Skip to content

Merge upstream - #7

Merged
mejedi merged 27 commits into
mainfrom
merge-upstream
Nov 9, 2026
Merged

mejedi merged 27 commits into
mainfrom
merge-upstream

Conversation

@mejedi

@mejedi mejedi commented Nov 9, 2026

Copy link
Copy Markdown
Collaborator

No description provided.

florianl and others added 26 commits April 15, 2026 12:47
Signed-off-by: Florian Lehner <dev@der-flo.net>
Signed-off-by: Florian Lehner <dev@der-flo.net>
Signed-off-by: Florian Lehner <dev@der-flo.net>
Signed-off-by: Florian Lehner <dev@der-flo.net>
In kernel 7.1, a new BTF feature called "layout" was added. This
requires pahole to have support for it. The latest pahole release
(v1.31) does not have support for this, so we need to install a
newer, not yet released version of pahole.

Signed-off-by: Dylan Reimerink <dylan.reimerink@isovalent.com>
BPF program signing support is gated by the kernel configuration option
`CONFIG_SYSTEM_DATA_VERIFICATION`. This kconfig is hidden, only enabled
when other kconfigs that depend on it are enabled.

Enabling `CONFIG_FS_VERITY` and `CONFIG_FS_VERITY_BUILTIN_SIGNATURES`
seemed to be the best way to enable support without to much side
effects.

Signed-off-by: Dylan Reimerink <dylan.reimerink@isovalent.com>
This commit enables RSA, ECDSA, and MLDSA signing algorithms as well as
SHA256, SHA512, and SHA3 hashing algorithms in the kernel configuration.

Enabling these in the kernel to allow verification of different
signature types for BPF programs.

Signed-off-by: Dylan Reimerink <dylan.reimerink@isovalent.com>
Signed-off-by: Timo Beckers <timo@isovalent.com>
Signed-off-by: Timo Beckers <timo@isovalent.com>
These still build libbpf for bpf_preload_umd and require those specific
ARM64 dependencies to be installed.

Signed-off-by: Timo Beckers <timo@isovalent.com>
With cilium@a8de719 the Linux kernel versions were updated and the 7.1 tag moved from 7.1.1 to 7.1.12.

This change was noticed, as in our CIs 7.1 is used for testing against amd64 and arm64 Linux kernel. But with the mentioned commit, no 7.1.12 arm64 image was created and our CI started to fail. 

To fix this, always build an arm64 image and don't limit it to static tags.

Signed-off-by: Florian Lehner <florianl@users.noreply.github.com>
…on fields

This makes the process a little more readable and understandable. Add a
'channel' field to versions.json to be used in a follow-up commit. Rename
static_tag to image_tag since it felt a little confusing.

AIL:3

Signed-off-by: Timo Beckers <timo@isovalent.com>
When testing locally, it can be a pain to figure out where the images are
again, especially after not having touched the repo in a few months.

This change puts it front and center and makes sure it stays updated.

AIL:3

Signed-off-by: Timo Beckers <timo@isovalent.com>
Since ae9bece, build_arm64 has been true on all call sites. Make it baseline.

Signed-off-by: Timo Beckers <timo@isovalent.com>
…mption

Signed-off-by: Timo Beckers <timo@isovalent.com>
Signed-off-by: Timo Beckers <timo@isovalent.com>
build-and-push was starting to show its age, and the workflow yaml was starting
to absorb a lot of logic as bash-in-yaml, which wasn't super robust. The root
directory was also getting crowded, which made it difficult to establish which
files were affecting the build output and which weren't. I ended up splitting
things into three directories:

- build/: for all files involved in the live build process: (kernel) configs,
  scripts and the Dockerfile.
- toolchain/: for building and configuring the GCC and LLVM toolchain involved
  in building kernels.
- scripts/: automation for managing versions.json and scripts invoked by GHA
  workflows.

As a result, we can now hash the contents of build/, which, paired with the
versions described in versions.json, can be used to drive an idempotent build
pipeline saving us a lot of CI and iteration time, and even allows for local
builds + pushes if necessary. The logic for making this all happen has been
extracted into local-friendly shell scripts that can be linted and run locally.

We can now make changes to scripts/ without letting CI spend an hour rebuilding
the world.

AIL:3 since I was and still am clueless about GHA.

Signed-off-by: Timo Beckers <timo@isovalent.com>
Signed-off-by: Timo Beckers <timo@isovalent.com>
Signed-off-by: Timo Beckers <timo@isovalent.com>
GHA's warnings about Node 20 deprecations were piling up and cluttering our
build summaries.

Signed-off-by: Timo Beckers <timo@isovalent.com>
To allow importers to follow along at their own pace, give the repository
tags automatically on merges to main when any changes are made to ./matrix.

Three tag tiers are maintained:

- v7.3.202609271941 as a unique, non-floating tag
- v7.3 floating tag to follow along with the current minor version
- v7 floating tag to follow along with the major version

Signed-off-by: Timo Beckers <timo@isovalent.com>
Signed-off-by: Timo Beckers <timo@isovalent.com>
Bumps the github-actions group with 2 updates: [actions/cache](https://github.com/actions/cache) and [actions/upload-artifact](https://github.com/actions/upload-artifact).


Updates `actions/cache` from 4 to 6
- [Release notes](https://github.com/actions/cache/releases)
- [Changelog](https://github.com/actions/cache/blob/main/RELEASES.md)
- [Commits](actions/cache@v4...v6)

Updates `actions/upload-artifact` from 4 to 7
- [Release notes](https://github.com/actions/upload-artifact/releases)
- [Commits](actions/upload-artifact@v4...v7)

---
updated-dependencies:
- dependency-name: actions/cache
  dependency-version: '6'
  dependency-type: direct:production
  update-type: version-update:semver-major
  dependency-group: github-actions
- dependency-name: actions/upload-artifact
  dependency-version: '7'
  dependency-type: direct:production
  update-type: version-update:semver-major
  dependency-group: github-actions
...

Signed-off-by: dependabot[bot] <support@github.com>
The modules_install RUN steps are separate shell invocations from build-vmlinux.sh,
so ARCH/CROSS_COMPILE exported by env.sh weren't set when they ran.

The build-vmlinux stage runs on $BUILDPLATFORM (amd64). Without ARCH, Kbuild
defaults to the host's uname -m, i.e. x86_64, even when targeting arm64.

Kbuild notices the arch change and re-runs syncconfig against the arm64 .config,
invalidating the build. For test_kmods (6.14+) the RUN also did the modules build,
so it tried to compile arm64 modules with the x86 toolchain and failed.

For bpf_testmod (pre-6.14) the modules were already built correctly by
build-vmlinux.sh, so only the install step was affected — which is why 6.12 arm64
images still ship the module.

The failure was masked: the RUN chained make ...; ln -s ..., so the exit status
was that of ln, not make. The arm64 images got published with a dangling /lib ->
usr/lib symlink and no modules.

This commit addresses these issues by requiring one kmod directory lookup to
succeed, and uses that lookup as a source of truth in all scenarios. build-vmlinux
builds the modules, Dockerfile installs them. Tolerate missing modules on 5.10,
since they were only introduced later.

Signed-off-by: Timo Beckers <timo@isovalent.com>
We want to exercise kprobe_multi on arm64 as well, so move it to the common
configuration.

Signed-off-by: Timo Beckers <timo@isovalent.com>
On arm64, kernels before 5.15 apparently didn't enable CONFIG_TMPFS_POSIX_ACL
by default, causing TMPFS_XATTR to remain unselected.

Enable it explicitly since vimto depends on it to mask the host's tmpfs and a
few other dirs.

Signed-off-by: Timo Beckers <timo@isovalent.com>
@mejedi
mejedi merged commit 67a1cc8 into main Nov 9, 2026
9 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants