On 21/12/2024 11:41, Masahiro Yamada wrote:
> On Tue, Dec 17, 2024 at 12:10 PM Li Zhijian <lizhijian@xxxxxxxxxxx> wrote:
>>
>> Fixes an issue where out-of-tree kselftest builds fail when building
>> the BPF and bpftools components. The failure occurs because the top-level
>> Makefile passes a relative srctree path ('..') to its sub-Makefiles, which
>> leads to errors in locating necessary files.
>>
>> For example, the following error is encountered:
>>
>> ```
>> $ make V=1 O=$build/ TARGETS=hid kselftest-all
>> ...
>> make -C ../tools/testing/selftests all
>> make[4]: Entering directory '/path/to/linux/tools/testing/selftests/hid'
>> make -C /path/to/linux/tools/testing/selftests/../../../tools/lib/bpf OUTPUT=/path/to/linux/O/kselftest/hid/tools/build/libbpf/ \
>> EXTRA_CFLAGS='-g -O0' \
>> DESTDIR=/path/to/linux/O/kselftest/hid/tools prefix= all install_headers
>> make[5]: Entering directory '/path/to/linux/tools/lib/bpf'
>> ...
>> make[5]: Entering directory '/path/to/linux/tools/bpf/bpftool'
>> Makefile:127: ../tools/build/Makefile.feature: No such file or directory
>> make[5]: *** No rule to make target '../tools/build/Makefile.feature'. Stop.
>> ```
>>
>> To resolve this, the srctree is exported as an absolute path (abs_srctree)
>> when performing an out-of-tree build. This ensures that all sub-Makefiles
>> have the correct path to the source tree, preventing directory resolution
>> errors.
>
> NACK.
> This negates 9da0763bdd82572be243fcf5161734f11568960f
>
> This is a recurring topic [1] because kselftest adopts a completely
> different build system.
> If kselftest cannot do this correctly, please do not hook it to the
> top-Makefile.
>
Hi Masahiro,
Thank you for your feedback
I understand and agree with your points, thanks again for these information.
> [1] https://lore.kernel.org/linux-kbuild/cover.1657614127.git.guillaume.tucker@xxxxxxxxxxxxx/
>
>
>
>> Signed-off-by: Li Zhijian <lizhijian@xxxxxxxxxxx>
>> ---
>> Request for Additional Testing
>>
>> We welcome all contributors and CI systems to test this change thoroughly.
>
> This is NACKed. I recommend not wasting CI system resources.
It sounds reasonable.
It makes sense to avoid using CI resources on this patch given the current circumstances.
Thanks
Zhijian
>
>
>
>
>
>
>> In theory, this change should not affect in-tree builds. However, to ensure
>> stability and compatibility, we encourage testing across different
>> configurations.
>>
>> What has been tested?
>> - out-of-tree kernel build
>> - out-of-tree kselftest-all
>> ---
>> Makefile | 3 ++-
>> 1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/Makefile b/Makefile
>> index e5b8a8832c0c..36e65806bb5e 100644
>> --- a/Makefile
>> +++ b/Makefile
>> @@ -275,7 +275,8 @@ else ifeq ($(srcroot)/,$(dir $(CURDIR)))
>> srcroot := ..
>> endif
>>
>> -export srctree := $(if $(KBUILD_EXTMOD),$(abs_srctree),$(srcroot))
>> +srctree := $(if $(KBUILD_EXTMOD),$(abs_srctree),$(srcroot))
>> +export srctree := $(if $(building_out_of_srctree),$(abs_srctree),$(srctree))
>>
>> ifdef building_out_of_srctree
>> export VPATH := $(srcroot)
>> --
>> 2.44.0
>>
>>
>
>
