On Thu, 22 Oct 2020 at 01:37, Jim Mattson <jmattson@xxxxxxxxxx> wrote:
>
> On Wed, Oct 21, 2020 at 7:57 AM Vitaly Kuznetsov <vkuznets@xxxxxxxxxx> wrote:
> >
> > Wanpeng Li <kernellwp@xxxxxxxxx> writes:
> >
> > > On Wed, 21 Oct 2020 at 14:47, Jim Mattson <jmattson@xxxxxxxxxx> wrote:
> > >>
> > >> Per the KVM_GET_SUPPORTED_CPUID ioctl, the KVM_HINTS_REALTIME CPUID
> > >> bit is never supported.
> > >>
> > >> Is this just an oversight?
> > >
> > > It is a performance hint, not a PV feature and doesn't depend on KVM.
> > >
> >
> > True, but personally I'd prefer it to be reported in
> > KVM_GET_SUPPORTED_CPUID too, that would mean that userspace is in its
> > right to set it, just like any other PV 'thingy', even if just for
> > consistency.
>
> I agree. The documentation says:
>
> This ioctl returns x86 cpuid features which are supported by both the
> hardware and kvm in its default configuration.
>
> If the bit isn't set, then it is not supported, and a well-behaved
> userspace will not set it.
>
> Now, if the bit were in a completely different CPUID leaf (not a
> KVM-defined leaf), then one could argue that it falls outside of the
> realm of KVM_GET_SUPPORTED_CPUID. But it isn't, is it?
Fair enough, just sent out a patch.
Wanpeng
