Subject: Re: question about Fastlane Z3 problems
To: None <hensen@wpos4.physik.uni-wuppertal.de>
From: Ignatios Souvatzis <ignatios@cs.uni-bonn.de>
List: port-amiga
Date: 02/06/1997 12:29:29
   From: Stefan Hensen <hensen@wpos4.physik.uni-wuppertal.de>
   Date: Thu, 6 Feb 1997 11:02:12 +0100 (MET)
   X-Mailer: ELM [version 2.4 PL23]
   Content-Type: text
   Sender: port-amiga-owner@NetBSD.ORG
   Precedence: list
   Delivered-To: port-amiga@NetBSD.ORG

   Hi,

   I have a question concerning the problems of the Fastlane Z3 with multiple
   SCSI devices on the bus, which are mentioned in the installation
   instructions. (Well, it's probably more about the general Emulex FAS
   problems.) There is something that surprised me about this:

[...]

   The big question is: What has happened to the Fastlane driver, that things
   that used to work with an older version do not work any more?

Well, as far as I know, the author of sfas has made changes, which
improved operation with _his_ set of devices, but obviously cause
problems with others. Since then, he didn't have any more time to work
on this.


   (Or could it occasionally be that the Quantum Fireball causes problems that
   the Empire does not?)


This is quite possible.

It is a state machine problem... to run SCSI, (at least ) three state
machines have to run synchronized: the software driver, the scsi chip+
DMA state, and the drive in question. Slightly different from what the
original author expected, but still legal (or even illegal) behaviour
of the connected drive seem to cause problems for our driver.

There might be additional problems with the Fastlane DMA driver,
because the Fastlane DMA isn't exactly the dream DMA a multitasking,
multi-purpose SCSI driver programmer would like to have, but a
big one was hunted down and squished some time ago, and the remaining
can't be isolated before the chip core driver isn't fixed.

THere are efforts underway to split the sparc/alpha/decstation ESP
driver into the chip core and DMA backends. When this has happened,
I'm pretty sure we can rewrite our backends to attach to _that_, and
get better results.

Regards,
	Ignatios SOuvatzis