From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932104Ab2EUT61 (ORCPT ); Mon, 21 May 2012 15:58:27 -0400 Received: from mga01.intel.com ([192.55.52.88]:64623 "EHLO mga01.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1756333Ab2EUT6Z (ORCPT ); Mon, 21 May 2012 15:58:25 -0400 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="4.71,315,1320652800"; d="scan'208";a="155318449" Subject: Re: [PATCH 2/3] x86: x2apic/cluster: Make use of lowest priority delivery mode From: Suresh Siddha Reply-To: Suresh Siddha To: Ingo Molnar Cc: Linus Torvalds , Alexander Gordeev , Arjan van de Ven , linux-kernel@vger.kernel.org, x86@kernel.org, Cyrill Gorcunov , Yinghai Lu , Peter Zijlstra Date: Mon, 21 May 2012 12:56:48 -0700 In-Reply-To: <20120521191546.GA28819@gmail.com> References: <20120518102640.GB31517@dhcp-26-207.brq.redhat.com> <20120521082240.GA31407@gmail.com> <20120521093648.GC28930@dhcp-26-207.brq.redhat.com> <20120521124025.GC17065@gmail.com> <20120521144812.GD28930@dhcp-26-207.brq.redhat.com> <20120521145904.GA7068@gmail.com> <20120521191546.GA28819@gmail.com> Organization: Intel Corp Content-Type: text/plain; charset="UTF-8" X-Mailer: Evolution 3.0.3 (3.0.3-1.fc15) Content-Transfer-Encoding: 7bit Message-ID: <1337630208.1997.141.camel@sbsiddha-desk.sc.intel.com> Mime-Version: 1.0 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Mon, 2012-05-21 at 21:15 +0200, Ingo Molnar wrote: > The one negative effect I mentioned, affine wakeups done by the > scheduler, could still bite us - this has to be measured and > affine wakeups have to be made less prominent if IRQ handlers > start jumping around. We definitely don't want tasks to follow > round-robin IRQs around. Actually this (ping-ponging around idle cpu's in a socket) happens already today, as the affine wakeups use select_idle_sibling() to wake up the task on an idle sibling (HT or core sibling) if the cpu on which the task is woken-up is busy with something else. But I agree, somehow all these need to work together. For example to save power, what we want is the idle core to stay in idle with least number of interruptions, with interrupts routed to busy cores, scheduler not aggressively waking the tasks up on idle cores etc. One quick thought is to use the cpufreq governor decisions and if all the cpu's in the package are in lower p-states, then we can default for power-saving decisions, otherwise if any cpu is at P0, switch for performance policy(in terms of irq routing, scheduler wake-up decisions etc) dynamically etc. thanks, suresh