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:   "Rafael J. Wysocki" <rjw@rjwysocki.net>
To:     Thomas Ilsche <thomas.ilsche@tu-dresden.de>
Cc:     Peter Zijlstra <peterz@infradead.org>,
        Linux PM <linux-pm@vger.kernel.org>,
        Frederic Weisbecker <fweisbec@gmail.com>,
        Thomas Gleixner <tglx@linutronix.de>,
        Paul McKenney <paulmck@linux.vnet.ibm.com>,
        Doug Smythies <dsmythies@telus.net>,
        Rik van Riel <riel@surriel.com>,
        Aubrey Li <aubrey.li@linux.intel.com>,
        Mike Galbraith <mgalbraith@suse.de>,
        LKML <linux-kernel@vger.kernel.org>
Subject: Re: [RFT][PATCH v7 6/8] sched: idle: Select idle state before stopping the tick
Date:   Wed, 28 Mar 2018 00:10:37 +0200
Message-ID: <4198010.6ArFqS34NK@aspire.rjw.lan>
In-Reply-To: <6462e44a-e207-6b97-22bf-ad4aed69afc2@tu-dresden.de>
References: <2390019.oHdSGtR3EE@aspire.rjw.lan> <2249320.0Z4q8AXauv@aspire.rjw.lan> <6462e44a-e207-6b97-22bf-ad4aed69afc2@tu-dresden.de>
MIME-Version: 1.0
Content-Transfer-Encoding: 7Bit
Content-Type: text/plain; charset="us-ascii"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tuesday, March 27, 2018 11:50:02 PM CEST Thomas Ilsche wrote:
> On 2018-03-20 16:45, Rafael J. Wysocki wrote:
> > From: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
> > 
> > In order to address the issue with short idle duration predictions
> > by the idle governor after the tick has been stopped, reorder the
> > code in cpuidle_idle_call() so that the governor idle state selection
> > runs before tick_nohz_idle_go_idle() and use the "nohz" hint returned
> > by cpuidle_select() to decide whether or not to stop the tick.
> > 
> > This isn't straightforward, because menu_select() invokes
> > tick_nohz_get_sleep_length() to get the time to the next timer
> > event and the number returned by the latter comes from
> > __tick_nohz_idle_enter().  Fortunately, however, it is possible
> > to compute that number without actually stopping the tick and with
> > the help of the existing code.
> 
> I think something is wrong with the new tick_nohz_get_sleep_length.
> It seems to return a value that is too large, ignoring immanent
> non-sched timer.

That's a very useful hint, let me have a look.

> I tested idle-loop-v7.3. It looks very similar to my previous results
> on the first idle-loop-git-version [1]. Idle and traditional synthetic
> powernightmares are mostly good.

OK

> But it selects too deep C-states for short idle periods, which is bad
> for power consumption [2].

That still needs to be improved, then.

> I tracked this down with additional tests using
> __attribute__((optimize("O0"))) menu_select
> and perf probe. With this the behavior seems slightly different, but it
> shows that data->next_timer_us is:
> v4.16-rc6: the expected ~500 us [3]
> idle-loop-v7.3: many milliseconds to minutes [4].
> This leads to the governor to wrongly selecting C6.
> 
> Checking with 372be9e and 6ea0577, I can confirm that the change is
> introduced by this patch.

Yes, that's where the most intrusive reordering happens.

Thanks for the feedback!