[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [libvirt] Support for kvm=off cpu option



On Fri, 2014-08-15 at 10:38 -0400, Cole Robinson wrote:
> On 08/11/2014 05:54 PM, Alex Williamson wrote:
> > As of QEMU 2.1 we now have a new -cpu option, kvm=on|off which controls
> > whether we expose KVM as the hypervisor via the MSR hypervisor nodes.
> > The default is on.  The off state is meant to hide kvm from standard
> > detection routines.  This allows us to disable paravirtualization
> > detection in the guest and can be used to bypass hypervisor checks in
> > certain guest drivers[1].
> > 
> > How would we like to enable this in libvirt?  One option would be a
> > processor feature, ex:
> > 
> >         <feature policy='disable' name='kvm'/>
> > 
> > But since this doesn't map to a cpuid feature, I don't know if that
> > would be acceptable.
> > 
> > Another option might be to a domain feature, similar to some of the
> > other hyper-v related flags.
> > 
> >         <domain type='kvm'>
> >           <features>
> >             <kvm>
> >               <kvm state='off'>
> >             </kvm>
> >             <hyperv>
> >               <relaxed state='on'>
> >               ...
> >             </hyperv>
> >           </features>
> >           ...
> >         </domain>
> > 
> > Any other suggestions?  Preference?  Thanks,
> 
> IMO the latter suggestion makes the most sense.

Yeah, that's what I'm leaning towards too.

> However <kvm state='off'/> could cause confusion, sounds like it turns off kvm
> acceleration. Maybe <signature state=on|off>, or hv_signature, something along
> those lines.

I was actually working with "hidden" so that the default would be off.
I think using an hv_ prefix might make it easily confused as a HyperV
option.  Thanks,

Alex


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]