esb with softdevice - clock initialization failes (urgent)

[Michal]

Hi,

For my application I was using ESB without softdevice and just recently I wanted to add dfu so I had to add softdevice as well.

After adding softdevice I found out that my piece of code that is starting lfclk and hfclk is now failing.

void clocks_start( void )
{
    nrfx_clock_enable();

    nrfx_clock_hfclk_start();
    while (!nrfx_clock_hfclk_is_running())
                 ;

    nrfx_clock_lfclk_start();
    while (!nrfx_clock_lfclk_is_running())
                 ;
}

First symptom is that hfclk_is_running function never returns true. After commenting out that while() loop when application reach lfclk_start it simply stops execution - I think it gets stuck somewhere in softdevice.

I already tried to use softdevice api for requesting clocks and added softdevice enabling, but then I got some other weird errors - probably softdevice was clashing with ESB, but didn't invetigate it much further. Then I read somewhere that if application does not use softdevice - like in my case - I can just ignore it. But then the question is why this clock function is failing?

Hoping for a quick answer !

Michał

[ovrebekk]

Hi Michał

When using the SoftDevice you shouldn't access the clock peripheral directly, which is what the nrfx_clock driver is doing. 

Starting LF clock is not necessary, since the SoftDevice starts this automatically, but if you want to change the LF clock source you have to do this as a part of the SoftDevice initialization. 

Starting the external HF clock can be done by calling sd_clock_hfclk_request(), after you have enabled the SoftDevice. 

Best regards
Torbjørn 

[Michal]

Hi Torbjørn,

I am a bit confused: I don't want to use softdevice in my main application so I am not enabling softdevice at all. Shouldn't in that case  "standard" nrfx_clock api be fine as suggested by Nguyen? I tried that but it doesn't work for me.

I tried to add softdevice to my application by using softdevice handler module, because I also need flash access. Then I am only requesting hfclk and that works fine. Unfortunately when initializing ESB protocol, exactly when accesing first RADIO register (which is TXPOWER but I don't this matters) application hangs. Is using ESB with softdevice enabled a valid use case? I guess all RADIO registers should be accessed via SD api when it is enabled but there no such ESB implementation which uses SD api.

Best regards

[ovrebekk]

Hi Michał

Sorry, I didn't realize that the issue happens even when the SoftDevice is disabled. I will have to check internally if this is a known issue. 

In the mean time, could you try to start the clocks by accessing the hardware registers directly and see if that works?

To start the LF clock just run the following code: 

NRF_CLOCK->EVENTS_LFCLKSTARTED = 0;
NRF_CLOCK->TASKS_LFCLKSTART = 1;
while(NRF_CLOCK->EVENTS_LFCLKSTARTED == 0);

You can also do something similar for the HF clock, just replace LF with HF in the code snippet above. 

Best regards
Torbjørn

[Michal]

Hi Torbjørn,

So I tried to access NRF_CLOCK registers directly as you suggested. I was observing CLOCK registers in SES ide and I can see that both clocks has status "running" so this seems to be fine. 

Still there seems to be a problem when accessing i2c or spi - device bricks then. As those peripherals require hfclk I believe there is still something wrong with hfclk initialization.

I noticed while debugging that code goes to 0x00000A60 addres which I believe is softdevice? Does it make sense ? I am using s140 sd 7.0.1 for nrf52840

Thanks

[Krzysztof Chruscinski]

can you try to add following code at the beginning of your application

sd_softdevice_vector_table_base_set(CODE_START);

If you are jumping to the application from the bootloader, softdevice might not forward interrupts to the application.

[Michal]

Hi Krzysztof :)

thank you for the input but that doesn't fix the issue. I am actually testing this without btl right now, just sd + application (and I get the same behaviour in both cases) so I wasn't expecting this to help in my case.

Any other suggestions?

Thanks,
Michał B

[ovrebekk]

Hi Michał 

A colleague of mine pointed out that when you have a SoftDevice added you actually have to reserve 8 bytes of RAM for the SoftDevice, even if you don't enable it. 

This is mentioned in the documentation, here. 

Could you try changing the RAM start address from 0x20000000 to 0x20000008 and see if it solves the issue?

Best regards
Torbjørn

[Michal]

Hi Torbjørn,

Actually it did help! I2C and spi initialized without a problem. But now after ESB init, code executes few more lines and then again jumps to that mysterious 0x00000A60 address and then I get HardFualt.

What else can be wrong here?

Best regards

EDIT: 

0x00000A60  is not mysterious anymore it is MBR's hardfault handler function address right?

Also I can see that application runs for few more cycles before I get this hardfault. Just sharing an observation.

EDIT2:

I found out that my tx works fine ! I was testing only with rx so far. Unfortunately rx still crashes with hardfault without clear reason

[Nguyen Hoan Hoang]

8 bytes RAM are not sufficient. You need the reserve at least the minimum required by Softdevice. I have written a firmware using ESB and Softdevice only used for DFU just like yours.   

[Michal]

Minimum RAM required by Softdevice when it is not enabled is 8 bytes as mentioned in documentation linked by Torbjørn.

But I also thought that maybe more bytes are required and tried to move RAM start by 80 and 160 bytes. But that didn't help and I was still getting hardfault.

Do you have any suggestion on this?

[Nguyen Hoan Hoang]

At a min, check in the softdevice doc for min ram required not just 80 or 160.  It is much more than that with SD7.  Secondly check the interrupt handler for ESB and any devices you are using.  Make sure you have them hookup. 

[Michal]

Here is the s140 documentation - link . It says that with softdevice disable RAM consumption is 8bytes. Anything I am missing?