Fatal error unhandled interrupt when using callback with nrfx_pwm_init

[Thomg]

Hi Nordic,

Recently we've been working on using the PWM library (nrfx_pwm library) to talk with WS2812C LEDs. This works great.

Now we want to add some callbacks to signal to (our own) library that the LEDs are once again ready for more transactions, I try to do so by passing a callback and semaphore with nrfx_pwm_init like so:

`nrfx_pwm_init(&pwm1, &cfg, ws2812_driver_nrfx_cb, &sem);`

Sadly this seems to result in an unhandled interrupt after a playback, full error below:

`[18730.11h.11m.21s.904] [00:00:05.002,777] <err> os: >>> ZEPHYR FATAL ERROR 1: Unhandled interrupt on CPU 0
[18730.11h.11m.21s.904] [00:00:05.002,807] <err> os: Current thread: 0x20010e28 (unknown)
[18730.11h.11m.21s.904] [00:00:05.018,188] <err> fatal_error: Resetting system`

This seems to happen immediately after the playback has finished, hinting that the interrupt is working properly, but the handling of it is not.

I'm using Nordic SDK v1.5.0 with Zephyr.

In the prj.conf I removed the initialization of the PWM in zephyr and instead manually enable the first PWM like so:

`CONFIG_PWM=n
CONFIG_NRFX_PWM0=y`

This saves me one additional uninit call before calling the init. This does not seem to affect the hardfault behaviour.

Kind regards,

Thom

[Sigurd]

Hi,

Did you add something like this to your code?

IRQ_DIRECT_CONNECT(PWM0_IRQn, 0, nrfx_pwm_0_irq_handler, 0);

If that did not help, could you add a simple sample here that reproduces the issue ?

[Thomg]

Hi Sigurd,

This fixed it. I wonder what went wrong in the initialization.

Thanks,

Thom

[Raymond]

Hi Sigurd,  

Why is the IRQ_DIRECT_CONNECT() required here? 

[Sigurd]

Typically when you have a Zephyr driver, the driver will handle this for you, but when using nrfx driver directly, you typically need to connect the IRQ yourself.

[MegaCpp]

Thank you! this did help a lot in getting the COMP interrupts to work in nRF Connect SDK.

Can you perhaps elaborate or give an example on how the IRQs are set up in a safe way using IRQ_DIRECT_CONNECT()?

As it is suggested to "implemented in such a way that it is safe to put them directly in the vector table. For ISRs written in C, The ISR_DIRECT_DECLARE() macro will do this automatically"

Is the ISR_DIRECT_DECLARE() macro enough or or other steps necessary to stay safe?

[Sigurd]

Hi,

Might be better to use IRQ_CONNECT(). IRQ_DIRECT_CONNECT() is for performance-critical interrupt handling.

Example usage of IRQ_CONNECT() here: https://github.com/nrfconnect/sdk-zephyr/blob/main/samples/boards/nrf/nrfx/src/main.c#L39