Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752269AbbEFNzZ (ORCPT ); Wed, 6 May 2015 09:55:25 -0400 Received: from cantor2.suse.de ([195.135.220.15]:37852 "EHLO mx2.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751472AbbEFNzW (ORCPT ); Wed, 6 May 2015 09:55:22 -0400 Date: Wed, 6 May 2015 15:55:20 +0200 From: Michal Hocko To: Vladimir Davydov Cc: Andrew Morton , Johannes Weiner , Tejun Heo , Christoph Lameter , Pekka Enberg , David Rientjes , Joonsoo Kim , Greg Thelen , linux-mm@kvack.org, cgroups@vger.kernel.org, linux-kernel@vger.kernel.org Subject: Re: [PATCH 1/2] gfp: add __GFP_NOACCOUNT Message-ID: <20150506135520.GN14550@dhcp22.suse.cz> References: <20150506115941.GH14550@dhcp22.suse.cz> <20150506122431.GA29387@esperanza> <20150506123541.GK14550@dhcp22.suse.cz> <20150506132510.GB29387@esperanza> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20150506132510.GB29387@esperanza> User-Agent: Mutt/1.5.23 (2014-03-12) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 3351 Lines: 81 On Wed 06-05-15 16:25:10, Vladimir Davydov wrote: > On Wed, May 06, 2015 at 02:35:41PM +0200, Michal Hocko wrote: > > On Wed 06-05-15 15:24:31, Vladimir Davydov wrote: [...] > > > I don't think making this flag per-cache is an option either, but for > > > another reason - it would not be possible to merge such a kmem cache > > > with caches without this flag set. As a result, total memory pressure > > > would increase, even for setups without kmem-active memory cgroups, > > > which does not sound acceptable to me. > > > > I am not sure I see the performance implications here because kmem > > accounted memcgs would have their copy of the cache anyway, no? > > It's orthogonal. > > Suppose there are two *global* kmem caches, A and B, which would > normally be merged, i.e. A=B. Then we find out that we don't want to > account allocations from A to memcg while still accounting allocations > from B. Obviously, cache A can no longer be merged with cache B so we > have two different caches instead of the only merged one, even if there > are *no* memory cgroups at all. That might result in increased memory > consumption due to fragmentation. Got your point. Thanks for the clarification! > Although it is not really critical, especially counting that SLAB > merging was introduced not long before, the idea that enabling an extra > feature, such as memcg, without actually using it, may affect the global > behavior does not sound good to me. Agreed. > > Anyway, I guess it would be good to document these reasons in the > > changelog. > > > > > > So I do not object to opt-out for kmemcg accounting but I really think > > > > the name should be changed. > > > > > > I named it __GFP_NOACCOUNT to match with __GFP_NOTRACK, which is a very > > > specific flag too (kmemcheck), nevertheless it has a rather generic > > > name. > > > > __GFP_NOTRACK is a bad name IMHO as well. One has to go and check the > > comment to see this is kmemleak related. > > I think it's a good practice to go to its definition and check comments > when encountering an unknown symbol anyway. With ctags/cscope it's > trivial :-) > > > > > > Anyways, what else apart from memcg can account kmem so that we have to > > > mention KMEMCG in the flag name explicitly? > > > > NOACCOUNT doesn't imply kmem at all so it is not clear who is in charge > > of the accounting. > > IMO it is a benefit. If one day for some reason we want to bypass memcg > accounting for some other type of allocation somewhere, we can simply > reuse it. But what if somebody, say a highlevel memory allocator in the kernel, want's to (ab)use this flag for its internal purpose as well? > > I do not insist on __GFP_NO_KMEMCG of course but it sounds quite > > specific about its meaning and scope. > > There is another argument against __GFP_NO_KMEMCG: it is not yet clear > if kmem is going to be accounted separately in the unified cgroup > hierarchy. As I've said, I do not insist on *KMEMCG. __GFP_NO_MEMCG would be generic enough to rule out MEMCG altogether as well. Be it kmem or user memory. -- Michal Hocko SUSE Labs -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/