RTC driver support on STM32F1 series

[ycsin]

I was trying to get the 32.768 kHz LSE work for my STM32F103RE custom board to achieve higher timing accuracy for k_msleep() (not sure if this is the correct direction), but then I found out that the STM32F1 series doesn't support the CONFIG_COUNTER_RTC_STM32 flag.

I've added:

#if defined(CONFIG_COUNTER_RTC_STM32)
#include <stm32f1xx_ll_rtc.h>
#include <stm32f1xx_ll_exti.h>
#include <stm32f1xx_ll_pwr.h>
#endif

to soc.h and relevant flags:

CONFIG_COUNTER=y
CONFIG_COUNTER_RTC_STM32=y
CONFIG_CLOCK_STM32_LSE=y
CONFIG_COUNTER_RTC_STM32_CLOCK_LSE=y

to *_defconfig. But the compilation failed due to stm32f1xx_ll_rtc.h lacks the LL_RTC_Date_Get(RTC_TypeDef *RTCx) function required by drivers\counter\counter_ll_stm32_rtc.c and a lots of other related things.

Is there a way of getting the RTC driver supported? The 32bit RTC register on the STM32F1 should be able to hold up to ~136 years. Or any possible workaround for the stm32f1xx_ll_rtc.h so that the COUNTER aspect of the LSE can work? (ie: Higher k_msleep() precision)

zephyr_compile_log.txt

[erwango]

@ycsin Just as a clarification, adding counter driver for F1 won't have any impact on k_sleep. k_sleep (and ms derivatives) are clocked by ARM Cortex Systick (cf drivers/timer/cortex_m_systick.c).

This doesn't mean it has no interest, but won't it provide the expected improvement.

I think what you rather look for is an implementation of timer api (to clock kernel ticks) using RTC. A L1 implementation is tempted here: https://github.com/zephyrproject-rtos/zephyr/pull/30790. Though, I'm not sure implementation can provide a better lock accuracy in ms, as the implementation of a ms ticker using RTC hardware is not easy (taking into account various cases expected by kernel).

[ycsin]

@ycsin Just as a clarification, adding counter driver for F1 won't have any impact on k_sleep. k_sleep (and ms derivatives) are clocked by ARM Cortex Systick (cf drivers/timer/cortex_m_systick.c).

This doesn't mean it has no interest, but won't it provide the expected improvement.

I think what you rather look for is an implementation of timer api (to clock kernel ticks) using RTC. A L1 implementation is tempted here: #30790. Though, I'm not sure implementation can provide a better lock accuracy in ms, as the implementation of a ms ticker using RTC hardware is not easy (taking into account various cases expected by kernel).

I see, thanks for the clarification!

[erwango]

@ycsin, I'll close the current point as it doesn't answer your need.