APPLIED/cmnt: [B][SRU][PATCH 0/5] Fix kselftest/net test build for the upcoming granularity improvement

[Kleber Souza]

On 05.11.21 11:45, Po-Hsu Lin wrote:
> [Impact]
> With 4.15 kernel, not all binary files inside a subtest directory in
> tools/testing/selftests of a kernel tree can be built with command:
> 
>      make TARGETS=$subtest
> 
> For example, the following binaries in net directory won't be built:
>   * reuseport_bpf
>   * reuseport_bpf_cpu
>   * reuseport_dualstack
>   * reuseaddr_conflict
>   * reuseport_bpf_numa
> 
> They can only be build with:
> 
>      make TARGETS=net run_tests
> 
> However, this will also start all of the tests in this subset and it's
> is not ideal for the granularity improvement, in which we will want to
> run tests one-by-one.
> 
> [Fix]
> 1. 14f1889f selftests: Fix loss of test output in run_kselftests.sh
> 2. 44f01352 selftests: Makefile set KSFT_TAP_LEVEL to prevent nested TAP headers
> 3. 8ce72dc3 selftests: fix headers_install circular dependency
> 4. 27d79a2b selftests: fix bpf build/test workflow regression when KBUILD_OUTPUT is set
> 5. e2e88325 selftests: vm: Fix test build failure when built by itself
> 
> The 1st and 5th patch can be cherry-picked. The 2nd needs to be
> backported with some context adjustment (can be applied with git -cp
> but not git am -C2). The 3rd and 4th patch would need to be applied
> with -C2.
> 
> The third patch (8ce72dc3) is the fix for this issue. To get it
> applied more easily we will need the first two.
> 
> The 4th (27d79a2b) and the 5th (e2e88325) patches are included here as
> they have been marked as a fix for 8ce72dc3. Since we don't use
> KBUILD_OUTPUT and the vm test, I think the risk of adding these two is
> pretty low.
> 
> [Test]
> Tested on a baremetal node "glameow" with 4.15.0-159-generic, the test
> cases number and results are identical before and after applying this
> change (compared side-by-side).
> 
> Please find attachment in the bug report for the test result.
> 
> [Where problems could occur]
> As this is just tested with generic Bionic 4.15 kernel on a bare-metal
> system, it's is possible to see other issues on different variants. If
> this change is incorrect it will cause issue to our automated-testing
> and regression-testing, but it won't affect the real kernel
> functionality.
> 
> 
> Michael Ellerman (1):
>    selftests: Fix loss of test output in run_kselftests.sh
> 
> Shuah Khan (4):
>    selftests: Makefile set KSFT_TAP_LEVEL to prevent nested TAP headers
>    selftests: fix headers_install circular dependency
>    selftests: fix bpf build/test workflow regression when KBUILD_OUTPUT
>      is set
>    selftests: vm: Fix test build failure when built by itself
> 
>   tools/testing/selftests/Makefile    | 71 +++++++++++++++++++++++++++++++------
>   tools/testing/selftests/lib.mk      | 34 ++++++++++++++++--
>   tools/testing/selftests/vm/Makefile |  4 ---
>   3 files changed, 92 insertions(+), 17 deletions(-)
> 

This patchset was applied on Nov-05 to bionic:linux but I forgot
to send the APPLIED reply. Sorry about that.

Thanks,
Kleber