Use relaxed atomics since there is no MO dependence

Replace the 3 uses of atomic_bit_set and atomic_bit_test_set with
atomic_fetch_or_relaxed.  Using relaxed MO is correct since the
atomics are used to ensure memory is released only once.

Reviewed-by: Florian Weimer <fweimer@redhat.com>

nptl/nptl_free_tcb.c

@@ -24,7 +24,8 @@ void
 __nptl_free_tcb (struct pthread *pd)
 {
   /* The thread is exiting now.  */
-  if (atomic_bit_test_set (&pd->cancelhandling, TERMINATED_BIT) == 0)
+  if ((atomic_fetch_or_relaxed (&pd->cancelhandling, TERMINATED_BITMASK)
+      & TERMINATED_BITMASK) == 0)
     {
       /* Free TPP data.  */
       if (pd->tpp != NULL)

nptl/pthread_create.c

@@ -487,7 +487,7 @@ start_thread (void *arg)
   /* The thread is exiting now.  Don't set this bit until after we've hit
      the event-reporting breakpoint, so that td_thr_get_info on us while at
      the breakpoint reports TD_THR_RUN state rather than TD_THR_ZOMBIE.  */
-  atomic_bit_set (&pd->cancelhandling, EXITING_BIT);
+  atomic_fetch_or_relaxed (&pd->cancelhandling, EXITING_BITMASK);
 
   if (__glibc_unlikely (atomic_decrement_and_test (&__nptl_nthreads)))
     /* This was the last thread.  */

sysdeps/nptl/pthreadP.h

@@ -43,12 +43,6 @@
   atomic_compare_and_exchange_val_acq (&(descr)->member, new, old)
 #endif
 
-#ifndef THREAD_ATOMIC_BIT_SET
-# define THREAD_ATOMIC_BIT_SET(descr, member, bit) \
-  atomic_bit_set (&(descr)->member, bit)
-#endif
-
-
 static inline short max_adaptive_count (void)
 {
 #if HAVE_TUNABLES
@@ -276,7 +270,7 @@ __do_cancel (void)
   struct pthread *self = THREAD_SELF;
 
   /* Make sure we get no more cancellations.  */
-  atomic_bit_set (&self->cancelhandling, EXITING_BIT);
+  atomic_fetch_or_relaxed (&self->cancelhandling, EXITING_BITMASK);
 
   __pthread_unwind ((__pthread_unwind_buf_t *)
 		    THREAD_GETMEM (self, cleanup_jmp_buf));