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

Re: [Xen-devel] [PATCH 0/4] CPU hotplug port from qemu-traditionnal to qemu-xen for 4.3.

On Fri, 2013-05-31 at 18:20 +0100, Anthony PERARD wrote:
> On 31/05/13 17:33, Anthony PERARD wrote:
> > Hi all,
> > 
> > This series ports few patches from qemu-xen-traditionnal to qemu-xen
> > (upstream). There is as little modification as possible on those patches so
> > they *can not* be upstream as is. So this is only for 4.3. A proper series 
> > will
> > be done for 4.4.
> > 
> > There make possible to have CPU hotplug on qemu-xen.
> > 
> > A patch for libxl will follow, which make use of the old command line
> > -vcpu_avail with qemu-xen. This command line will probably not be upstream.
> > 
> > BUG: There is an issue with SeaBIOS (used with qemu-xen). At SMP
> > initialisation, SeaBIOS will count the number of CPU running, and the 
> > number is
> > always equal to maxvcpu when SeaBIOS expect less. This happen when we have
> > something like:
> > vcpus = 2
> > maxvcpus = 8
> > in the VM config file. Linux is fine with this and will use only $vcpus.  So
> > the probleme is: an infinit loop in SeaBIOS.
> To let you know, I just realize that QEMU 1.5 (we are using 1.3, I
> think) already have everything needed for CPU hotplug. It will just be a
> matter of plumbing libxl to use the cpu-add QMP command.

Cool! Is the implementation be very different to what we have here? Is
it backportable?

At a minimum I'd prefer that whatever we do now interface wise will
allow libxl to work with the qemu upstream mechanisms in the future (see
my comment on your libxl patch).

> But cpu-remove QMP command seams to be missing from QEMU.

For a long time we didn't support CPU hot remove from HVM guests, in
fact I'm not even sure that we support it today with qemu-trad.

Perhaps we do for PVHVM guests, but I'd half expect that to be using the
PV path.


Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.