From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <linux-kernel-owner@vger.kernel.org>
Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand
	id <S265462AbRF1A6z>; Wed, 27 Jun 2001 20:58:55 -0400
Received: (majordomo@vger.kernel.org) by vger.kernel.org
	id <S265464AbRF1A6p>; Wed, 27 Jun 2001 20:58:45 -0400
Received: from perninha.conectiva.com.br ([200.250.58.156]:44303 "HELO
	perninha.conectiva.com.br") by vger.kernel.org with SMTP
	id <S265462AbRF1A6c>; Wed, 27 Jun 2001 20:58:32 -0400
Date: Wed, 27 Jun 2001 20:25:22 -0300 (BRT)
From: Marcelo Tosatti <marcelo@conectiva.com.br>
To: tcm@nac.net
Cc: linux-kernel@vger.kernel.org
Subject: Re: Freezing bug in all kernels greater than 2.4.5-ac13 *AND*
 2.4.6-pre2
In-Reply-To: <Pine.LNX.4.21.0106272010580.1836-100000@freak.distro.conectiva>
Message-ID: <Pine.LNX.4.21.0106272023430.1836-100000@freak.distro.conectiva>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Sender: linux-kernel-owner@vger.kernel.org
X-Mailing-List: linux-kernel@vger.kernel.org



On Wed, 27 Jun 2001, Marcelo Tosatti wrote:

> 
> 
> On Wed, 27 Jun 2001 tcm@nac.net wrote:
> 
> > I decided, for the hell of it, to test the pre series as I've been
> > nudged by many people to try it in favor of the ac kernel series that
> > I've been having problems with. Well, it turns out I have ran into
> > exactly the same problem I had with the ac kernel series, which quite
> > frankly is surprising the hell out of me.
> > 
> > To make the kernel freeze/slow down to a crawl with affected kernels on
> > my machine I do this test:
> > 
> > Load X (This fills up my ram and causes me to swap a bit)
> > run a rxvt and su to root (proboably unnecessary)
> > du /
> > 
> > Now, somewhere in this test I start swapping a little bit, nothing
> > big... then BAM. hard disk, mouse, keyboard, all completely and utterly
> > stop. Video continues to work, but my cpu's load goes absolutely INSANE.
> > (If it recovers, gkrellm generally says I've gotten a loadavg somewhere
> > between 3-20, depending on how long it was stuck) This can last for
> > seconds (usually) minutes (once) or it can simply get worse and hang the
> > machine (many, many many times)
> > 
> > When it recovers from this, I generally see a MASSIVE write to swap,
> > (I'm using gkrellm to monitor it) and the system continues on as if
> > nothing happened - until, of course, this happens again. A kernel
> > compile can cause it. a rm -R of a large directory can cause it. Loading
> > a large application can cause it.
> > 
> > On some kernels this is more noticable than others - ac15 does it the
> > worst, although pre3 rivals it, and the symptoms are different on
> > ac17/18 - it'll simply freeze randomly and with no recovery instead of
> > sometimes freezing or sometimes slowing down to a crawl and recovering
> > or freezing. (Which is worse? You decide.)
> > 
> > Now, as before, I tested this with swap and without swap. With swap, I
> > get the hangs/freezes in all the affected kernels. Without swap, I
> > don't. Nada.
> > 
> > Now, the big question of the day folks: What changed between 2.4.6-pre2
> > and 2.4.6-pre3 that ALSO changed between 2.4.5-ac13 and 2.4.5-ac14 - and
> > now, what part of those patches were the VM? Anyone? I don't see in
> > 2.4.6-pre3 what changed that was part of the VM... So I am trying to
> > narrow it down a bit :)
> > 
> > This bug is driving me slightly nuts, so I want it dead. Anyone got a
> > exterminator handy? =)
> 
> Rik's page_launder() changes. 

Eek. I mean Rik's page_launder() changes are _causing_ the problem. (its
the only VM change between 2.4.6-pre2->pre3/2.4.5-ac13->ac14)

Question:

Whats the size of the inactive dirty and clean lists when you're about to
crash.