Return-Path: <owner-ips@ece.cmu.edu>
X-Sieve: cmu-sieve 2.0
Return-Path: <owner-ips@ece.cmu.edu>
Received: (from majordom@localhost)
	by ece.cmu.edu (8.11.0/8.10.2) id h130IZ224964
	for ips-outgoing; Sun, 2 Feb 2003 19:18:35 -0500 (EST)
X-Authentication-Warning: ece.cmu.edu: majordom set sender to owner-ips@ece.cmu.edu using -f
Received: from lion.ninthwonder.com (lion.ninthwonder.com [151.199.66.147])
	by ece.cmu.edu (8.11.0/8.10.2) with ESMTP id h130IXW24959
	for <ips@ece.cmu.edu>; Sun, 2 Feb 2003 19:18:33 -0500 (EST)
Received: by lion.ninthwonder.com (Postfix, from userid 1021)
	id C77EA30706; Sun,  2 Feb 2003 19:18:32 -0500 (EST)
Date: Sun, 2 Feb 2003 16:17:24 -0800 (PST)
From: Bill Studenmund <wrstuden@wasabisystems.com>
X-X-Sender:  <wrstuden@candlekeep.home-net.internetconnect.net>
To: Eddy Quicksall <eddy_quicksall@ivivity.com>
Cc: "Elliott, Robert (Server Storage)" <Elliott@hp.com>, <ips@ece.cmu.edu>
Subject: RE: iSCSI: different question about terminated tasks
In-Reply-To: <Pine.NEB.4.33.0302021305030.408-100000@candlekeep.home-net.internetconnect.net>
Message-ID: <Pine.NEB.4.33.0302021603300.408-100000@candlekeep.home-net.internetconnect.net>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Sender: owner-ips@ece.cmu.edu
Precedence: bulk

On Sun, 2 Feb 2003, Bill Studenmund wrote:

> On Sat, 1 Feb 2003, Eddy Quicksall wrote:
>
> > Thanks, it was the abort that I really didn't understand.
> >
> > It appears as though there is not "one ACA for each terminated task" ... do
> > you agree with that?

Let me add some specifics, which might make a difference.

If QErr is 01b, then I can see only one ACA - the ACA itself killed the
others.

My question though is more what happens with QErr 00b. Because if the
other tasks get blocked before they can actually assert ACA, they are
still sitting there waiting to assert ACA as soon as the first aborted ACA
task finishes.

Part of where this is comming from a desire for symmetry. If we had two
totally independent tasks in the same session, both NACA=1, both error out
at the same time, as I understand it, we should get two ACA cases. Since
there should be a lock ensuring only one task asserts ACA at once, the
second ACA-desiring task will get blocked, waiting to assert ACA.

My desire for symmetry means that, since the spec says the tasks get
killed with (otherwise normal) SCSI consequences, I'd like the same thing
to happen here when we go killing tasks as would have happened any other
time we have multiple dying tasks.

If however there is a piece I've missed, please point it out.

Take care,

Bill

