Stack overflow in tickless mode with debug profile on NUCLEO-L476RG

[janjongboom]

Description

Same as #7222? This is on Mbed OS 5.9 (949cb49) with mbed-os-example-lorawan on NUCLEO-L476RG:

++ MbedOS Error Info ++
Error Status: 0x80020125 Code: 293 Module: 2
Error Message: CMSIS-RTOS error: Stack overflow
Location: 0x8015BF9
Error Value: 0x1
Current Thread: Id: 0x200020E8 Entry: 0x8015BAD StackSize: 0x200 StackMem: 0x20002178 SP: 0x10007EC0 
-- MbedOS Error Info --

Upping the idle thread stack size to 0x400 solves the issue.

Issue request type

[ ] Question
[ ] Enhancement
[X] Bug

[0xc0170]

cc @ARMmbed/team-st-mcd

[bcostm]

This stack size is defined in rtos/TARGET_CORTEX/mbed_rtx_conf.h

/** The idle thread's stack size can be configured by the application, if not explicitly specified, it'll default to 512 */
#ifndef MBED_CONF_APP_IDLE_THREAD_STACK_SIZE
#define MBED_CONF_APP_IDLE_THREAD_STACK_SIZE 512
#endif

As this value can be changed by the application I don't know what can we do on our side ?

[jeromecoutant]

Are you expecting something like:

#ifndef MBED_CONF_APP_IDLE_THREAD_STACK_SIZE
#ifdef MBED_TICKLESS
#define MBED_CONF_APP_IDLE_THREAD_STACK_SIZE 1024
#else
#define MBED_CONF_APP_IDLE_THREAD_STACK_SIZE 512
#endif
#endif

[janjongboom]

@bcostm I know that it can be set by the application, but enabling tickless (a supported feature) mode on a NUCLEO board (a supported board) should never return in an instant hard fault. I thought this was acknowledged and fixed in #7222. If we cannot reduce stack usage in debug builds when using tickless we should increase idle thread stack size as @jeromecoutant is suggesting (I'd rather optimize this in a way that it's not required, but OK).

[jeromecoutant]

Hi

For information, I am currently testing TICKLESS feature with NUCLEO-L476RG.
Tests are OK in default mode, but often crash in Stack overflow as soon as I added some stats (MBED_CPU_STATS_ENABLED in my case)...

Maybe ARM core experts can double check ?

Thx

[jeromecoutant]

@0xc0170 @deepikabhavnani

[adbridge]

Internal Jira reference: https://jira.arm.com/browse/IOTPART-6578

[jeromecoutant]

@c1728p9

[jeromecoutant]

@janjongboom do you any internal ARM contact to start the discussion about IDLE_THREAD_STACK_SIZE,
depending or not from TICKLESS ?
Thx

[janjongboom]

@bulislaw @0xc0170 ^

[c1728p9]

I took a look at the stack overflow on debug builds with the NUCLEO-L476RG and can confirm that the background thread is overflowing. It looks like most of the extra stack usage is coming from the low power ticker wrapper which is used on the background thread when tickless is enabled. Below is a breakdown of the stack usage.

Stack function	Stack pointer	Increase		Notes
update_present_time()	0x200015e0	120		Stack start = 0x20001070
schedule_interrupt()	0x20001658	56		Stack end = 0x20001870
ticker_insert_event_us()	0x20001690	32		Used = 656
mbed::TimerEvent::insert_absolute()	0x200016b0	32
mbed::Ticker::setup()	0x200016d0	32
mbed::Ticker::attach_us()	0x200016f0	24
LowPowerTickerWrapper::_schedule_match()	0x20001708	96
LowPowerTickerWrapper::set_interrupt()	0x20001768	16
lp_ticker_wrapper_set_interrupt()	0x20001778	16
schedule_interrupt()	0x20001788	56
ticker_insert_event_us()	0x200017c0	32
mbed::TimerEvent::insert_absolute()	0x200017e0	32
rtos::internal::SysTimer::schedule_tick()	0x20001800	56
rtos::internal::SysTimer::suspend()	0x20001838	16
default_idle_hook()	0x20001848	16
rtos_idle_loop()	0x20001858	8
osRtxIdleThread()	0x20001860	16
osThreadGetId()	0x20001870

[jeromecoutant]

@janjongboom Could you review #8551 ?
Thx

[jeromecoutant]

This issue should be fixed by #8551
Thx

[0xc0170]

Please test and reopen if there is still issue