1. Lecture 14: Pthreads Mutex and Condition Variables

2. Review: Mutual Exclusion
Software solution
Disabling interrupts
Strict alternation
Peterson’s solution
Hardware solution
TSL/XCHG
Semaphore

3. Review: Semaphore Implementation
down(&S):
If (S=0) then
Suspend thread, put into a waiting queue
Schedule another thread to run
Else decrement S and return
up(&S):
Increment S
If any threads in waiting queue, then
release one of them (make it ‘ready’)
Both the above are done atomically
by disabling interrupts
by TSL/XCHG

4. In this lecture
Pthreads APIs
Mutex
Condition variables

5. Some Pthreads APIs for mutex

6. Mutex usage in POSIX Threads
pthread_mutex_t m;
pthread_mutex_init(&m);
pthread_mutex_lock(&m);
critical_region();
pthread_mutex_unlock(&m);
To solve our synchronization problem, we introduce mutexes—a synchronization construct providing mutual exclusion. A mutex is used to insure either that only one thread is executing a particular piece of code at once (code locking) or that only one thread is accessing a particular data structure at once (data locking). A mutex belongs either to a particular thread or to no thread (i.e., it is either locked or unlocked). A thread may lock a mutex by calling pthread_mutex_lock. If no other thread has the mutex locked, then the calling thread obtains the lock on the mutex and returns. Otherwise it waits until no other thread has the mutex, and finally returns with the mutex locked. There may of course be multiple threads waiting for the mutex to be unlocked. Only one thread can lock the mutex at a time; there is no specified order for who gets the mutex next, though the ordering is assumed to be at least somewhat “fair.”
To unlock a mutex, a thread calls pthread_mutex_unlock. It is considered incorrect to unlock a mutex that is not held by the caller (i.e., to unlock someone else’s mutex). However, it is somewhat costly to check for this, so most implementations, if they check at all, do so only when certain degrees of debugging are turned on.
Like any other data structure, mutexes must be initialized. This can be done via a call to pthread_mutex_init or can be done statically by assigning PTHREAD_MUTEX_INITIALIZER to a mutex. The initial state of such initialized mutexes is unlocked. Of course, a mutex should be initialized only once! (I.e., make certain that, for each mutex, no more than one thread calls pthread_mutex_init.)

7. pthread_mutex_trylock
Return fails when the mutex is already locked
Used for implementing busy waiting

8. Mutex vs condition variables
Mutex is good to guarantee mutual exclusion
Allow and block access to the critical regions
Conditional variables
Block threads due to some condition not met

9. Condition Variables
Allows a thread to wait till a condition is satisfied
Testing the condition must be done within a mutex
With every condition variable, a mutex is associated

10. Pthreads APIs for condition variables

11. Comparison with Semaphores
If a signal is sent to a conditional variable on which no thread is waiting, the signal is lost
Semaphore will accumulate ‘signals’ by up()

12. Condition Variables pthread_cond_t condition_variable;
pthread_mutex_t mutex;
Signaling Thread:
pthread_mutex_lock(&mutex);
/* change variable value */
if (condition satisfied) {
pthread_cond_signal( &condition_variable); }
pthread_mutex_unlock(&mutex);
/* Alternative to cond_signal is
pthread_cond_broadcast( &condition_variable); */
Waiting Thread:
pthread_mutex_lock(&mutex);
while (condition not satisfied) {
pthread_cond_wait( &condition_variable, &mutex); }
pthread_mutex_unlock(&mutex);
Condition variables are another means for synchronization in POSIX; they represent queues of threads waiting to be woken by other threads and are used in conjunction with the routines shown in the slide. Though they are rather complicated at first glance, when used in standard ways (as discussed in upcoming slides), they are fairly straightforward to use.
A thread puts itself to sleep and joins the queue of threads associated with a condition variable by calling pthread_cond_wait. When it places this call, it must have some mutex locked, and it passes the mutex as the second argument. As part of the call, the mutex is unlocked and the thread is put to sleep, all in a single atomic step: i.e., nothing can happen that might affect the thread between the moments when the mutex is unlocked and when the thread goes to sleep. Threads queued on a condition variable are released in first-in-first-out order within priority levels.
So far, though complicated, the description is rational. Now for the weird part: a thread may be released from the condition-variable queue at any moment, perhaps spontaneously, perhaps due to sun spots. In addition, the first thread in line is released in response to some other thread’s calling pthread_cond_signal. All threads currently in line are released in response to a thread’s calling pthread_cond_broadcast. Though almost all of the time threads are released to calls to these routines, the official POSIX semantics allow threads to be released at any time without provocation. (This, apparently, makes the implementation easier on some platforms.) Note that if pthread_cond_signal or pthread_cond_broadcast is called and no thread is waiting, nothing happens: there is no memory that such a call has occurred.
Once a thread is released, things still aren’t simple. It does not return from pthread_cond_wait until it has retaken the lock on the mutex (the one it passed as its second argument).

13. Condition variable and mutex
A mutex is passed into wait: pthread_cond_wait(cond_var, mutex)
Mutex is unlocked before the thread sleeps
Mutex is locked again before pthread_cond_wait() returns
Safe to use pthread_cond_wait() in a while loop and check condition again before proceeding

14. Example Usage Write a program using two threads
Thread 1 prints “hello”
Thread 2 prints “world”
Thread 2 should wait till thread 1 finishes before printing
Use a condition variable

15. Using condition variables
int thread1_done = 0;
pthread_cond_t cv; pthread_mutex_t mutex;
Thread 1:
printf(“hello “);
pthread_mutex_lock(&mutex);
thread1_done = 1;
pthread_cond_signal(&cv);
pthread_mutex_unlock(&mutex);
Thread 2:
pthread_mutex_lock(&mutex);
pthread_cond_wait(&cv, &mutex);
printf(“ world\n“);
pthread_mutex_unlock(&mutex);
What is the problem in this implementation?

16. Using condition variables
int thread1_done = 0;
pthread_cond_t cv; pthread_mutex_t mutex;
Thread 1:
printf(“hello “);
pthread_mutex_lock(&mutex);
thread1_done = 1;
pthread_cond_signal(&cv);
pthread_mutex_unlock(&mutex);
Thread 2:
pthread_mutex_lock(&mutex);
while (thread1_done == 0) {
pthread_cond_wait(&cv, &mutex); }
printf(“ world\n“);
pthread_mutex_unlock(&mutex);

17. Multiple Locks thread a thread b proc1( ) { pthread_mutex_lock(&m1);
/* use object 1 */
pthread_mutex_lock(&m2);
/* use objects 1 and 2 */
pthread_mutex_unlock(&m2);
pthread_mutex_unlock(&m1); }
proc2( ) {
pthread_mutex_lock(&m2);
/* use object 2 */
pthread_mutex_lock(&m1);
/* use objects 1 and 2 */
pthread_mutex_unlock(&m1);
pthread_mutex_unlock(&m2); }
In this example our threads are using two mutexes to control access to two different objects. Thread 1, executing proc1, first takes mutex 1, then, while still holding mutex 1, obtains mutex 2. Thread 2, executing proc2, first takes mutex 2, then, while still holding mutex 2, obtains mutex 1. However, things do not always work out as planned. If thread 1 obtains mutex 1 and, at about the same time, thread 2 obtains mutex 2, then if thread 1 attempts to take mutex 2 and thread 2 attempts to take mutex 1, we have a deadlock.
thread a
thread b

18. Deadlock Thread 1 Mutex m1 Mutex m2 Thread 2
The slide shows what’s known as a resource graph: a directed graph with two sorts of nodes, representing threads and mutexes (protecting resources). There’s an arc from a mutex to a thread if the thread has that mutex locked. There’s an arc from a thread to a mutex if the thread is waiting to lock that mutex. Clearly, such a graph has a cycle if and only if there is a deadlock.
Thread 2