From mboxrd@z Thu Jan 1 00:00:00 1970
From: Pat LaVarre
Subject: Re: zeroes read back more often than appended
Date: 27 Oct 2003 14:55:43 -0700
Sender: linux-fsdevel-owner@vger.kernel.org
Message-ID: <1067291743.4396.5.camel@patehci2>
References: <1065553341.8172.45.camel@patehci2>
<20031007204951.A25423@falcon.csc.calpoly.edu>
<1065732882.5176.14.camel@patehci2> <1065747151.2314.12.camel@patehci2>
<1065803984.3037.7.camel@patehci2> <1065809754.3037.105.camel@patehci2>
<1066091916.3011.2.camel@patehci2> <1066096137.3314.32.camel@patehci2>
<1066692050.2814.33.camel@patehci2> <1066747637.8228.7.camel@patehci2>
<1066754816.12021.1.camel@patehci2> <1066935133.11806.0.camel@patehci2>
Mime-Version: 1.0
Content-Type: text/plain
Content-Transfer-Encoding: 7bit
Return-path:
Received: from email-out1.iomega.com ([147.178.1.82]:49861 "EHLO
email.iomega.com") by vger.kernel.org with ESMTP id S263590AbTJ0U4H
(ORCPT );
Mon, 27 Oct 2003 15:56:07 -0500
Received: from royntex01.iomegacorp.com (unknown [147.178.90.120])
by email.iomega.com (Postfix) with ESMTP id C30281DA3
for ; Mon, 27 Oct 2003 13:56:06 -0700 (MST)
To: linux-fsdevel@vger.kernel.org
In-Reply-To: <1066935133.11806.0.camel@patehci2>
List-Id: linux-fsdevel.vger.kernel.org
> > List: linux-fsdevel
> > Subject: given fs dies if SMP, if PREEMPT, if open two or more
> >
> > ... Can anyone here advise me specifically
> > what to try next? ... trouble ... says to me
> > something in udf.ko doesn't know how to
> > reenter itself, right?
Sure.
> > So my next step is to try reading while
> > writing, to see if that breaks also, right?
I did try, but I got the wrong answer.
Writing while reading does Not break so easily.
For example, writing via a loop of (read file 0, write file 1, read file
2) works ok, at least at my desk for width x height of 0xFF600 x 0xC8
tried three times in a row. Also I tried a read by dd in parallel with
a loop to append, again no joy.
So far, all that breaks is writing while writing. That does break in
-test9, no matter SMP/PREEMPT, as described by this thread.
> > in -test9 as described
Either `make oldconfig` somewhere near -test9 injects some noise, or I
erred in past configurations myself.
Specifically, to have my -test9 .config differ minimally from `make
defconfig`, I had to work to ensure I agreed with defconfig on two
points:
$ egrep 'SOFTWARE|SPINLOCK' .config
# CONFIG_SOFTWARE_SUSPEND is not set
CONFIG_DEBUG_SPINLOCK_SLEEP=y
$
Pat LaVarre