NVS in watchdog callback

Hi,

You only get ~2 32KHz clock cycles from the WDT IRQ is raised until the WD reset will be effectuated, and I doubt this is enough time to perform a flash write. I would rather recommend you to store the uptime to a non-initialized section and RAM as in the sample posted here: https://devzone.nordicsemi.com/f/nordic-q-a/80416/save-register-information-when-watchdog-reset/334194#334194. You may store the uptime from RAM to flash on the following boot. 

Best regards,

Vidar

Hi Vidar,

First of all I should mention that task watchdog is used if that changes anything.

In my watchdog callback I have tried to create a while loop printing the uptime to measure the amount of time available before the reset occours. This gave me around 40 ms which is plenty to save uptime in NVS.

Since topic creation, I have been succesfull in writing to nvs by taking the WDT sample and added the NVS functionality. However as soon as i set CONFIG_BT=y, the NVS is giving me the ETIMEDOUT error. Why does it not work with BT enabled? 

Hi,

Thanks for the clarification. It does make a difference actually, I assumed you where using the HW WDT (WDT — Watchdog timer) which is less flexible in that it won't let you abort or delay an impending reset.

It doesn't sound like you are getting the MPU fault as discussed here https://developer.nordicsemi.com/nRF_Connect_SDK/doc/latest/zephyr/services/storage/nvs/nvs.html#troubleshooting either. Have you tried to single step through the function to see what condition is causing it to return with this error?

Edit:
Do you have CONFIG_BT_SETTINGS=y enabled as well?

It seems to work if I signal a thread from the WDT callback to save the uptime. I suppose the NVS write function have some mutex/semaphore stuff that should not be handled in ISR. However, it is still strange that it work with CONFIG_BT=n.

I'm not sure whether it is to risky to rely on signaling a thread to save NVS after the watchdog has bitten as the system might be in an unstable state (the task could have crashed).

Wouldn't it be better just to allocate a flash area for storing uptime and inside the WDT CB write directly to the flash address without using the NVS driver?

It seems to work if I signal a thread from the WDT callback to save the uptime. I suppose the NVS write function have some mutex/semaphore stuff that should not be handled in ISR. However, it is still strange that it work with CONFIG_BT=n.

I'm not sure whether it is to risky to rely on signaling a thread to save NVS after the watchdog has bitten as the system might be in an unstable state (the task could have crashed).

Wouldn't it be better just to allocate a flash area for storing uptime and inside the WDT CB write directly to the flash address without using the NVS driver?