Curated device data for all AVR and STM32 devices
Switch branches/tags
Clone or download
salkinium Move tools/device/modm -> tools/device/modm_devices
This prevents shadowing of the modm package from PyPi.
Latest commit b02e24e Nov 15, 2018
Type Name Latest commit message Commit time
Failed to load latest commit information.
devices Update STM32 device files Oct 14, 2018
tools Move tools/device/modm -> tools/device/modm_devices Nov 14, 2018
.gitignore Initial commit Aug 30, 2016
.travis.yml [ci] Extend CI tests Oct 26, 2018
LICENSE Initial commit Aug 30, 2016 [dfg] Groom extractor makefile Oct 26, 2018

Curated data for AVR and STM32 devices

This repository contains tools for extracting data from vendor sources, filtering and reformatting them into a vendor-independent format.

This data is used by the modm project to generate its Hardware Abstraction Layer (HAL), startup code and additional support tools. Please have a look at modm's platform modules for examples on how to use this data.

These tools and this data set is maintained and curated by @salkinium only at modm-io/modm-devices. It is licensed under the MPLv2 license.

Currently data on all STM32 families is available, as well as all AVR devices (most of which are missing GPIO data though). Please open an issue or better yet a pull request for additional support.


git clone
cd modm-devices/tools/generator
# Clone the required dependencies
make init
# Extract and generate STM32 device data
make extract-stm32
make generate-stm32
# Extract and generate AVR device data
make extract-avr
make generate-avr

You need Python3 with lxml, jinja2, deepdiff and CppHeaderParser packages, as well as Java (for javac/java).

pip install lxml jinja2 deepdiff CppHeaderParser


The device data idea originally comes from xpcc, which is the predecessor to modm. Around 2013 we wanted to remove some of the repetitive steps for building a HAL for AVR and STM32 devices and we chose to extract some common data and collapse some peripheral drivers into Jinja2 templates.

This eventually evolved from manually extracted device data to fully generated device data as soon as we found machine readable data sources from vendors. For AVRs, we use the Atmel Target Description Files and for STM32, we use internal data extracted from the CubeMX code generator.

Thus the Device File Generator (DFG) was born. The DFG has been rewritten for modm to make it more maintainable and flexible as well as handling edge cases much better.

We've separated the device data from modm, so that it becomes easier for YOU to use this data for your own purposes. I've written an blog post with all the details.

Data quality

The quality of the resulting device data obviously depend heavily on the quality of the input data. I reluctantly maintain a manual patch list for the bugs I've encountered in the vendor sources, that I don't want to write a fix for in the DFG. I have sent some of these patches to a contact in ST, however, every new release of CubeMX changes a lot of data and can reintroduce some of these bugs. I don't have a contact at Atmel to send bug reports to.

In addition, the CubeMX and AVR data does not contain some very important information, which has to be assembled manually from hundreds of datasheets and is then injected into the DFG. This is extremely labor intensive.

Please be respectful in asking for more data: I do not like spending hours upon hours copying this additional data out of datasheets. It's also much more likely to introduce errors, so automating data extraction is much easier for me to maintain. You may of course open an issue about wrong data, but I'd prefer if you opened a pull request that fixes the problem in the DFG instead.

All fixes MUST BE REPRODUCIBLE by the DFG! This means you need to track down the bug to either the raw vendor data (=> update the manual patches) or in the DFG data pipeline (=> fix the DFG).

DO NOT UNDER ANY CIRCUMSTANCES PUBLISH THE RAW DATA EXTRACTED FROM CUBEMX ANYWHERE! It is subject to ST's copyright and you are not allowed to distribute it!

Data format

I initially wanted to format this data as device trees, however, since it is so tied to the Linux kernel, there isn't (or wasn't) much tool support available at the time (though now there is a Python parser in Zephyr), so we wrote our own tree-based format, which we called "device files" since we're so creative. It allows lossless overlaying of data trees to reduce the amount of duplicate data noise which makes it easier to comprehend as a human.

I do not intend to standardize this format, it may change at any time for any reason. This allows us maximum flexibility in encoding this complicated device information. If you want to engage in format discussions, please consider contributing to the device tree specification instead.

Since I may change this data format to accommodate new data, you should write your own formatter of this data, so that you have much better control over what your tools are expecting! So, if you need this data in the form of a Device Tree, please write your own data converter and maintain it yourself!

For modm we convert this format to a Python dictionary tree, for details see the DeviceFile class in tools/device/modm/