path: root/manual/threads.texi

@node Threads
@c @node Threads, Dynamic Linker, Debugging Support, Top
@c %MENU% Functions, constants, and data types for working with threads
@chapter Threads
@cindex threads

This chapter describes functions used for managing threads.
@Theglibc{} provides two threading implementations: ISO C threads and
POSIX threads.

@menu
* ISO C Threads::	Threads based on the ISO C specification.
* POSIX Threads::	Threads based on the POSIX specification.
@end menu


@node ISO C Threads
@section ISO C Threads
@cindex ISO C threads
@cindex C threads
@pindex threads.h

This section describes the @glibcadj{} ISO C threads implementation.
To have a deeper understanding of this API, it is strongly recommended
to read ISO/IEC 9899:2011, section 7.26, in which ISO C threads were
originally specified.  All types and function prototypes are declared
in the header file @file{threads.h}.

@menu
* ISO C Threads Return Values:: Symbolic constants that represent a
				function's return value.
* ISO C Thread Management::	Support for basic threading.
* Call Once::			Single-call functions and macros.
* ISO C Mutexes::		A low-level mechanism for mutual exclusion.
* ISO C Condition Variables::	High-level objects for thread synchronization.
* ISO C Thread-local Storage::	Functions to support thread-local storage.
@end menu


@node ISO C Threads Return Values
@subsection Return Values

The ISO C thread specification provides the following enumeration
constants for return values from functions in the API:

@vtable @code
@item thrd_timedout
@standards{C11, threads.h}
A specified time was reached without acquiring the requested resource,
usually a mutex or condition variable.

@item thrd_success
@standards{C11, threads.h}
The requested operation succeeded.

@item thrd_busy
@standards{C11, threads.h}
The requested operation failed because a requested resource is already
in use.

@item thrd_error
@standards{C11, threads.h}
The requested operation failed.

@item thrd_nomem
@standards{C11, threads.h}
The requested operation failed because it was unable to allocate
enough memory.
@end vtable


@node ISO C Thread Management
@subsection Creation and Control
@cindex thread creation
@cindex thread control
@cindex thread management

@Theglibc{} implements a set of functions that allow the user to easily
create and use threads.  Additional functionality is provided to control
the behavior of threads.

The following data types are defined for managing threads:

@deftp {Data Type} thrd_t
@standards{C11, threads.h}
A unique object that identifies a thread.
@end deftp

@deftp {Data Type} thrd_start_t
@standards{C11, threads.h}
This data type is an @code{int (*) (void *)} typedef that is passed to
@code{thrd_create} when creating a new thread.  It should point to the
first function that thread will run.
@end deftp

The following functions are used for working with threads:

@deftypefun int thrd_create (thrd_t *@var{thr}, thrd_start_t @var{func}, void *@var{arg})
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
@code{thrd_create} creates a new thread that will execute the function
@var{func}.  The object pointed to by @var{arg} will be used as the
argument to @var{func}.  If successful, @var{thr} is set to the new
thread identifier.

This function may return @code{thrd_success}, @code{thrd_nomem}, or
@code{thrd_error}.
@end deftypefun

@deftypefun thrd_t thrd_current (void)
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
This function returns the identifier of the calling thread.
@end deftypefun

@deftypefun int thrd_equal (thrd_t @var{lhs}, thrd_t @var{rhs})
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
@code{thrd_equal} checks whether @var{lhs} and @var{rhs} refer to the
same thread.  If @var{lhs} and @var{rhs} are different threads, this
function returns @math{0}; otherwise, the return value is non-zero.
@end deftypefun

@deftypefun int thrd_sleep (const struct timespec *@var{time_point}, struct timespec *@var{remaining})
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
@code{thrd_sleep} blocks the execution of the current thread for at
least until the elapsed time pointed to by @var{time_point} has been
reached.  This function does not take an absolute time, but a duration
that the thread is required to be blocked.  @xref{Time Basics}, and
@ref{Time Types}.

The thread may wake early if a signal that is not ignored is received.
In such a case, if @code{remaining} is not NULL, the remaining time
duration is stored in the object pointed to by
@var{remaining}.

@code{thrd_sleep} returns @math{0} if it blocked for at least the
amount of time in @code{time_point}, @math{-1} if it was interrupted
by a signal, or a negative number on failure.
@end deftypefun

@deftypefun void thrd_yield (void)
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
@code{thrd_yield} provides a hint to the implementation to reschedule
the execution of the current thread, allowing other threads to run.
@end deftypefun

@deftypefun {_Noreturn void} thrd_exit (int @var{res})
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
@code{thrd_exit} terminates execution of the calling thread and sets
its result code to @var{res}.

If this function is called from a single-threaded process, the call is
equivalent to calling @code{exit} with @code{EXIT_SUCCESS}
(@pxref{Normal Termination}).  Also note that returning from a
function that started a thread is equivalent to calling
@code{thrd_exit}.
@end deftypefun

@deftypefun int thrd_detach (thrd_t @var{thr})
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
@code{thrd_detach} detaches the thread identified by @code{thr} from
the current control thread.  The resources held by the detached thread
will be freed automatically once the thread exits.  The parent thread
will never be notified by any @var{thr} signal.

Calling @code{thrd_detach} on a thread that was previously detached or
joined by another thread results in undefined behavior.

This function returns either @code{thrd_success} or @code{thrd_error}.
@end deftypefun

@deftypefun int thrd_join (thrd_t @var{thr}, int *@var{res})
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
@code{thrd_join} blocks the current thread until the thread identified
by @code{thr} finishes execution.  If @code{res} is not NULL, the
result code of the thread is put into the location pointed to by
@var{res}.  The termination of the thread @dfn{synchronizes-with} the
completion of this function, meaning both threads have arrived at a
common point in their execution.

Calling @code{thrd_join} on a thread that was previously detached or
joined by another thread results in undefined behavior.

This function returns either @code{thrd_success} or @code{thrd_error}.
@end deftypefun


@node Call Once
@subsection Call Once
@cindex call once
@cindex single-call functions

In order to guarantee single access to a function, @theglibc{}
implements a @dfn{call once function} to ensure a function is only
called once in the presence of multiple, potentially calling threads.

@deftp {Data Type} once_flag
@standards{C11, threads.h}
A complete object type capable of holding a flag used by @code{call_once}.
@end deftp

@defvr Macro ONCE_FLAG_INIT
@standards{C11, threads.h}
This value is used to initialize an object of type @code{once_flag}.
@end defvr

@deftypefun void call_once (once_flag *@var{flag}, void (*@var{func}) (void))
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
@code{call_once} calls function @var{func} exactly once, even if
invoked from several threads.  The completion of the function
@var{func} synchronizes-with all previous or subsequent calls to
@code{call_once} with the same @code{flag} variable.
@end deftypefun


@node ISO C Mutexes
@subsection Mutexes
@cindex mutex
@cindex mutual exclusion

To have better control of resources and how threads access them,
@theglibc{} implements a @dfn{mutex} object, which can help avoid race
conditions and other concurrency issues.  The term ``mutex'' refers to
mutual exclusion.

The fundamental data type for a mutex is the @code{mtx_t}:

@deftp {Data Type} mtx_t
@standards{C11, threads.h}
The @code{mtx_t} data type uniquely identifies a mutex object.
@end deftp

The ISO C standard defines several types of mutexes.  They are
represented by the following symbolic constants:

@vtable @code
@item mtx_plain
@standards{C11, threads.h}
A mutex that does not support timeout, or test and return.

@item mtx_recursive
@standards{C11, threads.h}
A mutex that supports recursive locking, which means that the owning
thread can lock it more than once without causing deadlock.

@item mtx_timed
@standards{C11, threads.h}
A mutex that supports timeout.
@end vtable

The following functions are used for working with mutexes:

@deftypefun int mtx_init (mtx_t *@var{mutex}, int @var{type})
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
@code{mtx_init} creates a new mutex object with type @var{type}.  The
object pointed to by @var{mutex} is set to the identifier of the newly
created mutex.

Not all combinations of mutex types are valid for the @code{type}
argument.  Valid uses of mutex types for the @code{type} argument are:

@table @code
@item mtx_plain
A non-recursive mutex that does not support timeout.

@item mtx_timed
A non-recursive mutex that does support timeout.

@item mtx_plain | mtx_recursive
A recursive mutex that does not support timeout.

@item mtx_timed | mtx_recursive
A recursive mutex that does support timeout.
@end table

This function returns either @code{thrd_success} or @code{thrd_error}.
@end deftypefun

@deftypefun int mtx_lock (mtx_t *@var{mutex})
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@asulock{}}@acunsafe{@aculock{}}}
@code{mtx_lock} blocks the current thread until the mutex pointed to
by @var{mutex} is locked.  The behavior is undefined if the current
thread has already locked the mutex and the mutex is not recursive.

Prior calls to @code{mtx_unlock} on the same mutex synchronize-with
this operation (if this operation succeeds), and all lock/unlock
operations on any given mutex form a single total order (similar to
the modification order of an atomic).

This function returns either @code{thrd_success} or @code{thrd_error}.
@end deftypefun

@deftypefun int mtx_timedlock (mtx_t *restrict @var{mutex}, const struct timespec *restrict @var{time_point})
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@asulock{}}@acunsafe{@aculock{}}}
@code{mtx_timedlock} blocks the current thread until the mutex pointed
to by @var{mutex} is locked or until the calendar time pointed to by
@var{time_point} has been reached.  Since this function takes an
absolute time, if a duration is required, the calendar time must be
calculated manually.  @xref{Time Basics}, and @ref{Calendar Time}.

If the current thread has already locked the mutex and the mutex is
not recursive, or if the mutex does not support timeout, the behavior
of this function is undefined.

Prior calls to @code{mtx_unlock} on the same mutex synchronize-with
this operation (if this operation succeeds), and all lock/unlock
operations on any given mutex form a single total order (similar to
the modification order of an atomic).

This function returns either @code{thrd_success} or @code{thrd_error}.
@end deftypefun

@deftypefun int mtx_trylock (mtx_t *@var{mutex})
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@asulock{}}@acunsafe{@aculock{}}}
@code{mtx_trylock} tries to lock the mutex pointed to by @var{mutex}
without blocking.  It returns immediately if the mutex is already
locked.

Prior calls to @code{mtx_unlock} on the same mutex synchronize-with
this operation (if this operation succeeds), and all lock/unlock
operations on any given mutex form a single total order (similar to
the modification order of an atomic).

This function returns @code{thrd_success} if the lock was obtained,
@code{thrd_busy} if the mutex is already locked, and @code{thrd_error}
on failure.
@end deftypefun

@deftypefun int mtx_unlock (mtx_t *@var{mutex})
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
@code{mtx_unlock} unlocks the mutex pointed to by @var{mutex}.  The
behavior is undefined if the mutex is not locked by the calling
thread.

This function synchronizes-with subsequent @code{mtx_lock},
@code{mtx_trylock}, and @code{mtx_timedlock} calls on the same mutex.
All lock/unlock operations on any given mutex form a single total
order (similar to the modification order of an atomic).

This function returns either @code{thrd_success} or @code{thrd_error}.
@end deftypefun

@deftypefun void mtx_destroy (mtx_t *@var{mutex})
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
@code{mtx_destroy} destroys the mutex pointed to by @var{mutex}.  If
there are any threads waiting on the mutex, the behavior is
undefined.
@end deftypefun


@node ISO C Condition Variables
@subsection Condition Variables
@cindex condvar
@cindex condition variables

Mutexes are not the only synchronization mechanisms available.  For
some more complex tasks, @theglibc{} also implements @dfn{condition
variables}, which allow the programmer to think at a higher level when
solving complex synchronization problems.  They are used to
synchronize threads waiting on a certain condition to happen.

The fundamental data type for condition variables is the @code{cnd_t}:

@deftp {Data Type} cnd_t
@standards{C11, threads.h}
The @code{cnd_t} uniquely identifies a condition variable object.
@end deftp

The following functions are used for working with condition variables:

@deftypefun int cnd_init (cnd_t *@var{cond})
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
@code{cnd_init} initializes a new condition variable, identified by
@var{cond}.

This function may return @code{thrd_success}, @code{thrd_nomem}, or
@code{thrd_error}.
@end deftypefun

@deftypefun int cnd_signal (cnd_t *@var{cond})
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
@code{cnd_signal} unblocks one thread that is currently waiting on the
condition variable pointed to by @var{cond}.  If a thread is
successfully unblocked, this function returns @code{thrd_success}.  If
no threads are blocked, this function does nothing and returns
@code{thrd_success}.  Otherwise, this function returns
@code{thrd_error}.
@end deftypefun

@deftypefun int cnd_broadcast (cnd_t *@var{cond})
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
@code{cnd_broadcast} unblocks all the threads that are currently
waiting on the condition variable pointed to by @var{cond}.  This
function returns @code{thrd_success} on success.  If no threads are
blocked, this function does nothing and returns
@code{thrd_success}. Otherwise, this function returns
@code{thrd_error}.
@end deftypefun

@deftypefun int cnd_wait (cnd_t *@var{cond}, mtx_t *@var{mutex})
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@asulock{}}@acunsafe{@aculock{}}}
@code{cnd_wait} atomically unlocks the mutex pointed to by @var{mutex}
and blocks on the condition variable pointed to by @var{cond} until
the thread is signaled by @code{cnd_signal} or @code{cnd_broadcast}.
The mutex is locked again before the function returns.

This function returns either @code{thrd_success} or @code{thrd_error}.
@end deftypefun

@deftypefun int cnd_timedwait (cnd_t *restrict @var{cond}, mtx_t *restrict @var{mutex}, const struct timespec *restrict @var{time_point})
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@asulock{}}@acunsafe{@aculock{}}}
@code{cnd_timedwait} atomically unlocks the mutex pointed to by
@var{mutex} and blocks on the condition variable pointed to by
@var{cond} until the thread is signaled by @code{cnd_signal} or
@code{cnd_broadcast}, or until the calendar time pointed to by
@var{time_point} has been reached.  The mutex is locked again before
the function returns.

As for @code{mtx_timedlock}, since this function takes an absolute
time, if a duration is required, the calendar time must be calculated
manually.  @xref{Time Basics}, and @ref{Calendar Time}.

This function may return @code{thrd_success}, @code{thrd_nomem}, or
@code{thrd_error}.
@end deftypefun

@deftypefun void cnd_destroy (cnd_t *@var{cond})
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
@code{cnd_destroy} destroys the condition variable pointed to by
@var{cond}.  If there are threads waiting on @var{cond}, the behavior
is undefined.
@end deftypefun


@node ISO C Thread-local Storage
@subsection Thread-local Storage
@cindex thread-local storage

@Theglibc{} implements functions to provide @dfn{thread-local
storage}, a mechanism by which variables can be defined to have unique
per-thread storage, lifetimes that match the thread lifetime, and
destructors that cleanup the unique per-thread storage.

Several data types and macros exist for working with thread-local
storage:

@deftp {Data Type} tss_t
@standards{C11, threads.h}
The @code{tss_t} data type identifies a thread-specific storage
object.  Even if shared, every thread will have its own instance of
the variable, with different values.
@end deftp

@deftp {Data Type} tss_dtor_t
@standards{C11, threads.h}
The @code{tss_dtor_t} is a function pointer of type @code{void (*)
(void *)}, to be used as a thread-specific storage destructor.  The
function will be called when the current thread calls @code{thrd_exit}
(but never when calling @code{tss_delete} or @code{exit}).
@end deftp

@defvr Macro thread_local
@standards{C11, threads.h}
@code{thread_local} is used to mark a variable with thread storage
duration, which means it is created when the thread starts and cleaned
up when the thread ends.

@emph{Note:} For C++, C++11 or later is required to use the
@code{thread_local} keyword.
@end defvr

@defvr Macro TSS_DTOR_ITERATIONS
@standards{C11, threads.h}
@code{TSS_DTOR_ITERATIONS} is an integer constant expression
representing the maximum number of iterations over all thread-local
destructors at the time of thread termination.  This value provides a
bounded limit to the destruction of thread-local storage; e.g.,
consider a destructor that creates more thread-local storage.
@end defvr

The following functions are used to manage thread-local storage:

@deftypefun int tss_create (tss_t *@var{tss_key}, tss_dtor_t @var{destructor})
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
@code{tss_create} creates a new thread-specific storage key and stores
it in the object pointed to by @var{tss_key}.  Although the same key
value may be used by different threads, the values bound to the key by
@code{tss_set} are maintained on a per-thread basis and persist for
the life of the calling thread.

If @code{destructor} is not NULL, a destructor function will be set,
and called when the thread finishes its execution by calling
@code{thrd_exit}.

This function returns @code{thrd_success} if @code{tss_key} is
successfully set to a unique value for the thread; otherwise,
@code{thrd_error} is returned and the value of @code{tss_key} is
undefined.
@end deftypefun

@deftypefun int tss_set (tss_t @var{tss_key}, void *@var{val})
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
@code{tss_set} sets the value of the thread-specific storage
identified by @var{tss_key} for the current thread to @var{val}.
Different threads may set different values to the same key.

This function returns either @code{thrd_success} or @code{thrd_error}.
@end deftypefun

@deftypefun {void *} tss_get (tss_t @var{tss_key})
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
@code{tss_get} returns the value identified by @var{tss_key} held in
thread-specific storage for the current thread.  Different threads may
get different values identified by the same key.  On failure,
@code{tss_get} returns zero.
@end deftypefun

@deftypefun void tss_delete (tss_t @var{tss_key})
@standards{C11, threads.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
@code{tss_delete} destroys the thread-specific storage identified by
@var{tss_key}.
@end deftypefun


@node POSIX Threads
@section POSIX Threads
@cindex pthreads

This section describes the @glibcadj{} POSIX Threads implementation.

@menu
* Thread-specific Data::          Support for creating and
				  managing thread-specific data
* Non-POSIX Extensions::          Additional functions to extend
				  POSIX Thread functionality
@end menu

@node Thread-specific Data
@subsection Thread-specific Data

The @glibcadj{} implements functions to allow users to create and manage
data specific to a thread.  Such data may be destroyed at thread exit,
if a destructor is provided.  The following functions are defined:

@deftypefun int pthread_key_create (pthread_key_t *@var{key}, void (*@var{destructor})(void*))
@standards{POSIX, pthread.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
@c pthread_key_create ok
@c  KEY_UNUSED ok
@c  KEY_USABLE ok
Create a thread-specific data key for the calling thread, referenced by
@var{key}.

Objects declared with the C++11 @code{thread_local} keyword are destroyed
before thread-specific data, so they should not be used in thread-specific
data destructors or even as members of the thread-specific data, since the
latter is passed as an argument to the destructor function.
@end deftypefun

@deftypefun int pthread_key_delete (pthread_key_t @var{key})
@standards{POSIX, pthread.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
@c pthread_key_delete ok
@c   This uses atomic compare and exchange to increment the seq number
@c   after testing it's not a KEY_UNUSED seq number.
@c  KEY_UNUSED dup ok
Destroy the thread-specific data @var{key} in the calling thread.  The
destructor for the thread-specific data is not called during destruction, nor
is it called during thread exit.
@end deftypefun

@deftypefun {void *} pthread_getspecific (pthread_key_t @var{key})
@standards{POSIX, pthread.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
@c pthread_getspecific ok
Return the thread-specific data associated with @var{key} in the calling
thread.
@end deftypefun

@deftypefun int pthread_setspecific (pthread_key_t @var{key}, const void *@var{value})
@standards{POSIX, pthread.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@asucorrupt{} @ascuheap{}}@acunsafe{@acucorrupt{} @acsmem{}}}
@c pthread_setspecific @asucorrupt @ascuheap @acucorrupt @acsmem
@c   a level2 block may be allocated by a signal handler after
@c   another call already made a decision to allocate it, thus losing
@c   the allocated value.  the seq number is updated before the
@c   value, which might cause an earlier-generation value to seem
@c   current if setspecific is cancelled or interrupted by a signal
@c  KEY_UNUSED ok
@c  calloc dup @ascuheap @acsmem
Associate the thread-specific @var{value} with @var{key} in the calling thread.
@end deftypefun


@node Non-POSIX Extensions
@subsection Non-POSIX Extensions

In addition to implementing the POSIX API for threads, @theglibc{} provides
additional functions and interfaces to provide functionality not specified in
the standard.

@menu
* Default Thread Attributes::             Setting default attributes for
					  threads in a process.
* Initial Thread Signal Mask::            Setting the initial mask of threads.
* Waiting with Explicit Clocks::          Functions for waiting with an
                                          explicit clock specification.
* Single-Threaded::                       Detecting single-threaded execution.
* Restartable Sequences::                 Linux-specific restartable sequences
                                          integration.
@end menu

@node Default Thread Attributes
@subsubsection Setting Process-wide defaults for thread attributes

@Theglibc{} provides non-standard API functions to set and get the default
attributes used in the creation of threads in a process.

@deftypefun int pthread_getattr_default_np (pthread_attr_t *@var{attr})
@standards{GNU, pthread.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@asulock{}}@acunsafe{@aculock{}}}
@c Takes lock around read from default_pthread_attr.
Get the default attribute values and set @var{attr} to match.  This
function returns @math{0} on success and a non-zero error code on
failure.
@end deftypefun

@deftypefun int pthread_setattr_default_np (pthread_attr_t *@var{attr})
@standards{GNU, pthread.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@ascuheap{} @asulock{}}@acunsafe{@aculock{} @acsmem{}}}
@c pthread_setattr_default_np @ascuheap @asulock @aculock @acsmem
@c  check_sched_policy_attr ok
@c  check_sched_priority_attr ok
@c   sched_get_priority_min dup ok
@c   sched_get_priority_max dup ok
@c  check_stacksize_attr ok
@c  lll_lock @asulock @aculock
@c  free dup @ascuheap @acsmem
@c  realloc dup @ascuheap @acsmem
@c  memcpy dup ok
@c  lll_unlock @asulock @aculock
Set the default attribute values to match the values in @var{attr}.  The
function returns @math{0} on success and a non-zero error code on failure.
The following error codes are defined for this function:

@table @code
@item EINVAL
At least one of the values in @var{attr} does not qualify as valid for the
attributes or the stack address is set in the attribute.
@item ENOMEM
The system does not have sufficient memory.
@end table
@end deftypefun

@node Initial Thread Signal Mask
@subsubsection Controlling the Initial Signal Mask of a New Thread

@Theglibc{} provides a way to specify the initial signal mask of a
thread created using @code{pthread_create}, passing a thread attribute
object configured for this purpose.

@deftypefun int pthread_attr_setsigmask_np (pthread_attr_t *@var{attr}, const sigset_t *@var{sigmask})
@standards{GNU, pthread.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@ascuheap{}}@acunsafe{@acsmem{}}}
Change the initial signal mask specified by @var{attr}.  If
@var{sigmask} is not @code{NULL}, the initial signal mask for new
threads created with @var{attr} is set to @code{*@var{sigmask}}.  If
@var{sigmask} is @code{NULL}, @var{attr} will no longer specify an
explicit signal mask, so that the initial signal mask of the new
thread is inherited from the thread that calls @code{pthread_create}.

This function returns zero on success, and @code{ENOMEM} on memory
allocation failure.
@end deftypefun

@deftypefun int pthread_attr_getsigmask_np (const pthread_attr_t *@var{attr}, sigset_t *@var{sigmask})
@standards{GNU, pthread.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@ascuheap{}}@acunsafe{@acsmem{}}}
Retrieve the signal mask stored in @var{attr} and copy it to
@code{*@var{sigmask}}.  If the signal mask has not been set, return
the special constant @code{PTHREAD_ATTR_NO_SIGMASK_NP}, otherwise
return zero.

@c Move this to the documentation of pthread_getattr_np once it exists.
Obtaining the signal mask only works if it has been previously stored
by @code{pthread_attr_setsigmask_np}.  For example, the
@code{pthread_getattr_np} function does not obtain the current signal
mask of the specified thread, and @code{pthread_attr_getsigmask_np}
will subsequently report the signal mask as unset.
@end deftypefun

@deftypevr Macro int PTHREAD_ATTR_NO_SIGMASK_NP
The special value returned by @code{pthread_attr_setsigmask_np} to
indicate that no signal mask has been set for the attribute.
@end deftypevr

It is possible to create a new thread with a specific signal mask
without using these functions.  On the thread that calls
@code{pthread_create}, the required steps for the general case are:

@enumerate 1
@item
Mask all signals, and save the old signal mask, using
@code{pthread_sigmask}.  This ensures that the new thread will be
created with all signals masked, so that no signals can be delivered
to the thread until the desired signal mask is set.

@item
Call @code{pthread_create} to create the new thread, passing the
desired signal mask to the thread start routine (which could be a
wrapper function for the actual thread start routine).  It may be
necessary to make a copy of the desired signal mask on the heap, so
that the life-time of the copy extends to the point when the start
routine needs to access the signal mask.

@item
Restore the thread's signal mask, to the set that was saved in the
first step.
@end enumerate

The start routine for the created thread needs to locate the desired
signal mask and use @code{pthread_sigmask} to apply it to the thread.
If the signal mask was copied to a heap allocation, the copy should be
freed.

@node Waiting with Explicit Clocks
@subsubsection Functions for Waiting According to a Specific Clock

@Theglibc{} provides several waiting functions that expect an explicit
@code{clockid_t} argument.

@comment semaphore.h
@comment POSIX-proposed
@deftypefun int sem_clockwait (sem_t *@var{sem}, clockid_t @var{clockid}, const struct timespec *@var{abstime})
@safety{@prelim{}@mtsafe{}@asunsafe{@asulock{}}@acunsafe{@aculock{}}}
Behaves like @code{sem_timedwait} except the time @var{abstime} is measured
against the clock specified by @var{clockid} rather than
@code{CLOCK_REALTIME}.  Currently, @var{clockid} must be either
@code{CLOCK_MONOTONIC} or @code{CLOCK_REALTIME}.
@end deftypefun

@comment pthread.h
@comment POSIX-proposed
@deftypefun int pthread_cond_clockwait (pthread_cond_t *@var{cond}, pthread_mutex_t *@var{mutex}, clockid_t @var{clockid}, const struct timespec *@var{abstime})
@safety{@prelim{}@mtsafe{}@asunsafe{@asulock{}}@acunsafe{@aculock{}}}
@c If exactly the same function with arguments is called from a signal
@c handler that interrupts between the mutex unlock and sleep then it
@c will unlock the mutex twice resulting in undefined behaviour.  Keep
@c in mind that the unlock and sleep are only atomic with respect to other
@c threads (really a happens-after relationship for pthread_cond_broadcast
@c and pthread_cond_signal).
@c In the AC case we would cancel the thread and the mutex would remain
@c locked and we can't recover from that.
Behaves like @code{pthread_cond_timedwait} except the time @var{abstime} is
measured against the clock specified by @var{clockid} rather than the clock
specified or defaulted when @code{pthread_cond_init} was called.  Currently,
@var{clockid} must be either @code{CLOCK_MONOTONIC} or
@code{CLOCK_REALTIME}.
@end deftypefun

@comment pthread.h
@comment POSIX-proposed
@deftypefun int pthread_rwlock_clockrdlock (pthread_rwlock_t *@var{rwlock}, clockid_t @var{clockid}, const struct timespec *@var{abstime})
@safety{@prelim{}@mtsafe{}@asunsafe{@asulock{}}@acunsafe{@aculock{}}}
Behaves like @code{pthread_rwlock_timedrdlock} except the time
@var{abstime} is measured against the clock specified by @var{clockid}
rather than @code{CLOCK_REALTIME}.  Currently, @var{clockid} must be either
@code{CLOCK_MONOTONIC} or @code{CLOCK_REALTIME}, otherwise @code{EINVAL} is
returned.
@end deftypefun

@comment pthread.h
@comment POSIX-proposed
@deftypefun int pthread_rwlock_clockwrlock (pthread_rwlock_t *@var{rwlock}, clockid_t @var{clockid}, const struct timespec *@var{abstime})
@safety{@prelim{}@mtsafe{}@asunsafe{@asulock{}}@acunsafe{@aculock{}}}
Behaves like @code{pthread_rwlock_timedwrlock} except the time
@var{abstime} is measured against the clock specified by @var{clockid}
rather than @code{CLOCK_REALTIME}.  Currently, @var{clockid} must be either
@code{CLOCK_MONOTONIC} or @code{CLOCK_REALTIME}, otherwise @code{EINVAL} is
returned.
@end deftypefun

@comment pthread.h
@comment GNU extension
@deftypefun int pthread_tryjoin_np (pthread_t *@var{thread}, void **@var{thread_return})
@standards{GNU, pthread.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@asulock{}}@acunsafe{@aculock{}}}
Behaves like @code{pthread_join} except that it will return @code{EBUSY}
immediately if the thread specified by @var{thread} has not yet terminated.
@end deftypefun

@comment pthread.h
@comment GNU extension
@deftypefun int pthread_timedjoin_np (pthread_t *@var{thread}, void **@var{thread_return}, const struct timespec *@var{abstime})
@standards{GNU, pthread.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@asulock{}}@acunsafe{@aculock{}}}
Behaves like @code{pthread_tryjoin_np} except that it will block until the
absolute time @var{abstime} measured against @code{CLOCK_REALTIME} is
reached if the thread has not terminated by that time and return
@code{EBUSY}. If @var{abstime} is equal to @code{NULL} then the function
will wait forever in the same way as @code{pthread_join}.
@end deftypefun

@comment pthread.h
@comment GNU extension
@deftypefun int pthread_clockjoin_np (pthread_t *@var{thread}, void **@var{thread_return}, clockid_t @var{clockid}, const struct timespec *@var{abstime})
@standards{GNU, pthread.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@asulock{}}@acunsafe{@aculock{}}}
Behaves like @code{pthread_timedjoin_np} except that the absolute time in
@var{abstime} is measured against the clock specified by @var{clockid}.
Currently, @var{clockid} must be either @code{CLOCK_MONOTONIC} or
@code{CLOCK_REALTIME}.
@end deftypefun

@node Single-Threaded
@subsubsection Detecting Single-Threaded Execution

Multi-threaded programs require synchronization among threads.  This
synchronization can be costly even if there is just a single thread
and no data is shared between multiple processors.  @Theglibc{} offers
an interface to detect whether the process is in single-threaded mode.
Applications can use this information to avoid synchronization, for
example by using regular instructions to load and store memory instead
of atomic instructions, or using relaxed memory ordering instead of
stronger memory ordering.

@deftypevar char __libc_single_threaded
@standards{GNU, sys/single_threaded.h}
This variable is non-zero if the current process is definitely
single-threaded.  If it is zero, the process may be multi-threaded,
or @theglibc{} cannot determine at this point of the program execution
whether the process is single-threaded or not.

Applications must never write to this variable.
@end deftypevar

Most applications should perform the same actions whether or not
@code{__libc_single_threaded} is true, except with less
synchronization.  If this rule is followed, a process that
subsequently becomes multi-threaded is already in a consistent state.
For example, in order to increment a reference count, the following
code can be used:

@smallexample
if (__libc_single_threaded)
  atomic_fetch_add (&reference_count, 1, memory_order_relaxed);
else
  atomic_fetch_add (&reference_count, 1, memory_order_acq_rel);
@end smallexample

@c Note: No memory order on __libc_single_threaded.  The
@c implementation must ensure that exit of the critical
@c (second-to-last) thread happens-before setting
@c __libc_single_threaded to true.  Otherwise, acquire MO might be
@c needed for reading the variable in some scenarios, and that would
@c completely defeat its purpose.

This still requires some form of synchronization on the
single-threaded branch, so it can be beneficial not to declare the
reference count as @code{_Atomic}, and use the GCC @code{__atomic}
built-ins.  @xref{__atomic Builtins,, Built-in Functions for Memory
Model Aware Atomic Operations, gcc, Using the GNU Compiler Collection
(GCC)}.  Then the code to increment a reference count looks like this:

@smallexample
if (__libc_single_threaded)
  ++reference_count;
else
  __atomic_fetch_add (&reference_count, 1, __ATOMIC_ACQ_REL);
@end smallexample

(Depending on the data associated with the reference count, it may be
possible to use the weaker @code{__ATOMIC_RELAXED} memory ordering on
the multi-threaded branch.)

Several functions in @theglibc{} can change the value of the
@code{__libc_single_threaded} variable.  For example, creating new
threads using the @code{pthread_create} or @code{thrd_create} function
sets the variable to false.  This can also happen indirectly, say via
a call to @code{dlopen}.  Therefore, applications need to make a copy
of the value of @code{__libc_single_threaded} if after such a function
call, behavior must match the value as it was before the call, like
this:

@smallexample
bool single_threaded = __libc_single_threaded;
if (single_threaded)
  prepare_single_threaded ();
else
  prepare_multi_thread ();

void *handle = dlopen (shared_library_name, RTLD_NOW);
lookup_symbols (handle);

if (single_threaded)
  cleanup_single_threaded ();
else
  cleanup_multi_thread ();
@end smallexample

Since the value of @code{__libc_single_threaded} can change from true
to false during the execution of the program, it is not useful for
selecting optimized function implementations in IFUNC resolvers.

Atomic operations can also be used on mappings shared among
single-threaded processes.  This means that a compiler must not use
@code{__libc_single_threaded} to optimize atomic operations, unless it
is able to prove that the memory is not shared.

@strong{Implementation Note:} The @code{__libc_single_threaded}
variable is not declared as @code{volatile} because it is expected
that compilers optimize a sequence of single-threaded checks into one
check, for example if several reference counts are updated.  The
current implementation in @theglibc{} does not set the
@code{__libc_single_threaded} variable to a true value if a process
turns single-threaded again.  Future versions of @theglibc{} may do
this, but only as the result of function calls which imply an acquire
(compiler) barrier.  (Some compilers assume that well-known functions
such as @code{malloc} do not write to global variables, and setting
@code{__libc_single_threaded} would introduce a data race and
undefined behavior.)  In any case, an application must not write to
@code{__libc_single_threaded} even if it has joined the last
application-created thread because future versions of @theglibc{} may
create background threads after the first thread has been created, and
the application has no way of knowing that these threads are present.

@node Restartable Sequences
@subsubsection Restartable Sequences

This section describes restartable sequences integration for
@theglibc{}.  This functionality is only available on Linux.

@deftp {Data Type} {struct rseq}
@standards{Linux, sys/rseq.h}
The type of the restartable sequences area.  Future versions
of Linux may add additional fields to the end of this structure.


Users need to obtain the address of the restartable sequences area using
the thread pointer and the @code{__rseq_offset} variable, described
below.

One use of the restartable sequences area is to read the current CPU
number from its @code{cpu_id} field, as an inline version of
@code{sched_getcpu}.  @Theglibc{} sets the @code{cpu_id} field to
@code{RSEQ_CPU_ID_REGISTRATION_FAILED} if registration failed or was
explicitly disabled.

Furthermore, users can store the address of a @code{struct rseq_cs}
object into the @code{rseq_cs} field of @code{struct rseq}, thus
informing the kernel that the thread enters a restartable sequence
critical section.  This pointer and the code areas it itself points to
must not be left pointing to memory areas which are freed or re-used.
Several approaches can guarantee this.  If the application or library
can guarantee that the memory used to hold the @code{struct rseq_cs} and
the code areas it refers to are never freed or re-used, no special
action must be taken.  Else, before that memory is re-used of freed, the
application is responsible for setting the @code{rseq_cs} field to
@code{NULL} in each thread's restartable sequence area to guarantee that
it does not leak dangling references.  Because the application does not
typically have knowledge of libraries' use of restartable sequences, it
is recommended that libraries using restartable sequences which may end
up freeing or re-using their memory set the @code{rseq_cs} field to
@code{NULL} before returning from library functions which use
restartable sequences.

The manual for the @code{rseq} system call can be found
at @uref{https://git.kernel.org/pub/scm/libs/librseq/librseq.git/tree/doc/man/rseq.2}.
@end deftp

@deftypevar {ptrdiff_t} __rseq_offset
@standards{Linux, sys/rseq.h}
This variable contains the offset between the thread pointer (as defined
by @code{__builtin_thread_pointer} or the thread pointer register for
the architecture) and the restartable sequences area.  This value is the
same for all threads in the process.  If the restartable sequences area
is located at a lower address than the location to which the thread
pointer points, the value is negative.
@end deftypevar

@deftypevar {unsigned int} __rseq_size
@standards{Linux, sys/rseq.h}
This variable is either zero (if restartable sequence registration
failed or has been disabled) or the size of the restartable sequence
registration.  This can be different from the size of @code{struct rseq}
if the kernel has extended the size of the registration.  If
registration is successful, @code{__rseq_size} is at least 20 (the
initially active size of @code{struct rseq}).

Previous versions of @theglibc{} set this to 32 even if the kernel only
supported the initial area of 20 bytes because the value included unused
padding at the end of the restartable sequence area.
@end deftypevar

@deftypevar {unsigned int} __rseq_flags
@standards{Linux, sys/rseq.h}
The flags used during restartable sequence registration with the kernel.
Currently zero.
@end deftypevar

@deftypevr Macro int RSEQ_SIG
@standards{Linux, sys/rseq.h}
Each supported architecture provides a @code{RSEQ_SIG} macro in
@file{sys/rseq.h} which contains a signature.  That signature is
expected to be present in the code before each restartable sequences
abort handler.  Failure to provide the expected signature may terminate
the process with a segmentation fault.
@end deftypevr

@c FIXME these are undocumented:
@c pthread_atfork
@c pthread_attr_destroy
@c pthread_attr_getaffinity_np
@c pthread_attr_getdetachstate
@c pthread_attr_getguardsize
@c pthread_attr_getinheritsched
@c pthread_attr_getschedparam
@c pthread_attr_getschedpolicy
@c pthread_attr_getscope
@c pthread_attr_getstack
@c pthread_attr_getstackaddr
@c pthread_attr_getstacksize
@c pthread_attr_init
@c pthread_attr_setaffinity_np
@c pthread_attr_setdetachstate
@c pthread_attr_setguardsize
@c pthread_attr_setinheritsched
@c pthread_attr_setschedparam
@c pthread_attr_setschedpolicy
@c pthread_attr_setscope
@c pthread_attr_setstack
@c pthread_attr_setstackaddr
@c pthread_attr_setstacksize
@c pthread_barrierattr_destroy
@c pthread_barrierattr_getpshared
@c pthread_barrierattr_init
@c pthread_barrierattr_setpshared
@c pthread_barrier_destroy
@c pthread_barrier_init
@c pthread_barrier_wait
@c pthread_cancel
@c pthread_cleanup_push
@c pthread_cleanup_pop
@c pthread_condattr_destroy
@c pthread_condattr_getclock
@c pthread_condattr_getpshared
@c pthread_condattr_init
@c pthread_condattr_setclock
@c pthread_condattr_setpshared
@c pthread_cond_broadcast
@c pthread_cond_destroy
@c pthread_cond_init
@c pthread_cond_signal
@c pthread_cond_timedwait
@c pthread_cond_wait
@c pthread_create
@c pthread_detach
@c pthread_equal
@c pthread_exit
@c pthread_getaffinity_np
@c pthread_getattr_np
@c pthread_getconcurrency
@c pthread_getcpuclockid
@c pthread_getname_np
@c pthread_getschedparam
@c pthread_join
@c pthread_kill
@c pthread_kill_other_threads_np
@c pthread_mutexattr_destroy
@c pthread_mutexattr_getkind_np
@c pthread_mutexattr_getprioceiling
@c pthread_mutexattr_getprotocol
@c pthread_mutexattr_getpshared
@c pthread_mutexattr_getrobust
@c pthread_mutexattr_getrobust_np
@c pthread_mutexattr_gettype
@c pthread_mutexattr_init
@c pthread_mutexattr_setkind_np
@c pthread_mutexattr_setprioceiling
@c pthread_mutexattr_setprotocol
@c pthread_mutexattr_setpshared
@c pthread_mutexattr_setrobust
@c pthread_mutexattr_setrobust_np
@c pthread_mutexattr_settype
@c pthread_mutex_consistent
@c pthread_mutex_consistent_np
@c pthread_mutex_destroy
@c pthread_mutex_getprioceiling
@c pthread_mutex_init
@c pthread_mutex_lock
@c pthread_mutex_setprioceiling
@c pthread_mutex_timedlock
@c pthread_mutex_trylock
@c pthread_mutex_unlock
@c pthread_once
@c pthread_rwlockattr_destroy
@c pthread_rwlockattr_getkind_np
@c pthread_rwlockattr_getpshared
@c pthread_rwlockattr_init
@c pthread_rwlockattr_setkind_np
@c pthread_rwlockattr_setpshared
@c pthread_rwlock_destroy
@c pthread_rwlock_init
@c pthread_rwlock_rdlock
@c pthread_rwlock_timedrdlock
@c pthread_rwlock_timedwrlock
@c pthread_rwlock_tryrdlock
@c pthread_rwlock_trywrlock
@c pthread_rwlock_unlock
@c pthread_rwlock_wrlock
@c pthread_self
@c pthread_setaffinity_np
@c pthread_setcancelstate
@c pthread_setcanceltype
@c pthread_setconcurrency
@c pthread_setname_np
@c pthread_setschedparam
@c pthread_setschedprio
@c pthread_sigmask
@c pthread_sigqueue
@c pthread_spin_destroy
@c pthread_spin_init
@c pthread_spin_lock
@c pthread_spin_trylock
@c pthread_spin_unlock
@c pthread_testcancel
@c pthread_yield