After hearing that sharks vastly prefer DHCP to BOOTP, I set up dhcpd on a
stable i386/1.4.1 machine and tried once again to NFS boot my $30 swap-meet
shark.

Here is the dhcpd.conf, a little weird since this server is on both my ISDN
line and also on my internal 10. network, but dhcpd complains if I do not
have a subnet entry for every active interface. (I'm not doing packet
forwarding of any kind on this machine, in case it matters.)

"obi-wan" (10.0.2.1) is the shark.
"toddpw-gw" (10.0.0.1) is the relevant interface of the DHCP server.

subnet 10.0.0.0 netmask 255.0.0.0 {
}
subnet 147.11.31.136 netmask 255.255.255.248 {
}

host obi-wan {
	hardware ethernet 08:00:2b:81:63:47;
	fixed-address obi-wan;
	filename "/tftpboot/netbsd_netboot";
	option domain-name "toddpw.net";
	option root-path "/data/export/obi-wan/root/";
#	option boot-size 1234;
#	option host-name "obi-wan";
#	server-name "toddpw-gw";
#	always-reply-rfc1048 true;
#	option subnet-mask 255.0.0.0;
#	option broadcast-address 10.255.255.255;
#	option routers toddpw-gw;
#	option log-servers toddpw-gw;
}

The shark is able to load netbsd_netboot fine (it could with bootpd also),
but once booted it gets less happy. Here's the suspicious dmesg excerpt:

...
cs0 at ofisa0 (ethernet@i300): CS8900
cs0: CS8900 rev. F, address 08:00:2b:81:63:47, media UTP
cs0: driver only supports memory mode
...
wdc0 at ofisa0 (ide@i1f0): error getting interrupt data
...
boot device: <unknown>
nfs_boot: 'cs0' not found
cannot mount root, error = 6
root device (default cs0):

This happens with 1.4, 1.4.1, and -current from a couple days ago.

Due to the dubious origins of this shark, there is a very real possibility
that it is damaged, however the innards of the thing look great (although
the outside is scratched up a bit). I am hoping that I'm just lacking some
magic incantation which is not in a FAQ yet...

Todd Whitesel
toddpw @ best.com