Re: [PATCH bpf-next v10 02/11] selftests/bpf: Fix test_kmods KDIR to honor O= and distro kernels

["Ricardo B. Marlière" <rbm@suse.com>]

On Thu Apr 30, 2026 at 10:30 AM -03, bot+bpf-ci wrote:
>> diff --git a/tools/testing/selftests/bpf/Makefile b/tools/testing/selftests/bpf/Makefile
>> index 6094fe99b5f6..cc6ee7a2df93 100644
>> --- a/tools/testing/selftests/bpf/Makefile
>> +++ b/tools/testing/selftests/bpf/Makefile
>
> [ ... ]
>
>> diff --git a/tools/testing/selftests/bpf/test_kmods/Makefile b/tools/testing/selftests/bpf/test_kmods/Makefile
>> index 63c4d3f6a12f..031c7454ce65 100644
>> --- a/tools/testing/selftests/bpf/test_kmods/Makefile
>> +++ b/tools/testing/selftests/bpf/test_kmods/Makefile
>> @@ -1,5 +1,16 @@
>>  TEST_KMOD_DIR := $(realpath $(dir $(abspath $(lastword $(MAKEFILE_LIST)))))
>> -KDIR ?= $(abspath $(TEST_KMOD_DIR)/../../../../..)
>> +SRCTREE_KDIR := $(abspath $(TEST_KMOD_DIR)/../../../../..)
>> +# Honor O=/KBUILD_OUTPUT only if they point at a prepared kernel build
>> +# directory (one containing Module.symvers); otherwise treat the value as a
>> +# selftests-only output directory and fall back to in-tree or distro headers.
>> +# The parent bpf/Makefile resolves O=/KBUILD_OUTPUT to absolute paths before
>> +# invoking this sub-make so relative paths still anchor to the user's
>> +# invocation directory.
>> +KMOD_O := $(or $(O),$(KBUILD_OUTPUT))
>> +KMOD_O_VALID := $(if $(KMOD_O),$(if $(wildcard $(KMOD_O)/Module.symvers),$(KMOD_O)))
>> +KDIR ?= $(if $(KMOD_O_VALID),$(SRCTREE_KDIR), \
>> +	    $(if $(wildcard $(SRCTREE_KDIR)/Module.symvers),$(SRCTREE_KDIR), \
>> +		/lib/modules/$(shell uname -r)/build))
>
> Will falling back to the host kernel headers cause compilation errors
> when building in-tree test modules like bpf_testmod?
>
> Since bpf_testmod relies heavily on internal BPF structures,
> experimental macros, and kfuncs that change frequently between kernel
> releases, compiling it against /lib/modules/$(uname -r)/build on a
> system where the running kernel does not exactly match the upstream
> source tree will likely fail.

This is only the fallback case (inspired by livepatch collection which uses
this KDIR as default), the targeted invocation for this series is
something like:

  make -C /lib/modules/$(uname -r)/source/tools/testing/selftests \
       O=/lib/modules/$(uname -r)/build SKIP_TARGETS= TARGETS=bpf \
       BPF_STRICT_BUILD=0

The usual in-tree way of building vmlinux and then the selftests strictly
would keep the same current behavior.

>
> If the host headers are present, the permissive skip check further down
> in the file will pass since the directory exists. Because the
> compilation failure is not ignored by the parent bpf/Makefile, won't
> this abort the entire BPF selftests build with confusing C compilation
> errors instead of silently skipping the unconfigured kernel tree?
>
> Should in-tree test modules avoid falling back to host distro headers to
> prevent these version mismatches?

Without the fallback, a user running from a tree matched to their installed
kernel but without a prior in-tree build would lose bpf_testmod. Small
niche, but the workaround (O=/lib/modules/$(uname -r)/build) might not be
obvious. The fallback covers them at zero cost to everyone else, so I think
it should stay.

>
> This was raised by Sashiko AI reviewer in v9:
>
>     https://lore.kernel.org/bpf/20260429194107.806B7C19425@smtp.kernel.org/
>
> The issue was acknowledged in v9 but has not been addressed in v10.
>
> [ ... ]
>
>
> ---
> AI reviewed your patch. Please fix the bug or email reply why it's not a bug.
> See: https://github.com/kernel-patches/vmtest/blob/master/ci/claude/README.md
>
> CI run summary: https://github.com/kernel-patches/bpf/actions/runs/25167006036