Frank van der Linden writes:
> 
> Judging from what other people are seeing, it's a geometry problem.
> The output that you show confirms this. What you need to do is
> find out what geometry your controller's BIOS really uses. You
> might be able to do that in the BIOS's setup, or you might
> need to boot a DOS floppy and run a utility like 'pfdisk'
> that will print the geometry for you.
 
I was forced to use pfdisk.  SCSI disks have proven to be a lot
harder to size than IDE disks, which is what I'm used to installing
NetBSD on.

> Remember this geometry. Then, do a re-install. At the point where
> it asks if the detected geometry is OK (the first time it talks
> about geometry), choose 'no', and specify the geometry that you
> remembered. Things should work then.

And they did, thank you Frank and all the other people who pointed
out the inconsistancies in the output from fdisk to me.  Here's
what I get now:

  ******* Working on device /dev/rsd0d *******
  parameters extracted from in-core disklabel are:
  cylinders=3117 heads=4 sectors/track=83 (332 sectors/cylinder)
  
  parameters to be used for BIOS calculations are:
  cylinders=511 heads=64 sectors/track=32 (2048 sectors/cylinder)
  
  Information from DOS bootblock is:
  0: sysid 0 (unused)
     [...]
  3: sysid 165 (NetBSD or FreeBSD or 386BSD)
      start 32, size 1046496 (510 MB), flag 0x80
          beg: cylinder    0, head   1, sector  1
          end: cylinder  510, head  63, sector 32


This all boils down to RTFM and mea culpa!  I'm sorry to have wasted
more list bandwidth.  I can only say in my defense that I was *sure*
that the INSTALL doc implied that if I was installing NetBSD onto
the whole drive that it wouldn't matter if I used the drive's physical
geometry rather than the LBA geometry.  But clearly I didn't take
into account the foibles of the BIOS boot code.  Let that be a
lesson to me.

-bmw