>I think the DAD code is broken anyway...  It doesn't make certain the
>query packet is actually sent, for instance.
>On the laptop I have, ne2 times out for the first few packets sent,
>then recovers and works normally.  DAD reports "no duplicates found"
>long before it starts to reliably transmit/receive...

	I agree that the current code does not work with drivers that takes
	too much time for initialization.  This is because, with the
	current code, upper-layer cannot be sure when the driver gets
	initialized perfectly.  

	DAD code tries to guess that by looking at IFF_UP and IFF_RUNNING,
	and add some delay before sending packets (based on RFC2462 5.4.2,
	last paragraph).  If you know any other way to know when the driver
	is completely initialized, please let me know.

itojun