KVM: x86: only copy XSAVE state for the supported features

This makes the interface more deterministic for userspace, which can expect
(after configuring only the features it supports) to get exactly the same
state from the kernel, independent of the host CPU and kernel version.
Suggested-by: NGleb Natapov <gleb@redhat.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>
Signed-off-by: NGleb Natapov <gleb@redhat.com>

arch/x86/include/asm/kvm_host.h

@@ -390,6 +390,7 @@ struct kvm_vcpu_arch {
 	struct fpu guest_fpu;
 	u64 xcr0;
 	u64 guest_supported_xcr0;
+	u32 guest_xstate_size;
 
 	struct kvm_pio_request pio;
 	void *pio_data;

arch/x86/kvm/cpuid.c

@@ -23,6 +23,26 @@
 #include "mmu.h"
 #include "trace.h"
 
+static u32 xstate_required_size(u64 xstate_bv)
+{
+	int feature_bit = 0;
+	u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
+
+	xstate_bv &= ~XSTATE_FPSSE;
+	while (xstate_bv) {
+		if (xstate_bv & 0x1) {
+		        u32 eax, ebx, ecx, edx;
+		        cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
+			ret = max(ret, eax + ebx);
+		}
+
+		xstate_bv >>= 1;
+		feature_bit++;
+	}
+
+	return ret;
+}
+
 void kvm_update_cpuid(struct kvm_vcpu *vcpu)
 {
 	struct kvm_cpuid_entry2 *best;
@@ -47,12 +67,16 @@ void kvm_update_cpuid(struct kvm_vcpu *vcpu)
 	}
 
 	best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
-	if (!best)
+	if (!best) {
 		vcpu->arch.guest_supported_xcr0 = 0;
-	else
+		vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
+	} else {
 		vcpu->arch.guest_supported_xcr0 =
 			(best->eax | ((u64)best->edx << 32)) &
 			host_xcr0 & KVM_SUPPORTED_XCR0;
+		vcpu->arch.guest_xstate_size =
+			xstate_required_size(vcpu->arch.guest_supported_xcr0);
+	}
 
 	kvm_pmu_cpuid_update(vcpu);
 }

arch/x86/kvm/x86.c

@@ -2984,11 +2984,13 @@ static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
 					 struct kvm_xsave *guest_xsave)
 {
-	if (cpu_has_xsave)
+	if (cpu_has_xsave) {
 		memcpy(guest_xsave->region,
 			&vcpu->arch.guest_fpu.state->xsave,
-			xstate_size);
-	else {
+			vcpu->arch.guest_xstate_size);
+		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] &=
+			vcpu->arch.guest_supported_xcr0 | XSTATE_FPSSE;
+	} else {
 		memcpy(guest_xsave->region,
 			&vcpu->arch.guest_fpu.state->fxsave,
 			sizeof(struct i387_fxsave_struct));
@@ -3014,7 +3016,7 @@ static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
 		if (xstate_bv & ~host_xcr0)
 			return -EINVAL;
 		memcpy(&vcpu->arch.guest_fpu.state->xsave,
-			guest_xsave->region, xstate_size);
+			guest_xsave->region, vcpu->arch.guest_xstate_size);
 	} else {
 		if (xstate_bv & ~XSTATE_FPSSE)
 			return -EINVAL;
@@ -6951,6 +6953,7 @@ int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 	vcpu->arch.pv_time_enabled = false;
 
 	vcpu->arch.guest_supported_xcr0 = 0;
+	vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
 
 	kvm_async_pf_hash_reset(vcpu);
 	kvm_pmu_init(vcpu);