Skip to contentSkip to conversation

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

DISCUSSION Brad Barnett 1 participant 2 messages Nov 13, 2006 – Nov 27, 2006

lore
Share & tools
lore
#0 lore raw

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

#1 lore raw
Quoted 25 lines · On Mon, 13 Nov 2006 16:45:07 -0500 Brad Barnett <[email protected]> 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.

Quoted 9 lines

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.