[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]


From: Vicente Hernando Ara
Subject: Hurd_condition_wait
Date: 16 Nov 2002 20:58:26 +0100

Hi all!

The following is a hurd_condition_wait replacement with pthread code.

This would work assuming I can freely replace __spin_lock and
__spin_unlock glibc functions by pthread_spin_lock and
pthread_spin_unlock, and they are compatible.

hurd_pthread_cond_wait function:

/* Just like pthread_condition_wait, but cancellable.  Returns true if
cancelled.  */
hurd_pthread_cond_wait (pthread_cond_t c, pthread_mutex_t m)
  /* This function will be called by hurd_thread_cancel while we 
are blocked in the pthread_cond_wait. We wake up all threads blocked 
on C,so our thread will wake up and notice the cancellation flag.  */
  void cancel_me (void)
      pthread_cond_broadcast (c);
  struct hurd_sigstate *ss = _hurd_self_sigstate ();
  int cancel;
  assert (ss->intr_port == MACH_PORT_NULL); /* Sanity check for signal
bugs. */
  pthread_spin_lock (&ss->lock); /* With this lock we avoid being
suspended. */
  cancel = ss->cancel;
  if (cancel)
    /* We were cancelled before doing anything.  Don't block at all.  */
    ss->cancel = 0;
      /* Tell hurd_thread_cancel how to unblock us.  */
      ss->cancel_hook = &cancel_me;
  pthread_spin_unlock (&ss->lock);  
  if (cancel)
      /* Cancelled on entry.  Just leave the mutex locked.  */
      m = NULL;
      /* Now unlock the mutex and block until woken.  */
      pthread_mutex_unlock (m);
      pthread_cond_wait(c, m);
  pthread_spin_lock (&ss->lock);
  /* Clear the hook, now that we are done blocking.  */
  ss->cancel_hook = NULL;
  /* Check the cancellation flag; we might have unblocked due to      
cancellation rather than a normal condition_signal or
condition_broadcast (or we might have just happened to get cancelled
right after waking up).  */
  cancel |= ss->cancel;
  ss->cancel = 0;
  pthread_spin_unlock (&ss->lock);

  if (m)
    /* Reacquire the mutex and return.  */
    pthread_mutex_lock (m);

  return cancel;

I think there is a race condition in this implementation.
After setting the hook function and unlocking ss->lock, we can receive
a signal broadcast before we have done a pthread_cond_wait.

I see several solutions:

a) Including pthread_cond_wait before we unlock ss->lock spinlock.
   Can this have side effects with the thread suspension?

b) Add a new mutex or spin_lock to hurd_thread_cancel before calling the
hook function. 

c) A different implementation using __pthread_enqueue function.

Can you advice me? I think c is the best, but not sure.

At the end there is the original hurd_condition_wait function attached.



Attachment: hurd_condition_wait.c
Description: Text Data

reply via email to

[Prev in Thread] Current Thread [Next in Thread]