Re: [Qemu-devel] [PATCH 0/3] target-i386: Don't trigger "check" warnings by default in KVM mode

[Eduardo Habkost]

On Tue, Nov 03, 2015 at 06:49:08PM +0100, Paolo Bonzini wrote:
> 
> 
> On 03/11/2015 18:41, Eduardo Habkost wrote:
> > I just got a report from David that the lack of ABM on Sandy Bridge
> > hosts will trigger the warning, too. I guess we don't want to require a
> > Haswell CPU to run the default CPU model with KVM, so ABM would be
> > disabled by default in KVM mode too.
> > 
> > We really should make the defaults different in TCG and KVM mode. The
> > question is what to do about the TCG defaults in QEMU 2.5.
> > 
> > I propose we remove ABM and SSE4E from qemu64 in 2.5,
> 
> But if we do that for all machine types we break guest ABI for >=Haswell
> on old machine types; if we do that for 2.5 only we get "-cpu check"
> warnings on old machine types.

I don't propose we remove it from all machine-types, but just from
pc-2.5 and newer.

> 
> This is true even if we implement different defaults for KVM and TCG.
> Should we disable "-cpu check" for old machine types (or do we already
> do that)?  That would suck, but it would let us drop ABM and SSE4A from
> the 2.4 machine type without much hassle.

You mean dropping from the 2.5 machine-type, right?

> 
> > and implement a
> > proper mechanism to have different KVM and TCG defaults after 2.5. We
> > never pretended that qemu64 had all TCG-supported features enabled
> > anyway: the last time the feature set of qemu64 was changed was in 2009
> > (when we added ABM).
> 
> That's true.

So, trying to summarize what I would like to do:

* Remove ABM and SEE4A from qemu64 in pc-2.5
* Keep ABM and SSE4A in qemu64 in pc-2.4 and older
* Disable "check" by default in pc-2.4 and older, to avoid spurious
  warnings

-- 
Eduardo