pw_sync#

The pw_sync module contains utilities for synchronizing between threads and/or interrupts through signaling primitives and critical section lock primitives.

Warning

This module is still under construction, the API is not yet stable.

Note

The objects in this module do not have an Init() style public API which is common in many RTOS C APIs. Instead, they rely on being able to invoke the native initialization APIs for synchronization primitives during C++ construction.

In order to support global statically constructed synchronization without constexpr constructors, the user and/or backend MUST ensure that any initialization required in your environment is done prior to the creation and/or initialization of the native synchronization primitives (e.g. kernel initialization).

Critical Section Lock Primitives#

The critical section lock primitives provided by this module comply with BasicLockable, Lockable, and where relevant TimedLockable C++ named requirements. This means that they are compatible with existing helpers in the STL’s <mutex> thread support library. For example std::lock_guard and std::unique_lock can be directly used.

Mutex#

The Mutex is a synchronization primitive that can be used to protect shared data from being simultaneously accessed by multiple threads. It offers exclusive, non-recursive ownership semantics where priority inheritance is used to solve the classic priority-inversion problem.

The Mutex’s API is C++11 STL std::mutex like, meaning it is a BasicLockable and Lockable.

Supported on	Backend module
FreeRTOS	pw_sync_freertos
ThreadX	pw_sync_threadx
embOS	pw_sync_embos
STL	pw_sync_stl
Baremetal	Planned
Zephyr	Planned
CMSIS-RTOS API v2 & RTX5	Planned

C++#

class Mutex#

The Mutex is a synchronization primitive that can be used to protect shared data from being simultaneously accessed by multiple threads. It offers exclusive, non-recursive ownership semantics where priority inheritance is used to solve the classic priority-inversion problem. This is thread safe, but NOT IRQ safe.

Warning

In order to support global statically constructed Mutexes, the user and/or backend MUST ensure that any initialization required in your environment is done prior to the creation and/or initialization of the native synchronization primitives (e.g. kernel initialization).

Subclassed by pw::sync::TimedMutex

Public Functions

void lock()#

Locks the mutex, blocking indefinitely. Failures are fatal.

PRECONDITION: The lock isn’t already held by this thread. Recursive locking is undefined behavior.

bool try_lock()#

Attempts to lock the mutex in a non-blocking manner. Returns true if the mutex was successfully acquired.

PRECONDITION: The lock isn’t already held by this thread. Recursive locking is undefined behavior.

void unlock()#

Unlocks the mutex. Failures are fatal.

PRECONDITION: The mutex is held by this thread.

Safe to use in context	Thread	Interrupt	NMI
`pw::sync::Mutex::Mutex`	✔
`pw::sync::Mutex::~Mutex()`	✔
`lock()`	✔
`try_lock()`	✔
`unlock()`	✔

Examples in C++#

#include "pw_sync/mutex.h"

pw::sync::Mutex mutex;

void ThreadSafeCriticalSection() {
  mutex.lock();
  NotThreadSafeCriticalSection();
  mutex.unlock();
}

Alternatively you can use C++’s RAII helpers to ensure you always unlock.

#include <mutex>

#include "pw_sync/mutex.h"

pw::sync::Mutex mutex;

void ThreadSafeCriticalSection() {
  std::lock_guard lock(mutex);
  NotThreadSafeCriticalSection();
}

C#

The Mutex must be created in C++, however it can be passed into C using the pw_sync_Mutex opaque struct alias.

void pw_sync_Mutex_Lock(pw_sync_Mutex *mutex)#: Invokes the Mutex::lock member function on the given mutex.

bool pw_sync_Mutex_TryLock(pw_sync_Mutex *mutex)#: Invokes the Mutex::try_lock member function on the given mutex.

void pw_sync_Mutex_Unlock(pw_sync_Mutex *mutex)#: Invokes the Mutex::unlock member function on the given mutex.

Safe to use in context	Thread	Interrupt	NMI
`void pw_sync_Mutex_Lock`	✔
`bool pw_sync_Mutex_TryLock`	✔
`void pw_sync_Mutex_Unlock`	✔

Example in C#

#include "pw_sync/mutex.h"

pw::sync::Mutex mutex;

extern pw_sync_Mutex mutex;  // This can only be created in C++.

void ThreadSafeCriticalSection(void) {
  pw_sync_Mutex_Lock(&mutex);
  NotThreadSafeCriticalSection();
  pw_sync_Mutex_Unlock(&mutex);
}

TimedMutex#

The TimedMutex is an extension of the Mutex which offers timeout and deadline based semantics.

The TimedMutex’s API is C++11 STL std::timed_mutex like, meaning it is a BasicLockable, Lockable, and TimedLockable.

Note that the TimedMutex is a derived Mutex class, meaning that a TimedMutex can be used by someone who needs the basic Mutex. This is in contrast to the C++ STL’s std::timed_mutex.

Supported on	Backend module
FreeRTOS	pw_sync_freertos
ThreadX	pw_sync_threadx
embOS	pw_sync_embos
STL	pw_sync_stl
Zephyr	Planned
CMSIS-RTOS API v2 & RTX5	Planned

C++#

class TimedMutex : public pw::sync::Mutex #

The TimedMutex is a synchronization primitive that can be used to protect shared data from being simultaneously accessed by multiple threads with timeouts and deadlines, extending the Mutex. It offers exclusive, non-recursive ownership semantics where priority inheritance is used to solve the classic priority-inversion problem. This is thread safe, but NOT IRQ safe.

Warning

In order to support global statically constructed TimedMutexes, the user and/or backend MUST ensure that any initialization required in your environment is done prior to the creation and/or initialization of the native synchronization primitives (e.g. kernel initialization).

Public Functions

bool try_lock_for(chrono::SystemClock::duration timeout)#

Tries to lock the mutex. Blocks until specified the timeout has elapsed or the lock is acquired, whichever comes first. Returns true if the mutex was successfully acquired.

PRECONDITION: The lock isn’t already held by this thread. Recursive locking is undefined behavior.

bool try_lock_until(chrono::SystemClock::time_point deadline)#

Tries to lock the mutex. Blocks until specified deadline has been reached or the lock is acquired, whichever comes first. Returns true if the mutex was successfully acquired.

PRECONDITION: The lock isn’t already held by this thread. Recursive locking is undefined behavior.

Safe to use in context	Thread	Interrupt	NMI
`pw::sync::TimedMutex::TimedMutex`	✔
`pw::sync::TimedMutex::~TimedMutex()`	✔
`pw::sync::Mutex::lock()`	✔
`pw::sync::Mutex::try_lock()`	✔
`try_lock_for()`	✔
`try_lock_until()`	✔
`pw::sync::Mutex::unlock()`	✔

Examples in C++#

#include "pw_chrono/system_clock.h"
#include "pw_sync/timed_mutex.h"

pw::sync::TimedMutex mutex;

bool ThreadSafeCriticalSectionWithTimeout(
    const SystemClock::duration timeout) {
  if (!mutex.try_lock_for(timeout)) {
    return false;
  }
  NotThreadSafeCriticalSection();
  mutex.unlock();
  return true;
}

Alternatively you can use C++’s RAII helpers to ensure you always unlock.

#include <mutex>

#include "pw_chrono/system_clock.h"
#include "pw_sync/timed_mutex.h"

pw::sync::TimedMutex mutex;

bool ThreadSafeCriticalSectionWithTimeout(
    const SystemClock::duration timeout) {
  std::unique_lock lock(mutex, std::defer_lock);
  if (!lock.try_lock_for(timeout)) {
    return false;
  }
  NotThreadSafeCriticalSection();
  return true;
}

C#

The TimedMutex must be created in C++, however it can be passed into C using the pw_sync_TimedMutex opaque struct alias.

Functions

void pw_sync_TimedMutex_Lock(pw_sync_TimedMutex *mutex)#: Invokes the TimedMutex::lock member function on the given mutex.

bool pw_sync_TimedMutex_TryLock(pw_sync_TimedMutex *mutex)#: Invokes the TimedMutex::try_lock member function on the given mutex.

bool pw_sync_TimedMutex_TryLockFor(pw_sync_TimedMutex *mutex, pw_chrono_SystemClock_Duration timeout)#: Invokes the TimedMutex::try_lock_for member function on the given mutex.

bool pw_sync_TimedMutex_TryLockUntil(pw_sync_TimedMutex *mutex, pw_chrono_SystemClock_TimePoint deadline)#: Invokes the TimedMutex::try_lock_until member function on the given mutex.

void pw_sync_TimedMutex_Unlock(pw_sync_TimedMutex *mutex)#: Invokes the TimedMutex::unlock member function on the given mutex.

Safe to use in context	Thread	Interrupt	NMI
`pw_sync_TimedMutex_Lock()`	✔
`pw_sync_TimedMutex_TryLock()`	✔
`pw_sync_TimedMutex_TryLockFor()`	✔
`pw_sync_TimedMutex_TryLockUntil()`	✔
`pw_sync_TimedMutex_Unlock()`	✔

Example in C#

#include "pw_chrono/system_clock.h"
#include "pw_sync/timed_mutex.h"

pw::sync::TimedMutex mutex;

extern pw_sync_TimedMutex mutex;  // This can only be created in C++.

bool ThreadSafeCriticalSectionWithTimeout(
    const pw_chrono_SystemClock_Duration timeout) {
  if (!pw_sync_TimedMutex_TryLockFor(&mutex, timeout)) {
    return false;
  }
  NotThreadSafeCriticalSection();
  pw_sync_TimedMutex_Unlock(&mutex);
  return true;
}

RecursiveMutex#

pw_sync provides pw::sync::RecursiveMutex, a recursive mutex implementation. At this time, this facade can only be used internally by Pigweed.

InterruptSpinLock#

The InterruptSpinLock is a synchronization primitive that can be used to protect shared data from being simultaneously accessed by multiple threads and/or interrupts as a targeted global lock, with the exception of Non-Maskable Interrupts (NMIs). It offers exclusive, non-recursive ownership semantics where IRQs up to a backend defined level of “NMIs” will be masked to solve priority-inversion.

This InterruptSpinLock relies on built-in local interrupt masking to make it interrupt safe without requiring the caller to separately mask and unmask interrupts when using this primitive.

Unlike global interrupt locks, this also works safely and efficiently on SMP systems. On systems which are not SMP, spinning is not required but some state may still be used to detect recursion.

The InterruptSpinLock is a BasicLockable and Lockable.

Supported on	Backend module
FreeRTOS	pw_sync_freertos
ThreadX	pw_sync_threadx
embOS	pw_sync_embos
STL	pw_sync_stl
Baremetal	Planned, not ready for use
Zephyr	Planned
CMSIS-RTOS API v2 & RTX5	Planned

C++#

class InterruptSpinLock#

The InterruptSpinLock is a synchronization primitive that can be used to protect shared data from being simultaneously accessed by multiple threads and/or interrupts as a targeted global lock, with the exception of Non-Maskable Interrupts (NMIs). It offers exclusive, non-recursive ownership semantics where IRQs up to a backend defined level of “NMIs” will be masked to solve priority-inversion.

Unlike global interrupt locks, this also works safely and efficiently on SMP systems. On systems which are not SMP, spinning is not required and it’s possible that only interrupt masking occurs but some state may still be used to detect recursion.

This entire API is IRQ safe, but NOT NMI safe.

Precondition: Code that holds a specific InterruptSpinLock must not try to re-acquire it. However, it is okay to nest distinct spinlocks.

Note

This InterruptSpinLock relies on built-in local interrupt masking to make it interrupt safe without requiring the caller to separately mask and unmask interrupts when using this primitive.

Public Functions

void lock()#

Locks the spinlock, blocking indefinitely. Failures are fatal.

Precondition: Recursive locking is undefined behavior.

bool try_lock()#

Tries to lock the spinlock in a non-blocking manner. Returns true if the spinlock was successfully acquired.

Precondition: Recursive locking is undefined behavior.

void unlock()#

Unlocks the spinlock. Failures are fatal.

Precondition: The spinlock is held by the caller.

Safe to use in context	Thread	Interrupt
`pw::sync::InterruptSpinLock::InterruptSpinLock`	✔	✔
`pw::sync::InterruptSpinLock::~InterruptSpinLock()`	✔	✔
`lock()`	✔	✔
`try_lock()`	✔	✔
`unlock()`	✔	✔

Examples in C++#

#include "pw_sync/interrupt_spin_lock.h"

pw::sync::InterruptSpinLock interrupt_spin_lock;

void InterruptSafeCriticalSection() {
  interrupt_spin_lock.lock();
  NotThreadSafeCriticalSection();
  interrupt_spin_lock.unlock();
}

Alternatively you can use C++’s RAII helpers to ensure you always unlock.

#include <mutex>

#include "pw_sync/interrupt_spin_lock.h"

pw::sync::InterruptSpinLock interrupt_spin_lock;

void InterruptSafeCriticalSection() {
  std::lock_guard lock(interrupt_spin_lock);
  NotThreadSafeCriticalSection();
}

C#

The InterruptSpinLock must be created in C++, however it can be passed into C using the pw_sync_InterruptSpinLock opaque struct alias.

void pw_sync_InterruptSpinLock_Lock(pw_sync_InterruptSpinLock *spin_lock)#: Invokes the InterruptSpinLock::lock member function on the given interrupt_spin_lock.

bool pw_sync_InterruptSpinLock_TryLock(pw_sync_InterruptSpinLock *spin_lock)#: Invokes the InterruptSpinLock::try_lock member function on the given interrupt_spin_lock.

void pw_sync_InterruptSpinLock_Unlock(pw_sync_InterruptSpinLock *spin_lock)#: Invokes the InterruptSpinLock::unlock member function on the given interrupt_spin_lock.

Safe to use in context	Thread	Interrupt
`pw_sync_InterruptSpinLock_Lock()`	✔	✔
`pw_sync_InterruptSpinLock_TryLock()`	✔	✔
`pw_sync_InterruptSpinLock_Unlock()`	✔	✔

Example in C#

#include "pw_chrono/system_clock.h"
#include "pw_sync/interrupt_spin_lock.h"

pw::sync::InterruptSpinLock interrupt_spin_lock;

extern pw_sync_InterruptSpinLock interrupt_spin_lock;  // This can only be created in C++.

void InterruptSafeCriticalSection(void) {
  pw_sync_InterruptSpinLock_Lock(&interrupt_spin_lock);
  NotThreadSafeCriticalSection();
  pw_sync_InterruptSpinLock_Unlock(&interrupt_spin_lock);
}

Thread Safety Lock Annotations#

Pigweed’s critical section lock primitives support Clang’s thread safety analysis extension for C++. The analysis is completely static at compile-time. This is only supported when building with Clang. The annotations are no-ops when using different compilers.

Pigweed provides the pw_sync/lock_annotations.h header file with macro definitions to allow developers to document the locking policies of multi-threaded code. The annotations can also help program analysis tools to identify potential thread safety issues.

More information on Clang’s thread safety analysis system can be found here.

Enabling Clang’s Analysis#

In order to enable the analysis, Clang requires that the -Wthread-safety compilation flag be used. In addition, if any STL components like std::lock_guard are used, the STL’s built in annotations have to be manually enabled, typically by setting the _LIBCPP_ENABLE_THREAD_SAFETY_ANNOTATIONS macro.

If using GN, the pw_build:clang_thread_safety_warnings config is provided to do this for you, when added to your clang toolchain definition’s default configs.

Why use lock annotations?#

Lock annotations can help warn you about potential race conditions in your code when using locks: you have to remember to grab lock(s) before entering a critical section, yuou have to remember to unlock it when you leave, and you have to avoid deadlocks.

Clang’s lock annotations let you inform the compiler and anyone reading your code which variables are guarded by which locks, which locks should or cannot be held when calling which function, which order locks should be acquired in, etc.

Using Lock Annotations#

When referring to locks in the arguments of the attributes, you should use variable names or more complex expressions (e.g. my_object->lock_) that evaluate to a concrete lock object whenever possible. If the lock you want to refer to is not in scope, you may use a member pointer (e.g. &MyClass::lock_) to refer to a lock in some (unknown) object.

Annotating Lock Usage#

PW_GUARDED_BY(x)#

Documents if a shared field or global variable needs to be protected by a lock. PW_GUARDED_BY() allows the user to specify a particular lock that should be held when accessing the annotated variable.

Although this annotation (and PW_PT_GUARDED_BY(), below) cannot be applied to local variables, a local variable and its associated lock can often be combined into a small class or struct, thereby allowing the annotation.

Example:

class Foo {
  Mutex mu_;
  int p1_ PW_GUARDED_BY(mu_);
  ...
};

PW_PT_GUARDED_BY(x)#

Documents if the memory location pointed to by a pointer should be guarded by a lock when dereferencing the pointer.

Example:

class Foo {
  Mutex mu_;
  int *p1_ PW_PT_GUARDED_BY(mu_);
  ...
};

Example:

// `q_`, guarded by `mu1_`, points to a shared memory location that is
// guarded by `mu2_`:
int *q_ PW_GUARDED_BY(mu1_) PW_PT_GUARDED_BY(mu2_);

Note

A pointer variable to a shared memory location could itself be a shared variable.

PW_ACQUIRED_AFTER(...)#

PW_ACQUIRED_BEFORE(...)#

Documents the acquisition order between locks that can be held simultaneously by a thread. For any two locks that need to be annotated to establish an acquisition order, only one of them needs the annotation. (i.e. You don’t have to annotate both locks with both PW_ACQUIRED_AFTER() and PW_ACQUIRED_BEFORE().)

As with PW_GUARDED_BY(), this is only applicable to locks that are shared fields or global variables.

Example:

Mutex m1_;
Mutex m2_ PW_ACQUIRED_AFTER(m1_);

PW_EXCLUSIVE_LOCKS_REQUIRED(...)#

PW_SHARED_LOCKS_REQUIRED(...)#

Documents a function that expects a lock to be held prior to entry. The lock is expected to be held both on entry to, and exit from, the function.

An exclusive lock allows read-write access to the guarded data member(s), and only one thread can acquire a lock exclusively at any one time. A shared lock allows read-only access, and any number of threads can acquire a shared lock concurrently.

Generally, non-const methods should be annotated with PW_EXCLUSIVE_LOCKS_REQUIRED(), while const methods should be annotated with PW_SHARED_LOCKS_REQUIRED().

Example:

Mutex mu1, mu2;
int a PW_GUARDED_BY(mu1);
int b PW_GUARDED_BY(mu2);

void foo() PW_EXCLUSIVE_LOCKS_REQUIRED(mu1, mu2) { ... }
void bar() const PW_SHARED_LOCKS_REQUIRED(mu1, mu2) { ... }

PW_LOCKS_EXCLUDED(...)#

Documents that the caller must not hold the given lock. This annotation is often used to prevent deadlocks. Pigweed’s mutex implementation is not re-entrant, so a deadlock will occur if the function acquires the mutex a second time.

Example:

Mutex mu;
int a PW_GUARDED_BY(mu);

void foo() PW_LOCKS_EXCLUDED(mu) {
  mu.lock();
  ...
  mu.unlock();
}

PW_LOCK_RETURNED(x)#

Documents a function that returns a lock without acquiring it. For example, a public getter method that returns a pointer to a private lock should be annotated with PW_LOCK_RETURNED().

Example:

class Foo {
 public:
  Mutex* mu() PW_LOCK_RETURNED(mu) { return &mu; }

 private:
  Mutex mu;
};

PW_LOCKABLE(name)#: Documents if a class/type is a lockable type (such as the pw::sync::Mutex class). The name is used in the warning messages. This can also be useful on classes which have locking like semantics but aren’t actually locks.

PW_SCOPED_LOCKABLE#

Documents if a class does RAII locking. The name is used in the warning messages.

The constructor should use LOCK_FUNCTION() to specify the lock that is acquired, and the destructor should use UNLOCK_FUNCTION() with no arguments; the analysis will assume that the destructor unlocks whatever the constructor locked.

PW_EXCLUSIVE_LOCK_FUNCTION(...)#: Documents functions that acquire a lock in the body of a function, and do not release it.

PW_SHARED_LOCK_FUNCTION(...)#: Documents functions that acquire a shared (reader) lock in the body of a function, and do not release it.

PW_UNLOCK_FUNCTION(...)#: Documents functions that expect a lock to be held on entry to the function, and release it in the body of the function.

PW_EXCLUSIVE_TRYLOCK_FUNCTION(...)#

PW_SHARED_TRYLOCK_FUNCTION(...)#: Documents functions that try to acquire a lock, and return success or failure (or a non-boolean value that can be interpreted as a boolean). The first argument should be true for functions that return true on success, or false for functions that return false on success. The second argument specifies the lock that is locked on success. If unspecified, this lock is assumed to be this.

PW_ASSERT_EXCLUSIVE_LOCK(...)#

PW_ASSERT_SHARED_LOCK(...)#: Documents functions that dynamically check to see if a lock is held, and fail if it is not held.

PW_NO_LOCK_SAFETY_ANALYSIS#: Turns off thread safety checking within the body of a particular function. This annotation is used to mark functions that are known to be correct, but the locking behavior is more complicated than the analyzer can handle.

Annotating Lock Objects#

In order of lock usage annotation to work, the lock objects themselves need to be annotated as well. In case you are providing your own lock or psuedo-lock object, you can use the macros in this section to annotate it.

As an example we’ve annotated a Lock and a RAII ScopedLocker object for you, see the macro documentation after for more details:

class PW_LOCKABLE("Lock") Lock {
 public:
  void Lock() PW_EXCLUSIVE_LOCK_FUNCTION();

  void ReaderLock() PW_SHARED_LOCK_FUNCTION();

  void Unlock() PW_UNLOCK_FUNCTION();

  void ReaderUnlock() PW_SHARED_TRYLOCK_FUNCTION();

  bool TryLock() PW_EXCLUSIVE_TRYLOCK_FUNCTION(true);

  bool ReaderTryLock() PW_SHARED_TRYLOCK_FUNCTION(true);

  void AssertHeld() PW_ASSERT_EXCLUSIVE_LOCK();

  void AssertReaderHeld() PW_ASSERT_SHARED_LOCK();
};


// Tag types for selecting a constructor.
struct adopt_lock_t {} inline constexpr adopt_lock = {};
struct defer_lock_t {} inline constexpr defer_lock = {};
struct shared_lock_t {} inline constexpr shared_lock = {};

class PW_SCOPED_LOCKABLE ScopedLocker {
  // Acquire lock, implicitly acquire *this and associate it with lock.
  ScopedLocker(Lock *lock) PW_EXCLUSIVE_LOCK_FUNCTION(lock)
      : lock_(lock), locked(true) {
    lock->Lock();
  }

  // Assume lock is held, implicitly acquire *this and associate it with lock.
  ScopedLocker(Lock *lock, adopt_lock_t) PW_EXCLUSIVE_LOCKS_REQUIRED(lock)
      : lock_(lock), locked(true) {}

  // Acquire lock in shared mode, implicitly acquire *this and associate it
  // with lock.
  ScopedLocker(Lock *lock, shared_lock_t) PW_SHARED_LOCK_FUNCTION(lock)
      : lock_(lock), locked(true) {
    lock->ReaderLock();
  }

  // Assume lock is held in shared mode, implicitly acquire *this and associate
  // it with lock.
  ScopedLocker(Lock *lock, adopt_lock_t, shared_lock_t)
      PW_SHARED_LOCKS_REQUIRED(lock) : lock_(lock), locked(true) {}

  // Assume lock is not held, implicitly acquire *this and associate it with
  // lock.
  ScopedLocker(Lock *lock, defer_lock_t) PW_LOCKS_EXCLUDED(lock)
      : lock_(lock), locked(false) {}

  // Release *this and all associated locks, if they are still held.
  // There is no warning if the scope was already unlocked before.
  ~ScopedLocker() PW_UNLOCK_FUNCTION() {
    if (locked)
      lock_->GenericUnlock();
  }

  // Acquire all associated locks exclusively.
  void Lock() PW_EXCLUSIVE_LOCK_FUNCTION() {
    lock_->Lock();
    locked = true;
  }

  // Try to acquire all associated locks exclusively.
  bool TryLock() PW_EXCLUSIVE_TRYLOCK_FUNCTION(true) {
    return locked = lock_->TryLock();
  }

  // Acquire all associated locks in shared mode.
  void ReaderLock() PW_SHARED_LOCK_FUNCTION() {
    lock_->ReaderLock();
    locked = true;
  }

  // Try to acquire all associated locks in shared mode.
  bool ReaderTryLock() PW_SHARED_TRYLOCK_FUNCTION(true) {
    return locked = lock_->ReaderTryLock();
  }

  // Release all associated locks. Warn on double unlock.
  void Unlock() PW_UNLOCK_FUNCTION() {
    lock_->Unlock();
    locked = false;
  }

  // Release all associated locks. Warn on double unlock.
  void ReaderUnlock() PW_UNLOCK_FUNCTION() {
    lock_->ReaderUnlock();
    locked = false;
  }

 private:
  Lock* lock_;
  bool locked_;
};

Critical Section Lock Helpers#

Virtual Lock Interfaces#

Virtual lock interfaces can be useful when lock selection cannot be templated.

Why use virtual locks?#

Virtual locks enable depending on locks without templating implementation code on the type, while retaining flexibility with respect to the concrete lock type. Pigweed tries to avoid pushing policy on to users, and virtual locks are one way to accomplish that without templating everything.

A case when virtual locks are useful is when the concrete lock type changes at run time. For example, access to flash may be protected at run time by an internal mutex, however at crash time we may want to switch to a no-op lock. A virtual lock interface could be used here to minimize the code-size cost that would occur otherwise if the flash driver were templated.

VirtualBasicLockable#

The VirtualBasicLockable interface meets the BasicLockable C++ named requirement. Our critical section lock primitives offer optional virtual versions, including:

pw::sync::VirtualMutex()
pw::sync::VirtualTimedMutex()
pw::sync::VirtualInterruptSpinLock()

GenericBasicLockable#

GenericBasicLockable is a helper construct that can be used to declare virtual versions of a critical section lock primitive that meets the BasicLockable C++ named requirement. For example, given a Mutex type with lock() and unlock() methods, a VirtualMutex type that derives from VirtualBasicLockable can be declared as follows:

class VirtualMutex : public GenericBasicLockable<Mutex> {};

Borrowable#

Borrowable is a helper construct that enables callers to borrow an object which is guarded by a lock, enabling a containerized style of external locking.

Users who need access to the guarded object can ask to acquire a BorrowedPointer which permits access while the lock is held.

This class is compatible with locks which comply with BasicLockable, Lockable, and TimedLockable C++ named requirements.

By default the selected lock type is a pw::sync::VirtualBasicLockable. If this virtual interface is used, the templated lock parameter can be skipped.

External vs Internal locking#

Before we explain why Borrowable is useful, it’s important to understand the trade-offs when deciding on using internal and/or external locking.

Internal locking is when the lock is hidden from the caller entirely and is used internally to the API. For example:

class BankAccount {
 public:
  void Deposit(int amount) {
    std::lock_guard lock(mutex_);
    balance_ += amount;
  }

  void Withdraw(int amount) {
    std::lock_guard lock(mutex_);
    balance_ -= amount;
  }

  void Balance() const {
    std::lock_guard lock(mutex_);
    return balance_;
  }

 private:
  int balance_ PW_GUARDED_BY(mutex_);
  pw::sync::Mutex mutex_;
};

Internal locking guarantees that any concurrent calls to its public member functions don’t corrupt an instance of that class. This is typically ensured by having each member function acquire a lock on the object upon entry. This way, for any instance, there can only be one member function call active at any moment, serializing the operations.

One common issue that pops up is that member functions may have to call other member functions which also require locks. This typically results in a duplication of the public API into an internal mirror where the lock is already held. This along with having to modify every thread-safe public member function may results in an increased code size.

However, with the per-method locking approach, it is not possible to perform a multi-method thread-safe transaction. For example, what if we only wanted to withdraw money if the balance was high enough? With the current API there would be a risk that money is withdrawn after we’ve checked the balance.

This is usually why external locking is used. This is when the lock is exposed to the caller and may be used externally to the public API. External locking can take may forms which may even include mixing internal and external locking. In its most simplistic form it is an external lock used along side each instance, e.g.:

class BankAccount {
 public:
  void Deposit(int amount) {
    balance_ += amount;
  }

  void Withdraw(int amount) {
    balance_ -= amount;
  }

  void Balance() const {
    return balance_;
  }

 private:
  int balance_;
};

pw::sync::Mutex bobs_account_mutex;
BankAccount bobs_account PW_GUARDED_BY(bobs_account_mutex);

The lock is acquired before the bank account is used for a transaction. In addition, we do not have to modify every public function and its trivial to call other public member functions from a public member function. However, as you can imagine instantiating and passing around the instances and their locks can become error prone.

This is why Borrowable exists.

Why use Borrowable?#

Borrowable offers code-size efficient way to enable external locking that is easy and safe to use. It is effectively a container which holds references to a protected instance and its lock which provides RAII-style access.

pw::sync::Mutex bobs_account_mutex;
BankAccount bobs_account PW_GUARDED_BY(bobs_account_mutex);
pw::sync::Borrowable<BankAccount, pw::sync::Mutex> bobs_acount(
    bobs_account, bobs_account_mutex);

This construct is useful when sharing objects or data which are transactional in nature where making individual operations threadsafe is insufficient. See the section on internal vs external locking tradeoffs above.

It can also offer a code-size and stack-usage efficient way to separate timeout constraints between the acquiring of the shared object and timeouts used for the shared object’s API. For example, imagine you have an I2c bus which is used by several threads and you’d like to specify an ACK timeout of 50ms. It’d be ideal if the duration it takes to gain exclusive access to the I2c bus does not eat into the ACK timeout you’d like to use for the transaction. Borrowable can help you do exactly this if you provide access to the I2c bus through a Borrowable.

Note

Borrowable has semantics similar to a pointer and should be passed by value. Furthermore, a Borrowable<U> can be assigned to a Borrowable<T> if U is a subclass of T.

C++#

template<typename GuardedType, typename Lock = pw::sync::VirtualBasicLockable> class BorrowedPointer#

The BorrowedPointer is an RAII handle which wraps a pointer to a borrowed object along with a held lock which is guarding the object. When destroyed, the lock is released.

Public Functions

inline ~BorrowedPointer()#: Release the lock on destruction.

template<typename OtherType, typename OtherLock> inline BorrowedPointer(BorrowedPointer<OtherType, OtherLock> &&other)#

Move-constructs a BorrowedPointer<T> from a BorrowedPointer<U>.

This allows not only pure move construction where GuardedType == OtherType and Lock == OtherLock, but also converting construction where GuardedType is a base class of OtherType and Lock is a base class of OtherLock, like `BorrowedPointer<Base> base_ptr(derived_borrowable.acquire());

Postcondition: The other BorrowedPointer is no longer valid and will assert if the GuardedType is accessed.

template<typename OtherType, typename OtherLock> inline BorrowedPointer &operator=( BorrowedPointer<OtherType, OtherLock> &&other, )#

Move-assigns a BorrowedPointer<T> from a BorrowedPointer<U>.

Postcondition: The other BorrowedPointer is no longer valid and will assert if the GuardedType is accessed.

inline GuardedType *operator->()#: Provides access to the borrowed object’s members.

inline const GuardedType *operator->() const#: Const overload.

inline GuardedType &operator*()#

Provides access to the borrowed object directly.

Note

The member of pointer member access operator, operator->(), is recommended over this API as this is prone to leaking references. However, this is sometimes necessary.

inline const GuardedType &operator*() const#: Const overload.

template<typename GuardedType, typename Lock = pw::sync::VirtualBasicLockable> class Borrowable#

The Borrowable is a helper construct that enables callers to borrow an object which is guarded by a lock.

Users who need access to the guarded object can ask to acquire a BorrowedPointer which permits access while the lock is held.

Thread-safety analysis is not supported for this class, as the BorrowedPointers it creates conditionally releases the lock. See also https://clang.llvm.org/docs/ThreadSafetyAnalysis.html#no-conditionally-held-locks

This class is compatible with locks which comply with BasicLockable, Lockable, and TimedLockable C++ named requirements.

Borrowable<T> is covariant with respect to T, so that Borrowable<U> can be converted to Borrowable<T>, if U is a subclass of T.

Borrowable has pointer-like semantics and should be passed by value.

Subclassed by pw::sync::InlineBorrowable< GuardedType, Lock, LockInterface >

Public Functions

inline BorrowedPointer<GuardedType, Lock> acquire() const#: Blocks indefinitely until the object can be borrowed. Failures are fatal.

template<int&... ExplicitArgumentBarrier, typename T = Lock, typename = std::enable_if_t<is_lockable_v<T>>> inline std::optional<BorrowedPointer<GuardedType, Lock>> try_acquire( ) const#: Tries to borrow the object in a non-blocking manner. Returns a BorrowedPointer on success, otherwise std::nullopt (nothing).

template<class Rep, class Period, int&... ExplicitArgumentBarrier, typename T = Lock, typename = std::enable_if_t<is_lockable_for_v<T, std::chrono::duration<Rep, Period>>>> inline std::optional<BorrowedPointer<GuardedType, Lock>> try_acquire_for( std::chrono::duration<Rep, Period> timeout, ) const#: Tries to borrow the object. Blocks until the specified timeout has elapsed or the object has been borrowed, whichever comes first. Returns a BorrowedPointer on success, otherwise std::nullopt (nothing).

template<class Clock, class Duration, int&... ExplicitArgumentBarrier, typename T = Lock, typename = std::enable_if_t<is_lockable_until_v<T, std::chrono::time_point<Clock, Duration>>>> inline std::optional<BorrowedPointer<GuardedType, Lock>> try_acquire_until( std::chrono::time_point<Clock, Duration> deadline, ) const#: Tries to borrow the object. Blocks until the specified deadline has passed or the object has been borrowed, whichever comes first. Returns a BorrowedPointer on success, otherwise std::nullopt (nothing).

Example in C++#

#include <chrono>

#include "pw_bytes/span.h"
#include "pw_i2c/initiator.h"
#include "pw_status/try.h"
#include "pw_status/result.h"
#include "pw_sync/borrow.h"
#include "pw_sync/mutex.h"

class ExampleI2c : public pw::i2c::Initiator;

pw::sync::VirtualMutex i2c_mutex;
ExampleI2c i2c;
pw::sync::Borrowable<ExampleI2c> borrowable_i2c(i2c, i2c_mutex);

pw::Result<ConstByteSpan> ReadI2cData(ByteSpan buffer) {
  // Block indefinitely waiting to borrow the i2c bus.
  pw::sync::BorrowedPointer<ExampleI2c> borrowed_i2c =
      borrowable_i2c.acquire();

  // Execute a sequence of transactions to get the needed data.
  PW_TRY(borrowed_i2c->WriteFor(kFirstWrite, std::chrono::milliseconds(50)));
  PW_TRY(borrowed_i2c->WriteReadFor(kSecondWrite, buffer,
                                    std::chrono::milliseconds(10)));

  // Borrowed i2c pointer is returned when the scope exits.
  return buffer;
}

InlineBorrowable#

InlineBorrowable is a helper to simplify the common use case where an object is wrapped in a Borrowable for its entire lifetime. The InlineBorrowable owns the guarded object and the lock object.

InlineBorrowable has a separate parameter for the concrete lock type that is instantiated and a (possibly virtual) lock interface type that is referenced by users of the guarded object. The default lock is pw::sync::VirtualMutex and the default lock interface is pw::sync::VirtualBasicLockable.

An InlineBorrowable is a Borrowable with the same guarded object and lock interface types, and it can be passed directly to APIs that expect a Borrowable reference.

Why use InlineBorrowable?#

It is a safer and simpler way to guard an object for its entire lifetime. The unguarded object is never exposed and doesn’t need to be stored in a separate variable or data member. The guarded object and its lock are guaranteed to have the same lifetime, and the lock cannot be re-used for any other purpose.

Constructing objects in-place#

The guarded object and its lock are constructed in-place by the InlineBorrowable, and any constructor parameters required by the object or its lock must be passed through the InlineBorrowable constructor. There are several ways to do this:

Pass the parameters for the guarded object inline to the constructor. This is the recommended way to construct the object when the lock does not require any constructor parameters. Use the std::in_place marker to invoke the inline constructor.
```
InlineBorrowable<Foo> foo(std::in_place, foo_arg1, foo_arg2);
InlineBorrowable<std::array<int, 2>> foo_array(std::in_place, 1, 2);
```

Pass the parameters inside tuples:

InlineBorrowable<Foo> foo(std::forward_as_tuple(foo_arg1, foo_arg2));

InlineBorrowable<Foo, MyLock> foo_lock(
    std::forward_as_tuple(foo_arg1, foo_arg2),
    std::forward_as_tuple(lock_arg1, lock_arg2));

Note

This approach only supports list initialization starting with C++20.

Use callables to construct the guarded object and lock object:

InlineBorrowable<Foo> foo([&]{ return Foo{foo_arg1, foo_arg2}; });

InlineBorrowable<Foo, MyLock> foo_lock(
    [&]{ return Foo{foo_arg1, foo_arg2}; }
    [&]{ return MyLock{lock_arg1, lock_arg2}; }

Note

It is possible to construct and return objects that are not copyable or movable, thanks to mandatory copy ellision (return value optimization).

C++#

template<typename GuardedType, typename Lock = pw::sync::VirtualMutex, typename LockInterface = pw::sync::VirtualBasicLockable> class InlineBorrowable : private internal::BorrowableStorage<GuardedType, pw::sync::VirtualMutex>, public pw::sync::Borrowable<GuardedType, pw::sync::VirtualBasicLockable>#

InlineBorrowable holds an object of GuardedType and a Lock that guards access to the object. It should be used when an object should be guarded for its entire lifecycle by a single lock.

This object should be shared with other componetns as a reference of type Borrowable<GuardedType, LockInterface>.

Public Functions

inline constexpr InlineBorrowable()#: Construct the guarded object and lock using their default constructors.

template<typename ...Args> inline explicit constexpr InlineBorrowable(std::in_place_t, Args&&... args)#

Construct the guarded object by providing its constructor arguments inline. The lock is constructed using its default constructor.

This constructor supports list initialization for arrays, structs, and other objects such as std::array.

Example:

InlineBorrowable<Foo> foo(std::in_place, foo_arg1, foo_arg2);

InlineBorrowable<std::array<int, 2>> foo_array(std::in_place, 1, 2);

template<typename ...ObjectArgs, typename ...LockArgs> inline explicit constexpr InlineBorrowable( std::tuple<ObjectArgs...> &&object_args, std::tuple<LockArgs...> &&lock_args = std::make_tuple(), )#

Construct the guarded object and lock by providing their construction parameters using separate tuples. The 2nd tuple can be ommitted to construct the lock using its default constructor.

Example:

InlineBorrowable<Foo> foo(std::forward_as_tuple(foo_arg1, foo_arg2));

InlineBorrowable<Foo, MyLock> foo_lock(
    std::forward_as_tuple(foo_arg1, foo_arg2),
    std::forward_as_tuple(lock_arg1, lock_arg2));

Note

This constructor only supports list initialization with C++20 or later, because it requires https://wg21.link/p0960.

template<typename ObjectConstructor, typename LockConstructor = Lock(), typename = std::enable_if_t<std::is_invocable_r_v<GuardedType&&, ObjectConstructor>>, typename = std::enable_if_t<std::is_invocable_r_v<Lock&&, LockConstructor>>> inline explicit constexpr InlineBorrowable( const ObjectConstructor &object_ctor, const LockConstructor &lock_ctor = internal::DefaultConstruct<Lock>, )#

Construct the guarded object and lock by providing factory functions. The 2nd callable can be ommitted to construct the lock using its default constructor.

Example:

InlineBorrowable<Foo> foo([&]{ return Foo{foo_arg1, foo_arg2}; });

InlineBorrowable<Foo, MyLock> foo_lock(
    [&]{ return Foo{foo_arg1, foo_arg2}; }
    [&]{ return MyLock{lock_arg1, lock_arg2}; }

Example in C++#

#include <utility>

#include "pw_bytes/span.h"
#include "pw_i2c/initiator.h"
#include "pw_status/result.h"
#include "pw_sync/inline_borrowable.h"

struct I2cOptions;

class ExampleI2c : public pw::i2c::Initiator {
 public:
  ExampleI2c(int bus_id, I2cOptions options);
  // ...
};

int kBusId;
I2cOptions opts;

pw::sync::InlineBorrowable<ExampleI2c> i2c(std::in_place, kBusId, opts);

pw::Result<ConstByteSpan> ReadI2cData(
  pw::sync::Borrowable<pw::i2c::Initiator> initiator,
  ByteSpan buffer);

pw::Result<ConstByteSpan> ReadData(ByteSpan buffer) {
  return ReadI2cData(i2c, buffer);
}

Signaling Primitives#

Native signaling primitives tend to vary more compared to critial section locks across different platforms. For example, although common signaling primtives like semaphores are in most if not all RTOSes and even POSIX, it was not in the STL before C++20. Likewise many C++ developers are surprised that conditional variables tend to not be natively supported on RTOSes. Although you can usually build any signaling primitive based on other native signaling primitives, this may come with non-trivial added overhead in ROM, RAM, and execution efficiency.

For this reason, Pigweed intends to provide some simpler signaling primitives which exist to solve a narrow programming need but can be implemented as efficiently as possible for the platform that it is used on.

This simpler but highly portable class of signaling primitives is intended to ensure that a portability efficiency tradeoff does not have to be made up front. Today this is class of simpler signaling primitives is limited to the pw::sync::ThreadNotification and pw::sync::TimedThreadNotification.

ThreadNotification#

The ThreadNotification is a synchronization primitive that can be used to permit a SINGLE thread to block and consume a latching, saturating notification from multiple notifiers.

Note

Although only a single thread can block on a ThreadNotification at a time, many instances may be used by a single thread just like binary semaphores. This is in contrast to some native RTOS APIs, such as direct task notifications, which re-use the same state within a thread’s context.

Warning

This is a single consumer/waiter, multiple producer/notifier API! The acquire APIs must only be invoked by a single consuming thread. As a result, having multiple threads receiving notifications via the acquire API is unsupported.

This is effectively a subset of the BinarySemaphore API, except that only a single thread can be notified and block at a time.

The single consumer aspect of the API permits the use of a smaller and/or faster native APIs such as direct thread signaling. This should be backed by the most efficient native primitive for a target, regardless of whether that is a semaphore, event flag group, condition variable, or something else.

The ThreadNotification is initialized to being empty (latch is not set).

Generic BinarySemaphore-based Backend#

This module provides a generic backend for pw::sync::ThreadNotification via pw_sync:binary_semaphore_thread_notification which uses a pw::sync::BinarySemaphore as the backing primitive. See BinarySemaphore for backend availability.

Optimized Backend#

Supported on	Optimized backend module
FreeRTOS	`pw_sync_freertos:thread_notification`
ThreadX	Not possible, use `pw_sync:binary_semaphore_thread_notification`
embOS	Not needed, use `pw_sync:binary_semaphore_thread_notification`
STL	Not planned, use `pw_sync:binary_semaphore_thread_notification`
Baremetal	Planned
Zephyr	Planned
CMSIS-RTOS API v2 & RTX5	Planned

C++#

class ThreadNotification#

The ThreadNotification is a synchronization primitive that can be used to permit a SINGLE thread to block and consume a latching, saturating notification from multiple notifiers.

IMPORTANT: This is a single consumer/waiter, multiple producer/notifier API! The acquire APIs must only be invoked by a single consuming thread. As a result, having multiple threads receiving notifications via the acquire API is unsupported.

This is effectively a subset of a binary semaphore API, except that only a single thread can be notified and block at a time.

The single consumer aspect of the API permits the use of a smaller and/or faster native APIs such as direct thread signaling.

The ThreadNotification is initialized to being empty (latch is not set).

Subclassed by pw::sync::TimedThreadNotification

Public Functions

void acquire()#

Blocks indefinitely until the thread is notified, i.e. until the notification latch can be cleared because it was set.

Clears the notification latch.

IMPORTANT: This should only be used by a single consumer thread.

bool try_acquire()#

Returns whether the thread has been notified, i.e. whether the notificion latch was set and resets the latch regardless.

Clears the notification latch.

Returns true if the thread was notified, meaning the the internal latch was reset successfully.

IMPORTANT: This should only be used by a single consumer thread.

void release()#

Notifies the thread in a saturating manner, setting the notification latch.

Raising the notification multiple time without it being acquired by the consuming thread is equivalent to raising the notification once to the thread. The notification is latched in case the thread was not waiting at the time.

This is IRQ and thread safe.

Safe to use in context	Thread	Interrupt
`pw::sync::ThreadNotification::ThreadNotification`	✔
`pw::sync::ThreadNotification::~ThreadNotification()`	✔
`acquire()`	✔
`try_acquire()`	✔
`release()`	✔	✔

Examples in C++#

#include "pw_sync/thread_notification.h"
#include "pw_thread/thread_core.h"

class FooHandler() : public pw::thread::ThreadCore {
 // Public API invoked by other threads and/or interrupts.
 void NewFooAvailable() {
   new_foo_notification_.release();
 }

 private:
  pw::sync::ThreadNotification new_foo_notification_;

  // Thread function.
  void Run() override {
    while (true) {
      new_foo_notification_.acquire();
      HandleFoo();
    }
  }

  void HandleFoo();
}

TimedThreadNotification#

The TimedThreadNotification is an extension of the ThreadNotification which offers timeout and deadline based semantics.

The TimedThreadNotification is initialized to being empty (latch is not set).

Warning

Generic BinarySemaphore-based Backend#

This module provides a generic backend for pw::sync::TimedThreadNotification via pw_sync:binary_semaphore_timed_thread_notification which uses a pw::sync::BinarySemaphore as the backing primitive. See BinarySemaphore for backend availability.

Optimized Backend#

Supported on	Backend module
FreeRTOS	`pw_sync_freertos:timed_thread_notification`
ThreadX	Not possible, use `pw_sync:binary_semaphore_timed_thread_notification`
embOS	Not needed, use `pw_sync:binary_semaphore_timed_thread_notification`
STL	Not planned, use `pw_sync:binary_semaphore_timed_thread_notification`
Zephyr	Planned
CMSIS-RTOS API v2 & RTX5	Planned

C++#

class TimedThreadNotification : public pw::sync::ThreadNotification #

The TimedThreadNotification is a synchronization primitive that can be used to permit a SINGLE thread to block and consume a latching, saturating notification from multiple notifiers.

IMPORTANT: This is a single consumer/waiter, multiple producer/notifier API! The acquire APIs must only be invoked by a single consuming thread. As a result, having multiple threads receiving notifications via the acquire API is unsupported.

This is effectively a subset of a binary semaphore API, except that only a single thread can be notified and block at a time.

The single consumer aspect of the API permits the use of a smaller and/or faster native APIs such as direct thread signaling.

The TimedThreadNotification is initialized to being empty (latch is not set).

Public Functions

bool try_acquire_for(chrono::SystemClock::duration timeout)#

Blocks until the specified timeout duration has elapsed or the thread has been notified (i.e. notification latch can be cleared because it was set), whichever comes first.

Clears the notification latch.

Returns true if the thread was notified, meaning the the internal latch was reset successfully.

IMPORTANT: This should only be used by a single consumer thread.

bool try_acquire_until(chrono::SystemClock::time_point deadline)#

Blocks until the specified deadline time has been reached the thread has been notified (i.e. notification latch can be cleared because it was set), whichever comes first.

Clears the notification latch.

Returns true if the thread was notified, meaning the the internal latch was reset successfully.

IMPORTANT: This should only be used by a single consumer thread.

Safe to use in context	Thread	Interrupt
`pw::sync::TimedThreadNotification::TimedThreadNotification`	✔
`pw::sync::TimedThreadNotification::~TimedThreadNotification()`	✔
`acquire()`	✔
`try_acquire()`	✔
`try_acquire_for()`	✔
`try_acquire_until()`	✔
`release()`	✔	✔

Examples in C++#

#include "pw_sync/timed_thread_notification.h"
#include "pw_thread/thread_core.h"

class FooHandler() : public pw::thread::ThreadCore {
 // Public API invoked by other threads and/or interrupts.
 void NewFooAvailable() {
   new_foo_notification_.release();
 }

 private:
  pw::sync::TimedThreadNotification new_foo_notification_;

  // Thread function.
  void Run() override {
    while (true) {
      if (new_foo_notification_.try_acquire_for(kNotificationTimeout)) {
        HandleFoo();
      }
      DoOtherStuff();
    }
  }

  void HandleFoo();
  void DoOtherStuff();
}

CountingSemaphore#

Note that unlike Mutex, priority inheritance is not used by semaphores meaning semaphores are subject to unbounded priority inversions. Due to this, Pigweed does not recommend semaphores for mutual exclusion.

The CountingSemaphore is initialized to being empty or having no tokens.

The entire API is thread safe, but only a subset is interrupt safe.

Note

If there is only a single consuming thread, use a ThreadNotification instead which can be much more efficient on some RTOSes such as FreeRTOS.

Warning

Releasing multiple tokens is often not natively supported, meaning you may end up invoking the native kernel API many times, i.e. once per token you are releasing!

Supported on	Backend module
FreeRTOS	pw_sync_freertos
ThreadX	pw_sync_threadx
embOS	pw_sync_embos
STL	pw_sync_stl
Zephyr	Planned
CMSIS-RTOS API v2 & RTX5	Planned

C++#

class CountingSemaphore#

The CountingSemaphore is a synchronization primitive that can be used for counting events and/or resource management where receiver(s) can block on acquire until notifier(s) signal by invoking release. Note that unlike Mutexes, priority inheritance is not used by semaphores meaning semaphores are subject to unbounded priority inversions. Pigweed does not recommend semaphores for mutual exclusion. The entire API is thread safe but only a subset is IRQ safe.

The CountingSemaphore is initialized to being empty or having no tokens.

Warning

In order to support global statically constructed CountingSemaphores the user and/or backend MUST ensure that any initialization required in your environment is done prior to the creation and/or initialization of the native synchronization primitives (e.g. kernel initialization).

Public Functions

void release(ptrdiff_t update = 1)#

Atomically increments the internal counter by the value of update. Any thread(s) waiting for the counter to be greater than 0, i.e. blocked in acquire, will subsequently be unblocked. This is IRQ safe.

Precondition: update >= 0

Precondition: update <= max() - counter

void acquire()#

Decrements the internal counter by 1 or blocks indefinitely until it can.

This is thread safe, but not IRQ safe.

bool try_acquire() noexcept#

Tries to decrement by the internal counter by 1 without blocking. Returns true if the internal counter was decremented successfully.

This is IRQ safe.

bool try_acquire_for(chrono::SystemClock::duration timeout)#

Tries to decrement the internal counter by 1. Blocks until the specified timeout has elapsed or the counter was decremented by 1, whichever comes first.

Returns true if the internal counter was decremented successfully. This is thread safe, but not IRQ safe.

bool try_acquire_until(chrono::SystemClock::time_point deadline)#

Tries to decrement the internal counter by 1. Blocks until the specified deadline has been reached or the counter was decremented by 1, whichever comes first.

Returns true if the internal counter was decremented successfully.

This is thread safe, but not IRQ safe.

Public Static Functions

static inline constexpr ptrdiff_t max() noexcept#: Returns the internal counter’s maximum possible value.

Safe to use in context	Thread	Interrupt	NMI
`pw::sync::CountingSemaphore::CountingSemaphore`	✔
`pw::sync::CountingSemaphore::~CountingSemaphore()`	✔
`acquire()`	✔
`try_acquire()`	✔	✔
`try_acquire_for()`	✔
`try_acquire_until()`	✔
`release()`	✔	✔
`max()`	✔	✔	✔

Examples in C++#

As an example, a counting sempahore can be useful to run periodic tasks at frequencies near or higher than the system clock tick rate in a way which lets you detect whether you ever fall behind.

#include "pw_sync/counting_semaphore.h"
#include "pw_thread/thread_core.h"

class PeriodicWorker() : public pw::thread::ThreadCore {
 // Public API invoked by a higher frequency timer interrupt.
 void TimeToExecute() {
   periodic_run_semaphore_.release();
 }

 private:
  pw::sync::CountingSemaphore periodic_run_semaphore_;

  // Thread function.
  void Run() override {
    while (true) {
      size_t behind_by_n_cycles = 0;
      periodic_run_semaphore_.acquire(); // Wait to run until it's time.
      while (periodic_run_semaphore_.try_acquire()) {
        ++behind_by_n_cycles;
      }
      if (behind_by_n_cycles > 0) {
        PW_LOG_WARNING("Not keeping up, behind by %d cycles",
                       behind_by_n_cycles);
      }
      DoPeriodicWork();
    }
  }

  void DoPeriodicWork();
}

BinarySemaphore#

BinarySemaphore is a specialization of CountingSemaphore with an arbitrary token limit of 1. Note that that max() is >= 1, meaning it may be released up to max() times but only acquired once for those N releases.

Implementations of BinarySemaphore are typically more efficient than the default implementation of CountingSemaphore.

The BinarySemaphore is initialized to being empty or having no tokens.

The entire API is thread safe, but only a subset is interrupt safe.

Note

If there is only a single consuming thread, use a ThreadNotification instead which can be much more efficient on some RTOSes such as FreeRTOS.

Supported on	Backend module
FreeRTOS	pw_sync_freertos
ThreadX	pw_sync_threadx
embOS	pw_sync_embos
STL	pw_sync_stl
Zephyr	Planned
CMSIS-RTOS API v2 & RTX5	Planned

C++#

class BinarySemaphore#

BinarySemaphore is a specialization of CountingSemaphore with an arbitrary token limit of 1. Note that that max() is >= 1, meaning it may be released up to max() times but only acquired once for those N releases. Implementations of BinarySemaphore are typically more efficient than the default implementation of CountingSemaphore. The entire API is thread safe but only a subset is IRQ safe.

WARNING: In order to support global statically constructed BinarySemaphores, the user and/or backend MUST ensure that any initialization required in your environment is done prior to the creation and/or initialization of the native synchronization primitives (e.g. kernel initialization).

The BinarySemaphore is initialized to being empty or having no tokens.

Public Functions

void release()#

Atomically increments the internal counter by 1. Any thread(s) waiting for the counter to be greater than 0, i.e. blocked in acquire, will subsequently be unblocked. This is thread and IRQ safe.

There exists an overflow risk if one releases more than max() times between acquires because many RTOS implementations internally increment the counter past one where it is only cleared when acquired.

PRECONDITION: 1 <= max() - counter

void acquire()#

Decrements the internal counter to 0 or blocks indefinitely until it can.

This is thread safe, but not IRQ safe.

bool try_acquire() noexcept#

Tries to decrement by the internal counter to 0 without blocking.

This is thread and IRQ safe.

Return values:: true – if the internal counter was reset successfully.

bool try_acquire_for(chrono::SystemClock::duration timeout)#

Tries to decrement the internal counter to 0. Blocks until the specified timeout has elapsed or the counter was decremented to 0, whichever comes first.

This is thread safe, but not IRQ safe.

Return values:: true – if the internal counter was decremented successfully.

bool try_acquire_until(chrono::SystemClock::time_point deadline)#

Tries to decrement the internal counter to 0. Blocks until the specified deadline has been reached or the counter was decremented to 0, whichever comes first.

This is thread safe, but not IRQ safe.

Return values:: true – if the internal counter was decremented successfully.

Public Static Functions

static inline constexpr ptrdiff_t max() noexcept#

Return values:: backend::kBinarySemaphoreMaxValue – the internal counter’s maximum possible value.

Safe to use in context	Thread	Interrupt	NMI
`pw::sync::BinarySemaphore::BinarySemaphore`	✔
`pw::sync::BinarySemaphore::~BinarySemaphore()`	✔
`acquire()`	✔
`try_acquire()`	✔	✔
`try_acquire_for()`	✔
`try_acquire_until()`	✔
`release()`	✔	✔
`max()`	✔	✔	✔

Examples in C++#

#include "pw_sync/binary_semaphore.h"
#include "pw_thread/thread_core.h"

class FooHandler() : public pw::thread::ThreadCore {
 // Public API invoked by other threads and/or interrupts.
 void NewFooAvailable() {
   new_foo_semaphore_.release();
 }

 private:
  pw::sync::BinarySemaphore new_foo_semaphore_;

  // Thread function.
  void Run() override {
    while (true) {
      if (new_foo_semaphore_.try_acquire_for(kNotificationTimeout)) {
        HandleFoo();
      }
      DoOtherStuff();
    }
  }

  void HandleFoo();
  void DoOtherStuff();
}

Condition Variables#

pw::sync::ConditionVariable provides a condition variable implementation that provides semantics and an API very similar to std::condition_variable in the C++ Standard Library.

Warning

Condition variables are not a good abstraction for embedded due to spurious wakeups. As a result, the only pw_sync backend provided by Pigweed that supports condition variables is pw_sync_stl. Consider using a ThreadNotification instead, as these do not cause spurious wakeups and can be used in an interrupt context.

Limitations#

As a blocking operation, condition variables should not be waited on in an interrupt context. Less intuitively, condition variables should not be notified in an interrupt context. Notifying a condition variable involves checking the corresponding condition to decide whether to resume waiting threads. This check can happen either on the signaling thread or the waiting thread:

If the signaling thread checks the condition, it needs to exclusively access the waiters and their associated conditions. Access to this list must be synchronized with calls to wait on the variable. Additional state checked by the conditions may also need to be synchronized. As a result, checking the conditions on the signaling thread may involve blocking and is not suitable for a interrupt context.
If the waiting threads check their conditions, access to the list of waiters still needs to be synchronized. Additionally, a thread may find that its condition is not satisfied, and that it needs to resume waiting. Waking threads only to resume waiting is costly in terms of both power and performance.

The second approach leads to spurious wakeups in a thread context as well. The first approach may also have spurious wakeups if the condition changes between signaling the waiter and the waiter reacquiring its lock.