Re: ARM Snapshots Not Backwards-Compatible

[Andrew Jones]

On Wed, Feb 03, 2021 at 10:52:59AM +0000, Peter Maydell wrote:
> On Wed, 3 Feb 2021 at 10:49, Dr. David Alan Gilbert <dgilbert@redhat.com> 
> wrote:
> >
> > * Peter Maydell (peter.maydell@linaro.org) wrote:
> > > On Wed, 3 Feb 2021 at 10:28, Dr. David Alan Gilbert <dgilbert@redhat.com> 
> > > wrote:
> > > >
> > > > * Philippe Mathieu-Daudé (philmd@redhat.com) wrote:
> > > > > Cc'ing migration team and qemu-arm@ list.
> > > >
> > > > I'll have to leave the detail of that to the ARM peole; but from a
> > > > migration point of view I think we do want the 64 bit ARM migrations to
> > > > be stable now.  Please tie incompatible changes to machine types.
> > >
> > > That is the intention, but because there's no upstream testing
> > > of migration compat, we never notice if we get it wrong.
> > > What is x86 doing to keep cross-version migration working ?
> >
> > I know there used to be some of our team running Avocado tests for
> > compatibility regularly, I'm not sure of the current status.
> > It's something we also do regularly around when we do downstream
> > releases, so we tend to catch them then, although even on x86 that
> > often turns out to be a bit late.
> 
> So downstream testing only?

Not even downstream for the Arm architecture, at least not at Red Hat. The
support we have for Arm Virt is still limited to the use cases for which
it has been deployed. To this day that hasn't included migration[*].

> I think that unless we either (a) start
> doing migration-compat testing consistently upstream or

This is the best choice and it can certainly be an additional approach
regardless of what goes on downstream. I can look into this. A pointer
to the x86 tests would be a good start. It's pretty simple to do a
quick migration test between two versions of QEMU, but we need the
whole build two versions of QEMU stuff first, which I hope already
exists.

> (b) RedHat or
> some other downstream start testing and reporting compat issues
> to us for aarch64 as they do for x86-64,

Red Hat isn't currently allocating resources to this type of testing
for AArch64. And, even if it were to start, it would likely focus on
compat testing of RHEL arm-virt machine types, which aren't exactly the
same thing as the upstream arm-virt machine types. In fact, currently,
there are only two (virt-rhel8.2.0 and virt-rhel8.3.0).

Thanks,
drew

[*] Migration support hasn't been a real high priority for AArch64 KVM,
    because it currently requires identical host CPU types and identical
    host kernels to work. It'd be nice if the destination host kernel
    could at least be a later version, but not even that is guaranteed
    to work at this time.