fix deadlock in condition wait whenever there are multiple waiters

it's amazing none of the conformance tests i've run even bothered to
check whether something so basic works...

src/internal/pthread_impl.h

@@ -66,6 +66,7 @@ struct __timer {
 #define _m_count __u.__i[5]
 #define _c_block __u.__i[0]
 #define _c_clock __u.__i[1]
+#define _c_waiters __u.__i[2]
 #define _rw_lock __u.__i[0]
 #define _rw_waiters __u.__i[1]
 #define _b_inst __u.__p[0]

src/thread/pthread_cond_broadcast.c

@@ -2,7 +2,8 @@
 
 int pthread_cond_broadcast(pthread_cond_t *c)
 {
-	if (a_swap(&c->_c_block, 0))
+	int w = c->_c_waiters;
+	if (a_swap(&c->_c_block, 0) || w)
 		__wake(&c->_c_block, -1, 0);
 	return 0;
 }

src/thread/pthread_cond_signal.c

@@ -2,7 +2,8 @@
 
 int pthread_cond_signal(pthread_cond_t *c)
 {
-	if (a_swap(&c->_c_block, 0));
+	int w = c->_c_waiters;
+	if (a_swap(&c->_c_block, 0) || w)
 		__wake(&c->_c_block, 1, 0);
 	return 0;
 }

src/thread/pthread_cond_timedwait.c

 #include "pthread_impl.h"
 
-static void relock(void *m)
+struct cm {
+	pthread_cond_t *c;
+	pthread_mutex_t *m;
+};
+
+static void cleanup(void *p)
 {
-	pthread_mutex_lock(m);
+	struct cm *cm = p;
+	a_dec(&cm->c->_c_waiters);
+	pthread_mutex_lock(cm->m);
 }
 
 int pthread_cond_timedwait(pthread_cond_t *c, pthread_mutex_t *m, const struct timespec *ts)
 {
+	struct cm cm = { .c=c, .m=m };
 	int r, e=0;
 
 	if (ts && ts->tv_nsec >= 1000000000UL)
@@ -17,8 +25,10 @@ int pthread_cond_timedwait(pthread_cond_t *c, pthread_mutex_t *m, const struct t
 	c->_c_block = 1;
 	if ((r=pthread_mutex_unlock(m))) return r;
 
-	do e = __timedwait(&c->_c_block, 1, c->_c_clock, ts, relock, m, 0);
+	a_inc(&c->_c_waiters);
+	do e = __timedwait(&c->_c_block, 1, c->_c_clock, ts, cleanup, &cm, 0);
 	while (e == EINTR);
+	a_dec(&c->_c_waiters);
 
 	if ((r=pthread_mutex_lock(m))) return r;