[PATCH] cpufreq: Avoid unnecessary frequency updates due to mismatch

[Vincent Guittot]

On Thu, 5 May 2022 at 09:44, Viresh Kumar <viresh.kumar at linaro.org> wrote:
>
> On 05-05-22, 09:28, Vincent Guittot wrote:
> > On Wed, 4 May 2022 at 10:21, Viresh Kumar <viresh.kumar at linaro.org> wrote:
> > >
> > > For some platforms, the frequency returned by hardware may be slightly
> > > different from what is provided in the frequency table. For example,
> >
> > Do you have more details ?
>
> This is where the problem was discussed.
>
> https://lore.kernel.org/lkml/20220422075239.16437-8-rex-bc.chen@mediatek.com/

Thanks for the link

>
> > Do you mean that between 2 consecutives reads you can get either
> > 500Mhz or 499Mhz ?
>
> No, the hardware always returns something like 499,999,726 Hz, but the

Part of your problem is that cpufreq use khz whereas clock uses hz

Would it be better to do something like below in cpufreq_generic_get

(clk_get_rate(policy->clk) + 500) / 1000

so you round to closest instead of always floor rounding


> OPP table contains the value 500 MHz. The field policy->cur is set
> based on opp table eventually (target_index) and so contains 500MHz,
> almost always. But when cpufreq_get() is called, it finds the current
> freq is 499 MHz, instead of 500 MHz. And so the issue.
>
> > Or is it a fixed mismatch between the table and the freq returned by HW ?
>
> Yes.
>
> --
> viresh