On Thu, Jun 14, 2018 at 05:40:41PM +0200, Paolo Bonzini wrote:
> On 14/06/2018 10:18, Daniel P. Berrangé wrote:
> > I don't think
> > the -realtime flag should ever have been introduced, and we certainly
> > shouldn't add more stuff under it.
> >
> > "-realtime" is referring to a very specific use case, while the
> > properties listed under it are all general purpose features. Real
> > time guests just happen to be one possible use case, but it is
> > valid to use them for non-real time guests.
> >
> > IOW, I think we should just have this as an option under -cpu or
> > some other *functionally* named option, not a option named after
> > a specific usage scenario.
>
> "-cpu" is certainly wrong for KVM_CAP_X86_DISABLE_EXITS. "-cpu" is a
> device option, while this is about host behavior. "-realtime"'s name is
> awful, but I still think it's the best place for this option. Maybe we
> could call it "-realtime power-mgmt={host|guest}".
I don't feel it's a good name. The new flag allows guest to put CPU
into a low power state without host knowing about it. This will affect
other guests on the same host CPU. It doesn't however limit power
management to guest only, things like speedstep remain under host
control.
> A separate issue is whether the same flag should control both
> KVM_CAP_X86_DISABLE_EXITS and the monitor/mwait CPUID leaf. Eduardo,
> what do you think?
>
> Paolo
It's just for -cpu host.
