Softdevice fault

RE: Softdevice fault

Wojciech Jasko — Thu, 21 Feb 2019 07:01:10 GMT

Hi,

Michal managed to fix this issue. I would like to share with you our findings.

It looks like workaround from MDK for Errata 36 does not cover all corner cases.

Our hypothesis: Calibration is not terminated by reset.

We extended system_nrf52.c with this code:

void SystemInit(void)
{
 (...)
    /* Workaround for Errata 36 "CLOCK: Some registers are not reset when expected" found at the Errata document
       for your device located at https://infocenter.nordicsemi.com/  */
    if (errata_36()){
        NRF_CLOCK->EVENTS_DONE = 0;
        NRF_CLOCK->EVENTS_CTTO = 0;
        NRF_CLOCK->CTIV = 0;
    }

    lfclk_calibrate();
 (...)
}
 

static void lfclk_calibrate(void)
{
    // Turn on HFCLK XTAL
    NRF_CLOCK->EVENTS_HFCLKSTARTED = 0;
    NRF_CLOCK->TASKS_HFCLKSTART = 1;
    while (NRF_CLOCK->EVENTS_HFCLKSTARTED == 0)
    {
    }

    // Start LFCLK (32kHz) RC oscillator.
    NRF_CLOCK->LFCLKSRC = CLOCK_LFCLKSRC_SRC_RC << CLOCK_LFCLKSRC_SRC_Pos;
    NRF_CLOCK->EVENTS_LFCLKSTARTED =0;
    NRF_CLOCK->TASKS_LFCLKSTART = 1;
    while (NRF_CLOCK->EVENTS_LFCLKSTARTED == 0);
    NRF_CLOCK->EVENTS_LFCLKSTARTED = 0;

    // Calibrate
    NRF_CLOCK->EVENTS_DONE = 0;
    NRF_CLOCK->TASKS_CAL = 1;
    while (NRF_CLOCK->EVENTS_DONE == 0)
    {
    }
    // Turn off HFCLK XTAL
    NRF_CLOCK->TASKS_HFCLKSTOP = 1;

    // Turn off LFCLK
    NRF_CLOCK->TASKS_LFCLKSTOP = 1;
}

This solved this problem.

RE: Softdevice fault

Sigurd — Mon, 05 Nov 2018 09:51:00 GMT

"I see that you are using SDK_NRF_SDH_CLOCK_LF_XTAL_ACCURACY set to 0( 250ppm). The internal LF clock on the nRF52810/nnRF52832 is 500ppm. So let's try to set SDK_NRF_SDH_CLOCK_LF_XTAL_ACCURACY to 1 (500) ppm, and see if that solves the issue."

RE: Softdevice fault

Michal — Fri, 02 Nov 2018 09:45:11 GMT

set(SDK_NRF_SDH_CLOCK_LF_RC_CTIV           16 CACHE STRING "")
set(SDK_NRF_SDH_CLOCK_LF_RC_TEMP_CTIV       2 CACHE STRING "")
set(SDK_NRF_SDH_CLOCK_LF_SRC                0 CACHE STRING "")
set(SDK_NRF_SDH_CLOCK_LF_XTAL_ACCURACY      0 CACHE STRING "")

Yes, it is always the same address.

I am using fds, I reserved 5 pages (they are placed before bootloader).

I dont think it would be possible as source is confidential.

Today I took one of devices after tests and attached debug session. Turns out for some reason after random number of tests (write to flash + power off) softdevice starts to return NRF_ERROR_INVALID_STATE when calling:

sd_softdevice_enable

It is weird as I dont think softdevice is using any flash so it should at least fix after power cycle but it doesnt. I checked flash content against binary but neither softdevice nor application were overwritten.

EDIT: The above error was caused by some problem with debugger, it is not a real case.

RE: Softdevice fault

Sigurd — Wed, 31 Oct 2018 16:08:59 GMT

Hi,

What is the value of NRF_SDH_CLOCK_LF_SRC, NRF_SDH_CLOCK_LF_RC_CTIV, NRF_SDH_CLOCK_LF_RC_TEMP_CTIV and NRF_SDH_CLOCK_LF_ACCURACY , found in sdk_config.h ?

Is it always the same assert code(0x000089B2) ?

Where in the flash are you writing values to with the sd flash write function?

Could you share the project that creates this issue ? If you don't want to share the project publicly, then we could convert this case into a private one, or you could create a new private question where you upload the code/project, and refers to this thread.

RE: Softdevice fault

Michal — Wed, 31 Oct 2018 07:42:49 GMT

I was running tests for last hour on 4 different bords and it did not fail. I will give feedback after night. If it really helps probably should it be mentioned to not use clock_drv with sd.

EDIT: Hello. I was running tests on 4 boards overnight but they also did fail after 10, 30, 120 and 130 minutes.

RE: Softdevice fault

Michal — Tue, 30 Oct 2018 13:25:13 GMT

I will comment it out and run tests over night to see if it helped.

RE: Softdevice fault

Sigurd — Tue, 30 Oct 2018 13:19:03 GMT

[quote userid="14528" url="~/f/nordic-q-a/39941/softdevice-fault/155091"]4) I only call nrf_drv_clock_init at the beginning[/quote]

We generally don't use that driver together with the SoftDevice, since the SoftDevice controls the NRF_CLOCK when the SoftDeivce is enabled.

Could you try to comment that function out, and see if the assert disappears?

RE: Softdevice fault

Michal — Tue, 30 Oct 2018 12:02:41 GMT

Hello,

1) 2) I have long term tests that are writing data and after some period of time (different values 0-500ms) execute power off on the device. This is rather short test and I was executing it in parallel for 1 hour on 6 different devices. 3 did not fail and other were failing after 20, 40 and 45 minutes.

3) nrf52832 development kit

4) I only call nrf_drv_clock_init at the beginning

EDIT: I am also considering option that I am writing flash too early - before fds return init event. Can that be an option in this case, given SD fault address ?

RE: Softdevice fault

Sigurd — Tue, 30 Oct 2018 10:08:09 GMT

Hi,

1) How often does this happen?

2) Are you able to reliably reproduce this?

3) What hardware are you running on? nRF52832 or nRF52810? custom board or development kit?

4) Are you accessing the NRF_CLOCK peripheral in your application?