Version 0.40 with 3c900 sporadically spaces out

[William Pietri]

I'm using Red Hat Linux 4.2 on a box that has, until recently, been very
well behaved. I just moved it to a new location, though, and every few
hours the ethernet interface seems to go wonky and refuse to talk to the
outside world anymore. I've tried a couple of different cables and
different hubs in an effort to fix the problem. Things are complicated by
the fact that the box resides in a colocated cabinet 2000 miles away from
me.

When I boot up, the driver has this to say:

kernel: 3c900.c:v0.40 4/16/97 becker@cesdis.gsfc.nasa.gov
kernel: loading device 'eth0'...
kernel: eth0: 3Com 3c900 Boomerang 10baseT at 0x1400, 00:60:08:3d:87:25,
IRQ 11
kernel:   8K word-wide RAM 3:5 Rx:Tx split, autoselect/10baseT interface.
kernel:   Rx Pacing bug exists, disabling bus-master receives.
kernel: eth0: Overriding PCI latency timer (CFLT) setting of 64, new value
is 248.

Things work fine for anywhere from a few minutes to a few hours, and then I
get something like this in the logs:

kernel: eth0: transmit timed out, tx_status 88 status e000.
kernel: eth0: Transmitter encountered 16 collisions -- network network
cable problem?
kernel:   Flags; bus-master 0, full 0; dirty 0 current 0.
kernel:   Down list 00000000 vs. 01a70a10.
kernel:   0: 01a70a10  length 00000000 status 00000000
kernel:   1: 01a70a20  length 00000000 status 00000000
kernel:   2: 01a70a30  length 00000000 status 00000000

And then I get a lot of transmit timed out errors from then on. It seems to
be at this point that the interface stops working. Rebooting always fixes
the problem, and nothing else seems to do much. Of course, it's hard to
tell, as working on it remotely is a bit tricky.


I can think of several reasons that this might be happening. The driver
version is old, so I could try upgrading. The Red Hat hardware
compatability list mutters darkly about this particular model of card, so
replacing the card might be in order. Or, there could be some sort of
ethernet problem that the other (eight or nine) computers on this segment
don't notice. I could even believe that something (configuration or
hardware) got goofed up in the move. 

If nobody has seen this before, I'll just start trying things in roughly
this order. But I'd love to hear suggestions from anybody who has seen this
before!


Many thanks,

William