Re: [PATCH] PM / devfreq: use put_device() instead of kfree()

[Chanwoo Choi]

On 2018ë 04ì 13ì 11:15, arvindY wrote:
> Hi Chanwoo,
> 
> On Friday 13 April 2018 06:43 AM, Chanwoo Choi wrote:
>> On 2018ë 04ì 13ì 10:03, Chanwoo Choi wrote:
>>> Hi,
>>>
>>> I'm sorry for the late reply.
>>>
>>> On 2018ë 03ì 30ì 20:44, Arvind Yadav wrote:
>>>> Never directly free @dev after calling device_register() or
>>>> device_unregister(), even if device_register() returned an error.
>>>> Always use put_device() to give up the reference initialized.
>>>>
>>>> Signed-off-by: Arvind Yadav <arvind.yadav.cs@xxxxxxxxx>
>>>> ---
>>>>   drivers/devfreq/devfreq.c | 4 +++-
>>>>   1 file changed, 3 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/devfreq/devfreq.c b/drivers/devfreq/devfreq.c
>>>> index fe2af6a..a225b94 100644
>>>> --- a/drivers/devfreq/devfreq.c
>>>> +++ b/drivers/devfreq/devfreq.c
>>>> @@ -625,7 +625,8 @@ struct devfreq *devfreq_add_device(struct device *dev,
>>>>       err = device_register(&devfreq->dev);
>>>>       if (err) {
>>>>           mutex_unlock(&devfreq->lock);
>>>> -        goto err_dev;
>>>> +        put_device(&devfreq->dev);
>>>> +        goto err_out;
>>> why do you change the goto postion?
>>> err_out is correct to free the memory of devfreq instance.
>> Sorry. err_dev is correct instead of err_out.
> 
> If you will see the comment for device_register(drivers/base/core.c)
> there is mentioned that 'NOTE: _Never_ directly free @dev
> after calling this function, even if it returned an error!
> Always use put_device() to give up the reference initialized
> in this function instead. Here  put_device() will decrement
> the last reference and then free the memory by calling dev->release.
> Internally put_device() -> kobject_put() -> kobject_cleanup() which
> is responsible to call 'dev -> release' and also free other kobject resources.
> 
> We are releasing devfreq in devfreq_dev_release(). So no need
> to call kfree() again. It'll be redundant.  'err_out' is correct.

You're right. err_out is correct.
put_device() -> dev->release() -> devfreq_dev_release() -> kfree(devfreq)

>>
>>>>       }
>>>>         devfreq->trans_table =    devm_kzalloc(&devfreq->dev,
>>>> @@ -671,6 +672,7 @@ struct devfreq *devfreq_add_device(struct device *dev,
>>>>       mutex_unlock(&devfreq_list_lock);
>>>>         device_unregister(&devfreq->dev);
>>>> +    devfreq = NULL;
>>> It is wrong. If you initialize the devfreq as NULL,
>>> never free the 'devfreq' instance.
> 
> No need to release memory after device_unregister().
> driver core will take care of this. It will release memory
> So no need to call free again.

If you have to initialize the devfreq instance as NULL,
I think that you better to init in the devfreq_dev_release()
as following:

diff --git a/drivers/devfreq/devfreq.c b/drivers/devfreq/devfreq.c
index fe2af6aa88fc..8c52a13d3887 100644
--- a/drivers/devfreq/devfreq.c
+++ b/drivers/devfreq/devfreq.c
@@ -543,6 +543,7 @@ static void devfreq_dev_release(struct device *dev)
 
        mutex_destroy(&devfreq->lock);
        kfree(devfreq);
+       devfreq = NULL;
 }


>>>
>>>>   err_dev:
>>>>       if (devfreq)
>>>>           kfree(devfreq);
>>>>
>>>
>>
> 
> ~arvind
> 
> 
> 


-- 
Best Regards,
Chanwoo Choi
Samsung Electronics