nRF52 GPIOTE interrupt latency

Question

Hi

I have a problem with long latency on interrupts and long delay on retriggered interrupt.

I have a nRF52832 running with external HFCLK source. Using the SDK 14.2.0. The softdevice is disabled before using the GOIOTE and re-enabled afterwards.

The GPIOTE is setup like this:

Fullscreen

1
2
3
4
5
6
7
8
    if (!nrf_drv_gpiote_is_init())
    {
            err_code = nrf_drv_gpiote_init();
            APP_ERROR_CHECK(err_code);
    }   
    nrf_drv_gpiote_in_config_t config = GPIOTE_CONFIG_IN_SENSE_HITOLO(true);
    nrf_drv_gpiote_in_init(RFID_PIN_DOUT, &config, gpiote_void_handler); 
    nrf_drv_gpiote_in_event_enable(RFID_PIN_DOUT, true);
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX

	if (!nrf_drv_gpiote_is_init())
	{
			err_code = nrf_drv_gpiote_init();
			APP_ERROR_CHECK(err_code);
	}	
	nrf_drv_gpiote_in_config_t config = GPIOTE_CONFIG_IN_SENSE_HITOLO(true);
	nrf_drv_gpiote_in_init(RFID_PIN_DOUT, &config, gpiote_void_handler); 
	nrf_drv_gpiote_in_event_enable(RFID_PIN_DOUT, true);

As you can see I have skipped the original interrupt service routine from the SDK and wrote my own to avoid the overhead.

My interrupt service routine is like this:

Fullscreen

1
2
3
4
5
6
7
8
9
10
11
12
13
void GPIOTE_IRQHandler(void) // DOUT call back handler
{
// Clear the GPIO event
    NRF_GPIOTE->EVENTS_IN[0] = 0;   
    
// Mark start of interrrupt 
    NRF_P0->OUTSET = 1 << RFID_PIN_TEST1;
/* Interrupt code here. Takes around 2µs */
// Mark end of interrrupt   
    NRF_P0->OUTCLR = 1 << RFID_PIN_TEST1;
}
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX

void GPIOTE_IRQHandler(void) // DOUT call back handler
{
// Clear the GPIO event
	NRF_GPIOTE->EVENTS_IN[0] = 0;	
	
// Mark start of interrrupt	
	NRF_P0->OUTSET = 1 << RFID_PIN_TEST1;

/* Interrupt code here. Takes around 2µs */

// Mark end of interrrupt	
	NRF_P0->OUTCLR = 1 << RFID_PIN_TEST1;
}

I got a latency of 1.6 µs from the falling edge on GPIO pin RFID_PIN_DOUT to the rising edge of RFID_PIN_TEST1.

An if an additional falling edge occurs before the interrupt service routine, there is a delay between the falling edge of RFID_PIN_TEST1 to the next rising edge also on 1.6 µs.

I had expected much lower latency on a 64MHz processor. It should take around 16 clock cycles to enter or leave the interrupt service routine, which is 0.25µs.

The big question: What does the MCU do en the rest of the time?