linux-kernel - Re: [PATCH 3/3] mm: oom: show unreclaimable slab info when unreclaimable slabs

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <0353184c-7cad-7222-6fea-2c5df3dbe851@alibaba-inc.com>
Date:   Fri, 20 Oct 2017 07:12:56 +0800
From:   "Yang Shi" <yang.s@...baba-inc.com>
To:     Michal Hocko <mhocko@...nel.org>,
        David Rientjes <rientjes@...gle.com>
Cc:     cl@...ux.com, penberg@...nel.org, iamjoonsoo.kim@....com,
        akpm@...ux-foundation.org, linux-mm@...ck.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH 3/3] mm: oom: show unreclaimable slab info when
 unreclaimable slabs > user memory



On 10/19/17 12:28 AM, Michal Hocko wrote:
> On Tue 17-10-17 15:39:08, David Rientjes wrote:
>> On Wed, 18 Oct 2017, Yang Shi wrote:
>>
>>>> Yes, this should catch occurrences of "huge unreclaimable slabs", right?
>>>
>>> Yes, it sounds so. Although single "huge" unreclaimable slab might not result
>>> in excessive slabs use in a whole, but this would help to filter out "small"
>>> unreclaimable slab.
>>>
>>
>> Keep in mind this is regardless of SLAB_RECLAIM_ACCOUNT: your patch has
>> value beyond only unreclaimable slab, it can also be used to show
>> instances where the oom killer was invoked without properly reclaiming
>> slab.  If the total footprint of a slab cache exceeds 5%, I think a line
>> should be emitted unconditionally to the kernel log.
> 
> agreed. I am not sure 5% is the greatest fit but we can tune that later.

5% might be too few. For example, on a machine with 200G memory, if 
there is 80G page cache, radix_tree_node might consume 10G. IMHO, 10% 
might be better.

Yang

>