HV: Use NMI-window exiting to address req missing issue

There is a window where we may miss the current request in the
notification period when the work flow is as the following:

      CPUx +                   + CPUr
           |                   |
           |                   +--+
           |                   |  | Handle pending req
           |                   <--+
           +--+                |
           |  | Set req flag   |
           <--+                |
           +------------------>---+
           |     Send NMI      |  | Handle NMI
           |                   <--+
           |                   |
           |                   |
           |                   +--> vCPU enter
           |                   |
           +                   +

So, this patch enables the NMI-window exiting to trigger the next vmexit
once there is no "virtual-NMI blocking" after vCPU enter into VMX non-root
mode. Then we can process the pending request on time.

Tracked-On: #3886
Acked-by: Eddie Dong <eddie.dong@intel.com>
Signed-off-by: Kaige Fu <kaige.fu@intel.com>

hypervisor/arch/x86/guest/virq.c

@@ -531,3 +531,20 @@ int32_t exception_vmexit_handler(struct acrn_vcpu *vcpu)
 
 	return status;
 }
+
+int32_t nmi_window_vmexit_handler(struct acrn_vcpu *vcpu)
+{
+	uint32_t value32;
+
+	/*
+	 * Disable NMI-window exiting here. We will process
+	 * the pending request in acrn_handle_pending_request later
+	 */
+	value32 = exec_vmread32(VMX_PROC_VM_EXEC_CONTROLS);
+	value32 &= ~VMX_PROCBASED_CTLS_NMI_WINEXIT;
+	exec_vmwrite32(VMX_PROC_VM_EXEC_CONTROLS, value32);
+
+	vcpu_retain_rip(vcpu);
+
+	return 0;
+}

hypervisor/arch/x86/guest/vmcs.c

@@ -582,7 +582,7 @@ void switch_apicv_mode_x2apic(struct acrn_vcpu *vcpu)
 		 * directly without vmexit. So, here we enable NMI-exiting and use NMI
 		 * as notification signal after passthroughing the lapic to vCPU.
 		 */
-		value32 |= VMX_PINBASED_CTLS_NMI_EXIT;
+		value32 |= VMX_PINBASED_CTLS_NMI_EXIT | VMX_PINBASED_CTLS_VIRT_NMI;
 		exec_vmwrite32(VMX_PIN_VM_EXEC_CONTROLS, value32);
 
 		value32 = exec_vmread32(VMX_EXIT_CONTROLS);

hypervisor/arch/x86/guest/vmexit.c

@@ -51,7 +51,7 @@ static const struct vm_exit_dispatch dispatch_table[NR_VMX_EXIT_REASONS] = {
 	[VMX_EXIT_REASON_INTERRUPT_WINDOW] = {
 		.handler = interrupt_window_vmexit_handler},
 	[VMX_EXIT_REASON_NMI_WINDOW] = {
-		.handler = unhandled_vmexit_handler},
+		.handler = nmi_window_vmexit_handler},
 	[VMX_EXIT_REASON_TASK_SWITCH] = {
 		.handler = unhandled_vmexit_handler},
 	[VMX_EXIT_REASON_CPUID] = {

hypervisor/arch/x86/irq.c

@@ -18,6 +18,7 @@
 #include <vboot.h>
 #include <dump.h>
 #include <logmsg.h>
+#include <vmx.h>
 
 static spinlock_t exception_spinlock = { .head = 0U, .tail = 0U, };
 static spinlock_t irq_alloc_spinlock = { .head = 0U, .tail = 0U, };
@@ -384,10 +385,36 @@ void dispatch_exception(struct intr_excp_ctx *ctx)
 
 void handle_nmi(__unused struct intr_excp_ctx *ctx)
 {
+	uint32_t value32;
+
 	/*
-	 * Just ignore the NMI here for now.
+	 * There is a window where we may miss the current request in this
-	 * TODO: implement specific NMI handling function.
+	 * notification period when the work flow is as the following:
+	 *
+	 *       CPUx +                   + CPUr
+	 *            |                   |
+	 *            |                   +--+
+	 *            |                   |  | Handle pending req
+	 *            |                   <--+
+	 *            +--+                |
+	 *            |  | Set req flag   |
+	 *            <--+                |
+	 *            +------------------>---+
+	 *            |     Send NMI      |  | Handle NMI
+	 *            |                   <--+
+	 *            |                   |
+	 *            |                   |
+	 *            |                   +--> vCPU enter
+	 *            |                   |
+	 *            +                   +
+	 *
+	 * So, here we enable the NMI-window exiting to trigger the next vmexit
+	 * once there is no "virtual-NMI blocking" after vCPU enter into VMX non-root
+	 * mode. Then we can process the pending request on time.
 	 */
+	value32 = exec_vmread32(VMX_PROC_VM_EXEC_CONTROLS);
+	value32 |= VMX_PROCBASED_CTLS_NMI_WINEXIT;
+	exec_vmwrite32(VMX_PROC_VM_EXEC_CONTROLS, value32);
 }
 
 static void init_irq_descs(void)

hypervisor/include/arch/x86/irq.h

@@ -209,6 +209,7 @@ void vcpu_make_request(struct acrn_vcpu *vcpu, uint16_t eventid);
  * @pre vcpu != NULL
  */
 int32_t exception_vmexit_handler(struct acrn_vcpu *vcpu);
+int32_t nmi_window_vmexit_handler(struct acrn_vcpu *vcpu);
 int32_t interrupt_window_vmexit_handler(struct acrn_vcpu *vcpu);
 int32_t external_interrupt_vmexit_handler(struct acrn_vcpu *vcpu);
 int32_t acrn_handle_pending_request(struct acrn_vcpu *vcpu);