Subject: Re: Another changer, another changer problem
To: NetBSD-current Discussion List <current-users@netbsd.org>
From: Greg A. Woods <woods@most.weird.com>
List: current-users
Date: 10/07/1998 23:39:28
[ On Wed, October 7, 1998 at 17:29:01 (-0700), Curt Sampson wrote: ]
> Subject: Re: Another changer, another changer problem
>
> Ok, so I've got a pair of NCR 53c810 SCSI controllers in slots on
> my PCI bus, at devices 9 and 11. They probe as scsibus0 and scsibus1.
> Now how do you deal with me putting in a third controller in a
> random slot that just happens to map to device 10?
> 
> If you probe in slot order, you've just moved my second controller
> from scsibus1 to scsibus2.

Damn!  You're right, of course.

The scsibusN assignment must be done on the basis of whatever physical
identification there is.  For hardware using slot assignment that means
you have to configure the kernel as if there could be a scsi controller
in every slot, thus if you use "pseudo-device scscibus N" then N must be
equal to the maximum number of slots in the machine and the mapping of
controller assignments to scsibusN must be done using the "slot" number.

This will of course still work OK for "small" machines such as PCI based
systems with only one or two PCI buses, or small Sbus machines, etc.,
but we're right back to needing devfs for bigger servers and as der
Mouse points out we'll also need sparse tables in a couple other places
too.

-- 
							Greg A. Woods

+1 416 218-0098      VE3TCP      <gwoods@acm.org>      <robohack!woods>
Planix, Inc. <woods@planix.com>; Secrets of the Weird <woods@weird.com>