Re: cpufreq: Unable to set cpufreq to maximum

[Stefan Wahren <stefan.wahren@i2se.com>]

Am 06.10.2014 um 11:26 schrieb Viresh Kumar:
> On 6 October 2014 14:43, Stefan Wahren <stefan.wahren@i2se.com> wrote:
>> Sorry, about the confusion. The main problem about setting cpufreq to
>> maximum has been fixed by DT settings. Above i talked about an idea to
>> fix the confusing warning after fixing the main problem:
>>
>> root@duckbill:~# modprobe cpufreq-cpu0
>> [   48.573808] cpufreq: __cpufreq_add_dev: CPU0: Running at unlisted
>> freq: 454736 KHz
>> [   48.604871] cpufreq: __cpufreq_add_dev: CPU0: Unlisted initial
>> frequency changed to: 454737 KHz
>>
>> But the best solution would be a more intelligent function like
>> dev_pm_opp_find_freq_nearest() which find the OPP with the smallest
>> frequency distance. So nobody of the DTS users has to care about
>> frequency rounding.
>>
>> Currently i'm trying to adapt the DTS file to the driver and that's not
>> the intention of devicetree.
> It looks that one of us hasn't understood how things are working :)

I think i know him ;-)

>
>  So here is my part of theory:
> - Search for CPUFREQ_NEED_INITIAL_FREQ_CHECK in cpufreq.c
> - When starting cpufreq core for every CPU we check if the current
> frequency at which is running is mentioned in DT. If not, we throw above
> warnings and fix the freq.

Okay i.MX28 cpu is running initial on 454736842 Hz. 454736 kHz in the DT
will make cpufreq core happy, but cpufreq-cpu0 will select next lower
frequency. 454737 kHz in the DT will make cpufreq sad, but it's
selectable by cpufreq-cpu0.

Is it possible cpufreq core round the clock frequency down?

> - Until this time cpufreq-driver's ->target_index() routine hasn't come
> into picture and so NO opp routine will fix that..
>
> Does this sound reasonable ?

Thanks for your explanation.

Stefan