Re: sem_post() does not signal waiting threads

Am Freitag den, 30. August 2002, um 11:33, schrieb Dr. Uwe Girlich:
>
> On Fri, Aug 30, 2002 at 09:56:20AM +0200, Dr. Uwe Girlich wrote:
>> There is an error in the POSIX semaphore implementation in NGPT:
> ...
>> The attached patch file corrects this strange behaviour.
> This correction corrects the behaviour of process shared semaphores.
> Process local semaphores are still wrong. I'll look into it.

Does your code uses sem_trywait()?
In that function the mutex is not released in every code path.

Another question: Is that double locking required? When all functions
accessing the semaphore use __sem_waiting->__lock, why is
__sem_waiting->__sem_lock needed? Just for destroying it safely?


--- semaphore.c.orig    Fri Aug 30 15:04:23 2002
+++ semaphore.c Fri Aug 30 15:06:02 2002
@@ -100,9 +100,11 @@

      retval = pthread_mutex_trylock(&sem->__sem_waiting->__lock);
      if (retval == 0) {
-       if (sem->__sem_value == 0)
-           retval = -1;
-       else {
+       if (sem->__sem_value == 0){
+               return_errno(-1, EAGAIN);
+               pthread_mutex_unlock(&sem->__sem_waiting->__lock);
+               return -1;
+       } else {
             pth_acquire_lock(&sem->__sem_waiting->__sem_lock);
             sem->__sem_value--;
             pth_release_lock(&sem->__sem_waiting->__sem_lock);