From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752898AbbFAIZL (ORCPT ); Mon, 1 Jun 2015 04:25:11 -0400 Received: from bombadil.infradead.org ([198.137.202.9]:58204 "EHLO bombadil.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752225AbbFAIY2 (ORCPT ); Mon, 1 Jun 2015 04:24:28 -0400 Date: Mon, 1 Jun 2015 10:24:23 +0200 From: Peter Zijlstra To: Mike Galbraith Cc: LKML , Ingo Molnar , ktkhai@parallels.com Subject: Re: sched_setscheduler() vs idle_balance() race Message-ID: <20150601082423.GS19282@twins.programming.kicks-ass.net> References: <1432799032.3237.119.camel@gmail.com> <20150528135355.GK3644@twins.programming.kicks-ass.net> <1432991306.3395.35.camel@gmail.com> <1433054344.6545.20.camel@gmail.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <1433054344.6545.20.camel@gmail.com> User-Agent: Mutt/1.5.21 (2012-12-30) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Sun, May 31, 2015 at 08:39:04AM +0200, Mike Galbraith wrote: > I don't see why we can't just say no in can_migrate_task() if ->pi_lock > is held. I suppose we could do that; what I really want to avoid is also requiring pi_lock for scheduling. The down-side of looking at pi_lock for migration is that there is no common point for migrating tasks, its all inside the classes, so we'd get to sprinkle it all over the place.