From: Boszormenyi Zoltan <zboszor@MOL.HU>
  To  : ggi-develop@eskimo.com
  Date: Wed, 1 Jul 1998 09:56:34 +0100

Re: fbcon-KGI bridge

At 08.04 1998.07.01. +0100, Paananen Osmo wrote:
>Would it be a bad idea to start using device names which would resemble more 
>Solaris device naming than Linux?  
>
>For example the device names of hard disks in linux go: sda1 sda2 sda3 and so 
>on, but in Solaris they go on: c0t0d0 c0t1d0 c0t2d0
>
>Where c = controller, t = scsi target and d = lun.  So when I add one device 
>the numbering stays same, and c0t0d0 refers to same disk unlike in Linux where 
>the mapping can change (more easily than in Solaris).

Have you tried the devfs patch yet? The naming scheme is already like what
you just described.

>I haven't been thinking this for long, so there may not be any sense in this 
>but please give it a tought.

It make sense. Consider the following case (this is my machine :) ):
I have two SCSI controllers: a Diamond Fireport 40 and a parallel port ZIP
drive. I compiled the SCSI support, SCSI disk support and the BSD NCR
driver into the kernel and the ZIP driver into module. Now the ZIP
cannot be automatically loaded. I have to say "modprobe ppa" as root
and it is not comfortable to say the least. (The same happens by not
compiling in the NCR driver and all others stay as above.)

With the device naming scheme used by devfs it would be possible to
cut all such drivers into two: one part registers the low level
(SCSI and other) drivers at compile time into a list and thus the
kernel knows which driver it should load when one want to access
/dev/sd/c1t6u0p4 (controller 1, target 6, LUN 0, partition 4) -
e.g. a DOS ZIP disk.

Zoltan Boszormenyi