On Sat, May 2, 2026 at 12:30 AM Peter Maydell <[email protected]> wrote:
>
> On Fri, 1 May 2026 at 15:02, Leif Lindholm
> <[email protected]> wrote:
> >
> > On Fri, 1 May 2026 at 04:28, Alistair Francis <[email protected]> wrote:
> > > The first line of the README says:
> > >
> > > """
> > > This document is capturing discussions at the Server Platform TG.
> > > These are not official specifications and everything in this document
> > > may change.
> > > """

Ah, this was just ratified. So now there is a version 1.0 and no "this
may change at any point" notice.

> > >
> > > So this machine needs to start out with version support (like the ARM
> > > virt board)
> >
> > We ended up not doing that for sbsa-ref, partially based on advice
> > from Peter (added to cc),
> > which I can't right now bring to mind in any great detail.
> >
> > I would expect the same argument holds here. Peter?
>
> Roughly, QEMU versioned machine types are a big maintenance pain,
> so we should only have them where the use-case is that users
> really want to be able to run and migrate their workloads from
> old QEMU to new QEMU. The use case for sbsa-ref is "I am developing
> the full firmware/system software stack and I want a reference
> platform for that" -- nobody wants to be able to run the older
> version of the firmware/software stack on the newer QEMU,
> and certainly nobody is going to be doing live migration of the
> software between QEMU versions. The small population of people
> doing this development work will be running the bleeding edge of
> all the components involved anyway, and to the extent they care
> about running on an older / newer QEMU for convenience they can
> handle that in the firmware -- sbsa-ref passes the firmware a
> "machine-version-major" and "machine-version-minor" that we
> bump when we add something that would be a compat break with the
> firmware.

Now that the spec is ratified and this isn't expected to change weekly
in user breaking ways I agree, thanks Peter

Alistair

Reply via email to