Thread Safety in mbed OS 5
About this document
One of the major improvements introduced in mbed OS 5 is a new programming model based on a real time operating system (RTOS). Some earlier versions of mbed have had optional support for an RTOS, but with version 5 we are making this a standard feature of the platform, so that developers can take advantage of a more flexible programming model based on multiple threads.
As with any multi-threaded environment, mbed developers need to use various synchronization primitives to ensure their code doesn’t include race conditions or other concurrency problems. They also need to understand what thread-safety guarantees are provided by the mbed OS 5 APIs when they use them. This is particularly important for code that runs in response to a hardware interrupt service routine (ISR), which needs to be carefully designed so as not to compromise the thread safety of the whole system.
The mbed OS library contains internal synchronization to provide various levels of thread safety. This document describes the mechanisms provided by mbed OS 5 to build thread safe applications.
Different components within mbed OS 5 provide different levels of synchronization:
- Interrupt safe - safe for use from multiple threads and interrupts; operation is done atomically or in a critical section. The behavior is well defined when used from both interrupts and threads.
- Thread safe - safe for use from multiple threads; operation is protected by an RTOS primitive and can be used from multiple threads, but will cause problems if used from an interrupt service routine.
- Not protected - Operation does not protect against concurrent access and needs to be synchronized externally. If you call from multiple threads without some other form of synchronization, data can become corrupted and behavior is undefined.
Tip: The API reference indicates the level of synchronization of each function.
All supported toolchains are thread safe when using the full version of their standard library:
- Multi-threading support:
- GCC Newlib - GCC toolchain
- IAR Standard Library - IAR toolchain
- ARMCC Standard Library - ARM toolchain
- Single thread support only:
- Newlib Nano - GCC_ARM toolchain
- Micro ARMCC - uARM toolchain
Note: GCC and ARMCC provide smaller variants of their libraries. These smaller versions are not thread safe and projects using them should always use only one thread.
Most drivers are thread safe. Some notable exceptions are listed below; check the relevant handbook page or doxygen for more specific details on the driver being used.
Drivers that are interrupt safe:
- DigitialIn, DigitalOut, DigitalInOut
- PortIn, PortOut, PortInOut
- Ticker, TimerEvent, Timeout
Drivers that are not protected:
HAL C API
The HAL C API is the porting layer of mbed OS 5 and is not thread safe. Developers should not typically use this API directly, instead using the higher-level drivers and libraries. If you program directly to the HAL C API it is your responsibility to synchronize operations with an appropriate mechanism, such as a mutex.
The mbed RTOS C++ API provides the
mutex class, which is one of the simplest primitives used to make code thread safe. Using a mutex is the recommended way of protecting an object’s data, and most of the common mbed APIs are made thread safe through the use of mutexes.
However, there are times when a simple mutex is not an appropriate mechanism for achieving thread safety. In particular, mutexes should not be used within interrupt service routines (ISRs). The correct way to work with ISRs is not to let them do any real processing when responding to an interrupt. Instead, interrupts should send an event or message to a thread; the interrupt then completes, the thread is rescheduled and, subsequently, it is the thread that will perform the processing. This allows the thread to acquire the mutex safely when it needs it for processing.
The RTOS provides several mechanisms to move interrupt processing onto a thread. These include, but are not limited to:
Warning: In mbed OS 5, if you attempt to use a mutex from within an interrupt nothing happens; attempts to lock a mutex will succeed immediately, regardless of whether the lock is actually free. In other words, if you acquire a mutex lock in an interrupt, you can break the thread safety mechanisms and introduce race-conditions into an otherwise safe piece of code. Future versions of mbed OS will provide warnings and ultimately prevent this from happening.
For more information see rtos/rtos/Mutex.h.
mbed OS provides atomic functions to make code interrupt safe. If you must modify an object or data structure from interrupts, you can use these atomic functions to synchronize the access.
For more information see hal/api/critical.h.
Critical sections disable interrupts to provide uninterrupted access to a resource, so you can use critical sections to make code interrupt safe. However, you should avoid using critical sections if you can, because they must execute quickly, or they will cause system instability.
- Do not perform time consuming operations inside critical sections. This will negatively affect the timing of the entire system, because all interrupts are disabled during critical sections.
- Do not invoke any standard lib or RTOS functions within a critical section; it could result in a hard fault because RTOS performs SVC calls.
For more information see hal/api/critical.h.
Major mbed OS libraries
- Network Socket API - Thread safe: public calls are protected by a mutex.
- Nanostack - Not protected: in general, we recommend that mbed developers use our 6LoWPAN stack through the Network Socket API. If you wish to use Nanostack directly, you need to be aware of how it uses threads. The core of Nanostack runs on a tasklet mechanism, scheduled on a single underlying OS thread. Developers who wish to call directly to Nanostack must create their own tasklet and make Nanostack calls from there. Since there is only one OS thread servicing all tasklets, there is no need for further synchronization between tasklets see the 6LoWPAN documentation.
- mbed-tls - Not protected: function calls are safe from any thread as long as the objects they operate on are properly protected - see the documentation.
- mbed client - Thread safe: public calls are protected by a mutex.
- BLE - Not protected: the expected use case for BLE is to run on one thread and serialize all events to that thread. This is analogous to the way Nanostack uses a thread. We provide a number of examples that showcase how to use BLE in a thread-safe way, including the Eddystone Service found here.
Additional considerations when writing thread-safe code
Synchronization will not provide protection when deleting an object; when the object is deleted, the mutex protecting it is deleted as well. This mean that when an object is deleted, you must make sure that all other threads are done using it. .
You need to synchronize access to hardware from two objects using the same pins. For example, if two instances of Serial are created on the same pins, writing to both of the objects at the same time could cause instabilities.
Considerations when porting
Porting new platforms to mbed OS 5 is nearly the same as it is in mbed 2. In general, no synchronization mechanisms are needed in drivers that operate below the C HAL layer, since this is already provided at a higher level. The only exceptions to this are the functions
gpio_write, which are expected to use processor-specific
clear registers rather than performing a read-modify-write sequence. An example of this can be found here.
For more information, see the CMSIS-RTOS tutorial.