Re: [PATCH 1/2] iio: adc: ti_am335x_adc: do not free the kfifo twice

[Lars-Peter Clausen <lars@metafoo.de>]

On 11/09/2013 01:34 PM, Jonathan Cameron wrote:
> On 10/24/13 12:50, Jonathan Cameron wrote:
>> On 10/24/13 11:44, Lars-Peter Clausen wrote:
>>> On 10/24/2013 12:34 PM, Sebastian Andrzej Siewior wrote:
>>>> On 10/24/2013 11:41 AM, Lars-Peter Clausen wrote:
>>>>> The driver seems to be missing the iio_buffer_attach() call. Something like
>>>>> this should fix the problem:
>>>>>
>>>>> diff --git a/drivers/iio/adc/ti_am335x_adc.c b/drivers/iio/adc/ti_am335x_adc.c
>>>>> index ef54d8a..bf9c89c 100644
>>>>> --- a/drivers/iio/adc/ti_am335x_adc.c
>>>>> +++ b/drivers/iio/adc/ti_am335x_adc.c
>>>>> @@ -229,12 +229,15 @@ static int tiadc_iio_buffered_hardware_setup(struct
>>>>> iio_dev *indio_dev,
>>>>>  	unsigned long flags,
>>>>>  	const struct iio_buffer_setup_ops *setup_ops)
>>>>>  {
>>>>> +	struct iio_buffer *buffer;
>>>>>  	int ret;
>>>>>
>>>>> -	indio_dev->buffer = iio_kfifo_allocate(indio_dev);
>>>>> -	if (!indio_dev->buffer)
>>>>> +	buffer = iio_kfifo_allocate(indio_dev);
>>>>> +	if (!buffer)
>>>>>  		return -ENOMEM;
>>>>>
>>>>> +	iio_device_attach_buffer(indio_dev, buffer);
>>>>> +
>>>>>  	ret = request_threaded_irq(irq,	pollfunc_th, pollfunc_bh,
>>>>>  				flags, indio_dev->name, indio_dev);
>>>>>  	if (ret)
>>>>
>>>> Yep, that works, thanks.
>>>>
>>>> Shouldn't the two
>>>>
>>>>          tiadc_iio_buffered_hardware_remove(indio_dev);
>>>>          tiadc_channels_remove(indio_dev);
>>>>
>>>> in tiadc_remove() be reversed in their call order? The second alter is
>>>> accessing the buffer which is released by the former one.
>>>>
>>>
>>> As far as I can see tiadc_channels_remove() only does a
>>> kfree(indio_dev->channels), so it does not access the buffer at all.
>> Certainly seems to be true...
>>>
>>>> btw: is all this ref counting really required? I mean I would assume
>>>> allocate buffer in one place (at probe time) release it remove time
>>>> should be enough.
>>>
>>> It is required. Userspace may still be reading from the buffer when the
>>> driver frees it. So we need proper refcounting here.
>>
>> Lars, can you do a clean version of the above with a reported-by from Sebastian
>> then Sebastian can you ack (if you are happy with it of course!)
>>
> Lars, I've turned the above into a coherent patch and applied it to the fixes-togreg
> branch.  I've added your Signed-off-by: Shout if you would prefer not.
> 
> Also a reported by for Sebastian.
> 
> I didn't want this patch to fall through the cracks given it's been around a while now.

It was still on my radar. The plan was to send it once the merge window has
closed.

- Lars