On Fri, May 05, 2017 at 02:59:24PM -0300, Eduardo Habkost wrote:
> On Thu, May 04, 2017 at 12:16:38PM +0200, Kashyap Chamarthy wrote:
> > On Tue, May 02, 2017 at 05:31:15PM -0300, Eduardo Habkost wrote:
> >
> > Hi,
> >
> > Yet to try this series, a small question in-line.
> >
> > > Introduce a new CPUFeatureSetting QAPI data type, and use it to support
> > > feature=force on -cpu.
> > >
> > > Signed-off-by: Eduardo Habkost <ehabk...@redhat.com>
> [...]
> > > @@ -336,5 +340,13 @@ int main(int argc, char **argv)
> > > "-machine accel=kvm:tcg -cpu max,mmx=off",
> > > 1, 0, "EDX", 23, false);
> > >
> > > + {
> > > + FeatureTestArgs *a;
> > > + a = add_feature_test("x86/cpuid/features/monitor-force",
> > > + "-machine accel=kvm:tcg -cpu 486,monitor=force",
> >
> > Following your above test example, should the 'force' boolean also work
> > (understood: only for testing / debugging) as below, for a recognized
> > CPUID flag, taking the example of 'INVPCID'?
> >
> > $ qemu-system-x86_64 [...] -cpu Haswell-noTSX,invpcid=force [...]
>
> This would be valid syntax and would enable invpcid on CPUID,
> yes. But on most cases this means you will get a broken VCPU
> because the host won't virtualize the feature properly (if it
> did, the feature would be reported as available in
> GET_SUPPORTED_CPUID and "feature=on" would already work).
Thanks for the clarification. (/me today learnt that
GET_SUPPORTED_CPUID is a KVM ioctl(2).
--
/kashyap
