1.6.1.1 Optionally Implemented POSIX.1 Routines

• Reported by the POSIX.1 _POSIX_THREAD_PRIO_PROTECT macro:

  pthread_mutex_getprioceiling()
  pthread_mutex_setprioceiling()
  pthread_mutexattr_getprioceiling()
  pthread_mutexattr_setprioceiling()

  
  
• Reported by the POSIX.1 _POSIX_THREAD_PRIO_PROTECT and
  _POSIX_THREAD_PRIO_INHERIT macros:

  pthread_mutexattr_getprotocol()
  pthread_mutexattr_setprotocol()

  
  
• (Not supported for OpenVMS systems) Reported by the POSIX.1 _POSIX_THREAD_PROCESS_SHARED macro:

  pthread_condattr_getpshared()
  pthread_condattr_setpshared()
  pthread_mutexattr_getpshared()
  pthread_mutexattr_setpshared()
  pthread_rwlockattr_getpshared()
  pthread_rwlockattr_setpshared()

The POSIX.1 standard directs the Threads Library to provide the macros named _POSIX_THREAD_PROCESS_SHARED , _POSIX_THREAD_PRIO_PROTECT , and _POSIX_THREAD_PRIO_INHERIT to report whether optionally implemented routines are present.

The Threads Library does provide the following macros specified in the POSIX.1 standard:

_POSIX_THREADS : threads are supported
_POSIX_THREAD_SAFE_FUNCTIONS : thread-safe libraries are supported
_POSIX_THREAD_ATTR_STACKSIZE : can specify stack size
_POSIX_THREAD_ATTR_STACKADDR : can specify stack address
_POSIX_THREAD_PRIORITY_SCHEDULING : real-time scheduling control is supported
_POSIX_THREAD_PROCESS_SHARED : cross-process synchronization is supported (Tru64 UNIX only)

The HP proprietary tis interface offers a set of thread-independent services. Use these routines to build software that performs processing that requires synchronization, but without requiring the use of pthreads. That is, use tis routines to build thread-safe code libraries whose routines can be called from either a single-threaded or a multithreaded environment.

In the absence of threads, tis routines impose minimal overhead on the calling program. For instance, tis routines avoid the use of interlocked instructions and memory barriers.

When threads are present, tis routines provide full support for synchronization. Note that there are no tis routines for creating threads or thread objects, because these routines would have no meaning if called from a single-threaded environment.

The tis routines can be classified into these functional categories:

• General routines
• Thread cancelation routines
• Thread-specific data key routines
• Mutex routines
• Condition variable routines
• Read-write lock routines

Table 1-2 summarizes these groups of tis routines.

Table 1-2 tis Routines Summary

Routine	Description
General Routines
tis_once()	Calls an initialization routine to be executed only once
tis_self()	Obtains the identifier of the calling thread
tis_yield()	Notifies the scheduler that the calling thread is willing to release its processor to other threads of the same or higher priority
Thread Cancelation Routines
tis_setcancelstate()	Sets the calling thread's cancelability state to enable or disable the delivery of cancelation requests
tis_testcancel()	Requests delivery of any pending cancelation request to the calling thread
Thread-Specific Data Key Routines
tis_getspecific()	Obtains the thread-specific data associated with the specified key for the calling thread
tis_key_create()	Generates a unique thread-specific data key
tis_key_delete()	Deletes a thread-specific data key
tis_setspecific()	Changes the thread-specific data value associated with the specified key for the calling thread
Mutex Routines
tis_lock_global()	Locks the global mutex
tis_mutex_destroy()	Destroys the specified mutex object
tis_mutex_init()	Initializes a mutex object
tis_mutex_lock()	Locks the specified mutex, if unlocked
tis_mutex_trylock()	Tries to lock the specified mutex
tis_mutex_unlock()	Unlocks the specified mutex when locked by the calling thread
tis_unlock_global()	Unlocks the global mutex
Condition Variable Routines
tis_cond_broadcast()	Wakes all threads currently waiting on the specified condition variable
tis_cond_destroy()	Destroys the specified condition variable object
tis_cond_init()	Initializes a condition variable object
tis_cond_signal()	Wakes at least one thread that is waiting on the specified condition variable
tis_cond_timedwait()	Causes a thread to wait a specified period of time for a condition variable to be signaled or broadcast
tis_cond_wait()	Causes the calling thread to wait for the specified condition variable to be signaled or broadcast
tis_get_expiration()	Calculates a timeout for a timed condition variable wait
OpenVMS I/O Completion Routines
tis_io_complete()	Completion AST service routine
tis_sync()	Thread-synchronous replacement for $SYNC system service
Read-Write Lock Routines
tis_read_lock()	Acquires the specified read-write lock for read access
tis_read_trylock()	Acquires the specified read-write lock for read access; returns immediately if already locked
tis_read_unlock()	Unlocks the specified read-write lock already acquired for read access by the calling thread
tis_rwlock_destroy()	Destroys the specified read-write lock object
tis_rwlock_init()	Initializes the specified read-write lock object
tis_write_lock()	Acquires the specified read-write lock for write access
tis_write_trylock()	Acquires the specified read-write lock for write access; returns immediately if already locked
tis_write_unlock()	Unlocks the specified read-write lock already acquired for write access by the calling thread

1.6.3 Undocumented and Obsolete Interfaces

Previous versions of the Threads Library offered interfaces that under this version are no longer documented.

1.6.3.1 The cma Interface

This version of the Threads Library supports the HP proprietary CMA (or cma) interface. The cma interface reports errors by raising exceptions. This interface is layered on top of the pthread interface. This interface is usually available only on HP or Compaq platforms.

HP will continue to support existing applications that were developed using the cma interface. Binary compatibility will be supported indefinitely. Nonetheless, HP recommends that, as soon as possible, you migrate any cma code in your existing applications to the latest pthread interface, to take advantage of its standard features, portability, and future enhancements.

Routines of the cma interface are not documented in this guide.

1.6.3.2 The d4 (DCEthread) Interfaces

For backward compatibility only, this version of the Threads Library retains full binary support for the d4 interfaces. These interfaces are implementations of the IEEE POSIX 1003.4a/Draft 4 document, and are also known as "DCE threads".

These interfaces include both a "standard" interface that reports errors by setting errno and returning a value of -1, and an "exception-returning" interface that, like the cma interface, reports errors by raising exceptions.

The d4 interfaces will not be provided in a future release of the Threads Library. HP recommends that you migrate any d4 code in your existing applications to the latest pthread interface, to take advantage of its standard features, portability, and future enhancements.

Routines of the d4 interfaces are not documented in this guide. In this guide see Appendix D for information to help you migrate your d4-based programs and applications to the latest pthread interface.

This chapter describes operations that act upon the objects supported in the pthread interface.

2.1 Threads and Synchronization Objects

A multithreaded program typically manipulates these objects:

• A thread object describes a thread, which refers to a distinct flow of control within a process. After a thread object is created, the Threads Library uses it to maintain information about the thread's state and its associated attributes.
• A mutex serves as a lock for data that is shared among the program's threads. To access data that is guarded by a mutex, a thread must acquire the mutex, access the data, and then release the mutex. Each instance of acquiring a mutex is called a lock acquisition. While a mutex is locked, if other threads attempt to acquire that mutex, those threads must wait for the mutex to be released.
• For data that is shared among a program's threads but is more frequently read than written, use a read-write lock to guard access to the data. Unlike a mutex, more than one thread can acquire the same read-write lock for read access at the same time.
• When associated with a shared data object and its mutex, a condition variable provides a mechanism that allows a thread to wait until a piece of shared data protected by a mutex is placed into a particular state.

When your program creates a thread, mutex, read-write lock or condition variable, it can accept the default attributes for that object or specify an existing attributes object (previously created by your program) that contains particular attribute values. You can also change some of the attributes of a thread after it has begun execution---for example, you can change the thread's priority. However, other attributes, such as stack size, are fixed at execution.

To initialize an attributes object, you can use one of the following routines, depending on the type of object to which the attributes apply:

• pthread_attr_init() for thread attributes
• pthread_mutexattr_init() for mutex attributes
• pthread_rwlockattr_init() for read-write lock attributes
• pthread_condattr_init() for condition variable attributes

These routines initialize an attributes object with default values for the individual attributes. To modify any attribute values in an attributes object, use one of the " attr_set " routines, such as pthread_attr_setinheritsched() , described in later sections.

Initializing an attributes object (or changing the values in an attributes object) does not affect the attributes of existing threads, mutexes, read-write locks and condition variables.

To destroy an attributes object, use one of the following routines:

• pthread_attr_destroy() for thread attributes objects
• pthread_condattr_destroy() for condition variable attributes objects
• pthread_mutexattr_destroy() for mutex attributes objects
• pthread_rwlockattr_destroy() for read-write lock attributes objects

Deleting an attributes object does not affect the attributes of objects previously created with that attributes object.

2.3 Thread Operations

The following sections describe these operations on threads:

• Creating a thread
• Setting the attributes for a new thread
• Terminating a thread
• Detaching and destroying a thread
• Joining with another thread
• Controlling how a thread is scheduled
• Canceling a thread

Your program creates a thread using the pthread_create() routine. This routine creates a thread based on the settings of the thread attributes object if specified, which your program must have previously initialized. If called without a specified thread attributes object, pthread_create creates a new thread that has the default attributes.

The Threads Library creates a thread in the ready state and prepares the thread to begin executing its start routine, the function passed to the pthread_create() routine. Depending on the presence of other threads and their scheduling attributes, the new thread might preempt its creator (that is, it might start before the call to pthread_create() returns). The caller of pthread_create() can synchronize with the new thread using any mutually agreed upon mechanism or await its termination using pthread_join() .

The Threads Library assigns each new thread a thread identifier, which is written into the address specified as the pthread_create() routine's thread argument. The new thread's identifier is written before the new thread executes.

You can create a thread that is detached. To do so, create a thread using a thread attributes object whose detachstate attribute has been set, using the pthread_attr_setdetachstate() routine, to PTHREAD_CREATE_DETACHED . This is useful for creating a thread that your program knows will not be joined by any other thread. That is, when such a thread terminates, the thread and its thread object are automatically destroyed.

For more detailed information about thread creation, see the reference description of the pthread_create() routine in Part 2.

2.3.2 Setting the Attributes of a New Thread

When creating a thread, your program can optionally specify the attributes of the new thread using a thread attributes object. To do so, your program must:

1. Allocate a thread attributes object and then initialize it by calling the pthread_attr_init() routine. (Normally, you will initialize an extern or local variable of the appropriate type.)
2. Set values for the individual attributes of the thread attributes object. (The POSIX standard provides a separate routine for setting each attribute in the thread attributes object.)
3. When ready to create the new thread, pass the address of the thread attributes object as an argument to the pthread_create() routine.

After your program creates a thread attributes object, it can be reused for each new thread that the program creates. For the details about creating and deleting a thread attributes object, see the descriptions in Part 2 of the pthread_attr_init() and pthread_attr_destroy() routines.

Using the thread attributes object, your program can specify these attributes of a new thread:

• Scheduling inheritance
• Scheduling policy
• Scheduling parameters
• Stack size
• Stack location
• Stack guard size
• Contention scope

By default, a new thread is created with the scheduling attributes (policy, parameters and contention scope) of its creator. If an attributes object is specified, the scheduling attribute values are ignored. When you want to create a thread with different scheduling attributes, you must set the attribute values, and also set the value of the inheritsched attribute to PTHREAD_EXPLICIT_SCHED . You do this by calling the pthread_attr_setinheritsched() routine. The default value is PTHREAD_INHERIT_SCHED .

2.3.2.2 Setting the Scheduling Policy Attribute

The scheduling policy attribute describes how new threads are scheduled for execution relative to the other threads in the process.

A thread has one of the following scheduling policies:

• SCHED_FIFO (first-in/first-out or FIFO)---The highest-priority thread runs until it blocks. If there is more than one thread with the same priority and that priority is the highest among other threads, the first thread to begin running continues until it blocks. If a thread with this policy becomes ready, and it has a higher priority than the currently running thread, then the current thread is preempted and the higher priority thread immediately begins running.
• SCHED_RR (round-robin or RR)---The highest-priority thread runs until it blocks; however, threads of equal priority are time sliced. If a thread with this policy becomes ready, and it has a higher priority than the currently running thread, then the current thread is preempted and the higher priority thread immediately begins running.
  On a multiprocessor, threads of varying policy and priority may run simultaneously. A high priority thread is not guaranteed exclusive use of a multiprocessor system. You must use synchronization, not scheduling attributes, to ensure exclusive access.
• SCHED_OTHER (Foreground or "throughput"; also known as SCHED_FG_NP )---This is the default scheduling policy. All threads are time sliced, and no thread with this policy will completely starve any other thread with this policy, regardless of any thread's priority. (Time slicing is a mechanism that ensures that every thread is allowed time to execute by preempting running threads at fixed intervals.) However, higher-priority threads tend to receive more execution time than lower-priority threads, if the threads behave similarly.
  Threads with this scheduling policy can be denied execution time by first-in/first-out (FIFO) or round-robin (RR) threads. Threads in this policy do not preempt other threads.

Section 2.3.6 describes and shows the effect of the scheduling policy on thread scheduling.