I've been looking at the source of the bootloader code in 1.1.  When it
asks you for the boot device, it only allows unit numbers 0 or 1. I
think this is too limiting, since I recently ran into a situation where
this is not sufficient.

I was installing NetBSD on a SCSI MO drive on a machine that also had 1
SCSI hd and 2 IDE drives installed. The Adaptec SCSI BIOS assigned drive
number 0x83 to the MO drive, and an unmodified boot loader did not allow
me to boot from this drive, because it did not accept wd(3,a) (or
variants with sd or hd) as a valid drive.

After this, I noticed that I also had to include the -r option
(RB_DFLTROOT), since the kernel wants to figure out which drive it had
booted from. Of course, the previous wd(3,a) was sd1a by now and it
guessed sd3a which obviously failed. I am wondering why this
second-guessing is taking place even though I specifically configured
the root device to be sd1a. I can understand this to be needed for for
generic kernels, but I wouln't expect this otherwise.

So, I would propose to remove the aforementioned limit on the unit
number in the boot loader and make RB_DFLTROOT a default boot option.
Both of these are trivial changes.

Discussion?

-Olaf.
--
___ Olaf 'Rhialto' Seibert      D787B44DFC896063 4CBB95A5BD1DAA96 
\X/ There are no lemurs in this post    rhialto@polder.ubc.kun.nl