X-Git-Url: http://nsz.repo.hu/git/?a=blobdiff_plain;f=src%2Fthread%2Fpthread_barrier_wait.c;h=e15abb843afdea5cba6b76f9730084216ded0d4e;hb=c63c98a6067030a25a42703db1209ccbcc74803a;hp=9b1edbf94ad19a46a93fc9d68e497db662148ff4;hpb=95b147967216a057d51a1b970da777486be716e3;p=musl

diff --git a/src/thread/pthread_barrier_wait.c b/src/thread/pthread_barrier_wait.c
index 9b1edbf9..e15abb84 100644
--- a/src/thread/pthread_barrier_wait.c
+++ b/src/thread/pthread_barrier_wait.c
@@ -1,27 +1,11 @@
 #include "pthread_impl.h"
 
-static int vmlock[2];
-
-void __vm_lock(int inc)
-{
-	for (;;) {
-		int v = vmlock[0];
-		if (inc*v < 0) __wait(vmlock, vmlock+1, v, 1);
-		else if (a_cas(vmlock, v, v+inc)==v) break;
-	}
-}
-
-void __vm_unlock(void)
-{
-	if (vmlock[0]>0) a_dec(vmlock);
-	else a_inc(vmlock);
-	if (vmlock[1]) __wake(vmlock, 1, 1);
-}
+void __vm_lock_impl(int);
+void __vm_unlock_impl(void);
 
 static int pshared_barrier_wait(pthread_barrier_t *b)
 {
 	int limit = (b->_b_limit & INT_MAX) + 1;
-	int seq;
 	int ret = 0;
 	int v, w;
 
@@ -30,37 +14,40 @@ static int pshared_barrier_wait(pthread_barrier_t *b)
 	while ((v=a_cas(&b->_b_lock, 0, limit)))
 		__wait(&b->_b_lock, &b->_b_waiters, v, 0);
 
-	seq = b->_b_seq;
-
+	/* Wait for <limit> threads to get to the barrier */
 	if (++b->_b_count == limit) {
+		a_store(&b->_b_count, 0);
 		ret = PTHREAD_BARRIER_SERIAL_THREAD;
-		b->_b_seq++;
-		__wake(&b->_b_seq, -1, 0);
+		if (b->_b_waiters2) __wake(&b->_b_count, -1, 0);
 	} else {
 		a_store(&b->_b_lock, 0);
 		if (b->_b_waiters) __wake(&b->_b_lock, 1, 0);
-		__wait(&b->_b_seq, 0, seq, 0);
+		while ((v=b->_b_count)>0)
+			__wait(&b->_b_count, &b->_b_waiters2, v, 0);
 	}
 
-	__vm_lock(+1);
+	__vm_lock_impl(+1);
 
-	if (a_fetch_add(&b->_b_count, -1)==1) {
-		b->_b_seq++;
-		__wake(&b->_b_seq, -1, 0);
+	/* Ensure all threads have a vm lock before proceeding */
+	if (a_fetch_add(&b->_b_count, -1)==1-limit) {
+		a_store(&b->_b_count, 0);
+		if (b->_b_waiters2) __wake(&b->_b_count, -1, 0);
 	} else {
-		__wait(&b->_b_seq, 0, seq+1, 0);
+		while ((v=b->_b_count))
+			__wait(&b->_b_count, &b->_b_waiters2, v, 0);
 	}
 	
 	/* Perform a recursive unlock suitable for self-sync'd destruction */
 	do {
 		v = b->_b_lock;
 		w = b->_b_waiters;
-	} while (a_cas(&b->_b_lock, v, v-1 & INT_MAX) != v);
+	} while (a_cas(&b->_b_lock, v, v==INT_MIN+1 ? 0 : v-1) != v);
 
+	/* Wake a thread waiting to reuse or destroy the barrier */
 	if (v==INT_MIN+1 || (v==1 && w))
 		__wake(&b->_b_lock, 1, 0);
 
-	__vm_unlock();
+	__vm_unlock_impl();
 
 	return ret;
 }
@@ -92,7 +79,7 @@ int pthread_barrier_wait(pthread_barrier_t *b)
 	/* First thread to enter the barrier becomes the "instance owner" */
 	if (!inst) {
 		struct instance new_inst = { 0 };
-		int spins = 10000;
+		int spins = 200;
 		b->_b_inst = inst = &new_inst;
 		a_store(&b->_b_lock, 0);
 		if (b->_b_waiters) __wake(&b->_b_lock, 1, 1);
@@ -100,7 +87,8 @@ int pthread_barrier_wait(pthread_barrier_t *b)
 			a_spin();
 		a_inc(&inst->finished);
 		while (inst->finished == 1)
-			__syscall(SYS_futex, &inst->finished, FUTEX_WAIT,1,0);
+			__syscall(SYS_futex,&inst->finished,FUTEX_WAIT|128,1,0) != -ENOSYS
+			|| __syscall(SYS_futex,&inst->finished,FUTEX_WAIT,1,0);
 		return PTHREAD_BARRIER_SERIAL_THREAD;
 	}