1. 1 CENG334 Introduction to Operating Systems Erol Sahin Dept of Computer Eng. Middle East Technical University Ankara, TURKEY URL: http://kovan.ceng.metu.edu.tr/~erol/Courses/CENG334 Synchronization Topics: Synchronization problem Race conditions and Critical Sections Mutual exclusion Locks Spinlocks Mutexes Some of the following slides are adapted from Matt Welsh, Harvard Univ.

2. 2 Single and Multithreaded Processes

3. 3 Synchronization Threads cooperate in multithreaded programs in several ways: Access to shared state e.g., multiple threads accessing a memory cache in a Web server To coordinate their execution e.g., Pressing stop button on browser cancels download of current page “stop button thread” has to signal the “download thread” For correctness, we have to control this cooperation Must assume threads interleave executions arbitrarily and at different rates scheduling is not under application’s control We control cooperation using synchronization enables us to restrict the interleaving of executions

4. 4 Shared Resources We’ll focus on coordinating access to shared resources Basic problem: Two concurrent threads are accessing a shared variable If the variable is read/modified/written by both threads, then access to the variable must be controlled Otherwise, unexpected results may occur We’ll look at: Mechanisms to control access to shared resources Low-level mechanisms: locks Higher level mechanisms: mutexes, semaphores, monitors, and condition variables Patterns for coordinating access to shared resources bounded buffer, producer-consumer, … This stuff is complicated and rife with pitfalls Details are important for completing assignments Expect questions on the midterm/final!

5. 5 Shared Variable Example Suppose we implement a function to withdraw money from a bank account: int withdraw(account, amount) { balance = get_balance(account); balance = balance - amount; put_balance(account, balance); return balance; } Now suppose that you and your friend share a bank account with a balance of 1000.00TL What happens if you both go to separate ATM machines, and simultaneously withdraw 100.00TL from the account?

6. 6 Example continued We represent the situation by creating a separate thread for each ATM user doing a withdrawal Both threads run on the same bank server system Thread 1 Thread 2 What’s the problem with this? What are the possible balance values after each thread runs? int withdraw(account, amount) { balance = get_balance(account); balance -= amount; put_balance(account, balance); return balance; } int withdraw(account, amount) { balance = get_balance(account); balance -= amount; put_balance(account, balance); return balance; }

7. 7 Interleaved Execution The execution of the two threads can be interleaved Assume preemptive scheduling Each thread can context switch after each instruction We need to worry about the worst-case scenario! What’s the account balance after this sequence? And who's happier, the bank or you??? balance = get_balance(account); balance -= amount; balance = get_balance(account); balance -= amount; put_balance(account, balance); Execution sequence as seen by CPU context switch

8. 8 Interleaved Execution The execution of the two threads can be interleaved Assume preemptive scheduling Each thread can context switch after each instruction We need to worry about the worst-case scenario! What’s the account balance after this sequence? And who's happier, the bank or you??? balance = get_balance(account); balance -= amount; balance = get_balance(account); balance -= amount; put_balance(account, balance); Execution sequence as seen by CPU Balance = 1000TL Balance = 900TL Balance = 900TL! Local = 900TL

9. 9 Race Conditions The problem is that two concurrent threads access a shared resource without any synchronization This is called a race condition The result of the concurrent access is non-deterministic Result depends on: Timing When context switches occurred Which thread ran at context switch What the threads were doing We need mechanisms for controlling access to shared resources in the face of concurrency This allows us to reason about the operation of programs Essentially, we want to re-introduce determinism into the thread's execution Synchronization is necessary for any shared data structure buffers, queues, lists, hash tables, …

10. 10 Which resources are shared? Local variables in a function are not shared They exist on the stack, and each thread has its own stack You can't safely pass a pointer from a local variable to another thread Why? Global variables are shared Stored in static data portion of the address space Accessible by any thread Dynamically-allocated data is shared Stored in the heap, accessible by any thread Stack for thread 0 Heap Initialized vars (data segment) Code (text segment) Uninitialized vars (BSS segment) Stack for thread 1 Stack for thread 2 (Reserved for OS) Shared Unshared

11. 11 Mutual Exclusion We want to use mutual exclusion to synchronize access to shared resources Meaning: When only one thread can access a shared resource at a time. Code that uses mutual exclusion to synchronize its execution is called a critical section Only one thread at a time can execute code in the critical section All other threads are forced to wait on entry When one thread leaves the critical section, another can enter Critical Section Thread 1 (modify account balance) Adapted from Matt Welsh’s (Harvard University) slides.

12. 12 Mutual Exclusion We want to use mutual exclusion to synchronize access to shared resources Meaning: When only one thread can access a shared resource at a time. Code that uses mutual exclusion to synchronize its execution is called a critical section Only one thread at a time can execute code in the critical section All other threads are forced to wait on entry When one thread leaves the critical section, another can enter Thread 2 Critical Section Thread 1 (modify account balance) 2 nd thread must wait for critical section to clear Adapted from Matt Welsh’s (Harvard University) slides.

13. 13 Mutual Exclusion We want to use mutual exclusion to synchronize access to shared resources Meaning: When only one thread can access a shared resource at a time. Code that uses mutual exclusion to synchronize its execution is called a critical section Only one thread at a time can execute code in the critical section All other threads are forced to wait on entry When one thread leaves the critical section, another can enter Critical Section Thread 1 (modify account balance) 1 st thread leaves critical section Thread 2 2 nd thread free to enter Adapted from Matt Welsh’s (Harvard University) slides.

14. 14 Critical Section Requirements Mutual exclusion At most one thread is currently executing in the critical section Progress If thread T1 is outside the critical section, then T1 cannot prevent T2 from entering the critical section Bounded waiting (no starvation) If thread T1 is waiting on the critical section, then T1 will eventually enter the critical section Assumes threads eventually leave critical sections Performance The overhead of entering and exiting the critical section is small with respect to the work being done within it Adapted from Matt Welsh’s (Harvard University) slides.

15. 15 Locks A lock is a object (in memory) that provides the following two operations: –acquire( ): a thread calls this before entering a critical section May require waiting to enter the critical section –release( ): a thread calls this after leaving a critical section Allows another thread to enter the critical section A call to acquire( ) must have a corresponding call to release( ) Between acquire( ) and release( ), the thread holds the lock acquire( ) does not return until the caller holds the lock At most one thread can hold a lock at a time (usually!) We'll talk about the exceptions later... What can happen if acquire( ) and release( ) calls are not paired? Adapted from Matt Welsh’s (Harvard University) slides.

16. 16 Using Locks int withdraw(account, amount) { acquire(lock); balance = get_balance(account); balance -= amount; put_balance(account, balance); release(lock); return balance; } critical section Adapted from Matt Welsh’s (Harvard University) slides.

17. 17 Execution with Locks What happens when the blue thread tries to acquire the lock? acquire(lock); balance = get_balance(account); balance -= amount; balance = get_balance(account); balance -= amount; put_balance(account, balance); release(lock); put_balance(account, balance); release(lock); acquire(lock); Thread 1 runs Thread 2 waits on lock Thread 1 completes Thread 2 resumes Adapted from Matt Welsh’s (Harvard University) slides.

18. 18 Spinlocks Very simple way to implement a lock: Why doesn't this work? Where is the race condition? struct lock { int held = 0; } void acquire(lock) { while (lock->held); lock->held = 1; } void release(lock) { lock->held = 0; } The caller busy waits for the lock to be released Adapted from Matt Welsh’s (Harvard University) slides.

19. 19 Implementing Spinlocks Problem is that the internals of the lock acquire/release have critical sections too! The acquire( ) and release( ) actions must be atomic Atomic means that the code cannot be interrupted during execution “All or nothing” execution struct lock { int held = 0; } void acquire(lock) { while (lock->held); lock->held = 1; } void release(lock) { lock->held = 0; } What can happen if there is a context switch here? Adapted from Matt Welsh’s (Harvard University) slides.

20. 20 Implementing Spinlocks Problem is that the internals of the lock acquire/release have critical sections too! The acquire( ) and release( ) actions must be atomic Atomic means that the code cannot be interrupted during execution “All or nothing” execution struct lock { int held = 0; } void acquire(lock) { while (lock->held); lock->held = 1; } void release(lock) { lock->held = 0; } This sequence needs to be atomic Adapted from Matt Welsh’s (Harvard University) slides.

21. 21 Implementing Spinlocks Problem is that the internals of the lock acquire/release have critical sections too! The acquire( ) and release( ) actions must be atomic Atomic means that the code cannot be interrupted during execution “All or nothing” execution Doing this requires help from hardware! Disabling interrupts Why does this prevent a context switch from occurring? Atomic instructions – CPU guarantees entire action will execute atomically Test-and-set Compare-and-swap Adapted from Matt Welsh’s (Harvard University) slides.

22. 22 Spinlocks using test-and-set CPU provides the following as one atomic instruction: So to fix our broken spinlocks, we do this: bool test_and_set(bool *flag) { … // Hardware dependent implementation } struct lock { int held = 0; } void acquire(lock) { while(test_and_set(&lock->held)); } void release(lock) { lock->held = 0; } Adapted from Matt Welsh’s (Harvard University) slides.

23. 23 What's wrong with spinlocks? OK, so spinlocks work (if you implement them correctly), and they are simple. So what's the catch? struct lock { int held = 0; } void acquire(lock) { while(test_and_set(&lock->held)); } void release(lock) { lock->held = 0; } Adapted from Matt Welsh’s (Harvard University) slides.

24. 24 Problems with spinlocks Horribly wasteful! Threads waiting to acquire locks spin on the CPU Eats up lots of cycles, slows down progress of other threads Note that other threads can still run... how? What happens if you have a lot of threads trying to acquire the lock? Only want spinlocks as primitives to build higher-level synchronization constructs Adapted from Matt Welsh’s (Harvard University) slides.

25. 25 Disabling Interrupts An alternative to spinlocks: Can two threads disable/reenable interrupts at the same time? What's wrong with this approach? struct lock { // Note – no state! } void acquire(lock) { cli(); // disable interrupts } void release(lock) { sti(); // reenable interupts } Adapted from Matt Welsh’s (Harvard University) slides.

26. 26 Disabling Interrupts An alternative to spinlocks: Can two threads disable/reenable interrupts at the same time? What's wrong with this approach? Can only be implemented at kernel level (why?) Inefficient on a multiprocessor system (why?) All locks in the system are mutually exclusive No separation between different locks for different bank accounts struct lock { // Note – no state! } void acquire(lock) { cli(); // disable interrupts } void release(lock) { sti(); // reenable interupts } Adapted from Matt Welsh’s (Harvard University) slides.

27. 27 Peterson’s Algorithm flag[0] = 0; flag[1] = 0; turn; P0: flag[0] = 1; turn = 1; while (flag[1] == 1 && turn == 1) { // busy wait } // critical section …. // end of critical section flag[0] = 0; Adapted from Matt Welsh’s (Harvard University) slides. P1: flag[1] = 1; turn = 0; while (flag[0] == 1 && turn == 0) { // busy wait } // critical section …. // end of critical section flag[1] = 0; The algorithm uses two variables, flag and turn. A flag value of 1 indicates that the process wants to enter the critical section. The variable turn holds the ID of the process whose turn it is. Entrance to the critical section is granted for process P0 if P1 does not want to enter its critical section or if P1 has given priority to P0 by setting turn to 0.

28. 28 Mutexes – Blocking Locks Really want a thread waiting to enter a critical section to block Put the thread to sleep until it can enter the critical section Frees up the CPU for other threads to run Straightforward to implement using our TCB queues! Lock wait queue Thread 1 unlocked Lock state Ø 1) Check lock state ??? Adapted from Matt Welsh’s (Harvard University) slides.

29. 29 Mutexes – Blocking Locks Really want a thread waiting to enter a critical section to block Put the thread to sleep until it can enter the critical section Frees up the CPU for other threads to run Straightforward to implement using our TCB queues! Lock wait queue Thread 1 Ø 1) Check lock state 2) Set state to locked 3) Enter critical section locked Lock state Adapted from Matt Welsh’s (Harvard University) slides.

30. 30 Mutexes – Blocking Locks Really want a thread waiting to enter a critical section to block Put the thread to sleep until it can enter the critical section Frees up the CPU for other threads to run Straightforward to implement using our TCB queues! Lock wait queue Thread 1 Ø 1) Check lock state locked Lock state Thread 2 ??? Adapted from Matt Welsh’s (Harvard University) slides.

31. 31 Mutexes – Blocking Locks Really want a thread waiting to enter a critical section to block Put the thread to sleep until it can enter the critical section Frees up the CPU for other threads to run Straightforward to implement using our TCB queues! Lock wait queue Thread 1 1) Check lock state locked Lock state Thread 2 2) Add self to wait queue (sleep) Thread 2 Ø Adapted from Matt Welsh’s (Harvard University) slides.

32. 32 Mutexes – Blocking Locks Really want a thread waiting to enter a critical section to block Put the thread to sleep until it can enter the critical section Frees up the CPU for other threads to run Straightforward to implement using our TCB queues! Lock wait queue Thread 1 1) Check lock state locked Lock state Thread 3 2) Add self to wait queue (sleep) Thread 2 ??? Thread 3 Adapted from Matt Welsh’s (Harvard University) slides.

33. 33 Mutexes – Blocking Locks Really want a thread waiting to enter a critical section to block Put the thread to sleep until it can enter the critical section Frees up the CPU for other threads to run Straightforward to implement using our TCB queues! Lock wait queue Thread 1 1) Thread 1 finishes critical section locked Lock state Thread 2Thread 3 Adapted from Matt Welsh’s (Harvard University) slides.

34. 34 Mutexes – Blocking Locks Really want a thread waiting to enter a critical section to block Put the thread to sleep until it can enter the critical section Frees up the CPU for other threads to run Straightforward to implement using our TCB queues! Lock wait queue Thread 1 1) Thread 1 finishes critical section 2) Reset lock state to unlocked Thread 2Thread 3 3) Wake one thread from wait queue unlocked Lock state Thread 3 Adapted from Matt Welsh’s (Harvard University) slides.

35. 35 Mutexes – Blocking Locks Really want a thread waiting to enter a critical section to block Put the thread to sleep until it can enter the critical section Frees up the CPU for other threads to run Straightforward to implement using our TCB queues! Lock wait queue Thread 3 can now grab lock and enter critical section Thread 2Thread 3 locked Lock state Adapted from Matt Welsh’s (Harvard University) slides.

36. 36 Limitations of locks Locks are great, and simple. What can they not easily accomplish? What if you have a data structure where it's OK for many threads to read the data, but only one thread to write the data? Bank account example. Locks only let one thread access the data structure at a time. Adapted from Matt Welsh’s (Harvard University) slides.

37. 37 Limitations of locks Locks are great, and simple. What can they not easily accomplish? What if you have a data structure where it's OK for many threads to read the data, but only one thread to write the data? Bank account example. Locks only let one thread access the data structure at a time. What if you want to protect access to two (or more) data structures at a time? e.g., Transferring money from one bank account to another. Simple approach: Use a separate lock for each. What happens if you have transfer from account A -> account B, at the same time as transfer from account B -> account A? Hmmmmm... tricky. We will get into this next time. Adapted from Matt Welsh’s (Harvard University) slides.

38. 38 Next Lecture Higher level synchronization primitives: How do to fancier stuff than just locks Semaphores, monitors, and condition variables Implemented using basic locks as a primitive Allow applications to perform more complicated coordination schemes Adapted from Matt Welsh’s (Harvard University) slides.

39. 39 39 CENG334 Introduction to Operating Systems Erol Sahin Dept of Computer Eng. Middle East Technical University Ankara, TURKEY URL: http://kovan.ceng.metu.edu.tr/~erol/Courses/CENG334 Semaphores Topics: Need for higher-level synchronization primitives Semaphores and their implementation The Producer/Consumer problem and its solution with semaphores The Reader/Writer problem and its solution with semaphores Some of the following slides are adapted from Matt Welsh, Harvard Univ.

40. 40 Higher-level synchronization primitives We have looked at one synchronization primitive: locks Locks are useful for many things, but sometimes programs have different requirements. Examples? Say we had a shared variable where we wanted any number of threads to read the variable, but only one thread to write it. How would you do this with locks? What's wrong with this code? Reader() { lock.acquire(); mycopy = shared_var; lock.release(); return mycopy; } Writer() { lock.acquire(); shared_var = NEW_VALUE; lock.release(); } Adapted from Matt Welsh’s (Harvard University) slides.

41. 41 Semaphores Higher-level synchronization construct Designed by Edsger Dijkstra in the 1960's, part of the THE operating system (classic stuff!) Semaphore is a shared counter Two operations on semaphores: P() or wait() or down() From Dutch “proeberen”, meaning “test” Atomic action: Wait for semaphore value to become > 0, then decrement it V() or signal() or up() From Dutch “verhogen”, meaning “increment” Atomic action: Increments semaphore value by 1. Semaphore Adapted from Matt Welsh’s (Harvard University) slides.

42. 42 Semaphore Example Semaphores can be used to implement locks: A semaphore where the counter value is only 0 or 1 is called a binary semaphore. int withdraw(account, amount) { down(my_semaphore); balance = get_balance(account); balance -= amount; put_balance(account, balance); up(my_semaphore); return balance; } critical section Semaphore my_semaphore = 1; // Initialize to nonzero Adapted from Matt Welsh’s (Harvard University) slides.

43. 43 Simple Semaphore Implementation What's wrong with this picture??? struct semaphore { int val; threadlist L; // List of threads waiting for semaphore } down(semaphore S): // Wait until > 0 then decrement while (S.val <= 0) { add this thread to S.L; block(this thread); } S.val = S.val -1; return; up(semaphore S): // Increment value and wake up next thread S.val = S.val + 1; if (S.L is nonempty) { remove a thread T from S.L; wakeup(T); } Why is this a while loop and not just an if statement? Adapted from Matt Welsh’s (Harvard University) slides.

44. 44 Simple Semaphore Implementation down() and up() must be atomic actions! struct semaphore { int val; threadlist L; // List of threads waiting for semaphore } down(semaphore S): // Wait until > 0 then decrement while (S.val <= 0) { add this thread to S.L; block(this thread); } S.val = S.val -1; return; up(semaphore S): // Increment value and wake up next thread S.val = S.val + 1; if (S.L is nonempty) { remove a thread T from S.L; wakeup(T); } Adapted from Matt Welsh’s (Harvard University) slides.

45. 45 Semaphore Implementation How do we ensure that the semaphore implementation is atomic? Adapted from Matt Welsh’s (Harvard University) slides.

46. 46 Semaphore Implementation How do we ensure that the semaphore implementation is atomic? One approach: Make them system calls, and ensure only one down() or up() operation can be executed by any process at a time. This effectively puts a lock around the down() and up() operations themselves! Easy to do by disabling interrupts in the down() and up() calls. Another approach: Use hardware support Say your CPU had atomic down and up instructions Adapted from Matt Welsh’s (Harvard University) slides.

47. 47 OK, but why are semaphores useful? A binary semaphore (counter is always 0 or 1) is basically a lock. The real value of semaphores becomes apparent when the counter can be initialized to a value other than 0 or 1. Say we initialize a semaphore's counter to 50. What does this mean about down() and up() operations? Adapted from Matt Welsh’s (Harvard University) slides.

48. 48 The Producer/Consumer Problem Also called the Bounded Buffer problem. Producer pushes items into the buffer. Consumer pulls items from the buffer. Producer needs to wait when buffer is full. Consumer needs to wait when the buffer is empty. Producer Consumer Mmmm... donuts Adapted from Matt Welsh’s (Harvard University) slides.

49. 49 The Producer/Consumer Problem Also called the Bounded Buffer problem. Producer pushes items into the buffer. Consumer pulls items from the buffer. Producer needs to wait when buffer is full. Consumer needs to wait when the buffer is empty. Producer Consumer zzzzz.... Adapted from Matt Welsh’s (Harvard University) slides.

50. 50 One implementation... Producer Consumer int count = 0; Producer() { int item; while (TRUE) { item = bake(); if (count == N) sleep(); insert_item(item); count = count + 1; if (count == 1) wakeup(consumer); } Consumer() { int item; while (TRUE) { if (count == 0) sleep(); item = remove_item(); count = count – 1; if (count == N-1) wakeup(producer); eat(item); } What's wrong with this code? What if we context switch right here?? Adapted from Matt Welsh’s (Harvard University) slides.

51. 51 A fix using semaphores Producer Consumer Semaphore mutex = 1; Semaphore empty = N; Semaphore full = 0; Producer() { int item; while (TRUE) { item = bake(); down(empty); down(mutex); insert_item(item); up(mutex); up(full); } Consumer() { int item; while (TRUE) { down(full); down(mutex); item = remove_item(); up(mutex); up(empty); eat(item); } Adapted from Matt Welsh’s (Harvard University) slides.

52. 52 Reader/Writers Let's go back to the problem at the beginning of lecture. Single shared object Want to allow any number of threads to read simultaneously But, only one thread should be able to write to the object at a time (And, not interfere with any readers...) Semaphore mutex = 1; Semaphore wrt = 1; int readcount = 0; Writer() { down(wrt); do_write(); up(wrt); } Reader() { down(mutex); readcount++; if (readcount == 1) { down(wrt); } up(mutex); do_read(); down(mutex); readcount--; if (readcount == 0) { up(wrt); } up(mutex); } Why the test here?? Adapted from Matt Welsh’s (Harvard University) slides. ✔ A Reader should only wait for a Writer to complete its do_write(). ✔ A Reader should not wait for other Readers to complete their do_read(). ✔ The Writer should wait for the other Writers to complete their do_write(). ✔ The Writer should wait for all the Readers to complete their do_read().

53. 53 Issues with Semaphores Much of the power of semaphores derives from calls to down() and up() that are unmatched See previous example! Unlike locks, acquire() and release() are not always paired. This means it is a lot easier to get into trouble with semaphores. “More rope” Would be nice if we had some clean, well-defined language support for synchronization... Java does! Adapted from Matt Welsh’s (Harvard University) slides.