Ok, something happend to the tulip driver in the recent testN kernels.
I haven't found a reason why it happens and I can't easily reproduce it
but what happens is noted on the subject line.

I have three machines now that do this.  It's unfixable until reboot (I
don't have these as modules).

I have a suspicion that it's triggered by large amounts of traffic.  My
brother feels he is able to trigger it repeatedly by doing big transfers
between a w9x box and a linux box.

-d

--
"The difference between 'involvement' and 'commitment' is like an
eggs-and-ham breakfast: the chicken was 'involved' - the pig was
'committed'."