How-To

How to add a new native test suite

This guide contains all the steps needed to add a native test suite to KernelCI. It will cover the LAVA use-case in particular as it is currently the most popular framework to run tests for KernelCI.

Background

KernelCI is still very much about running tests on hardware platforms. More abstract tests such as static analysis and KUnit are on the horizon but they are not quite supported yet. So this guide will only cover functional testing on “real” hardware.

The only moving part is the kernel, which will get built in many flavours for every revision covered by KernelCI. All the tests are part of fixed root file systems (rootfs), which get updated typically once a week. Adding a test therefore involves either reusing an existing rootfs or creating a new one.

A good way to start prototyping things is to use the plain Debian Buster NFS rootfs and install or compile anything at runtime on the target platform itself. This is of course slower and less reliable than using a tailored rootfs with everything already set up, but it allows a lot more flexibility. It is the approach followed in this guide: first using a generic rootfs and then creating a dedicated one.

A simple test

For the sake of this guide, here’s a very simple test to check the current OS is Linux:

[ $(uname -o) = "GNU/Linux" ]

Let’s test this locally first, just to prove it works:

$ [ $(uname -o) = "GNU/Linux" ]
$ echo $?
0

and to prove it would return an error if the test failed:

$ [ $(uname -o) = "OpenBSD" ]
$ echo $?
1

All the steps required to enable this test to run in KernelCI are detailed below. There is also a sample Git branch with the changes:

https://github.com/kernelci/kernelci-core/commits/how-to

Step 1: Enable basic test plan

The first step is to make the minimal changes required to run the command mentioned above.

LAVA job template

See commit on the how-to branch: config/lava: add uname test plan for the How-To guide

KernelCI LAVA templates use Jinja. To add this uname test plan, create a template file config/lava/uname/uname.jinja2:

- test:
    timeout:
      minutes: 1
    definitions:
    - repository:
        metadata:
          format: Lava-Test Test Definition 1.0
          name: {{ plan }}
          description: "uname"
          os:
          - oe
          scope:
          - functional
        run:
          steps:
          - lava-test-case uname-os --shell '[ $(uname -o) = "GNU/Linux" ]'
      from: inline
      name: {{ plan }}
      path: inline/{{ plan }}.yaml
      lava-signal: kmsg

This is pretty much all boiler plate, except for the lava-test-case line which runs the test and uses the exit code to set the result (0 is pass, 1 is fail). Some extra templates need to be added for each boot method, such as GRUB, U-Boot and Depthcharge. For example, here’s the Depthcharge one to use on Chromebooks generic-depthcharge-tftp-nfs-uname-template.jinja2:

{% extends 'boot-nfs/generic-depthcharge-tftp-nfs-template.jinja2' %}
{% block actions %}
{{ super () }}

{% include 'uname/uname.jinja2' %}

{% endblock %}

The name of the template follows a convention to automatically generate the variant required for a particular platform. This one is for the uname test plan on a platform using depthcharge, with the kernel downloaded over tftp and the rootfs available over nfs.

KernelCI YAML configuration

See commit on the how-to branch: config/core: enable uname test plan using Buster NFS

Once the LAVA templates have been created, the next step is to enable the test plan in the KernelCI YAML configuration.

First add the uname test plan with the chosen rootfs (Debian Buster NFS in this case) in test-configs.yaml:

test_plans:

  uname:
    rootfs: debian_buster_nfs

Then define which platforms should run this test plan, still in test-configs.yaml:

test_configs:

  - device_type: hp-11A-G6-EE-grunt
    test_plans:
      - uname

  - device_type: minnowboard-max-E3825
    test_plans:
      - uname

Each test plan also needs to be enabled to run in particular test labs in lab-configs.yaml. Some labs such as the Collabora one allow all tests to be run, and it contains the platforms listed above so no extra changes are required at this point.

These changes are enough to make an intial pull request in kernelci-core, and the test will automatically get run on staging. Then the results will appear on the web dashboard, for example:

Results for uname: «staging-next-20210805.0» on «minnowboard-max-E3825» (kernelci / staging-next)

Note First-time contributors needed to be added to the list of trusted GitHub users by a maintainer before their pull requests get merged and deployed on staging.

Step 2: Modify the file system at runtime

Most tests will require more than what is already available in a plain Buster rootfs. Let’s see how this can be done in a simple way.

Add a C file: uname-os.c

See commit on the how-to branch: config/lava: add uname-os.c

For example, we could have the test implemented as a C program rather than a shell script. See the uname-os.c file.

To test it locally:

$ gcc -o uname-os uname-os.c
$ ./uname-os
System: 'Linux', expected: 'Linux', result: PASS
$ echo $?
0

and to test it would fail if the OS name was not the expected one:

$ ./uname-os OpenBSD
System: 'Linux', expected: 'OpenBSD', result: FAIL
$ echo $?
1

Now, let’s see how this can be used with KernelCI.

Build it and run the C implementation

See commit on the how-to branch: config/lava: download and build uname-os.c and use it

Any arbitrary commands can be added to the uname.jinja2 template before running the actual test cases. In this example, we can install Debian packages then download the uname-os.c file and compile it to be able to finally run it as a test case:

          steps:
          - apt update
          - apt install -y wget gcc
          - wget https://raw.githubusercontent.com/kernelci/kernelci-core/how-to/config/lava/uname/uname-os.c
          - gcc -o uname-os uname-os.c
          - lava-test-case uname-os-shell --shell '[ $(uname -o) = "GNU/Linux" ]'
          - lava-test-case uname-os-c --shell './uname-os'

We now have 2 test cases, one with the shell version and one with the C version. After updating the pull request on GitHub, this will also get tested automatically on staging. Here’s some sample results:

Results for uname: «staging-mainline-20210805.0» on «minnowboard-max-E3825» (kernelci / staging-mainline)

Note If one of the steps fails, the job will abort. So if apt install or wget fails, the tests won’t be run and the LAVA job status will show an error.

Note Some test labs don’t enable a route to the Internet from their hardware platforms so installing things at runtime may not always work. This would typically be discussed as part of the pull request review, depending on what the jobs are trying to do and which device types have been enabled to run them.

Step 3: Going further

With Step 2, pretty much anything can already be run within the limitations of the CPU and network bandwidth on the target platform. Even if this doesn’t take too long to run, there are many reasons why it’s not really suitable to enable in production. There’s no point installing the same packages and building the same source code over and over again, it will add up as significant wasted resources and extra causes for test job failures. Once the steps required to run a test suite are well defined, having a rootfs image with everything pre-installed solves these issues.

Then for more complex tests, results may be produced in other forms than an exit code from a command. Some file may need to be parsed, or any extra logic may need to be added. For example, the v4l2-parser.sh script will run v4l2-compliance and parse the output to then call lava-test-case for each result found. It also uses LAVA test sets, which is a more advanced feature for grouping test results together inside a test suite.

Adding a rootfs variant

Root file systems are built using the kci_rootfs command. All the variants are defined in the config/core/rootfs-configs.yaml file with some parameters. There are also extra dedicated files in config/rootfs such as additional build scripts.

Let’s take a look at the buster-v4l2 rootfs for example:

rootfs_configs:
  buster-v4l2:
    rootfs_type: debos
    debian_release: buster
    arch_list:
      - amd64
      - arm64
      - armhf
    extra_packages:
      - libasound2
      - libelf1
      - libjpeg62-turbo
      - libudev1
    extra_packages_remove:
      - bash
      - e2fslibs
      - e2fsprogs
    script: "scripts/buster-v4l2.sh"
    test_overlay: "overlays/v4l2"
  • arch_list is to define for which architectures the rootfs should be built.

  • extra_packages is a list passed to the package manager to install them.

  • extra_packages_remove is a list passed to the package manager to remove them.

  • script is an arbitrary script to be run after packages have been installed. In this case, it will build and install the v4l2 tools to be able to run v4l2-compliance.

  • test_overlay is the path to a directory with extra files to be copied on top of the file system. In this case, it will install the v4l2-parser.sh script to parse the output of the test suite and report test case results to LAVA:

    $ tree config/rootfs/debos/overlays/v4l2/
    config/rootfs/debos/overlays/v4l2/
    └── usr
        └── bin
            └── v4l2-parser.sh
    

Here’s a sample command using kci_rootfs to build the buster-v4l2 root file system for amd64:

$ docker run -it \
  -v $PWD:/tmp/kernelci-core \
  --privileged \
  --device /dev/kvm \
  kernelci/debos
root@759fc147da29:/# cd /tmp/kernelci-core
root@759fc147da29:~/kernelci-core# ./kci_rootfs build \
  --rootfs-config=buster-v4l2 \
   --arch=amd64

Writing more advanced test definitions

Running fully featured test suites can involve more than just invoking a few commands with the lave-test-case helper. This very much depends on the test itself. Existing KernelCI native tests such as v4l2-compliance, ltp, kselftest, igt and others provide good examples for how to do this. The test-definitions repository (forked from Linaro for KernelCI) can also be used as a reference, and new tests may even be added there to be able to use them in LAVA outside of KernelCI. Finally, the LAVA documentation about writing tests describes all the available features in detail.

Last modified August 6, 2021