27. Self-Contained Objects

27.1. Introduction

One of the original design goals of RTEMS was the support for heterogeneous computing based on message passing. This was realized by synchronization objects with an architecture-independent identifier provided by the system during object creation (a 32-bit unsigned integer used as a bitfield) and a user-defined four character name. This approach in the so called Classic API has some weaknesses:

  • Dynamic memory (the workspace) is used to allocate object pools. This requires a complex configuration with heavy use of the C pre-processor. The unlimited objects support optionally expands and shrinks the object pool. Dynamic memory is strongly discouraged by some coding standards, e.g. MISRA C:2012 [BBB+13].

  • Objects are created via function calls which return an object identifier. The object operations use this identifier and map it internally to an object representation.

  • The object identifier is only known at run-time. This hinders compiler optimizations and static analysis.

  • The objects reside in a table, e.g. they are subject to false sharing of cache lines [Dre07].

  • The object operations use a rich set of options and attributes. For each object operation these parameters must be evaluated and validated at run-time to figure out what to do exactly for this operation.

For applications that use fine grained locking the mapping of the identifier to the object representation and the parameter evaluation are a significant overhead that may degrade the performance dramatically. An example is the new network stack (libbsd) which uses hundreds of locks in a basic setup. Another example is the OpenMP support (libgomp).

To overcome these issues new self-contained synchronization objects are available since RTEMS 4.11. Self-contained synchronization objects encapsulate all their state in exactly one data structure. The user must provide the storage space for this structure and nothing more. The user is responsible for the object life-cycle. Initialization and destruction of self-contained synchronization objects cannot fail provided all function parameters are valid. In particular, a not enough memory error cannot happen. It is possible to statically initialize self-contained synchronization objects. This allows an efficient use of static analysis tools.

Several header files define self-contained synchronization objects. The Newlib <sys/lock.h> header file provides

  • mutexes,

  • recursive mutexes,

  • condition variables,

  • counting semaphores,

  • binary semaphores, and

  • Futex synchronization [FRK02].

They are used internally in Newlib (e.g. for FILE objects), for the C++11 threads and the OpenMP support (libgomp). The Newlib provided self-contained synchronization objects focus on performance. There are no error checks to catch software errors, e.g. invalid parameters. The application configuration is significantly simplified, since it is no longer necessary to account for lock objects used by Newlib and GCC. The Newlib defined self-contained synchronization objects can be a statically initialized and reside in the .bss section. Destruction is a no-operation.

The header file <pthread.h> provides

  • POSIX barriers (pthread_barrier_t),

  • POSIX condition variables (pthread_cond_t),

  • POSIX mutexes (pthread_mutex_t),

  • POSIX reader/writer locks (pthread_rwlock_t), and

  • POSIX spinlocks (pthread_spinlock_t)

as self-contained synchronization objects. The POSIX synchronization objects are used for example by the Ada run-time support. The header file <semaphore.h> provides self-contained

  • POSIX unnamed semaphores (sem_t initialized via sem_init()).

27.2. RTEMS Thread API

To give RTEMS users access to self-contained synchronization objects an API is necessary. One option would be to simply use the POSIX threads API (pthreads), C11 threads or C++11 threads. However, these standard APIs lack for example binary semaphores which are important for task/interrupt synchronization. The timed operations use in general time values specified by seconds and nanoseconds. Setting up the time values in seconds (time_t has 64 bits) and nanoseconds is burdened with a high overhead compared to time values in clock ticks for relative timeouts. The POSIX API mutexes can be configured for various protocols and options, this adds a run-time overhead. There are a variety of error conditions. This is a problem in combination with some coding standards, e.g. MISRA C:2012. APIs used by Linux (e.g. <linux/mutex.h>) or the FreeBSD kernel (e.g. MUTEX(9)) are better suited as a template for high-performance synchronization objects. The goal of the RTEMS Thread API is to offer the highest performance with the lowest space-overhead on RTEMS. It should be suitable for device drivers.

27.3. Mutual Exclusion

The rtems_mutex and rtems_recursive_mutex objects provide mutual-exclusion synchronization using the Priority Inheritance Protocol in uniprocessor configurations or the O(m) Independence-Preserving Protocol (OMIP) in SMP configurations. Recursive locking should be used with care [Wil12]. The storage space for these object must be provided by the user. There are no defined comparison or assignment operators for these type. Only the object itself may be used for performing synchronization. The result of referring to copies of the object in calls to

  • rtems_mutex_lock(),

  • rtems_recursive_mutex_lock(),

  • rtems_mutex_try_lock(),

  • rtems_recursive_mutex_try_lock(),

  • rtems_mutex_unlock(),

  • rtems_recursive_mutex_unlock(),

  • rtems_mutex_set_name(),

  • rtems_recursive_mutex_set_name(),

  • rtems_mutex_get_name(),

  • rtems_recursive_mutex_get_name(),

  • rtems_mutex_destroy(), and

  • rtems_recursive_mutex_destroy()

is undefined. Objects of the type rtems_mutex must be initialized via

  • RTEMS_MUTEX_INITIALIZER(), or

  • rtems_mutex_init().

They must be destroyed via

  • rtems_mutex_destroy().

Objects of the type rtems_recursive_mutex must be initialized via

  • RTEMS_RECURSIVE_MUTEX_INITIALIZER(), or

  • rtems_recursive_mutex_init().

They must be destroyed via

  • rtems_recursive_mutex_destroy().

27.3.1. Static mutex initialization

CALLING SEQUENCE:
rtems_mutex mutex = RTEMS_MUTEX_INITIALIZER(
  name
);

rtems_recursive_mutex mutex = RTEMS_RECURSIVE_MUTEX_INITIALIZER(
  name
);
DESCRIPTION:

An initializer for static initialization. It is equivalent to a call to rtems_mutex_init() or rtems_recursive_mutex_init() respectively.

NOTES:

Global mutexes with a name of NULL may reside in the .bss section.

27.3.2. Run-time mutex initialization

CALLING SEQUENCE:
void rtems_mutex_init(
  rtems_mutex *mutex,
  const char  *name
);

void rtems_recursive_mutex_init(
  rtems_recursive_mutex *mutex,
  const char            *name
);
DESCRIPTION:

Initializes the mutex with the specified name.

NOTES:

The name must be persistent throughout the life-time of the mutex. A name of NULL is valid. The mutex is unlocked after initialization.

27.3.3. Lock the mutex

CALLING SEQUENCE:
void rtems_mutex_lock(
  rtems_mutex *mutex
);

void rtems_recursive_mutex_lock(
  rtems_recursive_mutex *mutex
);
DESCRIPTION:

Locks the mutex.

NOTES:

This function must be called from thread context with interrupts enabled. In case the mutex is currently locked by another thread, then the thread is blocked until it becomes the mutex owner. Threads wait in priority order.

A recursive lock happens in case the mutex owner tries to lock the mutex again. The result of recursively locking a mutex depends on the mutex variant. For a normal (non-recursive) mutex (rtems_mutex) the result is unpredictable. It could block the owner indefinetly or lead to a fatal deadlock error. A recursive mutex (rtems_recursive_mutex) can be locked recursively by the mutex owner.

Each mutex lock operation must have a corresponding unlock operation.

27.3.4. Try to lock the mutex

CALLING SEQUENCE:
int rtems_mutex_try_lock(
  rtems_mutex *mutex
);

int rtems_recursive_mutex_try_lock(
  rtems_recursive_mutex *mutex
);
DESCRIPTION:

Tries to lock the mutex. In case the mutex is not locked, it will be locked and the function returns with a return value of 0. If the mutex is already locked, the function will return with a value of EBUSY.

NOTES:

This function must be called from thread context with interrupts enabled.

For recursively locking a mutex, please also see the notes for rtems_mutex_lock() and rtems_recursive_mutex_lock().

Each mutex lock operation must have a corresponding unlock operation.

27.3.5. Unlock the mutex

CALLING SEQUENCE:
void rtems_mutex_unlock(
  rtems_mutex *mutex
);

void rtems_recursive_mutex_unlock(
  rtems_recursive_mutex *mutex
);
DESCRIPTION:

Unlocks the mutex.

NOTES:

This function must be called from thread context with interrupts enabled. In case the currently executing thread is not the owner of the mutex, then the result is unpredictable.

Exactly the outer-most unlock will make a recursive mutex available to other threads.

27.3.6. Set mutex name

CALLING SEQUENCE:
void rtems_mutex_set_name(
  rtems_mutex *mutex,
  const char  *name
);

void rtems_recursive_mutex_set_name(
  rtems_recursive_mutex *mutex,
  const char            *name
);
DESCRIPTION:

Sets the mutex name to name.

NOTES:

The name must be persistent throughout the life-time of the mutex. A name of NULL is valid.

27.3.7. Get mutex name

CALLING SEQUENCE:
const char *rtems_mutex_get_name(
  const rtems_mutex *mutex
);

const char *rtems_recursive_mutex_get_name(
  const rtems_recursive_mutex *mutex
);
DESCRIPTION:

Returns the mutex name.

NOTES:

The name may be NULL.

27.3.8. Mutex destruction

CALLING SEQUENCE:
void rtems_mutex_destroy(
  rtems_mutex *mutex
);

void rtems_recursive_mutex_destroy(
  rtems_recursive_mutex *mutex
);
DESCRIPTION:

Destroys the mutex.

NOTES:

In case the mutex is locked or still in use, then the result is unpredictable.

27.4. Condition Variables

The rtems_condition_variable object provides a condition variable synchronization object. The storage space for this object must be provided by the user. There are no defined comparison or assignment operators for this type. Only the object itself may be used for performing synchronization. The result of referring to copies of the object in calls to

  • rtems_condition_variable_wait(),

  • rtems_condition_variable_signal(),

  • rtems_condition_variable_broadcast(),

  • rtems_condition_variable_set_name(),

  • rtems_condition_variable_get_name(), and

  • rtems_condition_variable_destroy()

is undefined. Objects of this type must be initialized via

  • RTEMS_CONDITION_VARIABLE_INITIALIZER(), or

  • rtems_condition_variable_init().

They must be destroyed via

  • rtems_condition_variable_destroy().

27.4.1. Static condition variable initialization

CALLING SEQUENCE:
rtems_condition_variable condition_variable = RTEMS_CONDITION_VARIABLE_INITIALIZER(
  name
);
DESCRIPTION:

An initializer for static initialization. It is equivalent to a call to rtems_condition_variable_init().

NOTES:

Global condition variables with a name of NULL may reside in the .bss section.

27.4.2. Run-time condition variable initialization

CALLING SEQUENCE:
void rtems_condition_variable_init(
  rtems_condition_variable *condition_variable,
  const char               *name
);
DESCRIPTION:

Initializes the condition_variable with the specified name.

NOTES:

The name must be persistent throughout the life-time of the condition variable. A name of NULL is valid.

27.4.3. Wait for condition signal

CALLING SEQUENCE:
void rtems_condition_variable_wait(
  rtems_condition_variable *condition_variable,
  rtems_mutex              *mutex
);
DESCRIPTION:

Atomically waits for a condition signal and unlocks the mutex. Once the condition is signalled to the thread it wakes up and locks the mutex again.

NOTES:

This function must be called from thread context with interrupts enabled. Threads wait in priority order.

27.4.4. Signals a condition change

CALLING SEQUENCE:
void rtems_condition_variable_signal(
  rtems_condition_variable *condition_variable
);
DESCRIPTION:

Signals a condition change to the highest priority waiting thread. If no threads wait currently on this condition variable, then nothing happens.

27.4.5. Broadcasts a condition change

CALLING SEQUENCE:
void rtems_condition_variable_broadcast(
  rtems_condition_variable *condition_variable
);
DESCRIPTION:

Signals a condition change to all waiting thread. If no threads wait currently on this condition variable, then nothing happens.

27.4.6. Set condition variable name

CALLING SEQUENCE:
void rtems_condition_variable_set_name(
  rtems_condition_variable *condition_variable,
  const char               *name
);
DESCRIPTION:

Sets the condition_variable name to name.

NOTES:

The name must be persistent throughout the life-time of the condition variable. A name of NULL is valid.

27.4.7. Get condition variable name

CALLING SEQUENCE:
const char *rtems_condition_variable_get_name(
  const rtems_condition_variable *condition_variable
);
DESCRIPTION:

Returns the condition_variable name.

NOTES:

The name may be NULL.

27.4.8. Condition variable destruction

CALLING SEQUENCE:
void rtems_condition_variable_destroy(
  rtems_condition_variable *condition_variable
);
DESCRIPTION:

Destroys the condition_variable.

NOTES:

In case the condition variable still in use, then the result is unpredictable.

27.5. Counting Semaphores

The rtems_counting_semaphore object provides a counting semaphore synchronization object. The storage space for this object must be provided by the user. There are no defined comparison or assignment operators for this type. Only the object itself may be used for performing synchronization. The result of referring to copies of the object in calls to

  • rtems_counting_semaphore_wait(),

  • rtems_counting_semaphore_post(),

  • rtems_counting_semaphore_set_name(),

  • rtems_counting_semaphore_get_name(), and

  • rtems_counting_semaphore_destroy()

is undefined. Objects of this type must be initialized via

  • RTEMS_COUNTING_SEMAPHORE_INITIALIZER(), or

  • rtems_counting_semaphore_init().

They must be destroyed via

  • rtems_counting_semaphore_destroy().

27.5.1. Static counting semaphore initialization

CALLING SEQUENCE:
rtems_counting_semaphore counting_semaphore = RTEMS_COUNTING_SEMAPHORE_INITIALIZER(
  name,
  value
);
DESCRIPTION:

An initializer for static initialization. It is equivalent to a call to rtems_counting_semaphore_init().

NOTES:

Global counting semaphores with a name of NULL may reside in the .bss section.

27.5.2. Run-time counting semaphore initialization

CALLING SEQUENCE:
void rtems_counting_semaphore_init(
  rtems_counting_semaphore *counting_semaphore,
  const char               *name,
  unsigned int              value
);
DESCRIPTION:

Initializes the counting_semaphore with the specified name and value. The initial value is set to value.

NOTES:

The name must be persistent throughout the life-time of the counting semaphore. A name of NULL is valid.

27.5.3. Wait for a counting semaphore

CALLING SEQUENCE:
void rtems_counting_semaphore_wait(
  rtems_counting_semaphore *counting_semaphore
);
DESCRIPTION:

Waits for the counting_semaphore. In case the current semaphore value is positive, then the value is decremented and the function returns immediately, otherwise the thread is blocked waiting for a semaphore post.

NOTES:

This function must be called from thread context with interrupts enabled. Threads wait in priority order.

27.5.4. Post a counting semaphore

CALLING SEQUENCE:
void rtems_counting_semaphore_post(
  rtems_counting_semaphore *counting_semaphore
);
DESCRIPTION:

Posts the counting_semaphore. In case at least one thread is waiting on the counting semaphore, then the highest priority thread is woken up, otherwise the current value is incremented.

NOTES:

This function may be called from interrupt context. In case it is called from thread context, then interrupts must be enabled.

27.5.5. Set counting semaphore name

CALLING SEQUENCE:
void rtems_counting_semaphore_set_name(
  rtems_counting_semaphore *counting_semaphore,
  const char               *name
);
DESCRIPTION:

Sets the counting_semaphore name to name.

NOTES:

The name must be persistent throughout the life-time of the counting semaphore. A name of NULL is valid.

27.5.6. Get counting semaphore name

CALLING SEQUENCE:
const char *rtems_counting_semaphore_get_name(
  const rtems_counting_semaphore *counting_semaphore
);
DESCRIPTION:

Returns the counting_semaphore name.

NOTES:

The name may be NULL.

27.5.7. Counting semaphore destruction

CALLING SEQUENCE:
void rtems_counting_semaphore_destroy(
  rtems_counting_semaphore *counting_semaphore
);
DESCRIPTION:

Destroys the counting_semaphore.

NOTES:

In case the counting semaphore still in use, then the result is unpredictable.

27.6. Binary Semaphores

The rtems_binary_semaphore object provides a binary semaphore synchronization object. The storage space for this object must be provided by the user. There are no defined comparison or assignment operators for this type. Only the object itself may be used for performing synchronization. The result of referring to copies of the object in calls to

  • rtems_binary_semaphore_wait(),

  • rtems_binary_semaphore_wait_timed_ticks(),

  • rtems_binary_semaphore_try_wait(),

  • rtems_binary_semaphore_post(),

  • rtems_binary_semaphore_set_name(),

  • rtems_binary_semaphore_get_name(), and

  • rtems_binary_semaphore_destroy()

is undefined. Objects of this type must be initialized via

  • RTEMS_BINARY_SEMAPHORE_INITIALIZER(), or

  • rtems_binary_semaphore_init().

They must be destroyed via

  • rtems_binary_semaphore_destroy().

27.6.1. Static binary semaphore initialization

CALLING SEQUENCE:
rtems_binary_semaphore binary_semaphore = RTEMS_BINARY_SEMAPHORE_INITIALIZER(
  name
);
DESCRIPTION:

An initializer for static initialization. It is equivalent to a call to rtems_binary_semaphore_init().

NOTES:

Global binary semaphores with a name of NULL may reside in the .bss section.

27.6.2. Run-time binary semaphore initialization

CALLING SEQUENCE:
void rtems_binary_semaphore_init(
  rtems_binary_semaphore *binary_semaphore,
  const char             *name
);
DESCRIPTION:

Initializes the binary_semaphore with the specified name. The initial value is set to zero.

NOTES:

The name must be persistent throughout the life-time of the binary semaphore. A name of NULL is valid.

27.6.3. Wait for a binary semaphore

CALLING SEQUENCE:
void rtems_binary_semaphore_wait(
  rtems_binary_semaphore *binary_semaphore
);
DESCRIPTION:

Waits for the binary_semaphore. In case the current semaphore value is one, then the value is set to zero and the function returns immediately, otherwise the thread is blocked waiting for a semaphore post.

NOTES:

This function must be called from thread context with interrupts enabled. Threads wait in priority order.

27.6.4. Wait for a binary semaphore with timeout in ticks

CALLING SEQUENCE:
int rtems_binary_semaphore_wait_timed_ticks(
  rtems_binary_semaphore *binary_semaphore,
  uint32_t                ticks
);
DIRECTIVE STATUS CODES:

0

The semaphore wait was successful.

ETIMEDOUT

The semaphore wait timed out.
DESCRIPTION:

Waits for the binary_semaphore with a timeout in ticks. In case the current semaphore value is one, then the value is set to zero and the function returns immediately with a return value of 0, otherwise the thread is blocked waiting for a semaphore post. The time waiting for a semaphore post is limited by ticks. A ticks value of zero specifies an infinite timeout.

NOTES:

This function must be called from thread context with interrupts enabled. Threads wait in priority order.

27.6.5. Tries to wait for a binary semaphore

CALLING SEQUENCE:
int rtems_binary_semaphore_try_wait(
  rtems_binary_semaphore *binary_semaphore
);
DIRECTIVE STATUS CODES:

0

The semaphore wait was successful.

EAGAIN

The semaphore wait failed.
DESCRIPTION:

Tries to wait for the binary_semaphore. In case the current semaphore value is one, then the value is set to zero and the function returns immediately with a return value of 0, otherwise it returns immediately with a return value of EAGAIN.

NOTES:

This function may be called from interrupt context. In case it is called from thread context, then interrupts must be enabled.

27.6.6. Post a binary semaphore

CALLING SEQUENCE:
void rtems_binary_semaphore_post(
  rtems_binary_semaphore *binary_semaphore
);
DESCRIPTION:

Posts the binary_semaphore. In case at least one thread is waiting on the binary semaphore, then the highest priority thread is woken up, otherwise the current value is set to one.

NOTES:

This function may be called from interrupt context. In case it is called from thread context, then interrupts must be enabled.

27.6.7. Set binary semaphore name

CALLING SEQUENCE:
void rtems_binary_semaphore_set_name(
  rtems_binary_semaphore *binary_semaphore,
  const char             *name
);
DESCRIPTION:

Sets the binary_semaphore name to name.

NOTES:

The name must be persistent throughout the life-time of the binary semaphore. A name of NULL is valid.

27.6.8. Get binary semaphore name

CALLING SEQUENCE:
const char *rtems_binary_semaphore_get_name(
  const rtems_binary_semaphore *binary_semaphore
);
DESCRIPTION:

Returns the binary_semaphore name.

NOTES:

The name may be NULL.

27.6.9. Binary semaphore destruction

CALLING SEQUENCE:
void rtems_binary_semaphore_destroy(
  rtems_binary_semaphore *binary_semaphore
);
DESCRIPTION:

Destroys the binary_semaphore.

NOTES:

In case the binary semaphore still in use, then the result is unpredictable.

27.7. Threads

Warning

The self-contained threads support is work in progress. In contrast to the synchronization objects the self-contained thread support is not just an API glue layer to already existing implementations.

The rtems_thread object provides a thread of execution.

CALLING SEQUENCE:
RTEMS_THREAD_INITIALIZER(
  name,
  thread_size,
  priority,
  flags,
  entry,
  arg
);

void rtems_thread_start(
  rtems_thread *thread,
  const char   *name,
  size_t        thread_size,
  uint32_t      priority,
  uint32_t      flags,
  void       ( *entry )( void * ),
  void         *arg
);

void rtems_thread_restart(
  rtems_thread *thread,
  void         *arg
) RTEMS_NO_RETURN;

void rtems_thread_event_send(
  rtems_thread *thread,
  uint32_t      events
);

uint32_t rtems_thread_event_poll(
  rtems_thread *thread,
  uint32_t      events_of_interest
);

uint32_t rtems_thread_event_wait_all(
  rtems_thread *thread,
  uint32_t      events_of_interest
);

uint32_t rtems_thread_event_wait_any(
  rtems_thread *thread,
  uint32_t      events_of_interest
);

void rtems_thread_destroy(
  rtems_thread *thread
);

void rtems_thread_destroy_self(
  void
) RTEMS_NO_RETURN;