Somehow, both if_le.c and ether_input() are doing an ntohs()
on the ethertype of arriving packets. This doesn't work, in general.

The following fix  works for me. With it I can telnet out (using the
Ultrix binary) from a 3max. Ultrix nfs_mount can find mount handles,
and Ultrix nfs mounts should work *read-write* soon.

I don't know  what the NetBSD Official Architectural Position is on
whether it's the device-specific drivers or ether_input() who should
be byteswapping eh->ether_type. Could someone who knows for sure
make the appropriate fix to if_le.c:leread() if I guessed the wrong
way?

*** sys/net/if_ethersubr.c.DIST	Wed Jun 29 03:30:26 1994
--- sys/net/if_ethersubr.c	Tue Aug 16 18:05:20 1994
***************
*** 332,338 ****
--- 332,342 ----
  	if (m->m_flags & (M_BCAST|M_MCAST))
  		ifp->if_imcasts++;
  
+ #ifdef BRAINDAMAGE
  	etype = ntohs(eh->ether_type);
+ #else
+ 	etype = eh->ether_type;
+ #endif
  	switch (etype) {
  #ifdef INET
  	case ETHERTYPE_IP:

------------------------------------------------------------------------------