kselftest Makefiles don't support separate output directory build

[musamaanjum]

Kselftest Makefiles have hard-coded paths. Out of tree build of kselftest fails of some tests if they depend on latest uapi in tree headers.

To reproduce:
Run make -C tools/testing/selftests/ O=<build_dir> inside kernel top most directory. The build of some tests will get fail in this way.

Workaround:
Either build kselftests in tree in same directory or build kselftest from the build directory with slightly changed command, make -C source/tools/testing/selftests.

[musamaanjum]

Patch series has been sent: https://lore.kernel.org/lkml/20220118112909.1885705-1-usama.anjum@collabora.com/T/#mf68787ae303e49314db5a5053832e915c71b512d
V2: https://lore.kernel.org/lkml/20220119101531.2850400-1-usama.anjum@collabora.com/

[musamaanjum]

2 not directly relatad patches were also sent while working on this:
https://lore.kernel.org/lkml/20220118122053.1941915-1-usama.anjum@collabora.com/
https://lore.kernel.org/lkml/20220119103332.2887370-1-usama.anjum@collabora.com/

[gctucker]

OK thanks, all your patches are now added to the staging-next branch:
https://github.com/kernelci/linux/commits/staging-next

[musamaanjum]

The patch series has been accepted. This also fixes kernelci/kernelci-core#981.

[gctucker]

I'd like to see the build issues disappear at least with linux-next in the KernelCI results before closing this issue. That's the difference between the "Fixed" and "Verified" status.

[musamaanjum]

The support of separate directory was added and build of intended kselftests (futex, kvm, net, mptcp, kvm, vm etc) has been fixed on the linux-next. Here are logs: https://storage.staging.kernelci.org/next/master/next-20220128/x86_64/x86_64_defconfig+x86-chromebook+kselftest/gcc-10/logs/kselftest.log

There are a few more failures because of some unrelated other errors. Separate tickets for those:
kernelci/kernelci-core#1013
#95

[gctucker]

This could be the basis for a nice blog post, to explain how KernelCI highlighted some kernel issues and how they got fixed and then eventually verified with KernelCI results. If you don't have the time then I could make a Tweet about it at least.

[gctucker]

If you don't have the time then I could make a Tweet about it at least.

Or you may make one and the KernelCI account will retweet it... Let me know if you want to do any of that.