Handling NRF_ERROR_INVALID_STATE error code

[Boaz]

There are a bunch of questions (1, 2, 3) about handling the NRF_ERROR_INVALID_STATE code returned by the soft-device. This code is returned when the stack is in an "Invalid state to perform operation", e.g. start-scan while scanning, disconnect while disconnected, etc.

Most answers recommend to rely on SD events to keep track of its state, and only call methods proper for the given stack state. However, in some cases, race conditions are inevitable. For example, a master calls sd_ble_gap_disconnect() from the main tread, while at the same time its peer disconnected.

Sample code examples will call APP_ERROR_CHECK() on the faulty return code, which will ultimately call the weak app_error_fault_handler() and soft-reset the device, aborting the application.

I don't think aborting the application in those cases is a reasonable action.

Some questions:

• Am I correct in stating that some race conditions, related to the soft-device are inevitable (though may be unlikely)?
• If so, can I program fine-grained exception rules for APP_ERROR_CHECK(), e.g. by overriding app_error_fault_handler()?
• Is there sample code for achieving this?

[Susheel Nuguru]

Hi Boaz,

1. There must be no race conditions in the softdevice (ideally). The example you gave is already documented that it is possible to have a ongoing disconnection procedure when you call sd_ble_gap_disconnect and the call will return NRF_ERROR_INVALID_STATE. This is normal and this recoverable error so should not make your system restart.

2. Correct. This is why the definition of app_error_fault_handler is __WEAK. So that your application can override it and handle specific errors in the way the application likes instead of restarting the system. The default definition in the SDK is for unhandled errors so it is mostly useful for debugging. It is always recommended for the application to override it.

3. some peripheral examples have overridden this function. But unfortunately we have no examples to show in BLE.

[bart]

Hi, I'm also running into some invalid state errors, and found this thread. You say it's documented that it returns this error, but it would be nice if the documentation also explicitly says whether it's safe to ignore the error.

[Susheel Nuguru]

noted bart. Will let the team know about this.

[bart]

Appreciated!