On (Fri) 09 Dec 2011 [11:55:21], Daniel P. Berrange wrote:
> On Fri, Dec 09, 2011 at 05:19:35PM +0530, Amit Shah wrote:
> > Hello,
> >
> > These patches make qemu generate an error on failure in setting a
> > balloon value:
> >
> > (qemu) balloon 400
> > Device 'balloon' has not been activated
> >
> > This can happen when the guest hasn't yet loaded a balloon driver.
>
> This is a pretty significant change in behaviour/semantics of the
> balloon driver interface.
>
> eg, when libvirt starts a guest, it launches QEMU paused, and
> immediately sets a balloon target, which may be lower than the
> memory limit, and then starts CPUS to let the guest boot. The
> guest will utilize this target when it loads the balloon driver
> during bootup.
>
> With this proposed change, libvirt will immediately get
> an error, and have to guess when the balloon driver might be
> loaded by the guest, in order to set the limit post-boot.
>
> > Previously, such a ballooning attempt didn't produce an error, but the
> > virtio-balloon driver remembered the value nevertheless, causing a
> > module load in the guest using the attempted balloon value
> > originally. This may not be the desired course of action.
>
> This is *exactly* desired behaviour. Please don't change this.
OK, thanks.
I'll retract this patch series.
Amit
