> On Mar 3, Ken Harrenstien wrote
> > Hi,
> > 
> > I tried to install 1.3.3 via FTP and gave up after several hours of
> > frustration.
> > 
> > I did read the entire INSTALL document, pulled over the boot floppy
> > image, took off all drives except a fresh virginal 2.1G drive, and
> > did my best, but:
> > [...]
> 
> This may be a problem with how NetBSD handles the BIOS geometry:
> it gets it from the MBR, and not from the BIOS (should be fixed for
> 1.4). So, with a virgin disk it may get wrong geometries values.
> The following ususally solves the problem:
> boot from a dos floppy with the fdisk utilitie. Run 'fdisk /mbr'; then
> fdisk to create a partition (which may be erased later by the NetBSD install
> if you want, the goal here is just to get fdisk to write a partition table).
> Then reboot with the NetBSD install floppy, it should now get the BIOS
> geometry rigth.

This worked!  I temporarily moved the disk to another system where I
could clear it by overwriting the first hundred or so sectors with zeros,
then tried the NetBSD install again.  This time it told me it couldn't
find the MBR and suggested either aborting to create one first, or
entering the geometry by hand.  I aborted, found a DOS "fdisk" that
understood the /MBR switch (the first one I tried didn't), and tried
again.  This time it found a geometry and all went well.

In particular, the partition edit program finally worked correctly.

My conclusion is that the original disk must have had some pre-existing
data on it that was being misinterpreted as a MBR.  I don't know enough
to know how easy it would be to detect or guard against this kind of
thing, but clearly there was not a lot of sanity checking going on.  I
certainly recommend that your response above should at least be put
into the doc as an example of how to recover from this fix.

One other suggestion for improvement: the screen about "real" versus
"fake" geometries needs to be re-worded and made consistent with the
doc (which isn't bad).  In retrospect it is talking about the fact that
the SCSI-reported geometry doesn't add up to the SCSI-reported total
number of sectors, as is typical for ZBR disks.  This has nothing to do
with the MBR or BIOS but that isn't clear.

Furthermore the use of the word "real" is very confusing because in the
doc it means "the geometry NetBSD uses" which hasn't yet been decided!
I had to struggle with alternative theories before deciding that what
it really meant on this screen was "as reported from the drive"; i.e.
not "as reported from BIOS" or "as reported from the MBR" or "NetBSD
real geometry".

Oh well...  hopefully these comments will help make things better for
the next hapless newcomer.  Thanks again for getting me unstuck!

--Ken