STM32L0 MCU hang after some deepsleep/wakeup cycles

[atoy40]

Description

• Type: Bug
• Priority: Major

---

Bug

Target
NUCLEO_L073RZ

Toolchain:
GCC_ARM

mbed-os sha:
eca67ca (HEAD, tag: mbed-os-5.7.2, tag: latest, origin/mbed-os-5.7)

Problem

A gpio is used as an interrupt to wakeup after a deep sleep. After some sleep/wakeup cycle, the processor hangs (it can hangs after less than ten, or a lot more, no identified "rules").

#include "mbed.h"

DigitalOut led(LED1);
InterruptIn btn(PC_13);
 
void callback(void) {
    led = 1 - led;
}

int main() {
    btn.rise(&callback);
    while(1) {
        sleep();
    }
}

Initially, I've try the exact same code with a LowPowerTicker instead of the InteruptIn. It gives the exact same behavior.

[0xc0170]

@ARMmbed/team-st-mcd please review

[LMESTM]

@atoy40 thanks for reporting the issue
@0xc0170 with new sleep architecture, is it expected that sleep() is called from main thread ? Shouldn't it be called directly from background loop when thread don't have anything to do ?
Basically, doing wait for a long time in the main instead of sleep() should result in the same sleep mode right ?

[0xc0170]

@0xc0170 with new sleep architecture, is it expected that sleep() is called from main thread ? Shouldn't it be called directly from background loop when thread don't have anything to do ?
Basically, doing wait for a long time in the main instead of sleep() should result in the same sleep mode right ?

I overlooked the sleep there ! True should not be. In case it is, it does not provide time for other threads, go to sleep immediately. I assume this should be functional though.

@atoy40 Please read https://os.mbed.com/docs/v5.7/reference/sleep-manager.html Instead of sleep, use wait function.

[atoy40]

@0xc0170 @LMESTM one more information to add : I've tested the exact same code with a "mbedlib" (rev 159 ) project, it's had the same behavior : seems dead after some deepsleep/wakeup
And stop me if I'm wrong, but if I take a look into the wait code in mbed5.7, it locks for deepsleep before calling Thread::wait. So when you ask to test with wait, you mean Thread::wait ?
Thanks.

[update] I've just read the code of the default idle_hook (https://github.com/ARMmbed/mbed-os/blob/master/rtos/TARGET_CORTEX/mbed_rtx_idle.cpp#L223) and it seems to lock deepsleep also, true ?

[LMESTM]

@atoy40 yes I think we should better use wait(100) that sleep now. wait would end up in Thread::wait indeed I think. I agree that this doesn't mean it would solve your issue - I'll have to test, but cannot do it now.

[atoy40]

@LMESTM yes, but I think Thread::wait() - and so wait() - will never put the STM into deep sleep (STM STOP mode) as the idle code lock it and so the sample at the bottom of https://os.mbed.com/docs/v5.7/reference/sleep-manager.html sound false to me as it suggests it works (or may be I misundestood something in the code :))

[0xc0170]

@0xc0170 @LMESTM one more information to add : I've tested the exact same code with a "mbedlib" (rev 159 ) project, it's had the same behavior : seems dead after some deepsleep/wakeup

Using mbed 2 then. sleep there looks fine then and you can ignore sleep manager as that one is involved in mbed OS 5.

https://os.mbed.com/docs/v5.7/reference/sleep-manager.html sound false to me as it suggests it works (or may be I misundestood something in the code :))

What is false?

[update] I've just read the code of the default idle_hook (https://github.com/ARMmbed/mbed-os/blob/master/rtos/TARGET_CORTEX/mbed_rtx_idle.cpp#L223) and it seems to lock deepsleep also, true ?

That is correct, as this is default ide look as it was (backward compatible), as RTOS OS timer is running and does not allow deepsleep (tickless changes things, however on a line above is a bug that I'll address soon).

The code in your initial report looks correct to me. @LMESTM Can you confirm that this should work for STM32L0

[atoy40]

@0xc0170 What sound false to me is the example, where :

// Deep sleep for 1 second
    printf("Deep sleep allowed: %i\r\n", sleep_manager_can_deep_sleep());
    wait(1.0);

the comment suggests the wait call will put the MCU into deepsleep, but it's not the case as it's lock into the idle hook itself.

[jeromecoutant]

Hi
I have tried the code from the description,
and it doesn't hang for me...

[jeromecoutant]

Thanks to sleep() command, target goes well in deepsleep mode.
With some wait command, as deepsleep is locked, target goes only in sleep mode.
Without any command (only a while(1);), target stays in normal mode.

I didn't try OS2 (non RTOS).

[atoy40]

@jeromecoutant can you try this one please ? :

#include "mbed.h"

DigitalOut led(LED1);
LowPowerTicker ticker;
 
void callback(void) {
    led = 1 - led;
}

int main() {
    ticker.attach(&callback, 2.0f);
    while(1) {
        sleep();
    }
}

and wait a bunch of minutes to see if it continues to blink ? In my case it stop (led on or off, it depends) to blink.

For your description on how mbed choose STOP/SLEEP mode, I agree with you, but imho, SLEEP mode cannot be consider as a deep sleep, this is why I think the doc sample can be misleading.

[0xc0170]

For your description on how mbed choose STOP/SLEEP mode, I agree with you, but imho, SLEEP mode cannot be consider as a deep sleep, this is why I think the doc sample can be misleading.

I'll fix the documentation now. Already proposed a fix for tickless deep sleep enable.

[atoy40]

@0xc0170 you mean tickless will be enable on STM based targets ? in 5.7.2 I think only some targets enabled it.

[0xc0170]

#5866 - discussing how to address this what you were asking (deepsleep locked by default)