[scheduler] Fix waitUntilQueued() failure.

Author: gaomy3832

src/scheduler.h

@@ -638,13 +638,16 @@ class Scheduler : public GlobAlloc, public Callee {
                 if (futex_res == 0 || th->futexWord != 1) break;
             }
             //info("%d out of sched wait, got cid = %d, needsJoin = %d", th->gid, th->cid, th->needsJoin);
+            // NOTE(mgao): After wakeup, waker assumes the wakee will wait on schedLock to join. See waitUntilQueued().
+            // So we should get the lock regardless of needsJoin.
+            futex_lock(&schedLock);
             if (th->needsJoin) {
-                futex_lock(&schedLock);
-                assert(th->needsJoin); //re-check after the lock
+                //assert(th->needsJoin); //re-check after the lock
                 zinfo->cores[th->cid]->join();
                 bar.join(th->cid, &schedLock);
                 //info("%d join done", th->gid);
             }
+            futex_unlock(&schedLock);
         }
 
         void wakeup(ThreadInfo* th, bool needsJoin) {