From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932426AbdJZQ1J (ORCPT ); Thu, 26 Oct 2017 12:27:09 -0400 Received: from mx2.suse.de ([195.135.220.15]:57857 "EHLO mx2.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932127AbdJZQ1E (ORCPT ); Thu, 26 Oct 2017 12:27:04 -0400 Date: Thu, 26 Oct 2017 18:27:01 +0200 From: Michal Hocko To: Yang Shi Cc: cl@linux.com, penberg@kernel.org, rientjes@google.com, iamjoonsoo.kim@lge.com, akpm@linux-foundation.org, linux-mm@kvack.org, linux-kernel@vger.kernel.org Subject: Re: [PATCH 2/2] mm: oom: dump single excessive slab cache when oom Message-ID: <20171026162701.re4lclnqkngczpcl@dhcp22.suse.cz> References: <1508971740-118317-1-git-send-email-yang.s@alibaba-inc.com> <1508971740-118317-3-git-send-email-yang.s@alibaba-inc.com> <20171026145312.6svuzriij33vzgw7@dhcp22.suse.cz> <44577b73-2e2d-5571-4c8b-3233e3776a52@alibaba-inc.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <44577b73-2e2d-5571-4c8b-3233e3776a52@alibaba-inc.com> User-Agent: NeoMutt/20170609 (1.8.3) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Fri 27-10-17 00:15:17, Yang Shi wrote: > > > On 10/26/17 7:53 AM, Michal Hocko wrote: > > On Thu 26-10-17 06:49:00, Yang Shi wrote: > > > Per the discussion with David [1], it looks more reasonable to just dump > > > > Please try to avoid external references in the changelog as much as > > possible. > > OK. > > > > > > the single excessive slab cache instead of dumping all slab caches when > > > oom. > > > > You meant to say > > "to just dump all slab caches which excess 10% of the total memory." > > > > While we are at it. Abusing calc_mem_size seems to be rather clumsy and > > tt is not nodemask aware so you the whole thing is dubious for NUMA > > constrained OOMs. > > Since we just need the total memory size of the node for NUMA constrained > OOM, we should be able to use show_mem_node_skip() to bring in nodemask. yes > > The more I think about this the more I am convinced that this is just > > fiddling with the code without a good reason and without much better > > outcome. > > I don't get you. Do you mean the benefit is not that much with just dumping > excessive slab caches? Yes, I am not sure it makes sense to touch it without further experiences. I am not saying this is a wrong approach I would just give it some more time to see how it behaves in the wild and then make changes based on that experience. -- Michal Hocko SUSE Labs