STM32 PWM driver generates signal with wrong frequency on STM32G4

[martinjaeger]

Describe the bug

The following code snipped should generate a 5 kHz PWM signal with 50% duty cycle.

#define PWM_CONTROLLER "PWM_8"
#define PWM_PERIOD 200000 /* 200 us period = 5 kHz */
#define PWM_CHANNEL 1

void start_pwm()
{
    const struct device *pwm_dev = device_get_binding(PWM_CONTROLLER);
    pwm_pin_set_nsec(pwm_dev, PWM_CHANNEL, PWM_PERIOD, PWM_PERIOD / 2, 0);
}

The actual signal generated is 7.5 kHz (duty cycle of 50% is correct).

Tested MCUs: STM32G431 and STM32G474.

Environment:

• OS: Linux
• Zephyr master, SHA: 1b680c8

@gmarull I'm trying to find a fix for the issue myself, but if you already have an idea what could be wrong, please let me know.

[gmarull]

It may be related with https://github.com/zephyrproject-rtos/zephyr/blob/master/drivers/pwm/pwm_stm32.c#L109 function. Timer clock frequency calculation has some differences amongst STM32 series.

[martinjaeger]

It turns out that the reason was a too high prescaler value, which I left at the default value of 10000.

zephyr/dts/arm/st/g4/stm32g4.dtsi

Line 425 in 55df308

st,prescaler = <10000>;

So the issue was my fault and it's solved with a smaller prescaler.

@gmarull Do you know the reason why the prescaler for STM32 PWM nodes is sometimes set to 10000 and sometimes to 0? It doesn't make much sense to me... as far as I understand, 10000 does not even lead to a clean division of the system clock frequency by 10000 as the counter starts at 0 and we'd have to substract 1.

[gmarull]

@martinjaeger I don't know, I've always seen it set to 10000. @erwango do you know why?

[martinjaeger]

Yeah, 10000 is much more common, but there are also quite a few occasions of 0

$ git grep "st,prescaler = <10000>;" | wc -l
107
$ git grep "st,prescaler = <0>;" | wc -l
20

[martinjaeger]

Raised the question of correct prescaler setting also in #30740, so I'm closing this issue as the original problem is resolved.