diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 6cbf97a2ebc4c07167d0005fe5ec85d2234e03b9..89d1fdb39c46d66acb88cdabb4b6795a24cd636f 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -647,6 +647,7 @@ struct kvm_vcpu_arch {
 	u64 smi_count;
 	bool tpr_access_reporting;
 	bool xsaves_enabled;
+	bool xfd_no_write_intercept;
 	u64 ia32_xss;
 	u64 microcode_version;
 	u64 arch_capabilities;
diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
index b8b7f5c7b3df54cc83ed527618e58bbecd1d3f2b..15e30602782b3cd388a8c84e1f721a9791883586 100644
--- a/arch/x86/kvm/vmx/vmx.c
+++ b/arch/x86/kvm/vmx/vmx.c
@@ -162,6 +162,7 @@ static u32 vmx_possible_passthrough_msrs[MAX_POSSIBLE_PASSTHROUGH_MSRS] = {
 	MSR_FS_BASE,
 	MSR_GS_BASE,
 	MSR_KERNEL_GS_BASE,
+	MSR_IA32_XFD,
 	MSR_IA32_XFD_ERR,
 #endif
 	MSR_IA32_SYSENTER_CS,
@@ -764,10 +765,11 @@ void vmx_update_exception_bitmap(struct kvm_vcpu *vcpu)
 	}
 
 	/*
-	 * Trap #NM if guest xfd contains a non-zero value so guest XFD_ERR
-	 * can be saved timely.
+	 * Disabling xfd interception indicates that dynamic xfeatures
+	 * might be used in the guest. Always trap #NM in this case
+	 * to save guest xfd_err timely.
 	 */
-	if (vcpu->arch.guest_fpu.fpstate->xfd)
+	if (vcpu->arch.xfd_no_write_intercept)
 		eb |= (1u << NM_VECTOR);
 
 	vmcs_write32(EXCEPTION_BITMAP, eb);
@@ -1978,9 +1980,21 @@ static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 		break;
 	case MSR_IA32_XFD:
 		ret = kvm_set_msr_common(vcpu, msr_info);
-		/* Update #NM interception according to guest xfd */
-		if (!ret)
+		/*
+		 * Always intercepting WRMSR could incur non-negligible
+		 * overhead given xfd might be changed frequently in
+		 * guest context switch. Disable write interception
+		 * upon the first write with a non-zero value (indicating
+		 * potential usage on dynamic xfeatures). Also update
+		 * exception bitmap to trap #NM for proper virtualization
+		 * of guest xfd_err.
+		 */
+		if (!ret && data) {
+			vmx_disable_intercept_for_msr(vcpu, MSR_IA32_XFD,
+						      MSR_TYPE_RW);
+			vcpu->arch.xfd_no_write_intercept = true;
 			vmx_update_exception_bitmap(vcpu);
+		}
 		break;
 #endif
 	case MSR_IA32_SYSENTER_CS:
diff --git a/arch/x86/kvm/vmx/vmx.h b/arch/x86/kvm/vmx/vmx.h
index 69dd2f85abdc3b51ad5caa9a5621e93663f43a8f..f8fc7441baea93da21762f92b93e45e45849cf81 100644
--- a/arch/x86/kvm/vmx/vmx.h
+++ b/arch/x86/kvm/vmx/vmx.h
@@ -349,7 +349,7 @@ struct vcpu_vmx {
 	struct lbr_desc lbr_desc;
 
 	/* Save desired MSR intercept (read: pass-through) state */
-#define MAX_POSSIBLE_PASSTHROUGH_MSRS	14
+#define MAX_POSSIBLE_PASSTHROUGH_MSRS	15
 	struct {
 		DECLARE_BITMAP(read, MAX_POSSIBLE_PASSTHROUGH_MSRS);
 		DECLARE_BITMAP(write, MAX_POSSIBLE_PASSTHROUGH_MSRS);
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index bde18ca657db6318b1dadb5e1bc6b2399d627120..60da2331ec32c0dfcb9ad5e2a379003c870c39dc 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -10083,6 +10083,14 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
 	vcpu->mode = OUTSIDE_GUEST_MODE;
 	smp_wmb();
 
+	/*
+	 * Sync xfd before calling handle_exit_irqoff() which may
+	 * rely on the fact that guest_fpu::xfd is up-to-date (e.g.
+	 * in #NM irqoff handler).
+	 */
+	if (vcpu->arch.xfd_no_write_intercept)
+		fpu_sync_guest_vmexit_xfd_state();
+
 	static_call(kvm_x86_handle_exit_irqoff)(vcpu);
 
 	if (vcpu->arch.guest_fpu.xfd_err)