The crash in the title occurs on my workstation under very well-defined
circumstances: ~4-5 minutes of constant load average 1, in my case an
l3enc reading and writing to a SCSI disk, everything else idle (load
average is really 1.00). The l3enc process is niced 19, although I don't
think this makes any difference - I believe the last time time it crashed
it was running with 0 niceness.
What happens: all of a sudden the kernel starts spewing out back-to-back
oopses to the console, very fast. They seem to be followed by a _long_
stack dump, most of which appears to be garbage (again, it all scrolls up
very fast). Nothing is logged, since user space programs never get a
chance to run again, yet the software watchdog does kick in after a minute
so at least the timers are still functional inside the kernel.
I suspect some stack overflow, but at this point it's a simple
supposition. The machine is otherwise stable and can stay up for weeks
without problems. Also, I had it do exactly the same thing (l3enc) under
2.0.30 and 2.0.31-pre2-davem and it was all fine and dandy by that time.
Configuration: P5/166, Triton II with IDE busmastering DMA enabled, one
IDE drive (main system drive), one wide-SCSI drive (swap) and one
narrow-SCSI drive (that l3enc was working on) attached to an AHA2740UW, a
tulip 21041 (Kingston) 10MB attached to a light-traffic network.
Custom kernel patches: abort code for AIC7xxx, pentium clock frequency,
dmsdos (not used, compiled as module), my own /proc/swaps and a few other
really small and irrelevant patches. The whole thing is compiled with pgcc
snapshot 970924.
I'll probably try a few more things, such as to attach a serial console to
catch the oops output, run l3enc without nice, recompile the kernel with
regular gcc 2.7.2.. and I'll see if it makes any difference.
OTOH, if somebody has other ideas/suggestions, or wants my .config, feel
free to send email..
Thanks,
Ionut
--
It is better to keep your mouth shut and be thought a fool,
than to open it and remove all doubt.