I'm tempted to incriminate the disk/filesystem code, since
- it always happens in times of high disk activity,
- even if I wait for 15 minutes, when I have errors during the following
fsck while I reboot, they are always related to the temporary file that
shoudl have been used (the errors being the classical zero dtime and
sometimes a bad blocksize).
I also seem to remember that I had freezes in the late 2.1.3x and 2.1.4x
kernels when doing a rm -rf on big directories (kernel source actually, it
looked like self-preservation instinct at the time).
>From what I've been able to test, I'd say the only things that still work
are VT switching and (num|caps|scroll)lock. There is no disk activity, no
screen activity (except for the blinking cursor, but I believe it's only
controlled by the videocard), and ctrl-alt-del does not work.
I'm giving up, but I'm still willing to test patches if someone has an
idea.
Mat