Softdevice assert with TWI

[ftjandra]

On SDK16.0.0 + Mesh SDK 4.1.0

I am getting random softdevice asserts when running our application with TWI accessing a temperature/humidity sensor (HTU21D/HPP845). Sometimes within a few minutes and sometimes it takes hours.

<t:   16498108>, app_error_weak.c,   96, Softdevice assert: 154516:0

I tried both blocking and non-blocking with a handler. I set .interrupt_priority = APP_IRQ_PRIORITY_LOW.

static void twi_handler(nrf_drv_twi_evt_t const * p_event, void * p_context) {
    //Not really being used
}

static void twi_init(void) {
    __LOG(LOG_SRC_APP, LOG_LEVEL_INFO, "twi_init()\n");
    ret_code_t err_code;

    const nrf_drv_twi_config_t twi_config = {
       .scl                = I2C_SCL_PIN,
       .sda                = I2C_SDA_PIN,
       .frequency          = NRF_DRV_TWI_FREQ_400K,
       .interrupt_priority = APP_IRQ_PRIORITY_LOW,
       .clear_bus_init     = false
    };

    err_code = nrf_drv_twi_init(&m_twi, &twi_config, twi_handler, NULL); //event handler, non-blocking mode
    //err_code = nrf_drv_twi_init(&m_twi, &twi_config, NULL, NULL); //no event handler, blocking mode
    ERROR_CHECK(err_code);

    nrf_drv_twi_enable(&m_twi);
}

static void read_hpp845(void) {
    ret_code_t ret_code = 0;
    uint8_t command_address;
    uint16_t rawHumidity;
    uint16_t rawTemperature;
    static uint8_t hpp845_stage = 0;

    switch(hpp845_stage) {
        case 0: //start_hpp845_humidity
            command_address = HPP845_START_HUMIDITY_ADDR;
            ret_code = nrf_drv_twi_tx(&m_twi, HPP845_ADDR, &command_address, 1, false);
            break;
        case 1: //fetch_hpp845_humidity
            memset(i2cBuffer_RX, 0, sizeof(i2cBuffer_RX));
            ret_code = nrf_drv_twi_rx(&m_twi, HPP845_ADDR, i2cBuffer_RX, 3); //Get raw humidity data   
            break;
        case 2: //continue fetch_hpp845_humidity
            rawHumidity = ((unsigned int)i2cBuffer_RX[0] << 8) | ((unsigned int)i2cBuffer_RX[1]);
            float tempRH = rawHumidity / (float)65536;
            float rh = -6 + (125 * tempRH);

            if(rh < 0.0) {
                rh = 0.0;
            }
            else if(rh > 100.0) {
                rh = 100.0;
            }

            hpp845_humidity = (uint8_t)rh;

            __LOG(LOG_SRC_APP, LOG_LEVEL_INFO, "HPP845 Humidity: %u\n", hpp845_humidity);
            break;
        case 3: //start_hpp845_temperature
            command_address = HPP845_START_TEMPERATURE_ADDR;
            ret_code = nrf_drv_twi_tx(&m_twi, HPP845_ADDR, &command_address, 1, false);
            break;
        case 4: //fetch_hpp845_temperature
            memset(i2cBuffer_RX, 0, sizeof(i2cBuffer_RX));
            ret_code = nrf_drv_twi_rx(&m_twi, HPP845_ADDR, i2cBuffer_RX, 3); //Get raw temperature data
            break;
        case 5: //continue fetch_hpp845_temperature
            rawTemperature = ((unsigned int)i2cBuffer_RX[0] << 8) | ((unsigned int)i2cBuffer_RX[1]);
            float tempTemperature =  (float)rawTemperature / (float)65536;
            float temperatureC = (float)-46.85 + ((float)175.72 *(float)tempTemperature);

            //set range limits
            if(temperatureC < -40.0) {
                temperatureC = -40.0;
            }
            else if(temperatureC > 125.0) {
                temperatureC = 125.0;
            }

            hpp845_temperature = (uint16_t)((temperatureC*3.0) + 120.0); //in one-third degC and offset by 120 to work in uint16_t

            __LOG(LOG_SRC_APP, LOG_LEVEL_INFO, "HPP845 Temperature: %u in one-third degC (offset +120)\n", hpp845_temperature);
            break;
        default:
            __LOG(LOG_SRC_APP, LOG_LEVEL_INFO, "ERROR: hpp845_stage out-of-range: %u\n", hpp845_stage);
            hpp845_stage = 0; //reset, should never get here
            break;
    }

    if((ret_code == 0) && (hpp845_stage != 5)) {
        hpp845_stage++;
    }
    else if(ret_code != 0) {
        __LOG(LOG_SRC_APP, LOG_LEVEL_INFO, "ERROR: hpp845_stage: %u, ret_code: %u\n", hpp845_stage, ret_code);
        hpp845_stage = 0; //reset, something went wrong
    }
    else {
        hpp845_stage = 0; //reset, need to cycle back to start
    }
}

I am calling read_hpp845() from an app_timer and I am only running one TWI instruction per app_timer cycle. If I comment out the call, the application works fine. The app_timer is set at 100ms and so I am not even using the twi event handler to check that it was successful, for now I am just assuming that it was.

How do I figure out what is causing the assert?

Thanks.

[tesc]

Hi,

Since it is a SoftDevice assert it sounds like wrong settings for interrupt priorities, or otherwise too much time used in high priority interrupt context blocking the mesh stack from releasing resources back to the SoftDevice.

Please double check Setting interrupt priority levels, and be careful with running interrupts at a higher priority than that of the mesh stack. The schedule handling library is an option for moving interrupt triggered code to main context.

Regards,
Terje

[ftjandra]

I changed the interrupt priority of the TWI to lowest and that seemed to help a little, but not completely.

So now I am trying to implement the scheduler. I followed the tutorial here: https://devzone.nordicsemi.com/nordic/short-range-guides/b/software-development-kit/posts/scheduler-tutorial

But now I get 'app_error_weak.c,  115, Mesh error 9 at 0x00000000 (:0)' on power up.

Does running mesh require additional changes?

Thanks.

[tesc]

Hi,

From Setting interrupt priority levels:

"However, if you use the nRF5 SDK's app_scheduler module to run the application from the main loop, the low priority parts of the mesh stack should also run from the main loop."

In other words, when using the app scheduler the irq_priority field must be NRF_MESH_IRQ_PRIORITY_THREAD (also stated in the same section of the setting interrupt priority levels documentation page.)

Regards,
Terje