[3327] in linux-scsi channel archive
Re: SCSI host numbering (patch)
daemon@ATHENA.MIT.EDU (Andrzej Krzysztofowicz)
Sat Feb 14 10:46:58 1998
From: Andrzej Krzysztofowicz <ankry@green.mif.pg.gda.pl>
To: Dario_Ballabio@milano.europe.dg.com (Dario_Ballabio)
Date: Sat, 14 Feb 1998 16:42:00 +0100 (CET)
Cc: linux-scsi@vger.rutgers.edu, rgooch@atnf.CSIRO.AU
In-Reply-To: <199802141429.PAA18915@milano.europe.dg.com> from "Dario_Ballabio" at Feb 14, 98 03:29:33 pm
>
> Since I have systems with multiple omogeneous scsi boards (either
> Ultrastor or DPT), it is a very valuable feature for my driver
> development to test the drivers as modules, but to have the
> system running I must have the driver in the kernel as well
> because all the scsi boards are from the same brand and I do not
> use IDE. Anyway if the driver is properly written and the only
> external data are the legitimate scsi structures, for the kernel
> loading it as a module is just the same of another scsi driver
> and there are no conflicts. If it does not work, it is an
> indication that either the driver need a fix or that the
> scsi structures are not general enough and need to be
> circumvented in some way by the driver.
> Another good reason is that on an external bus I have extremely
> various scsi devices, not turned on all the time, and I need to
> plug and unplug them in a clean way without rebooting the system.
> I do not see any justification why the above should be allowed
> if I had two different brands of scsi controllers but forbidden
> if all the scsi boards have the same brand.
> Anyway if the patch could take into account only modules there should
> be no conflicts at all.
I think it is better when the list include all drivers, but it is no problem
to sign which drivers are in modules and the other will not be touched.
It may be usefull for implementing a kernel/module option in future
assigning host numbers to appropriate drivers/boards (eg. I could force
that my ncr is always scsi1, even if I remove/disable adaptec being scsi0;
or in reverse - I don't know the autodetection sequence) It would cause
(if one wish) disks in devfs having always the same names (c#b#...)
independently on current hardware configuration.
By the way, I think that the names of drivers currently placed in
drivers' templates are too long to be functional for managing drivers.
So I have two ways
1. add a short field to sd_template (better in my opinion)
2. add a conversion table into the main scsi driver (it might be a large
structure)
What is your opinion in this matter?
with regards
--
=======================================================================
Andrzej M. Krzysztofowicz ankry@mif.pg.gda.pl
phone (48)(58) 347 14 61
Faculty of Applied Phys. & Math., Technical University of Gdansk
