Re: [RFC v2 2/3] dma-buf/fence-array: add fence_array_teardown()

["Christian König" <christian.koenig@amd.com>]

Am 28.06.2016 um 16:17 schrieb Gustavo Padovan:
> 2016-06-28 Christian König <christian.koenig@amd.com>:
>
>> Am 27.06.2016 um 21:29 schrieb Gustavo Padovan:
>>> From: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
>>>
>>> As the array of fence callbacks held by an active struct fence_array
>>> each has a reference to the struct fence_array, when the owner of the
>>> fence_array is freed it must dispose of the callback references before
>>> it can free the fence_array. This can not happen simply during
>>> fence_release() because of the extra references and so we need a new
>>> function to run before the final fence_put().
>> As I said previously as well, this is completely superfluous.
>>
>> The fence array keeps a reference to itself as long as not all callbacks are
>> signaled.
>>
>> So you only need to unregister your callback from the array itself and drop
>> your reference when you don't need it any more in the sync file.
> Exactly, this should be called from sync_file_free() because of the
> following use case:
>
> 	1. You create 2 sync_file with 1 fence each
> 	2. Merge both fences, which creates a fence array
> 	3. Close the sync_file fd without waiting for the fences to
> 	signal
>
> At this point you leak the fence-array because the final fence_put()
> does not release it because of the extra references from the non
> signalled fences so we need to clean up this somehow.

No, there won't be any leak and you don't need to cleanup anything.

The fences contained in the fence array are already enabled for 
signaling. So they will eventually signal sooner or later and drop the 
reference to the fence array freeing it when the last one finished.

This was done to avoid the need to call fence_remove_callback() all the 
time on the fences in the array because of the warning on the function.

Regards,
Christian.

>
> 	Gustavo