Linux Kernel Testing: Providing Better Results to the Upstream Community
summary
KernelCI adds three new kselftest probes (one for device-tree platforms, one for ACPI platforms, one using a per-board YAML binding) that run against a 200+ device lab and landed partially upstream (tests appear from 6.8). Tests rely on sysfs to verify driver binding, use stable generated ID lists plus a tiny ignore list, and emit ktap-compliant output so CI can parse results. Results already find real regressions and cut down false negatives by removing fragile name- and static-hardware checks.
AI-written summary · about
chapters (21)
- 0:00Linux kernel testing intro
- 1:04Testing quality in the kernel
- 2:50Kosci CI system challenges
- 4:18False negatives in CI
- 6:24Legacy device probe problems
- 7:28High-quality kernel tests
- 9:13Prioritizing in-tree tests
- 10:17Analyzing failures to improve
- 11:24Device probe KUnit tests
- 13:30Device Tree probe tests
- 15:15ACPI probe tests and pitfalls
- 17:42Edge cases in ACPI probing
- 19:28Testing dynamically discovered devices
- 21:17Kernel testing limitations
- 22:42Fixing KUnit output conformance
- 23:52Expanding test coverage
- 24:55Machine-readable kernel errors
- 25:57Plans for new KCI system
- 26:40Community engagement
- 27:46Recap and call for involvement
- 28:30Q&A
related talks
34mLinux Kernel Testing: Providing Better Results to the Upstream Community
19mChecking your work: Linux kernel testing and CI
40mKernelCI for Embedded: Enhancing Reliability through Scalable Test Automation
33mMerging an existing framework into KernelCI
41mKeeping up with LTS: Linux Kernel Functional Testing (LKFT) on Devices
30mKernel Testing with KUnit: Bridging the Gap
33mKernel CI – How Far Can It Go?
23mExperiences and lessons we learned using kselftest and pote