Re: [PATCH] coresight: use put_device() instead of kfree()

[Mathieu Poirier]

On 27 March 2018 at 10:28, arvindY <arvind.yadav.cs@xxxxxxxxx> wrote:
>
>
> On Tuesday 27 March 2018 09:37 PM, Mathieu Poirier wrote:
>>
>> On 26 March 2018 at 20:30, arvindY <arvind.yadav.cs@xxxxxxxxx> wrote:
>>>
>>>
>>> On Tuesday 27 March 2018 03:16 AM, Mathieu Poirier wrote:
>>>>
>>>> drivers/hwtracing/coresight/coresight.c
>>>> On 18 March 2018 at 01:38, Arvind Yadav <arvind.yadav.cs@xxxxxxxxx>
>>>> wrote:
>>>>>
>>>>> Never directly free @dev after calling device_register(), even
>>>>> if it returned an error. Always use put_device() to give up the
>>>>> reference initialized.
>>>>>
>>>>> Signed-off-by: Arvind Yadav <arvind.yadav.cs@xxxxxxxxx>
>>>>> ---
>>>>>    drivers/hwtracing/coresight/coresight.c | 8 ++++----
>>>>>    1 file changed, 4 insertions(+), 4 deletions(-)
>>>>>
>>>>> diff --git a/drivers/hwtracing/coresight/coresight.c
>>>>> b/drivers/hwtracing/coresight/coresight.c
>>>>> index 389c4ba..132dfbc 100644
>>>>> --- a/drivers/hwtracing/coresight/coresight.c
>>>>> +++ b/drivers/hwtracing/coresight/coresight.c
>>>>> @@ -1026,8 +1026,10 @@ struct coresight_device
>>>>> *coresight_register(struct
>>>>> coresight_desc *desc)
>>>>>           dev_set_name(&csdev->dev, "%s", desc->pdata->name);
>>>>>
>>>>>           ret = device_register(&csdev->dev);
>>>>> -       if (ret)
>>>>> -               goto err_device_register;
>>>>> +       if (ret) {
>>>>> +               put_device(&csdev->dev);
>>>>> +               goto err_kzalloc_csdev;
>>>>> +       }
>>>>>
>>>>>           mutex_lock(&coresight_mutex);
>>>>>
>>>>> @@ -1038,8 +1040,6 @@ struct coresight_device
>>>>> *coresight_register(struct
>>>>> coresight_desc *desc)
>>>>>
>>>>>           return csdev;
>>>>>
>>>>> -err_device_register:
>>>>> -       kfree(conns);
>>>>
>>>> Apologies for the late reply, I was travelling.
>>>>
>>>> I suggest to simply replace kfree() with put_device() in order to
>>>> concentrate the error handling code in the same area and make sure
>>>> that memory allocated for @conns and @refcnts is freed.
>>>>
>>>> Thanks,
>>>> Mathieu
>>>
>>>
>>> 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.
>>
>> I have read the notice and in full agreement with the put_device() part.
>>
>>> 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.
>>
>> Memory would be automatically freed if it would have been allocated
>> with the devm_XYZ() helpers but it is not the case here.  As such it
>> has to be freed explicitly.  Reading your patch again (and the jetlag
>> somewhat fading) I think you've done the right thing except for the
>> "goto" that should still point to "err_device_register".
>>
>
> Take rest :)
> 'goto' should not point to "err_device_register" because
> put_device() will call 'dev->release' which is coresight_device_release().
> It's release @conns, @refcnt and @csdev . If we will keep same 'goto
> then kfree() will be redundant for all.

You are correct - your patch does exactly what should be done.

>
>
>>> we should always avoid kfree() if device_register()
>>> returned an error. Otherwise it'll not do clean up of other
>>> kobject resources.
>>>
>>> ~arvind
>>>>>
>>>>>    err_kzalloc_conns:
>>>>>           kfree(refcnts);
>>>>>    err_kzalloc_refcnts:
>>>>> --
>>>>> 2.7.4
>>>>>
>