On Mon, Nov 23, 2015 at 02:04:46PM +0100, Christian Borntraeger wrote:
> On 11/23/2015 12:54 PM, Peter Krempa wrote:
> > On Fri, Nov 20, 2015 at 18:24:29 +0530, Bharata B Rao wrote:
> >> This patchset adds CPU hotplug support for sPAPR PowerPC guests using
> >> device_add and device_del commands
> >>
> >> (qemu) device_add POWER8-powerpc64-cpu,id=cpu0
> > 
> > Is there a reason why this uses 'device_add' rather than the 'cpu_add'
> > command? Libvirt uses two separate approaches already. Due to legacy
> > reasons we support the HMP 'cpu_set' command, and lately we added
> > support for QMP 'cpu-add'. Using device_add here will introduce a
> > different approach and will require yet another compatibility layer in
> > libvirt to support this.
> 
> s390 and powerpc both started with cpu_add patches. Andreas Faerber
> suggested then to only implement device_add. This was apparently discussed
> at the last KVM forum.

This may not be the only reason, but one problem with cpu_add is that
the interface is strictly per-vcpu, i.e. per-thread.  That doesn't
work for Power guests, because the PAPR interface used to communicate
to the guest can only work at a core granularity.

-- 
David Gibson                    | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au  | minimalist, thank you.  NOT _the_ _other_
                                | _way_ _around_!
http://www.ozlabs.org/~dgibson

Attachment: signature.asc
Description: PGP signature

Reply via email to