linux-kernel - [RFC PATCH 0/4] Reduce tree_lock contention during swap and reclaim of a single file v1

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

Date:   Fri,  9 Sep 2016 10:59:31 +0100
From:   Mel Gorman <mgorman@...hsingularity.net>
To:     LKML <linux-kernel@...r.kernel.org>, Linux-MM <linux-mm@...ck.org>,
        Mel Gorman <mgorman@...hsingularity.net>
Cc:     Dave Chinner <david@...morbit.com>,
        Linus Torvalds <torvalds@...ux-foundation.org>,
        Ying Huang <ying.huang@...el.com>,
        Michal Hocko <mhocko@...nel.org>
Subject: [RFC PATCH 0/4] Reduce tree_lock contention during swap and reclaim of a single file v1

This is a follow-on series from the thread "[lkp] [xfs] 68a9f5e700:
aim7.jobs-per-min -13.6% regression" with active parties cc'd.  I've
pushed the series to git.kernel.org where the LKP robot should pick it
up automatically.

git://git.kernel.org/pub/scm/linux/kernel/git/mel/linux.git mm-reclaim-contention-v1r15

The progression of this series has been unsatisfactory. Dave originally
reported a problem with tree_lock contention and while it can be fixed
by pushing reclaim to direct reclaim, it slows swap considerably and was
not a universal win. This series is the best balance I've found so far
between the swapping and large rewriter cases.

I never reliably produced the same contentions that Dave did so testing
is needed.  Dave, ideally you would test patches 1+2 and patches 1+4 but
a test of patches 1+3 would also be nice if you have the time. Minimally,
I'm expected that patches 1+2 will help the swapping-to-fast-storage case
(LKP to confirm independently) and may be worth considering on their own
even if Dave's test case is not helped.

 drivers/block/brd.c |   1 +
 mm/vmscan.c         | 209 +++++++++++++++++++++++++++++++++++++++++++++-------
 2 files changed, 182 insertions(+), 28 deletions(-)

-- 
2.6.4