Skip to content

Build configurations

Guillaume Tucker edited this page Aug 3, 2021 · 10 revisions

Status: moved to kernelci.org/docs


KernelCI job configuration is defined in YAML files for builds and tests. This page covers the builds, see also the related page about tests.

All the top-level build configurations are contained in a YAML file: build-configs.yaml. This defines all the kernels that need to be built, with some attributes to cover all the details and variants.

There are several types of entries in this file:

  • trees are git repositories
  • config fragments are modular lists of kernel config values
  • build environments define the available compilers
  • architectures are to keep all the properties related to each CPU architecture
  • build variants are to define a set of builds done within a build configuration
  • build configurations are combinations of all of the above

The build configurations are the main entries as they define what is actually going to be be built. They must have one tree and one branch defined, and optionally more details such as which compilers to use, which architectures and defconfigs to build etc... There should not be more than one build config with a given tree / branch combination to avoid monitoring the same branch more than once. Each build entry can have several build variants as sub-sections to define multiple ways of building a same branch.

How to add trees and branches

Each tree is defined in the trees section. So to add a new tree, that is to say a new git repository, add an entry such as this one with the name of the tree and its URL:

  renesas:
    url: "https://git.kernel.org/pub/scm/linux/kernel/git/geert/renesas-devel.git"

Then to start monitoring and building branches from that tree, a build config needs to be added for each branch. For example, in this tree we're monitoring 2 branches:

  renesas:
    tree: renesas
    branch: 'devel'

  renesas_next:
    tree: renesas
    branch: 'next'

The build configuration names typically follow the tree name and the branch name if there's more than one being monitored. It's only a convention to avoid having the same branch monitored twice and to make it easy to maintain, the name is not encoding any actual information as the actual tree and branch names are defined in attributes.

How to override the default build variants

There is a build_configs_defaults entry which acts as a default for the build variants used in build configs. If a variants dictionary is defined in a build config, it will override the default. Build variants are defined as a dictionary to allow several different ones for each tree / branch configuration. They need to have a build_environment attribute referring to a build environment definition and a list of architectures to build with any extra attributes needed to define what needs to be built.

Here's an example for the media tree which only builds 4 architectures and enables a special config fragment virtualvideo:

  media:
    tree: media
    branch: 'master'
    variants:
      gcc-7:
        build_environment: gcc-7
        fragments: [virtualvideo]
        architectures:
          i386: *i386_arch
          x86_64: *x86_64_arch
          arm: *arm_arch
          arm64: *arm64_arch

It uses YAML anchors such as *arm_arch to refer to the default definitions for that architecture. It's also possible to override that to only build a subset of the default configuration, for example to only build the defconfig on arm64 and x86_64:

        architectures:
          arm64:
            filters:
              - whitelist: {defconfig: ['defconfig']}
          x86_64:
            base_defconfig: 'x86_64_defconfig'
            filters:
              - whitelist: {defconfig: ['x86_64_defconfig']}