[RFC][PATCH] KVM: nVMX: Leave VMX mode on apparent CPU reset

[Jan Kiszka <jan.kiszka@siemens.com>]

As long as we do not expose all the VMX related states to user space,
there is no way to properly reset a VCPU when VMX is enabled. Emulate
this for now by catching host-side clearings of the feature control MSR.
This allows to reboot a VM while it is running some hypervisor code.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
---

Better ideas? Or continue to leave it as it is?

 arch/x86/kvm/vmx.c | 35 +++++++++++++++++++++++++++++++++++
 1 file changed, 35 insertions(+)

diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index f90320b..da04247 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -2455,6 +2455,8 @@ static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
 	return 1;
 }
 
+static void vmx_reset_nested(struct kvm_vcpu *vcpu);
+
 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 {
 	u32 msr_index = msr_info->index;
@@ -2470,6 +2472,12 @@ static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 				& FEATURE_CONTROL_LOCKED)
 			return 0;
 		to_vmx(vcpu)->nested.msr_ia32_feature_control = data;
+		/*
+		 * Detect reset and allow to leave VMX mode this way until we
+		 * expose all related states to user space.
+		 */
+		if (host_initialized && data == 0)
+			vmx_reset_nested(vcpu);
 		return 1;
 	}
 
@@ -8487,6 +8495,33 @@ static void nested_vmx_vmexit(struct kvm_vcpu *vcpu)
 		vmx->nested.sync_shadow_vmcs = true;
 }
 
+static void vmx_reset_nested(struct kvm_vcpu *vcpu)
+{
+	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
+	struct vcpu_vmx *vmx = to_vmx(vcpu);
+
+	if (!vmx->nested.vmxon)
+		return;
+
+	if (is_guest_mode(vcpu)) {
+		vmcs12->host_cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
+		vmcs12->host_cr3 = 0;
+		vmcs12->host_cr4 = 0;
+		vmcs12->host_rsp = 0;
+		vmcs12->vm_exit_controls = 0;
+		nested_vmx_vmexit(vcpu);
+	}
+
+	free_nested(vmx);
+
+	/*
+	 * If we were in guest mode, the reset state user space wrote so far
+	 * is now inconsistent. If we were in host mode, some state update may
+	 * have been rejected. So simply repeat the reset her.
+	 */
+	vmx_vcpu_reset(vcpu);
+}
+
 /*
  * L1's failure to enter L2 is a subset of a normal exit, as explained in
  * 23.7 "VM-entry failures during or after loading guest state" (this also
-- 
1.8.1.1.298.ge7eed54