EFM32: Fixed pwmout_all_inactive being inversed

[seppestas]

Description

Fixes a bug causing the pwmout_all_inactive function to return the inverse of the actual state. If one of the CC channel pins is enabled it means a channel is active so it should return true.

The bug seems to have been introduced in mbed 3 (a81fdc4). Before pwmout_all_inactive just checked if the route was enabled (which doesn't really work...)

This PR also contains some cleanup in pwm_init.

Status

READY

Steps to test or reproduce

Note: I only tested this on the EFM32 Happy Gecko MCU (EFM32HG_STK3400 target).

Since pwm_free is not called in ~PWMOut (see #3106), this bug currently does not cause any problems. However, when enabling this functionality, e.g by adding the folowing destructor to PWMOut:

~PwmOut() {
        core_util_critical_section_enter();
        pwmout_free(&_pwm);
        core_util_critical_section_exit();
}

this bug causes the PWM timer to be disabled when a PWMOut object is destroyed, even if other PWMOut objects are still being used. This causes the other PWMOut object to end up in an invalid state (I'm not sure why).

An excerpt of the code I used to test this:

#include "mbed.h"

InterruptIn activate_button(PC10);
AnalogIn in(PD6);
PwmOut *motor_forward;
PwmOut *motor_reverse;
volatile bool reverse = false;

static void activate_button_pressed() {
    reverse = ! reverse;
}

int main() {
    activate_button.rise(activate_button_pressed);

    while(true) {
        while (activate_button == 1) {
            if (motor_forward == NULL) {
                motor_forward = new PwmOut(PA0);
                motor_forward->period_us(25);
            }
            if (motor_reverse == NULL) {
                motor_reverse = new PwmOut(PA1);
                motor_reverse->period_us(25);
            }
            if (reverse) {
                motor_reverse->write(in.read());
                motor_forward->write(0);
            } else {
                motor_forward->write(in.read());
                motor_reverse->write(0);
            }
        }
        motor_reverse->write(0);
        motor_forward->write(0);
        if (motor_forward != NULL) {
            delete motor_forward;
            motor_forward = NULL;
        }
        if (motor_reverse != NULL) {
            delete motor_reverse;
            motor_reverse = NULL;
        }
        sleep();
    }
}

Without the fix of this PR, motor_reverse stops working after it gets destructed (even when it is initialized again).

[0xc0170]

cc @stevew817 @akselsm

[stevew817]

Thanks @seppestas !

[0xc0170]

/morph test

[seppestas]

How should I write tests for this? As far as I can tell there are no host-runnable / local unit test for the mbed SDK, only Greentea tests... Is there any intent to change this? If so, I see both Unity and CppUTest being used in other parts of the project. Is there any preference (maybe I should add googletest, cgreen or mutest XD).

[bridadan]

/morph test

@seppestas in case there was confusion, this is us triggering a CI run with a command, not telling you to run/write tests (though @0xc0170 may still ask you to do so, he just won't call you "morph" 😄 )

[mbed-bot]

Result: FAILURE

Your command has finished executing! Here's what you wrote!

/morph test

Output

mbed Build Number: 1868

Build failed!

[seppestas]

@bridadan I go that XD.

It's weird the CI is failing for the UBLOX C030 target, pretty sure that wasn't me 😁

[bridadan]

@seppestas Nope, wasn't you 😄 We had a breakage on master yesterday but this has since been fixed.

/morph test

[mbed-bot]

Result: SUCCESS

Your command has finished executing! Here's what you wrote!

/morph test

Output

mbed Build Number: 1884

All builds and test passed!