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;
Date:   Wed, 8 May 2019 11:22:22 +0200
From:   Juri Lelli <juri.lelli@redhat.com>
To:     luca abeni <luca.abeni@santannapisa.it>
Cc:     linux-kernel@vger.kernel.org,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        "Rafael J . Wysocki" <rafael@kernel.org>,
        Ingo Molnar <mingo@redhat.com>,
        Peter Zijlstra <peterz@infradead.org>,
        Vincent Guittot <vincent.guittot@linaro.org>,
        "Paul E . McKenney" <paulmck@linux.ibm.com>,
        Joel Fernandes <joel@joelfernandes.org>,
        Quentin Perret <quentin.perret@arm.com>,
        Luc Van Oostenryck <luc.vanoostenryck@gmail.com>,
        Morten Rasmussen <morten.rasmussen@arm.com>,
        Daniel Bristot de Oliveira <bristot@redhat.com>,
        Patrick Bellasi <patrick.bellasi@arm.com>,
        Tommaso Cucinotta <tommaso.cucinotta@santannapisa.it>
Subject: Re: [RFC PATCH 3/6] sched/dl: Try better placement even for deadline
 tasks that do not block
Message-ID: <20190508092222.GH6551@localhost.localdomain>
References: <20190506044836.2914-1-luca.abeni@santannapisa.it>
 <20190506044836.2914-4-luca.abeni@santannapisa.it>
 <20190508080116.GE6551@localhost.localdomain>
 <20190508101414.1c968810@nowhere>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20190508101414.1c968810@nowhere>
User-Agent: Mutt/1.11.3 (2019-02-01)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 08/05/19 10:14, luca abeni wrote:
> Hi Juri,
> 
> On Wed, 8 May 2019 10:01:16 +0200
> Juri Lelli <juri.lelli@redhat.com> wrote:
> 
> > Hi Luca,
> > 
> > On 06/05/19 06:48, Luca Abeni wrote:
> > > From: luca abeni <luca.abeni@santannapisa.it>
> > > 
> > > Currently, the scheduler tries to find a proper placement for
> > > SCHED_DEADLINE tasks when they are pushed out of a core or when
> > > they wake up. Hence, if there is a single SCHED_DEADLINE task
> > > that never blocks and wakes up, such a task is never migrated to
> > > an appropriate CPU core, but continues to execute on its original
> > > core.
> > > 
> > > This commit addresses the issue by trying to migrate a
> > > SCHED_DEADLINE task (searching for an appropriate CPU core) the
> > > first time it is throttled.  
> > 
> > Why we failed to put the task on a CPU with enough (max) capacity
> > right after it passed admission control? The very first time the task
> > was scheduled I mean.
> 
> I think the currently executing task cannot be pushed out of a
> CPU/core, right?
> 
> So, if a task switches from SCHED_OTHER to SCHED_DEADLINE while it is
> executing on a fast core, the only way to migrate it would be to
> preempt it (by using the stop_sched_class, I think), no?
> (the typical situation here is a "cpu hog" task that switches from
> SCHED_OTHER to SCHED_DEADLINE, and it is the only SCHED_DEADLINE
> task... The task never blocks, so push/pull functions are never invoked)
> 
> Or am I missing something?

OK, but "ideally" we should not be waiting to it to be throttled, right?

I wonder if you could queue a balance callback in switched_to_dl() (from
check_class_changed()), so that it is picked up by balance_callback()
before setscheduler returns.