1 // SPDX-License-Identifier: GPL-2.0
4 #include "mmu_internal.h"
10 #include <asm/cmpxchg.h>
11 #include <trace/events/kvm.h>
13 static bool __read_mostly tdp_mmu_enabled = true;
14 module_param_named(tdp_mmu, tdp_mmu_enabled, bool, 0644);
16 /* Initializes the TDP MMU for the VM, if enabled. */
17 bool kvm_mmu_init_tdp_mmu(struct kvm *kvm)
19 if (!tdp_enabled || !READ_ONCE(tdp_mmu_enabled))
22 /* This should not be changed for the lifetime of the VM. */
23 kvm->arch.tdp_mmu_enabled = true;
25 INIT_LIST_HEAD(&kvm->arch.tdp_mmu_roots);
26 spin_lock_init(&kvm->arch.tdp_mmu_pages_lock);
27 INIT_LIST_HEAD(&kvm->arch.tdp_mmu_pages);
32 static __always_inline void kvm_lockdep_assert_mmu_lock_held(struct kvm *kvm,
36 lockdep_assert_held_read(&kvm->mmu_lock);
38 lockdep_assert_held_write(&kvm->mmu_lock);
41 void kvm_mmu_uninit_tdp_mmu(struct kvm *kvm)
43 if (!kvm->arch.tdp_mmu_enabled)
46 WARN_ON(!list_empty(&kvm->arch.tdp_mmu_pages));
47 WARN_ON(!list_empty(&kvm->arch.tdp_mmu_roots));
50 * Ensure that all the outstanding RCU callbacks to free shadow pages
51 * can run before the VM is torn down.
56 static bool zap_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root,
57 gfn_t start, gfn_t end, bool can_yield, bool flush,
60 static void tdp_mmu_free_sp(struct kvm_mmu_page *sp)
62 free_page((unsigned long)sp->spt);
63 kmem_cache_free(mmu_page_header_cache, sp);
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
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.
74 static void tdp_mmu_free_sp_rcu_callback(struct rcu_head *head)
76 struct kvm_mmu_page *sp = container_of(head, struct kvm_mmu_page,
82 void kvm_tdp_mmu_put_root(struct kvm *kvm, struct kvm_mmu_page *root,
85 kvm_lockdep_assert_mmu_lock_held(kvm, shared);
87 if (!refcount_dec_and_test(&root->tdp_mmu_root_count))
90 WARN_ON(!root->tdp_mmu_page);
92 spin_lock(&kvm->arch.tdp_mmu_pages_lock);
93 list_del_rcu(&root->link);
94 spin_unlock(&kvm->arch.tdp_mmu_pages_lock);
96 zap_gfn_range(kvm, root, 0, -1ull, false, false, shared);
98 call_rcu(&root->rcu_head, tdp_mmu_free_sp_rcu_callback);
102 * Finds the next valid root after root (or the first valid root if root
103 * is NULL), takes a reference on it, and returns that next root. If root
104 * is not NULL, this thread should have already taken a reference on it, and
105 * that reference will be dropped. If no valid root is found, this
106 * function will return NULL.
108 static struct kvm_mmu_page *tdp_mmu_next_root(struct kvm *kvm,
109 struct kvm_mmu_page *prev_root,
112 struct kvm_mmu_page *next_root;
117 next_root = list_next_or_null_rcu(&kvm->arch.tdp_mmu_roots,
119 typeof(*prev_root), link);
121 next_root = list_first_or_null_rcu(&kvm->arch.tdp_mmu_roots,
122 typeof(*next_root), link);
124 while (next_root && !kvm_tdp_mmu_get_root(kvm, next_root))
125 next_root = list_next_or_null_rcu(&kvm->arch.tdp_mmu_roots,
126 &next_root->link, typeof(*next_root), link);
131 kvm_tdp_mmu_put_root(kvm, prev_root, shared);
137 * Note: this iterator gets and puts references to the roots it iterates over.
138 * This makes it safe to release the MMU lock and yield within the loop, but
139 * if exiting the loop early, the caller must drop the reference to the most
140 * recent root. (Unless keeping a live reference is desirable.)
142 * If shared is set, this function is operating under the MMU lock in read
143 * mode. In the unlikely event that this thread must free a root, the lock
144 * will be temporarily dropped and reacquired in write mode.
146 #define for_each_tdp_mmu_root_yield_safe(_kvm, _root, _as_id, _shared) \
147 for (_root = tdp_mmu_next_root(_kvm, NULL, _shared); \
149 _root = tdp_mmu_next_root(_kvm, _root, _shared)) \
150 if (kvm_mmu_page_as_id(_root) != _as_id) { \
153 #define for_each_tdp_mmu_root(_kvm, _root, _as_id) \
154 list_for_each_entry_rcu(_root, &_kvm->arch.tdp_mmu_roots, link, \
155 lockdep_is_held_type(&kvm->mmu_lock, 0) || \
156 lockdep_is_held(&kvm->arch.tdp_mmu_pages_lock)) \
157 if (kvm_mmu_page_as_id(_root) != _as_id) { \
160 static union kvm_mmu_page_role page_role_for_level(struct kvm_vcpu *vcpu,
163 union kvm_mmu_page_role role;
165 role = vcpu->arch.mmu->mmu_role.base;
168 role.gpte_is_8_bytes = true;
169 role.access = ACC_ALL;
174 static struct kvm_mmu_page *alloc_tdp_mmu_page(struct kvm_vcpu *vcpu, gfn_t gfn,
177 struct kvm_mmu_page *sp;
179 sp = kvm_mmu_memory_cache_alloc(&vcpu->arch.mmu_page_header_cache);
180 sp->spt = kvm_mmu_memory_cache_alloc(&vcpu->arch.mmu_shadow_page_cache);
181 set_page_private(virt_to_page(sp->spt), (unsigned long)sp);
183 sp->role.word = page_role_for_level(vcpu, level).word;
185 sp->tdp_mmu_page = true;
187 trace_kvm_mmu_get_page(sp, true);
192 hpa_t kvm_tdp_mmu_get_vcpu_root_hpa(struct kvm_vcpu *vcpu)
194 union kvm_mmu_page_role role;
195 struct kvm *kvm = vcpu->kvm;
196 struct kvm_mmu_page *root;
198 lockdep_assert_held_write(&kvm->mmu_lock);
200 role = page_role_for_level(vcpu, vcpu->arch.mmu->shadow_root_level);
202 /* Check for an existing root before allocating a new one. */
203 for_each_tdp_mmu_root(kvm, root, kvm_mmu_role_as_id(role)) {
204 if (root->role.word == role.word &&
205 kvm_tdp_mmu_get_root(kvm, root))
209 root = alloc_tdp_mmu_page(vcpu, 0, vcpu->arch.mmu->shadow_root_level);
210 refcount_set(&root->tdp_mmu_root_count, 1);
212 spin_lock(&kvm->arch.tdp_mmu_pages_lock);
213 list_add_rcu(&root->link, &kvm->arch.tdp_mmu_roots);
214 spin_unlock(&kvm->arch.tdp_mmu_pages_lock);
217 return __pa(root->spt);
220 static void handle_changed_spte(struct kvm *kvm, int as_id, gfn_t gfn,
221 u64 old_spte, u64 new_spte, int level,
224 static void handle_changed_spte_acc_track(u64 old_spte, u64 new_spte, int level)
226 if (!is_shadow_present_pte(old_spte) || !is_last_spte(old_spte, level))
229 if (is_accessed_spte(old_spte) &&
230 (!is_shadow_present_pte(new_spte) || !is_accessed_spte(new_spte) ||
231 spte_to_pfn(old_spte) != spte_to_pfn(new_spte)))
232 kvm_set_pfn_accessed(spte_to_pfn(old_spte));
235 static void handle_changed_spte_dirty_log(struct kvm *kvm, int as_id, gfn_t gfn,
236 u64 old_spte, u64 new_spte, int level)
239 struct kvm_memory_slot *slot;
241 if (level > PG_LEVEL_4K)
244 pfn_changed = spte_to_pfn(old_spte) != spte_to_pfn(new_spte);
246 if ((!is_writable_pte(old_spte) || pfn_changed) &&
247 is_writable_pte(new_spte)) {
248 slot = __gfn_to_memslot(__kvm_memslots(kvm, as_id), gfn);
249 mark_page_dirty_in_slot(kvm, slot, gfn);
254 * tdp_mmu_link_page - Add a new page to the list of pages used by the TDP MMU
258 * @account_nx: This page replaces a NX large page and should be marked for
261 static void tdp_mmu_link_page(struct kvm *kvm, struct kvm_mmu_page *sp,
264 spin_lock(&kvm->arch.tdp_mmu_pages_lock);
265 list_add(&sp->link, &kvm->arch.tdp_mmu_pages);
267 account_huge_nx_page(kvm, sp);
268 spin_unlock(&kvm->arch.tdp_mmu_pages_lock);
272 * tdp_mmu_unlink_page - Remove page from the list of pages used by the TDP MMU
275 * @sp: the page to be removed
276 * @shared: This operation may not be running under the exclusive use of
277 * the MMU lock and the operation must synchronize with other
278 * threads that might be adding or removing pages.
280 static void tdp_mmu_unlink_page(struct kvm *kvm, struct kvm_mmu_page *sp,
284 spin_lock(&kvm->arch.tdp_mmu_pages_lock);
286 lockdep_assert_held_write(&kvm->mmu_lock);
289 if (sp->lpage_disallowed)
290 unaccount_huge_nx_page(kvm, sp);
293 spin_unlock(&kvm->arch.tdp_mmu_pages_lock);
297 * handle_removed_tdp_mmu_page - handle a pt removed from the TDP structure
300 * @pt: the page removed from the paging structure
301 * @shared: This operation may not be running under the exclusive use
302 * of the MMU lock and the operation must synchronize with other
303 * threads that might be modifying SPTEs.
305 * Given a page table that has been removed from the TDP paging structure,
306 * iterates through the page table to clear SPTEs and free child page tables.
308 * Note that pt is passed in as a tdp_ptep_t, but it does not need RCU
309 * protection. Since this thread removed it from the paging structure,
310 * this thread will be responsible for ensuring the page is freed. Hence the
311 * early rcu_dereferences in the function.
313 static void handle_removed_tdp_mmu_page(struct kvm *kvm, tdp_ptep_t pt,
316 struct kvm_mmu_page *sp = sptep_to_sp(rcu_dereference(pt));
317 int level = sp->role.level;
318 gfn_t base_gfn = sp->gfn;
324 trace_kvm_mmu_prepare_zap_page(sp);
326 tdp_mmu_unlink_page(kvm, sp, shared);
328 for (i = 0; i < PT64_ENT_PER_PAGE; i++) {
329 sptep = rcu_dereference(pt) + i;
330 gfn = base_gfn + i * KVM_PAGES_PER_HPAGE(level);
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.
342 old_child_spte = xchg(sptep, REMOVED_SPTE);
343 if (!is_removed_spte(old_child_spte))
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
357 old_child_spte = READ_ONCE(*sptep);
358 if (!is_shadow_present_pte(old_child_spte))
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
369 WRITE_ONCE(*sptep, REMOVED_SPTE);
371 handle_changed_spte(kvm, kvm_mmu_page_as_id(sp), gfn,
372 old_child_spte, REMOVED_SPTE, level,
376 kvm_flush_remote_tlbs_with_address(kvm, gfn,
377 KVM_PAGES_PER_HPAGE(level + 1));
379 call_rcu(&sp->rcu_head, tdp_mmu_free_sp_rcu_callback);
383 * __handle_changed_spte - handle bookkeeping associated with an SPTE change
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
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.
394 * Handle bookkeeping that might result from the modification of a SPTE.
395 * This function must be called for all TDP SPTE modifications.
397 static void __handle_changed_spte(struct kvm *kvm, int as_id, gfn_t gfn,
398 u64 old_spte, u64 new_spte, int level,
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);
407 WARN_ON(level > PT64_ROOT_MAX_LEVEL);
408 WARN_ON(level < PG_LEVEL_4K);
409 WARN_ON(gfn & (KVM_PAGES_PER_HPAGE(level) - 1));
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
415 * present leaf SPTE pointing to a different PFN. A notifier handler
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.
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"
424 "as_id: %d gfn: %llx old_spte: %llx new_spte: %llx level: %d",
425 as_id, gfn, old_spte, new_spte, level);
428 * Crash the host to prevent error propagation and guest data
434 if (old_spte == new_spte)
437 trace_kvm_tdp_mmu_spte_changed(as_id, gfn, level, old_spte, new_spte);
440 * The only times a SPTE should be changed from a non-present to
441 * non-present state is when an MMIO entry is installed/modified/
442 * removed. In that case, there is nothing to do here.
444 if (!was_present && !is_present) {
446 * If this change does not involve a MMIO SPTE or removed SPTE,
447 * it is unexpected. Log the change, though it should not
448 * impact the guest since both the former and current SPTEs
451 if (WARN_ON(!is_mmio_spte(old_spte) &&
452 !is_mmio_spte(new_spte) &&
453 !is_removed_spte(new_spte)))
454 pr_err("Unexpected SPTE change! Nonpresent SPTEs\n"
455 "should not be replaced with another,\n"
456 "different nonpresent SPTE, unless one or both\n"
457 "are MMIO SPTEs, or the new SPTE is\n"
458 "a temporary removed SPTE.\n"
459 "as_id: %d gfn: %llx old_spte: %llx new_spte: %llx level: %d",
460 as_id, gfn, old_spte, new_spte, level);
464 if (is_leaf != was_leaf)
465 kvm_update_page_stats(kvm, level, is_leaf ? 1 : -1);
467 if (was_leaf && is_dirty_spte(old_spte) &&
468 (!is_present || !is_dirty_spte(new_spte) || pfn_changed))
469 kvm_set_pfn_dirty(spte_to_pfn(old_spte));
472 * Recursively handle child PTs if the change removed a subtree from
473 * the paging structure.
475 if (was_present && !was_leaf && (pfn_changed || !is_present))
476 handle_removed_tdp_mmu_page(kvm,
477 spte_to_child_pt(old_spte, level), shared);
480 static void handle_changed_spte(struct kvm *kvm, int as_id, gfn_t gfn,
481 u64 old_spte, u64 new_spte, int level,
484 __handle_changed_spte(kvm, as_id, gfn, old_spte, new_spte, level,
486 handle_changed_spte_acc_track(old_spte, new_spte, level);
487 handle_changed_spte_dirty_log(kvm, as_id, gfn, old_spte,
492 * tdp_mmu_set_spte_atomic_no_dirty_log - Set a TDP MMU SPTE atomically
493 * and handle the associated bookkeeping, but do not mark the page dirty
494 * in KVM's dirty bitmaps.
497 * @iter: a tdp_iter instance currently on the SPTE that should be set
498 * @new_spte: The value the SPTE should be set to
499 * Returns: true if the SPTE was set, false if it was not. If false is returned,
500 * this function will have no side-effects.
502 static inline bool tdp_mmu_set_spte_atomic_no_dirty_log(struct kvm *kvm,
503 struct tdp_iter *iter,
506 lockdep_assert_held_read(&kvm->mmu_lock);
509 * Do not change removed SPTEs. Only the thread that froze the SPTE
512 if (is_removed_spte(iter->old_spte))
516 * Note, fast_pf_fix_direct_spte() can also modify TDP MMU SPTEs and
517 * does not hold the mmu_lock.
519 if (cmpxchg64(rcu_dereference(iter->sptep), iter->old_spte,
520 new_spte) != iter->old_spte)
523 __handle_changed_spte(kvm, iter->as_id, iter->gfn, iter->old_spte,
524 new_spte, iter->level, true);
525 handle_changed_spte_acc_track(iter->old_spte, new_spte, iter->level);
531 * tdp_mmu_map_set_spte_atomic - Set a leaf TDP MMU SPTE atomically to resolve a
534 * @vcpu: The vcpu instance that took the TDP page fault.
535 * @iter: a tdp_iter instance currently on the SPTE that should be set
536 * @new_spte: The value the SPTE should be set to
538 * Returns: true if the SPTE was set, false if it was not. If false is returned,
539 * this function will have no side-effects.
541 static inline bool tdp_mmu_map_set_spte_atomic(struct kvm_vcpu *vcpu,
542 struct tdp_iter *iter,
545 struct kvm *kvm = vcpu->kvm;
547 if (!tdp_mmu_set_spte_atomic_no_dirty_log(kvm, iter, new_spte))
551 * Use kvm_vcpu_gfn_to_memslot() instead of going through
552 * handle_changed_spte_dirty_log() to leverage vcpu->last_used_slot.
554 if (is_writable_pte(new_spte)) {
555 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, iter->gfn);
557 if (slot && kvm_slot_dirty_track_enabled(slot)) {
558 /* Enforced by kvm_mmu_hugepage_adjust. */
559 WARN_ON_ONCE(iter->level > PG_LEVEL_4K);
560 mark_page_dirty_in_slot(kvm, slot, iter->gfn);
567 static inline bool tdp_mmu_zap_spte_atomic(struct kvm *kvm,
568 struct tdp_iter *iter)
571 * Freeze the SPTE by setting it to a special,
572 * non-present value. This will stop other threads from
573 * immediately installing a present entry in its place
574 * before the TLBs are flushed.
576 if (!tdp_mmu_set_spte_atomic_no_dirty_log(kvm, iter, REMOVED_SPTE))
579 kvm_flush_remote_tlbs_with_address(kvm, iter->gfn,
580 KVM_PAGES_PER_HPAGE(iter->level));
583 * No other thread can overwrite the removed SPTE as they
584 * must either wait on the MMU lock or use
585 * tdp_mmu_set_spte_atomic which will not overwrite the
586 * special removed SPTE value. No bookkeeping is needed
587 * here since the SPTE is going from non-present
590 WRITE_ONCE(*rcu_dereference(iter->sptep), 0);
597 * __tdp_mmu_set_spte - Set a TDP MMU SPTE and handle the associated bookkeeping
599 * @iter: a tdp_iter instance currently on the SPTE that should be set
600 * @new_spte: The value the SPTE should be set to
601 * @record_acc_track: Notify the MM subsystem of changes to the accessed state
602 * of the page. Should be set unless handling an MMU
603 * notifier for access tracking. Leaving record_acc_track
604 * unset in that case prevents page accesses from being
606 * @record_dirty_log: Record the page as dirty in the dirty bitmap if
607 * appropriate for the change being made. Should be set
608 * unless performing certain dirty logging operations.
609 * Leaving record_dirty_log unset in that case prevents page
610 * writes from being double counted.
612 static inline void __tdp_mmu_set_spte(struct kvm *kvm, struct tdp_iter *iter,
613 u64 new_spte, bool record_acc_track,
614 bool record_dirty_log)
616 lockdep_assert_held_write(&kvm->mmu_lock);
619 * No thread should be using this function to set SPTEs to the
620 * temporary removed SPTE value.
621 * If operating under the MMU lock in read mode, tdp_mmu_set_spte_atomic
622 * should be used. If operating under the MMU lock in write mode, the
623 * use of the removed SPTE should not be necessary.
625 WARN_ON(is_removed_spte(iter->old_spte));
627 WRITE_ONCE(*rcu_dereference(iter->sptep), new_spte);
629 __handle_changed_spte(kvm, iter->as_id, iter->gfn, iter->old_spte,
630 new_spte, iter->level, false);
631 if (record_acc_track)
632 handle_changed_spte_acc_track(iter->old_spte, new_spte,
634 if (record_dirty_log)
635 handle_changed_spte_dirty_log(kvm, iter->as_id, iter->gfn,
636 iter->old_spte, new_spte,
640 static inline void tdp_mmu_set_spte(struct kvm *kvm, struct tdp_iter *iter,
643 __tdp_mmu_set_spte(kvm, iter, new_spte, true, true);
646 static inline void tdp_mmu_set_spte_no_acc_track(struct kvm *kvm,
647 struct tdp_iter *iter,
650 __tdp_mmu_set_spte(kvm, iter, new_spte, false, true);
653 static inline void tdp_mmu_set_spte_no_dirty_log(struct kvm *kvm,
654 struct tdp_iter *iter,
657 __tdp_mmu_set_spte(kvm, iter, new_spte, true, false);
660 #define tdp_root_for_each_pte(_iter, _root, _start, _end) \
661 for_each_tdp_pte(_iter, _root->spt, _root->role.level, _start, _end)
663 #define tdp_root_for_each_leaf_pte(_iter, _root, _start, _end) \
664 tdp_root_for_each_pte(_iter, _root, _start, _end) \
665 if (!is_shadow_present_pte(_iter.old_spte) || \
666 !is_last_spte(_iter.old_spte, _iter.level)) \
670 #define tdp_mmu_for_each_pte(_iter, _mmu, _start, _end) \
671 for_each_tdp_pte(_iter, __va(_mmu->root_hpa), \
672 _mmu->shadow_root_level, _start, _end)
675 * Yield if the MMU lock is contended or this thread needs to return control
678 * If this function should yield and flush is set, it will perform a remote
679 * TLB flush before yielding.
681 * If this function yields, it will also reset the tdp_iter's walk over the
682 * paging structure and the calling function should skip to the next
683 * iteration to allow the iterator to continue its traversal from the
684 * paging structure root.
686 * Return true if this function yielded and the iterator's traversal was reset.
687 * Return false if a yield was not needed.
689 static inline bool tdp_mmu_iter_cond_resched(struct kvm *kvm,
690 struct tdp_iter *iter, bool flush,
693 /* Ensure forward progress has been made before yielding. */
694 if (iter->next_last_level_gfn == iter->yielded_gfn)
697 if (need_resched() || rwlock_needbreak(&kvm->mmu_lock)) {
701 kvm_flush_remote_tlbs(kvm);
704 cond_resched_rwlock_read(&kvm->mmu_lock);
706 cond_resched_rwlock_write(&kvm->mmu_lock);
710 WARN_ON(iter->gfn > iter->next_last_level_gfn);
712 tdp_iter_restart(iter);
721 * Tears down the mappings for the range of gfns, [start, end), and frees the
722 * non-root pages mapping GFNs strictly within that range. Returns true if
723 * SPTEs have been cleared and a TLB flush is needed before releasing the
726 * If can_yield is true, will release the MMU lock and reschedule if the
727 * scheduler needs the CPU or there is contention on the MMU lock. If this
728 * function cannot yield, it will not release the MMU lock or reschedule and
729 * the caller must ensure it does not supply too large a GFN range, or the
730 * operation can cause a soft lockup.
732 * If shared is true, this thread holds the MMU lock in read mode and must
733 * account for the possibility that other threads are modifying the paging
734 * structures concurrently. If shared is false, this thread should hold the
735 * MMU lock in write mode.
737 static bool zap_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root,
738 gfn_t start, gfn_t end, bool can_yield, bool flush,
741 gfn_t max_gfn_host = 1ULL << (shadow_phys_bits - PAGE_SHIFT);
742 bool zap_all = (start == 0 && end >= max_gfn_host);
743 struct tdp_iter iter;
746 * No need to try to step down in the iterator when zapping all SPTEs,
747 * zapping the top-level non-leaf SPTEs will recurse on their children.
749 int min_level = zap_all ? root->role.level : PG_LEVEL_4K;
752 * Bound the walk at host.MAXPHYADDR, guest accesses beyond that will
753 * hit a #PF(RSVD) and never get to an EPT Violation/Misconfig / #NPF,
754 * and so KVM will never install a SPTE for such addresses.
756 end = min(end, max_gfn_host);
758 kvm_lockdep_assert_mmu_lock_held(kvm, shared);
762 for_each_tdp_pte_min_level(iter, root->spt, root->role.level,
763 min_level, start, end) {
766 tdp_mmu_iter_cond_resched(kvm, &iter, flush, shared)) {
771 if (!is_shadow_present_pte(iter.old_spte))
775 * If this is a non-last-level SPTE that covers a larger range
776 * than should be zapped, continue, and zap the mappings at a
777 * lower level, except when zapping all SPTEs.
781 iter.gfn + KVM_PAGES_PER_HPAGE(iter.level) > end) &&
782 !is_last_spte(iter.old_spte, iter.level))
786 tdp_mmu_set_spte(kvm, &iter, 0);
788 } else if (!tdp_mmu_zap_spte_atomic(kvm, &iter)) {
790 * The iter must explicitly re-read the SPTE because
791 * the atomic cmpxchg failed.
793 iter.old_spte = READ_ONCE(*rcu_dereference(iter.sptep));
803 * Tears down the mappings for the range of gfns, [start, end), and frees the
804 * non-root pages mapping GFNs strictly within that range. Returns true if
805 * SPTEs have been cleared and a TLB flush is needed before releasing the
808 bool __kvm_tdp_mmu_zap_gfn_range(struct kvm *kvm, int as_id, gfn_t start,
809 gfn_t end, bool can_yield, bool flush)
811 struct kvm_mmu_page *root;
813 for_each_tdp_mmu_root_yield_safe(kvm, root, as_id, false)
814 flush = zap_gfn_range(kvm, root, start, end, can_yield, flush,
820 void kvm_tdp_mmu_zap_all(struct kvm *kvm)
825 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
826 flush = kvm_tdp_mmu_zap_gfn_range(kvm, i, 0, -1ull, flush);
829 kvm_flush_remote_tlbs(kvm);
832 static struct kvm_mmu_page *next_invalidated_root(struct kvm *kvm,
833 struct kvm_mmu_page *prev_root)
835 struct kvm_mmu_page *next_root;
838 next_root = list_next_or_null_rcu(&kvm->arch.tdp_mmu_roots,
840 typeof(*prev_root), link);
842 next_root = list_first_or_null_rcu(&kvm->arch.tdp_mmu_roots,
843 typeof(*next_root), link);
845 while (next_root && !(next_root->role.invalid &&
846 refcount_read(&next_root->tdp_mmu_root_count)))
847 next_root = list_next_or_null_rcu(&kvm->arch.tdp_mmu_roots,
849 typeof(*next_root), link);
855 * Since kvm_tdp_mmu_zap_all_fast has acquired a reference to each
856 * invalidated root, they will not be freed until this function drops the
857 * reference. Before dropping that reference, tear down the paging
858 * structure so that whichever thread does drop the last reference
859 * only has to do a trivial amount of work. Since the roots are invalid,
860 * no new SPTEs should be created under them.
862 void kvm_tdp_mmu_zap_invalidated_roots(struct kvm *kvm)
864 struct kvm_mmu_page *next_root;
865 struct kvm_mmu_page *root;
868 lockdep_assert_held_read(&kvm->mmu_lock);
872 root = next_invalidated_root(kvm, NULL);
875 next_root = next_invalidated_root(kvm, root);
879 flush = zap_gfn_range(kvm, root, 0, -1ull, true, flush, true);
882 * Put the reference acquired in
883 * kvm_tdp_mmu_invalidate_roots
885 kvm_tdp_mmu_put_root(kvm, root, true);
895 kvm_flush_remote_tlbs(kvm);
899 * Mark each TDP MMU root as invalid so that other threads
900 * will drop their references and allow the root count to
903 * Also take a reference on all roots so that this thread
904 * can do the bulk of the work required to free the roots
905 * once they are invalidated. Without this reference, a
906 * vCPU thread might drop the last reference to a root and
907 * get stuck with tearing down the entire paging structure.
909 * Roots which have a zero refcount should be skipped as
910 * they're already being torn down.
911 * Already invalid roots should be referenced again so that
912 * they aren't freed before kvm_tdp_mmu_zap_all_fast is
915 * This has essentially the same effect for the TDP MMU
916 * as updating mmu_valid_gen does for the shadow MMU.
918 void kvm_tdp_mmu_invalidate_all_roots(struct kvm *kvm)
920 struct kvm_mmu_page *root;
922 lockdep_assert_held_write(&kvm->mmu_lock);
923 list_for_each_entry(root, &kvm->arch.tdp_mmu_roots, link)
924 if (refcount_inc_not_zero(&root->tdp_mmu_root_count))
925 root->role.invalid = true;
929 * Installs a last-level SPTE to handle a TDP page fault.
930 * (NPT/EPT violation/misconfiguration)
932 static int tdp_mmu_map_handle_target_level(struct kvm_vcpu *vcpu, int write,
934 struct tdp_iter *iter,
935 kvm_pfn_t pfn, bool prefault)
938 int ret = RET_PF_FIXED;
939 int make_spte_ret = 0;
941 if (unlikely(is_noslot_pfn(pfn)))
942 new_spte = make_mmio_spte(vcpu, iter->gfn, ACC_ALL);
944 make_spte_ret = make_spte(vcpu, ACC_ALL, iter->level, iter->gfn,
945 pfn, iter->old_spte, prefault, true,
946 map_writable, !shadow_accessed_mask,
949 if (new_spte == iter->old_spte)
950 ret = RET_PF_SPURIOUS;
951 else if (!tdp_mmu_map_set_spte_atomic(vcpu, iter, new_spte))
955 * If the page fault was caused by a write but the page is write
956 * protected, emulation is needed. If the emulation was skipped,
957 * the vCPU would have the same fault again.
959 if (make_spte_ret & SET_SPTE_WRITE_PROTECTED_PT) {
961 ret = RET_PF_EMULATE;
962 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
965 /* If a MMIO SPTE is installed, the MMIO will need to be emulated. */
966 if (unlikely(is_mmio_spte(new_spte))) {
967 trace_mark_mmio_spte(rcu_dereference(iter->sptep), iter->gfn,
969 ret = RET_PF_EMULATE;
971 trace_kvm_mmu_set_spte(iter->level, iter->gfn,
972 rcu_dereference(iter->sptep));
976 * Increase pf_fixed in both RET_PF_EMULATE and RET_PF_FIXED to be
977 * consistent with legacy MMU behavior.
979 if (ret != RET_PF_SPURIOUS)
980 vcpu->stat.pf_fixed++;
986 * Handle a TDP page fault (NPT/EPT violation/misconfiguration) by installing
987 * page tables and SPTEs to translate the faulting guest physical address.
989 int kvm_tdp_mmu_map(struct kvm_vcpu *vcpu, gpa_t gpa, u32 error_code,
990 int map_writable, int max_level, kvm_pfn_t pfn,
993 bool nx_huge_page_workaround_enabled = is_nx_huge_page_enabled();
994 bool write = error_code & PFERR_WRITE_MASK;
995 bool exec = error_code & PFERR_FETCH_MASK;
996 bool huge_page_disallowed = exec && nx_huge_page_workaround_enabled;
997 struct kvm_mmu *mmu = vcpu->arch.mmu;
998 struct tdp_iter iter;
999 struct kvm_mmu_page *sp;
1003 gfn_t gfn = gpa >> PAGE_SHIFT;
1007 level = kvm_mmu_hugepage_adjust(vcpu, gfn, max_level, &pfn,
1008 huge_page_disallowed, &req_level);
1010 trace_kvm_mmu_spte_requested(gpa, level, pfn);
1014 tdp_mmu_for_each_pte(iter, mmu, gfn, gfn + 1) {
1015 if (nx_huge_page_workaround_enabled)
1016 disallowed_hugepage_adjust(iter.old_spte, gfn,
1017 iter.level, &pfn, &level);
1019 if (iter.level == level)
1023 * If there is an SPTE mapping a large page at a higher level
1024 * than the target, that SPTE must be cleared and replaced
1025 * with a non-leaf SPTE.
1027 if (is_shadow_present_pte(iter.old_spte) &&
1028 is_large_pte(iter.old_spte)) {
1029 if (!tdp_mmu_zap_spte_atomic(vcpu->kvm, &iter))
1033 * The iter must explicitly re-read the spte here
1034 * because the new value informs the !present
1037 iter.old_spte = READ_ONCE(*rcu_dereference(iter.sptep));
1040 if (!is_shadow_present_pte(iter.old_spte)) {
1042 * If SPTE has been frozen by another thread, just
1043 * give up and retry, avoiding unnecessary page table
1044 * allocation and free.
1046 if (is_removed_spte(iter.old_spte))
1049 sp = alloc_tdp_mmu_page(vcpu, iter.gfn, iter.level - 1);
1052 new_spte = make_nonleaf_spte(child_pt,
1053 !shadow_accessed_mask);
1055 if (tdp_mmu_set_spte_atomic_no_dirty_log(vcpu->kvm, &iter, new_spte)) {
1056 tdp_mmu_link_page(vcpu->kvm, sp,
1057 huge_page_disallowed &&
1058 req_level >= iter.level);
1060 trace_kvm_mmu_get_page(sp, true);
1062 tdp_mmu_free_sp(sp);
1068 if (iter.level != level) {
1070 return RET_PF_RETRY;
1073 ret = tdp_mmu_map_handle_target_level(vcpu, write, map_writable, &iter,
1080 bool kvm_tdp_mmu_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range,
1083 struct kvm_mmu_page *root;
1085 for_each_tdp_mmu_root(kvm, root, range->slot->as_id)
1086 flush |= zap_gfn_range(kvm, root, range->start, range->end,
1087 range->may_block, flush, false);
1092 typedef bool (*tdp_handler_t)(struct kvm *kvm, struct tdp_iter *iter,
1093 struct kvm_gfn_range *range);
1095 static __always_inline bool kvm_tdp_mmu_handle_gfn(struct kvm *kvm,
1096 struct kvm_gfn_range *range,
1097 tdp_handler_t handler)
1099 struct kvm_mmu_page *root;
1100 struct tdp_iter iter;
1106 * Don't support rescheduling, none of the MMU notifiers that funnel
1107 * into this helper allow blocking; it'd be dead, wasteful code.
1109 for_each_tdp_mmu_root(kvm, root, range->slot->as_id) {
1110 tdp_root_for_each_leaf_pte(iter, root, range->start, range->end)
1111 ret |= handler(kvm, &iter, range);
1120 * Mark the SPTEs range of GFNs [start, end) unaccessed and return non-zero
1121 * if any of the GFNs in the range have been accessed.
1123 static bool age_gfn_range(struct kvm *kvm, struct tdp_iter *iter,
1124 struct kvm_gfn_range *range)
1128 /* If we have a non-accessed entry we don't need to change the pte. */
1129 if (!is_accessed_spte(iter->old_spte))
1132 new_spte = iter->old_spte;
1134 if (spte_ad_enabled(new_spte)) {
1135 new_spte &= ~shadow_accessed_mask;
1138 * Capture the dirty status of the page, so that it doesn't get
1139 * lost when the SPTE is marked for access tracking.
1141 if (is_writable_pte(new_spte))
1142 kvm_set_pfn_dirty(spte_to_pfn(new_spte));
1144 new_spte = mark_spte_for_access_track(new_spte);
1147 tdp_mmu_set_spte_no_acc_track(kvm, iter, new_spte);
1152 bool kvm_tdp_mmu_age_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range)
1154 return kvm_tdp_mmu_handle_gfn(kvm, range, age_gfn_range);
1157 static bool test_age_gfn(struct kvm *kvm, struct tdp_iter *iter,
1158 struct kvm_gfn_range *range)
1160 return is_accessed_spte(iter->old_spte);
1163 bool kvm_tdp_mmu_test_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range)
1165 return kvm_tdp_mmu_handle_gfn(kvm, range, test_age_gfn);
1168 static bool set_spte_gfn(struct kvm *kvm, struct tdp_iter *iter,
1169 struct kvm_gfn_range *range)
1173 /* Huge pages aren't expected to be modified without first being zapped. */
1174 WARN_ON(pte_huge(range->pte) || range->start + 1 != range->end);
1176 if (iter->level != PG_LEVEL_4K ||
1177 !is_shadow_present_pte(iter->old_spte))
1181 * Note, when changing a read-only SPTE, it's not strictly necessary to
1182 * zero the SPTE before setting the new PFN, but doing so preserves the
1183 * invariant that the PFN of a present * leaf SPTE can never change.
1184 * See __handle_changed_spte().
1186 tdp_mmu_set_spte(kvm, iter, 0);
1188 if (!pte_write(range->pte)) {
1189 new_spte = kvm_mmu_changed_pte_notifier_make_spte(iter->old_spte,
1190 pte_pfn(range->pte));
1192 tdp_mmu_set_spte(kvm, iter, new_spte);
1199 * Handle the changed_pte MMU notifier for the TDP MMU.
1200 * data is a pointer to the new pte_t mapping the HVA specified by the MMU
1202 * Returns non-zero if a flush is needed before releasing the MMU lock.
1204 bool kvm_tdp_mmu_set_spte_gfn(struct kvm *kvm, struct kvm_gfn_range *range)
1206 bool flush = kvm_tdp_mmu_handle_gfn(kvm, range, set_spte_gfn);
1208 /* FIXME: return 'flush' instead of flushing here. */
1210 kvm_flush_remote_tlbs_with_address(kvm, range->start, 1);
1216 * Remove write access from all SPTEs at or above min_level that map GFNs
1217 * [start, end). Returns true if an SPTE has been changed and the TLBs need to
1220 static bool wrprot_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root,
1221 gfn_t start, gfn_t end, int min_level)
1223 struct tdp_iter iter;
1225 bool spte_set = false;
1229 BUG_ON(min_level > KVM_MAX_HUGEPAGE_LEVEL);
1231 for_each_tdp_pte_min_level(iter, root->spt, root->role.level,
1232 min_level, start, end) {
1234 if (tdp_mmu_iter_cond_resched(kvm, &iter, false, true))
1237 if (!is_shadow_present_pte(iter.old_spte) ||
1238 !is_last_spte(iter.old_spte, iter.level) ||
1239 !(iter.old_spte & PT_WRITABLE_MASK))
1242 new_spte = iter.old_spte & ~PT_WRITABLE_MASK;
1244 if (!tdp_mmu_set_spte_atomic_no_dirty_log(kvm, &iter,
1247 * The iter must explicitly re-read the SPTE because
1248 * the atomic cmpxchg failed.
1250 iter.old_spte = READ_ONCE(*rcu_dereference(iter.sptep));
1261 * Remove write access from all the SPTEs mapping GFNs in the memslot. Will
1262 * only affect leaf SPTEs down to min_level.
1263 * Returns true if an SPTE has been changed and the TLBs need to be flushed.
1265 bool kvm_tdp_mmu_wrprot_slot(struct kvm *kvm,
1266 const struct kvm_memory_slot *slot, int min_level)
1268 struct kvm_mmu_page *root;
1269 bool spte_set = false;
1271 lockdep_assert_held_read(&kvm->mmu_lock);
1273 for_each_tdp_mmu_root_yield_safe(kvm, root, slot->as_id, true)
1274 spte_set |= wrprot_gfn_range(kvm, root, slot->base_gfn,
1275 slot->base_gfn + slot->npages, min_level);
1281 * Clear the dirty status of all the SPTEs mapping GFNs in the memslot. If
1282 * AD bits are enabled, this will involve clearing the dirty bit on each SPTE.
1283 * If AD bits are not enabled, this will require clearing the writable bit on
1284 * each SPTE. Returns true if an SPTE has been changed and the TLBs need to
1287 static bool clear_dirty_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root,
1288 gfn_t start, gfn_t end)
1290 struct tdp_iter iter;
1292 bool spte_set = false;
1296 tdp_root_for_each_leaf_pte(iter, root, start, end) {
1298 if (tdp_mmu_iter_cond_resched(kvm, &iter, false, true))
1301 if (spte_ad_need_write_protect(iter.old_spte)) {
1302 if (is_writable_pte(iter.old_spte))
1303 new_spte = iter.old_spte & ~PT_WRITABLE_MASK;
1307 if (iter.old_spte & shadow_dirty_mask)
1308 new_spte = iter.old_spte & ~shadow_dirty_mask;
1313 if (!tdp_mmu_set_spte_atomic_no_dirty_log(kvm, &iter,
1316 * The iter must explicitly re-read the SPTE because
1317 * the atomic cmpxchg failed.
1319 iter.old_spte = READ_ONCE(*rcu_dereference(iter.sptep));
1330 * Clear the dirty status of all the SPTEs mapping GFNs in the memslot. If
1331 * AD bits are enabled, this will involve clearing the dirty bit on each SPTE.
1332 * If AD bits are not enabled, this will require clearing the writable bit on
1333 * each SPTE. Returns true if an SPTE has been changed and the TLBs need to
1336 bool kvm_tdp_mmu_clear_dirty_slot(struct kvm *kvm,
1337 const struct kvm_memory_slot *slot)
1339 struct kvm_mmu_page *root;
1340 bool spte_set = false;
1342 lockdep_assert_held_read(&kvm->mmu_lock);
1344 for_each_tdp_mmu_root_yield_safe(kvm, root, slot->as_id, true)
1345 spte_set |= clear_dirty_gfn_range(kvm, root, slot->base_gfn,
1346 slot->base_gfn + slot->npages);
1352 * Clears the dirty status of all the 4k SPTEs mapping GFNs for which a bit is
1353 * set in mask, starting at gfn. The given memslot is expected to contain all
1354 * the GFNs represented by set bits in the mask. If AD bits are enabled,
1355 * clearing the dirty status will involve clearing the dirty bit on each SPTE
1356 * or, if AD bits are not enabled, clearing the writable bit on each SPTE.
1358 static void clear_dirty_pt_masked(struct kvm *kvm, struct kvm_mmu_page *root,
1359 gfn_t gfn, unsigned long mask, bool wrprot)
1361 struct tdp_iter iter;
1366 tdp_root_for_each_leaf_pte(iter, root, gfn + __ffs(mask),
1367 gfn + BITS_PER_LONG) {
1371 if (iter.level > PG_LEVEL_4K ||
1372 !(mask & (1UL << (iter.gfn - gfn))))
1375 mask &= ~(1UL << (iter.gfn - gfn));
1377 if (wrprot || spte_ad_need_write_protect(iter.old_spte)) {
1378 if (is_writable_pte(iter.old_spte))
1379 new_spte = iter.old_spte & ~PT_WRITABLE_MASK;
1383 if (iter.old_spte & shadow_dirty_mask)
1384 new_spte = iter.old_spte & ~shadow_dirty_mask;
1389 tdp_mmu_set_spte_no_dirty_log(kvm, &iter, new_spte);
1396 * Clears the dirty status of all the 4k SPTEs mapping GFNs for which a bit is
1397 * set in mask, starting at gfn. The given memslot is expected to contain all
1398 * the GFNs represented by set bits in the mask. If AD bits are enabled,
1399 * clearing the dirty status will involve clearing the dirty bit on each SPTE
1400 * or, if AD bits are not enabled, clearing the writable bit on each SPTE.
1402 void kvm_tdp_mmu_clear_dirty_pt_masked(struct kvm *kvm,
1403 struct kvm_memory_slot *slot,
1404 gfn_t gfn, unsigned long mask,
1407 struct kvm_mmu_page *root;
1409 lockdep_assert_held_write(&kvm->mmu_lock);
1410 for_each_tdp_mmu_root(kvm, root, slot->as_id)
1411 clear_dirty_pt_masked(kvm, root, gfn, mask, wrprot);
1415 * Clear leaf entries which could be replaced by large mappings, for
1416 * GFNs within the slot.
1418 static bool zap_collapsible_spte_range(struct kvm *kvm,
1419 struct kvm_mmu_page *root,
1420 const struct kvm_memory_slot *slot,
1423 gfn_t start = slot->base_gfn;
1424 gfn_t end = start + slot->npages;
1425 struct tdp_iter iter;
1430 tdp_root_for_each_pte(iter, root, start, end) {
1432 if (tdp_mmu_iter_cond_resched(kvm, &iter, flush, true)) {
1437 if (!is_shadow_present_pte(iter.old_spte) ||
1438 !is_last_spte(iter.old_spte, iter.level))
1441 pfn = spte_to_pfn(iter.old_spte);
1442 if (kvm_is_reserved_pfn(pfn) ||
1443 iter.level >= kvm_mmu_max_mapping_level(kvm, slot, iter.gfn,
1447 if (!tdp_mmu_zap_spte_atomic(kvm, &iter)) {
1449 * The iter must explicitly re-read the SPTE because
1450 * the atomic cmpxchg failed.
1452 iter.old_spte = READ_ONCE(*rcu_dereference(iter.sptep));
1464 * Clear non-leaf entries (and free associated page tables) which could
1465 * be replaced by large mappings, for GFNs within the slot.
1467 bool kvm_tdp_mmu_zap_collapsible_sptes(struct kvm *kvm,
1468 const struct kvm_memory_slot *slot,
1471 struct kvm_mmu_page *root;
1473 lockdep_assert_held_read(&kvm->mmu_lock);
1475 for_each_tdp_mmu_root_yield_safe(kvm, root, slot->as_id, true)
1476 flush = zap_collapsible_spte_range(kvm, root, slot, flush);
1482 * Removes write access on the last level SPTE mapping this GFN and unsets the
1483 * MMU-writable bit to ensure future writes continue to be intercepted.
1484 * Returns true if an SPTE was set and a TLB flush is needed.
1486 static bool write_protect_gfn(struct kvm *kvm, struct kvm_mmu_page *root,
1487 gfn_t gfn, int min_level)
1489 struct tdp_iter iter;
1491 bool spte_set = false;
1493 BUG_ON(min_level > KVM_MAX_HUGEPAGE_LEVEL);
1497 for_each_tdp_pte_min_level(iter, root->spt, root->role.level,
1498 min_level, gfn, gfn + 1) {
1499 if (!is_shadow_present_pte(iter.old_spte) ||
1500 !is_last_spte(iter.old_spte, iter.level))
1503 if (!is_writable_pte(iter.old_spte))
1506 new_spte = iter.old_spte &
1507 ~(PT_WRITABLE_MASK | shadow_mmu_writable_mask);
1509 tdp_mmu_set_spte(kvm, &iter, new_spte);
1519 * Removes write access on the last level SPTE mapping this GFN and unsets the
1520 * MMU-writable bit to ensure future writes continue to be intercepted.
1521 * Returns true if an SPTE was set and a TLB flush is needed.
1523 bool kvm_tdp_mmu_write_protect_gfn(struct kvm *kvm,
1524 struct kvm_memory_slot *slot, gfn_t gfn,
1527 struct kvm_mmu_page *root;
1528 bool spte_set = false;
1530 lockdep_assert_held_write(&kvm->mmu_lock);
1531 for_each_tdp_mmu_root(kvm, root, slot->as_id)
1532 spte_set |= write_protect_gfn(kvm, root, gfn, min_level);
1538 * Return the level of the lowest level SPTE added to sptes.
1539 * That SPTE may be non-present.
1541 * Must be called between kvm_tdp_mmu_walk_lockless_{begin,end}.
1543 int kvm_tdp_mmu_get_walk(struct kvm_vcpu *vcpu, u64 addr, u64 *sptes,
1546 struct tdp_iter iter;
1547 struct kvm_mmu *mmu = vcpu->arch.mmu;
1548 gfn_t gfn = addr >> PAGE_SHIFT;
1551 *root_level = vcpu->arch.mmu->shadow_root_level;
1553 tdp_mmu_for_each_pte(iter, mmu, gfn, gfn + 1) {
1555 sptes[leaf] = iter.old_spte;
1562 * Returns the last level spte pointer of the shadow page walk for the given
1563 * gpa, and sets *spte to the spte value. This spte may be non-preset. If no
1564 * walk could be performed, returns NULL and *spte does not contain valid data.
1567 * - Must be called between kvm_tdp_mmu_walk_lockless_{begin,end}.
1568 * - The returned sptep must not be used after kvm_tdp_mmu_walk_lockless_end.
1570 * WARNING: This function is only intended to be called during fast_page_fault.
1572 u64 *kvm_tdp_mmu_fast_pf_get_last_sptep(struct kvm_vcpu *vcpu, u64 addr,
1575 struct tdp_iter iter;
1576 struct kvm_mmu *mmu = vcpu->arch.mmu;
1577 gfn_t gfn = addr >> PAGE_SHIFT;
1578 tdp_ptep_t sptep = NULL;
1580 tdp_mmu_for_each_pte(iter, mmu, gfn, gfn + 1) {
1581 *spte = iter.old_spte;
1586 * Perform the rcu_dereference to get the raw spte pointer value since
1587 * we are passing it up to fast_page_fault, which is shared with the
1588 * legacy MMU and thus does not retain the TDP MMU-specific __rcu
1591 * This is safe since fast_page_fault obeys the contracts of this
1592 * function as well as all TDP MMU contracts around modifying SPTEs
1593 * outside of mmu_lock.
1595 return rcu_dereference(sptep);