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 = false;
14 module_param_named(tdp_mmu, tdp_mmu_enabled, bool, 0644);
16 /* Initializes the TDP MMU for the VM, if enabled. */
17 void 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);
30 void kvm_mmu_uninit_tdp_mmu(struct kvm *kvm)
32 if (!kvm->arch.tdp_mmu_enabled)
35 WARN_ON(!list_empty(&kvm->arch.tdp_mmu_roots));
38 * Ensure that all the outstanding RCU callbacks to free shadow pages
39 * can run before the VM is torn down.
44 static void tdp_mmu_put_root(struct kvm *kvm, struct kvm_mmu_page *root)
46 if (kvm_mmu_put_root(kvm, root))
47 kvm_tdp_mmu_free_root(kvm, root);
50 static inline bool tdp_mmu_next_root_valid(struct kvm *kvm,
51 struct kvm_mmu_page *root)
53 lockdep_assert_held_write(&kvm->mmu_lock);
55 if (list_entry_is_head(root, &kvm->arch.tdp_mmu_roots, link))
58 kvm_mmu_get_root(kvm, root);
63 static inline struct kvm_mmu_page *tdp_mmu_next_root(struct kvm *kvm,
64 struct kvm_mmu_page *root)
66 struct kvm_mmu_page *next_root;
68 next_root = list_next_entry(root, link);
69 tdp_mmu_put_root(kvm, root);
74 * Note: this iterator gets and puts references to the roots it iterates over.
75 * This makes it safe to release the MMU lock and yield within the loop, but
76 * if exiting the loop early, the caller must drop the reference to the most
77 * recent root. (Unless keeping a live reference is desirable.)
79 #define for_each_tdp_mmu_root_yield_safe(_kvm, _root) \
80 for (_root = list_first_entry(&_kvm->arch.tdp_mmu_roots, \
81 typeof(*_root), link); \
82 tdp_mmu_next_root_valid(_kvm, _root); \
83 _root = tdp_mmu_next_root(_kvm, _root))
85 #define for_each_tdp_mmu_root(_kvm, _root) \
86 list_for_each_entry(_root, &_kvm->arch.tdp_mmu_roots, link)
88 static bool zap_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root,
89 gfn_t start, gfn_t end, bool can_yield, bool flush);
91 void kvm_tdp_mmu_free_root(struct kvm *kvm, struct kvm_mmu_page *root)
93 gfn_t max_gfn = 1ULL << (shadow_phys_bits - PAGE_SHIFT);
95 lockdep_assert_held_write(&kvm->mmu_lock);
97 WARN_ON(root->root_count);
98 WARN_ON(!root->tdp_mmu_page);
100 list_del(&root->link);
102 zap_gfn_range(kvm, root, 0, max_gfn, false, false);
104 free_page((unsigned long)root->spt);
105 kmem_cache_free(mmu_page_header_cache, root);
108 static union kvm_mmu_page_role page_role_for_level(struct kvm_vcpu *vcpu,
111 union kvm_mmu_page_role role;
113 role = vcpu->arch.mmu->mmu_role.base;
116 role.gpte_is_8_bytes = true;
117 role.access = ACC_ALL;
122 static struct kvm_mmu_page *alloc_tdp_mmu_page(struct kvm_vcpu *vcpu, gfn_t gfn,
125 struct kvm_mmu_page *sp;
127 sp = kvm_mmu_memory_cache_alloc(&vcpu->arch.mmu_page_header_cache);
128 sp->spt = kvm_mmu_memory_cache_alloc(&vcpu->arch.mmu_shadow_page_cache);
129 set_page_private(virt_to_page(sp->spt), (unsigned long)sp);
131 sp->role.word = page_role_for_level(vcpu, level).word;
133 sp->tdp_mmu_page = true;
135 trace_kvm_mmu_get_page(sp, true);
140 static struct kvm_mmu_page *get_tdp_mmu_vcpu_root(struct kvm_vcpu *vcpu)
142 union kvm_mmu_page_role role;
143 struct kvm *kvm = vcpu->kvm;
144 struct kvm_mmu_page *root;
146 role = page_role_for_level(vcpu, vcpu->arch.mmu->shadow_root_level);
148 write_lock(&kvm->mmu_lock);
150 /* Check for an existing root before allocating a new one. */
151 for_each_tdp_mmu_root(kvm, root) {
152 if (root->role.word == role.word) {
153 kvm_mmu_get_root(kvm, root);
154 write_unlock(&kvm->mmu_lock);
159 root = alloc_tdp_mmu_page(vcpu, 0, vcpu->arch.mmu->shadow_root_level);
160 root->root_count = 1;
162 list_add(&root->link, &kvm->arch.tdp_mmu_roots);
164 write_unlock(&kvm->mmu_lock);
169 hpa_t kvm_tdp_mmu_get_vcpu_root_hpa(struct kvm_vcpu *vcpu)
171 struct kvm_mmu_page *root;
173 root = get_tdp_mmu_vcpu_root(vcpu);
177 return __pa(root->spt);
180 static void tdp_mmu_free_sp(struct kvm_mmu_page *sp)
182 free_page((unsigned long)sp->spt);
183 kmem_cache_free(mmu_page_header_cache, sp);
187 * This is called through call_rcu in order to free TDP page table memory
188 * safely with respect to other kernel threads that may be operating on
190 * By only accessing TDP MMU page table memory in an RCU read critical
191 * section, and freeing it after a grace period, lockless access to that
192 * memory won't use it after it is freed.
194 static void tdp_mmu_free_sp_rcu_callback(struct rcu_head *head)
196 struct kvm_mmu_page *sp = container_of(head, struct kvm_mmu_page,
202 static void handle_changed_spte(struct kvm *kvm, int as_id, gfn_t gfn,
203 u64 old_spte, u64 new_spte, int level,
206 static void handle_changed_spte_acc_track(u64 old_spte, u64 new_spte, int level)
208 bool pfn_changed = spte_to_pfn(old_spte) != spte_to_pfn(new_spte);
210 if (!is_shadow_present_pte(old_spte) || !is_last_spte(old_spte, level))
213 if (is_accessed_spte(old_spte) &&
214 (!is_accessed_spte(new_spte) || pfn_changed))
215 kvm_set_pfn_accessed(spte_to_pfn(old_spte));
218 static void handle_changed_spte_dirty_log(struct kvm *kvm, int as_id, gfn_t gfn,
219 u64 old_spte, u64 new_spte, int level)
222 struct kvm_memory_slot *slot;
224 if (level > PG_LEVEL_4K)
227 pfn_changed = spte_to_pfn(old_spte) != spte_to_pfn(new_spte);
229 if ((!is_writable_pte(old_spte) || pfn_changed) &&
230 is_writable_pte(new_spte)) {
231 slot = __gfn_to_memslot(__kvm_memslots(kvm, as_id), gfn);
232 mark_page_dirty_in_slot(kvm, slot, gfn);
237 * tdp_mmu_link_page - Add a new page to the list of pages used by the TDP MMU
241 * @shared: This operation may not be running under the exclusive use of
242 * the MMU lock and the operation must synchronize with other
243 * threads that might be adding or removing pages.
244 * @account_nx: This page replaces a NX large page and should be marked for
247 static void tdp_mmu_link_page(struct kvm *kvm, struct kvm_mmu_page *sp,
248 bool shared, bool account_nx)
251 spin_lock(&kvm->arch.tdp_mmu_pages_lock);
253 lockdep_assert_held_write(&kvm->mmu_lock);
255 list_add(&sp->link, &kvm->arch.tdp_mmu_pages);
257 account_huge_nx_page(kvm, sp);
260 spin_unlock(&kvm->arch.tdp_mmu_pages_lock);
264 * tdp_mmu_unlink_page - Remove page from the list of pages used by the TDP MMU
267 * @sp: the page to be removed
268 * @shared: This operation may not be running under the exclusive use of
269 * the MMU lock and the operation must synchronize with other
270 * threads that might be adding or removing pages.
272 static void tdp_mmu_unlink_page(struct kvm *kvm, struct kvm_mmu_page *sp,
276 spin_lock(&kvm->arch.tdp_mmu_pages_lock);
278 lockdep_assert_held_write(&kvm->mmu_lock);
281 if (sp->lpage_disallowed)
282 unaccount_huge_nx_page(kvm, sp);
285 spin_unlock(&kvm->arch.tdp_mmu_pages_lock);
289 * handle_removed_tdp_mmu_page - handle a pt removed from the TDP structure
292 * @pt: the page removed from the paging structure
293 * @shared: This operation may not be running under the exclusive use
294 * of the MMU lock and the operation must synchronize with other
295 * threads that might be modifying SPTEs.
297 * Given a page table that has been removed from the TDP paging structure,
298 * iterates through the page table to clear SPTEs and free child page tables.
300 * Note that pt is passed in as a tdp_ptep_t, but it does not need RCU
301 * protection. Since this thread removed it from the paging structure,
302 * this thread will be responsible for ensuring the page is freed. Hence the
303 * early rcu_dereferences in the function.
305 static void handle_removed_tdp_mmu_page(struct kvm *kvm, tdp_ptep_t pt,
308 struct kvm_mmu_page *sp = sptep_to_sp(rcu_dereference(pt));
309 int level = sp->role.level;
310 gfn_t base_gfn = sp->gfn;
316 trace_kvm_mmu_prepare_zap_page(sp);
318 tdp_mmu_unlink_page(kvm, sp, shared);
320 for (i = 0; i < PT64_ENT_PER_PAGE; i++) {
321 sptep = rcu_dereference(pt) + i;
322 gfn = base_gfn + (i * KVM_PAGES_PER_HPAGE(level - 1));
326 * Set the SPTE to a nonpresent value that other
327 * threads will not overwrite. If the SPTE was
328 * already marked as removed then another thread
329 * handling a page fault could overwrite it, so
330 * set the SPTE until it is set from some other
331 * value to the removed SPTE value.
334 old_child_spte = xchg(sptep, REMOVED_SPTE);
335 if (!is_removed_spte(old_child_spte))
341 * If the SPTE is not MMU-present, there is no backing
342 * page associated with the SPTE and so no side effects
343 * that need to be recorded, and exclusive ownership of
344 * mmu_lock ensures the SPTE can't be made present.
345 * Note, zapping MMIO SPTEs is also unnecessary as they
346 * are guarded by the memslots generation, not by being
349 old_child_spte = READ_ONCE(*sptep);
350 if (!is_shadow_present_pte(old_child_spte))
354 * Marking the SPTE as a removed SPTE is not
355 * strictly necessary here as the MMU lock will
356 * stop other threads from concurrently modifying
357 * this SPTE. Using the removed SPTE value keeps
358 * the two branches consistent and simplifies
361 WRITE_ONCE(*sptep, REMOVED_SPTE);
363 handle_changed_spte(kvm, kvm_mmu_page_as_id(sp), gfn,
364 old_child_spte, REMOVED_SPTE, level - 1,
368 kvm_flush_remote_tlbs_with_address(kvm, gfn,
369 KVM_PAGES_PER_HPAGE(level));
371 call_rcu(&sp->rcu_head, tdp_mmu_free_sp_rcu_callback);
375 * handle_changed_spte - handle bookkeeping associated with an SPTE change
377 * @as_id: the address space of the paging structure the SPTE was a part of
378 * @gfn: the base GFN that was mapped by the SPTE
379 * @old_spte: The value of the SPTE before the change
380 * @new_spte: The value of the SPTE after the change
381 * @level: the level of the PT the SPTE is part of in the paging structure
382 * @shared: This operation may not be running under the exclusive use of
383 * the MMU lock and the operation must synchronize with other
384 * threads that might be modifying SPTEs.
386 * Handle bookkeeping that might result from the modification of a SPTE.
387 * This function must be called for all TDP SPTE modifications.
389 static void __handle_changed_spte(struct kvm *kvm, int as_id, gfn_t gfn,
390 u64 old_spte, u64 new_spte, int level,
393 bool was_present = is_shadow_present_pte(old_spte);
394 bool is_present = is_shadow_present_pte(new_spte);
395 bool was_leaf = was_present && is_last_spte(old_spte, level);
396 bool is_leaf = is_present && is_last_spte(new_spte, level);
397 bool pfn_changed = spte_to_pfn(old_spte) != spte_to_pfn(new_spte);
399 WARN_ON(level > PT64_ROOT_MAX_LEVEL);
400 WARN_ON(level < PG_LEVEL_4K);
401 WARN_ON(gfn & (KVM_PAGES_PER_HPAGE(level) - 1));
404 * If this warning were to trigger it would indicate that there was a
405 * missing MMU notifier or a race with some notifier handler.
406 * A present, leaf SPTE should never be directly replaced with another
407 * present leaf SPTE pointing to a different PFN. A notifier handler
408 * should be zapping the SPTE before the main MM's page table is
409 * changed, or the SPTE should be zeroed, and the TLBs flushed by the
410 * thread before replacement.
412 if (was_leaf && is_leaf && pfn_changed) {
413 pr_err("Invalid SPTE change: cannot replace a present leaf\n"
414 "SPTE with another present leaf SPTE mapping a\n"
416 "as_id: %d gfn: %llx old_spte: %llx new_spte: %llx level: %d",
417 as_id, gfn, old_spte, new_spte, level);
420 * Crash the host to prevent error propagation and guest data
426 if (old_spte == new_spte)
429 trace_kvm_tdp_mmu_spte_changed(as_id, gfn, level, old_spte, new_spte);
432 * The only times a SPTE should be changed from a non-present to
433 * non-present state is when an MMIO entry is installed/modified/
434 * removed. In that case, there is nothing to do here.
436 if (!was_present && !is_present) {
438 * If this change does not involve a MMIO SPTE or removed SPTE,
439 * it is unexpected. Log the change, though it should not
440 * impact the guest since both the former and current SPTEs
443 if (WARN_ON(!is_mmio_spte(old_spte) &&
444 !is_mmio_spte(new_spte) &&
445 !is_removed_spte(new_spte)))
446 pr_err("Unexpected SPTE change! Nonpresent SPTEs\n"
447 "should not be replaced with another,\n"
448 "different nonpresent SPTE, unless one or both\n"
449 "are MMIO SPTEs, or the new SPTE is\n"
450 "a temporary removed SPTE.\n"
451 "as_id: %d gfn: %llx old_spte: %llx new_spte: %llx level: %d",
452 as_id, gfn, old_spte, new_spte, level);
457 if (was_leaf && is_dirty_spte(old_spte) &&
458 (!is_dirty_spte(new_spte) || pfn_changed))
459 kvm_set_pfn_dirty(spte_to_pfn(old_spte));
462 * Recursively handle child PTs if the change removed a subtree from
463 * the paging structure.
465 if (was_present && !was_leaf && (pfn_changed || !is_present))
466 handle_removed_tdp_mmu_page(kvm,
467 spte_to_child_pt(old_spte, level), shared);
470 static void handle_changed_spte(struct kvm *kvm, int as_id, gfn_t gfn,
471 u64 old_spte, u64 new_spte, int level,
474 __handle_changed_spte(kvm, as_id, gfn, old_spte, new_spte, level,
476 handle_changed_spte_acc_track(old_spte, new_spte, level);
477 handle_changed_spte_dirty_log(kvm, as_id, gfn, old_spte,
482 * tdp_mmu_set_spte_atomic - Set a TDP MMU SPTE atomically and handle the
483 * associated bookkeeping
486 * @iter: a tdp_iter instance currently on the SPTE that should be set
487 * @new_spte: The value the SPTE should be set to
488 * Returns: true if the SPTE was set, false if it was not. If false is returned,
489 * this function will have no side-effects.
491 static inline bool tdp_mmu_set_spte_atomic(struct kvm *kvm,
492 struct tdp_iter *iter,
495 lockdep_assert_held_read(&kvm->mmu_lock);
498 * Do not change removed SPTEs. Only the thread that froze the SPTE
501 if (iter->old_spte == REMOVED_SPTE)
504 if (cmpxchg64(rcu_dereference(iter->sptep), iter->old_spte,
505 new_spte) != iter->old_spte)
508 handle_changed_spte(kvm, iter->as_id, iter->gfn, iter->old_spte,
509 new_spte, iter->level, true);
514 static inline bool tdp_mmu_zap_spte_atomic(struct kvm *kvm,
515 struct tdp_iter *iter)
518 * Freeze the SPTE by setting it to a special,
519 * non-present value. This will stop other threads from
520 * immediately installing a present entry in its place
521 * before the TLBs are flushed.
523 if (!tdp_mmu_set_spte_atomic(kvm, iter, REMOVED_SPTE))
526 kvm_flush_remote_tlbs_with_address(kvm, iter->gfn,
527 KVM_PAGES_PER_HPAGE(iter->level));
530 * No other thread can overwrite the removed SPTE as they
531 * must either wait on the MMU lock or use
532 * tdp_mmu_set_spte_atomic which will not overwrite the
533 * special removed SPTE value. No bookkeeping is needed
534 * here since the SPTE is going from non-present
537 WRITE_ONCE(*rcu_dereference(iter->sptep), 0);
544 * __tdp_mmu_set_spte - Set a TDP MMU SPTE and handle the associated bookkeeping
546 * @iter: a tdp_iter instance currently on the SPTE that should be set
547 * @new_spte: The value the SPTE should be set to
548 * @record_acc_track: Notify the MM subsystem of changes to the accessed state
549 * of the page. Should be set unless handling an MMU
550 * notifier for access tracking. Leaving record_acc_track
551 * unset in that case prevents page accesses from being
553 * @record_dirty_log: Record the page as dirty in the dirty bitmap if
554 * appropriate for the change being made. Should be set
555 * unless performing certain dirty logging operations.
556 * Leaving record_dirty_log unset in that case prevents page
557 * writes from being double counted.
559 static inline void __tdp_mmu_set_spte(struct kvm *kvm, struct tdp_iter *iter,
560 u64 new_spte, bool record_acc_track,
561 bool record_dirty_log)
563 lockdep_assert_held_write(&kvm->mmu_lock);
566 * No thread should be using this function to set SPTEs to the
567 * temporary removed SPTE value.
568 * If operating under the MMU lock in read mode, tdp_mmu_set_spte_atomic
569 * should be used. If operating under the MMU lock in write mode, the
570 * use of the removed SPTE should not be necessary.
572 WARN_ON(iter->old_spte == REMOVED_SPTE);
574 WRITE_ONCE(*rcu_dereference(iter->sptep), new_spte);
576 __handle_changed_spte(kvm, iter->as_id, iter->gfn, iter->old_spte,
577 new_spte, iter->level, false);
578 if (record_acc_track)
579 handle_changed_spte_acc_track(iter->old_spte, new_spte,
581 if (record_dirty_log)
582 handle_changed_spte_dirty_log(kvm, iter->as_id, iter->gfn,
583 iter->old_spte, new_spte,
587 static inline void tdp_mmu_set_spte(struct kvm *kvm, struct tdp_iter *iter,
590 __tdp_mmu_set_spte(kvm, iter, new_spte, true, true);
593 static inline void tdp_mmu_set_spte_no_acc_track(struct kvm *kvm,
594 struct tdp_iter *iter,
597 __tdp_mmu_set_spte(kvm, iter, new_spte, false, true);
600 static inline void tdp_mmu_set_spte_no_dirty_log(struct kvm *kvm,
601 struct tdp_iter *iter,
604 __tdp_mmu_set_spte(kvm, iter, new_spte, true, false);
607 #define tdp_root_for_each_pte(_iter, _root, _start, _end) \
608 for_each_tdp_pte(_iter, _root->spt, _root->role.level, _start, _end)
610 #define tdp_root_for_each_leaf_pte(_iter, _root, _start, _end) \
611 tdp_root_for_each_pte(_iter, _root, _start, _end) \
612 if (!is_shadow_present_pte(_iter.old_spte) || \
613 !is_last_spte(_iter.old_spte, _iter.level)) \
617 #define tdp_mmu_for_each_pte(_iter, _mmu, _start, _end) \
618 for_each_tdp_pte(_iter, __va(_mmu->root_hpa), \
619 _mmu->shadow_root_level, _start, _end)
622 * Yield if the MMU lock is contended or this thread needs to return control
625 * If this function should yield and flush is set, it will perform a remote
626 * TLB flush before yielding.
628 * If this function yields, it will also reset the tdp_iter's walk over the
629 * paging structure and the calling function should skip to the next
630 * iteration to allow the iterator to continue its traversal from the
631 * paging structure root.
633 * Return true if this function yielded and the iterator's traversal was reset.
634 * Return false if a yield was not needed.
636 static inline bool tdp_mmu_iter_cond_resched(struct kvm *kvm,
637 struct tdp_iter *iter, bool flush)
639 /* Ensure forward progress has been made before yielding. */
640 if (iter->next_last_level_gfn == iter->yielded_gfn)
643 if (need_resched() || rwlock_needbreak(&kvm->mmu_lock)) {
647 kvm_flush_remote_tlbs(kvm);
649 cond_resched_rwlock_write(&kvm->mmu_lock);
652 WARN_ON(iter->gfn > iter->next_last_level_gfn);
654 tdp_iter_restart(iter);
663 * Tears down the mappings for the range of gfns, [start, end), and frees the
664 * non-root pages mapping GFNs strictly within that range. Returns true if
665 * SPTEs have been cleared and a TLB flush is needed before releasing the
667 * If can_yield is true, will release the MMU lock and reschedule if the
668 * scheduler needs the CPU or there is contention on the MMU lock. If this
669 * function cannot yield, it will not release the MMU lock or reschedule and
670 * the caller must ensure it does not supply too large a GFN range, or the
671 * operation can cause a soft lockup. Note, in some use cases a flush may be
672 * required by prior actions. Ensure the pending flush is performed prior to
675 static bool zap_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root,
676 gfn_t start, gfn_t end, bool can_yield, bool flush)
678 struct tdp_iter iter;
682 tdp_root_for_each_pte(iter, root, start, end) {
684 tdp_mmu_iter_cond_resched(kvm, &iter, flush)) {
689 if (!is_shadow_present_pte(iter.old_spte))
693 * If this is a non-last-level SPTE that covers a larger range
694 * than should be zapped, continue, and zap the mappings at a
697 if ((iter.gfn < start ||
698 iter.gfn + KVM_PAGES_PER_HPAGE(iter.level) > end) &&
699 !is_last_spte(iter.old_spte, iter.level))
702 tdp_mmu_set_spte(kvm, &iter, 0);
711 * Tears down the mappings for the range of gfns, [start, end), and frees the
712 * non-root pages mapping GFNs strictly within that range. Returns true if
713 * SPTEs have been cleared and a TLB flush is needed before releasing the
716 bool __kvm_tdp_mmu_zap_gfn_range(struct kvm *kvm, gfn_t start, gfn_t end,
719 struct kvm_mmu_page *root;
722 for_each_tdp_mmu_root_yield_safe(kvm, root)
723 flush = zap_gfn_range(kvm, root, start, end, can_yield, flush);
728 void kvm_tdp_mmu_zap_all(struct kvm *kvm)
730 gfn_t max_gfn = 1ULL << (shadow_phys_bits - PAGE_SHIFT);
733 flush = kvm_tdp_mmu_zap_gfn_range(kvm, 0, max_gfn);
735 kvm_flush_remote_tlbs(kvm);
739 * Installs a last-level SPTE to handle a TDP page fault.
740 * (NPT/EPT violation/misconfiguration)
742 static int tdp_mmu_map_handle_target_level(struct kvm_vcpu *vcpu, int write,
744 struct tdp_iter *iter,
745 kvm_pfn_t pfn, bool prefault)
749 int make_spte_ret = 0;
751 if (unlikely(is_noslot_pfn(pfn)))
752 new_spte = make_mmio_spte(vcpu, iter->gfn, ACC_ALL);
754 make_spte_ret = make_spte(vcpu, ACC_ALL, iter->level, iter->gfn,
755 pfn, iter->old_spte, prefault, true,
756 map_writable, !shadow_accessed_mask,
759 if (new_spte == iter->old_spte)
760 ret = RET_PF_SPURIOUS;
761 else if (!tdp_mmu_set_spte_atomic(vcpu->kvm, iter, new_spte))
765 * If the page fault was caused by a write but the page is write
766 * protected, emulation is needed. If the emulation was skipped,
767 * the vCPU would have the same fault again.
769 if (make_spte_ret & SET_SPTE_WRITE_PROTECTED_PT) {
771 ret = RET_PF_EMULATE;
772 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
775 /* If a MMIO SPTE is installed, the MMIO will need to be emulated. */
776 if (unlikely(is_mmio_spte(new_spte))) {
777 trace_mark_mmio_spte(rcu_dereference(iter->sptep), iter->gfn,
779 ret = RET_PF_EMULATE;
781 trace_kvm_mmu_set_spte(iter->level, iter->gfn,
782 rcu_dereference(iter->sptep));
784 trace_kvm_mmu_set_spte(iter->level, iter->gfn,
785 rcu_dereference(iter->sptep));
787 vcpu->stat.pf_fixed++;
793 * Handle a TDP page fault (NPT/EPT violation/misconfiguration) by installing
794 * page tables and SPTEs to translate the faulting guest physical address.
796 int kvm_tdp_mmu_map(struct kvm_vcpu *vcpu, gpa_t gpa, u32 error_code,
797 int map_writable, int max_level, kvm_pfn_t pfn,
800 bool nx_huge_page_workaround_enabled = is_nx_huge_page_enabled();
801 bool write = error_code & PFERR_WRITE_MASK;
802 bool exec = error_code & PFERR_FETCH_MASK;
803 bool huge_page_disallowed = exec && nx_huge_page_workaround_enabled;
804 struct kvm_mmu *mmu = vcpu->arch.mmu;
805 struct tdp_iter iter;
806 struct kvm_mmu_page *sp;
810 gfn_t gfn = gpa >> PAGE_SHIFT;
814 if (WARN_ON(!VALID_PAGE(vcpu->arch.mmu->root_hpa)))
816 if (WARN_ON(!is_tdp_mmu_root(vcpu->kvm, vcpu->arch.mmu->root_hpa)))
819 level = kvm_mmu_hugepage_adjust(vcpu, gfn, max_level, &pfn,
820 huge_page_disallowed, &req_level);
822 trace_kvm_mmu_spte_requested(gpa, level, pfn);
826 tdp_mmu_for_each_pte(iter, mmu, gfn, gfn + 1) {
827 if (nx_huge_page_workaround_enabled)
828 disallowed_hugepage_adjust(iter.old_spte, gfn,
829 iter.level, &pfn, &level);
831 if (iter.level == level)
835 * If there is an SPTE mapping a large page at a higher level
836 * than the target, that SPTE must be cleared and replaced
837 * with a non-leaf SPTE.
839 if (is_shadow_present_pte(iter.old_spte) &&
840 is_large_pte(iter.old_spte)) {
841 if (!tdp_mmu_zap_spte_atomic(vcpu->kvm, &iter))
845 * The iter must explicitly re-read the spte here
846 * because the new value informs the !present
849 iter.old_spte = READ_ONCE(*rcu_dereference(iter.sptep));
852 if (!is_shadow_present_pte(iter.old_spte)) {
853 sp = alloc_tdp_mmu_page(vcpu, iter.gfn, iter.level);
856 new_spte = make_nonleaf_spte(child_pt,
857 !shadow_accessed_mask);
859 if (tdp_mmu_set_spte_atomic(vcpu->kvm, &iter,
861 tdp_mmu_link_page(vcpu->kvm, sp, true,
862 huge_page_disallowed &&
863 req_level >= iter.level);
865 trace_kvm_mmu_get_page(sp, true);
873 if (iter.level != level) {
878 ret = tdp_mmu_map_handle_target_level(vcpu, write, map_writable, &iter,
885 static __always_inline int
886 kvm_tdp_mmu_handle_hva_range(struct kvm *kvm,
890 int (*handler)(struct kvm *kvm,
891 struct kvm_memory_slot *slot,
892 struct kvm_mmu_page *root,
897 struct kvm_memslots *slots;
898 struct kvm_memory_slot *memslot;
899 struct kvm_mmu_page *root;
903 for_each_tdp_mmu_root_yield_safe(kvm, root) {
904 as_id = kvm_mmu_page_as_id(root);
905 slots = __kvm_memslots(kvm, as_id);
906 kvm_for_each_memslot(memslot, slots) {
907 unsigned long hva_start, hva_end;
908 gfn_t gfn_start, gfn_end;
910 hva_start = max(start, memslot->userspace_addr);
911 hva_end = min(end, memslot->userspace_addr +
912 (memslot->npages << PAGE_SHIFT));
913 if (hva_start >= hva_end)
916 * {gfn(page) | page intersects with [hva_start, hva_end)} =
917 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
919 gfn_start = hva_to_gfn_memslot(hva_start, memslot);
920 gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
922 ret |= handler(kvm, memslot, root, gfn_start,
930 static int zap_gfn_range_hva_wrapper(struct kvm *kvm,
931 struct kvm_memory_slot *slot,
932 struct kvm_mmu_page *root, gfn_t start,
933 gfn_t end, unsigned long unused)
935 return zap_gfn_range(kvm, root, start, end, false, false);
938 int kvm_tdp_mmu_zap_hva_range(struct kvm *kvm, unsigned long start,
941 return kvm_tdp_mmu_handle_hva_range(kvm, start, end, 0,
942 zap_gfn_range_hva_wrapper);
946 * Mark the SPTEs range of GFNs [start, end) unaccessed and return non-zero
947 * if any of the GFNs in the range have been accessed.
949 static int age_gfn_range(struct kvm *kvm, struct kvm_memory_slot *slot,
950 struct kvm_mmu_page *root, gfn_t start, gfn_t end,
951 unsigned long unused)
953 struct tdp_iter iter;
959 tdp_root_for_each_leaf_pte(iter, root, start, end) {
961 * If we have a non-accessed entry we don't need to change the
964 if (!is_accessed_spte(iter.old_spte))
967 new_spte = iter.old_spte;
969 if (spte_ad_enabled(new_spte)) {
970 clear_bit((ffs(shadow_accessed_mask) - 1),
971 (unsigned long *)&new_spte);
974 * Capture the dirty status of the page, so that it doesn't get
975 * lost when the SPTE is marked for access tracking.
977 if (is_writable_pte(new_spte))
978 kvm_set_pfn_dirty(spte_to_pfn(new_spte));
980 new_spte = mark_spte_for_access_track(new_spte);
982 new_spte &= ~shadow_dirty_mask;
984 tdp_mmu_set_spte_no_acc_track(kvm, &iter, new_spte);
987 trace_kvm_age_page(iter.gfn, iter.level, slot, young);
995 int kvm_tdp_mmu_age_hva_range(struct kvm *kvm, unsigned long start,
998 return kvm_tdp_mmu_handle_hva_range(kvm, start, end, 0,
1002 static int test_age_gfn(struct kvm *kvm, struct kvm_memory_slot *slot,
1003 struct kvm_mmu_page *root, gfn_t gfn, gfn_t unused,
1004 unsigned long unused2)
1006 struct tdp_iter iter;
1008 tdp_root_for_each_leaf_pte(iter, root, gfn, gfn + 1)
1009 if (is_accessed_spte(iter.old_spte))
1015 int kvm_tdp_mmu_test_age_hva(struct kvm *kvm, unsigned long hva)
1017 return kvm_tdp_mmu_handle_hva_range(kvm, hva, hva + 1, 0,
1022 * Handle the changed_pte MMU notifier for the TDP MMU.
1023 * data is a pointer to the new pte_t mapping the HVA specified by the MMU
1025 * Returns non-zero if a flush is needed before releasing the MMU lock.
1027 static int set_tdp_spte(struct kvm *kvm, struct kvm_memory_slot *slot,
1028 struct kvm_mmu_page *root, gfn_t gfn, gfn_t unused,
1031 struct tdp_iter iter;
1032 pte_t *ptep = (pte_t *)data;
1039 WARN_ON(pte_huge(*ptep));
1041 new_pfn = pte_pfn(*ptep);
1043 tdp_root_for_each_pte(iter, root, gfn, gfn + 1) {
1044 if (iter.level != PG_LEVEL_4K)
1047 if (!is_shadow_present_pte(iter.old_spte))
1050 tdp_mmu_set_spte(kvm, &iter, 0);
1052 kvm_flush_remote_tlbs_with_address(kvm, iter.gfn, 1);
1054 if (!pte_write(*ptep)) {
1055 new_spte = kvm_mmu_changed_pte_notifier_make_spte(
1056 iter.old_spte, new_pfn);
1058 tdp_mmu_set_spte(kvm, &iter, new_spte);
1065 kvm_flush_remote_tlbs_with_address(kvm, gfn, 1);
1072 int kvm_tdp_mmu_set_spte_hva(struct kvm *kvm, unsigned long address,
1075 return kvm_tdp_mmu_handle_hva_range(kvm, address, address + 1,
1076 (unsigned long)host_ptep,
1081 * Remove write access from all the SPTEs mapping GFNs [start, end). If
1082 * skip_4k is set, SPTEs that map 4k pages, will not be write-protected.
1083 * Returns true if an SPTE has been changed and the TLBs need to be flushed.
1085 static bool wrprot_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root,
1086 gfn_t start, gfn_t end, int min_level)
1088 struct tdp_iter iter;
1090 bool spte_set = false;
1094 BUG_ON(min_level > KVM_MAX_HUGEPAGE_LEVEL);
1096 for_each_tdp_pte_min_level(iter, root->spt, root->role.level,
1097 min_level, start, end) {
1098 if (tdp_mmu_iter_cond_resched(kvm, &iter, false))
1101 if (!is_shadow_present_pte(iter.old_spte) ||
1102 !is_last_spte(iter.old_spte, iter.level) ||
1103 !(iter.old_spte & PT_WRITABLE_MASK))
1106 new_spte = iter.old_spte & ~PT_WRITABLE_MASK;
1108 tdp_mmu_set_spte_no_dirty_log(kvm, &iter, new_spte);
1117 * Remove write access from all the SPTEs mapping GFNs in the memslot. Will
1118 * only affect leaf SPTEs down to min_level.
1119 * Returns true if an SPTE has been changed and the TLBs need to be flushed.
1121 bool kvm_tdp_mmu_wrprot_slot(struct kvm *kvm, struct kvm_memory_slot *slot,
1124 struct kvm_mmu_page *root;
1126 bool spte_set = false;
1128 for_each_tdp_mmu_root_yield_safe(kvm, root) {
1129 root_as_id = kvm_mmu_page_as_id(root);
1130 if (root_as_id != slot->as_id)
1133 spte_set |= wrprot_gfn_range(kvm, root, slot->base_gfn,
1134 slot->base_gfn + slot->npages, min_level);
1141 * Clear the dirty status of all the SPTEs mapping GFNs in the memslot. If
1142 * AD bits are enabled, this will involve clearing the dirty bit on each SPTE.
1143 * If AD bits are not enabled, this will require clearing the writable bit on
1144 * each SPTE. Returns true if an SPTE has been changed and the TLBs need to
1147 static bool clear_dirty_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root,
1148 gfn_t start, gfn_t end)
1150 struct tdp_iter iter;
1152 bool spte_set = false;
1156 tdp_root_for_each_leaf_pte(iter, root, start, end) {
1157 if (tdp_mmu_iter_cond_resched(kvm, &iter, false))
1160 if (spte_ad_need_write_protect(iter.old_spte)) {
1161 if (is_writable_pte(iter.old_spte))
1162 new_spte = iter.old_spte & ~PT_WRITABLE_MASK;
1166 if (iter.old_spte & shadow_dirty_mask)
1167 new_spte = iter.old_spte & ~shadow_dirty_mask;
1172 tdp_mmu_set_spte_no_dirty_log(kvm, &iter, new_spte);
1181 * Clear the dirty status of all the SPTEs mapping GFNs in the memslot. If
1182 * AD bits are enabled, this will involve clearing the dirty bit on each SPTE.
1183 * If AD bits are not enabled, this will require clearing the writable bit on
1184 * each SPTE. Returns true if an SPTE has been changed and the TLBs need to
1187 bool kvm_tdp_mmu_clear_dirty_slot(struct kvm *kvm, struct kvm_memory_slot *slot)
1189 struct kvm_mmu_page *root;
1191 bool spte_set = false;
1193 for_each_tdp_mmu_root_yield_safe(kvm, root) {
1194 root_as_id = kvm_mmu_page_as_id(root);
1195 if (root_as_id != slot->as_id)
1198 spte_set |= clear_dirty_gfn_range(kvm, root, slot->base_gfn,
1199 slot->base_gfn + slot->npages);
1206 * Clears the dirty status of all the 4k SPTEs mapping GFNs for which a bit is
1207 * set in mask, starting at gfn. The given memslot is expected to contain all
1208 * the GFNs represented by set bits in the mask. If AD bits are enabled,
1209 * clearing the dirty status will involve clearing the dirty bit on each SPTE
1210 * or, if AD bits are not enabled, clearing the writable bit on each SPTE.
1212 static void clear_dirty_pt_masked(struct kvm *kvm, struct kvm_mmu_page *root,
1213 gfn_t gfn, unsigned long mask, bool wrprot)
1215 struct tdp_iter iter;
1220 tdp_root_for_each_leaf_pte(iter, root, gfn + __ffs(mask),
1221 gfn + BITS_PER_LONG) {
1225 if (iter.level > PG_LEVEL_4K ||
1226 !(mask & (1UL << (iter.gfn - gfn))))
1229 mask &= ~(1UL << (iter.gfn - gfn));
1231 if (wrprot || spte_ad_need_write_protect(iter.old_spte)) {
1232 if (is_writable_pte(iter.old_spte))
1233 new_spte = iter.old_spte & ~PT_WRITABLE_MASK;
1237 if (iter.old_spte & shadow_dirty_mask)
1238 new_spte = iter.old_spte & ~shadow_dirty_mask;
1243 tdp_mmu_set_spte_no_dirty_log(kvm, &iter, new_spte);
1250 * Clears the dirty status of all the 4k SPTEs mapping GFNs for which a bit is
1251 * set in mask, starting at gfn. The given memslot is expected to contain all
1252 * the GFNs represented by set bits in the mask. If AD bits are enabled,
1253 * clearing the dirty status will involve clearing the dirty bit on each SPTE
1254 * or, if AD bits are not enabled, clearing the writable bit on each SPTE.
1256 void kvm_tdp_mmu_clear_dirty_pt_masked(struct kvm *kvm,
1257 struct kvm_memory_slot *slot,
1258 gfn_t gfn, unsigned long mask,
1261 struct kvm_mmu_page *root;
1264 lockdep_assert_held_write(&kvm->mmu_lock);
1265 for_each_tdp_mmu_root(kvm, root) {
1266 root_as_id = kvm_mmu_page_as_id(root);
1267 if (root_as_id != slot->as_id)
1270 clear_dirty_pt_masked(kvm, root, gfn, mask, wrprot);
1275 * Clear leaf entries which could be replaced by large mappings, for
1276 * GFNs within the slot.
1278 static void zap_collapsible_spte_range(struct kvm *kvm,
1279 struct kvm_mmu_page *root,
1280 struct kvm_memory_slot *slot)
1282 gfn_t start = slot->base_gfn;
1283 gfn_t end = start + slot->npages;
1284 struct tdp_iter iter;
1286 bool spte_set = false;
1290 tdp_root_for_each_pte(iter, root, start, end) {
1291 if (tdp_mmu_iter_cond_resched(kvm, &iter, spte_set)) {
1296 if (!is_shadow_present_pte(iter.old_spte) ||
1297 !is_last_spte(iter.old_spte, iter.level))
1300 pfn = spte_to_pfn(iter.old_spte);
1301 if (kvm_is_reserved_pfn(pfn) ||
1302 iter.level >= kvm_mmu_max_mapping_level(kvm, slot, iter.gfn,
1306 tdp_mmu_set_spte(kvm, &iter, 0);
1313 kvm_flush_remote_tlbs(kvm);
1317 * Clear non-leaf entries (and free associated page tables) which could
1318 * be replaced by large mappings, for GFNs within the slot.
1320 void kvm_tdp_mmu_zap_collapsible_sptes(struct kvm *kvm,
1321 struct kvm_memory_slot *slot)
1323 struct kvm_mmu_page *root;
1326 for_each_tdp_mmu_root_yield_safe(kvm, root) {
1327 root_as_id = kvm_mmu_page_as_id(root);
1328 if (root_as_id != slot->as_id)
1331 zap_collapsible_spte_range(kvm, root, slot);
1336 * Removes write access on the last level SPTE mapping this GFN and unsets the
1337 * SPTE_MMU_WRITABLE bit to ensure future writes continue to be intercepted.
1338 * Returns true if an SPTE was set and a TLB flush is needed.
1340 static bool write_protect_gfn(struct kvm *kvm, struct kvm_mmu_page *root,
1343 struct tdp_iter iter;
1345 bool spte_set = false;
1349 tdp_root_for_each_leaf_pte(iter, root, gfn, gfn + 1) {
1350 if (!is_writable_pte(iter.old_spte))
1353 new_spte = iter.old_spte &
1354 ~(PT_WRITABLE_MASK | SPTE_MMU_WRITEABLE);
1356 tdp_mmu_set_spte(kvm, &iter, new_spte);
1366 * Removes write access on the last level SPTE mapping this GFN and unsets the
1367 * SPTE_MMU_WRITABLE bit to ensure future writes continue to be intercepted.
1368 * Returns true if an SPTE was set and a TLB flush is needed.
1370 bool kvm_tdp_mmu_write_protect_gfn(struct kvm *kvm,
1371 struct kvm_memory_slot *slot, gfn_t gfn)
1373 struct kvm_mmu_page *root;
1375 bool spte_set = false;
1377 lockdep_assert_held_write(&kvm->mmu_lock);
1378 for_each_tdp_mmu_root(kvm, root) {
1379 root_as_id = kvm_mmu_page_as_id(root);
1380 if (root_as_id != slot->as_id)
1383 spte_set |= write_protect_gfn(kvm, root, gfn);
1389 * Return the level of the lowest level SPTE added to sptes.
1390 * That SPTE may be non-present.
1392 int kvm_tdp_mmu_get_walk(struct kvm_vcpu *vcpu, u64 addr, u64 *sptes,
1395 struct tdp_iter iter;
1396 struct kvm_mmu *mmu = vcpu->arch.mmu;
1397 gfn_t gfn = addr >> PAGE_SHIFT;
1400 *root_level = vcpu->arch.mmu->shadow_root_level;
1404 tdp_mmu_for_each_pte(iter, mmu, gfn, gfn + 1) {
1406 sptes[leaf] = iter.old_spte;