FYI, there is definitely some issue with the configuration loading routine. I created a much simplified loading routine by converting the serial bit-bang vectors into a constant configuration that can be used with the automatic serial transfer, and that seems to work. But we will also need to provide some routines for the Nucleo board that allow the Caravel development board to be run while it's still sitting on (and being powered by) the Nucleo board. Removing the Caravel board from the Nucleo board and putting it on a different supply changes the environment enough that some of the hold violations will shift and cross the boundary between an independent and a dependent hold violation, which invalidates the whole calibration process and causes everything to stop working.

@Tim Edwards Could you provide your "much simplified loading routine" please?

@Philipp Gühring: This program reads the "gpio_config_data.py" file written by the calibration process, and converts it into an equivalent static configuration in C code that can be dropped into a program. Instead of running

gpio_config_io()

to bit-bang the serial transfer, just set the configuration data per the output of

gpio_convert.py

, then run

reg_mprj_xfer = 1;  while (reg_mprj_xfer == 1);

to run the automatic serial transfer. Follow that transfer by setting the configuration values to what they were intended to be, so that the housekeeping module understands which GPIOs are supposed to be inputs and which are supposed to be outputs.