From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755172AbbAZNkS (ORCPT ); Mon, 26 Jan 2015 08:40:18 -0500 Received: from gum.cmpxchg.org ([85.214.110.215]:49198 "EHLO gum.cmpxchg.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751244AbbAZNkO (ORCPT ); Mon, 26 Jan 2015 08:40:14 -0500 Date: Mon, 26 Jan 2015 08:37:12 -0500 From: Johannes Weiner To: Valdis.Kletnieks@vt.edu Cc: Christoph Lameter , Guenter Roeck , akpm@linux-foundation.org, mm-commits@vger.kernel.org, linux-kernel@vger.kernel.org, linux-mm@kvack.org, linux-fsdevel@vger.kernel.org, linux-next@vger.kernel.org, sfr@canb.auug.org.au, mhocko@suse.cz Subject: Re: mmotm 2015-01-22-15-04: qemu failure due to 'mm: memcontrol: remove unnecessary soft limit tree node test' Message-ID: <20150126133712.GB9738@cmpxchg.org> References: <54c1822d.RtdGfWPekQVAw8Ly%akpm@linux-foundation.org> <20150123050802.GB22751@roeck-us.net> <20150123141817.GA22926@phnom.home.cmpxchg.org> <54C2B01D.4070303@roeck-us.net> <20150124071623.GA17705@phnom.home.cmpxchg.org> <109548.1422221788@turing-police.cc.vt.edu> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <109548.1422221788@turing-police.cc.vt.edu> Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Sun, Jan 25, 2015 at 04:36:28PM -0500, Valdis.Kletnieks@vt.edu wrote: > On Sat, 24 Jan 2015 02:16:23 -0500, Johannes Weiner said: > > > I would generally agree, but this code, which implements a userspace > > interface, is already grotesquely inefficient and heavyhanded. It's > > also superseded in the next release, so we can just keep this simple > > at this point. > > Wait, what? Userspace interface that's superceded in the next release? The existing interface and its implementation are going to remain in place, obviously, we can't break userspace. But the semantics are ill-defined and the implementation bad to a point where we decided to fix both by adding a second interface and encouraging users to switch. Now if a user were to report that these off-node allocations are actually creating problems in real life I would fix it. But I'm fairly certain that remote access costs are overshadowed by the reclaim stalls this mechanism creates. So what I was trying to say above is that I don't see a point in complicating the v1 implementation for a questionable minor optimization when v2 is already being added to address much more severe shortcomings in v1.