Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Patrick Bellasi <patrick.bellasi@arm.com>
To:     linux-kernel@vger.kernel.org
Cc:     Ingo Molnar <mingo@redhat.com>,
        Peter Zijlstra <peterz@infradead.org>,
        Vincent Guittot <vincent.guittot@linaro.org>,
        Quentin Perret <quentin.perret@arm.com>,
        Dietmar Eggemann <dietmar.eggemann@arm.com>,
        Morten Rasmussen <morten.rasmussen@arm.com>,
        Juri Lelli <juri.lelli@redhat.com>,
        Todd Kjos <tkjos@google.com>,
        Steve Muckle <smuckle@google.com>,
        Suren Baghdasaryan <surenb@google.com>
Subject: [PATCH v2 0/3] util_est regression fixup and cleanups
Date:   Mon,  5 Nov 2018 14:53:57 +0000
Message-Id: <20181105145400.935-1-patrick.bellasi@arm.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

This is a respin of:

   https://lore.kernel.org/lkml/20181030160947.19581-1-patrick.bellasi@arm.com/

rebased on v4.20-rc1, which addresses Peter's comments by also adding a
couple of additional cleanup patches on top.

Tests on a 40 CPUs Intel(R) Xeon(R) CPU E5-2690 v2 @ 3.00GHz system
still reports the ~10-15% Execl Throughput improvements after applying
the first patch. Those benefits are not there if we remove the
additional test on "current == p" which Peter was asking about.

I guess the race condition described in the new inline comment I've now
added could be the reason for the additional test being required, but I
did not really verified that guess.
I've just kept both conditions but swapped them since we will probably
be more likely to call cpu_util_without() with a task which is
eventually marked as task_on_rq_queued().

The second patch is pretty simple, while the last one implements what
Peter suggested in the previous review. I did not used something similar
to sub_positive, as suggested by Peter, just because in my tests that
implementation seems to affect negatively the Execl Throughput tests
results by reducing the speedup we get with the proposed version.

Best Patrick

Patrick Bellasi (3):
  sched/fair: util_est: fix cpu_util_wake for execl
  sched/fair: util_est: mask UTIL_AVG_UNCHANGED usages
  sched/fair: add lsub_positive and use it consistently

 kernel/sched/fair.c | 85 ++++++++++++++++++++++++++++++++++-----------
 1 file changed, 64 insertions(+), 21 deletions(-)

-- 
2.18.0