x86/kvm: move kvm_load/put_guest_xcr0 into atomic context

[ Upstream commit 1811d979 ]

guest xcr0 could leak into host when MCE happens in guest mode. Because
do_machine_check() could schedule out at a few places.

For example:

kvm_load_guest_xcr0
...
kvm_x86_ops->run(vcpu) {
  vmx_vcpu_run
    vmx_complete_atomic_exit
      kvm_machine_check
        do_machine_check
          do_memory_failure
            memory_failure
              lock_page

In this case, host_xcr0 is 0x2ff, guest vcpu xcr0 is 0xff. After schedule
out, host cpu has guest xcr0 loaded (0xff).

In __switch_to {
     switch_fpu_finish
       copy_kernel_to_fpregs
         XRSTORS

If any bit i in XSTATE_BV[i] == 1 and xcr0[i] == 0, XRSTORS will
generate #GP (In this case, bit 9). Then ex_handler_fprestore kicks in
and tries to reinitialize fpu by restoring init fpu state. Same story as
last #GP, except we get DOUBLE FAULT this time.

Cc: stable@vger.kernel.org
Signed-off-by: WANG Chao <chao.wang@ucloud.cn>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>

arch/x86/kvm/svm.c

@@ -5622,6 +5622,7 @@ static void svm_vcpu_run(struct kvm_vcpu *vcpu)
 	svm->vmcb->save.cr2 = vcpu->arch.cr2;
 
 	clgi();
+	kvm_load_guest_xcr0(vcpu);
 
 	/*
 	 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
@@ -5769,6 +5770,7 @@ static void svm_vcpu_run(struct kvm_vcpu *vcpu)
 	if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
 		kvm_before_interrupt(&svm->vcpu);
 
+	kvm_put_guest_xcr0(vcpu);
 	stgi();
 
 	/* Any pending NMI will happen here */

arch/x86/kvm/vmx.c

@@ -10756,6 +10756,8 @@ static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
 		vmx_set_interrupt_shadow(vcpu, 0);
 
+	kvm_load_guest_xcr0(vcpu);
+
 	if (static_cpu_has(X86_FEATURE_PKU) &&
 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
 	    vcpu->arch.pkru != vmx->host_pkru)
@@ -10971,6 +10973,8 @@ static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
 			__write_pkru(vmx->host_pkru);
 	}
 
+	kvm_put_guest_xcr0(vcpu);
+
 	vmx->nested.nested_run_pending = 0;
 	vmx->idt_vectoring_info = 0;
 

arch/x86/kvm/x86.c

@@ -713,7 +713,7 @@ void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 }
 EXPORT_SYMBOL_GPL(kvm_lmsw);
 
-static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
+void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
 {
 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
 			!vcpu->guest_xcr0_loaded) {
@@ -723,8 +723,9 @@ static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
 		vcpu->guest_xcr0_loaded = 1;
 	}
 }
+EXPORT_SYMBOL_GPL(kvm_load_guest_xcr0);
 
-static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
+void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
 {
 	if (vcpu->guest_xcr0_loaded) {
 		if (vcpu->arch.xcr0 != host_xcr0)
@@ -732,6 +733,7 @@ static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
 		vcpu->guest_xcr0_loaded = 0;
 	}
 }
+EXPORT_SYMBOL_GPL(kvm_put_guest_xcr0);
 
 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
 {
@@ -7649,8 +7651,6 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
 		goto cancel_injection;
 	}
 
-	kvm_load_guest_xcr0(vcpu);
-
 	if (req_immediate_exit) {
 		kvm_make_request(KVM_REQ_EVENT, vcpu);
 		kvm_x86_ops->request_immediate_exit(vcpu);
@@ -7703,8 +7703,6 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
 	vcpu->mode = OUTSIDE_GUEST_MODE;
 	smp_wmb();
 
-	kvm_put_guest_xcr0(vcpu);
-
 	kvm_before_interrupt(vcpu);
 	kvm_x86_ops->handle_external_intr(vcpu);
 	kvm_after_interrupt(vcpu);

arch/x86/kvm/x86.h

@@ -345,4 +345,6 @@ static inline void kvm_after_interrupt(struct kvm_vcpu *vcpu)
 	__this_cpu_write(current_vcpu, NULL);
 }
 
+void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu);
+void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu);
 #endif