From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752364AbbGMWXz (ORCPT ); Mon, 13 Jul 2015 18:23:55 -0400 Received: from bombadil.infradead.org ([198.137.202.9]:47551 "EHLO bombadil.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751450AbbGMWXx (ORCPT ); Mon, 13 Jul 2015 18:23:53 -0400 Date: Tue, 14 Jul 2015 00:23:46 +0200 From: Peter Zijlstra To: "Paul E. McKenney" Cc: Will Deacon , "linux-arch@vger.kernel.org" , "linux-kernel@vger.kernel.org" , Benjamin Herrenschmidt Subject: Re: [RFC PATCH v2] memory-barriers: remove smp_mb__after_unlock_lock() Message-ID: <20150713222346.GE19282@twins.programming.kicks-ass.net> References: <1436789704-10086-1-git-send-email-will.deacon@arm.com> <20150713131143.GY19282@twins.programming.kicks-ass.net> <20150713140915.GD2632@arm.com> <20150713142109.GE2632@arm.com> <20150713155447.GB19282@twins.programming.kicks-ass.net> <20150713175029.GO2632@arm.com> <20150713202032.GZ3717@linux.vnet.ibm.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20150713202032.GZ3717@linux.vnet.ibm.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 Mon, Jul 13, 2015 at 01:20:32PM -0700, Paul E. McKenney wrote: > On Mon, Jul 13, 2015 at 06:50:29PM +0100, Will Deacon wrote: > > So if I'm following along, smp_mb__after_unlock_lock *does* provide > > transitivity when used with UNLOCK + LOCK, which is stronger than your > > example here. > > Yes, that is indeed the intent. Maybe good to state this explicitly somewhere. > > I don't think we want to make the same guarantee for general RELEASE + > > ACQUIRE, because we'd end up forcing most architectures to implement the > > expensive macro for a case that currently has no users. > > Agreed, smp_mb__after_unlock_lock() makes a limited guarantee. I'm still not seeing how the archs that implement load_acquire and store_release with smp_mb() are a problem. If we look at the inside of the critical section again -- similar argument as before: *A = a smp_mb() store M load N smp_mb() *B = b A and B are fully ordered, and in this case even transitivity is provided. I'm stating that the order of M and N don't matter, only the load/stores that are inside the acquire/release are constrained. IOW, I think smp_mb__after_unlock_lock() already works as advertised with all our acquire/release thingies -- as is stated by the documentation. That said, I'm not aware of anybody but RCU actually using this, so its not used in that capacity. > > In which case, it boils down to the question of how expensive it would > > be to implement an SC UNLOCK operation on PowerPC and whether that justifies > > the existence of a complicated barrier macro that isn't used outside of > > RCU. > > Given that it is either smp_mb() or nothing, I am not seeing the > "complicated" part... The 'complicated' part is that we need think about it; that is we need to realized and remember that UNLOCK+LOCK is a load-store barrier but fails to provide transitivity.