Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Qais Yousef <qyousef@layalina.io>
To:     Ingo Molnar <mingo@kernel.org>,
        Peter Zijlstra <peterz@infradead.org>,
        Vincent Guittot <vincent.guittot@linaro.org>,
        Dietmar Eggemann <dietmar.eggemann@arm.com>
Cc:     linux-kernel@vger.kernel.org, Lukasz Luba <lukasz.luba@arm.com>,
        Wei Wang <wvw@google.com>, Xuewen Yan <xuewen.yan94@gmail.com>,
        Hank <han.lin@mediatek.com>,
        Jonathan JMChen <Jonathan.JMChen@mediatek.com>,
        Hongyan Xia <hongyan.xia2@arm.com>,
        Qais Yousef <qyousef@layalina.io>
Subject: [PATCH v3 0/3] Fix a couple of corner cases in feec() when using uclamp_max
Date:   Mon, 17 Jul 2023 22:57:14 +0100
Message-Id: <20230717215717.309174-1-qyousef@layalina.io>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

Changes in v3:

	* Fix sign comparison problem in patch 1 (Thanks Vincent!)
	* Simplify comparison and remove function in patch 2 (Thanks Dietmar!)

Changes in v2:

	* Use long instead of unsigned long to keep the comparison simple
	  in spite of being inconsistent with how capacity type.
	* Fix missing termination parenthesis that caused build error.
	* Rebase on latest tip/sched/core and Vincent v5 of Unlink misift patch.

v1 link: https://lore.kernel.org/lkml/20230129161444.1674958-1-qyousef@layalina.io/
v2 link: https://lore.kernel.org/lkml/20230205224318.2035646-1-qyousef@layalina.io/

In v2 Dietmar has raised concerns about limitation in current EM calculations
that can end up packing more tasks on a cluster. While this is not ideal
situation and we need to fix it, but it is another independent problem that is
not introduced by this fix. I don't see a reason why we should couple them
rather than work on each problem independently. The packing behavior in
practice is actually not bad as if something is capped really hard, there's
a desire to keep them on this less performant clusters.

Patch 1 addresses a bug because forcing a task on a small CPU to honour
uclamp_max hint means we can end up with spare_capacity = 0; but the logic is
constructed such that spare_capacity = 0 leads to ignoring this CPU as
a candidate to compute_energy().

Patch 2 addresses a bug due to an optimization in feec() that could lead to
ignoring tasks whose uclamp_max = 0 but task_util(0) != 0.

Patch 3 adds a new tracepoint in compute_energy() as it was helpful in
debugging analyzing these two problems. It's generally useful to trace down
task placement decision based on EAS.

This is based on tip/sched/core.

Qais Yousef (3):
  sched/uclamp: Set max_spare_cap_cpu even if max_spare_cap is 0
  sched/uclamp: Ignore (util == 0) optimization in feec() when
    p_util_max = 0
  sched/tp: Add new tracepoint to track compute energy computation

 include/trace/events/sched.h |  4 ++++
 kernel/sched/core.c          |  1 +
 kernel/sched/fair.c          | 36 ++++++++++++------------------------
 3 files changed, 17 insertions(+), 24 deletions(-)

-- 
2.25.1