2.0.27 major problems #1 -- 3c59x driver.

Chris Evans (chris@ferret.lmh.ox.ac.uk)
Tue, 11 Feb 1997 14:16:41 +0000 (GMT)

AKA -- "The quest for a decent stable PCI network card and driver"

I'm at my wits end with the 8k 3c590 board we have. A cheap NE2000 clone
will replace it soon.

With a frequency of greater than about once a week, the driver spews out
"eth0: transmitter access conflict" at regular ~1 second intervals and
stops transmitting. The thing is fixable by stopping network functions
then unloading the module, reloading, and starting up the network.

I'm running 2.0.27 and v0.30c-all of the 3c59x driver. I've started seeing
this since we increased network traffic by serving a load of windows files
over samba.

Ideally the driver can be fixed/worked around.

Surely it must be possibly to temporarily do this;

When the network card driver detects an "access conflict", it knows it is
probably about to buy it. Hence, it could reset the card back into a sane
state, and return some form of "packet transmit failed -- retry" to a
higher networking layer.

As a second question, what is the board recommended for PCI 10BaseT under
Linux, with a solid, well hammered driver etc? As always, Alan's opinion
on this one would be valued.

cc: replies to me if possible, please!


PS Do 2.0.28 and 2.0.29 fix any serious OOPS-causing kernel-killing bugs?
Apart from the 2.0.29 nfs fix (we dont use nfs).