Re: [PATCH 00/12] KVM: arm64: nv: Implement FEAT_XNX and FEAT_HAF

[Marc Zyngier <maz@kernel.org>]

On Wed, 12 Nov 2025 18:33:54 +0000,
Oliver Upton <oupton@kernel.org> wrote:
> 
> This series closes a couple of gaps between our shadow stage-2
> implementation and the architecture:
> 
> FEAT_XNX - KVM doesn't make use of the feature for itself but this is
> a rather low hanging fruit which entails expressing privileged and
> unprivileged execute permissions in our pseudo-TLB

I think this part looks good, no objection from my end.

> 
> FEAT_HAF - This one is a bit more involved, requiring the PTW
> implementations to atomically update descriptors in user memory to set
> the Access Flag. I've made the implementation choice that AT
> instructions also update the AF which was done to avoid evaluating the
> access type in the PTW.

This one needs a bit more work. There is a couple of amusing bugs that
need addressing. Hopefully you can respin it shortly so that we can
have this series in 6.19.

> There's more work to be done for FEAT_HAFDBS (dirty state updates),
> since updates to DBM are conditional based on the walk context. Lastly,
> this is all horribly untested since I don't have an NV-capable machine
> with FEAT_HAF that is easy to hack on. Although I do plan on adding some
> selftests coverage soon.

Yup, it looks like systems accessible to the common mortals don't have
any of it (my QC box doesn't either). Oh well...

Thanks,

	M.

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