Question/bug: recent CMSIS RTOS initialisation/memory/behaviour change (possible issue with #2642)

[RobMeades]

I'm in the process of integrating a new target into mbed-os. A week ago, at #7669d7f, I had a working Pull Request that passed all the htrun tests ready to go. Because of RTX file change collisions I've had to rebase to #c1c1492 and now CMSIS RTOS doesn't even get into mbed_sdk_init(): rt_tsk_create() returns 0.

Can anyone point me at any recent changes to do with memory configuration, initialisation or RTOS behaviour before I start diffing the world and tearing my hair out?

Rob

[0xc0170]

Looking at the history, https://github.com/ARMmbed/mbed-os/commits/master/rtos , there's stack check enabled, some targets fixes, Thread class clean up.

[RobMeades]

Thanks Martin. It's definitely one of those as, if I revert the changes in that directory, my target works again. Digging...

[RobMeades]

It appears to be the Add RTX API to report stack usage commit, #2642, that's causing my problem. Investigating...

[RobMeades]

If I include the void *argv parameter in the OS_TCB struct in rt_TypeDef.h, rt_task_create() fails (even if I comment out all uses of argv) , if I add it back in then rt_task_create() returns 1 and things work. So something doesn't fit with these extra 4 bytes in the TCB.

[RobMeades]

It looks as though there might be an issue with #2642 (at least, there is for me). With the void *argv pointer added to the OS_TCB struct it is now 64 bytes in size. In RTX_CM_lib.h I see #define OS_TCB_SIZE 60, which I'm guessing defines the size of this struct (why not use sizeof()?). From the map file, with this pointer added, the struct is now 64 bytes big. With the #define left as it is rt_task_create() returns 0, if I increase the #define to 64 rt_task_create() returns 1 and the RTOS starts.

Now I'm not exactly sure what I'm doing here as this code is somewhat obtuse, and I've no idea why this isn't a problem for anyone else (possibly only those with MBED_RTOS_SINGLE_THREAD, of which I'm guessing there are some on CI?), so I'm very happy to be told that it's actually my problem but I do think I need someone else to take a look now.

Help!

[RobMeades]

I've proposed changing OS_TCB_SIZE to 64 in the above pull request. If a different fix is required, please feel free to reject the pull request.

[0xc0170]

Thanks for resolving this.