On Wed, May 17, 2023 at 09:14:02AM +0200, Michal Prívozník wrote:
> On 5/16/23 18:32, Andrea Bolognani wrote:
> > Last but not least, the way detection has been implemented is not
> > accurate: as of today, QEMU does *not* support enabling MTE with KVM.
> > Patches adding this feature have been posted[1] and are going to be
> > merged soon, but even then just looking at the machine type property
> > is not going to be enough to determine whether MTE can actually be
> > used.
>
> Then it's a matter of:
>
> + if (kvm_arm_mte_supported()) {
> + object_class_property_add_bool(oc, "mte", virt_get_mte,
> virt_set_mte);
> + object_class_property_set_description(oc, "mte",
> + "Set on/off to enable/disable
> emulating a "
> + "guest CPU which implements
> the ARM "
> + "Memory Tagging Extension");
> + }
I don't think this would work: even if KVM doesn't support MTE, TCG
can still emulate it, so the property still needs to show up.
> Or querying KVM extensions in libvirt (we already do that for some features).
That would tell us whether KVM itself is MTE-capable, but not whether
the QEMU binary can make use of that feature.
> > Committing to any specific interface in libvirt at this point in time
> > feels premature, as it's pretty much guaranteed that it will no
> > longer fit once the questions above have been answered.
>
> Fair enough, feel free to revert my patches.
Let's keep the discussion going for now, but if we get very close to
the freeze without a clear consensus on the one you have implemented
being the interface that we want I'll probably post a revert.
--
Andrea Bolognani / Red Hat / Virtualization
