Re: [PATCH v4 0/3] KVM: arm64: Errata management for VM Live migration

[Oliver Upton <oliver.upton@linux.dev>]

On Thu, Dec 19, 2024 at 10:07:39AM +0000, Marc Zyngier wrote:
> Hi Shameer,
> 
> On Wed, 18 Dec 2024 10:53:42 +0000,
> Shameer Kolothum <shameerali.kolothum.thodi@huawei.com> wrote:
> > 
> > Hi,
> > 
> > v3 --> v4(Minor updates)
> > https://lore.kernel.org/kvmarm/20241209115311.40496-1-shameerali.kolothum.thodi@huawei.com/
> > 
> >  -Changed MIDR/REVIDR to 64 bits based on feedback from Connie
> >   and Marc(Patch #3).
> >  -Added R-by tags from Sebastian (Thanks!).
> 
> Thanks again for putting this together.
> 
> I think it would be really good to have a sample userspace
> implementation of this extension so that we can play with it for real
> before fully committing to it.
> 
> I am also wondering is we should make it mandatory that a guest is
> presented with an MIDR_EL1.Implementer value set to 0, which denotes
> SW use, and would make it plain to the guest (and crucially, guest
> userspace) that we are going to play tricks.
> 
> Thoughts?

I see no issues with giving userspace the option of doing this, but I'd
rather not enforce this to use the PV interface.

At least in a cloud setting it seems highly likely that you'd be running
a mix of guests on the same VM definition, some aware of the PV
interface and others not. If we use a software MIDR, all old VMs will
lose errata mitigations and whatever else userspace might key off of MIDR.

Userspace can make the call whether or not a VM can be migrated onto a
different implementation based on whether or not the guest used the
hypercall interface. And maybe just terminate the VM if it didn't.

-- 
Thanks,
Oliver