Re: [PATCH v4 07/11] KVM: VMX: Handle NMI Source report in VM exit

[Sean Christopherson <seanjc@google.com>]

On Tue, Jul 09, 2024, Jacob Pan wrote:
> From: Zeng Guang <guang.zeng@intel.com>
> 
> If the "NMI exiting" VM-execution control is 1, the value of the 16-bit NMI
> source vector is saved in the exit-qualification field in the VMCS when VM
> exits occur on CPUs that support NMI source.
> 
> KVM that is aware of NMI-source reporting will push the bitmask of NMI source
> vectors as the exceptoin event data field on the stack for then entry of FRED
> exception. Subsequently, the host NMI exception handler is invoked which
> will process NMI source information in the event data. This operation is
> independent of vCPU FRED enabling status.
> 
> Signed-off-by: Zeng Guang <guang.zeng@intel.com>
> Signed-off-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
> ---
>  arch/x86/kvm/vmx/vmx.c | 11 ++++++++---
>  1 file changed, 8 insertions(+), 3 deletions(-)
> 
> diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
> index 4e7b36081b76..6719c598fa5f 100644
> --- a/arch/x86/kvm/vmx/vmx.c
> +++ b/arch/x86/kvm/vmx/vmx.c
> @@ -7331,10 +7331,15 @@ static noinstr void vmx_vcpu_enter_exit(struct kvm_vcpu *vcpu,
>  	if ((u16)vmx->exit_reason.basic == EXIT_REASON_EXCEPTION_NMI &&
>  	    is_nmi(vmx_get_intr_info(vcpu))) {
>  		kvm_before_interrupt(vcpu, KVM_HANDLING_NMI);
> -		if (cpu_feature_enabled(X86_FEATURE_FRED))
> -			fred_entry_from_kvm(EVENT_TYPE_NMI, NMI_VECTOR, 0);
> -		else
> +		if (cpu_feature_enabled(X86_FEATURE_FRED)) {
> +			unsigned long edata = 0;
> +
> +			if (cpu_feature_enabled(X86_FEATURE_NMI_SOURCE))
> +				edata = vmx_get_exit_qual(vcpu);
> +			fred_entry_from_kvm(EVENT_TYPE_NMI, NMI_VECTOR, edata);

Eh, I would just east the extra VMREAD if KVM happens to run on a CPU that supports
FRED but not NMI source reporting.  Per spec 7.0, the EXIT_QUALIFICATION is 0 if
NMI source isn't supported, i.e. it's guaranteed to not cause functional problems.

I assume there won't be many (any?) CPUs with FRED but not NMI_SOURCE (and if there
are, why!?!?!).  On the other hand, gating this on NMI_SOURCE means KVM will need
another update if some other event data is ever added for NMIs.  And readers don't
have to hunt through the FRED spec with a fine-toothed comb to suss out that
EXIT_QUALIFICATION == event data, but that KVM is getting cute and only reading
it when it can be non-zero (as far as KVM knows).

As a bonus, the code is less ugly:

		if (cpu_feature_enabled(X86_FEATURE_FRED))
			fred_entry_from_kvm(EVENT_TYPE_NMI, NMI_VECTOR,
					    vmx_get_exit_qual(vcpu));
		else
			vmx_do_nmi_irqoff();

> +		} else {
>  			vmx_do_nmi_irqoff();
> +		}
>  		kvm_after_interrupt(vcpu);
>  	}
>  
> -- 
> 2.25.1
>