X-Git-Url: http://nsz.repo.hu/git/?p=musl;a=blobdiff_plain;f=src%2Fthread%2Fpthread_barrier_wait.c;h=6052925a819a5280f627572d8a6b6455dcfa0b71;hp=1e638d9681234cb67586137b2fa06d91f8b2fd50;hb=e6d765a8b1278e9e5f507638ccdec9fe40e52364;hpb=f16a3089be33a75ef8e75b2dd5ec3095996bbb87

diff --git a/src/thread/pthread_barrier_wait.c b/src/thread/pthread_barrier_wait.c
index 1e638d96..6052925a 100644
--- a/src/thread/pthread_barrier_wait.c
+++ b/src/thread/pthread_barrier_wait.c
@@ -1,5 +1,72 @@
 #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)
+{
+	int inc = vmlock[0]>0 ? -1 : 1;
+	if (a_fetch_add(vmlock, inc)==-inc && vmlock[1])
+		__wake(vmlock, -1, 1);
+}
+
+static int pshared_barrier_wait(pthread_barrier_t *b)
+{
+	int limit = (b->_b_limit & INT_MAX) + 1;
+	int ret = 0;
+	int v, w;
+
+	if (limit==1) return PTHREAD_BARRIER_SERIAL_THREAD;
+
+	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;
+		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);
+		while ((v=b->_b_count)>0)
+			__wait(&b->_b_count, &b->_b_waiters2, v, 0);
+	}
+
+	__vm_lock(+1);
+
+	/* 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 {
+		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;
+}
+
 struct instance
 {
 	int count;
@@ -16,9 +83,12 @@ int pthread_barrier_wait(pthread_barrier_t *b)
 	/* Trivial case: count was set at 1 */
 	if (!limit) return PTHREAD_BARRIER_SERIAL_THREAD;
 
+	/* Process-shared barriers require a separate, inefficient wait */
+	if (limit < 0) return pshared_barrier_wait(b);
+
 	/* Otherwise we need a lock on the barrier object */
 	while (a_swap(&b->_b_lock, 1))
-		__wait(&b->_b_lock, &b->_b_waiters, 1, 0);
+		__wait(&b->_b_lock, &b->_b_waiters, 1, 1);
 	inst = b->_b_inst;
 
 	/* First thread to enter the barrier becomes the "instance owner" */
@@ -27,7 +97,7 @@ int pthread_barrier_wait(pthread_barrier_t *b)
 		int spins = 10000;
 		b->_b_inst = inst = &new_inst;
 		a_store(&b->_b_lock, 0);
-		if (b->_b_waiters) __wake(&b->_b_lock, 1, 0);
+		if (b->_b_waiters) __wake(&b->_b_lock, 1, 1);
 		while (spins-- && !inst->finished)
 			a_spin();
 		a_inc(&inst->finished);
@@ -38,22 +108,21 @@ int pthread_barrier_wait(pthread_barrier_t *b)
 
 	/* Last thread to enter the barrier wakes all non-instance-owners */
 	if (++inst->count == limit) {
-a_spin(); a_spin();
 		b->_b_inst = 0;
 		a_store(&b->_b_lock, 0);
-		if (b->_b_waiters) __wake(&b->_b_lock, 1, 0);
+		if (b->_b_waiters) __wake(&b->_b_lock, 1, 1);
 		a_store(&inst->last, 1);
 		if (inst->waiters)
-			__wake(&inst->last, -1, 0);
+			__wake(&inst->last, -1, 1);
 	} else {
 		a_store(&b->_b_lock, 0);
-		if (b->_b_waiters) __wake(&b->_b_lock, 1, 0);
-		__wait(&inst->last, &inst->waiters, 0, 0);
+		if (b->_b_waiters) __wake(&b->_b_lock, 1, 1);
+		__wait(&inst->last, &inst->waiters, 0, 1);
 	}
 
 	/* Last thread to exit the barrier wakes the instance owner */
 	if (a_fetch_add(&inst->count,-1)==1 && a_fetch_add(&inst->finished,1))
-		__wake(&inst->finished, 1, 0);
+		__wake(&inst->finished, 1, 1);
 
 	return 0;
 }