First off, the free HP stuff:  I have an HP320 available to anyone who
wants it.  The frame buffer may be bad, and the LAN card may be bad, or
they may both be good.  Power supply and case seem alright!  ;-)  Memory is
8 meg, and may be good.  I'm keeping the CPU+FPU for my class next
semester.  Also available is a tape drive (32 track?) that ate the only
tape I tried in it.  Feel free to forward this to other NetBSDers.  I'll
pay shipping to core team members, or HP port developers + I'm in the SF
Bay Area now.

> From: "Charles M. Hannum" <mycroft@gnu.ai.mit.edu>
> I don't know about the other problems you mentioned, but why are you
> offsetting the `c' partition?  Only the i386 port does something
> non-standard with the `c' partition; on all other ports, it's supposed
> to be the whole disk.

Aha!  This was it.  My guess is that the HP looks at the beginning of the
disk for the disklabel + found none, because I was putting it at offset a.
If I disklabeled it, it knew where it put it, but on reboot, it looks only
at the beginning of the disk.  I changed that, and it works great now.  (Do
we offset partition a to avoid clobbering the boostrap code?)

Does anyone know if the following is correct:
HP:
|------------d----------------|
|------------c----------------|
B&D|--a---||--b---||----e-----|

i386:
|------------d-----------------------------|
P  |---------c----------------||-other-os--|
   |--a---||--b---||----e-----|
B=BSD Boot Blocks
D=BSD Disk Label
P=PC BIOS (DOS) Partition table

d=whole disk always  c=part used by BSD (on hp = whole disk, on i386, we
need to leave room for PC BIOS partition table that it looks for)

Okay, so where do the B(oot sectors) and D(isk label) go on the i386?
If it put them at the beginning of d, it would clobber the DOS partition
table and boot loader, and if it put them at the beginning of c, it would
clobber BSD partition a.  Hm.

> From: mike@cs.utah.edu (Mike Hibler)
> You probably need to specify "-r" when doing the disklabel command.
> That ensures that you change the on-disk version of the label and not
> just the in-core version.

The disklabel -w -B flags make it do a -r too.  I didn't believe the man
page either, so I tried -w -r -B as well!  ;-)

Thanks, all for your help.

Mark