1PTHREAD_CREATE(3P) POSIX Programmer's Manual PTHREAD_CREATE(3P)
2
6 This manual page is part of the POSIX Programmer's Manual. The Linux
7 implementation of this interface may differ (consult the corresponding
8 Linux manual page for details of Linux behavior), or the interface may
9 not be implemented on Linux.
10
12 pthread_create - thread creation
13
15 #include <pthread.h>
16 int pthread_create(pthread_t *restrict thread,
18 const pthread_attr_t *restrict attr,
19 void *(*start_routine)(void*), void *restrict arg);
20
23 The pthread_create() function shall create a new thread, with
24 attributes specified by attr, within a process. If attr is NULL, the
25 default attributes shall be used. If the attributes specified by attr
26 are modified later, the thread's attributes shall not be affected. Upon
27 successful completion, pthread_create() shall store the ID of the cre‐
28 ated thread in the location referenced by thread.
29 The thread is created executing start_routine with arg as its sole
31 argument. If the start_routine returns, the effect shall be as if there
32 was an implicit call to pthread_exit() using the return value of
33 start_routine as the exit status. Note that the thread in which main()
34 was originally invoked differs from this. When it returns from main(),
35 the effect shall be as if there was an implicit call to exit() using
36 the return value of main() as the exit status.
37 The signal state of the new thread shall be initialized as follows:
39 * The signal mask shall be inherited from the creating thread.
41 * The set of signals pending for the new thread shall be empty.
43 The alternate stack shall not be inherited.
45 The floating-point environment shall be inherited from the creating
47 thread.
48 If pthread_create() fails, no new thread is created and the contents of
50 the location referenced by thread are undefined.
51 If _POSIX_THREAD_CPUTIME is defined, the new thread shall have a CPU-
53 time clock accessible, and the initial value of this clock shall be set
54 to zero.
55
57 If successful, the pthread_create() function shall return zero; other‐
58 wise, an error number shall be returned to indicate the error.
59
61 The pthread_create() function shall fail if:
62 EAGAIN The system lacked the necessary resources to create another
64 thread, or the system-imposed limit on the total number of
65 threads in a process {PTHREAD_THREADS_MAX} would be exceeded.
66 EINVAL The value specified by attr is invalid.
68 EPERM The caller does not have appropriate permission to set the
70 required scheduling parameters or scheduling policy.
71 The pthread_create() function shall not return an error code of
74 [EINTR].
75 The following sections are informative.
77
79 None.
80
82 None.
83
85 A suggested alternative to pthread_create() would be to define two sep‐
86 arate operations: create and start. Some applications would find such
87 behavior more natural. Ada, in particular, separates the "creation" of
88 a task from its "activation".
89 Splitting the operation was rejected by the standard developers for
91 many reasons:
92 * The number of calls required to start a thread would increase from
94 one to two and thus place an additional burden on applications that
95 do not require the additional synchronization. The second call, how‐
96 ever, could be avoided by the additional complication of a start-up
97 state attribute.
98 * An extra state would be introduced: "created but not started". This
100 would require the standard to specify the behavior of the thread
101 operations when the target has not yet started executing.
102 * For those applications that require such behavior, it is possible to
104 simulate the two separate steps with the facilities that are cur‐
105 rently provided. The start_routine() can synchronize by waiting on a
106 condition variable that is signaled by the start operation.
107 An Ada implementor can choose to create the thread at either of two
109 points in the Ada program: when the task object is created, or when the
110 task is activated (generally at a "begin"). If the first approach is
111 adopted, the start_routine() needs to wait on a condition variable to
112 receive the order to begin "activation". The second approach requires
113 no such condition variable or extra synchronization. In either
114 approach, a separate Ada task control block would need to be created
115 when the task object is created to hold rendezvous queues, and so on.
116 An extension of the preceding model would be to allow the state of the
118 thread to be modified between the create and start. This would allow
119 the thread attributes object to be eliminated. This has been rejected
120 because:
121 * All state in the thread attributes object has to be able to be set
123 for the thread. This would require the definition of functions to
124 modify thread attributes. There would be no reduction in the number
125 of function calls required to set up the thread. In fact, for an
126 application that creates all threads using identical attributes, the
127 number of function calls required to set up the threads would be
128 dramatically increased. Use of a thread attributes object permits
129 the application to make one set of attribute setting function calls.
130 Otherwise, the set of attribute setting function calls needs to be
131 made for each thread creation.
132 * Depending on the implementation architecture, functions to set
134 thread state would require kernel calls, or for other implementation
135 reasons would not be able to be implemented as macros, thereby
136 increasing the cost of thread creation.
137 * The ability for applications to segregate threads by class would be
139 lost.
140 Another suggested alternative uses a model similar to that for process
142 creation, such as "thread fork". The fork semantics would provide more
143 flexibility and the "create" function can be implemented simply by
144 doing a thread fork followed immediately by a call to the desired
145 "start routine" for the thread. This alternative has these problems:
146 * For many implementations, the entire stack of the calling thread
148 would need to be duplicated, since in many architectures there is no
149 way to determine the size of the calling frame.
150 * Efficiency is reduced since at least some part of the stack has to
152 be copied, even though in most cases the thread never needs the
153 copied context, since it merely calls the desired start routine.
154
156 None.
157
159 fork(), pthread_exit(), pthread_join(), the Base Definitions volume of
160 IEEE Std 1003.1-2001, <pthread.h>
161
163 Portions of this text are reprinted and reproduced in electronic form
164 from IEEE Std 1003.1, 2003 Edition, Standard for Information Technology
165 -- Portable Operating System Interface (POSIX), The Open Group Base
166 Specifications Issue 6, Copyright (C) 2001-2003 by the Institute of
167 Electrical and Electronics Engineers, Inc and The Open Group. In the
168 event of any discrepancy between this version and the original IEEE and
169 The Open Group Standard, the original IEEE and The Open Group Standard
170 is the referee document. The original Standard can be obtained online
171 at http://www.opengroup.org/unix/online.html .
172IEEE/The Open Group 2003 PTHREAD_CREATE(3P)