I have been having problems with my Lenovo T61 and Rawhide. Currently the system will do all kinds of watchdog resets when I am booting the 3.0 kernels but not with the 2.6.38 kernel. I am trying to figure out how to best capture and debug them as they tend to scroll off the screen faster than I can catch anything. For the rc4 kernels I think it is dieing when trying to remount the root disk but on the rc5 kernel it gets past that and crashes during nfs or smartd. The funny thing is.. when I installed kdump and put in 256MB for a kernel save.. the problem goes away on reboot.
Does turning on kdump and installing debuginfo kernel stuff change things in a way that a watchdog oops won't happen?
I have been having problems with my Lenovo T61 and Rawhide. Currently the system will do all kinds of watchdog resets when I am booting the 3.0 kernels but not with the 2.6.38 kernel. I am trying to figure out how to best capture and debug them as they tend to scroll off the screen faster than I can catch anything.
Does serial console (Documentation/serial-console.txt) work over the modem port of a T61? You'll need a null modem cable and another machine running minicom. What about netconsole (Documentation/network/netconsole.txt)? What about Documentation/sysrq.txt ? Those are the lowest-level techniques for getting info out of a kernel.
--
John Reiser jreiser@bitwagon.com wrote:
Does serial console (Documentation/serial-console.txt) work over the modem port of a T61?
Isn't the modem port actually a soundcard channel on modern laptops?
David
On Thu, Jun 30, 2011 at 10:17 PM, Stephen John Smoogen smooge@gmail.com wrote:
Does turning on kdump and installing debuginfo kernel stuff change things in a way that a watchdog oops won't happen?
I doubt the debuginfo package install has anything to do with it, as that is really only used by things like crash (and maybe perf?). Enabling kdump does cause some changes though, as the kernel will reserve a section of memory to put the kdump kernel in. It's plausible that your machine is tripping on some memory issues and enabling kdump is forcing the kernel to not touch that memory at runtime. It's something of a long shot but it's plausible.
If you haven't already, try running memtest86+ for a while and see if it comes back with any errors.
josh
On Fri, Jul 1, 2011 at 04:37, Josh Boyer jwboyer@gmail.com wrote:
On Thu, Jun 30, 2011 at 10:17 PM, Stephen John Smoogen smooge@gmail.com wrote:
Does turning on kdump and installing debuginfo kernel stuff change things in a way that a watchdog oops won't happen?
I doubt the debuginfo package install has anything to do with it, as that is really only used by things like crash (and maybe perf?). Enabling kdump does cause some changes though, as the kernel will reserve a section of memory to put the kdump kernel in. It's plausible that your machine is tripping on some memory issues and enabling kdump is forcing the kernel to not touch that memory at runtime. It's something of a long shot but it's plausible.
I am going to try the memtest. The system this morning was incredibly sluggish with a yum install taking several times longer than in the past. A reboot of the system this morning had it oops a lot on agetty versus working.. but it locked up solid so a power reboot was needed so I couldn't get anything 'saved' from ram. sigh.
Will run memtest86 for the day and see if that gets anything. I had tested the system earlier on my rawhide experience with the IBM maintenance tools but they may miss something.
If you haven't already, try running memtest86+ for a while and see if it comes back with any errors.
josh
devel mailing list devel@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/devel
On Fri, Jul 1, 2011 at 10:44, Stephen John Smoogen smooge@gmail.com wrote:
On Fri, Jul 1, 2011 at 04:37, Josh Boyer jwboyer@gmail.com wrote:
On Thu, Jun 30, 2011 at 10:17 PM, Stephen John Smoogen smooge@gmail.com wrote:
Does turning on kdump and installing debuginfo kernel stuff change things in a way that a watchdog oops won't happen?
I doubt the debuginfo package install has anything to do with it, as that is really only used by things like crash (and maybe perf?). Enabling kdump does cause some changes though, as the kernel will reserve a section of memory to put the kdump kernel in. It's plausible that your machine is tripping on some memory issues and enabling kdump is forcing the kernel to not touch that memory at runtime. It's something of a long shot but it's plausible.
I am going to try the memtest. The system this morning was incredibly sluggish with a yum install taking several times longer than in the past. A reboot of the system this morning had it oops a lot on agetty versus working.. but it locked up solid so a power reboot was needed so I couldn't get anything 'saved' from ram. sigh.
Will run memtest86 for the day and see if that gets anything. I had tested the system earlier on my rawhide experience with the IBM maintenance tools but they may miss something.
So ran memtest86+ for the morning, and nothing showed up. Rebooted into 3.0 and it watchdog halted on me in a way that FN-CNTRL-SHIFT-SYSRQ-B didn't do anything. Power cycling caused it to reboot into another watchdog issue so I went into single user mode. There I was able to turn off watchdog resets and exit (this was not meant to be a solution as much to see what would be causing stuff).
What seems to happen is I get slower and slower accesses of some sort to disks. [Sorry for the wishy-washy wording I need to figure out how to better debug this for you guys.] System came up and I logged in after 5 minutes. Doing a yum update took about 4 seconds to get to the (y/N) part and then afterwords sat there and then did a long set of disk accesses. Doing it again took even longer with long pauses. My guess is something is sitting in the CPU and the watchdog saw nothing happening for 10 seconds and whacked. On the third time the system just sat there and never returned (keyboard doesn't seem to allow for a reset so I don't know what is going on.)
At the moment, the system is usable with the 2.6.38 from Fedora-15 (though systemd will not allow a reboot due to constantly trying to restart systemd-kmsg something) Will look into getting a modem null cable though I may just go for a picture of the watchdog issue.
-
David Howells -
John Reiser -
Josh Boyer -
Stephen John Smoogen