Re: [PATCH 1/3] PM / OPP: Add support for descending order for cpufreq table

[Nishanth Menon <nm@ti.com>]

On Wed, May 7, 2014 at 8:22 PM, Jonghwan Choi <jhbird.choi@samsung.com> wrote:
>> @Jonghwan: Please consider doing this:
>> - Don't play with the order of frequencies in table.
>> - Instead initialize .driver_data filed with values that you need to write
>> in the registers for all frequencies. i.e. 0 for highest frequency and
>> FREQ_COUNT-1 for lowest one.
>
> -> For that, I changed like this.
> For initializing .driver_data, I changed dev_pm_opp_init_cpufreq_table function().
>
>
> --- a/drivers/base/power/opp.c
> +++ b/drivers/base/power/opp.c
> @@ -622,12 +622,12 @@ EXPORT_SYMBOL_GPL(dev_pm_opp_disable);
>   * or in contexts where mutex locking cannot be used.
>   */
>  int dev_pm_opp_init_cpufreq_table(struct device *dev,
> -                           struct cpufreq_frequency_table **table)
> +               struct cpufreq_frequency_table **table, int order)
>  {
>         struct device_opp *dev_opp;
>         struct dev_pm_opp *opp;
>         struct cpufreq_frequency_table *freq_table;
> -       int i = 0;
> +       int i = 0, index = 0;
>
>         /* Pretend as if I am an updater */
>         mutex_lock(&dev_opp_list_lock);
> @@ -649,16 +649,22 @@ int dev_pm_opp_init_cpufreq_table(struct device *dev,
>                 return -ENOMEM;
>         }
>
> +       if (OPP_TABLE_ORDER_DESCENDING == order)
> +               index = dev_pm_opp_get_opp_count(dev) - 1;
> +
>         list_for_each_entry(opp, &dev_opp->opp_list, node) {
>                 if (opp->available) {
> -                       freq_table[i].driver_data = i;
> +                       if (OPP_TABLE_ORDER_DESCENDING == order)
> +                               freq_table[i].driver_data = index--;
> +                       else
> +                               freq_table[i].driver_data = index++;
>                         freq_table[i].frequency = opp->rate / 1000;
>                         i++;
>                 }
>         }
>         mutex_unlock(&dev_opp_list_lock);
>
> -       freq_table[i].driver_data = i;
> +       freq_table[i].driver_data = index;
>         freq_table[i].frequency = CPUFREQ_TABLE_END;
>
>         *table = &freq_table[0];
>
>
> Is it acceptiable?

Personally, I feel that filling up driver_data should be left to the
driver(caller of dev_pm_opp_init_cpufreq_table). for example providing
a function pointer which decides what that value should be (be it
index or some magical register value).. Viresh might have better
opinions.

Regards,
Nishanth Menon