]>
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 | */ |
e495606d | 17 | |
af585b92 | 18 | #include "irq.h" |
88197e6a | 19 | #include "ioapic.h" |
1d737c8a | 20 | #include "mmu.h" |
6ca9a6f3 | 21 | #include "mmu_internal.h" |
836a1b3c | 22 | #include "x86.h" |
6de4f3ad | 23 | #include "kvm_cache_regs.h" |
2f728d66 | 24 | #include "kvm_emulate.h" |
5f7dde7b | 25 | #include "cpuid.h" |
e495606d | 26 | |
edf88417 | 27 | #include <linux/kvm_host.h> |
6aa8b732 AK |
28 | #include <linux/types.h> |
29 | #include <linux/string.h> | |
6aa8b732 AK |
30 | #include <linux/mm.h> |
31 | #include <linux/highmem.h> | |
1767e931 PG |
32 | #include <linux/moduleparam.h> |
33 | #include <linux/export.h> | |
448353ca | 34 | #include <linux/swap.h> |
05da4558 | 35 | #include <linux/hugetlb.h> |
2f333bcb | 36 | #include <linux/compiler.h> |
bc6678a3 | 37 | #include <linux/srcu.h> |
5a0e3ad6 | 38 | #include <linux/slab.h> |
3f07c014 | 39 | #include <linux/sched/signal.h> |
bf998156 | 40 | #include <linux/uaccess.h> |
114df303 | 41 | #include <linux/hash.h> |
f160c7b7 | 42 | #include <linux/kern_levels.h> |
1aa9b957 | 43 | #include <linux/kthread.h> |
6aa8b732 | 44 | |
e495606d | 45 | #include <asm/page.h> |
eb243d1d | 46 | #include <asm/memtype.h> |
e495606d | 47 | #include <asm/cmpxchg.h> |
0c55671f | 48 | #include <asm/e820/api.h> |
4e542370 | 49 | #include <asm/io.h> |
13673a90 | 50 | #include <asm/vmx.h> |
3d0c27ad | 51 | #include <asm/kvm_page_track.h> |
1261bfa3 | 52 | #include "trace.h" |
6aa8b732 | 53 | |
b8e8c830 PB |
54 | extern bool itlb_multihit_kvm_mitigation; |
55 | ||
56 | static int __read_mostly nx_huge_pages = -1; | |
13fb5927 PB |
57 | #ifdef CONFIG_PREEMPT_RT |
58 | /* Recovery can cause latency spikes, disable it for PREEMPT_RT. */ | |
59 | static uint __read_mostly nx_huge_pages_recovery_ratio = 0; | |
60 | #else | |
1aa9b957 | 61 | static uint __read_mostly nx_huge_pages_recovery_ratio = 60; |
13fb5927 | 62 | #endif |
b8e8c830 PB |
63 | |
64 | static int set_nx_huge_pages(const char *val, const struct kernel_param *kp); | |
1aa9b957 | 65 | static int set_nx_huge_pages_recovery_ratio(const char *val, const struct kernel_param *kp); |
b8e8c830 | 66 | |
d5d6c18d | 67 | static const struct kernel_param_ops nx_huge_pages_ops = { |
b8e8c830 PB |
68 | .set = set_nx_huge_pages, |
69 | .get = param_get_bool, | |
70 | }; | |
71 | ||
d5d6c18d | 72 | static const struct kernel_param_ops nx_huge_pages_recovery_ratio_ops = { |
1aa9b957 JS |
73 | .set = set_nx_huge_pages_recovery_ratio, |
74 | .get = param_get_uint, | |
75 | }; | |
76 | ||
b8e8c830 PB |
77 | module_param_cb(nx_huge_pages, &nx_huge_pages_ops, &nx_huge_pages, 0644); |
78 | __MODULE_PARM_TYPE(nx_huge_pages, "bool"); | |
1aa9b957 JS |
79 | module_param_cb(nx_huge_pages_recovery_ratio, &nx_huge_pages_recovery_ratio_ops, |
80 | &nx_huge_pages_recovery_ratio, 0644); | |
81 | __MODULE_PARM_TYPE(nx_huge_pages_recovery_ratio, "uint"); | |
b8e8c830 | 82 | |
71fe7013 SC |
83 | static bool __read_mostly force_flush_and_sync_on_reuse; |
84 | module_param_named(flush_on_reuse, force_flush_and_sync_on_reuse, bool, 0644); | |
85 | ||
18552672 JR |
86 | /* |
87 | * When setting this variable to true it enables Two-Dimensional-Paging | |
88 | * where the hardware walks 2 page tables: | |
89 | * 1. the guest-virtual to guest-physical | |
90 | * 2. while doing 1. it walks guest-physical to host-physical | |
91 | * If the hardware supports that we don't need to do shadow paging. | |
92 | */ | |
2f333bcb | 93 | bool tdp_enabled = false; |
18552672 | 94 | |
1d92d2e8 | 95 | static int max_huge_page_level __read_mostly; |
83013059 | 96 | static int max_tdp_level __read_mostly; |
703c335d | 97 | |
8b1fe17c XG |
98 | enum { |
99 | AUDIT_PRE_PAGE_FAULT, | |
100 | AUDIT_POST_PAGE_FAULT, | |
101 | AUDIT_PRE_PTE_WRITE, | |
6903074c XG |
102 | AUDIT_POST_PTE_WRITE, |
103 | AUDIT_PRE_SYNC, | |
104 | AUDIT_POST_SYNC | |
8b1fe17c | 105 | }; |
37a7d8b0 | 106 | |
8b1fe17c | 107 | #undef MMU_DEBUG |
37a7d8b0 AK |
108 | |
109 | #ifdef MMU_DEBUG | |
fa4a2c08 PB |
110 | static bool dbg = 0; |
111 | module_param(dbg, bool, 0644); | |
37a7d8b0 AK |
112 | |
113 | #define pgprintk(x...) do { if (dbg) printk(x); } while (0) | |
114 | #define rmap_printk(x...) do { if (dbg) printk(x); } while (0) | |
fa4a2c08 | 115 | #define MMU_WARN_ON(x) WARN_ON(x) |
37a7d8b0 | 116 | #else |
37a7d8b0 AK |
117 | #define pgprintk(x...) do { } while (0) |
118 | #define rmap_printk(x...) do { } while (0) | |
fa4a2c08 | 119 | #define MMU_WARN_ON(x) do { } while (0) |
d6c69ee9 | 120 | #endif |
6aa8b732 | 121 | |
957ed9ef XG |
122 | #define PTE_PREFETCH_NUM 8 |
123 | ||
00763e41 | 124 | #define PT_FIRST_AVAIL_BITS_SHIFT 10 |
6eeb4ef0 PB |
125 | #define PT64_SECOND_AVAIL_BITS_SHIFT 54 |
126 | ||
127 | /* | |
128 | * The mask used to denote special SPTEs, which can be either MMIO SPTEs or | |
129 | * Access Tracking SPTEs. | |
130 | */ | |
131 | #define SPTE_SPECIAL_MASK (3ULL << 52) | |
132 | #define SPTE_AD_ENABLED_MASK (0ULL << 52) | |
133 | #define SPTE_AD_DISABLED_MASK (1ULL << 52) | |
1f4e5fc8 | 134 | #define SPTE_AD_WRPROT_ONLY_MASK (2ULL << 52) |
6eeb4ef0 | 135 | #define SPTE_MMIO_MASK (3ULL << 52) |
6aa8b732 | 136 | |
6aa8b732 AK |
137 | #define PT64_LEVEL_BITS 9 |
138 | ||
139 | #define PT64_LEVEL_SHIFT(level) \ | |
d77c26fc | 140 | (PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS) |
6aa8b732 | 141 | |
6aa8b732 AK |
142 | #define PT64_INDEX(address, level)\ |
143 | (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1)) | |
144 | ||
145 | ||
146 | #define PT32_LEVEL_BITS 10 | |
147 | ||
148 | #define PT32_LEVEL_SHIFT(level) \ | |
d77c26fc | 149 | (PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS) |
6aa8b732 | 150 | |
e04da980 JR |
151 | #define PT32_LVL_OFFSET_MASK(level) \ |
152 | (PT32_BASE_ADDR_MASK & ((1ULL << (PAGE_SHIFT + (((level) - 1) \ | |
153 | * PT32_LEVEL_BITS))) - 1)) | |
6aa8b732 AK |
154 | |
155 | #define PT32_INDEX(address, level)\ | |
156 | (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1)) | |
157 | ||
158 | ||
8acc0993 KH |
159 | #ifdef CONFIG_DYNAMIC_PHYSICAL_MASK |
160 | #define PT64_BASE_ADDR_MASK (physical_mask & ~(u64)(PAGE_SIZE-1)) | |
161 | #else | |
162 | #define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1)) | |
163 | #endif | |
e04da980 JR |
164 | #define PT64_LVL_ADDR_MASK(level) \ |
165 | (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + (((level) - 1) \ | |
166 | * PT64_LEVEL_BITS))) - 1)) | |
167 | #define PT64_LVL_OFFSET_MASK(level) \ | |
168 | (PT64_BASE_ADDR_MASK & ((1ULL << (PAGE_SHIFT + (((level) - 1) \ | |
169 | * PT64_LEVEL_BITS))) - 1)) | |
6aa8b732 AK |
170 | |
171 | #define PT32_BASE_ADDR_MASK PAGE_MASK | |
172 | #define PT32_DIR_BASE_ADDR_MASK \ | |
173 | (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1)) | |
e04da980 JR |
174 | #define PT32_LVL_ADDR_MASK(level) \ |
175 | (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + (((level) - 1) \ | |
176 | * PT32_LEVEL_BITS))) - 1)) | |
6aa8b732 | 177 | |
53166229 | 178 | #define PT64_PERM_MASK (PT_PRESENT_MASK | PT_WRITABLE_MASK | shadow_user_mask \ |
d0ec49d4 | 179 | | shadow_x_mask | shadow_nx_mask | shadow_me_mask) |
6aa8b732 | 180 | |
fe135d2c AK |
181 | #define ACC_EXEC_MASK 1 |
182 | #define ACC_WRITE_MASK PT_WRITABLE_MASK | |
183 | #define ACC_USER_MASK PT_USER_MASK | |
184 | #define ACC_ALL (ACC_EXEC_MASK | ACC_WRITE_MASK | ACC_USER_MASK) | |
185 | ||
f160c7b7 JS |
186 | /* The mask for the R/X bits in EPT PTEs */ |
187 | #define PT64_EPT_READABLE_MASK 0x1ull | |
188 | #define PT64_EPT_EXECUTABLE_MASK 0x4ull | |
189 | ||
90bb6fc5 AK |
190 | #include <trace/events/kvm.h> |
191 | ||
49fde340 XG |
192 | #define SPTE_HOST_WRITEABLE (1ULL << PT_FIRST_AVAIL_BITS_SHIFT) |
193 | #define SPTE_MMU_WRITEABLE (1ULL << (PT_FIRST_AVAIL_BITS_SHIFT + 1)) | |
1403283a | 194 | |
135f8c2b AK |
195 | #define SHADOW_PT_INDEX(addr, level) PT64_INDEX(addr, level) |
196 | ||
220f773a TY |
197 | /* make pte_list_desc fit well in cache line */ |
198 | #define PTE_LIST_EXT 3 | |
199 | ||
9b8ebbdb | 200 | /* |
c4371c2a SC |
201 | * Return values of handle_mmio_page_fault, mmu.page_fault, and fast_page_fault(). |
202 | * | |
9b8ebbdb PB |
203 | * RET_PF_RETRY: let CPU fault again on the address. |
204 | * RET_PF_EMULATE: mmio page fault, emulate the instruction directly. | |
9b8ebbdb | 205 | * RET_PF_INVALID: the spte is invalid, let the real page fault path update it. |
c4371c2a SC |
206 | * RET_PF_FIXED: The faulting entry has been fixed. |
207 | * RET_PF_SPURIOUS: The faulting entry was already fixed, e.g. by another vCPU. | |
9b8ebbdb PB |
208 | */ |
209 | enum { | |
210 | RET_PF_RETRY = 0, | |
c4371c2a SC |
211 | RET_PF_EMULATE, |
212 | RET_PF_INVALID, | |
213 | RET_PF_FIXED, | |
214 | RET_PF_SPURIOUS, | |
9b8ebbdb PB |
215 | }; |
216 | ||
53c07b18 XG |
217 | struct pte_list_desc { |
218 | u64 *sptes[PTE_LIST_EXT]; | |
219 | struct pte_list_desc *more; | |
cd4a4e53 AK |
220 | }; |
221 | ||
2d11123a AK |
222 | struct kvm_shadow_walk_iterator { |
223 | u64 addr; | |
224 | hpa_t shadow_addr; | |
2d11123a | 225 | u64 *sptep; |
dd3bfd59 | 226 | int level; |
2d11123a AK |
227 | unsigned index; |
228 | }; | |
229 | ||
7eb77e9f JS |
230 | #define for_each_shadow_entry_using_root(_vcpu, _root, _addr, _walker) \ |
231 | for (shadow_walk_init_using_root(&(_walker), (_vcpu), \ | |
232 | (_root), (_addr)); \ | |
233 | shadow_walk_okay(&(_walker)); \ | |
234 | shadow_walk_next(&(_walker))) | |
235 | ||
236 | #define for_each_shadow_entry(_vcpu, _addr, _walker) \ | |
2d11123a AK |
237 | for (shadow_walk_init(&(_walker), _vcpu, _addr); \ |
238 | shadow_walk_okay(&(_walker)); \ | |
239 | shadow_walk_next(&(_walker))) | |
240 | ||
c2a2ac2b XG |
241 | #define for_each_shadow_entry_lockless(_vcpu, _addr, _walker, spte) \ |
242 | for (shadow_walk_init(&(_walker), _vcpu, _addr); \ | |
243 | shadow_walk_okay(&(_walker)) && \ | |
244 | ({ spte = mmu_spte_get_lockless(_walker.sptep); 1; }); \ | |
245 | __shadow_walk_next(&(_walker), spte)) | |
246 | ||
53c07b18 | 247 | static struct kmem_cache *pte_list_desc_cache; |
d3d25b04 | 248 | static struct kmem_cache *mmu_page_header_cache; |
45221ab6 | 249 | static struct percpu_counter kvm_total_used_mmu_pages; |
b5a33a75 | 250 | |
7b52345e SY |
251 | static u64 __read_mostly shadow_nx_mask; |
252 | static u64 __read_mostly shadow_x_mask; /* mutual exclusive with nx_mask */ | |
253 | static u64 __read_mostly shadow_user_mask; | |
254 | static u64 __read_mostly shadow_accessed_mask; | |
255 | static u64 __read_mostly shadow_dirty_mask; | |
dcdca5fe | 256 | static u64 __read_mostly shadow_mmio_value; |
4af77151 | 257 | static u64 __read_mostly shadow_mmio_access_mask; |
ffb128c8 | 258 | static u64 __read_mostly shadow_present_mask; |
d0ec49d4 | 259 | static u64 __read_mostly shadow_me_mask; |
ce88decf | 260 | |
f160c7b7 | 261 | /* |
6eeb4ef0 PB |
262 | * SPTEs used by MMUs without A/D bits are marked with SPTE_AD_DISABLED_MASK; |
263 | * shadow_acc_track_mask is the set of bits to be cleared in non-accessed | |
264 | * pages. | |
f160c7b7 JS |
265 | */ |
266 | static u64 __read_mostly shadow_acc_track_mask; | |
f160c7b7 JS |
267 | |
268 | /* | |
269 | * The mask/shift to use for saving the original R/X bits when marking the PTE | |
270 | * as not-present for access tracking purposes. We do not save the W bit as the | |
271 | * PTEs being access tracked also need to be dirty tracked, so the W bit will be | |
272 | * restored only when a write is attempted to the page. | |
273 | */ | |
274 | static const u64 shadow_acc_track_saved_bits_mask = PT64_EPT_READABLE_MASK | | |
275 | PT64_EPT_EXECUTABLE_MASK; | |
276 | static const u64 shadow_acc_track_saved_bits_shift = PT64_SECOND_AVAIL_BITS_SHIFT; | |
277 | ||
28a1f3ac JS |
278 | /* |
279 | * This mask must be set on all non-zero Non-Present or Reserved SPTEs in order | |
280 | * to guard against L1TF attacks. | |
281 | */ | |
282 | static u64 __read_mostly shadow_nonpresent_or_rsvd_mask; | |
283 | ||
284 | /* | |
285 | * The number of high-order 1 bits to use in the mask above. | |
286 | */ | |
287 | static const u64 shadow_nonpresent_or_rsvd_mask_len = 5; | |
288 | ||
daa07cbc SC |
289 | /* |
290 | * In some cases, we need to preserve the GFN of a non-present or reserved | |
291 | * SPTE when we usurp the upper five bits of the physical address space to | |
292 | * defend against L1TF, e.g. for MMIO SPTEs. To preserve the GFN, we'll | |
293 | * shift bits of the GFN that overlap with shadow_nonpresent_or_rsvd_mask | |
294 | * left into the reserved bits, i.e. the GFN in the SPTE will be split into | |
295 | * high and low parts. This mask covers the lower bits of the GFN. | |
296 | */ | |
297 | static u64 __read_mostly shadow_nonpresent_or_rsvd_lower_gfn_mask; | |
298 | ||
f3ecb59d KH |
299 | /* |
300 | * The number of non-reserved physical address bits irrespective of features | |
301 | * that repurpose legal bits, e.g. MKTME. | |
302 | */ | |
303 | static u8 __read_mostly shadow_phys_bits; | |
daa07cbc | 304 | |
ce88decf | 305 | static void mmu_spte_set(u64 *sptep, u64 spte); |
335e192a | 306 | static bool is_executable_pte(u64 spte); |
9fa72119 JS |
307 | static union kvm_mmu_page_role |
308 | kvm_mmu_calc_root_page_role(struct kvm_vcpu *vcpu); | |
ce88decf | 309 | |
335e192a PB |
310 | #define CREATE_TRACE_POINTS |
311 | #include "mmutrace.h" | |
312 | ||
40ef75a7 LT |
313 | |
314 | static inline bool kvm_available_flush_tlb_with_range(void) | |
315 | { | |
afaf0b2f | 316 | return kvm_x86_ops.tlb_remote_flush_with_range; |
40ef75a7 LT |
317 | } |
318 | ||
319 | static void kvm_flush_remote_tlbs_with_range(struct kvm *kvm, | |
320 | struct kvm_tlb_range *range) | |
321 | { | |
322 | int ret = -ENOTSUPP; | |
323 | ||
afaf0b2f SC |
324 | if (range && kvm_x86_ops.tlb_remote_flush_with_range) |
325 | ret = kvm_x86_ops.tlb_remote_flush_with_range(kvm, range); | |
40ef75a7 LT |
326 | |
327 | if (ret) | |
328 | kvm_flush_remote_tlbs(kvm); | |
329 | } | |
330 | ||
331 | static void kvm_flush_remote_tlbs_with_address(struct kvm *kvm, | |
332 | u64 start_gfn, u64 pages) | |
333 | { | |
334 | struct kvm_tlb_range range; | |
335 | ||
336 | range.start_gfn = start_gfn; | |
337 | range.pages = pages; | |
338 | ||
339 | kvm_flush_remote_tlbs_with_range(kvm, &range); | |
340 | } | |
341 | ||
e7581cac | 342 | void kvm_mmu_set_mmio_spte_mask(u64 mmio_value, u64 access_mask) |
ce88decf | 343 | { |
4af77151 | 344 | BUG_ON((u64)(unsigned)access_mask != access_mask); |
d43e2675 PB |
345 | WARN_ON(mmio_value & (shadow_nonpresent_or_rsvd_mask << shadow_nonpresent_or_rsvd_mask_len)); |
346 | WARN_ON(mmio_value & shadow_nonpresent_or_rsvd_lower_gfn_mask); | |
6eeb4ef0 | 347 | shadow_mmio_value = mmio_value | SPTE_MMIO_MASK; |
4af77151 | 348 | shadow_mmio_access_mask = access_mask; |
ce88decf XG |
349 | } |
350 | EXPORT_SYMBOL_GPL(kvm_mmu_set_mmio_spte_mask); | |
351 | ||
26c44a63 SC |
352 | static bool is_mmio_spte(u64 spte) |
353 | { | |
e7581cac | 354 | return (spte & SPTE_SPECIAL_MASK) == SPTE_MMIO_MASK; |
26c44a63 SC |
355 | } |
356 | ||
ac8d57e5 PF |
357 | static inline bool sp_ad_disabled(struct kvm_mmu_page *sp) |
358 | { | |
359 | return sp->role.ad_disabled; | |
360 | } | |
361 | ||
1f4e5fc8 PB |
362 | static inline bool kvm_vcpu_ad_need_write_protect(struct kvm_vcpu *vcpu) |
363 | { | |
364 | /* | |
365 | * When using the EPT page-modification log, the GPAs in the log | |
366 | * would come from L2 rather than L1. Therefore, we need to rely | |
367 | * on write protection to record dirty pages. This also bypasses | |
368 | * PML, since writes now result in a vmexit. | |
369 | */ | |
370 | return vcpu->arch.mmu == &vcpu->arch.guest_mmu; | |
371 | } | |
372 | ||
ac8d57e5 PF |
373 | static inline bool spte_ad_enabled(u64 spte) |
374 | { | |
26c44a63 | 375 | MMU_WARN_ON(is_mmio_spte(spte)); |
1f4e5fc8 PB |
376 | return (spte & SPTE_SPECIAL_MASK) != SPTE_AD_DISABLED_MASK; |
377 | } | |
378 | ||
379 | static inline bool spte_ad_need_write_protect(u64 spte) | |
380 | { | |
381 | MMU_WARN_ON(is_mmio_spte(spte)); | |
382 | return (spte & SPTE_SPECIAL_MASK) != SPTE_AD_ENABLED_MASK; | |
ac8d57e5 PF |
383 | } |
384 | ||
b8e8c830 PB |
385 | static bool is_nx_huge_page_enabled(void) |
386 | { | |
387 | return READ_ONCE(nx_huge_pages); | |
388 | } | |
389 | ||
ac8d57e5 PF |
390 | static inline u64 spte_shadow_accessed_mask(u64 spte) |
391 | { | |
26c44a63 | 392 | MMU_WARN_ON(is_mmio_spte(spte)); |
ac8d57e5 PF |
393 | return spte_ad_enabled(spte) ? shadow_accessed_mask : 0; |
394 | } | |
395 | ||
396 | static inline u64 spte_shadow_dirty_mask(u64 spte) | |
397 | { | |
26c44a63 | 398 | MMU_WARN_ON(is_mmio_spte(spte)); |
ac8d57e5 PF |
399 | return spte_ad_enabled(spte) ? shadow_dirty_mask : 0; |
400 | } | |
401 | ||
f160c7b7 JS |
402 | static inline bool is_access_track_spte(u64 spte) |
403 | { | |
ac8d57e5 | 404 | return !spte_ad_enabled(spte) && (spte & shadow_acc_track_mask) == 0; |
f160c7b7 JS |
405 | } |
406 | ||
f2fd125d | 407 | /* |
cae7ed3c SC |
408 | * Due to limited space in PTEs, the MMIO generation is a 19 bit subset of |
409 | * the memslots generation and is derived as follows: | |
ee3d1570 | 410 | * |
164bf7e5 SC |
411 | * Bits 0-8 of the MMIO generation are propagated to spte bits 3-11 |
412 | * Bits 9-18 of the MMIO generation are propagated to spte bits 52-61 | |
cae7ed3c | 413 | * |
164bf7e5 SC |
414 | * The KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS flag is intentionally not included in |
415 | * the MMIO generation number, as doing so would require stealing a bit from | |
416 | * the "real" generation number and thus effectively halve the maximum number | |
417 | * of MMIO generations that can be handled before encountering a wrap (which | |
418 | * requires a full MMU zap). The flag is instead explicitly queried when | |
419 | * checking for MMIO spte cache hits. | |
f2fd125d | 420 | */ |
56871d44 | 421 | #define MMIO_SPTE_GEN_MASK GENMASK_ULL(17, 0) |
f2fd125d | 422 | |
cae7ed3c SC |
423 | #define MMIO_SPTE_GEN_LOW_START 3 |
424 | #define MMIO_SPTE_GEN_LOW_END 11 | |
425 | #define MMIO_SPTE_GEN_LOW_MASK GENMASK_ULL(MMIO_SPTE_GEN_LOW_END, \ | |
426 | MMIO_SPTE_GEN_LOW_START) | |
f2fd125d | 427 | |
56871d44 PB |
428 | #define MMIO_SPTE_GEN_HIGH_START PT64_SECOND_AVAIL_BITS_SHIFT |
429 | #define MMIO_SPTE_GEN_HIGH_END 62 | |
cae7ed3c SC |
430 | #define MMIO_SPTE_GEN_HIGH_MASK GENMASK_ULL(MMIO_SPTE_GEN_HIGH_END, \ |
431 | MMIO_SPTE_GEN_HIGH_START) | |
56871d44 | 432 | |
5192f9b9 | 433 | static u64 generation_mmio_spte_mask(u64 gen) |
f2fd125d XG |
434 | { |
435 | u64 mask; | |
436 | ||
cae7ed3c | 437 | WARN_ON(gen & ~MMIO_SPTE_GEN_MASK); |
56871d44 | 438 | BUILD_BUG_ON((MMIO_SPTE_GEN_HIGH_MASK | MMIO_SPTE_GEN_LOW_MASK) & SPTE_SPECIAL_MASK); |
f2fd125d | 439 | |
cae7ed3c SC |
440 | mask = (gen << MMIO_SPTE_GEN_LOW_START) & MMIO_SPTE_GEN_LOW_MASK; |
441 | mask |= (gen << MMIO_SPTE_GEN_HIGH_START) & MMIO_SPTE_GEN_HIGH_MASK; | |
f2fd125d XG |
442 | return mask; |
443 | } | |
444 | ||
5192f9b9 | 445 | static u64 get_mmio_spte_generation(u64 spte) |
f2fd125d | 446 | { |
5192f9b9 | 447 | u64 gen; |
f2fd125d | 448 | |
cae7ed3c SC |
449 | gen = (spte & MMIO_SPTE_GEN_LOW_MASK) >> MMIO_SPTE_GEN_LOW_START; |
450 | gen |= (spte & MMIO_SPTE_GEN_HIGH_MASK) >> MMIO_SPTE_GEN_HIGH_START; | |
f2fd125d XG |
451 | return gen; |
452 | } | |
453 | ||
8f79b064 | 454 | static u64 make_mmio_spte(struct kvm_vcpu *vcpu, u64 gfn, unsigned int access) |
ce88decf | 455 | { |
8f79b064 | 456 | |
cae7ed3c | 457 | u64 gen = kvm_vcpu_memslots(vcpu)->generation & MMIO_SPTE_GEN_MASK; |
f8f55942 | 458 | u64 mask = generation_mmio_spte_mask(gen); |
28a1f3ac | 459 | u64 gpa = gfn << PAGE_SHIFT; |
95b0430d | 460 | |
4af77151 | 461 | access &= shadow_mmio_access_mask; |
28a1f3ac JS |
462 | mask |= shadow_mmio_value | access; |
463 | mask |= gpa | shadow_nonpresent_or_rsvd_mask; | |
464 | mask |= (gpa & shadow_nonpresent_or_rsvd_mask) | |
465 | << shadow_nonpresent_or_rsvd_mask_len; | |
f2fd125d | 466 | |
8f79b064 BG |
467 | return mask; |
468 | } | |
469 | ||
470 | static void mark_mmio_spte(struct kvm_vcpu *vcpu, u64 *sptep, u64 gfn, | |
471 | unsigned int access) | |
472 | { | |
473 | u64 mask = make_mmio_spte(vcpu, gfn, access); | |
474 | unsigned int gen = get_mmio_spte_generation(mask); | |
475 | ||
476 | access = mask & ACC_ALL; | |
477 | ||
f8f55942 | 478 | trace_mark_mmio_spte(sptep, gfn, access, gen); |
f2fd125d | 479 | mmu_spte_set(sptep, mask); |
ce88decf XG |
480 | } |
481 | ||
ce88decf XG |
482 | static gfn_t get_mmio_spte_gfn(u64 spte) |
483 | { | |
daa07cbc | 484 | u64 gpa = spte & shadow_nonpresent_or_rsvd_lower_gfn_mask; |
28a1f3ac JS |
485 | |
486 | gpa |= (spte >> shadow_nonpresent_or_rsvd_mask_len) | |
487 | & shadow_nonpresent_or_rsvd_mask; | |
488 | ||
489 | return gpa >> PAGE_SHIFT; | |
ce88decf XG |
490 | } |
491 | ||
492 | static unsigned get_mmio_spte_access(u64 spte) | |
493 | { | |
4af77151 | 494 | return spte & shadow_mmio_access_mask; |
ce88decf XG |
495 | } |
496 | ||
54bf36aa | 497 | static bool set_mmio_spte(struct kvm_vcpu *vcpu, u64 *sptep, gfn_t gfn, |
0a2b64c5 | 498 | kvm_pfn_t pfn, unsigned int access) |
ce88decf XG |
499 | { |
500 | if (unlikely(is_noslot_pfn(pfn))) { | |
54bf36aa | 501 | mark_mmio_spte(vcpu, sptep, gfn, access); |
ce88decf XG |
502 | return true; |
503 | } | |
504 | ||
505 | return false; | |
506 | } | |
c7addb90 | 507 | |
54bf36aa | 508 | static bool check_mmio_spte(struct kvm_vcpu *vcpu, u64 spte) |
f8f55942 | 509 | { |
cae7ed3c | 510 | u64 kvm_gen, spte_gen, gen; |
089504c0 | 511 | |
cae7ed3c SC |
512 | gen = kvm_vcpu_memslots(vcpu)->generation; |
513 | if (unlikely(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS)) | |
514 | return false; | |
089504c0 | 515 | |
cae7ed3c | 516 | kvm_gen = gen & MMIO_SPTE_GEN_MASK; |
089504c0 XG |
517 | spte_gen = get_mmio_spte_generation(spte); |
518 | ||
519 | trace_check_mmio_spte(spte, kvm_gen, spte_gen); | |
520 | return likely(kvm_gen == spte_gen); | |
f8f55942 XG |
521 | } |
522 | ||
cd313569 MG |
523 | static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access, |
524 | struct x86_exception *exception) | |
525 | { | |
ec7771ab | 526 | /* Check if guest physical address doesn't exceed guest maximum */ |
dc46515c | 527 | if (kvm_vcpu_is_illegal_gpa(vcpu, gpa)) { |
ec7771ab MG |
528 | exception->error_code |= PFERR_RSVD_MASK; |
529 | return UNMAPPED_GVA; | |
530 | } | |
531 | ||
cd313569 MG |
532 | return gpa; |
533 | } | |
534 | ||
ce00053b PF |
535 | /* |
536 | * Sets the shadow PTE masks used by the MMU. | |
537 | * | |
538 | * Assumptions: | |
539 | * - Setting either @accessed_mask or @dirty_mask requires setting both | |
540 | * - At least one of @accessed_mask or @acc_track_mask must be set | |
541 | */ | |
7b52345e | 542 | void kvm_mmu_set_mask_ptes(u64 user_mask, u64 accessed_mask, |
f160c7b7 | 543 | u64 dirty_mask, u64 nx_mask, u64 x_mask, u64 p_mask, |
d0ec49d4 | 544 | u64 acc_track_mask, u64 me_mask) |
7b52345e | 545 | { |
ce00053b PF |
546 | BUG_ON(!dirty_mask != !accessed_mask); |
547 | BUG_ON(!accessed_mask && !acc_track_mask); | |
6eeb4ef0 | 548 | BUG_ON(acc_track_mask & SPTE_SPECIAL_MASK); |
312b616b | 549 | |
7b52345e SY |
550 | shadow_user_mask = user_mask; |
551 | shadow_accessed_mask = accessed_mask; | |
552 | shadow_dirty_mask = dirty_mask; | |
553 | shadow_nx_mask = nx_mask; | |
554 | shadow_x_mask = x_mask; | |
ffb128c8 | 555 | shadow_present_mask = p_mask; |
f160c7b7 | 556 | shadow_acc_track_mask = acc_track_mask; |
d0ec49d4 | 557 | shadow_me_mask = me_mask; |
7b52345e SY |
558 | } |
559 | EXPORT_SYMBOL_GPL(kvm_mmu_set_mask_ptes); | |
560 | ||
f3ecb59d KH |
561 | static u8 kvm_get_shadow_phys_bits(void) |
562 | { | |
563 | /* | |
7adacf5e PB |
564 | * boot_cpu_data.x86_phys_bits is reduced when MKTME or SME are detected |
565 | * in CPU detection code, but the processor treats those reduced bits as | |
566 | * 'keyID' thus they are not reserved bits. Therefore KVM needs to look at | |
567 | * the physical address bits reported by CPUID. | |
f3ecb59d | 568 | */ |
7adacf5e PB |
569 | if (likely(boot_cpu_data.extended_cpuid_level >= 0x80000008)) |
570 | return cpuid_eax(0x80000008) & 0xff; | |
f3ecb59d | 571 | |
7adacf5e PB |
572 | /* |
573 | * Quite weird to have VMX or SVM but not MAXPHYADDR; probably a VM with | |
574 | * custom CPUID. Proceed with whatever the kernel found since these features | |
575 | * aren't virtualizable (SME/SEV also require CPUIDs higher than 0x80000008). | |
576 | */ | |
577 | return boot_cpu_data.x86_phys_bits; | |
f3ecb59d KH |
578 | } |
579 | ||
28a1f3ac | 580 | static void kvm_mmu_reset_all_pte_masks(void) |
f160c7b7 | 581 | { |
daa07cbc SC |
582 | u8 low_phys_bits; |
583 | ||
f160c7b7 JS |
584 | shadow_user_mask = 0; |
585 | shadow_accessed_mask = 0; | |
586 | shadow_dirty_mask = 0; | |
587 | shadow_nx_mask = 0; | |
588 | shadow_x_mask = 0; | |
f160c7b7 JS |
589 | shadow_present_mask = 0; |
590 | shadow_acc_track_mask = 0; | |
28a1f3ac | 591 | |
f3ecb59d KH |
592 | shadow_phys_bits = kvm_get_shadow_phys_bits(); |
593 | ||
28a1f3ac JS |
594 | /* |
595 | * If the CPU has 46 or less physical address bits, then set an | |
596 | * appropriate mask to guard against L1TF attacks. Otherwise, it is | |
597 | * assumed that the CPU is not vulnerable to L1TF. | |
61455bf2 KH |
598 | * |
599 | * Some Intel CPUs address the L1 cache using more PA bits than are | |
600 | * reported by CPUID. Use the PA width of the L1 cache when possible | |
601 | * to achieve more effective mitigation, e.g. if system RAM overlaps | |
602 | * the most significant bits of legal physical address space. | |
28a1f3ac | 603 | */ |
61455bf2 | 604 | shadow_nonpresent_or_rsvd_mask = 0; |
d43e2675 PB |
605 | low_phys_bits = boot_cpu_data.x86_phys_bits; |
606 | if (boot_cpu_has_bug(X86_BUG_L1TF) && | |
607 | !WARN_ON_ONCE(boot_cpu_data.x86_cache_bits >= | |
608 | 52 - shadow_nonpresent_or_rsvd_mask_len)) { | |
609 | low_phys_bits = boot_cpu_data.x86_cache_bits | |
610 | - shadow_nonpresent_or_rsvd_mask_len; | |
28a1f3ac | 611 | shadow_nonpresent_or_rsvd_mask = |
d43e2675 PB |
612 | rsvd_bits(low_phys_bits, boot_cpu_data.x86_cache_bits - 1); |
613 | } | |
61455bf2 | 614 | |
daa07cbc SC |
615 | shadow_nonpresent_or_rsvd_lower_gfn_mask = |
616 | GENMASK_ULL(low_phys_bits - 1, PAGE_SHIFT); | |
f160c7b7 JS |
617 | } |
618 | ||
6aa8b732 AK |
619 | static int is_cpuid_PSE36(void) |
620 | { | |
621 | return 1; | |
622 | } | |
623 | ||
73b1087e AK |
624 | static int is_nx(struct kvm_vcpu *vcpu) |
625 | { | |
f6801dff | 626 | return vcpu->arch.efer & EFER_NX; |
73b1087e AK |
627 | } |
628 | ||
c7addb90 AK |
629 | static int is_shadow_present_pte(u64 pte) |
630 | { | |
f160c7b7 | 631 | return (pte != 0) && !is_mmio_spte(pte); |
c7addb90 AK |
632 | } |
633 | ||
05da4558 MT |
634 | static int is_large_pte(u64 pte) |
635 | { | |
636 | return pte & PT_PAGE_SIZE_MASK; | |
637 | } | |
638 | ||
776e6633 MT |
639 | static int is_last_spte(u64 pte, int level) |
640 | { | |
3bae0459 | 641 | if (level == PG_LEVEL_4K) |
776e6633 | 642 | return 1; |
852e3c19 | 643 | if (is_large_pte(pte)) |
776e6633 MT |
644 | return 1; |
645 | return 0; | |
646 | } | |
647 | ||
d3e328f2 JS |
648 | static bool is_executable_pte(u64 spte) |
649 | { | |
650 | return (spte & (shadow_x_mask | shadow_nx_mask)) == shadow_x_mask; | |
651 | } | |
652 | ||
ba049e93 | 653 | static kvm_pfn_t spte_to_pfn(u64 pte) |
0b49ea86 | 654 | { |
35149e21 | 655 | return (pte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT; |
0b49ea86 AK |
656 | } |
657 | ||
da928521 AK |
658 | static gfn_t pse36_gfn_delta(u32 gpte) |
659 | { | |
660 | int shift = 32 - PT32_DIR_PSE36_SHIFT - PAGE_SHIFT; | |
661 | ||
662 | return (gpte & PT32_DIR_PSE36_MASK) << shift; | |
663 | } | |
664 | ||
603e0651 | 665 | #ifdef CONFIG_X86_64 |
d555c333 | 666 | static void __set_spte(u64 *sptep, u64 spte) |
e663ee64 | 667 | { |
b19ee2ff | 668 | WRITE_ONCE(*sptep, spte); |
e663ee64 AK |
669 | } |
670 | ||
603e0651 | 671 | static void __update_clear_spte_fast(u64 *sptep, u64 spte) |
a9221dd5 | 672 | { |
b19ee2ff | 673 | WRITE_ONCE(*sptep, spte); |
603e0651 XG |
674 | } |
675 | ||
676 | static u64 __update_clear_spte_slow(u64 *sptep, u64 spte) | |
677 | { | |
678 | return xchg(sptep, spte); | |
679 | } | |
c2a2ac2b XG |
680 | |
681 | static u64 __get_spte_lockless(u64 *sptep) | |
682 | { | |
6aa7de05 | 683 | return READ_ONCE(*sptep); |
c2a2ac2b | 684 | } |
a9221dd5 | 685 | #else |
603e0651 XG |
686 | union split_spte { |
687 | struct { | |
688 | u32 spte_low; | |
689 | u32 spte_high; | |
690 | }; | |
691 | u64 spte; | |
692 | }; | |
a9221dd5 | 693 | |
c2a2ac2b XG |
694 | static void count_spte_clear(u64 *sptep, u64 spte) |
695 | { | |
57354682 | 696 | struct kvm_mmu_page *sp = sptep_to_sp(sptep); |
c2a2ac2b XG |
697 | |
698 | if (is_shadow_present_pte(spte)) | |
699 | return; | |
700 | ||
701 | /* Ensure the spte is completely set before we increase the count */ | |
702 | smp_wmb(); | |
703 | sp->clear_spte_count++; | |
704 | } | |
705 | ||
603e0651 XG |
706 | static void __set_spte(u64 *sptep, u64 spte) |
707 | { | |
708 | union split_spte *ssptep, sspte; | |
a9221dd5 | 709 | |
603e0651 XG |
710 | ssptep = (union split_spte *)sptep; |
711 | sspte = (union split_spte)spte; | |
712 | ||
713 | ssptep->spte_high = sspte.spte_high; | |
714 | ||
715 | /* | |
716 | * If we map the spte from nonpresent to present, We should store | |
717 | * the high bits firstly, then set present bit, so cpu can not | |
718 | * fetch this spte while we are setting the spte. | |
719 | */ | |
720 | smp_wmb(); | |
721 | ||
b19ee2ff | 722 | WRITE_ONCE(ssptep->spte_low, sspte.spte_low); |
a9221dd5 AK |
723 | } |
724 | ||
603e0651 XG |
725 | static void __update_clear_spte_fast(u64 *sptep, u64 spte) |
726 | { | |
727 | union split_spte *ssptep, sspte; | |
728 | ||
729 | ssptep = (union split_spte *)sptep; | |
730 | sspte = (union split_spte)spte; | |
731 | ||
b19ee2ff | 732 | WRITE_ONCE(ssptep->spte_low, sspte.spte_low); |
603e0651 XG |
733 | |
734 | /* | |
735 | * If we map the spte from present to nonpresent, we should clear | |
736 | * present bit firstly to avoid vcpu fetch the old high bits. | |
737 | */ | |
738 | smp_wmb(); | |
739 | ||
740 | ssptep->spte_high = sspte.spte_high; | |
c2a2ac2b | 741 | count_spte_clear(sptep, spte); |
603e0651 XG |
742 | } |
743 | ||
744 | static u64 __update_clear_spte_slow(u64 *sptep, u64 spte) | |
745 | { | |
746 | union split_spte *ssptep, sspte, orig; | |
747 | ||
748 | ssptep = (union split_spte *)sptep; | |
749 | sspte = (union split_spte)spte; | |
750 | ||
751 | /* xchg acts as a barrier before the setting of the high bits */ | |
752 | orig.spte_low = xchg(&ssptep->spte_low, sspte.spte_low); | |
41bc3186 ZJ |
753 | orig.spte_high = ssptep->spte_high; |
754 | ssptep->spte_high = sspte.spte_high; | |
c2a2ac2b | 755 | count_spte_clear(sptep, spte); |
603e0651 XG |
756 | |
757 | return orig.spte; | |
758 | } | |
c2a2ac2b XG |
759 | |
760 | /* | |
761 | * The idea using the light way get the spte on x86_32 guest is from | |
39656e83 | 762 | * gup_get_pte (mm/gup.c). |
accaefe0 XG |
763 | * |
764 | * An spte tlb flush may be pending, because kvm_set_pte_rmapp | |
765 | * coalesces them and we are running out of the MMU lock. Therefore | |
766 | * we need to protect against in-progress updates of the spte. | |
767 | * | |
768 | * Reading the spte while an update is in progress may get the old value | |
769 | * for the high part of the spte. The race is fine for a present->non-present | |
770 | * change (because the high part of the spte is ignored for non-present spte), | |
771 | * but for a present->present change we must reread the spte. | |
772 | * | |
773 | * All such changes are done in two steps (present->non-present and | |
774 | * non-present->present), hence it is enough to count the number of | |
775 | * present->non-present updates: if it changed while reading the spte, | |
776 | * we might have hit the race. This is done using clear_spte_count. | |
c2a2ac2b XG |
777 | */ |
778 | static u64 __get_spte_lockless(u64 *sptep) | |
779 | { | |
57354682 | 780 | struct kvm_mmu_page *sp = sptep_to_sp(sptep); |
c2a2ac2b XG |
781 | union split_spte spte, *orig = (union split_spte *)sptep; |
782 | int count; | |
783 | ||
784 | retry: | |
785 | count = sp->clear_spte_count; | |
786 | smp_rmb(); | |
787 | ||
788 | spte.spte_low = orig->spte_low; | |
789 | smp_rmb(); | |
790 | ||
791 | spte.spte_high = orig->spte_high; | |
792 | smp_rmb(); | |
793 | ||
794 | if (unlikely(spte.spte_low != orig->spte_low || | |
795 | count != sp->clear_spte_count)) | |
796 | goto retry; | |
797 | ||
798 | return spte.spte; | |
799 | } | |
603e0651 XG |
800 | #endif |
801 | ||
ea4114bc | 802 | static bool spte_can_locklessly_be_made_writable(u64 spte) |
c7ba5b48 | 803 | { |
feb3eb70 GN |
804 | return (spte & (SPTE_HOST_WRITEABLE | SPTE_MMU_WRITEABLE)) == |
805 | (SPTE_HOST_WRITEABLE | SPTE_MMU_WRITEABLE); | |
c7ba5b48 XG |
806 | } |
807 | ||
8672b721 XG |
808 | static bool spte_has_volatile_bits(u64 spte) |
809 | { | |
f160c7b7 JS |
810 | if (!is_shadow_present_pte(spte)) |
811 | return false; | |
812 | ||
c7ba5b48 | 813 | /* |
6a6256f9 | 814 | * Always atomically update spte if it can be updated |
c7ba5b48 XG |
815 | * out of mmu-lock, it can ensure dirty bit is not lost, |
816 | * also, it can help us to get a stable is_writable_pte() | |
817 | * to ensure tlb flush is not missed. | |
818 | */ | |
f160c7b7 JS |
819 | if (spte_can_locklessly_be_made_writable(spte) || |
820 | is_access_track_spte(spte)) | |
c7ba5b48 XG |
821 | return true; |
822 | ||
ac8d57e5 | 823 | if (spte_ad_enabled(spte)) { |
f160c7b7 JS |
824 | if ((spte & shadow_accessed_mask) == 0 || |
825 | (is_writable_pte(spte) && (spte & shadow_dirty_mask) == 0)) | |
826 | return true; | |
827 | } | |
8672b721 | 828 | |
f160c7b7 | 829 | return false; |
8672b721 XG |
830 | } |
831 | ||
83ef6c81 | 832 | static bool is_accessed_spte(u64 spte) |
4132779b | 833 | { |
ac8d57e5 PF |
834 | u64 accessed_mask = spte_shadow_accessed_mask(spte); |
835 | ||
836 | return accessed_mask ? spte & accessed_mask | |
837 | : !is_access_track_spte(spte); | |
4132779b XG |
838 | } |
839 | ||
83ef6c81 | 840 | static bool is_dirty_spte(u64 spte) |
7e71a59b | 841 | { |
ac8d57e5 PF |
842 | u64 dirty_mask = spte_shadow_dirty_mask(spte); |
843 | ||
844 | return dirty_mask ? spte & dirty_mask : spte & PT_WRITABLE_MASK; | |
7e71a59b KH |
845 | } |
846 | ||
1df9f2dc XG |
847 | /* Rules for using mmu_spte_set: |
848 | * Set the sptep from nonpresent to present. | |
849 | * Note: the sptep being assigned *must* be either not present | |
850 | * or in a state where the hardware will not attempt to update | |
851 | * the spte. | |
852 | */ | |
853 | static void mmu_spte_set(u64 *sptep, u64 new_spte) | |
854 | { | |
855 | WARN_ON(is_shadow_present_pte(*sptep)); | |
856 | __set_spte(sptep, new_spte); | |
857 | } | |
858 | ||
f39a058d JS |
859 | /* |
860 | * Update the SPTE (excluding the PFN), but do not track changes in its | |
861 | * accessed/dirty status. | |
1df9f2dc | 862 | */ |
f39a058d | 863 | static u64 mmu_spte_update_no_track(u64 *sptep, u64 new_spte) |
b79b93f9 | 864 | { |
c7ba5b48 | 865 | u64 old_spte = *sptep; |
4132779b | 866 | |
afd28fe1 | 867 | WARN_ON(!is_shadow_present_pte(new_spte)); |
b79b93f9 | 868 | |
6e7d0354 XG |
869 | if (!is_shadow_present_pte(old_spte)) { |
870 | mmu_spte_set(sptep, new_spte); | |
f39a058d | 871 | return old_spte; |
6e7d0354 | 872 | } |
4132779b | 873 | |
c7ba5b48 | 874 | if (!spte_has_volatile_bits(old_spte)) |
603e0651 | 875 | __update_clear_spte_fast(sptep, new_spte); |
4132779b | 876 | else |
603e0651 | 877 | old_spte = __update_clear_spte_slow(sptep, new_spte); |
4132779b | 878 | |
83ef6c81 JS |
879 | WARN_ON(spte_to_pfn(old_spte) != spte_to_pfn(new_spte)); |
880 | ||
f39a058d JS |
881 | return old_spte; |
882 | } | |
883 | ||
884 | /* Rules for using mmu_spte_update: | |
885 | * Update the state bits, it means the mapped pfn is not changed. | |
886 | * | |
887 | * Whenever we overwrite a writable spte with a read-only one we | |
888 | * should flush remote TLBs. Otherwise rmap_write_protect | |
889 | * will find a read-only spte, even though the writable spte | |
890 | * might be cached on a CPU's TLB, the return value indicates this | |
891 | * case. | |
892 | * | |
893 | * Returns true if the TLB needs to be flushed | |
894 | */ | |
895 | static bool mmu_spte_update(u64 *sptep, u64 new_spte) | |
896 | { | |
897 | bool flush = false; | |
898 | u64 old_spte = mmu_spte_update_no_track(sptep, new_spte); | |
899 | ||
900 | if (!is_shadow_present_pte(old_spte)) | |
901 | return false; | |
902 | ||
c7ba5b48 XG |
903 | /* |
904 | * For the spte updated out of mmu-lock is safe, since | |
6a6256f9 | 905 | * we always atomically update it, see the comments in |
c7ba5b48 XG |
906 | * spte_has_volatile_bits(). |
907 | */ | |
ea4114bc | 908 | if (spte_can_locklessly_be_made_writable(old_spte) && |
7f31c959 | 909 | !is_writable_pte(new_spte)) |
83ef6c81 | 910 | flush = true; |
4132779b | 911 | |
7e71a59b | 912 | /* |
83ef6c81 | 913 | * Flush TLB when accessed/dirty states are changed in the page tables, |
7e71a59b KH |
914 | * to guarantee consistency between TLB and page tables. |
915 | */ | |
7e71a59b | 916 | |
83ef6c81 JS |
917 | if (is_accessed_spte(old_spte) && !is_accessed_spte(new_spte)) { |
918 | flush = true; | |
4132779b | 919 | kvm_set_pfn_accessed(spte_to_pfn(old_spte)); |
83ef6c81 JS |
920 | } |
921 | ||
922 | if (is_dirty_spte(old_spte) && !is_dirty_spte(new_spte)) { | |
923 | flush = true; | |
4132779b | 924 | kvm_set_pfn_dirty(spte_to_pfn(old_spte)); |
83ef6c81 | 925 | } |
6e7d0354 | 926 | |
83ef6c81 | 927 | return flush; |
b79b93f9 AK |
928 | } |
929 | ||
1df9f2dc XG |
930 | /* |
931 | * Rules for using mmu_spte_clear_track_bits: | |
932 | * It sets the sptep from present to nonpresent, and track the | |
933 | * state bits, it is used to clear the last level sptep. | |
83ef6c81 | 934 | * Returns non-zero if the PTE was previously valid. |
1df9f2dc XG |
935 | */ |
936 | static int mmu_spte_clear_track_bits(u64 *sptep) | |
937 | { | |
ba049e93 | 938 | kvm_pfn_t pfn; |
1df9f2dc XG |
939 | u64 old_spte = *sptep; |
940 | ||
941 | if (!spte_has_volatile_bits(old_spte)) | |
603e0651 | 942 | __update_clear_spte_fast(sptep, 0ull); |
1df9f2dc | 943 | else |
603e0651 | 944 | old_spte = __update_clear_spte_slow(sptep, 0ull); |
1df9f2dc | 945 | |
afd28fe1 | 946 | if (!is_shadow_present_pte(old_spte)) |
1df9f2dc XG |
947 | return 0; |
948 | ||
949 | pfn = spte_to_pfn(old_spte); | |
86fde74c XG |
950 | |
951 | /* | |
952 | * KVM does not hold the refcount of the page used by | |
953 | * kvm mmu, before reclaiming the page, we should | |
954 | * unmap it from mmu first. | |
955 | */ | |
bf4bea8e | 956 | WARN_ON(!kvm_is_reserved_pfn(pfn) && !page_count(pfn_to_page(pfn))); |
86fde74c | 957 | |
83ef6c81 | 958 | if (is_accessed_spte(old_spte)) |
1df9f2dc | 959 | kvm_set_pfn_accessed(pfn); |
83ef6c81 JS |
960 | |
961 | if (is_dirty_spte(old_spte)) | |
1df9f2dc | 962 | kvm_set_pfn_dirty(pfn); |
83ef6c81 | 963 | |
1df9f2dc XG |
964 | return 1; |
965 | } | |
966 | ||
967 | /* | |
968 | * Rules for using mmu_spte_clear_no_track: | |
969 | * Directly clear spte without caring the state bits of sptep, | |
970 | * it is used to set the upper level spte. | |
971 | */ | |
972 | static void mmu_spte_clear_no_track(u64 *sptep) | |
973 | { | |
603e0651 | 974 | __update_clear_spte_fast(sptep, 0ull); |
1df9f2dc XG |
975 | } |
976 | ||
c2a2ac2b XG |
977 | static u64 mmu_spte_get_lockless(u64 *sptep) |
978 | { | |
979 | return __get_spte_lockless(sptep); | |
980 | } | |
981 | ||
f160c7b7 JS |
982 | static u64 mark_spte_for_access_track(u64 spte) |
983 | { | |
ac8d57e5 | 984 | if (spte_ad_enabled(spte)) |
f160c7b7 JS |
985 | return spte & ~shadow_accessed_mask; |
986 | ||
ac8d57e5 | 987 | if (is_access_track_spte(spte)) |
f160c7b7 JS |
988 | return spte; |
989 | ||
990 | /* | |
20d65236 JS |
991 | * Making an Access Tracking PTE will result in removal of write access |
992 | * from the PTE. So, verify that we will be able to restore the write | |
993 | * access in the fast page fault path later on. | |
f160c7b7 JS |
994 | */ |
995 | WARN_ONCE((spte & PT_WRITABLE_MASK) && | |
996 | !spte_can_locklessly_be_made_writable(spte), | |
997 | "kvm: Writable SPTE is not locklessly dirty-trackable\n"); | |
998 | ||
999 | WARN_ONCE(spte & (shadow_acc_track_saved_bits_mask << | |
1000 | shadow_acc_track_saved_bits_shift), | |
1001 | "kvm: Access Tracking saved bit locations are not zero\n"); | |
1002 | ||
1003 | spte |= (spte & shadow_acc_track_saved_bits_mask) << | |
1004 | shadow_acc_track_saved_bits_shift; | |
1005 | spte &= ~shadow_acc_track_mask; | |
f160c7b7 JS |
1006 | |
1007 | return spte; | |
1008 | } | |
1009 | ||
d3e328f2 JS |
1010 | /* Restore an acc-track PTE back to a regular PTE */ |
1011 | static u64 restore_acc_track_spte(u64 spte) | |
1012 | { | |
1013 | u64 new_spte = spte; | |
1014 | u64 saved_bits = (spte >> shadow_acc_track_saved_bits_shift) | |
1015 | & shadow_acc_track_saved_bits_mask; | |
1016 | ||
ac8d57e5 | 1017 | WARN_ON_ONCE(spte_ad_enabled(spte)); |
d3e328f2 JS |
1018 | WARN_ON_ONCE(!is_access_track_spte(spte)); |
1019 | ||
1020 | new_spte &= ~shadow_acc_track_mask; | |
1021 | new_spte &= ~(shadow_acc_track_saved_bits_mask << | |
1022 | shadow_acc_track_saved_bits_shift); | |
1023 | new_spte |= saved_bits; | |
1024 | ||
1025 | return new_spte; | |
1026 | } | |
1027 | ||
f160c7b7 JS |
1028 | /* Returns the Accessed status of the PTE and resets it at the same time. */ |
1029 | static bool mmu_spte_age(u64 *sptep) | |
1030 | { | |
1031 | u64 spte = mmu_spte_get_lockless(sptep); | |
1032 | ||
1033 | if (!is_accessed_spte(spte)) | |
1034 | return false; | |
1035 | ||
ac8d57e5 | 1036 | if (spte_ad_enabled(spte)) { |
f160c7b7 JS |
1037 | clear_bit((ffs(shadow_accessed_mask) - 1), |
1038 | (unsigned long *)sptep); | |
1039 | } else { | |
1040 | /* | |
1041 | * Capture the dirty status of the page, so that it doesn't get | |
1042 | * lost when the SPTE is marked for access tracking. | |
1043 | */ | |
1044 | if (is_writable_pte(spte)) | |
1045 | kvm_set_pfn_dirty(spte_to_pfn(spte)); | |
1046 | ||
1047 | spte = mark_spte_for_access_track(spte); | |
1048 | mmu_spte_update_no_track(sptep, spte); | |
1049 | } | |
1050 | ||
1051 | return true; | |
1052 | } | |
1053 | ||
c2a2ac2b XG |
1054 | static void walk_shadow_page_lockless_begin(struct kvm_vcpu *vcpu) |
1055 | { | |
c142786c AK |
1056 | /* |
1057 | * Prevent page table teardown by making any free-er wait during | |
1058 | * kvm_flush_remote_tlbs() IPI to all active vcpus. | |
1059 | */ | |
1060 | local_irq_disable(); | |
36ca7e0a | 1061 | |
c142786c AK |
1062 | /* |
1063 | * Make sure a following spte read is not reordered ahead of the write | |
1064 | * to vcpu->mode. | |
1065 | */ | |
36ca7e0a | 1066 | smp_store_mb(vcpu->mode, READING_SHADOW_PAGE_TABLES); |
c2a2ac2b XG |
1067 | } |
1068 | ||
1069 | static void walk_shadow_page_lockless_end(struct kvm_vcpu *vcpu) | |
1070 | { | |
c142786c AK |
1071 | /* |
1072 | * Make sure the write to vcpu->mode is not reordered in front of | |
9a984586 | 1073 | * reads to sptes. If it does, kvm_mmu_commit_zap_page() can see us |
c142786c AK |
1074 | * OUTSIDE_GUEST_MODE and proceed to free the shadow page table. |
1075 | */ | |
36ca7e0a | 1076 | smp_store_release(&vcpu->mode, OUTSIDE_GUEST_MODE); |
c142786c | 1077 | local_irq_enable(); |
c2a2ac2b XG |
1078 | } |
1079 | ||
378f5cd6 | 1080 | static int mmu_topup_memory_caches(struct kvm_vcpu *vcpu, bool maybe_indirect) |
714b93da | 1081 | { |
e2dec939 AK |
1082 | int r; |
1083 | ||
531281ad | 1084 | /* 1 rmap, 1 parent PTE per level, and the prefetched rmaps. */ |
94ce87ef SC |
1085 | r = kvm_mmu_topup_memory_cache(&vcpu->arch.mmu_pte_list_desc_cache, |
1086 | 1 + PT64_ROOT_MAX_LEVEL + PTE_PREFETCH_NUM); | |
d3d25b04 | 1087 | if (r) |
284aa868 | 1088 | return r; |
94ce87ef SC |
1089 | r = kvm_mmu_topup_memory_cache(&vcpu->arch.mmu_shadow_page_cache, |
1090 | PT64_ROOT_MAX_LEVEL); | |
d3d25b04 | 1091 | if (r) |
171a90d7 | 1092 | return r; |
378f5cd6 | 1093 | if (maybe_indirect) { |
94ce87ef SC |
1094 | r = kvm_mmu_topup_memory_cache(&vcpu->arch.mmu_gfn_array_cache, |
1095 | PT64_ROOT_MAX_LEVEL); | |
378f5cd6 SC |
1096 | if (r) |
1097 | return r; | |
1098 | } | |
94ce87ef SC |
1099 | return kvm_mmu_topup_memory_cache(&vcpu->arch.mmu_page_header_cache, |
1100 | PT64_ROOT_MAX_LEVEL); | |
714b93da AK |
1101 | } |
1102 | ||
1103 | static void mmu_free_memory_caches(struct kvm_vcpu *vcpu) | |
1104 | { | |
94ce87ef SC |
1105 | kvm_mmu_free_memory_cache(&vcpu->arch.mmu_pte_list_desc_cache); |
1106 | kvm_mmu_free_memory_cache(&vcpu->arch.mmu_shadow_page_cache); | |
1107 | kvm_mmu_free_memory_cache(&vcpu->arch.mmu_gfn_array_cache); | |
1108 | kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_header_cache); | |
714b93da AK |
1109 | } |
1110 | ||
53c07b18 | 1111 | static struct pte_list_desc *mmu_alloc_pte_list_desc(struct kvm_vcpu *vcpu) |
714b93da | 1112 | { |
94ce87ef | 1113 | return kvm_mmu_memory_cache_alloc(&vcpu->arch.mmu_pte_list_desc_cache); |
714b93da AK |
1114 | } |
1115 | ||
53c07b18 | 1116 | static void mmu_free_pte_list_desc(struct pte_list_desc *pte_list_desc) |
714b93da | 1117 | { |
53c07b18 | 1118 | kmem_cache_free(pte_list_desc_cache, pte_list_desc); |
714b93da AK |
1119 | } |
1120 | ||
2032a93d LJ |
1121 | static gfn_t kvm_mmu_page_get_gfn(struct kvm_mmu_page *sp, int index) |
1122 | { | |
1123 | if (!sp->role.direct) | |
1124 | return sp->gfns[index]; | |
1125 | ||
1126 | return sp->gfn + (index << ((sp->role.level - 1) * PT64_LEVEL_BITS)); | |
1127 | } | |
1128 | ||
1129 | static void kvm_mmu_page_set_gfn(struct kvm_mmu_page *sp, int index, gfn_t gfn) | |
1130 | { | |
e9f2a760 | 1131 | if (!sp->role.direct) { |
2032a93d | 1132 | sp->gfns[index] = gfn; |
e9f2a760 PB |
1133 | return; |
1134 | } | |
1135 | ||
1136 | if (WARN_ON(gfn != kvm_mmu_page_get_gfn(sp, index))) | |
1137 | pr_err_ratelimited("gfn mismatch under direct page %llx " | |
1138 | "(expected %llx, got %llx)\n", | |
1139 | sp->gfn, | |
1140 | kvm_mmu_page_get_gfn(sp, index), gfn); | |
2032a93d LJ |
1141 | } |
1142 | ||
05da4558 | 1143 | /* |
d4dbf470 TY |
1144 | * Return the pointer to the large page information for a given gfn, |
1145 | * handling slots that are not large page aligned. | |
05da4558 | 1146 | */ |
d4dbf470 TY |
1147 | static struct kvm_lpage_info *lpage_info_slot(gfn_t gfn, |
1148 | struct kvm_memory_slot *slot, | |
1149 | int level) | |
05da4558 MT |
1150 | { |
1151 | unsigned long idx; | |
1152 | ||
fb03cb6f | 1153 | idx = gfn_to_index(gfn, slot->base_gfn, level); |
db3fe4eb | 1154 | return &slot->arch.lpage_info[level - 2][idx]; |
05da4558 MT |
1155 | } |
1156 | ||
547ffaed XG |
1157 | static void update_gfn_disallow_lpage_count(struct kvm_memory_slot *slot, |
1158 | gfn_t gfn, int count) | |
1159 | { | |
1160 | struct kvm_lpage_info *linfo; | |
1161 | int i; | |
1162 | ||
3bae0459 | 1163 | for (i = PG_LEVEL_2M; i <= KVM_MAX_HUGEPAGE_LEVEL; ++i) { |
547ffaed XG |
1164 | linfo = lpage_info_slot(gfn, slot, i); |
1165 | linfo->disallow_lpage += count; | |
1166 | WARN_ON(linfo->disallow_lpage < 0); | |
1167 | } | |
1168 | } | |
1169 | ||
1170 | void kvm_mmu_gfn_disallow_lpage(struct kvm_memory_slot *slot, gfn_t gfn) | |
1171 | { | |
1172 | update_gfn_disallow_lpage_count(slot, gfn, 1); | |
1173 | } | |
1174 | ||
1175 | void kvm_mmu_gfn_allow_lpage(struct kvm_memory_slot *slot, gfn_t gfn) | |
1176 | { | |
1177 | update_gfn_disallow_lpage_count(slot, gfn, -1); | |
1178 | } | |
1179 | ||
3ed1a478 | 1180 | static void account_shadowed(struct kvm *kvm, struct kvm_mmu_page *sp) |
05da4558 | 1181 | { |
699023e2 | 1182 | struct kvm_memslots *slots; |
d25797b2 | 1183 | struct kvm_memory_slot *slot; |
3ed1a478 | 1184 | gfn_t gfn; |
05da4558 | 1185 | |
56ca57f9 | 1186 | kvm->arch.indirect_shadow_pages++; |
3ed1a478 | 1187 | gfn = sp->gfn; |
699023e2 PB |
1188 | slots = kvm_memslots_for_spte_role(kvm, sp->role); |
1189 | slot = __gfn_to_memslot(slots, gfn); | |
56ca57f9 XG |
1190 | |
1191 | /* the non-leaf shadow pages are keeping readonly. */ | |
3bae0459 | 1192 | if (sp->role.level > PG_LEVEL_4K) |
56ca57f9 XG |
1193 | return kvm_slot_page_track_add_page(kvm, slot, gfn, |
1194 | KVM_PAGE_TRACK_WRITE); | |
1195 | ||
547ffaed | 1196 | kvm_mmu_gfn_disallow_lpage(slot, gfn); |
05da4558 MT |
1197 | } |
1198 | ||
b8e8c830 PB |
1199 | static void account_huge_nx_page(struct kvm *kvm, struct kvm_mmu_page *sp) |
1200 | { | |
1201 | if (sp->lpage_disallowed) | |
1202 | return; | |
1203 | ||
1204 | ++kvm->stat.nx_lpage_splits; | |
1aa9b957 JS |
1205 | list_add_tail(&sp->lpage_disallowed_link, |
1206 | &kvm->arch.lpage_disallowed_mmu_pages); | |
b8e8c830 PB |
1207 | sp->lpage_disallowed = true; |
1208 | } | |
1209 | ||
3ed1a478 | 1210 | static void unaccount_shadowed(struct kvm *kvm, struct kvm_mmu_page *sp) |
05da4558 | 1211 | { |
699023e2 | 1212 | struct kvm_memslots *slots; |
d25797b2 | 1213 | struct kvm_memory_slot *slot; |
3ed1a478 | 1214 | gfn_t gfn; |
05da4558 | 1215 | |
56ca57f9 | 1216 | kvm->arch.indirect_shadow_pages--; |
3ed1a478 | 1217 | gfn = sp->gfn; |
699023e2 PB |
1218 | slots = kvm_memslots_for_spte_role(kvm, sp->role); |
1219 | slot = __gfn_to_memslot(slots, gfn); | |
3bae0459 | 1220 | if (sp->role.level > PG_LEVEL_4K) |
56ca57f9 XG |
1221 | return kvm_slot_page_track_remove_page(kvm, slot, gfn, |
1222 | KVM_PAGE_TRACK_WRITE); | |
1223 | ||
547ffaed | 1224 | kvm_mmu_gfn_allow_lpage(slot, gfn); |
05da4558 MT |
1225 | } |
1226 | ||
b8e8c830 PB |
1227 | static void unaccount_huge_nx_page(struct kvm *kvm, struct kvm_mmu_page *sp) |
1228 | { | |
1229 | --kvm->stat.nx_lpage_splits; | |
1230 | sp->lpage_disallowed = false; | |
1aa9b957 | 1231 | list_del(&sp->lpage_disallowed_link); |
b8e8c830 PB |
1232 | } |
1233 | ||
5d163b1c XG |
1234 | static struct kvm_memory_slot * |
1235 | gfn_to_memslot_dirty_bitmap(struct kvm_vcpu *vcpu, gfn_t gfn, | |
1236 | bool no_dirty_log) | |
05da4558 MT |
1237 | { |
1238 | struct kvm_memory_slot *slot; | |
5d163b1c | 1239 | |
54bf36aa | 1240 | slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); |
91b0d268 PB |
1241 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
1242 | return NULL; | |
1243 | if (no_dirty_log && slot->dirty_bitmap) | |
1244 | return NULL; | |
5d163b1c XG |
1245 | |
1246 | return slot; | |
1247 | } | |
1248 | ||
290fc38d | 1249 | /* |
018aabb5 | 1250 | * About rmap_head encoding: |
cd4a4e53 | 1251 | * |
018aabb5 TY |
1252 | * If the bit zero of rmap_head->val is clear, then it points to the only spte |
1253 | * in this rmap chain. Otherwise, (rmap_head->val & ~1) points to a struct | |
53c07b18 | 1254 | * pte_list_desc containing more mappings. |
018aabb5 TY |
1255 | */ |
1256 | ||
1257 | /* | |
1258 | * Returns the number of pointers in the rmap chain, not counting the new one. | |
cd4a4e53 | 1259 | */ |
53c07b18 | 1260 | static int pte_list_add(struct kvm_vcpu *vcpu, u64 *spte, |
018aabb5 | 1261 | struct kvm_rmap_head *rmap_head) |
cd4a4e53 | 1262 | { |
53c07b18 | 1263 | struct pte_list_desc *desc; |
53a27b39 | 1264 | int i, count = 0; |
cd4a4e53 | 1265 | |
018aabb5 | 1266 | if (!rmap_head->val) { |
53c07b18 | 1267 | rmap_printk("pte_list_add: %p %llx 0->1\n", spte, *spte); |
018aabb5 TY |
1268 | rmap_head->val = (unsigned long)spte; |
1269 | } else if (!(rmap_head->val & 1)) { | |
53c07b18 XG |
1270 | rmap_printk("pte_list_add: %p %llx 1->many\n", spte, *spte); |
1271 | desc = mmu_alloc_pte_list_desc(vcpu); | |
018aabb5 | 1272 | desc->sptes[0] = (u64 *)rmap_head->val; |
d555c333 | 1273 | desc->sptes[1] = spte; |
018aabb5 | 1274 | rmap_head->val = (unsigned long)desc | 1; |
cb16a7b3 | 1275 | ++count; |
cd4a4e53 | 1276 | } else { |
53c07b18 | 1277 | rmap_printk("pte_list_add: %p %llx many->many\n", spte, *spte); |
018aabb5 | 1278 | desc = (struct pte_list_desc *)(rmap_head->val & ~1ul); |
53c07b18 | 1279 | while (desc->sptes[PTE_LIST_EXT-1] && desc->more) { |
cd4a4e53 | 1280 | desc = desc->more; |
53c07b18 | 1281 | count += PTE_LIST_EXT; |
53a27b39 | 1282 | } |
53c07b18 XG |
1283 | if (desc->sptes[PTE_LIST_EXT-1]) { |
1284 | desc->more = mmu_alloc_pte_list_desc(vcpu); | |
cd4a4e53 AK |
1285 | desc = desc->more; |
1286 | } | |
d555c333 | 1287 | for (i = 0; desc->sptes[i]; ++i) |
cb16a7b3 | 1288 | ++count; |
d555c333 | 1289 | desc->sptes[i] = spte; |
cd4a4e53 | 1290 | } |
53a27b39 | 1291 | return count; |
cd4a4e53 AK |
1292 | } |
1293 | ||
53c07b18 | 1294 | static void |
018aabb5 TY |
1295 | pte_list_desc_remove_entry(struct kvm_rmap_head *rmap_head, |
1296 | struct pte_list_desc *desc, int i, | |
1297 | struct pte_list_desc *prev_desc) | |
cd4a4e53 AK |
1298 | { |
1299 | int j; | |
1300 | ||
53c07b18 | 1301 | for (j = PTE_LIST_EXT - 1; !desc->sptes[j] && j > i; --j) |
cd4a4e53 | 1302 | ; |
d555c333 AK |
1303 | desc->sptes[i] = desc->sptes[j]; |
1304 | desc->sptes[j] = NULL; | |
cd4a4e53 AK |
1305 | if (j != 0) |
1306 | return; | |
1307 | if (!prev_desc && !desc->more) | |
fe3c2b4c | 1308 | rmap_head->val = 0; |
cd4a4e53 AK |
1309 | else |
1310 | if (prev_desc) | |
1311 | prev_desc->more = desc->more; | |
1312 | else | |
018aabb5 | 1313 | rmap_head->val = (unsigned long)desc->more | 1; |
53c07b18 | 1314 | mmu_free_pte_list_desc(desc); |
cd4a4e53 AK |
1315 | } |
1316 | ||
8daf3462 | 1317 | static void __pte_list_remove(u64 *spte, struct kvm_rmap_head *rmap_head) |
cd4a4e53 | 1318 | { |
53c07b18 XG |
1319 | struct pte_list_desc *desc; |
1320 | struct pte_list_desc *prev_desc; | |
cd4a4e53 AK |
1321 | int i; |
1322 | ||
018aabb5 | 1323 | if (!rmap_head->val) { |
8daf3462 | 1324 | pr_err("%s: %p 0->BUG\n", __func__, spte); |
cd4a4e53 | 1325 | BUG(); |
018aabb5 | 1326 | } else if (!(rmap_head->val & 1)) { |
8daf3462 | 1327 | rmap_printk("%s: %p 1->0\n", __func__, spte); |
018aabb5 | 1328 | if ((u64 *)rmap_head->val != spte) { |
8daf3462 | 1329 | pr_err("%s: %p 1->BUG\n", __func__, spte); |
cd4a4e53 AK |
1330 | BUG(); |
1331 | } | |
018aabb5 | 1332 | rmap_head->val = 0; |
cd4a4e53 | 1333 | } else { |
8daf3462 | 1334 | rmap_printk("%s: %p many->many\n", __func__, spte); |
018aabb5 | 1335 | desc = (struct pte_list_desc *)(rmap_head->val & ~1ul); |
cd4a4e53 AK |
1336 | prev_desc = NULL; |
1337 | while (desc) { | |
018aabb5 | 1338 | for (i = 0; i < PTE_LIST_EXT && desc->sptes[i]; ++i) { |
d555c333 | 1339 | if (desc->sptes[i] == spte) { |
018aabb5 TY |
1340 | pte_list_desc_remove_entry(rmap_head, |
1341 | desc, i, prev_desc); | |
cd4a4e53 AK |
1342 | return; |
1343 | } | |
018aabb5 | 1344 | } |
cd4a4e53 AK |
1345 | prev_desc = desc; |
1346 | desc = desc->more; | |
1347 | } | |
8daf3462 | 1348 | pr_err("%s: %p many->many\n", __func__, spte); |
cd4a4e53 AK |
1349 | BUG(); |
1350 | } | |
1351 | } | |
1352 | ||
e7912386 WY |
1353 | static void pte_list_remove(struct kvm_rmap_head *rmap_head, u64 *sptep) |
1354 | { | |
1355 | mmu_spte_clear_track_bits(sptep); | |
1356 | __pte_list_remove(sptep, rmap_head); | |
1357 | } | |
1358 | ||
018aabb5 TY |
1359 | static struct kvm_rmap_head *__gfn_to_rmap(gfn_t gfn, int level, |
1360 | struct kvm_memory_slot *slot) | |
53c07b18 | 1361 | { |
77d11309 | 1362 | unsigned long idx; |
53c07b18 | 1363 | |
77d11309 | 1364 | idx = gfn_to_index(gfn, slot->base_gfn, level); |
3bae0459 | 1365 | return &slot->arch.rmap[level - PG_LEVEL_4K][idx]; |
53c07b18 XG |
1366 | } |
1367 | ||
018aabb5 TY |
1368 | static struct kvm_rmap_head *gfn_to_rmap(struct kvm *kvm, gfn_t gfn, |
1369 | struct kvm_mmu_page *sp) | |
9b9b1492 | 1370 | { |
699023e2 | 1371 | struct kvm_memslots *slots; |
9b9b1492 TY |
1372 | struct kvm_memory_slot *slot; |
1373 | ||
699023e2 PB |
1374 | slots = kvm_memslots_for_spte_role(kvm, sp->role); |
1375 | slot = __gfn_to_memslot(slots, gfn); | |
e4cd1da9 | 1376 | return __gfn_to_rmap(gfn, sp->role.level, slot); |
9b9b1492 TY |
1377 | } |
1378 | ||
f759e2b4 XG |
1379 | static bool rmap_can_add(struct kvm_vcpu *vcpu) |
1380 | { | |
356ec69a | 1381 | struct kvm_mmu_memory_cache *mc; |
f759e2b4 | 1382 | |
356ec69a | 1383 | mc = &vcpu->arch.mmu_pte_list_desc_cache; |
94ce87ef | 1384 | return kvm_mmu_memory_cache_nr_free_objects(mc); |
f759e2b4 XG |
1385 | } |
1386 | ||
53c07b18 XG |
1387 | static int rmap_add(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn) |
1388 | { | |
1389 | struct kvm_mmu_page *sp; | |
018aabb5 | 1390 | struct kvm_rmap_head *rmap_head; |
53c07b18 | 1391 | |
57354682 | 1392 | sp = sptep_to_sp(spte); |
53c07b18 | 1393 | kvm_mmu_page_set_gfn(sp, spte - sp->spt, gfn); |
018aabb5 TY |
1394 | rmap_head = gfn_to_rmap(vcpu->kvm, gfn, sp); |
1395 | return pte_list_add(vcpu, spte, rmap_head); | |
53c07b18 XG |
1396 | } |
1397 | ||
53c07b18 XG |
1398 | static void rmap_remove(struct kvm *kvm, u64 *spte) |
1399 | { | |
1400 | struct kvm_mmu_page *sp; | |
1401 | gfn_t gfn; | |
018aabb5 | 1402 | struct kvm_rmap_head *rmap_head; |
53c07b18 | 1403 | |
57354682 | 1404 | sp = sptep_to_sp(spte); |
53c07b18 | 1405 | gfn = kvm_mmu_page_get_gfn(sp, spte - sp->spt); |
018aabb5 | 1406 | rmap_head = gfn_to_rmap(kvm, gfn, sp); |
8daf3462 | 1407 | __pte_list_remove(spte, rmap_head); |
53c07b18 XG |
1408 | } |
1409 | ||
1e3f42f0 TY |
1410 | /* |
1411 | * Used by the following functions to iterate through the sptes linked by a | |
1412 | * rmap. All fields are private and not assumed to be used outside. | |
1413 | */ | |
1414 | struct rmap_iterator { | |
1415 | /* private fields */ | |
1416 | struct pte_list_desc *desc; /* holds the sptep if not NULL */ | |
1417 | int pos; /* index of the sptep */ | |
1418 | }; | |
1419 | ||
1420 | /* | |
1421 | * Iteration must be started by this function. This should also be used after | |
1422 | * removing/dropping sptes from the rmap link because in such cases the | |
0a03cbda | 1423 | * information in the iterator may not be valid. |
1e3f42f0 TY |
1424 | * |
1425 | * Returns sptep if found, NULL otherwise. | |
1426 | */ | |
018aabb5 TY |
1427 | static u64 *rmap_get_first(struct kvm_rmap_head *rmap_head, |
1428 | struct rmap_iterator *iter) | |
1e3f42f0 | 1429 | { |
77fbbbd2 TY |
1430 | u64 *sptep; |
1431 | ||
018aabb5 | 1432 | if (!rmap_head->val) |
1e3f42f0 TY |
1433 | return NULL; |
1434 | ||
018aabb5 | 1435 | if (!(rmap_head->val & 1)) { |
1e3f42f0 | 1436 | iter->desc = NULL; |
77fbbbd2 TY |
1437 | sptep = (u64 *)rmap_head->val; |
1438 | goto out; | |
1e3f42f0 TY |
1439 | } |
1440 | ||
018aabb5 | 1441 | iter->desc = (struct pte_list_desc *)(rmap_head->val & ~1ul); |
1e3f42f0 | 1442 | iter->pos = 0; |
77fbbbd2 TY |
1443 | sptep = iter->desc->sptes[iter->pos]; |
1444 | out: | |
1445 | BUG_ON(!is_shadow_present_pte(*sptep)); | |
1446 | return sptep; | |
1e3f42f0 TY |
1447 | } |
1448 | ||
1449 | /* | |
1450 | * Must be used with a valid iterator: e.g. after rmap_get_first(). | |
1451 | * | |
1452 | * Returns sptep if found, NULL otherwise. | |
1453 | */ | |
1454 | static u64 *rmap_get_next(struct rmap_iterator *iter) | |
1455 | { | |
77fbbbd2 TY |
1456 | u64 *sptep; |
1457 | ||
1e3f42f0 TY |
1458 | if (iter->desc) { |
1459 | if (iter->pos < PTE_LIST_EXT - 1) { | |
1e3f42f0 TY |
1460 | ++iter->pos; |
1461 | sptep = iter->desc->sptes[iter->pos]; | |
1462 | if (sptep) | |
77fbbbd2 | 1463 | goto out; |
1e3f42f0 TY |
1464 | } |
1465 | ||
1466 | iter->desc = iter->desc->more; | |
1467 | ||
1468 | if (iter->desc) { | |
1469 | iter->pos = 0; | |
1470 | /* desc->sptes[0] cannot be NULL */ | |
77fbbbd2 TY |
1471 | sptep = iter->desc->sptes[iter->pos]; |
1472 | goto out; | |
1e3f42f0 TY |
1473 | } |
1474 | } | |
1475 | ||
1476 | return NULL; | |
77fbbbd2 TY |
1477 | out: |
1478 | BUG_ON(!is_shadow_present_pte(*sptep)); | |
1479 | return sptep; | |
1e3f42f0 TY |
1480 | } |
1481 | ||
018aabb5 TY |
1482 | #define for_each_rmap_spte(_rmap_head_, _iter_, _spte_) \ |
1483 | for (_spte_ = rmap_get_first(_rmap_head_, _iter_); \ | |
77fbbbd2 | 1484 | _spte_; _spte_ = rmap_get_next(_iter_)) |
0d536790 | 1485 | |
c3707958 | 1486 | static void drop_spte(struct kvm *kvm, u64 *sptep) |
e4b502ea | 1487 | { |
1df9f2dc | 1488 | if (mmu_spte_clear_track_bits(sptep)) |
eb45fda4 | 1489 | rmap_remove(kvm, sptep); |
be38d276 AK |
1490 | } |
1491 | ||
8e22f955 XG |
1492 | |
1493 | static bool __drop_large_spte(struct kvm *kvm, u64 *sptep) | |
1494 | { | |
1495 | if (is_large_pte(*sptep)) { | |
57354682 | 1496 | WARN_ON(sptep_to_sp(sptep)->role.level == PG_LEVEL_4K); |
8e22f955 XG |
1497 | drop_spte(kvm, sptep); |
1498 | --kvm->stat.lpages; | |
1499 | return true; | |
1500 | } | |
1501 | ||
1502 | return false; | |
1503 | } | |
1504 | ||
1505 | static void drop_large_spte(struct kvm_vcpu *vcpu, u64 *sptep) | |
1506 | { | |
c3134ce2 | 1507 | if (__drop_large_spte(vcpu->kvm, sptep)) { |
57354682 | 1508 | struct kvm_mmu_page *sp = sptep_to_sp(sptep); |
c3134ce2 LT |
1509 | |
1510 | kvm_flush_remote_tlbs_with_address(vcpu->kvm, sp->gfn, | |
1511 | KVM_PAGES_PER_HPAGE(sp->role.level)); | |
1512 | } | |
8e22f955 XG |
1513 | } |
1514 | ||
1515 | /* | |
49fde340 | 1516 | * Write-protect on the specified @sptep, @pt_protect indicates whether |
c126d94f | 1517 | * spte write-protection is caused by protecting shadow page table. |
49fde340 | 1518 | * |
b4619660 | 1519 | * Note: write protection is difference between dirty logging and spte |
49fde340 XG |
1520 | * protection: |
1521 | * - for dirty logging, the spte can be set to writable at anytime if | |
1522 | * its dirty bitmap is properly set. | |
1523 | * - for spte protection, the spte can be writable only after unsync-ing | |
1524 | * shadow page. | |
8e22f955 | 1525 | * |
c126d94f | 1526 | * Return true if tlb need be flushed. |
8e22f955 | 1527 | */ |
c4f138b4 | 1528 | static bool spte_write_protect(u64 *sptep, bool pt_protect) |
d13bc5b5 XG |
1529 | { |
1530 | u64 spte = *sptep; | |
1531 | ||
49fde340 | 1532 | if (!is_writable_pte(spte) && |
ea4114bc | 1533 | !(pt_protect && spte_can_locklessly_be_made_writable(spte))) |
d13bc5b5 XG |
1534 | return false; |
1535 | ||
1536 | rmap_printk("rmap_write_protect: spte %p %llx\n", sptep, *sptep); | |
1537 | ||
49fde340 XG |
1538 | if (pt_protect) |
1539 | spte &= ~SPTE_MMU_WRITEABLE; | |
d13bc5b5 | 1540 | spte = spte & ~PT_WRITABLE_MASK; |
49fde340 | 1541 | |
c126d94f | 1542 | return mmu_spte_update(sptep, spte); |
d13bc5b5 XG |
1543 | } |
1544 | ||
018aabb5 TY |
1545 | static bool __rmap_write_protect(struct kvm *kvm, |
1546 | struct kvm_rmap_head *rmap_head, | |
245c3912 | 1547 | bool pt_protect) |
98348e95 | 1548 | { |
1e3f42f0 TY |
1549 | u64 *sptep; |
1550 | struct rmap_iterator iter; | |
d13bc5b5 | 1551 | bool flush = false; |
374cbac0 | 1552 | |
018aabb5 | 1553 | for_each_rmap_spte(rmap_head, &iter, sptep) |
c4f138b4 | 1554 | flush |= spte_write_protect(sptep, pt_protect); |
855149aa | 1555 | |
d13bc5b5 | 1556 | return flush; |
a0ed4607 TY |
1557 | } |
1558 | ||
c4f138b4 | 1559 | static bool spte_clear_dirty(u64 *sptep) |
f4b4b180 KH |
1560 | { |
1561 | u64 spte = *sptep; | |
1562 | ||
1563 | rmap_printk("rmap_clear_dirty: spte %p %llx\n", sptep, *sptep); | |
1564 | ||
1f4e5fc8 | 1565 | MMU_WARN_ON(!spte_ad_enabled(spte)); |
f4b4b180 | 1566 | spte &= ~shadow_dirty_mask; |
f4b4b180 KH |
1567 | return mmu_spte_update(sptep, spte); |
1568 | } | |
1569 | ||
1f4e5fc8 | 1570 | static bool spte_wrprot_for_clear_dirty(u64 *sptep) |
ac8d57e5 PF |
1571 | { |
1572 | bool was_writable = test_and_clear_bit(PT_WRITABLE_SHIFT, | |
1573 | (unsigned long *)sptep); | |
1f4e5fc8 | 1574 | if (was_writable && !spte_ad_enabled(*sptep)) |
ac8d57e5 PF |
1575 | kvm_set_pfn_dirty(spte_to_pfn(*sptep)); |
1576 | ||
1577 | return was_writable; | |
1578 | } | |
1579 | ||
1580 | /* | |
1581 | * Gets the GFN ready for another round of dirty logging by clearing the | |
1582 | * - D bit on ad-enabled SPTEs, and | |
1583 | * - W bit on ad-disabled SPTEs. | |
1584 | * Returns true iff any D or W bits were cleared. | |
1585 | */ | |
018aabb5 | 1586 | static bool __rmap_clear_dirty(struct kvm *kvm, struct kvm_rmap_head *rmap_head) |
f4b4b180 KH |
1587 | { |
1588 | u64 *sptep; | |
1589 | struct rmap_iterator iter; | |
1590 | bool flush = false; | |
1591 | ||
018aabb5 | 1592 | for_each_rmap_spte(rmap_head, &iter, sptep) |
1f4e5fc8 PB |
1593 | if (spte_ad_need_write_protect(*sptep)) |
1594 | flush |= spte_wrprot_for_clear_dirty(sptep); | |
ac8d57e5 | 1595 | else |
1f4e5fc8 | 1596 | flush |= spte_clear_dirty(sptep); |
f4b4b180 KH |
1597 | |
1598 | return flush; | |
1599 | } | |
1600 | ||
c4f138b4 | 1601 | static bool spte_set_dirty(u64 *sptep) |
f4b4b180 KH |
1602 | { |
1603 | u64 spte = *sptep; | |
1604 | ||
1605 | rmap_printk("rmap_set_dirty: spte %p %llx\n", sptep, *sptep); | |
1606 | ||
1f4e5fc8 | 1607 | /* |
afaf0b2f | 1608 | * Similar to the !kvm_x86_ops.slot_disable_log_dirty case, |
1f4e5fc8 PB |
1609 | * do not bother adding back write access to pages marked |
1610 | * SPTE_AD_WRPROT_ONLY_MASK. | |
1611 | */ | |
f4b4b180 KH |
1612 | spte |= shadow_dirty_mask; |
1613 | ||
1614 | return mmu_spte_update(sptep, spte); | |
1615 | } | |
1616 | ||
018aabb5 | 1617 | static bool __rmap_set_dirty(struct kvm *kvm, struct kvm_rmap_head *rmap_head) |
f4b4b180 KH |
1618 | { |
1619 | u64 *sptep; | |
1620 | struct rmap_iterator iter; | |
1621 | bool flush = false; | |
1622 | ||
018aabb5 | 1623 | for_each_rmap_spte(rmap_head, &iter, sptep) |
ac8d57e5 PF |
1624 | if (spte_ad_enabled(*sptep)) |
1625 | flush |= spte_set_dirty(sptep); | |
f4b4b180 KH |
1626 | |
1627 | return flush; | |
1628 | } | |
1629 | ||
5dc99b23 | 1630 | /** |
3b0f1d01 | 1631 | * kvm_mmu_write_protect_pt_masked - write protect selected PT level pages |
5dc99b23 TY |
1632 | * @kvm: kvm instance |
1633 | * @slot: slot to protect | |
1634 | * @gfn_offset: start of the BITS_PER_LONG pages we care about | |
1635 | * @mask: indicates which pages we should protect | |
1636 | * | |
1637 | * Used when we do not need to care about huge page mappings: e.g. during dirty | |
1638 | * logging we do not have any such mappings. | |
1639 | */ | |
3b0f1d01 | 1640 | static void kvm_mmu_write_protect_pt_masked(struct kvm *kvm, |
5dc99b23 TY |
1641 | struct kvm_memory_slot *slot, |
1642 | gfn_t gfn_offset, unsigned long mask) | |
a0ed4607 | 1643 | { |
018aabb5 | 1644 | struct kvm_rmap_head *rmap_head; |
a0ed4607 | 1645 | |
5dc99b23 | 1646 | while (mask) { |
018aabb5 | 1647 | rmap_head = __gfn_to_rmap(slot->base_gfn + gfn_offset + __ffs(mask), |
3bae0459 | 1648 | PG_LEVEL_4K, slot); |
018aabb5 | 1649 | __rmap_write_protect(kvm, rmap_head, false); |
05da4558 | 1650 | |
5dc99b23 TY |
1651 | /* clear the first set bit */ |
1652 | mask &= mask - 1; | |
1653 | } | |
374cbac0 AK |
1654 | } |
1655 | ||
f4b4b180 | 1656 | /** |
ac8d57e5 PF |
1657 | * kvm_mmu_clear_dirty_pt_masked - clear MMU D-bit for PT level pages, or write |
1658 | * protect the page if the D-bit isn't supported. | |
f4b4b180 KH |
1659 | * @kvm: kvm instance |
1660 | * @slot: slot to clear D-bit | |
1661 | * @gfn_offset: start of the BITS_PER_LONG pages we care about | |
1662 | * @mask: indicates which pages we should clear D-bit | |
1663 | * | |
1664 | * Used for PML to re-log the dirty GPAs after userspace querying dirty_bitmap. | |
1665 | */ | |
1666 | void kvm_mmu_clear_dirty_pt_masked(struct kvm *kvm, | |
1667 | struct kvm_memory_slot *slot, | |
1668 | gfn_t gfn_offset, unsigned long mask) | |
1669 | { | |
018aabb5 | 1670 | struct kvm_rmap_head *rmap_head; |
f4b4b180 KH |
1671 | |
1672 | while (mask) { | |
018aabb5 | 1673 | rmap_head = __gfn_to_rmap(slot->base_gfn + gfn_offset + __ffs(mask), |
3bae0459 | 1674 | PG_LEVEL_4K, slot); |
018aabb5 | 1675 | __rmap_clear_dirty(kvm, rmap_head); |
f4b4b180 KH |
1676 | |
1677 | /* clear the first set bit */ | |
1678 | mask &= mask - 1; | |
1679 | } | |
1680 | } | |
1681 | EXPORT_SYMBOL_GPL(kvm_mmu_clear_dirty_pt_masked); | |
1682 | ||
3b0f1d01 KH |
1683 | /** |
1684 | * kvm_arch_mmu_enable_log_dirty_pt_masked - enable dirty logging for selected | |
1685 | * PT level pages. | |
1686 | * | |
1687 | * It calls kvm_mmu_write_protect_pt_masked to write protect selected pages to | |
1688 | * enable dirty logging for them. | |
1689 | * | |
1690 | * Used when we do not need to care about huge page mappings: e.g. during dirty | |
1691 | * logging we do not have any such mappings. | |
1692 | */ | |
1693 | void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm, | |
1694 | struct kvm_memory_slot *slot, | |
1695 | gfn_t gfn_offset, unsigned long mask) | |
1696 | { | |
afaf0b2f SC |
1697 | if (kvm_x86_ops.enable_log_dirty_pt_masked) |
1698 | kvm_x86_ops.enable_log_dirty_pt_masked(kvm, slot, gfn_offset, | |
88178fd4 KH |
1699 | mask); |
1700 | else | |
1701 | kvm_mmu_write_protect_pt_masked(kvm, slot, gfn_offset, mask); | |
3b0f1d01 KH |
1702 | } |
1703 | ||
aeecee2e XG |
1704 | bool kvm_mmu_slot_gfn_write_protect(struct kvm *kvm, |
1705 | struct kvm_memory_slot *slot, u64 gfn) | |
95d4c16c | 1706 | { |
018aabb5 | 1707 | struct kvm_rmap_head *rmap_head; |
5dc99b23 | 1708 | int i; |
2f84569f | 1709 | bool write_protected = false; |
95d4c16c | 1710 | |
3bae0459 | 1711 | for (i = PG_LEVEL_4K; i <= KVM_MAX_HUGEPAGE_LEVEL; ++i) { |
018aabb5 | 1712 | rmap_head = __gfn_to_rmap(gfn, i, slot); |
aeecee2e | 1713 | write_protected |= __rmap_write_protect(kvm, rmap_head, true); |
5dc99b23 TY |
1714 | } |
1715 | ||
1716 | return write_protected; | |
95d4c16c TY |
1717 | } |
1718 | ||
aeecee2e XG |
1719 | static bool rmap_write_protect(struct kvm_vcpu *vcpu, u64 gfn) |
1720 | { | |
1721 | struct kvm_memory_slot *slot; | |
1722 | ||
1723 | slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); | |
1724 | return kvm_mmu_slot_gfn_write_protect(vcpu->kvm, slot, gfn); | |
1725 | } | |
1726 | ||
018aabb5 | 1727 | static bool kvm_zap_rmapp(struct kvm *kvm, struct kvm_rmap_head *rmap_head) |
e930bffe | 1728 | { |
1e3f42f0 TY |
1729 | u64 *sptep; |
1730 | struct rmap_iterator iter; | |
6a49f85c | 1731 | bool flush = false; |
e930bffe | 1732 | |
018aabb5 | 1733 | while ((sptep = rmap_get_first(rmap_head, &iter))) { |
6a49f85c | 1734 | rmap_printk("%s: spte %p %llx.\n", __func__, sptep, *sptep); |
1e3f42f0 | 1735 | |
e7912386 | 1736 | pte_list_remove(rmap_head, sptep); |
6a49f85c | 1737 | flush = true; |
e930bffe | 1738 | } |
1e3f42f0 | 1739 | |
6a49f85c XG |
1740 | return flush; |
1741 | } | |
1742 | ||
018aabb5 | 1743 | static int kvm_unmap_rmapp(struct kvm *kvm, struct kvm_rmap_head *rmap_head, |
6a49f85c XG |
1744 | struct kvm_memory_slot *slot, gfn_t gfn, int level, |
1745 | unsigned long data) | |
1746 | { | |
018aabb5 | 1747 | return kvm_zap_rmapp(kvm, rmap_head); |
e930bffe AA |
1748 | } |
1749 | ||
cb3eedab PB |
1750 | static u64 kvm_mmu_changed_pte_notifier_make_spte(u64 old_spte, kvm_pfn_t new_pfn) |
1751 | { | |
1752 | u64 new_spte; | |
1753 | ||
1754 | new_spte = old_spte & ~PT64_BASE_ADDR_MASK; | |
1755 | new_spte |= (u64)new_pfn << PAGE_SHIFT; | |
1756 | ||
1757 | new_spte &= ~PT_WRITABLE_MASK; | |
1758 | new_spte &= ~SPTE_HOST_WRITEABLE; | |
1759 | ||
1760 | new_spte = mark_spte_for_access_track(new_spte); | |
1761 | ||
1762 | return new_spte; | |
1763 | } | |
1764 | ||
018aabb5 | 1765 | static int kvm_set_pte_rmapp(struct kvm *kvm, struct kvm_rmap_head *rmap_head, |
8a9522d2 ALC |
1766 | struct kvm_memory_slot *slot, gfn_t gfn, int level, |
1767 | unsigned long data) | |
3da0dd43 | 1768 | { |
1e3f42f0 TY |
1769 | u64 *sptep; |
1770 | struct rmap_iterator iter; | |
3da0dd43 | 1771 | int need_flush = 0; |
1e3f42f0 | 1772 | u64 new_spte; |
3da0dd43 | 1773 | pte_t *ptep = (pte_t *)data; |
ba049e93 | 1774 | kvm_pfn_t new_pfn; |
3da0dd43 IE |
1775 | |
1776 | WARN_ON(pte_huge(*ptep)); | |
1777 | new_pfn = pte_pfn(*ptep); | |
1e3f42f0 | 1778 | |
0d536790 | 1779 | restart: |
018aabb5 | 1780 | for_each_rmap_spte(rmap_head, &iter, sptep) { |
8a9522d2 | 1781 | rmap_printk("kvm_set_pte_rmapp: spte %p %llx gfn %llx (%d)\n", |
f160c7b7 | 1782 | sptep, *sptep, gfn, level); |
1e3f42f0 | 1783 | |
3da0dd43 | 1784 | need_flush = 1; |
1e3f42f0 | 1785 | |
3da0dd43 | 1786 | if (pte_write(*ptep)) { |
e7912386 | 1787 | pte_list_remove(rmap_head, sptep); |
0d536790 | 1788 | goto restart; |
3da0dd43 | 1789 | } else { |
cb3eedab PB |
1790 | new_spte = kvm_mmu_changed_pte_notifier_make_spte( |
1791 | *sptep, new_pfn); | |
1e3f42f0 TY |
1792 | |
1793 | mmu_spte_clear_track_bits(sptep); | |
1794 | mmu_spte_set(sptep, new_spte); | |
3da0dd43 IE |
1795 | } |
1796 | } | |
1e3f42f0 | 1797 | |
3cc5ea94 LT |
1798 | if (need_flush && kvm_available_flush_tlb_with_range()) { |
1799 | kvm_flush_remote_tlbs_with_address(kvm, gfn, 1); | |
1800 | return 0; | |
1801 | } | |
1802 | ||
0cf853c5 | 1803 | return need_flush; |
3da0dd43 IE |
1804 | } |
1805 | ||
6ce1f4e2 XG |
1806 | struct slot_rmap_walk_iterator { |
1807 | /* input fields. */ | |
1808 | struct kvm_memory_slot *slot; | |
1809 | gfn_t start_gfn; | |
1810 | gfn_t end_gfn; | |
1811 | int start_level; | |
1812 | int end_level; | |
1813 | ||
1814 | /* output fields. */ | |
1815 | gfn_t gfn; | |
018aabb5 | 1816 | struct kvm_rmap_head *rmap; |
6ce1f4e2 XG |
1817 | int level; |
1818 | ||
1819 | /* private field. */ | |
018aabb5 | 1820 | struct kvm_rmap_head *end_rmap; |
6ce1f4e2 XG |
1821 | }; |
1822 | ||
1823 | static void | |
1824 | rmap_walk_init_level(struct slot_rmap_walk_iterator *iterator, int level) | |
1825 | { | |
1826 | iterator->level = level; | |
1827 | iterator->gfn = iterator->start_gfn; | |
1828 | iterator->rmap = __gfn_to_rmap(iterator->gfn, level, iterator->slot); | |
1829 | iterator->end_rmap = __gfn_to_rmap(iterator->end_gfn, level, | |
1830 | iterator->slot); | |
1831 | } | |
1832 | ||
1833 | static void | |
1834 | slot_rmap_walk_init(struct slot_rmap_walk_iterator *iterator, | |
1835 | struct kvm_memory_slot *slot, int start_level, | |
1836 | int end_level, gfn_t start_gfn, gfn_t end_gfn) | |
1837 | { | |
1838 | iterator->slot = slot; | |
1839 | iterator->start_level = start_level; | |
1840 | iterator->end_level = end_level; | |
1841 | iterator->start_gfn = start_gfn; | |
1842 | iterator->end_gfn = end_gfn; | |
1843 | ||
1844 | rmap_walk_init_level(iterator, iterator->start_level); | |
1845 | } | |
1846 | ||
1847 | static bool slot_rmap_walk_okay(struct slot_rmap_walk_iterator *iterator) | |
1848 | { | |
1849 | return !!iterator->rmap; | |
1850 | } | |
1851 | ||
1852 | static void slot_rmap_walk_next(struct slot_rmap_walk_iterator *iterator) | |
1853 | { | |
1854 | if (++iterator->rmap <= iterator->end_rmap) { | |
1855 | iterator->gfn += (1UL << KVM_HPAGE_GFN_SHIFT(iterator->level)); | |
1856 | return; | |
1857 | } | |
1858 | ||
1859 | if (++iterator->level > iterator->end_level) { | |
1860 | iterator->rmap = NULL; | |
1861 | return; | |
1862 | } | |
1863 | ||
1864 | rmap_walk_init_level(iterator, iterator->level); | |
1865 | } | |
1866 | ||
1867 | #define for_each_slot_rmap_range(_slot_, _start_level_, _end_level_, \ | |
1868 | _start_gfn, _end_gfn, _iter_) \ | |
1869 | for (slot_rmap_walk_init(_iter_, _slot_, _start_level_, \ | |
1870 | _end_level_, _start_gfn, _end_gfn); \ | |
1871 | slot_rmap_walk_okay(_iter_); \ | |
1872 | slot_rmap_walk_next(_iter_)) | |
1873 | ||
84504ef3 TY |
1874 | static int kvm_handle_hva_range(struct kvm *kvm, |
1875 | unsigned long start, | |
1876 | unsigned long end, | |
1877 | unsigned long data, | |
1878 | int (*handler)(struct kvm *kvm, | |
018aabb5 | 1879 | struct kvm_rmap_head *rmap_head, |
048212d0 | 1880 | struct kvm_memory_slot *slot, |
8a9522d2 ALC |
1881 | gfn_t gfn, |
1882 | int level, | |
84504ef3 | 1883 | unsigned long data)) |
e930bffe | 1884 | { |
bc6678a3 | 1885 | struct kvm_memslots *slots; |
be6ba0f0 | 1886 | struct kvm_memory_slot *memslot; |
6ce1f4e2 XG |
1887 | struct slot_rmap_walk_iterator iterator; |
1888 | int ret = 0; | |
9da0e4d5 | 1889 | int i; |
bc6678a3 | 1890 | |
9da0e4d5 PB |
1891 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
1892 | slots = __kvm_memslots(kvm, i); | |
1893 | kvm_for_each_memslot(memslot, slots) { | |
1894 | unsigned long hva_start, hva_end; | |
1895 | gfn_t gfn_start, gfn_end; | |
e930bffe | 1896 | |
9da0e4d5 PB |
1897 | hva_start = max(start, memslot->userspace_addr); |
1898 | hva_end = min(end, memslot->userspace_addr + | |
1899 | (memslot->npages << PAGE_SHIFT)); | |
1900 | if (hva_start >= hva_end) | |
1901 | continue; | |
1902 | /* | |
1903 | * {gfn(page) | page intersects with [hva_start, hva_end)} = | |
1904 | * {gfn_start, gfn_start+1, ..., gfn_end-1}. | |
1905 | */ | |
1906 | gfn_start = hva_to_gfn_memslot(hva_start, memslot); | |
1907 | gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot); | |
1908 | ||
3bae0459 | 1909 | for_each_slot_rmap_range(memslot, PG_LEVEL_4K, |
e662ec3e | 1910 | KVM_MAX_HUGEPAGE_LEVEL, |
9da0e4d5 PB |
1911 | gfn_start, gfn_end - 1, |
1912 | &iterator) | |
1913 | ret |= handler(kvm, iterator.rmap, memslot, | |
1914 | iterator.gfn, iterator.level, data); | |
1915 | } | |
e930bffe AA |
1916 | } |
1917 | ||
f395302e | 1918 | return ret; |
e930bffe AA |
1919 | } |
1920 | ||
84504ef3 TY |
1921 | static int kvm_handle_hva(struct kvm *kvm, unsigned long hva, |
1922 | unsigned long data, | |
018aabb5 TY |
1923 | int (*handler)(struct kvm *kvm, |
1924 | struct kvm_rmap_head *rmap_head, | |
048212d0 | 1925 | struct kvm_memory_slot *slot, |
8a9522d2 | 1926 | gfn_t gfn, int level, |
84504ef3 TY |
1927 | unsigned long data)) |
1928 | { | |
1929 | return kvm_handle_hva_range(kvm, hva, hva + 1, data, handler); | |
e930bffe AA |
1930 | } |
1931 | ||
fdfe7cbd WD |
1932 | int kvm_unmap_hva_range(struct kvm *kvm, unsigned long start, unsigned long end, |
1933 | unsigned flags) | |
b3ae2096 TY |
1934 | { |
1935 | return kvm_handle_hva_range(kvm, start, end, 0, kvm_unmap_rmapp); | |
1936 | } | |
1937 | ||
748c0e31 | 1938 | int kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte) |
3da0dd43 | 1939 | { |
0cf853c5 | 1940 | return kvm_handle_hva(kvm, hva, (unsigned long)&pte, kvm_set_pte_rmapp); |
e930bffe AA |
1941 | } |
1942 | ||
018aabb5 | 1943 | static int kvm_age_rmapp(struct kvm *kvm, struct kvm_rmap_head *rmap_head, |
8a9522d2 ALC |
1944 | struct kvm_memory_slot *slot, gfn_t gfn, int level, |
1945 | unsigned long data) | |
e930bffe | 1946 | { |
1e3f42f0 | 1947 | u64 *sptep; |
3f649ab7 | 1948 | struct rmap_iterator iter; |
e930bffe AA |
1949 | int young = 0; |
1950 | ||
f160c7b7 JS |
1951 | for_each_rmap_spte(rmap_head, &iter, sptep) |
1952 | young |= mmu_spte_age(sptep); | |
0d536790 | 1953 | |
8a9522d2 | 1954 | trace_kvm_age_page(gfn, level, slot, young); |
e930bffe AA |
1955 | return young; |
1956 | } | |
1957 | ||
018aabb5 | 1958 | static int kvm_test_age_rmapp(struct kvm *kvm, struct kvm_rmap_head *rmap_head, |
8a9522d2 ALC |
1959 | struct kvm_memory_slot *slot, gfn_t gfn, |
1960 | int level, unsigned long data) | |
8ee53820 | 1961 | { |
1e3f42f0 TY |
1962 | u64 *sptep; |
1963 | struct rmap_iterator iter; | |
8ee53820 | 1964 | |
83ef6c81 JS |
1965 | for_each_rmap_spte(rmap_head, &iter, sptep) |
1966 | if (is_accessed_spte(*sptep)) | |
1967 | return 1; | |
83ef6c81 | 1968 | return 0; |
8ee53820 AA |
1969 | } |
1970 | ||
53a27b39 MT |
1971 | #define RMAP_RECYCLE_THRESHOLD 1000 |
1972 | ||
852e3c19 | 1973 | static void rmap_recycle(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn) |
53a27b39 | 1974 | { |
018aabb5 | 1975 | struct kvm_rmap_head *rmap_head; |
852e3c19 JR |
1976 | struct kvm_mmu_page *sp; |
1977 | ||
57354682 | 1978 | sp = sptep_to_sp(spte); |
53a27b39 | 1979 | |
018aabb5 | 1980 | rmap_head = gfn_to_rmap(vcpu->kvm, gfn, sp); |
53a27b39 | 1981 | |
018aabb5 | 1982 | kvm_unmap_rmapp(vcpu->kvm, rmap_head, NULL, gfn, sp->role.level, 0); |
c3134ce2 LT |
1983 | kvm_flush_remote_tlbs_with_address(vcpu->kvm, sp->gfn, |
1984 | KVM_PAGES_PER_HPAGE(sp->role.level)); | |
53a27b39 MT |
1985 | } |
1986 | ||
57128468 | 1987 | int kvm_age_hva(struct kvm *kvm, unsigned long start, unsigned long end) |
e930bffe | 1988 | { |
57128468 | 1989 | return kvm_handle_hva_range(kvm, start, end, 0, kvm_age_rmapp); |
e930bffe AA |
1990 | } |
1991 | ||
8ee53820 AA |
1992 | int kvm_test_age_hva(struct kvm *kvm, unsigned long hva) |
1993 | { | |
1994 | return kvm_handle_hva(kvm, hva, 0, kvm_test_age_rmapp); | |
1995 | } | |
1996 | ||
d6c69ee9 | 1997 | #ifdef MMU_DEBUG |
47ad8e68 | 1998 | static int is_empty_shadow_page(u64 *spt) |
6aa8b732 | 1999 | { |
139bdb2d AK |
2000 | u64 *pos; |
2001 | u64 *end; | |
2002 | ||
47ad8e68 | 2003 | for (pos = spt, end = pos + PAGE_SIZE / sizeof(u64); pos != end; pos++) |
3c915510 | 2004 | if (is_shadow_present_pte(*pos)) { |
b8688d51 | 2005 | printk(KERN_ERR "%s: %p %llx\n", __func__, |
139bdb2d | 2006 | pos, *pos); |
6aa8b732 | 2007 | return 0; |
139bdb2d | 2008 | } |
6aa8b732 AK |
2009 | return 1; |
2010 | } | |
d6c69ee9 | 2011 | #endif |
6aa8b732 | 2012 | |
45221ab6 DH |
2013 | /* |
2014 | * This value is the sum of all of the kvm instances's | |
2015 | * kvm->arch.n_used_mmu_pages values. We need a global, | |
2016 | * aggregate version in order to make the slab shrinker | |
2017 | * faster | |
2018 | */ | |
bc8a3d89 | 2019 | static inline void kvm_mod_used_mmu_pages(struct kvm *kvm, unsigned long nr) |
45221ab6 DH |
2020 | { |
2021 | kvm->arch.n_used_mmu_pages += nr; | |
2022 | percpu_counter_add(&kvm_total_used_mmu_pages, nr); | |
2023 | } | |
2024 | ||
834be0d8 | 2025 | static void kvm_mmu_free_page(struct kvm_mmu_page *sp) |
260746c0 | 2026 | { |
fa4a2c08 | 2027 | MMU_WARN_ON(!is_empty_shadow_page(sp->spt)); |
7775834a | 2028 | hlist_del(&sp->hash_link); |
bd4c86ea XG |
2029 | list_del(&sp->link); |
2030 | free_page((unsigned long)sp->spt); | |
834be0d8 GN |
2031 | if (!sp->role.direct) |
2032 | free_page((unsigned long)sp->gfns); | |
e8ad9a70 | 2033 | kmem_cache_free(mmu_page_header_cache, sp); |
260746c0 AK |
2034 | } |
2035 | ||
cea0f0e7 AK |
2036 | static unsigned kvm_page_table_hashfn(gfn_t gfn) |
2037 | { | |
114df303 | 2038 | return hash_64(gfn, KVM_MMU_HASH_SHIFT); |
cea0f0e7 AK |
2039 | } |
2040 | ||
714b93da | 2041 | static void mmu_page_add_parent_pte(struct kvm_vcpu *vcpu, |
4db35314 | 2042 | struct kvm_mmu_page *sp, u64 *parent_pte) |
cea0f0e7 | 2043 | { |
cea0f0e7 AK |
2044 | if (!parent_pte) |
2045 | return; | |
cea0f0e7 | 2046 | |
67052b35 | 2047 | pte_list_add(vcpu, parent_pte, &sp->parent_ptes); |
cea0f0e7 AK |
2048 | } |
2049 | ||
4db35314 | 2050 | static void mmu_page_remove_parent_pte(struct kvm_mmu_page *sp, |
cea0f0e7 AK |
2051 | u64 *parent_pte) |
2052 | { | |
8daf3462 | 2053 | __pte_list_remove(parent_pte, &sp->parent_ptes); |
cea0f0e7 AK |
2054 | } |
2055 | ||
bcdd9a93 XG |
2056 | static void drop_parent_pte(struct kvm_mmu_page *sp, |
2057 | u64 *parent_pte) | |
2058 | { | |
2059 | mmu_page_remove_parent_pte(sp, parent_pte); | |
1df9f2dc | 2060 | mmu_spte_clear_no_track(parent_pte); |
bcdd9a93 XG |
2061 | } |
2062 | ||
47005792 | 2063 | static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu, int direct) |
ad8cfbe3 | 2064 | { |
67052b35 | 2065 | struct kvm_mmu_page *sp; |
7ddca7e4 | 2066 | |
94ce87ef SC |
2067 | sp = kvm_mmu_memory_cache_alloc(&vcpu->arch.mmu_page_header_cache); |
2068 | sp->spt = kvm_mmu_memory_cache_alloc(&vcpu->arch.mmu_shadow_page_cache); | |
67052b35 | 2069 | if (!direct) |
94ce87ef | 2070 | sp->gfns = kvm_mmu_memory_cache_alloc(&vcpu->arch.mmu_gfn_array_cache); |
67052b35 | 2071 | set_page_private(virt_to_page(sp->spt), (unsigned long)sp); |
002c5f73 SC |
2072 | |
2073 | /* | |
2074 | * active_mmu_pages must be a FIFO list, as kvm_zap_obsolete_pages() | |
2075 | * depends on valid pages being added to the head of the list. See | |
2076 | * comments in kvm_zap_obsolete_pages(). | |
2077 | */ | |
ca333add | 2078 | sp->mmu_valid_gen = vcpu->kvm->arch.mmu_valid_gen; |
67052b35 | 2079 | list_add(&sp->link, &vcpu->kvm->arch.active_mmu_pages); |
67052b35 XG |
2080 | kvm_mod_used_mmu_pages(vcpu->kvm, +1); |
2081 | return sp; | |
ad8cfbe3 MT |
2082 | } |
2083 | ||
67052b35 | 2084 | static void mark_unsync(u64 *spte); |
1047df1f | 2085 | static void kvm_mmu_mark_parents_unsync(struct kvm_mmu_page *sp) |
0074ff63 | 2086 | { |
74c4e63a TY |
2087 | u64 *sptep; |
2088 | struct rmap_iterator iter; | |
2089 | ||
2090 | for_each_rmap_spte(&sp->parent_ptes, &iter, sptep) { | |
2091 | mark_unsync(sptep); | |
2092 | } | |
0074ff63 MT |
2093 | } |
2094 | ||
67052b35 | 2095 | static void mark_unsync(u64 *spte) |
0074ff63 | 2096 | { |
67052b35 | 2097 | struct kvm_mmu_page *sp; |
1047df1f | 2098 | unsigned int index; |
0074ff63 | 2099 | |
57354682 | 2100 | sp = sptep_to_sp(spte); |
1047df1f XG |
2101 | index = spte - sp->spt; |
2102 | if (__test_and_set_bit(index, sp->unsync_child_bitmap)) | |
0074ff63 | 2103 | return; |
1047df1f | 2104 | if (sp->unsync_children++) |
0074ff63 | 2105 | return; |
1047df1f | 2106 | kvm_mmu_mark_parents_unsync(sp); |
0074ff63 MT |
2107 | } |
2108 | ||
e8bc217a | 2109 | static int nonpaging_sync_page(struct kvm_vcpu *vcpu, |
a4a8e6f7 | 2110 | struct kvm_mmu_page *sp) |
e8bc217a | 2111 | { |
1f50f1b3 | 2112 | return 0; |
e8bc217a MT |
2113 | } |
2114 | ||
0f53b5b1 XG |
2115 | static void nonpaging_update_pte(struct kvm_vcpu *vcpu, |
2116 | struct kvm_mmu_page *sp, u64 *spte, | |
7c562522 | 2117 | const void *pte) |
0f53b5b1 XG |
2118 | { |
2119 | WARN_ON(1); | |
2120 | } | |
2121 | ||
60c8aec6 MT |
2122 | #define KVM_PAGE_ARRAY_NR 16 |
2123 | ||
2124 | struct kvm_mmu_pages { | |
2125 | struct mmu_page_and_offset { | |
2126 | struct kvm_mmu_page *sp; | |
2127 | unsigned int idx; | |
2128 | } page[KVM_PAGE_ARRAY_NR]; | |
2129 | unsigned int nr; | |
2130 | }; | |
2131 | ||
cded19f3 HE |
2132 | static int mmu_pages_add(struct kvm_mmu_pages *pvec, struct kvm_mmu_page *sp, |
2133 | int idx) | |
4731d4c7 | 2134 | { |
60c8aec6 | 2135 | int i; |
4731d4c7 | 2136 | |
60c8aec6 MT |
2137 | if (sp->unsync) |
2138 | for (i=0; i < pvec->nr; i++) | |
2139 | if (pvec->page[i].sp == sp) | |
2140 | return 0; | |
2141 | ||
2142 | pvec->page[pvec->nr].sp = sp; | |
2143 | pvec->page[pvec->nr].idx = idx; | |
2144 | pvec->nr++; | |
2145 | return (pvec->nr == KVM_PAGE_ARRAY_NR); | |
2146 | } | |
2147 | ||
fd951457 TY |
2148 | static inline void clear_unsync_child_bit(struct kvm_mmu_page *sp, int idx) |
2149 | { | |
2150 | --sp->unsync_children; | |
2151 | WARN_ON((int)sp->unsync_children < 0); | |
2152 | __clear_bit(idx, sp->unsync_child_bitmap); | |
2153 | } | |
2154 | ||
60c8aec6 MT |
2155 | static int __mmu_unsync_walk(struct kvm_mmu_page *sp, |
2156 | struct kvm_mmu_pages *pvec) | |
2157 | { | |
2158 | int i, ret, nr_unsync_leaf = 0; | |
4731d4c7 | 2159 | |
37178b8b | 2160 | for_each_set_bit(i, sp->unsync_child_bitmap, 512) { |
7a8f1a74 | 2161 | struct kvm_mmu_page *child; |
4731d4c7 MT |
2162 | u64 ent = sp->spt[i]; |
2163 | ||
fd951457 TY |
2164 | if (!is_shadow_present_pte(ent) || is_large_pte(ent)) { |
2165 | clear_unsync_child_bit(sp, i); | |
2166 | continue; | |
2167 | } | |
7a8f1a74 | 2168 | |
e47c4aee | 2169 | child = to_shadow_page(ent & PT64_BASE_ADDR_MASK); |
7a8f1a74 XG |
2170 | |
2171 | if (child->unsync_children) { | |
2172 | if (mmu_pages_add(pvec, child, i)) | |
2173 | return -ENOSPC; | |
2174 | ||
2175 | ret = __mmu_unsync_walk(child, pvec); | |
fd951457 TY |
2176 | if (!ret) { |
2177 | clear_unsync_child_bit(sp, i); | |
2178 | continue; | |
2179 | } else if (ret > 0) { | |
7a8f1a74 | 2180 | nr_unsync_leaf += ret; |
fd951457 | 2181 | } else |
7a8f1a74 XG |
2182 | return ret; |
2183 | } else if (child->unsync) { | |
2184 | nr_unsync_leaf++; | |
2185 | if (mmu_pages_add(pvec, child, i)) | |
2186 | return -ENOSPC; | |
2187 | } else | |
fd951457 | 2188 | clear_unsync_child_bit(sp, i); |
4731d4c7 MT |
2189 | } |
2190 | ||
60c8aec6 MT |
2191 | return nr_unsync_leaf; |
2192 | } | |
2193 | ||
e23d3fef XG |
2194 | #define INVALID_INDEX (-1) |
2195 | ||
60c8aec6 MT |
2196 | static int mmu_unsync_walk(struct kvm_mmu_page *sp, |
2197 | struct kvm_mmu_pages *pvec) | |
2198 | { | |
0a47cd85 | 2199 | pvec->nr = 0; |
60c8aec6 MT |
2200 | if (!sp->unsync_children) |
2201 | return 0; | |
2202 | ||
e23d3fef | 2203 | mmu_pages_add(pvec, sp, INVALID_INDEX); |
60c8aec6 | 2204 | return __mmu_unsync_walk(sp, pvec); |
4731d4c7 MT |
2205 | } |
2206 | ||
4731d4c7 MT |
2207 | static void kvm_unlink_unsync_page(struct kvm *kvm, struct kvm_mmu_page *sp) |
2208 | { | |
2209 | WARN_ON(!sp->unsync); | |
5e1b3ddb | 2210 | trace_kvm_mmu_sync_page(sp); |
4731d4c7 MT |
2211 | sp->unsync = 0; |
2212 | --kvm->stat.mmu_unsync; | |
2213 | } | |
2214 | ||
83cdb568 SC |
2215 | static bool kvm_mmu_prepare_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp, |
2216 | struct list_head *invalid_list); | |
7775834a XG |
2217 | static void kvm_mmu_commit_zap_page(struct kvm *kvm, |
2218 | struct list_head *invalid_list); | |
4731d4c7 | 2219 | |
ac101b7c SC |
2220 | #define for_each_valid_sp(_kvm, _sp, _list) \ |
2221 | hlist_for_each_entry(_sp, _list, hash_link) \ | |
fac026da | 2222 | if (is_obsolete_sp((_kvm), (_sp))) { \ |
f3414bc7 | 2223 | } else |
1044b030 TY |
2224 | |
2225 | #define for_each_gfn_indirect_valid_sp(_kvm, _sp, _gfn) \ | |
ac101b7c SC |
2226 | for_each_valid_sp(_kvm, _sp, \ |
2227 | &(_kvm)->arch.mmu_page_hash[kvm_page_table_hashfn(_gfn)]) \ | |
f3414bc7 | 2228 | if ((_sp)->gfn != (_gfn) || (_sp)->role.direct) {} else |
7ae680eb | 2229 | |
47c42e6b SC |
2230 | static inline bool is_ept_sp(struct kvm_mmu_page *sp) |
2231 | { | |
2232 | return sp->role.cr0_wp && sp->role.smap_andnot_wp; | |
2233 | } | |
2234 | ||
f918b443 | 2235 | /* @sp->gfn should be write-protected at the call site */ |
1f50f1b3 PB |
2236 | static bool __kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp, |
2237 | struct list_head *invalid_list) | |
4731d4c7 | 2238 | { |
47c42e6b SC |
2239 | if ((!is_ept_sp(sp) && sp->role.gpte_is_8_bytes != !!is_pae(vcpu)) || |
2240 | vcpu->arch.mmu->sync_page(vcpu, sp) == 0) { | |
d98ba053 | 2241 | kvm_mmu_prepare_zap_page(vcpu->kvm, sp, invalid_list); |
1f50f1b3 | 2242 | return false; |
4731d4c7 MT |
2243 | } |
2244 | ||
1f50f1b3 | 2245 | return true; |
4731d4c7 MT |
2246 | } |
2247 | ||
a2113634 SC |
2248 | static bool kvm_mmu_remote_flush_or_zap(struct kvm *kvm, |
2249 | struct list_head *invalid_list, | |
2250 | bool remote_flush) | |
2251 | { | |
cfd32acf | 2252 | if (!remote_flush && list_empty(invalid_list)) |
a2113634 SC |
2253 | return false; |
2254 | ||
2255 | if (!list_empty(invalid_list)) | |
2256 | kvm_mmu_commit_zap_page(kvm, invalid_list); | |
2257 | else | |
2258 | kvm_flush_remote_tlbs(kvm); | |
2259 | return true; | |
2260 | } | |
2261 | ||
35a70510 PB |
2262 | static void kvm_mmu_flush_or_zap(struct kvm_vcpu *vcpu, |
2263 | struct list_head *invalid_list, | |
2264 | bool remote_flush, bool local_flush) | |
1d9dc7e0 | 2265 | { |
a2113634 | 2266 | if (kvm_mmu_remote_flush_or_zap(vcpu->kvm, invalid_list, remote_flush)) |
35a70510 | 2267 | return; |
d98ba053 | 2268 | |
a2113634 | 2269 | if (local_flush) |
8c8560b8 | 2270 | kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu); |
1d9dc7e0 XG |
2271 | } |
2272 | ||
e37fa785 XG |
2273 | #ifdef CONFIG_KVM_MMU_AUDIT |
2274 | #include "mmu_audit.c" | |
2275 | #else | |
2276 | static void kvm_mmu_audit(struct kvm_vcpu *vcpu, int point) { } | |
2277 | static void mmu_audit_disable(void) { } | |
2278 | #endif | |
2279 | ||
002c5f73 SC |
2280 | static bool is_obsolete_sp(struct kvm *kvm, struct kvm_mmu_page *sp) |
2281 | { | |
fac026da SC |
2282 | return sp->role.invalid || |
2283 | unlikely(sp->mmu_valid_gen != kvm->arch.mmu_valid_gen); | |
002c5f73 SC |
2284 | } |
2285 | ||
1f50f1b3 | 2286 | static bool kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp, |
d98ba053 | 2287 | struct list_head *invalid_list) |
1d9dc7e0 | 2288 | { |
9a43c5d9 PB |
2289 | kvm_unlink_unsync_page(vcpu->kvm, sp); |
2290 | return __kvm_sync_page(vcpu, sp, invalid_list); | |
1d9dc7e0 XG |
2291 | } |
2292 | ||
9f1a122f | 2293 | /* @gfn should be write-protected at the call site */ |
2a74003a PB |
2294 | static bool kvm_sync_pages(struct kvm_vcpu *vcpu, gfn_t gfn, |
2295 | struct list_head *invalid_list) | |
9f1a122f | 2296 | { |
9f1a122f | 2297 | struct kvm_mmu_page *s; |
2a74003a | 2298 | bool ret = false; |
9f1a122f | 2299 | |
b67bfe0d | 2300 | for_each_gfn_indirect_valid_sp(vcpu->kvm, s, gfn) { |
7ae680eb | 2301 | if (!s->unsync) |
9f1a122f XG |
2302 | continue; |
2303 | ||
3bae0459 | 2304 | WARN_ON(s->role.level != PG_LEVEL_4K); |
2a74003a | 2305 | ret |= kvm_sync_page(vcpu, s, invalid_list); |
9f1a122f XG |
2306 | } |
2307 | ||
2a74003a | 2308 | return ret; |
9f1a122f XG |
2309 | } |
2310 | ||
60c8aec6 | 2311 | struct mmu_page_path { |
2a7266a8 YZ |
2312 | struct kvm_mmu_page *parent[PT64_ROOT_MAX_LEVEL]; |
2313 | unsigned int idx[PT64_ROOT_MAX_LEVEL]; | |
4731d4c7 MT |
2314 | }; |
2315 | ||
60c8aec6 | 2316 | #define for_each_sp(pvec, sp, parents, i) \ |
0a47cd85 | 2317 | for (i = mmu_pages_first(&pvec, &parents); \ |
60c8aec6 MT |
2318 | i < pvec.nr && ({ sp = pvec.page[i].sp; 1;}); \ |
2319 | i = mmu_pages_next(&pvec, &parents, i)) | |
2320 | ||
cded19f3 HE |
2321 | static int mmu_pages_next(struct kvm_mmu_pages *pvec, |
2322 | struct mmu_page_path *parents, | |
2323 | int i) | |
60c8aec6 MT |
2324 | { |
2325 | int n; | |
2326 | ||
2327 | for (n = i+1; n < pvec->nr; n++) { | |
2328 | struct kvm_mmu_page *sp = pvec->page[n].sp; | |
0a47cd85 PB |
2329 | unsigned idx = pvec->page[n].idx; |
2330 | int level = sp->role.level; | |
60c8aec6 | 2331 | |
0a47cd85 | 2332 | parents->idx[level-1] = idx; |
3bae0459 | 2333 | if (level == PG_LEVEL_4K) |
0a47cd85 | 2334 | break; |
60c8aec6 | 2335 | |
0a47cd85 | 2336 | parents->parent[level-2] = sp; |
60c8aec6 MT |
2337 | } |
2338 | ||
2339 | return n; | |
2340 | } | |
2341 | ||
0a47cd85 PB |
2342 | static int mmu_pages_first(struct kvm_mmu_pages *pvec, |
2343 | struct mmu_page_path *parents) | |
2344 | { | |
2345 | struct kvm_mmu_page *sp; | |
2346 | int level; | |
2347 | ||
2348 | if (pvec->nr == 0) | |
2349 | return 0; | |
2350 | ||
e23d3fef XG |
2351 | WARN_ON(pvec->page[0].idx != INVALID_INDEX); |
2352 | ||
0a47cd85 PB |
2353 | sp = pvec->page[0].sp; |
2354 | level = sp->role.level; | |
3bae0459 | 2355 | WARN_ON(level == PG_LEVEL_4K); |
0a47cd85 PB |
2356 | |
2357 | parents->parent[level-2] = sp; | |
2358 | ||
2359 | /* Also set up a sentinel. Further entries in pvec are all | |
2360 | * children of sp, so this element is never overwritten. | |
2361 | */ | |
2362 | parents->parent[level-1] = NULL; | |
2363 | return mmu_pages_next(pvec, parents, 0); | |
2364 | } | |
2365 | ||
cded19f3 | 2366 | static void mmu_pages_clear_parents(struct mmu_page_path *parents) |
4731d4c7 | 2367 | { |
60c8aec6 MT |
2368 | struct kvm_mmu_page *sp; |
2369 | unsigned int level = 0; | |
2370 | ||
2371 | do { | |
2372 | unsigned int idx = parents->idx[level]; | |
60c8aec6 MT |
2373 | sp = parents->parent[level]; |
2374 | if (!sp) | |
2375 | return; | |
2376 | ||
e23d3fef | 2377 | WARN_ON(idx == INVALID_INDEX); |
fd951457 | 2378 | clear_unsync_child_bit(sp, idx); |
60c8aec6 | 2379 | level++; |
0a47cd85 | 2380 | } while (!sp->unsync_children); |
60c8aec6 | 2381 | } |
4731d4c7 | 2382 | |
60c8aec6 MT |
2383 | static void mmu_sync_children(struct kvm_vcpu *vcpu, |
2384 | struct kvm_mmu_page *parent) | |
2385 | { | |
2386 | int i; | |
2387 | struct kvm_mmu_page *sp; | |
2388 | struct mmu_page_path parents; | |
2389 | struct kvm_mmu_pages pages; | |
d98ba053 | 2390 | LIST_HEAD(invalid_list); |
50c9e6f3 | 2391 | bool flush = false; |
60c8aec6 | 2392 | |
60c8aec6 | 2393 | while (mmu_unsync_walk(parent, &pages)) { |
2f84569f | 2394 | bool protected = false; |
b1a36821 MT |
2395 | |
2396 | for_each_sp(pages, sp, parents, i) | |
54bf36aa | 2397 | protected |= rmap_write_protect(vcpu, sp->gfn); |
b1a36821 | 2398 | |
50c9e6f3 | 2399 | if (protected) { |
b1a36821 | 2400 | kvm_flush_remote_tlbs(vcpu->kvm); |
50c9e6f3 PB |
2401 | flush = false; |
2402 | } | |
b1a36821 | 2403 | |
60c8aec6 | 2404 | for_each_sp(pages, sp, parents, i) { |
1f50f1b3 | 2405 | flush |= kvm_sync_page(vcpu, sp, &invalid_list); |
60c8aec6 MT |
2406 | mmu_pages_clear_parents(&parents); |
2407 | } | |
50c9e6f3 PB |
2408 | if (need_resched() || spin_needbreak(&vcpu->kvm->mmu_lock)) { |
2409 | kvm_mmu_flush_or_zap(vcpu, &invalid_list, false, flush); | |
2410 | cond_resched_lock(&vcpu->kvm->mmu_lock); | |
2411 | flush = false; | |
2412 | } | |
60c8aec6 | 2413 | } |
50c9e6f3 PB |
2414 | |
2415 | kvm_mmu_flush_or_zap(vcpu, &invalid_list, false, flush); | |
4731d4c7 MT |
2416 | } |
2417 | ||
a30f47cb XG |
2418 | static void __clear_sp_write_flooding_count(struct kvm_mmu_page *sp) |
2419 | { | |
e5691a81 | 2420 | atomic_set(&sp->write_flooding_count, 0); |
a30f47cb XG |
2421 | } |
2422 | ||
2423 | static void clear_sp_write_flooding_count(u64 *spte) | |
2424 | { | |
57354682 | 2425 | __clear_sp_write_flooding_count(sptep_to_sp(spte)); |
a30f47cb XG |
2426 | } |
2427 | ||
cea0f0e7 AK |
2428 | static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu, |
2429 | gfn_t gfn, | |
2430 | gva_t gaddr, | |
2431 | unsigned level, | |
f6e2c02b | 2432 | int direct, |
0a2b64c5 | 2433 | unsigned int access) |
cea0f0e7 | 2434 | { |
fb58a9c3 | 2435 | bool direct_mmu = vcpu->arch.mmu->direct_map; |
cea0f0e7 | 2436 | union kvm_mmu_page_role role; |
ac101b7c | 2437 | struct hlist_head *sp_list; |
cea0f0e7 | 2438 | unsigned quadrant; |
9f1a122f | 2439 | struct kvm_mmu_page *sp; |
9f1a122f | 2440 | bool need_sync = false; |
2a74003a | 2441 | bool flush = false; |
f3414bc7 | 2442 | int collisions = 0; |
2a74003a | 2443 | LIST_HEAD(invalid_list); |
cea0f0e7 | 2444 | |
36d9594d | 2445 | role = vcpu->arch.mmu->mmu_role.base; |
cea0f0e7 | 2446 | role.level = level; |
f6e2c02b | 2447 | role.direct = direct; |
84b0c8c6 | 2448 | if (role.direct) |
47c42e6b | 2449 | role.gpte_is_8_bytes = true; |
41074d07 | 2450 | role.access = access; |
fb58a9c3 | 2451 | if (!direct_mmu && vcpu->arch.mmu->root_level <= PT32_ROOT_LEVEL) { |
cea0f0e7 AK |
2452 | quadrant = gaddr >> (PAGE_SHIFT + (PT64_PT_BITS * level)); |
2453 | quadrant &= (1 << ((PT32_PT_BITS - PT64_PT_BITS) * level)) - 1; | |
2454 | role.quadrant = quadrant; | |
2455 | } | |
ac101b7c SC |
2456 | |
2457 | sp_list = &vcpu->kvm->arch.mmu_page_hash[kvm_page_table_hashfn(gfn)]; | |
2458 | for_each_valid_sp(vcpu->kvm, sp, sp_list) { | |
f3414bc7 DM |
2459 | if (sp->gfn != gfn) { |
2460 | collisions++; | |
2461 | continue; | |
2462 | } | |
2463 | ||
7ae680eb XG |
2464 | if (!need_sync && sp->unsync) |
2465 | need_sync = true; | |
4731d4c7 | 2466 | |
7ae680eb XG |
2467 | if (sp->role.word != role.word) |
2468 | continue; | |
4731d4c7 | 2469 | |
fb58a9c3 SC |
2470 | if (direct_mmu) |
2471 | goto trace_get_page; | |
2472 | ||
2a74003a PB |
2473 | if (sp->unsync) { |
2474 | /* The page is good, but __kvm_sync_page might still end | |
2475 | * up zapping it. If so, break in order to rebuild it. | |
2476 | */ | |
2477 | if (!__kvm_sync_page(vcpu, sp, &invalid_list)) | |
2478 | break; | |
2479 | ||
2480 | WARN_ON(!list_empty(&invalid_list)); | |
8c8560b8 | 2481 | kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu); |
2a74003a | 2482 | } |
e02aa901 | 2483 | |
98bba238 | 2484 | if (sp->unsync_children) |
f6f6195b | 2485 | kvm_make_request(KVM_REQ_MMU_SYNC, vcpu); |
e02aa901 | 2486 | |
a30f47cb | 2487 | __clear_sp_write_flooding_count(sp); |
fb58a9c3 SC |
2488 | |
2489 | trace_get_page: | |
7ae680eb | 2490 | trace_kvm_mmu_get_page(sp, false); |
f3414bc7 | 2491 | goto out; |
7ae680eb | 2492 | } |
47005792 | 2493 | |
dfc5aa00 | 2494 | ++vcpu->kvm->stat.mmu_cache_miss; |
47005792 TY |
2495 | |
2496 | sp = kvm_mmu_alloc_page(vcpu, direct); | |
2497 | ||
4db35314 AK |
2498 | sp->gfn = gfn; |
2499 | sp->role = role; | |
ac101b7c | 2500 | hlist_add_head(&sp->hash_link, sp_list); |
f6e2c02b | 2501 | if (!direct) { |
56ca57f9 XG |
2502 | /* |
2503 | * we should do write protection before syncing pages | |
2504 | * otherwise the content of the synced shadow page may | |
2505 | * be inconsistent with guest page table. | |
2506 | */ | |
2507 | account_shadowed(vcpu->kvm, sp); | |
3bae0459 | 2508 | if (level == PG_LEVEL_4K && rmap_write_protect(vcpu, gfn)) |
c3134ce2 | 2509 | kvm_flush_remote_tlbs_with_address(vcpu->kvm, gfn, 1); |
9f1a122f | 2510 | |
3bae0459 | 2511 | if (level > PG_LEVEL_4K && need_sync) |
2a74003a | 2512 | flush |= kvm_sync_pages(vcpu, gfn, &invalid_list); |
4731d4c7 | 2513 | } |
f691fe1d | 2514 | trace_kvm_mmu_get_page(sp, true); |
2a74003a PB |
2515 | |
2516 | kvm_mmu_flush_or_zap(vcpu, &invalid_list, false, flush); | |
f3414bc7 DM |
2517 | out: |
2518 | if (collisions > vcpu->kvm->stat.max_mmu_page_hash_collisions) | |
2519 | vcpu->kvm->stat.max_mmu_page_hash_collisions = collisions; | |
4db35314 | 2520 | return sp; |
cea0f0e7 AK |
2521 | } |
2522 | ||
7eb77e9f JS |
2523 | static void shadow_walk_init_using_root(struct kvm_shadow_walk_iterator *iterator, |
2524 | struct kvm_vcpu *vcpu, hpa_t root, | |
2525 | u64 addr) | |
2d11123a AK |
2526 | { |
2527 | iterator->addr = addr; | |
7eb77e9f | 2528 | iterator->shadow_addr = root; |
44dd3ffa | 2529 | iterator->level = vcpu->arch.mmu->shadow_root_level; |
81407ca5 | 2530 | |
2a7266a8 | 2531 | if (iterator->level == PT64_ROOT_4LEVEL && |
44dd3ffa VK |
2532 | vcpu->arch.mmu->root_level < PT64_ROOT_4LEVEL && |
2533 | !vcpu->arch.mmu->direct_map) | |
81407ca5 JR |
2534 | --iterator->level; |
2535 | ||
2d11123a | 2536 | if (iterator->level == PT32E_ROOT_LEVEL) { |
7eb77e9f JS |
2537 | /* |
2538 | * prev_root is currently only used for 64-bit hosts. So only | |
2539 | * the active root_hpa is valid here. | |
2540 | */ | |
44dd3ffa | 2541 | BUG_ON(root != vcpu->arch.mmu->root_hpa); |
7eb77e9f | 2542 | |
2d11123a | 2543 | iterator->shadow_addr |
44dd3ffa | 2544 | = vcpu->arch.mmu->pae_root[(addr >> 30) & 3]; |
2d11123a AK |
2545 | iterator->shadow_addr &= PT64_BASE_ADDR_MASK; |
2546 | --iterator->level; | |
2547 | if (!iterator->shadow_addr) | |
2548 | iterator->level = 0; | |
2549 | } | |
2550 | } | |
2551 | ||
7eb77e9f JS |
2552 | static void shadow_walk_init(struct kvm_shadow_walk_iterator *iterator, |
2553 | struct kvm_vcpu *vcpu, u64 addr) | |
2554 | { | |
44dd3ffa | 2555 | shadow_walk_init_using_root(iterator, vcpu, vcpu->arch.mmu->root_hpa, |
7eb77e9f JS |
2556 | addr); |
2557 | } | |
2558 | ||
2d11123a AK |
2559 | static bool shadow_walk_okay(struct kvm_shadow_walk_iterator *iterator) |
2560 | { | |
3bae0459 | 2561 | if (iterator->level < PG_LEVEL_4K) |
2d11123a | 2562 | return false; |
4d88954d | 2563 | |
2d11123a AK |
2564 | iterator->index = SHADOW_PT_INDEX(iterator->addr, iterator->level); |
2565 | iterator->sptep = ((u64 *)__va(iterator->shadow_addr)) + iterator->index; | |
2566 | return true; | |
2567 | } | |
2568 | ||
c2a2ac2b XG |
2569 | static void __shadow_walk_next(struct kvm_shadow_walk_iterator *iterator, |
2570 | u64 spte) | |
2d11123a | 2571 | { |
c2a2ac2b | 2572 | if (is_last_spte(spte, iterator->level)) { |
052331be XG |
2573 | iterator->level = 0; |
2574 | return; | |
2575 | } | |
2576 | ||
c2a2ac2b | 2577 | iterator->shadow_addr = spte & PT64_BASE_ADDR_MASK; |
2d11123a AK |
2578 | --iterator->level; |
2579 | } | |
2580 | ||
c2a2ac2b XG |
2581 | static void shadow_walk_next(struct kvm_shadow_walk_iterator *iterator) |
2582 | { | |
bb606a9b | 2583 | __shadow_walk_next(iterator, *iterator->sptep); |
c2a2ac2b XG |
2584 | } |
2585 | ||
cc4674d0 | 2586 | static u64 make_nonleaf_spte(u64 *child_pt, bool ad_disabled) |
32ef26a3 AK |
2587 | { |
2588 | u64 spte; | |
2589 | ||
cc4674d0 | 2590 | spte = __pa(child_pt) | shadow_present_mask | PT_WRITABLE_MASK | |
d0ec49d4 | 2591 | shadow_user_mask | shadow_x_mask | shadow_me_mask; |
ac8d57e5 | 2592 | |
cc4674d0 | 2593 | if (ad_disabled) |
6eeb4ef0 | 2594 | spte |= SPTE_AD_DISABLED_MASK; |
ac8d57e5 PF |
2595 | else |
2596 | spte |= shadow_accessed_mask; | |
24db2734 | 2597 | |
cc4674d0 BG |
2598 | return spte; |
2599 | } | |
2600 | ||
2601 | static void link_shadow_page(struct kvm_vcpu *vcpu, u64 *sptep, | |
2602 | struct kvm_mmu_page *sp) | |
2603 | { | |
2604 | u64 spte; | |
2605 | ||
2606 | BUILD_BUG_ON(VMX_EPT_WRITABLE_MASK != PT_WRITABLE_MASK); | |
2607 | ||
2608 | spte = make_nonleaf_spte(sp->spt, sp_ad_disabled(sp)); | |
2609 | ||
1df9f2dc | 2610 | mmu_spte_set(sptep, spte); |
98bba238 TY |
2611 | |
2612 | mmu_page_add_parent_pte(vcpu, sp, sptep); | |
2613 | ||
2614 | if (sp->unsync_children || sp->unsync) | |
2615 | mark_unsync(sptep); | |
32ef26a3 AK |
2616 | } |
2617 | ||
a357bd22 AK |
2618 | static void validate_direct_spte(struct kvm_vcpu *vcpu, u64 *sptep, |
2619 | unsigned direct_access) | |
2620 | { | |
2621 | if (is_shadow_present_pte(*sptep) && !is_large_pte(*sptep)) { | |
2622 | struct kvm_mmu_page *child; | |
2623 | ||
2624 | /* | |
2625 | * For the direct sp, if the guest pte's dirty bit | |
2626 | * changed form clean to dirty, it will corrupt the | |
2627 | * sp's access: allow writable in the read-only sp, | |
2628 | * so we should update the spte at this point to get | |
2629 | * a new sp with the correct access. | |
2630 | */ | |
e47c4aee | 2631 | child = to_shadow_page(*sptep & PT64_BASE_ADDR_MASK); |
a357bd22 AK |
2632 | if (child->role.access == direct_access) |
2633 | return; | |
2634 | ||
bcdd9a93 | 2635 | drop_parent_pte(child, sptep); |
c3134ce2 | 2636 | kvm_flush_remote_tlbs_with_address(vcpu->kvm, child->gfn, 1); |
a357bd22 AK |
2637 | } |
2638 | } | |
2639 | ||
2de4085c BG |
2640 | /* Returns the number of zapped non-leaf child shadow pages. */ |
2641 | static int mmu_page_zap_pte(struct kvm *kvm, struct kvm_mmu_page *sp, | |
2642 | u64 *spte, struct list_head *invalid_list) | |
38e3b2b2 XG |
2643 | { |
2644 | u64 pte; | |
2645 | struct kvm_mmu_page *child; | |
2646 | ||
2647 | pte = *spte; | |
2648 | if (is_shadow_present_pte(pte)) { | |
505aef8f | 2649 | if (is_last_spte(pte, sp->role.level)) { |
c3707958 | 2650 | drop_spte(kvm, spte); |
505aef8f XG |
2651 | if (is_large_pte(pte)) |
2652 | --kvm->stat.lpages; | |
2653 | } else { | |
e47c4aee | 2654 | child = to_shadow_page(pte & PT64_BASE_ADDR_MASK); |
bcdd9a93 | 2655 | drop_parent_pte(child, spte); |
2de4085c BG |
2656 | |
2657 | /* | |
2658 | * Recursively zap nested TDP SPs, parentless SPs are | |
2659 | * unlikely to be used again in the near future. This | |
2660 | * avoids retaining a large number of stale nested SPs. | |
2661 | */ | |
2662 | if (tdp_enabled && invalid_list && | |
2663 | child->role.guest_mode && !child->parent_ptes.val) | |
2664 | return kvm_mmu_prepare_zap_page(kvm, child, | |
2665 | invalid_list); | |
38e3b2b2 | 2666 | } |
ace569e0 | 2667 | } else if (is_mmio_spte(pte)) { |
ce88decf | 2668 | mmu_spte_clear_no_track(spte); |
ace569e0 | 2669 | } |
2de4085c | 2670 | return 0; |
38e3b2b2 XG |
2671 | } |
2672 | ||
2de4085c BG |
2673 | static int kvm_mmu_page_unlink_children(struct kvm *kvm, |
2674 | struct kvm_mmu_page *sp, | |
2675 | struct list_head *invalid_list) | |
a436036b | 2676 | { |
2de4085c | 2677 | int zapped = 0; |
697fe2e2 | 2678 | unsigned i; |
697fe2e2 | 2679 | |
38e3b2b2 | 2680 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) |
2de4085c BG |
2681 | zapped += mmu_page_zap_pte(kvm, sp, sp->spt + i, invalid_list); |
2682 | ||
2683 | return zapped; | |
a436036b AK |
2684 | } |
2685 | ||
31aa2b44 | 2686 | static void kvm_mmu_unlink_parents(struct kvm *kvm, struct kvm_mmu_page *sp) |
a436036b | 2687 | { |
1e3f42f0 TY |
2688 | u64 *sptep; |
2689 | struct rmap_iterator iter; | |
a436036b | 2690 | |
018aabb5 | 2691 | while ((sptep = rmap_get_first(&sp->parent_ptes, &iter))) |
1e3f42f0 | 2692 | drop_parent_pte(sp, sptep); |
31aa2b44 AK |
2693 | } |
2694 | ||
60c8aec6 | 2695 | static int mmu_zap_unsync_children(struct kvm *kvm, |
7775834a XG |
2696 | struct kvm_mmu_page *parent, |
2697 | struct list_head *invalid_list) | |
4731d4c7 | 2698 | { |
60c8aec6 MT |
2699 | int i, zapped = 0; |
2700 | struct mmu_page_path parents; | |
2701 | struct kvm_mmu_pages pages; | |
4731d4c7 | 2702 | |
3bae0459 | 2703 | if (parent->role.level == PG_LEVEL_4K) |
4731d4c7 | 2704 | return 0; |
60c8aec6 | 2705 | |
60c8aec6 MT |
2706 | while (mmu_unsync_walk(parent, &pages)) { |
2707 | struct kvm_mmu_page *sp; | |
2708 | ||
2709 | for_each_sp(pages, sp, parents, i) { | |
7775834a | 2710 | kvm_mmu_prepare_zap_page(kvm, sp, invalid_list); |
60c8aec6 | 2711 | mmu_pages_clear_parents(&parents); |
77662e00 | 2712 | zapped++; |
60c8aec6 | 2713 | } |
60c8aec6 MT |
2714 | } |
2715 | ||
2716 | return zapped; | |
4731d4c7 MT |
2717 | } |
2718 | ||
83cdb568 SC |
2719 | static bool __kvm_mmu_prepare_zap_page(struct kvm *kvm, |
2720 | struct kvm_mmu_page *sp, | |
2721 | struct list_head *invalid_list, | |
2722 | int *nr_zapped) | |
31aa2b44 | 2723 | { |
83cdb568 | 2724 | bool list_unstable; |
f691fe1d | 2725 | |
7775834a | 2726 | trace_kvm_mmu_prepare_zap_page(sp); |
31aa2b44 | 2727 | ++kvm->stat.mmu_shadow_zapped; |
83cdb568 | 2728 | *nr_zapped = mmu_zap_unsync_children(kvm, sp, invalid_list); |
2de4085c | 2729 | *nr_zapped += kvm_mmu_page_unlink_children(kvm, sp, invalid_list); |
31aa2b44 | 2730 | kvm_mmu_unlink_parents(kvm, sp); |
5304b8d3 | 2731 | |
83cdb568 SC |
2732 | /* Zapping children means active_mmu_pages has become unstable. */ |
2733 | list_unstable = *nr_zapped; | |
2734 | ||
f6e2c02b | 2735 | if (!sp->role.invalid && !sp->role.direct) |
3ed1a478 | 2736 | unaccount_shadowed(kvm, sp); |
5304b8d3 | 2737 | |
4731d4c7 MT |
2738 | if (sp->unsync) |
2739 | kvm_unlink_unsync_page(kvm, sp); | |
4db35314 | 2740 | if (!sp->root_count) { |
54a4f023 | 2741 | /* Count self */ |
83cdb568 | 2742 | (*nr_zapped)++; |
f95eec9b SC |
2743 | |
2744 | /* | |
2745 | * Already invalid pages (previously active roots) are not on | |
2746 | * the active page list. See list_del() in the "else" case of | |
2747 | * !sp->root_count. | |
2748 | */ | |
2749 | if (sp->role.invalid) | |
2750 | list_add(&sp->link, invalid_list); | |
2751 | else | |
2752 | list_move(&sp->link, invalid_list); | |
aa6bd187 | 2753 | kvm_mod_used_mmu_pages(kvm, -1); |
2e53d63a | 2754 | } else { |
f95eec9b SC |
2755 | /* |
2756 | * Remove the active root from the active page list, the root | |
2757 | * will be explicitly freed when the root_count hits zero. | |
2758 | */ | |
2759 | list_del(&sp->link); | |
05988d72 | 2760 | |
10605204 SC |
2761 | /* |
2762 | * Obsolete pages cannot be used on any vCPUs, see the comment | |
2763 | * in kvm_mmu_zap_all_fast(). Note, is_obsolete_sp() also | |
2764 | * treats invalid shadow pages as being obsolete. | |
2765 | */ | |
2766 | if (!is_obsolete_sp(kvm, sp)) | |
05988d72 | 2767 | kvm_reload_remote_mmus(kvm); |
2e53d63a | 2768 | } |
7775834a | 2769 | |
b8e8c830 PB |
2770 | if (sp->lpage_disallowed) |
2771 | unaccount_huge_nx_page(kvm, sp); | |
2772 | ||
7775834a | 2773 | sp->role.invalid = 1; |
83cdb568 SC |
2774 | return list_unstable; |
2775 | } | |
2776 | ||
2777 | static bool kvm_mmu_prepare_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp, | |
2778 | struct list_head *invalid_list) | |
2779 | { | |
2780 | int nr_zapped; | |
2781 | ||
2782 | __kvm_mmu_prepare_zap_page(kvm, sp, invalid_list, &nr_zapped); | |
2783 | return nr_zapped; | |
a436036b AK |
2784 | } |
2785 | ||
7775834a XG |
2786 | static void kvm_mmu_commit_zap_page(struct kvm *kvm, |
2787 | struct list_head *invalid_list) | |
2788 | { | |
945315b9 | 2789 | struct kvm_mmu_page *sp, *nsp; |
7775834a XG |
2790 | |
2791 | if (list_empty(invalid_list)) | |
2792 | return; | |
2793 | ||
c142786c | 2794 | /* |
9753f529 LT |
2795 | * We need to make sure everyone sees our modifications to |
2796 | * the page tables and see changes to vcpu->mode here. The barrier | |
2797 | * in the kvm_flush_remote_tlbs() achieves this. This pairs | |
2798 | * with vcpu_enter_guest and walk_shadow_page_lockless_begin/end. | |
2799 | * | |
2800 | * In addition, kvm_flush_remote_tlbs waits for all vcpus to exit | |
2801 | * guest mode and/or lockless shadow page table walks. | |
c142786c AK |
2802 | */ |
2803 | kvm_flush_remote_tlbs(kvm); | |
c2a2ac2b | 2804 | |
945315b9 | 2805 | list_for_each_entry_safe(sp, nsp, invalid_list, link) { |
7775834a | 2806 | WARN_ON(!sp->role.invalid || sp->root_count); |
aa6bd187 | 2807 | kvm_mmu_free_page(sp); |
945315b9 | 2808 | } |
7775834a XG |
2809 | } |
2810 | ||
6b82ef2c SC |
2811 | static unsigned long kvm_mmu_zap_oldest_mmu_pages(struct kvm *kvm, |
2812 | unsigned long nr_to_zap) | |
5da59607 | 2813 | { |
6b82ef2c SC |
2814 | unsigned long total_zapped = 0; |
2815 | struct kvm_mmu_page *sp, *tmp; | |
ba7888dd | 2816 | LIST_HEAD(invalid_list); |
6b82ef2c SC |
2817 | bool unstable; |
2818 | int nr_zapped; | |
5da59607 TY |
2819 | |
2820 | if (list_empty(&kvm->arch.active_mmu_pages)) | |
ba7888dd SC |
2821 | return 0; |
2822 | ||
6b82ef2c SC |
2823 | restart: |
2824 | list_for_each_entry_safe(sp, tmp, &kvm->arch.active_mmu_pages, link) { | |
2825 | /* | |
2826 | * Don't zap active root pages, the page itself can't be freed | |
2827 | * and zapping it will just force vCPUs to realloc and reload. | |
2828 | */ | |
2829 | if (sp->root_count) | |
2830 | continue; | |
2831 | ||
2832 | unstable = __kvm_mmu_prepare_zap_page(kvm, sp, &invalid_list, | |
2833 | &nr_zapped); | |
2834 | total_zapped += nr_zapped; | |
2835 | if (total_zapped >= nr_to_zap) | |
ba7888dd SC |
2836 | break; |
2837 | ||
6b82ef2c SC |
2838 | if (unstable) |
2839 | goto restart; | |
ba7888dd | 2840 | } |
5da59607 | 2841 | |
6b82ef2c SC |
2842 | kvm_mmu_commit_zap_page(kvm, &invalid_list); |
2843 | ||
2844 | kvm->stat.mmu_recycled += total_zapped; | |
2845 | return total_zapped; | |
2846 | } | |
2847 | ||
afe8d7e6 SC |
2848 | static inline unsigned long kvm_mmu_available_pages(struct kvm *kvm) |
2849 | { | |
2850 | if (kvm->arch.n_max_mmu_pages > kvm->arch.n_used_mmu_pages) | |
2851 | return kvm->arch.n_max_mmu_pages - | |
2852 | kvm->arch.n_used_mmu_pages; | |
2853 | ||
2854 | return 0; | |
5da59607 TY |
2855 | } |
2856 | ||
ba7888dd SC |
2857 | static int make_mmu_pages_available(struct kvm_vcpu *vcpu) |
2858 | { | |
6b82ef2c | 2859 | unsigned long avail = kvm_mmu_available_pages(vcpu->kvm); |
ba7888dd | 2860 | |
6b82ef2c | 2861 | if (likely(avail >= KVM_MIN_FREE_MMU_PAGES)) |
ba7888dd SC |
2862 | return 0; |
2863 | ||
6b82ef2c | 2864 | kvm_mmu_zap_oldest_mmu_pages(vcpu->kvm, KVM_REFILL_PAGES - avail); |
ba7888dd SC |
2865 | |
2866 | if (!kvm_mmu_available_pages(vcpu->kvm)) | |
2867 | return -ENOSPC; | |
2868 | return 0; | |
2869 | } | |
2870 | ||
82ce2c96 IE |
2871 | /* |
2872 | * Changing the number of mmu pages allocated to the vm | |
49d5ca26 | 2873 | * Note: if goal_nr_mmu_pages is too small, you will get dead lock |
82ce2c96 | 2874 | */ |
bc8a3d89 | 2875 | void kvm_mmu_change_mmu_pages(struct kvm *kvm, unsigned long goal_nr_mmu_pages) |
82ce2c96 | 2876 | { |
b34cb590 TY |
2877 | spin_lock(&kvm->mmu_lock); |
2878 | ||
49d5ca26 | 2879 | if (kvm->arch.n_used_mmu_pages > goal_nr_mmu_pages) { |
6b82ef2c SC |
2880 | kvm_mmu_zap_oldest_mmu_pages(kvm, kvm->arch.n_used_mmu_pages - |
2881 | goal_nr_mmu_pages); | |
82ce2c96 | 2882 | |
49d5ca26 | 2883 | goal_nr_mmu_pages = kvm->arch.n_used_mmu_pages; |
82ce2c96 | 2884 | } |
82ce2c96 | 2885 | |
49d5ca26 | 2886 | kvm->arch.n_max_mmu_pages = goal_nr_mmu_pages; |
b34cb590 TY |
2887 | |
2888 | spin_unlock(&kvm->mmu_lock); | |
82ce2c96 IE |
2889 | } |
2890 | ||
1cb3f3ae | 2891 | int kvm_mmu_unprotect_page(struct kvm *kvm, gfn_t gfn) |
a436036b | 2892 | { |
4db35314 | 2893 | struct kvm_mmu_page *sp; |
d98ba053 | 2894 | LIST_HEAD(invalid_list); |
a436036b AK |
2895 | int r; |
2896 | ||
9ad17b10 | 2897 | pgprintk("%s: looking for gfn %llx\n", __func__, gfn); |
a436036b | 2898 | r = 0; |
1cb3f3ae | 2899 | spin_lock(&kvm->mmu_lock); |
b67bfe0d | 2900 | for_each_gfn_indirect_valid_sp(kvm, sp, gfn) { |
9ad17b10 | 2901 | pgprintk("%s: gfn %llx role %x\n", __func__, gfn, |
7ae680eb XG |
2902 | sp->role.word); |
2903 | r = 1; | |
f41d335a | 2904 | kvm_mmu_prepare_zap_page(kvm, sp, &invalid_list); |
7ae680eb | 2905 | } |
d98ba053 | 2906 | kvm_mmu_commit_zap_page(kvm, &invalid_list); |
1cb3f3ae XG |
2907 | spin_unlock(&kvm->mmu_lock); |
2908 | ||
a436036b | 2909 | return r; |
cea0f0e7 | 2910 | } |
1cb3f3ae | 2911 | EXPORT_SYMBOL_GPL(kvm_mmu_unprotect_page); |
cea0f0e7 | 2912 | |
5c520e90 | 2913 | static void kvm_unsync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp) |
9cf5cf5a XG |
2914 | { |
2915 | trace_kvm_mmu_unsync_page(sp); | |
2916 | ++vcpu->kvm->stat.mmu_unsync; | |
2917 | sp->unsync = 1; | |
2918 | ||
2919 | kvm_mmu_mark_parents_unsync(sp); | |
9cf5cf5a XG |
2920 | } |
2921 | ||
3d0c27ad XG |
2922 | static bool mmu_need_write_protect(struct kvm_vcpu *vcpu, gfn_t gfn, |
2923 | bool can_unsync) | |
4731d4c7 | 2924 | { |
5c520e90 | 2925 | struct kvm_mmu_page *sp; |
4731d4c7 | 2926 | |
3d0c27ad XG |
2927 | if (kvm_page_track_is_active(vcpu, gfn, KVM_PAGE_TRACK_WRITE)) |
2928 | return true; | |
9cf5cf5a | 2929 | |
5c520e90 | 2930 | for_each_gfn_indirect_valid_sp(vcpu->kvm, sp, gfn) { |
36a2e677 | 2931 | if (!can_unsync) |
3d0c27ad | 2932 | return true; |
36a2e677 | 2933 | |
5c520e90 XG |
2934 | if (sp->unsync) |
2935 | continue; | |
9cf5cf5a | 2936 | |
3bae0459 | 2937 | WARN_ON(sp->role.level != PG_LEVEL_4K); |
5c520e90 | 2938 | kvm_unsync_page(vcpu, sp); |
4731d4c7 | 2939 | } |
3d0c27ad | 2940 | |
578e1c4d JS |
2941 | /* |
2942 | * We need to ensure that the marking of unsync pages is visible | |
2943 | * before the SPTE is updated to allow writes because | |
2944 | * kvm_mmu_sync_roots() checks the unsync flags without holding | |
2945 | * the MMU lock and so can race with this. If the SPTE was updated | |
2946 | * before the page had been marked as unsync-ed, something like the | |
2947 | * following could happen: | |
2948 | * | |
2949 | * CPU 1 CPU 2 | |
2950 | * --------------------------------------------------------------------- | |
2951 | * 1.2 Host updates SPTE | |
2952 | * to be writable | |
2953 | * 2.1 Guest writes a GPTE for GVA X. | |
2954 | * (GPTE being in the guest page table shadowed | |
2955 | * by the SP from CPU 1.) | |
2956 | * This reads SPTE during the page table walk. | |
2957 | * Since SPTE.W is read as 1, there is no | |
2958 | * fault. | |
2959 | * | |
2960 | * 2.2 Guest issues TLB flush. | |
2961 | * That causes a VM Exit. | |
2962 | * | |
2963 | * 2.3 kvm_mmu_sync_pages() reads sp->unsync. | |
2964 | * Since it is false, so it just returns. | |
2965 | * | |
2966 | * 2.4 Guest accesses GVA X. | |
2967 | * Since the mapping in the SP was not updated, | |
2968 | * so the old mapping for GVA X incorrectly | |
2969 | * gets used. | |
2970 | * 1.1 Host marks SP | |
2971 | * as unsync | |
2972 | * (sp->unsync = true) | |
2973 | * | |
2974 | * The write barrier below ensures that 1.1 happens before 1.2 and thus | |
2975 | * the situation in 2.4 does not arise. The implicit barrier in 2.2 | |
2976 | * pairs with this write barrier. | |
2977 | */ | |
2978 | smp_wmb(); | |
2979 | ||
3d0c27ad | 2980 | return false; |
4731d4c7 MT |
2981 | } |
2982 | ||
ba049e93 | 2983 | static bool kvm_is_mmio_pfn(kvm_pfn_t pfn) |
d1fe9219 PB |
2984 | { |
2985 | if (pfn_valid(pfn)) | |
aa2e063a HZ |
2986 | return !is_zero_pfn(pfn) && PageReserved(pfn_to_page(pfn)) && |
2987 | /* | |
2988 | * Some reserved pages, such as those from NVDIMM | |
2989 | * DAX devices, are not for MMIO, and can be mapped | |
2990 | * with cached memory type for better performance. | |
2991 | * However, the above check misconceives those pages | |
2992 | * as MMIO, and results in KVM mapping them with UC | |
2993 | * memory type, which would hurt the performance. | |
2994 | * Therefore, we check the host memory type in addition | |
2995 | * and only treat UC/UC-/WC pages as MMIO. | |
2996 | */ | |
2997 | (!pat_enabled() || pat_pfn_immune_to_uc_mtrr(pfn)); | |
d1fe9219 | 2998 | |
0c55671f KA |
2999 | return !e820__mapped_raw_any(pfn_to_hpa(pfn), |
3000 | pfn_to_hpa(pfn + 1) - 1, | |
3001 | E820_TYPE_RAM); | |
d1fe9219 PB |
3002 | } |
3003 | ||
5ce4786f JS |
3004 | /* Bits which may be returned by set_spte() */ |
3005 | #define SET_SPTE_WRITE_PROTECTED_PT BIT(0) | |
3006 | #define SET_SPTE_NEED_REMOTE_TLB_FLUSH BIT(1) | |
12703759 | 3007 | #define SET_SPTE_SPURIOUS BIT(2) |
5ce4786f | 3008 | |
799a4190 BG |
3009 | static int make_spte(struct kvm_vcpu *vcpu, unsigned int pte_access, int level, |
3010 | gfn_t gfn, kvm_pfn_t pfn, u64 old_spte, bool speculative, | |
3011 | bool can_unsync, bool host_writable, bool ad_disabled, | |
3012 | u64 *new_spte) | |
1c4f1fd6 | 3013 | { |
ffb128c8 | 3014 | u64 spte = 0; |
1e73f9dd | 3015 | int ret = 0; |
64d4d521 | 3016 | |
799a4190 | 3017 | if (ad_disabled) |
6eeb4ef0 | 3018 | spte |= SPTE_AD_DISABLED_MASK; |
1f4e5fc8 PB |
3019 | else if (kvm_vcpu_ad_need_write_protect(vcpu)) |
3020 | spte |= SPTE_AD_WRPROT_ONLY_MASK; | |
ac8d57e5 | 3021 | |
d95c5568 BD |
3022 | /* |
3023 | * For the EPT case, shadow_present_mask is 0 if hardware | |
3024 | * supports exec-only page table entries. In that case, | |
3025 | * ACC_USER_MASK and shadow_user_mask are used to represent | |
3026 | * read access. See FNAME(gpte_access) in paging_tmpl.h. | |
3027 | */ | |
ffb128c8 | 3028 | spte |= shadow_present_mask; |
947da538 | 3029 | if (!speculative) |
ac8d57e5 | 3030 | spte |= spte_shadow_accessed_mask(spte); |
640d9b0d | 3031 | |
3bae0459 | 3032 | if (level > PG_LEVEL_4K && (pte_access & ACC_EXEC_MASK) && |
b8e8c830 PB |
3033 | is_nx_huge_page_enabled()) { |
3034 | pte_access &= ~ACC_EXEC_MASK; | |
3035 | } | |
3036 | ||
7b52345e SY |
3037 | if (pte_access & ACC_EXEC_MASK) |
3038 | spte |= shadow_x_mask; | |
3039 | else | |
3040 | spte |= shadow_nx_mask; | |
49fde340 | 3041 | |
1c4f1fd6 | 3042 | if (pte_access & ACC_USER_MASK) |
7b52345e | 3043 | spte |= shadow_user_mask; |
49fde340 | 3044 | |
3bae0459 | 3045 | if (level > PG_LEVEL_4K) |
05da4558 | 3046 | spte |= PT_PAGE_SIZE_MASK; |
b0bc3ee2 | 3047 | if (tdp_enabled) |
afaf0b2f | 3048 | spte |= kvm_x86_ops.get_mt_mask(vcpu, gfn, |
d1fe9219 | 3049 | kvm_is_mmio_pfn(pfn)); |
1c4f1fd6 | 3050 | |
9bdbba13 | 3051 | if (host_writable) |
1403283a | 3052 | spte |= SPTE_HOST_WRITEABLE; |
f8e453b0 XG |
3053 | else |
3054 | pte_access &= ~ACC_WRITE_MASK; | |
1403283a | 3055 | |
daaf216c TL |
3056 | if (!kvm_is_mmio_pfn(pfn)) |
3057 | spte |= shadow_me_mask; | |
3058 | ||
35149e21 | 3059 | spte |= (u64)pfn << PAGE_SHIFT; |
1c4f1fd6 | 3060 | |
c2288505 | 3061 | if (pte_access & ACC_WRITE_MASK) { |
49fde340 | 3062 | spte |= PT_WRITABLE_MASK | SPTE_MMU_WRITEABLE; |
1c4f1fd6 | 3063 | |
ecc5589f MT |
3064 | /* |
3065 | * Optimization: for pte sync, if spte was writable the hash | |
3066 | * lookup is unnecessary (and expensive). Write protection | |
3067 | * is responsibility of mmu_get_page / kvm_sync_page. | |
3068 | * Same reasoning can be applied to dirty page accounting. | |
3069 | */ | |
799a4190 BG |
3070 | if (!can_unsync && is_writable_pte(old_spte)) |
3071 | goto out; | |
ecc5589f | 3072 | |
4731d4c7 | 3073 | if (mmu_need_write_protect(vcpu, gfn, can_unsync)) { |
9ad17b10 | 3074 | pgprintk("%s: found shadow page for %llx, marking ro\n", |
b8688d51 | 3075 | __func__, gfn); |
5ce4786f | 3076 | ret |= SET_SPTE_WRITE_PROTECTED_PT; |
1c4f1fd6 | 3077 | pte_access &= ~ACC_WRITE_MASK; |
49fde340 | 3078 | spte &= ~(PT_WRITABLE_MASK | SPTE_MMU_WRITEABLE); |
1c4f1fd6 AK |
3079 | } |
3080 | } | |
3081 | ||
799a4190 | 3082 | if (pte_access & ACC_WRITE_MASK) |
ac8d57e5 | 3083 | spte |= spte_shadow_dirty_mask(spte); |
1c4f1fd6 | 3084 | |
f160c7b7 JS |
3085 | if (speculative) |
3086 | spte = mark_spte_for_access_track(spte); | |
3087 | ||
799a4190 BG |
3088 | out: |
3089 | *new_spte = spte; | |
3090 | return ret; | |
3091 | } | |
3092 | ||
3093 | static int set_spte(struct kvm_vcpu *vcpu, u64 *sptep, | |
3094 | unsigned int pte_access, int level, | |
3095 | gfn_t gfn, kvm_pfn_t pfn, bool speculative, | |
3096 | bool can_unsync, bool host_writable) | |
3097 | { | |
3098 | u64 spte; | |
3099 | struct kvm_mmu_page *sp; | |
3100 | int ret; | |
3101 | ||
3102 | if (set_mmio_spte(vcpu, sptep, gfn, pfn, pte_access)) | |
3103 | return 0; | |
3104 | ||
3105 | sp = sptep_to_sp(sptep); | |
3106 | ||
3107 | ret = make_spte(vcpu, pte_access, level, gfn, pfn, *sptep, speculative, | |
3108 | can_unsync, host_writable, sp_ad_disabled(sp), &spte); | |
3109 | ||
3110 | if (spte & PT_WRITABLE_MASK) | |
3111 | kvm_vcpu_mark_page_dirty(vcpu, gfn); | |
3112 | ||
12703759 SC |
3113 | if (*sptep == spte) |
3114 | ret |= SET_SPTE_SPURIOUS; | |
3115 | else if (mmu_spte_update(sptep, spte)) | |
5ce4786f | 3116 | ret |= SET_SPTE_NEED_REMOTE_TLB_FLUSH; |
1e73f9dd MT |
3117 | return ret; |
3118 | } | |
3119 | ||
0a2b64c5 | 3120 | static int mmu_set_spte(struct kvm_vcpu *vcpu, u64 *sptep, |
e88b8093 | 3121 | unsigned int pte_access, bool write_fault, int level, |
0a2b64c5 BG |
3122 | gfn_t gfn, kvm_pfn_t pfn, bool speculative, |
3123 | bool host_writable) | |
1e73f9dd MT |
3124 | { |
3125 | int was_rmapped = 0; | |
53a27b39 | 3126 | int rmap_count; |
5ce4786f | 3127 | int set_spte_ret; |
c4371c2a | 3128 | int ret = RET_PF_FIXED; |
c2a4eadf | 3129 | bool flush = false; |
1e73f9dd | 3130 | |
f7616203 XG |
3131 | pgprintk("%s: spte %llx write_fault %d gfn %llx\n", __func__, |
3132 | *sptep, write_fault, gfn); | |
1e73f9dd | 3133 | |
afd28fe1 | 3134 | if (is_shadow_present_pte(*sptep)) { |
1e73f9dd MT |
3135 | /* |
3136 | * If we overwrite a PTE page pointer with a 2MB PMD, unlink | |
3137 | * the parent of the now unreachable PTE. | |
3138 | */ | |
3bae0459 | 3139 | if (level > PG_LEVEL_4K && !is_large_pte(*sptep)) { |
1e73f9dd | 3140 | struct kvm_mmu_page *child; |
d555c333 | 3141 | u64 pte = *sptep; |
1e73f9dd | 3142 | |
e47c4aee | 3143 | child = to_shadow_page(pte & PT64_BASE_ADDR_MASK); |
bcdd9a93 | 3144 | drop_parent_pte(child, sptep); |
c2a4eadf | 3145 | flush = true; |
d555c333 | 3146 | } else if (pfn != spte_to_pfn(*sptep)) { |
9ad17b10 | 3147 | pgprintk("hfn old %llx new %llx\n", |
d555c333 | 3148 | spte_to_pfn(*sptep), pfn); |
c3707958 | 3149 | drop_spte(vcpu->kvm, sptep); |
c2a4eadf | 3150 | flush = true; |
6bed6b9e JR |
3151 | } else |
3152 | was_rmapped = 1; | |
1e73f9dd | 3153 | } |
852e3c19 | 3154 | |
5ce4786f JS |
3155 | set_spte_ret = set_spte(vcpu, sptep, pte_access, level, gfn, pfn, |
3156 | speculative, true, host_writable); | |
3157 | if (set_spte_ret & SET_SPTE_WRITE_PROTECTED_PT) { | |
1e73f9dd | 3158 | if (write_fault) |
9b8ebbdb | 3159 | ret = RET_PF_EMULATE; |
8c8560b8 | 3160 | kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu); |
a378b4e6 | 3161 | } |
c3134ce2 | 3162 | |
c2a4eadf | 3163 | if (set_spte_ret & SET_SPTE_NEED_REMOTE_TLB_FLUSH || flush) |
c3134ce2 LT |
3164 | kvm_flush_remote_tlbs_with_address(vcpu->kvm, gfn, |
3165 | KVM_PAGES_PER_HPAGE(level)); | |
1e73f9dd | 3166 | |
029499b4 | 3167 | if (unlikely(is_mmio_spte(*sptep))) |
9b8ebbdb | 3168 | ret = RET_PF_EMULATE; |
ce88decf | 3169 | |
12703759 SC |
3170 | /* |
3171 | * The fault is fully spurious if and only if the new SPTE and old SPTE | |
3172 | * are identical, and emulation is not required. | |
3173 | */ | |
3174 | if ((set_spte_ret & SET_SPTE_SPURIOUS) && ret == RET_PF_FIXED) { | |
3175 | WARN_ON_ONCE(!was_rmapped); | |
3176 | return RET_PF_SPURIOUS; | |
3177 | } | |
3178 | ||
d555c333 | 3179 | pgprintk("%s: setting spte %llx\n", __func__, *sptep); |
335e192a | 3180 | trace_kvm_mmu_set_spte(level, gfn, sptep); |
d555c333 | 3181 | if (!was_rmapped && is_large_pte(*sptep)) |
05da4558 MT |
3182 | ++vcpu->kvm->stat.lpages; |
3183 | ||
ffb61bb3 | 3184 | if (is_shadow_present_pte(*sptep)) { |
ffb61bb3 XG |
3185 | if (!was_rmapped) { |
3186 | rmap_count = rmap_add(vcpu, sptep, gfn); | |
3187 | if (rmap_count > RMAP_RECYCLE_THRESHOLD) | |
3188 | rmap_recycle(vcpu, sptep, gfn); | |
3189 | } | |
1c4f1fd6 | 3190 | } |
cb9aaa30 | 3191 | |
9b8ebbdb | 3192 | return ret; |
1c4f1fd6 AK |
3193 | } |
3194 | ||
ba049e93 | 3195 | static kvm_pfn_t pte_prefetch_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn, |
957ed9ef XG |
3196 | bool no_dirty_log) |
3197 | { | |
3198 | struct kvm_memory_slot *slot; | |
957ed9ef | 3199 | |
5d163b1c | 3200 | slot = gfn_to_memslot_dirty_bitmap(vcpu, gfn, no_dirty_log); |
903816fa | 3201 | if (!slot) |
6c8ee57b | 3202 | return KVM_PFN_ERR_FAULT; |
957ed9ef | 3203 | |
037d92dc | 3204 | return gfn_to_pfn_memslot_atomic(slot, gfn); |
957ed9ef XG |
3205 | } |
3206 | ||
3207 | static int direct_pte_prefetch_many(struct kvm_vcpu *vcpu, | |
3208 | struct kvm_mmu_page *sp, | |
3209 | u64 *start, u64 *end) | |
3210 | { | |
3211 | struct page *pages[PTE_PREFETCH_NUM]; | |
d9ef13c2 | 3212 | struct kvm_memory_slot *slot; |
0a2b64c5 | 3213 | unsigned int access = sp->role.access; |
957ed9ef XG |
3214 | int i, ret; |
3215 | gfn_t gfn; | |
3216 | ||
3217 | gfn = kvm_mmu_page_get_gfn(sp, start - sp->spt); | |
d9ef13c2 PB |
3218 | slot = gfn_to_memslot_dirty_bitmap(vcpu, gfn, access & ACC_WRITE_MASK); |
3219 | if (!slot) | |
957ed9ef XG |
3220 | return -1; |
3221 | ||
d9ef13c2 | 3222 | ret = gfn_to_page_many_atomic(slot, gfn, pages, end - start); |
957ed9ef XG |
3223 | if (ret <= 0) |
3224 | return -1; | |
3225 | ||
43fdcda9 | 3226 | for (i = 0; i < ret; i++, gfn++, start++) { |
e88b8093 | 3227 | mmu_set_spte(vcpu, start, access, false, sp->role.level, gfn, |
029499b4 | 3228 | page_to_pfn(pages[i]), true, true); |
43fdcda9 JS |
3229 | put_page(pages[i]); |
3230 | } | |
957ed9ef XG |
3231 | |
3232 | return 0; | |
3233 | } | |
3234 | ||
3235 | static void __direct_pte_prefetch(struct kvm_vcpu *vcpu, | |
3236 | struct kvm_mmu_page *sp, u64 *sptep) | |
3237 | { | |
3238 | u64 *spte, *start = NULL; | |
3239 | int i; | |
3240 | ||
3241 | WARN_ON(!sp->role.direct); | |
3242 | ||
3243 | i = (sptep - sp->spt) & ~(PTE_PREFETCH_NUM - 1); | |
3244 | spte = sp->spt + i; | |
3245 | ||
3246 | for (i = 0; i < PTE_PREFETCH_NUM; i++, spte++) { | |
c3707958 | 3247 | if (is_shadow_present_pte(*spte) || spte == sptep) { |
957ed9ef XG |
3248 | if (!start) |
3249 | continue; | |
3250 | if (direct_pte_prefetch_many(vcpu, sp, start, spte) < 0) | |
3251 | break; | |
3252 | start = NULL; | |
3253 | } else if (!start) | |
3254 | start = spte; | |
3255 | } | |
3256 | } | |
3257 | ||
3258 | static void direct_pte_prefetch(struct kvm_vcpu *vcpu, u64 *sptep) | |
3259 | { | |
3260 | struct kvm_mmu_page *sp; | |
3261 | ||
57354682 | 3262 | sp = sptep_to_sp(sptep); |
ac8d57e5 | 3263 | |
957ed9ef | 3264 | /* |
ac8d57e5 PF |
3265 | * Without accessed bits, there's no way to distinguish between |
3266 | * actually accessed translations and prefetched, so disable pte | |
3267 | * prefetch if accessed bits aren't available. | |
957ed9ef | 3268 | */ |
ac8d57e5 | 3269 | if (sp_ad_disabled(sp)) |
957ed9ef XG |
3270 | return; |
3271 | ||
3bae0459 | 3272 | if (sp->role.level > PG_LEVEL_4K) |
957ed9ef XG |
3273 | return; |
3274 | ||
3275 | __direct_pte_prefetch(vcpu, sp, sptep); | |
3276 | } | |
3277 | ||
db543216 | 3278 | static int host_pfn_mapping_level(struct kvm_vcpu *vcpu, gfn_t gfn, |
293e306e | 3279 | kvm_pfn_t pfn, struct kvm_memory_slot *slot) |
db543216 | 3280 | { |
db543216 SC |
3281 | unsigned long hva; |
3282 | pte_t *pte; | |
3283 | int level; | |
3284 | ||
e851265a | 3285 | if (!PageCompound(pfn_to_page(pfn)) && !kvm_is_zone_device_pfn(pfn)) |
3bae0459 | 3286 | return PG_LEVEL_4K; |
db543216 | 3287 | |
293e306e SC |
3288 | /* |
3289 | * Note, using the already-retrieved memslot and __gfn_to_hva_memslot() | |
3290 | * is not solely for performance, it's also necessary to avoid the | |
3291 | * "writable" check in __gfn_to_hva_many(), which will always fail on | |
3292 | * read-only memslots due to gfn_to_hva() assuming writes. Earlier | |
3293 | * page fault steps have already verified the guest isn't writing a | |
3294 | * read-only memslot. | |
3295 | */ | |
db543216 SC |
3296 | hva = __gfn_to_hva_memslot(slot, gfn); |
3297 | ||
3298 | pte = lookup_address_in_mm(vcpu->kvm->mm, hva, &level); | |
3299 | if (unlikely(!pte)) | |
3bae0459 | 3300 | return PG_LEVEL_4K; |
db543216 SC |
3301 | |
3302 | return level; | |
3303 | } | |
3304 | ||
83f06fa7 | 3305 | static int kvm_mmu_hugepage_adjust(struct kvm_vcpu *vcpu, gfn_t gfn, |
3cf06612 SC |
3306 | int max_level, kvm_pfn_t *pfnp, |
3307 | bool huge_page_disallowed, int *req_level) | |
0885904d | 3308 | { |
293e306e | 3309 | struct kvm_memory_slot *slot; |
2c0629f4 | 3310 | struct kvm_lpage_info *linfo; |
0885904d | 3311 | kvm_pfn_t pfn = *pfnp; |
17eff019 | 3312 | kvm_pfn_t mask; |
83f06fa7 | 3313 | int level; |
17eff019 | 3314 | |
3cf06612 SC |
3315 | *req_level = PG_LEVEL_4K; |
3316 | ||
3bae0459 SC |
3317 | if (unlikely(max_level == PG_LEVEL_4K)) |
3318 | return PG_LEVEL_4K; | |
17eff019 | 3319 | |
e851265a | 3320 | if (is_error_noslot_pfn(pfn) || kvm_is_reserved_pfn(pfn)) |
3bae0459 | 3321 | return PG_LEVEL_4K; |
17eff019 | 3322 | |
293e306e SC |
3323 | slot = gfn_to_memslot_dirty_bitmap(vcpu, gfn, true); |
3324 | if (!slot) | |
3bae0459 | 3325 | return PG_LEVEL_4K; |
293e306e | 3326 | |
1d92d2e8 | 3327 | max_level = min(max_level, max_huge_page_level); |
3bae0459 | 3328 | for ( ; max_level > PG_LEVEL_4K; max_level--) { |
2c0629f4 SC |
3329 | linfo = lpage_info_slot(gfn, slot, max_level); |
3330 | if (!linfo->disallow_lpage) | |
293e306e SC |
3331 | break; |
3332 | } | |
3333 | ||
3bae0459 SC |
3334 | if (max_level == PG_LEVEL_4K) |
3335 | return PG_LEVEL_4K; | |
293e306e SC |
3336 | |
3337 | level = host_pfn_mapping_level(vcpu, gfn, pfn, slot); | |
3bae0459 | 3338 | if (level == PG_LEVEL_4K) |
83f06fa7 | 3339 | return level; |
17eff019 | 3340 | |
3cf06612 SC |
3341 | *req_level = level = min(level, max_level); |
3342 | ||
3343 | /* | |
3344 | * Enforce the iTLB multihit workaround after capturing the requested | |
3345 | * level, which will be used to do precise, accurate accounting. | |
3346 | */ | |
3347 | if (huge_page_disallowed) | |
3348 | return PG_LEVEL_4K; | |
0885904d SC |
3349 | |
3350 | /* | |
17eff019 SC |
3351 | * mmu_notifier_retry() was successful and mmu_lock is held, so |
3352 | * the pmd can't be split from under us. | |
0885904d | 3353 | */ |
17eff019 SC |
3354 | mask = KVM_PAGES_PER_HPAGE(level) - 1; |
3355 | VM_BUG_ON((gfn & mask) != (pfn & mask)); | |
3356 | *pfnp = pfn & ~mask; | |
83f06fa7 SC |
3357 | |
3358 | return level; | |
0885904d SC |
3359 | } |
3360 | ||
b8e8c830 PB |
3361 | static void disallowed_hugepage_adjust(struct kvm_shadow_walk_iterator it, |
3362 | gfn_t gfn, kvm_pfn_t *pfnp, int *levelp) | |
3363 | { | |
3364 | int level = *levelp; | |
3365 | u64 spte = *it.sptep; | |
3366 | ||
3bae0459 | 3367 | if (it.level == level && level > PG_LEVEL_4K && |
b8e8c830 PB |
3368 | is_shadow_present_pte(spte) && |
3369 | !is_large_pte(spte)) { | |
3370 | /* | |
3371 | * A small SPTE exists for this pfn, but FNAME(fetch) | |
3372 | * and __direct_map would like to create a large PTE | |
3373 | * instead: just force them to go down another level, | |
3374 | * patching back for them into pfn the next 9 bits of | |
3375 | * the address. | |
3376 | */ | |
3377 | u64 page_mask = KVM_PAGES_PER_HPAGE(level) - KVM_PAGES_PER_HPAGE(level - 1); | |
3378 | *pfnp |= gfn & page_mask; | |
3379 | (*levelp)--; | |
3380 | } | |
3381 | } | |
3382 | ||
6c2fd34f | 3383 | static int __direct_map(struct kvm_vcpu *vcpu, gpa_t gpa, u32 error_code, |
83f06fa7 | 3384 | int map_writable, int max_level, kvm_pfn_t pfn, |
6c2fd34f | 3385 | bool prefault, bool is_tdp) |
140754bc | 3386 | { |
6c2fd34f SC |
3387 | bool nx_huge_page_workaround_enabled = is_nx_huge_page_enabled(); |
3388 | bool write = error_code & PFERR_WRITE_MASK; | |
3389 | bool exec = error_code & PFERR_FETCH_MASK; | |
3390 | bool huge_page_disallowed = exec && nx_huge_page_workaround_enabled; | |
3fcf2d1b | 3391 | struct kvm_shadow_walk_iterator it; |
140754bc | 3392 | struct kvm_mmu_page *sp; |
3cf06612 | 3393 | int level, req_level, ret; |
3fcf2d1b PB |
3394 | gfn_t gfn = gpa >> PAGE_SHIFT; |
3395 | gfn_t base_gfn = gfn; | |
6aa8b732 | 3396 | |
0c7a98e3 | 3397 | if (WARN_ON(!VALID_PAGE(vcpu->arch.mmu->root_hpa))) |
3fcf2d1b | 3398 | return RET_PF_RETRY; |
989c6b34 | 3399 | |
3cf06612 SC |
3400 | level = kvm_mmu_hugepage_adjust(vcpu, gfn, max_level, &pfn, |
3401 | huge_page_disallowed, &req_level); | |
4cd071d1 | 3402 | |
335e192a | 3403 | trace_kvm_mmu_spte_requested(gpa, level, pfn); |
3fcf2d1b | 3404 | for_each_shadow_entry(vcpu, gpa, it) { |
b8e8c830 PB |
3405 | /* |
3406 | * We cannot overwrite existing page tables with an NX | |
3407 | * large page, as the leaf could be executable. | |
3408 | */ | |
dcc70651 SC |
3409 | if (nx_huge_page_workaround_enabled) |
3410 | disallowed_hugepage_adjust(it, gfn, &pfn, &level); | |
b8e8c830 | 3411 | |
3fcf2d1b PB |
3412 | base_gfn = gfn & ~(KVM_PAGES_PER_HPAGE(it.level) - 1); |
3413 | if (it.level == level) | |
9f652d21 | 3414 | break; |
6aa8b732 | 3415 | |
3fcf2d1b PB |
3416 | drop_large_spte(vcpu, it.sptep); |
3417 | if (!is_shadow_present_pte(*it.sptep)) { | |
3418 | sp = kvm_mmu_get_page(vcpu, base_gfn, it.addr, | |
3419 | it.level - 1, true, ACC_ALL); | |
c9fa0b3b | 3420 | |
3fcf2d1b | 3421 | link_shadow_page(vcpu, it.sptep, sp); |
5bcaf3e1 SC |
3422 | if (is_tdp && huge_page_disallowed && |
3423 | req_level >= it.level) | |
b8e8c830 | 3424 | account_huge_nx_page(vcpu->kvm, sp); |
9f652d21 AK |
3425 | } |
3426 | } | |
3fcf2d1b PB |
3427 | |
3428 | ret = mmu_set_spte(vcpu, it.sptep, ACC_ALL, | |
3429 | write, level, base_gfn, pfn, prefault, | |
3430 | map_writable); | |
12703759 SC |
3431 | if (ret == RET_PF_SPURIOUS) |
3432 | return ret; | |
3433 | ||
3fcf2d1b PB |
3434 | direct_pte_prefetch(vcpu, it.sptep); |
3435 | ++vcpu->stat.pf_fixed; | |
3436 | return ret; | |
6aa8b732 AK |
3437 | } |
3438 | ||
77db5cbd | 3439 | static void kvm_send_hwpoison_signal(unsigned long address, struct task_struct *tsk) |
bf998156 | 3440 | { |
585a8b9b | 3441 | send_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, PAGE_SHIFT, tsk); |
bf998156 HY |
3442 | } |
3443 | ||
ba049e93 | 3444 | static int kvm_handle_bad_page(struct kvm_vcpu *vcpu, gfn_t gfn, kvm_pfn_t pfn) |
bf998156 | 3445 | { |
4d8b81ab XG |
3446 | /* |
3447 | * Do not cache the mmio info caused by writing the readonly gfn | |
3448 | * into the spte otherwise read access on readonly gfn also can | |
3449 | * caused mmio page fault and treat it as mmio access. | |
4d8b81ab XG |
3450 | */ |
3451 | if (pfn == KVM_PFN_ERR_RO_FAULT) | |
9b8ebbdb | 3452 | return RET_PF_EMULATE; |
4d8b81ab | 3453 | |
e6c1502b | 3454 | if (pfn == KVM_PFN_ERR_HWPOISON) { |
54bf36aa | 3455 | kvm_send_hwpoison_signal(kvm_vcpu_gfn_to_hva(vcpu, gfn), current); |
9b8ebbdb | 3456 | return RET_PF_RETRY; |
d7c55201 | 3457 | } |
edba23e5 | 3458 | |
2c151b25 | 3459 | return -EFAULT; |
bf998156 HY |
3460 | } |
3461 | ||
d7c55201 | 3462 | static bool handle_abnormal_pfn(struct kvm_vcpu *vcpu, gva_t gva, gfn_t gfn, |
0a2b64c5 BG |
3463 | kvm_pfn_t pfn, unsigned int access, |
3464 | int *ret_val) | |
d7c55201 | 3465 | { |
d7c55201 | 3466 | /* The pfn is invalid, report the error! */ |
81c52c56 | 3467 | if (unlikely(is_error_pfn(pfn))) { |
d7c55201 | 3468 | *ret_val = kvm_handle_bad_page(vcpu, gfn, pfn); |
798e88b3 | 3469 | return true; |
d7c55201 XG |
3470 | } |
3471 | ||
ce88decf | 3472 | if (unlikely(is_noslot_pfn(pfn))) |
4af77151 SC |
3473 | vcpu_cache_mmio_info(vcpu, gva, gfn, |
3474 | access & shadow_mmio_access_mask); | |
d7c55201 | 3475 | |
798e88b3 | 3476 | return false; |
d7c55201 XG |
3477 | } |
3478 | ||
e5552fd2 | 3479 | static bool page_fault_can_be_fast(u32 error_code) |
c7ba5b48 | 3480 | { |
1c118b82 XG |
3481 | /* |
3482 | * Do not fix the mmio spte with invalid generation number which | |
3483 | * need to be updated by slow page fault path. | |
3484 | */ | |
3485 | if (unlikely(error_code & PFERR_RSVD_MASK)) | |
3486 | return false; | |
3487 | ||
f160c7b7 JS |
3488 | /* See if the page fault is due to an NX violation */ |
3489 | if (unlikely(((error_code & (PFERR_FETCH_MASK | PFERR_PRESENT_MASK)) | |
3490 | == (PFERR_FETCH_MASK | PFERR_PRESENT_MASK)))) | |
3491 | return false; | |
3492 | ||
c7ba5b48 | 3493 | /* |
f160c7b7 JS |
3494 | * #PF can be fast if: |
3495 | * 1. The shadow page table entry is not present, which could mean that | |
3496 | * the fault is potentially caused by access tracking (if enabled). | |
3497 | * 2. The shadow page table entry is present and the fault | |
3498 | * is caused by write-protect, that means we just need change the W | |
3499 | * bit of the spte which can be done out of mmu-lock. | |
3500 | * | |
3501 | * However, if access tracking is disabled we know that a non-present | |
3502 | * page must be a genuine page fault where we have to create a new SPTE. | |
3503 | * So, if access tracking is disabled, we return true only for write | |
3504 | * accesses to a present page. | |
c7ba5b48 | 3505 | */ |
c7ba5b48 | 3506 | |
f160c7b7 JS |
3507 | return shadow_acc_track_mask != 0 || |
3508 | ((error_code & (PFERR_WRITE_MASK | PFERR_PRESENT_MASK)) | |
3509 | == (PFERR_WRITE_MASK | PFERR_PRESENT_MASK)); | |
c7ba5b48 XG |
3510 | } |
3511 | ||
97dceba2 JS |
3512 | /* |
3513 | * Returns true if the SPTE was fixed successfully. Otherwise, | |
3514 | * someone else modified the SPTE from its original value. | |
3515 | */ | |
c7ba5b48 | 3516 | static bool |
92a476cb | 3517 | fast_pf_fix_direct_spte(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp, |
d3e328f2 | 3518 | u64 *sptep, u64 old_spte, u64 new_spte) |
c7ba5b48 | 3519 | { |
c7ba5b48 XG |
3520 | gfn_t gfn; |
3521 | ||
3522 | WARN_ON(!sp->role.direct); | |
3523 | ||
9b51a630 KH |
3524 | /* |
3525 | * Theoretically we could also set dirty bit (and flush TLB) here in | |
3526 | * order to eliminate unnecessary PML logging. See comments in | |
3527 | * set_spte. But fast_page_fault is very unlikely to happen with PML | |
3528 | * enabled, so we do not do this. This might result in the same GPA | |
3529 | * to be logged in PML buffer again when the write really happens, and | |
3530 | * eventually to be called by mark_page_dirty twice. But it's also no | |
3531 | * harm. This also avoids the TLB flush needed after setting dirty bit | |
3532 | * so non-PML cases won't be impacted. | |
3533 | * | |
3534 | * Compare with set_spte where instead shadow_dirty_mask is set. | |
3535 | */ | |
f160c7b7 | 3536 | if (cmpxchg64(sptep, old_spte, new_spte) != old_spte) |
97dceba2 JS |
3537 | return false; |
3538 | ||
d3e328f2 | 3539 | if (is_writable_pte(new_spte) && !is_writable_pte(old_spte)) { |
f160c7b7 JS |
3540 | /* |
3541 | * The gfn of direct spte is stable since it is | |
3542 | * calculated by sp->gfn. | |
3543 | */ | |
3544 | gfn = kvm_mmu_page_get_gfn(sp, sptep - sp->spt); | |
3545 | kvm_vcpu_mark_page_dirty(vcpu, gfn); | |
3546 | } | |
c7ba5b48 XG |
3547 | |
3548 | return true; | |
3549 | } | |
3550 | ||
d3e328f2 JS |
3551 | static bool is_access_allowed(u32 fault_err_code, u64 spte) |
3552 | { | |
3553 | if (fault_err_code & PFERR_FETCH_MASK) | |
3554 | return is_executable_pte(spte); | |
3555 | ||
3556 | if (fault_err_code & PFERR_WRITE_MASK) | |
3557 | return is_writable_pte(spte); | |
3558 | ||
3559 | /* Fault was on Read access */ | |
3560 | return spte & PT_PRESENT_MASK; | |
3561 | } | |
3562 | ||
c7ba5b48 | 3563 | /* |
c4371c2a | 3564 | * Returns one of RET_PF_INVALID, RET_PF_FIXED or RET_PF_SPURIOUS. |
c7ba5b48 | 3565 | */ |
c4371c2a SC |
3566 | static int fast_page_fault(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa, |
3567 | u32 error_code) | |
c7ba5b48 XG |
3568 | { |
3569 | struct kvm_shadow_walk_iterator iterator; | |
92a476cb | 3570 | struct kvm_mmu_page *sp; |
c4371c2a | 3571 | int ret = RET_PF_INVALID; |
c7ba5b48 | 3572 | u64 spte = 0ull; |
97dceba2 | 3573 | uint retry_count = 0; |
c7ba5b48 | 3574 | |
e5552fd2 | 3575 | if (!page_fault_can_be_fast(error_code)) |
c4371c2a | 3576 | return ret; |
c7ba5b48 XG |
3577 | |
3578 | walk_shadow_page_lockless_begin(vcpu); | |
c7ba5b48 | 3579 | |
97dceba2 | 3580 | do { |
d3e328f2 | 3581 | u64 new_spte; |
c7ba5b48 | 3582 | |
736c291c | 3583 | for_each_shadow_entry_lockless(vcpu, cr2_or_gpa, iterator, spte) |
f9fa2509 | 3584 | if (!is_shadow_present_pte(spte)) |
d162f30a JS |
3585 | break; |
3586 | ||
57354682 | 3587 | sp = sptep_to_sp(iterator.sptep); |
97dceba2 JS |
3588 | if (!is_last_spte(spte, sp->role.level)) |
3589 | break; | |
c7ba5b48 | 3590 | |
97dceba2 | 3591 | /* |
f160c7b7 JS |
3592 | * Check whether the memory access that caused the fault would |
3593 | * still cause it if it were to be performed right now. If not, | |
3594 | * then this is a spurious fault caused by TLB lazily flushed, | |
3595 | * or some other CPU has already fixed the PTE after the | |
3596 | * current CPU took the fault. | |
97dceba2 JS |
3597 | * |
3598 | * Need not check the access of upper level table entries since | |
3599 | * they are always ACC_ALL. | |
3600 | */ | |
d3e328f2 | 3601 | if (is_access_allowed(error_code, spte)) { |
c4371c2a | 3602 | ret = RET_PF_SPURIOUS; |
d3e328f2 JS |
3603 | break; |
3604 | } | |
f160c7b7 | 3605 | |
d3e328f2 JS |
3606 | new_spte = spte; |
3607 | ||
3608 | if (is_access_track_spte(spte)) | |
3609 | new_spte = restore_acc_track_spte(new_spte); | |
3610 | ||
3611 | /* | |
3612 | * Currently, to simplify the code, write-protection can | |
3613 | * be removed in the fast path only if the SPTE was | |
3614 | * write-protected for dirty-logging or access tracking. | |
3615 | */ | |
3616 | if ((error_code & PFERR_WRITE_MASK) && | |
e6302698 | 3617 | spte_can_locklessly_be_made_writable(spte)) { |
d3e328f2 | 3618 | new_spte |= PT_WRITABLE_MASK; |
f160c7b7 JS |
3619 | |
3620 | /* | |
d3e328f2 JS |
3621 | * Do not fix write-permission on the large spte. Since |
3622 | * we only dirty the first page into the dirty-bitmap in | |
3623 | * fast_pf_fix_direct_spte(), other pages are missed | |
3624 | * if its slot has dirty logging enabled. | |
3625 | * | |
3626 | * Instead, we let the slow page fault path create a | |
3627 | * normal spte to fix the access. | |
3628 | * | |
3629 | * See the comments in kvm_arch_commit_memory_region(). | |
f160c7b7 | 3630 | */ |
3bae0459 | 3631 | if (sp->role.level > PG_LEVEL_4K) |
f160c7b7 | 3632 | break; |
97dceba2 | 3633 | } |
c7ba5b48 | 3634 | |
f160c7b7 | 3635 | /* Verify that the fault can be handled in the fast path */ |
d3e328f2 JS |
3636 | if (new_spte == spte || |
3637 | !is_access_allowed(error_code, new_spte)) | |
97dceba2 JS |
3638 | break; |
3639 | ||
3640 | /* | |
3641 | * Currently, fast page fault only works for direct mapping | |
3642 | * since the gfn is not stable for indirect shadow page. See | |
3ecad8c2 | 3643 | * Documentation/virt/kvm/locking.rst to get more detail. |
97dceba2 | 3644 | */ |
c4371c2a SC |
3645 | if (fast_pf_fix_direct_spte(vcpu, sp, iterator.sptep, spte, |
3646 | new_spte)) { | |
3647 | ret = RET_PF_FIXED; | |
97dceba2 | 3648 | break; |
c4371c2a | 3649 | } |
97dceba2 JS |
3650 | |
3651 | if (++retry_count > 4) { | |
3652 | printk_once(KERN_WARNING | |
3653 | "kvm: Fast #PF retrying more than 4 times.\n"); | |
3654 | break; | |
3655 | } | |
3656 | ||
97dceba2 | 3657 | } while (true); |
c126d94f | 3658 | |
736c291c | 3659 | trace_fast_page_fault(vcpu, cr2_or_gpa, error_code, iterator.sptep, |
c4371c2a | 3660 | spte, ret); |
c7ba5b48 XG |
3661 | walk_shadow_page_lockless_end(vcpu); |
3662 | ||
c4371c2a | 3663 | return ret; |
c7ba5b48 XG |
3664 | } |
3665 | ||
74b566e6 JS |
3666 | static void mmu_free_root_page(struct kvm *kvm, hpa_t *root_hpa, |
3667 | struct list_head *invalid_list) | |
17ac10ad | 3668 | { |
4db35314 | 3669 | struct kvm_mmu_page *sp; |
17ac10ad | 3670 | |
74b566e6 | 3671 | if (!VALID_PAGE(*root_hpa)) |
7b53aa56 | 3672 | return; |
35af577a | 3673 | |
e47c4aee | 3674 | sp = to_shadow_page(*root_hpa & PT64_BASE_ADDR_MASK); |
74b566e6 JS |
3675 | --sp->root_count; |
3676 | if (!sp->root_count && sp->role.invalid) | |
3677 | kvm_mmu_prepare_zap_page(kvm, sp, invalid_list); | |
17ac10ad | 3678 | |
74b566e6 JS |
3679 | *root_hpa = INVALID_PAGE; |
3680 | } | |
3681 | ||
08fb59d8 | 3682 | /* roots_to_free must be some combination of the KVM_MMU_ROOT_* flags */ |
6a82cd1c VK |
3683 | void kvm_mmu_free_roots(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, |
3684 | ulong roots_to_free) | |
74b566e6 | 3685 | { |
4d710de9 | 3686 | struct kvm *kvm = vcpu->kvm; |
74b566e6 JS |
3687 | int i; |
3688 | LIST_HEAD(invalid_list); | |
08fb59d8 | 3689 | bool free_active_root = roots_to_free & KVM_MMU_ROOT_CURRENT; |
74b566e6 | 3690 | |
b94742c9 | 3691 | BUILD_BUG_ON(KVM_MMU_NUM_PREV_ROOTS >= BITS_PER_LONG); |
74b566e6 | 3692 | |
08fb59d8 | 3693 | /* Before acquiring the MMU lock, see if we need to do any real work. */ |
b94742c9 JS |
3694 | if (!(free_active_root && VALID_PAGE(mmu->root_hpa))) { |
3695 | for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) | |
3696 | if ((roots_to_free & KVM_MMU_ROOT_PREVIOUS(i)) && | |
3697 | VALID_PAGE(mmu->prev_roots[i].hpa)) | |
3698 | break; | |
3699 | ||
3700 | if (i == KVM_MMU_NUM_PREV_ROOTS) | |
3701 | return; | |
3702 | } | |
35af577a | 3703 | |
4d710de9 | 3704 | spin_lock(&kvm->mmu_lock); |
17ac10ad | 3705 | |
b94742c9 JS |
3706 | for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) |
3707 | if (roots_to_free & KVM_MMU_ROOT_PREVIOUS(i)) | |
4d710de9 | 3708 | mmu_free_root_page(kvm, &mmu->prev_roots[i].hpa, |
b94742c9 | 3709 | &invalid_list); |
7c390d35 | 3710 | |
08fb59d8 JS |
3711 | if (free_active_root) { |
3712 | if (mmu->shadow_root_level >= PT64_ROOT_4LEVEL && | |
3713 | (mmu->root_level >= PT64_ROOT_4LEVEL || mmu->direct_map)) { | |
4d710de9 | 3714 | mmu_free_root_page(kvm, &mmu->root_hpa, &invalid_list); |
08fb59d8 JS |
3715 | } else { |
3716 | for (i = 0; i < 4; ++i) | |
3717 | if (mmu->pae_root[i] != 0) | |
4d710de9 | 3718 | mmu_free_root_page(kvm, |
08fb59d8 JS |
3719 | &mmu->pae_root[i], |
3720 | &invalid_list); | |
3721 | mmu->root_hpa = INVALID_PAGE; | |
3722 | } | |
be01e8e2 | 3723 | mmu->root_pgd = 0; |
17ac10ad | 3724 | } |
74b566e6 | 3725 | |
4d710de9 SC |
3726 | kvm_mmu_commit_zap_page(kvm, &invalid_list); |
3727 | spin_unlock(&kvm->mmu_lock); | |
17ac10ad | 3728 | } |
74b566e6 | 3729 | EXPORT_SYMBOL_GPL(kvm_mmu_free_roots); |
17ac10ad | 3730 | |
8986ecc0 MT |
3731 | static int mmu_check_root(struct kvm_vcpu *vcpu, gfn_t root_gfn) |
3732 | { | |
3733 | int ret = 0; | |
3734 | ||
995decb6 | 3735 | if (!kvm_vcpu_is_visible_gfn(vcpu, root_gfn)) { |
a8eeb04a | 3736 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); |
8986ecc0 MT |
3737 | ret = 1; |
3738 | } | |
3739 | ||
3740 | return ret; | |
3741 | } | |
3742 | ||
8123f265 SC |
3743 | static hpa_t mmu_alloc_root(struct kvm_vcpu *vcpu, gfn_t gfn, gva_t gva, |
3744 | u8 level, bool direct) | |
651dd37a JR |
3745 | { |
3746 | struct kvm_mmu_page *sp; | |
8123f265 SC |
3747 | |
3748 | spin_lock(&vcpu->kvm->mmu_lock); | |
3749 | ||
3750 | if (make_mmu_pages_available(vcpu)) { | |
3751 | spin_unlock(&vcpu->kvm->mmu_lock); | |
3752 | return INVALID_PAGE; | |
3753 | } | |
3754 | sp = kvm_mmu_get_page(vcpu, gfn, gva, level, direct, ACC_ALL); | |
3755 | ++sp->root_count; | |
3756 | ||
3757 | spin_unlock(&vcpu->kvm->mmu_lock); | |
3758 | return __pa(sp->spt); | |
3759 | } | |
3760 | ||
3761 | static int mmu_alloc_direct_roots(struct kvm_vcpu *vcpu) | |
3762 | { | |
3763 | u8 shadow_root_level = vcpu->arch.mmu->shadow_root_level; | |
3764 | hpa_t root; | |
7ebaf15e | 3765 | unsigned i; |
651dd37a | 3766 | |
8123f265 SC |
3767 | if (shadow_root_level >= PT64_ROOT_4LEVEL) { |
3768 | root = mmu_alloc_root(vcpu, 0, 0, shadow_root_level, true); | |
3769 | if (!VALID_PAGE(root)) | |
ed52870f | 3770 | return -ENOSPC; |
8123f265 SC |
3771 | vcpu->arch.mmu->root_hpa = root; |
3772 | } else if (shadow_root_level == PT32E_ROOT_LEVEL) { | |
651dd37a | 3773 | for (i = 0; i < 4; ++i) { |
8123f265 | 3774 | MMU_WARN_ON(VALID_PAGE(vcpu->arch.mmu->pae_root[i])); |
651dd37a | 3775 | |
8123f265 SC |
3776 | root = mmu_alloc_root(vcpu, i << (30 - PAGE_SHIFT), |
3777 | i << 30, PT32_ROOT_LEVEL, true); | |
3778 | if (!VALID_PAGE(root)) | |
ed52870f | 3779 | return -ENOSPC; |
44dd3ffa | 3780 | vcpu->arch.mmu->pae_root[i] = root | PT_PRESENT_MASK; |
651dd37a | 3781 | } |
44dd3ffa | 3782 | vcpu->arch.mmu->root_hpa = __pa(vcpu->arch.mmu->pae_root); |
651dd37a JR |
3783 | } else |
3784 | BUG(); | |
3651c7fc | 3785 | |
be01e8e2 SC |
3786 | /* root_pgd is ignored for direct MMUs. */ |
3787 | vcpu->arch.mmu->root_pgd = 0; | |
651dd37a JR |
3788 | |
3789 | return 0; | |
3790 | } | |
3791 | ||
3792 | static int mmu_alloc_shadow_roots(struct kvm_vcpu *vcpu) | |
17ac10ad | 3793 | { |
81407ca5 | 3794 | u64 pdptr, pm_mask; |
be01e8e2 | 3795 | gfn_t root_gfn, root_pgd; |
8123f265 | 3796 | hpa_t root; |
81407ca5 | 3797 | int i; |
3bb65a22 | 3798 | |
be01e8e2 SC |
3799 | root_pgd = vcpu->arch.mmu->get_guest_pgd(vcpu); |
3800 | root_gfn = root_pgd >> PAGE_SHIFT; | |
17ac10ad | 3801 | |
651dd37a JR |
3802 | if (mmu_check_root(vcpu, root_gfn)) |
3803 | return 1; | |
3804 | ||
3805 | /* | |
3806 | * Do we shadow a long mode page table? If so we need to | |
3807 | * write-protect the guests page table root. | |
3808 | */ | |
44dd3ffa | 3809 | if (vcpu->arch.mmu->root_level >= PT64_ROOT_4LEVEL) { |
8123f265 | 3810 | MMU_WARN_ON(VALID_PAGE(vcpu->arch.mmu->root_hpa)); |
651dd37a | 3811 | |
8123f265 SC |
3812 | root = mmu_alloc_root(vcpu, root_gfn, 0, |
3813 | vcpu->arch.mmu->shadow_root_level, false); | |
3814 | if (!VALID_PAGE(root)) | |
ed52870f | 3815 | return -ENOSPC; |
44dd3ffa | 3816 | vcpu->arch.mmu->root_hpa = root; |
be01e8e2 | 3817 | goto set_root_pgd; |
17ac10ad | 3818 | } |
f87f9288 | 3819 | |
651dd37a JR |
3820 | /* |
3821 | * We shadow a 32 bit page table. This may be a legacy 2-level | |
81407ca5 JR |
3822 | * or a PAE 3-level page table. In either case we need to be aware that |
3823 | * the shadow page table may be a PAE or a long mode page table. | |
651dd37a | 3824 | */ |
81407ca5 | 3825 | pm_mask = PT_PRESENT_MASK; |
44dd3ffa | 3826 | if (vcpu->arch.mmu->shadow_root_level == PT64_ROOT_4LEVEL) |
81407ca5 JR |
3827 | pm_mask |= PT_ACCESSED_MASK | PT_WRITABLE_MASK | PT_USER_MASK; |
3828 | ||
17ac10ad | 3829 | for (i = 0; i < 4; ++i) { |
8123f265 | 3830 | MMU_WARN_ON(VALID_PAGE(vcpu->arch.mmu->pae_root[i])); |
44dd3ffa VK |
3831 | if (vcpu->arch.mmu->root_level == PT32E_ROOT_LEVEL) { |
3832 | pdptr = vcpu->arch.mmu->get_pdptr(vcpu, i); | |
812f30b2 | 3833 | if (!(pdptr & PT_PRESENT_MASK)) { |
44dd3ffa | 3834 | vcpu->arch.mmu->pae_root[i] = 0; |
417726a3 AK |
3835 | continue; |
3836 | } | |
6de4f3ad | 3837 | root_gfn = pdptr >> PAGE_SHIFT; |
f87f9288 JR |
3838 | if (mmu_check_root(vcpu, root_gfn)) |
3839 | return 1; | |
5a7388c2 | 3840 | } |
8facbbff | 3841 | |
8123f265 SC |
3842 | root = mmu_alloc_root(vcpu, root_gfn, i << 30, |
3843 | PT32_ROOT_LEVEL, false); | |
3844 | if (!VALID_PAGE(root)) | |
3845 | return -ENOSPC; | |
44dd3ffa | 3846 | vcpu->arch.mmu->pae_root[i] = root | pm_mask; |
17ac10ad | 3847 | } |
44dd3ffa | 3848 | vcpu->arch.mmu->root_hpa = __pa(vcpu->arch.mmu->pae_root); |
81407ca5 JR |
3849 | |
3850 | /* | |
3851 | * If we shadow a 32 bit page table with a long mode page | |
3852 | * table we enter this path. | |
3853 | */ | |
44dd3ffa VK |
3854 | if (vcpu->arch.mmu->shadow_root_level == PT64_ROOT_4LEVEL) { |
3855 | if (vcpu->arch.mmu->lm_root == NULL) { | |
81407ca5 JR |
3856 | /* |
3857 | * The additional page necessary for this is only | |
3858 | * allocated on demand. | |
3859 | */ | |
3860 | ||
3861 | u64 *lm_root; | |
3862 | ||
254272ce | 3863 | lm_root = (void*)get_zeroed_page(GFP_KERNEL_ACCOUNT); |
81407ca5 JR |
3864 | if (lm_root == NULL) |
3865 | return 1; | |
3866 | ||
44dd3ffa | 3867 | lm_root[0] = __pa(vcpu->arch.mmu->pae_root) | pm_mask; |
81407ca5 | 3868 | |
44dd3ffa | 3869 | vcpu->arch.mmu->lm_root = lm_root; |
81407ca5 JR |
3870 | } |
3871 | ||
44dd3ffa | 3872 | vcpu->arch.mmu->root_hpa = __pa(vcpu->arch.mmu->lm_root); |
81407ca5 JR |
3873 | } |
3874 | ||
be01e8e2 SC |
3875 | set_root_pgd: |
3876 | vcpu->arch.mmu->root_pgd = root_pgd; | |
ad7dc69a | 3877 | |
8986ecc0 | 3878 | return 0; |
17ac10ad AK |
3879 | } |
3880 | ||
651dd37a JR |
3881 | static int mmu_alloc_roots(struct kvm_vcpu *vcpu) |
3882 | { | |
44dd3ffa | 3883 | if (vcpu->arch.mmu->direct_map) |
651dd37a JR |
3884 | return mmu_alloc_direct_roots(vcpu); |
3885 | else | |
3886 | return mmu_alloc_shadow_roots(vcpu); | |
3887 | } | |
3888 | ||
578e1c4d | 3889 | void kvm_mmu_sync_roots(struct kvm_vcpu *vcpu) |
0ba73cda MT |
3890 | { |
3891 | int i; | |
3892 | struct kvm_mmu_page *sp; | |
3893 | ||
44dd3ffa | 3894 | if (vcpu->arch.mmu->direct_map) |
81407ca5 JR |
3895 | return; |
3896 | ||
44dd3ffa | 3897 | if (!VALID_PAGE(vcpu->arch.mmu->root_hpa)) |
0ba73cda | 3898 | return; |
6903074c | 3899 | |
56f17dd3 | 3900 | vcpu_clear_mmio_info(vcpu, MMIO_GVA_ANY); |
578e1c4d | 3901 | |
44dd3ffa VK |
3902 | if (vcpu->arch.mmu->root_level >= PT64_ROOT_4LEVEL) { |
3903 | hpa_t root = vcpu->arch.mmu->root_hpa; | |
e47c4aee | 3904 | sp = to_shadow_page(root); |
578e1c4d JS |
3905 | |
3906 | /* | |
3907 | * Even if another CPU was marking the SP as unsync-ed | |
3908 | * simultaneously, any guest page table changes are not | |
3909 | * guaranteed to be visible anyway until this VCPU issues a TLB | |
3910 | * flush strictly after those changes are made. We only need to | |
3911 | * ensure that the other CPU sets these flags before any actual | |
3912 | * changes to the page tables are made. The comments in | |
3913 | * mmu_need_write_protect() describe what could go wrong if this | |
3914 | * requirement isn't satisfied. | |
3915 | */ | |
3916 | if (!smp_load_acquire(&sp->unsync) && | |
3917 | !smp_load_acquire(&sp->unsync_children)) | |
3918 | return; | |
3919 | ||
3920 | spin_lock(&vcpu->kvm->mmu_lock); | |
3921 | kvm_mmu_audit(vcpu, AUDIT_PRE_SYNC); | |
3922 | ||
0ba73cda | 3923 | mmu_sync_children(vcpu, sp); |
578e1c4d | 3924 | |
0375f7fa | 3925 | kvm_mmu_audit(vcpu, AUDIT_POST_SYNC); |
578e1c4d | 3926 | spin_unlock(&vcpu->kvm->mmu_lock); |
0ba73cda MT |
3927 | return; |
3928 | } | |
578e1c4d JS |
3929 | |
3930 | spin_lock(&vcpu->kvm->mmu_lock); | |
3931 | kvm_mmu_audit(vcpu, AUDIT_PRE_SYNC); | |
3932 | ||
0ba73cda | 3933 | for (i = 0; i < 4; ++i) { |
44dd3ffa | 3934 | hpa_t root = vcpu->arch.mmu->pae_root[i]; |
0ba73cda | 3935 | |
8986ecc0 | 3936 | if (root && VALID_PAGE(root)) { |
0ba73cda | 3937 | root &= PT64_BASE_ADDR_MASK; |
e47c4aee | 3938 | sp = to_shadow_page(root); |
0ba73cda MT |
3939 | mmu_sync_children(vcpu, sp); |
3940 | } | |
3941 | } | |
0ba73cda | 3942 | |
578e1c4d | 3943 | kvm_mmu_audit(vcpu, AUDIT_POST_SYNC); |
6cffe8ca | 3944 | spin_unlock(&vcpu->kvm->mmu_lock); |
0ba73cda | 3945 | } |
bfd0a56b | 3946 | EXPORT_SYMBOL_GPL(kvm_mmu_sync_roots); |
0ba73cda | 3947 | |
736c291c | 3948 | static gpa_t nonpaging_gva_to_gpa(struct kvm_vcpu *vcpu, gpa_t vaddr, |
ab9ae313 | 3949 | u32 access, struct x86_exception *exception) |
6aa8b732 | 3950 | { |
ab9ae313 AK |
3951 | if (exception) |
3952 | exception->error_code = 0; | |
6aa8b732 AK |
3953 | return vaddr; |
3954 | } | |
3955 | ||
736c291c | 3956 | static gpa_t nonpaging_gva_to_gpa_nested(struct kvm_vcpu *vcpu, gpa_t vaddr, |
ab9ae313 AK |
3957 | u32 access, |
3958 | struct x86_exception *exception) | |
6539e738 | 3959 | { |
ab9ae313 AK |
3960 | if (exception) |
3961 | exception->error_code = 0; | |
54987b7a | 3962 | return vcpu->arch.nested_mmu.translate_gpa(vcpu, vaddr, access, exception); |
6539e738 JR |
3963 | } |
3964 | ||
d625b155 XG |
3965 | static bool |
3966 | __is_rsvd_bits_set(struct rsvd_bits_validate *rsvd_check, u64 pte, int level) | |
3967 | { | |
b5c3c1b3 | 3968 | int bit7 = (pte >> 7) & 1; |
d625b155 | 3969 | |
b5c3c1b3 | 3970 | return pte & rsvd_check->rsvd_bits_mask[bit7][level-1]; |
d625b155 XG |
3971 | } |
3972 | ||
b5c3c1b3 | 3973 | static bool __is_bad_mt_xwr(struct rsvd_bits_validate *rsvd_check, u64 pte) |
d625b155 | 3974 | { |
b5c3c1b3 | 3975 | return rsvd_check->bad_mt_xwr & BIT_ULL(pte & 0x3f); |
d625b155 XG |
3976 | } |
3977 | ||
ded58749 | 3978 | static bool mmio_info_in_cache(struct kvm_vcpu *vcpu, u64 addr, bool direct) |
ce88decf | 3979 | { |
9034e6e8 PB |
3980 | /* |
3981 | * A nested guest cannot use the MMIO cache if it is using nested | |
3982 | * page tables, because cr2 is a nGPA while the cache stores GPAs. | |
3983 | */ | |
3984 | if (mmu_is_nested(vcpu)) | |
3985 | return false; | |
3986 | ||
ce88decf XG |
3987 | if (direct) |
3988 | return vcpu_match_mmio_gpa(vcpu, addr); | |
3989 | ||
3990 | return vcpu_match_mmio_gva(vcpu, addr); | |
3991 | } | |
3992 | ||
47ab8751 XG |
3993 | /* return true if reserved bit is detected on spte. */ |
3994 | static bool | |
3995 | walk_shadow_page_get_mmio_spte(struct kvm_vcpu *vcpu, u64 addr, u64 *sptep) | |
ce88decf XG |
3996 | { |
3997 | struct kvm_shadow_walk_iterator iterator; | |
2a7266a8 | 3998 | u64 sptes[PT64_ROOT_MAX_LEVEL], spte = 0ull; |
b5c3c1b3 | 3999 | struct rsvd_bits_validate *rsvd_check; |
47ab8751 XG |
4000 | int root, leaf; |
4001 | bool reserved = false; | |
ce88decf | 4002 | |
b5c3c1b3 | 4003 | rsvd_check = &vcpu->arch.mmu->shadow_zero_check; |
37f6a4e2 | 4004 | |
ce88decf | 4005 | walk_shadow_page_lockless_begin(vcpu); |
47ab8751 | 4006 | |
29ecd660 PB |
4007 | for (shadow_walk_init(&iterator, vcpu, addr), |
4008 | leaf = root = iterator.level; | |
47ab8751 XG |
4009 | shadow_walk_okay(&iterator); |
4010 | __shadow_walk_next(&iterator, spte)) { | |
47ab8751 XG |
4011 | spte = mmu_spte_get_lockless(iterator.sptep); |
4012 | ||
4013 | sptes[leaf - 1] = spte; | |
29ecd660 | 4014 | leaf--; |
47ab8751 | 4015 | |
ce88decf XG |
4016 | if (!is_shadow_present_pte(spte)) |
4017 | break; | |
47ab8751 | 4018 | |
b5c3c1b3 SC |
4019 | /* |
4020 | * Use a bitwise-OR instead of a logical-OR to aggregate the | |
4021 | * reserved bit and EPT's invalid memtype/XWR checks to avoid | |
4022 | * adding a Jcc in the loop. | |
4023 | */ | |
4024 | reserved |= __is_bad_mt_xwr(rsvd_check, spte) | | |
4025 | __is_rsvd_bits_set(rsvd_check, spte, iterator.level); | |
47ab8751 XG |
4026 | } |
4027 | ||
ce88decf XG |
4028 | walk_shadow_page_lockless_end(vcpu); |
4029 | ||
47ab8751 XG |
4030 | if (reserved) { |
4031 | pr_err("%s: detect reserved bits on spte, addr 0x%llx, dump hierarchy:\n", | |
4032 | __func__, addr); | |
29ecd660 | 4033 | while (root > leaf) { |
47ab8751 XG |
4034 | pr_err("------ spte 0x%llx level %d.\n", |
4035 | sptes[root - 1], root); | |
4036 | root--; | |
4037 | } | |
4038 | } | |
ddce6208 | 4039 | |
47ab8751 XG |
4040 | *sptep = spte; |
4041 | return reserved; | |
ce88decf XG |
4042 | } |
4043 | ||
e08d26f0 | 4044 | static int handle_mmio_page_fault(struct kvm_vcpu *vcpu, u64 addr, bool direct) |
ce88decf XG |
4045 | { |
4046 | u64 spte; | |
47ab8751 | 4047 | bool reserved; |
ce88decf | 4048 | |
ded58749 | 4049 | if (mmio_info_in_cache(vcpu, addr, direct)) |
9b8ebbdb | 4050 | return RET_PF_EMULATE; |
ce88decf | 4051 | |
47ab8751 | 4052 | reserved = walk_shadow_page_get_mmio_spte(vcpu, addr, &spte); |
450869d6 | 4053 | if (WARN_ON(reserved)) |
9b8ebbdb | 4054 | return -EINVAL; |
ce88decf XG |
4055 | |
4056 | if (is_mmio_spte(spte)) { | |
4057 | gfn_t gfn = get_mmio_spte_gfn(spte); | |
0a2b64c5 | 4058 | unsigned int access = get_mmio_spte_access(spte); |
ce88decf | 4059 | |
54bf36aa | 4060 | if (!check_mmio_spte(vcpu, spte)) |
9b8ebbdb | 4061 | return RET_PF_INVALID; |
f8f55942 | 4062 | |
ce88decf XG |
4063 | if (direct) |
4064 | addr = 0; | |
4f022648 XG |
4065 | |
4066 | trace_handle_mmio_page_fault(addr, gfn, access); | |
ce88decf | 4067 | vcpu_cache_mmio_info(vcpu, addr, gfn, access); |
9b8ebbdb | 4068 | return RET_PF_EMULATE; |
ce88decf XG |
4069 | } |
4070 | ||
ce88decf XG |
4071 | /* |
4072 | * If the page table is zapped by other cpus, let CPU fault again on | |
4073 | * the address. | |
4074 | */ | |
9b8ebbdb | 4075 | return RET_PF_RETRY; |
ce88decf | 4076 | } |
ce88decf | 4077 | |
3d0c27ad XG |
4078 | static bool page_fault_handle_page_track(struct kvm_vcpu *vcpu, |
4079 | u32 error_code, gfn_t gfn) | |
4080 | { | |
4081 | if (unlikely(error_code & PFERR_RSVD_MASK)) | |
4082 | return false; | |
4083 | ||
4084 | if (!(error_code & PFERR_PRESENT_MASK) || | |
4085 | !(error_code & PFERR_WRITE_MASK)) | |
4086 | return false; | |
4087 | ||
4088 | /* | |
4089 | * guest is writing the page which is write tracked which can | |
4090 | * not be fixed by page fault handler. | |
4091 | */ | |
4092 | if (kvm_page_track_is_active(vcpu, gfn, KVM_PAGE_TRACK_WRITE)) | |
4093 | return true; | |
4094 | ||
4095 | return false; | |
4096 | } | |
4097 | ||
e5691a81 XG |
4098 | static void shadow_page_table_clear_flood(struct kvm_vcpu *vcpu, gva_t addr) |
4099 | { | |
4100 | struct kvm_shadow_walk_iterator iterator; | |
4101 | u64 spte; | |
4102 | ||
e5691a81 XG |
4103 | walk_shadow_page_lockless_begin(vcpu); |
4104 | for_each_shadow_entry_lockless(vcpu, addr, iterator, spte) { | |
4105 | clear_sp_write_flooding_count(iterator.sptep); | |
4106 | if (!is_shadow_present_pte(spte)) | |
4107 | break; | |
4108 | } | |
4109 | walk_shadow_page_lockless_end(vcpu); | |
4110 | } | |
4111 | ||
e8c22266 VK |
4112 | static bool kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa, |
4113 | gfn_t gfn) | |
af585b92 GN |
4114 | { |
4115 | struct kvm_arch_async_pf arch; | |
fb67e14f | 4116 | |
7c90705b | 4117 | arch.token = (vcpu->arch.apf.id++ << 12) | vcpu->vcpu_id; |
af585b92 | 4118 | arch.gfn = gfn; |
44dd3ffa | 4119 | arch.direct_map = vcpu->arch.mmu->direct_map; |
d8dd54e0 | 4120 | arch.cr3 = vcpu->arch.mmu->get_guest_pgd(vcpu); |
af585b92 | 4121 | |
9f1a8526 SC |
4122 | return kvm_setup_async_pf(vcpu, cr2_or_gpa, |
4123 | kvm_vcpu_gfn_to_hva(vcpu, gfn), &arch); | |
af585b92 GN |
4124 | } |
4125 | ||
78b2c54a | 4126 | static bool try_async_pf(struct kvm_vcpu *vcpu, bool prefault, gfn_t gfn, |
9f1a8526 SC |
4127 | gpa_t cr2_or_gpa, kvm_pfn_t *pfn, bool write, |
4128 | bool *writable) | |
af585b92 | 4129 | { |
c36b7150 | 4130 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); |
af585b92 GN |
4131 | bool async; |
4132 | ||
c36b7150 PB |
4133 | /* Don't expose private memslots to L2. */ |
4134 | if (is_guest_mode(vcpu) && !kvm_is_visible_memslot(slot)) { | |
3a2936de | 4135 | *pfn = KVM_PFN_NOSLOT; |
c583eed6 | 4136 | *writable = false; |
3a2936de JM |
4137 | return false; |
4138 | } | |
4139 | ||
3520469d PB |
4140 | async = false; |
4141 | *pfn = __gfn_to_pfn_memslot(slot, gfn, false, &async, write, writable); | |
af585b92 GN |
4142 | if (!async) |
4143 | return false; /* *pfn has correct page already */ | |
4144 | ||
9bc1f09f | 4145 | if (!prefault && kvm_can_do_async_pf(vcpu)) { |
9f1a8526 | 4146 | trace_kvm_try_async_get_page(cr2_or_gpa, gfn); |
af585b92 | 4147 | if (kvm_find_async_pf_gfn(vcpu, gfn)) { |
9f1a8526 | 4148 | trace_kvm_async_pf_doublefault(cr2_or_gpa, gfn); |
af585b92 GN |
4149 | kvm_make_request(KVM_REQ_APF_HALT, vcpu); |
4150 | return true; | |
9f1a8526 | 4151 | } else if (kvm_arch_setup_async_pf(vcpu, cr2_or_gpa, gfn)) |
af585b92 GN |
4152 | return true; |
4153 | } | |
4154 | ||
3520469d | 4155 | *pfn = __gfn_to_pfn_memslot(slot, gfn, false, NULL, write, writable); |
af585b92 GN |
4156 | return false; |
4157 | } | |
4158 | ||
0f90e1c1 SC |
4159 | static int direct_page_fault(struct kvm_vcpu *vcpu, gpa_t gpa, u32 error_code, |
4160 | bool prefault, int max_level, bool is_tdp) | |
6aa8b732 | 4161 | { |
367fd790 | 4162 | bool write = error_code & PFERR_WRITE_MASK; |
0f90e1c1 | 4163 | bool map_writable; |
6aa8b732 | 4164 | |
0f90e1c1 SC |
4165 | gfn_t gfn = gpa >> PAGE_SHIFT; |
4166 | unsigned long mmu_seq; | |
4167 | kvm_pfn_t pfn; | |
83f06fa7 | 4168 | int r; |
ce88decf | 4169 | |
3d0c27ad | 4170 | if (page_fault_handle_page_track(vcpu, error_code, gfn)) |
9b8ebbdb | 4171 | return RET_PF_EMULATE; |
ce88decf | 4172 | |
c4371c2a SC |
4173 | r = fast_page_fault(vcpu, gpa, error_code); |
4174 | if (r != RET_PF_INVALID) | |
4175 | return r; | |
83291445 | 4176 | |
378f5cd6 | 4177 | r = mmu_topup_memory_caches(vcpu, false); |
e2dec939 AK |
4178 | if (r) |
4179 | return r; | |
714b93da | 4180 | |
367fd790 SC |
4181 | mmu_seq = vcpu->kvm->mmu_notifier_seq; |
4182 | smp_rmb(); | |
4183 | ||
4184 | if (try_async_pf(vcpu, prefault, gfn, gpa, &pfn, write, &map_writable)) | |
4185 | return RET_PF_RETRY; | |
4186 | ||
0f90e1c1 | 4187 | if (handle_abnormal_pfn(vcpu, is_tdp ? 0 : gpa, gfn, pfn, ACC_ALL, &r)) |
367fd790 | 4188 | return r; |
6aa8b732 | 4189 | |
367fd790 SC |
4190 | r = RET_PF_RETRY; |
4191 | spin_lock(&vcpu->kvm->mmu_lock); | |
4192 | if (mmu_notifier_retry(vcpu->kvm, mmu_seq)) | |
4193 | goto out_unlock; | |
7bd7ded6 SC |
4194 | r = make_mmu_pages_available(vcpu); |
4195 | if (r) | |
367fd790 | 4196 | goto out_unlock; |
6c2fd34f SC |
4197 | r = __direct_map(vcpu, gpa, error_code, map_writable, max_level, pfn, |
4198 | prefault, is_tdp); | |
0f90e1c1 | 4199 | |
367fd790 SC |
4200 | out_unlock: |
4201 | spin_unlock(&vcpu->kvm->mmu_lock); | |
4202 | kvm_release_pfn_clean(pfn); | |
4203 | return r; | |
6aa8b732 AK |
4204 | } |
4205 | ||
0f90e1c1 SC |
4206 | static int nonpaging_page_fault(struct kvm_vcpu *vcpu, gpa_t gpa, |
4207 | u32 error_code, bool prefault) | |
4208 | { | |
4209 | pgprintk("%s: gva %lx error %x\n", __func__, gpa, error_code); | |
4210 | ||
4211 | /* This path builds a PAE pagetable, we can map 2mb pages at maximum. */ | |
4212 | return direct_page_fault(vcpu, gpa & PAGE_MASK, error_code, prefault, | |
3bae0459 | 4213 | PG_LEVEL_2M, false); |
0f90e1c1 SC |
4214 | } |
4215 | ||
1261bfa3 | 4216 | int kvm_handle_page_fault(struct kvm_vcpu *vcpu, u64 error_code, |
d0006530 | 4217 | u64 fault_address, char *insn, int insn_len) |
1261bfa3 WL |
4218 | { |
4219 | int r = 1; | |
9ce372b3 | 4220 | u32 flags = vcpu->arch.apf.host_apf_flags; |
1261bfa3 | 4221 | |
736c291c SC |
4222 | #ifndef CONFIG_X86_64 |
4223 | /* A 64-bit CR2 should be impossible on 32-bit KVM. */ | |
4224 | if (WARN_ON_ONCE(fault_address >> 32)) | |
4225 | return -EFAULT; | |
4226 | #endif | |
4227 | ||
c595ceee | 4228 | vcpu->arch.l1tf_flush_l1d = true; |
9ce372b3 | 4229 | if (!flags) { |
1261bfa3 WL |
4230 | trace_kvm_page_fault(fault_address, error_code); |
4231 | ||
d0006530 | 4232 | if (kvm_event_needs_reinjection(vcpu)) |
1261bfa3 WL |
4233 | kvm_mmu_unprotect_page_virt(vcpu, fault_address); |
4234 | r = kvm_mmu_page_fault(vcpu, fault_address, error_code, insn, | |
4235 | insn_len); | |
9ce372b3 | 4236 | } else if (flags & KVM_PV_REASON_PAGE_NOT_PRESENT) { |
68fd66f1 | 4237 | vcpu->arch.apf.host_apf_flags = 0; |
1261bfa3 | 4238 | local_irq_disable(); |
6bca69ad | 4239 | kvm_async_pf_task_wait_schedule(fault_address); |
1261bfa3 | 4240 | local_irq_enable(); |
9ce372b3 VK |
4241 | } else { |
4242 | WARN_ONCE(1, "Unexpected host async PF flags: %x\n", flags); | |
1261bfa3 | 4243 | } |
9ce372b3 | 4244 | |
1261bfa3 WL |
4245 | return r; |
4246 | } | |
4247 | EXPORT_SYMBOL_GPL(kvm_handle_page_fault); | |
4248 | ||
7a02674d SC |
4249 | int kvm_tdp_page_fault(struct kvm_vcpu *vcpu, gpa_t gpa, u32 error_code, |
4250 | bool prefault) | |
fb72d167 | 4251 | { |
cb9b88c6 | 4252 | int max_level; |
fb72d167 | 4253 | |
e662ec3e | 4254 | for (max_level = KVM_MAX_HUGEPAGE_LEVEL; |
3bae0459 | 4255 | max_level > PG_LEVEL_4K; |
cb9b88c6 SC |
4256 | max_level--) { |
4257 | int page_num = KVM_PAGES_PER_HPAGE(max_level); | |
0f90e1c1 | 4258 | gfn_t base = (gpa >> PAGE_SHIFT) & ~(page_num - 1); |
ce88decf | 4259 | |
cb9b88c6 SC |
4260 | if (kvm_mtrr_check_gfn_range_consistency(vcpu, base, page_num)) |
4261 | break; | |
fd136902 | 4262 | } |
852e3c19 | 4263 | |
0f90e1c1 SC |
4264 | return direct_page_fault(vcpu, gpa, error_code, prefault, |
4265 | max_level, true); | |
fb72d167 JR |
4266 | } |
4267 | ||
8a3c1a33 PB |
4268 | static void nonpaging_init_context(struct kvm_vcpu *vcpu, |
4269 | struct kvm_mmu *context) | |
6aa8b732 | 4270 | { |
6aa8b732 | 4271 | context->page_fault = nonpaging_page_fault; |
6aa8b732 | 4272 | context->gva_to_gpa = nonpaging_gva_to_gpa; |
e8bc217a | 4273 | context->sync_page = nonpaging_sync_page; |
5efac074 | 4274 | context->invlpg = NULL; |
0f53b5b1 | 4275 | context->update_pte = nonpaging_update_pte; |
cea0f0e7 | 4276 | context->root_level = 0; |
6aa8b732 | 4277 | context->shadow_root_level = PT32E_ROOT_LEVEL; |
c5a78f2b | 4278 | context->direct_map = true; |
2d48a985 | 4279 | context->nx = false; |
6aa8b732 AK |
4280 | } |
4281 | ||
be01e8e2 | 4282 | static inline bool is_root_usable(struct kvm_mmu_root_info *root, gpa_t pgd, |
0be44352 SC |
4283 | union kvm_mmu_page_role role) |
4284 | { | |
be01e8e2 | 4285 | return (role.direct || pgd == root->pgd) && |
e47c4aee SC |
4286 | VALID_PAGE(root->hpa) && to_shadow_page(root->hpa) && |
4287 | role.word == to_shadow_page(root->hpa)->role.word; | |
0be44352 SC |
4288 | } |
4289 | ||
b94742c9 | 4290 | /* |
be01e8e2 | 4291 | * Find out if a previously cached root matching the new pgd/role is available. |
b94742c9 JS |
4292 | * The current root is also inserted into the cache. |
4293 | * If a matching root was found, it is assigned to kvm_mmu->root_hpa and true is | |
4294 | * returned. | |
4295 | * Otherwise, the LRU root from the cache is assigned to kvm_mmu->root_hpa and | |
4296 | * false is returned. This root should now be freed by the caller. | |
4297 | */ | |
be01e8e2 | 4298 | static bool cached_root_available(struct kvm_vcpu *vcpu, gpa_t new_pgd, |
b94742c9 JS |
4299 | union kvm_mmu_page_role new_role) |
4300 | { | |
4301 | uint i; | |
4302 | struct kvm_mmu_root_info root; | |
44dd3ffa | 4303 | struct kvm_mmu *mmu = vcpu->arch.mmu; |
b94742c9 | 4304 | |
be01e8e2 | 4305 | root.pgd = mmu->root_pgd; |
b94742c9 JS |
4306 | root.hpa = mmu->root_hpa; |
4307 | ||
be01e8e2 | 4308 | if (is_root_usable(&root, new_pgd, new_role)) |
0be44352 SC |
4309 | return true; |
4310 | ||
b94742c9 JS |
4311 | for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) { |
4312 | swap(root, mmu->prev_roots[i]); | |
4313 | ||
be01e8e2 | 4314 | if (is_root_usable(&root, new_pgd, new_role)) |
b94742c9 JS |
4315 | break; |
4316 | } | |
4317 | ||
4318 | mmu->root_hpa = root.hpa; | |
be01e8e2 | 4319 | mmu->root_pgd = root.pgd; |
b94742c9 JS |
4320 | |
4321 | return i < KVM_MMU_NUM_PREV_ROOTS; | |
4322 | } | |
4323 | ||
be01e8e2 | 4324 | static bool fast_pgd_switch(struct kvm_vcpu *vcpu, gpa_t new_pgd, |
b869855b | 4325 | union kvm_mmu_page_role new_role) |
6aa8b732 | 4326 | { |
44dd3ffa | 4327 | struct kvm_mmu *mmu = vcpu->arch.mmu; |
7c390d35 JS |
4328 | |
4329 | /* | |
4330 | * For now, limit the fast switch to 64-bit hosts+VMs in order to avoid | |
4331 | * having to deal with PDPTEs. We may add support for 32-bit hosts/VMs | |
4332 | * later if necessary. | |
4333 | */ | |
4334 | if (mmu->shadow_root_level >= PT64_ROOT_4LEVEL && | |
b869855b | 4335 | mmu->root_level >= PT64_ROOT_4LEVEL) |
fe9304d3 | 4336 | return cached_root_available(vcpu, new_pgd, new_role); |
7c390d35 JS |
4337 | |
4338 | return false; | |
6aa8b732 AK |
4339 | } |
4340 | ||
be01e8e2 | 4341 | static void __kvm_mmu_new_pgd(struct kvm_vcpu *vcpu, gpa_t new_pgd, |
ade61e28 | 4342 | union kvm_mmu_page_role new_role, |
4a632ac6 | 4343 | bool skip_tlb_flush, bool skip_mmu_sync) |
6aa8b732 | 4344 | { |
be01e8e2 | 4345 | if (!fast_pgd_switch(vcpu, new_pgd, new_role)) { |
b869855b SC |
4346 | kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, KVM_MMU_ROOT_CURRENT); |
4347 | return; | |
4348 | } | |
4349 | ||
4350 | /* | |
4351 | * It's possible that the cached previous root page is obsolete because | |
4352 | * of a change in the MMU generation number. However, changing the | |
4353 | * generation number is accompanied by KVM_REQ_MMU_RELOAD, which will | |
4354 | * free the root set here and allocate a new one. | |
4355 | */ | |
4356 | kvm_make_request(KVM_REQ_LOAD_MMU_PGD, vcpu); | |
4357 | ||
71fe7013 | 4358 | if (!skip_mmu_sync || force_flush_and_sync_on_reuse) |
b869855b | 4359 | kvm_make_request(KVM_REQ_MMU_SYNC, vcpu); |
71fe7013 | 4360 | if (!skip_tlb_flush || force_flush_and_sync_on_reuse) |
b869855b | 4361 | kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu); |
b869855b SC |
4362 | |
4363 | /* | |
4364 | * The last MMIO access's GVA and GPA are cached in the VCPU. When | |
4365 | * switching to a new CR3, that GVA->GPA mapping may no longer be | |
4366 | * valid. So clear any cached MMIO info even when we don't need to sync | |
4367 | * the shadow page tables. | |
4368 | */ | |
4369 | vcpu_clear_mmio_info(vcpu, MMIO_GVA_ANY); | |
4370 | ||
e47c4aee | 4371 | __clear_sp_write_flooding_count(to_shadow_page(vcpu->arch.mmu->root_hpa)); |
6aa8b732 AK |
4372 | } |
4373 | ||
be01e8e2 | 4374 | void kvm_mmu_new_pgd(struct kvm_vcpu *vcpu, gpa_t new_pgd, bool skip_tlb_flush, |
4a632ac6 | 4375 | bool skip_mmu_sync) |
0aab33e4 | 4376 | { |
be01e8e2 | 4377 | __kvm_mmu_new_pgd(vcpu, new_pgd, kvm_mmu_calc_root_page_role(vcpu), |
4a632ac6 | 4378 | skip_tlb_flush, skip_mmu_sync); |
0aab33e4 | 4379 | } |
be01e8e2 | 4380 | EXPORT_SYMBOL_GPL(kvm_mmu_new_pgd); |
0aab33e4 | 4381 | |
5777ed34 JR |
4382 | static unsigned long get_cr3(struct kvm_vcpu *vcpu) |
4383 | { | |
9f8fe504 | 4384 | return kvm_read_cr3(vcpu); |
5777ed34 JR |
4385 | } |
4386 | ||
54bf36aa | 4387 | static bool sync_mmio_spte(struct kvm_vcpu *vcpu, u64 *sptep, gfn_t gfn, |
0a2b64c5 | 4388 | unsigned int access, int *nr_present) |
ce88decf XG |
4389 | { |
4390 | if (unlikely(is_mmio_spte(*sptep))) { | |
4391 | if (gfn != get_mmio_spte_gfn(*sptep)) { | |
4392 | mmu_spte_clear_no_track(sptep); | |
4393 | return true; | |
4394 | } | |
4395 | ||
4396 | (*nr_present)++; | |
54bf36aa | 4397 | mark_mmio_spte(vcpu, sptep, gfn, access); |
ce88decf XG |
4398 | return true; |
4399 | } | |
4400 | ||
4401 | return false; | |
4402 | } | |
4403 | ||
6bb69c9b PB |
4404 | static inline bool is_last_gpte(struct kvm_mmu *mmu, |
4405 | unsigned level, unsigned gpte) | |
6fd01b71 | 4406 | { |
6bb69c9b PB |
4407 | /* |
4408 | * The RHS has bit 7 set iff level < mmu->last_nonleaf_level. | |
4409 | * If it is clear, there are no large pages at this level, so clear | |
4410 | * PT_PAGE_SIZE_MASK in gpte if that is the case. | |
4411 | */ | |
4412 | gpte &= level - mmu->last_nonleaf_level; | |
4413 | ||
829ee279 | 4414 | /* |
3bae0459 SC |
4415 | * PG_LEVEL_4K always terminates. The RHS has bit 7 set |
4416 | * iff level <= PG_LEVEL_4K, which for our purpose means | |
4417 | * level == PG_LEVEL_4K; set PT_PAGE_SIZE_MASK in gpte then. | |
829ee279 | 4418 | */ |
3bae0459 | 4419 | gpte |= level - PG_LEVEL_4K - 1; |
829ee279 | 4420 | |
6bb69c9b | 4421 | return gpte & PT_PAGE_SIZE_MASK; |
6fd01b71 AK |
4422 | } |
4423 | ||
37406aaa NHE |
4424 | #define PTTYPE_EPT 18 /* arbitrary */ |
4425 | #define PTTYPE PTTYPE_EPT | |
4426 | #include "paging_tmpl.h" | |
4427 | #undef PTTYPE | |
4428 | ||
6aa8b732 AK |
4429 | #define PTTYPE 64 |
4430 | #include "paging_tmpl.h" | |
4431 | #undef PTTYPE | |
4432 | ||
4433 | #define PTTYPE 32 | |
4434 | #include "paging_tmpl.h" | |
4435 | #undef PTTYPE | |
4436 | ||
6dc98b86 XG |
4437 | static void |
4438 | __reset_rsvds_bits_mask(struct kvm_vcpu *vcpu, | |
4439 | struct rsvd_bits_validate *rsvd_check, | |
4440 | int maxphyaddr, int level, bool nx, bool gbpages, | |
6fec2144 | 4441 | bool pse, bool amd) |
82725b20 | 4442 | { |
82725b20 | 4443 | u64 exb_bit_rsvd = 0; |
5f7dde7b | 4444 | u64 gbpages_bit_rsvd = 0; |
a0c0feb5 | 4445 | u64 nonleaf_bit8_rsvd = 0; |
82725b20 | 4446 | |
a0a64f50 | 4447 | rsvd_check->bad_mt_xwr = 0; |
25d92081 | 4448 | |
6dc98b86 | 4449 | if (!nx) |
82725b20 | 4450 | exb_bit_rsvd = rsvd_bits(63, 63); |
6dc98b86 | 4451 | if (!gbpages) |
5f7dde7b | 4452 | gbpages_bit_rsvd = rsvd_bits(7, 7); |
a0c0feb5 PB |
4453 | |
4454 | /* | |
4455 | * Non-leaf PML4Es and PDPEs reserve bit 8 (which would be the G bit for | |
4456 | * leaf entries) on AMD CPUs only. | |
4457 | */ | |
6fec2144 | 4458 | if (amd) |
a0c0feb5 PB |
4459 | nonleaf_bit8_rsvd = rsvd_bits(8, 8); |
4460 | ||
6dc98b86 | 4461 | switch (level) { |
82725b20 DE |
4462 | case PT32_ROOT_LEVEL: |
4463 | /* no rsvd bits for 2 level 4K page table entries */ | |
a0a64f50 XG |
4464 | rsvd_check->rsvd_bits_mask[0][1] = 0; |
4465 | rsvd_check->rsvd_bits_mask[0][0] = 0; | |
4466 | rsvd_check->rsvd_bits_mask[1][0] = | |
4467 | rsvd_check->rsvd_bits_mask[0][0]; | |
f815bce8 | 4468 | |
6dc98b86 | 4469 | if (!pse) { |
a0a64f50 | 4470 | rsvd_check->rsvd_bits_mask[1][1] = 0; |
f815bce8 XG |
4471 | break; |
4472 | } | |
4473 | ||
82725b20 DE |
4474 | if (is_cpuid_PSE36()) |
4475 | /* 36bits PSE 4MB page */ | |
a0a64f50 | 4476 | rsvd_check->rsvd_bits_mask[1][1] = rsvd_bits(17, 21); |
82725b20 DE |
4477 | else |
4478 | /* 32 bits PSE 4MB page */ | |
a0a64f50 | 4479 | rsvd_check->rsvd_bits_mask[1][1] = rsvd_bits(13, 21); |
82725b20 DE |
4480 | break; |
4481 | case PT32E_ROOT_LEVEL: | |
a0a64f50 | 4482 | rsvd_check->rsvd_bits_mask[0][2] = |
20c466b5 | 4483 | rsvd_bits(maxphyaddr, 63) | |
cd9ae5fe | 4484 | rsvd_bits(5, 8) | rsvd_bits(1, 2); /* PDPTE */ |
a0a64f50 | 4485 | rsvd_check->rsvd_bits_mask[0][1] = exb_bit_rsvd | |
4c26b4cd | 4486 | rsvd_bits(maxphyaddr, 62); /* PDE */ |
a0a64f50 | 4487 | rsvd_check->rsvd_bits_mask[0][0] = exb_bit_rsvd | |
82725b20 | 4488 | rsvd_bits(maxphyaddr, 62); /* PTE */ |
a0a64f50 | 4489 | rsvd_check->rsvd_bits_mask[1][1] = exb_bit_rsvd | |
82725b20 DE |
4490 | rsvd_bits(maxphyaddr, 62) | |
4491 | rsvd_bits(13, 20); /* large page */ | |
a0a64f50 XG |
4492 | rsvd_check->rsvd_bits_mask[1][0] = |
4493 | rsvd_check->rsvd_bits_mask[0][0]; | |
82725b20 | 4494 | break; |
855feb67 YZ |
4495 | case PT64_ROOT_5LEVEL: |
4496 | rsvd_check->rsvd_bits_mask[0][4] = exb_bit_rsvd | | |
4497 | nonleaf_bit8_rsvd | rsvd_bits(7, 7) | | |
4498 | rsvd_bits(maxphyaddr, 51); | |
4499 | rsvd_check->rsvd_bits_mask[1][4] = | |
4500 | rsvd_check->rsvd_bits_mask[0][4]; | |
df561f66 | 4501 | fallthrough; |
2a7266a8 | 4502 | case PT64_ROOT_4LEVEL: |
a0a64f50 XG |
4503 | rsvd_check->rsvd_bits_mask[0][3] = exb_bit_rsvd | |
4504 | nonleaf_bit8_rsvd | rsvd_bits(7, 7) | | |
4c26b4cd | 4505 | rsvd_bits(maxphyaddr, 51); |
a0a64f50 | 4506 | rsvd_check->rsvd_bits_mask[0][2] = exb_bit_rsvd | |
5ecad245 | 4507 | gbpages_bit_rsvd | |
82725b20 | 4508 | rsvd_bits(maxphyaddr, 51); |
a0a64f50 XG |
4509 | rsvd_check->rsvd_bits_mask[0][1] = exb_bit_rsvd | |
4510 | rsvd_bits(maxphyaddr, 51); | |
4511 | rsvd_check->rsvd_bits_mask[0][0] = exb_bit_rsvd | | |
4512 | rsvd_bits(maxphyaddr, 51); | |
4513 | rsvd_check->rsvd_bits_mask[1][3] = | |
4514 | rsvd_check->rsvd_bits_mask[0][3]; | |
4515 | rsvd_check->rsvd_bits_mask[1][2] = exb_bit_rsvd | | |
5f7dde7b | 4516 | gbpages_bit_rsvd | rsvd_bits(maxphyaddr, 51) | |
e04da980 | 4517 | rsvd_bits(13, 29); |
a0a64f50 | 4518 | rsvd_check->rsvd_bits_mask[1][1] = exb_bit_rsvd | |
4c26b4cd SY |
4519 | rsvd_bits(maxphyaddr, 51) | |
4520 | rsvd_bits(13, 20); /* large page */ | |
a0a64f50 XG |
4521 | rsvd_check->rsvd_bits_mask[1][0] = |
4522 | rsvd_check->rsvd_bits_mask[0][0]; | |
82725b20 DE |
4523 | break; |
4524 | } | |
4525 | } | |
4526 | ||
6dc98b86 XG |
4527 | static void reset_rsvds_bits_mask(struct kvm_vcpu *vcpu, |
4528 | struct kvm_mmu *context) | |
4529 | { | |
4530 | __reset_rsvds_bits_mask(vcpu, &context->guest_rsvd_check, | |
4531 | cpuid_maxphyaddr(vcpu), context->root_level, | |
d6321d49 RK |
4532 | context->nx, |
4533 | guest_cpuid_has(vcpu, X86_FEATURE_GBPAGES), | |
23493d0a SC |
4534 | is_pse(vcpu), |
4535 | guest_cpuid_is_amd_or_hygon(vcpu)); | |
6dc98b86 XG |
4536 | } |
4537 | ||
81b8eebb XG |
4538 | static void |
4539 | __reset_rsvds_bits_mask_ept(struct rsvd_bits_validate *rsvd_check, | |
4540 | int maxphyaddr, bool execonly) | |
25d92081 | 4541 | { |
951f9fd7 | 4542 | u64 bad_mt_xwr; |
25d92081 | 4543 | |
855feb67 YZ |
4544 | rsvd_check->rsvd_bits_mask[0][4] = |
4545 | rsvd_bits(maxphyaddr, 51) | rsvd_bits(3, 7); | |
a0a64f50 | 4546 | rsvd_check->rsvd_bits_mask[0][3] = |
25d92081 | 4547 | rsvd_bits(maxphyaddr, 51) | rsvd_bits(3, 7); |
a0a64f50 | 4548 | rsvd_check->rsvd_bits_mask[0][2] = |
25d92081 | 4549 | rsvd_bits(maxphyaddr, 51) | rsvd_bits(3, 6); |
a0a64f50 | 4550 | rsvd_check->rsvd_bits_mask[0][1] = |
25d92081 | 4551 | rsvd_bits(maxphyaddr, 51) | rsvd_bits(3, 6); |
a0a64f50 | 4552 | rsvd_check->rsvd_bits_mask[0][0] = rsvd_bits(maxphyaddr, 51); |
25d92081 YZ |
4553 | |
4554 | /* large page */ | |
855feb67 | 4555 | rsvd_check->rsvd_bits_mask[1][4] = rsvd_check->rsvd_bits_mask[0][4]; |
a0a64f50 XG |
4556 | rsvd_check->rsvd_bits_mask[1][3] = rsvd_check->rsvd_bits_mask[0][3]; |
4557 | rsvd_check->rsvd_bits_mask[1][2] = | |
25d92081 | 4558 | rsvd_bits(maxphyaddr, 51) | rsvd_bits(12, 29); |
a0a64f50 | 4559 | rsvd_check->rsvd_bits_mask[1][1] = |
25d92081 | 4560 | rsvd_bits(maxphyaddr, 51) | rsvd_bits(12, 20); |
a0a64f50 | 4561 | rsvd_check->rsvd_bits_mask[1][0] = rsvd_check->rsvd_bits_mask[0][0]; |
25d92081 | 4562 | |
951f9fd7 PB |
4563 | bad_mt_xwr = 0xFFull << (2 * 8); /* bits 3..5 must not be 2 */ |
4564 | bad_mt_xwr |= 0xFFull << (3 * 8); /* bits 3..5 must not be 3 */ | |
4565 | bad_mt_xwr |= 0xFFull << (7 * 8); /* bits 3..5 must not be 7 */ | |
4566 | bad_mt_xwr |= REPEAT_BYTE(1ull << 2); /* bits 0..2 must not be 010 */ | |
4567 | bad_mt_xwr |= REPEAT_BYTE(1ull << 6); /* bits 0..2 must not be 110 */ | |
4568 | if (!execonly) { | |
4569 | /* bits 0..2 must not be 100 unless VMX capabilities allow it */ | |
4570 | bad_mt_xwr |= REPEAT_BYTE(1ull << 4); | |
25d92081 | 4571 | } |
951f9fd7 | 4572 | rsvd_check->bad_mt_xwr = bad_mt_xwr; |
25d92081 YZ |
4573 | } |
4574 | ||
81b8eebb XG |
4575 | static void reset_rsvds_bits_mask_ept(struct kvm_vcpu *vcpu, |
4576 | struct kvm_mmu *context, bool execonly) | |
4577 | { | |
4578 | __reset_rsvds_bits_mask_ept(&context->guest_rsvd_check, | |
4579 | cpuid_maxphyaddr(vcpu), execonly); | |
4580 | } | |
4581 | ||
c258b62b XG |
4582 | /* |
4583 | * the page table on host is the shadow page table for the page | |
4584 | * table in guest or amd nested guest, its mmu features completely | |
4585 | * follow the features in guest. | |
4586 | */ | |
4587 | void | |
4588 | reset_shadow_zero_bits_mask(struct kvm_vcpu *vcpu, struct kvm_mmu *context) | |
4589 | { | |
36d9594d VK |
4590 | bool uses_nx = context->nx || |
4591 | context->mmu_role.base.smep_andnot_wp; | |
ea2800dd BS |
4592 | struct rsvd_bits_validate *shadow_zero_check; |
4593 | int i; | |
5f0b8199 | 4594 | |
6fec2144 PB |
4595 | /* |
4596 | * Passing "true" to the last argument is okay; it adds a check | |
4597 | * on bit 8 of the SPTEs which KVM doesn't use anyway. | |
4598 | */ | |
ea2800dd BS |
4599 | shadow_zero_check = &context->shadow_zero_check; |
4600 | __reset_rsvds_bits_mask(vcpu, shadow_zero_check, | |
f3ecb59d | 4601 | shadow_phys_bits, |
5f0b8199 | 4602 | context->shadow_root_level, uses_nx, |
d6321d49 RK |
4603 | guest_cpuid_has(vcpu, X86_FEATURE_GBPAGES), |
4604 | is_pse(vcpu), true); | |
ea2800dd BS |
4605 | |
4606 | if (!shadow_me_mask) | |
4607 | return; | |
4608 | ||
4609 | for (i = context->shadow_root_level; --i >= 0;) { | |
4610 | shadow_zero_check->rsvd_bits_mask[0][i] &= ~shadow_me_mask; | |
4611 | shadow_zero_check->rsvd_bits_mask[1][i] &= ~shadow_me_mask; | |
4612 | } | |
4613 | ||
c258b62b XG |
4614 | } |
4615 | EXPORT_SYMBOL_GPL(reset_shadow_zero_bits_mask); | |
4616 | ||
6fec2144 PB |
4617 | static inline bool boot_cpu_is_amd(void) |
4618 | { | |
4619 | WARN_ON_ONCE(!tdp_enabled); | |
4620 | return shadow_x_mask == 0; | |
4621 | } | |
4622 | ||
c258b62b XG |
4623 | /* |
4624 | * the direct page table on host, use as much mmu features as | |
4625 | * possible, however, kvm currently does not do execution-protection. | |
4626 | */ | |
4627 | static void | |
4628 | reset_tdp_shadow_zero_bits_mask(struct kvm_vcpu *vcpu, | |
4629 | struct kvm_mmu *context) | |
4630 | { | |
ea2800dd BS |
4631 | struct rsvd_bits_validate *shadow_zero_check; |
4632 | int i; | |
4633 | ||
4634 | shadow_zero_check = &context->shadow_zero_check; | |
4635 | ||
6fec2144 | 4636 | if (boot_cpu_is_amd()) |
ea2800dd | 4637 | __reset_rsvds_bits_mask(vcpu, shadow_zero_check, |
f3ecb59d | 4638 | shadow_phys_bits, |
c258b62b | 4639 | context->shadow_root_level, false, |
b8291adc BP |
4640 | boot_cpu_has(X86_FEATURE_GBPAGES), |
4641 | true, true); | |
c258b62b | 4642 | else |
ea2800dd | 4643 | __reset_rsvds_bits_mask_ept(shadow_zero_check, |
f3ecb59d | 4644 | shadow_phys_bits, |
c258b62b XG |
4645 | false); |
4646 | ||
ea2800dd BS |
4647 | if (!shadow_me_mask) |
4648 | return; | |
4649 | ||
4650 | for (i = context->shadow_root_level; --i >= 0;) { | |
4651 | shadow_zero_check->rsvd_bits_mask[0][i] &= ~shadow_me_mask; | |
4652 | shadow_zero_check->rsvd_bits_mask[1][i] &= ~shadow_me_mask; | |
4653 | } | |
c258b62b XG |
4654 | } |
4655 | ||
4656 | /* | |
4657 | * as the comments in reset_shadow_zero_bits_mask() except it | |
4658 | * is the shadow page table for intel nested guest. | |
4659 | */ | |
4660 | static void | |
4661 | reset_ept_shadow_zero_bits_mask(struct kvm_vcpu *vcpu, | |
4662 | struct kvm_mmu *context, bool execonly) | |
4663 | { | |
4664 | __reset_rsvds_bits_mask_ept(&context->shadow_zero_check, | |
f3ecb59d | 4665 | shadow_phys_bits, execonly); |
c258b62b XG |
4666 | } |
4667 | ||
09f037aa PB |
4668 | #define BYTE_MASK(access) \ |
4669 | ((1 & (access) ? 2 : 0) | \ | |
4670 | (2 & (access) ? 4 : 0) | \ | |
4671 | (3 & (access) ? 8 : 0) | \ | |
4672 | (4 & (access) ? 16 : 0) | \ | |
4673 | (5 & (access) ? 32 : 0) | \ | |
4674 | (6 & (access) ? 64 : 0) | \ | |
4675 | (7 & (access) ? 128 : 0)) | |
4676 | ||
4677 | ||
edc90b7d XG |
4678 | static void update_permission_bitmask(struct kvm_vcpu *vcpu, |
4679 | struct kvm_mmu *mmu, bool ept) | |
97d64b78 | 4680 | { |
09f037aa PB |
4681 | unsigned byte; |
4682 | ||
4683 | const u8 x = BYTE_MASK(ACC_EXEC_MASK); | |
4684 | const u8 w = BYTE_MASK(ACC_WRITE_MASK); | |
4685 | const u8 u = BYTE_MASK(ACC_USER_MASK); | |
4686 | ||
4687 | bool cr4_smep = kvm_read_cr4_bits(vcpu, X86_CR4_SMEP) != 0; | |
4688 | bool cr4_smap = kvm_read_cr4_bits(vcpu, X86_CR4_SMAP) != 0; | |
4689 | bool cr0_wp = is_write_protection(vcpu); | |
97d64b78 | 4690 | |
97d64b78 | 4691 | for (byte = 0; byte < ARRAY_SIZE(mmu->permissions); ++byte) { |
09f037aa PB |
4692 | unsigned pfec = byte << 1; |
4693 | ||
97ec8c06 | 4694 | /* |
09f037aa PB |
4695 | * Each "*f" variable has a 1 bit for each UWX value |
4696 | * that causes a fault with the given PFEC. | |
97ec8c06 | 4697 | */ |
97d64b78 | 4698 | |
09f037aa | 4699 | /* Faults from writes to non-writable pages */ |
a6a6d3b1 | 4700 | u8 wf = (pfec & PFERR_WRITE_MASK) ? (u8)~w : 0; |
09f037aa | 4701 | /* Faults from user mode accesses to supervisor pages */ |
a6a6d3b1 | 4702 | u8 uf = (pfec & PFERR_USER_MASK) ? (u8)~u : 0; |
09f037aa | 4703 | /* Faults from fetches of non-executable pages*/ |
a6a6d3b1 | 4704 | u8 ff = (pfec & PFERR_FETCH_MASK) ? (u8)~x : 0; |
09f037aa PB |
4705 | /* Faults from kernel mode fetches of user pages */ |
4706 | u8 smepf = 0; | |
4707 | /* Faults from kernel mode accesses of user pages */ | |
4708 | u8 smapf = 0; | |
4709 | ||
4710 | if (!ept) { | |
4711 | /* Faults from kernel mode accesses to user pages */ | |
4712 | u8 kf = (pfec & PFERR_USER_MASK) ? 0 : u; | |
4713 | ||
4714 | /* Not really needed: !nx will cause pte.nx to fault */ | |
4715 | if (!mmu->nx) | |
4716 | ff = 0; | |
4717 | ||
4718 | /* Allow supervisor writes if !cr0.wp */ | |
4719 | if (!cr0_wp) | |
4720 | wf = (pfec & PFERR_USER_MASK) ? wf : 0; | |
4721 | ||
4722 | /* Disallow supervisor fetches of user code if cr4.smep */ | |
4723 | if (cr4_smep) | |
4724 | smepf = (pfec & PFERR_FETCH_MASK) ? kf : 0; | |
4725 | ||
4726 | /* | |
4727 | * SMAP:kernel-mode data accesses from user-mode | |
4728 | * mappings should fault. A fault is considered | |
4729 | * as a SMAP violation if all of the following | |
39337ad1 | 4730 | * conditions are true: |
09f037aa PB |
4731 | * - X86_CR4_SMAP is set in CR4 |
4732 | * - A user page is accessed | |
4733 | * - The access is not a fetch | |
4734 | * - Page fault in kernel mode | |
4735 | * - if CPL = 3 or X86_EFLAGS_AC is clear | |
4736 | * | |
4737 | * Here, we cover the first three conditions. | |
4738 | * The fourth is computed dynamically in permission_fault(); | |
4739 | * PFERR_RSVD_MASK bit will be set in PFEC if the access is | |
4740 | * *not* subject to SMAP restrictions. | |
4741 | */ | |
4742 | if (cr4_smap) | |
4743 | smapf = (pfec & (PFERR_RSVD_MASK|PFERR_FETCH_MASK)) ? 0 : kf; | |
97d64b78 | 4744 | } |
09f037aa PB |
4745 | |
4746 | mmu->permissions[byte] = ff | uf | wf | smepf | smapf; | |
97d64b78 AK |
4747 | } |
4748 | } | |
4749 | ||
2d344105 HH |
4750 | /* |
4751 | * PKU is an additional mechanism by which the paging controls access to | |
4752 | * user-mode addresses based on the value in the PKRU register. Protection | |
4753 | * key violations are reported through a bit in the page fault error code. | |
4754 | * Unlike other bits of the error code, the PK bit is not known at the | |
4755 | * call site of e.g. gva_to_gpa; it must be computed directly in | |
4756 | * permission_fault based on two bits of PKRU, on some machine state (CR4, | |
4757 | * CR0, EFER, CPL), and on other bits of the error code and the page tables. | |
4758 | * | |
4759 | * In particular the following conditions come from the error code, the | |
4760 | * page tables and the machine state: | |
4761 | * - PK is always zero unless CR4.PKE=1 and EFER.LMA=1 | |
4762 | * - PK is always zero if RSVD=1 (reserved bit set) or F=1 (instruction fetch) | |
4763 | * - PK is always zero if U=0 in the page tables | |
4764 | * - PKRU.WD is ignored if CR0.WP=0 and the access is a supervisor access. | |
4765 | * | |
4766 | * The PKRU bitmask caches the result of these four conditions. The error | |
4767 | * code (minus the P bit) and the page table's U bit form an index into the | |
4768 | * PKRU bitmask. Two bits of the PKRU bitmask are then extracted and ANDed | |
4769 | * with the two bits of the PKRU register corresponding to the protection key. | |
4770 | * For the first three conditions above the bits will be 00, thus masking | |
4771 | * away both AD and WD. For all reads or if the last condition holds, WD | |
4772 | * only will be masked away. | |
4773 | */ | |
4774 | static void update_pkru_bitmask(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, | |
4775 | bool ept) | |
4776 | { | |
4777 | unsigned bit; | |
4778 | bool wp; | |
4779 | ||
4780 | if (ept) { | |
4781 | mmu->pkru_mask = 0; | |
4782 | return; | |
4783 | } | |
4784 | ||
4785 | /* PKEY is enabled only if CR4.PKE and EFER.LMA are both set. */ | |
4786 | if (!kvm_read_cr4_bits(vcpu, X86_CR4_PKE) || !is_long_mode(vcpu)) { | |
4787 | mmu->pkru_mask = 0; | |
4788 | return; | |
4789 | } | |
4790 | ||
4791 | wp = is_write_protection(vcpu); | |
4792 | ||
4793 | for (bit = 0; bit < ARRAY_SIZE(mmu->permissions); ++bit) { | |
4794 | unsigned pfec, pkey_bits; | |
4795 | bool check_pkey, check_write, ff, uf, wf, pte_user; | |
4796 | ||
4797 | pfec = bit << 1; | |
4798 | ff = pfec & PFERR_FETCH_MASK; | |
4799 | uf = pfec & PFERR_USER_MASK; | |
4800 | wf = pfec & PFERR_WRITE_MASK; | |
4801 | ||
4802 | /* PFEC.RSVD is replaced by ACC_USER_MASK. */ | |
4803 | pte_user = pfec & PFERR_RSVD_MASK; | |
4804 | ||
4805 | /* | |
4806 | * Only need to check the access which is not an | |
4807 | * instruction fetch and is to a user page. | |
4808 | */ | |
4809 | check_pkey = (!ff && pte_user); | |
4810 | /* | |
4811 | * write access is controlled by PKRU if it is a | |
4812 | * user access or CR0.WP = 1. | |
4813 | */ | |
4814 | check_write = check_pkey && wf && (uf || wp); | |
4815 | ||
4816 | /* PKRU.AD stops both read and write access. */ | |
4817 | pkey_bits = !!check_pkey; | |
4818 | /* PKRU.WD stops write access. */ | |
4819 | pkey_bits |= (!!check_write) << 1; | |
4820 | ||
4821 | mmu->pkru_mask |= (pkey_bits & 3) << pfec; | |
4822 | } | |
4823 | } | |
4824 | ||
6bb69c9b | 4825 | static void update_last_nonleaf_level(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu) |
6fd01b71 | 4826 | { |
6bb69c9b PB |
4827 | unsigned root_level = mmu->root_level; |
4828 | ||
4829 | mmu->last_nonleaf_level = root_level; | |
4830 | if (root_level == PT32_ROOT_LEVEL && is_pse(vcpu)) | |
4831 | mmu->last_nonleaf_level++; | |
6fd01b71 AK |
4832 | } |
4833 | ||
8a3c1a33 PB |
4834 | static void paging64_init_context_common(struct kvm_vcpu *vcpu, |
4835 | struct kvm_mmu *context, | |
4836 | int level) | |
6aa8b732 | 4837 | { |
2d48a985 | 4838 | context->nx = is_nx(vcpu); |
4d6931c3 | 4839 | context->root_level = level; |
2d48a985 | 4840 | |
4d6931c3 | 4841 | reset_rsvds_bits_mask(vcpu, context); |
25d92081 | 4842 | update_permission_bitmask(vcpu, context, false); |
2d344105 | 4843 | update_pkru_bitmask(vcpu, context, false); |
6bb69c9b | 4844 | update_last_nonleaf_level(vcpu, context); |
6aa8b732 | 4845 | |
fa4a2c08 | 4846 | MMU_WARN_ON(!is_pae(vcpu)); |
6aa8b732 | 4847 | context->page_fault = paging64_page_fault; |
6aa8b732 | 4848 | context->gva_to_gpa = paging64_gva_to_gpa; |
e8bc217a | 4849 | context->sync_page = paging64_sync_page; |
a7052897 | 4850 | context->invlpg = paging64_invlpg; |
0f53b5b1 | 4851 | context->update_pte = paging64_update_pte; |
17ac10ad | 4852 | context->shadow_root_level = level; |
c5a78f2b | 4853 | context->direct_map = false; |
6aa8b732 AK |
4854 | } |
4855 | ||
8a3c1a33 PB |
4856 | static void paging64_init_context(struct kvm_vcpu *vcpu, |
4857 | struct kvm_mmu *context) | |
17ac10ad | 4858 | { |
855feb67 YZ |
4859 | int root_level = is_la57_mode(vcpu) ? |
4860 | PT64_ROOT_5LEVEL : PT64_ROOT_4LEVEL; | |
4861 | ||
4862 | paging64_init_context_common(vcpu, context, root_level); | |
17ac10ad AK |
4863 | } |
4864 | ||
8a3c1a33 PB |
4865 | static void paging32_init_context(struct kvm_vcpu *vcpu, |
4866 | struct kvm_mmu *context) | |
6aa8b732 | 4867 | { |
2d48a985 | 4868 | context->nx = false; |
4d6931c3 | 4869 | context->root_level = PT32_ROOT_LEVEL; |
2d48a985 | 4870 | |
4d6931c3 | 4871 | reset_rsvds_bits_mask(vcpu, context); |
25d92081 | 4872 | update_permission_bitmask(vcpu, context, false); |
2d344105 | 4873 | update_pkru_bitmask(vcpu, context, false); |
6bb69c9b | 4874 | update_last_nonleaf_level(vcpu, context); |
6aa8b732 | 4875 | |
6aa8b732 | 4876 | context->page_fault = paging32_page_fault; |
6aa8b732 | 4877 | context->gva_to_gpa = paging32_gva_to_gpa; |
e8bc217a | 4878 | context->sync_page = paging32_sync_page; |
a7052897 | 4879 | context->invlpg = paging32_invlpg; |
0f53b5b1 | 4880 | context->update_pte = paging32_update_pte; |
6aa8b732 | 4881 | context->shadow_root_level = PT32E_ROOT_LEVEL; |
c5a78f2b | 4882 | context->direct_map = false; |
6aa8b732 AK |
4883 | } |
4884 | ||
8a3c1a33 PB |
4885 | static void paging32E_init_context(struct kvm_vcpu *vcpu, |
4886 | struct kvm_mmu *context) | |
6aa8b732 | 4887 | { |
8a3c1a33 | 4888 | paging64_init_context_common(vcpu, context, PT32E_ROOT_LEVEL); |
6aa8b732 AK |
4889 | } |
4890 | ||
a336282d VK |
4891 | static union kvm_mmu_extended_role kvm_calc_mmu_role_ext(struct kvm_vcpu *vcpu) |
4892 | { | |
4893 | union kvm_mmu_extended_role ext = {0}; | |
4894 | ||
7dcd5755 | 4895 | ext.cr0_pg = !!is_paging(vcpu); |
0699c64a | 4896 | ext.cr4_pae = !!is_pae(vcpu); |
a336282d VK |
4897 | ext.cr4_smep = !!kvm_read_cr4_bits(vcpu, X86_CR4_SMEP); |
4898 | ext.cr4_smap = !!kvm_read_cr4_bits(vcpu, X86_CR4_SMAP); | |
4899 | ext.cr4_pse = !!is_pse(vcpu); | |
4900 | ext.cr4_pke = !!kvm_read_cr4_bits(vcpu, X86_CR4_PKE); | |
de3ccd26 | 4901 | ext.maxphyaddr = cpuid_maxphyaddr(vcpu); |
a336282d VK |
4902 | |
4903 | ext.valid = 1; | |
4904 | ||
4905 | return ext; | |
4906 | } | |
4907 | ||
7dcd5755 VK |
4908 | static union kvm_mmu_role kvm_calc_mmu_role_common(struct kvm_vcpu *vcpu, |
4909 | bool base_only) | |
4910 | { | |
4911 | union kvm_mmu_role role = {0}; | |
4912 | ||
4913 | role.base.access = ACC_ALL; | |
4914 | role.base.nxe = !!is_nx(vcpu); | |
7dcd5755 VK |
4915 | role.base.cr0_wp = is_write_protection(vcpu); |
4916 | role.base.smm = is_smm(vcpu); | |
4917 | role.base.guest_mode = is_guest_mode(vcpu); | |
4918 | ||
4919 | if (base_only) | |
4920 | return role; | |
4921 | ||
4922 | role.ext = kvm_calc_mmu_role_ext(vcpu); | |
4923 | ||
4924 | return role; | |
4925 | } | |
4926 | ||
d468d94b SC |
4927 | static inline int kvm_mmu_get_tdp_level(struct kvm_vcpu *vcpu) |
4928 | { | |
4929 | /* Use 5-level TDP if and only if it's useful/necessary. */ | |
83013059 | 4930 | if (max_tdp_level == 5 && cpuid_maxphyaddr(vcpu) <= 48) |
d468d94b SC |
4931 | return 4; |
4932 | ||
83013059 | 4933 | return max_tdp_level; |
d468d94b SC |
4934 | } |
4935 | ||
7dcd5755 VK |
4936 | static union kvm_mmu_role |
4937 | kvm_calc_tdp_mmu_root_page_role(struct kvm_vcpu *vcpu, bool base_only) | |
9fa72119 | 4938 | { |
7dcd5755 | 4939 | union kvm_mmu_role role = kvm_calc_mmu_role_common(vcpu, base_only); |
9fa72119 | 4940 | |
7dcd5755 | 4941 | role.base.ad_disabled = (shadow_accessed_mask == 0); |
d468d94b | 4942 | role.base.level = kvm_mmu_get_tdp_level(vcpu); |
7dcd5755 | 4943 | role.base.direct = true; |
47c42e6b | 4944 | role.base.gpte_is_8_bytes = true; |
9fa72119 JS |
4945 | |
4946 | return role; | |
4947 | } | |
4948 | ||
8a3c1a33 | 4949 | static void init_kvm_tdp_mmu(struct kvm_vcpu *vcpu) |
fb72d167 | 4950 | { |
8c008659 | 4951 | struct kvm_mmu *context = &vcpu->arch.root_mmu; |
7dcd5755 VK |
4952 | union kvm_mmu_role new_role = |
4953 | kvm_calc_tdp_mmu_root_page_role(vcpu, false); | |
fb72d167 | 4954 | |
7dcd5755 VK |
4955 | if (new_role.as_u64 == context->mmu_role.as_u64) |
4956 | return; | |
4957 | ||
4958 | context->mmu_role.as_u64 = new_role.as_u64; | |
7a02674d | 4959 | context->page_fault = kvm_tdp_page_fault; |
e8bc217a | 4960 | context->sync_page = nonpaging_sync_page; |
5efac074 | 4961 | context->invlpg = NULL; |
0f53b5b1 | 4962 | context->update_pte = nonpaging_update_pte; |
d468d94b | 4963 | context->shadow_root_level = kvm_mmu_get_tdp_level(vcpu); |
c5a78f2b | 4964 | context->direct_map = true; |
d8dd54e0 | 4965 | context->get_guest_pgd = get_cr3; |
e4e517b4 | 4966 | context->get_pdptr = kvm_pdptr_read; |
cb659db8 | 4967 | context->inject_page_fault = kvm_inject_page_fault; |
fb72d167 JR |
4968 | |
4969 | if (!is_paging(vcpu)) { | |
2d48a985 | 4970 | context->nx = false; |
fb72d167 JR |
4971 | context->gva_to_gpa = nonpaging_gva_to_gpa; |
4972 | context->root_level = 0; | |
4973 | } else if (is_long_mode(vcpu)) { | |
2d48a985 | 4974 | context->nx = is_nx(vcpu); |
855feb67 YZ |
4975 | context->root_level = is_la57_mode(vcpu) ? |
4976 | PT64_ROOT_5LEVEL : PT64_ROOT_4LEVEL; | |
4d6931c3 DB |
4977 | reset_rsvds_bits_mask(vcpu, context); |
4978 | context->gva_to_gpa = paging64_gva_to_gpa; | |
fb72d167 | 4979 | } else if (is_pae(vcpu)) { |
2d48a985 | 4980 | context->nx = is_nx(vcpu); |
fb72d167 | 4981 | context->root_level = PT32E_ROOT_LEVEL; |
4d6931c3 DB |
4982 | reset_rsvds_bits_mask(vcpu, context); |
4983 | context->gva_to_gpa = paging64_gva_to_gpa; | |
fb72d167 | 4984 | } else { |
2d48a985 | 4985 | context->nx = false; |
fb72d167 | 4986 | context->root_level = PT32_ROOT_LEVEL; |
4d6931c3 DB |
4987 | reset_rsvds_bits_mask(vcpu, context); |
4988 | context->gva_to_gpa = paging32_gva_to_gpa; | |
fb72d167 JR |
4989 | } |
4990 | ||
25d92081 | 4991 | update_permission_bitmask(vcpu, context, false); |
2d344105 | 4992 | update_pkru_bitmask(vcpu, context, false); |
6bb69c9b | 4993 | update_last_nonleaf_level(vcpu, context); |
c258b62b | 4994 | reset_tdp_shadow_zero_bits_mask(vcpu, context); |
fb72d167 JR |
4995 | } |
4996 | ||
7dcd5755 | 4997 | static union kvm_mmu_role |
59505b55 | 4998 | kvm_calc_shadow_root_page_role_common(struct kvm_vcpu *vcpu, bool base_only) |
7dcd5755 VK |
4999 | { |
5000 | union kvm_mmu_role role = kvm_calc_mmu_role_common(vcpu, base_only); | |
5001 | ||
5002 | role.base.smep_andnot_wp = role.ext.cr4_smep && | |
5003 | !is_write_protection(vcpu); | |
5004 | role.base.smap_andnot_wp = role.ext.cr4_smap && | |
5005 | !is_write_protection(vcpu); | |
47c42e6b | 5006 | role.base.gpte_is_8_bytes = !!is_pae(vcpu); |
9fa72119 | 5007 | |
59505b55 SC |
5008 | return role; |
5009 | } | |
5010 | ||
5011 | static union kvm_mmu_role | |
5012 | kvm_calc_shadow_mmu_root_page_role(struct kvm_vcpu *vcpu, bool base_only) | |
5013 | { | |
5014 | union kvm_mmu_role role = | |
5015 | kvm_calc_shadow_root_page_role_common(vcpu, base_only); | |
5016 | ||
5017 | role.base.direct = !is_paging(vcpu); | |
5018 | ||
9fa72119 | 5019 | if (!is_long_mode(vcpu)) |
7dcd5755 | 5020 | role.base.level = PT32E_ROOT_LEVEL; |
9fa72119 | 5021 | else if (is_la57_mode(vcpu)) |
7dcd5755 | 5022 | role.base.level = PT64_ROOT_5LEVEL; |
9fa72119 | 5023 | else |
7dcd5755 | 5024 | role.base.level = PT64_ROOT_4LEVEL; |
9fa72119 JS |
5025 | |
5026 | return role; | |
5027 | } | |
5028 | ||
8c008659 PB |
5029 | static void shadow_mmu_init_context(struct kvm_vcpu *vcpu, struct kvm_mmu *context, |
5030 | u32 cr0, u32 cr4, u32 efer, | |
5031 | union kvm_mmu_role new_role) | |
9fa72119 | 5032 | { |
929d1cfa | 5033 | if (!(cr0 & X86_CR0_PG)) |
8a3c1a33 | 5034 | nonpaging_init_context(vcpu, context); |
929d1cfa | 5035 | else if (efer & EFER_LMA) |
8a3c1a33 | 5036 | paging64_init_context(vcpu, context); |
929d1cfa | 5037 | else if (cr4 & X86_CR4_PAE) |
8a3c1a33 | 5038 | paging32E_init_context(vcpu, context); |
6aa8b732 | 5039 | else |
8a3c1a33 | 5040 | paging32_init_context(vcpu, context); |
a770f6f2 | 5041 | |
7dcd5755 | 5042 | context->mmu_role.as_u64 = new_role.as_u64; |
c258b62b | 5043 | reset_shadow_zero_bits_mask(vcpu, context); |
52fde8df | 5044 | } |
0f04a2ac VK |
5045 | |
5046 | static void kvm_init_shadow_mmu(struct kvm_vcpu *vcpu, u32 cr0, u32 cr4, u32 efer) | |
5047 | { | |
8c008659 | 5048 | struct kvm_mmu *context = &vcpu->arch.root_mmu; |
0f04a2ac VK |
5049 | union kvm_mmu_role new_role = |
5050 | kvm_calc_shadow_mmu_root_page_role(vcpu, false); | |
5051 | ||
5052 | if (new_role.as_u64 != context->mmu_role.as_u64) | |
8c008659 | 5053 | shadow_mmu_init_context(vcpu, context, cr0, cr4, efer, new_role); |
0f04a2ac VK |
5054 | } |
5055 | ||
59505b55 SC |
5056 | static union kvm_mmu_role |
5057 | kvm_calc_shadow_npt_root_page_role(struct kvm_vcpu *vcpu) | |
5058 | { | |
5059 | union kvm_mmu_role role = | |
5060 | kvm_calc_shadow_root_page_role_common(vcpu, false); | |
5061 | ||
5062 | role.base.direct = false; | |
d468d94b | 5063 | role.base.level = kvm_mmu_get_tdp_level(vcpu); |
59505b55 SC |
5064 | |
5065 | return role; | |
5066 | } | |
5067 | ||
0f04a2ac VK |
5068 | void kvm_init_shadow_npt_mmu(struct kvm_vcpu *vcpu, u32 cr0, u32 cr4, u32 efer, |
5069 | gpa_t nested_cr3) | |
5070 | { | |
8c008659 | 5071 | struct kvm_mmu *context = &vcpu->arch.guest_mmu; |
59505b55 | 5072 | union kvm_mmu_role new_role = kvm_calc_shadow_npt_root_page_role(vcpu); |
0f04a2ac | 5073 | |
096586fd SC |
5074 | context->shadow_root_level = new_role.base.level; |
5075 | ||
a506fdd2 VK |
5076 | __kvm_mmu_new_pgd(vcpu, nested_cr3, new_role.base, false, false); |
5077 | ||
0f04a2ac | 5078 | if (new_role.as_u64 != context->mmu_role.as_u64) |
8c008659 | 5079 | shadow_mmu_init_context(vcpu, context, cr0, cr4, efer, new_role); |
0f04a2ac VK |
5080 | } |
5081 | EXPORT_SYMBOL_GPL(kvm_init_shadow_npt_mmu); | |
52fde8df | 5082 | |
a336282d VK |
5083 | static union kvm_mmu_role |
5084 | kvm_calc_shadow_ept_root_page_role(struct kvm_vcpu *vcpu, bool accessed_dirty, | |
bb1fcc70 | 5085 | bool execonly, u8 level) |
9fa72119 | 5086 | { |
552c69b1 | 5087 | union kvm_mmu_role role = {0}; |
14c07ad8 | 5088 | |
47c42e6b SC |
5089 | /* SMM flag is inherited from root_mmu */ |
5090 | role.base.smm = vcpu->arch.root_mmu.mmu_role.base.smm; | |
9fa72119 | 5091 | |
bb1fcc70 | 5092 | role.base.level = level; |
47c42e6b | 5093 | role.base.gpte_is_8_bytes = true; |
a336282d VK |
5094 | role.base.direct = false; |
5095 | role.base.ad_disabled = !accessed_dirty; | |
5096 | role.base.guest_mode = true; | |
5097 | role.base.access = ACC_ALL; | |
9fa72119 | 5098 | |
47c42e6b SC |
5099 | /* |
5100 | * WP=1 and NOT_WP=1 is an impossible combination, use WP and the | |
5101 | * SMAP variation to denote shadow EPT entries. | |
5102 | */ | |
5103 | role.base.cr0_wp = true; | |
5104 | role.base.smap_andnot_wp = true; | |
5105 | ||
552c69b1 | 5106 | role.ext = kvm_calc_mmu_role_ext(vcpu); |
a336282d | 5107 | role.ext.execonly = execonly; |
9fa72119 JS |
5108 | |
5109 | return role; | |
5110 | } | |
5111 | ||
ae1e2d10 | 5112 | void kvm_init_shadow_ept_mmu(struct kvm_vcpu *vcpu, bool execonly, |
50c28f21 | 5113 | bool accessed_dirty, gpa_t new_eptp) |
155a97a3 | 5114 | { |
8c008659 | 5115 | struct kvm_mmu *context = &vcpu->arch.guest_mmu; |
bb1fcc70 | 5116 | u8 level = vmx_eptp_page_walk_level(new_eptp); |
a336282d VK |
5117 | union kvm_mmu_role new_role = |
5118 | kvm_calc_shadow_ept_root_page_role(vcpu, accessed_dirty, | |
bb1fcc70 | 5119 | execonly, level); |
a336282d | 5120 | |
be01e8e2 | 5121 | __kvm_mmu_new_pgd(vcpu, new_eptp, new_role.base, true, true); |
a336282d | 5122 | |
a336282d VK |
5123 | if (new_role.as_u64 == context->mmu_role.as_u64) |
5124 | return; | |
ad896af0 | 5125 | |
bb1fcc70 | 5126 | context->shadow_root_level = level; |
155a97a3 NHE |
5127 | |
5128 | context->nx = true; | |
ae1e2d10 | 5129 | context->ept_ad = accessed_dirty; |
155a97a3 NHE |
5130 | context->page_fault = ept_page_fault; |
5131 | context->gva_to_gpa = ept_gva_to_gpa; | |
5132 | context->sync_page = ept_sync_page; | |
5133 | context->invlpg = ept_invlpg; | |
5134 | context->update_pte = ept_update_pte; | |
bb1fcc70 | 5135 | context->root_level = level; |
155a97a3 | 5136 | context->direct_map = false; |
a336282d | 5137 | context->mmu_role.as_u64 = new_role.as_u64; |
3dc773e7 | 5138 | |
155a97a3 | 5139 | update_permission_bitmask(vcpu, context, true); |
2d344105 | 5140 | update_pkru_bitmask(vcpu, context, true); |
fd19d3b4 | 5141 | update_last_nonleaf_level(vcpu, context); |
155a97a3 | 5142 | reset_rsvds_bits_mask_ept(vcpu, context, execonly); |
c258b62b | 5143 | reset_ept_shadow_zero_bits_mask(vcpu, context, execonly); |
155a97a3 NHE |
5144 | } |
5145 | EXPORT_SYMBOL_GPL(kvm_init_shadow_ept_mmu); | |
5146 | ||
8a3c1a33 | 5147 | static void init_kvm_softmmu(struct kvm_vcpu *vcpu) |
52fde8df | 5148 | { |
8c008659 | 5149 | struct kvm_mmu *context = &vcpu->arch.root_mmu; |
ad896af0 | 5150 | |
929d1cfa PB |
5151 | kvm_init_shadow_mmu(vcpu, |
5152 | kvm_read_cr0_bits(vcpu, X86_CR0_PG), | |
5153 | kvm_read_cr4_bits(vcpu, X86_CR4_PAE), | |
5154 | vcpu->arch.efer); | |
5155 | ||
d8dd54e0 | 5156 | context->get_guest_pgd = get_cr3; |
ad896af0 PB |
5157 | context->get_pdptr = kvm_pdptr_read; |
5158 | context->inject_page_fault = kvm_inject_page_fault; | |
6aa8b732 AK |
5159 | } |
5160 | ||
8a3c1a33 | 5161 | static void init_kvm_nested_mmu(struct kvm_vcpu *vcpu) |
02f59dc9 | 5162 | { |
bf627a92 | 5163 | union kvm_mmu_role new_role = kvm_calc_mmu_role_common(vcpu, false); |
02f59dc9 JR |
5164 | struct kvm_mmu *g_context = &vcpu->arch.nested_mmu; |
5165 | ||
bf627a92 VK |
5166 | if (new_role.as_u64 == g_context->mmu_role.as_u64) |
5167 | return; | |
5168 | ||
5169 | g_context->mmu_role.as_u64 = new_role.as_u64; | |
d8dd54e0 | 5170 | g_context->get_guest_pgd = get_cr3; |
e4e517b4 | 5171 | g_context->get_pdptr = kvm_pdptr_read; |
02f59dc9 JR |
5172 | g_context->inject_page_fault = kvm_inject_page_fault; |
5173 | ||
5efac074 PB |
5174 | /* |
5175 | * L2 page tables are never shadowed, so there is no need to sync | |
5176 | * SPTEs. | |
5177 | */ | |
5178 | g_context->invlpg = NULL; | |
5179 | ||
02f59dc9 | 5180 | /* |
44dd3ffa | 5181 | * Note that arch.mmu->gva_to_gpa translates l2_gpa to l1_gpa using |
0af2593b DM |
5182 | * L1's nested page tables (e.g. EPT12). The nested translation |
5183 | * of l2_gva to l1_gpa is done by arch.nested_mmu.gva_to_gpa using | |
5184 | * L2's page tables as the first level of translation and L1's | |
5185 | * nested page tables as the second level of translation. Basically | |
5186 | * the gva_to_gpa functions between mmu and nested_mmu are swapped. | |
02f59dc9 JR |
5187 | */ |
5188 | if (!is_paging(vcpu)) { | |
2d48a985 | 5189 | g_context->nx = false; |
02f59dc9 JR |
5190 | g_context->root_level = 0; |
5191 | g_context->gva_to_gpa = nonpaging_gva_to_gpa_nested; | |
5192 | } else if (is_long_mode(vcpu)) { | |
2d48a985 | 5193 | g_context->nx = is_nx(vcpu); |
855feb67 YZ |
5194 | g_context->root_level = is_la57_mode(vcpu) ? |
5195 | PT64_ROOT_5LEVEL : PT64_ROOT_4LEVEL; | |
4d6931c3 | 5196 | reset_rsvds_bits_mask(vcpu, g_context); |
02f59dc9 JR |
5197 | g_context->gva_to_gpa = paging64_gva_to_gpa_nested; |
5198 | } else if (is_pae(vcpu)) { | |
2d48a985 | 5199 | g_context->nx = is_nx(vcpu); |
02f59dc9 | 5200 | g_context->root_level = PT32E_ROOT_LEVEL; |
4d6931c3 | 5201 | reset_rsvds_bits_mask(vcpu, g_context); |
02f59dc9 JR |
5202 | g_context->gva_to_gpa = paging64_gva_to_gpa_nested; |
5203 | } else { | |
2d48a985 | 5204 | g_context->nx = false; |
02f59dc9 | 5205 | g_context->root_level = PT32_ROOT_LEVEL; |
4d6931c3 | 5206 | reset_rsvds_bits_mask(vcpu, g_context); |
02f59dc9 JR |
5207 | g_context->gva_to_gpa = paging32_gva_to_gpa_nested; |
5208 | } | |
5209 | ||
25d92081 | 5210 | update_permission_bitmask(vcpu, g_context, false); |
2d344105 | 5211 | update_pkru_bitmask(vcpu, g_context, false); |
6bb69c9b | 5212 | update_last_nonleaf_level(vcpu, g_context); |
02f59dc9 JR |
5213 | } |
5214 | ||
1c53da3f | 5215 | void kvm_init_mmu(struct kvm_vcpu *vcpu, bool reset_roots) |
fb72d167 | 5216 | { |
1c53da3f | 5217 | if (reset_roots) { |
b94742c9 JS |
5218 | uint i; |
5219 | ||
44dd3ffa | 5220 | vcpu->arch.mmu->root_hpa = INVALID_PAGE; |
b94742c9 JS |
5221 | |
5222 | for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) | |
44dd3ffa | 5223 | vcpu->arch.mmu->prev_roots[i] = KVM_MMU_ROOT_INFO_INVALID; |
1c53da3f JS |
5224 | } |
5225 | ||
02f59dc9 | 5226 | if (mmu_is_nested(vcpu)) |
e0c6db3e | 5227 | init_kvm_nested_mmu(vcpu); |
02f59dc9 | 5228 | else if (tdp_enabled) |
e0c6db3e | 5229 | init_kvm_tdp_mmu(vcpu); |
fb72d167 | 5230 | else |
e0c6db3e | 5231 | init_kvm_softmmu(vcpu); |
fb72d167 | 5232 | } |
1c53da3f | 5233 | EXPORT_SYMBOL_GPL(kvm_init_mmu); |
fb72d167 | 5234 | |
9fa72119 JS |
5235 | static union kvm_mmu_page_role |
5236 | kvm_mmu_calc_root_page_role(struct kvm_vcpu *vcpu) | |
5237 | { | |
7dcd5755 VK |
5238 | union kvm_mmu_role role; |
5239 | ||
9fa72119 | 5240 | if (tdp_enabled) |
7dcd5755 | 5241 | role = kvm_calc_tdp_mmu_root_page_role(vcpu, true); |
9fa72119 | 5242 | else |
7dcd5755 VK |
5243 | role = kvm_calc_shadow_mmu_root_page_role(vcpu, true); |
5244 | ||
5245 | return role.base; | |
9fa72119 | 5246 | } |
fb72d167 | 5247 | |
8a3c1a33 | 5248 | void kvm_mmu_reset_context(struct kvm_vcpu *vcpu) |
6aa8b732 | 5249 | { |
95f93af4 | 5250 | kvm_mmu_unload(vcpu); |
1c53da3f | 5251 | kvm_init_mmu(vcpu, true); |
17c3ba9d | 5252 | } |
8668a3c4 | 5253 | EXPORT_SYMBOL_GPL(kvm_mmu_reset_context); |
17c3ba9d AK |
5254 | |
5255 | int kvm_mmu_load(struct kvm_vcpu *vcpu) | |
6aa8b732 | 5256 | { |
714b93da AK |
5257 | int r; |
5258 | ||
378f5cd6 | 5259 | r = mmu_topup_memory_caches(vcpu, !vcpu->arch.mmu->direct_map); |
17c3ba9d AK |
5260 | if (r) |
5261 | goto out; | |
8986ecc0 | 5262 | r = mmu_alloc_roots(vcpu); |
e2858b4a | 5263 | kvm_mmu_sync_roots(vcpu); |
8986ecc0 MT |
5264 | if (r) |
5265 | goto out; | |
727a7e27 | 5266 | kvm_mmu_load_pgd(vcpu); |
8c8560b8 | 5267 | kvm_x86_ops.tlb_flush_current(vcpu); |
714b93da AK |
5268 | out: |
5269 | return r; | |
6aa8b732 | 5270 | } |
17c3ba9d AK |
5271 | EXPORT_SYMBOL_GPL(kvm_mmu_load); |
5272 | ||
5273 | void kvm_mmu_unload(struct kvm_vcpu *vcpu) | |
5274 | { | |
14c07ad8 VK |
5275 | kvm_mmu_free_roots(vcpu, &vcpu->arch.root_mmu, KVM_MMU_ROOTS_ALL); |
5276 | WARN_ON(VALID_PAGE(vcpu->arch.root_mmu.root_hpa)); | |
5277 | kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL); | |
5278 | WARN_ON(VALID_PAGE(vcpu->arch.guest_mmu.root_hpa)); | |
17c3ba9d | 5279 | } |
4b16184c | 5280 | EXPORT_SYMBOL_GPL(kvm_mmu_unload); |
6aa8b732 | 5281 | |
0028425f | 5282 | static void mmu_pte_write_new_pte(struct kvm_vcpu *vcpu, |
7c562522 XG |
5283 | struct kvm_mmu_page *sp, u64 *spte, |
5284 | const void *new) | |
0028425f | 5285 | { |
3bae0459 | 5286 | if (sp->role.level != PG_LEVEL_4K) { |
7e4e4056 JR |
5287 | ++vcpu->kvm->stat.mmu_pde_zapped; |
5288 | return; | |
30945387 | 5289 | } |
0028425f | 5290 | |
4cee5764 | 5291 | ++vcpu->kvm->stat.mmu_pte_updated; |
44dd3ffa | 5292 | vcpu->arch.mmu->update_pte(vcpu, sp, spte, new); |
0028425f AK |
5293 | } |
5294 | ||
79539cec AK |
5295 | static bool need_remote_flush(u64 old, u64 new) |
5296 | { | |
5297 | if (!is_shadow_present_pte(old)) | |
5298 | return false; | |
5299 | if (!is_shadow_present_pte(new)) | |
5300 | return true; | |
5301 | if ((old ^ new) & PT64_BASE_ADDR_MASK) | |
5302 | return true; | |
53166229 GN |
5303 | old ^= shadow_nx_mask; |
5304 | new ^= shadow_nx_mask; | |
79539cec AK |
5305 | return (old & ~new & PT64_PERM_MASK) != 0; |
5306 | } | |
5307 | ||
889e5cbc | 5308 | static u64 mmu_pte_write_fetch_gpte(struct kvm_vcpu *vcpu, gpa_t *gpa, |
0e0fee5c | 5309 | int *bytes) |
da4a00f0 | 5310 | { |
0e0fee5c | 5311 | u64 gentry = 0; |
889e5cbc | 5312 | int r; |
72016f3a | 5313 | |
72016f3a AK |
5314 | /* |
5315 | * Assume that the pte write on a page table of the same type | |
49b26e26 XG |
5316 | * as the current vcpu paging mode since we update the sptes only |
5317 | * when they have the same mode. | |
72016f3a | 5318 | */ |
889e5cbc | 5319 | if (is_pae(vcpu) && *bytes == 4) { |
72016f3a | 5320 | /* Handle a 32-bit guest writing two halves of a 64-bit gpte */ |
889e5cbc XG |
5321 | *gpa &= ~(gpa_t)7; |
5322 | *bytes = 8; | |
08e850c6 AK |
5323 | } |
5324 | ||
0e0fee5c JS |
5325 | if (*bytes == 4 || *bytes == 8) { |
5326 | r = kvm_vcpu_read_guest_atomic(vcpu, *gpa, &gentry, *bytes); | |
5327 | if (r) | |
5328 | gentry = 0; | |
72016f3a AK |
5329 | } |
5330 | ||
889e5cbc XG |
5331 | return gentry; |
5332 | } | |
5333 | ||
5334 | /* | |
5335 | * If we're seeing too many writes to a page, it may no longer be a page table, | |
5336 | * or we may be forking, in which case it is better to unmap the page. | |
5337 | */ | |
a138fe75 | 5338 | static bool detect_write_flooding(struct kvm_mmu_page *sp) |
889e5cbc | 5339 | { |
a30f47cb XG |
5340 | /* |
5341 | * Skip write-flooding detected for the sp whose level is 1, because | |
5342 | * it can become unsync, then the guest page is not write-protected. | |
5343 | */ | |
3bae0459 | 5344 | if (sp->role.level == PG_LEVEL_4K) |
a30f47cb | 5345 | return false; |
3246af0e | 5346 | |
e5691a81 XG |
5347 | atomic_inc(&sp->write_flooding_count); |
5348 | return atomic_read(&sp->write_flooding_count) >= 3; | |
889e5cbc XG |
5349 | } |
5350 | ||
5351 | /* | |
5352 | * Misaligned accesses are too much trouble to fix up; also, they usually | |
5353 | * indicate a page is not used as a page table. | |
5354 | */ | |
5355 | static bool detect_write_misaligned(struct kvm_mmu_page *sp, gpa_t gpa, | |
5356 | int bytes) | |
5357 | { | |
5358 | unsigned offset, pte_size, misaligned; | |
5359 | ||
5360 | pgprintk("misaligned: gpa %llx bytes %d role %x\n", | |
5361 | gpa, bytes, sp->role.word); | |
5362 | ||
5363 | offset = offset_in_page(gpa); | |
47c42e6b | 5364 | pte_size = sp->role.gpte_is_8_bytes ? 8 : 4; |
5d9ca30e XG |
5365 | |
5366 | /* | |
5367 | * Sometimes, the OS only writes the last one bytes to update status | |
5368 | * bits, for example, in linux, andb instruction is used in clear_bit(). | |
5369 | */ | |
5370 | if (!(offset & (pte_size - 1)) && bytes == 1) | |
5371 | return false; | |
5372 | ||
889e5cbc XG |
5373 | misaligned = (offset ^ (offset + bytes - 1)) & ~(pte_size - 1); |
5374 | misaligned |= bytes < 4; | |
5375 | ||
5376 | return misaligned; | |
5377 | } | |
5378 | ||
5379 | static u64 *get_written_sptes(struct kvm_mmu_page *sp, gpa_t gpa, int *nspte) | |
5380 | { | |
5381 | unsigned page_offset, quadrant; | |
5382 | u64 *spte; | |
5383 | int level; | |
5384 | ||
5385 | page_offset = offset_in_page(gpa); | |
5386 | level = sp->role.level; | |
5387 | *nspte = 1; | |
47c42e6b | 5388 | if (!sp->role.gpte_is_8_bytes) { |
889e5cbc XG |
5389 | page_offset <<= 1; /* 32->64 */ |
5390 | /* | |
5391 | * A 32-bit pde maps 4MB while the shadow pdes map | |
5392 | * only 2MB. So we need to double the offset again | |
5393 | * and zap two pdes instead of one. | |
5394 | */ | |
5395 | if (level == PT32_ROOT_LEVEL) { | |
5396 | page_offset &= ~7; /* kill rounding error */ | |
5397 | page_offset <<= 1; | |
5398 | *nspte = 2; | |
5399 | } | |
5400 | quadrant = page_offset >> PAGE_SHIFT; | |
5401 | page_offset &= ~PAGE_MASK; | |
5402 | if (quadrant != sp->role.quadrant) | |
5403 | return NULL; | |
5404 | } | |
5405 | ||
5406 | spte = &sp->spt[page_offset / sizeof(*spte)]; | |
5407 | return spte; | |
5408 | } | |
5409 | ||
a102a674 SC |
5410 | /* |
5411 | * Ignore various flags when determining if a SPTE can be immediately | |
5412 | * overwritten for the current MMU. | |
5413 | * - level: explicitly checked in mmu_pte_write_new_pte(), and will never | |
5414 | * match the current MMU role, as MMU's level tracks the root level. | |
5415 | * - access: updated based on the new guest PTE | |
5416 | * - quadrant: handled by get_written_sptes() | |
5417 | * - invalid: always false (loop only walks valid shadow pages) | |
5418 | */ | |
5419 | static const union kvm_mmu_page_role role_ign = { | |
5420 | .level = 0xf, | |
5421 | .access = 0x7, | |
5422 | .quadrant = 0x3, | |
5423 | .invalid = 0x1, | |
5424 | }; | |
5425 | ||
13d268ca | 5426 | static void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa, |
d126363d JS |
5427 | const u8 *new, int bytes, |
5428 | struct kvm_page_track_notifier_node *node) | |
889e5cbc XG |
5429 | { |
5430 | gfn_t gfn = gpa >> PAGE_SHIFT; | |
889e5cbc | 5431 | struct kvm_mmu_page *sp; |
889e5cbc XG |
5432 | LIST_HEAD(invalid_list); |
5433 | u64 entry, gentry, *spte; | |
5434 | int npte; | |
b8c67b7a | 5435 | bool remote_flush, local_flush; |
889e5cbc XG |
5436 | |
5437 | /* | |
5438 | * If we don't have indirect shadow pages, it means no page is | |
5439 | * write-protected, so we can exit simply. | |
5440 | */ | |
6aa7de05 | 5441 | if (!READ_ONCE(vcpu->kvm->arch.indirect_shadow_pages)) |
889e5cbc XG |
5442 | return; |
5443 | ||
b8c67b7a | 5444 | remote_flush = local_flush = false; |
889e5cbc XG |
5445 | |
5446 | pgprintk("%s: gpa %llx bytes %d\n", __func__, gpa, bytes); | |
5447 | ||
889e5cbc XG |
5448 | /* |
5449 | * No need to care whether allocation memory is successful | |
5450 | * or not since pte prefetch is skiped if it does not have | |
5451 | * enough objects in the cache. | |
5452 | */ | |
378f5cd6 | 5453 | mmu_topup_memory_caches(vcpu, true); |
889e5cbc XG |
5454 | |
5455 | spin_lock(&vcpu->kvm->mmu_lock); | |
0e0fee5c JS |
5456 | |
5457 | gentry = mmu_pte_write_fetch_gpte(vcpu, &gpa, &bytes); | |
5458 | ||
889e5cbc | 5459 | ++vcpu->kvm->stat.mmu_pte_write; |
0375f7fa | 5460 | kvm_mmu_audit(vcpu, AUDIT_PRE_PTE_WRITE); |
889e5cbc | 5461 | |
b67bfe0d | 5462 | for_each_gfn_indirect_valid_sp(vcpu->kvm, sp, gfn) { |
a30f47cb | 5463 | if (detect_write_misaligned(sp, gpa, bytes) || |
a138fe75 | 5464 | detect_write_flooding(sp)) { |
b8c67b7a | 5465 | kvm_mmu_prepare_zap_page(vcpu->kvm, sp, &invalid_list); |
4cee5764 | 5466 | ++vcpu->kvm->stat.mmu_flooded; |
0e7bc4b9 AK |
5467 | continue; |
5468 | } | |
889e5cbc XG |
5469 | |
5470 | spte = get_written_sptes(sp, gpa, &npte); | |
5471 | if (!spte) | |
5472 | continue; | |
5473 | ||
0671a8e7 | 5474 | local_flush = true; |
ac1b714e | 5475 | while (npte--) { |
36d9594d VK |
5476 | u32 base_role = vcpu->arch.mmu->mmu_role.base.word; |
5477 | ||
79539cec | 5478 | entry = *spte; |
2de4085c | 5479 | mmu_page_zap_pte(vcpu->kvm, sp, spte, NULL); |
fa1de2bf | 5480 | if (gentry && |
a102a674 SC |
5481 | !((sp->role.word ^ base_role) & ~role_ign.word) && |
5482 | rmap_can_add(vcpu)) | |
7c562522 | 5483 | mmu_pte_write_new_pte(vcpu, sp, spte, &gentry); |
9bb4f6b1 | 5484 | if (need_remote_flush(entry, *spte)) |
0671a8e7 | 5485 | remote_flush = true; |
ac1b714e | 5486 | ++spte; |
9b7a0325 | 5487 | } |
9b7a0325 | 5488 | } |
b8c67b7a | 5489 | kvm_mmu_flush_or_zap(vcpu, &invalid_list, remote_flush, local_flush); |
0375f7fa | 5490 | kvm_mmu_audit(vcpu, AUDIT_POST_PTE_WRITE); |
aaee2c94 | 5491 | spin_unlock(&vcpu->kvm->mmu_lock); |
da4a00f0 AK |
5492 | } |
5493 | ||
a436036b AK |
5494 | int kvm_mmu_unprotect_page_virt(struct kvm_vcpu *vcpu, gva_t gva) |
5495 | { | |
10589a46 MT |
5496 | gpa_t gpa; |
5497 | int r; | |
a436036b | 5498 | |
44dd3ffa | 5499 | if (vcpu->arch.mmu->direct_map) |
60f24784 AK |
5500 | return 0; |
5501 | ||
1871c602 | 5502 | gpa = kvm_mmu_gva_to_gpa_read(vcpu, gva, NULL); |
10589a46 | 5503 | |
10589a46 | 5504 | r = kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT); |
1cb3f3ae | 5505 | |
10589a46 | 5506 | return r; |
a436036b | 5507 | } |
577bdc49 | 5508 | EXPORT_SYMBOL_GPL(kvm_mmu_unprotect_page_virt); |
a436036b | 5509 | |
736c291c | 5510 | int kvm_mmu_page_fault(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa, u64 error_code, |
dc25e89e | 5511 | void *insn, int insn_len) |
3067714c | 5512 | { |
92daa48b | 5513 | int r, emulation_type = EMULTYPE_PF; |
44dd3ffa | 5514 | bool direct = vcpu->arch.mmu->direct_map; |
3067714c | 5515 | |
6948199a | 5516 | if (WARN_ON(!VALID_PAGE(vcpu->arch.mmu->root_hpa))) |
ddce6208 SC |
5517 | return RET_PF_RETRY; |
5518 | ||
9b8ebbdb | 5519 | r = RET_PF_INVALID; |
e9ee956e | 5520 | if (unlikely(error_code & PFERR_RSVD_MASK)) { |
736c291c | 5521 | r = handle_mmio_page_fault(vcpu, cr2_or_gpa, direct); |
472faffa | 5522 | if (r == RET_PF_EMULATE) |
e9ee956e | 5523 | goto emulate; |
e9ee956e | 5524 | } |
3067714c | 5525 | |
9b8ebbdb | 5526 | if (r == RET_PF_INVALID) { |
7a02674d SC |
5527 | r = kvm_mmu_do_page_fault(vcpu, cr2_or_gpa, |
5528 | lower_32_bits(error_code), false); | |
7b367bc9 SC |
5529 | if (WARN_ON_ONCE(r == RET_PF_INVALID)) |
5530 | return -EIO; | |
9b8ebbdb PB |
5531 | } |
5532 | ||
3067714c | 5533 | if (r < 0) |
e9ee956e | 5534 | return r; |
83a2ba4c SC |
5535 | if (r != RET_PF_EMULATE) |
5536 | return 1; | |
3067714c | 5537 | |
14727754 TL |
5538 | /* |
5539 | * Before emulating the instruction, check if the error code | |
5540 | * was due to a RO violation while translating the guest page. | |
5541 | * This can occur when using nested virtualization with nested | |
5542 | * paging in both guests. If true, we simply unprotect the page | |
5543 | * and resume the guest. | |
14727754 | 5544 | */ |
44dd3ffa | 5545 | if (vcpu->arch.mmu->direct_map && |
eebed243 | 5546 | (error_code & PFERR_NESTED_GUEST_PAGE) == PFERR_NESTED_GUEST_PAGE) { |
736c291c | 5547 | kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(cr2_or_gpa)); |
14727754 TL |
5548 | return 1; |
5549 | } | |
5550 | ||
472faffa SC |
5551 | /* |
5552 | * vcpu->arch.mmu.page_fault returned RET_PF_EMULATE, but we can still | |
5553 | * optimistically try to just unprotect the page and let the processor | |
5554 | * re-execute the instruction that caused the page fault. Do not allow | |
5555 | * retrying MMIO emulation, as it's not only pointless but could also | |
5556 | * cause us to enter an infinite loop because the processor will keep | |
6c3dfeb6 SC |
5557 | * faulting on the non-existent MMIO address. Retrying an instruction |
5558 | * from a nested guest is also pointless and dangerous as we are only | |
5559 | * explicitly shadowing L1's page tables, i.e. unprotecting something | |
5560 | * for L1 isn't going to magically fix whatever issue cause L2 to fail. | |
472faffa | 5561 | */ |
736c291c | 5562 | if (!mmio_info_in_cache(vcpu, cr2_or_gpa, direct) && !is_guest_mode(vcpu)) |
92daa48b | 5563 | emulation_type |= EMULTYPE_ALLOW_RETRY_PF; |
e9ee956e | 5564 | emulate: |
736c291c | 5565 | return x86_emulate_instruction(vcpu, cr2_or_gpa, emulation_type, insn, |
60fc3d02 | 5566 | insn_len); |
3067714c AK |
5567 | } |
5568 | EXPORT_SYMBOL_GPL(kvm_mmu_page_fault); | |
5569 | ||
5efac074 PB |
5570 | void kvm_mmu_invalidate_gva(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, |
5571 | gva_t gva, hpa_t root_hpa) | |
a7052897 | 5572 | { |
b94742c9 | 5573 | int i; |
7eb77e9f | 5574 | |
5efac074 PB |
5575 | /* It's actually a GPA for vcpu->arch.guest_mmu. */ |
5576 | if (mmu != &vcpu->arch.guest_mmu) { | |
5577 | /* INVLPG on a non-canonical address is a NOP according to the SDM. */ | |
5578 | if (is_noncanonical_address(gva, vcpu)) | |
5579 | return; | |
5580 | ||
5581 | kvm_x86_ops.tlb_flush_gva(vcpu, gva); | |
5582 | } | |
5583 | ||
5584 | if (!mmu->invlpg) | |
faff8758 JS |
5585 | return; |
5586 | ||
5efac074 PB |
5587 | if (root_hpa == INVALID_PAGE) { |
5588 | mmu->invlpg(vcpu, gva, mmu->root_hpa); | |
956bf353 | 5589 | |
5efac074 PB |
5590 | /* |
5591 | * INVLPG is required to invalidate any global mappings for the VA, | |
5592 | * irrespective of PCID. Since it would take us roughly similar amount | |
5593 | * of work to determine whether any of the prev_root mappings of the VA | |
5594 | * is marked global, or to just sync it blindly, so we might as well | |
5595 | * just always sync it. | |
5596 | * | |
5597 | * Mappings not reachable via the current cr3 or the prev_roots will be | |
5598 | * synced when switching to that cr3, so nothing needs to be done here | |
5599 | * for them. | |
5600 | */ | |
5601 | for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) | |
5602 | if (VALID_PAGE(mmu->prev_roots[i].hpa)) | |
5603 | mmu->invlpg(vcpu, gva, mmu->prev_roots[i].hpa); | |
5604 | } else { | |
5605 | mmu->invlpg(vcpu, gva, root_hpa); | |
5606 | } | |
5607 | } | |
5608 | EXPORT_SYMBOL_GPL(kvm_mmu_invalidate_gva); | |
956bf353 | 5609 | |
5efac074 PB |
5610 | void kvm_mmu_invlpg(struct kvm_vcpu *vcpu, gva_t gva) |
5611 | { | |
5612 | kvm_mmu_invalidate_gva(vcpu, vcpu->arch.mmu, gva, INVALID_PAGE); | |
a7052897 MT |
5613 | ++vcpu->stat.invlpg; |
5614 | } | |
5615 | EXPORT_SYMBOL_GPL(kvm_mmu_invlpg); | |
5616 | ||
5efac074 | 5617 | |
eb4b248e JS |
5618 | void kvm_mmu_invpcid_gva(struct kvm_vcpu *vcpu, gva_t gva, unsigned long pcid) |
5619 | { | |
44dd3ffa | 5620 | struct kvm_mmu *mmu = vcpu->arch.mmu; |
faff8758 | 5621 | bool tlb_flush = false; |
b94742c9 | 5622 | uint i; |
eb4b248e JS |
5623 | |
5624 | if (pcid == kvm_get_active_pcid(vcpu)) { | |
7eb77e9f | 5625 | mmu->invlpg(vcpu, gva, mmu->root_hpa); |
faff8758 | 5626 | tlb_flush = true; |
eb4b248e JS |
5627 | } |
5628 | ||
b94742c9 JS |
5629 | for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) { |
5630 | if (VALID_PAGE(mmu->prev_roots[i].hpa) && | |
be01e8e2 | 5631 | pcid == kvm_get_pcid(vcpu, mmu->prev_roots[i].pgd)) { |
b94742c9 JS |
5632 | mmu->invlpg(vcpu, gva, mmu->prev_roots[i].hpa); |
5633 | tlb_flush = true; | |
5634 | } | |
956bf353 | 5635 | } |
ade61e28 | 5636 | |
faff8758 | 5637 | if (tlb_flush) |
afaf0b2f | 5638 | kvm_x86_ops.tlb_flush_gva(vcpu, gva); |
faff8758 | 5639 | |
eb4b248e JS |
5640 | ++vcpu->stat.invlpg; |
5641 | ||
5642 | /* | |
b94742c9 JS |
5643 | * Mappings not reachable via the current cr3 or the prev_roots will be |
5644 | * synced when switching to that cr3, so nothing needs to be done here | |
5645 | * for them. | |
eb4b248e JS |
5646 | */ |
5647 | } | |
5648 | EXPORT_SYMBOL_GPL(kvm_mmu_invpcid_gva); | |
5649 | ||
83013059 SC |
5650 | void kvm_configure_mmu(bool enable_tdp, int tdp_max_root_level, |
5651 | int tdp_huge_page_level) | |
18552672 | 5652 | { |
bde77235 | 5653 | tdp_enabled = enable_tdp; |
83013059 | 5654 | max_tdp_level = tdp_max_root_level; |
703c335d SC |
5655 | |
5656 | /* | |
1d92d2e8 | 5657 | * max_huge_page_level reflects KVM's MMU capabilities irrespective |
703c335d SC |
5658 | * of kernel support, e.g. KVM may be capable of using 1GB pages when |
5659 | * the kernel is not. But, KVM never creates a page size greater than | |
5660 | * what is used by the kernel for any given HVA, i.e. the kernel's | |
5661 | * capabilities are ultimately consulted by kvm_mmu_hugepage_adjust(). | |
5662 | */ | |
5663 | if (tdp_enabled) | |
1d92d2e8 | 5664 | max_huge_page_level = tdp_huge_page_level; |
703c335d | 5665 | else if (boot_cpu_has(X86_FEATURE_GBPAGES)) |
1d92d2e8 | 5666 | max_huge_page_level = PG_LEVEL_1G; |
703c335d | 5667 | else |
1d92d2e8 | 5668 | max_huge_page_level = PG_LEVEL_2M; |
18552672 | 5669 | } |
bde77235 | 5670 | EXPORT_SYMBOL_GPL(kvm_configure_mmu); |
85875a13 SC |
5671 | |
5672 | /* The return value indicates if tlb flush on all vcpus is needed. */ | |
5673 | typedef bool (*slot_level_handler) (struct kvm *kvm, struct kvm_rmap_head *rmap_head); | |
5674 | ||
5675 | /* The caller should hold mmu-lock before calling this function. */ | |
5676 | static __always_inline bool | |
5677 | slot_handle_level_range(struct kvm *kvm, struct kvm_memory_slot *memslot, | |
5678 | slot_level_handler fn, int start_level, int end_level, | |
5679 | gfn_t start_gfn, gfn_t end_gfn, bool lock_flush_tlb) | |
5680 | { | |
5681 | struct slot_rmap_walk_iterator iterator; | |
5682 | bool flush = false; | |
5683 | ||
5684 | for_each_slot_rmap_range(memslot, start_level, end_level, start_gfn, | |
5685 | end_gfn, &iterator) { | |
5686 | if (iterator.rmap) | |
5687 | flush |= fn(kvm, iterator.rmap); | |
5688 | ||
5689 | if (need_resched() || spin_needbreak(&kvm->mmu_lock)) { | |
5690 | if (flush && lock_flush_tlb) { | |
f285c633 BG |
5691 | kvm_flush_remote_tlbs_with_address(kvm, |
5692 | start_gfn, | |
5693 | iterator.gfn - start_gfn + 1); | |
85875a13 SC |
5694 | flush = false; |
5695 | } | |
5696 | cond_resched_lock(&kvm->mmu_lock); | |
5697 | } | |
5698 | } | |
5699 | ||
5700 | if (flush && lock_flush_tlb) { | |
f285c633 BG |
5701 | kvm_flush_remote_tlbs_with_address(kvm, start_gfn, |
5702 | end_gfn - start_gfn + 1); | |
85875a13 SC |
5703 | flush = false; |
5704 | } | |
5705 | ||
5706 | return flush; | |
5707 | } | |
5708 | ||
5709 | static __always_inline bool | |
5710 | slot_handle_level(struct kvm *kvm, struct kvm_memory_slot *memslot, | |
5711 | slot_level_handler fn, int start_level, int end_level, | |
5712 | bool lock_flush_tlb) | |
5713 | { | |
5714 | return slot_handle_level_range(kvm, memslot, fn, start_level, | |
5715 | end_level, memslot->base_gfn, | |
5716 | memslot->base_gfn + memslot->npages - 1, | |
5717 | lock_flush_tlb); | |
5718 | } | |
5719 | ||
5720 | static __always_inline bool | |
5721 | slot_handle_all_level(struct kvm *kvm, struct kvm_memory_slot *memslot, | |
5722 | slot_level_handler fn, bool lock_flush_tlb) | |
5723 | { | |
3bae0459 | 5724 | return slot_handle_level(kvm, memslot, fn, PG_LEVEL_4K, |
e662ec3e | 5725 | KVM_MAX_HUGEPAGE_LEVEL, lock_flush_tlb); |
85875a13 SC |
5726 | } |
5727 | ||
5728 | static __always_inline bool | |
5729 | slot_handle_large_level(struct kvm *kvm, struct kvm_memory_slot *memslot, | |
5730 | slot_level_handler fn, bool lock_flush_tlb) | |
5731 | { | |
3bae0459 | 5732 | return slot_handle_level(kvm, memslot, fn, PG_LEVEL_4K + 1, |
e662ec3e | 5733 | KVM_MAX_HUGEPAGE_LEVEL, lock_flush_tlb); |
85875a13 SC |
5734 | } |
5735 | ||
5736 | static __always_inline bool | |
5737 | slot_handle_leaf(struct kvm *kvm, struct kvm_memory_slot *memslot, | |
5738 | slot_level_handler fn, bool lock_flush_tlb) | |
5739 | { | |
3bae0459 SC |
5740 | return slot_handle_level(kvm, memslot, fn, PG_LEVEL_4K, |
5741 | PG_LEVEL_4K, lock_flush_tlb); | |
85875a13 SC |
5742 | } |
5743 | ||
1cfff4d9 | 5744 | static void free_mmu_pages(struct kvm_mmu *mmu) |
6aa8b732 | 5745 | { |
1cfff4d9 JP |
5746 | free_page((unsigned long)mmu->pae_root); |
5747 | free_page((unsigned long)mmu->lm_root); | |
6aa8b732 AK |
5748 | } |
5749 | ||
04d28e37 | 5750 | static int __kvm_mmu_create(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu) |
6aa8b732 | 5751 | { |
17ac10ad | 5752 | struct page *page; |
6aa8b732 AK |
5753 | int i; |
5754 | ||
04d28e37 SC |
5755 | mmu->root_hpa = INVALID_PAGE; |
5756 | mmu->root_pgd = 0; | |
5757 | mmu->translate_gpa = translate_gpa; | |
5758 | for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) | |
5759 | mmu->prev_roots[i] = KVM_MMU_ROOT_INFO_INVALID; | |
5760 | ||
17ac10ad | 5761 | /* |
b6b80c78 SC |
5762 | * When using PAE paging, the four PDPTEs are treated as 'root' pages, |
5763 | * while the PDP table is a per-vCPU construct that's allocated at MMU | |
5764 | * creation. When emulating 32-bit mode, cr3 is only 32 bits even on | |
5765 | * x86_64. Therefore we need to allocate the PDP table in the first | |
5766 | * 4GB of memory, which happens to fit the DMA32 zone. Except for | |
5767 | * SVM's 32-bit NPT support, TDP paging doesn't use PAE paging and can | |
5768 | * skip allocating the PDP table. | |
17ac10ad | 5769 | */ |
d468d94b | 5770 | if (tdp_enabled && kvm_mmu_get_tdp_level(vcpu) > PT32E_ROOT_LEVEL) |
b6b80c78 SC |
5771 | return 0; |
5772 | ||
254272ce | 5773 | page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_DMA32); |
17ac10ad | 5774 | if (!page) |
d7fa6ab2 WY |
5775 | return -ENOMEM; |
5776 | ||
1cfff4d9 | 5777 | mmu->pae_root = page_address(page); |
17ac10ad | 5778 | for (i = 0; i < 4; ++i) |
1cfff4d9 | 5779 | mmu->pae_root[i] = INVALID_PAGE; |
17ac10ad | 5780 | |
6aa8b732 | 5781 | return 0; |
6aa8b732 AK |
5782 | } |
5783 | ||
8018c27b | 5784 | int kvm_mmu_create(struct kvm_vcpu *vcpu) |
6aa8b732 | 5785 | { |
1cfff4d9 | 5786 | int ret; |
b94742c9 | 5787 | |
5962bfb7 | 5788 | vcpu->arch.mmu_pte_list_desc_cache.kmem_cache = pte_list_desc_cache; |
5f6078f9 SC |
5789 | vcpu->arch.mmu_pte_list_desc_cache.gfp_zero = __GFP_ZERO; |
5790 | ||
5962bfb7 | 5791 | vcpu->arch.mmu_page_header_cache.kmem_cache = mmu_page_header_cache; |
5f6078f9 | 5792 | vcpu->arch.mmu_page_header_cache.gfp_zero = __GFP_ZERO; |
5962bfb7 | 5793 | |
96880883 SC |
5794 | vcpu->arch.mmu_shadow_page_cache.gfp_zero = __GFP_ZERO; |
5795 | ||
44dd3ffa VK |
5796 | vcpu->arch.mmu = &vcpu->arch.root_mmu; |
5797 | vcpu->arch.walk_mmu = &vcpu->arch.root_mmu; | |
6aa8b732 | 5798 | |
14c07ad8 | 5799 | vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa; |
1cfff4d9 | 5800 | |
04d28e37 | 5801 | ret = __kvm_mmu_create(vcpu, &vcpu->arch.guest_mmu); |
1cfff4d9 JP |
5802 | if (ret) |
5803 | return ret; | |
5804 | ||
04d28e37 | 5805 | ret = __kvm_mmu_create(vcpu, &vcpu->arch.root_mmu); |
1cfff4d9 JP |
5806 | if (ret) |
5807 | goto fail_allocate_root; | |
5808 | ||
5809 | return ret; | |
5810 | fail_allocate_root: | |
5811 | free_mmu_pages(&vcpu->arch.guest_mmu); | |
5812 | return ret; | |
6aa8b732 AK |
5813 | } |
5814 | ||
fbb158cb | 5815 | #define BATCH_ZAP_PAGES 10 |
002c5f73 SC |
5816 | static void kvm_zap_obsolete_pages(struct kvm *kvm) |
5817 | { | |
5818 | struct kvm_mmu_page *sp, *node; | |
fbb158cb | 5819 | int nr_zapped, batch = 0; |
002c5f73 SC |
5820 | |
5821 | restart: | |
5822 | list_for_each_entry_safe_reverse(sp, node, | |
5823 | &kvm->arch.active_mmu_pages, link) { | |
5824 | /* | |
5825 | * No obsolete valid page exists before a newly created page | |
5826 | * since active_mmu_pages is a FIFO list. | |
5827 | */ | |
5828 | if (!is_obsolete_sp(kvm, sp)) | |
5829 | break; | |
5830 | ||
5831 | /* | |
f95eec9b SC |
5832 | * Invalid pages should never land back on the list of active |
5833 | * pages. Skip the bogus page, otherwise we'll get stuck in an | |
5834 | * infinite loop if the page gets put back on the list (again). | |
002c5f73 | 5835 | */ |
f95eec9b | 5836 | if (WARN_ON(sp->role.invalid)) |
002c5f73 SC |
5837 | continue; |
5838 | ||
4506ecf4 SC |
5839 | /* |
5840 | * No need to flush the TLB since we're only zapping shadow | |
5841 | * pages with an obsolete generation number and all vCPUS have | |
5842 | * loaded a new root, i.e. the shadow pages being zapped cannot | |
5843 | * be in active use by the guest. | |
5844 | */ | |
fbb158cb | 5845 | if (batch >= BATCH_ZAP_PAGES && |
4506ecf4 | 5846 | cond_resched_lock(&kvm->mmu_lock)) { |
fbb158cb | 5847 | batch = 0; |
002c5f73 SC |
5848 | goto restart; |
5849 | } | |
5850 | ||
10605204 SC |
5851 | if (__kvm_mmu_prepare_zap_page(kvm, sp, |
5852 | &kvm->arch.zapped_obsolete_pages, &nr_zapped)) { | |
fbb158cb | 5853 | batch += nr_zapped; |
002c5f73 | 5854 | goto restart; |
fbb158cb | 5855 | } |
002c5f73 SC |
5856 | } |
5857 | ||
4506ecf4 SC |
5858 | /* |
5859 | * Trigger a remote TLB flush before freeing the page tables to ensure | |
5860 | * KVM is not in the middle of a lockless shadow page table walk, which | |
5861 | * may reference the pages. | |
5862 | */ | |
10605204 | 5863 | kvm_mmu_commit_zap_page(kvm, &kvm->arch.zapped_obsolete_pages); |
002c5f73 SC |
5864 | } |
5865 | ||
5866 | /* | |
5867 | * Fast invalidate all shadow pages and use lock-break technique | |
5868 | * to zap obsolete pages. | |
5869 | * | |
5870 | * It's required when memslot is being deleted or VM is being | |
5871 | * destroyed, in these cases, we should ensure that KVM MMU does | |
5872 | * not use any resource of the being-deleted slot or all slots | |
5873 | * after calling the function. | |
5874 | */ | |
5875 | static void kvm_mmu_zap_all_fast(struct kvm *kvm) | |
5876 | { | |
ca333add SC |
5877 | lockdep_assert_held(&kvm->slots_lock); |
5878 | ||
002c5f73 | 5879 | spin_lock(&kvm->mmu_lock); |
14a3c4f4 | 5880 | trace_kvm_mmu_zap_all_fast(kvm); |
ca333add SC |
5881 | |
5882 | /* | |
5883 | * Toggle mmu_valid_gen between '0' and '1'. Because slots_lock is | |
5884 | * held for the entire duration of zapping obsolete pages, it's | |
5885 | * impossible for there to be multiple invalid generations associated | |
5886 | * with *valid* shadow pages at any given time, i.e. there is exactly | |
5887 | * one valid generation and (at most) one invalid generation. | |
5888 | */ | |
5889 | kvm->arch.mmu_valid_gen = kvm->arch.mmu_valid_gen ? 0 : 1; | |
002c5f73 | 5890 | |
4506ecf4 SC |
5891 | /* |
5892 | * Notify all vcpus to reload its shadow page table and flush TLB. | |
5893 | * Then all vcpus will switch to new shadow page table with the new | |
5894 | * mmu_valid_gen. | |
5895 | * | |
5896 | * Note: we need to do this under the protection of mmu_lock, | |
5897 | * otherwise, vcpu would purge shadow page but miss tlb flush. | |
5898 | */ | |
5899 | kvm_reload_remote_mmus(kvm); | |
5900 | ||
002c5f73 SC |
5901 | kvm_zap_obsolete_pages(kvm); |
5902 | spin_unlock(&kvm->mmu_lock); | |
5903 | } | |
5904 | ||
10605204 SC |
5905 | static bool kvm_has_zapped_obsolete_pages(struct kvm *kvm) |
5906 | { | |
5907 | return unlikely(!list_empty_careful(&kvm->arch.zapped_obsolete_pages)); | |
5908 | } | |
5909 | ||
b5f5fdca | 5910 | static void kvm_mmu_invalidate_zap_pages_in_memslot(struct kvm *kvm, |
d126363d JS |
5911 | struct kvm_memory_slot *slot, |
5912 | struct kvm_page_track_notifier_node *node) | |
b5f5fdca | 5913 | { |
002c5f73 | 5914 | kvm_mmu_zap_all_fast(kvm); |
1bad2b2a XG |
5915 | } |
5916 | ||
13d268ca | 5917 | void kvm_mmu_init_vm(struct kvm *kvm) |
1bad2b2a | 5918 | { |
13d268ca | 5919 | struct kvm_page_track_notifier_node *node = &kvm->arch.mmu_sp_tracker; |
1bad2b2a | 5920 | |
13d268ca | 5921 | node->track_write = kvm_mmu_pte_write; |
b5f5fdca | 5922 | node->track_flush_slot = kvm_mmu_invalidate_zap_pages_in_memslot; |
13d268ca | 5923 | kvm_page_track_register_notifier(kvm, node); |
1bad2b2a XG |
5924 | } |
5925 | ||
13d268ca | 5926 | void kvm_mmu_uninit_vm(struct kvm *kvm) |
1bad2b2a | 5927 | { |
13d268ca | 5928 | struct kvm_page_track_notifier_node *node = &kvm->arch.mmu_sp_tracker; |
1bad2b2a | 5929 | |
13d268ca | 5930 | kvm_page_track_unregister_notifier(kvm, node); |
1bad2b2a XG |
5931 | } |
5932 | ||
efdfe536 XG |
5933 | void kvm_zap_gfn_range(struct kvm *kvm, gfn_t gfn_start, gfn_t gfn_end) |
5934 | { | |
5935 | struct kvm_memslots *slots; | |
5936 | struct kvm_memory_slot *memslot; | |
9da0e4d5 | 5937 | int i; |
efdfe536 XG |
5938 | |
5939 | spin_lock(&kvm->mmu_lock); | |
9da0e4d5 PB |
5940 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
5941 | slots = __kvm_memslots(kvm, i); | |
5942 | kvm_for_each_memslot(memslot, slots) { | |
5943 | gfn_t start, end; | |
5944 | ||
5945 | start = max(gfn_start, memslot->base_gfn); | |
5946 | end = min(gfn_end, memslot->base_gfn + memslot->npages); | |
5947 | if (start >= end) | |
5948 | continue; | |
efdfe536 | 5949 | |
92da008f | 5950 | slot_handle_level_range(kvm, memslot, kvm_zap_rmapp, |
3bae0459 | 5951 | PG_LEVEL_4K, |
e662ec3e | 5952 | KVM_MAX_HUGEPAGE_LEVEL, |
92da008f | 5953 | start, end - 1, true); |
9da0e4d5 | 5954 | } |
efdfe536 XG |
5955 | } |
5956 | ||
5957 | spin_unlock(&kvm->mmu_lock); | |
5958 | } | |
5959 | ||
018aabb5 TY |
5960 | static bool slot_rmap_write_protect(struct kvm *kvm, |
5961 | struct kvm_rmap_head *rmap_head) | |
d77aa73c | 5962 | { |
018aabb5 | 5963 | return __rmap_write_protect(kvm, rmap_head, false); |
d77aa73c XG |
5964 | } |
5965 | ||
1c91cad4 | 5966 | void kvm_mmu_slot_remove_write_access(struct kvm *kvm, |
3c9bd400 JZ |
5967 | struct kvm_memory_slot *memslot, |
5968 | int start_level) | |
6aa8b732 | 5969 | { |
d77aa73c | 5970 | bool flush; |
6aa8b732 | 5971 | |
9d1beefb | 5972 | spin_lock(&kvm->mmu_lock); |
3c9bd400 | 5973 | flush = slot_handle_level(kvm, memslot, slot_rmap_write_protect, |
e662ec3e | 5974 | start_level, KVM_MAX_HUGEPAGE_LEVEL, false); |
9d1beefb | 5975 | spin_unlock(&kvm->mmu_lock); |
198c74f4 | 5976 | |
198c74f4 XG |
5977 | /* |
5978 | * We can flush all the TLBs out of the mmu lock without TLB | |
5979 | * corruption since we just change the spte from writable to | |
5980 | * readonly so that we only need to care the case of changing | |
5981 | * spte from present to present (changing the spte from present | |
5982 | * to nonpresent will flush all the TLBs immediately), in other | |
5983 | * words, the only case we care is mmu_spte_update() where we | |
bdd303cb | 5984 | * have checked SPTE_HOST_WRITEABLE | SPTE_MMU_WRITEABLE |
198c74f4 XG |
5985 | * instead of PT_WRITABLE_MASK, that means it does not depend |
5986 | * on PT_WRITABLE_MASK anymore. | |
5987 | */ | |
d91ffee9 | 5988 | if (flush) |
7f42aa76 | 5989 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); |
6aa8b732 | 5990 | } |
37a7d8b0 | 5991 | |
3ea3b7fa | 5992 | static bool kvm_mmu_zap_collapsible_spte(struct kvm *kvm, |
018aabb5 | 5993 | struct kvm_rmap_head *rmap_head) |
3ea3b7fa WL |
5994 | { |
5995 | u64 *sptep; | |
5996 | struct rmap_iterator iter; | |
5997 | int need_tlb_flush = 0; | |
ba049e93 | 5998 | kvm_pfn_t pfn; |
3ea3b7fa WL |
5999 | struct kvm_mmu_page *sp; |
6000 | ||
0d536790 | 6001 | restart: |
018aabb5 | 6002 | for_each_rmap_spte(rmap_head, &iter, sptep) { |
57354682 | 6003 | sp = sptep_to_sp(sptep); |
3ea3b7fa WL |
6004 | pfn = spte_to_pfn(*sptep); |
6005 | ||
6006 | /* | |
decf6333 XG |
6007 | * We cannot do huge page mapping for indirect shadow pages, |
6008 | * which are found on the last rmap (level = 1) when not using | |
6009 | * tdp; such shadow pages are synced with the page table in | |
6010 | * the guest, and the guest page table is using 4K page size | |
6011 | * mapping if the indirect sp has level = 1. | |
3ea3b7fa | 6012 | */ |
a78986aa | 6013 | if (sp->role.direct && !kvm_is_reserved_pfn(pfn) && |
e851265a SC |
6014 | (kvm_is_zone_device_pfn(pfn) || |
6015 | PageCompound(pfn_to_page(pfn)))) { | |
e7912386 | 6016 | pte_list_remove(rmap_head, sptep); |
40ef75a7 LT |
6017 | |
6018 | if (kvm_available_flush_tlb_with_range()) | |
6019 | kvm_flush_remote_tlbs_with_address(kvm, sp->gfn, | |
6020 | KVM_PAGES_PER_HPAGE(sp->role.level)); | |
6021 | else | |
6022 | need_tlb_flush = 1; | |
6023 | ||
0d536790 XG |
6024 | goto restart; |
6025 | } | |
3ea3b7fa WL |
6026 | } |
6027 | ||
6028 | return need_tlb_flush; | |
6029 | } | |
6030 | ||
6031 | void kvm_mmu_zap_collapsible_sptes(struct kvm *kvm, | |
f36f3f28 | 6032 | const struct kvm_memory_slot *memslot) |
3ea3b7fa | 6033 | { |
f36f3f28 | 6034 | /* FIXME: const-ify all uses of struct kvm_memory_slot. */ |
3ea3b7fa | 6035 | spin_lock(&kvm->mmu_lock); |
f36f3f28 PB |
6036 | slot_handle_leaf(kvm, (struct kvm_memory_slot *)memslot, |
6037 | kvm_mmu_zap_collapsible_spte, true); | |
3ea3b7fa WL |
6038 | spin_unlock(&kvm->mmu_lock); |
6039 | } | |
6040 | ||
b3594ffb SC |
6041 | void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm, |
6042 | struct kvm_memory_slot *memslot) | |
6043 | { | |
6044 | /* | |
7f42aa76 SC |
6045 | * All current use cases for flushing the TLBs for a specific memslot |
6046 | * are related to dirty logging, and do the TLB flush out of mmu_lock. | |
6047 | * The interaction between the various operations on memslot must be | |
6048 | * serialized by slots_locks to ensure the TLB flush from one operation | |
6049 | * is observed by any other operation on the same memslot. | |
b3594ffb SC |
6050 | */ |
6051 | lockdep_assert_held(&kvm->slots_lock); | |
cec37648 SC |
6052 | kvm_flush_remote_tlbs_with_address(kvm, memslot->base_gfn, |
6053 | memslot->npages); | |
b3594ffb SC |
6054 | } |
6055 | ||
f4b4b180 KH |
6056 | void kvm_mmu_slot_leaf_clear_dirty(struct kvm *kvm, |
6057 | struct kvm_memory_slot *memslot) | |
6058 | { | |
d77aa73c | 6059 | bool flush; |
f4b4b180 KH |
6060 | |
6061 | spin_lock(&kvm->mmu_lock); | |
d77aa73c | 6062 | flush = slot_handle_leaf(kvm, memslot, __rmap_clear_dirty, false); |
f4b4b180 KH |
6063 | spin_unlock(&kvm->mmu_lock); |
6064 | ||
f4b4b180 KH |
6065 | /* |
6066 | * It's also safe to flush TLBs out of mmu lock here as currently this | |
6067 | * function is only used for dirty logging, in which case flushing TLB | |
6068 | * out of mmu lock also guarantees no dirty pages will be lost in | |
6069 | * dirty_bitmap. | |
6070 | */ | |
6071 | if (flush) | |
7f42aa76 | 6072 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); |
f4b4b180 KH |
6073 | } |
6074 | EXPORT_SYMBOL_GPL(kvm_mmu_slot_leaf_clear_dirty); | |
6075 | ||
6076 | void kvm_mmu_slot_largepage_remove_write_access(struct kvm *kvm, | |
6077 | struct kvm_memory_slot *memslot) | |
6078 | { | |
d77aa73c | 6079 | bool flush; |
f4b4b180 KH |
6080 | |
6081 | spin_lock(&kvm->mmu_lock); | |
d77aa73c XG |
6082 | flush = slot_handle_large_level(kvm, memslot, slot_rmap_write_protect, |
6083 | false); | |
f4b4b180 KH |
6084 | spin_unlock(&kvm->mmu_lock); |
6085 | ||
f4b4b180 | 6086 | if (flush) |
7f42aa76 | 6087 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); |
f4b4b180 KH |
6088 | } |
6089 | EXPORT_SYMBOL_GPL(kvm_mmu_slot_largepage_remove_write_access); | |
6090 | ||
6091 | void kvm_mmu_slot_set_dirty(struct kvm *kvm, | |
6092 | struct kvm_memory_slot *memslot) | |
6093 | { | |
d77aa73c | 6094 | bool flush; |
f4b4b180 KH |
6095 | |
6096 | spin_lock(&kvm->mmu_lock); | |
d77aa73c | 6097 | flush = slot_handle_all_level(kvm, memslot, __rmap_set_dirty, false); |
f4b4b180 KH |
6098 | spin_unlock(&kvm->mmu_lock); |
6099 | ||
f4b4b180 | 6100 | if (flush) |
7f42aa76 | 6101 | kvm_arch_flush_remote_tlbs_memslot(kvm, memslot); |
f4b4b180 KH |
6102 | } |
6103 | EXPORT_SYMBOL_GPL(kvm_mmu_slot_set_dirty); | |
6104 | ||
92f58b5c | 6105 | void kvm_mmu_zap_all(struct kvm *kvm) |
5304b8d3 XG |
6106 | { |
6107 | struct kvm_mmu_page *sp, *node; | |
7390de1e | 6108 | LIST_HEAD(invalid_list); |
83cdb568 | 6109 | int ign; |
5304b8d3 | 6110 | |
7390de1e | 6111 | spin_lock(&kvm->mmu_lock); |
5304b8d3 | 6112 | restart: |
8a674adc | 6113 | list_for_each_entry_safe(sp, node, &kvm->arch.active_mmu_pages, link) { |
f95eec9b | 6114 | if (WARN_ON(sp->role.invalid)) |
4771450c | 6115 | continue; |
92f58b5c | 6116 | if (__kvm_mmu_prepare_zap_page(kvm, sp, &invalid_list, &ign)) |
5304b8d3 | 6117 | goto restart; |
24efe61f | 6118 | if (cond_resched_lock(&kvm->mmu_lock)) |
5304b8d3 XG |
6119 | goto restart; |
6120 | } | |
6121 | ||
4771450c | 6122 | kvm_mmu_commit_zap_page(kvm, &invalid_list); |
5304b8d3 XG |
6123 | spin_unlock(&kvm->mmu_lock); |
6124 | } | |
6125 | ||
15248258 | 6126 | void kvm_mmu_invalidate_mmio_sptes(struct kvm *kvm, u64 gen) |
f8f55942 | 6127 | { |
164bf7e5 | 6128 | WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS); |
e1359e2b | 6129 | |
164bf7e5 | 6130 | gen &= MMIO_SPTE_GEN_MASK; |
e1359e2b | 6131 | |
f8f55942 | 6132 | /* |
e1359e2b SC |
6133 | * Generation numbers are incremented in multiples of the number of |
6134 | * address spaces in order to provide unique generations across all | |
6135 | * address spaces. Strip what is effectively the address space | |
6136 | * modifier prior to checking for a wrap of the MMIO generation so | |
6137 | * that a wrap in any address space is detected. | |
6138 | */ | |
6139 | gen &= ~((u64)KVM_ADDRESS_SPACE_NUM - 1); | |
6140 | ||
f8f55942 | 6141 | /* |
e1359e2b | 6142 | * The very rare case: if the MMIO generation number has wrapped, |
f8f55942 | 6143 | * zap all shadow pages. |
f8f55942 | 6144 | */ |
e1359e2b | 6145 | if (unlikely(gen == 0)) { |
ae0f5499 | 6146 | kvm_debug_ratelimited("kvm: zapping shadow pages for mmio generation wraparound\n"); |
92f58b5c | 6147 | kvm_mmu_zap_all_fast(kvm); |
7a2e8aaf | 6148 | } |
f8f55942 XG |
6149 | } |
6150 | ||
70534a73 DC |
6151 | static unsigned long |
6152 | mmu_shrink_scan(struct shrinker *shrink, struct shrink_control *sc) | |
3ee16c81 IE |
6153 | { |
6154 | struct kvm *kvm; | |
1495f230 | 6155 | int nr_to_scan = sc->nr_to_scan; |
70534a73 | 6156 | unsigned long freed = 0; |
3ee16c81 | 6157 | |
0d9ce162 | 6158 | mutex_lock(&kvm_lock); |
3ee16c81 IE |
6159 | |
6160 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
3d56cbdf | 6161 | int idx; |
d98ba053 | 6162 | LIST_HEAD(invalid_list); |
3ee16c81 | 6163 | |
35f2d16b TY |
6164 | /* |
6165 | * Never scan more than sc->nr_to_scan VM instances. | |
6166 | * Will not hit this condition practically since we do not try | |
6167 | * to shrink more than one VM and it is very unlikely to see | |
6168 | * !n_used_mmu_pages so many times. | |
6169 | */ | |
6170 | if (!nr_to_scan--) | |
6171 | break; | |
19526396 GN |
6172 | /* |
6173 | * n_used_mmu_pages is accessed without holding kvm->mmu_lock | |
6174 | * here. We may skip a VM instance errorneosly, but we do not | |
6175 | * want to shrink a VM that only started to populate its MMU | |
6176 | * anyway. | |
6177 | */ | |
10605204 SC |
6178 | if (!kvm->arch.n_used_mmu_pages && |
6179 | !kvm_has_zapped_obsolete_pages(kvm)) | |
19526396 | 6180 | continue; |
19526396 | 6181 | |
f656ce01 | 6182 | idx = srcu_read_lock(&kvm->srcu); |
3ee16c81 | 6183 | spin_lock(&kvm->mmu_lock); |
3ee16c81 | 6184 | |
10605204 SC |
6185 | if (kvm_has_zapped_obsolete_pages(kvm)) { |
6186 | kvm_mmu_commit_zap_page(kvm, | |
6187 | &kvm->arch.zapped_obsolete_pages); | |
6188 | goto unlock; | |
6189 | } | |
6190 | ||
ebdb292d | 6191 | freed = kvm_mmu_zap_oldest_mmu_pages(kvm, sc->nr_to_scan); |
19526396 | 6192 | |
10605204 | 6193 | unlock: |
3ee16c81 | 6194 | spin_unlock(&kvm->mmu_lock); |
f656ce01 | 6195 | srcu_read_unlock(&kvm->srcu, idx); |
19526396 | 6196 | |
70534a73 DC |
6197 | /* |
6198 | * unfair on small ones | |
6199 | * per-vm shrinkers cry out | |
6200 | * sadness comes quickly | |
6201 | */ | |
19526396 GN |
6202 | list_move_tail(&kvm->vm_list, &vm_list); |
6203 | break; | |
3ee16c81 | 6204 | } |
3ee16c81 | 6205 | |
0d9ce162 | 6206 | mutex_unlock(&kvm_lock); |
70534a73 | 6207 | return freed; |
70534a73 DC |
6208 | } |
6209 | ||
6210 | static unsigned long | |
6211 | mmu_shrink_count(struct shrinker *shrink, struct shrink_control *sc) | |
6212 | { | |
45221ab6 | 6213 | return percpu_counter_read_positive(&kvm_total_used_mmu_pages); |
3ee16c81 IE |
6214 | } |
6215 | ||
6216 | static struct shrinker mmu_shrinker = { | |
70534a73 DC |
6217 | .count_objects = mmu_shrink_count, |
6218 | .scan_objects = mmu_shrink_scan, | |
3ee16c81 IE |
6219 | .seeks = DEFAULT_SEEKS * 10, |
6220 | }; | |
6221 | ||
2ddfd20e | 6222 | static void mmu_destroy_caches(void) |
b5a33a75 | 6223 | { |
c1bd743e TH |
6224 | kmem_cache_destroy(pte_list_desc_cache); |
6225 | kmem_cache_destroy(mmu_page_header_cache); | |
b5a33a75 AK |
6226 | } |
6227 | ||
7b6f8a06 KH |
6228 | static void kvm_set_mmio_spte_mask(void) |
6229 | { | |
6230 | u64 mask; | |
7b6f8a06 KH |
6231 | |
6232 | /* | |
6129ed87 SC |
6233 | * Set a reserved PA bit in MMIO SPTEs to generate page faults with |
6234 | * PFEC.RSVD=1 on MMIO accesses. 64-bit PTEs (PAE, x86-64, and EPT | |
6235 | * paging) support a maximum of 52 bits of PA, i.e. if the CPU supports | |
6236 | * 52-bit physical addresses then there are no reserved PA bits in the | |
6237 | * PTEs and so the reserved PA approach must be disabled. | |
7b6f8a06 | 6238 | */ |
6129ed87 SC |
6239 | if (shadow_phys_bits < 52) |
6240 | mask = BIT_ULL(51) | PT_PRESENT_MASK; | |
6241 | else | |
6242 | mask = 0; | |
7b6f8a06 | 6243 | |
e7581cac | 6244 | kvm_mmu_set_mmio_spte_mask(mask, ACC_WRITE_MASK | ACC_USER_MASK); |
7b6f8a06 KH |
6245 | } |
6246 | ||
b8e8c830 PB |
6247 | static bool get_nx_auto_mode(void) |
6248 | { | |
6249 | /* Return true when CPU has the bug, and mitigations are ON */ | |
6250 | return boot_cpu_has_bug(X86_BUG_ITLB_MULTIHIT) && !cpu_mitigations_off(); | |
6251 | } | |
6252 | ||
6253 | static void __set_nx_huge_pages(bool val) | |
6254 | { | |
6255 | nx_huge_pages = itlb_multihit_kvm_mitigation = val; | |
6256 | } | |
6257 | ||
6258 | static int set_nx_huge_pages(const char *val, const struct kernel_param *kp) | |
6259 | { | |
6260 | bool old_val = nx_huge_pages; | |
6261 | bool new_val; | |
6262 | ||
6263 | /* In "auto" mode deploy workaround only if CPU has the bug. */ | |
6264 | if (sysfs_streq(val, "off")) | |
6265 | new_val = 0; | |
6266 | else if (sysfs_streq(val, "force")) | |
6267 | new_val = 1; | |
6268 | else if (sysfs_streq(val, "auto")) | |
6269 | new_val = get_nx_auto_mode(); | |
6270 | else if (strtobool(val, &new_val) < 0) | |
6271 | return -EINVAL; | |
6272 | ||
6273 | __set_nx_huge_pages(new_val); | |
6274 | ||
6275 | if (new_val != old_val) { | |
6276 | struct kvm *kvm; | |
b8e8c830 PB |
6277 | |
6278 | mutex_lock(&kvm_lock); | |
6279 | ||
6280 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
ed69a6cb | 6281 | mutex_lock(&kvm->slots_lock); |
b8e8c830 | 6282 | kvm_mmu_zap_all_fast(kvm); |
ed69a6cb | 6283 | mutex_unlock(&kvm->slots_lock); |
1aa9b957 JS |
6284 | |
6285 | wake_up_process(kvm->arch.nx_lpage_recovery_thread); | |
b8e8c830 PB |
6286 | } |
6287 | mutex_unlock(&kvm_lock); | |
6288 | } | |
6289 | ||
6290 | return 0; | |
6291 | } | |
6292 | ||
b5a33a75 AK |
6293 | int kvm_mmu_module_init(void) |
6294 | { | |
ab271bd4 AB |
6295 | int ret = -ENOMEM; |
6296 | ||
b8e8c830 PB |
6297 | if (nx_huge_pages == -1) |
6298 | __set_nx_huge_pages(get_nx_auto_mode()); | |
6299 | ||
36d9594d VK |
6300 | /* |
6301 | * MMU roles use union aliasing which is, generally speaking, an | |
6302 | * undefined behavior. However, we supposedly know how compilers behave | |
6303 | * and the current status quo is unlikely to change. Guardians below are | |
6304 | * supposed to let us know if the assumption becomes false. | |
6305 | */ | |
6306 | BUILD_BUG_ON(sizeof(union kvm_mmu_page_role) != sizeof(u32)); | |
6307 | BUILD_BUG_ON(sizeof(union kvm_mmu_extended_role) != sizeof(u32)); | |
6308 | BUILD_BUG_ON(sizeof(union kvm_mmu_role) != sizeof(u64)); | |
6309 | ||
28a1f3ac | 6310 | kvm_mmu_reset_all_pte_masks(); |
f160c7b7 | 6311 | |
7b6f8a06 KH |
6312 | kvm_set_mmio_spte_mask(); |
6313 | ||
53c07b18 XG |
6314 | pte_list_desc_cache = kmem_cache_create("pte_list_desc", |
6315 | sizeof(struct pte_list_desc), | |
46bea48a | 6316 | 0, SLAB_ACCOUNT, NULL); |
53c07b18 | 6317 | if (!pte_list_desc_cache) |
ab271bd4 | 6318 | goto out; |
b5a33a75 | 6319 | |
d3d25b04 AK |
6320 | mmu_page_header_cache = kmem_cache_create("kvm_mmu_page_header", |
6321 | sizeof(struct kvm_mmu_page), | |
46bea48a | 6322 | 0, SLAB_ACCOUNT, NULL); |
d3d25b04 | 6323 | if (!mmu_page_header_cache) |
ab271bd4 | 6324 | goto out; |
d3d25b04 | 6325 | |
908c7f19 | 6326 | if (percpu_counter_init(&kvm_total_used_mmu_pages, 0, GFP_KERNEL)) |
ab271bd4 | 6327 | goto out; |
45bf21a8 | 6328 | |
ab271bd4 AB |
6329 | ret = register_shrinker(&mmu_shrinker); |
6330 | if (ret) | |
6331 | goto out; | |
3ee16c81 | 6332 | |
b5a33a75 AK |
6333 | return 0; |
6334 | ||
ab271bd4 | 6335 | out: |
3ee16c81 | 6336 | mmu_destroy_caches(); |
ab271bd4 | 6337 | return ret; |
b5a33a75 AK |
6338 | } |
6339 | ||
3ad82a7e | 6340 | /* |
39337ad1 | 6341 | * Calculate mmu pages needed for kvm. |
3ad82a7e | 6342 | */ |
bc8a3d89 | 6343 | unsigned long kvm_mmu_calculate_default_mmu_pages(struct kvm *kvm) |
3ad82a7e | 6344 | { |
bc8a3d89 BG |
6345 | unsigned long nr_mmu_pages; |
6346 | unsigned long nr_pages = 0; | |
bc6678a3 | 6347 | struct kvm_memslots *slots; |
be6ba0f0 | 6348 | struct kvm_memory_slot *memslot; |
9da0e4d5 | 6349 | int i; |
3ad82a7e | 6350 | |
9da0e4d5 PB |
6351 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
6352 | slots = __kvm_memslots(kvm, i); | |
90d83dc3 | 6353 | |
9da0e4d5 PB |
6354 | kvm_for_each_memslot(memslot, slots) |
6355 | nr_pages += memslot->npages; | |
6356 | } | |
3ad82a7e ZX |
6357 | |
6358 | nr_mmu_pages = nr_pages * KVM_PERMILLE_MMU_PAGES / 1000; | |
bc8a3d89 | 6359 | nr_mmu_pages = max(nr_mmu_pages, KVM_MIN_ALLOC_MMU_PAGES); |
3ad82a7e ZX |
6360 | |
6361 | return nr_mmu_pages; | |
6362 | } | |
6363 | ||
c42fffe3 XG |
6364 | void kvm_mmu_destroy(struct kvm_vcpu *vcpu) |
6365 | { | |
95f93af4 | 6366 | kvm_mmu_unload(vcpu); |
1cfff4d9 JP |
6367 | free_mmu_pages(&vcpu->arch.root_mmu); |
6368 | free_mmu_pages(&vcpu->arch.guest_mmu); | |
c42fffe3 | 6369 | mmu_free_memory_caches(vcpu); |
b034cf01 XG |
6370 | } |
6371 | ||
b034cf01 XG |
6372 | void kvm_mmu_module_exit(void) |
6373 | { | |
6374 | mmu_destroy_caches(); | |
6375 | percpu_counter_destroy(&kvm_total_used_mmu_pages); | |
6376 | unregister_shrinker(&mmu_shrinker); | |
c42fffe3 XG |
6377 | mmu_audit_disable(); |
6378 | } | |
1aa9b957 JS |
6379 | |
6380 | static int set_nx_huge_pages_recovery_ratio(const char *val, const struct kernel_param *kp) | |
6381 | { | |
6382 | unsigned int old_val; | |
6383 | int err; | |
6384 | ||
6385 | old_val = nx_huge_pages_recovery_ratio; | |
6386 | err = param_set_uint(val, kp); | |
6387 | if (err) | |
6388 | return err; | |
6389 | ||
6390 | if (READ_ONCE(nx_huge_pages) && | |
6391 | !old_val && nx_huge_pages_recovery_ratio) { | |
6392 | struct kvm *kvm; | |
6393 | ||
6394 | mutex_lock(&kvm_lock); | |
6395 | ||
6396 | list_for_each_entry(kvm, &vm_list, vm_list) | |
6397 | wake_up_process(kvm->arch.nx_lpage_recovery_thread); | |
6398 | ||
6399 | mutex_unlock(&kvm_lock); | |
6400 | } | |
6401 | ||
6402 | return err; | |
6403 | } | |
6404 | ||
6405 | static void kvm_recover_nx_lpages(struct kvm *kvm) | |
6406 | { | |
6407 | int rcu_idx; | |
6408 | struct kvm_mmu_page *sp; | |
6409 | unsigned int ratio; | |
6410 | LIST_HEAD(invalid_list); | |
6411 | ulong to_zap; | |
6412 | ||
6413 | rcu_idx = srcu_read_lock(&kvm->srcu); | |
6414 | spin_lock(&kvm->mmu_lock); | |
6415 | ||
6416 | ratio = READ_ONCE(nx_huge_pages_recovery_ratio); | |
6417 | to_zap = ratio ? DIV_ROUND_UP(kvm->stat.nx_lpage_splits, ratio) : 0; | |
7d919c7a SC |
6418 | for ( ; to_zap; --to_zap) { |
6419 | if (list_empty(&kvm->arch.lpage_disallowed_mmu_pages)) | |
6420 | break; | |
6421 | ||
1aa9b957 JS |
6422 | /* |
6423 | * We use a separate list instead of just using active_mmu_pages | |
6424 | * because the number of lpage_disallowed pages is expected to | |
6425 | * be relatively small compared to the total. | |
6426 | */ | |
6427 | sp = list_first_entry(&kvm->arch.lpage_disallowed_mmu_pages, | |
6428 | struct kvm_mmu_page, | |
6429 | lpage_disallowed_link); | |
6430 | WARN_ON_ONCE(!sp->lpage_disallowed); | |
6431 | kvm_mmu_prepare_zap_page(kvm, sp, &invalid_list); | |
6432 | WARN_ON_ONCE(sp->lpage_disallowed); | |
6433 | ||
7d919c7a | 6434 | if (need_resched() || spin_needbreak(&kvm->mmu_lock)) { |
1aa9b957 | 6435 | kvm_mmu_commit_zap_page(kvm, &invalid_list); |
7d919c7a | 6436 | cond_resched_lock(&kvm->mmu_lock); |
1aa9b957 JS |
6437 | } |
6438 | } | |
e8950569 | 6439 | kvm_mmu_commit_zap_page(kvm, &invalid_list); |
1aa9b957 JS |
6440 | |
6441 | spin_unlock(&kvm->mmu_lock); | |
6442 | srcu_read_unlock(&kvm->srcu, rcu_idx); | |
6443 | } | |
6444 | ||
6445 | static long get_nx_lpage_recovery_timeout(u64 start_time) | |
6446 | { | |
6447 | return READ_ONCE(nx_huge_pages) && READ_ONCE(nx_huge_pages_recovery_ratio) | |
6448 | ? start_time + 60 * HZ - get_jiffies_64() | |
6449 | : MAX_SCHEDULE_TIMEOUT; | |
6450 | } | |
6451 | ||
6452 | static int kvm_nx_lpage_recovery_worker(struct kvm *kvm, uintptr_t data) | |
6453 | { | |
6454 | u64 start_time; | |
6455 | long remaining_time; | |
6456 | ||
6457 | while (true) { | |
6458 | start_time = get_jiffies_64(); | |
6459 | remaining_time = get_nx_lpage_recovery_timeout(start_time); | |
6460 | ||
6461 | set_current_state(TASK_INTERRUPTIBLE); | |
6462 | while (!kthread_should_stop() && remaining_time > 0) { | |
6463 | schedule_timeout(remaining_time); | |
6464 | remaining_time = get_nx_lpage_recovery_timeout(start_time); | |
6465 | set_current_state(TASK_INTERRUPTIBLE); | |
6466 | } | |
6467 | ||
6468 | set_current_state(TASK_RUNNING); | |
6469 | ||
6470 | if (kthread_should_stop()) | |
6471 | return 0; | |
6472 | ||
6473 | kvm_recover_nx_lpages(kvm); | |
6474 | } | |
6475 | } | |
6476 | ||
6477 | int kvm_mmu_post_init_vm(struct kvm *kvm) | |
6478 | { | |
6479 | int err; | |
6480 | ||
6481 | err = kvm_vm_create_worker_thread(kvm, kvm_nx_lpage_recovery_worker, 0, | |
6482 | "kvm-nx-lpage-recovery", | |
6483 | &kvm->arch.nx_lpage_recovery_thread); | |
6484 | if (!err) | |
6485 | kthread_unpark(kvm->arch.nx_lpage_recovery_thread); | |
6486 | ||
6487 | return err; | |
6488 | } | |
6489 | ||
6490 | void kvm_mmu_pre_destroy_vm(struct kvm *kvm) | |
6491 | { | |
6492 | if (kvm->arch.nx_lpage_recovery_thread) | |
6493 | kthread_stop(kvm->arch.nx_lpage_recovery_thread); | |
6494 | } |