| blob | ba9836c1a37d3ea0e9151e7a195fa679e49ae337 |
2 POSIX Threads Programming
4 Author: Blaise Barney, Lawrence Livermore National Laboratory UCRL-MI-133316
6 Table of Contents
8 1. Abstract
9 2. Pthreads Overview
10 1. What is a Thread?
11 2. What are Pthreads?
12 3. Why Pthreads?
13 4. Designing Threaded Programs
14 3. The Pthreads API
15 4. Compiling Threaded Programs
16 5. Thread Management
17 1. Creating and Terminating Threads
18 2. Passing Arguments to Threads
19 3. Joining and Detaching Threads
20 4. Stack Management
21 5. Miscellaneous Routines
22 6. Mutex Variables
23 1. Mutex Variables Overview
24 2. Creating and Destroying Mutexes
25 3. Locking and Unlocking Mutexes
26 7. Condition Variables
27 1. Condition Variables Overview
28 2. Creating and Destroying Condition Variables
29 3. Waiting and Signaling on Condition Variables
30 8. LLNL Specific Information and Recommendations
31 9. Topics Not Covered
32 10. Pthread Library Routines Reference
33 11. References and More Information
34 12. Exercise
38 ┌─────────────────────────────────────────────────────────────────────────────┐
39 │ Abstract │
40 └─────────────────────────────────────────────────────────────────────────────┘
43 In shared memory multiprocessor architectures, such as SMPs, threads can be
44 used to implement parallelism. Historically, hardware vendors have implemented
45 their own proprietary versions of threads, making portability a concern for
46 software developers. For UNIX systems, a standardized C language threads
47 programming interface has been specified by the IEEE POSIX 1003.1c standard.
48 Implementations that adhere to this standard are referred to as POSIX threads,
49 or Pthreads.
51 The tutorial begins with an introduction to concepts, motivations, and design
52 considerations for using Pthreads. Each of the three major classes of routines
53 in the Pthreads API are then covered: Thread Management, Mutex Variables, and
54 Condition Variables. Example codes are used throughout to demonstrate how to
55 use most of the Pthreads routines needed by a new Pthreads programmer. The
56 tutorial concludes with a discussion of LLNL specifics and how to mix MPI with
57 pthreads. A lab exercise, with numerous example codes (C Language) is also
58 included.
60 Level/Prerequisites: Ideal for those who are new to parallel programming with
61 threads. A basic understanding of parallel programming in C is assumed. For
62 those who are unfamiliar with Parallel Programming in general, the material
63 covered in EC3500: Introduction To Parallel Computing would be helpful.
67 ┌─────────────────────────────────────────────────────────────────────────────┐
68 │ Pthreads Overview │
69 └─────────────────────────────────────────────────────────────────────────────┘
71 What is a Thread?
73 • Technically, a thread is defined as an independent stream of instructions
74 that can be scheduled to run as such by the operating system. But what does
75 this mean?
77 • To the software developer, the concept of a "procedure" that runs
78 independently from its main program may best describe a thread.
80 • To go one step further, imagine a main program (a.out) that contains a
81 number of procedures. Then imagine all of these procedures being able to be
82 scheduled to run simultaneously and/or independently by the operating
83 system. That would describe a "multi-threaded" program.
85 • How is this accomplished?
87 • Before understanding a thread, one first needs to understand a UNIX
88 process. A process is created by the operating system, and requires a fair
89 amount of "overhead". Processes contain information about program resources
90 and program execution state, including:
91 □ Process ID, process group ID, user ID, and group ID
92 □ Environment
93 □ Working directory.
94 □ Program instructions
95 □ Registers
96 □ Stack
97 □ Heap
98 □ File descriptors
99 □ Signal actions
100 □ Shared libraries
101 □ Inter-process communication tools (such as message queues, pipes,
102 semaphores, or shared memory).
104 Unix Process Process-thread relationship
105 UNIX PROCESS THREADS WITHIN A UNIX PROCESS
107 • Threads use and exist within these process resources, yet are able to be
108 scheduled by the operating system and run as independent entities largely
109 because they duplicate only the bare essential resources that enable them
110 to exist as executable code.
112 • This independent flow of control is accomplished because a thread maintains
113 its own:
114 □ Stack pointer
115 □ Registers
116 □ Scheduling properties (such as policy or priority)
117 □ Set of pending and blocked signals
118 □ Thread specific data.
120 • So, in summary, in the UNIX environment a thread:
121 □ Exists within a process and uses the process resources
122 □ Has its own independent flow of control as long as its parent process
123 exists and the OS supports it
124 □ Duplicates only the essential resources it needs to be independently
125 schedulable
126 □ May share the process resources with other threads that act equally
127 independently (and dependently)
128 □ Dies if the parent process dies - or something similar
129 □ Is "lightweight" because most of the overhead has already been
130 accomplished through the creation of its process.
132 • Because threads within the same process share resources:
133 □ Changes made by one thread to shared system resources (such as closing
134 a file) will be seen by all other threads.
135 □ Two pointers having the same value point to the same data.
136 □ Reading and writing to the same memory locations is possible, and
137 therefore requires explicit synchronization by the programmer.
141 ┌─────────────────────────────────────────────────────────────────────────────┐
142 │ Pthreads Overview │
143 └─────────────────────────────────────────────────────────────────────────────┘
145 What are Pthreads?
147 • Historically, hardware vendors have implemented their own proprietary
148 versions of threads. These implementations differed substantially from each
149 other making it difficult for programmers to develop portable threaded
150 applications.
152 • In order to take full advantage of the capabilities provided by threads, a
153 standardized programming interface was required.
154 □ For UNIX systems, this interface has been specified by the IEEE POSIX
155 1003.1c standard (1995).
156 □ Implementations adhering to this standard are referred to as POSIX
157 threads, or Pthreads.
158 □ Most hardware vendors now offer Pthreads in addition to their
159 proprietary API's.
161 • The POSIX standard has continued to evolve and undergo revisions, including
162 the Pthreads specification.
164 • Some useful links:
165 □ standards.ieee.org/findstds/standard/1003.1-2008.html
166 □ www.opengroup.org/austin/papers/posix_faq.html
167 □ www.unix.org/version3/ieee_std.html
169 • Pthreads are defined as a set of C language programming types and procedure
170 calls, implemented with a pthread.h header/include file and a thread
171 library - though this library may be part of another library, such as libc,
172 in some implementations.
176 ┌─────────────────────────────────────────────────────────────────────────────┐
177 │ Pthreads Overview │
178 └─────────────────────────────────────────────────────────────────────────────┘
180 Why Pthreads?
182 • The primary motivation for using Pthreads is to realize potential program
183 performance gains.
185 • When compared to the cost of creating and managing a process, a thread can
186 be created with much less operating system overhead. Managing threads
187 requires fewer system resources than managing processes.
189 For example, the following table compares timing results for the fork()
190 subroutine and the pthread_create() subroutine. Timings reflect 50,000
191 process/thread creations, were performed with the time utility, and units
192 are in seconds, no optimization flags.
194 Note: don't expect the sytem and user times to add up to real time, because
195 these are SMP systems with multiple CPUs working on the problem at the same
196 time. At best, these are approximations run on local machines, past and
197 present.
199 ┌───────────────────────┬─────────────────────┬───────────────────┐
200 │ │ fork() │ pthread_create() │
201 │ Platform ├───────┬──────┬──────┼──────┬──────┬─────┤
202 │ │ real │ user │ sys │ real │ user │ sys │
203 ├───────────────────────┼───────┼──────┼──────┼──────┼──────┼─────┤
204 │ Intel 2.8 GHz Xeon │ 4.4 │ 0.4 │ 4.3 │ 0.7 │ 0.2 │ 0.5 │
205 │ 5660 (12cpus/node) │ │ │ │ │ │ │
206 ├───────────────────────┼───────┼──────┼──────┼──────┼──────┼─────┤
207 │ AMD 2.3 GHz Opteron │ 12.5 │ 1.0 │ 12.5 │ 1.2 │ 0.2 │ 1.3 │
208 │ (16cpus/node) │ │ │ │ │ │ │
209 ├───────────────────────┼───────┼──────┼──────┼──────┼──────┼─────┤
210 │ AMD 2.4 GHz Opteron │ 17.6 │ 2.2 │ 15.7 │ 1.4 │ 0.3 │ 1.3 │
211 │ (8cpus/node) │ │ │ │ │ │ │
212 ├───────────────────────┼───────┼──────┼──────┼──────┼──────┼─────┤
213 │ IBM 4.0 GHz POWER6 │ 9.5 │ 0.6 │ 8.8 │ 1.6 │ 0.1 │ 0.4 │
214 │ (8cpus/node) │ │ │ │ │ │ │
215 ├───────────────────────┼───────┼──────┼──────┼──────┼──────┼─────┤
216 │ IBM 1.9 GHz POWER5 │ 64.2 │ 30.7 │ 27.6 │ 1.7 │ 0.6 │ 1.1 │
217 │ p5-575 (8cpus/node) │ │ │ │ │ │ │
218 ├───────────────────────┼───────┼──────┼──────┼──────┼──────┼─────┤
219 │ IBM 1.5 GHz POWER4 │ 104.5 │ 48.6 │ 47.2 │ 2.1 │ 1.0 │ 1.5 │
220 │ (8cpus/node) │ │ │ │ │ │ │
221 ├───────────────────────┼───────┼──────┼──────┼──────┼──────┼─────┤
222 │ INTEL 2.4 GHz Xeon (2 │ 54.9 │ 1.5 │ 20.8 │ 1.6 │ 0.7 │ 0.9 │
223 │ cpus/node) │ │ │ │ │ │ │
224 ├───────────────────────┼───────┼──────┼──────┼──────┼──────┼─────┤
225 │ INTEL 1.4 GHz │ 54.5 │ 1.1 │ 22.2 │ 2.0 │ 1.2 │ 0.6 │
226 │ Itanium2 (4 cpus/ │ │ │ │ │ │ │
227 │ node) │ │ │ │ │ │ │
228 └───────────────────────┴───────┴──────┴──────┴──────┴──────┴─────┘
229 View source code fork_vs_thread.txt
231 • All threads within a process share the same address space. Inter-thread
232 communication is more efficient and in many cases, easier to use than
233 inter-process communication.
235 • Threaded applications offer potential performance gains and practical
236 advantages over non-threaded applications in several other ways:
237 □ Overlapping CPU work with I/O: For example, a program may have sections
238 where it is performing a long I/O operation. While one thread is
239 waiting for an I/O system call to complete, CPU intensive work can be
240 performed by other threads.
241 □ Priority/real-time scheduling: tasks which are more important can be
242 scheduled to supersede or interrupt lower priority tasks.
243 □ Asynchronous event handling: tasks which service events of
244 indeterminate frequency and duration can be interleaved. For example, a
245 web server can both transfer data from previous requests and manage the
246 arrival of new requests.
248 • The primary motivation for considering the use of Pthreads on an SMP
249 architecture is to achieve optimum performance. In particular, if an
250 application is using MPI for on-node communications, there is a potential
251 that performance could be greatly improved by using Pthreads for on-node
252 data transfer instead.
254 • For example:
255 □ MPI libraries usually implement on-node task communication via shared
256 memory, which involves at least one memory copy operation (process to
257 process).
258 □ For Pthreads there is no intermediate memory copy required because
259 threads share the same address space within a single process. There is
260 no data transfer, per se. It becomes more of a cache-to-CPU or
261 memory-to-CPU bandwidth (worst case) situation. These speeds are much
262 higher.
263 □ Some local comparisons are shown below:
265 ┌──────────────────────┬────────────────────────┬─────────────────────┐
266 │ │ MPI Shared Memory │ Pthreads Worst Case │
267 │ Platform │ Bandwidth │ Memory-to-CPU │
268 │ │ (GB/sec) │ Bandwidth │
269 │ │ │ (GB/sec) │
270 ├──────────────────────┼────────────────────────┼─────────────────────┤
271 │ Intel 2.8 GHz Xeon │ 5.6 │ 32 │
272 │ 5660 │ │ │
273 ├──────────────────────┼────────────────────────┼─────────────────────┤
274 │ AMD 2.3 GHz Opteron │ 1.8 │ 5.3 │
275 ├──────────────────────┼────────────────────────┼─────────────────────┤
276 │ AMD 2.4 GHz Opteron │ 1.2 │ 5.3 │
277 ├──────────────────────┼────────────────────────┼─────────────────────┤
278 │ IBM 1.9 GHz POWER5 │ 4.1 │ 16 │
279 │ p5-575 │ │ │
280 ├──────────────────────┼────────────────────────┼─────────────────────┤
281 │ IBM 1.5 GHz POWER4 │ 2.1 │ 4 │
282 ├──────────────────────┼────────────────────────┼─────────────────────┤
283 │ Intel 2.4 GHz Xeon │ 0.3 │ 4.3 │
284 ├──────────────────────┼────────────────────────┼─────────────────────┤
285 │ Intel 1.4 GHz │ 1.8 │ 6.4 │
286 │ Itanium 2 │ │ │
287 └──────────────────────┴────────────────────────┴─────────────────────┘
291 ┌─────────────────────────────────────────────────────────────────────────────┐
292 │ Pthreads Overview │
293 └─────────────────────────────────────────────────────────────────────────────┘
295 Designing Threaded Programs
297 [arrowBulle] Parallel Programming:
299 • On modern, multi-cpu machines, pthreads are ideally suited for parallel
300 programming, and whatever applies to parallel programming in general,
301 applies to parallel pthreads programs.
303 • There are many considerations for designing parallel programs, such as:
304 □ What type of parallel programming model to use?
305 □ Problem partitioning
306 □ Load balancing
307 □ Communications
308 □ Data dependencies
309 □ Synchronization and race conditions
310 □ Memory issues
311 □ I/O issues
312 □ Program complexity
313 □ Programmer effort/costs/time
314 □ ...
316 • Covering these topics is beyond the scope of this tutorial, however
317 interested readers can obtain a quick overview in the Introduction to
318 Parallel Computing tutorial.
320 • In general though, in order for a program to take advantage of Pthreads, it
321 must be able to be organized into discrete, independent tasks which can
322 execute concurrently. For example, if routine1 and routine2 can be
323 interchanged, interleaved and/or overlapped in real time, they are
324 candidates for threading.
326 [concurrent]
328 • Programs having the following characteristics may be well suited for
329 pthreads:
330 □ Work that can be executed, or data that can be operated on, by multiple
331 tasks simultaneously
332 □ Block for potentially long I/O waits
333 □ Use many CPU cycles in some places but not others
334 □ Must respond to asynchronous events
335 □ Some work is more important than other work (priority interrupts)
337 • Pthreads can also be used for serial applications, to emulate parallel
338 execution. A perfect example is the typical web browser, which for most
339 people, runs on a single cpu desktop/laptop machine. Many things can
340 "appear" to be happening at the same time.
342 • Several common models for threaded programs exist:
344 □ Manager/worker: a single thread, the manager assigns work to other
345 threads, the workers. Typically, the manager handles all input and
346 parcels out work to the other tasks. At least two forms of the manager/
347 worker model are common: static worker pool and dynamic worker pool.
349 □ Pipeline: a task is broken into a series of suboperations, each of
350 which is handled in series, but concurrently, by a different thread. An
351 automobile assembly line best describes this model.
353 □ Peer: similar to the manager/worker model, but after the main thread
354 creates other threads, it participates in the work.
356 [arrowBulle] Shared Memory Model:
358 • All threads have access to the same global, shared memory
360 • Threads also have their own private data
362 • Programmers are responsible for synchronizing access (protecting) globally
363 shared data.
364 Shared Memory Model
366 [arrowBulle] Thread-safeness:
368 • Thread-safeness: in a nutshell, refers an application's ability to execute
369 multiple threads simultaneously without "clobbering" shared data or
370 creating "race" conditions.
372 • For example, suppose that your application creates several threads, each of
373 which makes a call to the same library routine:
374 □ This library routine accesses/modifies a global structure or location
375 in memory.
376 □ As each thread calls this routine it is possible that they may try to
377 modify this global structure/memory location at the same time.
378 □ If the routine does not employ some sort of synchronization constructs
379 to prevent data corruption, then it is not thread-safe.
381 threadunsafe
383 • The implication to users of external library routines is that if you aren't
384 100% certain the routine is thread-safe, then you take your chances with
385 problems that could arise.
387 • Recommendation: Be careful if your application uses libraries or other
388 objects that don't explicitly guarantee thread-safeness. When in doubt,
389 assume that they are not thread-safe until proven otherwise. This can be
390 done by "serializing" the calls to the uncertain routine, etc.
394 ┌─────────────────────────────────────────────────────────────────────────────┐
395 │ The Pthreads API │
396 └─────────────────────────────────────────────────────────────────────────────┘
399 • The original Pthreads API was defined in the ANSI/IEEE POSIX 1003.1 - 1995
400 standard. The POSIX standard has continued to evolve and undergo revisions,
401 including the Pthreads specification.
403 • Copies of the standard can be purchased from IEEE or downloaded for free
404 from other sites online.
406 • The subroutines which comprise the Pthreads API can be informally grouped
407 into four major groups:
409 1. Thread management: Routines that work directly on threads - creating,
410 detaching, joining, etc. They also include functions to set/query
411 thread attributes (joinable, scheduling etc.)
413 2. Mutexes: Routines that deal with synchronization, called a "mutex",
414 which is an abbreviation for "mutual exclusion". Mutex functions
415 provide for creating, destroying, locking and unlocking mutexes. These
416 are supplemented by mutex attribute functions that set or modify
417 attributes associated with mutexes.
419 3. Condition variables: Routines that address communications between
420 threads that share a mutex. Based upon programmer specified conditions.
421 This group includes functions to create, destroy, wait and signal based
422 upon specified variable values. Functions to set/query condition
423 variable attributes are also included.
425 4. Synchronization: Routines that manage read/write locks and barriers.
427 • Naming conventions: All identifiers in the threads library begin with
428 pthread_. Some examples are shown below.
430 ┌────────────────────┬────────────────────────────────────────────┐
431 │ Routine Prefix │ Functional Group │
432 ├────────────────────┼────────────────────────────────────────────┤
433 │ pthread_ │ Threads themselves and miscellaneous │
434 │ │ subroutines │
435 ├────────────────────┼────────────────────────────────────────────┤
436 │ pthread_attr_ │ Thread attributes objects │
437 ├────────────────────┼────────────────────────────────────────────┤
438 │ pthread_mutex_ │ Mutexes │
439 ├────────────────────┼────────────────────────────────────────────┤
440 │ pthread_mutexattr_ │ Mutex attributes objects. │
441 ├────────────────────┼────────────────────────────────────────────┤
442 │ pthread_cond_ │ Condition variables │
443 ├────────────────────┼────────────────────────────────────────────┤
444 │ pthread_condattr_ │ Condition attributes objects │
445 ├────────────────────┼────────────────────────────────────────────┤
446 │ pthread_key_ │ Thread-specific data keys │
447 ├────────────────────┼────────────────────────────────────────────┤
448 │ pthread_rwlock_ │ Read/write locks │
449 ├────────────────────┼────────────────────────────────────────────┤
450 │ pthread_barrier_ │ Synchronization barriers │
451 └────────────────────┴────────────────────────────────────────────┘
453 • The concept of opaque objects pervades the design of the API. The basic
454 calls work to create or modify opaque objects - the opaque objects can be
455 modified by calls to attribute functions, which deal with opaque
456 attributes.
458 • The Pthreads API contains around 100 subroutines. This tutorial will focus
459 on a subset of these - specifically, those which are most likely to be
460 immediately useful to the beginning Pthreads programmer.
462 • For portability, the pthread.h header file should be included in each
463 source file using the Pthreads library.
465 • The current POSIX standard is defined only for the C language. Fortran
466 programmers can use wrappers around C function calls. Some Fortran
467 compilers (like IBM AIX Fortran) may provide a Fortram pthreads API.
469 • A number of excellent books about Pthreads are available. Several of these
470 are listed in the References section of this tutorial.
472 ┌─────────────────────────────────────────────────────────────────────────────┐
473 │ Compiling Threaded Programs │
474 └─────────────────────────────────────────────────────────────────────────────┘
478 • Several examples of compile commands used for pthreads codes are listed in
479 the table below.
481 ┌─────────────────────┬────────────────────┬──────────────────────┐
482 │ Compiler / Platform │ Compiler Command │ Description │
483 ├─────────────────────┼────────────────────┼──────────────────────┤
484 │ INTEL │ icc -pthread │ C │
485 │Linux ├────────────────────┼──────────────────────┤
486 │ │ icpc -pthread │ C++ │
487 ├─────────────────────┼────────────────────┼──────────────────────┤
488 │ PathScale │ pathcc -pthread │ C │
489 │Linux ├────────────────────┼──────────────────────┤
490 │ │ pathCC -pthread │ C++ │
491 ├─────────────────────┼────────────────────┼──────────────────────┤
492 │ PGI │ pgcc -lpthread │ C │
493 │Linux ├────────────────────┼──────────────────────┤
494 │ │ pgCC -lpthread │ C++ │
495 ├─────────────────────┼────────────────────┼──────────────────────┤
496 │ GNU │ gcc -pthread │ GNU C │
497 │Linux, BG/L, BG/P ├────────────────────┼──────────────────────┤
498 │ │ g++ -pthread │ GNU C++ │
499 ├─────────────────────┼────────────────────┼──────────────────────┤
500 │ │ bgxlc_r / bgcc_r │ C (ANSI / │
501 │ IBM │ │ non-ANSI) │
502 │BG/L and BG/P ├────────────────────┼──────────────────────┤
503 │ │ bgxlC_r, bgxlc++_r │ C++ │
504 │ ├────────────────────┼──────────────────────┤
505 └─────────────────────┴────────────────────┴──────────────────────┘
509 ┌─────────────────────────────────────────────────────────────────────────────┐
510 │ Thread Management │
511 └─────────────────────────────────────────────────────────────────────────────┘
513 Creating and Terminating Threads
515 [arrowBulle] Routines:
518 ┌─────────────────────────────────────────────────────────────────┐
519 │ pthread_create (thread,attr,start_routine,arg) │
520 │ │
521 │ pthread_exit (status) │
522 │ │
523 │ pthread_cancel (thread) │
524 │ │
525 │ pthread_attr_init (attr) │
526 │ │
527 │ pthread_attr_destroy (attr) │
528 └─────────────────────────────────────────────────────────────────┘
530 [arrowBulle] Creating Threads:
533 • Initially, your main() program comprises a single, default thread. All
534 other threads must be explicitly created by the programmer.
536 • pthread_create creates a new thread and makes it executable. This routine
537 can be called any number of times from anywhere within your code.
539 • pthread_create arguments:
540 □ thread: An opaque, unique identifier for the new thread returned by the
541 subroutine.
542 □ attr: An opaque attribute object that may be used to set thread
543 attributes. You can specify a thread attributes object, or NULL for the
544 default values.
545 □ start_routine: the C routine that the thread will execute once it is
546 created.
547 □ arg: A single argument that may be passed to start_routine. It must be
548 passed by reference as a pointer cast of type void. NULL may be used if
549 no argument is to be passed.
551 • The maximum number of threads that may be created by a process is
552 implementation dependent.
554 • Once created, threads are peers, and may create other threads. There is no
555 implied hierarchy or dependency between threads.
557 Peer Threads
559 [arrowBulle] Thread Attributes:
562 • By default, a thread is created with certain attributes. Some of these
563 attributes can be changed by the programmer via the thread attribute
564 object.
566 • pthread_attr_init and pthread_attr_destroy are used to initialize/destroy
567 the thread attribute object.
569 • Other routines are then used to query/set specific attributes in the thread
570 attribute object. Attributes include:
571 □ Detached or joinable state
572 □ Scheduling inheritance
573 □ Scheduling policy
574 □ Scheduling parameters
575 □ Scheduling contention scope
576 □ Stack size
577 □ Stack address
578 □ Stack guard (overflow) size
580 • Some of these attributes will be discussed later.
582 [arrowBulle] Thread Binding and Scheduling:
584 ● Question: After a thread has been created, how do you know a)when it will
585 be scheduled to run by the operating system, and b)which processor/core it
586 will run on?
587 [Answer]
589 • The Pthreads API provides several routines that may be used to specify how
590 threads are scheduled for execution. For example, threads can be scheduled
591 to run FIFO (first-in first-out), RR (round-robin) or OTHER (operating
592 system determines). It also provides the ability to set a thread's
593 scheduling priority value.
595 • These topics are not covered here, however a good overview of "how things
596 work" under Linux can be found in the sched_setscheduler man page.
598 • The Pthreads API does not provide routines for binding threads to specific
599 cpus/cores. However, local implementations may include this functionality -
600 such as providing the non-standard pthread_setaffinity_np routine. Note
601 that "_np" in the name stands for "non-portable".
603 • Also, the local operating system may provide a way to do this. For example,
604 Linux provides the sched_setaffinity routine.
606 [arrowBulle] Terminating Threads & pthread_exit():
609 • There are several ways in which a thread may be terminated:
611 □ The thread returns normally from its starting routine. It's work is
612 done.
614 □ The thread makes a call to the pthread_exit subroutine - whether its
615 work is done or not.
617 □ The thread is canceled by another thread via the pthread_cancel
618 routine.
620 □ The entire process is terminated due to making a call to either the
621 exec() or exit()
623 □ If main() finishes first, without calling pthread_exit explicitly
624 itself
626 • The pthread_exit() routine allows the programmer to specify an optional
627 termination status parameter. This optional parameter is typically returned
628 to threads "joining" the terminated thread (covered later).
630 • In subroutines that execute to completion normally, you can often dispense
631 with calling pthread_exit() - unless, of course, you want to pass the
632 optional status code back.
634 • Cleanup: the pthread_exit() routine does not close files; any files opened
635 inside the thread will remain open after the thread is terminated.
637 • Discussion on calling pthread_exit() from main():
638 □ There is a definite problem if main() finishes before the threads it
639 spawned if you don't call pthread_exit() explicitly. All of the threads
640 it created will terminate because main() is done and no longer exists
641 to support the threads.
642 □ By having main() explicitly call pthread_exit() as the last thing it
643 does, main() will block and be kept alive to support the threads it
644 created until they are done.
646 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
648 Example: Pthread Creation and Termination
650 • This simple example code creates 5 threads with the pthread_create()
651 routine. Each thread prints a "Hello World!" message, and then terminates
652 with a call to pthread_exit().
654 ┌──────────────────────────────────────────────────────────────────────────┐
655 │ ● Example Code - Pthread Creation and Termination │
656 │ │
657 │ #include <pthread.h> │
658 │ #include <stdio.h> │
659 │ #define NUM_THREADS 5 │
660 │ │
661 │ void *PrintHello(void *threadid) │
662 │ { │
663 │ long tid; │
664 │ tid = (long)threadid; │
665 │ printf("Hello World! It's me, thread #%ld!\n", tid); │
666 │ pthread_exit(NULL); │
667 │ } │
668 │ │
669 │ int main (int argc, char *argv[]) │
670 │ { │
671 │ pthread_t threads[NUM_THREADS]; │
672 │ int rc; │
673 │ long t; │
674 │ for(t=0; t<NUM_THREADS; t++){ │
675 │ printf("In main: creating thread %ld\n", t); │
676 │ rc = pthread_create(&threads[t], NULL, PrintHello, (void *)t); │
677 │ if (rc){ │
678 │ printf("ERROR; return code from pthread_create() is %d\n", rc); │
679 │ exit(-1); │
680 │ } │
681 │ } │
682 │ │
683 │ /* Last thing that main() should do */ │
684 │ pthread_exit(NULL); │
685 │ } │
686 │ │
687 │ View source code View sample output │
688 └──────────────────────────────────────────────────────────────────────────┘
692 ┌─────────────────────────────────────────────────────────────────────────────┐
693 │ Thread Management │
694 └─────────────────────────────────────────────────────────────────────────────┘
696 Passing Arguments to Threads
698 • The pthread_create() routine permits the programmer to pass one argument to
699 the thread start routine. For cases where multiple arguments must be
700 passed, this limitation is easily overcome by creating a structure which
701 contains all of the arguments, and then passing a pointer to that structure
702 in the pthread_create() routine.
704 • All arguments must be passed by reference and cast to (void *).
706 ● Question: How can you safely pass data to newly created threads, given
707 their non-deterministic start-up and scheduling?
708 [Answer]
710 ┌─────────────────────────────────────────────────────────────────────────────┐
711 │ ● Example 1 - Thread Argument Passing │
712 │ │
713 │ This code fragment demonstrates how to pass a simple integer to each │
714 │ thread. The calling thread uses a unique data structure for each │
715 │ thread, insuring that each thread's argument remains intact throughout │
716 │ the program. │
717 │ │
718 │ ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ │
719 │ │
720 │ long *taskids[NUM_THREADS]; │
721 │ │
722 │ for(t=0; t<NUM_THREADS; t++) │
723 │ { │
724 │ taskids[t] = (long *) malloc(sizeof(long)); │
725 │ *taskids[t] = t; │
726 │ printf("Creating thread %ld\n", t); │
727 │ rc = pthread_create(&threads[t], NULL, PrintHello, (void *) taskids[t]); │
728 │ ... │
729 │ } │
730 │ │
731 │ View source code View sample output │
732 └─────────────────────────────────────────────────────────────────────────────┘
734 ┌─────────────────────────────────────────────────────────────────┐
735 │ ● Example 2 - Thread Argument Passing │
736 │ │
737 │ This example shows how to setup/pass multiple arguments via │
738 │ a structure. Each thread receives a unique instance of the │
739 │ structure. │
740 │ │
741 │ ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ │
742 │ │
743 │ struct thread_data{ │
744 │ int thread_id; │
745 │ int sum; │
746 │ char *message; │
747 │ }; │
748 │ │
749 │ struct thread_data thread_data_array[NUM_THREADS]; │
750 │ │
751 │ void *PrintHello(void *threadarg) │
752 │ { │
753 │ struct thread_data *my_data; │
754 │ ... │
755 │ my_data = (struct thread_data *) threadarg; │
756 │ taskid = my_data->thread_id; │
757 │ sum = my_data->sum; │
758 │ hello_msg = my_data->message; │
759 │ ... │
760 │ } │
761 │ │
762 │ int main (int argc, char *argv[]) │
763 │ { │
764 │ ... │
765 │ thread_data_array[t].thread_id = t; │
766 │ thread_data_array[t].sum = sum; │
767 │ thread_data_array[t].message = messages[t]; │
768 │ rc = pthread_create(&threads[t], NULL, PrintHello, │
769 │ (void *) &thread_data_array[t]); │
770 │ ... │
771 │ } │
772 │ │
773 │ View source code View sample output │
774 └─────────────────────────────────────────────────────────────────┘
776 ┌─────────────────────────────────────────────────────────────────────┐
777 │ ● Example 3 - Thread Argument Passing (Incorrect) │
778 │ │
779 │ This example performs argument passing incorrectly. It passes │
780 │ the address of variable t, which is shared memory space and │
781 │ visible to all threads. As the loop iterates, the value of this │
782 │ memory location changes, possibly before the created threads │
783 │ can access it. │
784 │ │
785 │ ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ │
786 │ │
787 │ int rc; │
788 │ long t; │
789 │ │
790 │ for(t=0; t<NUM_THREADS; t++) │
791 │ { │
792 │ printf("Creating thread %ld\n", t); │
793 │ rc = pthread_create(&threads[t], NULL, PrintHello, (void *) &t); │
794 │ ... │
795 │ } │
796 │ │
797 │ View source code View sample output │
798 └─────────────────────────────────────────────────────────────────────┘
802 ┌─────────────────────────────────────────────────────────────────────────────┐
803 │ Thread Management │
804 └─────────────────────────────────────────────────────────────────────────────┘
806 Joining and Detaching Threads
808 [arrowBulle] Routines:
811 ┌─────────────────────────────────────────────────────────────────┐
812 │ pthread_join (threadid,status) │
813 │ │
814 │ pthread_detach (threadid) │
815 │ │
816 │ pthread_attr_setdetachstate (attr,detachstate) │
817 │ │
818 │ pthread_attr_getdetachstate (attr,detachstate) │
819 └─────────────────────────────────────────────────────────────────┘
821 [arrowBulle] Joining:
824 • "Joining" is one way to accomplish synchronization between threads. For
825 example:
827 Joining
829 • The pthread_join() subroutine blocks the calling thread until the specified
830 threadid thread terminates.
832 • The programmer is able to obtain the target thread's termination return
833 status if it was specified in the target thread's call to pthread_exit().
835 • A joining thread can match one pthread_join() call. It is a logical error
836 to attempt multiple joins on the same thread.
838 • Two other synchronization methods, mutexes and condition variables, will be
839 discussed later.
841 [arrowBulle] Joinable or Not?
844 • When a thread is created, one of its attributes defines whether it is
845 joinable or detached. Only threads that are created as joinable can be
846 joined. If a thread is created as detached, it can never be joined.
848 • The final draft of the POSIX standard specifies that threads should be
849 created as joinable.
851 • To explicitly create a thread as joinable or detached, the attr argument in
852 the pthread_create() routine is used. The typical 4 step process is:
853 1. Declare a pthread attribute variable of the pthread_attr_t data type
854 2. Initialize the attribute variable with pthread_attr_init()
855 3. Set the attribute detached status with pthread_attr_setdetachstate()
856 4. When done, free library resources used by the attribute with
857 pthread_attr_destroy()
859 [arrowBulle] Detaching:
862 • The pthread_detach() routine can be used to explicitly detach a thread even
863 though it was created as joinable.
865 • There is no converse routine.
867 [arrowBulle] Recommendations:
870 • If a thread requires joining, consider explicitly creating it as joinable.
871 This provides portability as not all implementations may create threads as
872 joinable by default.
874 • If you know in advance that a thread will never need to join with another
875 thread, consider creating it in a detached state. Some system resources may
876 be able to be freed.
878 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
880 Example: Pthread Joining
882 ┌───────────────────────────────────────────────────────────────────────┐
883 │ ● Example Code - Pthread Joining │
884 │ │
885 │ This example demonstrates how to "wait" for thread completions by │
886 │ using the Pthread join routine. Since some implementations of │
887 │ Pthreads may not create threads in a joinable state, the threads │
888 │ in this example are explicitly created in a joinable state so │
889 │ that they can be joined later. │
890 │ │
891 │ ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ │
892 │ │
893 │ #include <pthread.h> │
894 │ #include <stdio.h> │
895 │ #include <stdlib.h> │
896 │ #include <math.h> │
897 │ #define NUM_THREADS 4 │
898 │ │
899 │ void *BusyWork(void *t) │
900 │ { │
901 │ int i; │
902 │ long tid; │
903 │ double result=0.0; │
904 │ tid = (long)t; │
905 │ printf("Thread %ld starting...\n",tid); │
906 │ for (i=0; i<1000000; i++) │
907 │ { │
908 │ result = result + sin(i) * tan(i); │
909 │ } │
910 │ printf("Thread %ld done. Result = %e\n",tid, result); │
911 │ pthread_exit((void*) t); │
912 │ } │
913 │ │
914 │ int main (int argc, char *argv[]) │
915 │ { │
916 │ pthread_t thread[NUM_THREADS]; │
917 │ pthread_attr_t attr; │
918 │ int rc; │
919 │ long t; │
920 │ void *status; │
921 │ │
922 │ /* Initialize and set thread detached attribute */ │
923 │ pthread_attr_init(&attr); │
924 │ pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE); │
925 │ │
926 │ for(t=0; t<NUM_THREADS; t++) { │
927 │ printf("Main: creating thread %ld\n", t); │
928 │ rc = pthread_create(&thread[t], &attr, BusyWork, (void *)t); │
929 │ if (rc) { │
930 │ printf("ERROR; return code from pthread_create() │
931 │ is %d\n", rc); │
932 │ exit(-1); │
933 │ } │
934 │ } │
935 │ │
936 │ /* Free attribute and wait for the other threads */ │
937 │ pthread_attr_destroy(&attr); │
938 │ for(t=0; t<NUM_THREADS; t++) { │
939 │ rc = pthread_join(thread[t], &status); │
940 │ if (rc) { │
941 │ printf("ERROR; return code from pthread_join() │
942 │ is %d\n", rc); │
943 │ exit(-1); │
944 │ } │
945 │ printf("Main: completed join with thread %ld having a status │
946 │ of %ld\n",t,(long)status); │
947 │ } │
948 │ │
949 │ printf("Main: program completed. Exiting.\n"); │
950 │ pthread_exit(NULL); │
951 │ } │
952 │ │
953 │ View source code View sample output │
954 └───────────────────────────────────────────────────────────────────────┘
958 ┌─────────────────────────────────────────────────────────────────────────────┐
959 │ Thread Management │
960 └─────────────────────────────────────────────────────────────────────────────┘
962 Stack Management
964 [arrowBulle] Routines:
967 ┌─────────────────────────────────────────────────────────────────┐
968 │ pthread_attr_getstacksize (attr, stacksize) │
969 │ │
970 │ pthread_attr_setstacksize (attr, stacksize) │
971 │ │
972 │ pthread_attr_getstackaddr (attr, stackaddr) │
973 │ │
974 │ pthread_attr_setstackaddr (attr, stackaddr) │
975 └─────────────────────────────────────────────────────────────────┘
977 [arrowBulle] Preventing Stack Problems:
980 • The POSIX standard does not dictate the size of a thread's stack. This is
981 implementation dependent and varies.
983 • Exceeding the default stack limit is often very easy to do, with the usual
984 results: program termination and/or corrupted data.
986 • Safe and portable programs do not depend upon the default stack limit, but
987 instead, explicitly allocate enough stack for each thread by using the
988 pthread_attr_setstacksize routine.
990 • The pthread_attr_getstackaddr and pthread_attr_setstackaddr routines can be
991 used by applications in an environment where the stack for a thread must be
992 placed in some particular region of memory.
994 [arrowBulle] Some Practical Examples at LC:
997 • Default thread stack size varies greatly. The maximum size that can be
998 obtained also varies greatly, and may depend upon the number of threads per
999 node.
1001 • Both past and present architectures are shown to demonstrate the wide
1002 variation in default thread stack size.
1004 ┌───────────────┬───────┬─────────────┬──────────────┐
1005 │ Node │ #CPUs │ Memory (GB) │ Default Size │
1006 │ Architecture │ │ │ (bytes) │
1007 ├───────────────┼───────┼─────────────┼──────────────┤
1008 │ AMD Xeon 5660 │ 12 │ 24 │ 2,097,152 │
1009 ├───────────────┼───────┼─────────────┼──────────────┤
1010 │ AMD Opteron │ 8 │ 16 │ 2,097,152 │
1011 ├───────────────┼───────┼─────────────┼──────────────┤
1012 │ Intel IA64 │ 4 │ 8 │ 33,554,432 │
1013 ├───────────────┼───────┼─────────────┼──────────────┤
1014 │ Intel IA32 │ 2 │ 4 │ 2,097,152 │
1015 ├───────────────┼───────┼─────────────┼──────────────┤
1016 │ IBM Power5 │ 8 │ 32 │ 196,608 │
1017 ├───────────────┼───────┼─────────────┼──────────────┤
1018 │ IBM Power4 │ 8 │ 16 │ 196,608 │
1019 ├───────────────┼───────┼─────────────┼──────────────┤
1020 │ IBM Power3 │ 16 │ 16 │ 98,304 │
1021 └───────────────┴───────┴─────────────┴──────────────┘
1023 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
1025 Example: Stack Management
1027 ┌──────────────────────────────────────────────────────────────────────────┐
1028 │ ● Example Code - Stack Management │
1029 │ │
1030 │ This example demonstrates how to query and set a thread's stack │
1031 │ size. │
1032 │ │
1033 │ ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ │
1034 │ │
1035 │ #include <pthread.h> │
1036 │ #include <stdio.h> │
1037 │ #define NTHREADS 4 │
1038 │ #define N 1000 │
1039 │ #define MEGEXTRA 1000000 │
1040 │ │
1041 │ pthread_attr_t attr; │
1042 │ │
1043 │ void *dowork(void *threadid) │
1044 │ { │
1045 │ double A[N][N]; │
1046 │ int i,j; │
1047 │ long tid; │
1048 │ size_t mystacksize; │
1049 │ │
1050 │ tid = (long)threadid; │
1051 │ pthread_attr_getstacksize (&attr, &mystacksize); │
1052 │ printf("Thread %ld: stack size = %li bytes \n", tid, mystacksize); │
1053 │ for (i=0; i<N; i++) │
1054 │ for (j=0; j<N; j++) │
1055 │ A[i][j] = ((i*j)/3.452) + (N-i); │
1056 │ pthread_exit(NULL); │
1057 │ } │
1058 │ │
1059 │ int main(int argc, char *argv[]) │
1060 │ { │
1061 │ pthread_t threads[NTHREADS]; │
1062 │ size_t stacksize; │
1063 │ int rc; │
1064 │ long t; │
1065 │ │
1066 │ pthread_attr_init(&attr); │
1067 │ pthread_attr_getstacksize (&attr, &stacksize); │
1068 │ printf("Default stack size = %li\n", stacksize); │
1069 │ stacksize = sizeof(double)*N*N+MEGEXTRA; │
1070 │ printf("Amount of stack needed per thread = %li\n",stacksize); │
1071 │ pthread_attr_setstacksize (&attr, stacksize); │
1072 │ printf("Creating threads with stack size = %li bytes\n",stacksize); │
1073 │ for(t=0; t<NTHREADS; t++){ │
1074 │ rc = pthread_create(&threads[t], &attr, dowork, (void *)t); │
1075 │ if (rc){ │
1076 │ printf("ERROR; return code from pthread_create() is %d\n", rc); │
1077 │ exit(-1); │
1078 │ } │
1079 │ } │
1080 │ printf("Created %ld threads.\n", t); │
1081 │ pthread_exit(NULL); │
1082 │ } │
1083 └──────────────────────────────────────────────────────────────────────────┘
1087 ┌─────────────────────────────────────────────────────────────────────────────┐
1088 │ Thread Management │
1089 └─────────────────────────────────────────────────────────────────────────────┘
1091 Miscellaneous Routines
1093 ┌─────────────────────────────────────────────────────────────────┐
1094 │ pthread_self () │
1095 │ │
1096 │ pthread_equal (thread1,thread2) │
1097 └─────────────────────────────────────────────────────────────────┘
1099 • pthread_self returns the unique, system assigned thread ID of the calling
1100 thread.
1102 • pthread_equal compares two thread IDs. If the two IDs are different 0 is
1103 returned, otherwise a non-zero value is returned.
1105 • Note that for both of these routines, the thread identifier objects are
1106 opaque and can not be easily inspected. Because thread IDs are opaque
1107 objects, the C language equivalence operator == should not be used to
1108 compare two thread IDs against each other, or to compare a single thread ID
1109 against another value.
1111 ┌─────────────────────────────────────────────────────────────────┐
1112 │ pthread_once (once_control, init_routine) │
1113 └─────────────────────────────────────────────────────────────────┘
1115 • pthread_once executes the init_routine exactly once in a process. The first
1116 call to this routine by any thread in the process executes the given
1117 init_routine, without parameters. Any subsequent call will have no effect.
1119 • The init_routine routine is typically an initialization routine.
1121 • The once_control parameter is a synchronization control structure that
1122 requires initialization prior to calling pthread_once. For example:
1124 pthread_once_t once_control = PTHREAD_ONCE_INIT;
1128 ┌─────────────────────────────────────────────────────────────────────────────┐
1129 │ Mutex Variables │
1130 └─────────────────────────────────────────────────────────────────────────────┘
1132 Overview
1134 • Mutex is an abbreviation for "mutual exclusion". Mutex variables are one of
1135 the primary means of implementing thread synchronization and for protecting
1136 shared data when multiple writes occur.
1138 • A mutex variable acts like a "lock" protecting access to a shared data
1139 resource. The basic concept of a mutex as used in Pthreads is that only one
1140 thread can lock (or own) a mutex variable at any given time. Thus, even if
1141 several threads try to lock a mutex only one thread will be successful. No
1142 other thread can own that mutex until the owning thread unlocks that mutex.
1143 Threads must "take turns" accessing protected data.
1145 • Mutexes can be used to prevent "race" conditions. An example of a race
1146 condition involving a bank transaction is shown below:
1148 ┌───────────────────────────┬───────────────────────────┬─────────┐
1149 │ Thread 1 │ Thread 2 │ Balance │
1150 ├───────────────────────────┼───────────────────────────┼─────────┤
1151 │ Read balance: $1000 │ │ $1000 │
1152 ├───────────────────────────┼───────────────────────────┼─────────┤
1153 │ │ Read balance: $1000 │ $1000 │
1154 ├───────────────────────────┼───────────────────────────┼─────────┤
1155 │ │ Deposit $200 │ $1000 │
1156 ├───────────────────────────┼───────────────────────────┼─────────┤
1157 │ Deposit $200 │ │ $1000 │
1158 ├───────────────────────────┼───────────────────────────┼─────────┤
1159 │ Update balance $1000+$200 │ │ $1200 │
1160 ├───────────────────────────┼───────────────────────────┼─────────┤
1161 │ │ Update balance $1000+$200 │ $1200 │
1162 └───────────────────────────┴───────────────────────────┴─────────┘
1164 • In the above example, a mutex should be used to lock the "Balance" while a
1165 thread is using this shared data resource.
1167 • Very often the action performed by a thread owning a mutex is the updating
1168 of global variables. This is a safe way to ensure that when several threads
1169 update the same variable, the final value is the same as what it would be
1170 if only one thread performed the update. The variables being updated belong
1171 to a "critical section".
1173 • A typical sequence in the use of a mutex is as follows:
1174 □ Create and initialize a mutex variable
1175 □ Several threads attempt to lock the mutex
1176 □ Only one succeeds and that thread owns the mutex
1177 □ The owner thread performs some set of actions
1178 □ The owner unlocks the mutex
1179 □ Another thread acquires the mutex and repeats the process
1180 □ Finally the mutex is destroyed
1182 • When several threads compete for a mutex, the losers block at that call -
1183 an unblocking call is available with "trylock" instead of the "lock" call.
1185 • When protecting shared data, it is the programmer's responsibility to make
1186 sure every thread that needs to use a mutex does so. For example, if 4
1187 threads are updating the same data, but only one uses a mutex, the data can
1188 still be corrupted.
1192 ┌─────────────────────────────────────────────────────────────────────────────┐
1193 │ Mutex Variables │
1194 └─────────────────────────────────────────────────────────────────────────────┘
1196 Creating and Destroying Mutexes
1198 [arrowBulle] Routines:
1201 ┌─────────────────────────────────────────────────────────────────┐
1202 │ pthread_mutex_init (mutex,attr) │
1203 │ │
1204 │ pthread_mutex_destroy (mutex) │
1205 │ │
1206 │ pthread_mutexattr_init (attr) │
1207 │ │
1208 │ pthread_mutexattr_destroy (attr) │
1209 └─────────────────────────────────────────────────────────────────┘
1211 [arrowBulle] Usage:
1214 • Mutex variables must be declared with type pthread_mutex_t, and must be
1215 initialized before they can be used. There are two ways to initialize a
1216 mutex variable:
1218 1. Statically, when it is declared. For example:
1219 pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
1221 2. Dynamically, with the pthread_mutex_init() routine. This method permits
1222 setting mutex object attributes, attr.
1224 The mutex is initially unlocked.
1226 • The attr object is used to establish properties for the mutex object, and
1227 must be of type pthread_mutexattr_t if used (may be specified as NULL to
1228 accept defaults). The Pthreads standard defines three optional mutex
1229 attributes:
1230 □ Protocol: Specifies the protocol used to prevent priority inversions
1231 for a mutex.
1232 □ Prioceiling: Specifies the priority ceiling of a mutex.
1233 □ Process-shared: Specifies the process sharing of a mutex.
1235 Note that not all implementations may provide the three optional mutex
1236 attributes.
1238 • The pthread_mutexattr_init() and pthread_mutexattr_destroy() routines are
1239 used to create and destroy mutex attribute objects respectively.
1241 • pthread_mutex_destroy() should be used to free a mutex object which is no
1242 longer needed.
1246 ┌─────────────────────────────────────────────────────────────────────────────┐
1247 │ Mutex Variables │
1248 └─────────────────────────────────────────────────────────────────────────────┘
1250 Locking and Unlocking Mutexes
1252 [arrowBulle] Routines:
1255 ┌─────────────────────────────────────────────────────────────────┐
1256 │ pthread_mutex_lock (mutex) │
1257 │ │
1258 │ pthread_mutex_trylock (mutex) │
1259 │ │
1260 │ pthread_mutex_unlock (mutex) │
1261 └─────────────────────────────────────────────────────────────────┘
1263 [arrowBulle] Usage:
1266 • The pthread_mutex_lock() routine is used by a thread to acquire a lock on
1267 the specified mutex variable. If the mutex is already locked by another
1268 thread, this call will block the calling thread until the mutex is
1269 unlocked.
1271 • pthread_mutex_trylock() will attempt to lock a mutex. However, if the mutex
1272 is already locked, the routine will return immediately with a "busy" error
1273 code. This routine may be useful in preventing deadlock conditions, as in a
1274 priority-inversion situation.
1276 • pthread_mutex_unlock() will unlock a mutex if called by the owning thread.
1277 Calling this routine is required after a thread has completed its use of
1278 protected data if other threads are to acquire the mutex for their work
1279 with the protected data. An error will be returned if:
1280 □ If the mutex was already unlocked
1281 □ If the mutex is owned by another thread
1283 • There is nothing "magical" about mutexes...in fact they are akin to a
1284 "gentlemen's agreement" between participating threads. It is up to the code
1285 writer to insure that the necessary threads all make the the mutex lock and
1286 unlock calls correctly. The following scenario demonstrates a logical
1287 error:
1289 Thread 1 Thread 2 Thread 3
1290 Lock Lock
1291 A = 2 A = A+1 A = A*B
1292 Unlock Unlock
1294 ● Question: When more than one thread is waiting for a locked mutex, which
1295 thread will be granted the lock first after it is released?
1296 [Answer]
1298 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
1300 Example: Using Mutexes
1302 ┌───────────────────────────────────────────────────────────────────────────┐
1303 │ ● Example Code - Using Mutexes │
1304 │ │
1305 │ This example program illustrates the use of mutex variables in a │
1306 │ threads program that performs a dot product. The main data is made │
1307 │ available to all threads through a globally accessible structure. │
1308 │ Each thread works on a different part of the data. The main thread │
1309 │ waits for all the threads to complete their computations, and then it │
1310 │ prints the resulting sum. │
1311 │ │
1312 │ ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ │
1313 │ │
1314 │ #include <pthread.h> │
1315 │ #include <stdio.h> │
1316 │ #include <stdlib.h> │
1317 │ │
1318 │ /* │
1319 │ The following structure contains the necessary information │
1320 │ to allow the function "dotprod" to access its input data and │
1321 │ place its output into the structure. │
1322 │ */ │
1323 │ │
1324 │ typedef struct │
1325 │ { │
1326 │ double *a; │
1327 │ double *b; │
1328 │ double sum; │
1329 │ int veclen; │
1330 │ } DOTDATA; │
1331 │ │
1332 │ /* Define globally accessible variables and a mutex */ │
1333 │ │
1334 │ #define NUMTHRDS 4 │
1335 │ #define VECLEN 100 │
1336 │ DOTDATA dotstr; │
1337 │ pthread_t callThd[NUMTHRDS]; │
1338 │ pthread_mutex_t mutexsum; │
1339 │ │
1340 │ /* │
1341 │ The function dotprod is activated when the thread is created. │
1342 │ All input to this routine is obtained from a structure │
1343 │ of type DOTDATA and all output from this function is written into │
1344 │ this structure. The benefit of this approach is apparent for the │
1345 │ multi-threaded program: when a thread is created we pass a single │
1346 │ argument to the activated function - typically this argument │
1347 │ is a thread number. All the other information required by the │
1348 │ function is accessed from the globally accessible structure. │
1349 │ */ │
1350 │ │
1351 │ void *dotprod(void *arg) │
1352 │ { │
1353 │ │
1354 │ /* Define and use local variables for convenience */ │
1355 │ │
1356 │ int i, start, end, len ; │
1357 │ long offset; │
1358 │ double mysum, *x, *y; │
1359 │ offset = (long)arg; │
1360 │ │
1361 │ len = dotstr.veclen; │
1362 │ start = offset*len; │
1363 │ end = start + len; │
1364 │ x = dotstr.a; │
1365 │ y = dotstr.b; │
1366 │ │
1367 │ /* │
1368 │ Perform the dot product and assign result │
1369 │ to the appropriate variable in the structure. │
1370 │ */ │
1371 │ │
1372 │ mysum = 0; │
1373 │ for (i=start; i<end ; i++) │
1374 │ { │
1375 │ mysum += (x[i] * y[i]); │
1376 │ } │
1377 │ │
1378 │ /* │
1379 │ Lock a mutex prior to updating the value in the shared │
1380 │ structure, and unlock it upon updating. │
1381 │ */ │
1382 │ pthread_mutex_lock (&mutexsum); │
1383 │ dotstr.sum += mysum; │
1384 │ pthread_mutex_unlock (&mutexsum); │
1385 │ │
1386 │ pthread_exit((void*) 0); │
1387 │ } │
1388 │ │
1389 │ /* │
1390 │ The main program creates threads which do all the work and then │
1391 │ print out result upon completion. Before creating the threads, │
1392 │ the input data is created. Since all threads update a shared structure, │
1393 │ we need a mutex for mutual exclusion. The main thread needs to wait for │
1394 │ all threads to complete, it waits for each one of the threads. We specify │
1395 │ a thread attribute value that allow the main thread to join with the │
1396 │ threads it creates. Note also that we free up handles when they are │
1397 │ no longer needed. │
1398 │ */ │
1399 │ │
1400 │ int main (int argc, char *argv[]) │
1401 │ { │
1402 │ long i; │
1403 │ double *a, *b; │
1404 │ void *status; │
1405 │ pthread_attr_t attr; │
1406 │ │
1407 │ /* Assign storage and initialize values */ │
1408 │ a = (double*) malloc (NUMTHRDS*VECLEN*sizeof(double)); │
1409 │ b = (double*) malloc (NUMTHRDS*VECLEN*sizeof(double)); │
1410 │ │
1411 │ for (i=0; i<VECLEN*NUMTHRDS; i++) │
1412 │ { │
1413 │ a[i]=1.0; │
1414 │ b[i]=a[i]; │
1415 │ } │
1416 │ │
1417 │ dotstr.veclen = VECLEN; │
1418 │ dotstr.a = a; │
1419 │ dotstr.b = b; │
1420 │ dotstr.sum=0; │
1421 │ │
1422 │ pthread_mutex_init(&mutexsum, NULL); │
1423 │ │
1424 │ /* Create threads to perform the dotproduct */ │
1425 │ pthread_attr_init(&attr); │
1426 │ pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE); │
1427 │ │
1428 │ for(i=0; i<NUMTHRDS; i++) │
1429 │ { │
1430 │ /* │
1431 │ Each thread works on a different set of data. │
1432 │ The offset is specified by 'i'. The size of │
1433 │ the data for each thread is indicated by VECLEN. │
1434 │ */ │
1435 │ pthread_create(&callThd[i], &attr, dotprod, (void *)i); │
1436 │ } │
1437 │ │
1438 │ pthread_attr_destroy(&attr); │
1439 │ │
1440 │ /* Wait on the other threads */ │
1441 │ for(i=0; i<NUMTHRDS; i++) │
1442 │ { │
1443 │ pthread_join(callThd[i], &status); │
1444 │ } │
1445 │ │
1446 │ /* After joining, print out the results and cleanup */ │
1447 │ printf ("Sum = %f \n", dotstr.sum); │
1448 │ free (a); │
1449 │ free (b); │
1450 │ pthread_mutex_destroy(&mutexsum); │
1451 │ pthread_exit(NULL); │
1452 │ } │
1453 │ │
1454 │ View source code Serial version │
1455 │ View source code Pthreads version │
1456 └───────────────────────────────────────────────────────────────────────────┘
1460 ┌─────────────────────────────────────────────────────────────────────────────┐
1461 │ Condition Variables │
1462 └─────────────────────────────────────────────────────────────────────────────┘
1464 Overview
1466 • Condition variables provide yet another way for threads to synchronize.
1467 While mutexes implement synchronization by controlling thread access to
1468 data, condition variables allow threads to synchronize based upon the
1469 actual value of data.
1471 • Without condition variables, the programmer would need to have threads
1472 continually polling (possibly in a critical section), to check if the
1473 condition is met. This can be very resource consuming since the thread
1474 would be continuously busy in this activity. A condition variable is a way
1475 to achieve the same goal without polling.
1477 • A condition variable is always used in conjunction with a mutex lock.
1479 • A representative sequence for using condition variables is shown below.
1481 ┌─────────────────────────────────────────────────────────────────┐
1482 │ Main Thread │
1483 │ │
1484 │ • Declare and initialize global data/variables which require │
1485 │ synchronization (such as "count") │
1486 │ • Declare and initialize a condition variable object │
1487 │ • Declare and initialize an associated mutex │
1488 │ • Create threads A and B to do work │
1489 ├────────────────────────────────┬────────────────────────────────┤
1490 │ Thread A │ Thread B │
1491 │ │ │
1492 │ • Do work up to the point │ • Do work │
1493 │ where a certain condition │ • Lock associated mutex │
1494 │ must occur (such as │ • Change the value of the │
1495 │ "count" must reach a │ global variable that │
1496 │ specified value) │ Thread-A is waiting upon. │
1497 │ • Lock associated mutex and │ • Check value of the global │
1498 │ check value of a global │ Thread-A wait variable. If │
1499 │ variable │ it fulfills the desired │
1500 │ • Call pthread_cond_wait() │ condition, signal │
1501 │ to perform a blocking wait │ Thread-A. │
1502 │ for signal from Thread-B. │ • Unlock mutex. │
1503 │ Note that a call to │ • Continue │
1504 │ pthread_cond_wait() │ │
1505 │ automatically and │ │
1506 │ atomically unlocks the │ │
1507 │ associated mutex variable │ │
1508 │ so that it can be used by │ │
1509 │ Thread-B. │ │
1510 │ • When signalled, wake up. │ │
1511 │ Mutex is automatically and │ │
1512 │ atomically locked. │ │
1513 │ • Explicitly unlock mutex │ │
1514 │ • Continue │ │
1515 ├────────────────────────────────┴────────────────────────────────┤
1516 │ Main Thread │
1517 │ │
1518 │ Join / Continue │
1519 └─────────────────────────────────────────────────────────────────┘
1523 ┌─────────────────────────────────────────────────────────────────────────────┐
1524 │ Condition Variables │
1525 └─────────────────────────────────────────────────────────────────────────────┘
1527 Creating and Destroying Condition Variables
1529 [arrowBulle] Routines:
1532 ┌─────────────────────────────────────────────────────────────────┐
1533 │ pthread_cond_init (condition,attr) │
1534 │ │
1535 │ pthread_cond_destroy (condition) │
1536 │ │
1537 │ pthread_condattr_init (attr) │
1538 │ │
1539 │ pthread_condattr_destroy (attr) │
1540 └─────────────────────────────────────────────────────────────────┘
1542 [arrowBulle] Usage:
1545 • Condition variables must be declared with type pthread_cond_t, and must be
1546 initialized before they can be used. There are two ways to initialize a
1547 condition variable:
1549 1. Statically, when it is declared. For example:
1550 pthread_cond_t myconvar = PTHREAD_COND_INITIALIZER;
1552 2. Dynamically, with the pthread_cond_init() routine. The ID of the
1553 created condition variable is returned to the calling thread through
1554 the condition parameter. This method permits setting condition variable
1555 object attributes, attr.
1557 • The optional attr object is used to set condition variable attributes.
1558 There is only one attribute defined for condition variables:
1559 process-shared, which allows the condition variable to be seen by threads
1560 in other processes. The attribute object, if used, must be of type
1561 pthread_condattr_t (may be specified as NULL to accept defaults).
1563 Note that not all implementations may provide the process-shared attribute.
1565 • The pthread_condattr_init() and pthread_condattr_destroy() routines are
1566 used to create and destroy condition variable attribute objects.
1568 • pthread_cond_destroy() should be used to free a condition variable that is
1569 no longer needed.
1573 ┌─────────────────────────────────────────────────────────────────────────────┐
1574 │ Condition Variables │
1575 └─────────────────────────────────────────────────────────────────────────────┘
1577 Waiting and Signaling on Condition Variables
1579 [arrowBulle] Routines:
1582 ┌─────────────────────────────────────────────────────────────────┐
1583 │ pthread_cond_wait (condition,mutex) │
1584 │ │
1585 │ pthread_cond_signal (condition) │
1586 │ │
1587 │ pthread_cond_broadcast (condition) │
1588 └─────────────────────────────────────────────────────────────────┘
1590 [arrowBulle] Usage:
1593 • pthread_cond_wait() blocks the calling thread until the specified condition
1594 is signalled. This routine should be called while mutex is locked, and it
1595 will automatically release the mutex while it waits. After signal is
1596 received and thread is awakened, mutex will be automatically locked for use
1597 by the thread. The programmer is then responsible for unlocking mutex when
1598 the thread is finished with it.
1600 • The pthread_cond_signal() routine is used to signal (or wake up) another
1601 thread which is waiting on the condition variable. It should be called
1602 after mutex is locked, and must unlock mutex in order for pthread_cond_wait
1603 () routine to complete.
1605 • The pthread_cond_broadcast() routine should be used instead of
1606 pthread_cond_signal() if more than one thread is in a blocking wait state.
1608 • It is a logical error to call pthread_cond_signal() before calling
1609 pthread_cond_wait().
1611 [warning5] Proper locking and unlocking of the associated mutex variable is
1612 essential when using these routines. For example:
1614 • Failing to lock the mutex before calling pthread_cond_wait() may
1615 cause it NOT to block.
1617 • Failing to unlock the mutex after calling pthread_cond_signal()
1618 may not allow a matching pthread_cond_wait() routine to complete
1619 (it will remain blocked).
1622 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
1624 Example: Using Condition Variables
1626 ┌──────────────────────────────────────────────────────────────────────────────┐
1627 │ ● Example Code - Using Condition Variables │
1628 │ │
1629 │ This simple example code demonstrates the use of several Pthread │
1630 │ condition variable routines. The main routine creates three threads. Two │
1631 │ of the threads perform work and update a "count" variable. The third │
1632 │ thread waits until the count variable reaches a specified value. │
1633 │ │
1634 │ ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ │
1635 │ │
1636 │ #include <pthread.h> │
1637 │ #include <stdio.h> │
1638 │ #include <stdlib.h> │
1639 │ │
1640 │ #define NUM_THREADS 3 │
1641 │ #define TCOUNT 10 │
1642 │ #define COUNT_LIMIT 12 │
1643 │ │
1644 │ int count = 0; │
1645 │ int thread_ids[3] = {0,1,2}; │
1646 │ pthread_mutex_t count_mutex; │
1647 │ pthread_cond_t count_threshold_cv; │
1648 │ │
1649 │ void *inc_count(void *t) │
1650 │ { │
1651 │ int i; │
1652 │ long my_id = (long)t; │
1653 │ │
1654 │ for (i=0; i<TCOUNT; i++) { │
1655 │ pthread_mutex_lock(&count_mutex); │
1656 │ count++; │
1657 │ │
1658 │ /* │
1659 │ Check the value of count and signal waiting thread when condition is │
1660 │ reached. Note that this occurs while mutex is locked. │
1661 │ */ │
1662 │ if (count == COUNT_LIMIT) { │
1663 │ pthread_cond_signal(&count_threshold_cv); │
1664 │ printf("inc_count(): thread %ld, count = %d Threshold reached.\n", │
1665 │ my_id, count); │
1666 │ } │
1667 │ printf("inc_count(): thread %ld, count = %d, unlocking mutex\n", │
1668 │ my_id, count); │
1669 │ pthread_mutex_unlock(&count_mutex); │
1670 │ │
1671 │ /* Do some "work" so threads can alternate on mutex lock */ │
1672 │ sleep(1); │
1673 │ } │
1674 │ pthread_exit(NULL); │
1675 │ } │
1676 │ │
1677 │ void *watch_count(void *t) │
1678 │ { │
1679 │ long my_id = (long)t; │
1680 │ │
1681 │ printf("Starting watch_count(): thread %ld\n", my_id); │
1682 │ │
1683 │ /* │
1684 │ Lock mutex and wait for signal. Note that the pthread_cond_wait │
1685 │ routine will automatically and atomically unlock mutex while it waits. │
1686 │ Also, note that if COUNT_LIMIT is reached before this routine is run by │
1687 │ the waiting thread, the loop will be skipped to prevent pthread_cond_wait │
1688 │ from never returning. │
1689 │ */ │
1690 │ pthread_mutex_lock(&count_mutex); │
1691 │ while (count<COUNT_LIMIT) { │
1692 │ pthread_cond_wait(&count_threshold_cv, &count_mutex); │
1693 │ printf("watch_count(): thread %ld Condition signal received.\n", my_id); │
1694 │ count += 125; │
1695 │ printf("watch_count(): thread %ld count now = %d.\n", my_id, count); │
1696 │ } │
1697 │ pthread_mutex_unlock(&count_mutex); │
1698 │ pthread_exit(NULL); │
1699 │ } │
1700 │ │
1701 │ int main (int argc, char *argv[]) │
1702 │ { │
1703 │ int i, rc; │
1704 │ long t1=1, t2=2, t3=3; │
1705 │ pthread_t threads[3]; │
1706 │ pthread_attr_t attr; │
1707 │ │
1708 │ /* Initialize mutex and condition variable objects */ │
1709 │ pthread_mutex_init(&count_mutex, NULL); │
1710 │ pthread_cond_init (&count_threshold_cv, NULL); │
1711 │ │
1712 │ /* For portability, explicitly create threads in a joinable state */ │
1713 │ pthread_attr_init(&attr); │
1714 │ pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE); │
1715 │ pthread_create(&threads[0], &attr, watch_count, (void *)t1); │
1716 │ pthread_create(&threads[1], &attr, inc_count, (void *)t2); │
1717 │ pthread_create(&threads[2], &attr, inc_count, (void *)t3); │
1718 │ │
1719 │ /* Wait for all threads to complete */ │
1720 │ for (i=0; i<NUM_THREADS; i++) { │
1721 │ pthread_join(threads[i], NULL); │
1722 │ } │
1723 │ printf ("Main(): Waited on %d threads. Done.\n", NUM_THREADS); │
1724 │ │
1725 │ /* Clean up and exit */ │
1726 │ pthread_attr_destroy(&attr); │
1727 │ pthread_mutex_destroy(&count_mutex); │
1728 │ pthread_cond_destroy(&count_threshold_cv); │
1729 │ pthread_exit(NULL); │
1730 │ │
1731 │ } │
1732 │ │
1733 │ View source code View sample output │
1734 └──────────────────────────────────────────────────────────────────────────────┘
1738 ┌─────────────────────────────────────────────────────────────────────────────┐
1739 │ LLNL Specific Information and Recommendations │
1740 └─────────────────────────────────────────────────────────────────────────────┘
1743 This section describes details specific to Livermore Computing's systems.
1745 [arrowBulle] Implementations:
1748 • All LC production systems include a Pthreads implementation that follows
1749 draft 10 (final) of the POSIX standard. This is the preferred
1750 implementation.
1752 • Implementations differ in the maximum number of threads that a process may
1753 create. They also differ in the default amount of thread stack space.
1755 [arrowBulle] Compiling:
1758 • LC maintains a number of compilers, and usually several different versions
1759 of each - see the LC's Supported Compilers web page.
1761 • The compiler commands described in the Compiling Threaded Programs section
1762 apply to LC systems.
1764 [arrowBulle] Mixing MPI with Pthreads:
1767 • This is the primary motivation for using Pthreads at LC.
1769 • Design:
1770 □ Each MPI process typically creates and then manages N threads, where N
1771 makes the best use of the available CPUs/node.
1772 □ Finding the best value for N will vary with the platform and your
1773 application's characteristics.
1774 □ For IBM SP systems with two communication adapters per node, it may
1775 prove more efficient to use two (or more) MPI tasks per node.
1776 □ In general, there may be problems if multiple threads make MPI calls.
1777 The program may fail or behave unexpectedly. If MPI calls must be made
1778 from within a thread, they should be made only by one thread.
1780 • Compiling:
1781 □ Use the appropriate MPI compile command for the platform and language
1782 of choice
1783 □ Be sure to include the required Pthreads flag as shown in the Compiling
1784 Threaded Programs section.
1786 • An example code that uses both MPI and Pthreads is available below. The
1787 serial, threads-only, MPI-only and MPI-with-threads versions demonstrate
1788 one possible progression.
1789 □ Serial
1790 □ Pthreads only
1791 □ MPI only
1792 □ MPI with pthreads
1793 □ makefile (for IBM SP)
1795 [arrowBulle] Monitoring and Debugging Threads:
1798 • Debuggers vary in their ability to handle threads. The TotalView debugger
1799 is LC's recommended debugger for parallel programs, and is well suited for
1800 debugging threaded programs. See the TotalView Debugger tutorial for
1801 details.
1803 • The Linux ps command provides several flags for viewing thread information.
1804 Some examples are shown below. See the man page for details.
1806 ┌──────────────────────────────────────────────────────────────────┐
1807 │ % ps -Lf │
1808 │ UID PID PPID LWP C NLWP STIME TTY TIME CMD │
1809 │ blaise 22529 28240 22529 0 5 11:31 pts/53 00:00:00 a.out │
1810 │ blaise 22529 28240 22530 99 5 11:31 pts/53 00:01:24 a.out │
1811 │ blaise 22529 28240 22531 99 5 11:31 pts/53 00:01:24 a.out │
1812 │ blaise 22529 28240 22532 99 5 11:31 pts/53 00:01:24 a.out │
1813 │ blaise 22529 28240 22533 99 5 11:31 pts/53 00:01:24 a.out │
1814 │ │
1815 │ % ps -T │
1816 │ PID SPID TTY TIME CMD │
1817 │ 22529 22529 pts/53 00:00:00 a.out │
1818 │ 22529 22530 pts/53 00:01:49 a.out │
1819 │ 22529 22531 pts/53 00:01:49 a.out │
1820 │ 22529 22532 pts/53 00:01:49 a.out │
1821 │ 22529 22533 pts/53 00:01:49 a.out │
1822 │ │
1823 │ % ps -Lm │
1824 │ PID LWP TTY TIME CMD │
1825 │ 22529 - pts/53 00:18:56 a.out │
1826 │ - 22529 - 00:00:00 - │
1827 │ - 22530 - 00:04:44 - │
1828 │ - 22531 - 00:04:44 - │
1829 │ - 22532 - 00:04:44 - │
1830 │ - 22533 - 00:04:44 - │
1831 └──────────────────────────────────────────────────────────────────┘
1833 • LC's Linux clusters also provide the top command to monitor processes on a
1834 node. If used with the -H flag, the threads contained within a process will
1835 be visible. An example of the top -H command is shown below. The parent
1836 process is PID 18010 which spawned three threads, shown as PIDs 18012,
1837 18013 and 18014.
1839 top -H command
1843 ┌─────────────────────────────────────────────────────────────────────────────┐
1844 │ Topics Not Covered │
1845 └─────────────────────────────────────────────────────────────────────────────┘
1848 Several features of the Pthreads API are not covered in this tutorial. These
1849 are listed below. See the Pthread Library Routines Reference section for more
1850 information.
1852 • Thread Scheduling
1853 □ Implementations will differ on how threads are scheduled to run. In
1854 most cases, the default mechanism is adequate.
1855 □ The Pthreads API provides routines to explicitly set thread scheduling
1856 policies and priorities which may override the default mechanisms.
1857 □ The API does not require implementations to support these features.
1859 • Keys: Thread-Specific Data
1860 □ As threads call and return from different routines, the local data on a
1861 thread's stack comes and goes.
1862 □ To preserve stack data you can usually pass it as an argument from one
1863 routine to the next, or else store the data in a global variable
1864 associated with a thread.
1865 □ Pthreads provides another, possibly more convenient and versatile, way
1866 of accomplishing this through keys.
1868 • Mutex Protocol Attributes and Mutex Priority Management for the handling of
1869 "priority inversion" problems.
1871 • Condition Variable Sharing - across processes
1873 • Thread Cancellation
1875 • Threads and Signals
1877 • Synchronization constructs - barriers and locks
1881 ┌─────────────────────────────────────────────────────────────────────────────┐
1882 │ Pthread Library Routines Reference │
1883 └─────────────────────────────────────────────────────────────────────────────┘
1886 For convenience, an alphabetical list of Pthread routines, linked to their
1887 corresponding man page, is provided below.
1889 pthread_atfork
1890 pthread_attr_destroy
1891 pthread_attr_getdetachstate
1892 pthread_attr_getguardsize
1893 pthread_attr_getinheritsched
1894 pthread_attr_getschedparam
1895 pthread_attr_getschedpolicy
1896 pthread_attr_getscope
1897 pthread_attr_getstack
1898 pthread_attr_getstackaddr
1899 pthread_attr_getstacksize
1900 pthread_attr_init
1901 pthread_attr_setdetachstate
1902 pthread_attr_setguardsize
1903 pthread_attr_setinheritsched
1904 pthread_attr_setschedparam
1905 pthread_attr_setschedpolicy
1906 pthread_attr_setscope
1907 pthread_attr_setstack
1908 pthread_attr_setstackaddr
1909 pthread_attr_setstacksize
1910 pthread_barrier_destroy
1911 pthread_barrier_init
1912 pthread_barrier_wait
1913 pthread_barrierattr_destroy
1914 pthread_barrierattr_getpshared
1915 pthread_barrierattr_init
1916 pthread_barrierattr_setpshared
1917 pthread_cancel
1918 pthread_cleanup_pop
1919 pthread_cleanup_push
1920 pthread_cond_broadcast
1921 pthread_cond_destroy
1922 pthread_cond_init
1923 pthread_cond_signal
1924 pthread_cond_timedwait
1925 pthread_cond_wait
1926 pthread_condattr_destroy
1927 pthread_condattr_getclock
1928 pthread_condattr_getpshared
1929 pthread_condattr_init
1930 pthread_condattr_setclock
1931 pthread_condattr_setpshared
1932 pthread_create
1933 pthread_detach
1934 pthread_equal
1935 pthread_exit
1936 pthread_getconcurrency
1937 pthread_getcpuclockid
1938 pthread_getschedparam
1939 pthread_getspecific
1940 pthread_join
1941 pthread_key_create
1942 pthread_key_delete
1943 pthread_kill
1944 pthread_mutex_destroy
1945 pthread_mutex_getprioceiling
1946 pthread_mutex_init
1947 pthread_mutex_lock
1948 pthread_mutex_setprioceiling
1949 pthread_mutex_timedlock
1950 pthread_mutex_trylock
1951 pthread_mutex_unlock
1952 pthread_mutexattr_destroy
1953 pthread_mutexattr_getprioceiling
1954 pthread_mutexattr_getprotocol
1955 pthread_mutexattr_getpshared
1956 pthread_mutexattr_gettype
1957 pthread_mutexattr_init
1958 pthread_mutexattr_setprioceiling
1959 pthread_mutexattr_setprotocol
1960 pthread_mutexattr_setpshared
1961 pthread_mutexattr_settype
1962 pthread_once
1963 pthread_rwlock_destroy
1964 pthread_rwlock_init
1965 pthread_rwlock_rdlock
1966 pthread_rwlock_timedrdlock
1967 pthread_rwlock_timedwrlock
1968 pthread_rwlock_tryrdlock
1969 pthread_rwlock_trywrlock
1970 pthread_rwlock_unlock
1971 pthread_rwlock_wrlock
1972 pthread_rwlockattr_destroy
1973 pthread_rwlockattr_getpshared
1974 pthread_rwlockattr_init
1975 pthread_rwlockattr_setpshared
1976 pthread_self
1977 pthread_setcancelstate
1978 pthread_setcanceltype
1979 pthread_setconcurrency
1980 pthread_setschedparam
1981 pthread_setschedprio
1982 pthread_setspecific
1983 pthread_sigmask
1984 pthread_spin_destroy
1985 pthread_spin_init
1986 pthread_spin_lock
1987 pthread_spin_trylock
1988 pthread_spin_unlock
1989 pthread_testcancel
1991 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
1993 This completes the tutorial.
1995 Evaluation Please complete the online evaluation form - unless you are doing
1996 Form the exercise, in which case please complete it at the end of the
1997 exercise.
1999 Where would you like to go now?
2001 • Exercise
2002 • Agenda
2003 • Back to the top
2007 ┌─────────────────────────────────────────────────────────────────────────────┐
2008 │ References and More Information │
2009 └─────────────────────────────────────────────────────────────────────────────┘
2012 • Author: Blaise Barney, Livermore Computing.
2014 • POSIX Standard: www.unix.org/version3/ieee_std.html
2016 • "Pthreads Programming". B. Nichols et al. O'Reilly and Associates.
2018 • "Threads Primer". B. Lewis and D. Berg. Prentice Hall
2020 • "Programming With POSIX Threads". D. Butenhof. Addison Wesley
2021 www.awl.com/cseng/titles/0-201-63392-2
2023 • "Programming With Threads". S. Kleiman et al. Prentice Hall