kvm: x86: Skip shadow page resync on CR3 switch when indicated by guest

When the guest indicates that the TLB doesn't need to be flushed in a
CR3 switch, we can also skip resyncing the shadow page tables since an
out-of-sync shadow page table is equivalent to an out-of-sync TLB.

Signed-off-by: Junaid Shahid <junaids@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

3 files changed, 34 insertions, 7 deletions

diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index 0f6965ce016a..9446a36a4ab7 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -4098,9 +4098,19 @@ static bool fast_cr3_switch(struct kvm_vcpu *vcpu, gpa_t new_cr3,
 			 */
 
 			kvm_make_request(KVM_REQ_LOAD_CR3, vcpu);
-			kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
-			if (!skip_tlb_flush)
+			if (!skip_tlb_flush) {
+				kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
 				kvm_x86_ops->tlb_flush(vcpu, true);
+			}
+
+			/*
+			 * The last MMIO access's GVA and GPA are cached in the
+			 * VCPU. When switching to a new CR3, that GVA->GPA
+			 * mapping may no longer be valid. So clear any cached
+			 * MMIO info even when we don't need to sync the shadow
+			 * page tables.
+			 */
+			vcpu_clear_mmio_info(vcpu, MMIO_GVA_ANY);
 
 			__clear_sp_write_flooding_count(
 				page_header(mmu->root_hpa));
@@ -5217,6 +5227,21 @@ void kvm_mmu_invlpg(struct kvm_vcpu *vcpu, gva_t gva)
 	struct kvm_mmu *mmu = &vcpu->arch.mmu;
 
 	mmu->invlpg(vcpu, gva, mmu->root_hpa);
+
+	/*
+	 * INVLPG is required to invalidate any global mappings for the VA,
+	 * irrespective of PCID. Since it would take us roughly similar amount
+	 * of work to determine whether the prev_root mapping of the VA is
+	 * marked global, or to just sync it blindly, so we might as well just
+	 * always sync it.
+	 *
+	 * Mappings not reachable via the current cr3 or the prev_root.cr3 will
+	 * be synced when switching to that cr3, so nothing needs to be done
+	 * here for them.
+	 */
+	if (VALID_PAGE(mmu->prev_root.hpa))
+		mmu->invlpg(vcpu, gva, mmu->prev_root.hpa);
+
 	kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 	++vcpu->stat.invlpg;
 }
@@ -5232,8 +5257,10 @@ void kvm_mmu_invpcid_gva(struct kvm_vcpu *vcpu, gva_t gva, unsigned long pcid)
 	}
 
 	if (VALID_PAGE(mmu->prev_root.hpa) &&
-	    pcid == kvm_get_pcid(vcpu, mmu->prev_root.cr3))
+	    pcid == kvm_get_pcid(vcpu, mmu->prev_root.cr3)) {
+		mmu->invlpg(vcpu, gva, mmu->prev_root.hpa);
 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
+	}
 
 	++vcpu->stat.invlpg;
 
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index 6151418cec32..b81210051133 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -8821,7 +8821,7 @@ static int handle_invpcid(struct kvm_vcpu *vcpu)
 
 		if (kvm_get_pcid(vcpu, vcpu->arch.mmu.prev_root.cr3)
 		    == operand.pcid)
-			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
+			kvm_mmu_free_roots(vcpu, KVM_MMU_ROOT_PREVIOUS);
 
 		/*
 		 * If neither the current cr3 nor the prev_root.cr3 use the
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 493afbf12e78..aa5d96b4b386 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -858,10 +858,10 @@ int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 #endif
 
 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
-		kvm_mmu_sync_roots(vcpu);
-
-		if (!skip_tlb_flush)
+		if (!skip_tlb_flush) {
+			kvm_mmu_sync_roots(vcpu);
 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
+		}
 		return 0;
 	}