6020 | * the C-bit even when shadowing 32-bit NPT, and SME isn't supported | |
6021 | * by 32-bit kernels (when KVM itself uses 32-bit NPT). | |
6022 | */ | |
6023 | if (!tdp_enabled) | |
6024 | set_memory_decrypted((unsigned long)mmu->pae_root, 1); | |
6025 | else | |
e54f1ff2 | 6026 | WARN_ON_ONCE(shadow_me_value); |
4a98623d | 6027 | |
17ac10ad | 6028 | for (i = 0; i < 4; ++i) |
c834e5e4 | 6029 | mmu->pae_root[i] = INVALID_PAE_ROOT; |
17ac10ad | 6030 | |
6aa8b732 | 6031 | return 0; |
6aa8b732 AK |
6032 | } |
6033 | ||
8018c27b | 6034 | int kvm_mmu_create(struct kvm_vcpu *vcpu) |
6aa8b732 | 6035 | { |
1cfff4d9 | 6036 | int ret; |
b94742c9 | 6037 | |
5962bfb7 | 6038 | vcpu->arch.mmu_pte_list_desc_cache.kmem_cache = pte_list_desc_cache; |
5f6078f9 SC |
6039 | vcpu->arch.mmu_pte_list_desc_cache.gfp_zero = __GFP_ZERO; |
6040 | ||
5962bfb7 | 6041 | vcpu->arch.mmu_page_header_cache.kmem_cache = mmu_page_header_cache; |
5f6078f9 | 6042 | vcpu->arch.mmu_page_header_cache.gfp_zero = __GFP_ZERO; |
5962bfb7 | 6043 | |
96880883 SC |
6044 | vcpu->arch.mmu_shadow_page_cache.gfp_zero = __GFP_ZERO; |
6045 | ||
44dd3ffa VK |
6046 | vcpu->arch.mmu = &vcpu->arch.root_mmu; |
6047 | vcpu->arch.walk_mmu = &vcpu->arch.root_mmu; | |
6aa8b732 | 6048 | |
04d28e37 | 6049 | ret = __kvm_mmu_create(vcpu, &vcpu->arch.guest_mmu); |
1cfff4d9 JP |
6050 | if (ret) |
6051 | return ret; | |
6052 | ||
04d28e37 | 6053 | ret = __kvm_mmu_create(vcpu, &vcpu->arch.root_mmu); |
1cfff4d9 JP |
6054 | if (ret) |
6055 | goto fail_allocate_root; | |
6056 | ||
6057 | return ret; | |
6058 | fail_allocate_root: | |
6059 | free_mmu_pages(&vcpu->arch.guest_mmu); | |
6060 | return ret; | |
6aa8b732 AK |
6061 | } |
6062 | ||
fbb158cb | 6063 | #define BATCH_ZAP_PAGES 10 |
002c5f73 SC |
6064 | static void kvm_zap_obsolete_pages(struct kvm *kvm) |
6065 | { | |
6066 | struct kvm_mmu_page *sp, *node; | |
fbb158cb | 6067 | int nr_zapped, batch = 0; |
b28cb0cd | 6068 | bool unstable; |
002c5f73 SC |
6069 | |
6070 | restart: | |
6071 | list_for_each_entry_safe_reverse(sp, node, | |
6072 | &kvm->arch.active_mmu_pages, link) { | |
6073 | /* | |
6074 | * No obsolete valid page exists before a newly created page | |
6075 | * since active_mmu_pages is a FIFO list. | |
6076 | */ | |
6077 | if (!is_obsolete_sp(kvm, sp)) | |
6078 | break; | |
6079 | ||
6080 | /* | |
f95eec9b SC |
6081 | * Invalid pages should never land back on the list of active |
6082 | * pages. Skip the bogus page, otherwise we'll get stuck in an | |
6083 | * infinite loop if the page gets put back on the list (again). | |
002c5f73 | 6084 | */ |
20ba462d | 6085 | if (WARN_ON_ONCE(sp->role.invalid)) |
002c5f73 SC |
6086 | continue; |
6087 | ||
4506ecf4 SC |
6088 | /* |
6089 | * No need to flush the TLB since we're only zapping shadow | |
6090 | * pages with an obsolete generation number and all vCPUS have | |
6091 | * loaded a new root, i.e. the shadow pages being zapped cannot | |
6092 | * be in active use by the guest. | |
6093 | */ | |
fbb158cb | 6094 | if (batch >= BATCH_ZAP_PAGES && |
531810ca | 6095 | cond_resched_rwlock_write(&kvm->mmu_lock)) { |
fbb158cb | 6096 | batch = 0; |
002c5f73 SC |
6097 | goto restart; |
6098 | } | |
6099 | ||
b28cb0cd SC |
6100 | unstable = __kvm_mmu_prepare_zap_page(kvm, sp, |
6101 | &kvm->arch.zapped_obsolete_pages, &nr_zapped); | |
6102 | batch += nr_zapped; | |
6103 | ||
6104 | if (unstable) | |
002c5f73 SC |
6105 | goto restart; |
6106 | } | |
6107 | ||
4506ecf4 | 6108 | /* |
7ae5840e SC |
6109 | * Kick all vCPUs (via remote TLB flush) before freeing the page tables |
6110 | * to ensure KVM is not in the middle of a lockless shadow page table | |
6111 | * walk, which may reference the pages. The remote TLB flush itself is | |
6112 | * not required and is simply a convenient way to kick vCPUs as needed. | |
6113 | * KVM performs a local TLB flush when allocating a new root (see | |
6114 | * kvm_mmu_load()), and the reload in the caller ensure no vCPUs are | |
6115 | * running with an obsolete MMU. | |
4506ecf4 | 6116 | */ |
10605204 | 6117 | kvm_mmu_commit_zap_page(kvm, &kvm->arch.zapped_obsolete_pages); |
002c5f73 SC |
6118 | } |
6119 | ||
6120 | /* | |
6121 | * Fast invalidate all shadow pages and use lock-break technique | |
6122 | * to zap obsolete pages. | |
6123 | * | |
6124 | * It's required when memslot is being deleted or VM is being | |
6125 | * destroyed, in these cases, we should ensure that KVM MMU does | |
6126 | * not use any resource of the being-deleted slot or all slots | |
6127 | * after calling the function. | |
6128 | */ | |
6129 | static void kvm_mmu_zap_all_fast(struct kvm *kvm) | |
6130 | { | |
ca333add SC |
6131 | lockdep_assert_held(&kvm->slots_lock); |
6132 | ||
531810ca | 6133 | write_lock(&kvm->mmu_lock); |
14a3c4f4 | 6134 | trace_kvm_mmu_zap_all_fast(kvm); |
ca333add SC |
6135 | |
6136 | /* | |
6137 | * Toggle mmu_valid_gen between '0' and '1'. Because slots_lock is | |
6138 | * held for the entire duration of zapping obsolete pages, it's | |
6139 | * impossible for there to be multiple invalid generations associated | |
6140 | * with *valid* shadow pages at any given time, i.e. there is exactly | |
6141 | * one valid generation and (at most) one invalid generation. | |
6142 | */ | |
6143 | kvm->arch.mmu_valid_gen = kvm->arch.mmu_valid_gen ? 0 : 1; | |
002c5f73 | 6144 | |
2f6f66cc SC |
6145 | /* |
6146 | * In order to ensure all vCPUs drop their soon-to-be invalid roots, | |
6147 | * invalidating TDP MMU roots must be done while holding mmu_lock for | |
6148 | * write and in the same critical section as making the reload request, | |
6149 | * e.g. before kvm_zap_obsolete_pages() could drop mmu_lock and yield. | |
b7cccd39 | 6150 | */ |
1f98f2bd | 6151 | if (tdp_mmu_enabled) |
b7cccd39 BG |
6152 | kvm_tdp_mmu_invalidate_all_roots(kvm); |
6153 | ||
4506ecf4 SC |
6154 | /* |
6155 | * Notify all vcpus to reload its shadow page table and flush TLB. | |
6156 | * Then all vcpus will switch to new shadow page table with the new | |
6157 | * mmu_valid_gen. | |
6158 | * | |
6159 | * Note: we need to do this under the protection of mmu_lock, | |
6160 | * otherwise, vcpu would purge shadow page but miss tlb flush. | |
6161 | */ | |
527d5cd7 | 6162 | kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_FREE_OBSOLETE_ROOTS); |
4506ecf4 | 6163 | |
002c5f73 | 6164 | kvm_zap_obsolete_pages(kvm); |
faaf05b0 | 6165 | |
531810ca | 6166 | write_unlock(&kvm->mmu_lock); |
4c6654bd | 6167 | |
f28e9c7f SC |
6168 | /* |
6169 | * Zap the invalidated TDP MMU roots, all SPTEs must be dropped before | |
6170 | * returning to the caller, e.g. if the zap is in response to a memslot | |
6171 | * deletion, mmu_notifier callbacks will be unable to reach the SPTEs | |
6172 | * associated with the deleted memslot once the update completes, and | |
6173 | * Deferring the zap until the final reference to the root is put would | |
6174 | * lead to use-after-free. | |
6175 | */ | |
1f98f2bd | 6176 | if (tdp_mmu_enabled) |
4c6654bd | 6177 | kvm_tdp_mmu_zap_invalidated_roots(kvm); |
002c5f73 SC |
6178 | } |
6179 | ||
10605204 SC |
6180 | static bool kvm_has_zapped_obsolete_pages(struct kvm *kvm) |
6181 | { | |
6182 | return unlikely(!list_empty_careful(&kvm->arch.zapped_obsolete_pages)); | |
6183 | } | |
6184 | ||
0df9dab8 | 6185 | void kvm_mmu_init_vm(struct kvm *kvm) |
1bad2b2a | 6186 | { |
a1a39128 PB |
6187 | INIT_LIST_HEAD(&kvm->arch.active_mmu_pages); |
6188 | INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages); | |
55c510e2 | 6189 | INIT_LIST_HEAD(&kvm->arch.possible_nx_huge_pages); |
ce25681d SC |
6190 | spin_lock_init(&kvm->arch.mmu_unsync_pages_lock); |
6191 | ||
0df9dab8 SC |
6192 | if (tdp_mmu_enabled) |
6193 | kvm_mmu_init_tdp_mmu(kvm); | |
fe5db27d | 6194 | |
ada51a9d DM |
6195 | kvm->arch.split_page_header_cache.kmem_cache = mmu_page_header_cache; |
6196 | kvm->arch.split_page_header_cache.gfp_zero = __GFP_ZERO; | |
6197 | ||
6198 | kvm->arch.split_shadow_page_cache.gfp_zero = __GFP_ZERO; | |
6199 | ||
6200 | kvm->arch.split_desc_cache.kmem_cache = pte_list_desc_cache; | |
6201 | kvm->arch.split_desc_cache.gfp_zero = __GFP_ZERO; | |
1bad2b2a XG |
6202 | } |
6203 | ||
ada51a9d DM |
6204 | static void mmu_free_vm_memory_caches(struct kvm *kvm) |
6205 | { | |
6206 | kvm_mmu_free_memory_cache(&kvm->arch.split_desc_cache); | |
6207 | kvm_mmu_free_memory_cache(&kvm->arch.split_page_header_cache); | |
6208 | kvm_mmu_free_memory_cache(&kvm->arch.split_shadow_page_cache); | |
6209 | } | |
6210 | ||
13d268ca | 6211 | void kvm_mmu_uninit_vm(struct kvm *kvm) |
1bad2b2a | 6212 | { |
09732d2b DM |
6213 | if (tdp_mmu_enabled) |
6214 | kvm_mmu_uninit_tdp_mmu(kvm); | |
ada51a9d DM |
6215 | |
6216 | mmu_free_vm_memory_caches(kvm); | |
1bad2b2a XG |
6217 | } |
6218 | ||
2833eda0 | 6219 | static bool kvm_rmap_zap_gfn_range(struct kvm *kvm, gfn_t gfn_start, gfn_t gfn_end) |
21fa3246 SC |
6220 | { |
6221 | const struct kvm_memory_slot *memslot; | |
6222 | struct kvm_memslots *slots; | |
f4209439 | 6223 | struct kvm_memslot_iter iter; |
21fa3246 SC |
6224 | bool flush = false; |
6225 | gfn_t start, end; | |
f4209439 | 6226 | int i; |
21fa3246 SC |
6227 | |
6228 | if (!kvm_memslots_have_rmaps(kvm)) | |
6229 | return flush; | |
6230 | ||
6231 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { | |
6232 | slots = __kvm_memslots(kvm, i); | |
f4209439 MS |
6233 | |
6234 | kvm_for_each_memslot_in_gfn_range(&iter, slots, gfn_start, gfn_end) { | |
6235 | memslot = iter.slot; | |
21fa3246 SC |
6236 | start = max(gfn_start, memslot->base_gfn); |
6237 | end = min(gfn_end, memslot->base_gfn + memslot->npages); | |
f4209439 | 6238 | if (WARN_ON_ONCE(start >= end)) |
21fa3246 SC |
6239 | continue; |
6240 | ||
727ae377 SC |
6241 | flush = __walk_slot_rmaps(kvm, memslot, __kvm_zap_rmap, |
6242 | PG_LEVEL_4K, KVM_MAX_HUGEPAGE_LEVEL, | |
6243 | start, end - 1, true, flush); | |
21fa3246 SC |
6244 | } |
6245 | } | |
6246 | ||
6247 | return flush; | |
6248 | } | |
6249 | ||
88f58535 ML |
6250 | /* |
6251 | * Invalidate (zap) SPTEs that cover GFNs from gfn_start and up to gfn_end | |
6252 | * (not including it) | |
6253 | */ | |
efdfe536 XG |
6254 | void kvm_zap_gfn_range(struct kvm *kvm, gfn_t gfn_start, gfn_t gfn_end) |
6255 | { | |
21fa3246 | 6256 | bool flush; |
efdfe536 | 6257 | |
f4209439 MS |
6258 | if (WARN_ON_ONCE(gfn_end <= gfn_start)) |
6259 | return; | |
6260 | ||
5a324c24 SC |
6261 | write_lock(&kvm->mmu_lock); |
6262 | ||
6d3085e4 | 6263 | kvm_mmu_invalidate_begin(kvm, 0, -1ul); |
edb298c6 | 6264 | |
2833eda0 | 6265 | flush = kvm_rmap_zap_gfn_range(kvm, gfn_start, gfn_end); |
efdfe536 | 6266 | |
441a5dfc PB |
6267 | if (tdp_mmu_enabled) |
6268 | flush = kvm_tdp_mmu_zap_leafs(kvm, gfn_start, gfn_end, flush); | |
5a324c24 SC |
6269 | |
6270 | if (flush) | |
8c63e8c2 | 6271 | kvm_flush_remote_tlbs_range(kvm, gfn_start, gfn_end - gfn_start); |
5a324c24 | 6272 | |
6d3085e4 | 6273 | kvm_mmu_invalidate_end(kvm, 0, -1ul); |
edb298c6 | 6274 | |
5a324c24 | 6275 | write_unlock(&kvm->mmu_lock); |
efdfe536 XG |
6276 | } |
6277 | ||
018aabb5 | 6278 | static bool slot_rmap_write_protect(struct kvm *kvm, |
0a234f5d | 6279 | struct kvm_rmap_head *rmap_head, |
269e9552 | 6280 | const struct kvm_memory_slot *slot) |
d77aa73c | 6281 | { |
1346bbb6 | 6282 | return rmap_write_protect(rmap_head, false); |
d77aa73c XG |
6283 | } |
6284 | ||
1c91cad4 | 6285 | void kvm_mmu_slot_remove_write_access(struct kvm *kvm, |
269e9552 | 6286 | const struct kvm_memory_slot *memslot, |
3c9bd400 | 6287 | int start_level) |
6aa8b732 | 6288 | { |
e2209710 BG |
6289 | if (kvm_memslots_have_rmaps(kvm)) { |
6290 | write_lock(&kvm->mmu_lock); | |
727ae377 SC |
6291 | walk_slot_rmaps(kvm, memslot, slot_rmap_write_protect, |
6292 | start_level, KVM_MAX_HUGEPAGE_LEVEL, false); | |
e2209710 BG |
6293 | write_unlock(&kvm->mmu_lock); |
6294 | } | |
198c74f4 | 6295 | |
1f98f2bd | 6296 | if (tdp_mmu_enabled) { |
24ae4cfa | 6297 | read_lock(&kvm->mmu_lock); |
b64d740e | 6298 | kvm_tdp_mmu_wrprot_slot(kvm, memslot, start_level); |
24ae4cfa BG |
6299 | read_unlock(&kvm->mmu_lock); |
6300 | } | |
6aa8b732 | 6301 | } |
37a7d8b0 | 6302 | |
ada51a9d DM |
6303 | static inline bool need_topup(struct kvm_mmu_memory_cache *cache, int min) |
6304 | { | |
6305 | return kvm_mmu_memory_cache_nr_free_objects(cache) < min; | |
6306 | } | |
6307 | ||
6308 | static bool need_topup_split_caches_or_resched(struct kvm *kvm) | |
6309 | { | |
6310 | if (need_resched() || rwlock_needbreak(&kvm->mmu_lock)) | |
6311 | return true; | |
6312 | ||
6313 | /* | |
6314 | * In the worst case, SPLIT_DESC_CACHE_MIN_NR_OBJECTS descriptors are needed | |
6315 | * to split a single huge page. Calculating how many are actually needed | |
6316 | * is possible but not worth the complexity. | |
6317 | */ | |
6318 | return need_topup(&kvm->arch.split_desc_cache, SPLIT_DESC_CACHE_MIN_NR_OBJECTS) || | |
6319 | need_topup(&kvm->arch.split_page_header_cache, 1) || | |
6320 | need_topup(&kvm->arch.split_shadow_page_cache, 1); | |
6321 | } | |
6322 | ||
6323 | static int topup_split_caches(struct kvm *kvm) | |
6324 | { | |
b9b71f43 SC |
6325 | /* |
6326 | * Allocating rmap list entries when splitting huge pages for nested | |
dfd4eb44 | 6327 | * MMUs is uncommon as KVM needs to use a list if and only if there is |
b9b71f43 | 6328 | * more than one rmap entry for a gfn, i.e. requires an L1 gfn to be |
dfd4eb44 SC |
6329 | * aliased by multiple L2 gfns and/or from multiple nested roots with |
6330 | * different roles. Aliasing gfns when using TDP is atypical for VMMs; | |
6331 | * a few gfns are often aliased during boot, e.g. when remapping BIOS, | |
6332 | * but aliasing rarely occurs post-boot or for many gfns. If there is | |
6333 | * only one rmap entry, rmap->val points directly at that one entry and | |
6334 | * doesn't need to allocate a list. Buffer the cache by the default | |
6335 | * capacity so that KVM doesn't have to drop mmu_lock to topup if KVM | |
b9b71f43 SC |
6336 | * encounters an aliased gfn or two. |
6337 | */ | |
6338 | const int capacity = SPLIT_DESC_CACHE_MIN_NR_OBJECTS + | |
6339 | KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE; | |
ada51a9d DM |
6340 | int r; |
6341 | ||
6342 | lockdep_assert_held(&kvm->slots_lock); | |
6343 | ||
b9b71f43 | 6344 | r = __kvm_mmu_topup_memory_cache(&kvm->arch.split_desc_cache, capacity, |
ada51a9d DM |
6345 | SPLIT_DESC_CACHE_MIN_NR_OBJECTS); |
6346 | if (r) | |
6347 | return r; | |
6348 | ||
6349 | r = kvm_mmu_topup_memory_cache(&kvm->arch.split_page_header_cache, 1); | |
6350 | if (r) | |
6351 | return r; | |
6352 | ||
6353 | return kvm_mmu_topup_memory_cache(&kvm->arch.split_shadow_page_cache, 1); | |
6354 | } | |
6355 | ||
6356 | static struct kvm_mmu_page *shadow_mmu_get_sp_for_split(struct kvm *kvm, u64 *huge_sptep) | |
6357 | { | |
6358 | struct kvm_mmu_page *huge_sp = sptep_to_sp(huge_sptep); | |
6359 | struct shadow_page_caches caches = {}; | |
6360 | union kvm_mmu_page_role role; | |
6361 | unsigned int access; | |
6362 | gfn_t gfn; | |
6363 | ||
79e48cec SC |
6364 | gfn = kvm_mmu_page_get_gfn(huge_sp, spte_index(huge_sptep)); |
6365 | access = kvm_mmu_page_get_access(huge_sp, spte_index(huge_sptep)); | |
ada51a9d DM |
6366 | |
6367 | /* | |
6368 | * Note, huge page splitting always uses direct shadow pages, regardless | |
6369 | * of whether the huge page itself is mapped by a direct or indirect | |
6370 | * shadow page, since the huge page region itself is being directly | |
6371 | * mapped with smaller pages. | |
6372 | */ | |
6373 | role = kvm_mmu_child_role(huge_sptep, /*direct=*/true, access); | |
6374 | ||
6375 | /* Direct SPs do not require a shadowed_info_cache. */ | |
6376 | caches.page_header_cache = &kvm->arch.split_page_header_cache; | |
6377 | caches.shadow_page_cache = &kvm->arch.split_shadow_page_cache; | |
6378 | ||
6379 | /* Safe to pass NULL for vCPU since requesting a direct SP. */ | |
6380 | return __kvm_mmu_get_shadow_page(kvm, NULL, &caches, gfn, role); | |
6381 | } | |
6382 | ||
6383 | static void shadow_mmu_split_huge_page(struct kvm *kvm, | |
6384 | const struct kvm_memory_slot *slot, | |
6385 | u64 *huge_sptep) | |
6386 | ||
6387 | { | |
6388 | struct kvm_mmu_memory_cache *cache = &kvm->arch.split_desc_cache; | |
6389 | u64 huge_spte = READ_ONCE(*huge_sptep); | |
6390 | struct kvm_mmu_page *sp; | |
03787394 | 6391 | bool flush = false; |
ada51a9d DM |
6392 | u64 *sptep, spte; |
6393 | gfn_t gfn; | |
6394 | int index; | |
6395 | ||
6396 | sp = shadow_mmu_get_sp_for_split(kvm, huge_sptep); | |
6397 | ||
6398 | for (index = 0; index < SPTE_ENT_PER_PAGE; index++) { | |
6399 | sptep = &sp->spt[index]; | |
6400 | gfn = kvm_mmu_page_get_gfn(sp, index); | |
6401 | ||
6402 | /* | |
6403 | * The SP may already have populated SPTEs, e.g. if this huge | |
6404 | * page is aliased by multiple sptes with the same access | |
6405 | * permissions. These entries are guaranteed to map the same | |
6406 | * gfn-to-pfn translation since the SP is direct, so no need to | |
6407 | * modify them. | |
6408 | * | |
03787394 PB |
6409 | * However, if a given SPTE points to a lower level page table, |
6410 | * that lower level page table may only be partially populated. | |
6411 | * Installing such SPTEs would effectively unmap a potion of the | |
6412 | * huge page. Unmapping guest memory always requires a TLB flush | |
6413 | * since a subsequent operation on the unmapped regions would | |
6414 | * fail to detect the need to flush. | |
ada51a9d | 6415 | */ |
03787394 PB |
6416 | if (is_shadow_present_pte(*sptep)) { |
6417 | flush |= !is_last_spte(*sptep, sp->role.level); | |
ada51a9d | 6418 | continue; |
03787394 | 6419 | } |
ada51a9d DM |
6420 | |
6421 | spte = make_huge_page_split_spte(kvm, huge_spte, sp->role, index); | |
6422 | mmu_spte_set(sptep, spte); | |
6423 | __rmap_add(kvm, cache, slot, sptep, gfn, sp->role.access); | |
6424 | } | |
6425 | ||
03787394 | 6426 | __link_shadow_page(kvm, cache, huge_sptep, sp, flush); |
ada51a9d DM |
6427 | } |
6428 | ||
6429 | static int shadow_mmu_try_split_huge_page(struct kvm *kvm, | |
6430 | const struct kvm_memory_slot *slot, | |
6431 | u64 *huge_sptep) | |
6432 | { | |
6433 | struct kvm_mmu_page *huge_sp = sptep_to_sp(huge_sptep); | |
6434 | int level, r = 0; | |
6435 | gfn_t gfn; | |
6436 | u64 spte; | |
6437 | ||
6438 | /* Grab information for the tracepoint before dropping the MMU lock. */ | |
79e48cec | 6439 | gfn = kvm_mmu_page_get_gfn(huge_sp, spte_index(huge_sptep)); |
ada51a9d DM |
6440 | level = huge_sp->role.level; |
6441 | spte = *huge_sptep; | |
6442 | ||
6443 | if (kvm_mmu_available_pages(kvm) <= KVM_MIN_FREE_MMU_PAGES) { | |
6444 | r = -ENOSPC; | |
6445 | goto out; | |
6446 | } | |
6447 | ||
6448 | if (need_topup_split_caches_or_resched(kvm)) { | |
6449 | write_unlock(&kvm->mmu_lock); | |
6450 | cond_resched(); | |
6451 | /* | |
6452 | * If the topup succeeds, return -EAGAIN to indicate that the | |
6453 | * rmap iterator should be restarted because the MMU lock was | |
6454 | * dropped. | |
6455 | */ | |
6456 | r = topup_split_caches(kvm) ?: -EAGAIN; | |
6457 | write_lock(&kvm->mmu_lock); | |
6458 | goto out; | |
6459 | } | |
6460 | ||
6461 | shadow_mmu_split_huge_page(kvm, slot, huge_sptep); | |
6462 | ||
6463 | out: | |
6464 | trace_kvm_mmu_split_huge_page(gfn, spte, level, r); | |
6465 | return r; | |
6466 | } | |
6467 | ||
6468 | static bool shadow_mmu_try_split_huge_pages(struct kvm *kvm, | |
6469 | struct kvm_rmap_head *rmap_head, | |
6470 | const struct kvm_memory_slot *slot) | |
6471 | { | |
6472 | struct rmap_iterator iter; | |
6473 | struct kvm_mmu_page *sp; | |
6474 | u64 *huge_sptep; | |
6475 | int r; | |
6476 | ||
6477 | restart: | |
6478 | for_each_rmap_spte(rmap_head, &iter, huge_sptep) { | |
6479 | sp = sptep_to_sp(huge_sptep); | |
6480 | ||
6481 | /* TDP MMU is enabled, so rmap only contains nested MMU SPs. */ | |
6482 | if (WARN_ON_ONCE(!sp->role.guest_mode)) | |
6483 | continue; | |
6484 | ||
6485 | /* The rmaps should never contain non-leaf SPTEs. */ | |
6486 | if (WARN_ON_ONCE(!is_large_pte(*huge_sptep))) | |
6487 | continue; | |
6488 | ||
6489 | /* SPs with level >PG_LEVEL_4K should never by unsync. */ | |
6490 | if (WARN_ON_ONCE(sp->unsync)) | |
6491 | continue; | |
6492 | ||
6493 | /* Don't bother splitting huge pages on invalid SPs. */ | |
6494 | if (sp->role.invalid) | |
6495 | continue; | |
6496 | ||
6497 | r = shadow_mmu_try_split_huge_page(kvm, slot, huge_sptep); | |
6498 | ||
6499 | /* | |
6500 | * The split succeeded or needs to be retried because the MMU | |
6501 | * lock was dropped. Either way, restart the iterator to get it | |
6502 | * back into a consistent state. | |
6503 | */ | |
6504 | if (!r || r == -EAGAIN) | |
6505 | goto restart; | |
6506 | ||
6507 | /* The split failed and shouldn't be retried (e.g. -ENOMEM). */ | |
6508 | break; | |
6509 | } | |
6510 | ||
6511 | return false; | |
6512 | } | |
6513 | ||
6514 | static void kvm_shadow_mmu_try_split_huge_pages(struct kvm *kvm, | |
6515 | const struct kvm_memory_slot *slot, | |
6516 | gfn_t start, gfn_t end, | |
6517 | int target_level) | |
6518 | { | |
6519 | int level; | |
6520 | ||
6521 | /* | |
6522 | * Split huge pages starting with KVM_MAX_HUGEPAGE_LEVEL and working | |
6523 | * down to the target level. This ensures pages are recursively split | |
6524 | * all the way to the target level. There's no need to split pages | |
6525 | * already at the target level. | |
6526 | */ | |
727ae377 SC |
6527 | for (level = KVM_MAX_HUGEPAGE_LEVEL; level > target_level; level--) |
6528 | __walk_slot_rmaps(kvm, slot, shadow_mmu_try_split_huge_pages, | |
6529 | level, level, start, end - 1, true, false); | |
ada51a9d DM |
6530 | } |
6531 | ||
cb00a70b DM |
6532 | /* Must be called with the mmu_lock held in write-mode. */ |
6533 | void kvm_mmu_try_split_huge_pages(struct kvm *kvm, | |
6534 | const struct kvm_memory_slot *memslot, | |
6535 | u64 start, u64 end, | |
6536 | int target_level) | |
6537 | { | |
1f98f2bd | 6538 | if (!tdp_mmu_enabled) |
ada51a9d DM |
6539 | return; |
6540 | ||
6541 | if (kvm_memslots_have_rmaps(kvm)) | |
6542 | kvm_shadow_mmu_try_split_huge_pages(kvm, memslot, start, end, target_level); | |
6543 | ||
6544 | kvm_tdp_mmu_try_split_huge_pages(kvm, memslot, start, end, target_level, false); | |
cb00a70b DM |
6545 | |
6546 | /* | |
54aa699e | 6547 | * A TLB flush is unnecessary at this point for the same reasons as in |
cb00a70b DM |
6548 | * kvm_mmu_slot_try_split_huge_pages(). |
6549 | */ | |
6550 | } | |
6551 | ||
a3fe5dbd | 6552 | void kvm_mmu_slot_try_split_huge_pages(struct kvm *kvm, |
cb00a70b DM |
6553 | const struct kvm_memory_slot *memslot, |
6554 | int target_level) | |
a3fe5dbd DM |
6555 | { |
6556 | u64 start = memslot->base_gfn; | |
6557 | u64 end = start + memslot->npages; | |
6558 | ||
1f98f2bd | 6559 | if (!tdp_mmu_enabled) |
ada51a9d DM |
6560 | return; |
6561 | ||
6562 | if (kvm_memslots_have_rmaps(kvm)) { | |
6563 | write_lock(&kvm->mmu_lock); | |
6564 | kvm_shadow_mmu_try_split_huge_pages(kvm, memslot, start, end, target_level); | |
6565 | write_unlock(&kvm->mmu_lock); | |
a3fe5dbd DM |
6566 | } |
6567 | ||
ada51a9d DM |
6568 | read_lock(&kvm->mmu_lock); |
6569 | kvm_tdp_mmu_try_split_huge_pages(kvm, memslot, start, end, target_level, true); | |
6570 | read_unlock(&kvm->mmu_lock); | |
6571 | ||
a3fe5dbd DM |
6572 | /* |
6573 | * No TLB flush is necessary here. KVM will flush TLBs after | |
6574 | * write-protecting and/or clearing dirty on the newly split SPTEs to | |
6575 | * ensure that guest writes are reflected in the dirty log before the | |
6576 | * ioctl to enable dirty logging on this memslot completes. Since the | |
6577 | * split SPTEs retain the write and dirty bits of the huge SPTE, it is | |
6578 | * safe for KVM to decide if a TLB flush is necessary based on the split | |
6579 | * SPTEs. | |
6580 | */ | |
6581 | } | |
6582 | ||
3ea3b7fa | 6583 | static bool kvm_mmu_zap_collapsible_spte(struct kvm *kvm, |
0a234f5d | 6584 | struct kvm_rmap_head *rmap_head, |
269e9552 | 6585 | const struct kvm_memory_slot *slot) |
3ea3b7fa WL |
6586 | { |
6587 | u64 *sptep; | |
6588 | struct rmap_iterator iter; | |
6589 | int need_tlb_flush = 0; | |
3ea3b7fa WL |
6590 | struct kvm_mmu_page *sp; |
6591 | ||
0d536790 | 6592 | restart: |
018aabb5 | 6593 | for_each_rmap_spte(rmap_head, &iter, sptep) { |
57354682 | 6594 | sp = sptep_to_sp(sptep); |
3ea3b7fa WL |
6595 | |
6596 | /* | |
decf6333 XG |
6597 | * We cannot do huge page mapping for indirect shadow pages, |
6598 | * which are found on the last rmap (level = 1) when not using | |
6599 | * tdp; such shadow pages are synced with the page table in | |
6600 | * the guest, and the guest page table is using 4K page size | |
6601 | * mapping if the indirect sp has level = 1. | |
3ea3b7fa | 6602 | */ |
5d49f08c | 6603 | if (sp->role.direct && |
9eba50f8 | 6604 | sp->role.level < kvm_mmu_max_mapping_level(kvm, slot, sp->gfn, |
a8ac499b | 6605 | PG_LEVEL_NUM)) { |
9202aee8 | 6606 | kvm_zap_one_rmap_spte(kvm, rmap_head, sptep); |
40ef75a7 | 6607 | |
8a1300ff | 6608 | if (kvm_available_flush_remote_tlbs_range()) |
1b2dc736 | 6609 | kvm_flush_remote_tlbs_sptep(kvm, sptep); |
40ef75a7 LT |
6610 | else |
6611 | need_tlb_flush = 1; | |
6612 | ||
0d536790 XG |
6613 | goto restart; |
6614 | } | |
3ea3b7fa WL |
6615 | } |
6616 | ||
6617 | return need_tlb_flush; | |
6618 | } | |
6619 | ||
20d49186 DM |
6620 | static void kvm_rmap_zap_collapsible_sptes(struct kvm *kvm, |
6621 | const struct kvm_memory_slot *slot) | |
6622 | { | |
6623 | /* | |
6624 | * Note, use KVM_MAX_HUGEPAGE_LEVEL - 1 since there's no need to zap | |
6625 | * pages that are already mapped at the maximum hugepage level. | |
6626 | */ | |
727ae377 SC |
6627 | if (walk_slot_rmaps(kvm, slot, kvm_mmu_zap_collapsible_spte, |
6628 | PG_LEVEL_4K, KVM_MAX_HUGEPAGE_LEVEL - 1, true)) | |
619b5072 | 6629 | kvm_flush_remote_tlbs_memslot(kvm, slot); |
20d49186 DM |
6630 | } |
6631 | ||
3ea3b7fa | 6632 | void kvm_mmu_zap_collapsible_sptes(struct kvm *kvm, |
269e9552 | 6633 | const struct kvm_memory_slot *slot) |
3ea3b7fa | 6634 | { |
e2209710 BG |
6635 | if (kvm_memslots_have_rmaps(kvm)) { |
6636 | write_lock(&kvm->mmu_lock); | |
20d49186 | 6637 | kvm_rmap_zap_collapsible_sptes(kvm, slot); |
e2209710 BG |
6638 | write_unlock(&kvm->mmu_lock); |
6639 | } | |
2db6f772 | 6640 | |
1f98f2bd | 6641 | if (tdp_mmu_enabled) { |
2db6f772 | 6642 | read_lock(&kvm->mmu_lock); |
4b85c921 | 6643 | kvm_tdp_mmu_zap_collapsible_sptes(kvm, slot); |
2db6f772 BG |
6644 | read_unlock(&kvm->mmu_lock); |
6645 | } | |
3ea3b7fa WL |
6646 | } |
6647 | ||
f4b4b180 | 6648 | void kvm_mmu_slot_leaf_clear_dirty(struct kvm *kvm, |
269e9552 | 6649 | const struct kvm_memory_slot *memslot) |
f4b4b180 | 6650 | { |
e2209710 BG |
6651 | if (kvm_memslots_have_rmaps(kvm)) { |
6652 | write_lock(&kvm->mmu_lock); | |
610265ea DM |
6653 | /* |
6654 | * Clear dirty bits only on 4k SPTEs since the legacy MMU only | |
6655 | * support dirty logging at a 4k granularity. | |
6656 | */ | |
727ae377 | 6657 | walk_slot_rmaps_4k(kvm, memslot, __rmap_clear_dirty, false); |
e2209710 BG |
6658 | write_unlock(&kvm->mmu_lock); |
6659 | } | |
f4b4b180 | 6660 | |
1f98f2bd | 6661 | if (tdp_mmu_enabled) { |
24ae4cfa | 6662 | read_lock(&kvm->mmu_lock); |
b64d740e | 6663 | kvm_tdp_mmu_clear_dirty_slot(kvm, memslot); |
24ae4cfa BG |
6664 | read_unlock(&kvm->mmu_lock); |
6665 | } | |
6666 | ||
f4b4b180 | 6667 | /* |
b64d740e JS |
6668 | * The caller will flush the TLBs after this function returns. |
6669 | * | |
f4b4b180 KH |
6670 | * It's also safe to flush TLBs out of mmu lock here as currently this |
6671 | * function is only used for dirty logging, in which case flushing TLB | |
6672 | * out of mmu lock also guarantees no dirty pages will be lost in | |
6673 | * dirty_bitmap. | |
6674 | */ | |
f4b4b180 | 6675 | } |
f4b4b180 | 6676 | |
db0d70e6 | 6677 | static void kvm_mmu_zap_all(struct kvm *kvm) |
5304b8d3 XG |
6678 | { |
6679 | struct kvm_mmu_page *sp, *node; | |
7390de1e | 6680 | LIST_HEAD(invalid_list); |
83cdb568 | 6681 | int ign; |
5304b8d3 | 6682 | |
531810ca | 6683 | write_lock(&kvm->mmu_lock); |
5304b8d3 | 6684 | restart: |
8a674adc | 6685 | list_for_each_entry_safe(sp, node, &kvm->arch.active_mmu_pages, link) { |
20ba462d | 6686 | if (WARN_ON_ONCE(sp->role.invalid)) |
4771450c | 6687 | continue; |
92f58b5c | 6688 | if (__kvm_mmu_prepare_zap_page(kvm, sp, &invalid_list, &ign)) |
5304b8d3 | 6689 | goto restart; |
531810ca | 6690 | if (cond_resched_rwlock_write(&kvm->mmu_lock)) |
5304b8d3 XG |
6691 | goto restart; |
6692 | } | |
6693 | ||
4771450c | 6694 | kvm_mmu_commit_zap_page(kvm, &invalid_list); |
faaf05b0 | 6695 | |
1f98f2bd | 6696 | if (tdp_mmu_enabled) |
faaf05b0 BG |
6697 | kvm_tdp_mmu_zap_all(kvm); |
6698 | ||
531810ca | 6699 | write_unlock(&kvm->mmu_lock); |
5304b8d3 XG |
6700 | } |
6701 | ||
db0d70e6 SC |
6702 | void kvm_arch_flush_shadow_all(struct kvm *kvm) |
6703 | { | |
6704 | kvm_mmu_zap_all(kvm); | |
6705 | } | |
6706 | ||
6707 | void kvm_arch_flush_shadow_memslot(struct kvm *kvm, | |
6708 | struct kvm_memory_slot *slot) | |
6709 | { | |
eeb87272 | 6710 | kvm_mmu_zap_all_fast(kvm); |
db0d70e6 SC |
6711 | } |
6712 | ||
15248258 | 6713 | void kvm_mmu_invalidate_mmio_sptes(struct kvm *kvm, u64 gen) |
f8f55942 | 6714 | { |
20ba462d | 6715 | WARN_ON_ONCE(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS); |
e1359e2b | 6716 | |
164bf7e5 | 6717 | gen &= MMIO_SPTE_GEN_MASK; |
e1359e2b | 6718 | |
f8f55942 | 6719 | /* |
e1359e2b SC |
6720 | * Generation numbers are incremented in multiples of the number of |
6721 | * address spaces in order to provide unique generations across all | |
6722 | * address spaces. Strip what is effectively the address space | |
6723 | * modifier prior to checking for a wrap of the MMIO generation so | |
6724 | * that a wrap in any address space is detected. | |
6725 | */ | |
6726 | gen &= ~((u64)KVM_ADDRESS_SPACE_NUM - 1); | |
6727 | ||
f8f55942 | 6728 | /* |
e1359e2b | 6729 | * The very rare case: if the MMIO generation number has wrapped, |
f8f55942 | 6730 | * zap all shadow pages. |
f8f55942 | 6731 | */ |
e1359e2b | 6732 | if (unlikely(gen == 0)) { |
8d20bd63 | 6733 | kvm_debug_ratelimited("zapping shadow pages for mmio generation wraparound\n"); |
92f58b5c | 6734 | kvm_mmu_zap_all_fast(kvm); |
7a2e8aaf | 6735 | } |
f8f55942 XG |
6736 | } |
6737 | ||
f3d90f90 SC |
6738 | static unsigned long mmu_shrink_scan(struct shrinker *shrink, |
6739 | struct shrink_control *sc) | |
3ee16c81 IE |
6740 | { |
6741 | struct kvm *kvm; | |
1495f230 | 6742 | int nr_to_scan = sc->nr_to_scan; |
70534a73 | 6743 | unsigned long freed = 0; |
3ee16c81 | 6744 | |
0d9ce162 | 6745 | mutex_lock(&kvm_lock); |
3ee16c81 IE |
6746 | |
6747 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
3d56cbdf | 6748 | int idx; |
d98ba053 | 6749 | LIST_HEAD(invalid_list); |
3ee16c81 | 6750 | |
35f2d16b TY |
6751 | /* |
6752 | * Never scan more than sc->nr_to_scan VM instances. | |
6753 | * Will not hit this condition practically since we do not try | |
6754 | * to shrink more than one VM and it is very unlikely to see | |
6755 | * !n_used_mmu_pages so many times. | |
6756 | */ | |
6757 | if (!nr_to_scan--) | |
6758 | break; | |
19526396 GN |
6759 | /* |
6760 | * n_used_mmu_pages is accessed without holding kvm->mmu_lock | |
6761 | * here. We may skip a VM instance errorneosly, but we do not | |
6762 | * want to shrink a VM that only started to populate its MMU | |
6763 | * anyway. | |
6764 | */ | |
10605204 SC |
6765 | if (!kvm->arch.n_used_mmu_pages && |
6766 | !kvm_has_zapped_obsolete_pages(kvm)) | |
19526396 | 6767 | continue; |
19526396 | 6768 | |
f656ce01 | 6769 | idx = srcu_read_lock(&kvm->srcu); |
531810ca | 6770 | write_lock(&kvm->mmu_lock); |
3ee16c81 | 6771 | |
10605204 SC |
6772 | if (kvm_has_zapped_obsolete_pages(kvm)) { |
6773 | kvm_mmu_commit_zap_page(kvm, | |
6774 | &kvm->arch.zapped_obsolete_pages); | |
6775 | goto unlock; | |
6776 | } | |
6777 | ||
ebdb292d | 6778 | freed = kvm_mmu_zap_oldest_mmu_pages(kvm, sc->nr_to_scan); |
19526396 | 6779 | |
10605204 | 6780 | unlock: |
531810ca | 6781 | write_unlock(&kvm->mmu_lock); |
f656ce01 | 6782 | srcu_read_unlock(&kvm->srcu, idx); |
19526396 | 6783 | |
70534a73 DC |
6784 | /* |
6785 | * unfair on small ones | |
6786 | * per-vm shrinkers cry out | |
6787 | * sadness comes quickly | |
6788 | */ | |
19526396 GN |
6789 | list_move_tail(&kvm->vm_list, &vm_list); |
6790 | break; | |
3ee16c81 | 6791 | } |
3ee16c81 | 6792 | |
0d9ce162 | 6793 | mutex_unlock(&kvm_lock); |
70534a73 | 6794 | return freed; |
70534a73 DC |
6795 | } |
6796 | ||
f3d90f90 SC |
6797 | static unsigned long mmu_shrink_count(struct shrinker *shrink, |
6798 | struct shrink_control *sc) | |
70534a73 | 6799 | { |
45221ab6 | 6800 | return percpu_counter_read_positive(&kvm_total_used_mmu_pages); |
3ee16c81 IE |
6801 | } |
6802 | ||
e5985c40 | 6803 | static struct shrinker *mmu_shrinker; |
3ee16c81 | 6804 | |
2ddfd20e | 6805 | static void mmu_destroy_caches(void) |
b5a33a75 | 6806 | { |
c1bd743e TH |
6807 | kmem_cache_destroy(pte_list_desc_cache); |
6808 | kmem_cache_destroy(mmu_page_header_cache); | |
b5a33a75 AK |
6809 | } |
6810 | ||
0b210faf SC |
6811 | static int get_nx_huge_pages(char *buffer, const struct kernel_param *kp) |
6812 | { | |
6813 | if (nx_hugepage_mitigation_hard_disabled) | |
1d6664fa | 6814 | return sysfs_emit(buffer, "never\n"); |
0b210faf SC |
6815 | |
6816 | return param_get_bool(buffer, kp); | |
6817 | } | |
6818 | ||
b8e8c830 PB |
6819 | static bool get_nx_auto_mode(void) |
6820 | { | |
6821 | /* Return true when CPU has the bug, and mitigations are ON */ | |
6822 | return boot_cpu_has_bug(X86_BUG_ITLB_MULTIHIT) && !cpu_mitigations_off(); | |
6823 | } | |
6824 | ||
6825 | static void __set_nx_huge_pages(bool val) | |
6826 | { | |
6827 | nx_huge_pages = itlb_multihit_kvm_mitigation = val; | |
6828 | } | |
6829 | ||
6830 | static int set_nx_huge_pages(const char *val, const struct kernel_param *kp) | |
6831 | { | |
6832 | bool old_val = nx_huge_pages; | |
6833 | bool new_val; | |
6834 | ||
0b210faf SC |
6835 | if (nx_hugepage_mitigation_hard_disabled) |
6836 | return -EPERM; | |
6837 | ||
b8e8c830 | 6838 | /* In "auto" mode deploy workaround only if CPU has the bug. */ |
0b210faf | 6839 | if (sysfs_streq(val, "off")) { |
b8e8c830 | 6840 | new_val = 0; |
0b210faf | 6841 | } else if (sysfs_streq(val, "force")) { |
b8e8c830 | 6842 | new_val = 1; |
0b210faf | 6843 | } else if (sysfs_streq(val, "auto")) { |
b8e8c830 | 6844 | new_val = get_nx_auto_mode(); |
0b210faf SC |
6845 | } else if (sysfs_streq(val, "never")) { |
6846 | new_val = 0; | |
6847 | ||
6848 | mutex_lock(&kvm_lock); | |
6849 | if (!list_empty(&vm_list)) { | |
6850 | mutex_unlock(&kvm_lock); | |
6851 | return -EBUSY; | |
6852 | } | |
6853 | nx_hugepage_mitigation_hard_disabled = true; | |
6854 | mutex_unlock(&kvm_lock); | |
6855 | } else if (kstrtobool(val, &new_val) < 0) { | |
b8e8c830 | 6856 | return -EINVAL; |
0b210faf | 6857 | } |
b8e8c830 PB |
6858 | |
6859 | __set_nx_huge_pages(new_val); | |
6860 | ||
6861 | if (new_val != old_val) { | |
6862 | struct kvm *kvm; | |
b8e8c830 PB |
6863 | |
6864 | mutex_lock(&kvm_lock); | |
6865 | ||
6866 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
ed69a6cb | 6867 | mutex_lock(&kvm->slots_lock); |
b8e8c830 | 6868 | kvm_mmu_zap_all_fast(kvm); |
ed69a6cb | 6869 | mutex_unlock(&kvm->slots_lock); |
1aa9b957 | 6870 | |
55c510e2 | 6871 | wake_up_process(kvm->arch.nx_huge_page_recovery_thread); |
b8e8c830 PB |
6872 | } |
6873 | mutex_unlock(&kvm_lock); | |
6874 | } | |
6875 | ||
6876 | return 0; | |
6877 | } | |
6878 | ||
1d0e8480 SC |
6879 | /* |
6880 | * nx_huge_pages needs to be resolved to true/false when kvm.ko is loaded, as | |
6881 | * its default value of -1 is technically undefined behavior for a boolean. | |
c3e0c8c2 SC |
6882 | * Forward the module init call to SPTE code so that it too can handle module |
6883 | * params that need to be resolved/snapshot. | |
1d0e8480 | 6884 | */ |
982bae43 | 6885 | void __init kvm_mmu_x86_module_init(void) |
b5a33a75 | 6886 | { |
b8e8c830 PB |
6887 | if (nx_huge_pages == -1) |
6888 | __set_nx_huge_pages(get_nx_auto_mode()); | |
c3e0c8c2 | 6889 | |
1f98f2bd DM |
6890 | /* |
6891 | * Snapshot userspace's desire to enable the TDP MMU. Whether or not the | |
6892 | * TDP MMU is actually enabled is determined in kvm_configure_mmu() | |
6893 | * when the vendor module is loaded. | |
6894 | */ | |
6895 | tdp_mmu_allowed = tdp_mmu_enabled; | |
6896 | ||
c3e0c8c2 | 6897 | kvm_mmu_spte_module_init(); |
1d0e8480 SC |
6898 | } |
6899 | ||
6900 | /* | |
6901 | * The bulk of the MMU initialization is deferred until the vendor module is | |
6902 | * loaded as many of the masks/values may be modified by VMX or SVM, i.e. need | |
6903 | * to be reset when a potentially different vendor module is loaded. | |
6904 | */ | |
6905 | int kvm_mmu_vendor_module_init(void) | |
6906 | { | |
6907 | int ret = -ENOMEM; | |
b8e8c830 | 6908 | |
36d9594d VK |
6909 | /* |
6910 | * MMU roles use union aliasing which is, generally speaking, an | |
6911 | * undefined behavior. However, we supposedly know how compilers behave | |
6912 | * and the current status quo is unlikely to change. Guardians below are | |
6913 | * supposed to let us know if the assumption becomes false. | |
6914 | */ | |
6915 | BUILD_BUG_ON(sizeof(union kvm_mmu_page_role) != sizeof(u32)); | |
6916 | BUILD_BUG_ON(sizeof(union kvm_mmu_extended_role) != sizeof(u32)); | |
7a7ae829 | 6917 | BUILD_BUG_ON(sizeof(union kvm_cpu_role) != sizeof(u64)); |
36d9594d | 6918 | |
28a1f3ac | 6919 | kvm_mmu_reset_all_pte_masks(); |
f160c7b7 | 6920 | |
53c07b18 XG |
6921 | pte_list_desc_cache = kmem_cache_create("pte_list_desc", |
6922 | sizeof(struct pte_list_desc), | |
46bea48a | 6923 | 0, SLAB_ACCOUNT, NULL); |
53c07b18 | 6924 | if (!pte_list_desc_cache) |
ab271bd4 | 6925 | goto out; |
b5a33a75 | 6926 | |
d3d25b04 AK |
6927 | mmu_page_header_cache = kmem_cache_create("kvm_mmu_page_header", |
6928 | sizeof(struct kvm_mmu_page), | |
46bea48a | 6929 | 0, SLAB_ACCOUNT, NULL); |
d3d25b04 | 6930 | if (!mmu_page_header_cache) |
ab271bd4 | 6931 | goto out; |
d3d25b04 | 6932 | |
908c7f19 | 6933 | if (percpu_counter_init(&kvm_total_used_mmu_pages, 0, GFP_KERNEL)) |
ab271bd4 | 6934 | goto out; |
45bf21a8 | 6935 | |
e5985c40 QZ |
6936 | mmu_shrinker = shrinker_alloc(0, "x86-mmu"); |
6937 | if (!mmu_shrinker) | |
d7c9bfb9 | 6938 | goto out_shrinker; |
3ee16c81 | 6939 | |
e5985c40 QZ |
6940 | mmu_shrinker->count_objects = mmu_shrink_count; |
6941 | mmu_shrinker->scan_objects = mmu_shrink_scan; | |
6942 | mmu_shrinker->seeks = DEFAULT_SEEKS * 10; | |
6943 | ||
6944 | shrinker_register(mmu_shrinker); | |
6945 | ||
b5a33a75 AK |
6946 | return 0; |
6947 | ||
d7c9bfb9 ML |
6948 | out_shrinker: |
6949 | percpu_counter_destroy(&kvm_total_used_mmu_pages); | |
ab271bd4 | 6950 | out: |
3ee16c81 | 6951 | mmu_destroy_caches(); |
ab271bd4 | 6952 | return ret; |
b5a33a75 AK |
6953 | } |
6954 | ||
c42fffe3 XG |
6955 | void kvm_mmu_destroy(struct kvm_vcpu *vcpu) |
6956 | { | |
95f93af4 | 6957 | kvm_mmu_unload(vcpu); |
1cfff4d9 JP |
6958 | free_mmu_pages(&vcpu->arch.root_mmu); |
6959 | free_mmu_pages(&vcpu->arch.guest_mmu); | |
c42fffe3 | 6960 | mmu_free_memory_caches(vcpu); |
b034cf01 XG |
6961 | } |
6962 | ||
1d0e8480 | 6963 | void kvm_mmu_vendor_module_exit(void) |
b034cf01 XG |
6964 | { |
6965 | mmu_destroy_caches(); | |
6966 | percpu_counter_destroy(&kvm_total_used_mmu_pages); | |
e5985c40 | 6967 | shrinker_free(mmu_shrinker); |
c42fffe3 | 6968 | } |
1aa9b957 | 6969 | |
f47491d7 SC |
6970 | /* |
6971 | * Calculate the effective recovery period, accounting for '0' meaning "let KVM | |
6972 | * select a halving time of 1 hour". Returns true if recovery is enabled. | |
6973 | */ | |
6974 | static bool calc_nx_huge_pages_recovery_period(uint *period) | |
6975 | { | |
6976 | /* | |
6977 | * Use READ_ONCE to get the params, this may be called outside of the | |
6978 | * param setters, e.g. by the kthread to compute its next timeout. | |
6979 | */ | |
6980 | bool enabled = READ_ONCE(nx_huge_pages); | |
6981 | uint ratio = READ_ONCE(nx_huge_pages_recovery_ratio); | |
6982 | ||
6983 | if (!enabled || !ratio) | |
6984 | return false; | |
6985 | ||
6986 | *period = READ_ONCE(nx_huge_pages_recovery_period_ms); | |
6987 | if (!*period) { | |
6988 | /* Make sure the period is not less than one second. */ | |
6989 | ratio = min(ratio, 3600u); | |
6990 | *period = 60 * 60 * 1000 / ratio; | |
6991 | } | |
6992 | return true; | |
6993 | } | |
6994 | ||
4dfe4f40 | 6995 | static int set_nx_huge_pages_recovery_param(const char *val, const struct kernel_param *kp) |
1aa9b957 | 6996 | { |
4dfe4f40 JS |
6997 | bool was_recovery_enabled, is_recovery_enabled; |
6998 | uint old_period, new_period; | |
1aa9b957 JS |
6999 | int err; |
7000 | ||
0b210faf SC |
7001 | if (nx_hugepage_mitigation_hard_disabled) |
7002 | return -EPERM; | |
7003 | ||
f47491d7 | 7004 | was_recovery_enabled = calc_nx_huge_pages_recovery_period(&old_period); |
4dfe4f40 | 7005 | |
1aa9b957 JS |
7006 | err = param_set_uint(val, kp); |
7007 | if (err) | |
7008 | return err; | |
7009 | ||
f47491d7 | 7010 | is_recovery_enabled = calc_nx_huge_pages_recovery_period(&new_period); |
4dfe4f40 | 7011 | |
f47491d7 | 7012 | if (is_recovery_enabled && |
4dfe4f40 | 7013 | (!was_recovery_enabled || old_period > new_period)) { |
1aa9b957 JS |
7014 | struct kvm *kvm; |
7015 | ||
7016 | mutex_lock(&kvm_lock); | |
7017 | ||
7018 | list_for_each_entry(kvm, &vm_list, vm_list) | |
55c510e2 | 7019 | wake_up_process(kvm->arch.nx_huge_page_recovery_thread); |
1aa9b957 JS |
7020 | |
7021 | mutex_unlock(&kvm_lock); | |
7022 | } | |
7023 | ||
7024 | return err; | |
7025 | } | |
7026 | ||
55c510e2 | 7027 | static void kvm_recover_nx_huge_pages(struct kvm *kvm) |
1aa9b957 | 7028 | { |
ade74e14 | 7029 | unsigned long nx_lpage_splits = kvm->stat.nx_lpage_splits; |
eb298605 | 7030 | struct kvm_memory_slot *slot; |
1aa9b957 JS |
7031 | int rcu_idx; |
7032 | struct kvm_mmu_page *sp; | |
7033 | unsigned int ratio; | |
7034 | LIST_HEAD(invalid_list); | |
048f4980 | 7035 | bool flush = false; |
1aa9b957 JS |
7036 | ulong to_zap; |
7037 | ||
7038 | rcu_idx = srcu_read_lock(&kvm->srcu); | |
531810ca | 7039 | write_lock(&kvm->mmu_lock); |
1aa9b957 | 7040 | |
bb95dfb9 SC |
7041 | /* |
7042 | * Zapping TDP MMU shadow pages, including the remote TLB flush, must | |
7043 | * be done under RCU protection, because the pages are freed via RCU | |
7044 | * callback. | |
7045 | */ | |
7046 | rcu_read_lock(); | |
7047 | ||
1aa9b957 | 7048 | ratio = READ_ONCE(nx_huge_pages_recovery_ratio); |
ade74e14 | 7049 | to_zap = ratio ? DIV_ROUND_UP(nx_lpage_splits, ratio) : 0; |
7d919c7a | 7050 | for ( ; to_zap; --to_zap) { |
55c510e2 | 7051 | if (list_empty(&kvm->arch.possible_nx_huge_pages)) |
7d919c7a SC |
7052 | break; |
7053 | ||
1aa9b957 JS |
7054 | /* |
7055 | * We use a separate list instead of just using active_mmu_pages | |
55c510e2 SC |
7056 | * because the number of shadow pages that be replaced with an |
7057 | * NX huge page is expected to be relatively small compared to | |
7058 | * the total number of shadow pages. And because the TDP MMU | |
7059 | * doesn't use active_mmu_pages. | |
1aa9b957 | 7060 | */ |
55c510e2 | 7061 | sp = list_first_entry(&kvm->arch.possible_nx_huge_pages, |
1aa9b957 | 7062 | struct kvm_mmu_page, |
55c510e2 SC |
7063 | possible_nx_huge_page_link); |
7064 | WARN_ON_ONCE(!sp->nx_huge_page_disallowed); | |
eb298605 DM |
7065 | WARN_ON_ONCE(!sp->role.direct); |
7066 | ||
eb298605 DM |
7067 | /* |
7068 | * Unaccount and do not attempt to recover any NX Huge Pages | |
7069 | * that are being dirty tracked, as they would just be faulted | |
7070 | * back in as 4KiB pages. The NX Huge Pages in this slot will be | |
7071 | * recovered, along with all the other huge pages in the slot, | |
7072 | * when dirty logging is disabled. | |
6c7b2202 PB |
7073 | * |
7074 | * Since gfn_to_memslot() is relatively expensive, it helps to | |
7075 | * skip it if it the test cannot possibly return true. On the | |
7076 | * other hand, if any memslot has logging enabled, chances are | |
7077 | * good that all of them do, in which case unaccount_nx_huge_page() | |
7078 | * is much cheaper than zapping the page. | |
7079 | * | |
7080 | * If a memslot update is in progress, reading an incorrect value | |
7081 | * of kvm->nr_memslots_dirty_logging is not a problem: if it is | |
7082 | * becoming zero, gfn_to_memslot() will be done unnecessarily; if | |
7083 | * it is becoming nonzero, the page will be zapped unnecessarily. | |
7084 | * Either way, this only affects efficiency in racy situations, | |
7085 | * and not correctness. | |
eb298605 | 7086 | */ |
6c7b2202 PB |
7087 | slot = NULL; |
7088 | if (atomic_read(&kvm->nr_memslots_dirty_logging)) { | |
817fa998 SC |
7089 | struct kvm_memslots *slots; |
7090 | ||
7091 | slots = kvm_memslots_for_spte_role(kvm, sp->role); | |
7092 | slot = __gfn_to_memslot(slots, sp->gfn); | |
6c7b2202 PB |
7093 | WARN_ON_ONCE(!slot); |
7094 | } | |
7095 | ||
eb298605 DM |
7096 | if (slot && kvm_slot_dirty_track_enabled(slot)) |
7097 | unaccount_nx_huge_page(kvm, sp); | |
7098 | else if (is_tdp_mmu_page(sp)) | |
315f02c6 | 7099 | flush |= kvm_tdp_mmu_zap_sp(kvm, sp); |
3a056757 | 7100 | else |
29cf0f50 | 7101 | kvm_mmu_prepare_zap_page(kvm, sp, &invalid_list); |
3a056757 | 7102 | WARN_ON_ONCE(sp->nx_huge_page_disallowed); |
1aa9b957 | 7103 | |
531810ca | 7104 | if (need_resched() || rwlock_needbreak(&kvm->mmu_lock)) { |
048f4980 | 7105 | kvm_mmu_remote_flush_or_zap(kvm, &invalid_list, flush); |
bb95dfb9 SC |
7106 | rcu_read_unlock(); |
7107 | ||
531810ca | 7108 | cond_resched_rwlock_write(&kvm->mmu_lock); |
048f4980 | 7109 | flush = false; |
bb95dfb9 SC |
7110 | |
7111 | rcu_read_lock(); | |
1aa9b957 JS |
7112 | } |
7113 | } | |
048f4980 | 7114 | kvm_mmu_remote_flush_or_zap(kvm, &invalid_list, flush); |
1aa9b957 | 7115 | |
bb95dfb9 SC |
7116 | rcu_read_unlock(); |
7117 | ||
531810ca | 7118 | write_unlock(&kvm->mmu_lock); |
1aa9b957 JS |
7119 | srcu_read_unlock(&kvm->srcu, rcu_idx); |
7120 | } | |
7121 | ||
55c510e2 | 7122 | static long get_nx_huge_page_recovery_timeout(u64 start_time) |
1aa9b957 | 7123 | { |
f47491d7 SC |
7124 | bool enabled; |
7125 | uint period; | |
4dfe4f40 | 7126 | |
f47491d7 | 7127 | enabled = calc_nx_huge_pages_recovery_period(&period); |
4dfe4f40 | 7128 | |
f47491d7 SC |
7129 | return enabled ? start_time + msecs_to_jiffies(period) - get_jiffies_64() |
7130 | : MAX_SCHEDULE_TIMEOUT; | |
1aa9b957 JS |
7131 | } |
7132 | ||
55c510e2 | 7133 | static int kvm_nx_huge_page_recovery_worker(struct kvm *kvm, uintptr_t data) |
1aa9b957 JS |
7134 | { |
7135 | u64 start_time; | |
7136 | long remaining_time; | |
7137 | ||
7138 | while (true) { | |
7139 | start_time = get_jiffies_64(); | |
55c510e2 | 7140 | remaining_time = get_nx_huge_page_recovery_timeout(start_time); |
1aa9b957 JS |
7141 | |
7142 | set_current_state(TASK_INTERRUPTIBLE); | |
7143 | while (!kthread_should_stop() && remaining_time > 0) { | |
7144 | schedule_timeout(remaining_time); | |
55c510e2 | 7145 | remaining_time = get_nx_huge_page_recovery_timeout(start_time); |
1aa9b957 JS |
7146 | set_current_state(TASK_INTERRUPTIBLE); |
7147 | } | |
7148 | ||
7149 | set_current_state(TASK_RUNNING); | |
7150 | ||
7151 | if (kthread_should_stop()) | |
7152 | return 0; | |
7153 | ||
55c510e2 | 7154 | kvm_recover_nx_huge_pages(kvm); |
1aa9b957 JS |
7155 | } |
7156 | } | |
7157 | ||
7158 | int kvm_mmu_post_init_vm(struct kvm *kvm) | |
7159 | { | |
7160 | int err; | |
7161 | ||
0b210faf SC |
7162 | if (nx_hugepage_mitigation_hard_disabled) |
7163 | return 0; | |
7164 | ||
55c510e2 | 7165 | err = kvm_vm_create_worker_thread(kvm, kvm_nx_huge_page_recovery_worker, 0, |
1aa9b957 | 7166 | "kvm-nx-lpage-recovery", |
55c510e2 | 7167 | &kvm->arch.nx_huge_page_recovery_thread); |
1aa9b957 | 7168 | if (!err) |
55c510e2 | 7169 | kthread_unpark(kvm->arch.nx_huge_page_recovery_thread); |
1aa9b957 JS |
7170 | |
7171 | return err; | |
7172 | } | |
7173 | ||
7174 | void kvm_mmu_pre_destroy_vm(struct kvm *kvm) | |
7175 | { | |
55c510e2 SC |
7176 | if (kvm->arch.nx_huge_page_recovery_thread) |
7177 | kthread_stop(kvm->arch.nx_huge_page_recovery_thread); | |
1aa9b957 | 7178 | } |