[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <20250506225533.2580386-1-shakeel.butt@linux.dev>
Date: Tue, 6 May 2025 15:55:29 -0700
From: Shakeel Butt <shakeel.butt@...ux.dev>
To: Andrew Morton <akpm@...ux-foundation.org>
Cc: Johannes Weiner <hannes@...xchg.org>,
Michal Hocko <mhocko@...nel.org>,
Roman Gushchin <roman.gushchin@...ux.dev>,
Muchun Song <muchun.song@...ux.dev>,
Alexei Starovoitov <ast@...nel.org>,
Vlastimil Babka <vbabka@...e.cz>,
Sebastian Andrzej Siewior <bigeasy@...utronix.de>,
Jakub Kicinski <kuba@...nel.org>,
Eric Dumazet <edumazet@...gle.com>,
linux-mm@...ck.org,
cgroups@...r.kernel.org,
linux-kernel@...r.kernel.org,
bpf@...r.kernel.org,
netdev@...r.kernel.org,
Meta kernel team <kernel-team@...a.com>
Subject: [PATCH v3 0/4] memcg: decouple memcg and objcg stocks
The per-cpu memcg charge cache and objcg charge cache are coupled in a
single struct memcg_stock_pcp and a single local lock is used to protect
both of the caches. This makes memcg charging and objcg charging nmi
safe challenging. Decoupling memcg and objcg stocks would allow us to
make them nmi safe and even work without disabling irqs independently.
This series completely decouples memcg and objcg stocks.
To evaluate the impact of this series with and without PREEMPT_RT
config, we ran varying number of netperf clients in different cgroups on
a 72 CPU machine.
$ netserver -6
$ netperf -6 -H ::1 -l 60 -t TCP_SENDFILE -- -m 10K
PREEMPT_RT config:
------------------
number of clients | Without series | With series
6 | 38559.1 Mbps | 38652.6 Mbps
12 | 37388.8 Mbps | 37560.1 Mbps
18 | 30707.5 Mbps | 31378.3 Mbps
24 | 25908.4 Mbps | 26423.9 Mbps
30 | 22347.7 Mbps | 22326.5 Mbps
36 | 20235.1 Mbps | 20165.0 Mbps
!PREEMPT_RT config:
-------------------
number of clients | Without series | With series
6 | 50235.7 Mbps | 51415.4 Mbps
12 | 49336.5 Mbps | 49901.4 Mbps
18 | 46306.8 Mbps | 46482.7 Mbps
24 | 38145.7 Mbps | 38729.4 Mbps
30 | 30347.6 Mbps | 31698.2 Mbps
36 | 26976.6 Mbps | 27364.4 Mbps
No performance regression was observed.
Changes since v2:
- Ran and included network intensive benchmarking results
- Brought back the simplify patch dropped in v2 after perf experiment.
Changes since v1:
- Drop first patch as requested by Alexei.
- Remove preempt_disable() usage as suggested by Vlastimil.
Shakeel Butt (4):
memcg: simplify consume_stock
memcg: separate local_trylock for memcg and obj
memcg: completely decouple memcg and obj stocks
memcg: no irq disable for memcg stock lock
mm/memcontrol.c | 175 ++++++++++++++++++++++++++++--------------------
1 file changed, 102 insertions(+), 73 deletions(-)
--
2.47.1
Powered by blists - more mailing lists