Received-SPF: pass (google.com: domain of linux-kernel+bounces-162216-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) client-ip=147.75.80.249;
Date: Mon, 29 Apr 2024 14:14:22 +0200
From: Peter Zijlstra <peterz@infradead.org>
To: Mike Galbraith <efault@gmx.de>
Cc: K Prateek Nayak <kprateek.nayak@amd.com>, mingo@redhat.com,
	juri.lelli@redhat.com, vincent.guittot@linaro.org,
	dietmar.eggemann@arm.com, rostedt@goodmis.org, bsegall@google.com,
	mgorman@suse.de, bristot@redhat.com, vschneid@redhat.com,
	linux-kernel@vger.kernel.org, wuyun.abel@bytedance.com,
	tglx@linutronix.de, Chen Yu <yu.c.chen@intel.com>,
	Oliver Sang <oliver.sang@intel.com>
Subject: Re: [RFC][PATCH 08/10] sched/fair: Implement delayed dequeue
Message-ID: <20240429121422.GU30852@noisy.programming.kicks-ass.net>
References: <ec6f811b9977eeef1b3f1b3fb951fda066fd95f5.camel@gmx.de>
 <14330cf4-8d9e-1e55-7717-653b800e5cee@amd.com>
 <747627a1414f1f33d0c237f555494149d6937800.camel@gmx.de>
 <2b9f7617f2b2130bb6270504ec3858f15d463f1d.camel@gmx.de>
 <20240425112855.GF21980@noisy.programming.kicks-ass.net>
 <20240426105607.GK12673@noisy.programming.kicks-ass.net>
 <20240426111607.GL12673@noisy.programming.kicks-ass.net>
 <18f557cd69fc65f5b43daddd8ad3e43278dfa44e.camel@gmx.de>
 <8d2fd91db108428680cd4ed988b36cf3f579c5ec.camel@gmx.de>
 <b344235ed5e2af6c573ff371d3687f0fa7bc5c5e.camel@gmx.de>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <b344235ed5e2af6c573ff371d3687f0fa7bc5c5e.camel@gmx.de>

On Sun, Apr 28, 2024 at 06:32:49PM +0200, Mike Galbraith wrote:
> On Sat, 2024-04-27 at 08:42 +0200, Mike Galbraith wrote:
> > On Fri, 2024-04-26 at 18:03 +0200, Mike Galbraith wrote:
> > > fwiw, tbench liked the previous version better.
> >
> > The culprit lies somewhere in the new PREEMPT_SHORT patch.
> 
> <squint>
> 
> Ah, moving the curr eligibility check into pick_curr() left an
> ineligible curr to be seen/used further down in pick_eevdf().
> 
> found:
>         if (!best || (curr && entity_before(curr, best)))
>                 best = curr;

Hmm yes, over aggressive cleanup that. Let me try again.

> Nit: PREEMPT_SHORT depending on RUN_TO_PARITY looks odd.

The thinking was that without RUN_TO_PARITY we'll always do a full pick
and it will always pick a (new) shorter deadline task over current.

The PREEMPRT_SHORT thing really is an exception to avoid RUN_TO_PARITY
from ruining latency game for short tasks.