First of all, thanks for all the help I received so far...I've made
much more progress with this in a short period of time than I
expected, and I'm learning a lot along the way.

So far I've managed to do the following:
- Compile the old bootblocks with -DONEDISK to avoid the "Insert file
system floppy" prompt.  (thanks to David Sainty)
- Compile and install the new boot blocks, which also avoid this
problem (thanks to Perry E. Metzger and David Sainty)
- Compiled the libsa-based netboot code to a DOS .com file so I could
test it out (figured this one out by myself :-P)
- Compiled netboot without compression support (commented out the
line "SAMISCMAKEFLAGS= SA_USE_CREAD=yes") so that it can fit in a 32k
PROM (thanks to Matthias Drochner)

So now I have two hurdles to overcome before I can burn a chip and
boot this machine without any support on a local disk:

- What type of chip does the 3c509 use for its boot prom?  All I know
is it's 32K in size...but I can't find a part number in the manual or
on 3COM's web site.  Is this a standard chip or do I need to buy them
from 3COM?  Any suggestions for a PROM burner that will do this chip?

- When I boot diskless using netboot.com from DOS, the kernel chugs
along and configures all the devices.  Then it stops and asks me where
the root filesystem is and what type it is (it defaults to nfs).  I
type in "ep0" and it continues OK.  I tried changing my config line
from:
	config netbsd root on ? type nfs
to:
	config netbsd root on ep0 type nfs
but config gives the following error:
	tardis# config FOO
	FOO85: netbsd says root on ep0, but there's no ep0
	*** Stop.
What's the correct syntax?  Am I even barking up the right tree?
The system is -current from 5/19 if that matters.

Here's one other minor bit of flakiness I experienced.  When booting
diskless (with any bootblocks, and also with netboot.com) the kernel
boots, the system starts going multiuser, and then with no apparent
pattern the system has about a 50% chance of either coming up or
crashing to the kernel debugger.  I traced the problem to the line in
/etc/rc where it says:
	mount -a -t nfs
... At this point, all my filesystems are mounted (both / and /usr are
nfs filesystems, and /etc/rc has mounted them earlier on) so this
*should* be a no-op.  When I commented this line in /etc/rc, the
system boots perfectly all the time.
... The crash occurs at _nfs_reply+0x09 (that's from memory so don't
quote me on it).  If this is not a known problem I'll attempt to put
together a coherent bug report to send-pr.

Cheers,
entropy

--
entropy -- it's not just a good idea, it's the second law.