+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;
+
+ if (!limit) 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;
+
+ if (++b->_b_count == limit) {
+ ret = PTHREAD_BARRIER_SERIAL_THREAD;
+ b->_b_seq++;
+ __wake(&b->_b_seq, -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);
+ }
+
+ __vm_lock(+1);
+
+ 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);
+ } else {
+ __wait(&b->_b_seq, 0, seq+1, 0);
+ }
+
+ __vm_unlock();
+
+ return 0;
+}
+