On Tue, Aug 02, 2016 at 16:20:50 +1000, David Gibson wrote: > On Tue, Aug 02, 2016 at 10:34:34AM +0530, Bharata B Rao wrote: > > On Tue, Aug 02, 2016 at 02:25:08PM +1000, David Gibson wrote: > > > On Power, support for vCPU hotplug is new in qemu 2.7. However, we > > > currently implement the query_hotpluggable_cpus hook the same for all > > > pseries machine type versions. > > > > > > However, the old-style CPU initialization doesn't work with the new query > > > code, meaning that attempting to use query-hotpluggable-cpus on a > > > pseries-2.6 or earlier VM will cause qemu to SEGV. > > > > > > This fixes the problem by simply disabling the hook for earlier machine > > > types. > > > > I had sent a patch to fix this and a couple of other related issues > > some time back and it indeed was accepted into your ppc-for-2.7 branch. > > > > https://lists.gnu.org/archive/html/qemu-devel/2016-07/msg01539.html > > > > Only now I am realizing that somehow that patch didn't make it to mainline. > > Oh.. good point. Sorry, that one somehow slipped through the cracks. > > So, the remaining question is, what's the preferred behaviour for > older machine types: > > 1) should query-hotpluggable-cpus give an error, the same as it does > on machine types which have never supported it (this is what my > patch does) > > or > > 2) Should query-hotpluggable-cpus succeed, but return an empty list? > (this is what Bharata's patch does) > > Igor and / or Peter, do you have an opinion on which behaviour is preferable?
I don't really care which option you select as long as the 'hotpluggable-cpus' field which is repored in 'query-machines' is set to false for machine types which don't support it. Libvirt then won't even call query-hotpluggable-cpus. Peter
