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 /* 102*d62007edSSean Christopherson * Returns the next root after @prev_root (or the first root if @prev_root is 103*d62007edSSean Christopherson * NULL). A reference to the returned root is acquired, and the reference to 104*d62007edSSean Christopherson * @prev_root is released (the caller obviously must hold a reference to 105*d62007edSSean Christopherson * @prev_root if it's non-NULL). 106*d62007edSSean Christopherson * 107*d62007edSSean Christopherson * If @only_valid is true, invalid roots are skipped. 108*d62007edSSean Christopherson * 109*d62007edSSean 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, 113*d62007edSSean 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) { 128*d62007edSSean Christopherson if ((!only_valid || !next_root->role.invalid) && 12904dc4e6cSSean Christopherson kvm_tdp_mmu_get_root(kvm, 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 */ 154*d62007edSSean Christopherson #define __for_each_tdp_mmu_root_yield_safe(_kvm, _root, _as_id, _shared, _only_valid)\ 155*d62007edSSean Christopherson for (_root = tdp_mmu_next_root(_kvm, NULL, _shared, _only_valid); \ 156cfc10997SBen Gardon _root; \ 157*d62007edSSean 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 161*d62007edSSean Christopherson #define for_each_valid_tdp_mmu_root_yield_safe(_kvm, _root, _as_id, _shared) \ 162*d62007edSSean Christopherson __for_each_tdp_mmu_root_yield_safe(_kvm, _root, _as_id, _shared, true) 163*d62007edSSean Christopherson 164*d62007edSSean Christopherson #define for_each_tdp_mmu_root_yield_safe(_kvm, _root, _as_id, _shared) \ 165*d62007edSSean Christopherson __for_each_tdp_mmu_root_yield_safe(_kvm, _root, _as_id, _shared, false) 166*d62007edSSean 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 && 223fb101293SBen Gardon kvm_tdp_mmu_get_root(kvm, 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 4552f2fad08SBen Gardon /* 4562f2fad08SBen Gardon * The only times a SPTE should be changed from a non-present to 4572f2fad08SBen Gardon * non-present state is when an MMIO entry is installed/modified/ 4582f2fad08SBen Gardon * removed. In that case, there is nothing to do here. 4592f2fad08SBen Gardon */ 4602f2fad08SBen Gardon if (!was_present && !is_present) { 4612f2fad08SBen Gardon /* 46208f07c80SBen Gardon * If this change does not involve a MMIO SPTE or removed SPTE, 46308f07c80SBen Gardon * it is unexpected. Log the change, though it should not 46408f07c80SBen Gardon * impact the guest since both the former and current SPTEs 46508f07c80SBen Gardon * are nonpresent. 4662f2fad08SBen Gardon */ 46708f07c80SBen Gardon if (WARN_ON(!is_mmio_spte(old_spte) && 46808f07c80SBen Gardon !is_mmio_spte(new_spte) && 46908f07c80SBen Gardon !is_removed_spte(new_spte))) 4702f2fad08SBen Gardon pr_err("Unexpected SPTE change! Nonpresent SPTEs\n" 4712f2fad08SBen Gardon "should not be replaced with another,\n" 4722f2fad08SBen Gardon "different nonpresent SPTE, unless one or both\n" 47308f07c80SBen Gardon "are MMIO SPTEs, or the new SPTE is\n" 47408f07c80SBen Gardon "a temporary removed SPTE.\n" 4752f2fad08SBen Gardon "as_id: %d gfn: %llx old_spte: %llx new_spte: %llx level: %d", 4762f2fad08SBen Gardon as_id, gfn, old_spte, new_spte, level); 4772f2fad08SBen Gardon return; 4782f2fad08SBen Gardon } 4792f2fad08SBen Gardon 48071f51d2cSMingwei Zhang if (is_leaf != was_leaf) 48171f51d2cSMingwei Zhang kvm_update_page_stats(kvm, level, is_leaf ? 1 : -1); 4822f2fad08SBen Gardon 4832f2fad08SBen Gardon if (was_leaf && is_dirty_spte(old_spte) && 48464bb2769SSean Christopherson (!is_present || !is_dirty_spte(new_spte) || pfn_changed)) 4852f2fad08SBen Gardon kvm_set_pfn_dirty(spte_to_pfn(old_spte)); 4862f2fad08SBen Gardon 4872f2fad08SBen Gardon /* 4882f2fad08SBen Gardon * Recursively handle child PTs if the change removed a subtree from 4892f2fad08SBen Gardon * the paging structure. 4902f2fad08SBen Gardon */ 491a066e61fSBen Gardon if (was_present && !was_leaf && (pfn_changed || !is_present)) 492a066e61fSBen Gardon handle_removed_tdp_mmu_page(kvm, 4939a77daacSBen Gardon spte_to_child_pt(old_spte, level), shared); 4942f2fad08SBen Gardon } 4952f2fad08SBen Gardon 4962f2fad08SBen Gardon static void handle_changed_spte(struct kvm *kvm, int as_id, gfn_t gfn, 4979a77daacSBen Gardon u64 old_spte, u64 new_spte, int level, 4989a77daacSBen Gardon bool shared) 4992f2fad08SBen Gardon { 5009a77daacSBen Gardon __handle_changed_spte(kvm, as_id, gfn, old_spte, new_spte, level, 5019a77daacSBen Gardon shared); 502f8e14497SBen Gardon handle_changed_spte_acc_track(old_spte, new_spte, level); 503a6a0b05dSBen Gardon handle_changed_spte_dirty_log(kvm, as_id, gfn, old_spte, 504a6a0b05dSBen Gardon new_spte, level); 5052f2fad08SBen Gardon } 506faaf05b0SBen Gardon 507fe43fa2fSBen Gardon /* 5086ccf4438SPaolo Bonzini * tdp_mmu_set_spte_atomic - Set a TDP MMU SPTE atomically 5096ccf4438SPaolo Bonzini * and handle the associated bookkeeping. Do not mark the page dirty 51024ae4cfaSBen Gardon * in KVM's dirty bitmaps. 5119a77daacSBen Gardon * 5129a77daacSBen Gardon * @kvm: kvm instance 5139a77daacSBen Gardon * @iter: a tdp_iter instance currently on the SPTE that should be set 5149a77daacSBen Gardon * @new_spte: The value the SPTE should be set to 5159a77daacSBen Gardon * Returns: true if the SPTE was set, false if it was not. If false is returned, 5169a77daacSBen Gardon * this function will have no side-effects. 5179a77daacSBen Gardon */ 5186ccf4438SPaolo Bonzini static inline bool tdp_mmu_set_spte_atomic(struct kvm *kvm, 5199a77daacSBen Gardon struct tdp_iter *iter, 5209a77daacSBen Gardon u64 new_spte) 5219a77daacSBen Gardon { 5223a0f64deSSean Christopherson WARN_ON_ONCE(iter->yielded); 5233a0f64deSSean Christopherson 5249a77daacSBen Gardon lockdep_assert_held_read(&kvm->mmu_lock); 5259a77daacSBen Gardon 52608f07c80SBen Gardon /* 52708f07c80SBen Gardon * Do not change removed SPTEs. Only the thread that froze the SPTE 52808f07c80SBen Gardon * may modify it. 52908f07c80SBen Gardon */ 5307a51393aSSean Christopherson if (is_removed_spte(iter->old_spte)) 53108f07c80SBen Gardon return false; 53208f07c80SBen Gardon 5336e8eb206SDavid Matlack /* 5346e8eb206SDavid Matlack * Note, fast_pf_fix_direct_spte() can also modify TDP MMU SPTEs and 5356e8eb206SDavid Matlack * does not hold the mmu_lock. 5366e8eb206SDavid Matlack */ 5379a77daacSBen Gardon if (cmpxchg64(rcu_dereference(iter->sptep), iter->old_spte, 5389a77daacSBen Gardon new_spte) != iter->old_spte) 5399a77daacSBen Gardon return false; 5409a77daacSBen Gardon 54124ae4cfaSBen Gardon __handle_changed_spte(kvm, iter->as_id, iter->gfn, iter->old_spte, 54208889894SSean Christopherson new_spte, iter->level, true); 54324ae4cfaSBen Gardon handle_changed_spte_acc_track(iter->old_spte, new_spte, iter->level); 5449a77daacSBen Gardon 5459a77daacSBen Gardon return true; 5469a77daacSBen Gardon } 5479a77daacSBen Gardon 54808f07c80SBen Gardon static inline bool tdp_mmu_zap_spte_atomic(struct kvm *kvm, 54908f07c80SBen Gardon struct tdp_iter *iter) 55008f07c80SBen Gardon { 55108f07c80SBen Gardon /* 55208f07c80SBen Gardon * Freeze the SPTE by setting it to a special, 55308f07c80SBen Gardon * non-present value. This will stop other threads from 55408f07c80SBen Gardon * immediately installing a present entry in its place 55508f07c80SBen Gardon * before the TLBs are flushed. 55608f07c80SBen Gardon */ 5576ccf4438SPaolo Bonzini if (!tdp_mmu_set_spte_atomic(kvm, iter, REMOVED_SPTE)) 55808f07c80SBen Gardon return false; 55908f07c80SBen Gardon 56008f07c80SBen Gardon kvm_flush_remote_tlbs_with_address(kvm, iter->gfn, 56108f07c80SBen Gardon KVM_PAGES_PER_HPAGE(iter->level)); 56208f07c80SBen Gardon 56308f07c80SBen Gardon /* 56408f07c80SBen Gardon * No other thread can overwrite the removed SPTE as they 56508f07c80SBen Gardon * must either wait on the MMU lock or use 566d9f6e12fSIngo Molnar * tdp_mmu_set_spte_atomic which will not overwrite the 56708f07c80SBen Gardon * special removed SPTE value. No bookkeeping is needed 56808f07c80SBen Gardon * here since the SPTE is going from non-present 56908f07c80SBen Gardon * to non-present. 57008f07c80SBen Gardon */ 57114f6fec2SBen Gardon WRITE_ONCE(*rcu_dereference(iter->sptep), 0); 57208f07c80SBen Gardon 57308f07c80SBen Gardon return true; 57408f07c80SBen Gardon } 57508f07c80SBen Gardon 5769a77daacSBen Gardon 5779a77daacSBen Gardon /* 578fe43fa2fSBen Gardon * __tdp_mmu_set_spte - Set a TDP MMU SPTE and handle the associated bookkeeping 579fe43fa2fSBen Gardon * @kvm: kvm instance 580fe43fa2fSBen Gardon * @iter: a tdp_iter instance currently on the SPTE that should be set 581fe43fa2fSBen Gardon * @new_spte: The value the SPTE should be set to 582fe43fa2fSBen Gardon * @record_acc_track: Notify the MM subsystem of changes to the accessed state 583fe43fa2fSBen Gardon * of the page. Should be set unless handling an MMU 584fe43fa2fSBen Gardon * notifier for access tracking. Leaving record_acc_track 585fe43fa2fSBen Gardon * unset in that case prevents page accesses from being 586fe43fa2fSBen Gardon * double counted. 587fe43fa2fSBen Gardon * @record_dirty_log: Record the page as dirty in the dirty bitmap if 588fe43fa2fSBen Gardon * appropriate for the change being made. Should be set 589fe43fa2fSBen Gardon * unless performing certain dirty logging operations. 590fe43fa2fSBen Gardon * Leaving record_dirty_log unset in that case prevents page 591fe43fa2fSBen Gardon * writes from being double counted. 592fe43fa2fSBen Gardon */ 593f8e14497SBen Gardon static inline void __tdp_mmu_set_spte(struct kvm *kvm, struct tdp_iter *iter, 594a6a0b05dSBen Gardon u64 new_spte, bool record_acc_track, 595a6a0b05dSBen Gardon bool record_dirty_log) 596faaf05b0SBen Gardon { 5973a0f64deSSean Christopherson WARN_ON_ONCE(iter->yielded); 5983a0f64deSSean Christopherson 599531810caSBen Gardon lockdep_assert_held_write(&kvm->mmu_lock); 6003a9a4aa5SBen Gardon 60108f07c80SBen Gardon /* 60208f07c80SBen Gardon * No thread should be using this function to set SPTEs to the 60308f07c80SBen Gardon * temporary removed SPTE value. 60408f07c80SBen Gardon * If operating under the MMU lock in read mode, tdp_mmu_set_spte_atomic 60508f07c80SBen Gardon * should be used. If operating under the MMU lock in write mode, the 60608f07c80SBen Gardon * use of the removed SPTE should not be necessary. 60708f07c80SBen Gardon */ 6087a51393aSSean Christopherson WARN_ON(is_removed_spte(iter->old_spte)); 60908f07c80SBen Gardon 6107cca2d0bSBen Gardon WRITE_ONCE(*rcu_dereference(iter->sptep), new_spte); 611faaf05b0SBen Gardon 61208889894SSean Christopherson __handle_changed_spte(kvm, iter->as_id, iter->gfn, iter->old_spte, 61308889894SSean Christopherson new_spte, iter->level, false); 614f8e14497SBen Gardon if (record_acc_track) 615f8e14497SBen Gardon handle_changed_spte_acc_track(iter->old_spte, new_spte, 616f8e14497SBen Gardon iter->level); 617a6a0b05dSBen Gardon if (record_dirty_log) 61808889894SSean Christopherson handle_changed_spte_dirty_log(kvm, iter->as_id, iter->gfn, 619a6a0b05dSBen Gardon iter->old_spte, new_spte, 620a6a0b05dSBen Gardon iter->level); 621f8e14497SBen Gardon } 622f8e14497SBen Gardon 623f8e14497SBen Gardon static inline void tdp_mmu_set_spte(struct kvm *kvm, struct tdp_iter *iter, 624f8e14497SBen Gardon u64 new_spte) 625f8e14497SBen Gardon { 626a6a0b05dSBen Gardon __tdp_mmu_set_spte(kvm, iter, new_spte, true, true); 627f8e14497SBen Gardon } 628f8e14497SBen Gardon 629f8e14497SBen Gardon static inline void tdp_mmu_set_spte_no_acc_track(struct kvm *kvm, 630f8e14497SBen Gardon struct tdp_iter *iter, 631f8e14497SBen Gardon u64 new_spte) 632f8e14497SBen Gardon { 633a6a0b05dSBen Gardon __tdp_mmu_set_spte(kvm, iter, new_spte, false, true); 634a6a0b05dSBen Gardon } 635a6a0b05dSBen Gardon 636a6a0b05dSBen Gardon static inline void tdp_mmu_set_spte_no_dirty_log(struct kvm *kvm, 637a6a0b05dSBen Gardon struct tdp_iter *iter, 638a6a0b05dSBen Gardon u64 new_spte) 639a6a0b05dSBen Gardon { 640a6a0b05dSBen Gardon __tdp_mmu_set_spte(kvm, iter, new_spte, true, false); 641faaf05b0SBen Gardon } 642faaf05b0SBen Gardon 643faaf05b0SBen Gardon #define tdp_root_for_each_pte(_iter, _root, _start, _end) \ 644faaf05b0SBen Gardon for_each_tdp_pte(_iter, _root->spt, _root->role.level, _start, _end) 645faaf05b0SBen Gardon 646f8e14497SBen Gardon #define tdp_root_for_each_leaf_pte(_iter, _root, _start, _end) \ 647f8e14497SBen Gardon tdp_root_for_each_pte(_iter, _root, _start, _end) \ 648f8e14497SBen Gardon if (!is_shadow_present_pte(_iter.old_spte) || \ 649f8e14497SBen Gardon !is_last_spte(_iter.old_spte, _iter.level)) \ 650f8e14497SBen Gardon continue; \ 651f8e14497SBen Gardon else 652f8e14497SBen Gardon 653bb18842eSBen Gardon #define tdp_mmu_for_each_pte(_iter, _mmu, _start, _end) \ 654bb18842eSBen Gardon for_each_tdp_pte(_iter, __va(_mmu->root_hpa), \ 655bb18842eSBen Gardon _mmu->shadow_root_level, _start, _end) 656bb18842eSBen Gardon 657faaf05b0SBen Gardon /* 658e28a436cSBen Gardon * Yield if the MMU lock is contended or this thread needs to return control 659e28a436cSBen Gardon * to the scheduler. 660e28a436cSBen Gardon * 661e139a34eSBen Gardon * If this function should yield and flush is set, it will perform a remote 662e139a34eSBen Gardon * TLB flush before yielding. 663e139a34eSBen Gardon * 6643a0f64deSSean Christopherson * If this function yields, iter->yielded is set and the caller must skip to 6653a0f64deSSean Christopherson * the next iteration, where tdp_iter_next() will reset the tdp_iter's walk 6663a0f64deSSean Christopherson * over the paging structures to allow the iterator to continue its traversal 6673a0f64deSSean Christopherson * from the paging structure root. 668e28a436cSBen Gardon * 6693a0f64deSSean Christopherson * Returns true if this function yielded. 670e28a436cSBen Gardon */ 6713a0f64deSSean Christopherson static inline bool __must_check tdp_mmu_iter_cond_resched(struct kvm *kvm, 6723a0f64deSSean Christopherson struct tdp_iter *iter, 6733a0f64deSSean Christopherson bool flush, bool shared) 674a6a0b05dSBen Gardon { 6753a0f64deSSean Christopherson WARN_ON(iter->yielded); 6763a0f64deSSean Christopherson 677ed5e484bSBen Gardon /* Ensure forward progress has been made before yielding. */ 678ed5e484bSBen Gardon if (iter->next_last_level_gfn == iter->yielded_gfn) 679ed5e484bSBen Gardon return false; 680ed5e484bSBen Gardon 681531810caSBen Gardon if (need_resched() || rwlock_needbreak(&kvm->mmu_lock)) { 6827cca2d0bSBen Gardon rcu_read_unlock(); 6837cca2d0bSBen Gardon 684e139a34eSBen Gardon if (flush) 685e139a34eSBen Gardon kvm_flush_remote_tlbs(kvm); 686e139a34eSBen Gardon 6876103bc07SBen Gardon if (shared) 6886103bc07SBen Gardon cond_resched_rwlock_read(&kvm->mmu_lock); 6896103bc07SBen Gardon else 690531810caSBen Gardon cond_resched_rwlock_write(&kvm->mmu_lock); 6916103bc07SBen Gardon 6927cca2d0bSBen Gardon rcu_read_lock(); 693ed5e484bSBen Gardon 694ed5e484bSBen Gardon WARN_ON(iter->gfn > iter->next_last_level_gfn); 695ed5e484bSBen Gardon 6963a0f64deSSean Christopherson iter->yielded = true; 697a6a0b05dSBen Gardon } 698e28a436cSBen Gardon 6993a0f64deSSean Christopherson return iter->yielded; 700a6a0b05dSBen Gardon } 701a6a0b05dSBen Gardon 702faaf05b0SBen Gardon /* 703faaf05b0SBen Gardon * Tears down the mappings for the range of gfns, [start, end), and frees the 704faaf05b0SBen Gardon * non-root pages mapping GFNs strictly within that range. Returns true if 705faaf05b0SBen Gardon * SPTEs have been cleared and a TLB flush is needed before releasing the 706faaf05b0SBen Gardon * MMU lock. 7076103bc07SBen Gardon * 708063afacdSBen Gardon * If can_yield is true, will release the MMU lock and reschedule if the 709063afacdSBen Gardon * scheduler needs the CPU or there is contention on the MMU lock. If this 710063afacdSBen Gardon * function cannot yield, it will not release the MMU lock or reschedule and 711063afacdSBen Gardon * the caller must ensure it does not supply too large a GFN range, or the 7126103bc07SBen Gardon * operation can cause a soft lockup. 7136103bc07SBen Gardon * 7146103bc07SBen Gardon * If shared is true, this thread holds the MMU lock in read mode and must 7156103bc07SBen Gardon * account for the possibility that other threads are modifying the paging 7166103bc07SBen Gardon * structures concurrently. If shared is false, this thread should hold the 7176103bc07SBen Gardon * MMU lock in write mode. 718faaf05b0SBen Gardon */ 719faaf05b0SBen Gardon static bool zap_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root, 7206103bc07SBen Gardon gfn_t start, gfn_t end, bool can_yield, bool flush, 7216103bc07SBen Gardon bool shared) 722faaf05b0SBen Gardon { 723524a1e4eSSean Christopherson gfn_t max_gfn_host = 1ULL << (shadow_phys_bits - PAGE_SHIFT); 724524a1e4eSSean Christopherson bool zap_all = (start == 0 && end >= max_gfn_host); 725faaf05b0SBen Gardon struct tdp_iter iter; 726faaf05b0SBen Gardon 727524a1e4eSSean Christopherson /* 7280103098fSSean Christopherson * No need to try to step down in the iterator when zapping all SPTEs, 7290103098fSSean Christopherson * zapping the top-level non-leaf SPTEs will recurse on their children. 7300103098fSSean Christopherson */ 7310103098fSSean Christopherson int min_level = zap_all ? root->role.level : PG_LEVEL_4K; 7320103098fSSean Christopherson 7330103098fSSean Christopherson /* 734524a1e4eSSean Christopherson * Bound the walk at host.MAXPHYADDR, guest accesses beyond that will 735524a1e4eSSean Christopherson * hit a #PF(RSVD) and never get to an EPT Violation/Misconfig / #NPF, 736524a1e4eSSean Christopherson * and so KVM will never install a SPTE for such addresses. 737524a1e4eSSean Christopherson */ 738524a1e4eSSean Christopherson end = min(end, max_gfn_host); 739524a1e4eSSean Christopherson 7406103bc07SBen Gardon kvm_lockdep_assert_mmu_lock_held(kvm, shared); 7416103bc07SBen Gardon 7427cca2d0bSBen Gardon rcu_read_lock(); 7437cca2d0bSBen Gardon 7440103098fSSean Christopherson for_each_tdp_pte_min_level(iter, root->spt, root->role.level, 7450103098fSSean Christopherson min_level, start, end) { 7466103bc07SBen Gardon retry: 7471af4a960SBen Gardon if (can_yield && 7486103bc07SBen Gardon tdp_mmu_iter_cond_resched(kvm, &iter, flush, shared)) { 749a835429cSSean Christopherson flush = false; 7501af4a960SBen Gardon continue; 7511af4a960SBen Gardon } 7521af4a960SBen Gardon 753faaf05b0SBen Gardon if (!is_shadow_present_pte(iter.old_spte)) 754faaf05b0SBen Gardon continue; 755faaf05b0SBen Gardon 756faaf05b0SBen Gardon /* 757faaf05b0SBen Gardon * If this is a non-last-level SPTE that covers a larger range 758faaf05b0SBen Gardon * than should be zapped, continue, and zap the mappings at a 759524a1e4eSSean Christopherson * lower level, except when zapping all SPTEs. 760faaf05b0SBen Gardon */ 761524a1e4eSSean Christopherson if (!zap_all && 762524a1e4eSSean Christopherson (iter.gfn < start || 763faaf05b0SBen Gardon iter.gfn + KVM_PAGES_PER_HPAGE(iter.level) > end) && 764faaf05b0SBen Gardon !is_last_spte(iter.old_spte, iter.level)) 765faaf05b0SBen Gardon continue; 766faaf05b0SBen Gardon 7676103bc07SBen Gardon if (!shared) { 768faaf05b0SBen Gardon tdp_mmu_set_spte(kvm, &iter, 0); 769a835429cSSean Christopherson flush = true; 7706103bc07SBen Gardon } else if (!tdp_mmu_zap_spte_atomic(kvm, &iter)) { 7716103bc07SBen Gardon /* 7726103bc07SBen Gardon * The iter must explicitly re-read the SPTE because 7736103bc07SBen Gardon * the atomic cmpxchg failed. 7746103bc07SBen Gardon */ 7756103bc07SBen Gardon iter.old_spte = READ_ONCE(*rcu_dereference(iter.sptep)); 7766103bc07SBen Gardon goto retry; 7776103bc07SBen Gardon } 778faaf05b0SBen Gardon } 7797cca2d0bSBen Gardon 7807cca2d0bSBen Gardon rcu_read_unlock(); 781a835429cSSean Christopherson return flush; 782faaf05b0SBen Gardon } 783faaf05b0SBen Gardon 784faaf05b0SBen Gardon /* 785faaf05b0SBen Gardon * Tears down the mappings for the range of gfns, [start, end), and frees the 786faaf05b0SBen Gardon * non-root pages mapping GFNs strictly within that range. Returns true if 787faaf05b0SBen Gardon * SPTEs have been cleared and a TLB flush is needed before releasing the 788faaf05b0SBen Gardon * MMU lock. 789faaf05b0SBen Gardon */ 7902b9663d8SSean Christopherson bool __kvm_tdp_mmu_zap_gfn_range(struct kvm *kvm, int as_id, gfn_t start, 7915a324c24SSean Christopherson gfn_t end, bool can_yield, bool flush) 792faaf05b0SBen Gardon { 793faaf05b0SBen Gardon struct kvm_mmu_page *root; 794faaf05b0SBen Gardon 7955a324c24SSean Christopherson for_each_tdp_mmu_root_yield_safe(kvm, root, as_id, false) 7966103bc07SBen Gardon flush = zap_gfn_range(kvm, root, start, end, can_yield, flush, 7975a324c24SSean Christopherson false); 798faaf05b0SBen Gardon 799faaf05b0SBen Gardon return flush; 800faaf05b0SBen Gardon } 801faaf05b0SBen Gardon 802faaf05b0SBen Gardon void kvm_tdp_mmu_zap_all(struct kvm *kvm) 803faaf05b0SBen Gardon { 8042b9663d8SSean Christopherson bool flush = false; 8052b9663d8SSean Christopherson int i; 806faaf05b0SBen Gardon 8072b9663d8SSean Christopherson for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) 8085a324c24SSean Christopherson flush = kvm_tdp_mmu_zap_gfn_range(kvm, i, 0, -1ull, flush); 8092b9663d8SSean Christopherson 810faaf05b0SBen Gardon if (flush) 811faaf05b0SBen Gardon kvm_flush_remote_tlbs(kvm); 812faaf05b0SBen Gardon } 813bb18842eSBen Gardon 8144c6654bdSBen Gardon static struct kvm_mmu_page *next_invalidated_root(struct kvm *kvm, 8154c6654bdSBen Gardon struct kvm_mmu_page *prev_root) 8164c6654bdSBen Gardon { 8174c6654bdSBen Gardon struct kvm_mmu_page *next_root; 8184c6654bdSBen Gardon 8194c6654bdSBen Gardon if (prev_root) 8204c6654bdSBen Gardon next_root = list_next_or_null_rcu(&kvm->arch.tdp_mmu_roots, 8214c6654bdSBen Gardon &prev_root->link, 8224c6654bdSBen Gardon typeof(*prev_root), link); 8234c6654bdSBen Gardon else 8244c6654bdSBen Gardon next_root = list_first_or_null_rcu(&kvm->arch.tdp_mmu_roots, 8254c6654bdSBen Gardon typeof(*next_root), link); 8264c6654bdSBen Gardon 8274c6654bdSBen Gardon while (next_root && !(next_root->role.invalid && 8284c6654bdSBen Gardon refcount_read(&next_root->tdp_mmu_root_count))) 8294c6654bdSBen Gardon next_root = list_next_or_null_rcu(&kvm->arch.tdp_mmu_roots, 8304c6654bdSBen Gardon &next_root->link, 8314c6654bdSBen Gardon typeof(*next_root), link); 8324c6654bdSBen Gardon 8334c6654bdSBen Gardon return next_root; 8344c6654bdSBen Gardon } 8354c6654bdSBen Gardon 8364c6654bdSBen Gardon /* 8374c6654bdSBen Gardon * Since kvm_tdp_mmu_zap_all_fast has acquired a reference to each 8384c6654bdSBen Gardon * invalidated root, they will not be freed until this function drops the 8394c6654bdSBen Gardon * reference. Before dropping that reference, tear down the paging 8404c6654bdSBen Gardon * structure so that whichever thread does drop the last reference 8414c6654bdSBen Gardon * only has to do a trivial amount of work. Since the roots are invalid, 8424c6654bdSBen Gardon * no new SPTEs should be created under them. 8434c6654bdSBen Gardon */ 8444c6654bdSBen Gardon void kvm_tdp_mmu_zap_invalidated_roots(struct kvm *kvm) 8454c6654bdSBen Gardon { 8464c6654bdSBen Gardon struct kvm_mmu_page *next_root; 8474c6654bdSBen Gardon struct kvm_mmu_page *root; 8484c6654bdSBen Gardon bool flush = false; 8494c6654bdSBen Gardon 8504c6654bdSBen Gardon lockdep_assert_held_read(&kvm->mmu_lock); 8514c6654bdSBen Gardon 8524c6654bdSBen Gardon rcu_read_lock(); 8534c6654bdSBen Gardon 8544c6654bdSBen Gardon root = next_invalidated_root(kvm, NULL); 8554c6654bdSBen Gardon 8564c6654bdSBen Gardon while (root) { 8574c6654bdSBen Gardon next_root = next_invalidated_root(kvm, root); 8584c6654bdSBen Gardon 8594c6654bdSBen Gardon rcu_read_unlock(); 8604c6654bdSBen Gardon 861524a1e4eSSean Christopherson flush = zap_gfn_range(kvm, root, 0, -1ull, true, flush, true); 8624c6654bdSBen Gardon 8634c6654bdSBen Gardon /* 8644c6654bdSBen Gardon * Put the reference acquired in 8654c6654bdSBen Gardon * kvm_tdp_mmu_invalidate_roots 8664c6654bdSBen Gardon */ 8674c6654bdSBen Gardon kvm_tdp_mmu_put_root(kvm, root, true); 8684c6654bdSBen Gardon 8694c6654bdSBen Gardon root = next_root; 8704c6654bdSBen Gardon 8714c6654bdSBen Gardon rcu_read_lock(); 8724c6654bdSBen Gardon } 8734c6654bdSBen Gardon 8744c6654bdSBen Gardon rcu_read_unlock(); 8754c6654bdSBen Gardon 8764c6654bdSBen Gardon if (flush) 8774c6654bdSBen Gardon kvm_flush_remote_tlbs(kvm); 8784c6654bdSBen Gardon } 8794c6654bdSBen Gardon 880bb18842eSBen Gardon /* 881b7cccd39SBen Gardon * Mark each TDP MMU root as invalid so that other threads 882b7cccd39SBen Gardon * will drop their references and allow the root count to 883b7cccd39SBen Gardon * go to 0. 884b7cccd39SBen Gardon * 8854c6654bdSBen Gardon * Also take a reference on all roots so that this thread 8864c6654bdSBen Gardon * can do the bulk of the work required to free the roots 8874c6654bdSBen Gardon * once they are invalidated. Without this reference, a 8884c6654bdSBen Gardon * vCPU thread might drop the last reference to a root and 8894c6654bdSBen Gardon * get stuck with tearing down the entire paging structure. 8904c6654bdSBen Gardon * 8914c6654bdSBen Gardon * Roots which have a zero refcount should be skipped as 8924c6654bdSBen Gardon * they're already being torn down. 8934c6654bdSBen Gardon * Already invalid roots should be referenced again so that 8944c6654bdSBen Gardon * they aren't freed before kvm_tdp_mmu_zap_all_fast is 8954c6654bdSBen Gardon * done with them. 8964c6654bdSBen Gardon * 897b7cccd39SBen Gardon * This has essentially the same effect for the TDP MMU 898b7cccd39SBen Gardon * as updating mmu_valid_gen does for the shadow MMU. 899b7cccd39SBen Gardon */ 900b7cccd39SBen Gardon void kvm_tdp_mmu_invalidate_all_roots(struct kvm *kvm) 901b7cccd39SBen Gardon { 902b7cccd39SBen Gardon struct kvm_mmu_page *root; 903b7cccd39SBen Gardon 904b7cccd39SBen Gardon lockdep_assert_held_write(&kvm->mmu_lock); 905b7cccd39SBen Gardon list_for_each_entry(root, &kvm->arch.tdp_mmu_roots, link) 9064c6654bdSBen Gardon if (refcount_inc_not_zero(&root->tdp_mmu_root_count)) 907b7cccd39SBen Gardon root->role.invalid = true; 908b7cccd39SBen Gardon } 909b7cccd39SBen Gardon 910bb18842eSBen Gardon /* 911bb18842eSBen Gardon * Installs a last-level SPTE to handle a TDP page fault. 912bb18842eSBen Gardon * (NPT/EPT violation/misconfiguration) 913bb18842eSBen Gardon */ 914cdc47767SPaolo Bonzini static int tdp_mmu_map_handle_target_level(struct kvm_vcpu *vcpu, 915cdc47767SPaolo Bonzini struct kvm_page_fault *fault, 916cdc47767SPaolo Bonzini struct tdp_iter *iter) 917bb18842eSBen Gardon { 918c435d4b7SSean Christopherson struct kvm_mmu_page *sp = sptep_to_sp(rcu_dereference(iter->sptep)); 919bb18842eSBen Gardon u64 new_spte; 92057a3e96dSKai Huang int ret = RET_PF_FIXED; 921ad67e480SPaolo Bonzini bool wrprot = false; 922bb18842eSBen Gardon 9237158bee4SPaolo Bonzini WARN_ON(sp->role.level != fault->goal_level); 924e710c5f6SDavid Matlack if (unlikely(!fault->slot)) 925bb18842eSBen Gardon new_spte = make_mmio_spte(vcpu, iter->gfn, ACC_ALL); 9269a77daacSBen Gardon else 92753597858SDavid Matlack wrprot = make_spte(vcpu, sp, fault->slot, ACC_ALL, iter->gfn, 9282839180cSPaolo Bonzini fault->pfn, iter->old_spte, fault->prefetch, true, 9297158bee4SPaolo Bonzini fault->map_writable, &new_spte); 930bb18842eSBen Gardon 931bb18842eSBen Gardon if (new_spte == iter->old_spte) 932bb18842eSBen Gardon ret = RET_PF_SPURIOUS; 9336ccf4438SPaolo Bonzini else if (!tdp_mmu_set_spte_atomic(vcpu->kvm, iter, new_spte)) 9349a77daacSBen Gardon return RET_PF_RETRY; 935bb18842eSBen Gardon 936bb18842eSBen Gardon /* 937bb18842eSBen Gardon * If the page fault was caused by a write but the page is write 938bb18842eSBen Gardon * protected, emulation is needed. If the emulation was skipped, 939bb18842eSBen Gardon * the vCPU would have the same fault again. 940bb18842eSBen Gardon */ 941ad67e480SPaolo Bonzini if (wrprot) { 942cdc47767SPaolo Bonzini if (fault->write) 943bb18842eSBen Gardon ret = RET_PF_EMULATE; 944bb18842eSBen Gardon } 945bb18842eSBen Gardon 946bb18842eSBen Gardon /* If a MMIO SPTE is installed, the MMIO will need to be emulated. */ 9479a77daacSBen Gardon if (unlikely(is_mmio_spte(new_spte))) { 9489a77daacSBen Gardon trace_mark_mmio_spte(rcu_dereference(iter->sptep), iter->gfn, 9499a77daacSBen Gardon new_spte); 950bb18842eSBen Gardon ret = RET_PF_EMULATE; 9513849e092SSean Christopherson } else { 9529a77daacSBen Gardon trace_kvm_mmu_set_spte(iter->level, iter->gfn, 9539a77daacSBen Gardon rcu_dereference(iter->sptep)); 9543849e092SSean Christopherson } 955bb18842eSBen Gardon 956857f8474SKai Huang /* 957857f8474SKai Huang * Increase pf_fixed in both RET_PF_EMULATE and RET_PF_FIXED to be 958857f8474SKai Huang * consistent with legacy MMU behavior. 959857f8474SKai Huang */ 960857f8474SKai Huang if (ret != RET_PF_SPURIOUS) 961bb18842eSBen Gardon vcpu->stat.pf_fixed++; 962bb18842eSBen Gardon 963bb18842eSBen Gardon return ret; 964bb18842eSBen Gardon } 965bb18842eSBen Gardon 966bb18842eSBen Gardon /* 967bb18842eSBen Gardon * Handle a TDP page fault (NPT/EPT violation/misconfiguration) by installing 968bb18842eSBen Gardon * page tables and SPTEs to translate the faulting guest physical address. 969bb18842eSBen Gardon */ 9702f6305ddSPaolo Bonzini int kvm_tdp_mmu_map(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault) 971bb18842eSBen Gardon { 972bb18842eSBen Gardon struct kvm_mmu *mmu = vcpu->arch.mmu; 973bb18842eSBen Gardon struct tdp_iter iter; 97489c0fd49SBen Gardon struct kvm_mmu_page *sp; 975bb18842eSBen Gardon u64 *child_pt; 976bb18842eSBen Gardon u64 new_spte; 977bb18842eSBen Gardon int ret; 978bb18842eSBen Gardon 97973a3c659SPaolo Bonzini kvm_mmu_hugepage_adjust(vcpu, fault); 980bb18842eSBen Gardon 981f0066d94SPaolo Bonzini trace_kvm_mmu_spte_requested(fault); 9827cca2d0bSBen Gardon 9837cca2d0bSBen Gardon rcu_read_lock(); 9847cca2d0bSBen Gardon 9852f6305ddSPaolo Bonzini tdp_mmu_for_each_pte(iter, mmu, fault->gfn, fault->gfn + 1) { 98673a3c659SPaolo Bonzini if (fault->nx_huge_page_workaround_enabled) 987536f0e6aSPaolo Bonzini disallowed_hugepage_adjust(fault, iter.old_spte, iter.level); 988bb18842eSBen Gardon 98973a3c659SPaolo Bonzini if (iter.level == fault->goal_level) 990bb18842eSBen Gardon break; 991bb18842eSBen Gardon 992bb18842eSBen Gardon /* 993bb18842eSBen Gardon * If there is an SPTE mapping a large page at a higher level 994bb18842eSBen Gardon * than the target, that SPTE must be cleared and replaced 995bb18842eSBen Gardon * with a non-leaf SPTE. 996bb18842eSBen Gardon */ 997bb18842eSBen Gardon if (is_shadow_present_pte(iter.old_spte) && 998bb18842eSBen Gardon is_large_pte(iter.old_spte)) { 99908f07c80SBen Gardon if (!tdp_mmu_zap_spte_atomic(vcpu->kvm, &iter)) 10009a77daacSBen Gardon break; 1001bb18842eSBen Gardon 1002bb18842eSBen Gardon /* 1003bb18842eSBen Gardon * The iter must explicitly re-read the spte here 1004bb18842eSBen Gardon * because the new value informs the !present 1005bb18842eSBen Gardon * path below. 1006bb18842eSBen Gardon */ 10077cca2d0bSBen Gardon iter.old_spte = READ_ONCE(*rcu_dereference(iter.sptep)); 1008bb18842eSBen Gardon } 1009bb18842eSBen Gardon 1010bb18842eSBen Gardon if (!is_shadow_present_pte(iter.old_spte)) { 1011ff76d506SKai Huang /* 1012c4342633SIngo Molnar * If SPTE has been frozen by another thread, just 1013ff76d506SKai Huang * give up and retry, avoiding unnecessary page table 1014ff76d506SKai Huang * allocation and free. 1015ff76d506SKai Huang */ 1016ff76d506SKai Huang if (is_removed_spte(iter.old_spte)) 1017ff76d506SKai Huang break; 1018ff76d506SKai Huang 1019f1b83255SKai Huang sp = alloc_tdp_mmu_page(vcpu, iter.gfn, iter.level - 1); 102089c0fd49SBen Gardon child_pt = sp->spt; 1021a9442f59SBen Gardon 1022bb18842eSBen Gardon new_spte = make_nonleaf_spte(child_pt, 1023bb18842eSBen Gardon !shadow_accessed_mask); 1024bb18842eSBen Gardon 10256ccf4438SPaolo Bonzini if (tdp_mmu_set_spte_atomic(vcpu->kvm, &iter, new_spte)) { 10269653f2daSSean Christopherson tdp_mmu_link_page(vcpu->kvm, sp, 102773a3c659SPaolo Bonzini fault->huge_page_disallowed && 102873a3c659SPaolo Bonzini fault->req_level >= iter.level); 10299a77daacSBen Gardon 1030bb18842eSBen Gardon trace_kvm_mmu_get_page(sp, true); 10319a77daacSBen Gardon } else { 10329a77daacSBen Gardon tdp_mmu_free_sp(sp); 10339a77daacSBen Gardon break; 10349a77daacSBen Gardon } 1035bb18842eSBen Gardon } 1036bb18842eSBen Gardon } 1037bb18842eSBen Gardon 103873a3c659SPaolo Bonzini if (iter.level != fault->goal_level) { 10397cca2d0bSBen Gardon rcu_read_unlock(); 1040bb18842eSBen Gardon return RET_PF_RETRY; 10417cca2d0bSBen Gardon } 1042bb18842eSBen Gardon 1043cdc47767SPaolo Bonzini ret = tdp_mmu_map_handle_target_level(vcpu, fault, &iter); 10447cca2d0bSBen Gardon rcu_read_unlock(); 1045bb18842eSBen Gardon 1046bb18842eSBen Gardon return ret; 1047bb18842eSBen Gardon } 1048063afacdSBen Gardon 10493039bcc7SSean Christopherson bool kvm_tdp_mmu_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range, 10503039bcc7SSean Christopherson bool flush) 1051063afacdSBen Gardon { 105283b83a02SSean Christopherson return __kvm_tdp_mmu_zap_gfn_range(kvm, range->slot->as_id, range->start, 105383b83a02SSean Christopherson range->end, range->may_block, flush); 10543039bcc7SSean Christopherson } 10553039bcc7SSean Christopherson 10563039bcc7SSean Christopherson typedef bool (*tdp_handler_t)(struct kvm *kvm, struct tdp_iter *iter, 10573039bcc7SSean Christopherson struct kvm_gfn_range *range); 10583039bcc7SSean Christopherson 10593039bcc7SSean Christopherson static __always_inline bool kvm_tdp_mmu_handle_gfn(struct kvm *kvm, 10603039bcc7SSean Christopherson struct kvm_gfn_range *range, 1061c1b91493SSean Christopherson tdp_handler_t handler) 1062063afacdSBen Gardon { 1063063afacdSBen Gardon struct kvm_mmu_page *root; 10643039bcc7SSean Christopherson struct tdp_iter iter; 10653039bcc7SSean Christopherson bool ret = false; 1066063afacdSBen Gardon 10673039bcc7SSean Christopherson rcu_read_lock(); 1068063afacdSBen Gardon 1069063afacdSBen Gardon /* 1070e1eed584SSean Christopherson * Don't support rescheduling, none of the MMU notifiers that funnel 1071e1eed584SSean Christopherson * into this helper allow blocking; it'd be dead, wasteful code. 1072063afacdSBen Gardon */ 10733039bcc7SSean Christopherson for_each_tdp_mmu_root(kvm, root, range->slot->as_id) { 10743039bcc7SSean Christopherson tdp_root_for_each_leaf_pte(iter, root, range->start, range->end) 10753039bcc7SSean Christopherson ret |= handler(kvm, &iter, range); 10763039bcc7SSean Christopherson } 1077063afacdSBen Gardon 10783039bcc7SSean Christopherson rcu_read_unlock(); 1079063afacdSBen Gardon 1080063afacdSBen Gardon return ret; 1081063afacdSBen Gardon } 1082063afacdSBen Gardon 1083f8e14497SBen Gardon /* 1084f8e14497SBen Gardon * Mark the SPTEs range of GFNs [start, end) unaccessed and return non-zero 1085f8e14497SBen Gardon * if any of the GFNs in the range have been accessed. 1086f8e14497SBen Gardon */ 10873039bcc7SSean Christopherson static bool age_gfn_range(struct kvm *kvm, struct tdp_iter *iter, 10883039bcc7SSean Christopherson struct kvm_gfn_range *range) 1089f8e14497SBen Gardon { 1090f8e14497SBen Gardon u64 new_spte = 0; 1091f8e14497SBen Gardon 10923039bcc7SSean Christopherson /* If we have a non-accessed entry we don't need to change the pte. */ 10933039bcc7SSean Christopherson if (!is_accessed_spte(iter->old_spte)) 10943039bcc7SSean Christopherson return false; 10957cca2d0bSBen Gardon 10963039bcc7SSean Christopherson new_spte = iter->old_spte; 1097f8e14497SBen Gardon 1098f8e14497SBen Gardon if (spte_ad_enabled(new_spte)) { 10998f8f52a4SSean Christopherson new_spte &= ~shadow_accessed_mask; 1100f8e14497SBen Gardon } else { 1101f8e14497SBen Gardon /* 1102f8e14497SBen Gardon * Capture the dirty status of the page, so that it doesn't get 1103f8e14497SBen Gardon * lost when the SPTE is marked for access tracking. 1104f8e14497SBen Gardon */ 1105f8e14497SBen Gardon if (is_writable_pte(new_spte)) 1106f8e14497SBen Gardon kvm_set_pfn_dirty(spte_to_pfn(new_spte)); 1107f8e14497SBen Gardon 1108f8e14497SBen Gardon new_spte = mark_spte_for_access_track(new_spte); 1109f8e14497SBen Gardon } 1110f8e14497SBen Gardon 11113039bcc7SSean Christopherson tdp_mmu_set_spte_no_acc_track(kvm, iter, new_spte); 111233dd3574SBen Gardon 11133039bcc7SSean Christopherson return true; 1114f8e14497SBen Gardon } 1115f8e14497SBen Gardon 11163039bcc7SSean Christopherson bool kvm_tdp_mmu_age_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range) 1117f8e14497SBen Gardon { 11183039bcc7SSean Christopherson return kvm_tdp_mmu_handle_gfn(kvm, range, age_gfn_range); 1119f8e14497SBen Gardon } 1120f8e14497SBen Gardon 11213039bcc7SSean Christopherson static bool test_age_gfn(struct kvm *kvm, struct tdp_iter *iter, 11223039bcc7SSean Christopherson struct kvm_gfn_range *range) 1123f8e14497SBen Gardon { 11243039bcc7SSean Christopherson return is_accessed_spte(iter->old_spte); 1125f8e14497SBen Gardon } 1126f8e14497SBen Gardon 11273039bcc7SSean Christopherson bool kvm_tdp_mmu_test_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range) 1128f8e14497SBen Gardon { 11293039bcc7SSean Christopherson return kvm_tdp_mmu_handle_gfn(kvm, range, test_age_gfn); 11303039bcc7SSean Christopherson } 11313039bcc7SSean Christopherson 11323039bcc7SSean Christopherson static bool set_spte_gfn(struct kvm *kvm, struct tdp_iter *iter, 11333039bcc7SSean Christopherson struct kvm_gfn_range *range) 11343039bcc7SSean Christopherson { 11353039bcc7SSean Christopherson u64 new_spte; 11363039bcc7SSean Christopherson 11373039bcc7SSean Christopherson /* Huge pages aren't expected to be modified without first being zapped. */ 11383039bcc7SSean Christopherson WARN_ON(pte_huge(range->pte) || range->start + 1 != range->end); 11393039bcc7SSean Christopherson 11403039bcc7SSean Christopherson if (iter->level != PG_LEVEL_4K || 11413039bcc7SSean Christopherson !is_shadow_present_pte(iter->old_spte)) 11423039bcc7SSean Christopherson return false; 11433039bcc7SSean Christopherson 11443039bcc7SSean Christopherson /* 11453039bcc7SSean Christopherson * Note, when changing a read-only SPTE, it's not strictly necessary to 11463039bcc7SSean Christopherson * zero the SPTE before setting the new PFN, but doing so preserves the 11473039bcc7SSean Christopherson * invariant that the PFN of a present * leaf SPTE can never change. 11483039bcc7SSean Christopherson * See __handle_changed_spte(). 11493039bcc7SSean Christopherson */ 11503039bcc7SSean Christopherson tdp_mmu_set_spte(kvm, iter, 0); 11513039bcc7SSean Christopherson 11523039bcc7SSean Christopherson if (!pte_write(range->pte)) { 11533039bcc7SSean Christopherson new_spte = kvm_mmu_changed_pte_notifier_make_spte(iter->old_spte, 11543039bcc7SSean Christopherson pte_pfn(range->pte)); 11553039bcc7SSean Christopherson 11563039bcc7SSean Christopherson tdp_mmu_set_spte(kvm, iter, new_spte); 11573039bcc7SSean Christopherson } 11583039bcc7SSean Christopherson 11593039bcc7SSean Christopherson return true; 1160f8e14497SBen Gardon } 11611d8dd6b3SBen Gardon 11621d8dd6b3SBen Gardon /* 11631d8dd6b3SBen Gardon * Handle the changed_pte MMU notifier for the TDP MMU. 11641d8dd6b3SBen Gardon * data is a pointer to the new pte_t mapping the HVA specified by the MMU 11651d8dd6b3SBen Gardon * notifier. 11661d8dd6b3SBen Gardon * Returns non-zero if a flush is needed before releasing the MMU lock. 11671d8dd6b3SBen Gardon */ 11683039bcc7SSean Christopherson bool kvm_tdp_mmu_set_spte_gfn(struct kvm *kvm, struct kvm_gfn_range *range) 11691d8dd6b3SBen Gardon { 11703039bcc7SSean Christopherson bool flush = kvm_tdp_mmu_handle_gfn(kvm, range, set_spte_gfn); 11711d8dd6b3SBen Gardon 11723039bcc7SSean Christopherson /* FIXME: return 'flush' instead of flushing here. */ 11733039bcc7SSean Christopherson if (flush) 11743039bcc7SSean Christopherson kvm_flush_remote_tlbs_with_address(kvm, range->start, 1); 11757cca2d0bSBen Gardon 11763039bcc7SSean Christopherson return false; 11771d8dd6b3SBen Gardon } 11781d8dd6b3SBen Gardon 1179a6a0b05dSBen Gardon /* 1180bedd9195SDavid Matlack * Remove write access from all SPTEs at or above min_level that map GFNs 1181bedd9195SDavid Matlack * [start, end). Returns true if an SPTE has been changed and the TLBs need to 1182bedd9195SDavid Matlack * be flushed. 1183a6a0b05dSBen Gardon */ 1184a6a0b05dSBen Gardon static bool wrprot_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root, 1185a6a0b05dSBen Gardon gfn_t start, gfn_t end, int min_level) 1186a6a0b05dSBen Gardon { 1187a6a0b05dSBen Gardon struct tdp_iter iter; 1188a6a0b05dSBen Gardon u64 new_spte; 1189a6a0b05dSBen Gardon bool spte_set = false; 1190a6a0b05dSBen Gardon 11917cca2d0bSBen Gardon rcu_read_lock(); 11927cca2d0bSBen Gardon 1193a6a0b05dSBen Gardon BUG_ON(min_level > KVM_MAX_HUGEPAGE_LEVEL); 1194a6a0b05dSBen Gardon 1195a6a0b05dSBen Gardon for_each_tdp_pte_min_level(iter, root->spt, root->role.level, 1196a6a0b05dSBen Gardon min_level, start, end) { 119724ae4cfaSBen Gardon retry: 119824ae4cfaSBen Gardon if (tdp_mmu_iter_cond_resched(kvm, &iter, false, true)) 11991af4a960SBen Gardon continue; 12001af4a960SBen Gardon 1201a6a0b05dSBen Gardon if (!is_shadow_present_pte(iter.old_spte) || 12020f99ee2cSBen Gardon !is_last_spte(iter.old_spte, iter.level) || 12030f99ee2cSBen Gardon !(iter.old_spte & PT_WRITABLE_MASK)) 1204a6a0b05dSBen Gardon continue; 1205a6a0b05dSBen Gardon 1206a6a0b05dSBen Gardon new_spte = iter.old_spte & ~PT_WRITABLE_MASK; 1207a6a0b05dSBen Gardon 12086ccf4438SPaolo Bonzini if (!tdp_mmu_set_spte_atomic(kvm, &iter, new_spte)) { 120924ae4cfaSBen Gardon /* 121024ae4cfaSBen Gardon * The iter must explicitly re-read the SPTE because 121124ae4cfaSBen Gardon * the atomic cmpxchg failed. 121224ae4cfaSBen Gardon */ 121324ae4cfaSBen Gardon iter.old_spte = READ_ONCE(*rcu_dereference(iter.sptep)); 121424ae4cfaSBen Gardon goto retry; 121524ae4cfaSBen Gardon } 1216a6a0b05dSBen Gardon spte_set = true; 1217a6a0b05dSBen Gardon } 12187cca2d0bSBen Gardon 12197cca2d0bSBen Gardon rcu_read_unlock(); 1220a6a0b05dSBen Gardon return spte_set; 1221a6a0b05dSBen Gardon } 1222a6a0b05dSBen Gardon 1223a6a0b05dSBen Gardon /* 1224a6a0b05dSBen Gardon * Remove write access from all the SPTEs mapping GFNs in the memslot. Will 1225a6a0b05dSBen Gardon * only affect leaf SPTEs down to min_level. 1226a6a0b05dSBen Gardon * Returns true if an SPTE has been changed and the TLBs need to be flushed. 1227a6a0b05dSBen Gardon */ 1228269e9552SHamza Mahfooz bool kvm_tdp_mmu_wrprot_slot(struct kvm *kvm, 1229269e9552SHamza Mahfooz const struct kvm_memory_slot *slot, int min_level) 1230a6a0b05dSBen Gardon { 1231a6a0b05dSBen Gardon struct kvm_mmu_page *root; 1232a6a0b05dSBen Gardon bool spte_set = false; 1233a6a0b05dSBen Gardon 123424ae4cfaSBen Gardon lockdep_assert_held_read(&kvm->mmu_lock); 1235a6a0b05dSBen Gardon 1236*d62007edSSean Christopherson for_each_valid_tdp_mmu_root_yield_safe(kvm, root, slot->as_id, true) 1237a6a0b05dSBen Gardon spte_set |= wrprot_gfn_range(kvm, root, slot->base_gfn, 1238a6a0b05dSBen Gardon slot->base_gfn + slot->npages, min_level); 1239a6a0b05dSBen Gardon 1240a6a0b05dSBen Gardon return spte_set; 1241a6a0b05dSBen Gardon } 1242a6a0b05dSBen Gardon 1243a6a0b05dSBen Gardon /* 1244a6a0b05dSBen Gardon * Clear the dirty status of all the SPTEs mapping GFNs in the memslot. If 1245a6a0b05dSBen Gardon * AD bits are enabled, this will involve clearing the dirty bit on each SPTE. 1246a6a0b05dSBen Gardon * If AD bits are not enabled, this will require clearing the writable bit on 1247a6a0b05dSBen Gardon * each SPTE. Returns true if an SPTE has been changed and the TLBs need to 1248a6a0b05dSBen Gardon * be flushed. 1249a6a0b05dSBen Gardon */ 1250a6a0b05dSBen Gardon static bool clear_dirty_gfn_range(struct kvm *kvm, struct kvm_mmu_page *root, 1251a6a0b05dSBen Gardon gfn_t start, gfn_t end) 1252a6a0b05dSBen Gardon { 1253a6a0b05dSBen Gardon struct tdp_iter iter; 1254a6a0b05dSBen Gardon u64 new_spte; 1255a6a0b05dSBen Gardon bool spte_set = false; 1256a6a0b05dSBen Gardon 12577cca2d0bSBen Gardon rcu_read_lock(); 12587cca2d0bSBen Gardon 1259a6a0b05dSBen Gardon tdp_root_for_each_leaf_pte(iter, root, start, end) { 126024ae4cfaSBen Gardon retry: 126124ae4cfaSBen Gardon if (tdp_mmu_iter_cond_resched(kvm, &iter, false, true)) 12621af4a960SBen Gardon continue; 12631af4a960SBen Gardon 1264a6a0b05dSBen Gardon if (spte_ad_need_write_protect(iter.old_spte)) { 1265a6a0b05dSBen Gardon if (is_writable_pte(iter.old_spte)) 1266a6a0b05dSBen Gardon new_spte = iter.old_spte & ~PT_WRITABLE_MASK; 1267a6a0b05dSBen Gardon else 1268a6a0b05dSBen Gardon continue; 1269a6a0b05dSBen Gardon } else { 1270a6a0b05dSBen Gardon if (iter.old_spte & shadow_dirty_mask) 1271a6a0b05dSBen Gardon new_spte = iter.old_spte & ~shadow_dirty_mask; 1272a6a0b05dSBen Gardon else 1273a6a0b05dSBen Gardon continue; 1274a6a0b05dSBen Gardon } 1275a6a0b05dSBen Gardon 12766ccf4438SPaolo Bonzini if (!tdp_mmu_set_spte_atomic(kvm, &iter, new_spte)) { 127724ae4cfaSBen Gardon /* 127824ae4cfaSBen Gardon * The iter must explicitly re-read the SPTE because 127924ae4cfaSBen Gardon * the atomic cmpxchg failed. 128024ae4cfaSBen Gardon */ 128124ae4cfaSBen Gardon iter.old_spte = READ_ONCE(*rcu_dereference(iter.sptep)); 128224ae4cfaSBen Gardon goto retry; 128324ae4cfaSBen Gardon } 1284a6a0b05dSBen Gardon spte_set = true; 1285a6a0b05dSBen Gardon } 12867cca2d0bSBen Gardon 12877cca2d0bSBen Gardon rcu_read_unlock(); 1288a6a0b05dSBen Gardon return spte_set; 1289a6a0b05dSBen Gardon } 1290a6a0b05dSBen Gardon 1291a6a0b05dSBen Gardon /* 1292a6a0b05dSBen Gardon * Clear the dirty status of all the SPTEs mapping GFNs in the memslot. If 1293a6a0b05dSBen Gardon * AD bits are enabled, this will involve clearing the dirty bit on each SPTE. 1294a6a0b05dSBen Gardon * If AD bits are not enabled, this will require clearing the writable bit on 1295a6a0b05dSBen Gardon * each SPTE. Returns true if an SPTE has been changed and the TLBs need to 1296a6a0b05dSBen Gardon * be flushed. 1297a6a0b05dSBen Gardon */ 1298269e9552SHamza Mahfooz bool kvm_tdp_mmu_clear_dirty_slot(struct kvm *kvm, 1299269e9552SHamza Mahfooz const struct kvm_memory_slot *slot) 1300a6a0b05dSBen Gardon { 1301a6a0b05dSBen Gardon struct kvm_mmu_page *root; 1302a6a0b05dSBen Gardon bool spte_set = false; 1303a6a0b05dSBen Gardon 130424ae4cfaSBen Gardon lockdep_assert_held_read(&kvm->mmu_lock); 1305a6a0b05dSBen Gardon 1306*d62007edSSean Christopherson for_each_valid_tdp_mmu_root_yield_safe(kvm, root, slot->as_id, true) 1307a6a0b05dSBen Gardon spte_set |= clear_dirty_gfn_range(kvm, root, slot->base_gfn, 1308a6a0b05dSBen Gardon slot->base_gfn + slot->npages); 1309a6a0b05dSBen Gardon 1310a6a0b05dSBen Gardon return spte_set; 1311a6a0b05dSBen Gardon } 1312a6a0b05dSBen Gardon 1313a6a0b05dSBen Gardon /* 1314a6a0b05dSBen Gardon * Clears the dirty status of all the 4k SPTEs mapping GFNs for which a bit is 1315a6a0b05dSBen Gardon * set in mask, starting at gfn. The given memslot is expected to contain all 1316a6a0b05dSBen Gardon * the GFNs represented by set bits in the mask. If AD bits are enabled, 1317a6a0b05dSBen Gardon * clearing the dirty status will involve clearing the dirty bit on each SPTE 1318a6a0b05dSBen Gardon * or, if AD bits are not enabled, clearing the writable bit on each SPTE. 1319a6a0b05dSBen Gardon */ 1320a6a0b05dSBen Gardon static void clear_dirty_pt_masked(struct kvm *kvm, struct kvm_mmu_page *root, 1321a6a0b05dSBen Gardon gfn_t gfn, unsigned long mask, bool wrprot) 1322a6a0b05dSBen Gardon { 1323a6a0b05dSBen Gardon struct tdp_iter iter; 1324a6a0b05dSBen Gardon u64 new_spte; 1325a6a0b05dSBen Gardon 13267cca2d0bSBen Gardon rcu_read_lock(); 13277cca2d0bSBen Gardon 1328a6a0b05dSBen Gardon tdp_root_for_each_leaf_pte(iter, root, gfn + __ffs(mask), 1329a6a0b05dSBen Gardon gfn + BITS_PER_LONG) { 1330a6a0b05dSBen Gardon if (!mask) 1331a6a0b05dSBen Gardon break; 1332a6a0b05dSBen Gardon 1333a6a0b05dSBen Gardon if (iter.level > PG_LEVEL_4K || 1334a6a0b05dSBen Gardon !(mask & (1UL << (iter.gfn - gfn)))) 1335a6a0b05dSBen Gardon continue; 1336a6a0b05dSBen Gardon 1337f1b3b06aSBen Gardon mask &= ~(1UL << (iter.gfn - gfn)); 1338f1b3b06aSBen Gardon 1339a6a0b05dSBen Gardon if (wrprot || spte_ad_need_write_protect(iter.old_spte)) { 1340a6a0b05dSBen Gardon if (is_writable_pte(iter.old_spte)) 1341a6a0b05dSBen Gardon new_spte = iter.old_spte & ~PT_WRITABLE_MASK; 1342a6a0b05dSBen Gardon else 1343a6a0b05dSBen Gardon continue; 1344a6a0b05dSBen Gardon } else { 1345a6a0b05dSBen Gardon if (iter.old_spte & shadow_dirty_mask) 1346a6a0b05dSBen Gardon new_spte = iter.old_spte & ~shadow_dirty_mask; 1347a6a0b05dSBen Gardon else 1348a6a0b05dSBen Gardon continue; 1349a6a0b05dSBen Gardon } 1350a6a0b05dSBen Gardon 1351a6a0b05dSBen Gardon tdp_mmu_set_spte_no_dirty_log(kvm, &iter, new_spte); 1352a6a0b05dSBen Gardon } 13537cca2d0bSBen Gardon 13547cca2d0bSBen Gardon rcu_read_unlock(); 1355a6a0b05dSBen Gardon } 1356a6a0b05dSBen Gardon 1357a6a0b05dSBen Gardon /* 1358a6a0b05dSBen Gardon * Clears the dirty status of all the 4k SPTEs mapping GFNs for which a bit is 1359a6a0b05dSBen Gardon * set in mask, starting at gfn. The given memslot is expected to contain all 1360a6a0b05dSBen Gardon * the GFNs represented by set bits in the mask. If AD bits are enabled, 1361a6a0b05dSBen Gardon * clearing the dirty status will involve clearing the dirty bit on each SPTE 1362a6a0b05dSBen Gardon * or, if AD bits are not enabled, clearing the writable bit on each SPTE. 1363a6a0b05dSBen Gardon */ 1364a6a0b05dSBen Gardon void kvm_tdp_mmu_clear_dirty_pt_masked(struct kvm *kvm, 1365a6a0b05dSBen Gardon struct kvm_memory_slot *slot, 1366a6a0b05dSBen Gardon gfn_t gfn, unsigned long mask, 1367a6a0b05dSBen Gardon bool wrprot) 1368a6a0b05dSBen Gardon { 1369a6a0b05dSBen Gardon struct kvm_mmu_page *root; 1370a6a0b05dSBen Gardon 1371531810caSBen Gardon lockdep_assert_held_write(&kvm->mmu_lock); 1372a3f15bdaSSean Christopherson for_each_tdp_mmu_root(kvm, root, slot->as_id) 1373a6a0b05dSBen Gardon clear_dirty_pt_masked(kvm, root, gfn, mask, wrprot); 1374a6a0b05dSBen Gardon } 1375a6a0b05dSBen Gardon 1376a6a0b05dSBen Gardon /* 137787aa9ec9SBen Gardon * Clear leaf entries which could be replaced by large mappings, for 137887aa9ec9SBen Gardon * GFNs within the slot. 137914881998SBen Gardon */ 13804b85c921SSean Christopherson static void zap_collapsible_spte_range(struct kvm *kvm, 138114881998SBen Gardon struct kvm_mmu_page *root, 13824b85c921SSean Christopherson const struct kvm_memory_slot *slot) 138314881998SBen Gardon { 13849eba50f8SSean Christopherson gfn_t start = slot->base_gfn; 13859eba50f8SSean Christopherson gfn_t end = start + slot->npages; 138614881998SBen Gardon struct tdp_iter iter; 138714881998SBen Gardon kvm_pfn_t pfn; 138814881998SBen Gardon 13897cca2d0bSBen Gardon rcu_read_lock(); 13907cca2d0bSBen Gardon 139114881998SBen Gardon tdp_root_for_each_pte(iter, root, start, end) { 13922db6f772SBen Gardon retry: 13934b85c921SSean Christopherson if (tdp_mmu_iter_cond_resched(kvm, &iter, false, true)) 13941af4a960SBen Gardon continue; 13951af4a960SBen Gardon 139614881998SBen Gardon if (!is_shadow_present_pte(iter.old_spte) || 139787aa9ec9SBen Gardon !is_last_spte(iter.old_spte, iter.level)) 139814881998SBen Gardon continue; 139914881998SBen Gardon 140014881998SBen Gardon pfn = spte_to_pfn(iter.old_spte); 140114881998SBen Gardon if (kvm_is_reserved_pfn(pfn) || 14029eba50f8SSean Christopherson iter.level >= kvm_mmu_max_mapping_level(kvm, slot, iter.gfn, 14039eba50f8SSean Christopherson pfn, PG_LEVEL_NUM)) 140414881998SBen Gardon continue; 140514881998SBen Gardon 14064b85c921SSean Christopherson /* Note, a successful atomic zap also does a remote TLB flush. */ 14072db6f772SBen Gardon if (!tdp_mmu_zap_spte_atomic(kvm, &iter)) { 14082db6f772SBen Gardon /* 14092db6f772SBen Gardon * The iter must explicitly re-read the SPTE because 14102db6f772SBen Gardon * the atomic cmpxchg failed. 14112db6f772SBen Gardon */ 14122db6f772SBen Gardon iter.old_spte = READ_ONCE(*rcu_dereference(iter.sptep)); 14132db6f772SBen Gardon goto retry; 14142db6f772SBen Gardon } 141514881998SBen Gardon } 141614881998SBen Gardon 14177cca2d0bSBen Gardon rcu_read_unlock(); 141814881998SBen Gardon } 141914881998SBen Gardon 142014881998SBen Gardon /* 142114881998SBen Gardon * Clear non-leaf entries (and free associated page tables) which could 142214881998SBen Gardon * be replaced by large mappings, for GFNs within the slot. 142314881998SBen Gardon */ 14244b85c921SSean Christopherson void kvm_tdp_mmu_zap_collapsible_sptes(struct kvm *kvm, 14254b85c921SSean Christopherson const struct kvm_memory_slot *slot) 142614881998SBen Gardon { 142714881998SBen Gardon struct kvm_mmu_page *root; 142814881998SBen Gardon 14292db6f772SBen Gardon lockdep_assert_held_read(&kvm->mmu_lock); 143014881998SBen Gardon 1431*d62007edSSean Christopherson for_each_valid_tdp_mmu_root_yield_safe(kvm, root, slot->as_id, true) 14324b85c921SSean Christopherson zap_collapsible_spte_range(kvm, root, slot); 143314881998SBen Gardon } 143446044f72SBen Gardon 143546044f72SBen Gardon /* 143646044f72SBen Gardon * Removes write access on the last level SPTE mapping this GFN and unsets the 14375fc3424fSSean Christopherson * MMU-writable bit to ensure future writes continue to be intercepted. 143846044f72SBen Gardon * Returns true if an SPTE was set and a TLB flush is needed. 143946044f72SBen Gardon */ 144046044f72SBen Gardon static bool write_protect_gfn(struct kvm *kvm, struct kvm_mmu_page *root, 14413ad93562SKeqian Zhu gfn_t gfn, int min_level) 144246044f72SBen Gardon { 144346044f72SBen Gardon struct tdp_iter iter; 144446044f72SBen Gardon u64 new_spte; 144546044f72SBen Gardon bool spte_set = false; 144646044f72SBen Gardon 14473ad93562SKeqian Zhu BUG_ON(min_level > KVM_MAX_HUGEPAGE_LEVEL); 14483ad93562SKeqian Zhu 14497cca2d0bSBen Gardon rcu_read_lock(); 14507cca2d0bSBen Gardon 14513ad93562SKeqian Zhu for_each_tdp_pte_min_level(iter, root->spt, root->role.level, 14523ad93562SKeqian Zhu min_level, gfn, gfn + 1) { 14533ad93562SKeqian Zhu if (!is_shadow_present_pte(iter.old_spte) || 14543ad93562SKeqian Zhu !is_last_spte(iter.old_spte, iter.level)) 14553ad93562SKeqian Zhu continue; 14563ad93562SKeqian Zhu 145746044f72SBen Gardon new_spte = iter.old_spte & 14585fc3424fSSean Christopherson ~(PT_WRITABLE_MASK | shadow_mmu_writable_mask); 145946044f72SBen Gardon 14607c8a4742SDavid Matlack if (new_spte == iter.old_spte) 14617c8a4742SDavid Matlack break; 14627c8a4742SDavid Matlack 146346044f72SBen Gardon tdp_mmu_set_spte(kvm, &iter, new_spte); 146446044f72SBen Gardon spte_set = true; 146546044f72SBen Gardon } 146646044f72SBen Gardon 14677cca2d0bSBen Gardon rcu_read_unlock(); 14687cca2d0bSBen Gardon 146946044f72SBen Gardon return spte_set; 147046044f72SBen Gardon } 147146044f72SBen Gardon 147246044f72SBen Gardon /* 147346044f72SBen Gardon * Removes write access on the last level SPTE mapping this GFN and unsets the 14745fc3424fSSean Christopherson * MMU-writable bit to ensure future writes continue to be intercepted. 147546044f72SBen Gardon * Returns true if an SPTE was set and a TLB flush is needed. 147646044f72SBen Gardon */ 147746044f72SBen Gardon bool kvm_tdp_mmu_write_protect_gfn(struct kvm *kvm, 14783ad93562SKeqian Zhu struct kvm_memory_slot *slot, gfn_t gfn, 14793ad93562SKeqian Zhu int min_level) 148046044f72SBen Gardon { 148146044f72SBen Gardon struct kvm_mmu_page *root; 148246044f72SBen Gardon bool spte_set = false; 148346044f72SBen Gardon 1484531810caSBen Gardon lockdep_assert_held_write(&kvm->mmu_lock); 1485a3f15bdaSSean Christopherson for_each_tdp_mmu_root(kvm, root, slot->as_id) 14863ad93562SKeqian Zhu spte_set |= write_protect_gfn(kvm, root, gfn, min_level); 1487a3f15bdaSSean Christopherson 148846044f72SBen Gardon return spte_set; 148946044f72SBen Gardon } 149046044f72SBen Gardon 149195fb5b02SBen Gardon /* 149295fb5b02SBen Gardon * Return the level of the lowest level SPTE added to sptes. 149395fb5b02SBen Gardon * That SPTE may be non-present. 1494c5c8c7c5SDavid Matlack * 1495c5c8c7c5SDavid Matlack * Must be called between kvm_tdp_mmu_walk_lockless_{begin,end}. 149695fb5b02SBen Gardon */ 149739b4d43eSSean Christopherson int kvm_tdp_mmu_get_walk(struct kvm_vcpu *vcpu, u64 addr, u64 *sptes, 149839b4d43eSSean Christopherson int *root_level) 149995fb5b02SBen Gardon { 150095fb5b02SBen Gardon struct tdp_iter iter; 150195fb5b02SBen Gardon struct kvm_mmu *mmu = vcpu->arch.mmu; 150295fb5b02SBen Gardon gfn_t gfn = addr >> PAGE_SHIFT; 15032aa07893SSean Christopherson int leaf = -1; 150495fb5b02SBen Gardon 150539b4d43eSSean Christopherson *root_level = vcpu->arch.mmu->shadow_root_level; 150695fb5b02SBen Gardon 150795fb5b02SBen Gardon tdp_mmu_for_each_pte(iter, mmu, gfn, gfn + 1) { 150895fb5b02SBen Gardon leaf = iter.level; 1509dde81f94SSean Christopherson sptes[leaf] = iter.old_spte; 151095fb5b02SBen Gardon } 151195fb5b02SBen Gardon 151295fb5b02SBen Gardon return leaf; 151395fb5b02SBen Gardon } 15146e8eb206SDavid Matlack 15156e8eb206SDavid Matlack /* 15166e8eb206SDavid Matlack * Returns the last level spte pointer of the shadow page walk for the given 15176e8eb206SDavid Matlack * gpa, and sets *spte to the spte value. This spte may be non-preset. If no 15186e8eb206SDavid Matlack * walk could be performed, returns NULL and *spte does not contain valid data. 15196e8eb206SDavid Matlack * 15206e8eb206SDavid Matlack * Contract: 15216e8eb206SDavid Matlack * - Must be called between kvm_tdp_mmu_walk_lockless_{begin,end}. 15226e8eb206SDavid Matlack * - The returned sptep must not be used after kvm_tdp_mmu_walk_lockless_end. 15236e8eb206SDavid Matlack * 15246e8eb206SDavid Matlack * WARNING: This function is only intended to be called during fast_page_fault. 15256e8eb206SDavid Matlack */ 15266e8eb206SDavid Matlack u64 *kvm_tdp_mmu_fast_pf_get_last_sptep(struct kvm_vcpu *vcpu, u64 addr, 15276e8eb206SDavid Matlack u64 *spte) 15286e8eb206SDavid Matlack { 15296e8eb206SDavid Matlack struct tdp_iter iter; 15306e8eb206SDavid Matlack struct kvm_mmu *mmu = vcpu->arch.mmu; 15316e8eb206SDavid Matlack gfn_t gfn = addr >> PAGE_SHIFT; 15326e8eb206SDavid Matlack tdp_ptep_t sptep = NULL; 15336e8eb206SDavid Matlack 15346e8eb206SDavid Matlack tdp_mmu_for_each_pte(iter, mmu, gfn, gfn + 1) { 15356e8eb206SDavid Matlack *spte = iter.old_spte; 15366e8eb206SDavid Matlack sptep = iter.sptep; 15376e8eb206SDavid Matlack } 15386e8eb206SDavid Matlack 15396e8eb206SDavid Matlack /* 15406e8eb206SDavid Matlack * Perform the rcu_dereference to get the raw spte pointer value since 15416e8eb206SDavid Matlack * we are passing it up to fast_page_fault, which is shared with the 15426e8eb206SDavid Matlack * legacy MMU and thus does not retain the TDP MMU-specific __rcu 15436e8eb206SDavid Matlack * annotation. 15446e8eb206SDavid Matlack * 15456e8eb206SDavid Matlack * This is safe since fast_page_fault obeys the contracts of this 15466e8eb206SDavid Matlack * function as well as all TDP MMU contracts around modifying SPTEs 15476e8eb206SDavid Matlack * outside of mmu_lock. 15486e8eb206SDavid Matlack */ 15496e8eb206SDavid Matlack return rcu_dereference(sptep); 15506e8eb206SDavid Matlack } 1551