Recent LP Ticker Changes Break Tickless

[mattbrown015]

Description

I've just tried to update to today's master and it has broken my software. :-(

I'm building for STM32L4 and using tickless in RTC mode.

I suspect the problem is something to do with recent LP ticker changes.

My software runs, does its initialisation and then just appears to stop. I think it is stuck in a ticker handler.

My current theory is the problem comes from the changes made by this PR, Fixes for tickless and LPTICKER_DELAY_TICKS #7524 and, in particular, this commit Update low power ticker wrapper.

Sorry about being vague but this is a big problem for me and I wanted to get an issue created. I shall continue to investigate and try and add more detail.

Issue request type

[ ] Question
[ ] Enhancement
[X] Bug

[0xc0170]

My current theory is the problem comes from the changes made by this PR, Fixes for tickless and LPTICKER_DELAY_TICKS #7524 and, in particular, this commit Fixes for tickless and LPTICKER_DELAY_TICKS #7524.

If you revert that commit or PR, works as expected?

Sorry about being vague but this is a big problem for me and I wanted to get an issue created. I shall continue to investigate and try and add more detail.

👍
[Mirrored to Jira]

[0xc0170]

cc @c1728p9
[Mirrored to Jira]

[mattbrown015]

If you revert that commit or PR, works as expected?

I believe so but it is all a bit complicated and I went around a few circles. I'm currently trying to retrace my steps methodically in order to be sure.
[Mirrored to Jira]

[mattbrown015]

I deployed https://github.com/ARMmbed/mbed-os/#7c22dca302e76d9fa1cb41efff31925187a4b876, LoRaWAN: Restoring default RX2 data rate #7831, and built and tested my software without any problems.

Then I deployed https://github.com/ARMmbed/mbed-os/#e02466a77a6a7db6e043b6b87cefe8aa6088f7ec, Fixes for tickless and LPTICKER_DELAY_TICKS #7524, and did a complete rebuild. My software hangs.

Above, I pointed my finger at a particular commit within Fixes for tickless and LPTICKER_DELAY_TICKS #7524 but I haven't double-checked this yet.

I have a further suspicion that the hang occurs when Mbed OS deepsleeps for the first time. I.e. all of the initialisation gets done and then it hangs.
[Mirrored to Jira]

[c1728p9]

Hi @mattbrown015 can you provide steps or an example project I can use to reproduce this failure? Also what board variant of the STM32L4 are you using?
[Mirrored to Jira]

[mattbrown015]

Hi @c1728p9,

I think the most important nugget of info is that I'm using the RTC for my lp ticker i.e. "target.lpticker_lptim": 0. Obviously this isn't the default and hence feels like a likely source of problems.

We've created a custom target which is inherited from NUCLEO_L432KC. The MCU is actually a STM32L442KC but I'm pretty sure the precise MCU isn't important.

I don't have a cut down example for you yet but here's what I've just done:

1. Deploy 7c22dca and check everything is OK.
2. git cherry-pick adc64cccacfd93f04f17dd2f01f278c020b8831d i.e. 'Update low power ticker wrapper'
3. Rebuild and watch it hang
4. Select LPTIM for the lp ticker i.e. change to "target.lpticker_lptim": 1
5. This did not appear to make any difference
6. Power cycle and now it works

So there was some inconsistent behaviour which is making it difficult for me to be certain of anything. I.e. why did it not work until I turned it off and on again? Once possible explanation is that flashing and resetting doesn't actually reset the RTC.

I shall keep investigating for another 30 minutes and then I'm off home! :-)
[Mirrored to Jira]

[mattbrown015]

I didn't put STM32L4 in the title because I don't know that only STM32L4s are affected, but it is quite possibly a STM32L4 problem. Hence I wonder if @jeromecoutant has any comment?

I notice that @jeromecoutant has made some RTC changes since adc64cc but I tried being right up to date and that doesn't work either.

I've got meetings this morning but my plan is to get my NUCLEO-L433RC-P and try mbed-os-example-blinky in tickless mode using the RTC as the lp ticker.
[Mirrored to Jira]

[0xc0170]

cc @ARMmbed/team-st-mcd
[Mirrored to Jira]

[mattbrown015]

Oh, LPTICKER_DELAY_TICKS!

I added "target.lpticker_delay_ticks": 0 to mbed_app.json and my software no longer hangs. :-)

Is this the correct thing to do?

There's plenty of discussion about LPTICKER_DELAY_TICKS in Fixes for tickless and LPTICKER_DELAY_TICKS #7524 but I'm afraid I don't completely understand.
[Mirrored to Jira]

[0xc0170]

@mattbrown015 What's your target ? What is the current value now?
[Mirrored to Jira]

[mattbrown015]

My target is a custom target inherited from NUCLEO_L432KC.

The default value of LPTICKER_DELAY_TICKS on the NUCLEO_L432KC is 3, in common with most of the other STM32 devices.

But, I am also changing MBED_CONF_TARGET_LPTICKER_LPTIM to 0; its default is 1.

So I think my question is...

What is the correct LPTICKER_DELAY_TICKS value when using the RTC as the lp ticker on STM32?
[Mirrored to Jira]

[c1728p9]

Hi @mattbrown015 I think I was able to reproduce your issue on a NUCLEO_L432KC. By adding the macro MBED_TICKLESS and setting lpticker_lptim to 0 I was able to get the program below to lockup randomly. (Note - I built with the default profile and the ARM toolchain) Is this similar to the failure you were seeing?

To investigate this further I instrumented the ticker wrapper and looked at the signals with a logic analyzer. I found when the program locked up it was actually stuck in sleep mode. I couldn't find anything wrong with the ticker wrapper layer - the 3 cycle delay between back-to-back calls to attach_us was being met. The last ticker related call before entering sleep mode was to lp_ticker_set_interrupt. This makes it seem like this function failed to fire the match interrupt. @jeromecoutant @LMESTM do you have any thoughts on this or what could be going wrong?

Program which reproduces the failure:

#include "mbed.h"

int main() {
    uint32_t count = 0;
    while (true) {
        led1 = !led1;
        wait(0.1);
        count++;
        printf("Count: %i\r\n", count);
    }
}

[Mirrored to Jira]