Re: [PATCH v4 5/7] cpufreq: Calculate number of busy CPUs

["Rafael J. Wysocki" <rjw@sisk.pl>]

On Wednesday, June 19, 2013 10:58:32 PM Lukasz Majewski wrote:
> On Wed, 19 Jun 2013 11:01:07 -0700
> Dirk Brandewie <dirk.brandewie@gmail.com> wrote:
> 
> > On 06/19/2013 10:12 AM, Lukasz Majewski wrote:
> > > In the core governor code, per cpu load value is calculated. This
> > > patch uses it to mark processor as a "busy" one, when load value is
> > > higher than 90%.
> > >
> > > New cpufreq sysfs attribute is created (busy_cpus). It is read only
> > > and provides information about number of actually busy CPU.
> > >
> > 
> > What is the intended use of this interface?
> 
> The kernel API would be handy with boost managed by software (like it is
> done with exynos) and with LAB governor.
> 
> The intention is to have two save valves for boost:
> 
> 1. Enable SW controlled boost only when we have just one busy CPU.
> 
> 2. Use the Thermal subsystem to switch off SW managed boost when it
> detects that SoC is overheating.
> 
> The problem with 2 is that, the boost code is compiled in to the cpufreq
> core (no CONFIG_BOOST flag). Thereof we cannot guarantee, that thermal
> would be always enabled (the KConfig select option). 
> I think that thermal subsystem is a suitable place to disable SW boost
> at emergency. However in my opinion this is not enough and precaution
> defined at 1 is needed. 
> 
> I can remove exporting the "busy_cpu" sysfs attribute if you think, that
> it would confuse userspace. Its purpose is mainly informational.
> 
> > 
> > For drivers that have internal governors it will be misleading/wrong
> 
> Would you be so kind and give me an example of such an internal
> governor?
> 
> Maybe I've overlooked some important information/usecase.

Please look at intel_pstate.c.

Thanks,
Rafael


-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.