X-Git-Url: http://nsz.repo.hu/git/?a=blobdiff_plain;f=src%2Fthread%2Fpthread_barrier_wait.c;h=cc2a8bbf58a9cfee193d7f30bf2bcfd778debe7a;hb=269d193820342dc109f39909d78fb30f4c978f76;hp=a97f6d56a0f3ba1eae0042cce194644557b8b763;hpb=067bdc2c180885af831f55f971f5eb17d2c9dc27;p=musl

diff --git a/src/thread/pthread_barrier_wait.c b/src/thread/pthread_barrier_wait.c
index a97f6d56..cc2a8bbf 100644
--- a/src/thread/pthread_barrier_wait.c
+++ b/src/thread/pthread_barrier_wait.c
@@ -1,57 +1,49 @@
 #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);
-}
-
 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;
 
 	if (limit==1) return PTHREAD_BARRIER_SERIAL_THREAD;
 
-	while (a_swap(&b->_b_lock, 1))
-		__wait(&b->_b_lock, &b->_b_waiters, 1, 0);
-
-	seq = b->_b_seq;
+	while ((v=a_cas(&b->_b_lock, 0, limit)))
+		__wait(&b->_b_lock, &b->_b_waiters, v, 0);
 
+	/* 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();
 
-	if (a_fetch_add(&b->_b_count, -1)==1) {
-		b->_b_seq++;
-		__wake(&b->_b_seq, -1, 0);
-		a_store(&b->_b_lock, 0);
-		if (b->_b_waiters) __wake(&b->_b_lock, 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==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();
 
 	return ret;
@@ -59,10 +51,10 @@ static int pshared_barrier_wait(pthread_barrier_t *b)
 
 struct instance
 {
-	int count;
-	int last;
-	int waiters;
-	int finished;
+	volatile int count;
+	volatile int last;
+	volatile int waiters;
+	volatile int finished;
 };
 
 int pthread_barrier_wait(pthread_barrier_t *b)
@@ -84,7 +76,7 @@ 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);
@@ -92,7 +84,8 @@ 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|FUTEX_PRIVATE,1,0) != -ENOSYS
+			|| __syscall(SYS_futex,&inst->finished,FUTEX_WAIT,1,0);
 		return PTHREAD_BARRIER_SERIAL_THREAD;
 	}