[PATCH 1/2] OPP: Add API to update EM after adjustment of voltage for OPPs

[Lukasz Luba]

On 12/21/23 07:28, Xuewen Yan wrote:
> On Wed, Dec 20, 2023 at 7:02 PM Lukasz Luba <lukasz.luba at arm.com> wrote:
>>
>> There are device drivers which can modify voltage values for OPPs. It
>> could be due to the chip binning and those drivers have specific chip
>> knowledge about this. This adjustment can happen after Energy Model is
>> registered, thus EM can have stale data about power.
>>
>> Introduce new API function which can be used by device driver which
>> adjusted the voltage for OPPs. The implementation takes care about
>> calculating needed internal details in the new EM table ('cost' field).
>> It plugs in the new EM table to the framework so other subsystems would
>> use the correct data.
>>
>> Signed-off-by: Lukasz Luba <lukasz.luba at arm.com>
>> ---
>>   drivers/opp/of.c       | 69 ++++++++++++++++++++++++++++++++++++++++++
>>   include/linux/pm_opp.h |  6 ++++
>>   2 files changed, 75 insertions(+)
>>
>> diff --git a/drivers/opp/of.c b/drivers/opp/of.c
>> index 81fa27599d58..992434c0b711 100644
>> --- a/drivers/opp/of.c
>> +++ b/drivers/opp/of.c
>> @@ -1596,3 +1596,72 @@ int dev_pm_opp_of_register_em(struct device *dev, struct cpumask *cpus)
>>          return ret;
>>   }
>>   EXPORT_SYMBOL_GPL(dev_pm_opp_of_register_em);
>> +
>> +/**
>> + * dev_pm_opp_of_update_em() - Update Energy Model with new power values
>> + * @dev                : Device for which an Energy Model has to be registered
>> + *
>> + * This uses the "dynamic-power-coefficient" devicetree property to calculate
>> + * power values for EM. It uses the new adjusted voltage values known for OPPs
>> + * which have changed after boot.
>> + */
>> +int dev_pm_opp_of_update_em(struct device *dev)
>> +{
>> +       struct em_perf_table __rcu *runtime_table;
>> +       struct em_perf_state *table, *new_table;
>> +       struct em_perf_domain *pd;
>> +       int ret, table_size, i;
>> +
>> +       if (IS_ERR_OR_NULL(dev))
>> +               return -EINVAL;
>> +
>> +       pd = em_pd_get(dev);
>> +       if (!pd) {
>> +               dev_warn(dev, "Couldn't find Energy Model %d\n", ret);
>> +               return -EINVAL;
>> +       }
>> +
>> +       runtime_table = em_allocate_table(pd);
>> +       if (!runtime_table) {
>> +               dev_warn(dev, "new EM allocation failed\n");
>> +               return -ENOMEM;
>> +       }
>> +
>> +       new_table = runtime_table->state;
>> +
>> +       table = em_get_table(pd);
>> +       /* Initialize data based on older EM table */
>> +       table_size = sizeof(struct em_perf_state) * pd->nr_perf_states;
>> +       memcpy(new_table, table, table_size);
>> +
>> +       em_put_table();
>> +
>> +       /* Update power values which might change due to new voltage in OPPs */
>> +       for (i = 0; i < pd->nr_perf_states; i++) {
>> +               unsigned long freq = new_table[i].frequency;
>> +               unsigned long power;
>> +
>> +               ret = _get_power(dev, &power, &freq);
>> +               if (ret)
>> +                       goto failed;
> 
> Need we use the EM_SET_ACTIVE_POWER_CB(em_cb, _get_power) and call
> em_cb->active_power?
> 

No, not in this case. It's not like registration of EM, when there
is a need to also pass the callback function. As you can see this code
operates locally and the call _get_power() just simply gets the
power in straight way. Later the whole 'runtime_table' is passed to the
EM framework to 'swap' the pointers under RCU.

Thanks Xuewen for having a look at this!