如何在Mac OS X下调试死锁？

Question

I'm developing a port of audio player to Mac environment. It works fine under windows, but not with Mac. I got an unpredictable deadlock when I'm playing the music with the player, the only information I can get is to dump the stack of threads of the hanging process. Here is the call stack I get:

Sampling process 1808 for 1 seconds with 1 millisecond of run time between samples
Sampling completed, processing symbols...
Analysis of sampling Python (pid 1808) every 1 millisecond
Call graph:
    785 Thread_32664   DispatchQueue_1: com.apple.main-thread  (serial)
      785 0x1ea9
        785 0x1f82
          785 Py_Main
            785 PyRun_SimpleFileExFlags
              785 PyRun_FileExFlags
                785 PyEval_EvalCode
                  785 PyEval_EvalCodeEx
                    785 PyEval_EvalFrameEx
                      785 time_sleep
                        785 select
                          785 select$DARWIN_EXTSN$NOCANCEL
    785 Thread_32672   DispatchQueue_2: com.apple.libdispatch-manager  (serial)
      785 start_wqthread
        785 _pthread_wqthread
          785 _dispatch_worker_thread2
            785 _dispatch_queue_invoke
              785 _dispatch_mgr_invoke
                785 kevent
    785 Thread_32679
      785 thread_start
        785 _pthread_start
          785 CAPThread::Entry(CAPThread*)
            785 HP_IOThread::ThreadEntry(HP_IOThread*)
              785 HP_IOThread::WorkLoop()
                785 HP_IOThread::PerformIO(AudioTimeStamp const&, double)
                  785 IOA_Device::CallIOProcs(AudioTimeStamp const&, AudioTimeStamp const&, AudioTimeStamp const&)
                    785 HP_IOProc::Call(AudioTimeStamp const&, AudioTimeStamp const&, AudioBufferList const*, AudioTimeStamp const&, AudioBufferList*)
                      785 BASS_ChannelIsSliding
                        785 MPEnterCriticalRegion
                          785 TSWaitOnConditionTimedRelative
                            785 TSWaitOnCondition
                              785 pthread_cond_wait$UNIX2003
                                785 _pthread_cond_wait
                                  785 __semwait_signal
    785 Thread_32680
      785 0xa4fffad
        785 dict_repr
          785 0x1307ccd
            785 MPEnterCriticalRegion
              785 TSWaitOnConditionTimedRelative
                785 TSWaitOnCondition
                  785 pthread_cond_wait$UNIX2003
                    785 _pthread_cond_wait
                      785 __semwait_signal
    785 Thread_32681
      785 MPEnterCriticalRegion
        785 TSWaitOnConditionTimedRelative
          785 TSWaitOnCondition
            785 pthread_cond_wait$UNIX2003
              785 _pthread_cond_wait
                785 __semwait_signal
    785 Thread_36905
      785 start_wqthread
        785 _pthread_wqthread
          785 __workq_kernreturn

Total number in stack (recursive counted multiple, when >=5):

Sort by top of stack, same collapsed (when >= 5):
        __semwait_signal        2355
        __workq_kernreturn        785
        kevent        785
        select$DARWIN_EXTSN$NOCANCEL        785
Sample analysis of process 1808 written to file /dev/stdout

I can only know some threads are waiting for others, but I have no idea what locks they are, who is waiting for. Here comes my question, how can I debug the deadlock under Mac environment?

Answer 1

Debugging multi-threaded apps is inherently difficult. A simple (and possibly simplistic ) approach is to write all of your lock about-to-grab/blocked/released events to a log file. Nothing particularly OS X about that, but I've used it to good effect on similar problems.