System locks up with RealTek 8139 and kernel 2.2.20
I've seen this problem referred to in the archives of this newsgroup,
but no clear resolution seems to have emerged. I'm running potato on
2.2.20 on about 5 servers connected via RealTek 8139s to the end of
DSL connection. The loads are nothing extraordinary, but I seem to be
having this lockup problem where all I get on the console is:
kmem_alloc: Bad slab magic (corrupt) (name=size -32)
eth1: Memory squeeze, deferring packet
and an absolutely frozen server. I would love to include stuff from
/var/log/syslog (or other log files) but it seems that the machine
locks up before syslog has a chance to write anything.
After reboot, the problem seems to repeat itself every 1-3 days. The
curious thing is that it usually takes down 1-3 servers (of the total
5) at a time. Note that I have other servers with tulip-based cards
that don't seem affected.
Here are the solutions that I am contemplating:
1. replace all the realtek 8139 cards; since each host is multihomed,
this implies replacing 8 cards. Does anyone have a suggestion as to a
rock-solid card for linux in the similar price range? I used to use
netgears before they dropped the tulip support...
2. move to 2.4 and hope this solves the problem. I have already gone
from 2.2.17 to 2.2.20 to try to fix it, but maybe the move to 2.4 will
be more significant.
Any help would be greatly appreciated...
For the previous discussion on this: