[Sorry for all the clutter tonight.]

I found a way to make the fxp timeouts happen on Intel boxes too.

This is a full duplex point-point link. (no hubs or switches)  (The 
machines are right next to eachother.  It's just a short crossover 
jumper.) 

I'm trying to fill up both directions on that link.  I'm getting 
a total of ~190 megabits. 


Aug 13 01:29:29 hgm96b /netbsd: fxp1: device timeout
Aug 13 01:30:00 hgm96b last message repeated 6 times
Aug 13 01:32:04 hgm96b last message repeated 20 times
Aug 13 01:42:07 hgm96b last message repeated 100 times
Aug 13 01:52:10 hgm96b last message repeated 102 times
Aug 13 01:58:23 hgm96b last message repeated 61 times


Nothing on the other end.  That's slightly surprising.  The test 
is almost symetric.  It's using UDP.  The "server" side does recvfrom 
and sendto.  The "client" side does a bind then send/recv.