On 11/02/2019 11:38, Hans Verkuil (hansverk) wrote:
> On 09/02/2019 03:48, Wen Yang wrote:
>> put_device() should be called in cec_notifier_release(),
>> since the dev is being passed down to cec_notifier_get_conn(),
>> which holds reference. On cec_notifier destruction, it
>> should drop the reference to the device.
>>
>> Fixes: 6917a7b77413 ("[media] media: add CEC notifier support")
>> Signed-off-by: Wen Yang <[email protected]>
>> ---
>> drivers/media/cec/cec-notifier.c | 1 +
>> 1 file changed, 1 insertion(+)
>>
>> diff --git a/drivers/media/cec/cec-notifier.c b/drivers/media/cec/cec-notifier.c
>> index dd2078b..621d4ae 100644
>> --- a/drivers/media/cec/cec-notifier.c
>> +++ b/drivers/media/cec/cec-notifier.c
>> @@ -66,6 +66,7 @@ static void cec_notifier_release(struct kref *kref)
>> container_of(kref, struct cec_notifier, kref);
>>
>> list_del(&n->head);
>> + put_device(n->dev);
>> kfree(n->conn);
>> kfree(n);
>> }
>>
>
> Sorry, no. The dev pointer is just a search key that the notifier code looks
> for. It is not the notifier's responsibility to take a reference, that would
> be the responsibility of the hdmi and cec drivers.
Correction: the cec driver should never take a reference of the hdmi device.
It never accesses the HDMI device, it only needs the HDMI device pointer as
a key in the notifier list.
The real problem is that several CEC drivers take a reference of the HDMI device
and never release it. So those drivers need to be fixed.
Regards,
Hans
>
> If you can demonstrate that there is an object reference leak, then please
> provide the details: it is likely a bug elsewhere and not in the notifier
> code.
>
> BTW, your patch series didn't arrive on the linux-media mailinglist for
> some reason.
>
> Regards,
>
> Hans
>