STM32WL I2C doesn't work with LPM

[honestech74]

Describe the bug
When enabled the low power management using CONFIG_PM, STM32WL I2C doesn't work with J-Link but sometimes works with STM32CubeProgrammer using openocd (not stm32cubeprogrammer). There is the same issue #37352.
Also It's not working when using low power STOP mode manually without CONFIG_PM.

To Reproduce
Steps to reproduce the behavior:

1. Update the board.cmake (boards/arm/nucleo_wl55jc/board.cmake) for nucleo_wl55jc board to use J-Link

# SPDX-License-Identifier: Apache-2.0

board_runner_args(jlink "--device=STM32WL55JC" "--speed=4000" "--reset-after-load")
board_runner_args(stm32cubeprogrammer "--port=swd" "--reset=hw")
include(${ZEPHYR_BASE}/boards/common/jlink.board.cmake)
include(${ZEPHYR_BASE}/boards/common/openocd.board.cmake)
include(${ZEPHYR_BASE}/boards/common/stm32cubeprogrammer.board.cmake)

2. Prepare the BME280 sensor or some I2C sensor. If you will use another sensor, please build with your sensor sample.
3. Add CONFIG_PM to samples/sensor/bme280
4. Build and flash to test the I2C using BME280 sensor.

west build -b nucleo_wl55jc -p auto samples/sensor/bme280
west flash

3. Verify you are using J-Link in console logs like this

-- west flash: using runner jlink
-- runners.jlink: JLink version: 7.22b
-- runners.jlink: Flashing file: /home/honestech/zephyrproject/zephyr/build/zephyr/zephyr.hex

4. See logs after reset the board

Expected behavior
When disabled CONFIG_PM, it works well.
When enabled CONFIG_PM, it doesn't work at all. Even if the log shows the zero values, you can't see any I2C signals on I2C lines. I have checked the I2C line using the logic analyzer.

Log

*** Booting Zephyr OS build zephyr-v2.6.0-1680-gc4079e4be29a  ***
Found device "ENVIRONMENTAL_SENSOR", getting sensor data
temp: 0.000000; press: 0.000000; humidity: 0.000000
temp: 0.000000; press: 0.000000; humidity: 0.000000
temp: 0.000000; press: 0.000000; humidity: 0.000000

Environment (please complete the following information):

• OS: Ubuntu 18.04 LTS
• Toolchain: 0.12.4
• Commit SHA: c4079e4be2

[aurel32]

As a workaround, you might want to set CONFIG_I2C_STM32_INTERRUPT=n, this solution works fine on the STM32L4.

The correct solution for the STM32L4 is to prevent the SoC going to STOP2 mode where I2C can't function (well except I2C3) when waiting for the I2C interrupt. The I2C peripherals still need the HSI clock in STOP0 or STOP1 mode.

[honestech74]

@aurel32 You are right, I have removed the STOP2 mode from board dts, and then I2C works in LPM.
In fact, I am using I2C2 and it will lose the clock in STOP2 mode.

What I want is to work back after wake up from STOP2 mode. So I have called the I2C driver init(i2c_stm32_init()) after wake up, then it works well.
@erwango I think there might be some way to re-init enabled clocks or call driver init functions after wake up from suspend in power management. E.g. can we add something to enable clocks in power.c. I have no clue what's the best way.

__weak void pm_power_state_exit_post_ops(struct pm_state_info info)
{
	switch (info.state) {
	case PM_STATE_SUSPEND_TO_IDLE:
		LL_LPM_DisableSleepOnExit();
		LL_LPM_EnableSleep();
		/* need to restore the clock */
		stm32_clock_control_init(NULL);

		/* Add something */
#ifdef CONFIG_I2C_STM32
		// call i2c_stm32_init or enable i2c clocks due to dts
#endif

		break;
	case PM_STATE_SOFT_OFF:
		/* Nothing to do. */
		break;
	default:
		LOG_DBG("Unsupported power substate-id %u", info.state);
		break;
	}

	/*
	 * System is now in active mode.
	 * Reenable interrupts which were disabled
	 * when OS started idling code.
	 */
	irq_unlock(0);
}

[aurel32]

I am not sure it needs to be re-inited when exiting STOP2 mode. In my case (again STM32L4), the I2C module survives the STOP2 mode as long as there is no I2C transaction running when entering STOP2. That's why disabling CONFIG_I2C_STM32_INTERRUPT works, it makes the CPU busy wait instead of entering LPM.

Now I guess you indeed need to re-init the I2C module if the SoC entered STOP2 mode with a transaction running, which probably totally messes up its state.

[honestech74]

That's why disabling CONFIG_I2C_STM32_INTERRUPT works, it makes the CPU busy wait instead of entering LPM.

I haven't any issues with CONFIG_I2C_STM32_INTERRUPT, but agreed with it.

the I2C module survives the STOP2 mode as long as there is no I2C transaction running when entering STOP2.

I have no transaction when entering STOP2 mode. Now I have got running by configure the I2C2 again after wake up from STOP2 like this:

	const struct device *i2c_dev = DEVICE_DT_GET(DT_NODELABEL(i2c2));
	i2c_configure(i2c_dev, I2C_MODE_MASTER | I2C_SPEED_STANDARD << I2C_SPEED_SHIFT);

But it's the hard coded configuration for I2C, there is not easy way to convert the I2C clock-frequency of dts to I2C driver configuration like i2c_map_dt_bitrate in i2c-priv.h. It makes me to add some public function in I2C driver and also not easy to add this to LPM. Does anyone have any idea to configure I2C runtime using device tree configuration like i2c_configure(i2c_dev, DT_PROP(DT_NODELABEL(i2c), clock_frequency))? Or can we move the i2c_map_dt_bitrate to i2c.h?

[aurel32]

I have no transaction when entering STOP2 mode.

Do you mean you have a defined a custom power management policy? For me with the default PM residency policy, the SoC automatically enters the STOP2 mode during an I2C transaction.

Now I have got running by configure the I2C2 again after wake up from STOP2 like this:

	const struct device *i2c_dev = DEVICE_DT_GET(DT_NODELABEL(i2c2));
	i2c_configure(i2c_dev, I2C_MODE_MASTER | I2C_SPEED_STANDARD << I2C_SPEED_SHIFT);

But it's the hard coded configuration for I2C, there is not easy way to convert the I2C clock-frequency of dts to I2C driver configuration like i2c_map_dt_bitrate in i2c-priv.h. It makes me to add some public function in I2C driver and also not easy to add this to LPM. Does anyone have any idea to configure I2C runtime using device tree configuration like i2c_configure(i2c_dev, DT_PROP(DT_NODELABEL(i2c), clock_frequency))? Or can we move the i2c_map_dt_bitrate to i2c.h?

Not sure I can't really help you here, on the STM32L4 there is no need to reconfigure the I2C device when leaving the STOP2 mode. Maybe have a look at the STM32WL datasheet to have a better idea about what has to be restarted? I am afraid that i2c_configure does a lot more operations than really needed in your case.

[honestech74]

Do you mean you have a defined a custom power management policy? For me with the default PM residency policy, the SoC automatically enters the STOP2 mode during an I2C transaction.

I have to disabled LPM because there are similar issues with other peripherals like ADC. So I have entered STOP mode using STM32 LL library, not using Zephyr PM :-(. After digging the driver, I knew, should call stm32_i2c_configure_timing. I think it should be included to LPM or driver.

[UPDATE]
I got the I2C ARLO error when back to running mode from STOP2. If reconfigure the i2c timing using stm32_i2c_configure_timing, it's fixed but not sure if it's the best way.

[FRASTM]

It seems there is no i2c_stm32_pm_control function yet. Depending on the stm32 family : "the I2C must be disabled before entering Stop mode_" . On wakeUp from stop, if the HSI clock is restored, also for the I2C clock source, no specific code should be required for the I2C (registers are saved).

Note that keeping I2C clock is still possible for checking : "All U(S)ARTs, LPUARTs and I 2 Cs have the capability to enable the HSI16 oscillator even when the MCU is in Stop mode (if HSI16 is selected as the clock source for that peripheral)."