Re: [nVMX with: v3.9-11789-ge0fd9af] Stack trace when L2 guest is rebooted.

[Gleb Natapov <gleb@redhat.com>]

On Mon, May 13, 2013 at 08:57:32AM +0200, Jan Kiszka wrote:
> On 2013-05-13 08:45, Ren, Yongjie wrote:
> >> -----Original Message-----
> >> From: kvm-owner@vger.kernel.org [mailto:kvm-owner@vger.kernel.org]
> >> On Behalf Of Gleb Natapov
> >> Sent: Monday, May 13, 2013 2:39 PM
> >> To: Jan Kiszka
> >> Cc: Kashyap Chamarthy; Abel Gordon; Nakajima, Jun;
> >> kvm@vger.kernel.org; kvm-owner@vger.kernel.org
> >> Subject: Re: [nVMX with: v3.9-11789-ge0fd9af] Stack trace when L2 guest
> >> is rebooted.
> >>
> >> On Mon, May 13, 2013 at 08:31:33AM +0200, Jan Kiszka wrote:
> >>> On 2013-05-12 18:52, Kashyap Chamarthy wrote:
> >>>>>> --------------------
> >>>>>> ....
> >>>>>> [  217.938034] Uhhuh. NMI received for unknown reason 30 on CPU
> >> 0.
> >>>>>> [  217.938034] Do you have a strange power saving mode enabled?
> >>>>>> .[  222.523373] Uhhuh. NMI received for unknown reason 20 on
> >> CPU 0.
> >>>>>> [  222.524073] Do you have a strange power saving mode enabled?
> >>>>>> [  222.524073] Dazed and confused, but trying to continue
> >>>>>> [  243.860319] Uhhuh. NMI received for unknown reason 30 on CPU
> >> 0
> >>>>>> .....
> >>>>>> --------------------
> >>>>>> At the moment, L2 guest creation stuck at the above message
> >>>>>>
> >>>>> Are those in L2 dmesg or L1?
> >>>>
> >>>> L2 dmesg.
> >>>>
> >>>>
> >>>>>> $ cat /etc/grub2.cfg  | egrep -i 'hpet|nmi'
> >>>>>>
> >>>>> IIRC watchdog is enabled by default.
> >>>>
> >>>> Indeed, you're right. I disabled NMI on L1, and rebooted the newly
> >>>> created L2 guest starts just fine.
> >>>
> >>> NMI watchdogs go via some perf counters theses days IIRC. Can anyone
> >>> tell me which of those may be used in Kashyap's setup? I'm probably
> >>> lacking them for my guests and therefore do not see the errors.
> >>>
> >> Try running with -cpu host for L1. Your CPU definition probably lacks
> >> PMU leaf.
> >>
> > I met the same NMI issue in L2, too.
> > L1: -cpu host  (or -cpu Haswell,+vmx)
> > L2: -cpu qemu64 by default
> > 
> > If I use '-cpu qemu64,+vmx' to create L1, I'll not meet NMI issue in L2.
> > 
> 
> That, and it looks like my guest kernel was lacking
> CONFIG_LOCKUP_DETECTOR. Will rebuild and retest later.
> 
It looks like NMI injected by L0 to L1 are mistakenly injected into L2.
Can you test this by injecting NMI into L1 via qemu monitor?

--
			Gleb.