KVM: x86: only allocate gfn_track when necessary

Avoid allocating the gfn_track arrays if nothing needs them. If there
are no external to KVM users of the API (i.e. no GVT-g), then page
tracking is only needed for shadow page tables. This means that when tdp
is enabled and there are no external users, then the gfn_track arrays
can be lazily allocated when the shadow MMU is actually used. This avoid
allocations equal to .05% of guest memory when nested virtualization is
not used, if the kernel is compiled without GVT-g.

Signed-off-by: David Stevens <stevensd@chromium.org>
Message-Id: <20210922045859.2011227-3-stevensd@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

5 files changed, 93 insertions, 8 deletions

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 1b280292bdff..5271fce6cd65 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -1211,6 +1211,14 @@ struct kvm_arch {
 	 */
 	bool memslots_have_rmaps;
 
+	/*
+	 * Set when the KVM mmu needs guest write access page tracking. If
+	 * set, the necessary gfn_track arrays have been allocated for
+	 * all memslots and should be allocated for any newly created or
+	 * modified memslots.
+	 */
+	bool memslots_mmu_write_tracking;
+
 #if IS_ENABLED(CONFIG_HYPERV)
 	hpa_t	hv_root_tdp;
 	spinlock_t hv_root_tdp_lock;
diff --git a/arch/x86/include/asm/kvm_page_track.h b/arch/x86/include/asm/kvm_page_track.h
index 5c12f97ce934..79d84a94f8eb 100644
--- a/arch/x86/include/asm/kvm_page_track.h
+++ b/arch/x86/include/asm/kvm_page_track.h
@@ -49,8 +49,11 @@ struct kvm_page_track_notifier_node {
 int kvm_page_track_init(struct kvm *kvm);
 void kvm_page_track_cleanup(struct kvm *kvm);
 
+int kvm_page_track_enable_mmu_write_tracking(struct kvm *kvm);
+
 void kvm_page_track_free_memslot(struct kvm_memory_slot *slot);
-int kvm_page_track_create_memslot(struct kvm_memory_slot *slot,
+int kvm_page_track_create_memslot(struct kvm *kvm,
+				  struct kvm_memory_slot *slot,
 				  unsigned long npages);
 
 void kvm_slot_page_track_add_page(struct kvm *kvm,
@@ -59,7 +62,8 @@ void kvm_slot_page_track_add_page(struct kvm *kvm,
 void kvm_slot_page_track_remove_page(struct kvm *kvm,
 				     struct kvm_memory_slot *slot, gfn_t gfn,
 				     enum kvm_page_track_mode mode);
-bool kvm_slot_page_track_is_active(struct kvm_memory_slot *slot, gfn_t gfn,
+bool kvm_slot_page_track_is_active(struct kvm_vcpu *vcpu,
+				   struct kvm_memory_slot *slot, gfn_t gfn,
 				   enum kvm_page_track_mode mode);
 
 void
diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
index 91292009780a..24a9f4c3f5e7 100644
--- a/arch/x86/kvm/mmu/mmu.c
+++ b/arch/x86/kvm/mmu/mmu.c
@@ -2583,7 +2583,7 @@ int mmu_try_to_unsync_pages(struct kvm_vcpu *vcpu, struct kvm_memory_slot *slot,
 	 * track machinery is used to write-protect upper-level shadow pages,
 	 * i.e. this guards the role.level == 4K assertion below!
 	 */
-	if (kvm_slot_page_track_is_active(slot, gfn, KVM_PAGE_TRACK_WRITE))
+	if (kvm_slot_page_track_is_active(vcpu, slot, gfn, KVM_PAGE_TRACK_WRITE))
 		return -EPERM;
 
 	/*
@@ -3431,6 +3431,10 @@ static int mmu_alloc_shadow_roots(struct kvm_vcpu *vcpu)
 	if (r)
 		return r;
 
+	r = kvm_page_track_enable_mmu_write_tracking(vcpu->kvm);
+	if (r)
+		return r;
+
 	write_lock(&vcpu->kvm->mmu_lock);
 	r = make_mmu_pages_available(vcpu);
 	if (r < 0)
@@ -3790,7 +3794,7 @@ static bool page_fault_handle_page_track(struct kvm_vcpu *vcpu,
 	 * guest is writing the page which is write tracked which can
 	 * not be fixed by page fault handler.
 	 */
-	if (kvm_slot_page_track_is_active(fault->slot, fault->gfn, KVM_PAGE_TRACK_WRITE))
+	if (kvm_slot_page_track_is_active(vcpu, fault->slot, fault->gfn, KVM_PAGE_TRACK_WRITE))
 		return true;
 
 	return false;
@@ -5607,6 +5611,9 @@ void kvm_mmu_init_vm(struct kvm *kvm)
 		 */
 		kvm->arch.memslots_have_rmaps = true;
 
+	if (!tdp_enabled)
+		kvm->arch.memslots_mmu_write_tracking = true;
+
 	node->track_write = kvm_mmu_pte_write;
 	node->track_flush_slot = kvm_mmu_invalidate_zap_pages_in_memslot;
 	kvm_page_track_register_notifier(kvm, node);
diff --git a/arch/x86/kvm/mmu/page_track.c b/arch/x86/kvm/mmu/page_track.c
index 16e7176c97a5..bb5d60bd4dbf 100644
--- a/arch/x86/kvm/mmu/page_track.c
+++ b/arch/x86/kvm/mmu/page_track.c
@@ -19,6 +19,16 @@
 #include "mmu.h"
 #include "mmu_internal.h"
 
+static bool write_tracking_enabled(struct kvm *kvm)
+{
+	/*
+	 * Read memslots_mmu_write_tracking before gfn_track pointers. Pairs
+	 * with smp_store_release in kvm_page_track_enable_mmu_write_tracking.
+	 */
+	return IS_ENABLED(CONFIG_KVM_EXTERNAL_WRITE_TRACKING) ||
+	       smp_load_acquire(&kvm->arch.memslots_mmu_write_tracking);
+}
+
 void kvm_page_track_free_memslot(struct kvm_memory_slot *slot)
 {
 	int i;
@@ -29,12 +39,16 @@ void kvm_page_track_free_memslot(struct kvm_memory_slot *slot)
 	}
 }
 
-int kvm_page_track_create_memslot(struct kvm_memory_slot *slot,
+int kvm_page_track_create_memslot(struct kvm *kvm,
+				  struct kvm_memory_slot *slot,
 				  unsigned long npages)
 {
-	int  i;
+	int i;
 
 	for (i = 0; i < KVM_PAGE_TRACK_MAX; i++) {
+		if (i == KVM_PAGE_TRACK_WRITE && !write_tracking_enabled(kvm))
+			continue;
+
 		slot->arch.gfn_track[i] =
 			kvcalloc(npages, sizeof(*slot->arch.gfn_track[i]),
 				 GFP_KERNEL_ACCOUNT);
@@ -57,6 +71,46 @@ static inline bool page_track_mode_is_valid(enum kvm_page_track_mode mode)
 	return true;
 }
 
+int kvm_page_track_enable_mmu_write_tracking(struct kvm *kvm)
+{
+	struct kvm_memslots *slots;
+	struct kvm_memory_slot *slot;
+	unsigned short **gfn_track;
+	int i;
+
+	if (write_tracking_enabled(kvm))
+		return 0;
+
+	mutex_lock(&kvm->slots_arch_lock);
+
+	if (write_tracking_enabled(kvm)) {
+		mutex_unlock(&kvm->slots_arch_lock);
+		return 0;
+	}
+
+	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
+		slots = __kvm_memslots(kvm, i);
+		kvm_for_each_memslot(slot, slots) {
+			gfn_track = slot->arch.gfn_track + KVM_PAGE_TRACK_WRITE;
+			*gfn_track = kvcalloc(slot->npages, sizeof(*gfn_track),
+					      GFP_KERNEL_ACCOUNT);
+			if (*gfn_track == NULL) {
+				mutex_unlock(&kvm->slots_arch_lock);
+				return -ENOMEM;
+			}
+		}
+	}
+
+	/*
+	 * Ensure that memslots_mmu_write_tracking becomes true strictly
+	 * after all the pointers are set.
+	 */
+	smp_store_release(&kvm->arch.memslots_mmu_write_tracking, true);
+	mutex_unlock(&kvm->slots_arch_lock);
+
+	return 0;
+}
+
 static void update_gfn_track(struct kvm_memory_slot *slot, gfn_t gfn,
 			     enum kvm_page_track_mode mode, short count)
 {
@@ -92,6 +146,10 @@ void kvm_slot_page_track_add_page(struct kvm *kvm,
 	if (WARN_ON(!page_track_mode_is_valid(mode)))
 		return;
 
+	if (WARN_ON(mode == KVM_PAGE_TRACK_WRITE &&
+		    !write_tracking_enabled(kvm)))
+		return;
+
 	update_gfn_track(slot, gfn, mode, 1);
 
 	/*
@@ -126,6 +184,10 @@ void kvm_slot_page_track_remove_page(struct kvm *kvm,
 	if (WARN_ON(!page_track_mode_is_valid(mode)))
 		return;
 
+	if (WARN_ON(mode == KVM_PAGE_TRACK_WRITE &&
+		    !write_tracking_enabled(kvm)))
+		return;
+
 	update_gfn_track(slot, gfn, mode, -1);
 
 	/*
@@ -139,7 +201,8 @@ EXPORT_SYMBOL_GPL(kvm_slot_page_track_remove_page);
 /*
  * check if the corresponding access on the specified guest page is tracked.
  */
-bool kvm_slot_page_track_is_active(struct kvm_memory_slot *slot, gfn_t gfn,
+bool kvm_slot_page_track_is_active(struct kvm_vcpu *vcpu,
+				   struct kvm_memory_slot *slot, gfn_t gfn,
 				   enum kvm_page_track_mode mode)
 {
 	int index;
@@ -150,6 +213,9 @@ bool kvm_slot_page_track_is_active(struct kvm_memory_slot *slot, gfn_t gfn,
 	if (!slot)
 		return false;
 
+	if (mode == KVM_PAGE_TRACK_WRITE && !write_tracking_enabled(vcpu->kvm))
+		return false;
+
 	index = gfn_to_index(gfn, slot->base_gfn, PG_LEVEL_4K);
 	return !!READ_ONCE(slot->arch.gfn_track[mode][index]);
 }
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 03091a2e0822..db7fa1398f0d 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -11476,7 +11476,7 @@ static int kvm_alloc_memslot_metadata(struct kvm *kvm,
 		}
 	}
 
-	if (kvm_page_track_create_memslot(slot, npages))
+	if (kvm_page_track_create_memslot(kvm, slot, npages))
 		goto out_free;
 
 	return 0;