Merge tag 'kvm-x86-mmu-6.9' of https://github.com/kvm-x86/linux into HEAD

KVM x86 MMU changes for 6.9:

 - Clean up code related to unprotecting shadow pages when retrying a guest
   instruction after failed #PF-induced emulation.

 - Zap TDP MMU roots at 4KiB granularity to minimize the delay in yielding if
   a reschedule is needed, e.g. if a high priority task needs to run.  Because
   KVM doesn't support yielding in the middle of processing a zapped non-leaf
   SPTE, zapping at 1GiB granularity can result in multi-millisecond lag when
   attempting to schedule in a high priority.

 - Rework TDP MMU root unload, free, and alloc to run with mmu_lock held for
   read, e.g. to avoid serializing vCPUs when userspace deletes a memslot.

 - Allocate write-tracking metadata on-demand to avoid the memory overhead when
   running kernels built with KVMGT support (external write-tracking enabled),
   but for workloads that don't use nested virtualization (shadow paging) or
   KVMGT.

arch/x86/include/asm/kvm_host.h

@@ -1468,6 +1468,15 @@ struct kvm_arch {
 	 */
 	bool shadow_root_allocated;
 
+#ifdef CONFIG_KVM_EXTERNAL_WRITE_TRACKING
+	/*
+	 * If set, the VM has (or had) an external write tracking user, and
+	 * thus all write tracking metadata has been allocated, even if KVM
+	 * itself isn't using write tracking.
+	 */
+	bool external_write_tracking_enabled;
+#endif
+
 #if IS_ENABLED(CONFIG_HYPERV)
 	hpa_t	hv_root_tdp;
 	spinlock_t hv_root_tdp_lock;

arch/x86/kvm/mmu/mmu.c

@@ -3575,10 +3575,14 @@ static void mmu_free_root_page(struct kvm *kvm, hpa_t *root_hpa,
 	if (WARN_ON_ONCE(!sp))
 		return;
 
-	if (is_tdp_mmu_page(sp))
+	if (is_tdp_mmu_page(sp)) {
+		lockdep_assert_held_read(&kvm->mmu_lock);
 		kvm_tdp_mmu_put_root(kvm, sp);
-	else if (!--sp->root_count && sp->role.invalid)
+	} else {
+		lockdep_assert_held_write(&kvm->mmu_lock);
+		if (!--sp->root_count && sp->role.invalid)
 			kvm_mmu_prepare_zap_page(kvm, sp, invalid_list);
+	}
 
 	*root_hpa = INVALID_PAGE;
 }
@@ -3587,6 +3591,7 @@ static void mmu_free_root_page(struct kvm *kvm, hpa_t *root_hpa,
 void kvm_mmu_free_roots(struct kvm *kvm, struct kvm_mmu *mmu,
 			ulong roots_to_free)
 {
+	bool is_tdp_mmu = tdp_mmu_enabled && mmu->root_role.direct;
 	int i;
 	LIST_HEAD(invalid_list);
 	bool free_active_root;
@@ -3609,6 +3614,9 @@ void kvm_mmu_free_roots(struct kvm *kvm, struct kvm_mmu *mmu,
 			return;
 	}
 
+	if (is_tdp_mmu)
+		read_lock(&kvm->mmu_lock);
+	else
 		write_lock(&kvm->mmu_lock);
 
 	for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
@@ -3635,8 +3643,13 @@ void kvm_mmu_free_roots(struct kvm *kvm, struct kvm_mmu *mmu,
 		mmu->root.pgd = 0;
 	}
 
+	if (is_tdp_mmu) {
+		read_unlock(&kvm->mmu_lock);
+		WARN_ON_ONCE(!list_empty(&invalid_list));
+	} else {
 		kvm_mmu_commit_zap_page(kvm, &invalid_list);
 		write_unlock(&kvm->mmu_lock);
+	}
 }
 EXPORT_SYMBOL_GPL(kvm_mmu_free_roots);
 
@@ -3693,15 +3706,15 @@ static int mmu_alloc_direct_roots(struct kvm_vcpu *vcpu)
 	unsigned i;
 	int r;
 
+	if (tdp_mmu_enabled)
+		return kvm_tdp_mmu_alloc_root(vcpu);
+
 	write_lock(&vcpu->kvm->mmu_lock);
 	r = make_mmu_pages_available(vcpu);
 	if (r < 0)
 		goto out_unlock;
 
-	if (tdp_mmu_enabled) {
-		root = kvm_tdp_mmu_get_vcpu_root_hpa(vcpu);
-		mmu->root.hpa = root;
-	} else if (shadow_root_level >= PT64_ROOT_4LEVEL) {
+	if (shadow_root_level >= PT64_ROOT_4LEVEL) {
 		root = mmu_alloc_root(vcpu, 0, 0, shadow_root_level);
 		mmu->root.hpa = root;
 	} else if (shadow_root_level == PT32E_ROOT_LEVEL) {
@@ -6997,9 +7010,7 @@ int kvm_mmu_vendor_module_init(void)
 
 	kvm_mmu_reset_all_pte_masks();
 
-	pte_list_desc_cache = kmem_cache_create("pte_list_desc",
-					    sizeof(struct pte_list_desc),
-					    0, SLAB_ACCOUNT, NULL);
+	pte_list_desc_cache = KMEM_CACHE(pte_list_desc, SLAB_ACCOUNT);
 	if (!pte_list_desc_cache)
 		goto out;
 

arch/x86/kvm/mmu/page_track.c

@@ -20,10 +20,23 @@
 #include "mmu_internal.h"
 #include "page_track.h"
 
+static bool kvm_external_write_tracking_enabled(struct kvm *kvm)
+{
+#ifdef CONFIG_KVM_EXTERNAL_WRITE_TRACKING
+	/*
+	 * Read external_write_tracking_enabled before related pointers.  Pairs
+	 * with the smp_store_release in kvm_page_track_write_tracking_enable().
+	 */
+	return smp_load_acquire(&kvm->arch.external_write_tracking_enabled);
+#else
+	return false;
+#endif
+}
+
 bool kvm_page_track_write_tracking_enabled(struct kvm *kvm)
 {
-	return IS_ENABLED(CONFIG_KVM_EXTERNAL_WRITE_TRACKING) ||
-	       !tdp_enabled || kvm_shadow_root_allocated(kvm);
+	return kvm_external_write_tracking_enabled(kvm) ||
+	       kvm_shadow_root_allocated(kvm) || !tdp_enabled;
 }
 
 void kvm_page_track_free_memslot(struct kvm_memory_slot *slot)
@@ -153,6 +166,50 @@ int kvm_page_track_init(struct kvm *kvm)
 	return init_srcu_struct(&head->track_srcu);
 }
 
+static int kvm_enable_external_write_tracking(struct kvm *kvm)
+{
+	struct kvm_memslots *slots;
+	struct kvm_memory_slot *slot;
+	int r = 0, i, bkt;
+
+	mutex_lock(&kvm->slots_arch_lock);
+
+	/*
+	 * Check for *any* write tracking user (not just external users) under
+	 * lock.  This avoids unnecessary work, e.g. if KVM itself is using
+	 * write tracking, or if two external users raced when registering.
+	 */
+	if (kvm_page_track_write_tracking_enabled(kvm))
+		goto out_success;
+
+	for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) {
+		slots = __kvm_memslots(kvm, i);
+		kvm_for_each_memslot(slot, bkt, slots) {
+			/*
+			 * Intentionally do NOT free allocations on failure to
+			 * avoid having to track which allocations were made
+			 * now versus when the memslot was created.  The
+			 * metadata is guaranteed to be freed when the slot is
+			 * freed, and will be kept/used if userspace retries
+			 * the failed ioctl() instead of killing the VM.
+			 */
+			r = kvm_page_track_write_tracking_alloc(slot);
+			if (r)
+				goto out_unlock;
+		}
+	}
+
+out_success:
+	/*
+	 * Ensure that external_write_tracking_enabled becomes true strictly
+	 * after all the related pointers are set.
+	 */
+	smp_store_release(&kvm->arch.external_write_tracking_enabled, true);
+out_unlock:
+	mutex_unlock(&kvm->slots_arch_lock);
+	return r;
+}
+
 /*
  * register the notifier so that event interception for the tracked guest
  * pages can be received.
@@ -161,10 +218,17 @@ int kvm_page_track_register_notifier(struct kvm *kvm,
 				     struct kvm_page_track_notifier_node *n)
 {
 	struct kvm_page_track_notifier_head *head;
+	int r;
 
 	if (!kvm || kvm->mm != current->mm)
 		return -ESRCH;
 
+	if (!kvm_external_write_tracking_enabled(kvm)) {
+		r = kvm_enable_external_write_tracking(kvm);
+		if (r)
+			return r;
+	}
+
 	kvm_get_kvm(kvm);
 
 	head = &kvm->arch.track_notifier_head;

arch/x86/kvm/mmu/tdp_mmu.c

@@ -149,11 +149,11 @@ static struct kvm_mmu_page *tdp_mmu_next_root(struct kvm *kvm,
  * If shared is set, this function is operating under the MMU lock in read
  * mode.
  */
-#define __for_each_tdp_mmu_root_yield_safe(_kvm, _root, _as_id, _only_valid)\
+#define __for_each_tdp_mmu_root_yield_safe(_kvm, _root, _as_id, _only_valid)	\
 	for (_root = tdp_mmu_next_root(_kvm, NULL, _only_valid);		\
 	     ({ lockdep_assert_held(&(_kvm)->mmu_lock); }), _root;		\
 	     _root = tdp_mmu_next_root(_kvm, _root, _only_valid))		\
-		if (kvm_mmu_page_as_id(_root) != _as_id) {		\
+		if (_as_id >= 0 && kvm_mmu_page_as_id(_root) != _as_id) {	\
 		} else
 
 #define for_each_valid_tdp_mmu_root_yield_safe(_kvm, _root, _as_id)	\
@@ -171,12 +171,19 @@ static struct kvm_mmu_page *tdp_mmu_next_root(struct kvm *kvm,
  * Holding mmu_lock for write obviates the need for RCU protection as the list
  * is guaranteed to be stable.
  */
-#define for_each_tdp_mmu_root(_kvm, _root, _as_id)			\
+#define __for_each_tdp_mmu_root(_kvm, _root, _as_id, _only_valid)		\
 	list_for_each_entry(_root, &_kvm->arch.tdp_mmu_roots, link)		\
 		if (kvm_lockdep_assert_mmu_lock_held(_kvm, false) &&		\
-		    kvm_mmu_page_as_id(_root) != _as_id) {		\
+		    ((_as_id >= 0 && kvm_mmu_page_as_id(_root) != _as_id) ||	\
+		     ((_only_valid) && (_root)->role.invalid))) {		\
 		} else
 
+#define for_each_tdp_mmu_root(_kvm, _root, _as_id)			\
+	__for_each_tdp_mmu_root(_kvm, _root, _as_id, false)
+
+#define for_each_valid_tdp_mmu_root(_kvm, _root, _as_id)		\
+	__for_each_tdp_mmu_root(_kvm, _root, _as_id, true)
+
 static struct kvm_mmu_page *tdp_mmu_alloc_sp(struct kvm_vcpu *vcpu)
 {
 	struct kvm_mmu_page *sp;
@@ -216,22 +223,41 @@ static void tdp_mmu_init_child_sp(struct kvm_mmu_page *child_sp,
 	tdp_mmu_init_sp(child_sp, iter->sptep, iter->gfn, role);
 }
 
-hpa_t kvm_tdp_mmu_get_vcpu_root_hpa(struct kvm_vcpu *vcpu)
+int kvm_tdp_mmu_alloc_root(struct kvm_vcpu *vcpu)
 {
-	union kvm_mmu_page_role role = vcpu->arch.mmu->root_role;
+	struct kvm_mmu *mmu = vcpu->arch.mmu;
+	union kvm_mmu_page_role role = mmu->root_role;
+	int as_id = kvm_mmu_role_as_id(role);
 	struct kvm *kvm = vcpu->kvm;
 	struct kvm_mmu_page *root;
 
-	lockdep_assert_held_write(&kvm->mmu_lock);
+	/*
+	 * Check for an existing root before acquiring the pages lock to avoid
+	 * unnecessary serialization if multiple vCPUs are loading a new root.
+	 * E.g. when bringing up secondary vCPUs, KVM will already have created
+	 * a valid root on behalf of the primary vCPU.
+	 */
+	read_lock(&kvm->mmu_lock);
+
+	for_each_valid_tdp_mmu_root_yield_safe(kvm, root, as_id) {
+		if (root->role.word == role.word)
+			goto out_read_unlock;
+	}
+
+	spin_lock(&kvm->arch.tdp_mmu_pages_lock);
 
 	/*
-	 * Check for an existing root before allocating a new one.  Note, the
-	 * role check prevents consuming an invalid root.
+	 * Recheck for an existing root after acquiring the pages lock, another
+	 * vCPU may have raced ahead and created a new usable root.  Manually
+	 * walk the list of roots as the standard macros assume that the pages
+	 * lock is *not* held.  WARN if grabbing a reference to a usable root
+	 * fails, as the last reference to a root can only be put *after* the
+	 * root has been invalidated, which requires holding mmu_lock for write.
 	 */
-	for_each_tdp_mmu_root(kvm, root, kvm_mmu_role_as_id(role)) {
+	list_for_each_entry(root, &kvm->arch.tdp_mmu_roots, link) {
 		if (root->role.word == role.word &&
-		    kvm_tdp_mmu_get_root(root))
-			goto out;
+		    !WARN_ON_ONCE(!kvm_tdp_mmu_get_root(root)))
+			goto out_spin_unlock;
 	}
 
 	root = tdp_mmu_alloc_sp(vcpu);
@@ -245,13 +271,20 @@ hpa_t kvm_tdp_mmu_get_vcpu_root_hpa(struct kvm_vcpu *vcpu)
 	 * is ultimately put by kvm_tdp_mmu_zap_invalidated_roots().
 	 */
 	refcount_set(&root->tdp_mmu_root_count, 2);
-
-	spin_lock(&kvm->arch.tdp_mmu_pages_lock);
 	list_add_rcu(&root->link, &kvm->arch.tdp_mmu_roots);
-	spin_unlock(&kvm->arch.tdp_mmu_pages_lock);
 
-out:
-	return __pa(root->spt);
+out_spin_unlock:
+	spin_unlock(&kvm->arch.tdp_mmu_pages_lock);
+out_read_unlock:
+	read_unlock(&kvm->mmu_lock);
+	/*
+	 * Note, KVM_REQ_MMU_FREE_OBSOLETE_ROOTS will prevent entering the guest
+	 * and actually consuming the root if it's invalidated after dropping
+	 * mmu_lock, and the root can't be freed as this vCPU holds a reference.
+	 */
+	mmu->root.hpa = __pa(root->spt);
+	mmu->root.pgd = 0;
+	return 0;
 }
 
 static void handle_changed_spte(struct kvm *kvm, int as_id, gfn_t gfn,
@@ -734,15 +767,26 @@ static void tdp_mmu_zap_root(struct kvm *kvm, struct kvm_mmu_page *root,
 	rcu_read_lock();
 
 	/*
-	 * To avoid RCU stalls due to recursively removing huge swaths of SPs,
-	 * split the zap into two passes.  On the first pass, zap at the 1gb
-	 * level, and then zap top-level SPs on the second pass.  "1gb" is not
-	 * arbitrary, as KVM must be able to zap a 1gb shadow page without
-	 * inducing a stall to allow in-place replacement with a 1gb hugepage.
+	 * Zap roots in multiple passes of decreasing granularity, i.e. zap at
+	 * 4KiB=>2MiB=>1GiB=>root, in order to better honor need_resched() (all
+	 * preempt models) or mmu_lock contention (full or real-time models).
+	 * Zapping at finer granularity marginally increases the total time of
+	 * the zap, but in most cases the zap itself isn't latency sensitive.
 	 *
-	 * Because zapping a SP recurses on its children, stepping down to
-	 * PG_LEVEL_4K in the iterator itself is unnecessary.
-	 */
+	 * If KVM is configured to prove the MMU, skip the 4KiB and 2MiB zaps
+	 * in order to mimic the page fault path, which can replace a 1GiB page
+	 * table with an equivalent 1GiB hugepage, i.e. can get saddled with
+	 * zapping a 1GiB region that's fully populated with 4KiB SPTEs.  This
+	 * allows verifying that KVM can safely zap 1GiB regions, e.g. without
+	 * inducing RCU stalls, without relying on a relatively rare event
+	 * (zapping roots is orders of magnitude more common).  Note, because
+	 * zapping a SP recurses on its children, stepping down to PG_LEVEL_4K
+	 * in the iterator itself is unnecessary.
+	 */
+	if (!IS_ENABLED(CONFIG_KVM_PROVE_MMU)) {
+		__tdp_mmu_zap_root(kvm, root, shared, PG_LEVEL_4K);
+		__tdp_mmu_zap_root(kvm, root, shared, PG_LEVEL_2M);
+	}
 	__tdp_mmu_zap_root(kvm, root, shared, PG_LEVEL_1G);
 	__tdp_mmu_zap_root(kvm, root, shared, root->role.level);
 
@@ -800,6 +844,12 @@ static bool tdp_mmu_zap_leafs(struct kvm *kvm, struct kvm_mmu_page *root,
 			continue;
 
 		tdp_mmu_iter_set_spte(kvm, &iter, 0);
+
+		/*
+		 * Zappings SPTEs in invalid roots doesn't require a TLB flush,
+		 * see kvm_tdp_mmu_zap_invalidated_roots() for details.
+		 */
+		if (!root->role.invalid)
 			flush = true;
 	}
 
@@ -813,16 +863,16 @@ static bool tdp_mmu_zap_leafs(struct kvm *kvm, struct kvm_mmu_page *root,
 }
 
 /*
- * Zap leaf SPTEs for the range of gfns, [start, end), for all roots. Returns
- * true if a TLB flush is needed before releasing the MMU lock, i.e. if one or
- * more SPTEs were zapped since the MMU lock was last acquired.
+ * Zap leaf SPTEs for the range of gfns, [start, end), for all *VALID** roots.
+ * Returns true if a TLB flush is needed before releasing the MMU lock, i.e. if
+ * one or more SPTEs were zapped since the MMU lock was last acquired.
  */
 bool kvm_tdp_mmu_zap_leafs(struct kvm *kvm, gfn_t start, gfn_t end, bool flush)
 {
 	struct kvm_mmu_page *root;
 
 	lockdep_assert_held_write(&kvm->mmu_lock);
-	for_each_tdp_mmu_root_yield_safe(kvm, root)
+	for_each_valid_tdp_mmu_root_yield_safe(kvm, root, -1)
 		flush = tdp_mmu_zap_leafs(kvm, root, start, end, true, flush);
 
 	return flush;
@@ -896,7 +946,7 @@ void kvm_tdp_mmu_zap_invalidated_roots(struct kvm *kvm)
  * the VM is being destroyed).
  *
  * Note, kvm_tdp_mmu_zap_invalidated_roots() is gifted the TDP MMU's reference.
- * See kvm_tdp_mmu_get_vcpu_root_hpa().
+ * See kvm_tdp_mmu_alloc_root().
  */
 void kvm_tdp_mmu_invalidate_all_roots(struct kvm *kvm)
 {
@@ -1622,7 +1672,7 @@ void kvm_tdp_mmu_clear_dirty_pt_masked(struct kvm *kvm,
 {
 	struct kvm_mmu_page *root;
 
-	for_each_tdp_mmu_root(kvm, root, slot->as_id)
+	for_each_valid_tdp_mmu_root(kvm, root, slot->as_id)
 		clear_dirty_pt_masked(kvm, root, gfn, mask, wrprot);
 }
 
@@ -1740,7 +1790,7 @@ bool kvm_tdp_mmu_write_protect_gfn(struct kvm *kvm,
 	bool spte_set = false;
 
 	lockdep_assert_held_write(&kvm->mmu_lock);
-	for_each_tdp_mmu_root(kvm, root, slot->as_id)
+	for_each_valid_tdp_mmu_root(kvm, root, slot->as_id)
 		spte_set |= write_protect_gfn(kvm, root, gfn, min_level);
 
 	return spte_set;

arch/x86/kvm/mmu/tdp_mmu.h

@@ -10,7 +10,7 @@
 void kvm_mmu_init_tdp_mmu(struct kvm *kvm);
 void kvm_mmu_uninit_tdp_mmu(struct kvm *kvm);
 
-hpa_t kvm_tdp_mmu_get_vcpu_root_hpa(struct kvm_vcpu *vcpu);
+int kvm_tdp_mmu_alloc_root(struct kvm_vcpu *vcpu);
 
 __must_check static inline bool kvm_tdp_mmu_get_root(struct kvm_mmu_page *root)
 {

arch/x86/kvm/x86.c

@@ -8792,31 +8792,24 @@ static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
 
 	kvm_release_pfn_clean(pfn);
 
-	/* The instructions are well-emulated on direct mmu. */
-	if (vcpu->arch.mmu->root_role.direct) {
-		unsigned int indirect_shadow_pages;
-
-		write_lock(&vcpu->kvm->mmu_lock);
-		indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
-		write_unlock(&vcpu->kvm->mmu_lock);
-
-		if (indirect_shadow_pages)
-			kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
-
-		return true;
-	}
-
 	/*
-	 * if emulation was due to access to shadowed page table
-	 * and it failed try to unshadow page and re-enter the
-	 * guest to let CPU execute the instruction.
+	 * If emulation may have been triggered by a write to a shadowed page
+	 * table, unprotect the gfn (zap any relevant SPTEs) and re-enter the
+	 * guest to let the CPU re-execute the instruction in the hope that the
+	 * CPU can cleanly execute the instruction that KVM failed to emulate.
 	 */
+	if (vcpu->kvm->arch.indirect_shadow_pages)
 		kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
 
 	/*
-	 * If the access faults on its page table, it can not
-	 * be fixed by unprotecting shadow page and it should
-	 * be reported to userspace.
+	 * If the failed instruction faulted on an access to page tables that
+	 * are used to translate any part of the instruction, KVM can't resolve
+	 * the issue by unprotecting the gfn, as zapping the shadow page will
+	 * result in the instruction taking a !PRESENT page fault and thus put
+	 * the vCPU into an infinite loop of page faults.  E.g. KVM will create
+	 * a SPTE and write-protect the gfn to resolve the !PRESENT fault, and
+	 * then zap the SPTE to unprotect the gfn, and then do it all over
+	 * again.  Report the error to userspace.
 	 */
 	return !(emulation_type & EMULTYPE_WRITE_PF_TO_SP);
 }