Re: [GIT PULL] Kselftest update for Linux 5.20-rc1

From: Shuah Khan
Date: Wed Aug 03 2022 - 19:12:39 EST


On 8/3/22 2:22 PM, Guillaume Tucker wrote:
On 03/08/2022 16:32, Shuah Khan wrote:
On 8/2/22 10:44 PM, Guillaume Tucker wrote:
On 02/08/2022 15:29, Shuah Khan wrote:
On 8/2/22 3:51 AM, Mickaël Salaün wrote:
Hi,

This change breaks the Landlock test build when performed on the tools/testing/selftests/landlock directory because the (non-system/up-to-date) kernel headers aren't found. Looking at the use of top_srcdir and HDR_PATH, it seems that multiple subsystems are using this feature. I consider this change a regression.


I did a build test from the top level before sending the pull request
and didn't catch this breakage as a result. This breaks when build is
run from the test directory.

We have had several problems related to khdr_dir and target and decided
to move away from it with this change.

It also removes the check on up-to-date kernel headers (thanks to the Makefile's target timestamp).

I wasn't CCed for this change impacting Landlock [1]. Please keep in mind to add at least maintainers and related mailing-lists for changes on related subsystems.


That is the usual practice and if I notice missing maintainers, I add
them. We missed this one.

The following patch almost revert commit a917dd94b832 ("selftests/landlock: drop deprecated headers dependency") and partially fixes commit 49de12ba06ef ("selftests: drop KSFT_KHDR_INSTALL make target"):


Guillaume,

Will you be able to look at this and send a patch on top? I will
send another pull request before merge window closes?

Sure, I'll take a look today.


Thank you.

OK I just sent "selftests/landlock: fix broken include of linux/landlock.h"

This does fix the build when doing:

make -C tools/testing/selftests/landlock


Thank you for fixing this quickly.

However I've also noticed the landlock test is failing to build
when make is invoked from the top-level directory and using a
sub-directory for the build output, in other words my earlier
patches didn't fix the build for this test, but that's not a
regression.


Okay.

I'll see if that can be fixed too while also not breaking
the "-C" sub-make build.


Sounds good. Supporting all these use-cases makes it a bit hard.

Also I'll see if we can add some extra build tests in KernelCI
for the kselftest tree to catch issues like these automatically.



Great. Well I shall try and get that set up before making further changes ;)


Thanks.


P.S. The output of gen_tar is showing "-ne " on every line, is that expected?
For example: -ne Emit Tests for alsa


Hmm. I will try and let you. I haven't used this one in a bit.

thanks,
-- Shuah