Timer thread stack size too small for Berry on BK7231N

[niteria]

Symptoms / reproduction steps

Create a demo_module.be file (same as [1]) with contents:

# simple module
# use `import demo_module`

demo_module = module("demo_module")

demo_module.foo = "bar"

demo_module.say_hello = def ()
	print("Hello Berry!")
end

return demo_module      # return the module as the output of import

Running berry import demo_module will work correctly, but addRepeatingEvent 1 1 berry import demo_module will crash the device.

Diagnosis

berry import demo_module and addRepeatingEvent 1 1 berry import demo_module run in different RTOS threads.

One is a http handler thread with stack size defined here:

OpenBK7231T_App/src/httpserver/http_tcp_server.c

Line 19 in 92ea9cd

#define HTTP_CLIENT_STACK_SIZE 8192

Another runs inside QuickTick (

OpenBK7231T_App/src/user_main.c

Line 910 in 92ea9cd

void QuickTick(void* param)

) as a timer:

OpenBK7231T_App/src/user_main.c

Line 1041 in 92ea9cd

QuickTick,

The stack size for timers on BK7231N is defined here: https://github.com/openshwprojects/OpenBK7231N/blob/2a61aa189bad9c0120eb636b00bb12cfbb03dd95/platforms/bk7231n/bk7231n_os/beken378/os/include/FreeRTOSConfig.h#L88

Changing configTIMER_TASK_STACK_DEPTH to ( ( unsigned short ) (6000 / sizeof( portSTACK_TYPE )) ) allows addRepeatingEvent 1 1 berry import demo_module to run without crashing.

Helpful notes

The way I debugged this was by querying RTOS for current thread's stack size.

I've added these lines to bool berryRun(bvm *vm, const char *prog):

UBaseType_t highWaterMark = uxTaskGetStackHighWaterMark(NULL);
ADDLOG_INFO(LOG_FEATURE_BERRY, "berryRun Current task minimum free stack: %u words\n", (unsigned int)highWaterMark);

uxTaskGetStackHighWaterMark is not available by default. I needed to add this line to platforms/bk7231n/bk7231n_os/beken378/os/include/FreeRTOSConfig.h:

#define INCLUDE_uxTaskGetStackHighWaterMark 1

In the case of berry import demo_module I got this output:

Info:BERRY:berryRun Current task minimum free stack: 1784 words

and for addRepeatingEvent 1 1 berry import demo_module I got:

Info:BERRY:berryRun Current task minimum free stack: 222 words

Is doubling enough?

During my testing I discovered that the berry code compiled (be_loadstring function) successfully, but what would run out of stack is the actual execution (be_pcall function).

The main execution loop for berry is vm_exec [2] , it tries to avoid recursion for the most common cases by dispatch() being defined as #define dispatch() goto loop.
It's hard to tell if there are examples of berry code that would blow up the C stack, but they seem unlikely to arise from normal use.
My point is that I don't think that there's a fundamental flaw in berry that would force us to keep increasing the stack size for every bigger piece of code.

[1] https://berry.readthedocs.io/en/latest/source/en/Chapter-8.html#packaging-a-module
[2] https://github.com/berry-lang/berry/blob/72ee56f037261bb263178718fdf7a55654adf1fc/src/be_vm.c#L542