On 2011-05-03, at 9:32 AM, Shriram Rajagopalan wrote: On Tue, May 3, 2011 at 7:42 AM, Tim Deegan <Tim.Deegan@xxxxxxxxxx> wrote:
At 15:31 +0100 on 03 May (1304436713), Ian Campbell wrote:
> On Tue, 2011-05-03 at 15:09 +0100, Shriram Rajagopalan wrote:
>
> > Remus works with HVM domains via normal xenstore based suspend/resume.
> > Only PV-HVM support is "disabled" for the moment.
>
> I'm confused ;-)
>
> PV-HVM uses PV suspend, which can be triggered either by xenstore or the
> suspend evtchn, but the choice of which is mostly orthogonal to whether
> or not suspend cancellation is supported.
>
> The alternative to PV suspend is the inject an ACPI S3 (or whatever)
> event option, isn't it?
save and restore functions. For Remus, where downtime will be low and
storage consistency is handled elsewhere, that's good enough.
Tim.
Enabling it is a matter of just removing the check for pvhvm, which I ll remove
and send out a patch soon. :-)
I was waiting for the last remus patch (remus: fix incorrect error handling for
switch_qemu_logdirty...) to get pulled in before sending the patch.
As we discussed earlier, please do not do this without actually testing that pvhvm checkpointing works. On windows. PV suspend is a cooperative process and the reason it is disabled is that it hasn't been tested.
|
