linux-kernel - Re: [PATCH] cpufreq: ondemand: Change the calculation of target frequency

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Sat, 01 Jun 2013 19:06:23 +0300
From:	Stratos Karafotis <stratosk@...aphore.gr>
To:	Viresh Kumar <viresh.kumar@...aro.org>
CC:	"Rafael J. Wysocki" <rjw@...k.pl>, linux-kernel@...r.kernel.org,
	cpufreq@...r.kernel.org, linux-pm@...r.kernel.org
Subject: Re: [PATCH] cpufreq: ondemand: Change the calculation of target frequency

On 06/01/2013 05:56 PM, Viresh Kumar wrote:
> On 31 May 2013 22:03, Stratos Karafotis <stratosk@...aphore.gr> wrote:
>> On 05/31/2013 11:51 AM, Viresh Kumar wrote:
>>> I believe you should have removed other users of getavg() in a separate
>>> patch and also cc'd relevant people so that you can some review comments
>>> from  them.
>>
>> I will split the patch in two. If it's OK, I will keep the removal of
>> __cpufreq_driver_getavg in the original patch and move the clean up of
>> APERF/MPERF support in a second patch. I will also cc relevant people.
> 
> Even removal of __cpufreq_driver_getavg() should be done in a separate
> patch, so that it can be reverted easily if required later.

Thanks, Viresh. I will do the removal of that function in a seperate patch.
Should I use a third patch for it? Or should I include it in the patch which
will remove APERF/MPERF support?

>>> "Proportional to load" means C * load, so why is "policy->max / 100" *the* right C?
>>
>> I think, finally(?) I see your point. The right C should be "policy->cpuinfo.max_freq / 100".
> 
> Why are you changing it to cpuinfo.max_freq?? This is fixed once a driver is
> initialized.. but user may request a lower max freq for a governor or policy.
> Which is actually reflected in policy->max I believe.

My initial thought is to have a linear function to calculate the target freq 
proportional to load: (I will use 'C' as the function's slope as Rafael used it) 

freq_target = C * load

For simplicity, let's assume that load is between 0 and 1 as initially is calculated
in governor.
Ideally,  for a load = 0, we should have freq_target = 0 and for load = 1,
freq_target = cpuinfo.max

So, the slope C = cpuinfo.max 

I think, it's matter of definition about what policy->min and policy->max can do.
Should they change the slope C? Or only limit freq_target?
I don't think that the policy->max (or min) should affect HOW (slope C) governor
calculates freq_target but only limit the calculated result.

Maybe, we could have separate tunables to a affect the slope C.

If I'm wrong about the definition of policy->min, policy->max, I would change
the code accordingly.

> Over that why keeping following check is useful anymore?
> 
> if (load_freq > od_tuners->up_threshold)
>      goto max.
> 
> As, if load is over 95, then even policy->max * 95 / 100 will even give almost
> the same freq.
> 

I thought that too. But maybe user selects a lower value for up_threshold.
(For example, up_threshold = 60). In my opinion, we have to keep up_theshold,
to give the possibility to user to have max freq with small loads.

Thanks for your comments!
Stratos 
--
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/