[SOLVED] Industrialization: How to write specific NVS value during the flash process

Industrialization: How to write specific NVS value during the flash process

I wrote a firmware to run our ESP32-based custom PCB. The firmware holds a unique S/N (serial number) in the NVS thru Preferences API which is set thru the bluetooth app I wrote.

But now I have to produce tens of PCB and it takes time to connect and set the S/N thru the app.

The current process is a 2-step process which I want to streamline:

flash the generic firmware
set the unique S/N

I am wondering if I could write a script that could do both steps, providing the S/N as the script argument.

I could take advantage of Espressif esp tool write_flash for example.

How could I do that?

Solution

You can pre-generate the NVS data partition and flash it together with firmware. ESP IDF provides the NVS Partition Generator Utility specifically for for that purpose. Here's an overview of the process.

First you create a CSV file (let's name it mfgdata.csv) with your pre-generated data, e.g. serial number, product ID, keypair, whatever. Assume the NVS namespace is "mfgdata_ns".

key,type,encoding,value 
mfgdata_ns,namespace,,
serial,data,string,"ABC1234"
private_key,file,string,/path/key.pri

Then you generate the binary partition with NVS data on it (named mfgdata.bin) from this CSV file. Assume the NVS partition starts at address 0x3000.

$IDF_PATH/components/nvs_flash/nvs_partition_generator/nvs_partition_gen.py generate mfgdata.csv mfgdata.bin 0x3000

Finally you flash the NVS partition together with your firmware. Here's a sample flashing command for the NVS partition alone (assuming the partition is named "mfgdata_part" in your partition table).

$ $IDF_PATH/components/partition_table/parttool.py -p /dev/ttyUSB0 -b 921600 write_partition --partition-name=mfgdata_part --input=mfgdata.bin

Firmware starts up, loads the NVS namespace and finds all values that you specified in mfgdata.csv.

You have to be careful when creating the CSV file. Some 2 years ago they were using the python CSV module to parse this file and it worked perfectly, as expected. Then some not very bright person decided to ditch the python module and replace it by splitting each line on comma character, silently ignoring all parsing problems. I don't know if they've unf*cked it yet.