Skip to content

Latest commit

 

History

History
150 lines (120 loc) · 7.84 KB

README.md

File metadata and controls

150 lines (120 loc) · 7.84 KB
Raw

Build Sensorgnome Images and Packages

This repository contains scripts to build a complete Sensorgnome flash image, currently only for Raspberry Pi.

This README is about building the Sensorgnome software. Please refer to the documentation for information on installing and using a Sensorgnome.

The whole build process from scratch consists of 3 steps:

  • build a dockcross docker image to be able to cross-compile for the target architecture & OS
  • use the dockcross image to build all the sensorgnome packages
  • use the packages to build a bootable sensorgnome flash image

This repository only deals with the third step. The first step is found in the sensorgnome-dockcross repo, which results in a docker image on DockerHub (currently tvoneicken/sensorgnome-dockcross:armv7-rpi-bookworm-main). Unless this image needs to be rebuilt the easiest is to use the DockerHub one.

The second step should be handled within all the respective repositories and ideally each repo should have a github worflow that automatically produces the packages on push. Until all repos are converted to this new scheme, some have been built manually and uploaded to S3. The outcome of all this is a bunch of packages in some downloadable location, currently https://https://motus-builds.s3.us-east-2.amazonaws.com/<owner>/<repo>/<branch>/*.deb. See this crude file listing. The deb packages are also uploaded to the apt repo https://sensorgnome.s3.amazonaws.com

The final step is handled in this repo using pimod. Pimod uses a docker container and QEMU (a processor emulator) to customize an existing rPi or other ARM image. This is used to load up an official Raspberry Pi OS image, install all the packages produced in the previous step, apply some final tweaks, and save the result as a bootable Sensorgnome image. The output consists of the starting image, an intermediate "base" image, and the final "sg" image in S3 as https://https://motus-builds.s3.us-east-2.amazonaws.com/<owner>/sensorgnome-build/<branch>/sg-*.img.

Note: the previous version of all this used pi-gen to produce the image and there are a few other tools available to do the same. Overall the focus should be on installing the debian packages and they should each do everything necessary for their operation without requiring additional build steps here. The result ought to be that the specific tool used for the final image assembly is unimportant as long as a functional clean image pops out.

The switch from pi-gen was motivated by these comments where the maintainer states "pi-gen serves its primary purpose of building the official release images well enough", i.e., it's not designed for customizing. Pimod's focus is customization.

Operation

The image build process is organized as follows:

  • a `manifest-xxx' file specifies what is to be built, such as the official OS base image, the packagaes to install, the dockcross image to use, the name of the final image, etc.
  • a base-xxx.pifile tells pimod how to construct a "base image" from the official OS image; the intent being that only standard packages get installed that are unlikely to change.
  • a sg-xxx.pifile tells pimod how to construct the final sensorgnome image from the base image by adding all the sensorgnome stuff that may often change.
  • the build.sh script downloads the original OS image, the sensorgnome packages, and then runs pimod two times to generate first the base image and then the final image.
  • the split into a base image and a final image is motivated by the fact that building the base image takes a long time due to the many packages that get downloaded and updated and it's nice to be able to iterate more quickly on the sensorgnome stuff that changes.
  • the build.sh tries to avoid re-downloading or re-building things that are already there, to start from a clean slate remove the images and packages directories.

Debian repository

  • Sensorgnome packages get uploaded to a debian repository at https://sensorgnome.s3.amazonaws.com/
  • Each repository (sensorgnome-support, sensorgnome-control, fcd, ...) uses the generate-deb.yml workflow found in this repository to add its packages to the booktest codename.
  • In this repository, the generate-package.sh script generates a sensorgnome package that depends on the latest version of the packages in testing.
  • The version dependency of the sensorgnome package is ">=", which means that as soon as a new package is in the repo it can be upgraded via apt. The sensorgnome package itself does not need to be upgraded.
  • The sensorgnome package is published to the bookworm codename and images are produced to use the bookworm branch. This means that bookworm gets the version found in testing at the time the sensorgnome package is built&uploaded. This also means that all plain sensorgnome installs only upgrade to versions found in bookworm, i.e., versions from which newer images have been built.
  • To use the packages in booktest, one has to switch /etc/apt/sources.list.d/sensorgnome.list from bookworm to booktest.

Docker

The reason the build process uses docker is that it allows carefully constructed OS images with all the right tools to be run on almost any platform. Together with QEMU it, in addition, allows the final image construction to be run on an emulated ARM platform. This way running a command like apt install does exactly what it would do on a real rPi.

The reason two different docker images (dockcross and pimod) are used is that the former contains cross-compilers while the latter uses QEMU emulation. This means that when compiling programs using dockcross the compiler/link/etc run at full X64 speed and simply generate code for ARM. While when running any command in pimod/QEMU there is a whole processor emulation layer involved.

Tips

  • build the image by running ./build.sh on a unixy system.
  • refer to .gihub/workflows/build.yml for the exact process used by the automated build.
  • to mount an image on linux using loop-back mounts:
  sudo losetup -f images/blah.img --show
  sudo kpartx -avs /dev/loopN
  sudo mount /dev/mapper/loopNp1 /mnt
  • to enable the boot console on a CM3, add console=serial0,115200 to the kernel command line in /boot/cmdline.txt.

Keys to sign packages and update the debian repo

  • TvE has the sensorgnome-repo key in his gpg keyring, it is encrypted using a password found in his password store
  • The gpg key must be exported as ascii using
  gpg --list-secret-keys --keyid-format LONG; gpg --export-secret-keys --armor 11162C1D8661F9148480CDD98EFF151A5DDAE8F1
  • The result must be set as GPG_PRIVATE_KEY secret in github, and the passphrase as GPG_PASSPHRASE
  • To upload the package to S3, an AWS role must be configured to allow the github action to upload
  • To extend the validity of the keys:
gpg --edit-key 11162C1D8661F9148480CDD98EFF151A5DDAE8F1
gpg> expire
gpg> key 1
gpg> expire
gpg> save
  • To export the public key, which needs to be placed in /etc/apt/trusted.gpg.d/sensorgnome.gpg:
gpg --export 11162C1D8661F9148480CDD98EFF151A5DDAE8F1 >sensorgnome-pub-2024.gpg

Sensorstation boot

  • The jumper should be in the "enable USB boot" position
  • To mount the SS EMMC: run rpiboot, plug the micro-USB in, then power on the SS
  • To boot the SS normally: remove the micro-USB then power on the SS
  • To see the boot console, use a USB-serial adapter on the rPi expansion connector at 115200
  • If the console switches to NUL after the initial boot: that's gpsd trying to talk to a gps @9600 baud; this can be disabled in /opt/sensorgnome/gps-clock/init-adafruit-gps.sh