Re: continuous backtrace ... ?

[Andreas Dilger <adilger@clusterfs.com>]

On Jun 16, 2003  12:15 -0700, Robert White wrote:
> I have a 2.5.70 kernel image which, at boot, goes into a continuously
> looping backtrace.  It scrolls up the screen too fast and continuously to
> read but it is clearly a backtrace. If I just relax and watch I can see the
> [bracketed hex] and routine name shape of it whizzing by.
> 
> This happens while the file systems are read only and the system doesn't
> respond to the keyboard at all.
> 
> Is there any good way to capture this stream of data in any kind of useful
> way?  or at least pause the spew so that I can find the general locus of the
> problem?
> 
> Given the immediacy of the problem (it appears just-about-concurrently with
> attempt to draw the boot logo) I suspect that it has something to do with
> the console driver (radeon frame buffer VGA console in normal mode)
> interacting with my All-In-Wonder 9700 pro and the very new P4-3ghz in
> hyperthreading mode.

Probably a stack overflow.  Once you hit the do_IRQ() stack overflow
detection code, you get stuck in a loop because your console output is
slow enough that you immediately get another IRQ, rinse, repeat.

I fixed this locally by adding a static jiffies counter to rate-limit
the overflow stack dump to once per 5 seconds.  If you overflow your
stack more often than that, well too bad.  At least you have a chance
to look at your console instead of the machine essentially being wedged.

Cheers, Andreas
--
Andreas Dilger
http://sourceforge.net/projects/ext2resize/
http://www-mddsp.enel.ucalgary.ca/People/adilger/