> I have tried netbsd-GENERIC kernel again after your suggestion. It now
> boots fine, but it runs out of DMA receive buffers immediately.

good.  I now think it's probably just too much broadcast traffic which
doesn't get handled fast enough.

> Then I have tried your new SCSI kernel. I can't get it to boot with this
> one - it stops when it's doing DHCP/BOOTP request. I have compared
> network traffic against netbsd-GENERIC kernel. With generic one, I see
> two bootp requests - first one is from NeXT boot monitor, the second
> one - after about 40s - is from netbsd. But I only see one request with
> your SCSI kernel so it seems it's not sending anything out.

hmm, there should actually be three bootp requests:
- 1st from the NeXT rom monitor to load the boot program
- 2nd from the boot program to load the kernel
- 3rd from the kernel to mount root via nfs (after about 40s)
So it fails when the kernel tries to mount root?  When I tried this earlier,
this also happened here once but it worked when I made it try again (just
hit return a couple of times when it asks for root/dump devices/types and
such).  tcpdump on the server only showed the bootp request which worked.
If you're using twisted pair to connect the machine, then another
possibility would be the new code to select the ethernet media.  I've built
a kernel with the media selection disabled:
ftp://lola.pin.lu:21/pub/NetBSD/arch/next68k/test/SCSI-GORAN.gz
I've also added a printf, if you use twisted pair it should read "media: 90
d0" after the nfs_boot: trying DHCP/BOOTP line.

     christian