Synchronization, Part 2 Semaphores Operating Systems Synchronization, Part 2 Semaphores

Semaphores Enable simple synchronization An interface of atomic functions supplying mutual exclusion Programmers don’t need to bother with synchronization algorithms

Counting semaphore Init(i) down(S) [the ‘p’ operation] S = i down(S) [the ‘p’ operation] If (S≤0) the process is blocked. It will resume execution only after it is woken-up Else S-- up(S) [the `v’ operation] If (there are blocked processes) wake-up one of them Else S++ Semaphore’s interface doesn’t enforce the implementation of starvation freedom. All operations are ATOMIC! That is, up(s) is more than just s:=s+1 There is no way to access the semaphore’s internal value. Any attempt to access it is a mistake. Always remember to initialize the semaphore

Negative-valued semaphore down(S) S-- If (S<0) the process is blocked. It will resume execution only after it is woken-up up(S) [the `v’ operation] S++ If (there are blocked processes) wake-up one of them NOTE: If S is negative, there are |S| blocked processes

Counting Semaphores (Barz) Create a counting semaphore, S, by using binary semaphores Variables ? S.value=init_value binary-semaphore: S1=1, S2=min(1, init_value), ? ? down(S) down(S2); down(S1); S.value--; if (S.value>0) then up(S2); up(S1); up(S): down(S1); S.value++; if (S.value == 1) then up(S2); up(S1);

Question 1 Consider the following code snippets run by two processes, P and Q: Add the minimal number of semaphores, initialize them and place them in the code snippets, so the result of the calculation will be correct (52=25). Shared memory n=5; sqr=0; Process P loopP: if (n==0) goto endP; n=n-1; goto loopP; endP: print(sqr) Process Q loopQ: sqr = sqr + 2*n +1; goto loopQ;

Question 1 Note that we can break 25 in more than one way: 25=11+9+5=[(2*5+1)+(2*4+1)+(2*2+1)] 25=9+7+5+3+1=[(2*4+1)+(2*3+1)+(2*2+1)+(2*1+1)+(0+1)] Shared memory n=5; sqr=0; semA=1; semB=0; Process P loopP: if (n==0) goto endP; down(semA); n=n-1; up(semB); goto loopP; endP: print(sqr) Process Q loopQ: down(semB); sqr = sqr + 2*n +1; up(semA); goto loopQ; There are many more ways to break 25: 9+9+7 7+7+7+3+1 1+1+…+1 The solution 9+7+5+3+1 is easiest (and requires the least number of semaphores) because it results in a simple interleaved scheduling.

Question 2 (2007 Moed A) The following constraint graph is a DAG that defines a partial order over code lines. Each vertex is associated with a single line of code in a possible program. A directed edge e(u,v) is used to represent the precedence constraint: code line u must be completed prior to the beginning of code line v. For example, code line S1 should be completed before S2, S5 or S8 are executed, while S6 can only be executed after both S2 and S5 were completed. 8

Question 2a The following code is executed by two processes, A and B. Process A Process B S1 S2 S5 S3 S8 S6 S9 S4 S7

Question 2a Use two counting semaphores with properly initialized values so that in every execution of A and B the execution order of code lines will be consistent with the constraint graph defined above. That is, add up and down operations within A and B’s code lines so that no precedence constraint is violated. You may assume that in every execution both A and B run their code only once. Note: Partial scoring will be given to solutions using more than two semaphores. 10

Question 2a Semaphores: semA=0, semB=0 Process A S1 SemB.up S5 S8 S9 SemA.down S7 Process B SemB.down S2 S3 S6 SemA.up S4 11

Question 2b Give a concise but accurate explanation why a single semaphore is insufficient for maintaining consistency of the above constraint graph and the two processes. 12

Question 2b A single semaphore is insufficient because there are constraints which require that A waits until B completes one of its lines and vice versa. In such a state, having A signaling B and immediately waiting for it on the same semaphore will result in either a deadlock or a breaking of one of its constraints. 13

Question 3 – Producer Consumer Problem #define N 100 /* Buffer size */ semaphore mutex = 1; /* access control to critical section */ semaphore empty = N; /* counts empty buffer slots */ semaphore full = 0; /* full slots */ void producer(void) { int item; while(TRUE) { produce_item(&item); /* generate something... */ down(&empty); /* decrement count of empty */ down(&mutex); /* enter critical section */ enter_item(item); /* insert into buffer */ up(&mutex); /* leave critical section */ up(&full); /* increment count of full slots */ } }

Question 3 – Producer Consumer Problem void consumer(void){ int item; while(TRUE){ down(&full); /* decrement count of full */ down(&mutex); /* enter critical section */ remove_item(&item); /* take item from buffer */ up(&mutex); /* leave critical section */ up(&empty); /* update count of empty */ consume_item(item); /* do something... */ } }

Question 3 No mutual exclusion! 1. The red lines of the following code were swapped. How will this affect the algorithm? void producer(void) { int item; while(TRUE) { produce_item(&item); /* generate something... */ down(&empty); /* decrement count of empty */ down(&mutex); /* enter critical section */ up(&mutex); /* leave critical section */ enter_item(item); /* insert into buffer */ up(&full); /* increment count of full slots */ } No mutual exclusion!

Question 3 2. What will happen now? void producer(void) { int item; while(TRUE) { produce_item(&item); /* generate something... */ down(&empty); /* decrement count of empty */ down(&mutex); /* enter critical section */ enter_item(item); /* insert into buffer */ up(&full); /* increment count of full slots */ up(&mutex); /* leave critical section */ } No problems…

Question 3 3. And now? void consumer(void){ int item; while(TRUE){ down(&mutex); /* enter critical section */ down(&full); /* decrement count of full */ remove_item(&item); /* take item from buffer */ up(&mutex); /* leave critical section */ up(&empty); /* update count of empty */ consume_item(item); /* do something... */ } } Deadlock!

Question 4 (Moed B 2010) An unfair semaphore is a semaphore which does not guarantee that the wakeup order of processes is similar to their falling asleep order. It does, however, provide the following simple guarantee: if there are sleeping processes on the semaphore while an up operation is invoked, then one of these processes will be woken up (not necessarily the first amongst the waiting processes to do a down).

Question 4 Now you are required to implement a starvation free mutual exclusion algorithm for three processes using 3 unfair counting semaphores: R, S and T, initialized to 1. You are not allowed to use any other variable but these semaphores. Add your code and complete the entry and exit section of each process. Briefly explain why your code satisfies both mutual exclusion and starvation freedom.

unfair counting semaphores: R, S and T initialized to 1 Question 4 unfair counting semaphores: R, S and T initialized to 1 P1’s code: entry: Critical_section() exit: P2’s code: entry: Critical_section() exit: P3’s code: entry: Critical_section() exit:

unfair counting semaphores: R, S and T initialized to 1 Question 4 unfair counting semaphores: R, S and T initialized to 1 P1’s code: entry: Critical_section() exit: P2’s code: entry: Critical_section() exit: P3’s code: entry: Critical_section() exit: down (S) down (R) down (R) down (T) down (S) down (T) up(R) up(S) up(T) up(R) up(T) up (S)

Question 4 Mutual exclusion: Any process wishing to enter its critical section must successfully complete two ‘down’ operations on two distinct semaphores. Since any process competes over one different “successful down” with each of the other processes, only a single process may successfully enter the critical section at any given moment.

Question 4 Starvation freedom: We first note that there is no starvation problem when using an unfair semaphore with 2 processes (convince yourselves!). Since entrance to the critical section requires passing semaphores which are only shared in pairs no starvation problems will occur.

unfair counting semaphores: R, S and T initialized to 1 Question 4, supplement unfair counting semaphores: R, S and T initialized to 1 Will this solution work? P1’s code: entry: Critical_section() exit: P2’s code: entry: Critical_section() exit: P3’s code: entry: Critical_section() exit: down (S) down (R) down (R) down (T) down (T) down (S) up(R) up(S) up(T) up(R) up(S) up (T) Deadlock!

Question 5 (Midterm 2009) בסעיף זה עליכם לממש event counter תוך שימוש בסמאפורים בינאריים וברגיסטרים (משתנים התומכים בקריאות ובכתיבות בלבד). כפי שלמדתם, event counter E מכיל ערך שלם ומאותחל ל- 0. הוא תומך בשלוש פעולות אטומיות: פעולת Advance(E) מקדמת את ערכו של E ב-1 מערך v-1 לערך v (כאשר v-1 הוא ערכו של E לפני הפעולה) ומעירה את כל התהליכים אשר ישנו על E בהמתנה לערך v. פעולת Await(E,v) גורמת לתהליך הקורא לישון עד אשר ערכו של E מגיע ל-v. אם ערכו של E גדול או שווה ל-v בעת הקריאה לפעולת Await, אזי התהליך ממשיך בריצתו. פעולת Read(E) מחזירה את ערכו הנוכחי של E. ממשו event counter תוך שימוש בסמאפורים בינאריים (ניתן להניח כי הם הוגנים). הניחו כי ישנם N תהליכים המשתמשים ב-E וכי המזהים שלהם הינם 0, 1,…,N-1. כל תהליך יכול להשתמש במשתנה MyID השומר את המזהה שלו.

Question 5 Shared variables int waitval[N] (all are initialized to 0) int v = 0 semaphore wait[N] (all are initialized to 0) semaphore mutex (initialized to 1) Advance(E) down(mutex); E.v++; for (i = 0; i < N; i++) if (waitval[i] == E.v) up(wait[i]); up(mutex); Await(E,v) boolean wait = false; down(mutex) if (v > E.v) { wait = true; waitval[MyId] = v; } up(mutex); if (wait) down(wait[MyId]); Read(E) return E.v;

Question 5b Counting semaphore mutex=1 Counting semaphore b=0 למדתם כי אין זה פשוט לממש סמאפור כללי (counting semaphore) מסמאפורים בינאריים. מסתבר כי גם הכוון ההפוך אינו פשוט. להלן קטע קוד המכיל מימוש של סמאפור בינארי מסמאפורים כלליים (counting semaphores) ומרגיסטרים. Counting semaphore mutex=1 Counting semaphore b=0 register v=0, register waiting=0 procedure down( ) { mutex.down( ) if (v == 1){ v=0 mutex.up( )} else { waiting=waiting+1 mutex.up( ) b.down( ) }} procedure up( ){ mutex.down( ) if (waiting > 0){ waiting=waiting-1 b.up( )} else if (v == 0) v=1 mutex.up( ) }

Question 5b ניתן להניח כי הסמאפורים הכלליים בהם משתמש הקוד שלמעלה הוגנים. עליכם להסביר מדוע קטע הקוד שלמעלה אינו מממש סמאפור בינארי בצורה נכונה. הסבירו במדויק מהי הסמאנטיקה של סמאפור בינארי. בעל טווח הערכים 0, 1בלבד פעולת down: אם ערך הסמאפור הוא 0 אזי חוסם את התהליך, אחרת משנה את ערך הסמאפור ל-0. פעולת up: אם ישנו תהליך שממתין- מעיר אותו, אחרת משנה את ערך הסמאפור ל-1.

Question 5b תארו במדויק תסריט בו מספר תהליכים מבצעים פעולות על המימוש לעיל ובו המימוש אינו מקיים סמאנטיקה זו. Process p1 does down and stops prior to b.down Process p2 does down and stops prior to b.down Process p3 does up and finishes the operation (b == 1) Process p4 does up and finishes the operation (b == 2) Process p5 does down and finishes the operation (b == 1) Process p6 does down and finishes the operation (b == 0) התסריט אינו אפשרי כאשר משתמשים בסמפור בינארי מכיוון שלא יכול להיות מצב שבו שני תהליכים יעשו DOWN ושניהם יצליחו.

POSIX synchronization primitives POSIX defines the following: Semaphore (counting) Mutex Condition variable Read/Write locks

Mutex Resembles a binary semaphore, but has the notion of ownership. I.e., only the thread which locked the mutex is allowed to unlock it. Supports the following operations: pthread_mutex_init (mutex,attr) pthread_mutex_destroy (mutex) pthread_mutex_lock (mutex) pthread_mutex_trylock (mutex) pthread_mutex_unlock (mutex)

pthread_mutex_init (mutex,attr) The attributes allow you to set or get: the type (deadlocking, deadlock-detecting, recursive, etc). the robustness (what happens when you acquire a mutex and the original owner died while possessing it). the process-shared attribute (for sharing a mutex across process boundaries, useful for multi-threading and multi-processing). the protocol (how a thread behaves in terms of priority when a higher-priority thread wants the mutex). the priority ceiling (the minimum priority level at which the critical section will run, a way of preventing priority inversion).

pthread_mutex_lock VS. pthread_mutex_trylock The pthread_mutex_trylock() function shall be equivalent to pthread_mutex_lock(), except that if the mutex object referenced by mutex is currently locked the call shall return immediately. If successful, the pthread_mutex_lock() and pthread_mutex_unlock() functions shall return zero; otherwise, an error number shall be returned to indicate the error. The pthread_mutex_trylock() function shall return zero if a lock on the mutex object referenced by mutex is acquired. Otherwise, an error number is returned to indicate the error.

Condition Variables Are used to wait until a particular condition occurs. Enable threads to atomically release a lock and enter the sleeping state. A mutex must be associated with a condition variable. Supports the following operations: pthread_cond_init (condition, attr) pthread_cond_destroy (condition) pthread_cond_wait (condition, mutex) pthread_cond_signal (condition) pthread_cond_broadcast (condition)

Producer-consumer problem using mutexes and condition variables #include <pthread.h> #include <stdio.h> #define BSIZE 4 #define NUMITEMS 30 typedef struct { char buf[BSIZE]; int occupied; int nextin, nextout; pthread_mutex_t mutex; pthread_cond_t more; pthread_cond_t less; } buffer_t;

buffer_t buffer; void * producer(void *); void * consumer(void *); main( int argc, char *argv[] ) { int i; buffer.occupied = 0; buffer.nextin = buffer.nextout = 0; pthread_cond_init(&(buffer.more), NULL); pthread_cond_init(&(buffer.less), NULL); pthread_t tid[2]; pthread_create(&tid[1], NULL, consumer, NULL); pthread_create(&tid[0], NULL, producer, NULL); for ( i = 0; i < NUM_THREADS; i++) pthread_join(tid[i], NULL); printf("main() reporting that all %d threads have terminated\n", i); }

void * producer(void * parm){ char item[NUMITEMS] = "IT'S A SMALL WORLD, AFTER ALL."; int i; for(i=0;i<NUMITEMS;i++){ /* produce an item, one character from item[] */ if (item[i] == '\0') break; /* stop if at end of string. */ pthread_mutex_lock(&(buffer.mutex)); if (buffer.occupied >= BSIZE) printf("producer waiting.\n"); while (buffer.occupied >= BSIZE) pthread_cond_wait(&(buffer.less), &(buffer.mutex) ); printf("producer executing.\n"); buffer.buf[buffer.nextin++] = item[i]; buffer.nextin %= BSIZE; buffer.occupied++; pthread_cond_signal(&(buffer.more)); pthread_mutex_unlock(&(buffer.mutex)); } After each iteration there is 2 possibilities: either buffer.occupied < BSIZE and buffer.nextin is the index of the next empty slot in the buffer or buffer.occupied == BSIZE and buffer.nextin is the index of the next (occupied) slot that will be emptied by a consumer (such as buffer.nextin == buffer.nextout)

void * consumer(void * parm){ char item; int i; for(i=0;i<NUMITEMS;i++){ pthread_mutex_lock(&(buffer.mutex) ); if (buffer.occupied <= 0) printf("consumer waiting.\n"); while(buffer.occupied <= 0) pthread_cond_wait(&(buffer.more), &(buffer.mutex) ); printf("consumer executing.\n"); item = buffer.buf[buffer.nextout++]; printf("%c\n",item); buffer.nextout %= BSIZE; buffer.occupied--; pthread_cond_signal(&(buffer.less)); pthread_mutex_unlock(&(buffer.mutex)); } After each iteration there is 2 possibilities: either buffer.occupied > 0 and buffer.nextout is the index of the next occupied slot in the buffer or buffer.occupied == 0 and buffer.nextout is the index of the next (empty) slot that will be filled by a producer (such as buffer.nextout == buffer.nextin)