In /proc/sys/kernel you'll find the file "panic" and, starting
with 2.6 kernels, "panic_on_oops".
The "panic" file controls whether or not the kernel will attempt
to reboot after a panic. If "panic" is zero, it will just sit forever
waiting for you to do something. Obviously that's not good for an
unattended machine or a machine that is difficult to get to, or perhaps
a machine with no monitor. In those cases, you probably want to set
"panic" to some non-zero value. For example, setting it to
"30" means the kernel will reboot after 30 seconds. Assuming the
problem was transitory, you are back up and running. If not, well,
when you eventually get in front of the thing, you'll be able to
see the panic messages for 30 seconds. Remember that you need to
rewrite on each boot.
It might be interesting to write a little script that checks
the time it last wrote to "panic" and increments the amount if
it was recent.. thereby increasing the time between reboots in
the even the problem does not go away.
The "panic_on_oops" file is normally 0, but if set to 1, the
kernel will delay a bit before reponding to the panic, thereby
hopefully giving klogd a chance to write out what it knows about the