> Could you try the 2.6.7 code from XFS CVS for me? (with
> SLAB_DEBUG _enabled_). I think this may be fixed by the
> XFS buffer undelay fix recently, but I couldn't reproduce
> this problem initially so I'm not 100% sure (and I would
> really like to know :) -- thanks).
It is done but... Oops is still there - it occurs in shorter
period of time than on 2.6.6. run-parts /etc/cron.daily
gives here more than fifty percent chance of hitting this.
One time I also caught that before init managed to run gettys...
However - slab corruption messages did not show up (I have kdb
enabled here, so maybe that is the reason).
Also, some past observations lead to statement that disabling
CONFIG_PREEMPT reduces chance of hitting oops from minutes to
hours... but oops occurs in the very same form as with CONFIG_PREEMPT.