On Tue, Dec 03, 2024 at 01:14:07PM +0100, Jiri Denemark wrote:
> On Fri, Nov 29, 2024 at 09:03:11 +0000, Daniel P. Berrangé wrote:
> > On Fri, Nov 29, 2024 at 09:50:01AM +0100, Jiri Denemark wrote:
> > > On Fri, Nov 29, 2024 at 12:33:19 +0800, Han Han wrote:
> > > > Tested on this branch with qemu-kvm-9.1.0-5.el9.x86_64:
> > > > # for i in $(/usr/libexec/qemu-kvm -cpu help|grep deprecated -v|awk
> > > > '/Available CPUs/,/Recognized CPUID flags/'|grep '^ '|awk '{print $1}');do
> > > > if ! virsh cpu-models x86_64|grep -q $i;then echo $i;fi;done
> > > > Opteron_G4-v1
> > > > Opteron_G5-v1
> > > > base
> > > > host
> > > > max
> > > >
> > > > Opteron_G4-v1 and Opteron_G5-v1 are not deprecated. Expect to add them to
> > > > libvirt CPU models as well.
> > >
> > > I was not really sure which CPU models are deprecated. According to QEMU
> > > none of them is really deprecated (the only CPU model that was ever
> > > deprecated was Icelake-Client, which was later dropped completely). The
> > > info you use is apparently coming from downstream QEMU, because upstream
> > > shows nothing for "qemu-system-x86_64 -cpu help | grep deprecated".
> >
> > Correct, deprecation of CPUs is a decision RHEL makes downstream, and is
> > not relevant to libvirt's upstream usage.
> >
> > Libvirt queries QEMU deprecations, so if a user picks a deprecated CPU,
> > their VM will be tainted and show the deprecation message in logs, etc.
> >
> > > I guess we can use the info to say Opteron_G4 and Opteron_G5 should not
> > > be ignored by the script, I'm not sure we could use it the other way
> > > around for selecting which models are considered deprecated.
> >
> > We should always honour all CPUs QEMU reports as existing. Deprecated CPUs
> > are still supported for use, it is merely a warning that they /may/ go away
> > in future.
>
> So do you suggest we should not ignore even those ancient lower case CPU
> models and add their -v1 variants as well? That would seems to me like a
> lot of churn with no benefit. Although there's no technical reason for
> ignoring them.
My inclination is that we should not special case anything.
If I look at -cpu help there are 143 CPus currently. 3 are special (base,
host, max). Of the 140, 24 are the old lowercase models you mention, 12
traditional names, and 12 -v1 names. IOW, we've currently got 128 CPUs,
and we're excluding 12. That doesn't seems like a worthwhile thing to
specialcase.
With regards,
Daniel
--
|: https://berrange.com -o- https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org -o- https://fstop138.berrange.com :|
|: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|