Synchronize Application and Network Core

[idlemind_workshop]

I have created a barebones blinky example as a way to start developing for the dual cores on the nRF53, but I can't seem to synchronize them. The code on both cores is nearly identical. My main core releases the NETWORK.FORCEOFF flag and uses SysTick as the timing source (I've also tried a few other clocks). They start out in sync, but the network core gradually drifts out of phase.

Am I missing something? Could the issue be related to simultaneous memory access?

#include <stdbool.h>
#include <stdint.h>
#include "nrf.h"
#include "nrf_gpio.h"

#define APPLICATION_LED 28
#define NETWORK_LED     31

static volatile uint32_t time = 0;

void SysTick_Handler(void)
{
   time++;
}

void SysTick_Init(void)
{
    SysTick->LOAD = (64000 - 1);  // 1 ms interval (64MHz / 1000)
    SysTick->VAL = 0;             // Clear current value
    SysTick->CTRL = SysTick_CTRL_TICKINT_Msk | SysTick_CTRL_ENABLE_Msk | SysTick_CTRL_CLKSOURCE_Msk;
}

int main(void) 
{
  NRF_RESET->NETWORK.FORCEOFF = RESET_NETWORK_FORCEOFF_FORCEOFF_Release;
  SysTick_Init();

  // Assign pin to network core
  NRF_P0->PIN_CNF[NETWORK_LED] = (GPIO_PIN_CNF_MCUSEL_NetworkMCU << GPIO_PIN_CNF_MCUSEL_Pos);
  nrf_gpio_cfg_output(APPLICATION_LED);
  nrf_gpio_pin_clear(APPLICATION_LED);

  uint32_t last_time = time;
  while (1) {
    if((time - last_time) > 1000) {
      last_time = time;
      nrf_gpio_pin_toggle(APPLICATION_LED);
    }
  }
  return 0;
}

[Turbo J]

Systick is not a usable time source on NRFx MCUs, it gets turned off when the core is sleeping (unless the debugger was connected).

I can even (ab-) use that behavior to get a coarse "CPU usage %" number: Program systick to produce 100 cycles per second with clock running, then count the actual ticks you have gotten last second. But connecting the debugger always shows 100% usage...

[idlemind_workshop]

Hmm... A colleague also mentioned quirks in their power management. I've tried a few different approaches but posted this version for brevity. Instead of using SysTick_Handler, I’ve read the registers directly. I’ve also tested a timer-based example using the peripheral clock. The issue occurs both with and without a debugger attached.

[Turbo J]

Look into the PS, figure 18 "Clock control".

The network core has its own HFINT clock source - its only synced when the HFXO crystal was running. Also look at the frequency tolerances later in that chapter.

No, the HFXO is not enabled by default.

[idlemind_workshop]

Ah, that did it! Thanks for the second pair of eyes.