lkml.org 
[lkml]   [2013]   [Mar]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: Uhhuh. NMI received for unknown reason 2c on CPU 0.
On Wed, Mar 06, 2013 at 01:19:32AM +0100, Borislav Petkov wrote:
> On Wed, Mar 06, 2013 at 01:13:23AM +0100, Rafael J. Wysocki wrote:
> > I suspected that during resume from hibernation the boot kernel (the
> > one that loaded the image) did something to hardware and the restored
> > kernel didn't handle that change properly. It is hard do say what
> > piece of hardware that was, however (it might or might not be the NIC,
> > it may be pure coincidence that the NMI messages appear in the log at
> > this point).
>
> Agreed with the second part. About the first part, who communicates what
> to whom, come to think of it, it might not be related to any devices at
> all.
>
> Here's why I think so:
>
> So one of the things I did to trigger this is boot the machine, run
> powertop and set all the knobs in the "Tunables" tab to "Good". One of
> the tunables is turn-off-nmi-watchdog something which turns off the
> watchdog which is using the perf infrastructure which generates NMIs
> when the counter overflows.
>
> Now, imagine I do that in the "normal" kernel, then suspend,
> ...<something happens or does not happen>, then resume back into the
> normal kernel and it somehow "forgets" the fact that we disabled the NMI
> watchdog before the suspend cycle. And boom, it gets a single spurious
> NMI.
>
> Does it make sense? I dunno - I'm just connecting the dots here between
> the observation points which are most likely.
>
> Anyway, it's getting late, good night. :)

Exactly as I thought: so I'm running the machine with NMI watchdog
enabled, i.e. powertop says:


PowerTOP v2.0 Overview Idle stats Frequency stats Device stats Tunables

>> Bad NMI watchdog should be turned off
Good VM writeback timeout
....

and no more spurious NMIs.

I'd say the plot thickens: disabling NMIs and suspending to disk right
afterwards doesn't seem to really disable the watchdog. Or the disable
gets delayed leading to one last spurious NMI when resuming... I
probably need to go stare at the code though...

--
Regards/Gruss,
Boris.

Sent from a fat crate under my desk. Formatting is fine.
--


\
 
 \ /
  Last update: 2013-03-08 18:21    [W:0.063 / U:0.972 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site