From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752711Ab0JTNno (ORCPT ); Wed, 20 Oct 2010 09:43:44 -0400 Received: from canuck.infradead.org ([134.117.69.58]:48397 "EHLO canuck.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750875Ab0JTNnn convert rfc822-to-8bit (ORCPT ); Wed, 20 Oct 2010 09:43:43 -0400 Subject: Re: High CPU load when machine is idle (related to PROBLEM: Unusually high load average when idle in 2.6.35, 2.6.35.1 and later) From: Peter Zijlstra To: Damien Wyart Cc: Chase Douglas , Ingo Molnar , tmhikaru@gmail.com, Thomas Gleixner , linux-kernel@vger.kernel.org, Venkatesh Pallipadi In-Reply-To: <1287581440.3488.16.camel@twins> References: <20100929070153.GA2200@brouette> <20101014145813.GA2185@brouette> <20101020132732.GA30024@brouette> <1287581440.3488.16.camel@twins> Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 8BIT Date: Wed, 20 Oct 2010 15:43:28 +0200 Message-ID: <1287582208.3488.20.camel@twins> Mime-Version: 1.0 X-Mailer: Evolution 2.30.3 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Wed, 2010-10-20 at 15:30 +0200, Peter Zijlstra wrote: > On Wed, 2010-10-20 at 15:27 +0200, Damien Wyart wrote: > > > > Should'nt we enlarge the list of CC, because for now, responsivity has > > been close to 0 and it seems we will get a 2.6.36 with buggy load avg > > calculation. Even if it is only statistics, many supervision tools rely > > on the load avg, so for production environments, this is not a good > > thing. > > It already contains all the folks who know the code I'm afraid.. :/ > > I've been playing with it a bit more today, but haven't actually managed > to make it better, just differently worse.. Ah, I just remembered Venki recently poked at this code too, maybe he's got a bright idea.. Venki, there are cpu-load issues, the reported issue is that idle load is too high, and I think I can see that happening with the current code (due to 74f5187ac8). The flaw I can see in that commit is that we can go idle multiple times during the LOAD_FREQ window, which will basically inflate the idle contribution. All attempts from me to fix that so far have resulted in curious results.. Would you have a moment to also look at this?