On Fri, Apr 30 2021 at 11:06, Paolo Bonzini wrote:
> On 30/04/21 11:03, Thomas Gleixner wrote:
>> Lai,
>>
>> On Tue, Apr 27 2021 at 07:09, Lai Jiangshan wrote:
>>> u32 intr_info = vmx_get_intr_info(&vmx->vcpu);
>>> @@ -6427,12 +6417,19 @@ static void handle_exception_nmi_irqoff(struct vcpu_vmx *vmx)
>>> static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu)
>>> {
>>> u32 intr_info = vmx_get_intr_info(vcpu);
>>> + unsigned int vector;
>>> + gate_desc *desc;
>>>
>>> if (WARN_ONCE(!is_external_intr(intr_info),
>>> "KVM: unexpected VM-Exit interrupt info: 0x%x", intr_info))
>>> return;
>>>
>>> - handle_interrupt_nmi_irqoff(vcpu, intr_info);
>>> + vector = intr_info & INTR_INFO_VECTOR_MASK;
>>> + desc = (gate_desc *)host_idt_base + vector;
>>> +
>>> + kvm_before_interrupt(vcpu);
>>> + vmx_do_interrupt_nmi_irqoff(gate_offset(desc));
>>> + kvm_after_interrupt(vcpu);
>>
>> So the previous patch does:
>>
>> + kvm_before_interrupt(&vmx->vcpu);
>> + vmx_do_interrupt_nmi_irqoff((unsigned long)asm_noist_exc_nmi);
>> + kvm_after_interrupt(&vmx->vcpu);
>>
>> What is this idt gate descriptor dance for in this code?
>
> NMIs are sent through a different vmexit code (the same one as
> exceptions). This one is for interrupts.
Duh. Yes. The ability to read is clearly an advantage...
