[Core]: Add support for CMSIS RTOS 2 Thread synchronization #458
Conversation
ad3154
force-pushed
the
adrian/cmsis2-thread-sync
branch
from
af62b24
to
35cbc69
Compare
ad3154
force-pushed
the
adrian/cmsis2-thread-sync
branch
from
35cbc69
to
767458e
Compare
|
|
Marking as ready since it's now tested. It's a bit more invasive than I expected I guess, but latching onto the CMSIS abstraction for RTOSes seems very valuable, as it should give us very easy access to:
|
| #ifdef USE_CMSIS_RTOS2_THREADING | ||
| static void receive_thread_function(void *parent); | ||
| #else | ||
| void receive_thread_function(); | ||
| #endif |
There was a problem hiding this comment.
Hmm, since we'll be limited by C now I'd suggest we just refactor the old function instead of duplicating it. So we use the C void * pattern again which we previously refactored away from I believe 😆
| #if defined USE_CMSIS_RTOS2_THREADING | ||
| static void update_thread_function(void *); | ||
| #else | ||
| static void update_thread_function(); | ||
| #endif |
| if (!target->receive_can_frame()) | ||
| { | ||
| // There was no frame to receive, and osThreadYield may not exist | ||
| osDelay(1); |
There was a problem hiding this comment.
nice-to-have (non-blocking): I guess we can make generalize the thread "delay / sleep / yield" functions in the future
| // or reduce its usage by statically allocating your thread's stack(s) | ||
| } | ||
| } | ||
| return nullptr != handle; |
There was a problem hiding this comment.
nit: handle can never be nullptr here, not sure if this is here for future plans otherwise might be better to change the function to like return void and name it prepare_handle?
| nullptr, | ||
| 0 | ||
| }; | ||
| handle = osMutexNew(&attributes); |
| namespace std | ||
| { | ||
| using mutex = isobus::Mutex; | ||
| using recursive_mutex = isobus::RecursiveMutex; | ||
| using condition_variable = isobus::ConditionVariable; | ||
| } // namespace std |
There was a problem hiding this comment.
nit(non-blocking): I don't really like that we add parts to the std namespace here. If we want to keep expanding to more platforms that don't have the mutex / thread / condition_variable header file implemented, then it might be worth it to do more investigation there and see how to best structure the library such that we do it nice and organized (with proper documentation hehe 😄)
| { | ||
| /// @brief A wrapper around a CMSIS RTOS 2 mutex. | ||
| /// @details See definition at https://www.keil.com/pack/doc/CMSIS/RTOS2/html/group__CMSIS__RTOS.html | ||
| class Mutex |
There was a problem hiding this comment.
thought: Mutex is not a template class, so we should separate the implementation and definition?
| /// @brief A template class for a lock free queue. | ||
| /// @tparam T The item type for the queue. | ||
| template<typename T> | ||
| class LockFreeQueue |
There was a problem hiding this comment.
nit: LockFreeQueue is already in the file below. I think it makes sense if we just use a separate #ifdef directive there that covers both cases
| while (true) | ||
| { | ||
| // If your code is stuck in here, that means you did something | ||
| // very wrong, like recursively locked this mutex, or tried to | ||
| // lock the mutex before the OS was initialized, or called this in | ||
| // an interrupt service routine. | ||
| // osRetVal may contain more information. | ||
| } |
There was a problem hiding this comment.
I like the way you used these for loops. It's a clever trick to help with finding where things go wrong with debugging. Though I wonder if it's always the way to go in production? I'm not sure
Describe your changes
This is an attempt to provide support for embedded RTOSes that use the CMSIS RTOS 2 abstraction, such as FreeRTOS, Keil RTX, and others.
My thinking is that this is 50 percent of the battle to supporting STM32 and other small ARM devices, with the other half being hardware plugins with specific HAL integrations for each micro.
In order to actually consume this, you'd need to also implement a proper hardware time-base for the stack using a hardware timer on those devices, which maybe we could also provide as part of a CAN plugin eventually. Additionally, you'd also need to start the RTOS before using the stack, which I think is acceptable (osKernelInitialize, and osKernelStart).
Either way, the point is, CMSIS RTOS2 is a common RTOS abstraction which would be nice to support.
The CMSIS RTOS 2 API is documented here
I added some super obvious infinite loops to catch issues between us and the kernel. I used loops because assert may not be supported on the target platforms, and since this file is included in the logger, I can't emit any log output here.
How has this been tested?
This is a draft until I can test it. I have a cheap STM32L4 eval board coming to help test it. Will update once I have more to share.Tested on STM32L476 Nucleo dev kit with an external CAN transceiver running FreeRTOS and CMSIS 6.0 via the CMSIS_RTOS2 api.
The dev kit loaded the VT example and worked normally, even using > 1 thread as expected when used with the stm32l4 plugin which is not yet merged in,.