mm, sysctl: make NUMA stats configurable
authorKemi Wang <[email protected]>
Thu, 16 Nov 2017 01:38:22 +0000 (17:38 -0800)
committerLinus Torvalds <[email protected]>
Thu, 16 Nov 2017 02:21:07 +0000 (18:21 -0800)
commit4518085e127dff97e74f74a8780d7564e273bec8
tree1ad9218a5ca77345ee644b62ff74639771414106
parent9a8ec03ed022b79e56dca820cf04debbb240c7b3
mm, sysctl: make NUMA stats configurable

This is the second step which introduces a tunable interface that allow
numa stats configurable for optimizing zone_statistics(), as suggested
by Dave Hansen and Ying Huang.

=========================================================================

When page allocation performance becomes a bottleneck and you can
tolerate some possible tool breakage and decreased numa counter
precision, you can do:

echo 0 > /proc/sys/vm/numa_stat

In this case, numa counter update is ignored.  We can see about
*4.8%*(185->176) drop of cpu cycles per single page allocation and
reclaim on Jesper's page_bench01 (single thread) and *8.1%*(343->315)
drop of cpu cycles per single page allocation and reclaim on Jesper's
page_bench03 (88 threads) running on a 2-Socket Broadwell-based server
(88 threads, 126G memory).

Benchmark link provided by Jesper D Brouer (increase loop times to
10000000):

  https://github.com/netoptimizer/prototype-kernel/tree/master/kernel/mm/bench

=========================================================================

When page allocation performance is not a bottleneck and you want all
tooling to work, you can do:

echo 1 > /proc/sys/vm/numa_stat

This is system default setting.

Many thanks to Michal Hocko, Dave Hansen, Ying Huang and Vlastimil Babka
for comments to help improve the original patch.

[[email protected]: make sure mutex is a global static]
Link: http://lkml.kernel.org/r/20171107213809.GA4314@beast
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Kemi Wang <[email protected]>
Signed-off-by: Kees Cook <[email protected]>
Reported-by: Jesper Dangaard Brouer <[email protected]>
Suggested-by: Dave Hansen <[email protected]>
Suggested-by: Ying Huang <[email protected]>
Acked-by: Vlastimil Babka <[email protected]>
Acked-by: Michal Hocko <[email protected]>
Cc: "Luis R . Rodriguez" <[email protected]>
Cc: Kees Cook <[email protected]>
Cc: Jonathan Corbet <[email protected]>
Cc: Mel Gorman <[email protected]>
Cc: Johannes Weiner <[email protected]>
Cc: Christopher Lameter <[email protected]>
Cc: Sebastian Andrzej Siewior <[email protected]>
Cc: Andrey Ryabinin <[email protected]>
Cc: Tim Chen <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: Aaron Lu <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Documentation/sysctl/vm.txt
include/linux/vmstat.h
kernel/sysctl.c
mm/mempolicy.c
mm/page_alloc.c
mm/vmstat.c