>My quick but not quite elegant hack to fix this on my machines is to
>change /etc/rc.d/network from executing rtsol once to executing it
>twice with a sleep 1 in between. Apart from fixing the network
>drivers, does any of you have a somewhat more elegant solution that
>would be acceptable to include in the NetBSD-distribution to make the
>startup sequence robust against this type of failure?

	i think the correct solution is either of them:
	- change ex driver, so that the initialization code will not return
	  to the caller until the initialization gets perfectly finished.
	- set IFF_UP | IFF_RUNNING only when initialization is perfectly
	  finished.  (similar but allows upper-layers not to hang up)

	basically ND/DAD portion is not the place to workaround here.

itojun