ok, i looked at the logs once more and while i thought that it occured
twice it only occured once:
Jul 17 20:22:14 europe kernel: BUG skbuff_head_cache: Poison overwritten
... this would explain why my attempts to bisect and reproduce it
failed. I got too excited about it being seemingly reproducible and
possibly bisectable (memory corruption bugs rarely are). My overnight
reboot-the-same-kernel tests didnt show anything either.
It's a known-reliable system with thousands of bootups:
Jul 17 20:20:54 europe kernel: Linux version 2.6.26-tip (mingo@europe)
(gcc version 4.2.2) #3094 SMP Thu Jul 17 20:19:27 CEST 2008
... but a hw fluke is never out of question. (It wasnt a particularly
hot day but the evening was unusually humid, maybe that made the
So lets close this for now, it's not a reproducible regression that we
can act upon. I'll update this thread if anything new happens.
Netconsole is reliable on this system in any case.