Re: [PATCH] opp: of: Allow lazy-linking of required-opps to non genpd

[Chanwoo Choi]

On 6/16/21 6:09 PM, Viresh Kumar wrote:
> On 16-06-21, 17:47, Chanwoo Choi wrote:
>> On 6/16/21 4:55 PM, Viresh Kumar wrote:
>>> On 16-06-21, 13:33, Hsin-Yi Wang wrote:
>>>> Don't limit required_opp_table to genpd only. One possible use case is
>>>> cpufreq based devfreq governor, which can use required-opps property to
>>>> derive devfreq from cpufreq.
>>>>
>>>> Suggested-by: Chanwoo Choi <cw00.choi@xxxxxxxxxxx>
>>>> Signed-off-by: Hsin-Yi Wang <hsinyi@xxxxxxxxxxxx>
>>>> ---
>>>> This is tested with the non genpd case mt8183-cci with passive
>>>> governor[1].
>>>> [1] https://patchwork.kernel.org/project/linux-mediatek/patch/1616499241-4906-2-git-send-email-andrew-sh.cheng@xxxxxxxxxxxx/
>>>> ---
>>>>  drivers/opp/of.c | 20 +-------------------
>>>>  1 file changed, 1 insertion(+), 19 deletions(-)
>>>>
>>>> diff --git a/drivers/opp/of.c b/drivers/opp/of.c
>>>> index aa75a1caf08a3..9573facce53a5 100644
>>>> --- a/drivers/opp/of.c
>>>> +++ b/drivers/opp/of.c
>>>> @@ -201,17 +201,6 @@ static void _opp_table_alloc_required_tables(struct opp_table *opp_table,
>>>>  			lazy = true;
>>>>  			continue;
>>>>  		}
>>>> -
>>>> -		/*
>>>> -		 * We only support genpd's OPPs in the "required-opps" for now,
>>>> -		 * as we don't know how much about other cases. Error out if the
>>>> -		 * required OPP doesn't belong to a genpd.
>>>> -		 */
>>>> -		if (!required_opp_tables[i]->is_genpd) {
>>>> -			dev_err(dev, "required-opp doesn't belong to genpd: %pOF\n",
>>>> -				required_np);
>>>> -			goto free_required_tables;
>>>> -		}
>>>>  	}
>>>>  
>>>>  	/* Let's do the linking later on */
>>>> @@ -379,13 +368,6 @@ static void lazy_link_required_opp_table(struct opp_table *new_table)
>>>>  	struct dev_pm_opp *opp;
>>>>  	int i, ret;
>>>>  
>>>> -	/*
>>>> -	 * We only support genpd's OPPs in the "required-opps" for now,
>>>> -	 * as we don't know much about other cases.
>>>> -	 */
>>>> -	if (!new_table->is_genpd)
>>>> -		return;
>>>> -
>>>>  	mutex_lock(&opp_table_lock);
>>>>  
>>>>  	list_for_each_entry_safe(opp_table, temp, &lazy_opp_tables, lazy) {
>>>> @@ -873,7 +855,7 @@ static struct dev_pm_opp *_opp_add_static_v2(struct opp_table *opp_table,
>>>>  		return ERR_PTR(-ENOMEM);
>>>>  
>>>>  	ret = _read_opp_key(new_opp, opp_table, np, &rate_not_available);
>>>> -	if (ret < 0 && !opp_table->is_genpd) {
>>>> +	if (ret < 0) {
>>>>  		dev_err(dev, "%s: opp key field not found\n", __func__);
>>>>  		goto free_opp;
>>>>  	}
>>>
>>> Plus this and few changes to commit log.
>>>
>>> diff --git a/drivers/opp/core.c b/drivers/opp/core.c
>>> index e366218d6736..b335c077f215 100644
>>> --- a/drivers/opp/core.c
>>> +++ b/drivers/opp/core.c
>>> @@ -893,6 +893,16 @@ static int _set_required_opps(struct device *dev,
>>>         if (!required_opp_tables)
>>>                 return 0;
>>>  
>>> +       /*
>>> +        * We only support genpd's OPPs in the "required-opps" for now, as we
>>> +        * don't know much about other use cases. Error out if the required OPP
>>> +        * doesn't belong to a genpd.
>>> +        */
>>> +       if (unlikely(!required_opp_tables[0]->is_genpd)) {
>>> +               dev_err(dev, "required-opps don't belong to a genpd\n");
>>> +               return -ENOENT;
>>> +       }
>>> +
>>
>> If you add this checking statement, I think that 
>> when using dev_pm_opp_set_rate with required-opp property, it will be failed.
> 
> Yes, that is exactly what I am trying to do here. Hsin already
> confirmed that you guys won't use this API, isn't ?
> 
> The point here is that the _set_required_opps() function only updates
> the performance state of genpds today. So it won't work for you guys
> anyway.

The devfreq driver(exynos-bus.c) has used the dev_pm_opp_set_rate()
and used the passive governor without the required-opp property.

I have a plan to use the required-opp property
between devfreq drivers (exynos-bus.c) with dev_pm_opp_set_rate().

I'll support them on later if this approach doesn't break the any
rule of required-opp property.

-- 
Best Regards,
Chanwoo Choi
Samsung Electronics