On Wed, Nov 19, 2014 at 09:31:35AM -0700, Eric Blake wrote: > On 11/19/2014 09:06 AM, Michael S. Tsirkin wrote: > > >>>> This affects QMP right? > >>> > >>> I think later patches will tell how. CC'ing Eric. > >> > >> As far as I can tell, this is just correcting a reporting issue; the > >> existing QMP commands/events for tracking balloon size will now properly > >> account for hotplugged memory. > >> > >> What I don't know is if this change in semantics will affect any users. > >> Libvirt is not yet supporting memory hotplug, so ideally, fixing this > >> bug before libvirt uses memory hotplug means libvirt will never have to > >> worry about qemu versions that do incorrect reporting. > >> > >> The alternative is to declare that the existing QMP commands cannot > >> change in semantics for the existing members that it reports, and must > >> instead report additional dictionary members describing the amount of > >> hot-plugged memory, and then require that the client add the numbers > >> together itself. That sounds mean to the client, so I'm hoping we don't > >> have to go there. > > > > > > IOW you ack this patch for 2.2? > > > > Is memory hotplug one of the new features in 2.2? If so, then yes, we > should get its semantics right from the start (this is a bug fix to > avoid a release with broken semantics). On the other hand, if hotplug > existed in 2.1, then we already have a release with odd semantics, so > delaying this fix until 2.3 and leaving 2.2 with the same odd semantics > would not hurt, and it then becomes a judgment call of whether we are > rushing in a possibly incomplete solution by trying to get this into > 2.2. (Sorry I haven't been following the history of memory hotplug closer)
AFAIK it's there since 2.0. > > -- > Eric Blake eblake redhat com +1-919-301-3266 > Libvirt virtualization library http://libvirt.org >