> The problem in the above (Jim's) case is a bug that was present in 1.4.1,
> where partitions could end up with a off by one error (off by one sector
> for the end of the partition, to be exact)..

I believe I finally understand what you mean here.  (I know, slow.)

There is one other problem:  The geometry reported by my drive is the same as 
yours -- 16383 / 16 / 63 -- but for LBA mode this BIOS (Award 4.51PG) wants to 
use 1655 / 255 / 63, not 1024 / 255 / 63.  I don't know if the cylinder count
really makes any differnce, but....

After low-level formatting, I installed Linux, which uses the geometry the 
BIOS wants to use -- 1655 / 255 / 63 -- automatically.  I created the Linux 
partitions (with Linux cfdisk) using cylinders for units.

Then I went to install NetBSD.  I corrected fdisk to use 1655 / 255 / 63 
instead of 1024 / 255 / 63 for the BIOS geometry, created the NetBSD partition 
and the boot menu, wrote the MBR and proceeded.  After initializing the file 
systems I dropped into the handy utility shell, so I could run 'fdisk -u' to 
hand-edit the Linux partitions back to where they were _before_ the "off by 
one" error (so they end on cylinder boundaries again).  Again, I corrected 
fdisk to use 1655 / 255 / 63 as the BIOS geometry.

Once all of that was done, I rebooted to see how DOS and Linux were doing with 
it.  DOS fdisk sees all 8.4 BB it should see, and has no complaints.  Linux 
cfdisk has no complaints.  Linux fdisk complains about the NetBSD partition 
not ending on a cylinder boundary, but _not_ about it not _starting_ on a 
cylinder boundary (the CHS lines are both 1023 / 255 / 63, of course, which is 
correct and ought to be OK).


--Jim Howard  <jiho@mail.c-zone.net>