> I think there is an evil timing bug lurking somewhere.  I have 4 identical
> machines (ok, 2 with 300A celeron, 2 with 333A celeron) with Compaq branded
> fxp boards.  One (possibly two, can't remember) of them would lose link with
> my switch as soon as NetBSD probed the card.  Multiple reboots and kernels
> later it is fine.  I went back to the old kernels and seems fine.  So I think
> there is some kind of timing issue or race condition in the fxp driver or the
> phy driver.  I haven't had trouble this week though. :)  I will try to test
> it out some more and figure out if it is only on the machines with the same
> CPU speed, or anything else that might be a pattern.  It was happening every
> single reboot for a while.

Even better... I have an fxp under a machine that was running 1.5.
It was fine and dandy. Go to 1.5.2, and it suddenly can't autonegotiate
worth a damn.