Upreving Linux Kernel

Occasionally, the GitLab CI needs a Linux Kernel update to enable new kernel features, device drivers, bug fixes etc to CI jobs. Kernel uprevs in GitLab CI are relatively simple, but prone to lots of side-effects since many devices from different platforms are involved in the pipeline.

Kernel repository

The Linux Kernel used in the GitLab CI is stored at the following repository: https://gitlab.freedesktop.org/gfx-ci/linux

It is common that Mesa kernel brings some patches that were not merged on the Linux mainline, that is why Mesa has its own kernel version which should be used as the base for newer kernels.

So, one should base the kernel uprev from the last tag used in the Mesa CI, please refer to .gitlab-ci/container/gitlab-ci.yml KERNEL_URL variable. Every tag has a standard naming: vX.YZ-for-mesa-ci-<commit_short_SHA>, which can be created via the command:

git tag vX.YZ-for-mesa-ci-$(git rev-parse --short HEAD)

Building Kernel

When Mesa CI generates a new rootfs image, the Linux Kernel is built based on the script located at .gitlab-ci/container/build-kernel.sh.

Updating Kconfigs

When a Kernel uprev happens, it is worth compiling and cross-compiling the Kernel locally, in order to update the Kconfigs accordingly. Remember that the resulting Kconfig is a merge between Mesa CI Kconfig and Linux tree defconfig made via merge_config.sh script located at Linux Kernel tree.

Kconfigs location

Platform

Mesa CI Kconfig location

Linux tree defconfig

arm

.gitlab-ci/container/arm.config

arch/arm/configs/multi_v7_defconfig

arm64

.gitlab-ci/container/arm64.config

arch/arm64/configs/defconfig

x86-64

.gitlab-ci/container/x86_64.config

arch/x86/configs/x86_64_defconfig

Updating image tags

Every kernel uprev should update 3 image tags, located at two files.

.gitlab-ci/container/gitlab-ci.yml tag

  • KERNEL_URL for the location of the new kernel

.gitlab-ci/image-tags.yml tags

  • KERNEL_ROOTFS_TAG to rebuild rootfs with the new kernel

  • DEBIAN_X86_TEST_GL_TAG to ensure that the new rootfs is being used by the GitLab x86 jobs

Development routine

  1. Compile the newer kernel locally for each platform.

  2. Compile device trees for ARM platforms

  3. Update Kconfigs. Are new Kconfigs necessary? Is CONFIG_XYZ_BLA deprecated? Does the merge_config.sh override an important config?

  4. Push a new development branch to Kernel repository based on the latest kernel tag used in GitLab CI

  5. Hack build-kernel.sh script to clone kernel from your development branch

  6. Update image tags. See Updating image tags

  7. Run the entire CI pipeline, all the automatic jobs should be green. If some job is red or taking too long, you will need to investigate it and probably ask for help.

When the Kernel uprev is stable

  1. Push a new tag to Mesa CI Kernel repository

  2. Update KERNEL_URL debian/x86_test-gl job definition

  3. Open a merge request, if it is not opened yet

Tips and Tricks

Compare pipelines

To have the most confidence that a kernel uprev does not break anything in Mesa, it is suggested that one runs the entire CI pipeline to check if the update affected the manual CI jobs.

Step-by-step

  1. Create a local branch in the same git ref (should be the main branch) before branching to the kernel uprev kernel.

  2. Push this test branch

  3. Run the entire pipeline against the test branch, even the manual jobs

  4. Now do the same for the kernel uprev branch

  5. Compare the job results. If a CI job turned red on your uprev branch, it means that the kernel update broke the test. Otherwise, it should be fine.

Bare-metal custom kernels

Some CI jobs have support to plug in a custom kernel by simply changing a variable. This is great, since rebuilding the kernel and rootfs may takes dozens of minutes.

For example, Freedreno jobs gitlab.yml manifest support a variable named BM_KERNEL. If one puts a gz-compressed kernel URL there, the job will use that kernel to boot the Freedreno bare-metal devices. The same works for BM_DTB in the case of device tree binaries.

Careful reading of the job logs

Sometimes a job may turn to red for reasons unrelated to the kernel update, e.g. LAVA tftp timeout, problems with the freedesktop servers etc. So it is important to see the reason why the job turned red, and retry it if an infrastructure error has happened.