Message-id: <20130704074307.282a53dc@amdc308.digital.local>
Date:	Thu, 04 Jul 2013 07:43:07 +0200
From:	Lukasz Majewski <l.majewski@...sung.com>
To:	Viresh Kumar <viresh.kumar@...aro.org>
Cc:	"Rafael J. Wysocky" <rjw@...k.pl>,
	"cpufreq@...r.kernel.org" <cpufreq@...r.kernel.org>,
	Linux PM list <linux-pm@...r.kernel.org>,
	Vincent Guittot <vincent.guittot@...aro.org>,
	Jonghwa Lee <jonghwa3.lee@...sung.com>,
	Myungjoo Ham <myungjoo.ham@...sung.com>,
	linux-kernel <linux-kernel@...r.kernel.org>,
	Lukasz Majewski <l.majewski@...ess.pl>,
	Andre Przywara <andre.przywara@...aro.org>,
	Daniel Lezcano <daniel.lezcano@...aro.org>,
	Kukjin Kim <kgene.kim@...sung.com>,
	Zhang Rui <rui.zhang@...el.com>,
	Eduardo Valentin <eduardo.valentin@...com>
Subject: Re: [PATCH v4 5/7] cpufreq: Calculate number of busy CPUs

On Thu, 04 Jul 2013 10:36:53 +0530, Viresh Kumar wrote:
> Hi Lukasz,
> 
> Sorry for being late. Actually I didn't had an answer to your mail
> and wanted to go through it with some fresh mind. This is my
> first mail this morning, lets see if I can bring something good
> into the discussion.
> 
> On 1 July 2013 13:45, Lukasz Majewski <l.majewski@...sung.com> wrote:
> > Does anybody have any idea here? As written above, thermal is
> > suitable to disable boost.
> 
> See, one thing is very clear. User space applications aren't
> responsible for enabling boost again and again. There has to be a
> internal mechanism inside kernel for that.
> 
> > I'd like to bring those three options under discussion:
> >
> > 1. boost attr is always exported -> do not enable boost
> > automatically when disabled by thermal (as it was proposed at v4).
> 
> So, that's a problem. I see one more solution to that.
> - Create another Macro in cpufreq.c which would contain the time
> after which we will autoenable boost.
> - So, suppose thermal disabled it due to high temperature (Lets not
> change value of sysfs variable boost_enable, but create another
> variable like: skip_boost: which means skip boost temporarily).
> - Thermal would enable this variable skip_boost.
> - Then we will continue to get requests for next frequency and will
> check eveytime if we have exceeded time for autoenabling boost.
> - If yes, then we disable this variable and start boosting again..
> - Then thermal can disable it again later.
> 
> This variable (time for autoenable) looks to be more platform
> dependent for now, but lets don't make it like that unless somebody
> needs it.
> 
> > 2. boost attr is always exported -> find a way to enable boost after
> >    emergency disablement when thermal detects overheating (newest
> >    proposition).
> 
> My solution above probably.

This is a possible solution, but I've already modified thermal code a
bit and found a solution for the problem.

I use thermal workqueue (which is already in place anyway) to enable the
boost again.
Due to that I can provide behaviour similar to HW controlled boost.

Patches with this solution are already prepared. I will post them in a
few hours. Ok?

> 
> > 3. boost attr only exported at x86 (when supported)
> >    boost attr NOT exported via sysfs for SW controlled boost (e.g.
> >    Exynos ARM).
> >
> >    Then we only enable/disable boost at kernel and don't need to
> > take care of the user space interaction. This scenario is my use
> > case. I hadn't planned to expose boost to userspace and use it with
> > LAB as a kernel API.
> 
> Userspace must have control of this feature after kernel is built.
> That kernel image may run for ever without changing in a product.
> 
> @Rafael: How crazy do you think my solution is? :)


-- 
Best regards,

Lukasz Majewski

Samsung R&D Institute Poland (SRPOL) | Linux Platform Group
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists