Background: IBM ThinkPad 600 with IBM CardBus EtherJet + 56K modem in 10BT mode. Redhat Linux 6.0 with Kernel rev 2.2.12 (see below). Pcmcia-cs-3.1. (again see below). /var/log/message data is uninteresting (or at least unchanged from the posts on hyper.stanford.
Under Linux kernel rev 2.2.12 and pcmcia-cs-3.1.3 things went downhill. I couldn't ping anything, all I saw were destination unreachable messages from ping.
Figuring things were generally better but tulip_cb.c was a little worse, I took my modified source from 3.1.2 (tulip_cb.c v0.91g, patch at http://n2kiq.fdns.net/tulip_cb.c_91-91g.patch UNOFFICIAL combination of tulip_cb.c v0.91 and Donald Becker's tulip.c v0.91g experimental version.)
After re-compiling things got better, actually a fair bit better as I can ifconfig the interface down and back up and things work again. While searching for another update to the 0.91 series, I found that Mr. Becker created a tulip debugging program that dumped information about the card's state.
Running that at various times gives interesting information. When things are ok it gives the following status:
When things are bad (i.e. getting no response at all from ping) it gives this:
Doing the ifconfig down/up resets the RX process and things continue, for a while. It also says that the EEPROM is #6 and in an old style EEPROM layout, which may cause some features to be unavailable.
I've captured packets on another machine and the stuff the driver sends appears to be correct, it didn't say anything about corrupted packets. The stuff it did see was sometimes out of sequence, which I thought was strange.