Re: [PATCH v14 3/5] KVM: arm64: Manage GCS access and registers for guests

[Marc Zyngier <maz@kernel.org>]

On Sat, 05 Oct 2024 14:08:39 +0100,
Mark Brown <broonie@kernel.org> wrote:
> 
> On Sat, Oct 05, 2024 at 12:34:20PM +0100, Marc Zyngier wrote:
> > Mark Brown <broonie@kernel.org> wrote:
> 
> > > +	if (!kvm_has_gcs(kvm)) {
> > > +		kvm->arch.fgu[HFGxTR_GROUP] |= (HFGxTR_EL2_nGCS_EL0 |
> > > +						HFGxTR_EL2_nGCS_EL1);
> > > +		kvm->arch.fgu[HFGITR_GROUP] |= (HFGITR_EL2_nGCSEPP |
> > > +						HFGITR_EL2_nGCSSTR_EL1 |
> > > +						HFGITR_EL2_nGCSPUSHM_EL1);
> 
> > Where is the handling of traps resulting of HFGITR_EL2.nGCSSTR_EL1?
> 
> These will trap with an EC of 0x2d which isn't known so I was expecting
> this to get handled in the same way as for example a return of false
> from kvm_hyp_handle_fpsimd() for SVE when unsupported, or for the
> simiarly unknown SME EC, currently.  I gather from your comment that
> you're instead expecting to see an explicit exit handler for this EC
> that just injects the UNDEF directly?

Not just inject an UNDEF directly, but also track whether this needs
to be forwarded when the guest's HFGITR_EL2.nGCSSTR_EL1 is 0 while not
being not RES0. Basically following what the pseudocode describes.

	M.

-- 
Without deviation from the norm, progress is not possible.