Subject: Re: Config ...
To: C Kane <ckane@best.com>
From: Todd Vierling <tv@pobox.com>
List: port-i386
Date: 08/21/1998 19:24:17
On Fri, 21 Aug 1998, C Kane wrote:

: Something I've long disliked about NetBSD is the numbering
: scheme used to name disks (and tapes and many other peripherals).
: 
: Why is the first disk called "sd0" and the second disk "sd1"
: regardless of SCSI ID or controller?  I think even a new naive 
: user would understand that a disk on SCSI ID 4 is called sd4.

Because I have three SCSI controllers in one machine for speed, and don't
want to call ID 0, LUN 0 on the third controller "sd256"...?

The auto-detection is much more compact in a system which has static /dev
nodes.  If, of course, we had a _dynamic_ /dev system, a more specific
device naming scheme could be used.  But putting a bunch of controller-based
(and remember, LUN-based) nodes on a disk when you already have 8 per for
partitions is just way too much.

2 controllers * 16 IDs * 8 LUNs * 8 partitions * 2 types of nodes [chr&blk]
 = 32768 nodes just to hold two SCSI controllers.

You'll note that the GENERIC_SCSI3 kernel on the sparc port does some
hardwiring to simulate the ID's on SunOS (sd0 -> scsibus0/ID3 ... sd3 ->
scsibus0/ID0), but it's intended only for that compatibility.

-- 
-- Todd Vierling (Personal tv@pobox.com; Bus. todd_vierling@xn.xerox.com)