How to create C++ task class (a worker) with Keil's CMSIS RTOS2 thread and timer C funtions?

Question

How do I map C++ class concept to C functions osTimerNew() and osThreadNew() ?

How to use a C++ member function as a Keil RTOS2 osTimerNew() and osThreadNew() callback implementation.

Thanks

Answer 1

From an object oriented point of, I would suggest that you are addressing this is the wrong way. There as no direct relationship between a thread and a timer that indicates that they should be a in a single "class" and it is not a matter of mechanistically "mapping" functions to classes. Rather you need to identify the classes - i.e. the things you want to instantiate objects of, and then define the interfaces - the methods that define the functions and capabilities of those objects.

To that end, I would suggest that a thread (or task) and a timer are separate classes. You might create a higher level class of a periodic task that might then by composed and/or derived from these other classes. For example:

or

Let us consider the cTask class to start with. It would be wrong (or at least pointless) to simply wrap the osThreadNew() function in a class wrapper; rather you need to think a a task as a class and consider all the things that class may do. To that end, the CMSIS RTOS reference provides some inspiration in the organisation of its documentation. It has a section on Thread Management and Thread Flags that can be use to design the cTask interface.

A simple task class might have the following interface foir example:

    class cTask
    {
        public:
            typedef uint32_t tEventFlags ;  
            
            cTask();
            virtual ~cTask();

            eOSStatus spawn( const char* taskname, 
                             int taskpriority = DEFAULT_PRIORITY, 
                             int stack_size = DEFAULT_STACK, void* stack_ptr = 0 );

            void setEvent( tEventFlags flags ) const ;
            static void delay(int period);
            static void lock();
            static void unlock();

            int getPriority() const ;
            int setPriority(int new_priority);

    private :
            virtual void threadMain() = 0 ;
            tEventFlags eventWait( tEventFlags flags, int timeout ) ;

            static void entry_point( void* arg )
            { 
                cTask* instance = reinterpret_cast<cTask*>(argv) ;
                instance->threadMain() ;
            }
} ;

And you might then have a task:

class cMyThread : cTask()
{
    public :
        cMyThread()
        {
            spawn( "mythread" ) ;
        }

        void someEvent()
        {
            setEvent( 0x0001 ) ;
        }

        void someOtherEvent()
        {
            setEvent( 0x0002 ) ;
        }

    private: 
    
        void threadMain()
        {
            for(;;)
            {
                tEventFlags event eventWait( 0x0003, WAIT_FOREVER ) ;
                if( (event & 0x0001) != 0  )
                {
                    // process some event
                } 

                if( (event & 0x0002) != 0  )
                {
                    // process some other event
                } 
            }
        }
} ;

Such that you might instantiate and communicate with instance od cMyThread thus:

    cMyThread thread_a ;
    cMyThread thread_b ;

    thread_a.someEvent() ;
    thread_b.someOtherEvent() ;

Obviously the interface could be much more extensive, and you would want to add classes for semaphores, mutexes, message queues as well as timers.

The above is illustrative only; as you can see there is a lot of work perhaps to be done, but to answer your question osThreadNew()would be used here to implementcTask::spawn()and would start thecTask::threadMain()via the staticentry_point()function by passing it thethis` pointer.

You would take a similar approach to the cTimer class with respec to defining teh interface in terms of things a timer can do. such as start, cancel, wait, set event handler etc.

It is not necessary to slavishly provide an interface to every CMSIS RTOS function; the C++ layer provides an opportunity to abstract some of that detail into something easier to use and easier to port to some other RTOS API.

Answer 2

You give "this" to osTimerNew()/osThreadNew() in place of a "void * argument" parameter.

struct task
{
    task ( uint32_t timer_period_ms )
    {
        // ===  casting (needs must)
        using fp    = void ( task::* ) ( void * );
        using os_fp = void ( * )       ( void * );
        auto cast =
        [] ( fp in )
        {
            union {
                fp      in;
                os_fp   out;
            } u { in };

            return u.out;
        };


        auto timer_id = osTimerNew
        (
            cast ( &task::rtos_timer_callback  ),
            osTimerPeriodic,
            this,    // let RTOS know about the object
            nullptr
        );
 
        m_id_thread = osThreadNew
        (
            cast ( &task::rtos_thread_callBack ),
            this,    // let RTOS know about the object
            nullptr
        );

        osTimerStart ( timer_id, timer_period_ms );
    }
    
    virtual ~task() = default;

    virtual void do_work () = 0;

private:
    void rtos_timer_callback ( void * pvArg )
    {
        osThreadFlagsSet ( m_id_thread, 0x01 );
    }
    
    __NO_RETURN void rtos_thread_callBack ( void * pvArg )
    {
        while (1)
        {
            osThreadFlagsWait ( 0x01, osFlagsWaitAny, osWaitForever );
            do_work ();
        }
    }

private:
    osThreadId_t m_id_thread {};
};

Now use the task class:

struct d_task_0 : public task
{
    d_task_0 ( uint32_t timer_period_ms ) : task { timer_period_ms } {}
    void do_work () final
    {
        // called every d_task_0::timer_period_ms
    }
};

and create another task:

struct d_task_1 : public task
{
    d_task_1 ( uint32_t timer_period_ms ) : task { timer_period_ms } {}
    void do_work () final
    {
        // called every d_task_1::timer_period_ms
    }
};

And finally create workers:

d_task_0  worker0 { 500 }; // d_task_0::do_work () called every 500ms
d_task_1  worker1 { 800 }; // d_task_1::do_work () called every 800ms

RTOS2 documentation:

https://www.keil.com/pack/doc/CMSIS/RTOS2/html/group__CMSIS__RTOS__ThreadMgmt.html

https://www.keil.com/pack/doc/CMSIS/RTOS2/html/group__CMSIS__RTOS__TimerMgmt.html

and implementation:

https://github.com/ARM-software/CMSIS_5/tree/develop/CMSIS/RTOS2

My toolchain: Keil MDK-ARM Plus 5.33.0.0; ArmClang/Link v6.15

Casting solution came from here: Casting between void * and a pointer to member function

Another way of casting is:

using os_fp                       = void ( * ) ( void * );
void ( task::*pTimer ) ( void * ) = &task::rtos_timer_callback;
void * task_timer                 = ( void*& ) pTimer;

auto timer_id = osTimerNew
(
    reinterpret_cast<os_fp>(task_timer),
    osTimerPeriodic,
    this,    // let RTOS know about the object
    nullptr
);

Source: Get memory address of member function?