[mythtv-commits] Ticket #1153: Ringbuffer Race condition

MythTV mythtv at cvs.mythtv.org
Thu Jan 11 17:48:20 UTC 2007

#1153: Ringbuffer Race condition
 Reporter:  ben at vanilla.net  |        Owner:  danielk
     Type:  defect           |       Status:  new    
 Priority:  minor            |    Milestone:  0.21   
Component:  mythtv           |      Version:  head   
 Severity:  medium           |   Resolution:         
Comment (by david at hardeman.nu):

 I have managed to get the backtrace, but it is not very interesting,
 SwitchLiveTVRingBuffer is simply executed twice from the TVRec::RunTV

 However, I think I've managed to find the race condition.

 At the end of a program, TVRec::RunTV calls SwitchLiveTVRingBuffer with
 arguments discont 0, set_rec 1. discont = 0 means that
 recorder->CheckForRingBufferSwitch() will not be called in
 SwitchLiveTVRingBuffer. TVRec::RunTV then continues and reaches the end of
 the loop where it sleeps for 1000ms before the TVRec::RunTV loop is
 executed again.

 If everything goes well, TVRec::RecorderThread should (during the time
 that the TVRec::RunTV thread is sleeping) call RingBufferChanged. In the
 example of DVB, the call chain is as follows:

 TVRec::TuningNewRecorder -> pthread_create(&recorder_thread, NULL,
 TVRec::RecorderThread, recorder); -> TVRec::RecorderThread ->
 DVBRecorder::StartRecording ->  DVBRecorder::ProcessDataTS ->
 DVBRecorder::ProcessTSPacket -> DVBRecorder::ProcessTSPacket2 ->
 DTVRecorder::FindMPEG2Keyframes -> DTVRecorder::HandleKeyframe ->
 RecorderBase::CheckForRingBufferSwitch -> TVRec::RingBufferChanged

 So RingBufferChanged will only be called while the TVRec::RunTV thread is
 sleeping if a keyframe is found during that period. Since
 RingBufferChanged updates curRecording, TVRec::RunTV will run the same
 test with the same curRecording as last time the next time the loop is
 executed and re-run SwitchLiveTVRingBuffer if RingBufferChanged hasn't
 been executed (i.e. if no keyframe was found).

 I'm not sure what the best way to fix this would be....danielk?

Ticket URL: <http://svn.mythtv.org/trac/ticket/1153#comment:40>
MythTV <http://www.mythtv.org/>

More information about the mythtv-commits mailing list