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

Re: [Xen-devel] [PATCH] x86/hvm: implement save/restore for posted interrupts

> From: Jan Beulich [mailto:JBeulich@xxxxxxxx]
> Sent: Friday, July 25, 2014 7:00 AM
> >>> On 25.07.14 at 14:49, <kevin.tian@xxxxxxxxx> wrote:
> >>  From: Jan Beulich [mailto:JBeulich@xxxxxxxx]
> >> Sent: Friday, July 25, 2014 2:41 AM
> >>
> >> >>> On 03.07.14 at 17:09, <olaf@xxxxxxxxx> wrote:
> >> > Saving and restoring a PVonHVM guest on a host which has the VMX
> >> > "Posted Interrupt Processing" feature enabled will fail because the
> >> > xen-platform-pci device does not receive interrupts anymore after
> >> > restore. The reason is that the IRQs are not maintained in APIC_IRR,
> >> > but in a separate PIR array. This info is lost during the save
> >> > operation.
> >> >
> >> > Syncing the PIR state into IRR during save, and restoring the state
> >> > later will fix 'xm save/restore' of a PVonHVM guest.
> >> >
> >> > Signed-off-by: Olaf Hering <olaf@xxxxxxxxx>
> >>
> >> Guys, where are we with this? Afaict all simplification suggestions
> >> you provided so far yielded a non-working patch. Are you actively
> >> working on getting a working alternative, or can we get an ack on
> >> what Olaf submitted?
> >>
> >
> > I hope Olaf can figure out why sync_irr_to_pir is actually required before
> > I can ack.
> I was thinking it would be the other way around: You suggesting it's
> not needed (which is wrong according to observations) would offer
> an explanation and/or a working alternative fix. I don't think you can
> expect Olaf to explain why what you suggest doesn't work, even
> more so with _you_ being the VMX maintainer.

Well, my read of this patch is that it hides some problem other place by
forcing posted injection at restore. As Yang has pointed out, it is not
necessary once the notification has been synced to IRR it's done (will 
be noted before vmentry). So at restore path, it's just about how pending 
IRR is handled. Now the problem is that Yang can't reproduce the problem 
locally (let's see any change with Olaf's further information), so we need 
Olaf's help to figure out the real culprit with our input.


Xen-devel mailing list



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