[PATCH] kbuild: unset sub_make_done before calling kselftest build system

[PATCH] kbuild: unset sub_make_done before calling kselftest build system

[Thomas Weißschuh]

The kselftest build system may recourse back into kbuild when building
test modules. In that case kbuild needs to parse the new flags passed
from the command line, instead of using the ones inherited from the
kbuild invocation.

Force that command line reevaluation.

The same was done for scripts/install.sh in commit 14ccc638b02f9ec
("kbuild: cancel sub_make_done for the install target to fix DKMS")

Reported-by: Zelin Deng <zelin.deng@linux.alibaba.com>
Closes: https://lore.kernel.org/all/20260525083721.27857-1-zelin.deng@linux.alibaba.com/
Fixes: c9bb03ac2c66 ("kbuild: reduce output spam when building out of tree")
Signed-off-by: Thomas Weißschuh <thomas.weissschuh@linutronix.de>
---
 Makefile | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/Makefile b/Makefile
index b9c5792c79e0..f717a4dc96d6 100644
--- a/Makefile
+++ b/Makefile
@@ -1591,10 +1591,10 @@ tools/%: FORCE
 
 PHONY += kselftest
 kselftest: headers
-	$(Q)$(MAKE) -C $(srctree)/tools/testing/selftests run_tests
+	$(Q)unset sub_make_done; $(MAKE) -C $(srctree)/tools/testing/selftests run_tests
 
 kselftest-%: headers FORCE
-	$(Q)$(MAKE) -C $(srctree)/tools/testing/selftests $*
+	$(Q)unset sub_make_done; $(MAKE) -C $(srctree)/tools/testing/selftests $*
 
 PHONY += kselftest-merge
 kselftest-merge:

---
base-commit: dc59e4fea9d83f03bad6bddf3fa2e52491777482
change-id: 20260702-makefile-unset-submake-done-cbc0b9375617

Best regards,
--  
Thomas Weißschuh <thomas.weissschuh@linutronix.de>

Re: [PATCH] kbuild: unset sub_make_done before calling kselftest build system

[sashiko-bot]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] Incomplete fix and inconsistent state: The patch correctly unsets `sub_make_done` for the `kselftest` targets to allow Kbuild command-line reevaluation when building external kernel modules. However, it fails to apply the same logic to the `tools/` and `tools/%` targets, introducing a discrepancy between build entry points.
--

commit 06c418388d65deb3d8494a6e8e7071c0e709186b
Author: Thomas Weißschuh <thomas.weissschuh@linutronix.de>

kbuild: unset sub_make_done before calling kselftest build system

This commit unsets sub_make_done before invoking the kselftest build system to
ensure Kbuild reevaluates command-line flags when building test modules. It
mirrors a previous fix done for scripts/install.sh.

> diff --git a/Makefile b/Makefile
> index b9c5792c79e01..f717a4dc96d6a 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -1591,10 +1591,10 @@ tools/%: FORCE
>  
>  PHONY += kselftest
>  kselftest: headers
> -	$(Q)$(MAKE) -C $(srctree)/tools/testing/selftests run_tests
> +	$(Q)unset sub_make_done; $(MAKE) -C $(srctree)/tools/testing/selftests run_tests

[Severity: Medium]
Will make tools/selftests still fail after this change?

The tools/ and tools/% targets in Makefile delegate to the tools build
system, which can also trigger external kernel module builds (like bpf_testmod).

Because sub_make_done is not unset for these targets, invoking commands like
make tools/selftests or make tools/all will skip Kbuild's M= argument parsing.

Should sub_make_done also be unset in the tools/ and tools/% targets to maintain
consistency across the build entry points?

>  kselftest-%: headers FORCE
> -	$(Q)$(MAKE) -C $(srctree)/tools/testing/selftests $*
> +	$(Q)unset sub_make_done; $(MAKE) -C $(srctree)/tools/testing/selftests $*
>  
>  PHONY += kselftest-merge
>  kselftest-merge:

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260703-makefile-unset-submake-done-v1-1-6899248f3d6a@linutronix.de?part=1

Re: [PATCH] kbuild: unset sub_make_done before calling kselftest build system

[Petr Mladek]

On Fri 2026-07-03 08:04:08, Thomas Weißschuh wrote:
> The kselftest build system may recourse back into kbuild when building
> test modules. In that case kbuild needs to parse the new flags passed
> from the command line, instead of using the ones inherited from the
> kbuild invocation.
> 
> Force that command line reevaluation.
> 
> The same was done for scripts/install.sh in commit 14ccc638b02f9ec
> ("kbuild: cancel sub_make_done for the install target to fix DKMS")
>
> Reported-by: Zelin Deng <zelin.deng@linux.alibaba.com>
> Closes: https://lore.kernel.org/all/20260525083721.27857-1-zelin.deng@linux.alibaba.com/

It is great that you added references to the related commit and
the original thread.

> Fixes: c9bb03ac2c66 ("kbuild: reduce output spam when building out of tree")
> Signed-off-by: Thomas Weißschuh <thomas.weissschuh@linutronix.de>

I believe that this is reasonable solution:

Acked-by: Petr Mladek <pmladek@suse.com>

Zelin, could you please provide your Tested-by, please?

Best Regards,
Petr

PS: I could take this via livepatching tree. But feel free to take it
    via kbuild tree. I am going to wait few days until the discussion
    settles down.

Re: [PATCH] kbuild: unset sub_make_done before calling kselftest build system

[Petr Mladek]

On Fri 2026-07-03 06:14:48, sashiko-bot@kernel.org wrote:
> Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
> - [Medium] Incomplete fix and inconsistent state: The patch correctly unsets `sub_make_done` for the `kselftest` targets to allow Kbuild command-line reevaluation when building external kernel modules. However, it fails to apply the same logic to the `tools/` and `tools/%` targets, introducing a discrepancy between build entry points.
> --
> 
> commit 06c418388d65deb3d8494a6e8e7071c0e709186b
> Author: Thomas Weißschuh <thomas.weissschuh@linutronix.de>
> 
> kbuild: unset sub_make_done before calling kselftest build system
> 
> This commit unsets sub_make_done before invoking the kselftest build system to
> ensure Kbuild reevaluates command-line flags when building test modules. It
> mirrors a previous fix done for scripts/install.sh.
> 
> > diff --git a/Makefile b/Makefile
> > index b9c5792c79e01..f717a4dc96d6a 100644
> > --- a/Makefile
> > +++ b/Makefile
> > @@ -1591,10 +1591,10 @@ tools/%: FORCE
> >  
> >  PHONY += kselftest
> >  kselftest: headers
> > -	$(Q)$(MAKE) -C $(srctree)/tools/testing/selftests run_tests
> > +	$(Q)unset sub_make_done; $(MAKE) -C $(srctree)/tools/testing/selftests run_tests
> 
> [Severity: Medium]
> Will make tools/selftests still fail after this change?
> 
> The tools/ and tools/% targets in Makefile delegate to the tools build
> system, which can also trigger external kernel module builds (like bpf_testmod).
> 
> Because sub_make_done is not unset for these targets, invoking commands like
> make tools/selftests or make tools/all will skip Kbuild's M= argument parsing.
> 
> Should sub_make_done also be unset in the tools/ and tools/% targets to maintain
> consistency across the build entry points?

This might make sense. I see that the tools: and tools/%: targets are
defined right above the kselftest*: ones. And they are even advertised
in "make help" or "make tools/help".

Well it probably should be done in a separate patch. It would be better
for bisecting potential regressions. Thomas, Zelin, what do you think, please?

Best Regards,
Petr

Re: [PATCH] kbuild: unset sub_make_done before calling kselftest build system

[Miroslav Benes]

On Fri, 03 Jul 2026 08:04:08 +0200, Thomas Weißschuh <thomas.weissschuh@linutronix.de> wrote:
> kbuild: unset sub_make_done before calling kselftest build system

Acked-by: Miroslav Benes <mbenes@suse.cz>

-- 
Miroslav

Re: [PATCH] kbuild: unset sub_make_done before calling kselftest build system

[Thomas Weißschuh]

On Fri, Jul 03, 2026 at 10:26:07AM +0200, Petr Mladek wrote:
> On Fri 2026-07-03 06:14:48, sashiko-bot@kernel.org wrote:
> > Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
> > - [Medium] Incomplete fix and inconsistent state: The patch correctly unsets `sub_make_done` for the `kselftest` targets to allow Kbuild command-line reevaluation when building external kernel modules. However, it fails to apply the same logic to the `tools/` and `tools/%` targets, introducing a discrepancy between build entry points.
> > --
> > 
> > commit 06c418388d65deb3d8494a6e8e7071c0e709186b
> > Author: Thomas Weißschuh <thomas.weissschuh@linutronix.de>
> > 
> > kbuild: unset sub_make_done before calling kselftest build system
> > 
> > This commit unsets sub_make_done before invoking the kselftest build system to
> > ensure Kbuild reevaluates command-line flags when building test modules. It
> > mirrors a previous fix done for scripts/install.sh.
> > 
> > > diff --git a/Makefile b/Makefile
> > > index b9c5792c79e01..f717a4dc96d6a 100644
> > > --- a/Makefile
> > > +++ b/Makefile
> > > @@ -1591,10 +1591,10 @@ tools/%: FORCE
> > >  
> > >  PHONY += kselftest
> > >  kselftest: headers
> > > -	$(Q)$(MAKE) -C $(srctree)/tools/testing/selftests run_tests
> > > +	$(Q)unset sub_make_done; $(MAKE) -C $(srctree)/tools/testing/selftests run_tests
> > 
> > [Severity: Medium]
> > Will make tools/selftests still fail after this change?
> > 
> > The tools/ and tools/% targets in Makefile delegate to the tools build
> > system, which can also trigger external kernel module builds (like bpf_testmod).
> > 
> > Because sub_make_done is not unset for these targets, invoking commands like
> > make tools/selftests or make tools/all will skip Kbuild's M= argument parsing.
> > 
> > Should sub_make_done also be unset in the tools/ and tools/% targets to maintain
> > consistency across the build entry points?
> 
> This might make sense. I see that the tools: and tools/%: targets are
> defined right above the kselftest*: ones. And they are even advertised
> in "make help" or "make tools/help".

I am not sure how the tools/ targets can reach back into kbuild.
The example with bpf_testmod above is bogus, as tools/* is only for
top-level tools directories. Subdirectories in tools/testing/selftests/
don't seem to be reachable.

> Well it probably should be done in a separate patch. It would be better
> for bisecting potential regressions. Thomas, Zelin, what do you think, please?

If we do it for consistency it should be its own patch, agreed.


Thomas

Re: [PATCH] kbuild: unset sub_make_done before calling kselftest build system

[Petr Mladek]

On Fri 2026-07-03 11:32:30, Thomas Weißschuh wrote:
> On Fri, Jul 03, 2026 at 10:26:07AM +0200, Petr Mladek wrote:
> > On Fri 2026-07-03 06:14:48, sashiko-bot@kernel.org wrote:
> > > Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
> > > - [Medium] Incomplete fix and inconsistent state: The patch correctly unsets `sub_make_done` for the `kselftest` targets to allow Kbuild command-line reevaluation when building external kernel modules. However, it fails to apply the same logic to the `tools/` and `tools/%` targets, introducing a discrepancy between build entry points.
> > > --
> > > 
> > > commit 06c418388d65deb3d8494a6e8e7071c0e709186b
> > > Author: Thomas Weißschuh <thomas.weissschuh@linutronix.de>
> > > 
> > > kbuild: unset sub_make_done before calling kselftest build system
> > > 
> > > This commit unsets sub_make_done before invoking the kselftest build system to
> > > ensure Kbuild reevaluates command-line flags when building test modules. It
> > > mirrors a previous fix done for scripts/install.sh.
> > > 
> > > > diff --git a/Makefile b/Makefile
> > > > index b9c5792c79e01..f717a4dc96d6a 100644
> > > > --- a/Makefile
> > > > +++ b/Makefile
> > > > @@ -1591,10 +1591,10 @@ tools/%: FORCE
> > > >  
> > > >  PHONY += kselftest
> > > >  kselftest: headers
> > > > -	$(Q)$(MAKE) -C $(srctree)/tools/testing/selftests run_tests
> > > > +	$(Q)unset sub_make_done; $(MAKE) -C $(srctree)/tools/testing/selftests run_tests
> > > 
> > > [Severity: Medium]
> > > Will make tools/selftests still fail after this change?
> > > 
> > > The tools/ and tools/% targets in Makefile delegate to the tools build
> > > system, which can also trigger external kernel module builds (like bpf_testmod).
> > > 
> > > Because sub_make_done is not unset for these targets, invoking commands like
> > > make tools/selftests or make tools/all will skip Kbuild's M= argument parsing.
> > > 
> > > Should sub_make_done also be unset in the tools/ and tools/% targets to maintain
> > > consistency across the build entry points?
> > 
> > This might make sense. I see that the tools: and tools/%: targets are
> > defined right above the kselftest*: ones. And they are even advertised
> > in "make help" or "make tools/help".
> 
> I am not sure how the tools/ targets can reach back into kbuild.
> The example with bpf_testmod above is bogus, as tools/* is only for
> top-level tools directories. Subdirectories in tools/testing/selftests/
> don't seem to be reachable.

I see.

> > Well it probably should be done in a separate patch. It would be better
> > for bisecting potential regressions. Thomas, Zelin, what do you think, please?
> 
> If we do it for consistency it should be its own patch, agreed.

I would keep it as is until anyone reports a real problem.

Thanks a lot for explanation.

Best Regards,
Petr

Re: [PATCH] kbuild: unset sub_make_done before calling kselftest build system

[Nicolas Schier]

On Fri, Jul 03, 2026 at 12:06:46PM +0200, Petr Mladek wrote:
> On Fri 2026-07-03 11:32:30, Thomas Weißschuh wrote:
> > On Fri, Jul 03, 2026 at 10:26:07AM +0200, Petr Mladek wrote:
> > > On Fri 2026-07-03 06:14:48, sashiko-bot@kernel.org wrote:
> > > > Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
> > > > - [Medium] Incomplete fix and inconsistent state: The patch correctly unsets `sub_make_done` for the `kselftest` targets to allow Kbuild command-line reevaluation when building external kernel modules. However, it fails to apply the same logic to the `tools/` and `tools/%` targets, introducing a discrepancy between build entry points.
> > > > --
> > > > 
> > > > commit 06c418388d65deb3d8494a6e8e7071c0e709186b
> > > > Author: Thomas Weißschuh <thomas.weissschuh@linutronix.de>
> > > > 
> > > > kbuild: unset sub_make_done before calling kselftest build system
> > > > 
> > > > This commit unsets sub_make_done before invoking the kselftest build system to
> > > > ensure Kbuild reevaluates command-line flags when building test modules. It
> > > > mirrors a previous fix done for scripts/install.sh.
> > > > 
> > > > > diff --git a/Makefile b/Makefile
> > > > > index b9c5792c79e01..f717a4dc96d6a 100644
> > > > > --- a/Makefile
> > > > > +++ b/Makefile
> > > > > @@ -1591,10 +1591,10 @@ tools/%: FORCE
> > > > >  
> > > > >  PHONY += kselftest
> > > > >  kselftest: headers
> > > > > -	$(Q)$(MAKE) -C $(srctree)/tools/testing/selftests run_tests
> > > > > +	$(Q)unset sub_make_done; $(MAKE) -C $(srctree)/tools/testing/selftests run_tests
> > > > 
> > > > [Severity: Medium]
> > > > Will make tools/selftests still fail after this change?
> > > > 
> > > > The tools/ and tools/% targets in Makefile delegate to the tools build
> > > > system, which can also trigger external kernel module builds (like bpf_testmod).
> > > > 
> > > > Because sub_make_done is not unset for these targets, invoking commands like
> > > > make tools/selftests or make tools/all will skip Kbuild's M= argument parsing.
> > > > 
> > > > Should sub_make_done also be unset in the tools/ and tools/% targets to maintain
> > > > consistency across the build entry points?
> > > 
> > > This might make sense. I see that the tools: and tools/%: targets are
> > > defined right above the kselftest*: ones. And they are even advertised
> > > in "make help" or "make tools/help".
> > 
> > I am not sure how the tools/ targets can reach back into kbuild.
> > The example with bpf_testmod above is bogus, as tools/* is only for
> > top-level tools directories. Subdirectories in tools/testing/selftests/
> > don't seem to be reachable.
> 
> I see.
> 
> > > Well it probably should be done in a separate patch. It would be better
> > > for bisecting potential regressions. Thomas, Zelin, what do you think, please?
> > 
> > If we do it for consistency it should be its own patch, agreed.
> 
> I would keep it as is until anyone reports a real problem.

Ack.

Thomas, thanks for fix!

Reviewed-by: Nicolas Schier <nsc@kernel.org>


-- 
Nicolas

Re: [PATCH] kbuild: unset sub_make_done before calling kselftest build system

[Nicolas Schier]

On Fri, Jul 03, 2026 at 10:25:15AM +0200, Petr Mladek wrote:
[...]
> 
> PS: I could take this via livepatching tree. But feel free to take it
>     via kbuild tree. I am going to wait few days until the discussion
>     settles down.

Yes, take it via livepatching tree, thanks!

Kind regards,
Nicolas

Re: [PATCH] kbuild: unset sub_make_done before calling kselftest build system

[Zelin Deng]

On Fri, Jul 03, 2026 at 10:25:15AM +0200, Petr Mladek wrote:
> On Fri 2026-07-03 08:04:08, Thomas Weißschuh wrote:
> > The kselftest build system may recourse back into kbuild when building
> > test modules. In that case kbuild needs to parse the new flags passed
> > from the command line, instead of using the ones inherited from the
> > kbuild invocation.
> > 
> > Force that command line reevaluation.
> > 
> > The same was done for scripts/install.sh in commit 14ccc638b02f9ec
> > ("kbuild: cancel sub_make_done for the install target to fix DKMS")
> >
> > Reported-by: Zelin Deng <zelin.deng@linux.alibaba.com>
> > Closes: https://lore.kernel.org/all/20260525083721.27857-1-zelin.deng@linux.alibaba.com/
> 
> It is great that you added references to the related commit and
> the original thread.
> 
> > Fixes: c9bb03ac2c66 ("kbuild: reduce output spam when building out of tree")
> > Signed-off-by: Thomas Weißschuh <thomas.weissschuh@linutronix.de>
> 
> I believe that this is reasonable solution:
> 
> Acked-by: Petr Mladek <pmladek@suse.com>
> 
> Zelin, could you please provide your Tested-by, please?

Sorry for late response, I just started a vacation.
Sure I had verified, and
Tested-by: Zelin Deng <zelin.deng@linux.alibaba.com>

> 
> Best Regards,
> Petr
> 
> PS: I could take this via livepatching tree. But feel free to take it
>     via kbuild tree. I am going to wait few days until the discussion
>     settles down.