Added TODO regarding serious locking problem.

git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@28192 a95241bf-73f2-0310-859d-f6bbb57e9c96

src/system/kernel/posix/xsi_message_queue.cpp

@@ -127,6 +127,10 @@ public:
 		queueLocker->Unlock();
 
 		InterruptsSpinLocker _(gThreadSpinlock);
+// TODO: We've got a serious race condition: If BlockAndUnlock() returned due to
+// interruption, we will still be queued. A WakeUpThread() at this point will
+// call thread_unblock() and might thus screw with our trying to re-lock the
+// mutex.
 		return thread_block_locked(thread);
 	}
 
@@ -428,7 +432,7 @@ XsiMessageQueue::Insert(queued_message *message)
 		int32 oldCount = atomic_get(&sXsiMessageCount);
 		if (oldCount >= MAX_XSI_MESSAGE)
 			return true;
-		// If another thread updates the counter we keep 
+		// If another thread updates the counter we keep
 		// iterating
 		if (atomic_test_and_set(&sXsiMessageCount, oldCount + 1, oldCount)
 			== oldCount)

src/system/kernel/posix/xsi_semaphore.cpp

@@ -144,6 +144,10 @@ public:
 		setLocker->Unlock();
 
 		InterruptsSpinLocker _(gThreadSpinlock);
+// TODO: We've got a serious race condition: If BlockAndUnlock() returned due to
+// interruption, we will still be queued. A WakeUpThread() at this point will
+// call thread_unblock() and might thus screw with our trying to re-lock the
+// mutex.
 		return thread_block_locked(thread);
 	}
 
@@ -840,8 +844,8 @@ _user_xsi_semctl(int semaphoreID, int semaphoreNumber, int command,
 	}
 
 	// Lock the semaphore set itself and release both the semaphore
-	// set hash table lock and the ipc hash table lock _only_ if 
+	// set hash table lock and the ipc hash table lock _only_ if
-	// the command it's not IPC_RMID, this prevents undesidered 
+	// the command it's not IPC_RMID, this prevents undesidered
 	// situation from happening while (hopefully) improving the
 	// concurrency.
 	MutexLocker setLocker;