> > I'm trying to implement a mutex in shared memory but have run into a problem
> .
> > 
> > My system is QNX 6.3.0 SP3.
> > A single process creates a file with shm_open( , O_RDWR | O_CREAT, ) under /
> 
> > dev/shmem and maps it as mmap(, PROT_READ | PROT_WRITE, MAP_SHARED, ).
> > Then process initializes a mutex in shared section with 
> PTHREAD_PROCESS_SHARED
> >  and sets a boolean in shared section to indicate the mutex is initialized.
> > Then process locks the mutex and terminates on assertion failure afterwards.
> 
> > When I start the process for the second time the file already exists so the 
> 
> > process only opens it with shm_open(, O_RDWR, ) without ftruncate() and maps
>  
> > into memory.
> > It detects the flag that the mutex is already created so it does not do 
> > anything else to init the mutex.
> > After that pthread_mutex_trylock() returns EINVAL instead of expected EBUSY 
> 
> > and SyncMutexRevive() returns EINVAL too.
> > Is that the expected behavior? I was sure the mutex should persist even 
> after 
> > all the clients terminate until the file is unlinked. Otherwise there is not
>  
> > much sense in all those dances around SyncMutexRevive().
> 
> 
> Another interesting thing is that pthread_mutex_init() on that address seems 
> to be returning EBUSY. So, does not SyncMutexRevive() work without prior 
> SyncMutextEvent()? Why?

Today I was doing more experiments.
SyncMutexEvent succeeds for that mutex but the pulse is never delivered to the channel.
After the first process instance terminates the mutex is placed in {0x80000001, 0xfffffffe} state (not recursive, locked
, destroyed).

So, to summarize the open questions are:
1) Why is the mutex destroyed if the shared memory section that contains it still exists?
2) Why does an attempt to create a new mutex at that place fail with EBUSY if the previous one had been destroyed?
3) Why does SyncMutexEvent() succeed if the mutex has been destroyed?
4) [the most hard one] What is your suggested approach to make working with a mutex in shared memory feasible?