Re: [PATCH] KVM: SVM: Add a dedicated INVD intercept routine

[Tom Lendacky]

On 9/24/20 1:51 AM, Paolo Bonzini wrote:
> On 23/09/20 22:40, Tom Lendacky wrote:
> > > > +static int invd_interception(struct vcpu_svm *svm)
> > > > +{
> > > > +	/*
> > > > +	 * Can't do emulation on an SEV guest and INVD is emulated
> > > > +	 * as a NOP, so just skip the instruction.
> > > > +	 */
> > > > +	return (sev_guest(svm->vcpu.kvm))
> > > > +		? kvm_skip_emulated_instruction(&svm->vcpu)
> > > > +		: kvm_emulate_instruction(&svm->vcpu, 0);
> > >
> > > Is there any reason not to do kvm_skip_emulated_instruction() for both SEV
> > > and legacy?  VMX has the same odd kvm_emulate_instruction() call, but AFAICT
> > > that's completely unecessary, i.e. VMX can also convert to a straight skip.
> >
> > You could, I just figured I'd leave the legacy behavior just in case. Not
> > that I can think of a reason that behavior would ever change.
>
> Yeah, let's do skip for both SVM and VMX.

Ok, I'll submit a two patch series to change SVM and VMX. I'll do two 
patches because of the fixes tag to get the SVM fix back to stable. But, 
if you would prefer a single patch, let me know.

Thanks,
Tom

> Paolo