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;
MIME-Version: 1.0
References: <cb6c406e-1431-fcfd-ef82-87259760ead9@joelfernandes.org>
 <20220930134931.mpopdvri4xuponw2@wubuntu> <CALUeGD0yEimCRS2TQfZTUD2kwamyTZwM9Y9D7C=Xsd9t9mWsaw@mail.gmail.com>
 <a4a7a4de-c58d-d667-a4b3-0f7bfb2b09f1@joelfernandes.org> <CALUeGD21QY+-6oLUztdecg5C8AX9xToxpGpxx5M5D9VnhSkVMg@mail.gmail.com>
In-Reply-To: <CALUeGD21QY+-6oLUztdecg5C8AX9xToxpGpxx5M5D9VnhSkVMg@mail.gmail.com>
From:   Joel Fernandes <joel@joelfernandes.org>
Date:   Fri, 30 Sep 2022 14:45:10 -0400
Message-ID: <CAEXW_YQQ=CfQEOkXK-VFJtZTdY=_TsWz=Ndgg_BgJumFHJktkQ@mail.gmail.com>
Subject: Re: Sum of weights idea for CFS PI
To:     Youssef Esmat <youssefesmat@google.com>
Cc:     Qais Yousef <qais.yousef@arm.com>,
        Peter Zijlstra <peterz@infradead.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Steven Rostedt <rostedt@goodmis.org>, juri.lelli@redhat.com,
        vincent.guittot@linaro.org,
        Dietmar Eggemann <dietmar.eggemann@arm.com>,
        Thomas Gleixner <tglx@linutronix.de>, bristot@redhat.com,
        clark.williams@gmail.com, bigeasy@linutronix.de,
        "Paul E. McKenney" <paulmck@kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Fri, Sep 30, 2022 at 2:10 PM Youssef Esmat <youssefesmat@google.com> wrote:
[..]
> > > Hi Everyone!
> >
> > Hi Youssef,
> >
> > (Youssef is new to LKML though in no way new to OS or software development. I
> > gave him the usual 'dont-top-post' chat already - fyi).
> >
> > > I am not sure we should care about A's sleeping pattern. The case we
> > > care about is when A is running or wants to run but can't because it
> > > is blocked on C. In that case C should get the weight of A as if A was
> > > running.
> >
> > Just to clarify - Youssef did mean sum of weights of different things in the
> > chain, and not just weights (he confirmed on chat that that's what he meant).
> >
>
> Yeah thanks for clarifying, I meant that C should get the sum of
> weights as if A was running (3/5 in your example) since in this
> segment of time A would have been running if it was not blocked on the
> lock. I think it's safe to ignore the average and just use the sum of

For the onlooker, we are talking about the classical case of priority
inversion involving 3 tasks A, B and C which can be expanded to a
chain of tasks. Highest prio A blocks on a lock that lowest prio C
holds, while an unrelated medium prio B blocks C (or reduces progress
of it as in the case of CFS).

On the note of "A would have been running if it was not blocked on the
lock". I think that would be an assumption - we don't know if A would
be running. We only know the past, not the future. A could very well
make an I/O request for example. Hence there could be a need to use
A's past utilization, right?

thanks,

 - Joel