Re: [PATCH v3] PM / devfreq: Restart previous governor if new governor fails to start

[Chanwoo Choi]

2016-11-12 8:25 GMT+09:00 Saravana Kannan <skannan@xxxxxxxxxxxxxx>:
> On 11/09/2016 04:10 PM, Chanwoo Choi wrote:
>>
>> Hi,
>>
>> On 2016ë 11ì 10ì 05:34, Saravana Kannan wrote:
>>>
>>> On 11/08/2016 06:38 PM, Chanwoo Choi wrote:
>>>>
>>>> On 2016ë 11ì 09ì 11:36, Chanwoo Choi wrote:
>>>>>
>>>>> Hi,
>>>>>
>>>>> On 2016ë 11ì 09ì 10:33, Chanwoo Choi wrote:
>>>>>>
>>>>>> On 2016ë 11ì 09ì 05:52, Saravana Kannan wrote:
>>>>>>>
>>>>>>> On 11/08/2016 01:02 AM, Chanwoo Choi wrote:
>>>>>>>>
>>>>>>>> Hi,
>>>>>>>>
>>>>>>>> On 2016ë 11ì 08ì 03:57, Saravana Kannan wrote:
>>>>>>>>>
>>>>>>>>> On 10/26/2016 05:06 PM, Chanwoo Choi wrote:
>>>>>>>>>>
>>>>>>>>>> Hi,
>>>>>>>>>>
>>>>>>>>>> On 2016ë 10ì 27ì 04:17, Saravana Kannan wrote:
>>>>>>>>>>>
>>>>>>>>>>> If the new governor fails to start, switch back to old governor
>>>>>>>>>>> so that the
>>>>>>>>>>> devfreq state is not left in some weird limbo.
>>>>>>>>>>>
>>>>>>>>>>> Signed-off-by: Saravana Kannan <skannan@xxxxxxxxxxxxxx>
>>>>>>>>>>> ---
>>>>>>>>>>> * Fix NULL deref for real this time.
>>>>>>>>>>> * Addressed some style preferences.
>>>>>>>>>>>
>>>>>>>>>>>      drivers/devfreq/devfreq.c | 13 +++++++++++--
>>>>>>>>>>>      1 file changed, 11 insertions(+), 2 deletions(-)
>>>>>>>>>>>
>>>>>>>>>>> diff --git a/drivers/devfreq/devfreq.c
>>>>>>>>>>> b/drivers/devfreq/devfreq.c
>>>>>>>>>>> index bf3ea76..77c41a5 100644
>>>>>>>>>>> --- a/drivers/devfreq/devfreq.c
>>>>>>>>>>> +++ b/drivers/devfreq/devfreq.c
>>>>>>>>>>> @@ -919,7 +919,7 @@ static ssize_t governor_store(struct device
>>>>>>>>>>> *dev, struct device_attribute *attr,
>>>>>>>>>>>          struct devfreq *df = to_devfreq(dev);
>>>>>>>>>>>          int ret;
>>>>>>>>>>>          char str_governor[DEVFREQ_NAME_LEN + 1];
>>>>>>>>>>> -    struct devfreq_governor *governor;
>>>>>>>>>>> +    const struct devfreq_governor *governor, *prev_governor;
>>>>>>>>>>>
>>>>>>>>>>>          ret = sscanf(buf, "%" __stringify(DEVFREQ_NAME_LEN) "s",
>>>>>>>>>>> str_governor);
>>>>>>>>>>>          if (ret != 1)
>>>>>>>>>>> @@ -944,12 +944,21 @@ static ssize_t governor_store(struct device
>>>>>>>>>>> *dev, struct device_attribute *attr,
>>>>>>>>>>>                  goto out;
>>>>>>>>>>>              }
>>>>>>>>>>>          }
>>>>>>>>>>> +    prev_governor = df->governor;
>>>>>>>>>>>          df->governor = governor;
>>>>>>>>>>>          strncpy(df->governor_name, governor->name,
>>>>>>>>>>> DEVFREQ_NAME_LEN);
>>>>>>>>>>>          ret = df->governor->event_handler(df, DEVFREQ_GOV_START,
>>>>>>>>>>> NULL);
>>>>>>>>>>> -    if (ret)
>>>>>>>>>>> +    if (ret) {
>>>>>>>>>>>              dev_warn(dev, "%s: Governor %s not started(%d)\n",
>>>>>>>>>>>                   __func__, df->governor->name, ret);
>>>>>>>>>>> +        if (prev_governor) {
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> I think that prev_governor is always set. You don't need to check
>>>>>>>>>> it.
>>>>>>>>>> Why do check it?
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> Not true. Even higher up in this function, we check if df->governor
>>>>>>>>> != NULL. Simple example would be that the default governor passed in while
>>>>>>>>> adding the device isn't compiled in.
>>>>>>>>
>>>>>>>>
>>>>>>>> I don't understand. If device is not registered by
>>>>>>>> devfreq_add_device(),
>>>>>>>> you don't change the governor by using governor_store().
>>>>>>>>
>>>>>>>> If you can change the governor through governor_store(),
>>>>>>>> it means that the devfreq instance was added without any problem.
>>>>>>>> The added devfreq instance must have the sepecific governor
>>>>>>>> on df->governor variable.
>>>>>>>>
>>>>>>>> So, I think that df->governor is always set and then prev_governor
>>>>>>>> is always set.
>>>>>>>
>>>>>>>
>>>>>>> Read the code more closely. add_device doesn't (and shouldn't) fail
>>>>>>> if the governor isn't present. After that, the governor could be changed
>>>>>>> from user space.
>>>>>>
>>>>>>
>>>>>> If governor is not present during devfreq_add_device(),
>>>>>> the devfreq instance is not able to register the devfreq framework.
>>>>>> So, the user never touch the sysfs[1] to change the governor
>>>>>> because the sysfs[1] is not created by devfreq framework.
>>>>>> [1] /sys/class/devfreq/[device name]/governor
>>>>>>
>>>>>> In summary, if governor is not present during devfreq_add_device(),
>>>>>> the devfreq framework doesn't create tye sysfs[1] for governor.
>>>>>>
>>>>>> The user-space never change the governor through sysfs[1]
>>>>>> because there is no any sysfs entry[1].
>>>>>
>>>>>
>>>>> I checked the code of devfreq_add_device().
>>>>> As you mentioned, if there is no governor,
>>>>> devfreq_add_device() don't pass wihtout calling the
>>>>> devfreq->governor->even_handler().
>>>>
>>>>
>>>> Wrong expression / don't pass -> pass
>>>>
>>>
>>> I think it's correct as is. Just because the governor isn't there (or
>>> hasn't registered yet) doesn't mean the device add should fail.
>>>
>>> That aside, do you care to ack this patch for the way the code is
>>> currently?
>>
>>
>> Reviewed-by: Chanwoo Choi <cw00.choi@xxxxxxxxxxx>
>
>
> Thanks
>
>> After fixing the bug of devfreq_add_device(),
>> I'll remove the unneeded 'if statement' of prev_governor in
>> governor_store.
>>
>
> As mentioned earlier, I don't think it's a bug. It's fine as is.

No. I think it is a bug. The goal of devfreq should handle the frequency/voltage
with governor. If governor is not necessary of the specific device, the device
driver don't need to use the devfreq.

The default governor is always available and necessary of the device driver.

-- 
Best Regards,
Chanwoo Choi