Re: [PATCH v1 0/2] s390x/cpumodel: Introduce "best" model variants

[David Hildenbrand]

On 19.11.19 10:22, Peter Maydell wrote:
> On Mon, 18 Nov 2019 at 22:04, Eduardo Habkost <address@hidden> wrote:
> > On Mon, Nov 18, 2019 at 09:19:55PM +0000, Peter Maydell wrote:
> > > Why should it matter whether a feature is enabled
> > > or disabled by default in the CPU type? It ought to be probeable
> > > by QMP either way (ie there is a difference between
> > > "this CPU has this feature switch and it is on by default",
> > > "this CPU has this feature switch and it is off by default"
> > > and "this CPU does not have this feature switch at all",
> > > and presumably libvirt needs to distinguish them).
> >
> > Its use case is neither "this CPU has this feature switch" or for
> > "it is on|off by default".  We use it to probe for "this feature
> > can be enabled in this host hardware+kernel+QEMU combination".
>
> OK. Well, the answer to that depends on the name of the CPU,
> in general. So you can't use a fake CPU name to try to answer
> the question.
>
> > In other words, in x86 and s390x "max" is just a reserved name
> > for the query-cpu-model-expansion command arguments in s390x and
> > x86.  The fact that it is visible to users can be considered a
> > bug, and we can fix that.
>
> I think 'max' is useful to users, and we've provided it to users,
> so removing it again would be a compatibility break. I'm not
> entirely sure where we go from here...

I agree, right now on s390x "-cpu max" behaves almost like "-cpu 
qemu"/"-cpu host", but you don't have to care about the accelerator. 
Removing it might we evil. Removing it is not really an option.

> > If you still don't like how query-cpu-model-expansion works, we
> > can also discuss that.  But I'm not sure it would be a good use
> > of our (and libvirt developers') time.
>
> I don't hugely care about query-cpu-model-expansion. I
> just don't want it to have bad effects on the semantics
> of user-facing stuff like x- properties.

IMHO, max should really include all features (yes, also the bad 
x-features on arm :) ) and we should have a way to give users the 
opportunity to specify "just give me the best model independent of the 
accelerator" - something like a "best" model, but I don't care about the 
name.

E.g., we could introduce a new model for that purpose and start warning 
if "-cpu max" is used (and recommend them to use the other model) to run 
a VM and not simply to query model properties via the qmp interfaces. 
The users are aware of the model restrictions and can switch.

--

Thanks,

David / dhildenb