1fe5db27dSBen Gardon // SPDX-License-Identifier: GPL-2.0 2fe5db27dSBen Gardon 302c00b3aSBen Gardon #include "mmu.h" 402c00b3aSBen Gardon #include "mmu_internal.h" 5bb18842eSBen Gardon #include "mmutrace.h" 62f2fad08SBen Gardon #include "tdp_iter.h" 7fe5db27dSBen Gardon #include "tdp_mmu.h" 802c00b3aSBen Gardon #include "spte.h" 9fe5db27dSBen Gardon 109a77daacSBen Gardon #include <asm/cmpxchg.h> 1133dd3574SBen Gardon #include <trace/events/kvm.h> 1233dd3574SBen Gardon 1371ba3f31SPaolo Bonzini static bool __read_mostly tdp_mmu_enabled = true; 1495fb5b02SBen Gardon module_param_named(tdp_mmu, tdp_mmu_enabled, bool, 0644); 15fe5db27dSBen Gardon 16fe5db27dSBen Gardon /* Initializes the TDP MMU for the VM, if enabled. */ 17d501f747SBen Gardon bool kvm_mmu_init_tdp_mmu(struct kvm *kvm) 18fe5db27dSBen Gardon { 19897218ffSPaolo Bonzini if (!tdp_enabled || !READ_ONCE(tdp_mmu_enabled)) 20d501f747SBen Gardon return false; 21fe5db27dSBen Gardon 22fe5db27dSBen Gardon /* This should not be changed for the lifetime of the VM. */ 23fe5db27dSBen Gardon kvm->arch.tdp_mmu_enabled = true; 2402c00b3aSBen Gardon 2502c00b3aSBen Gardon INIT_LIST_HEAD(&kvm->arch.tdp_mmu_roots); 269a77daacSBen Gardon spin_lock_init(&kvm->arch.tdp_mmu_pages_lock); 2789c0fd49SBen Gardon INIT_LIST_HEAD(&kvm->arch.tdp_mmu_pages); 28d501f747SBen Gardon 29d501f747SBen Gardon return true; 30fe5db27dSBen Gardon } 31fe5db27dSBen Gardon 326103bc07SBen Gardon static __always_inline void kvm_lockdep_assert_mmu_lock_held(struct kvm *kvm, 336103bc07SBen Gardon bool shared) 346103bc07SBen Gardon { 356103bc07SBen Gardon if (shared) 366103bc07SBen Gardon lockdep_assert_held_read(&kvm->mmu_lock); 376103bc07SBen Gardon else 386103bc07SBen Gardon lockdep_assert_held_write(&kvm->mmu_lock); 396103bc07SBen Gardon } 406103bc07SBen Gardon 41fe5db27dSBen Gardon void kvm_mmu_uninit_tdp_mmu(struct kvm *kvm) 42fe5db27dSBen Gardon { 43fe5db27dSBen Gardon if (!kvm->arch.tdp_mmu_enabled) 44fe5db27dSBen Gardon return; 4502c00b3aSBen Gardon 46524a1e4eSSean Christopherson WARN_ON(!list_empty(&kvm->arch.tdp_mmu_pages)); 4702c00b3aSBen Gardon WARN_ON(!list_empty(&kvm->arch.tdp_mmu_roots)); 487cca2d0bSBen Gardon 497cca2d0bSBen Gardon /* 507cca2d0bSBen Gardon * Ensure that all the outstanding RCU callbacks to free shadow pages 517cca2d0bSBen Gardon * can run before the VM is torn down. 527cca2d0bSBen Gardon */ 537cca2d0bSBen Gardon rcu_barrier(); 5402c00b3aSBen Gardon } 5502c00b3aSBen Gardon 562bdb3d84SBen Gardon static bool zap_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root, 576103bc07SBen Gardon gfn_t start, gfn_t end, bool can_yield, bool flush, 586103bc07SBen Gardon bool shared); 592bdb3d84SBen Gardon 602bdb3d84SBen Gardon static void tdp_mmu_free_sp(struct kvm_mmu_page *sp) 61a889ea54SBen Gardon { 622bdb3d84SBen Gardon free_page((unsigned long)sp->spt); 632bdb3d84SBen Gardon kmem_cache_free(mmu_page_header_cache, sp); 64a889ea54SBen Gardon } 65a889ea54SBen Gardon 66c0e64238SBen Gardon /* 67c0e64238SBen Gardon * This is called through call_rcu in order to free TDP page table memory 68c0e64238SBen Gardon * safely with respect to other kernel threads that may be operating on 69c0e64238SBen Gardon * the memory. 70c0e64238SBen Gardon * By only accessing TDP MMU page table memory in an RCU read critical 71c0e64238SBen Gardon * section, and freeing it after a grace period, lockless access to that 72c0e64238SBen Gardon * memory won't use it after it is freed. 73c0e64238SBen Gardon */ 74c0e64238SBen Gardon static void tdp_mmu_free_sp_rcu_callback(struct rcu_head *head) 75a889ea54SBen Gardon { 76c0e64238SBen Gardon struct kvm_mmu_page *sp = container_of(head, struct kvm_mmu_page, 77c0e64238SBen Gardon rcu_head); 78a889ea54SBen Gardon 79c0e64238SBen Gardon tdp_mmu_free_sp(sp); 80a889ea54SBen Gardon } 81a889ea54SBen Gardon 826103bc07SBen Gardon void kvm_tdp_mmu_put_root(struct kvm *kvm, struct kvm_mmu_page *root, 836103bc07SBen Gardon bool shared) 842bdb3d84SBen Gardon { 856103bc07SBen Gardon kvm_lockdep_assert_mmu_lock_held(kvm, shared); 862bdb3d84SBen Gardon 8711cccf5cSBen Gardon if (!refcount_dec_and_test(&root->tdp_mmu_root_count)) 882bdb3d84SBen Gardon return; 892bdb3d84SBen Gardon 902bdb3d84SBen Gardon WARN_ON(!root->tdp_mmu_page); 912bdb3d84SBen Gardon 92c0e64238SBen Gardon spin_lock(&kvm->arch.tdp_mmu_pages_lock); 93c0e64238SBen Gardon list_del_rcu(&root->link); 94c0e64238SBen Gardon spin_unlock(&kvm->arch.tdp_mmu_pages_lock); 952bdb3d84SBen Gardon 96524a1e4eSSean Christopherson zap_gfn_range(kvm, root, 0, -1ull, false, false, shared); 972bdb3d84SBen Gardon 98c0e64238SBen Gardon call_rcu(&root->rcu_head, tdp_mmu_free_sp_rcu_callback); 99a889ea54SBen Gardon } 100a889ea54SBen Gardon 101cfc10997SBen Gardon /* 102d62007edSSean Christopherson * Returns the next root after @prev_root (or the first root if @prev_root is 103d62007edSSean Christopherson * NULL). A reference to the returned root is acquired, and the reference to 104d62007edSSean Christopherson * @prev_root is released (the caller obviously must hold a reference to 105d62007edSSean Christopherson * @prev_root if it's non-NULL). 106d62007edSSean Christopherson * 107d62007edSSean Christopherson * If @only_valid is true, invalid roots are skipped. 108d62007edSSean Christopherson * 109d62007edSSean Christopherson * Returns NULL if the end of tdp_mmu_roots was reached. 110cfc10997SBen Gardon */ 111cfc10997SBen Gardon static struct kvm_mmu_page *tdp_mmu_next_root(struct kvm *kvm, 1126103bc07SBen Gardon struct kvm_mmu_page *prev_root, 113d62007edSSean Christopherson bool shared, bool only_valid) 114a889ea54SBen Gardon { 115a889ea54SBen Gardon struct kvm_mmu_page *next_root; 116a889ea54SBen Gardon 117c0e64238SBen Gardon rcu_read_lock(); 118c0e64238SBen Gardon 119cfc10997SBen Gardon if (prev_root) 120c0e64238SBen Gardon next_root = list_next_or_null_rcu(&kvm->arch.tdp_mmu_roots, 121c0e64238SBen Gardon &prev_root->link, 122c0e64238SBen Gardon typeof(*prev_root), link); 123cfc10997SBen Gardon else 124c0e64238SBen Gardon next_root = list_first_or_null_rcu(&kvm->arch.tdp_mmu_roots, 125cfc10997SBen Gardon typeof(*next_root), link); 126cfc10997SBen Gardon 12704dc4e6cSSean Christopherson while (next_root) { 128d62007edSSean Christopherson if ((!only_valid || !next_root->role.invalid) && 129ad6d6b94SJinrong Liang kvm_tdp_mmu_get_root(next_root)) 13004dc4e6cSSean Christopherson break; 13104dc4e6cSSean Christopherson 132c0e64238SBen Gardon next_root = list_next_or_null_rcu(&kvm->arch.tdp_mmu_roots, 133c0e64238SBen Gardon &next_root->link, typeof(*next_root), link); 13404dc4e6cSSean Christopherson } 135fb101293SBen Gardon 136c0e64238SBen Gardon rcu_read_unlock(); 137cfc10997SBen Gardon 138cfc10997SBen Gardon if (prev_root) 1396103bc07SBen Gardon kvm_tdp_mmu_put_root(kvm, prev_root, shared); 140cfc10997SBen Gardon 141a889ea54SBen Gardon return next_root; 142a889ea54SBen Gardon } 143a889ea54SBen Gardon 144a889ea54SBen Gardon /* 145a889ea54SBen Gardon * Note: this iterator gets and puts references to the roots it iterates over. 146a889ea54SBen Gardon * This makes it safe to release the MMU lock and yield within the loop, but 147a889ea54SBen Gardon * if exiting the loop early, the caller must drop the reference to the most 148a889ea54SBen Gardon * recent root. (Unless keeping a live reference is desirable.) 1496103bc07SBen Gardon * 1506103bc07SBen Gardon * If shared is set, this function is operating under the MMU lock in read 1516103bc07SBen Gardon * mode. In the unlikely event that this thread must free a root, the lock 1526103bc07SBen Gardon * will be temporarily dropped and reacquired in write mode. 153a889ea54SBen Gardon */ 154d62007edSSean Christopherson #define __for_each_tdp_mmu_root_yield_safe(_kvm, _root, _as_id, _shared, _only_valid)\ 155d62007edSSean Christopherson for (_root = tdp_mmu_next_root(_kvm, NULL, _shared, _only_valid); \ 156cfc10997SBen Gardon _root; \ 157d62007edSSean Christopherson _root = tdp_mmu_next_root(_kvm, _root, _shared, _only_valid)) \ 158a3f15bdaSSean Christopherson if (kvm_mmu_page_as_id(_root) != _as_id) { \ 159a3f15bdaSSean Christopherson } else 160a889ea54SBen Gardon 161d62007edSSean Christopherson #define for_each_valid_tdp_mmu_root_yield_safe(_kvm, _root, _as_id, _shared) \ 162d62007edSSean Christopherson __for_each_tdp_mmu_root_yield_safe(_kvm, _root, _as_id, _shared, true) 163d62007edSSean Christopherson 164d62007edSSean Christopherson #define for_each_tdp_mmu_root_yield_safe(_kvm, _root, _as_id, _shared) \ 165d62007edSSean Christopherson __for_each_tdp_mmu_root_yield_safe(_kvm, _root, _as_id, _shared, false) 166d62007edSSean Christopherson 167a3f15bdaSSean Christopherson #define for_each_tdp_mmu_root(_kvm, _root, _as_id) \ 168c0e64238SBen Gardon list_for_each_entry_rcu(_root, &_kvm->arch.tdp_mmu_roots, link, \ 169c0e64238SBen Gardon lockdep_is_held_type(&kvm->mmu_lock, 0) || \ 170c0e64238SBen Gardon lockdep_is_held(&kvm->arch.tdp_mmu_pages_lock)) \ 171a3f15bdaSSean Christopherson if (kvm_mmu_page_as_id(_root) != _as_id) { \ 172a3f15bdaSSean Christopherson } else 17302c00b3aSBen Gardon 17402c00b3aSBen Gardon static union kvm_mmu_page_role page_role_for_level(struct kvm_vcpu *vcpu, 17502c00b3aSBen Gardon int level) 17602c00b3aSBen Gardon { 17702c00b3aSBen Gardon union kvm_mmu_page_role role; 17802c00b3aSBen Gardon 17902c00b3aSBen Gardon role = vcpu->arch.mmu->mmu_role.base; 18002c00b3aSBen Gardon role.level = level; 18102c00b3aSBen Gardon role.direct = true; 182bb3b394dSLai Jiangshan role.has_4_byte_gpte = false; 18302c00b3aSBen Gardon role.access = ACC_ALL; 18487e888eaSPaolo Bonzini role.ad_disabled = !shadow_accessed_mask; 18502c00b3aSBen Gardon 18602c00b3aSBen Gardon return role; 18702c00b3aSBen Gardon } 18802c00b3aSBen Gardon 18902c00b3aSBen Gardon static struct kvm_mmu_page *alloc_tdp_mmu_page(struct kvm_vcpu *vcpu, gfn_t gfn, 19002c00b3aSBen Gardon int level) 19102c00b3aSBen Gardon { 19202c00b3aSBen Gardon struct kvm_mmu_page *sp; 19302c00b3aSBen Gardon 19402c00b3aSBen Gardon sp = kvm_mmu_memory_cache_alloc(&vcpu->arch.mmu_page_header_cache); 19502c00b3aSBen Gardon sp->spt = kvm_mmu_memory_cache_alloc(&vcpu->arch.mmu_shadow_page_cache); 19602c00b3aSBen Gardon set_page_private(virt_to_page(sp->spt), (unsigned long)sp); 19702c00b3aSBen Gardon 19802c00b3aSBen Gardon sp->role.word = page_role_for_level(vcpu, level).word; 19902c00b3aSBen Gardon sp->gfn = gfn; 20002c00b3aSBen Gardon sp->tdp_mmu_page = true; 20102c00b3aSBen Gardon 20233dd3574SBen Gardon trace_kvm_mmu_get_page(sp, true); 20333dd3574SBen Gardon 20402c00b3aSBen Gardon return sp; 20502c00b3aSBen Gardon } 20602c00b3aSBen Gardon 2076e6ec584SSean Christopherson hpa_t kvm_tdp_mmu_get_vcpu_root_hpa(struct kvm_vcpu *vcpu) 20802c00b3aSBen Gardon { 20902c00b3aSBen Gardon union kvm_mmu_page_role role; 21002c00b3aSBen Gardon struct kvm *kvm = vcpu->kvm; 21102c00b3aSBen Gardon struct kvm_mmu_page *root; 21202c00b3aSBen Gardon 2136e6ec584SSean Christopherson lockdep_assert_held_write(&kvm->mmu_lock); 21402c00b3aSBen Gardon 21502c00b3aSBen Gardon role = page_role_for_level(vcpu, vcpu->arch.mmu->shadow_root_level); 21602c00b3aSBen Gardon 21704dc4e6cSSean Christopherson /* 21804dc4e6cSSean Christopherson * Check for an existing root before allocating a new one. Note, the 21904dc4e6cSSean Christopherson * role check prevents consuming an invalid root. 22004dc4e6cSSean Christopherson */ 221a3f15bdaSSean Christopherson for_each_tdp_mmu_root(kvm, root, kvm_mmu_role_as_id(role)) { 222fb101293SBen Gardon if (root->role.word == role.word && 223ad6d6b94SJinrong Liang kvm_tdp_mmu_get_root(root)) 2246e6ec584SSean Christopherson goto out; 22502c00b3aSBen Gardon } 22602c00b3aSBen Gardon 22702c00b3aSBen Gardon root = alloc_tdp_mmu_page(vcpu, 0, vcpu->arch.mmu->shadow_root_level); 22811cccf5cSBen Gardon refcount_set(&root->tdp_mmu_root_count, 1); 22902c00b3aSBen Gardon 230c0e64238SBen Gardon spin_lock(&kvm->arch.tdp_mmu_pages_lock); 231c0e64238SBen Gardon list_add_rcu(&root->link, &kvm->arch.tdp_mmu_roots); 232c0e64238SBen Gardon spin_unlock(&kvm->arch.tdp_mmu_pages_lock); 23302c00b3aSBen Gardon 2346e6ec584SSean Christopherson out: 23502c00b3aSBen Gardon return __pa(root->spt); 236fe5db27dSBen Gardon } 2372f2fad08SBen Gardon 2382f2fad08SBen Gardon static void handle_changed_spte(struct kvm *kvm, int as_id, gfn_t gfn, 2399a77daacSBen Gardon u64 old_spte, u64 new_spte, int level, 2409a77daacSBen Gardon bool shared); 2412f2fad08SBen Gardon 242f8e14497SBen Gardon static void handle_changed_spte_acc_track(u64 old_spte, u64 new_spte, int level) 243f8e14497SBen Gardon { 244f8e14497SBen Gardon if (!is_shadow_present_pte(old_spte) || !is_last_spte(old_spte, level)) 245f8e14497SBen Gardon return; 246f8e14497SBen Gardon 247f8e14497SBen Gardon if (is_accessed_spte(old_spte) && 24864bb2769SSean Christopherson (!is_shadow_present_pte(new_spte) || !is_accessed_spte(new_spte) || 24964bb2769SSean Christopherson spte_to_pfn(old_spte) != spte_to_pfn(new_spte))) 250f8e14497SBen Gardon kvm_set_pfn_accessed(spte_to_pfn(old_spte)); 251f8e14497SBen Gardon } 252f8e14497SBen Gardon 253a6a0b05dSBen Gardon static void handle_changed_spte_dirty_log(struct kvm *kvm, int as_id, gfn_t gfn, 254a6a0b05dSBen Gardon u64 old_spte, u64 new_spte, int level) 255a6a0b05dSBen Gardon { 256a6a0b05dSBen Gardon bool pfn_changed; 257a6a0b05dSBen Gardon struct kvm_memory_slot *slot; 258a6a0b05dSBen Gardon 259a6a0b05dSBen Gardon if (level > PG_LEVEL_4K) 260a6a0b05dSBen Gardon return; 261a6a0b05dSBen Gardon 262a6a0b05dSBen Gardon pfn_changed = spte_to_pfn(old_spte) != spte_to_pfn(new_spte); 263a6a0b05dSBen Gardon 264a6a0b05dSBen Gardon if ((!is_writable_pte(old_spte) || pfn_changed) && 265a6a0b05dSBen Gardon is_writable_pte(new_spte)) { 266a6a0b05dSBen Gardon slot = __gfn_to_memslot(__kvm_memslots(kvm, as_id), gfn); 267fb04a1edSPeter Xu mark_page_dirty_in_slot(kvm, slot, gfn); 268a6a0b05dSBen Gardon } 269a6a0b05dSBen Gardon } 270a6a0b05dSBen Gardon 2712f2fad08SBen Gardon /** 272a9442f59SBen Gardon * tdp_mmu_link_page - Add a new page to the list of pages used by the TDP MMU 273a9442f59SBen Gardon * 274a9442f59SBen Gardon * @kvm: kvm instance 275a9442f59SBen Gardon * @sp: the new page 276a9442f59SBen Gardon * @account_nx: This page replaces a NX large page and should be marked for 277a9442f59SBen Gardon * eventual reclaim. 278a9442f59SBen Gardon */ 279a9442f59SBen Gardon static void tdp_mmu_link_page(struct kvm *kvm, struct kvm_mmu_page *sp, 2809653f2daSSean Christopherson bool account_nx) 281a9442f59SBen Gardon { 2829a77daacSBen Gardon spin_lock(&kvm->arch.tdp_mmu_pages_lock); 283a9442f59SBen Gardon list_add(&sp->link, &kvm->arch.tdp_mmu_pages); 284a9442f59SBen Gardon if (account_nx) 285a9442f59SBen Gardon account_huge_nx_page(kvm, sp); 2869a77daacSBen Gardon spin_unlock(&kvm->arch.tdp_mmu_pages_lock); 287a9442f59SBen Gardon } 288a9442f59SBen Gardon 289a9442f59SBen Gardon /** 290a9442f59SBen Gardon * tdp_mmu_unlink_page - Remove page from the list of pages used by the TDP MMU 291a9442f59SBen Gardon * 292a9442f59SBen Gardon * @kvm: kvm instance 293a9442f59SBen Gardon * @sp: the page to be removed 2949a77daacSBen Gardon * @shared: This operation may not be running under the exclusive use of 2959a77daacSBen Gardon * the MMU lock and the operation must synchronize with other 2969a77daacSBen Gardon * threads that might be adding or removing pages. 297a9442f59SBen Gardon */ 2989a77daacSBen Gardon static void tdp_mmu_unlink_page(struct kvm *kvm, struct kvm_mmu_page *sp, 2999a77daacSBen Gardon bool shared) 300a9442f59SBen Gardon { 3019a77daacSBen Gardon if (shared) 3029a77daacSBen Gardon spin_lock(&kvm->arch.tdp_mmu_pages_lock); 3039a77daacSBen Gardon else 304a9442f59SBen Gardon lockdep_assert_held_write(&kvm->mmu_lock); 305a9442f59SBen Gardon 306a9442f59SBen Gardon list_del(&sp->link); 307a9442f59SBen Gardon if (sp->lpage_disallowed) 308a9442f59SBen Gardon unaccount_huge_nx_page(kvm, sp); 3099a77daacSBen Gardon 3109a77daacSBen Gardon if (shared) 3119a77daacSBen Gardon spin_unlock(&kvm->arch.tdp_mmu_pages_lock); 312a9442f59SBen Gardon } 313a9442f59SBen Gardon 314a9442f59SBen Gardon /** 315a066e61fSBen Gardon * handle_removed_tdp_mmu_page - handle a pt removed from the TDP structure 316a066e61fSBen Gardon * 317a066e61fSBen Gardon * @kvm: kvm instance 318a066e61fSBen Gardon * @pt: the page removed from the paging structure 3199a77daacSBen Gardon * @shared: This operation may not be running under the exclusive use 3209a77daacSBen Gardon * of the MMU lock and the operation must synchronize with other 3219a77daacSBen Gardon * threads that might be modifying SPTEs. 322a066e61fSBen Gardon * 323a066e61fSBen Gardon * Given a page table that has been removed from the TDP paging structure, 324a066e61fSBen Gardon * iterates through the page table to clear SPTEs and free child page tables. 32570fb3e41SBen Gardon * 32670fb3e41SBen Gardon * Note that pt is passed in as a tdp_ptep_t, but it does not need RCU 32770fb3e41SBen Gardon * protection. Since this thread removed it from the paging structure, 32870fb3e41SBen Gardon * this thread will be responsible for ensuring the page is freed. Hence the 32970fb3e41SBen Gardon * early rcu_dereferences in the function. 330a066e61fSBen Gardon */ 33170fb3e41SBen Gardon static void handle_removed_tdp_mmu_page(struct kvm *kvm, tdp_ptep_t pt, 3329a77daacSBen Gardon bool shared) 333a066e61fSBen Gardon { 33470fb3e41SBen Gardon struct kvm_mmu_page *sp = sptep_to_sp(rcu_dereference(pt)); 335a066e61fSBen Gardon int level = sp->role.level; 336e25f0e0cSBen Gardon gfn_t base_gfn = sp->gfn; 337a066e61fSBen Gardon int i; 338a066e61fSBen Gardon 339a066e61fSBen Gardon trace_kvm_mmu_prepare_zap_page(sp); 340a066e61fSBen Gardon 3419a77daacSBen Gardon tdp_mmu_unlink_page(kvm, sp, shared); 342a066e61fSBen Gardon 343a066e61fSBen Gardon for (i = 0; i < PT64_ENT_PER_PAGE; i++) { 344574c3c55SBen Gardon u64 *sptep = rcu_dereference(pt) + i; 345574c3c55SBen Gardon gfn_t gfn = base_gfn + i * KVM_PAGES_PER_HPAGE(level); 346574c3c55SBen Gardon u64 old_child_spte; 3479a77daacSBen Gardon 3489a77daacSBen Gardon if (shared) { 349e25f0e0cSBen Gardon /* 350e25f0e0cSBen Gardon * Set the SPTE to a nonpresent value that other 351e25f0e0cSBen Gardon * threads will not overwrite. If the SPTE was 352e25f0e0cSBen Gardon * already marked as removed then another thread 353e25f0e0cSBen Gardon * handling a page fault could overwrite it, so 354e25f0e0cSBen Gardon * set the SPTE until it is set from some other 355e25f0e0cSBen Gardon * value to the removed SPTE value. 356e25f0e0cSBen Gardon */ 357e25f0e0cSBen Gardon for (;;) { 358e25f0e0cSBen Gardon old_child_spte = xchg(sptep, REMOVED_SPTE); 359e25f0e0cSBen Gardon if (!is_removed_spte(old_child_spte)) 360e25f0e0cSBen Gardon break; 361e25f0e0cSBen Gardon cpu_relax(); 362e25f0e0cSBen Gardon } 3639a77daacSBen Gardon } else { 3648df9f1afSSean Christopherson /* 3658df9f1afSSean Christopherson * If the SPTE is not MMU-present, there is no backing 3668df9f1afSSean Christopherson * page associated with the SPTE and so no side effects 3678df9f1afSSean Christopherson * that need to be recorded, and exclusive ownership of 3688df9f1afSSean Christopherson * mmu_lock ensures the SPTE can't be made present. 3698df9f1afSSean Christopherson * Note, zapping MMIO SPTEs is also unnecessary as they 3708df9f1afSSean Christopherson * are guarded by the memslots generation, not by being 3718df9f1afSSean Christopherson * unreachable. 3728df9f1afSSean Christopherson */ 3739a77daacSBen Gardon old_child_spte = READ_ONCE(*sptep); 3748df9f1afSSean Christopherson if (!is_shadow_present_pte(old_child_spte)) 3758df9f1afSSean Christopherson continue; 376e25f0e0cSBen Gardon 377e25f0e0cSBen Gardon /* 378e25f0e0cSBen Gardon * Marking the SPTE as a removed SPTE is not 379e25f0e0cSBen Gardon * strictly necessary here as the MMU lock will 380e25f0e0cSBen Gardon * stop other threads from concurrently modifying 381e25f0e0cSBen Gardon * this SPTE. Using the removed SPTE value keeps 382e25f0e0cSBen Gardon * the two branches consistent and simplifies 383e25f0e0cSBen Gardon * the function. 384e25f0e0cSBen Gardon */ 385e25f0e0cSBen Gardon WRITE_ONCE(*sptep, REMOVED_SPTE); 3869a77daacSBen Gardon } 387e25f0e0cSBen Gardon handle_changed_spte(kvm, kvm_mmu_page_as_id(sp), gfn, 388f1b83255SKai Huang old_child_spte, REMOVED_SPTE, level, 389e25f0e0cSBen Gardon shared); 390a066e61fSBen Gardon } 391a066e61fSBen Gardon 392574c3c55SBen Gardon kvm_flush_remote_tlbs_with_address(kvm, base_gfn, 393f1b83255SKai Huang KVM_PAGES_PER_HPAGE(level + 1)); 394a066e61fSBen Gardon 3957cca2d0bSBen Gardon call_rcu(&sp->rcu_head, tdp_mmu_free_sp_rcu_callback); 396a066e61fSBen Gardon } 397a066e61fSBen Gardon 398a066e61fSBen Gardon /** 3997f6231a3SKai Huang * __handle_changed_spte - handle bookkeeping associated with an SPTE change 4002f2fad08SBen Gardon * @kvm: kvm instance 4012f2fad08SBen Gardon * @as_id: the address space of the paging structure the SPTE was a part of 4022f2fad08SBen Gardon * @gfn: the base GFN that was mapped by the SPTE 4032f2fad08SBen Gardon * @old_spte: The value of the SPTE before the change 4042f2fad08SBen Gardon * @new_spte: The value of the SPTE after the change 4052f2fad08SBen Gardon * @level: the level of the PT the SPTE is part of in the paging structure 4069a77daacSBen Gardon * @shared: This operation may not be running under the exclusive use of 4079a77daacSBen Gardon * the MMU lock and the operation must synchronize with other 4089a77daacSBen Gardon * threads that might be modifying SPTEs. 4092f2fad08SBen Gardon * 4102f2fad08SBen Gardon * Handle bookkeeping that might result from the modification of a SPTE. 4112f2fad08SBen Gardon * This function must be called for all TDP SPTE modifications. 4122f2fad08SBen Gardon */ 4132f2fad08SBen Gardon static void __handle_changed_spte(struct kvm *kvm, int as_id, gfn_t gfn, 4149a77daacSBen Gardon u64 old_spte, u64 new_spte, int level, 4159a77daacSBen Gardon bool shared) 4162f2fad08SBen Gardon { 4172f2fad08SBen Gardon bool was_present = is_shadow_present_pte(old_spte); 4182f2fad08SBen Gardon bool is_present = is_shadow_present_pte(new_spte); 4192f2fad08SBen Gardon bool was_leaf = was_present && is_last_spte(old_spte, level); 4202f2fad08SBen Gardon bool is_leaf = is_present && is_last_spte(new_spte, level); 4212f2fad08SBen Gardon bool pfn_changed = spte_to_pfn(old_spte) != spte_to_pfn(new_spte); 4222f2fad08SBen Gardon 4232f2fad08SBen Gardon WARN_ON(level > PT64_ROOT_MAX_LEVEL); 4242f2fad08SBen Gardon WARN_ON(level < PG_LEVEL_4K); 425764388ceSSean Christopherson WARN_ON(gfn & (KVM_PAGES_PER_HPAGE(level) - 1)); 4262f2fad08SBen Gardon 4272f2fad08SBen Gardon /* 4282f2fad08SBen Gardon * If this warning were to trigger it would indicate that there was a 4292f2fad08SBen Gardon * missing MMU notifier or a race with some notifier handler. 4302f2fad08SBen Gardon * A present, leaf SPTE should never be directly replaced with another 431d9f6e12fSIngo Molnar * present leaf SPTE pointing to a different PFN. A notifier handler 4322f2fad08SBen Gardon * should be zapping the SPTE before the main MM's page table is 4332f2fad08SBen Gardon * changed, or the SPTE should be zeroed, and the TLBs flushed by the 4342f2fad08SBen Gardon * thread before replacement. 4352f2fad08SBen Gardon */ 4362f2fad08SBen Gardon if (was_leaf && is_leaf && pfn_changed) { 4372f2fad08SBen Gardon pr_err("Invalid SPTE change: cannot replace a present leaf\n" 4382f2fad08SBen Gardon "SPTE with another present leaf SPTE mapping a\n" 4392f2fad08SBen Gardon "different PFN!\n" 4402f2fad08SBen Gardon "as_id: %d gfn: %llx old_spte: %llx new_spte: %llx level: %d", 4412f2fad08SBen Gardon as_id, gfn, old_spte, new_spte, level); 4422f2fad08SBen Gardon 4432f2fad08SBen Gardon /* 4442f2fad08SBen Gardon * Crash the host to prevent error propagation and guest data 445d9f6e12fSIngo Molnar * corruption. 4462f2fad08SBen Gardon */ 4472f2fad08SBen Gardon BUG(); 4482f2fad08SBen Gardon } 4492f2fad08SBen Gardon 4502f2fad08SBen Gardon if (old_spte == new_spte) 4512f2fad08SBen Gardon return; 4522f2fad08SBen Gardon 453b9a98c34SBen Gardon trace_kvm_tdp_mmu_spte_changed(as_id, gfn, level, old_spte, new_spte); 454b9a98c34SBen Gardon 455115111efSDavid Matlack if (is_leaf) 456115111efSDavid Matlack check_spte_writable_invariants(new_spte); 457115111efSDavid Matlack 4582f2fad08SBen Gardon /* 4592f2fad08SBen Gardon * The only times a SPTE should be changed from a non-present to 4602f2fad08SBen Gardon * non-present state is when an MMIO entry is installed/modified/ 4612f2fad08SBen Gardon * removed. In that case, there is nothing to do here. 4622f2fad08SBen Gardon */ 4632f2fad08SBen Gardon if (!was_present && !is_present) { 4642f2fad08SBen Gardon /* 46508f07c80SBen Gardon * If this change does not involve a MMIO SPTE or removed SPTE, 46608f07c80SBen Gardon * it is unexpected. Log the change, though it should not 46708f07c80SBen Gardon * impact the guest since both the former and current SPTEs 46808f07c80SBen Gardon * are nonpresent. 4692f2fad08SBen Gardon */ 47008f07c80SBen Gardon if (WARN_ON(!is_mmio_spte(old_spte) && 47108f07c80SBen Gardon !is_mmio_spte(new_spte) && 47208f07c80SBen Gardon !is_removed_spte(new_spte))) 4732f2fad08SBen Gardon pr_err("Unexpected SPTE change! Nonpresent SPTEs\n" 4742f2fad08SBen Gardon "should not be replaced with another,\n" 4752f2fad08SBen Gardon "different nonpresent SPTE, unless one or both\n" 47608f07c80SBen Gardon "are MMIO SPTEs, or the new SPTE is\n" 47708f07c80SBen Gardon "a temporary removed SPTE.\n" 4782f2fad08SBen Gardon "as_id: %d gfn: %llx old_spte: %llx new_spte: %llx level: %d", 4792f2fad08SBen Gardon as_id, gfn, old_spte, new_spte, level); 4802f2fad08SBen Gardon return; 4812f2fad08SBen Gardon } 4822f2fad08SBen Gardon 48371f51d2cSMingwei Zhang if (is_leaf != was_leaf) 48471f51d2cSMingwei Zhang kvm_update_page_stats(kvm, level, is_leaf ? 1 : -1); 4852f2fad08SBen Gardon 4862f2fad08SBen Gardon if (was_leaf && is_dirty_spte(old_spte) && 48764bb2769SSean Christopherson (!is_present || !is_dirty_spte(new_spte) || pfn_changed)) 4882f2fad08SBen Gardon kvm_set_pfn_dirty(spte_to_pfn(old_spte)); 4892f2fad08SBen Gardon 4902f2fad08SBen Gardon /* 4912f2fad08SBen Gardon * Recursively handle child PTs if the change removed a subtree from 4922f2fad08SBen Gardon * the paging structure. 4932f2fad08SBen Gardon */ 494a066e61fSBen Gardon if (was_present && !was_leaf && (pfn_changed || !is_present)) 495a066e61fSBen Gardon handle_removed_tdp_mmu_page(kvm, 4969a77daacSBen Gardon spte_to_child_pt(old_spte, level), shared); 4972f2fad08SBen Gardon } 4982f2fad08SBen Gardon 4992f2fad08SBen Gardon static void handle_changed_spte(struct kvm *kvm, int as_id, gfn_t gfn, 5009a77daacSBen Gardon u64 old_spte, u64 new_spte, int level, 5019a77daacSBen Gardon bool shared) 5022f2fad08SBen Gardon { 5039a77daacSBen Gardon __handle_changed_spte(kvm, as_id, gfn, old_spte, new_spte, level, 5049a77daacSBen Gardon shared); 505f8e14497SBen Gardon handle_changed_spte_acc_track(old_spte, new_spte, level); 506a6a0b05dSBen Gardon handle_changed_spte_dirty_log(kvm, as_id, gfn, old_spte, 507a6a0b05dSBen Gardon new_spte, level); 5082f2fad08SBen Gardon } 509faaf05b0SBen Gardon 510fe43fa2fSBen Gardon /* 5116ccf4438SPaolo Bonzini * tdp_mmu_set_spte_atomic - Set a TDP MMU SPTE atomically 5126ccf4438SPaolo Bonzini * and handle the associated bookkeeping. Do not mark the page dirty 51324ae4cfaSBen Gardon * in KVM's dirty bitmaps. 5149a77daacSBen Gardon * 5153255530aSDavid Matlack * If setting the SPTE fails because it has changed, iter->old_spte will be 5163255530aSDavid Matlack * refreshed to the current value of the spte. 5173255530aSDavid Matlack * 5189a77daacSBen Gardon * @kvm: kvm instance 5199a77daacSBen Gardon * @iter: a tdp_iter instance currently on the SPTE that should be set 5209a77daacSBen Gardon * @new_spte: The value the SPTE should be set to 521*3e72c791SDavid Matlack * Return: 522*3e72c791SDavid Matlack * * 0 - If the SPTE was set. 523*3e72c791SDavid Matlack * * -EBUSY - If the SPTE cannot be set. In this case this function will have 524*3e72c791SDavid Matlack * no side-effects other than setting iter->old_spte to the last 525*3e72c791SDavid Matlack * known value of the spte. 5269a77daacSBen Gardon */ 527*3e72c791SDavid Matlack static inline int tdp_mmu_set_spte_atomic(struct kvm *kvm, 5289a77daacSBen Gardon struct tdp_iter *iter, 5299a77daacSBen Gardon u64 new_spte) 5309a77daacSBen Gardon { 5313255530aSDavid Matlack u64 *sptep = rcu_dereference(iter->sptep); 5323255530aSDavid Matlack u64 old_spte; 5333255530aSDavid Matlack 5343a0f64deSSean Christopherson WARN_ON_ONCE(iter->yielded); 5353a0f64deSSean Christopherson 5369a77daacSBen Gardon lockdep_assert_held_read(&kvm->mmu_lock); 5379a77daacSBen Gardon 53808f07c80SBen Gardon /* 53908f07c80SBen Gardon * Do not change removed SPTEs. Only the thread that froze the SPTE 54008f07c80SBen Gardon * may modify it. 54108f07c80SBen Gardon */ 5427a51393aSSean Christopherson if (is_removed_spte(iter->old_spte)) 543*3e72c791SDavid Matlack return -EBUSY; 54408f07c80SBen Gardon 5456e8eb206SDavid Matlack /* 5466e8eb206SDavid Matlack * Note, fast_pf_fix_direct_spte() can also modify TDP MMU SPTEs and 5476e8eb206SDavid Matlack * does not hold the mmu_lock. 5486e8eb206SDavid Matlack */ 5493255530aSDavid Matlack old_spte = cmpxchg64(sptep, iter->old_spte, new_spte); 5503255530aSDavid Matlack if (old_spte != iter->old_spte) { 5513255530aSDavid Matlack /* 5523255530aSDavid Matlack * The page table entry was modified by a different logical 5533255530aSDavid Matlack * CPU. Refresh iter->old_spte with the current value so the 5543255530aSDavid Matlack * caller operates on fresh data, e.g. if it retries 5553255530aSDavid Matlack * tdp_mmu_set_spte_atomic(). 5563255530aSDavid Matlack */ 5573255530aSDavid Matlack iter->old_spte = old_spte; 558*3e72c791SDavid Matlack return -EBUSY; 5593255530aSDavid Matlack } 5609a77daacSBen Gardon 56124ae4cfaSBen Gardon __handle_changed_spte(kvm, iter->as_id, iter->gfn, iter->old_spte, 56208889894SSean Christopherson new_spte, iter->level, true); 56324ae4cfaSBen Gardon handle_changed_spte_acc_track(iter->old_spte, new_spte, iter->level); 5649a77daacSBen Gardon 565*3e72c791SDavid Matlack return 0; 5669a77daacSBen Gardon } 5679a77daacSBen Gardon 568*3e72c791SDavid Matlack static inline int tdp_mmu_zap_spte_atomic(struct kvm *kvm, 56908f07c80SBen Gardon struct tdp_iter *iter) 57008f07c80SBen Gardon { 571*3e72c791SDavid Matlack int ret; 572*3e72c791SDavid Matlack 57308f07c80SBen Gardon /* 57408f07c80SBen Gardon * Freeze the SPTE by setting it to a special, 57508f07c80SBen Gardon * non-present value. This will stop other threads from 57608f07c80SBen Gardon * immediately installing a present entry in its place 57708f07c80SBen Gardon * before the TLBs are flushed. 57808f07c80SBen Gardon */ 579*3e72c791SDavid Matlack ret = tdp_mmu_set_spte_atomic(kvm, iter, REMOVED_SPTE); 580*3e72c791SDavid Matlack if (ret) 581*3e72c791SDavid Matlack return ret; 58208f07c80SBen Gardon 58308f07c80SBen Gardon kvm_flush_remote_tlbs_with_address(kvm, iter->gfn, 58408f07c80SBen Gardon KVM_PAGES_PER_HPAGE(iter->level)); 58508f07c80SBen Gardon 58608f07c80SBen Gardon /* 58708f07c80SBen Gardon * No other thread can overwrite the removed SPTE as they 58808f07c80SBen Gardon * must either wait on the MMU lock or use 589d9f6e12fSIngo Molnar * tdp_mmu_set_spte_atomic which will not overwrite the 59008f07c80SBen Gardon * special removed SPTE value. No bookkeeping is needed 59108f07c80SBen Gardon * here since the SPTE is going from non-present 59208f07c80SBen Gardon * to non-present. 59308f07c80SBen Gardon */ 59414f6fec2SBen Gardon WRITE_ONCE(*rcu_dereference(iter->sptep), 0); 59508f07c80SBen Gardon 596*3e72c791SDavid Matlack return 0; 59708f07c80SBen Gardon } 59808f07c80SBen Gardon 5999a77daacSBen Gardon 6009a77daacSBen Gardon /* 601fe43fa2fSBen Gardon * __tdp_mmu_set_spte - Set a TDP MMU SPTE and handle the associated bookkeeping 602fe43fa2fSBen Gardon * @kvm: kvm instance 603fe43fa2fSBen Gardon * @iter: a tdp_iter instance currently on the SPTE that should be set 604fe43fa2fSBen Gardon * @new_spte: The value the SPTE should be set to 605fe43fa2fSBen Gardon * @record_acc_track: Notify the MM subsystem of changes to the accessed state 606fe43fa2fSBen Gardon * of the page. Should be set unless handling an MMU 607fe43fa2fSBen Gardon * notifier for access tracking. Leaving record_acc_track 608fe43fa2fSBen Gardon * unset in that case prevents page accesses from being 609fe43fa2fSBen Gardon * double counted. 610fe43fa2fSBen Gardon * @record_dirty_log: Record the page as dirty in the dirty bitmap if 611fe43fa2fSBen Gardon * appropriate for the change being made. Should be set 612fe43fa2fSBen Gardon * unless performing certain dirty logging operations. 613fe43fa2fSBen Gardon * Leaving record_dirty_log unset in that case prevents page 614fe43fa2fSBen Gardon * writes from being double counted. 615fe43fa2fSBen Gardon */ 616f8e14497SBen Gardon static inline void __tdp_mmu_set_spte(struct kvm *kvm, struct tdp_iter *iter, 617a6a0b05dSBen Gardon u64 new_spte, bool record_acc_track, 618a6a0b05dSBen Gardon bool record_dirty_log) 619faaf05b0SBen Gardon { 6203a0f64deSSean Christopherson WARN_ON_ONCE(iter->yielded); 6213a0f64deSSean Christopherson 622531810caSBen Gardon lockdep_assert_held_write(&kvm->mmu_lock); 6233a9a4aa5SBen Gardon 62408f07c80SBen Gardon /* 62508f07c80SBen Gardon * No thread should be using this function to set SPTEs to the 62608f07c80SBen Gardon * temporary removed SPTE value. 62708f07c80SBen Gardon * If operating under the MMU lock in read mode, tdp_mmu_set_spte_atomic 62808f07c80SBen Gardon * should be used. If operating under the MMU lock in write mode, the 62908f07c80SBen Gardon * use of the removed SPTE should not be necessary. 63008f07c80SBen Gardon */ 6317a51393aSSean Christopherson WARN_ON(is_removed_spte(iter->old_spte)); 63208f07c80SBen Gardon 6337cca2d0bSBen Gardon WRITE_ONCE(*rcu_dereference(iter->sptep), new_spte); 634faaf05b0SBen Gardon 63508889894SSean Christopherson __handle_changed_spte(kvm, iter->as_id, iter->gfn, iter->old_spte, 63608889894SSean Christopherson new_spte, iter->level, false); 637f8e14497SBen Gardon if (record_acc_track) 638f8e14497SBen Gardon handle_changed_spte_acc_track(iter->old_spte, new_spte, 639f8e14497SBen Gardon iter->level); 640a6a0b05dSBen Gardon if (record_dirty_log) 64108889894SSean Christopherson handle_changed_spte_dirty_log(kvm, iter->as_id, iter->gfn, 642a6a0b05dSBen Gardon iter->old_spte, new_spte, 643a6a0b05dSBen Gardon iter->level); 644f8e14497SBen Gardon } 645f8e14497SBen Gardon 646f8e14497SBen Gardon static inline void tdp_mmu_set_spte(struct kvm *kvm, struct tdp_iter *iter, 647f8e14497SBen Gardon u64 new_spte) 648f8e14497SBen Gardon { 649a6a0b05dSBen Gardon __tdp_mmu_set_spte(kvm, iter, new_spte, true, true); 650f8e14497SBen Gardon } 651f8e14497SBen Gardon 652f8e14497SBen Gardon static inline void tdp_mmu_set_spte_no_acc_track(struct kvm *kvm, 653f8e14497SBen Gardon struct tdp_iter *iter, 654f8e14497SBen Gardon u64 new_spte) 655f8e14497SBen Gardon { 656a6a0b05dSBen Gardon __tdp_mmu_set_spte(kvm, iter, new_spte, false, true); 657a6a0b05dSBen Gardon } 658a6a0b05dSBen Gardon 659a6a0b05dSBen Gardon static inline void tdp_mmu_set_spte_no_dirty_log(struct kvm *kvm, 660a6a0b05dSBen Gardon struct tdp_iter *iter, 661a6a0b05dSBen Gardon u64 new_spte) 662a6a0b05dSBen Gardon { 663a6a0b05dSBen Gardon __tdp_mmu_set_spte(kvm, iter, new_spte, true, false); 664faaf05b0SBen Gardon } 665faaf05b0SBen Gardon 666faaf05b0SBen Gardon #define tdp_root_for_each_pte(_iter, _root, _start, _end) \ 667faaf05b0SBen Gardon for_each_tdp_pte(_iter, _root->spt, _root->role.level, _start, _end) 668faaf05b0SBen Gardon 669f8e14497SBen Gardon #define tdp_root_for_each_leaf_pte(_iter, _root, _start, _end) \ 670f8e14497SBen Gardon tdp_root_for_each_pte(_iter, _root, _start, _end) \ 671f8e14497SBen Gardon if (!is_shadow_present_pte(_iter.old_spte) || \ 672f8e14497SBen Gardon !is_last_spte(_iter.old_spte, _iter.level)) \ 673f8e14497SBen Gardon continue; \ 674f8e14497SBen Gardon else 675f8e14497SBen Gardon 676bb18842eSBen Gardon #define tdp_mmu_for_each_pte(_iter, _mmu, _start, _end) \ 677bb18842eSBen Gardon for_each_tdp_pte(_iter, __va(_mmu->root_hpa), \ 678bb18842eSBen Gardon _mmu->shadow_root_level, _start, _end) 679bb18842eSBen Gardon 680faaf05b0SBen Gardon /* 681e28a436cSBen Gardon * Yield if the MMU lock is contended or this thread needs to return control 682e28a436cSBen Gardon * to the scheduler. 683e28a436cSBen Gardon * 684e139a34eSBen Gardon * If this function should yield and flush is set, it will perform a remote 685e139a34eSBen Gardon * TLB flush before yielding. 686e139a34eSBen Gardon * 6873a0f64deSSean Christopherson * If this function yields, iter->yielded is set and the caller must skip to 6883a0f64deSSean Christopherson * the next iteration, where tdp_iter_next() will reset the tdp_iter's walk 6893a0f64deSSean Christopherson * over the paging structures to allow the iterator to continue its traversal 6903a0f64deSSean Christopherson * from the paging structure root. 691e28a436cSBen Gardon * 6923a0f64deSSean Christopherson * Returns true if this function yielded. 693e28a436cSBen Gardon */ 6943a0f64deSSean Christopherson static inline bool __must_check tdp_mmu_iter_cond_resched(struct kvm *kvm, 6953a0f64deSSean Christopherson struct tdp_iter *iter, 6963a0f64deSSean Christopherson bool flush, bool shared) 697a6a0b05dSBen Gardon { 6983a0f64deSSean Christopherson WARN_ON(iter->yielded); 6993a0f64deSSean Christopherson 700ed5e484bSBen Gardon /* Ensure forward progress has been made before yielding. */ 701ed5e484bSBen Gardon if (iter->next_last_level_gfn == iter->yielded_gfn) 702ed5e484bSBen Gardon return false; 703ed5e484bSBen Gardon 704531810caSBen Gardon if (need_resched() || rwlock_needbreak(&kvm->mmu_lock)) { 7057cca2d0bSBen Gardon rcu_read_unlock(); 7067cca2d0bSBen Gardon 707e139a34eSBen Gardon if (flush) 708e139a34eSBen Gardon kvm_flush_remote_tlbs(kvm); 709e139a34eSBen Gardon 7106103bc07SBen Gardon if (shared) 7116103bc07SBen Gardon cond_resched_rwlock_read(&kvm->mmu_lock); 7126103bc07SBen Gardon else 713531810caSBen Gardon cond_resched_rwlock_write(&kvm->mmu_lock); 7146103bc07SBen Gardon 7157cca2d0bSBen Gardon rcu_read_lock(); 716ed5e484bSBen Gardon 717ed5e484bSBen Gardon WARN_ON(iter->gfn > iter->next_last_level_gfn); 718ed5e484bSBen Gardon 7193a0f64deSSean Christopherson iter->yielded = true; 720a6a0b05dSBen Gardon } 721e28a436cSBen Gardon 7223a0f64deSSean Christopherson return iter->yielded; 723a6a0b05dSBen Gardon } 724a6a0b05dSBen Gardon 725faaf05b0SBen Gardon /* 726faaf05b0SBen Gardon * Tears down the mappings for the range of gfns, [start, end), and frees the 727faaf05b0SBen Gardon * non-root pages mapping GFNs strictly within that range. Returns true if 728faaf05b0SBen Gardon * SPTEs have been cleared and a TLB flush is needed before releasing the 729faaf05b0SBen Gardon * MMU lock. 7306103bc07SBen Gardon * 731063afacdSBen Gardon * If can_yield is true, will release the MMU lock and reschedule if the 732063afacdSBen Gardon * scheduler needs the CPU or there is contention on the MMU lock. If this 733063afacdSBen Gardon * function cannot yield, it will not release the MMU lock or reschedule and 734063afacdSBen Gardon * the caller must ensure it does not supply too large a GFN range, or the 7356103bc07SBen Gardon * operation can cause a soft lockup. 7366103bc07SBen Gardon * 7376103bc07SBen Gardon * If shared is true, this thread holds the MMU lock in read mode and must 7386103bc07SBen Gardon * account for the possibility that other threads are modifying the paging 7396103bc07SBen Gardon * structures concurrently. If shared is false, this thread should hold the 7406103bc07SBen Gardon * MMU lock in write mode. 741faaf05b0SBen Gardon */ 742faaf05b0SBen Gardon static bool zap_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root, 7436103bc07SBen Gardon gfn_t start, gfn_t end, bool can_yield, bool flush, 7446103bc07SBen Gardon bool shared) 745faaf05b0SBen Gardon { 746524a1e4eSSean Christopherson gfn_t max_gfn_host = 1ULL << (shadow_phys_bits - PAGE_SHIFT); 747524a1e4eSSean Christopherson bool zap_all = (start == 0 && end >= max_gfn_host); 748faaf05b0SBen Gardon struct tdp_iter iter; 749faaf05b0SBen Gardon 750524a1e4eSSean Christopherson /* 7510103098fSSean Christopherson * No need to try to step down in the iterator when zapping all SPTEs, 7520103098fSSean Christopherson * zapping the top-level non-leaf SPTEs will recurse on their children. 7530103098fSSean Christopherson */ 7540103098fSSean Christopherson int min_level = zap_all ? root->role.level : PG_LEVEL_4K; 7550103098fSSean Christopherson 7560103098fSSean Christopherson /* 757524a1e4eSSean Christopherson * Bound the walk at host.MAXPHYADDR, guest accesses beyond that will 758524a1e4eSSean Christopherson * hit a #PF(RSVD) and never get to an EPT Violation/Misconfig / #NPF, 759524a1e4eSSean Christopherson * and so KVM will never install a SPTE for such addresses. 760524a1e4eSSean Christopherson */ 761524a1e4eSSean Christopherson end = min(end, max_gfn_host); 762524a1e4eSSean Christopherson 7636103bc07SBen Gardon kvm_lockdep_assert_mmu_lock_held(kvm, shared); 7646103bc07SBen Gardon 7657cca2d0bSBen Gardon rcu_read_lock(); 7667cca2d0bSBen Gardon 7670103098fSSean Christopherson for_each_tdp_pte_min_level(iter, root->spt, root->role.level, 7680103098fSSean Christopherson min_level, start, end) { 7696103bc07SBen Gardon retry: 7701af4a960SBen Gardon if (can_yield && 7716103bc07SBen Gardon tdp_mmu_iter_cond_resched(kvm, &iter, flush, shared)) { 772a835429cSSean Christopherson flush = false; 7731af4a960SBen Gardon continue; 7741af4a960SBen Gardon } 7751af4a960SBen Gardon 776faaf05b0SBen Gardon if (!is_shadow_present_pte(iter.old_spte)) 777faaf05b0SBen Gardon continue; 778faaf05b0SBen Gardon 779faaf05b0SBen Gardon /* 780faaf05b0SBen Gardon * If this is a non-last-level SPTE that covers a larger range 781faaf05b0SBen Gardon * than should be zapped, continue, and zap the mappings at a 782524a1e4eSSean Christopherson * lower level, except when zapping all SPTEs. 783faaf05b0SBen Gardon */ 784524a1e4eSSean Christopherson if (!zap_all && 785524a1e4eSSean Christopherson (iter.gfn < start || 786faaf05b0SBen Gardon iter.gfn + KVM_PAGES_PER_HPAGE(iter.level) > end) && 787faaf05b0SBen Gardon !is_last_spte(iter.old_spte, iter.level)) 788faaf05b0SBen Gardon continue; 789faaf05b0SBen Gardon 7906103bc07SBen Gardon if (!shared) { 791faaf05b0SBen Gardon tdp_mmu_set_spte(kvm, &iter, 0); 792a835429cSSean Christopherson flush = true; 793*3e72c791SDavid Matlack } else if (tdp_mmu_zap_spte_atomic(kvm, &iter)) { 7946103bc07SBen Gardon goto retry; 7956103bc07SBen Gardon } 796faaf05b0SBen Gardon } 7977cca2d0bSBen Gardon 7987cca2d0bSBen Gardon rcu_read_unlock(); 799a835429cSSean Christopherson return flush; 800faaf05b0SBen Gardon } 801faaf05b0SBen Gardon 802faaf05b0SBen Gardon /* 803faaf05b0SBen Gardon * Tears down the mappings for the range of gfns, [start, end), and frees the 804faaf05b0SBen Gardon * non-root pages mapping GFNs strictly within that range. Returns true if 805faaf05b0SBen Gardon * SPTEs have been cleared and a TLB flush is needed before releasing the 806faaf05b0SBen Gardon * MMU lock. 807faaf05b0SBen Gardon */ 8082b9663d8SSean Christopherson bool __kvm_tdp_mmu_zap_gfn_range(struct kvm *kvm, int as_id, gfn_t start, 8095a324c24SSean Christopherson gfn_t end, bool can_yield, bool flush) 810faaf05b0SBen Gardon { 811faaf05b0SBen Gardon struct kvm_mmu_page *root; 812faaf05b0SBen Gardon 8135a324c24SSean Christopherson for_each_tdp_mmu_root_yield_safe(kvm, root, as_id, false) 8146103bc07SBen Gardon flush = zap_gfn_range(kvm, root, start, end, can_yield, flush, 8155a324c24SSean Christopherson false); 816faaf05b0SBen Gardon 817faaf05b0SBen Gardon return flush; 818faaf05b0SBen Gardon } 819faaf05b0SBen Gardon 820faaf05b0SBen Gardon void kvm_tdp_mmu_zap_all(struct kvm *kvm) 821faaf05b0SBen Gardon { 8222b9663d8SSean Christopherson bool flush = false; 8232b9663d8SSean Christopherson int i; 824faaf05b0SBen Gardon 8252b9663d8SSean Christopherson for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) 8265a324c24SSean Christopherson flush = kvm_tdp_mmu_zap_gfn_range(kvm, i, 0, -1ull, flush); 8272b9663d8SSean Christopherson 828faaf05b0SBen Gardon if (flush) 829faaf05b0SBen Gardon kvm_flush_remote_tlbs(kvm); 830faaf05b0SBen Gardon } 831bb18842eSBen Gardon 8324c6654bdSBen Gardon static struct kvm_mmu_page *next_invalidated_root(struct kvm *kvm, 8334c6654bdSBen Gardon struct kvm_mmu_page *prev_root) 8344c6654bdSBen Gardon { 8354c6654bdSBen Gardon struct kvm_mmu_page *next_root; 8364c6654bdSBen Gardon 8374c6654bdSBen Gardon if (prev_root) 8384c6654bdSBen Gardon next_root = list_next_or_null_rcu(&kvm->arch.tdp_mmu_roots, 8394c6654bdSBen Gardon &prev_root->link, 8404c6654bdSBen Gardon typeof(*prev_root), link); 8414c6654bdSBen Gardon else 8424c6654bdSBen Gardon next_root = list_first_or_null_rcu(&kvm->arch.tdp_mmu_roots, 8434c6654bdSBen Gardon typeof(*next_root), link); 8444c6654bdSBen Gardon 8454c6654bdSBen Gardon while (next_root && !(next_root->role.invalid && 8464c6654bdSBen Gardon refcount_read(&next_root->tdp_mmu_root_count))) 8474c6654bdSBen Gardon next_root = list_next_or_null_rcu(&kvm->arch.tdp_mmu_roots, 8484c6654bdSBen Gardon &next_root->link, 8494c6654bdSBen Gardon typeof(*next_root), link); 8504c6654bdSBen Gardon 8514c6654bdSBen Gardon return next_root; 8524c6654bdSBen Gardon } 8534c6654bdSBen Gardon 8544c6654bdSBen Gardon /* 8554c6654bdSBen Gardon * Since kvm_tdp_mmu_zap_all_fast has acquired a reference to each 8564c6654bdSBen Gardon * invalidated root, they will not be freed until this function drops the 8574c6654bdSBen Gardon * reference. Before dropping that reference, tear down the paging 8584c6654bdSBen Gardon * structure so that whichever thread does drop the last reference 8594c6654bdSBen Gardon * only has to do a trivial amount of work. Since the roots are invalid, 8604c6654bdSBen Gardon * no new SPTEs should be created under them. 8614c6654bdSBen Gardon */ 8624c6654bdSBen Gardon void kvm_tdp_mmu_zap_invalidated_roots(struct kvm *kvm) 8634c6654bdSBen Gardon { 8644c6654bdSBen Gardon struct kvm_mmu_page *next_root; 8654c6654bdSBen Gardon struct kvm_mmu_page *root; 8664c6654bdSBen Gardon bool flush = false; 8674c6654bdSBen Gardon 8684c6654bdSBen Gardon lockdep_assert_held_read(&kvm->mmu_lock); 8694c6654bdSBen Gardon 8704c6654bdSBen Gardon rcu_read_lock(); 8714c6654bdSBen Gardon 8724c6654bdSBen Gardon root = next_invalidated_root(kvm, NULL); 8734c6654bdSBen Gardon 8744c6654bdSBen Gardon while (root) { 8754c6654bdSBen Gardon next_root = next_invalidated_root(kvm, root); 8764c6654bdSBen Gardon 8774c6654bdSBen Gardon rcu_read_unlock(); 8784c6654bdSBen Gardon 879524a1e4eSSean Christopherson flush = zap_gfn_range(kvm, root, 0, -1ull, true, flush, true); 8804c6654bdSBen Gardon 8814c6654bdSBen Gardon /* 8824c6654bdSBen Gardon * Put the reference acquired in 8834c6654bdSBen Gardon * kvm_tdp_mmu_invalidate_roots 8844c6654bdSBen Gardon */ 8854c6654bdSBen Gardon kvm_tdp_mmu_put_root(kvm, root, true); 8864c6654bdSBen Gardon 8874c6654bdSBen Gardon root = next_root; 8884c6654bdSBen Gardon 8894c6654bdSBen Gardon rcu_read_lock(); 8904c6654bdSBen Gardon } 8914c6654bdSBen Gardon 8924c6654bdSBen Gardon rcu_read_unlock(); 8934c6654bdSBen Gardon 8944c6654bdSBen Gardon if (flush) 8954c6654bdSBen Gardon kvm_flush_remote_tlbs(kvm); 8964c6654bdSBen Gardon } 8974c6654bdSBen Gardon 898bb18842eSBen Gardon /* 899b7cccd39SBen Gardon * Mark each TDP MMU root as invalid so that other threads 900b7cccd39SBen Gardon * will drop their references and allow the root count to 901b7cccd39SBen Gardon * go to 0. 902b7cccd39SBen Gardon * 9034c6654bdSBen Gardon * Also take a reference on all roots so that this thread 9044c6654bdSBen Gardon * can do the bulk of the work required to free the roots 9054c6654bdSBen Gardon * once they are invalidated. Without this reference, a 9064c6654bdSBen Gardon * vCPU thread might drop the last reference to a root and 9074c6654bdSBen Gardon * get stuck with tearing down the entire paging structure. 9084c6654bdSBen Gardon * 9094c6654bdSBen Gardon * Roots which have a zero refcount should be skipped as 9104c6654bdSBen Gardon * they're already being torn down. 9114c6654bdSBen Gardon * Already invalid roots should be referenced again so that 9124c6654bdSBen Gardon * they aren't freed before kvm_tdp_mmu_zap_all_fast is 9134c6654bdSBen Gardon * done with them. 9144c6654bdSBen Gardon * 915b7cccd39SBen Gardon * This has essentially the same effect for the TDP MMU 916b7cccd39SBen Gardon * as updating mmu_valid_gen does for the shadow MMU. 917b7cccd39SBen Gardon */ 918b7cccd39SBen Gardon void kvm_tdp_mmu_invalidate_all_roots(struct kvm *kvm) 919b7cccd39SBen Gardon { 920b7cccd39SBen Gardon struct kvm_mmu_page *root; 921b7cccd39SBen Gardon 922b7cccd39SBen Gardon lockdep_assert_held_write(&kvm->mmu_lock); 923b7cccd39SBen Gardon list_for_each_entry(root, &kvm->arch.tdp_mmu_roots, link) 9244c6654bdSBen Gardon if (refcount_inc_not_zero(&root->tdp_mmu_root_count)) 925b7cccd39SBen Gardon root->role.invalid = true; 926b7cccd39SBen Gardon } 927b7cccd39SBen Gardon 928bb18842eSBen Gardon /* 929bb18842eSBen Gardon * Installs a last-level SPTE to handle a TDP page fault. 930bb18842eSBen Gardon * (NPT/EPT violation/misconfiguration) 931bb18842eSBen Gardon */ 932cdc47767SPaolo Bonzini static int tdp_mmu_map_handle_target_level(struct kvm_vcpu *vcpu, 933cdc47767SPaolo Bonzini struct kvm_page_fault *fault, 934cdc47767SPaolo Bonzini struct tdp_iter *iter) 935bb18842eSBen Gardon { 936c435d4b7SSean Christopherson struct kvm_mmu_page *sp = sptep_to_sp(rcu_dereference(iter->sptep)); 937bb18842eSBen Gardon u64 new_spte; 93857a3e96dSKai Huang int ret = RET_PF_FIXED; 939ad67e480SPaolo Bonzini bool wrprot = false; 940bb18842eSBen Gardon 9417158bee4SPaolo Bonzini WARN_ON(sp->role.level != fault->goal_level); 942e710c5f6SDavid Matlack if (unlikely(!fault->slot)) 943bb18842eSBen Gardon new_spte = make_mmio_spte(vcpu, iter->gfn, ACC_ALL); 9449a77daacSBen Gardon else 94553597858SDavid Matlack wrprot = make_spte(vcpu, sp, fault->slot, ACC_ALL, iter->gfn, 9462839180cSPaolo Bonzini fault->pfn, iter->old_spte, fault->prefetch, true, 9477158bee4SPaolo Bonzini fault->map_writable, &new_spte); 948bb18842eSBen Gardon 949bb18842eSBen Gardon if (new_spte == iter->old_spte) 950bb18842eSBen Gardon ret = RET_PF_SPURIOUS; 951*3e72c791SDavid Matlack else if (tdp_mmu_set_spte_atomic(vcpu->kvm, iter, new_spte)) 9529a77daacSBen Gardon return RET_PF_RETRY; 953bb18842eSBen Gardon 954bb18842eSBen Gardon /* 955bb18842eSBen Gardon * If the page fault was caused by a write but the page is write 956bb18842eSBen Gardon * protected, emulation is needed. If the emulation was skipped, 957bb18842eSBen Gardon * the vCPU would have the same fault again. 958bb18842eSBen Gardon */ 959ad67e480SPaolo Bonzini if (wrprot) { 960cdc47767SPaolo Bonzini if (fault->write) 961bb18842eSBen Gardon ret = RET_PF_EMULATE; 962bb18842eSBen Gardon } 963bb18842eSBen Gardon 964bb18842eSBen Gardon /* If a MMIO SPTE is installed, the MMIO will need to be emulated. */ 9659a77daacSBen Gardon if (unlikely(is_mmio_spte(new_spte))) { 9669a77daacSBen Gardon trace_mark_mmio_spte(rcu_dereference(iter->sptep), iter->gfn, 9679a77daacSBen Gardon new_spte); 968bb18842eSBen Gardon ret = RET_PF_EMULATE; 9693849e092SSean Christopherson } else { 9709a77daacSBen Gardon trace_kvm_mmu_set_spte(iter->level, iter->gfn, 9719a77daacSBen Gardon rcu_dereference(iter->sptep)); 9723849e092SSean Christopherson } 973bb18842eSBen Gardon 974857f8474SKai Huang /* 975857f8474SKai Huang * Increase pf_fixed in both RET_PF_EMULATE and RET_PF_FIXED to be 976857f8474SKai Huang * consistent with legacy MMU behavior. 977857f8474SKai Huang */ 978857f8474SKai Huang if (ret != RET_PF_SPURIOUS) 979bb18842eSBen Gardon vcpu->stat.pf_fixed++; 980bb18842eSBen Gardon 981bb18842eSBen Gardon return ret; 982bb18842eSBen Gardon } 983bb18842eSBen Gardon 984bb18842eSBen Gardon /* 985bb18842eSBen Gardon * Handle a TDP page fault (NPT/EPT violation/misconfiguration) by installing 986bb18842eSBen Gardon * page tables and SPTEs to translate the faulting guest physical address. 987bb18842eSBen Gardon */ 9882f6305ddSPaolo Bonzini int kvm_tdp_mmu_map(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault) 989bb18842eSBen Gardon { 990bb18842eSBen Gardon struct kvm_mmu *mmu = vcpu->arch.mmu; 991bb18842eSBen Gardon struct tdp_iter iter; 99289c0fd49SBen Gardon struct kvm_mmu_page *sp; 993bb18842eSBen Gardon u64 *child_pt; 994bb18842eSBen Gardon u64 new_spte; 995bb18842eSBen Gardon int ret; 996bb18842eSBen Gardon 99773a3c659SPaolo Bonzini kvm_mmu_hugepage_adjust(vcpu, fault); 998bb18842eSBen Gardon 999f0066d94SPaolo Bonzini trace_kvm_mmu_spte_requested(fault); 10007cca2d0bSBen Gardon 10017cca2d0bSBen Gardon rcu_read_lock(); 10027cca2d0bSBen Gardon 10032f6305ddSPaolo Bonzini tdp_mmu_for_each_pte(iter, mmu, fault->gfn, fault->gfn + 1) { 100473a3c659SPaolo Bonzini if (fault->nx_huge_page_workaround_enabled) 1005536f0e6aSPaolo Bonzini disallowed_hugepage_adjust(fault, iter.old_spte, iter.level); 1006bb18842eSBen Gardon 100773a3c659SPaolo Bonzini if (iter.level == fault->goal_level) 1008bb18842eSBen Gardon break; 1009bb18842eSBen Gardon 1010bb18842eSBen Gardon /* 1011bb18842eSBen Gardon * If there is an SPTE mapping a large page at a higher level 1012bb18842eSBen Gardon * than the target, that SPTE must be cleared and replaced 1013bb18842eSBen Gardon * with a non-leaf SPTE. 1014bb18842eSBen Gardon */ 1015bb18842eSBen Gardon if (is_shadow_present_pte(iter.old_spte) && 1016bb18842eSBen Gardon is_large_pte(iter.old_spte)) { 1017*3e72c791SDavid Matlack if (tdp_mmu_zap_spte_atomic(vcpu->kvm, &iter)) 10189a77daacSBen Gardon break; 1019bb18842eSBen Gardon 1020bb18842eSBen Gardon /* 1021bb18842eSBen Gardon * The iter must explicitly re-read the spte here 1022bb18842eSBen Gardon * because the new value informs the !present 1023bb18842eSBen Gardon * path below. 1024bb18842eSBen Gardon */ 10257cca2d0bSBen Gardon iter.old_spte = READ_ONCE(*rcu_dereference(iter.sptep)); 1026bb18842eSBen Gardon } 1027bb18842eSBen Gardon 1028bb18842eSBen Gardon if (!is_shadow_present_pte(iter.old_spte)) { 1029ff76d506SKai Huang /* 1030c4342633SIngo Molnar * If SPTE has been frozen by another thread, just 1031ff76d506SKai Huang * give up and retry, avoiding unnecessary page table 1032ff76d506SKai Huang * allocation and free. 1033ff76d506SKai Huang */ 1034ff76d506SKai Huang if (is_removed_spte(iter.old_spte)) 1035ff76d506SKai Huang break; 1036ff76d506SKai Huang 1037f1b83255SKai Huang sp = alloc_tdp_mmu_page(vcpu, iter.gfn, iter.level - 1); 103889c0fd49SBen Gardon child_pt = sp->spt; 1039a9442f59SBen Gardon 1040bb18842eSBen Gardon new_spte = make_nonleaf_spte(child_pt, 1041bb18842eSBen Gardon !shadow_accessed_mask); 1042bb18842eSBen Gardon 1043*3e72c791SDavid Matlack if (!tdp_mmu_set_spte_atomic(vcpu->kvm, &iter, new_spte)) { 10449653f2daSSean Christopherson tdp_mmu_link_page(vcpu->kvm, sp, 104573a3c659SPaolo Bonzini fault->huge_page_disallowed && 104673a3c659SPaolo Bonzini fault->req_level >= iter.level); 10479a77daacSBen Gardon 1048bb18842eSBen Gardon trace_kvm_mmu_get_page(sp, true); 10499a77daacSBen Gardon } else { 10509a77daacSBen Gardon tdp_mmu_free_sp(sp); 10519a77daacSBen Gardon break; 10529a77daacSBen Gardon } 1053bb18842eSBen Gardon } 1054bb18842eSBen Gardon } 1055bb18842eSBen Gardon 105673a3c659SPaolo Bonzini if (iter.level != fault->goal_level) { 10577cca2d0bSBen Gardon rcu_read_unlock(); 1058bb18842eSBen Gardon return RET_PF_RETRY; 10597cca2d0bSBen Gardon } 1060bb18842eSBen Gardon 1061cdc47767SPaolo Bonzini ret = tdp_mmu_map_handle_target_level(vcpu, fault, &iter); 10627cca2d0bSBen Gardon rcu_read_unlock(); 1063bb18842eSBen Gardon 1064bb18842eSBen Gardon return ret; 1065bb18842eSBen Gardon } 1066063afacdSBen Gardon 10673039bcc7SSean Christopherson bool kvm_tdp_mmu_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range, 10683039bcc7SSean Christopherson bool flush) 1069063afacdSBen Gardon { 107083b83a02SSean Christopherson return __kvm_tdp_mmu_zap_gfn_range(kvm, range->slot->as_id, range->start, 107183b83a02SSean Christopherson range->end, range->may_block, flush); 10723039bcc7SSean Christopherson } 10733039bcc7SSean Christopherson 10743039bcc7SSean Christopherson typedef bool (*tdp_handler_t)(struct kvm *kvm, struct tdp_iter *iter, 10753039bcc7SSean Christopherson struct kvm_gfn_range *range); 10763039bcc7SSean Christopherson 10773039bcc7SSean Christopherson static __always_inline bool kvm_tdp_mmu_handle_gfn(struct kvm *kvm, 10783039bcc7SSean Christopherson struct kvm_gfn_range *range, 1079c1b91493SSean Christopherson tdp_handler_t handler) 1080063afacdSBen Gardon { 1081063afacdSBen Gardon struct kvm_mmu_page *root; 10823039bcc7SSean Christopherson struct tdp_iter iter; 10833039bcc7SSean Christopherson bool ret = false; 1084063afacdSBen Gardon 10853039bcc7SSean Christopherson rcu_read_lock(); 1086063afacdSBen Gardon 1087063afacdSBen Gardon /* 1088e1eed584SSean Christopherson * Don't support rescheduling, none of the MMU notifiers that funnel 1089e1eed584SSean Christopherson * into this helper allow blocking; it'd be dead, wasteful code. 1090063afacdSBen Gardon */ 10913039bcc7SSean Christopherson for_each_tdp_mmu_root(kvm, root, range->slot->as_id) { 10923039bcc7SSean Christopherson tdp_root_for_each_leaf_pte(iter, root, range->start, range->end) 10933039bcc7SSean Christopherson ret |= handler(kvm, &iter, range); 10943039bcc7SSean Christopherson } 1095063afacdSBen Gardon 10963039bcc7SSean Christopherson rcu_read_unlock(); 1097063afacdSBen Gardon 1098063afacdSBen Gardon return ret; 1099063afacdSBen Gardon } 1100063afacdSBen Gardon 1101f8e14497SBen Gardon /* 1102f8e14497SBen Gardon * Mark the SPTEs range of GFNs [start, end) unaccessed and return non-zero 1103f8e14497SBen Gardon * if any of the GFNs in the range have been accessed. 1104f8e14497SBen Gardon */ 11053039bcc7SSean Christopherson static bool age_gfn_range(struct kvm *kvm, struct tdp_iter *iter, 11063039bcc7SSean Christopherson struct kvm_gfn_range *range) 1107f8e14497SBen Gardon { 1108f8e14497SBen Gardon u64 new_spte = 0; 1109f8e14497SBen Gardon 11103039bcc7SSean Christopherson /* If we have a non-accessed entry we don't need to change the pte. */ 11113039bcc7SSean Christopherson if (!is_accessed_spte(iter->old_spte)) 11123039bcc7SSean Christopherson return false; 11137cca2d0bSBen Gardon 11143039bcc7SSean Christopherson new_spte = iter->old_spte; 1115f8e14497SBen Gardon 1116f8e14497SBen Gardon if (spte_ad_enabled(new_spte)) { 11178f8f52a4SSean Christopherson new_spte &= ~shadow_accessed_mask; 1118f8e14497SBen Gardon } else { 1119f8e14497SBen Gardon /* 1120f8e14497SBen Gardon * Capture the dirty status of the page, so that it doesn't get 1121f8e14497SBen Gardon * lost when the SPTE is marked for access tracking. 1122f8e14497SBen Gardon */ 1123f8e14497SBen Gardon if (is_writable_pte(new_spte)) 1124f8e14497SBen Gardon kvm_set_pfn_dirty(spte_to_pfn(new_spte)); 1125f8e14497SBen Gardon 1126f8e14497SBen Gardon new_spte = mark_spte_for_access_track(new_spte); 1127f8e14497SBen Gardon } 1128f8e14497SBen Gardon 11293039bcc7SSean Christopherson tdp_mmu_set_spte_no_acc_track(kvm, iter, new_spte); 113033dd3574SBen Gardon 11313039bcc7SSean Christopherson return true; 1132f8e14497SBen Gardon } 1133f8e14497SBen Gardon 11343039bcc7SSean Christopherson bool kvm_tdp_mmu_age_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range) 1135f8e14497SBen Gardon { 11363039bcc7SSean Christopherson return kvm_tdp_mmu_handle_gfn(kvm, range, age_gfn_range); 1137f8e14497SBen Gardon } 1138f8e14497SBen Gardon 11393039bcc7SSean Christopherson static bool test_age_gfn(struct kvm *kvm, struct tdp_iter *iter, 11403039bcc7SSean Christopherson struct kvm_gfn_range *range) 1141f8e14497SBen Gardon { 11423039bcc7SSean Christopherson return is_accessed_spte(iter->old_spte); 1143f8e14497SBen Gardon } 1144f8e14497SBen Gardon 11453039bcc7SSean Christopherson bool kvm_tdp_mmu_test_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range) 1146f8e14497SBen Gardon { 11473039bcc7SSean Christopherson return kvm_tdp_mmu_handle_gfn(kvm, range, test_age_gfn); 11483039bcc7SSean Christopherson } 11493039bcc7SSean Christopherson 11503039bcc7SSean Christopherson static bool set_spte_gfn(struct kvm *kvm, struct tdp_iter *iter, 11513039bcc7SSean Christopherson struct kvm_gfn_range *range) 11523039bcc7SSean Christopherson { 11533039bcc7SSean Christopherson u64 new_spte; 11543039bcc7SSean Christopherson 11553039bcc7SSean Christopherson /* Huge pages aren't expected to be modified without first being zapped. */ 11563039bcc7SSean Christopherson WARN_ON(pte_huge(range->pte) || range->start + 1 != range->end); 11573039bcc7SSean Christopherson 11583039bcc7SSean Christopherson if (iter->level != PG_LEVEL_4K || 11593039bcc7SSean Christopherson !is_shadow_present_pte(iter->old_spte)) 11603039bcc7SSean Christopherson return false; 11613039bcc7SSean Christopherson 11623039bcc7SSean Christopherson /* 11633039bcc7SSean Christopherson * Note, when changing a read-only SPTE, it's not strictly necessary to 11643039bcc7SSean Christopherson * zero the SPTE before setting the new PFN, but doing so preserves the 11653039bcc7SSean Christopherson * invariant that the PFN of a present * leaf SPTE can never change. 11663039bcc7SSean Christopherson * See __handle_changed_spte(). 11673039bcc7SSean Christopherson */ 11683039bcc7SSean Christopherson tdp_mmu_set_spte(kvm, iter, 0); 11693039bcc7SSean Christopherson 11703039bcc7SSean Christopherson if (!pte_write(range->pte)) { 11713039bcc7SSean Christopherson new_spte = kvm_mmu_changed_pte_notifier_make_spte(iter->old_spte, 11723039bcc7SSean Christopherson pte_pfn(range->pte)); 11733039bcc7SSean Christopherson 11743039bcc7SSean Christopherson tdp_mmu_set_spte(kvm, iter, new_spte); 11753039bcc7SSean Christopherson } 11763039bcc7SSean Christopherson 11773039bcc7SSean Christopherson return true; 1178f8e14497SBen Gardon } 11791d8dd6b3SBen Gardon 11801d8dd6b3SBen Gardon /* 11811d8dd6b3SBen Gardon * Handle the changed_pte MMU notifier for the TDP MMU. 11821d8dd6b3SBen Gardon * data is a pointer to the new pte_t mapping the HVA specified by the MMU 11831d8dd6b3SBen Gardon * notifier. 11841d8dd6b3SBen Gardon * Returns non-zero if a flush is needed before releasing the MMU lock. 11851d8dd6b3SBen Gardon */ 11863039bcc7SSean Christopherson bool kvm_tdp_mmu_set_spte_gfn(struct kvm *kvm, struct kvm_gfn_range *range) 11871d8dd6b3SBen Gardon { 11883039bcc7SSean Christopherson bool flush = kvm_tdp_mmu_handle_gfn(kvm, range, set_spte_gfn); 11891d8dd6b3SBen Gardon 11903039bcc7SSean Christopherson /* FIXME: return 'flush' instead of flushing here. */ 11913039bcc7SSean Christopherson if (flush) 11923039bcc7SSean Christopherson kvm_flush_remote_tlbs_with_address(kvm, range->start, 1); 11937cca2d0bSBen Gardon 11943039bcc7SSean Christopherson return false; 11951d8dd6b3SBen Gardon } 11961d8dd6b3SBen Gardon 1197a6a0b05dSBen Gardon /* 1198bedd9195SDavid Matlack * Remove write access from all SPTEs at or above min_level that map GFNs 1199bedd9195SDavid Matlack * [start, end). Returns true if an SPTE has been changed and the TLBs need to 1200bedd9195SDavid Matlack * be flushed. 1201a6a0b05dSBen Gardon */ 1202a6a0b05dSBen Gardon static bool wrprot_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root, 1203a6a0b05dSBen Gardon gfn_t start, gfn_t end, int min_level) 1204a6a0b05dSBen Gardon { 1205a6a0b05dSBen Gardon struct tdp_iter iter; 1206a6a0b05dSBen Gardon u64 new_spte; 1207a6a0b05dSBen Gardon bool spte_set = false; 1208a6a0b05dSBen Gardon 12097cca2d0bSBen Gardon rcu_read_lock(); 12107cca2d0bSBen Gardon 1211a6a0b05dSBen Gardon BUG_ON(min_level > KVM_MAX_HUGEPAGE_LEVEL); 1212a6a0b05dSBen Gardon 1213a6a0b05dSBen Gardon for_each_tdp_pte_min_level(iter, root->spt, root->role.level, 1214a6a0b05dSBen Gardon min_level, start, end) { 121524ae4cfaSBen Gardon retry: 121624ae4cfaSBen Gardon if (tdp_mmu_iter_cond_resched(kvm, &iter, false, true)) 12171af4a960SBen Gardon continue; 12181af4a960SBen Gardon 1219a6a0b05dSBen Gardon if (!is_shadow_present_pte(iter.old_spte) || 12200f99ee2cSBen Gardon !is_last_spte(iter.old_spte, iter.level) || 12210f99ee2cSBen Gardon !(iter.old_spte & PT_WRITABLE_MASK)) 1222a6a0b05dSBen Gardon continue; 1223a6a0b05dSBen Gardon 1224a6a0b05dSBen Gardon new_spte = iter.old_spte & ~PT_WRITABLE_MASK; 1225a6a0b05dSBen Gardon 1226*3e72c791SDavid Matlack if (tdp_mmu_set_spte_atomic(kvm, &iter, new_spte)) 122724ae4cfaSBen Gardon goto retry; 12283255530aSDavid Matlack 1229a6a0b05dSBen Gardon spte_set = true; 1230a6a0b05dSBen Gardon } 12317cca2d0bSBen Gardon 12327cca2d0bSBen Gardon rcu_read_unlock(); 1233a6a0b05dSBen Gardon return spte_set; 1234a6a0b05dSBen Gardon } 1235a6a0b05dSBen Gardon 1236a6a0b05dSBen Gardon /* 1237a6a0b05dSBen Gardon * Remove write access from all the SPTEs mapping GFNs in the memslot. Will 1238a6a0b05dSBen Gardon * only affect leaf SPTEs down to min_level. 1239a6a0b05dSBen Gardon * Returns true if an SPTE has been changed and the TLBs need to be flushed. 1240a6a0b05dSBen Gardon */ 1241269e9552SHamza Mahfooz bool kvm_tdp_mmu_wrprot_slot(struct kvm *kvm, 1242269e9552SHamza Mahfooz const struct kvm_memory_slot *slot, int min_level) 1243a6a0b05dSBen Gardon { 1244a6a0b05dSBen Gardon struct kvm_mmu_page *root; 1245a6a0b05dSBen Gardon bool spte_set = false; 1246a6a0b05dSBen Gardon 124724ae4cfaSBen Gardon lockdep_assert_held_read(&kvm->mmu_lock); 1248a6a0b05dSBen Gardon 1249d62007edSSean Christopherson for_each_valid_tdp_mmu_root_yield_safe(kvm, root, slot->as_id, true) 1250a6a0b05dSBen Gardon spte_set |= wrprot_gfn_range(kvm, root, slot->base_gfn, 1251a6a0b05dSBen Gardon slot->base_gfn + slot->npages, min_level); 1252a6a0b05dSBen Gardon 1253a6a0b05dSBen Gardon return spte_set; 1254a6a0b05dSBen Gardon } 1255a6a0b05dSBen Gardon 1256a6a0b05dSBen Gardon /* 1257a6a0b05dSBen Gardon * Clear the dirty status of all the SPTEs mapping GFNs in the memslot. If 1258a6a0b05dSBen Gardon * AD bits are enabled, this will involve clearing the dirty bit on each SPTE. 1259a6a0b05dSBen Gardon * If AD bits are not enabled, this will require clearing the writable bit on 1260a6a0b05dSBen Gardon * each SPTE. Returns true if an SPTE has been changed and the TLBs need to 1261a6a0b05dSBen Gardon * be flushed. 1262a6a0b05dSBen Gardon */ 1263a6a0b05dSBen Gardon static bool clear_dirty_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root, 1264a6a0b05dSBen Gardon gfn_t start, gfn_t end) 1265a6a0b05dSBen Gardon { 1266a6a0b05dSBen Gardon struct tdp_iter iter; 1267a6a0b05dSBen Gardon u64 new_spte; 1268a6a0b05dSBen Gardon bool spte_set = false; 1269a6a0b05dSBen Gardon 12707cca2d0bSBen Gardon rcu_read_lock(); 12717cca2d0bSBen Gardon 1272a6a0b05dSBen Gardon tdp_root_for_each_leaf_pte(iter, root, start, end) { 127324ae4cfaSBen Gardon retry: 127424ae4cfaSBen Gardon if (tdp_mmu_iter_cond_resched(kvm, &iter, false, true)) 12751af4a960SBen Gardon continue; 12761af4a960SBen Gardon 1277a6a0b05dSBen Gardon if (spte_ad_need_write_protect(iter.old_spte)) { 1278a6a0b05dSBen Gardon if (is_writable_pte(iter.old_spte)) 1279a6a0b05dSBen Gardon new_spte = iter.old_spte & ~PT_WRITABLE_MASK; 1280a6a0b05dSBen Gardon else 1281a6a0b05dSBen Gardon continue; 1282a6a0b05dSBen Gardon } else { 1283a6a0b05dSBen Gardon if (iter.old_spte & shadow_dirty_mask) 1284a6a0b05dSBen Gardon new_spte = iter.old_spte & ~shadow_dirty_mask; 1285a6a0b05dSBen Gardon else 1286a6a0b05dSBen Gardon continue; 1287a6a0b05dSBen Gardon } 1288a6a0b05dSBen Gardon 1289*3e72c791SDavid Matlack if (tdp_mmu_set_spte_atomic(kvm, &iter, new_spte)) 129024ae4cfaSBen Gardon goto retry; 12913255530aSDavid Matlack 1292a6a0b05dSBen Gardon spte_set = true; 1293a6a0b05dSBen Gardon } 12947cca2d0bSBen Gardon 12957cca2d0bSBen Gardon rcu_read_unlock(); 1296a6a0b05dSBen Gardon return spte_set; 1297a6a0b05dSBen Gardon } 1298a6a0b05dSBen Gardon 1299a6a0b05dSBen Gardon /* 1300a6a0b05dSBen Gardon * Clear the dirty status of all the SPTEs mapping GFNs in the memslot. If 1301a6a0b05dSBen Gardon * AD bits are enabled, this will involve clearing the dirty bit on each SPTE. 1302a6a0b05dSBen Gardon * If AD bits are not enabled, this will require clearing the writable bit on 1303a6a0b05dSBen Gardon * each SPTE. Returns true if an SPTE has been changed and the TLBs need to 1304a6a0b05dSBen Gardon * be flushed. 1305a6a0b05dSBen Gardon */ 1306269e9552SHamza Mahfooz bool kvm_tdp_mmu_clear_dirty_slot(struct kvm *kvm, 1307269e9552SHamza Mahfooz const struct kvm_memory_slot *slot) 1308a6a0b05dSBen Gardon { 1309a6a0b05dSBen Gardon struct kvm_mmu_page *root; 1310a6a0b05dSBen Gardon bool spte_set = false; 1311a6a0b05dSBen Gardon 131224ae4cfaSBen Gardon lockdep_assert_held_read(&kvm->mmu_lock); 1313a6a0b05dSBen Gardon 1314d62007edSSean Christopherson for_each_valid_tdp_mmu_root_yield_safe(kvm, root, slot->as_id, true) 1315a6a0b05dSBen Gardon spte_set |= clear_dirty_gfn_range(kvm, root, slot->base_gfn, 1316a6a0b05dSBen Gardon slot->base_gfn + slot->npages); 1317a6a0b05dSBen Gardon 1318a6a0b05dSBen Gardon return spte_set; 1319a6a0b05dSBen Gardon } 1320a6a0b05dSBen Gardon 1321a6a0b05dSBen Gardon /* 1322a6a0b05dSBen Gardon * Clears the dirty status of all the 4k SPTEs mapping GFNs for which a bit is 1323a6a0b05dSBen Gardon * set in mask, starting at gfn. The given memslot is expected to contain all 1324a6a0b05dSBen Gardon * the GFNs represented by set bits in the mask. If AD bits are enabled, 1325a6a0b05dSBen Gardon * clearing the dirty status will involve clearing the dirty bit on each SPTE 1326a6a0b05dSBen Gardon * or, if AD bits are not enabled, clearing the writable bit on each SPTE. 1327a6a0b05dSBen Gardon */ 1328a6a0b05dSBen Gardon static void clear_dirty_pt_masked(struct kvm *kvm, struct kvm_mmu_page *root, 1329a6a0b05dSBen Gardon gfn_t gfn, unsigned long mask, bool wrprot) 1330a6a0b05dSBen Gardon { 1331a6a0b05dSBen Gardon struct tdp_iter iter; 1332a6a0b05dSBen Gardon u64 new_spte; 1333a6a0b05dSBen Gardon 13347cca2d0bSBen Gardon rcu_read_lock(); 13357cca2d0bSBen Gardon 1336a6a0b05dSBen Gardon tdp_root_for_each_leaf_pte(iter, root, gfn + __ffs(mask), 1337a6a0b05dSBen Gardon gfn + BITS_PER_LONG) { 1338a6a0b05dSBen Gardon if (!mask) 1339a6a0b05dSBen Gardon break; 1340a6a0b05dSBen Gardon 1341a6a0b05dSBen Gardon if (iter.level > PG_LEVEL_4K || 1342a6a0b05dSBen Gardon !(mask & (1UL << (iter.gfn - gfn)))) 1343a6a0b05dSBen Gardon continue; 1344a6a0b05dSBen Gardon 1345f1b3b06aSBen Gardon mask &= ~(1UL << (iter.gfn - gfn)); 1346f1b3b06aSBen Gardon 1347a6a0b05dSBen Gardon if (wrprot || spte_ad_need_write_protect(iter.old_spte)) { 1348a6a0b05dSBen Gardon if (is_writable_pte(iter.old_spte)) 1349a6a0b05dSBen Gardon new_spte = iter.old_spte & ~PT_WRITABLE_MASK; 1350a6a0b05dSBen Gardon else 1351a6a0b05dSBen Gardon continue; 1352a6a0b05dSBen Gardon } else { 1353a6a0b05dSBen Gardon if (iter.old_spte & shadow_dirty_mask) 1354a6a0b05dSBen Gardon new_spte = iter.old_spte & ~shadow_dirty_mask; 1355a6a0b05dSBen Gardon else 1356a6a0b05dSBen Gardon continue; 1357a6a0b05dSBen Gardon } 1358a6a0b05dSBen Gardon 1359a6a0b05dSBen Gardon tdp_mmu_set_spte_no_dirty_log(kvm, &iter, new_spte); 1360a6a0b05dSBen Gardon } 13617cca2d0bSBen Gardon 13627cca2d0bSBen Gardon rcu_read_unlock(); 1363a6a0b05dSBen Gardon } 1364a6a0b05dSBen Gardon 1365a6a0b05dSBen Gardon /* 1366a6a0b05dSBen Gardon * Clears the dirty status of all the 4k SPTEs mapping GFNs for which a bit is 1367a6a0b05dSBen Gardon * set in mask, starting at gfn. The given memslot is expected to contain all 1368a6a0b05dSBen Gardon * the GFNs represented by set bits in the mask. If AD bits are enabled, 1369a6a0b05dSBen Gardon * clearing the dirty status will involve clearing the dirty bit on each SPTE 1370a6a0b05dSBen Gardon * or, if AD bits are not enabled, clearing the writable bit on each SPTE. 1371a6a0b05dSBen Gardon */ 1372a6a0b05dSBen Gardon void kvm_tdp_mmu_clear_dirty_pt_masked(struct kvm *kvm, 1373a6a0b05dSBen Gardon struct kvm_memory_slot *slot, 1374a6a0b05dSBen Gardon gfn_t gfn, unsigned long mask, 1375a6a0b05dSBen Gardon bool wrprot) 1376a6a0b05dSBen Gardon { 1377a6a0b05dSBen Gardon struct kvm_mmu_page *root; 1378a6a0b05dSBen Gardon 1379531810caSBen Gardon lockdep_assert_held_write(&kvm->mmu_lock); 1380a3f15bdaSSean Christopherson for_each_tdp_mmu_root(kvm, root, slot->as_id) 1381a6a0b05dSBen Gardon clear_dirty_pt_masked(kvm, root, gfn, mask, wrprot); 1382a6a0b05dSBen Gardon } 1383a6a0b05dSBen Gardon 1384a6a0b05dSBen Gardon /* 138587aa9ec9SBen Gardon * Clear leaf entries which could be replaced by large mappings, for 138687aa9ec9SBen Gardon * GFNs within the slot. 138714881998SBen Gardon */ 13884b85c921SSean Christopherson static void zap_collapsible_spte_range(struct kvm *kvm, 138914881998SBen Gardon struct kvm_mmu_page *root, 13904b85c921SSean Christopherson const struct kvm_memory_slot *slot) 139114881998SBen Gardon { 13929eba50f8SSean Christopherson gfn_t start = slot->base_gfn; 13939eba50f8SSean Christopherson gfn_t end = start + slot->npages; 139414881998SBen Gardon struct tdp_iter iter; 139514881998SBen Gardon kvm_pfn_t pfn; 139614881998SBen Gardon 13977cca2d0bSBen Gardon rcu_read_lock(); 13987cca2d0bSBen Gardon 139914881998SBen Gardon tdp_root_for_each_pte(iter, root, start, end) { 14002db6f772SBen Gardon retry: 14014b85c921SSean Christopherson if (tdp_mmu_iter_cond_resched(kvm, &iter, false, true)) 14021af4a960SBen Gardon continue; 14031af4a960SBen Gardon 140414881998SBen Gardon if (!is_shadow_present_pte(iter.old_spte) || 140587aa9ec9SBen Gardon !is_last_spte(iter.old_spte, iter.level)) 140614881998SBen Gardon continue; 140714881998SBen Gardon 140814881998SBen Gardon pfn = spte_to_pfn(iter.old_spte); 140914881998SBen Gardon if (kvm_is_reserved_pfn(pfn) || 14109eba50f8SSean Christopherson iter.level >= kvm_mmu_max_mapping_level(kvm, slot, iter.gfn, 14119eba50f8SSean Christopherson pfn, PG_LEVEL_NUM)) 141214881998SBen Gardon continue; 141314881998SBen Gardon 14144b85c921SSean Christopherson /* Note, a successful atomic zap also does a remote TLB flush. */ 1415*3e72c791SDavid Matlack if (tdp_mmu_zap_spte_atomic(kvm, &iter)) 14162db6f772SBen Gardon goto retry; 14172db6f772SBen Gardon } 141814881998SBen Gardon 14197cca2d0bSBen Gardon rcu_read_unlock(); 142014881998SBen Gardon } 142114881998SBen Gardon 142214881998SBen Gardon /* 142314881998SBen Gardon * Clear non-leaf entries (and free associated page tables) which could 142414881998SBen Gardon * be replaced by large mappings, for GFNs within the slot. 142514881998SBen Gardon */ 14264b85c921SSean Christopherson void kvm_tdp_mmu_zap_collapsible_sptes(struct kvm *kvm, 14274b85c921SSean Christopherson const struct kvm_memory_slot *slot) 142814881998SBen Gardon { 142914881998SBen Gardon struct kvm_mmu_page *root; 143014881998SBen Gardon 14312db6f772SBen Gardon lockdep_assert_held_read(&kvm->mmu_lock); 143214881998SBen Gardon 1433d62007edSSean Christopherson for_each_valid_tdp_mmu_root_yield_safe(kvm, root, slot->as_id, true) 14344b85c921SSean Christopherson zap_collapsible_spte_range(kvm, root, slot); 143514881998SBen Gardon } 143646044f72SBen Gardon 143746044f72SBen Gardon /* 143846044f72SBen Gardon * Removes write access on the last level SPTE mapping this GFN and unsets the 14395fc3424fSSean Christopherson * MMU-writable bit to ensure future writes continue to be intercepted. 144046044f72SBen Gardon * Returns true if an SPTE was set and a TLB flush is needed. 144146044f72SBen Gardon */ 144246044f72SBen Gardon static bool write_protect_gfn(struct kvm *kvm, struct kvm_mmu_page *root, 14433ad93562SKeqian Zhu gfn_t gfn, int min_level) 144446044f72SBen Gardon { 144546044f72SBen Gardon struct tdp_iter iter; 144646044f72SBen Gardon u64 new_spte; 144746044f72SBen Gardon bool spte_set = false; 144846044f72SBen Gardon 14493ad93562SKeqian Zhu BUG_ON(min_level > KVM_MAX_HUGEPAGE_LEVEL); 14503ad93562SKeqian Zhu 14517cca2d0bSBen Gardon rcu_read_lock(); 14527cca2d0bSBen Gardon 14533ad93562SKeqian Zhu for_each_tdp_pte_min_level(iter, root->spt, root->role.level, 14543ad93562SKeqian Zhu min_level, gfn, gfn + 1) { 14553ad93562SKeqian Zhu if (!is_shadow_present_pte(iter.old_spte) || 14563ad93562SKeqian Zhu !is_last_spte(iter.old_spte, iter.level)) 14573ad93562SKeqian Zhu continue; 14583ad93562SKeqian Zhu 145946044f72SBen Gardon new_spte = iter.old_spte & 14605fc3424fSSean Christopherson ~(PT_WRITABLE_MASK | shadow_mmu_writable_mask); 146146044f72SBen Gardon 14627c8a4742SDavid Matlack if (new_spte == iter.old_spte) 14637c8a4742SDavid Matlack break; 14647c8a4742SDavid Matlack 146546044f72SBen Gardon tdp_mmu_set_spte(kvm, &iter, new_spte); 146646044f72SBen Gardon spte_set = true; 146746044f72SBen Gardon } 146846044f72SBen Gardon 14697cca2d0bSBen Gardon rcu_read_unlock(); 14707cca2d0bSBen Gardon 147146044f72SBen Gardon return spte_set; 147246044f72SBen Gardon } 147346044f72SBen Gardon 147446044f72SBen Gardon /* 147546044f72SBen Gardon * Removes write access on the last level SPTE mapping this GFN and unsets the 14765fc3424fSSean Christopherson * MMU-writable bit to ensure future writes continue to be intercepted. 147746044f72SBen Gardon * Returns true if an SPTE was set and a TLB flush is needed. 147846044f72SBen Gardon */ 147946044f72SBen Gardon bool kvm_tdp_mmu_write_protect_gfn(struct kvm *kvm, 14803ad93562SKeqian Zhu struct kvm_memory_slot *slot, gfn_t gfn, 14813ad93562SKeqian Zhu int min_level) 148246044f72SBen Gardon { 148346044f72SBen Gardon struct kvm_mmu_page *root; 148446044f72SBen Gardon bool spte_set = false; 148546044f72SBen Gardon 1486531810caSBen Gardon lockdep_assert_held_write(&kvm->mmu_lock); 1487a3f15bdaSSean Christopherson for_each_tdp_mmu_root(kvm, root, slot->as_id) 14883ad93562SKeqian Zhu spte_set |= write_protect_gfn(kvm, root, gfn, min_level); 1489a3f15bdaSSean Christopherson 149046044f72SBen Gardon return spte_set; 149146044f72SBen Gardon } 149246044f72SBen Gardon 149395fb5b02SBen Gardon /* 149495fb5b02SBen Gardon * Return the level of the lowest level SPTE added to sptes. 149595fb5b02SBen Gardon * That SPTE may be non-present. 1496c5c8c7c5SDavid Matlack * 1497c5c8c7c5SDavid Matlack * Must be called between kvm_tdp_mmu_walk_lockless_{begin,end}. 149895fb5b02SBen Gardon */ 149939b4d43eSSean Christopherson int kvm_tdp_mmu_get_walk(struct kvm_vcpu *vcpu, u64 addr, u64 *sptes, 150039b4d43eSSean Christopherson int *root_level) 150195fb5b02SBen Gardon { 150295fb5b02SBen Gardon struct tdp_iter iter; 150395fb5b02SBen Gardon struct kvm_mmu *mmu = vcpu->arch.mmu; 150495fb5b02SBen Gardon gfn_t gfn = addr >> PAGE_SHIFT; 15052aa07893SSean Christopherson int leaf = -1; 150695fb5b02SBen Gardon 150739b4d43eSSean Christopherson *root_level = vcpu->arch.mmu->shadow_root_level; 150895fb5b02SBen Gardon 150995fb5b02SBen Gardon tdp_mmu_for_each_pte(iter, mmu, gfn, gfn + 1) { 151095fb5b02SBen Gardon leaf = iter.level; 1511dde81f94SSean Christopherson sptes[leaf] = iter.old_spte; 151295fb5b02SBen Gardon } 151395fb5b02SBen Gardon 151495fb5b02SBen Gardon return leaf; 151595fb5b02SBen Gardon } 15166e8eb206SDavid Matlack 15176e8eb206SDavid Matlack /* 15186e8eb206SDavid Matlack * Returns the last level spte pointer of the shadow page walk for the given 15196e8eb206SDavid Matlack * gpa, and sets *spte to the spte value. This spte may be non-preset. If no 15206e8eb206SDavid Matlack * walk could be performed, returns NULL and *spte does not contain valid data. 15216e8eb206SDavid Matlack * 15226e8eb206SDavid Matlack * Contract: 15236e8eb206SDavid Matlack * - Must be called between kvm_tdp_mmu_walk_lockless_{begin,end}. 15246e8eb206SDavid Matlack * - The returned sptep must not be used after kvm_tdp_mmu_walk_lockless_end. 15256e8eb206SDavid Matlack * 15266e8eb206SDavid Matlack * WARNING: This function is only intended to be called during fast_page_fault. 15276e8eb206SDavid Matlack */ 15286e8eb206SDavid Matlack u64 *kvm_tdp_mmu_fast_pf_get_last_sptep(struct kvm_vcpu *vcpu, u64 addr, 15296e8eb206SDavid Matlack u64 *spte) 15306e8eb206SDavid Matlack { 15316e8eb206SDavid Matlack struct tdp_iter iter; 15326e8eb206SDavid Matlack struct kvm_mmu *mmu = vcpu->arch.mmu; 15336e8eb206SDavid Matlack gfn_t gfn = addr >> PAGE_SHIFT; 15346e8eb206SDavid Matlack tdp_ptep_t sptep = NULL; 15356e8eb206SDavid Matlack 15366e8eb206SDavid Matlack tdp_mmu_for_each_pte(iter, mmu, gfn, gfn + 1) { 15376e8eb206SDavid Matlack *spte = iter.old_spte; 15386e8eb206SDavid Matlack sptep = iter.sptep; 15396e8eb206SDavid Matlack } 15406e8eb206SDavid Matlack 15416e8eb206SDavid Matlack /* 15426e8eb206SDavid Matlack * Perform the rcu_dereference to get the raw spte pointer value since 15436e8eb206SDavid Matlack * we are passing it up to fast_page_fault, which is shared with the 15446e8eb206SDavid Matlack * legacy MMU and thus does not retain the TDP MMU-specific __rcu 15456e8eb206SDavid Matlack * annotation. 15466e8eb206SDavid Matlack * 15476e8eb206SDavid Matlack * This is safe since fast_page_fault obeys the contracts of this 15486e8eb206SDavid Matlack * function as well as all TDP MMU contracts around modifying SPTEs 15496e8eb206SDavid Matlack * outside of mmu_lock. 15506e8eb206SDavid Matlack */ 15516e8eb206SDavid Matlack return rcu_dereference(sptep); 15526e8eb206SDavid Matlack } 1553