Re: [E-devel] Possible Unmatched LKL and LKU in _ecore_thread_shutdown

On Tue, Aug 7, 2012 at 11:47 AM, Cedric BAIL <cedric.bail <at> free.fr> wrote:
> On Tue, Aug 7, 2012 at 11:28 AM, Cedric BAIL <cedric.bail <at> free.fr> wrote:
>> On Tue, Aug 7, 2012 at 10:18 AM, Brian Wang <brian.wang.0721 <at> gmail.com> wrote:
>>> In ecore_thread.c:_ecore_thread_shutdown(),
>>>    if there is still some _ecore_running_job,
>>>    ecore_thread_cancel() will be called while
>>> _ecore_pending_job_threads_mutex is being held.
>>> In ecore_thread_cancel(),
>>>    it will try to acquire the same mutex, causing
>>> _ecore_thread_shutdown() to freeze.
>>>
>>> That's how I read the code while I was trying to debug why my app does
>>> not shutdown completely (there
>>> is a thread running, without being cancelled before calling elm_exit()).
>>>
>>> I do not know how to fix it other than moving LKU before calling
>>> ecore_thread_cancel; however, I do not know
>>> if it's safe to do or not.
>>
>> Sadly, it is not as we need to take lock as long as we walk on that
>> list. I will look at it and find a proper solution, but basically
>> ecore_thread_cancel doesn't access the same list. So either I use
>> another lock when accessing the _ecore_running_job or I tell somehow
>> ecore_thread_cancel to not even try to access the
>> _ecore_pending_job_threads and _ecore_pending_job_threads_feedback.
>> Need to think, quickly, about it. I need to fix that before EFL
>> release.
>
> Ok, fixed in r74953 by using another mutex.