From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id ; Mon, 10 Sep 2001 17:26:06 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id ; Mon, 10 Sep 2001 17:25:47 -0400 Received: from johnson.mail.mindspring.net ([207.69.200.177]:43806 "EHLO johnson.mail.mindspring.net") by vger.kernel.org with ESMTP id ; Mon, 10 Sep 2001 17:25:30 -0400 Subject: Re: Preemption patch, some more feedback From: Robert Love To: linux-kernel@vger.kernel.org In-Reply-To: In-Reply-To: X-Mailer: Evolution/0.13.99+cvs.2001.09.08.07.08 (Preview Release) Content-Type: text/plain Content-Transfer-Encoding: 7bit X-Mailer: Evolution/0.13.99+cvs.2001.09.08.07.08 (Preview Release) Date: 10 Sep 2001 17:26:08 -0400 Message-Id: <1000157173.18895.24.camel@phantasy> Mime-Version: 1.0 Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Mon, 2001-09-10 at 11:41, Francis Galiegue wrote: > Machine is Athlon 650, AMD Viper chipset, 256 MB RAM. Kernel is > 2.4.9-ac10 + preempt patch + irc_conntrack patch from iptables. > > The preempt patch largely improves multimedia latency (no surprise on > that), I can watch a DivX smoothly (with mplayer, gfx being Matrox G400) > and compile various stuff behind. > > However, a very simple command destroys this completely: > > cat /dev/zero >/dev/null > > DivX playback then becomes sluggish, no visible difference in this case > between stock kernel and "preempt" kernel. A long-term lock must be held for the duration of `cat /dev/zero > /dev/null' -- i dont know if it is in the access to /dev/null or /dev/zero or in the basic file operation itself. as long as a lock is held, preemption can not occur. what do we do? for the short term, and the benefit of everyone (UP, SMP, and preemption users) we need to eliminate long-held locks with a better solution. in the long term, we can look at having the preemption patch use various different types of locks (priority locks, spin then sleep locks, etc.) -- Robert M. Love rml at ufl.edu rml at tech9.net