From:   Kairui Song <ryncsn@gmail.com>
To:     Johannes Weiner <hannes@cmpxchg.org>,
        Suren Baghdasaryan <surenb@google.com>
Cc:     Chengming Zhou <zhouchengming@bytedance.com>,
        =?UTF-8?q?Michal=20Koutn=C3=BD?= <mkoutny@suse.com>,
        Tejun Heo <tj@kernel.org>, Ingo Molnar <mingo@redhat.com>,
        Peter Zijlstra <peterz@infradead.org>, cgroups@vger.kernel.org,
        linux-kernel@vger.kernel.org, Kairui Song <ryncsn@gmail.com>
Subject: [PATCH 0/2] sched/psi: Optimize PSI iteration
Date:   Thu,  9 Feb 2023 00:16:52 +0800
Message-Id: <20230208161654.99556-1-ryncsn@gmail.com>
Reply-To: Kairui Song <kasong@tencent.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

Hi all,

Patch 1/2 simplify cgroup_psi, I didn't see a measurable performance
change with this.

Patch 2/2 removed the cache, I just noticed it have the same hierarchy
as the cgroup it's in, so I wondered if it worth adding a cache just for
simplifying the branch, and if we can balance the branch simplification
and minimize the memory footprint in another way, it seems this is
doable.

After the two patches, I see a measurable performance gain
using mmtests/perfpipe:

(AVG of 100 test, ops/sec, the higher the better)
KVM guest on a i7-9700:
        psi=0         root cgroup   5 levels of cgroup
Before: 59221         55352         47821
After:  60100         56036         50884

KVM guest on a Ryzen 9 5900HX:
        psi=0         root cgroup   5 levels of cgroup
Before: 144566        138919        128888
After:  145812        139580        133514

Kairui Song (2):
  sched/psi: simplify cgroup psi retrieving
  sched/psi: iterate through cgroups directly

 include/linux/psi.h       |  2 +-
 include/linux/psi_types.h |  1 -
 kernel/cgroup/cgroup.c    |  7 +++++-
 kernel/sched/psi.c        | 45 ++++++++++++++++++++++++++++-----------
 4 files changed, 39 insertions(+), 16 deletions(-)

-- 
2.39.1