[PATCH RFT 1/5] iio: mxs-lradc: fix memory leak

[Marek Vasut]

On 04/19/2016 03:37 PM, Stefan Wahren wrote:
> Hi Marek,
> 
> Am 19.04.2016 um 12:32 schrieb Marek Vasut:
>> On 04/19/2016 08:33 AM, Stefan Wahren wrote:
>>> Hi Marek,
>>>
>>> Am 18.04.2016 um 19:16 schrieb Marek Vasut:
>>>> On 04/17/2016 12:08 PM, Jonathan Cameron wrote:
>>>>> On 14/04/16 21:01, Marek Vasut wrote:
>>>>>> On 04/14/2016 05:48 PM, Stefan Wahren wrote:
>>>>>>> After successful touchscreen registration the input device was
>>>>>>> never freed. So fix this issue by using devm_input_allocate_device().
>>>>>>>
>>>>>>> Signed-off-by: Stefan Wahren <stefan.wahren at i2se.com>
>>>>>>> ---
>>>>>>>  drivers/iio/adc/mxs-lradc.c |    8 ++------
>>>>>>>  1 file changed, 2 insertions(+), 6 deletions(-)
>>>>>>>
>>>>>>> diff --git a/drivers/iio/adc/mxs-lradc.c b/drivers/iio/adc/mxs-lradc.c
>>>>>>> index 33051b8..0576953 100644
>>>>>>> --- a/drivers/iio/adc/mxs-lradc.c
>>>>>>> +++ b/drivers/iio/adc/mxs-lradc.c
>>>>>>> @@ -1109,12 +1109,11 @@ static int mxs_lradc_ts_register(struct mxs_lradc *lradc)
>>>>>>>  {
>>>>>>>  	struct input_dev *input;
>>>>>>>  	struct device *dev = lradc->dev;
>>>>>>> -	int ret;
>>>>>>>  
>>>>>>>  	if (!lradc->use_touchscreen)
>>>>>>>  		return 0;
>>>>>>>  
>>>>>>> -	input = input_allocate_device();
>>>>>>> +	input = devm_input_allocate_device(dev);
>>>>>>>  	if (!input)
>>>>>>>  		return -ENOMEM;
>>>>>>>  
>>>>>>> @@ -1134,11 +1133,8 @@ static int mxs_lradc_ts_register(struct mxs_lradc *lradc)
>>>>>>>  
>>>>>>>  	lradc->ts_input = input;
>>>>>>>  	input_set_drvdata(input, lradc);
>>>>>>> -	ret = input_register_device(input);
>>>>>>> -	if (ret)
>>>>>>> -		input_free_device(lradc->ts_input);
>>>>>>>  
>>>>>>> -	return ret;
>>>>>>> +	return input_register_device(input);
>>>>>>>  }
>>>>>>>  
>>>>>>>  static void mxs_lradc_ts_unregister(struct mxs_lradc *lradc)
>>>>>>>
>>>>>> Nice find.
>>>>>>
>>>>>> Looks like at91_adc.c and exynos_adc.c suffer from the exact same issue.
>>>>>> The leak looks a bit more severe on exynos even, exynos_adc_ts_init()
>>>>>> could use a proper fail path. Do you want to send patches or shall I ?
>>>>>>
>>>>> As this has been there a long time I'm not going to rush it in as a fix.
>>>> I did take a proper look today and it seems they do the right thing
>>>> afterall. I checked them with kmemleak too to be sure.
>>> thanks, input_unregister_device already free the memory.
>>>
>>> Sorry for the mess :-(
>>>
>>> I think it would be the best to remove / revert this patch.
>> This one? Why exactly? Please elaborate some more, so it's possible to
>> understand the reasoning :)
>>
>>
> 
> as you already pointed out this is not a memory leak, because
> input_unregister_device free the input device. So the commit message is
> incorrect.

I see, yes, it's not a memleak.

> This commit also simplifies the code but in this case we should also
> remove input_unregister_device.

This matches the explanation in drivers/input/input.c above
input_register_device() , agreed.

> Regards
> Stefan
> 


-- 
Best regards,
Marek Vasut