From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753726Ab1EaBpv (ORCPT ); Mon, 30 May 2011 21:45:51 -0400 Received: from e1.ny.us.ibm.com ([32.97.182.141]:39165 "EHLO e1.ny.us.ibm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752164Ab1EaBpu (ORCPT ); Mon, 30 May 2011 21:45:50 -0400 Date: Mon, 30 May 2011 18:45:43 -0700 From: "Paul E. McKenney" To: Peter Zijlstra Cc: Damien Wyart , Ingo Molnar , Mike Galbraith , linux-kernel@vger.kernel.org Subject: Re: Very high CPU load when idle with 3.0-rc1 Message-ID: <20110531014543.GU2668@linux.vnet.ibm.com> Reply-To: paulmck@linux.vnet.ibm.com References: <20110530055924.GA9169@brouette> <1306755291.1200.2872.camel@twins> <20110530162354.GQ2668@linux.vnet.ibm.com> <1306775989.2497.527.camel@laptop> <20110530212833.GS2668@linux.vnet.ibm.com> <1306791219.23844.12.camel@twins> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <1306791219.23844.12.camel@twins> User-Agent: Mutt/1.5.20 (2009-06-14) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Mon, May 30, 2011 at 11:33:39PM +0200, Peter Zijlstra wrote: > On Mon, 2011-05-30 at 14:28 -0700, Paul E. McKenney wrote: > > On Mon, May 30, 2011 at 07:19:49PM +0200, Peter Zijlstra wrote: > > > On Mon, 2011-05-30 at 09:23 -0700, Paul E. McKenney wrote: > > > > sp.sched_priority = RCU_KTHREAD_PRIO; > > > > sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); > > > > > > Why are those things RT tasks anyway? The old ksoftirq runs as a regular > > > task. And once you start boosting things you can boost this into FIFO as > > > well... > > > > > > just wondering.. > > > > Because priority boosting doesn't help unless the callbacks also run > > RT priority. > > > > I could make it so that they ran as normal tasks if !RCU_BOOST, but > > they would still need to run as RT tasks for RCU_BOOST. I figured > > running them the same way in both cases would be simpler. > > Ah, I thought you'd boost the threads along with the waiters, to the > same prio so that they wouldn't disturb higher priority tasks for no > reason. I considered that, but working out when it is OK to deboost them is decidedly non-trivial. Thanx, Paul