Re: [PATCH] target/ppc/cpu-models: Update max alias to power10

From: "Daniel P. Berrangé" <berrange@redhat.com>
To: Greg Kurz <groug@kaod.org>
Cc: "Thomas Huth" <thuth@redhat.com>,
	"Murilo Opsfelder Araujo" <muriloo@linux.ibm.com>,
	qemu-devel@nongnu.org, qemu-ppc@nongnu.org,
	"Cédric Le Goater" <clg@kaod.org>,
	"Daniel Henrique Barboza" <danielhb413@gmail.com>,
	"David Gibson" <david@gibson.dropbear.id.au>,
	mopsfelder@gmail.com, "Fabiano Rosas" <farosas@linux.ibm.com>
Subject: Re: [PATCH] target/ppc/cpu-models: Update max alias to power10
Date: Mon, 6 Jun 2022 11:59:10 +0100	[thread overview]
Message-ID: <Yp3d/pdf/8Pwth88@redhat.com> (raw)
In-Reply-To: <20220601120324.66c77571@bahia>

On Wed, Jun 01, 2022 at 12:03:24PM +0200, Greg Kurz wrote:
> On Wed, 1 Jun 2022 11:25:43 +0200
> Thomas Huth <thuth@redhat.com> wrote:
> 
> > On 01/06/2022 10.38, Greg Kurz wrote:
> > > On Wed, 1 Jun 2022 09:27:31 +0200
> > > Thomas Huth <thuth@redhat.com> wrote:
> > > 
> > >> On 31/05/2022 19.27, Murilo Opsfelder Araujo wrote:
> > >>> Update max alias to power10 so users can take advantage of a more
> > >>> recent CPU model when '-cpu max' is provided.
> > ...
> > > We already have the concept of default CPU for the spapr
> > > machine types, that is usually popped up to the newer
> > > CPU model that is going to be supported in production.
> > > This goes with a bump of the machine type version as
> > > well for the sake of migration. This seems a lot more
> > > reliable than the "max" thingy IMHO.
> > > 
> > > Unless there's a very important use case I'm missing,
> > > I'd rather kill the thing instead of trying to resurrect
> > > it.
> > 
> > It's about making ppc similar to other architectures, which
> > have "-cpu max" as well, see:
> > 
> >   https://gitlab.com/qemu-project/qemu/-/issues/1038
> > 
> > It would be nice to get something similar on ppc.
> > 
> 
> Problem is that on ppc, given the variety of models and boards,
> the concept of "max" is quite fuzzy... i.e. a lot of cases to
> consider for a benefit that is unclear to me. Hence my questioning.
> If the idea is just to match what other targets do without a specific
> use case in mind, this looks quite useless to me.

Essentially "max" is a way for a mgmt application to ask for the
most feature rich CPU available for their given configuration.

With KVM this is trivially equiv to '-cpu host'.

With TCG on other archs this has been "all the features that
TCG implements". This implicitly assumes that CPU features are
generally additive and compatible with all historical machine
types. This is fine for x86, but if it doesn't work for PPC
we can come up with an alternative definition.

For example if you have a set of 3 possible CPU models that
work with a given machine board, then expand 'max' to be the
"best" of these 3 possible models.  This still satisfies the
goal of the mgmt app, which is to be able to ask for a good
CPU without having to know its architecture/machine type
specific name.

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 :|