[PATCH 07/11] PM / devfreq: check get_dev_status before start monitor

Chanwoo Choi cw00.choi at samsung.com
Thu Mar 18 09:54:26 GMT 2021


On 3/18/21 5:03 PM, Dong Aisheng wrote:
> Hi Chanwoo,
> 
> On Sat, Mar 13, 2021 at 2:45 PM Dong Aisheng <dongas86 at gmail.com> wrote:
>>
>> On Sat, Mar 13, 2021 at 12:09 AM Chanwoo Choi <cwchoi00 at gmail.com> wrote:
>>>
>>> On 21. 3. 12. 오후 7:57, Dong Aisheng wrote:
>>>> On Thu, Mar 11, 2021 at 2:54 PM Chanwoo Choi <cw00.choi at samsung.com> wrote:
>>>>>
>>>>> On 3/10/21 1:56 PM, Dong Aisheng wrote:
>>>>>> On Wed, Mar 10, 2021 at 11:08 AM Chanwoo Choi <cw00.choi at samsung.com> wrote:
>>>>>>>
>>>>>>> On 3/10/21 11:56 AM, Dong Aisheng wrote:
>>>>>>>> On Wed, Mar 10, 2021 at 12:12 AM Chanwoo Choi <cwchoi00 at gmail.com> wrote:
>>>>>>>>>
>>>>>>>>> On 21. 3. 10. 오전 12:58, Chanwoo Choi wrote:
>>>>>>>>>> On 21. 3. 9. 오후 9:58, Dong Aisheng wrote:
>>>>>>>>>>> The devfreq monitor depends on the device to provide load information
>>>>>>>>>>> by .get_dev_status() to calculate the next target freq.
>>>>>>>>>>>
>>>>>>>>>>> And this will cause changing governor to simple ondemand fail
>>>>>>>>>>> if device can't support.
>>>>>>>>>>>
>>>>>>>>>>> Signed-off-by: Dong Aisheng <aisheng.dong at nxp.com>
>>>>>>>>>>> ---
>>>>>>>>>>>    drivers/devfreq/devfreq.c                 | 10 +++++++---
>>>>>>>>>>>    drivers/devfreq/governor.h                |  2 +-
>>>>>>>>>>>    drivers/devfreq/governor_simpleondemand.c |  3 +--
>>>>>>>>>>>    3 files changed, 9 insertions(+), 6 deletions(-)
>>>>>>>>>>>
>>>>>>>>>>> diff --git a/drivers/devfreq/devfreq.c b/drivers/devfreq/devfreq.c
>>>>>>>>>>> index 7231fe6862a2..d1787b6c7d7c 100644
>>>>>>>>>>> --- a/drivers/devfreq/devfreq.c
>>>>>>>>>>> +++ b/drivers/devfreq/devfreq.c
>>>>>>>>>>> @@ -482,10 +482,13 @@ static void devfreq_monitor(struct work_struct
>>>>>>>>>>> *work)
>>>>>>>>>>>     * to be called from governor in response to DEVFREQ_GOV_START
>>>>>>>>>>>     * event when device is added to devfreq framework.
>>>>>>>>>>>     */
>>>>>>>>>>> -void devfreq_monitor_start(struct devfreq *devfreq)
>>>>>>>>>>> +int devfreq_monitor_start(struct devfreq *devfreq)
>>>>>>>>>>>    {
>>>>>>>>>>>        if (IS_SUPPORTED_FLAG(devfreq->governor->flags, IRQ_DRIVEN))
>>>>>>>>>>> -        return;
>>>>>>>>>>> +        return 0;
>>>>>>>>>>> +
>>>>>>>>>>> +    if (!devfreq->profile->get_dev_status)
>>>>>>>>>>> +        return -EINVAL;
>>>>>>>>>
>>>>>>>>> Again, I think that get_dev_status is not used for all governors.
>>>>>>>>> So that it cause the governor start fail. Don't check whether
>>>>>>>>> .get_dev_status is NULL or not.
>>>>>>>>>
>>>>>>>>
>>>>>>>> I'm not quite understand your point.
>>>>>>>> it is used by governor_simpleondemand.c and tegra_devfreq_governor.
>>>>>>>> get_target_freq -> devfreq_update_stats -> get_dev_status
>>>>>>>
>>>>>>> The devfreq can add the new governor by anyone.
>>>>>>> So these functions like devfreq_monitor_* have to support
>>>>>>> the governors and also must support the governor to be added
>>>>>>> in the future.
>>>>>>
>>>>>> Yes, but devfreq_monitor_* is only used by polling mode, right?
>>>>>> The governor using it has to implement get_dev_status unless
>>>>>> there's an exception in the future.
>>>>>>
>>>>>> Currently this patch wants to address the issue that user can switch
>>>>>> to ondemand governor (polling mode) by sysfs even devices does
>>>>>> not support it (no get_dev_status implemented).
>>>>>
>>>>> As I commented, I'll fix this issue. If devfreq driver doesn't implement
>>>>> the .get_dev_status, don't show it via available_governors. I think that
>>>>> it is fundamental solution to fix this issue.
>>>>
>>>> Sounds good
>>>>
>>>>> So on this version,
>>>>> don't add the this conditional statement on this function
>>>>>
>>>>
>>>> Almost all this patch did is adding a checking for get_dev_status.
>>>> So do you mean drop this patch?
>>>> I wonder it's still a necessary checking to explicitly tell devfreq monitor
>>>> users that get_dev_status is needed during governor startup.
>>>
>>> I think that the it is enough to check .get_dev_status in
>>> devfreq_update_stats. We have to check it on where it is used.
>>>
>>
>> I think the drawback of only checking .get_dev_status in
>> devfreq_update_stats is:
>> 1. devfreq will still register successfully and ondemand governor starts ok
>> 2. ondemand governor will still be shown in sysfs which is something
>> you want to fix
>> 3. devfreq will end up printing endless error messages in devfreq_monitor worker
>>     "dvfs failed with (%d) error" as the possible missing .get_dev_status

I think that devfreq_monitor_start have to handle only work instance.
This approach is too considering the deep check list.
I want to resolve this periodical error log with different solution.

Actually, we have to reject the registration of devfreq device
when calling devfreq_add_device instead of checking .get_dev_status
in devfreq_monitor_start().


>>
>> So i wonder if you don't like changing the common devfreq_monitor_start in order
>> to make it look common for all governors, then we probably still need
>> to fix it in
>> ondemand governor in order to avoid the possible above issues.
>>
>> static int devfreq_simple_ondemand_handler(struct devfreq *devfreq,
>>                                 unsigned int event, void *data)
>> {
>>         switch (event) {
>>         case DEVFREQ_GOV_START:
>>                 if (!devfreq->profile->get_dev_status)
>>                         return -EINVAL;
>>
>>                 return devfreq_monitor_start(devfreq);
>> ...
>> }
>>
>> How do you think?
> 
> Any feedback?
> 
> I'm waiting for your confirmation whether dropping this one,
> then I can re-sent the series.
> 
> Regards
> Aisheng
> 
>>
>> Regards
>> Aisheng
>>
>>
>>>>
>>>>> And on next version, please use the capital letter for first character
>>>>> on patch title as following:
>>>>>
>>>>> - PM / devfreq: Check get_dev_status before start monitor
>>>>>
>>>>
>>>> Okay to me.
>>>> Thanks for the suggestion.
>>>>
>>>> Regards
>>>> Aisheng
>>>>
>>>>>>
>>>>>> Regards
>>>>>> Aisheng
>>>>>>
>>>>>>>
>>>>>>>>
>>>>>>>> Without checking, device can switch to ondemand governor if it does not support.
>>>>>>>>
>>>>>>>> Am i missed something?
>>>>>>>>
>>>>>>>> Regards
>>>>>>>> Aisheng
>>>>>>>>
>>>>>>>>>>>        switch (devfreq->profile->timer) {
>>>>>>>>>>>        case DEVFREQ_TIMER_DEFERRABLE:
>>>>>>>>>>> @@ -495,12 +498,13 @@ void devfreq_monitor_start(struct devfreq *devfreq)
>>>>>>>>>>>            INIT_DELAYED_WORK(&devfreq->work, devfreq_monitor);
>>>>>>>>>>>            break;
>>>>>>>>>>>        default:
>>>>>>>>>>> -        return;
>>>>>>>>>>> +        return -EINVAL;
>>>>>>>>>>>        }
>>>>>>>>>>>        if (devfreq->profile->polling_ms)
>>>>>>>>>>>            queue_delayed_work(devfreq_wq, &devfreq->work,
>>>>>>>>>>>                msecs_to_jiffies(devfreq->profile->polling_ms));
>>>>>>>>>>> +    return 0;
>>>>>>>>>>>    }
>>>>>>>>>>>    EXPORT_SYMBOL(devfreq_monitor_start);
>>>>>>>>>>> diff --git a/drivers/devfreq/governor.h b/drivers/devfreq/governor.h
>>>>>>>>>>> index 5cee3f64fe2b..31af6d072a10 100644
>>>>>>>>>>> --- a/drivers/devfreq/governor.h
>>>>>>>>>>> +++ b/drivers/devfreq/governor.h
>>>>>>>>>>> @@ -75,7 +75,7 @@ struct devfreq_governor {
>>>>>>>>>>>                    unsigned int event, void *data);
>>>>>>>>>>>    };
>>>>>>>>>>> -void devfreq_monitor_start(struct devfreq *devfreq);
>>>>>>>>>>> +int devfreq_monitor_start(struct devfreq *devfreq);
>>>>>>>>>>>    void devfreq_monitor_stop(struct devfreq *devfreq);
>>>>>>>>>>>    void devfreq_monitor_suspend(struct devfreq *devfreq);
>>>>>>>>>>>    void devfreq_monitor_resume(struct devfreq *devfreq);
>>>>>>>>>>> diff --git a/drivers/devfreq/governor_simpleondemand.c
>>>>>>>>>>> b/drivers/devfreq/governor_simpleondemand.c
>>>>>>>>>>> index d57b82a2b570..ea287b57cbf3 100644
>>>>>>>>>>> --- a/drivers/devfreq/governor_simpleondemand.c
>>>>>>>>>>> +++ b/drivers/devfreq/governor_simpleondemand.c
>>>>>>>>>>> @@ -89,8 +89,7 @@ static int devfreq_simple_ondemand_handler(struct
>>>>>>>>>>> devfreq *devfreq,
>>>>>>>>>>>    {
>>>>>>>>>>>        switch (event) {
>>>>>>>>>>>        case DEVFREQ_GOV_START:
>>>>>>>>>>> -        devfreq_monitor_start(devfreq);
>>>>>>>>>>> -        break;
>>>>>>>>>>> +        return devfreq_monitor_start(devfreq);
>>>>>>>>>>>        case DEVFREQ_GOV_STOP:
>>>>>>>>>>>            devfreq_monitor_stop(devfreq);
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> Need to handle the all points of devfreq_monitor_start() usage.
>>>>>>>>>> please check the tegra30-devfreq.c for this update.
>>>>>>>>>>
>>>>>>>>>> $ grep -rn "devfreq_monitor_start" drivers/
>>>>>>>>>> drivers/devfreq/governor_simpleondemand.c:92:
>>>>>>>>>> devfreq_monitor_start(devfreq);
>>>>>>>>>> drivers/devfreq/tegra30-devfreq.c:744:
>>>>>>>>>> devfreq_monitor_start(devfreq);
>>>>>>>>>> ......
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> --
>>>>>>>>> Best Regards,
>>>>>>>>> Samsung Electronics
>>>>>>>>> Chanwoo Choi
>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> --
>>>>>>> Best Regards,
>>>>>>> Chanwoo Choi
>>>>>>> Samsung Electronics
>>>>>>
>>>>>>
>>>>>
>>>>>
>>>>> --
>>>>> Best Regards,
>>>>> Chanwoo Choi
>>>>> Samsung Electronics
>>>
>>>
>>> --
>>> Best Regards,
>>> Samsung Electronics
>>> Chanwoo Choi
> 
> 


-- 
Best Regards,
Chanwoo Choi
Samsung Electronics



More information about the linux-arm-kernel mailing list