Chapter 6: Synchronization Tools
Operating System Concepts – 10th Edition
Silberschatz, Galvin and Gagne ©2018
Chapter 6: Synchronization Tools
Background
The Critical-Section Problem
Peterson’s Solution
Hardware Support for Synchronization Mutex Locks
Semaphores
Monitors
Liveness
Evaluation
Operating System Concepts – 10th Edition 6.2
Silberschatz, Galvin and Gagne ©2018
Objectives
Describe the critical-section problem and illustrate a race condition
Illustrate hardware solutions to the critical-section problem using memory barriers, compare-and-swap operations, and atomic variables
Demonstrate how mutex locks, semaphores, monitors, and condition variables can be used to solve the critical section problem
Evaluate tools that solve the critical-section problem in low-. Moderate-, and high-contention scenarios
Operating System Concepts – 10th Edition 6.3 Silberschatz, Galvin and Gagne ©2018
Background
Processes can execute concurrently
May be interrupted at any time, partially completing
execution
Concurrent access to shared data may result in data inconsistency
Maintaining data consistency requires mechanisms to ensure the orderly execution of cooperating processes
Illustration of the problem:
Suppose that we wanted to provide a solution to the consumer-producer problem that fills all the buffers. We can do so by having an integer counter that keeps track of the number of full buffers. Initially, counter is set to 0. It is incremented by the producer after it produces a new buffer and is decremented by the consumer after it consumes a buffer.
Operating System Concepts – 10th Edition 6.4 Silberschatz, Galvin and Gagne ©2018
}
Producer
while (true) {
/* produce an item in next produced */
while (counter == BUFFER_SIZE) ; /* do nothing */
buffer[in] = next_produced; in = (in + 1) % BUFFER_SIZE; counter++;
Operating System Concepts – 10th Edition 6.5 Silberschatz, Galvin and Gagne ©2018
while (true) {
while (counter == 0)
}
Consumer
; /* do nothing */
next_consumed = buffer[out];
out = (out + 1) % BUFFER_SIZE;
counter–;
/* consume the item in next consumed */
Operating System Concepts – 10th Edition 6.6 Silberschatz, Galvin and Gagne ©2018
counter++ could be implemented as
Race Condition
register1 = counter
register1 = register1 + 1
counter = register1
counter– could be implemented as
register2 = counter
register2 = register2 – 1
counter = register2
Consider this execution interleaving with “count = 5” initially:
S0: producer execute register1 = counter
S1: producer execute register1 = register1 + 1 S2: consumer execute register2 = counter
S3: consumer execute register2 = register2 – 1 S4: producer execute counter = register1
S5: consumer execute counter = register2
{register1 = 5}
Operating System Concepts – 10th Edition 6.7
Silberschatz, Galvin and Gagne ©2018
{register1 = 6} {register2 = 5} {register2 = 4}
{counter = 6 } {counter = 4}
Race Condition
Processes P0 and P1 are creating child processs using the fork() system call
Race condition on kernel variable next_available_pid which represents the next available process identifier (pid)
Unless there is mutual exclusion, the same pid could be assigned to two different processes!
Operating System Concepts – 10th Edition 6.8 Silberschatz, Galvin and Gagne ©2018
Critical Section Problem
Consider system of n processes {p0, p1, … pn-1}
Each process has critical section segment of code
Process may be changing common variables, updating table, writing file, etc
When one process in critical section, no other may be in its critical section
Critical section problem is to design protocol to solve this
Each process must ask permission to enter critical section in entry section, may follow critical section with exit section, then remainder section
Operating System Concepts – 10th Edition 6.9 Silberschatz, Galvin and Gagne ©2018
General structure of process Pi
Critical Section
Operating System Concepts – 10th Edition 6.10 Silberschatz, Galvin and Gagne ©2018
1. 2.
Mutual Exclusion – If process Pi is executing in its critical section, then no other processes can be executing in their critical sections
Solution to Critical-Section Problem
Progress – If no process is executing in its critical section and there exist some processes that wish to enter their critical section, then the selection of the processes that will enter the critical section next cannot be postponed indefinitely
3. Bounded Waiting – A bound must exist on the number of times that other processes are allowed to enter their critical sections after a process has made a request to enter its critical section and before that request is granted
Assume that each process executes at a nonzero speed No assumption concerning relative speed of the n
processes
Operating System Concepts – 10th Edition 6.11 Silberschatz, Galvin and Gagne ©2018
Critical-Section Handling in OS
Two approaches depending on if kernel is preemptive or non- preemptive
Preemptive – allows preemption of process when running in kernel mode
Non-preemptive – runs until exits kernel mode, blocks, or voluntarily yields CPU
Essentially free of race conditions in kernel mode
Operating System Concepts – 10th Edition 6.12 Silberschatz, Galvin and Gagne ©2018
Peterson’s Solution
Not guaranteed to work on modern architectures! (But good
algorithmic description of solving the problem)
Two process solution
Assume that the load and store machine-language instructions are atomic; that is, cannot be interrupted
The two processes share two variables: int turn;
boolean flag[2]
The variable turn indicates whose turn it is to enter the critical
section
The flag array is used to indicate if a process is ready to enter the critical section. flag[i] = true implies that process Pi is ready!
Operating System Concepts – 10th Edition 6.13 Silberschatz, Galvin and Gagne ©2018
}
Algorithm for Process Pi
while (true){
flag[i] = true;
Operating System Concepts – 10th Edition 6.14
Silberschatz, Galvin and Gagne ©2018
turn = j;
while (flag[j] && turn = = j)
;
/* critical section */
flag[i] = false;
/* remainder section */
Peterson’s Solution (Cont.)
Provable that the three CS requirement are met: 1. Mutual exclusion is preserved
Pi enters CS only if:
either flag[j] = false or turn = i
2. Progress requirement is satisfied
3. Bounded-waiting requirement is met
Operating System Concepts – 10th Edition 6.15 Silberschatz, Galvin and Gagne ©2018
Peterson’s Solution
Although useful for demonstrating an algorithm, Peterson’s Solution is not guaranteed to work on modern architectures.
Understanding why it will not work is also useful for better understanding race conditions.
To improve performance, processors and/or compilers may reorder operations that have no dependencies.
For single-threaded this is ok as the result will always be the same.
For multithreaded the reordering may produce inconsistent or unexpected
results!
Operating System Concepts – 10th Edition 6.16 Silberschatz, Galvin and Gagne ©2018
Two threads share the data: boolean flag = false;
int x = 0;
Thread 1 performs
while (!flag)
;
print x
Thread 2 performs x = 100;
flag = true
What is the expected output?
Operating System Concepts – 10th Edition 6.17
Silberschatz, Galvin and Gagne ©2018
Peterson’s Solution
Peterson’s Solution
100 is the expected output.
However, the operations for Thread 2 may be reordered:
flag = true;
x = 100;
If this occurs, the output may be 0!
The effects of instruction reordering in Peterson’s Solution
This allows both processes to be in their critical section at the same time!
Operating System Concepts – 10th Edition 6.18 Silberschatz, Galvin and Gagne ©2018
Synchronization Hardware
Many systems provide hardware support for implementing the critical section code.
Uniprocessors – could disable interrupts
Currently running code would execute without preemption Generally too inefficient on multiprocessor systems
Operating systems using this not broadly scalable We will look at three forms of hardware support:
1. Memory barriers
2. Hardware instructions 3. Atomic variables
Operating System Concepts – 10th Edition 6.19 Silberschatz, Galvin and Gagne ©2018
Memory Barriers
A memory barrier is an instruction that causes a CPU or compiler to enforce an ordering constraint on memory operations issued before and after the barrier instruction. This typically means that operations issued prior to the barrier are guaranteed to be performed before operations issued after the barrier.
Memory barriers are necessary because most modern CPUs employ performance optimizations that can result in out-of-order execution. This reordering of memory operations (loads and stores) normally goes unnoticed within a single thread of execution, but can cause unpredictable behaviour in concurrent programs and device drivers unless carefully controlled.
Memory barriers are typically used when implementing low-level machine code that operates on memory shared by multiple devices. Memory barriers can be used to implement synchronization primitives on multiprocessor systems.
Operating System Concepts – 10th Edition 6.20 Silberschatz, Galvin and Gagne ©2018
Thread 1 now performs
while (!flag)
memory_barrier();
print x
Thread 2 now performs
x = 100;
memory_barrier();
flag = true
Operating System Concepts – 10th Edition 6.21
Silberschatz, Galvin and Gagne ©2018
Memory Barrier
We could add a memory barrier to the following instructions to ensure Thread 1 outputs 100:
Hardware Instructions
Special hardware instructions that allow us to either test-and-modify the content of a word, or two swap the contents of two words atomically (uninterruptibly.)
Test-and-Set instruction
Compare-and-Swap instruction
Operating System Concepts – 10th Edition 6.22 Silberschatz, Galvin and Gagne ©2018
1. 2. 3.
Executed atomically
Returns the original value of passed parameter Set the new value of passed parameter to true
Definition:
}
test_and_set Instruction
boolean test_and_set (boolean *target)
{
boolean rv = *target;
*target = true;
return rv:
Operating System Concepts – 10th Edition 6.23 Silberschatz, Galvin and Gagne ©2018
Solution using test_and_set()
Shared boolean variable lock, initialized to false Solution:
do {
while (test_and_set(&lock))
; /* do nothing */
} while (true);
/* critical section */
lock = false;
/* remainder section */
Operating System Concepts – 10th Edition 6.24 Silberschatz, Galvin and Gagne ©2018
Definition:
compare_and_swap Instruction
int compare _and_swap(int *value, int expected, int new_value) {
int temp = *value;
if (*value == expected)
*value = new_value;
return temp;
}
1. Executed atomically
2. Returns the original value of passed parameter value
3. Set the variable value the value of the passed parameter new_value but only if *value == expected is true. That is, the swap takes place only under this condition.
Operating System Concepts – 10th Edition 6.25 Silberschatz, Galvin and Gagne ©2018
Solution using compare_and_swap
Shared integer lock initialized to 0; Solution:
while (true){
while (compare_and_swap(&lock, 0, 1) != 0)
; /* do nothing */
/* critical section */
lock = 0;
/* remainder section */
}
Operating System Concepts – 10th Edition 6.26 Silberschatz, Galvin and Gagne ©2018
if (j == i)
lock = 0;
else
waiting[j] = false;
/* remainder section */
}
Operating System Concepts – 10th Edition 6.27
Silberschatz, Galvin and Gagne ©2018
Bounded-waiting Mutual Exclusion with compare-and-swap
while (true) {
waiting[i] = true;
key = 1;
while (waiting[i] && key == 1)
key = compare_and_swap(&lock,0,1);
waiting[i] = false;
/* critical section */
j = (i + 1) % n;
while ((j != i) && !waiting[j])
j = (j + 1) % n;
Atomic Variables
Typically, instructions such as compare-and-swap are used as building blocks for other synchronization tools.
One tool is an atomic variable that provides atomic (uninterruptible) updates on basic data types such as integers and booleans.
For example, the increment() operation on the atomic variable sequence ensures sequence is incremented without interruption:
increment(&sequence);
Operating System Concepts – 10th Edition 6.28 Silberschatz, Galvin and Gagne ©2018
}
Atomic Variables
The increment() function can be implemented as follows:
void increment(atomic_int *v)
{
int temp; do {
temp = *v;
while (temp != (compare_and_swap(v,temp,temp+1)); }
Operating System Concepts – 10th Edition 6.29 Silberschatz, Galvin and Gagne ©2018
Mutex Locks
Previous solutions are complicated and generally inaccessible to application programmers
OS designers build software tools to solve critical section problem
Simplest is mutex lock
Protect a critical section by first acquire() a lock then
release() the lock
Boolean variable indicating if lock is available or not
Calls to acquire() and release() must be atomic Usually implemented via hardware atomic instructions
such as compare-and-swap.
But this solution requires busy waiting
This lock therefore called a spinlock
Operating System Concepts – 10th Edition 6.30 Silberschatz, Galvin and Gagne ©2018
Solution to Critical-section Problem Using Locks
while (true) {
acquire lock
}
Operating System Concepts – 10th Edition 6.31
Silberschatz, Galvin and Gagne ©2018
critical section
release lock
remainder section
release() {
available = true;
}
}
Mutex Lock Definitions
acquire(){
while (!available)
; /* busy wait */
available = false;;
These two functions must be implemented atomically. Both test-and-set and compare-and-swap can be used to implement these functions.
Operating System Concepts – 10th Edition 6.32 Silberschatz, Galvin and Gagne ©2018
wait(S) {
while (S <= 0)
; // busy wait
S--;
}
Definition of the signal() operation
signal(S) { S++;
}
Semaphore
Synchronization tool that provides more sophisticated ways (than Mutex locks) for process to synchronize their activities.
Semaphore S – integer variable
Can only be accessed via two indivisible (atomic) operations
wait() and signal()
(Originally called P() and V())
Definition of the wait() operation
Operating System Concepts – 10th Edition 6.33 Silberschatz, Galvin and Gagne ©2018
S1;
signal(synch);
P2:
wait(synch);
S2;
Semaphore Usage
Counting semaphore – integer value can range over an unrestricted domain
Binary semaphore – integer value can range only between 0 and 1 Same as a mutex lock
Can solve various synchronization problems
Consider P1 and P2 that require S1 to happen before S2
Create a semaphore “synch” initialized to 0 P1:
Can implement a counting semaphore S as a binary semaphore
Operating System Concepts – 10th Edition 6.34 Silberschatz, Galvin and Gagne ©2018
Semaphore Implementation
Must guarantee that no two processes can execute the wait() and signal() on the same semaphore at the same time
Thus, the implementation becomes the critical section problem where the wait and signal code are placed in the critical
section
Could now have busy waiting in critical section implementation
But implementation code is short
Little busy waiting if critical section rarely occupied
Note that applications may spend lots of time in critical sections
and therefore this is not a good solution
Operating System Concepts – 10th Edition 6.35 Silberschatz, Galvin and Gagne ©2018
Semaphore Implementation with no Busy waiting
With each semaphore there is an associated waiting queue Each entry in a waiting queue has two data items:
value (of type integer)
pointer to next record in the list Two operations:
block – place the process invoking the operation on the appropriate waiting queue
wakeup – remove one of processes in the waiting queue and place it in the ready queue
typedef struct { int value;
struct process *list;
} semaphore;
Operating System Concepts – 10th Edition 6.36 Silberschatz, Galvin and Gagne ©2018
}
Implementation with no Busy waiting (Cont.)
wait(semaphore *S) {
S->value–;
if (S->value < 0) {
add this process to S->list;
block(); }
signal(semaphore *S) {
S->value++;
}
if (S->value <= 0) {
remove a process P from S->list;
wakeup(P); }
Operating System Concepts – 10th Edition 6.37 Silberschatz, Galvin and Gagne ©2018
Problems with Semaphores
Incorrect use of semaphore operations:
signal (mutex) …. wait (mutex) wait (mutex) … wait (mutex)
Omitting of wait (mutex) and/or signal (mutex)
These – and others – are examples of what can occur when sempahores and other synchronization tools are used incorrectly.
Operating System Concepts – 10th Edition 6.38 Silberschatz, Galvin and Gagne ©2018
monitor monitor-name
{
Monitors
A high-level abstraction that provides a convenient and effective mechanism for process synchronization
Abstract data type, internal variables only accessible by code within the procedure
Only one process may be active within the monitor at a time Pseudocode syntax of a monitor:
// shared variable declarations
function P1 (…) { …. }
function P2 (…) { …. }
function Pn (…) {……}
initialization code (…) { … }
}
Operating System Concepts – 10th Edition 6.39 Silberschatz, Galvin and Gagne ©2018
Schematic view of a Monitor
Operating System Concepts – 10th Edition 6.40 Silberschatz, Galvin and Gagne ©2018
Condition Variables
condition x, y;
Two operations are allowed on a condition variable:
x.wait() – a process that invokes the operation is suspended until x.signal()
x.signal() – resumes one of processes (if any) that invoked x.wait()
If no x.wait() on the variable, then it has no effect on the variable
Operating System Concepts – 10th Edition 6.41 Silberschatz, Galvin and Gagne ©2018
Monitor with Condition Variables
Operating System Concepts – 10th Edition 6.42 Silberschatz, Galvin and Gagne ©2018
Condition Variables Choices
If process P invokes x.signal(), and process Q is suspended in x.wait(), what should happen next?
Both Q and P cannot execute in parallel. If Q is resumed, then P must wait
Options include
Signal and wait – P waits until Q either leaves the monitor or it
waits for another condition
Signal and continue – Q waits until P either leaves the monitor or it waits for another condition
Both have pros and cons – language implementer can decide
Monitors implemented in Concurrent Pascal compromise
P executing signal immediately leaves the monitor, Q is resumed
Implemented in other languages including Mesa, C#, Java
Operating System Concepts – 10th Edition 6.43 Silberschatz, Galvin and Gagne ©2018
Monitor Implementation Using Semaphores
Variables
semaphore mutex; // (initially = 1)
semaphore next; // (initially = 0)
int next_count = 0;
Each function F will be replaced by
wait(mutex);
…
body of F; …
if (next_count > 0)
signal(next)
else
signal(mutex);
Mutual exclusion within a monitor is ensured
Operating System Concepts – 10th Edition 6.44 Silberschatz, Galvin and Gagne ©2018
Operating System Concepts – 10th Edition
6.45 Silberschatz, Galvin and Gagne ©2018
Monitor Implementation – Condition Variables
For each condition variable x, we have:
semaphore x_sem; // (initially = 0)
int x_count = 0;
The operation x.wait() can be implemented as:
x_count++;
if (next_count > 0)
signal(next);
else
signal(mutex);
wait(x_sem);
x_count–;
Monitor Implementation (Cont.)
The operation x.signal() can be implemented as:
if (x_count > 0) {
next_count++;
signal(x_sem);
wait(next);
next_count–;
}
Operating System Concepts – 10th Edition 6.46 Silberschatz, Galvin and Gagne ©2018
Resuming Processes within a Monitor
If several processes queued on condition variable x, and x.signal() is executed, which process should be resumed?
FCFS frequently not adequate
conditional-wait construct of the form x.wait(c)
Where c is priority number
Process with lowest number (highest priority) is
scheduled next
Operating System Concepts – 10th Edition 6.47 Silberschatz, Galvin and Gagne ©2018
Single Resource allocation
Allocate a single resource among competing processes using priority numbers that specify the maximum time a process plans to use the resource
R.acquire(t); …
access the resurce;
…
R.release;
Where R is an instance of type ResourceAllocator
Operating System Concepts – 10th Edition 6.48 Silberschatz, Galvin and Gagne ©2018
A Monitor to Allocate Single Resource
monitor ResourceAllocator
{
}
boolean busy;
condition x;
void acquire(int time) {
}
void release() {
busy = false;
}
if (busy)
x.wait(time);
busy = true;
busy = FALSE;
x.signal();
}
initialization code() {
Operating System Concepts – 10th Edition 6.49 Silberschatz, Galvin and Gagne ©2018
Liveness
Processes may have to wait indefinitely while trying to acquire a synchronization tool such as a mutex lock or semaphore.
Waiting indefinitely violates the progress and bounded-waiting criteria discussed at the beginning of this chapter.
Liveness refers to a set of properties that a system must satisfy to ensure processes make progress.
Indefinite waiting is an example of a liveness failure.
Operating System Concepts – 10th Edition 6.50 Silberschatz, Galvin and Gagne ©2018
Let S and Q be two semaphores initialized to 1
P0 wait(S);
P1 wait(Q);
wait(Q); …
wait(S); …
signal(S);
signal(Q);
signal(Q);
signal(S);
Liveness
Deadlock – two or more processes are waiting indefinitely for an event that can be caused by only one of the waiting processes
Consider if P0 executes wait(S) and P1 wait(Q). When P0 executes wait(Q), it must wait until P1 executes signal(Q)
However, P1 is waiting until P0 execute signal(S).
Since these signal() operations will never be executed, P0 and P1 are
deadlocked.
Operating System Concepts – 10th Edition 6.51 Silberschatz, Galvin and Gagne ©2018
Other forms of deadlock:
Starvation – indefinite blocking
Liveness
A process may never be removed from the semaphore queue in which it is suspended
Priority Inversion – Scheduling problem when lower-priority process holds a lock needed by higher-priority process
Solved via priority-inheritance protocol
Operating System Concepts – 10th Edition 6.52 Silberschatz, Galvin and Gagne ©2018
Priority Inheritance Protocol
Consider the scenario with three processes P1, P2, and P3. P1 has the highest priority, P2 the next highest, and P3 the lowest. Assume a resouce P3 is assigned a resource R that P1 wants. Thus, P1 must wait for P3 to finish using the resource. However, P2 becomes runnable and preempts P3. What has happened is that P2 – a process with a lower priority than P1 – has indirectly prevented P3 from gaining access to the resource.
To prevent this from occurring, a priority inheritance protocol is used. This simply allows the priority of the highest thread waiting to access a shared resource to be assigned to the thread currently using the resource. Thus, the current owner of the resource is assigned the priority of the highest priority thread wishing to acquire the resource.
Operating System Concepts – 10th Edition 6.53 Silberschatz, Galvin and Gagne ©2018
End of Chapter 6
Operating System Concepts – 10th Edition Silberschatz, Galvin and Gagne ©2018