1 // SPDX-License-Identifier: GPL-2.0
4 #include "mmu_internal.h"
11 static bool __read_mostly tdp_mmu_enabled = false;
12 module_param_named(tdp_mmu, tdp_mmu_enabled, bool, 0644);
15 static bool is_tdp_mmu_enabled(void)
18 return tdp_enabled && READ_ONCE(tdp_mmu_enabled);
21 #endif /* CONFIG_X86_64 */
24 /* Initializes the TDP MMU for the VM, if enabled. */
25 void kvm_mmu_init_tdp_mmu(struct kvm *kvm)
27 if (!is_tdp_mmu_enabled())
30 /* This should not be changed for the lifetime of the VM. */
31 kvm->arch.tdp_mmu_enabled = true;
33 INIT_LIST_HEAD(&kvm->arch.tdp_mmu_roots);
34 INIT_LIST_HEAD(&kvm->arch.tdp_mmu_pages);
37 void kvm_mmu_uninit_tdp_mmu(struct kvm *kvm)
39 if (!kvm->arch.tdp_mmu_enabled)
42 WARN_ON(!list_empty(&kvm->arch.tdp_mmu_roots));
45 #define for_each_tdp_mmu_root(_kvm, _root) \
46 list_for_each_entry(_root, &_kvm->arch.tdp_mmu_roots, link)
48 bool is_tdp_mmu_root(struct kvm *kvm, hpa_t hpa)
50 struct kvm_mmu_page *sp;
52 if (!kvm->arch.tdp_mmu_enabled)
54 if (WARN_ON(!VALID_PAGE(hpa)))
57 sp = to_shadow_page(hpa);
61 return sp->tdp_mmu_page && sp->root_count;
64 static bool zap_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root,
65 gfn_t start, gfn_t end, bool can_yield);
67 void kvm_tdp_mmu_free_root(struct kvm *kvm, struct kvm_mmu_page *root)
69 gfn_t max_gfn = 1ULL << (boot_cpu_data.x86_phys_bits - PAGE_SHIFT);
71 lockdep_assert_held(&kvm->mmu_lock);
73 WARN_ON(root->root_count);
74 WARN_ON(!root->tdp_mmu_page);
76 list_del(&root->link);
78 zap_gfn_range(kvm, root, 0, max_gfn, false);
80 free_page((unsigned long)root->spt);
81 kmem_cache_free(mmu_page_header_cache, root);
84 static union kvm_mmu_page_role page_role_for_level(struct kvm_vcpu *vcpu,
87 union kvm_mmu_page_role role;
89 role = vcpu->arch.mmu->mmu_role.base;
92 role.gpte_is_8_bytes = true;
93 role.access = ACC_ALL;
98 static struct kvm_mmu_page *alloc_tdp_mmu_page(struct kvm_vcpu *vcpu, gfn_t gfn,
101 struct kvm_mmu_page *sp;
103 sp = kvm_mmu_memory_cache_alloc(&vcpu->arch.mmu_page_header_cache);
104 sp->spt = kvm_mmu_memory_cache_alloc(&vcpu->arch.mmu_shadow_page_cache);
105 set_page_private(virt_to_page(sp->spt), (unsigned long)sp);
107 sp->role.word = page_role_for_level(vcpu, level).word;
109 sp->tdp_mmu_page = true;
114 static struct kvm_mmu_page *get_tdp_mmu_vcpu_root(struct kvm_vcpu *vcpu)
116 union kvm_mmu_page_role role;
117 struct kvm *kvm = vcpu->kvm;
118 struct kvm_mmu_page *root;
120 role = page_role_for_level(vcpu, vcpu->arch.mmu->shadow_root_level);
122 spin_lock(&kvm->mmu_lock);
124 /* Check for an existing root before allocating a new one. */
125 for_each_tdp_mmu_root(kvm, root) {
126 if (root->role.word == role.word) {
127 kvm_mmu_get_root(kvm, root);
128 spin_unlock(&kvm->mmu_lock);
133 root = alloc_tdp_mmu_page(vcpu, 0, vcpu->arch.mmu->shadow_root_level);
134 root->root_count = 1;
136 list_add(&root->link, &kvm->arch.tdp_mmu_roots);
138 spin_unlock(&kvm->mmu_lock);
143 hpa_t kvm_tdp_mmu_get_vcpu_root_hpa(struct kvm_vcpu *vcpu)
145 struct kvm_mmu_page *root;
147 root = get_tdp_mmu_vcpu_root(vcpu);
151 return __pa(root->spt);
154 static void handle_changed_spte(struct kvm *kvm, int as_id, gfn_t gfn,
155 u64 old_spte, u64 new_spte, int level);
157 static int kvm_mmu_page_as_id(struct kvm_mmu_page *sp)
159 return sp->role.smm ? 1 : 0;
162 static void handle_changed_spte_acc_track(u64 old_spte, u64 new_spte, int level)
164 bool pfn_changed = spte_to_pfn(old_spte) != spte_to_pfn(new_spte);
166 if (!is_shadow_present_pte(old_spte) || !is_last_spte(old_spte, level))
169 if (is_accessed_spte(old_spte) &&
170 (!is_accessed_spte(new_spte) || pfn_changed))
171 kvm_set_pfn_accessed(spte_to_pfn(old_spte));
174 static void handle_changed_spte_dirty_log(struct kvm *kvm, int as_id, gfn_t gfn,
175 u64 old_spte, u64 new_spte, int level)
178 struct kvm_memory_slot *slot;
180 if (level > PG_LEVEL_4K)
183 pfn_changed = spte_to_pfn(old_spte) != spte_to_pfn(new_spte);
185 if ((!is_writable_pte(old_spte) || pfn_changed) &&
186 is_writable_pte(new_spte)) {
187 slot = __gfn_to_memslot(__kvm_memslots(kvm, as_id), gfn);
188 mark_page_dirty_in_slot(slot, gfn);
193 * handle_changed_spte - handle bookkeeping associated with an SPTE change
195 * @as_id: the address space of the paging structure the SPTE was a part of
196 * @gfn: the base GFN that was mapped by the SPTE
197 * @old_spte: The value of the SPTE before the change
198 * @new_spte: The value of the SPTE after the change
199 * @level: the level of the PT the SPTE is part of in the paging structure
201 * Handle bookkeeping that might result from the modification of a SPTE.
202 * This function must be called for all TDP SPTE modifications.
204 static void __handle_changed_spte(struct kvm *kvm, int as_id, gfn_t gfn,
205 u64 old_spte, u64 new_spte, int level)
207 bool was_present = is_shadow_present_pte(old_spte);
208 bool is_present = is_shadow_present_pte(new_spte);
209 bool was_leaf = was_present && is_last_spte(old_spte, level);
210 bool is_leaf = is_present && is_last_spte(new_spte, level);
211 bool pfn_changed = spte_to_pfn(old_spte) != spte_to_pfn(new_spte);
213 struct kvm_mmu_page *sp;
217 WARN_ON(level > PT64_ROOT_MAX_LEVEL);
218 WARN_ON(level < PG_LEVEL_4K);
219 WARN_ON(gfn & (KVM_PAGES_PER_HPAGE(level) - 1));
222 * If this warning were to trigger it would indicate that there was a
223 * missing MMU notifier or a race with some notifier handler.
224 * A present, leaf SPTE should never be directly replaced with another
225 * present leaf SPTE pointing to a differnt PFN. A notifier handler
226 * should be zapping the SPTE before the main MM's page table is
227 * changed, or the SPTE should be zeroed, and the TLBs flushed by the
228 * thread before replacement.
230 if (was_leaf && is_leaf && pfn_changed) {
231 pr_err("Invalid SPTE change: cannot replace a present leaf\n"
232 "SPTE with another present leaf SPTE mapping a\n"
234 "as_id: %d gfn: %llx old_spte: %llx new_spte: %llx level: %d",
235 as_id, gfn, old_spte, new_spte, level);
238 * Crash the host to prevent error propagation and guest data
244 if (old_spte == new_spte)
248 * The only times a SPTE should be changed from a non-present to
249 * non-present state is when an MMIO entry is installed/modified/
250 * removed. In that case, there is nothing to do here.
252 if (!was_present && !is_present) {
254 * If this change does not involve a MMIO SPTE, it is
255 * unexpected. Log the change, though it should not impact the
256 * guest since both the former and current SPTEs are nonpresent.
258 if (WARN_ON(!is_mmio_spte(old_spte) && !is_mmio_spte(new_spte)))
259 pr_err("Unexpected SPTE change! Nonpresent SPTEs\n"
260 "should not be replaced with another,\n"
261 "different nonpresent SPTE, unless one or both\n"
263 "as_id: %d gfn: %llx old_spte: %llx new_spte: %llx level: %d",
264 as_id, gfn, old_spte, new_spte, level);
269 if (was_leaf && is_dirty_spte(old_spte) &&
270 (!is_dirty_spte(new_spte) || pfn_changed))
271 kvm_set_pfn_dirty(spte_to_pfn(old_spte));
274 * Recursively handle child PTs if the change removed a subtree from
275 * the paging structure.
277 if (was_present && !was_leaf && (pfn_changed || !is_present)) {
278 pt = spte_to_child_pt(old_spte, level);
279 sp = sptep_to_sp(pt);
283 if (sp->lpage_disallowed)
284 unaccount_huge_nx_page(kvm, sp);
286 for (i = 0; i < PT64_ENT_PER_PAGE; i++) {
287 old_child_spte = READ_ONCE(*(pt + i));
288 WRITE_ONCE(*(pt + i), 0);
289 handle_changed_spte(kvm, as_id,
290 gfn + (i * KVM_PAGES_PER_HPAGE(level - 1)),
291 old_child_spte, 0, level - 1);
294 kvm_flush_remote_tlbs_with_address(kvm, gfn,
295 KVM_PAGES_PER_HPAGE(level));
297 free_page((unsigned long)pt);
298 kmem_cache_free(mmu_page_header_cache, sp);
302 static void handle_changed_spte(struct kvm *kvm, int as_id, gfn_t gfn,
303 u64 old_spte, u64 new_spte, int level)
305 __handle_changed_spte(kvm, as_id, gfn, old_spte, new_spte, level);
306 handle_changed_spte_acc_track(old_spte, new_spte, level);
307 handle_changed_spte_dirty_log(kvm, as_id, gfn, old_spte,
311 static inline void __tdp_mmu_set_spte(struct kvm *kvm, struct tdp_iter *iter,
312 u64 new_spte, bool record_acc_track,
313 bool record_dirty_log)
315 u64 *root_pt = tdp_iter_root_pt(iter);
316 struct kvm_mmu_page *root = sptep_to_sp(root_pt);
317 int as_id = kvm_mmu_page_as_id(root);
319 WRITE_ONCE(*iter->sptep, new_spte);
321 __handle_changed_spte(kvm, as_id, iter->gfn, iter->old_spte, new_spte,
323 if (record_acc_track)
324 handle_changed_spte_acc_track(iter->old_spte, new_spte,
326 if (record_dirty_log)
327 handle_changed_spte_dirty_log(kvm, as_id, iter->gfn,
328 iter->old_spte, new_spte,
332 static inline void tdp_mmu_set_spte(struct kvm *kvm, struct tdp_iter *iter,
335 __tdp_mmu_set_spte(kvm, iter, new_spte, true, true);
338 static inline void tdp_mmu_set_spte_no_acc_track(struct kvm *kvm,
339 struct tdp_iter *iter,
342 __tdp_mmu_set_spte(kvm, iter, new_spte, false, true);
345 static inline void tdp_mmu_set_spte_no_dirty_log(struct kvm *kvm,
346 struct tdp_iter *iter,
349 __tdp_mmu_set_spte(kvm, iter, new_spte, true, false);
352 #define tdp_root_for_each_pte(_iter, _root, _start, _end) \
353 for_each_tdp_pte(_iter, _root->spt, _root->role.level, _start, _end)
355 #define tdp_root_for_each_leaf_pte(_iter, _root, _start, _end) \
356 tdp_root_for_each_pte(_iter, _root, _start, _end) \
357 if (!is_shadow_present_pte(_iter.old_spte) || \
358 !is_last_spte(_iter.old_spte, _iter.level)) \
362 #define tdp_mmu_for_each_pte(_iter, _mmu, _start, _end) \
363 for_each_tdp_pte(_iter, __va(_mmu->root_hpa), \
364 _mmu->shadow_root_level, _start, _end)
367 * Flush the TLB if the process should drop kvm->mmu_lock.
368 * Return whether the caller still needs to flush the tlb.
370 static bool tdp_mmu_iter_flush_cond_resched(struct kvm *kvm, struct tdp_iter *iter)
372 if (need_resched() || spin_needbreak(&kvm->mmu_lock)) {
373 kvm_flush_remote_tlbs(kvm);
374 cond_resched_lock(&kvm->mmu_lock);
375 tdp_iter_refresh_walk(iter);
382 static void tdp_mmu_iter_cond_resched(struct kvm *kvm, struct tdp_iter *iter)
384 if (need_resched() || spin_needbreak(&kvm->mmu_lock)) {
385 cond_resched_lock(&kvm->mmu_lock);
386 tdp_iter_refresh_walk(iter);
391 * Tears down the mappings for the range of gfns, [start, end), and frees the
392 * non-root pages mapping GFNs strictly within that range. Returns true if
393 * SPTEs have been cleared and a TLB flush is needed before releasing the
395 * If can_yield is true, will release the MMU lock and reschedule if the
396 * scheduler needs the CPU or there is contention on the MMU lock. If this
397 * function cannot yield, it will not release the MMU lock or reschedule and
398 * the caller must ensure it does not supply too large a GFN range, or the
399 * operation can cause a soft lockup.
401 static bool zap_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root,
402 gfn_t start, gfn_t end, bool can_yield)
404 struct tdp_iter iter;
405 bool flush_needed = false;
407 tdp_root_for_each_pte(iter, root, start, end) {
408 if (!is_shadow_present_pte(iter.old_spte))
412 * If this is a non-last-level SPTE that covers a larger range
413 * than should be zapped, continue, and zap the mappings at a
416 if ((iter.gfn < start ||
417 iter.gfn + KVM_PAGES_PER_HPAGE(iter.level) > end) &&
418 !is_last_spte(iter.old_spte, iter.level))
421 tdp_mmu_set_spte(kvm, &iter, 0);
424 flush_needed = tdp_mmu_iter_flush_cond_resched(kvm, &iter);
432 * Tears down the mappings for the range of gfns, [start, end), and frees the
433 * non-root pages mapping GFNs strictly within that range. Returns true if
434 * SPTEs have been cleared and a TLB flush is needed before releasing the
437 bool kvm_tdp_mmu_zap_gfn_range(struct kvm *kvm, gfn_t start, gfn_t end)
439 struct kvm_mmu_page *root;
442 for_each_tdp_mmu_root(kvm, root) {
444 * Take a reference on the root so that it cannot be freed if
445 * this thread releases the MMU lock and yields in this loop.
447 kvm_mmu_get_root(kvm, root);
449 flush |= zap_gfn_range(kvm, root, start, end, true);
451 kvm_mmu_put_root(kvm, root);
457 void kvm_tdp_mmu_zap_all(struct kvm *kvm)
459 gfn_t max_gfn = 1ULL << (boot_cpu_data.x86_phys_bits - PAGE_SHIFT);
462 flush = kvm_tdp_mmu_zap_gfn_range(kvm, 0, max_gfn);
464 kvm_flush_remote_tlbs(kvm);
468 * Installs a last-level SPTE to handle a TDP page fault.
469 * (NPT/EPT violation/misconfiguration)
471 static int tdp_mmu_map_handle_target_level(struct kvm_vcpu *vcpu, int write,
473 struct tdp_iter *iter,
474 kvm_pfn_t pfn, bool prefault)
478 int make_spte_ret = 0;
480 if (unlikely(is_noslot_pfn(pfn))) {
481 new_spte = make_mmio_spte(vcpu, iter->gfn, ACC_ALL);
482 trace_mark_mmio_spte(iter->sptep, iter->gfn, new_spte);
484 make_spte_ret = make_spte(vcpu, ACC_ALL, iter->level, iter->gfn,
485 pfn, iter->old_spte, prefault, true,
486 map_writable, !shadow_accessed_mask,
489 if (new_spte == iter->old_spte)
490 ret = RET_PF_SPURIOUS;
492 tdp_mmu_set_spte(vcpu->kvm, iter, new_spte);
495 * If the page fault was caused by a write but the page is write
496 * protected, emulation is needed. If the emulation was skipped,
497 * the vCPU would have the same fault again.
499 if (make_spte_ret & SET_SPTE_WRITE_PROTECTED_PT) {
501 ret = RET_PF_EMULATE;
502 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
505 /* If a MMIO SPTE is installed, the MMIO will need to be emulated. */
506 if (unlikely(is_mmio_spte(new_spte)))
507 ret = RET_PF_EMULATE;
509 trace_kvm_mmu_set_spte(iter->level, iter->gfn, iter->sptep);
511 vcpu->stat.pf_fixed++;
517 * Handle a TDP page fault (NPT/EPT violation/misconfiguration) by installing
518 * page tables and SPTEs to translate the faulting guest physical address.
520 int kvm_tdp_mmu_map(struct kvm_vcpu *vcpu, gpa_t gpa, u32 error_code,
521 int map_writable, int max_level, kvm_pfn_t pfn,
524 bool nx_huge_page_workaround_enabled = is_nx_huge_page_enabled();
525 bool write = error_code & PFERR_WRITE_MASK;
526 bool exec = error_code & PFERR_FETCH_MASK;
527 bool huge_page_disallowed = exec && nx_huge_page_workaround_enabled;
528 struct kvm_mmu *mmu = vcpu->arch.mmu;
529 struct tdp_iter iter;
530 struct kvm_mmu_page *sp;
534 gfn_t gfn = gpa >> PAGE_SHIFT;
538 if (WARN_ON(!VALID_PAGE(vcpu->arch.mmu->root_hpa)))
540 if (WARN_ON(!is_tdp_mmu_root(vcpu->kvm, vcpu->arch.mmu->root_hpa)))
543 level = kvm_mmu_hugepage_adjust(vcpu, gfn, max_level, &pfn,
544 huge_page_disallowed, &req_level);
546 trace_kvm_mmu_spte_requested(gpa, level, pfn);
547 tdp_mmu_for_each_pte(iter, mmu, gfn, gfn + 1) {
548 if (nx_huge_page_workaround_enabled)
549 disallowed_hugepage_adjust(iter.old_spte, gfn,
550 iter.level, &pfn, &level);
552 if (iter.level == level)
556 * If there is an SPTE mapping a large page at a higher level
557 * than the target, that SPTE must be cleared and replaced
558 * with a non-leaf SPTE.
560 if (is_shadow_present_pte(iter.old_spte) &&
561 is_large_pte(iter.old_spte)) {
562 tdp_mmu_set_spte(vcpu->kvm, &iter, 0);
564 kvm_flush_remote_tlbs_with_address(vcpu->kvm, iter.gfn,
565 KVM_PAGES_PER_HPAGE(iter.level));
568 * The iter must explicitly re-read the spte here
569 * because the new value informs the !present
572 iter.old_spte = READ_ONCE(*iter.sptep);
575 if (!is_shadow_present_pte(iter.old_spte)) {
576 sp = alloc_tdp_mmu_page(vcpu, iter.gfn, iter.level);
577 list_add(&sp->link, &vcpu->kvm->arch.tdp_mmu_pages);
579 clear_page(child_pt);
580 new_spte = make_nonleaf_spte(child_pt,
581 !shadow_accessed_mask);
583 trace_kvm_mmu_get_page(sp, true);
584 if (huge_page_disallowed && req_level >= iter.level)
585 account_huge_nx_page(vcpu->kvm, sp);
587 tdp_mmu_set_spte(vcpu->kvm, &iter, new_spte);
591 if (WARN_ON(iter.level != level))
594 ret = tdp_mmu_map_handle_target_level(vcpu, write, map_writable, &iter,
600 static int kvm_tdp_mmu_handle_hva_range(struct kvm *kvm, unsigned long start,
601 unsigned long end, unsigned long data,
602 int (*handler)(struct kvm *kvm, struct kvm_memory_slot *slot,
603 struct kvm_mmu_page *root, gfn_t start,
604 gfn_t end, unsigned long data))
606 struct kvm_memslots *slots;
607 struct kvm_memory_slot *memslot;
608 struct kvm_mmu_page *root;
612 for_each_tdp_mmu_root(kvm, root) {
614 * Take a reference on the root so that it cannot be freed if
615 * this thread releases the MMU lock and yields in this loop.
617 kvm_mmu_get_root(kvm, root);
619 as_id = kvm_mmu_page_as_id(root);
620 slots = __kvm_memslots(kvm, as_id);
621 kvm_for_each_memslot(memslot, slots) {
622 unsigned long hva_start, hva_end;
623 gfn_t gfn_start, gfn_end;
625 hva_start = max(start, memslot->userspace_addr);
626 hva_end = min(end, memslot->userspace_addr +
627 (memslot->npages << PAGE_SHIFT));
628 if (hva_start >= hva_end)
631 * {gfn(page) | page intersects with [hva_start, hva_end)} =
632 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
634 gfn_start = hva_to_gfn_memslot(hva_start, memslot);
635 gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
637 ret |= handler(kvm, memslot, root, gfn_start,
641 kvm_mmu_put_root(kvm, root);
647 static int zap_gfn_range_hva_wrapper(struct kvm *kvm,
648 struct kvm_memory_slot *slot,
649 struct kvm_mmu_page *root, gfn_t start,
650 gfn_t end, unsigned long unused)
652 return zap_gfn_range(kvm, root, start, end, false);
655 int kvm_tdp_mmu_zap_hva_range(struct kvm *kvm, unsigned long start,
658 return kvm_tdp_mmu_handle_hva_range(kvm, start, end, 0,
659 zap_gfn_range_hva_wrapper);
663 * Mark the SPTEs range of GFNs [start, end) unaccessed and return non-zero
664 * if any of the GFNs in the range have been accessed.
666 static int age_gfn_range(struct kvm *kvm, struct kvm_memory_slot *slot,
667 struct kvm_mmu_page *root, gfn_t start, gfn_t end,
668 unsigned long unused)
670 struct tdp_iter iter;
674 tdp_root_for_each_leaf_pte(iter, root, start, end) {
676 * If we have a non-accessed entry we don't need to change the
679 if (!is_accessed_spte(iter.old_spte))
682 new_spte = iter.old_spte;
684 if (spte_ad_enabled(new_spte)) {
685 clear_bit((ffs(shadow_accessed_mask) - 1),
686 (unsigned long *)&new_spte);
689 * Capture the dirty status of the page, so that it doesn't get
690 * lost when the SPTE is marked for access tracking.
692 if (is_writable_pte(new_spte))
693 kvm_set_pfn_dirty(spte_to_pfn(new_spte));
695 new_spte = mark_spte_for_access_track(new_spte);
697 new_spte &= ~shadow_dirty_mask;
699 tdp_mmu_set_spte_no_acc_track(kvm, &iter, new_spte);
706 int kvm_tdp_mmu_age_hva_range(struct kvm *kvm, unsigned long start,
709 return kvm_tdp_mmu_handle_hva_range(kvm, start, end, 0,
713 static int test_age_gfn(struct kvm *kvm, struct kvm_memory_slot *slot,
714 struct kvm_mmu_page *root, gfn_t gfn, gfn_t unused,
715 unsigned long unused2)
717 struct tdp_iter iter;
719 tdp_root_for_each_leaf_pte(iter, root, gfn, gfn + 1)
720 if (is_accessed_spte(iter.old_spte))
726 int kvm_tdp_mmu_test_age_hva(struct kvm *kvm, unsigned long hva)
728 return kvm_tdp_mmu_handle_hva_range(kvm, hva, hva + 1, 0,
733 * Handle the changed_pte MMU notifier for the TDP MMU.
734 * data is a pointer to the new pte_t mapping the HVA specified by the MMU
736 * Returns non-zero if a flush is needed before releasing the MMU lock.
738 static int set_tdp_spte(struct kvm *kvm, struct kvm_memory_slot *slot,
739 struct kvm_mmu_page *root, gfn_t gfn, gfn_t unused,
742 struct tdp_iter iter;
743 pte_t *ptep = (pte_t *)data;
748 WARN_ON(pte_huge(*ptep));
750 new_pfn = pte_pfn(*ptep);
752 tdp_root_for_each_pte(iter, root, gfn, gfn + 1) {
753 if (iter.level != PG_LEVEL_4K)
756 if (!is_shadow_present_pte(iter.old_spte))
759 tdp_mmu_set_spte(kvm, &iter, 0);
761 kvm_flush_remote_tlbs_with_address(kvm, iter.gfn, 1);
763 if (!pte_write(*ptep)) {
764 new_spte = kvm_mmu_changed_pte_notifier_make_spte(
765 iter.old_spte, new_pfn);
767 tdp_mmu_set_spte(kvm, &iter, new_spte);
774 kvm_flush_remote_tlbs_with_address(kvm, gfn, 1);
779 int kvm_tdp_mmu_set_spte_hva(struct kvm *kvm, unsigned long address,
782 return kvm_tdp_mmu_handle_hva_range(kvm, address, address + 1,
783 (unsigned long)host_ptep,
788 * Remove write access from all the SPTEs mapping GFNs [start, end). If
789 * skip_4k is set, SPTEs that map 4k pages, will not be write-protected.
790 * Returns true if an SPTE has been changed and the TLBs need to be flushed.
792 static bool wrprot_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root,
793 gfn_t start, gfn_t end, int min_level)
795 struct tdp_iter iter;
797 bool spte_set = false;
799 BUG_ON(min_level > KVM_MAX_HUGEPAGE_LEVEL);
801 for_each_tdp_pte_min_level(iter, root->spt, root->role.level,
802 min_level, start, end) {
803 if (!is_shadow_present_pte(iter.old_spte) ||
804 !is_last_spte(iter.old_spte, iter.level))
807 new_spte = iter.old_spte & ~PT_WRITABLE_MASK;
809 tdp_mmu_set_spte_no_dirty_log(kvm, &iter, new_spte);
812 tdp_mmu_iter_cond_resched(kvm, &iter);
818 * Remove write access from all the SPTEs mapping GFNs in the memslot. Will
819 * only affect leaf SPTEs down to min_level.
820 * Returns true if an SPTE has been changed and the TLBs need to be flushed.
822 bool kvm_tdp_mmu_wrprot_slot(struct kvm *kvm, struct kvm_memory_slot *slot,
825 struct kvm_mmu_page *root;
827 bool spte_set = false;
829 for_each_tdp_mmu_root(kvm, root) {
830 root_as_id = kvm_mmu_page_as_id(root);
831 if (root_as_id != slot->as_id)
835 * Take a reference on the root so that it cannot be freed if
836 * this thread releases the MMU lock and yields in this loop.
838 kvm_mmu_get_root(kvm, root);
840 spte_set |= wrprot_gfn_range(kvm, root, slot->base_gfn,
841 slot->base_gfn + slot->npages, min_level);
843 kvm_mmu_put_root(kvm, root);
850 * Clear the dirty status of all the SPTEs mapping GFNs in the memslot. If
851 * AD bits are enabled, this will involve clearing the dirty bit on each SPTE.
852 * If AD bits are not enabled, this will require clearing the writable bit on
853 * each SPTE. Returns true if an SPTE has been changed and the TLBs need to
856 static bool clear_dirty_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root,
857 gfn_t start, gfn_t end)
859 struct tdp_iter iter;
861 bool spte_set = false;
863 tdp_root_for_each_leaf_pte(iter, root, start, end) {
864 if (spte_ad_need_write_protect(iter.old_spte)) {
865 if (is_writable_pte(iter.old_spte))
866 new_spte = iter.old_spte & ~PT_WRITABLE_MASK;
870 if (iter.old_spte & shadow_dirty_mask)
871 new_spte = iter.old_spte & ~shadow_dirty_mask;
876 tdp_mmu_set_spte_no_dirty_log(kvm, &iter, new_spte);
879 tdp_mmu_iter_cond_resched(kvm, &iter);
885 * Clear the dirty status of all the SPTEs mapping GFNs in the memslot. If
886 * AD bits are enabled, this will involve clearing the dirty bit on each SPTE.
887 * If AD bits are not enabled, this will require clearing the writable bit on
888 * each SPTE. Returns true if an SPTE has been changed and the TLBs need to
891 bool kvm_tdp_mmu_clear_dirty_slot(struct kvm *kvm, struct kvm_memory_slot *slot)
893 struct kvm_mmu_page *root;
895 bool spte_set = false;
897 for_each_tdp_mmu_root(kvm, root) {
898 root_as_id = kvm_mmu_page_as_id(root);
899 if (root_as_id != slot->as_id)
903 * Take a reference on the root so that it cannot be freed if
904 * this thread releases the MMU lock and yields in this loop.
906 kvm_mmu_get_root(kvm, root);
908 spte_set |= clear_dirty_gfn_range(kvm, root, slot->base_gfn,
909 slot->base_gfn + slot->npages);
911 kvm_mmu_put_root(kvm, root);
918 * Clears the dirty status of all the 4k SPTEs mapping GFNs for which a bit is
919 * set in mask, starting at gfn. The given memslot is expected to contain all
920 * the GFNs represented by set bits in the mask. If AD bits are enabled,
921 * clearing the dirty status will involve clearing the dirty bit on each SPTE
922 * or, if AD bits are not enabled, clearing the writable bit on each SPTE.
924 static void clear_dirty_pt_masked(struct kvm *kvm, struct kvm_mmu_page *root,
925 gfn_t gfn, unsigned long mask, bool wrprot)
927 struct tdp_iter iter;
930 tdp_root_for_each_leaf_pte(iter, root, gfn + __ffs(mask),
931 gfn + BITS_PER_LONG) {
935 if (iter.level > PG_LEVEL_4K ||
936 !(mask & (1UL << (iter.gfn - gfn))))
939 if (wrprot || spte_ad_need_write_protect(iter.old_spte)) {
940 if (is_writable_pte(iter.old_spte))
941 new_spte = iter.old_spte & ~PT_WRITABLE_MASK;
945 if (iter.old_spte & shadow_dirty_mask)
946 new_spte = iter.old_spte & ~shadow_dirty_mask;
951 tdp_mmu_set_spte_no_dirty_log(kvm, &iter, new_spte);
953 mask &= ~(1UL << (iter.gfn - gfn));
958 * Clears the dirty status of all the 4k SPTEs mapping GFNs for which a bit is
959 * set in mask, starting at gfn. The given memslot is expected to contain all
960 * the GFNs represented by set bits in the mask. If AD bits are enabled,
961 * clearing the dirty status will involve clearing the dirty bit on each SPTE
962 * or, if AD bits are not enabled, clearing the writable bit on each SPTE.
964 void kvm_tdp_mmu_clear_dirty_pt_masked(struct kvm *kvm,
965 struct kvm_memory_slot *slot,
966 gfn_t gfn, unsigned long mask,
969 struct kvm_mmu_page *root;
972 lockdep_assert_held(&kvm->mmu_lock);
973 for_each_tdp_mmu_root(kvm, root) {
974 root_as_id = kvm_mmu_page_as_id(root);
975 if (root_as_id != slot->as_id)
978 clear_dirty_pt_masked(kvm, root, gfn, mask, wrprot);
983 * Set the dirty status of all the SPTEs mapping GFNs in the memslot. This is
984 * only used for PML, and so will involve setting the dirty bit on each SPTE.
985 * Returns true if an SPTE has been changed and the TLBs need to be flushed.
987 static bool set_dirty_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root,
988 gfn_t start, gfn_t end)
990 struct tdp_iter iter;
992 bool spte_set = false;
994 tdp_root_for_each_pte(iter, root, start, end) {
995 if (!is_shadow_present_pte(iter.old_spte))
998 new_spte = iter.old_spte | shadow_dirty_mask;
1000 tdp_mmu_set_spte(kvm, &iter, new_spte);
1003 tdp_mmu_iter_cond_resched(kvm, &iter);
1010 * Set the dirty status of all the SPTEs mapping GFNs in the memslot. This is
1011 * only used for PML, and so will involve setting the dirty bit on each SPTE.
1012 * Returns true if an SPTE has been changed and the TLBs need to be flushed.
1014 bool kvm_tdp_mmu_slot_set_dirty(struct kvm *kvm, struct kvm_memory_slot *slot)
1016 struct kvm_mmu_page *root;
1018 bool spte_set = false;
1020 for_each_tdp_mmu_root(kvm, root) {
1021 root_as_id = kvm_mmu_page_as_id(root);
1022 if (root_as_id != slot->as_id)
1026 * Take a reference on the root so that it cannot be freed if
1027 * this thread releases the MMU lock and yields in this loop.
1029 kvm_mmu_get_root(kvm, root);
1031 spte_set |= set_dirty_gfn_range(kvm, root, slot->base_gfn,
1032 slot->base_gfn + slot->npages);
1034 kvm_mmu_put_root(kvm, root);
1040 * Clear non-leaf entries (and free associated page tables) which could
1041 * be replaced by large mappings, for GFNs within the slot.
1043 static void zap_collapsible_spte_range(struct kvm *kvm,
1044 struct kvm_mmu_page *root,
1045 gfn_t start, gfn_t end)
1047 struct tdp_iter iter;
1049 bool spte_set = false;
1051 tdp_root_for_each_pte(iter, root, start, end) {
1052 if (!is_shadow_present_pte(iter.old_spte) ||
1053 is_last_spte(iter.old_spte, iter.level))
1056 pfn = spte_to_pfn(iter.old_spte);
1057 if (kvm_is_reserved_pfn(pfn) ||
1058 !PageTransCompoundMap(pfn_to_page(pfn)))
1061 tdp_mmu_set_spte(kvm, &iter, 0);
1063 spte_set = tdp_mmu_iter_flush_cond_resched(kvm, &iter);
1067 kvm_flush_remote_tlbs(kvm);
1071 * Clear non-leaf entries (and free associated page tables) which could
1072 * be replaced by large mappings, for GFNs within the slot.
1074 void kvm_tdp_mmu_zap_collapsible_sptes(struct kvm *kvm,
1075 const struct kvm_memory_slot *slot)
1077 struct kvm_mmu_page *root;
1080 for_each_tdp_mmu_root(kvm, root) {
1081 root_as_id = kvm_mmu_page_as_id(root);
1082 if (root_as_id != slot->as_id)
1086 * Take a reference on the root so that it cannot be freed if
1087 * this thread releases the MMU lock and yields in this loop.
1089 kvm_mmu_get_root(kvm, root);
1091 zap_collapsible_spte_range(kvm, root, slot->base_gfn,
1092 slot->base_gfn + slot->npages);
1094 kvm_mmu_put_root(kvm, root);
1099 * Removes write access on the last level SPTE mapping this GFN and unsets the
1100 * SPTE_MMU_WRITABLE bit to ensure future writes continue to be intercepted.
1101 * Returns true if an SPTE was set and a TLB flush is needed.
1103 static bool write_protect_gfn(struct kvm *kvm, struct kvm_mmu_page *root,
1106 struct tdp_iter iter;
1108 bool spte_set = false;
1110 tdp_root_for_each_leaf_pte(iter, root, gfn, gfn + 1) {
1111 if (!is_writable_pte(iter.old_spte))
1114 new_spte = iter.old_spte &
1115 ~(PT_WRITABLE_MASK | SPTE_MMU_WRITEABLE);
1117 tdp_mmu_set_spte(kvm, &iter, new_spte);
1125 * Removes write access on the last level SPTE mapping this GFN and unsets the
1126 * SPTE_MMU_WRITABLE bit to ensure future writes continue to be intercepted.
1127 * Returns true if an SPTE was set and a TLB flush is needed.
1129 bool kvm_tdp_mmu_write_protect_gfn(struct kvm *kvm,
1130 struct kvm_memory_slot *slot, gfn_t gfn)
1132 struct kvm_mmu_page *root;
1134 bool spte_set = false;
1136 lockdep_assert_held(&kvm->mmu_lock);
1137 for_each_tdp_mmu_root(kvm, root) {
1138 root_as_id = kvm_mmu_page_as_id(root);
1139 if (root_as_id != slot->as_id)
1142 spte_set |= write_protect_gfn(kvm, root, gfn);
1148 * Return the level of the lowest level SPTE added to sptes.
1149 * That SPTE may be non-present.
1151 int kvm_tdp_mmu_get_walk(struct kvm_vcpu *vcpu, u64 addr, u64 *sptes)
1153 struct tdp_iter iter;
1154 struct kvm_mmu *mmu = vcpu->arch.mmu;
1155 int leaf = vcpu->arch.mmu->shadow_root_level;
1156 gfn_t gfn = addr >> PAGE_SHIFT;
1158 tdp_mmu_for_each_pte(iter, mmu, gfn, gfn + 1) {
1160 sptes[leaf - 1] = iter.old_spte;