> I'm much rather be able to get the "real" (reported) geometry AND the
> BIOS geometry.

I believe this was clear in my 2nd draft.  If (and *only* if) we
change the boot block as I suggested, then the disklabel geometry is
only really used to calculate the CHS values for a partition in
fdisk(8).  I suppose we could change fdisk(8) to use some other
mechanism to get the geometry, but that seems pretty hoaky.

> I have heard that using some BIOS geometries can
> force FFS to use 4k frags/32k blocks.

Only if you make the cylinder groups ridiculously large.  This can be
tweaked.  I don't believe it's an issue.

> If we could reliably get the BIOS geometry, we'd never have to bother
> the user with geometry questions unless they wanted to know.

Uh, this makes me wonder if you've actually been paying attention to
the discussion.  There is *NO* `reliable' way to map a BIOS device to
a NetBSD device without writing to the disk.  And writing to the disk
during boot really isn't acceptable, for a large number of obvious
reasons.

So, `reliably get[ting] the BIOS geometry' isn't an option.