root NFS causes solid lockup, serial console no help, watchdog NMI won't function on Dual AMD system

root NFS causes solid lockup, serial console no help, watchdog NMI won't function on Dual AMD system

[Brad Barnett]

Hello,

I have a couple of dual AMD Opteron system with differing motherboards. 
On both of these differing pieces of hardware, I can not enable the NMI
watchdog timer, with nmi_watchdog=1.  I constantly get:

testing NMI watchdog ... CPU#0: NMI appears to be stuck (0->0)!

This happens when booting with both Debian's packaged kernels, 2.6.17-2
and 2.6.18-2.  According to this page:

http://www.mjmwired.net/kernel/Documentation/nmi_watchdog.txt

nmi_watchdog=2 does not work with AMD systems at this time.

Both boxes, aside from this, boot fine.  They use nfsroot, booting the
kernel off of a local drive, and switching to nfsroot during the bootup. 
They work flawlessly for anywhere from a week to a few days, then lock up
solid.  I can not seem to reproduce this problem unless I am using
nfsroot, and the above kernels seem fine without issue when booting and
running off of a local drive.

I can not get any information from a serial console, and just for fun I
redirected syslog to another box, yet I did not see anything about the
impending doom to come.  The console itself is simply a blank screen, and
the keyboard is dead... 

Does anyone have any suggestions for tracking this bug down?

Thanks

Re: root NFS causes solid lockup, serial console no help, watchdog NMI won't function on Dual AMD system

[Brad Barnett]

On Mon, 13 Nov 2006 16:45:07 -0500
Brad Barnett <lists@l8r.net> wrote:

> 
> 
> Hello,
> 
> I have a couple of dual AMD Opteron system with differing motherboards. 
> On both of these differing pieces of hardware, I can not enable the NMI
> watchdog timer, with nmi_watchdog=1.  I constantly get:
> 
> testing NMI watchdog ... CPU#0: NMI appears to be stuck (0->0)!
> 
> This happens when booting with both Debian's packaged kernels, 2.6.17-2
> and 2.6.18-2.  According to this page:
> 
> http://www.mjmwired.net/kernel/Documentation/nmi_watchdog.txt
> 
> nmi_watchdog=2 does not work with AMD systems at this time.
> 
> Both boxes, aside from this, boot fine.  They use nfsroot, booting the
> kernel off of a local drive, and switching to nfsroot during the bootup.
> 
> They work flawlessly for anywhere from a week to a few days, then lock
> up solid.  I can not seem to reproduce this problem unless I am using
> nfsroot, and the above kernels seem fine without issue when booting and
> running off of a local drive.
> 
> I can not get any information from a serial console, and just for fun I
> redirected syslog to another box, yet I did not see anything about the
> impending doom to come.  The console itself is simply a blank screen,
> and the keyboard is dead... 
> 
> Does anyone have any suggestions for tracking this bug down?
> 
> Thanks
> 


An update.

I just had a reoccurrence of this bug, without using NFS root.  This is
under Debian's packaged 2.6.17-2, and the same issue occurred with
2.6.18-2.  Total and complete system lockup.  Serial console is fruitless.

I have been forced to downgrade to 2.6.9.  Does anyone out there, have any
ideas for debugging this complete lockup in 2.6.17/2.6.18?  It seems to be
NFS activity related, but I can not verify this at this point in time.