Lect04
1 like763 views
Mutual exclusion is required when multiple threads or processes access shared resources concurrently to prevent race conditions. Critical regions define sections of code where shared resources are accessed exclusively by one thread at a time. Semaphores and monitors provide synchronization mechanisms for processes to coordinate access to shared resources and critical regions in a way that prevents race conditions and ensures progress.
Lect04
- 2. Learning Outcomes • Understand concurrency is an issue in operating systems and multithreaded applications • Know the concept of a critical region. • Understand how mutual exclusion of critical regions can be used to solve concurrency issues – Including how mutual exclusion can be implemented correctly and efficiently. • Be able to identify and solve a producer consumer bounded buffer problem. • Understand and apply standard synchronisation primitives to solve synchronisation problems. 2
- 3. Textbook • Sections 2.3 & 2.5 3
- 4. Making Single-Threaded Code Multithreaded Conflicts between threads over the use of a global variable 4
- 5. Inter- Thread and Process Communication We have a race condition Two processes want to access shared memory at same time 5
- 6. Critical Region • We can control access to the shared resource by controlling access to the code that accesses the resource. ⇒ A critical region is a region of code where shared resources are accessed. – Variables, memory, files, etc… • Uncoordinated entry to the critical region results in a race condition ⇒ Incorrect behaviour, deadlock, lost work,… 6
- 7. Critical Regions Mutual exclusion using critical regions 7
- 8. Example critical sections struct node { void insert(struct *item) int data; { struct node *next; item->next = head; }; head = item; struct node *head; } void init(void) struct node *remove(void) { { head = NULL; struct node *t; } t = head; if (t != NULL) { head = head->next; • Simple last-in-first-out queue } implemented as a linked list. return t; } 8
- 9. Example Race void insert(struct *item) void insert(struct *item) { { item->next = head; item->next = head; head = item; head = item; } } 9
- 10. Example critical sections struct node { void insert(struct *item) int data; { struct node *next; item->next = head; }; head = item; struct node *head; } void init(void) struct node *remove(void) { { head = NULL; struct node *t; } t = head; if (t != NULL) { head = head->next; • Critical sections } return t; } 10
- 11. Critical Regions Also called critical sections Conditions required of any solution to the critical region problem Mutual Exclusion: No two processes simultaneously in critical region No assumptions made about speeds or numbers of CPUs Progress No process running outside its critical region may block another process Bounded No process waits forever to enter its critical region 11
- 12. A solution? • A lock variable – If lock == 1, • somebody is in the critical section and we must wait – If lock == 0, • nobody is in the critical section and we are free to enter 12
- 13. A solution? while(TRUE) { while(TRUE) { while(lock == 1); while(lock == 1); lock = 1; lock = 1; critical(); critical(); lock = 0 lock = 0 non_critical(); non_critical(); } } 13
- 14. A problematic execution sequence while(TRUE) { while(TRUE) { while(lock == 1); while(lock == 1); lock = 1; lock = 1; critical(); critical(); lock = 0 non_critical(); } lock = 0 non_critical(); } 14
- 15. Observation • Unfortunately, it is usually easier to show something does not work, than it is to prove that it does work. – Ideally, we’d like to prove, or at least informally demonstrate, that our solutions work. 15
- 16. Mutual Exclusion by Taking Turns Proposed solution to critical region problem (a) Process 0. (b) Process 1. 16
- 17. 17
- 18. Mutual Exclusion by Taking Turns • Works due to strict alternation – Each process takes turns • Cons – Busy waiting – Process must wait its turn even while the other process is doing something else. • With many processes, must wait for everyone to have a turn – Does not guarantee progress if a process no longer needs a turn. • Poor solution when processes require the critical section at differing rates 18
- 19. Peterson’s Solution • See the textbook 19
- 20. Mutual Exclusion by Disabling Interrupts • Before entering a critical region, disable interrupts • After leaving the critical region, enable interrupts • Pros – simple • Cons – Only available in the kernel – Blocks everybody else, even with no contention • Slows interrupt response time – Does not work on a multiprocessor 20
- 21. Hardware Support for mutual exclusion • Test and set instruction – Can be used to implement lock variables correctly • It loads the value of the lock • If lock == 0, – set the lock to 1 – return the result 0 – we acquire the lock • If lock == 1 – return 1 – another thread/process has the lock – Hardware guarantees that the instruction executes atomically. • Atomically: As an indivisible unit. 21
- 22. Mutual Exclusion with Test-and-Set Entering and leaving a critical region using the TSL instruction 22
- 23. Test-and-Set • Pros – Simple (easy to show it’s correct) – Available at user-level • To any number of processors • To implement any number of lock variables • Cons – Busy waits (also termed a spin lock) • Consumes CPU • Livelock in the presence of priorities – If a low priority process has the lock and a high priority process attempts to get it, the high priority process will busy-wait forever. • Starvation is possible when a process leaves its critical section and more than one process is waiting. 23
- 24. Tackling the Busy-Wait Problem • Sleep / Wakeup – The idea • When process is waiting for an event, it calls sleep to block, instead of busy waiting. • The the event happens, the event generator (another process) calls wakeup to unblock the sleeping process. 24
- 25. The Producer-Consumer Problem • Also called the bounded buffer problem • A producer produces data items and stores the items in a buffer • A consumer takes the items out of the buffer and consumes them. Producer X X X Consumer 25
- 26. Issues • We must keep an accurate count of items in buffer – Producer • can sleep when the buffer is full, • and wakeup when there is empty space in the buffer – The consumer can call wakeup when it consumes the first entry of the full buffer – Consumer • Can sleep when the buffer is empty • And wake up when there are items available – Producer can call wakeup when it adds the first item to the buffer Producer X X X Consumer 26
- 27. Pseudo-code for producer and consumer int count = 0; con() { #define N 4 /* buf size */ while(TRUE) { prod() { if (count == 0) while(TRUE) { sleep(); item = produce() remove_item(); if (count == N) count--; sleep(); if (count == N-1) insert_item(); wakeup(prod); count++; } if (count == 1) } wakeup(con); } } 27
- 28. Problems int count = 0; con() { #define N 4 /* buf size */ while(TRUE) { prod() { if (count == 0) while(TRUE) { sleep(); item = produce() remove_item(); if (count == N) count--; sleep(); if (count == N-1) insert_item(); wakeup(prod); count++; } if (count == 1) } Concurrent wakeup(con); uncontrolled } access to the buffer } 28
- 29. Problems int count = 0; con() { #define N 4 /* buf size */ while(TRUE) { prod() { if (count == 0) while(TRUE) { sleep(); item = produce() remove_item(); if (count == N) count--; sleep(); if (count == N-1) insert_item(); wakeup(prod); count++; } if (count == 1) } Concurrent wakeup(con); uncontrolled } access to the counter } 29
- 30. Proposed Solution • Lets use a locking primitive based on test- and-set to protect the concurrent access 30
- 31. Proposed solution? int count = 0; con() { #define N 4 /* buf size */ while(TRUE) { prod() { if (count == 0) while(TRUE) { sleep(); item = produce() acquire_lock() if (count == N) sleep(); remove_item(); acquire_lock() count--; insert_item(); release_lock(); count++; if (count == N-1) release_lock() wakeup(prod); if (count == 1) } wakeup(con); } } } 31
- 32. Problematic execution sequence con() { while(TRUE) { if (count == 0) prod() { while(TRUE) { item = produce() if (count == N) wakeup without a sleep(); matching sleep is acquire_lock() lost insert_item(); count++; release_lock() if (count == 1) wakeup(con); sleep(); acquire_lock() remove_item(); count--; release_lock(); if (count == N-1) wakeup(prod); } 32 }
- 33. Problem • The test for some condition and actually going to sleep needs to be atomic • The following does not work acquire_lock() if (count == N) sleep(); release_lock() The lock is held while asleep ⇒ count will never change 33
- 34. Semaphores • Dijkstra (1965) introduced two primitives that are more powerful than simple sleep and wakeup alone. – P(): proberen, from Dutch to test. – V(): verhogen, from Dutch to increment. – Also called wait & signal, down & up. 34
- 35. How do they work • If a resource is not available, the corresponding semaphore blocks any process waiting for the resource • Blocked processes are put into a process queue maintained by the semaphore (avoids busy waiting!) • When a process releases a resource, it signals this by means of the semaphore • Signalling resumes a blocked process if there is any • Wait and signal operations cannot be interrupted • Complex coordination can be implemented by multiple semaphores 35
- 36. Semaphore Implementation • Define a semaphore as a record typedef struct { int count; struct process *L; } semaphore; • Assume two simple operations: – sleep suspends the process that invokes it. – wakeup(P) resumes the execution of a blocked process P. 36
- 37. • Semaphore operations now defined as wait(S): S.count--; if (S.count < 0) { add this process to S.L; sleep; } signal(S): S.count++; if (S.count <= 0) { remove a process P from S.L; wakeup(P); } • Each primitive is atomic 37
- 38. Semaphore as a General Synchronization Tool • Execute B in Pj only after A executed in Pi • Use semaphore count initialized to 0 • Code: Pi Pj M M A wait(flag) signal(flag) B 38
- 39. Semaphore Implementation of a Mutex • Mutex is short for Mutual Exclusion – Can also be called a lock semaphore mutex; mutex.count = 1; /* initialise mutex */ wait(mutex); /* enter the critcal region */ Blahblah(); signal(mutex); /* exit the critical region */ Notice that the initial count determines how many waits can progress before blocking and requiring a signal ⇒ mutex.count initialised as 1 39
- 40. Solving the producer-consumer problem with semaphores #define N = 4 semaphore mutex = 1; /* count empty slots */ semaphore empty = N; /* count full slots */ semaphore full = 0; 40
- 41. Solving the producer-consumer problem with semaphores prod() { con() { while(TRUE) { while(TRUE) { item = produce() wait(full); wait(empty); wait(mutex); wait(mutex) remove_item(); insert_item(); signal(mutex); signal(mutex); signal(empty); signal(full); } } } } 41
- 42. Summarising Semaphores • Semaphores can be used to solve a variety of concurrency problems • However, programming with then can be error-prone – E.g. must signal for every wait for mutexes • Too many, or too few signals or waits, or signals and waits in the wrong order, can have catastrophic results 42
- 43. Monitors • To ease concurrent programming, Hoare (1974) proposed monitors. – A higher level synchronisation primitive – Programming language construct • Idea – A set of procedures, variables, data types are grouped in a special kind of module, a monitor. • Variables and data types only accessed from within the monitor – Only one process/thread can be in the monitor at any one time • Mutual exclusion is implemented by the compiler (which should be less error prone) 43
- 44. Monitor • When a thread calls a monitor procedure that has a thread already inside, it is queued and it sleeps until the current thread exits the monitor. 44
- 45. Monitors Example of a monitor 45
- 46. Simple example monitor counter { Note: “paper” language int count; • Compiler guarantees procedure inc() { count = count + 1; only one thread can } be active in the procedure dec() { monitor at any one count = count –1; time } • Easy to see this } provides mutual exclusion – No race condition on count. 46
- 47. How do we block waiting for an event? • We need a mechanism to block waiting for an event (in addition to ensuring mutual exclusion) – e.g., for producer consumer problem when buffer is empty or full • Condition Variables 47
- 48. Condition Variable • To allow a process to wait within the monitor, a condition variable must be declared, as condition x, y; • Condition variable can only be used with the operations wait and signal. – The operation x.wait(); means that the process invoking this operation is suspended until another process invokes x.signal(); – The x.signal operation resumes exactly one suspended process. If no process is suspended, then the signal operation has no effect. 48
- 49. Condition Variables 49
- 50. Monitors • Outline of producer-consumer problem with monitors – only one monitor procedure active at one time 50 – buffer has N slots
- 51. OS/161 Provided Synchronisation Primitives • Locks • Semaphores • Condition Variables 51
- 52. Locks • Functions to create and destroy locks struct lock *lock_create(const char *name); void lock_destroy(struct lock *); • Functions to acquire and release them void lock_acquire(struct lock *); void lock_release(struct lock *); 52
- 53. Example use of locks int count; procedure inc() { struct lock *count_lock lock_acquire(count_lock); count = count + 1; main() { lock_release(count_lock); count = 0; } count_lock = procedure dec() { lock_create(“count lock_acquire(count_lock); lock”); count = count –1; if (count_lock == NULL) lock_release(count_lock); panic(“I’m dead”); } stuff(); } 53
- 54. Semaphores struct semaphore *sem_create(const char *name, int initial_count); void sem_destroy(struct semaphore *); void P(struct semaphore *); void V(struct semaphore *); 54
- 55. Example use of Semaphores int count; procedure inc() { struct semaphore P(count_mutex); *count_mutex; count = count + 1; V(count_mutex); main() { } count = 0; procedure dec() { count_mutex = P(count_mutex); sem_create(“count”, count = count –1; 1); V(count_mutex); if (count_mutex == NULL) } panic(“I’m dead”); stuff(); } 55
- 56. Condition Variables struct cv *cv_create(const char *name); void cv_destroy(struct cv *); void cv_wait(struct cv *cv, struct lock *lock); – Releases the lock and blocks – Upon resumption, it re-acquires the lock • Note: we must recheck the condition we slept on void cv_signal(struct cv *cv, struct lock *lock); void cv_broadcast(struct cv *cv, struct lock *lock); – Wakes one/all, does not release the lock – First “waiter” scheduled after signaller releases the lock will re- acquire the lock Note: All three variants must hold the lock passed in. 56
- 57. Condition Variables and Bounded Buffers Non-solution Solution lock_acquire(c_lock) lock_acquire(c_lock) while (count == 0) if (count == 0) cv_wait(c_cv, c_lock); sleep(); remove_item(); remove_item(); count--; count--; lock_release(c_lock); lock_release(c_lock); 57
- 58. A Producer-Consumer Solution Using OS/161 CVs int count = 0; #define N 4 /* buf size */ prod() { con() { while(TRUE) { while(TRUE) { item = produce() lock_acquire(l) lock_aquire(l) while (count == 0) while (count == N) cv_wait(e,l); cv_wait(f,l); item = remove_item(); insert_item(item); count--; count++; if (count == N-1) if (count == 1) cv_signal(f,l); cv_signal(e,l); lock_release(l); lock_release() consume(item); } } } } 58
- 59. Dining Philosophers • Philosophers eat/think • Eating needs 2 forks • Pick one fork at a time • How to prevent deadlock 59
- 60. Dining Philosophers 60 Solution to dining philosophers problem (part 1)
- 61. Dining Philosophers A nonsolution to the dining philosophers problem 61
- 62. Dining Philosophers 62 Solution to dining philosophers problem (part 2)
- 63. The Readers and Writers Problem • Models access to a database • E.g. airline reservation system – Can have more than one concurrent reader • To check schedules and reservations – Writers must have exclusive access • To book a ticket or update a schedule 63
- 64. The Readers and Writers Problem A solution to the readers and writers problem 64