Re: NULL pointer dereference in iio_buffer_register()

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

On 07/31/2012 02:55 PM, Fengguang Wu wrote:
>>> The panic happens while trying to dereference the NULL indio_dev->buffer:
>>>
>>>         266 int iio_buffer_register(struct iio_dev *indio_dev,
>>>         267                         const struct iio_chan_spec *channels,
>>>         268                         int num_channels)
>>>         269 {  
>>>         270         struct iio_dev_attr *p;
>>>         271         struct attribute **attr;
>>>         272         struct iio_buffer *buffer = indio_dev->buffer;
>>>         273         int ret, i, attrn, attrcount, attrcount_orig = 0;
>>>         274 
>>> ==>     275         if (buffer->attrs)
>>>         276                 indio_dev->groups[indio_dev->groupcounter++] = buffer->attrs;
>>>
>>> iio_dummy_probe() has the code to configure that buffer, however
>>> iio_simple_dummy_configure_buffer() is defined to do nothing on
>>> !CONFIG_IIO_SIMPLE_DUMMY_BUFFER..
>>>
>>>         448         /* Configure buffered capture support. */
>>> ==>     449         ret = iio_simple_dummy_configure_buffer(indio_dev);
>>>         450         if (ret < 0)
>>>         451                 goto error_unregister_events;
>>>         452 
>>>         453         /*
>>>         454          * Register the channels with the buffer, but avoid the output
>>>         455          * channel being registered by reducing the number of channels by 1.
>>>         456          */
>>>         457         ret = iio_buffer_register(indio_dev, iio_dummy_channels, 5);
>>>         458         if (ret < 0)
>>>         459                 goto error_unconfigure_buffer;
>>>
>>> Any ideas to fix it?
>>>
>>
>> Hi,
>>
>> I think the best would be to move the iio_buffer_register to
>> iio_simple_dummy_configure_buffer.
> 
> Lars, thanks for the quick reply! Hmm, that looks more like a code
> refactor recommendation than fix ;)  In the simplest form, can the
> bug fixed like this?
> 
>   static inline int iio_simple_dummy_configure_buffer(struct iio_dev *indio_dev)
>   {
> -         return 0;
> +         return -1;
>   };
> 

No, we want iio_simple_dummy_configure_buffer to be a noop if buffer support
is disabled, since the driver works fine without it. Except for the issue
you discovered. This issue only appears if CONFIG_IIO_BUFFER=y and
CONFIG_IIO_SIMPLE_DUMMY_BUFFER=n. E.g. if both are not set the driver works
fine without buffers.

I can prepare a patch which moves the iio_buffer_register to
iio_simple_dummy_configure_buffer if you want to.

- Lars