Commit | Line | Data |
---|---|---|
fe5db27d BG |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | ||
02c00b3a BG |
3 | #include "mmu.h" |
4 | #include "mmu_internal.h" | |
bb18842e | 5 | #include "mmutrace.h" |
2f2fad08 | 6 | #include "tdp_iter.h" |
fe5db27d | 7 | #include "tdp_mmu.h" |
02c00b3a | 8 | #include "spte.h" |
fe5db27d | 9 | |
9a77daac | 10 | #include <asm/cmpxchg.h> |
33dd3574 BG |
11 | #include <trace/events/kvm.h> |
12 | ||
71ba3f31 | 13 | static bool __read_mostly tdp_mmu_enabled = true; |
95fb5b02 | 14 | module_param_named(tdp_mmu, tdp_mmu_enabled, bool, 0644); |
fe5db27d BG |
15 | |
16 | /* Initializes the TDP MMU for the VM, if enabled. */ | |
d501f747 | 17 | bool kvm_mmu_init_tdp_mmu(struct kvm *kvm) |
fe5db27d | 18 | { |
897218ff | 19 | if (!tdp_enabled || !READ_ONCE(tdp_mmu_enabled)) |
d501f747 | 20 | return false; |
fe5db27d BG |
21 | |
22 | /* This should not be changed for the lifetime of the VM. */ | |
23 | kvm->arch.tdp_mmu_enabled = true; | |
02c00b3a BG |
24 | |
25 | INIT_LIST_HEAD(&kvm->arch.tdp_mmu_roots); | |
9a77daac | 26 | spin_lock_init(&kvm->arch.tdp_mmu_pages_lock); |
89c0fd49 | 27 | INIT_LIST_HEAD(&kvm->arch.tdp_mmu_pages); |
d501f747 BG |
28 | |
29 | return true; | |
fe5db27d BG |
30 | } |
31 | ||
6103bc07 BG |
32 | static __always_inline void kvm_lockdep_assert_mmu_lock_held(struct kvm *kvm, |
33 | bool shared) | |
34 | { | |
35 | if (shared) | |
36 | lockdep_assert_held_read(&kvm->mmu_lock); | |
37 | else | |
38 | lockdep_assert_held_write(&kvm->mmu_lock); | |
39 | } | |
40 | ||
fe5db27d BG |
41 | void kvm_mmu_uninit_tdp_mmu(struct kvm *kvm) |
42 | { | |
43 | if (!kvm->arch.tdp_mmu_enabled) | |
44 | return; | |
02c00b3a | 45 | |
524a1e4e | 46 | WARN_ON(!list_empty(&kvm->arch.tdp_mmu_pages)); |
02c00b3a | 47 | WARN_ON(!list_empty(&kvm->arch.tdp_mmu_roots)); |
7cca2d0b BG |
48 | |
49 | /* | |
50 | * Ensure that all the outstanding RCU callbacks to free shadow pages | |
51 | * can run before the VM is torn down. | |
52 | */ | |
53 | rcu_barrier(); | |
02c00b3a BG |
54 | } |
55 | ||
2bdb3d84 | 56 | static bool zap_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root, |
6103bc07 BG |
57 | gfn_t start, gfn_t end, bool can_yield, bool flush, |
58 | bool shared); | |
2bdb3d84 BG |
59 | |
60 | static void tdp_mmu_free_sp(struct kvm_mmu_page *sp) | |
a889ea54 | 61 | { |
2bdb3d84 BG |
62 | free_page((unsigned long)sp->spt); |
63 | kmem_cache_free(mmu_page_header_cache, sp); | |
a889ea54 BG |
64 | } |
65 | ||
c0e64238 BG |
66 | /* |
67 | * This is called through call_rcu in order to free TDP page table memory | |
68 | * safely with respect to other kernel threads that may be operating on | |
69 | * the memory. | |
70 | * By only accessing TDP MMU page table memory in an RCU read critical | |
71 | * section, and freeing it after a grace period, lockless access to that | |
72 | * memory won't use it after it is freed. | |
73 | */ | |
74 | static void tdp_mmu_free_sp_rcu_callback(struct rcu_head *head) | |
a889ea54 | 75 | { |
c0e64238 BG |
76 | struct kvm_mmu_page *sp = container_of(head, struct kvm_mmu_page, |
77 | rcu_head); | |
a889ea54 | 78 | |
c0e64238 BG |
79 | tdp_mmu_free_sp(sp); |
80 | } | |
a889ea54 | 81 | |
6103bc07 BG |
82 | void kvm_tdp_mmu_put_root(struct kvm *kvm, struct kvm_mmu_page *root, |
83 | bool shared) | |
2bdb3d84 | 84 | { |
6103bc07 | 85 | kvm_lockdep_assert_mmu_lock_held(kvm, shared); |
a889ea54 | 86 | |
11cccf5c | 87 | if (!refcount_dec_and_test(&root->tdp_mmu_root_count)) |
2bdb3d84 BG |
88 | return; |
89 | ||
90 | WARN_ON(!root->tdp_mmu_page); | |
91 | ||
c0e64238 BG |
92 | spin_lock(&kvm->arch.tdp_mmu_pages_lock); |
93 | list_del_rcu(&root->link); | |
94 | spin_unlock(&kvm->arch.tdp_mmu_pages_lock); | |
2bdb3d84 | 95 | |
524a1e4e | 96 | zap_gfn_range(kvm, root, 0, -1ull, false, false, shared); |
2bdb3d84 | 97 | |
c0e64238 | 98 | call_rcu(&root->rcu_head, tdp_mmu_free_sp_rcu_callback); |
a889ea54 BG |
99 | } |
100 | ||
cfc10997 | 101 | /* |
d62007ed SC |
102 | * Returns the next root after @prev_root (or the first root if @prev_root is |
103 | * NULL). A reference to the returned root is acquired, and the reference to | |
104 | * @prev_root is released (the caller obviously must hold a reference to | |
105 | * @prev_root if it's non-NULL). | |
106 | * | |
107 | * If @only_valid is true, invalid roots are skipped. | |
108 | * | |
109 | * Returns NULL if the end of tdp_mmu_roots was reached. | |
cfc10997 BG |
110 | */ |
111 | static struct kvm_mmu_page *tdp_mmu_next_root(struct kvm *kvm, | |
6103bc07 | 112 | struct kvm_mmu_page *prev_root, |
d62007ed | 113 | bool shared, bool only_valid) |
a889ea54 BG |
114 | { |
115 | struct kvm_mmu_page *next_root; | |
116 | ||
c0e64238 BG |
117 | rcu_read_lock(); |
118 | ||
cfc10997 | 119 | if (prev_root) |
c0e64238 BG |
120 | next_root = list_next_or_null_rcu(&kvm->arch.tdp_mmu_roots, |
121 | &prev_root->link, | |
122 | typeof(*prev_root), link); | |
cfc10997 | 123 | else |
c0e64238 BG |
124 | next_root = list_first_or_null_rcu(&kvm->arch.tdp_mmu_roots, |
125 | typeof(*next_root), link); | |
a889ea54 | 126 | |
04dc4e6c | 127 | while (next_root) { |
d62007ed | 128 | if ((!only_valid || !next_root->role.invalid) && |
ad6d6b94 | 129 | kvm_tdp_mmu_get_root(next_root)) |
04dc4e6c SC |
130 | break; |
131 | ||
c0e64238 BG |
132 | next_root = list_next_or_null_rcu(&kvm->arch.tdp_mmu_roots, |
133 | &next_root->link, typeof(*next_root), link); | |
04dc4e6c | 134 | } |
fb101293 | 135 | |
c0e64238 | 136 | rcu_read_unlock(); |
a889ea54 | 137 | |
cfc10997 | 138 | if (prev_root) |
6103bc07 | 139 | kvm_tdp_mmu_put_root(kvm, prev_root, shared); |
a889ea54 | 140 | |
a889ea54 BG |
141 | return next_root; |
142 | } | |
143 | ||
144 | /* | |
145 | * Note: this iterator gets and puts references to the roots it iterates over. | |
146 | * This makes it safe to release the MMU lock and yield within the loop, but | |
147 | * if exiting the loop early, the caller must drop the reference to the most | |
148 | * recent root. (Unless keeping a live reference is desirable.) | |
6103bc07 BG |
149 | * |
150 | * If shared is set, this function is operating under the MMU lock in read | |
151 | * mode. In the unlikely event that this thread must free a root, the lock | |
152 | * will be temporarily dropped and reacquired in write mode. | |
a889ea54 | 153 | */ |
d62007ed SC |
154 | #define __for_each_tdp_mmu_root_yield_safe(_kvm, _root, _as_id, _shared, _only_valid)\ |
155 | for (_root = tdp_mmu_next_root(_kvm, NULL, _shared, _only_valid); \ | |
156 | _root; \ | |
157 | _root = tdp_mmu_next_root(_kvm, _root, _shared, _only_valid)) \ | |
158 | if (kvm_mmu_page_as_id(_root) != _as_id) { \ | |
a3f15bda | 159 | } else |
a889ea54 | 160 | |
d62007ed SC |
161 | #define for_each_valid_tdp_mmu_root_yield_safe(_kvm, _root, _as_id, _shared) \ |
162 | __for_each_tdp_mmu_root_yield_safe(_kvm, _root, _as_id, _shared, true) | |
163 | ||
164 | #define for_each_tdp_mmu_root_yield_safe(_kvm, _root, _as_id, _shared) \ | |
165 | __for_each_tdp_mmu_root_yield_safe(_kvm, _root, _as_id, _shared, false) | |
166 | ||
c0e64238 BG |
167 | #define for_each_tdp_mmu_root(_kvm, _root, _as_id) \ |
168 | list_for_each_entry_rcu(_root, &_kvm->arch.tdp_mmu_roots, link, \ | |
169 | lockdep_is_held_type(&kvm->mmu_lock, 0) || \ | |
170 | lockdep_is_held(&kvm->arch.tdp_mmu_pages_lock)) \ | |
a3f15bda SC |
171 | if (kvm_mmu_page_as_id(_root) != _as_id) { \ |
172 | } else | |
02c00b3a | 173 | |
a82070b6 | 174 | static struct kvm_mmu_page *tdp_mmu_alloc_sp(struct kvm_vcpu *vcpu) |
02c00b3a BG |
175 | { |
176 | struct kvm_mmu_page *sp; | |
177 | ||
178 | sp = kvm_mmu_memory_cache_alloc(&vcpu->arch.mmu_page_header_cache); | |
179 | sp->spt = kvm_mmu_memory_cache_alloc(&vcpu->arch.mmu_shadow_page_cache); | |
a82070b6 DM |
180 | |
181 | return sp; | |
182 | } | |
183 | ||
184 | static void tdp_mmu_init_sp(struct kvm_mmu_page *sp, gfn_t gfn, | |
185 | union kvm_mmu_page_role role) | |
186 | { | |
02c00b3a BG |
187 | set_page_private(virt_to_page(sp->spt), (unsigned long)sp); |
188 | ||
a3aca4de | 189 | sp->role = role; |
02c00b3a BG |
190 | sp->gfn = gfn; |
191 | sp->tdp_mmu_page = true; | |
192 | ||
33dd3574 | 193 | trace_kvm_mmu_get_page(sp, true); |
02c00b3a BG |
194 | } |
195 | ||
a82070b6 DM |
196 | static void tdp_mmu_init_child_sp(struct kvm_mmu_page *child_sp, |
197 | struct tdp_iter *iter) | |
02c00b3a | 198 | { |
a3aca4de | 199 | struct kvm_mmu_page *parent_sp; |
02c00b3a | 200 | union kvm_mmu_page_role role; |
a3aca4de DM |
201 | |
202 | parent_sp = sptep_to_sp(rcu_dereference(iter->sptep)); | |
203 | ||
204 | role = parent_sp->role; | |
205 | role.level--; | |
206 | ||
a82070b6 | 207 | tdp_mmu_init_sp(child_sp, iter->gfn, role); |
a3aca4de DM |
208 | } |
209 | ||
210 | hpa_t kvm_tdp_mmu_get_vcpu_root_hpa(struct kvm_vcpu *vcpu) | |
211 | { | |
212 | union kvm_mmu_page_role role = vcpu->arch.mmu->mmu_role.base; | |
02c00b3a BG |
213 | struct kvm *kvm = vcpu->kvm; |
214 | struct kvm_mmu_page *root; | |
215 | ||
6e6ec584 | 216 | lockdep_assert_held_write(&kvm->mmu_lock); |
02c00b3a | 217 | |
04dc4e6c SC |
218 | /* |
219 | * Check for an existing root before allocating a new one. Note, the | |
220 | * role check prevents consuming an invalid root. | |
221 | */ | |
a3f15bda | 222 | for_each_tdp_mmu_root(kvm, root, kvm_mmu_role_as_id(role)) { |
fb101293 | 223 | if (root->role.word == role.word && |
ad6d6b94 | 224 | kvm_tdp_mmu_get_root(root)) |
6e6ec584 | 225 | goto out; |
02c00b3a BG |
226 | } |
227 | ||
a82070b6 DM |
228 | root = tdp_mmu_alloc_sp(vcpu); |
229 | tdp_mmu_init_sp(root, 0, role); | |
230 | ||
11cccf5c | 231 | refcount_set(&root->tdp_mmu_root_count, 1); |
02c00b3a | 232 | |
c0e64238 BG |
233 | spin_lock(&kvm->arch.tdp_mmu_pages_lock); |
234 | list_add_rcu(&root->link, &kvm->arch.tdp_mmu_roots); | |
235 | spin_unlock(&kvm->arch.tdp_mmu_pages_lock); | |
02c00b3a | 236 | |
6e6ec584 | 237 | out: |
02c00b3a | 238 | return __pa(root->spt); |
fe5db27d | 239 | } |
2f2fad08 BG |
240 | |
241 | static void handle_changed_spte(struct kvm *kvm, int as_id, gfn_t gfn, | |
9a77daac BG |
242 | u64 old_spte, u64 new_spte, int level, |
243 | bool shared); | |
2f2fad08 | 244 | |
f8e14497 BG |
245 | static void handle_changed_spte_acc_track(u64 old_spte, u64 new_spte, int level) |
246 | { | |
f8e14497 BG |
247 | if (!is_shadow_present_pte(old_spte) || !is_last_spte(old_spte, level)) |
248 | return; | |
249 | ||
250 | if (is_accessed_spte(old_spte) && | |
64bb2769 SC |
251 | (!is_shadow_present_pte(new_spte) || !is_accessed_spte(new_spte) || |
252 | spte_to_pfn(old_spte) != spte_to_pfn(new_spte))) | |
f8e14497 BG |
253 | kvm_set_pfn_accessed(spte_to_pfn(old_spte)); |
254 | } | |
255 | ||
a6a0b05d BG |
256 | static void handle_changed_spte_dirty_log(struct kvm *kvm, int as_id, gfn_t gfn, |
257 | u64 old_spte, u64 new_spte, int level) | |
258 | { | |
259 | bool pfn_changed; | |
260 | struct kvm_memory_slot *slot; | |
261 | ||
262 | if (level > PG_LEVEL_4K) | |
263 | return; | |
264 | ||
265 | pfn_changed = spte_to_pfn(old_spte) != spte_to_pfn(new_spte); | |
266 | ||
267 | if ((!is_writable_pte(old_spte) || pfn_changed) && | |
268 | is_writable_pte(new_spte)) { | |
269 | slot = __gfn_to_memslot(__kvm_memslots(kvm, as_id), gfn); | |
fb04a1ed | 270 | mark_page_dirty_in_slot(kvm, slot, gfn); |
a6a0b05d BG |
271 | } |
272 | } | |
273 | ||
a9442f59 | 274 | /** |
c298a30c | 275 | * tdp_mmu_unlink_sp() - Remove a shadow page from the list of used pages |
a9442f59 BG |
276 | * |
277 | * @kvm: kvm instance | |
278 | * @sp: the page to be removed | |
9a77daac BG |
279 | * @shared: This operation may not be running under the exclusive use of |
280 | * the MMU lock and the operation must synchronize with other | |
281 | * threads that might be adding or removing pages. | |
a9442f59 | 282 | */ |
c298a30c DM |
283 | static void tdp_mmu_unlink_sp(struct kvm *kvm, struct kvm_mmu_page *sp, |
284 | bool shared) | |
a9442f59 | 285 | { |
9a77daac BG |
286 | if (shared) |
287 | spin_lock(&kvm->arch.tdp_mmu_pages_lock); | |
288 | else | |
289 | lockdep_assert_held_write(&kvm->mmu_lock); | |
a9442f59 BG |
290 | |
291 | list_del(&sp->link); | |
292 | if (sp->lpage_disallowed) | |
293 | unaccount_huge_nx_page(kvm, sp); | |
9a77daac BG |
294 | |
295 | if (shared) | |
296 | spin_unlock(&kvm->arch.tdp_mmu_pages_lock); | |
a9442f59 BG |
297 | } |
298 | ||
a066e61f | 299 | /** |
0f53dfa3 | 300 | * handle_removed_pt() - handle a page table removed from the TDP structure |
a066e61f BG |
301 | * |
302 | * @kvm: kvm instance | |
303 | * @pt: the page removed from the paging structure | |
9a77daac BG |
304 | * @shared: This operation may not be running under the exclusive use |
305 | * of the MMU lock and the operation must synchronize with other | |
306 | * threads that might be modifying SPTEs. | |
a066e61f BG |
307 | * |
308 | * Given a page table that has been removed from the TDP paging structure, | |
309 | * iterates through the page table to clear SPTEs and free child page tables. | |
70fb3e41 BG |
310 | * |
311 | * Note that pt is passed in as a tdp_ptep_t, but it does not need RCU | |
312 | * protection. Since this thread removed it from the paging structure, | |
313 | * this thread will be responsible for ensuring the page is freed. Hence the | |
314 | * early rcu_dereferences in the function. | |
a066e61f | 315 | */ |
0f53dfa3 | 316 | static void handle_removed_pt(struct kvm *kvm, tdp_ptep_t pt, bool shared) |
a066e61f | 317 | { |
70fb3e41 | 318 | struct kvm_mmu_page *sp = sptep_to_sp(rcu_dereference(pt)); |
a066e61f | 319 | int level = sp->role.level; |
e25f0e0c | 320 | gfn_t base_gfn = sp->gfn; |
a066e61f BG |
321 | int i; |
322 | ||
323 | trace_kvm_mmu_prepare_zap_page(sp); | |
324 | ||
c298a30c | 325 | tdp_mmu_unlink_sp(kvm, sp, shared); |
a066e61f BG |
326 | |
327 | for (i = 0; i < PT64_ENT_PER_PAGE; i++) { | |
574c3c55 BG |
328 | u64 *sptep = rcu_dereference(pt) + i; |
329 | gfn_t gfn = base_gfn + i * KVM_PAGES_PER_HPAGE(level); | |
330 | u64 old_child_spte; | |
9a77daac BG |
331 | |
332 | if (shared) { | |
e25f0e0c BG |
333 | /* |
334 | * Set the SPTE to a nonpresent value that other | |
335 | * threads will not overwrite. If the SPTE was | |
336 | * already marked as removed then another thread | |
337 | * handling a page fault could overwrite it, so | |
338 | * set the SPTE until it is set from some other | |
339 | * value to the removed SPTE value. | |
340 | */ | |
341 | for (;;) { | |
342 | old_child_spte = xchg(sptep, REMOVED_SPTE); | |
343 | if (!is_removed_spte(old_child_spte)) | |
344 | break; | |
345 | cpu_relax(); | |
346 | } | |
9a77daac | 347 | } else { |
8df9f1af SC |
348 | /* |
349 | * If the SPTE is not MMU-present, there is no backing | |
350 | * page associated with the SPTE and so no side effects | |
351 | * that need to be recorded, and exclusive ownership of | |
352 | * mmu_lock ensures the SPTE can't be made present. | |
353 | * Note, zapping MMIO SPTEs is also unnecessary as they | |
354 | * are guarded by the memslots generation, not by being | |
355 | * unreachable. | |
356 | */ | |
9a77daac | 357 | old_child_spte = READ_ONCE(*sptep); |
8df9f1af SC |
358 | if (!is_shadow_present_pte(old_child_spte)) |
359 | continue; | |
e25f0e0c BG |
360 | |
361 | /* | |
362 | * Marking the SPTE as a removed SPTE is not | |
363 | * strictly necessary here as the MMU lock will | |
364 | * stop other threads from concurrently modifying | |
365 | * this SPTE. Using the removed SPTE value keeps | |
366 | * the two branches consistent and simplifies | |
367 | * the function. | |
368 | */ | |
369 | WRITE_ONCE(*sptep, REMOVED_SPTE); | |
9a77daac | 370 | } |
e25f0e0c | 371 | handle_changed_spte(kvm, kvm_mmu_page_as_id(sp), gfn, |
f1b83255 | 372 | old_child_spte, REMOVED_SPTE, level, |
e25f0e0c | 373 | shared); |
a066e61f BG |
374 | } |
375 | ||
574c3c55 | 376 | kvm_flush_remote_tlbs_with_address(kvm, base_gfn, |
f1b83255 | 377 | KVM_PAGES_PER_HPAGE(level + 1)); |
a066e61f | 378 | |
7cca2d0b | 379 | call_rcu(&sp->rcu_head, tdp_mmu_free_sp_rcu_callback); |
a066e61f BG |
380 | } |
381 | ||
2f2fad08 | 382 | /** |
7f6231a3 | 383 | * __handle_changed_spte - handle bookkeeping associated with an SPTE change |
2f2fad08 BG |
384 | * @kvm: kvm instance |
385 | * @as_id: the address space of the paging structure the SPTE was a part of | |
386 | * @gfn: the base GFN that was mapped by the SPTE | |
387 | * @old_spte: The value of the SPTE before the change | |
388 | * @new_spte: The value of the SPTE after the change | |
389 | * @level: the level of the PT the SPTE is part of in the paging structure | |
9a77daac BG |
390 | * @shared: This operation may not be running under the exclusive use of |
391 | * the MMU lock and the operation must synchronize with other | |
392 | * threads that might be modifying SPTEs. | |
2f2fad08 BG |
393 | * |
394 | * Handle bookkeeping that might result from the modification of a SPTE. | |
395 | * This function must be called for all TDP SPTE modifications. | |
396 | */ | |
397 | static void __handle_changed_spte(struct kvm *kvm, int as_id, gfn_t gfn, | |
9a77daac BG |
398 | u64 old_spte, u64 new_spte, int level, |
399 | bool shared) | |
2f2fad08 BG |
400 | { |
401 | bool was_present = is_shadow_present_pte(old_spte); | |
402 | bool is_present = is_shadow_present_pte(new_spte); | |
403 | bool was_leaf = was_present && is_last_spte(old_spte, level); | |
404 | bool is_leaf = is_present && is_last_spte(new_spte, level); | |
405 | bool pfn_changed = spte_to_pfn(old_spte) != spte_to_pfn(new_spte); | |
2f2fad08 BG |
406 | |
407 | WARN_ON(level > PT64_ROOT_MAX_LEVEL); | |
408 | WARN_ON(level < PG_LEVEL_4K); | |
764388ce | 409 | WARN_ON(gfn & (KVM_PAGES_PER_HPAGE(level) - 1)); |
2f2fad08 BG |
410 | |
411 | /* | |
412 | * If this warning were to trigger it would indicate that there was a | |
413 | * missing MMU notifier or a race with some notifier handler. | |
414 | * A present, leaf SPTE should never be directly replaced with another | |
d9f6e12f | 415 | * present leaf SPTE pointing to a different PFN. A notifier handler |
2f2fad08 BG |
416 | * should be zapping the SPTE before the main MM's page table is |
417 | * changed, or the SPTE should be zeroed, and the TLBs flushed by the | |
418 | * thread before replacement. | |
419 | */ | |
420 | if (was_leaf && is_leaf && pfn_changed) { | |
421 | pr_err("Invalid SPTE change: cannot replace a present leaf\n" | |
422 | "SPTE with another present leaf SPTE mapping a\n" | |
423 | "different PFN!\n" | |
424 | "as_id: %d gfn: %llx old_spte: %llx new_spte: %llx level: %d", | |
425 | as_id, gfn, old_spte, new_spte, level); | |
426 | ||
427 | /* | |
428 | * Crash the host to prevent error propagation and guest data | |
d9f6e12f | 429 | * corruption. |
2f2fad08 BG |
430 | */ |
431 | BUG(); | |
432 | } | |
433 | ||
434 | if (old_spte == new_spte) | |
435 | return; | |
436 | ||
b9a98c34 BG |
437 | trace_kvm_tdp_mmu_spte_changed(as_id, gfn, level, old_spte, new_spte); |
438 | ||
115111ef DM |
439 | if (is_leaf) |
440 | check_spte_writable_invariants(new_spte); | |
441 | ||
2f2fad08 BG |
442 | /* |
443 | * The only times a SPTE should be changed from a non-present to | |
444 | * non-present state is when an MMIO entry is installed/modified/ | |
445 | * removed. In that case, there is nothing to do here. | |
446 | */ | |
447 | if (!was_present && !is_present) { | |
448 | /* | |
08f07c80 BG |
449 | * If this change does not involve a MMIO SPTE or removed SPTE, |
450 | * it is unexpected. Log the change, though it should not | |
451 | * impact the guest since both the former and current SPTEs | |
452 | * are nonpresent. | |
2f2fad08 | 453 | */ |
08f07c80 BG |
454 | if (WARN_ON(!is_mmio_spte(old_spte) && |
455 | !is_mmio_spte(new_spte) && | |
456 | !is_removed_spte(new_spte))) | |
2f2fad08 BG |
457 | pr_err("Unexpected SPTE change! Nonpresent SPTEs\n" |
458 | "should not be replaced with another,\n" | |
459 | "different nonpresent SPTE, unless one or both\n" | |
08f07c80 BG |
460 | "are MMIO SPTEs, or the new SPTE is\n" |
461 | "a temporary removed SPTE.\n" | |
2f2fad08 BG |
462 | "as_id: %d gfn: %llx old_spte: %llx new_spte: %llx level: %d", |
463 | as_id, gfn, old_spte, new_spte, level); | |
464 | return; | |
465 | } | |
466 | ||
71f51d2c MZ |
467 | if (is_leaf != was_leaf) |
468 | kvm_update_page_stats(kvm, level, is_leaf ? 1 : -1); | |
2f2fad08 BG |
469 | |
470 | if (was_leaf && is_dirty_spte(old_spte) && | |
64bb2769 | 471 | (!is_present || !is_dirty_spte(new_spte) || pfn_changed)) |
2f2fad08 BG |
472 | kvm_set_pfn_dirty(spte_to_pfn(old_spte)); |
473 | ||
474 | /* | |
475 | * Recursively handle child PTs if the change removed a subtree from | |
476 | * the paging structure. | |
477 | */ | |
a066e61f | 478 | if (was_present && !was_leaf && (pfn_changed || !is_present)) |
0f53dfa3 | 479 | handle_removed_pt(kvm, spte_to_child_pt(old_spte, level), shared); |
2f2fad08 BG |
480 | } |
481 | ||
482 | static void handle_changed_spte(struct kvm *kvm, int as_id, gfn_t gfn, | |
9a77daac BG |
483 | u64 old_spte, u64 new_spte, int level, |
484 | bool shared) | |
2f2fad08 | 485 | { |
9a77daac BG |
486 | __handle_changed_spte(kvm, as_id, gfn, old_spte, new_spte, level, |
487 | shared); | |
f8e14497 | 488 | handle_changed_spte_acc_track(old_spte, new_spte, level); |
a6a0b05d BG |
489 | handle_changed_spte_dirty_log(kvm, as_id, gfn, old_spte, |
490 | new_spte, level); | |
2f2fad08 | 491 | } |
faaf05b0 | 492 | |
9a77daac | 493 | /* |
6ccf4438 PB |
494 | * tdp_mmu_set_spte_atomic - Set a TDP MMU SPTE atomically |
495 | * and handle the associated bookkeeping. Do not mark the page dirty | |
24ae4cfa | 496 | * in KVM's dirty bitmaps. |
9a77daac | 497 | * |
3255530a DM |
498 | * If setting the SPTE fails because it has changed, iter->old_spte will be |
499 | * refreshed to the current value of the spte. | |
500 | * | |
9a77daac BG |
501 | * @kvm: kvm instance |
502 | * @iter: a tdp_iter instance currently on the SPTE that should be set | |
503 | * @new_spte: The value the SPTE should be set to | |
3e72c791 DM |
504 | * Return: |
505 | * * 0 - If the SPTE was set. | |
506 | * * -EBUSY - If the SPTE cannot be set. In this case this function will have | |
507 | * no side-effects other than setting iter->old_spte to the last | |
508 | * known value of the spte. | |
9a77daac | 509 | */ |
3e72c791 DM |
510 | static inline int tdp_mmu_set_spte_atomic(struct kvm *kvm, |
511 | struct tdp_iter *iter, | |
512 | u64 new_spte) | |
9a77daac | 513 | { |
3255530a DM |
514 | u64 *sptep = rcu_dereference(iter->sptep); |
515 | u64 old_spte; | |
516 | ||
3a0f64de SC |
517 | WARN_ON_ONCE(iter->yielded); |
518 | ||
9a77daac BG |
519 | lockdep_assert_held_read(&kvm->mmu_lock); |
520 | ||
08f07c80 BG |
521 | /* |
522 | * Do not change removed SPTEs. Only the thread that froze the SPTE | |
523 | * may modify it. | |
524 | */ | |
7a51393a | 525 | if (is_removed_spte(iter->old_spte)) |
3e72c791 | 526 | return -EBUSY; |
08f07c80 | 527 | |
6e8eb206 DM |
528 | /* |
529 | * Note, fast_pf_fix_direct_spte() can also modify TDP MMU SPTEs and | |
530 | * does not hold the mmu_lock. | |
531 | */ | |
3255530a DM |
532 | old_spte = cmpxchg64(sptep, iter->old_spte, new_spte); |
533 | if (old_spte != iter->old_spte) { | |
534 | /* | |
535 | * The page table entry was modified by a different logical | |
536 | * CPU. Refresh iter->old_spte with the current value so the | |
537 | * caller operates on fresh data, e.g. if it retries | |
538 | * tdp_mmu_set_spte_atomic(). | |
539 | */ | |
540 | iter->old_spte = old_spte; | |
3e72c791 | 541 | return -EBUSY; |
3255530a | 542 | } |
9a77daac | 543 | |
24ae4cfa BG |
544 | __handle_changed_spte(kvm, iter->as_id, iter->gfn, iter->old_spte, |
545 | new_spte, iter->level, true); | |
546 | handle_changed_spte_acc_track(iter->old_spte, new_spte, iter->level); | |
9a77daac | 547 | |
3e72c791 | 548 | return 0; |
9a77daac BG |
549 | } |
550 | ||
3e72c791 DM |
551 | static inline int tdp_mmu_zap_spte_atomic(struct kvm *kvm, |
552 | struct tdp_iter *iter) | |
08f07c80 | 553 | { |
3e72c791 DM |
554 | int ret; |
555 | ||
08f07c80 BG |
556 | /* |
557 | * Freeze the SPTE by setting it to a special, | |
558 | * non-present value. This will stop other threads from | |
559 | * immediately installing a present entry in its place | |
560 | * before the TLBs are flushed. | |
561 | */ | |
3e72c791 DM |
562 | ret = tdp_mmu_set_spte_atomic(kvm, iter, REMOVED_SPTE); |
563 | if (ret) | |
564 | return ret; | |
08f07c80 BG |
565 | |
566 | kvm_flush_remote_tlbs_with_address(kvm, iter->gfn, | |
567 | KVM_PAGES_PER_HPAGE(iter->level)); | |
568 | ||
569 | /* | |
570 | * No other thread can overwrite the removed SPTE as they | |
571 | * must either wait on the MMU lock or use | |
d9f6e12f | 572 | * tdp_mmu_set_spte_atomic which will not overwrite the |
08f07c80 BG |
573 | * special removed SPTE value. No bookkeeping is needed |
574 | * here since the SPTE is going from non-present | |
575 | * to non-present. | |
576 | */ | |
14f6fec2 | 577 | WRITE_ONCE(*rcu_dereference(iter->sptep), 0); |
08f07c80 | 578 | |
3e72c791 | 579 | return 0; |
08f07c80 BG |
580 | } |
581 | ||
9a77daac | 582 | |
fe43fa2f BG |
583 | /* |
584 | * __tdp_mmu_set_spte - Set a TDP MMU SPTE and handle the associated bookkeeping | |
585 | * @kvm: kvm instance | |
586 | * @iter: a tdp_iter instance currently on the SPTE that should be set | |
587 | * @new_spte: The value the SPTE should be set to | |
588 | * @record_acc_track: Notify the MM subsystem of changes to the accessed state | |
589 | * of the page. Should be set unless handling an MMU | |
590 | * notifier for access tracking. Leaving record_acc_track | |
591 | * unset in that case prevents page accesses from being | |
592 | * double counted. | |
593 | * @record_dirty_log: Record the page as dirty in the dirty bitmap if | |
594 | * appropriate for the change being made. Should be set | |
595 | * unless performing certain dirty logging operations. | |
596 | * Leaving record_dirty_log unset in that case prevents page | |
597 | * writes from being double counted. | |
598 | */ | |
f8e14497 | 599 | static inline void __tdp_mmu_set_spte(struct kvm *kvm, struct tdp_iter *iter, |
a6a0b05d BG |
600 | u64 new_spte, bool record_acc_track, |
601 | bool record_dirty_log) | |
faaf05b0 | 602 | { |
3a0f64de SC |
603 | WARN_ON_ONCE(iter->yielded); |
604 | ||
531810ca | 605 | lockdep_assert_held_write(&kvm->mmu_lock); |
3a9a4aa5 | 606 | |
08f07c80 BG |
607 | /* |
608 | * No thread should be using this function to set SPTEs to the | |
609 | * temporary removed SPTE value. | |
610 | * If operating under the MMU lock in read mode, tdp_mmu_set_spte_atomic | |
611 | * should be used. If operating under the MMU lock in write mode, the | |
612 | * use of the removed SPTE should not be necessary. | |
613 | */ | |
7a51393a | 614 | WARN_ON(is_removed_spte(iter->old_spte)); |
08f07c80 | 615 | |
7cca2d0b | 616 | WRITE_ONCE(*rcu_dereference(iter->sptep), new_spte); |
f8e14497 | 617 | |
08889894 SC |
618 | __handle_changed_spte(kvm, iter->as_id, iter->gfn, iter->old_spte, |
619 | new_spte, iter->level, false); | |
f8e14497 BG |
620 | if (record_acc_track) |
621 | handle_changed_spte_acc_track(iter->old_spte, new_spte, | |
622 | iter->level); | |
a6a0b05d | 623 | if (record_dirty_log) |
08889894 | 624 | handle_changed_spte_dirty_log(kvm, iter->as_id, iter->gfn, |
a6a0b05d BG |
625 | iter->old_spte, new_spte, |
626 | iter->level); | |
f8e14497 BG |
627 | } |
628 | ||
629 | static inline void tdp_mmu_set_spte(struct kvm *kvm, struct tdp_iter *iter, | |
630 | u64 new_spte) | |
631 | { | |
a6a0b05d | 632 | __tdp_mmu_set_spte(kvm, iter, new_spte, true, true); |
f8e14497 | 633 | } |
faaf05b0 | 634 | |
f8e14497 BG |
635 | static inline void tdp_mmu_set_spte_no_acc_track(struct kvm *kvm, |
636 | struct tdp_iter *iter, | |
637 | u64 new_spte) | |
638 | { | |
a6a0b05d BG |
639 | __tdp_mmu_set_spte(kvm, iter, new_spte, false, true); |
640 | } | |
641 | ||
642 | static inline void tdp_mmu_set_spte_no_dirty_log(struct kvm *kvm, | |
643 | struct tdp_iter *iter, | |
644 | u64 new_spte) | |
645 | { | |
646 | __tdp_mmu_set_spte(kvm, iter, new_spte, true, false); | |
faaf05b0 BG |
647 | } |
648 | ||
649 | #define tdp_root_for_each_pte(_iter, _root, _start, _end) \ | |
77aa6075 | 650 | for_each_tdp_pte(_iter, _root, _start, _end) |
faaf05b0 | 651 | |
f8e14497 BG |
652 | #define tdp_root_for_each_leaf_pte(_iter, _root, _start, _end) \ |
653 | tdp_root_for_each_pte(_iter, _root, _start, _end) \ | |
654 | if (!is_shadow_present_pte(_iter.old_spte) || \ | |
655 | !is_last_spte(_iter.old_spte, _iter.level)) \ | |
656 | continue; \ | |
657 | else | |
658 | ||
bb18842e | 659 | #define tdp_mmu_for_each_pte(_iter, _mmu, _start, _end) \ |
77aa6075 | 660 | for_each_tdp_pte(_iter, to_shadow_page(_mmu->root_hpa), _start, _end) |
bb18842e | 661 | |
e28a436c BG |
662 | /* |
663 | * Yield if the MMU lock is contended or this thread needs to return control | |
664 | * to the scheduler. | |
665 | * | |
e139a34e BG |
666 | * If this function should yield and flush is set, it will perform a remote |
667 | * TLB flush before yielding. | |
668 | * | |
3a0f64de SC |
669 | * If this function yields, iter->yielded is set and the caller must skip to |
670 | * the next iteration, where tdp_iter_next() will reset the tdp_iter's walk | |
671 | * over the paging structures to allow the iterator to continue its traversal | |
672 | * from the paging structure root. | |
e28a436c | 673 | * |
3a0f64de | 674 | * Returns true if this function yielded. |
e28a436c | 675 | */ |
3a0f64de SC |
676 | static inline bool __must_check tdp_mmu_iter_cond_resched(struct kvm *kvm, |
677 | struct tdp_iter *iter, | |
678 | bool flush, bool shared) | |
a6a0b05d | 679 | { |
3a0f64de SC |
680 | WARN_ON(iter->yielded); |
681 | ||
ed5e484b BG |
682 | /* Ensure forward progress has been made before yielding. */ |
683 | if (iter->next_last_level_gfn == iter->yielded_gfn) | |
684 | return false; | |
685 | ||
531810ca | 686 | if (need_resched() || rwlock_needbreak(&kvm->mmu_lock)) { |
7cca2d0b BG |
687 | rcu_read_unlock(); |
688 | ||
e139a34e BG |
689 | if (flush) |
690 | kvm_flush_remote_tlbs(kvm); | |
691 | ||
6103bc07 BG |
692 | if (shared) |
693 | cond_resched_rwlock_read(&kvm->mmu_lock); | |
694 | else | |
695 | cond_resched_rwlock_write(&kvm->mmu_lock); | |
696 | ||
7cca2d0b | 697 | rcu_read_lock(); |
ed5e484b BG |
698 | |
699 | WARN_ON(iter->gfn > iter->next_last_level_gfn); | |
700 | ||
3a0f64de | 701 | iter->yielded = true; |
a6a0b05d | 702 | } |
e28a436c | 703 | |
3a0f64de | 704 | return iter->yielded; |
a6a0b05d BG |
705 | } |
706 | ||
faaf05b0 BG |
707 | /* |
708 | * Tears down the mappings for the range of gfns, [start, end), and frees the | |
709 | * non-root pages mapping GFNs strictly within that range. Returns true if | |
710 | * SPTEs have been cleared and a TLB flush is needed before releasing the | |
711 | * MMU lock. | |
6103bc07 | 712 | * |
063afacd BG |
713 | * If can_yield is true, will release the MMU lock and reschedule if the |
714 | * scheduler needs the CPU or there is contention on the MMU lock. If this | |
715 | * function cannot yield, it will not release the MMU lock or reschedule and | |
716 | * the caller must ensure it does not supply too large a GFN range, or the | |
6103bc07 BG |
717 | * operation can cause a soft lockup. |
718 | * | |
719 | * If shared is true, this thread holds the MMU lock in read mode and must | |
720 | * account for the possibility that other threads are modifying the paging | |
721 | * structures concurrently. If shared is false, this thread should hold the | |
722 | * MMU lock in write mode. | |
faaf05b0 BG |
723 | */ |
724 | static bool zap_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root, | |
6103bc07 BG |
725 | gfn_t start, gfn_t end, bool can_yield, bool flush, |
726 | bool shared) | |
faaf05b0 | 727 | { |
524a1e4e SC |
728 | gfn_t max_gfn_host = 1ULL << (shadow_phys_bits - PAGE_SHIFT); |
729 | bool zap_all = (start == 0 && end >= max_gfn_host); | |
faaf05b0 | 730 | struct tdp_iter iter; |
faaf05b0 | 731 | |
0103098f SC |
732 | /* |
733 | * No need to try to step down in the iterator when zapping all SPTEs, | |
734 | * zapping the top-level non-leaf SPTEs will recurse on their children. | |
735 | */ | |
736 | int min_level = zap_all ? root->role.level : PG_LEVEL_4K; | |
737 | ||
524a1e4e SC |
738 | /* |
739 | * Bound the walk at host.MAXPHYADDR, guest accesses beyond that will | |
740 | * hit a #PF(RSVD) and never get to an EPT Violation/Misconfig / #NPF, | |
741 | * and so KVM will never install a SPTE for such addresses. | |
742 | */ | |
743 | end = min(end, max_gfn_host); | |
744 | ||
6103bc07 BG |
745 | kvm_lockdep_assert_mmu_lock_held(kvm, shared); |
746 | ||
7cca2d0b BG |
747 | rcu_read_lock(); |
748 | ||
77aa6075 | 749 | for_each_tdp_pte_min_level(iter, root, min_level, start, end) { |
6103bc07 | 750 | retry: |
1af4a960 | 751 | if (can_yield && |
6103bc07 | 752 | tdp_mmu_iter_cond_resched(kvm, &iter, flush, shared)) { |
a835429c | 753 | flush = false; |
1af4a960 BG |
754 | continue; |
755 | } | |
756 | ||
faaf05b0 BG |
757 | if (!is_shadow_present_pte(iter.old_spte)) |
758 | continue; | |
759 | ||
760 | /* | |
761 | * If this is a non-last-level SPTE that covers a larger range | |
762 | * than should be zapped, continue, and zap the mappings at a | |
524a1e4e | 763 | * lower level, except when zapping all SPTEs. |
faaf05b0 | 764 | */ |
524a1e4e SC |
765 | if (!zap_all && |
766 | (iter.gfn < start || | |
faaf05b0 BG |
767 | iter.gfn + KVM_PAGES_PER_HPAGE(iter.level) > end) && |
768 | !is_last_spte(iter.old_spte, iter.level)) | |
769 | continue; | |
770 | ||
6103bc07 BG |
771 | if (!shared) { |
772 | tdp_mmu_set_spte(kvm, &iter, 0); | |
773 | flush = true; | |
3e72c791 | 774 | } else if (tdp_mmu_zap_spte_atomic(kvm, &iter)) { |
6103bc07 BG |
775 | goto retry; |
776 | } | |
faaf05b0 | 777 | } |
7cca2d0b BG |
778 | |
779 | rcu_read_unlock(); | |
a835429c | 780 | return flush; |
faaf05b0 BG |
781 | } |
782 | ||
783 | /* | |
784 | * Tears down the mappings for the range of gfns, [start, end), and frees the | |
785 | * non-root pages mapping GFNs strictly within that range. Returns true if | |
786 | * SPTEs have been cleared and a TLB flush is needed before releasing the | |
787 | * MMU lock. | |
788 | */ | |
2b9663d8 | 789 | bool __kvm_tdp_mmu_zap_gfn_range(struct kvm *kvm, int as_id, gfn_t start, |
5a324c24 | 790 | gfn_t end, bool can_yield, bool flush) |
faaf05b0 BG |
791 | { |
792 | struct kvm_mmu_page *root; | |
faaf05b0 | 793 | |
5a324c24 | 794 | for_each_tdp_mmu_root_yield_safe(kvm, root, as_id, false) |
6103bc07 | 795 | flush = zap_gfn_range(kvm, root, start, end, can_yield, flush, |
5a324c24 | 796 | false); |
faaf05b0 | 797 | |
faaf05b0 BG |
798 | return flush; |
799 | } | |
800 | ||
801 | void kvm_tdp_mmu_zap_all(struct kvm *kvm) | |
802 | { | |
2b9663d8 SC |
803 | bool flush = false; |
804 | int i; | |
805 | ||
806 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) | |
5a324c24 | 807 | flush = kvm_tdp_mmu_zap_gfn_range(kvm, i, 0, -1ull, flush); |
faaf05b0 | 808 | |
faaf05b0 BG |
809 | if (flush) |
810 | kvm_flush_remote_tlbs(kvm); | |
811 | } | |
bb18842e | 812 | |
4c6654bd BG |
813 | static struct kvm_mmu_page *next_invalidated_root(struct kvm *kvm, |
814 | struct kvm_mmu_page *prev_root) | |
815 | { | |
816 | struct kvm_mmu_page *next_root; | |
817 | ||
818 | if (prev_root) | |
819 | next_root = list_next_or_null_rcu(&kvm->arch.tdp_mmu_roots, | |
820 | &prev_root->link, | |
821 | typeof(*prev_root), link); | |
822 | else | |
823 | next_root = list_first_or_null_rcu(&kvm->arch.tdp_mmu_roots, | |
824 | typeof(*next_root), link); | |
825 | ||
826 | while (next_root && !(next_root->role.invalid && | |
827 | refcount_read(&next_root->tdp_mmu_root_count))) | |
828 | next_root = list_next_or_null_rcu(&kvm->arch.tdp_mmu_roots, | |
829 | &next_root->link, | |
830 | typeof(*next_root), link); | |
831 | ||
832 | return next_root; | |
833 | } | |
834 | ||
835 | /* | |
836 | * Since kvm_tdp_mmu_zap_all_fast has acquired a reference to each | |
837 | * invalidated root, they will not be freed until this function drops the | |
838 | * reference. Before dropping that reference, tear down the paging | |
839 | * structure so that whichever thread does drop the last reference | |
840 | * only has to do a trivial amount of work. Since the roots are invalid, | |
841 | * no new SPTEs should be created under them. | |
842 | */ | |
843 | void kvm_tdp_mmu_zap_invalidated_roots(struct kvm *kvm) | |
844 | { | |
4c6654bd BG |
845 | struct kvm_mmu_page *next_root; |
846 | struct kvm_mmu_page *root; | |
847 | bool flush = false; | |
848 | ||
849 | lockdep_assert_held_read(&kvm->mmu_lock); | |
850 | ||
851 | rcu_read_lock(); | |
852 | ||
853 | root = next_invalidated_root(kvm, NULL); | |
854 | ||
855 | while (root) { | |
856 | next_root = next_invalidated_root(kvm, root); | |
857 | ||
858 | rcu_read_unlock(); | |
859 | ||
524a1e4e | 860 | flush = zap_gfn_range(kvm, root, 0, -1ull, true, flush, true); |
4c6654bd BG |
861 | |
862 | /* | |
863 | * Put the reference acquired in | |
864 | * kvm_tdp_mmu_invalidate_roots | |
865 | */ | |
866 | kvm_tdp_mmu_put_root(kvm, root, true); | |
867 | ||
868 | root = next_root; | |
869 | ||
870 | rcu_read_lock(); | |
871 | } | |
872 | ||
873 | rcu_read_unlock(); | |
faaf05b0 | 874 | |
faaf05b0 BG |
875 | if (flush) |
876 | kvm_flush_remote_tlbs(kvm); | |
877 | } | |
bb18842e | 878 | |
b7cccd39 BG |
879 | /* |
880 | * Mark each TDP MMU root as invalid so that other threads | |
881 | * will drop their references and allow the root count to | |
882 | * go to 0. | |
883 | * | |
4c6654bd BG |
884 | * Also take a reference on all roots so that this thread |
885 | * can do the bulk of the work required to free the roots | |
886 | * once they are invalidated. Without this reference, a | |
887 | * vCPU thread might drop the last reference to a root and | |
888 | * get stuck with tearing down the entire paging structure. | |
889 | * | |
890 | * Roots which have a zero refcount should be skipped as | |
891 | * they're already being torn down. | |
892 | * Already invalid roots should be referenced again so that | |
893 | * they aren't freed before kvm_tdp_mmu_zap_all_fast is | |
894 | * done with them. | |
895 | * | |
b7cccd39 BG |
896 | * This has essentially the same effect for the TDP MMU |
897 | * as updating mmu_valid_gen does for the shadow MMU. | |
898 | */ | |
899 | void kvm_tdp_mmu_invalidate_all_roots(struct kvm *kvm) | |
900 | { | |
901 | struct kvm_mmu_page *root; | |
902 | ||
903 | lockdep_assert_held_write(&kvm->mmu_lock); | |
904 | list_for_each_entry(root, &kvm->arch.tdp_mmu_roots, link) | |
4c6654bd BG |
905 | if (refcount_inc_not_zero(&root->tdp_mmu_root_count)) |
906 | root->role.invalid = true; | |
b7cccd39 BG |
907 | } |
908 | ||
bb18842e BG |
909 | /* |
910 | * Installs a last-level SPTE to handle a TDP page fault. | |
911 | * (NPT/EPT violation/misconfiguration) | |
912 | */ | |
cdc47767 PB |
913 | static int tdp_mmu_map_handle_target_level(struct kvm_vcpu *vcpu, |
914 | struct kvm_page_fault *fault, | |
915 | struct tdp_iter *iter) | |
bb18842e | 916 | { |
c435d4b7 | 917 | struct kvm_mmu_page *sp = sptep_to_sp(rcu_dereference(iter->sptep)); |
bb18842e | 918 | u64 new_spte; |
57a3e96d | 919 | int ret = RET_PF_FIXED; |
ad67e480 | 920 | bool wrprot = false; |
bb18842e | 921 | |
7158bee4 | 922 | WARN_ON(sp->role.level != fault->goal_level); |
e710c5f6 | 923 | if (unlikely(!fault->slot)) |
bb18842e | 924 | new_spte = make_mmio_spte(vcpu, iter->gfn, ACC_ALL); |
9a77daac | 925 | else |
53597858 | 926 | wrprot = make_spte(vcpu, sp, fault->slot, ACC_ALL, iter->gfn, |
2839180c | 927 | fault->pfn, iter->old_spte, fault->prefetch, true, |
7158bee4 | 928 | fault->map_writable, &new_spte); |
bb18842e BG |
929 | |
930 | if (new_spte == iter->old_spte) | |
931 | ret = RET_PF_SPURIOUS; | |
3e72c791 | 932 | else if (tdp_mmu_set_spte_atomic(vcpu->kvm, iter, new_spte)) |
9a77daac | 933 | return RET_PF_RETRY; |
bb18842e BG |
934 | |
935 | /* | |
936 | * If the page fault was caused by a write but the page is write | |
937 | * protected, emulation is needed. If the emulation was skipped, | |
938 | * the vCPU would have the same fault again. | |
939 | */ | |
ad67e480 | 940 | if (wrprot) { |
cdc47767 | 941 | if (fault->write) |
bb18842e | 942 | ret = RET_PF_EMULATE; |
bb18842e BG |
943 | } |
944 | ||
945 | /* If a MMIO SPTE is installed, the MMIO will need to be emulated. */ | |
9a77daac BG |
946 | if (unlikely(is_mmio_spte(new_spte))) { |
947 | trace_mark_mmio_spte(rcu_dereference(iter->sptep), iter->gfn, | |
948 | new_spte); | |
bb18842e | 949 | ret = RET_PF_EMULATE; |
3849e092 | 950 | } else { |
9a77daac BG |
951 | trace_kvm_mmu_set_spte(iter->level, iter->gfn, |
952 | rcu_dereference(iter->sptep)); | |
3849e092 | 953 | } |
bb18842e | 954 | |
857f8474 KH |
955 | /* |
956 | * Increase pf_fixed in both RET_PF_EMULATE and RET_PF_FIXED to be | |
957 | * consistent with legacy MMU behavior. | |
958 | */ | |
959 | if (ret != RET_PF_SPURIOUS) | |
bb18842e BG |
960 | vcpu->stat.pf_fixed++; |
961 | ||
962 | return ret; | |
963 | } | |
964 | ||
7b7e1ab6 DM |
965 | /* |
966 | * tdp_mmu_link_sp_atomic - Atomically replace the given spte with an spte | |
967 | * pointing to the provided page table. | |
968 | * | |
969 | * @kvm: kvm instance | |
970 | * @iter: a tdp_iter instance currently on the SPTE that should be set | |
971 | * @sp: The new TDP page table to install. | |
972 | * @account_nx: True if this page table is being installed to split a | |
973 | * non-executable huge page. | |
974 | * | |
975 | * Returns: 0 if the new page table was installed. Non-0 if the page table | |
976 | * could not be installed (e.g. the atomic compare-exchange failed). | |
977 | */ | |
978 | static int tdp_mmu_link_sp_atomic(struct kvm *kvm, struct tdp_iter *iter, | |
979 | struct kvm_mmu_page *sp, bool account_nx) | |
980 | { | |
981 | u64 spte = make_nonleaf_spte(sp->spt, !shadow_accessed_mask); | |
982 | int ret; | |
983 | ||
984 | ret = tdp_mmu_set_spte_atomic(kvm, iter, spte); | |
985 | if (ret) | |
986 | return ret; | |
987 | ||
988 | spin_lock(&kvm->arch.tdp_mmu_pages_lock); | |
989 | list_add(&sp->link, &kvm->arch.tdp_mmu_pages); | |
990 | if (account_nx) | |
991 | account_huge_nx_page(kvm, sp); | |
992 | spin_unlock(&kvm->arch.tdp_mmu_pages_lock); | |
993 | ||
994 | return 0; | |
995 | } | |
996 | ||
bb18842e BG |
997 | /* |
998 | * Handle a TDP page fault (NPT/EPT violation/misconfiguration) by installing | |
999 | * page tables and SPTEs to translate the faulting guest physical address. | |
1000 | */ | |
2f6305dd | 1001 | int kvm_tdp_mmu_map(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault) |
bb18842e | 1002 | { |
bb18842e BG |
1003 | struct kvm_mmu *mmu = vcpu->arch.mmu; |
1004 | struct tdp_iter iter; | |
89c0fd49 | 1005 | struct kvm_mmu_page *sp; |
bb18842e | 1006 | int ret; |
bb18842e | 1007 | |
73a3c659 | 1008 | kvm_mmu_hugepage_adjust(vcpu, fault); |
bb18842e | 1009 | |
f0066d94 | 1010 | trace_kvm_mmu_spte_requested(fault); |
7cca2d0b BG |
1011 | |
1012 | rcu_read_lock(); | |
1013 | ||
2f6305dd | 1014 | tdp_mmu_for_each_pte(iter, mmu, fault->gfn, fault->gfn + 1) { |
73a3c659 | 1015 | if (fault->nx_huge_page_workaround_enabled) |
536f0e6a | 1016 | disallowed_hugepage_adjust(fault, iter.old_spte, iter.level); |
bb18842e | 1017 | |
73a3c659 | 1018 | if (iter.level == fault->goal_level) |
bb18842e BG |
1019 | break; |
1020 | ||
1021 | /* | |
1022 | * If there is an SPTE mapping a large page at a higher level | |
1023 | * than the target, that SPTE must be cleared and replaced | |
1024 | * with a non-leaf SPTE. | |
1025 | */ | |
1026 | if (is_shadow_present_pte(iter.old_spte) && | |
1027 | is_large_pte(iter.old_spte)) { | |
3e72c791 | 1028 | if (tdp_mmu_zap_spte_atomic(vcpu->kvm, &iter)) |
9a77daac | 1029 | break; |
bb18842e | 1030 | |
bb18842e BG |
1031 | /* |
1032 | * The iter must explicitly re-read the spte here | |
1033 | * because the new value informs the !present | |
1034 | * path below. | |
1035 | */ | |
7cca2d0b | 1036 | iter.old_spte = READ_ONCE(*rcu_dereference(iter.sptep)); |
bb18842e BG |
1037 | } |
1038 | ||
1039 | if (!is_shadow_present_pte(iter.old_spte)) { | |
7b7e1ab6 DM |
1040 | bool account_nx = fault->huge_page_disallowed && |
1041 | fault->req_level >= iter.level; | |
1042 | ||
ff76d506 | 1043 | /* |
c4342633 | 1044 | * If SPTE has been frozen by another thread, just |
ff76d506 KH |
1045 | * give up and retry, avoiding unnecessary page table |
1046 | * allocation and free. | |
1047 | */ | |
1048 | if (is_removed_spte(iter.old_spte)) | |
1049 | break; | |
1050 | ||
a82070b6 DM |
1051 | sp = tdp_mmu_alloc_sp(vcpu); |
1052 | tdp_mmu_init_child_sp(sp, &iter); | |
1053 | ||
7b7e1ab6 | 1054 | if (tdp_mmu_link_sp_atomic(vcpu->kvm, &iter, sp, account_nx)) { |
9a77daac BG |
1055 | tdp_mmu_free_sp(sp); |
1056 | break; | |
1057 | } | |
bb18842e BG |
1058 | } |
1059 | } | |
1060 | ||
73a3c659 | 1061 | if (iter.level != fault->goal_level) { |
7cca2d0b | 1062 | rcu_read_unlock(); |
bb18842e | 1063 | return RET_PF_RETRY; |
7cca2d0b | 1064 | } |
bb18842e | 1065 | |
cdc47767 | 1066 | ret = tdp_mmu_map_handle_target_level(vcpu, fault, &iter); |
7cca2d0b | 1067 | rcu_read_unlock(); |
bb18842e BG |
1068 | |
1069 | return ret; | |
1070 | } | |
063afacd | 1071 | |
3039bcc7 SC |
1072 | bool kvm_tdp_mmu_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range, |
1073 | bool flush) | |
063afacd | 1074 | { |
83b83a02 SC |
1075 | return __kvm_tdp_mmu_zap_gfn_range(kvm, range->slot->as_id, range->start, |
1076 | range->end, range->may_block, flush); | |
063afacd BG |
1077 | } |
1078 | ||
3039bcc7 SC |
1079 | typedef bool (*tdp_handler_t)(struct kvm *kvm, struct tdp_iter *iter, |
1080 | struct kvm_gfn_range *range); | |
063afacd | 1081 | |
3039bcc7 SC |
1082 | static __always_inline bool kvm_tdp_mmu_handle_gfn(struct kvm *kvm, |
1083 | struct kvm_gfn_range *range, | |
1084 | tdp_handler_t handler) | |
063afacd | 1085 | { |
3039bcc7 SC |
1086 | struct kvm_mmu_page *root; |
1087 | struct tdp_iter iter; | |
1088 | bool ret = false; | |
1089 | ||
1090 | rcu_read_lock(); | |
1091 | ||
e1eed584 SC |
1092 | /* |
1093 | * Don't support rescheduling, none of the MMU notifiers that funnel | |
1094 | * into this helper allow blocking; it'd be dead, wasteful code. | |
1095 | */ | |
3039bcc7 SC |
1096 | for_each_tdp_mmu_root(kvm, root, range->slot->as_id) { |
1097 | tdp_root_for_each_leaf_pte(iter, root, range->start, range->end) | |
1098 | ret |= handler(kvm, &iter, range); | |
1099 | } | |
1100 | ||
1101 | rcu_read_unlock(); | |
1102 | ||
1103 | return ret; | |
063afacd | 1104 | } |
f8e14497 BG |
1105 | |
1106 | /* | |
1107 | * Mark the SPTEs range of GFNs [start, end) unaccessed and return non-zero | |
1108 | * if any of the GFNs in the range have been accessed. | |
1109 | */ | |
3039bcc7 SC |
1110 | static bool age_gfn_range(struct kvm *kvm, struct tdp_iter *iter, |
1111 | struct kvm_gfn_range *range) | |
f8e14497 | 1112 | { |
f8e14497 BG |
1113 | u64 new_spte = 0; |
1114 | ||
3039bcc7 SC |
1115 | /* If we have a non-accessed entry we don't need to change the pte. */ |
1116 | if (!is_accessed_spte(iter->old_spte)) | |
1117 | return false; | |
7cca2d0b | 1118 | |
3039bcc7 SC |
1119 | new_spte = iter->old_spte; |
1120 | ||
1121 | if (spte_ad_enabled(new_spte)) { | |
1122 | new_spte &= ~shadow_accessed_mask; | |
1123 | } else { | |
f8e14497 | 1124 | /* |
3039bcc7 SC |
1125 | * Capture the dirty status of the page, so that it doesn't get |
1126 | * lost when the SPTE is marked for access tracking. | |
f8e14497 | 1127 | */ |
3039bcc7 SC |
1128 | if (is_writable_pte(new_spte)) |
1129 | kvm_set_pfn_dirty(spte_to_pfn(new_spte)); | |
f8e14497 | 1130 | |
3039bcc7 | 1131 | new_spte = mark_spte_for_access_track(new_spte); |
f8e14497 BG |
1132 | } |
1133 | ||
3039bcc7 | 1134 | tdp_mmu_set_spte_no_acc_track(kvm, iter, new_spte); |
7cca2d0b | 1135 | |
3039bcc7 | 1136 | return true; |
f8e14497 BG |
1137 | } |
1138 | ||
3039bcc7 | 1139 | bool kvm_tdp_mmu_age_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range) |
f8e14497 | 1140 | { |
3039bcc7 | 1141 | return kvm_tdp_mmu_handle_gfn(kvm, range, age_gfn_range); |
f8e14497 BG |
1142 | } |
1143 | ||
3039bcc7 SC |
1144 | static bool test_age_gfn(struct kvm *kvm, struct tdp_iter *iter, |
1145 | struct kvm_gfn_range *range) | |
f8e14497 | 1146 | { |
3039bcc7 | 1147 | return is_accessed_spte(iter->old_spte); |
f8e14497 BG |
1148 | } |
1149 | ||
3039bcc7 | 1150 | bool kvm_tdp_mmu_test_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range) |
f8e14497 | 1151 | { |
3039bcc7 | 1152 | return kvm_tdp_mmu_handle_gfn(kvm, range, test_age_gfn); |
f8e14497 | 1153 | } |
1d8dd6b3 | 1154 | |
3039bcc7 SC |
1155 | static bool set_spte_gfn(struct kvm *kvm, struct tdp_iter *iter, |
1156 | struct kvm_gfn_range *range) | |
1d8dd6b3 | 1157 | { |
1d8dd6b3 | 1158 | u64 new_spte; |
7cca2d0b | 1159 | |
3039bcc7 SC |
1160 | /* Huge pages aren't expected to be modified without first being zapped. */ |
1161 | WARN_ON(pte_huge(range->pte) || range->start + 1 != range->end); | |
1d8dd6b3 | 1162 | |
3039bcc7 SC |
1163 | if (iter->level != PG_LEVEL_4K || |
1164 | !is_shadow_present_pte(iter->old_spte)) | |
1165 | return false; | |
1d8dd6b3 | 1166 | |
3039bcc7 SC |
1167 | /* |
1168 | * Note, when changing a read-only SPTE, it's not strictly necessary to | |
1169 | * zero the SPTE before setting the new PFN, but doing so preserves the | |
1170 | * invariant that the PFN of a present * leaf SPTE can never change. | |
1171 | * See __handle_changed_spte(). | |
1172 | */ | |
1173 | tdp_mmu_set_spte(kvm, iter, 0); | |
1d8dd6b3 | 1174 | |
3039bcc7 SC |
1175 | if (!pte_write(range->pte)) { |
1176 | new_spte = kvm_mmu_changed_pte_notifier_make_spte(iter->old_spte, | |
1177 | pte_pfn(range->pte)); | |
1d8dd6b3 | 1178 | |
3039bcc7 | 1179 | tdp_mmu_set_spte(kvm, iter, new_spte); |
1d8dd6b3 BG |
1180 | } |
1181 | ||
3039bcc7 | 1182 | return true; |
1d8dd6b3 BG |
1183 | } |
1184 | ||
3039bcc7 SC |
1185 | /* |
1186 | * Handle the changed_pte MMU notifier for the TDP MMU. | |
1187 | * data is a pointer to the new pte_t mapping the HVA specified by the MMU | |
1188 | * notifier. | |
1189 | * Returns non-zero if a flush is needed before releasing the MMU lock. | |
1190 | */ | |
1191 | bool kvm_tdp_mmu_set_spte_gfn(struct kvm *kvm, struct kvm_gfn_range *range) | |
1d8dd6b3 | 1192 | { |
3039bcc7 SC |
1193 | bool flush = kvm_tdp_mmu_handle_gfn(kvm, range, set_spte_gfn); |
1194 | ||
1195 | /* FIXME: return 'flush' instead of flushing here. */ | |
1196 | if (flush) | |
1197 | kvm_flush_remote_tlbs_with_address(kvm, range->start, 1); | |
1198 | ||
1199 | return false; | |
1d8dd6b3 BG |
1200 | } |
1201 | ||
a6a0b05d | 1202 | /* |
bedd9195 DM |
1203 | * Remove write access from all SPTEs at or above min_level that map GFNs |
1204 | * [start, end). Returns true if an SPTE has been changed and the TLBs need to | |
1205 | * be flushed. | |
a6a0b05d BG |
1206 | */ |
1207 | static bool wrprot_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root, | |
1208 | gfn_t start, gfn_t end, int min_level) | |
1209 | { | |
1210 | struct tdp_iter iter; | |
1211 | u64 new_spte; | |
1212 | bool spte_set = false; | |
1213 | ||
7cca2d0b BG |
1214 | rcu_read_lock(); |
1215 | ||
a6a0b05d BG |
1216 | BUG_ON(min_level > KVM_MAX_HUGEPAGE_LEVEL); |
1217 | ||
77aa6075 | 1218 | for_each_tdp_pte_min_level(iter, root, min_level, start, end) { |
24ae4cfa BG |
1219 | retry: |
1220 | if (tdp_mmu_iter_cond_resched(kvm, &iter, false, true)) | |
1af4a960 BG |
1221 | continue; |
1222 | ||
a6a0b05d | 1223 | if (!is_shadow_present_pte(iter.old_spte) || |
0f99ee2c BG |
1224 | !is_last_spte(iter.old_spte, iter.level) || |
1225 | !(iter.old_spte & PT_WRITABLE_MASK)) | |
a6a0b05d BG |
1226 | continue; |
1227 | ||
1228 | new_spte = iter.old_spte & ~PT_WRITABLE_MASK; | |
1229 | ||
3e72c791 | 1230 | if (tdp_mmu_set_spte_atomic(kvm, &iter, new_spte)) |
24ae4cfa | 1231 | goto retry; |
3255530a | 1232 | |
a6a0b05d | 1233 | spte_set = true; |
a6a0b05d | 1234 | } |
7cca2d0b BG |
1235 | |
1236 | rcu_read_unlock(); | |
a6a0b05d BG |
1237 | return spte_set; |
1238 | } | |
1239 | ||
1240 | /* | |
1241 | * Remove write access from all the SPTEs mapping GFNs in the memslot. Will | |
1242 | * only affect leaf SPTEs down to min_level. | |
1243 | * Returns true if an SPTE has been changed and the TLBs need to be flushed. | |
1244 | */ | |
269e9552 HM |
1245 | bool kvm_tdp_mmu_wrprot_slot(struct kvm *kvm, |
1246 | const struct kvm_memory_slot *slot, int min_level) | |
a6a0b05d BG |
1247 | { |
1248 | struct kvm_mmu_page *root; | |
a6a0b05d BG |
1249 | bool spte_set = false; |
1250 | ||
24ae4cfa | 1251 | lockdep_assert_held_read(&kvm->mmu_lock); |
a6a0b05d | 1252 | |
d62007ed | 1253 | for_each_valid_tdp_mmu_root_yield_safe(kvm, root, slot->as_id, true) |
a6a0b05d BG |
1254 | spte_set |= wrprot_gfn_range(kvm, root, slot->base_gfn, |
1255 | slot->base_gfn + slot->npages, min_level); | |
a6a0b05d BG |
1256 | |
1257 | return spte_set; | |
1258 | } | |
1259 | ||
1260 | /* | |
1261 | * Clear the dirty status of all the SPTEs mapping GFNs in the memslot. If | |
1262 | * AD bits are enabled, this will involve clearing the dirty bit on each SPTE. | |
1263 | * If AD bits are not enabled, this will require clearing the writable bit on | |
1264 | * each SPTE. Returns true if an SPTE has been changed and the TLBs need to | |
1265 | * be flushed. | |
1266 | */ | |
1267 | static bool clear_dirty_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root, | |
1268 | gfn_t start, gfn_t end) | |
1269 | { | |
1270 | struct tdp_iter iter; | |
1271 | u64 new_spte; | |
1272 | bool spte_set = false; | |
1273 | ||
7cca2d0b BG |
1274 | rcu_read_lock(); |
1275 | ||
a6a0b05d | 1276 | tdp_root_for_each_leaf_pte(iter, root, start, end) { |
24ae4cfa BG |
1277 | retry: |
1278 | if (tdp_mmu_iter_cond_resched(kvm, &iter, false, true)) | |
1af4a960 BG |
1279 | continue; |
1280 | ||
a6a0b05d BG |
1281 | if (spte_ad_need_write_protect(iter.old_spte)) { |
1282 | if (is_writable_pte(iter.old_spte)) | |
1283 | new_spte = iter.old_spte & ~PT_WRITABLE_MASK; | |
1284 | else | |
1285 | continue; | |
1286 | } else { | |
1287 | if (iter.old_spte & shadow_dirty_mask) | |
1288 | new_spte = iter.old_spte & ~shadow_dirty_mask; | |
1289 | else | |
1290 | continue; | |
1291 | } | |
1292 | ||
3e72c791 | 1293 | if (tdp_mmu_set_spte_atomic(kvm, &iter, new_spte)) |
24ae4cfa | 1294 | goto retry; |
3255530a | 1295 | |
a6a0b05d | 1296 | spte_set = true; |
a6a0b05d | 1297 | } |
7cca2d0b BG |
1298 | |
1299 | rcu_read_unlock(); | |
a6a0b05d BG |
1300 | return spte_set; |
1301 | } | |
1302 | ||
1303 | /* | |
1304 | * Clear the dirty status of all the SPTEs mapping GFNs in the memslot. If | |
1305 | * AD bits are enabled, this will involve clearing the dirty bit on each SPTE. | |
1306 | * If AD bits are not enabled, this will require clearing the writable bit on | |
1307 | * each SPTE. Returns true if an SPTE has been changed and the TLBs need to | |
1308 | * be flushed. | |
1309 | */ | |
269e9552 HM |
1310 | bool kvm_tdp_mmu_clear_dirty_slot(struct kvm *kvm, |
1311 | const struct kvm_memory_slot *slot) | |
a6a0b05d BG |
1312 | { |
1313 | struct kvm_mmu_page *root; | |
a6a0b05d BG |
1314 | bool spte_set = false; |
1315 | ||
24ae4cfa | 1316 | lockdep_assert_held_read(&kvm->mmu_lock); |
a6a0b05d | 1317 | |
d62007ed | 1318 | for_each_valid_tdp_mmu_root_yield_safe(kvm, root, slot->as_id, true) |
a6a0b05d BG |
1319 | spte_set |= clear_dirty_gfn_range(kvm, root, slot->base_gfn, |
1320 | slot->base_gfn + slot->npages); | |
a6a0b05d BG |
1321 | |
1322 | return spte_set; | |
1323 | } | |
1324 | ||
1325 | /* | |
1326 | * Clears the dirty status of all the 4k SPTEs mapping GFNs for which a bit is | |
1327 | * set in mask, starting at gfn. The given memslot is expected to contain all | |
1328 | * the GFNs represented by set bits in the mask. If AD bits are enabled, | |
1329 | * clearing the dirty status will involve clearing the dirty bit on each SPTE | |
1330 | * or, if AD bits are not enabled, clearing the writable bit on each SPTE. | |
1331 | */ | |
1332 | static void clear_dirty_pt_masked(struct kvm *kvm, struct kvm_mmu_page *root, | |
1333 | gfn_t gfn, unsigned long mask, bool wrprot) | |
1334 | { | |
1335 | struct tdp_iter iter; | |
1336 | u64 new_spte; | |
1337 | ||
7cca2d0b BG |
1338 | rcu_read_lock(); |
1339 | ||
a6a0b05d BG |
1340 | tdp_root_for_each_leaf_pte(iter, root, gfn + __ffs(mask), |
1341 | gfn + BITS_PER_LONG) { | |
1342 | if (!mask) | |
1343 | break; | |
1344 | ||
1345 | if (iter.level > PG_LEVEL_4K || | |
1346 | !(mask & (1UL << (iter.gfn - gfn)))) | |
1347 | continue; | |
1348 | ||
f1b3b06a BG |
1349 | mask &= ~(1UL << (iter.gfn - gfn)); |
1350 | ||
a6a0b05d BG |
1351 | if (wrprot || spte_ad_need_write_protect(iter.old_spte)) { |
1352 | if (is_writable_pte(iter.old_spte)) | |
1353 | new_spte = iter.old_spte & ~PT_WRITABLE_MASK; | |
1354 | else | |
1355 | continue; | |
1356 | } else { | |
1357 | if (iter.old_spte & shadow_dirty_mask) | |
1358 | new_spte = iter.old_spte & ~shadow_dirty_mask; | |
1359 | else | |
1360 | continue; | |
1361 | } | |
1362 | ||
1363 | tdp_mmu_set_spte_no_dirty_log(kvm, &iter, new_spte); | |
a6a0b05d | 1364 | } |
7cca2d0b BG |
1365 | |
1366 | rcu_read_unlock(); | |
a6a0b05d BG |
1367 | } |
1368 | ||
1369 | /* | |
1370 | * Clears the dirty status of all the 4k SPTEs mapping GFNs for which a bit is | |
1371 | * set in mask, starting at gfn. The given memslot is expected to contain all | |
1372 | * the GFNs represented by set bits in the mask. If AD bits are enabled, | |
1373 | * clearing the dirty status will involve clearing the dirty bit on each SPTE | |
1374 | * or, if AD bits are not enabled, clearing the writable bit on each SPTE. | |
1375 | */ | |
1376 | void kvm_tdp_mmu_clear_dirty_pt_masked(struct kvm *kvm, | |
1377 | struct kvm_memory_slot *slot, | |
1378 | gfn_t gfn, unsigned long mask, | |
1379 | bool wrprot) | |
1380 | { | |
1381 | struct kvm_mmu_page *root; | |
a6a0b05d | 1382 | |
531810ca | 1383 | lockdep_assert_held_write(&kvm->mmu_lock); |
a3f15bda | 1384 | for_each_tdp_mmu_root(kvm, root, slot->as_id) |
a6a0b05d | 1385 | clear_dirty_pt_masked(kvm, root, gfn, mask, wrprot); |
a6a0b05d BG |
1386 | } |
1387 | ||
14881998 | 1388 | /* |
87aa9ec9 BG |
1389 | * Clear leaf entries which could be replaced by large mappings, for |
1390 | * GFNs within the slot. | |
14881998 | 1391 | */ |
4b85c921 | 1392 | static void zap_collapsible_spte_range(struct kvm *kvm, |
14881998 | 1393 | struct kvm_mmu_page *root, |
4b85c921 | 1394 | const struct kvm_memory_slot *slot) |
14881998 | 1395 | { |
9eba50f8 SC |
1396 | gfn_t start = slot->base_gfn; |
1397 | gfn_t end = start + slot->npages; | |
14881998 BG |
1398 | struct tdp_iter iter; |
1399 | kvm_pfn_t pfn; | |
14881998 | 1400 | |
7cca2d0b BG |
1401 | rcu_read_lock(); |
1402 | ||
14881998 | 1403 | tdp_root_for_each_pte(iter, root, start, end) { |
2db6f772 | 1404 | retry: |
4b85c921 | 1405 | if (tdp_mmu_iter_cond_resched(kvm, &iter, false, true)) |
1af4a960 | 1406 | continue; |
1af4a960 | 1407 | |
14881998 | 1408 | if (!is_shadow_present_pte(iter.old_spte) || |
87aa9ec9 | 1409 | !is_last_spte(iter.old_spte, iter.level)) |
14881998 BG |
1410 | continue; |
1411 | ||
1412 | pfn = spte_to_pfn(iter.old_spte); | |
1413 | if (kvm_is_reserved_pfn(pfn) || | |
9eba50f8 SC |
1414 | iter.level >= kvm_mmu_max_mapping_level(kvm, slot, iter.gfn, |
1415 | pfn, PG_LEVEL_NUM)) | |
14881998 BG |
1416 | continue; |
1417 | ||
4b85c921 | 1418 | /* Note, a successful atomic zap also does a remote TLB flush. */ |
3e72c791 | 1419 | if (tdp_mmu_zap_spte_atomic(kvm, &iter)) |
2db6f772 | 1420 | goto retry; |
14881998 BG |
1421 | } |
1422 | ||
7cca2d0b | 1423 | rcu_read_unlock(); |
14881998 BG |
1424 | } |
1425 | ||
1426 | /* | |
1427 | * Clear non-leaf entries (and free associated page tables) which could | |
1428 | * be replaced by large mappings, for GFNs within the slot. | |
1429 | */ | |
4b85c921 SC |
1430 | void kvm_tdp_mmu_zap_collapsible_sptes(struct kvm *kvm, |
1431 | const struct kvm_memory_slot *slot) | |
14881998 BG |
1432 | { |
1433 | struct kvm_mmu_page *root; | |
14881998 | 1434 | |
2db6f772 | 1435 | lockdep_assert_held_read(&kvm->mmu_lock); |
14881998 | 1436 | |
d62007ed | 1437 | for_each_valid_tdp_mmu_root_yield_safe(kvm, root, slot->as_id, true) |
4b85c921 | 1438 | zap_collapsible_spte_range(kvm, root, slot); |
14881998 | 1439 | } |
46044f72 BG |
1440 | |
1441 | /* | |
1442 | * Removes write access on the last level SPTE mapping this GFN and unsets the | |
5fc3424f | 1443 | * MMU-writable bit to ensure future writes continue to be intercepted. |
46044f72 BG |
1444 | * Returns true if an SPTE was set and a TLB flush is needed. |
1445 | */ | |
1446 | static bool write_protect_gfn(struct kvm *kvm, struct kvm_mmu_page *root, | |
3ad93562 | 1447 | gfn_t gfn, int min_level) |
46044f72 BG |
1448 | { |
1449 | struct tdp_iter iter; | |
1450 | u64 new_spte; | |
1451 | bool spte_set = false; | |
1452 | ||
3ad93562 KZ |
1453 | BUG_ON(min_level > KVM_MAX_HUGEPAGE_LEVEL); |
1454 | ||
7cca2d0b BG |
1455 | rcu_read_lock(); |
1456 | ||
77aa6075 | 1457 | for_each_tdp_pte_min_level(iter, root, min_level, gfn, gfn + 1) { |
3ad93562 KZ |
1458 | if (!is_shadow_present_pte(iter.old_spte) || |
1459 | !is_last_spte(iter.old_spte, iter.level)) | |
1460 | continue; | |
1461 | ||
46044f72 | 1462 | new_spte = iter.old_spte & |
5fc3424f | 1463 | ~(PT_WRITABLE_MASK | shadow_mmu_writable_mask); |
46044f72 | 1464 | |
7c8a4742 DM |
1465 | if (new_spte == iter.old_spte) |
1466 | break; | |
1467 | ||
46044f72 BG |
1468 | tdp_mmu_set_spte(kvm, &iter, new_spte); |
1469 | spte_set = true; | |
1470 | } | |
1471 | ||
7cca2d0b BG |
1472 | rcu_read_unlock(); |
1473 | ||
46044f72 BG |
1474 | return spte_set; |
1475 | } | |
1476 | ||
1477 | /* | |
1478 | * Removes write access on the last level SPTE mapping this GFN and unsets the | |
5fc3424f | 1479 | * MMU-writable bit to ensure future writes continue to be intercepted. |
46044f72 BG |
1480 | * Returns true if an SPTE was set and a TLB flush is needed. |
1481 | */ | |
1482 | bool kvm_tdp_mmu_write_protect_gfn(struct kvm *kvm, | |
3ad93562 KZ |
1483 | struct kvm_memory_slot *slot, gfn_t gfn, |
1484 | int min_level) | |
46044f72 BG |
1485 | { |
1486 | struct kvm_mmu_page *root; | |
46044f72 BG |
1487 | bool spte_set = false; |
1488 | ||
531810ca | 1489 | lockdep_assert_held_write(&kvm->mmu_lock); |
a3f15bda | 1490 | for_each_tdp_mmu_root(kvm, root, slot->as_id) |
3ad93562 | 1491 | spte_set |= write_protect_gfn(kvm, root, gfn, min_level); |
a3f15bda | 1492 | |
46044f72 BG |
1493 | return spte_set; |
1494 | } | |
1495 | ||
95fb5b02 BG |
1496 | /* |
1497 | * Return the level of the lowest level SPTE added to sptes. | |
1498 | * That SPTE may be non-present. | |
c5c8c7c5 DM |
1499 | * |
1500 | * Must be called between kvm_tdp_mmu_walk_lockless_{begin,end}. | |
95fb5b02 | 1501 | */ |
39b4d43e SC |
1502 | int kvm_tdp_mmu_get_walk(struct kvm_vcpu *vcpu, u64 addr, u64 *sptes, |
1503 | int *root_level) | |
95fb5b02 BG |
1504 | { |
1505 | struct tdp_iter iter; | |
1506 | struct kvm_mmu *mmu = vcpu->arch.mmu; | |
95fb5b02 | 1507 | gfn_t gfn = addr >> PAGE_SHIFT; |
2aa07893 | 1508 | int leaf = -1; |
95fb5b02 | 1509 | |
39b4d43e | 1510 | *root_level = vcpu->arch.mmu->shadow_root_level; |
95fb5b02 BG |
1511 | |
1512 | tdp_mmu_for_each_pte(iter, mmu, gfn, gfn + 1) { | |
1513 | leaf = iter.level; | |
dde81f94 | 1514 | sptes[leaf] = iter.old_spte; |
95fb5b02 BG |
1515 | } |
1516 | ||
1517 | return leaf; | |
1518 | } | |
6e8eb206 DM |
1519 | |
1520 | /* | |
1521 | * Returns the last level spte pointer of the shadow page walk for the given | |
1522 | * gpa, and sets *spte to the spte value. This spte may be non-preset. If no | |
1523 | * walk could be performed, returns NULL and *spte does not contain valid data. | |
1524 | * | |
1525 | * Contract: | |
1526 | * - Must be called between kvm_tdp_mmu_walk_lockless_{begin,end}. | |
1527 | * - The returned sptep must not be used after kvm_tdp_mmu_walk_lockless_end. | |
1528 | * | |
1529 | * WARNING: This function is only intended to be called during fast_page_fault. | |
1530 | */ | |
1531 | u64 *kvm_tdp_mmu_fast_pf_get_last_sptep(struct kvm_vcpu *vcpu, u64 addr, | |
1532 | u64 *spte) | |
1533 | { | |
1534 | struct tdp_iter iter; | |
1535 | struct kvm_mmu *mmu = vcpu->arch.mmu; | |
1536 | gfn_t gfn = addr >> PAGE_SHIFT; | |
1537 | tdp_ptep_t sptep = NULL; | |
1538 | ||
1539 | tdp_mmu_for_each_pte(iter, mmu, gfn, gfn + 1) { | |
1540 | *spte = iter.old_spte; | |
1541 | sptep = iter.sptep; | |
1542 | } | |
1543 | ||
1544 | /* | |
1545 | * Perform the rcu_dereference to get the raw spte pointer value since | |
1546 | * we are passing it up to fast_page_fault, which is shared with the | |
1547 | * legacy MMU and thus does not retain the TDP MMU-specific __rcu | |
1548 | * annotation. | |
1549 | * | |
1550 | * This is safe since fast_page_fault obeys the contracts of this | |
1551 | * function as well as all TDP MMU contracts around modifying SPTEs | |
1552 | * outside of mmu_lock. | |
1553 | */ | |
1554 | return rcu_dereference(sptep); | |
1555 | } |