linux-kernel - Re: [RFC PATCH 2/3] mm, compaction: use MIN_COMPACT_COSTLY

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

Message-ID: <a0f01341-a5d8-d015-c37e-4932eaafd868@suse.cz>
Date:   Fri, 2 Aug 2019 12:20:33 +0200
From:   Vlastimil Babka <vbabka@...e.cz>
To:     Mike Kravetz <mike.kravetz@...cle.com>, linux-mm@...ck.org,
        linux-kernel@...r.kernel.org
Cc:     Hillf Danton <hdanton@...a.com>, Michal Hocko <mhocko@...nel.org>,
        Mel Gorman <mgorman@...e.de>,
        Johannes Weiner <hannes@...xchg.org>,
        Andrew Morton <akpm@...ux-foundation.org>
Subject: Re: [RFC PATCH 2/3] mm, compaction: use MIN_COMPACT_COSTLY_PRIORITY
 everywhere for costly orders

On 8/1/19 10:33 PM, Mike Kravetz wrote:
> On 8/1/19 6:01 AM, Vlastimil Babka wrote:
>> Could you try testing the patch below instead? It should hopefully
>> eliminate the stalls. If it makes hugepage allocation give up too early,
>> we'll know we have to involve __GFP_RETRY_MAYFAIL in allowing the
>> MIN_COMPACT_PRIORITY priority. Thanks!
> 
> Thanks.  This patch does eliminate the stalls I was seeing.

Great, thanks! I'll send a proper patch then.

> In my testing, there is little difference in how many hugetlb pages are
> allocated.  It does not appear to be giving up/failing too early.  But,
> this is only with __GFP_RETRY_MAYFAIL.  The real concern would with THP
> requests.  Any suggestions on how to test that?

AFAICS the default THP defrag mode is unaffected, as GFP_TRANSHUGE_LIGHT doesn't
include __GFP_DIRECT_RECLAIM, so it never reaches this code. Madvised THP
allocations will be affected, which should best be tested the same way as Andrea
and Mel did in the __GFP_THISNODE debate.