Re: [PATCH RFC v1 07/20] KVM: nVMX: Support the extended instruction info field

[Paolo Bonzini]

On Tue, Nov 18, 2025 at 2:39 AM Chao Gao <chao.gao@xxxxxxxxx> wrote:
>
> On Tue, Nov 18, 2025 at 12:29:19AM +0100, Paolo Bonzini wrote:
> >Il mer 12 nov 2025, 02:54 Chao Gao <chao.gao@xxxxxxxxx> ha scritto:
> >>
> >> Shouldn't we check guest's capabilities rather than host's,
> >>
> >> i.e., guest_cpu_cap_has(X86_FEATURE_APX)?
> >
> >As the manual says, you're free to use the extended field if
> >available, and it's faster.
>
> The point is, from the guest's perspective, the field is available iff the vCPU
> supports APX. KVM (L0) doesn't need to virtualize VMCS12's EII field if the vCPU
> doesn't have APX.

Well, it would be faster to just do it. But you're right, checking the
guest CPUID is consistent with other code, for example

        if (nested_cpu_has_vid(vmcs12))
                vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);

Paolo