Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751693AbdFEJ11 (ORCPT ); Mon, 5 Jun 2017 05:27:27 -0400 Received: from mx2.suse.de ([195.135.220.15]:32989 "EHLO mx1.suse.de" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1751290AbdFEJ1Z (ORCPT ); Mon, 5 Jun 2017 05:27:25 -0400 Subject: Re: [PATCH] mm: make PR_SET_THP_DISABLE immediately active To: Andrew Morton Cc: Mike Rapoport , Linux API , Michal Hocko , Andrea Arcangeli , Arnd Bergmann , "Kirill A. Shutemov" , Pavel Emelyanov , linux-mm , lkml , Michal Hocko References: <1496415802-30944-1-git-send-email-rppt@linux.vnet.ibm.com> <20170602125059.66209870607085b84c257593@linux-foundation.org> <8a810c81-6a72-2af0-a450-6f03c71d8cca@suse.cz> <20170602134038.13728cb77678ae1a7d7128a4@linux-foundation.org> <20170602141041.baace0cfa370b6bec6d411b4@linux-foundation.org> From: Vlastimil Babka Message-ID: <0457fa18-fdaa-6572-819d-f918c49c0c6f@suse.cz> Date: Mon, 5 Jun 2017 11:27:20 +0200 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101 Thunderbird/52.1.1 MIME-Version: 1.0 In-Reply-To: <20170602141041.baace0cfa370b6bec6d411b4@linux-foundation.org> Content-Type: text/plain; charset=utf-8 Content-Language: en-US Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 2814 Lines: 68 On 06/02/2017 11:10 PM, Andrew Morton wrote: > On Fri, 2 Jun 2017 22:55:12 +0200 Vlastimil Babka wrote: > >> On 06/02/2017 10:40 PM, Andrew Morton wrote: >>> On Fri, 2 Jun 2017 22:31:47 +0200 Vlastimil Babka wrote: >>>>> Perhaps we should be adding new prctl modes to select this new >>>>> behaviour and leave the existing PR_SET_THP_DISABLE behaviour as-is? >>>> >>>> I think we can reasonably assume that most users of the prctl do just >>>> the fork() & exec() thing, so they will be unaffected. >>> >>> That sounds optimistic. Perhaps people are using the current behaviour >>> to set on particular mapping to MMF_DISABLE_THP, with >>> >>> prctl(PR_SET_THP_DISABLE) >>> mmap() >>> prctl(PR_CLR_THP_DISABLE) >>> >>> ? >>> >>> Seems a reasonable thing to do. >> >> Using madvise(MADV_NOHUGEPAGE) seems reasonabler to me, with the same >> effect. And it's older (2.6.38). >> >>> But who knows - people do all sorts of >>> inventive things. >> >> Yeah :( but we can hope they don't even know that the prctl currently >> behaves they way it does - man page doesn't suggest it would, and most >> of us in this thread found it surprising. > > Well. There might be such people and sometimes we do make people > unhappy. it partly depends on how traumatic it would be to leave the > current behaviour as-is. Have you evaluated such a patch? You mean introducing a new prctl instead of changing the existing one? I can evaluate that as being ugly :) Well, maybe we could use arg3, because currently we have: case PR_SET_THP_DISABLE: if (arg3 || arg4 || arg5) return -EINVAL; We could make non-zero arg3 (or specific value of arg3) set the new "immediate" behavior. This would also take care of the discovery of kernels that support the fixed/altered behavior, without having to check uname etc - just check if we got -EINVAL. I'm just not sure how to implement PR_GET_THP_DISABLE properly in such scenario. Or what happens when somebody calls SET with arg3==0 and then arg3==1 (or vice versa). But we would have to think about it even when we introduced a newly named option. Reminds me of the MLOCK_ONFAULT discussions... >>>> And as usual, if >>>> somebody does complain in the end, we revert and try the other way? >>> >>> But by then it's too late - the new behaviour will be out in the field. >> >> Revert in stable then? >> But I don't think this patch should go to stable. I understand right >> that CRIU will switch to the UFFDIO_COPY approach and doesn't need the >> prctl change/new madvise anymore? > > What I mean is that the new behaviour will go out in 4.12 and it may > be many months before we find out that we broke someone. By then, we > can't go back because others may be assuming the new behaviour. >