Re: linux-next: build failure after merge of the kunit-next tree

[Shuah Khan]

On 8/17/25 20:08, Stephen Rothwell wrote:
> Hi all,
>
> After merging the kunit-next tree, today's linux-next build (x86_64
> allmodconfig) failed like this:

Thank you Stephen. I did a allmodconfig build on 6.17-rc1 base - didn't
see the error.

Marie, David, can you take a look this. Looks like conflict with drm
in next?

>
> In file included from include/kunit/static_stub.h:18,
> from drivers/gpu/drm/xe/xe_bo.c:20:
> drivers/gpu/drm/xe/tests/xe_bo.c:610:48: error: initialization of 'const void * (*)(struct kunit *, const void *, char *)' from incompatible pointer type 'const void * (*)(const void *, char *)' [-Wincompatible-pointer-types]
> 610 | KUNIT_CASE_PARAM(xe_ccs_migrate_kunit, xe_pci_live_device_gen_param),
> | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
> include/kunit/test.h:210:38: note: in definition of macro 'KUNIT_CASE_PARAM'
> 210 | .generate_params = gen_params, .module_name = KBUILD_MODNAME}
> | ^~~~~~~~~~
> drivers/gpu/drm/xe/tests/xe_bo.c:610:48: note: (near initialization for 'xe_bo_tests[0].generate_params')
> 610 | KUNIT_CASE_PARAM(xe_ccs_migrate_kunit, xe_pci_live_device_gen_param),
> | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
> include/kunit/test.h:210:38: note: in definition of macro 'KUNIT_CASE_PARAM'
> 210 | .generate_params = gen_params, .module_name = KBUILD_MODNAME}
> | ^~~~~~~~~~
> drivers/gpu/drm/xe/tests/xe_bo.c:611:45: error: initialization of 'const void * (*)(struct kunit *, const void *, char *)' from incompatible pointer type 'const void * (*)(const void *, char *)' [-Wincompatible-pointer-types]
> 611 | KUNIT_CASE_PARAM(xe_bo_evict_kunit, xe_pci_live_device_gen_param),
> | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
> include/kunit/test.h:210:38: note: in definition of macro 'KUNIT_CASE_PARAM'
> 210 | .generate_params = gen_params, .module_name = KBUILD_MODNAME}
> | ^~~~~~~~~~
> drivers/gpu/drm/xe/tests/xe_bo.c:611:45: note: (near initialization for 'xe_bo_tests[1].generate_params')
> 611 | KUNIT_CASE_PARAM(xe_bo_evict_kunit, xe_pci_live_device_gen_param),
> | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
> include/kunit/test.h:210:38: note: in definition of macro 'KUNIT_CASE_PARAM'
> 210 | .generate_params = gen_params, .module_name = KBUILD_MODNAME}
> | ^~~~~~~~~~
> drivers/gpu/drm/xe/tests/xe_bo.c:624:51: error: initialization of 'const void * (*)(struct kunit *, const void *, char *)' from incompatible pointer type 'const void * (*)(const void *, char *)' [-Wincompatible-pointer-types]
> 624 | KUNIT_CASE_PARAM_ATTR(xe_bo_shrink_kunit, xe_pci_live_device_gen_param,
> | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
> include/kunit/test.h:223:38: note: in definition of macro 'KUNIT_CASE_PARAM_ATTR'
> 223 | .generate_params = gen_params, \
> | ^~~~~~~~~~
> drivers/gpu/drm/xe/tests/xe_bo.c:624:51: note: (near initialization for 'xe_bo_shrink_test[0].generate_params')
> 624 | KUNIT_CASE_PARAM_ATTR(xe_bo_shrink_kunit, xe_pci_live_device_gen_param,
> | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
> include/kunit/test.h:223:38: note: in definition of macro 'KUNIT_CASE_PARAM_ATTR'
> 223 | .generate_params = gen_params, \
> | ^~~~~~~~~~
>
> Caused by commit
>
> 444be9072fca ("kunit: Pass parameterized test context to generate_params()")
>
> I have used the kunit-next tree from next-20250815 for today.
>

thanks,
-- Shuah

[David Gow]

On Tue, 19 Aug 2025 at 00:32, Shuah Khan <sk...@linuxfoundation.org> wrote:
>
> On 8/17/25 20:08, Stephen Rothwell wrote:
> > Hi all,
> >
> > After merging the kunit-next tree, today's linux-next build (x86_64
> > allmodconfig) failed like this:
>
> Thank you Stephen. I did a allmodconfig build on 6.17-rc1 base - didn't
> see the error.
>
> Marie, David, can you take a look this. Looks like conflict with drm
> in next?
>

Thanks, Shuah. I've managed to reproduce this with:
./tools/testing/kunit/kunit.py run --arch x86_64 --kunitconfig
drivers/gpu/drm/xe

These patches fix it (and a corresponding drm/xe test failure):
https://lore.kernel.org/linux-next/20250819073434.1...@google.com/T/#t

Ideally, they'll be squashed into the corresponding patches, as
otherwise there'd be some temporary breakage during bisections. I can
squash these into the original series and re-send it out if that works
best for you.

Cheers,
-- David

[Stephen Rothwell]

Hi all,

I am still seeing this build failure.

--
Cheers,
Stephen Rothwell

[Shuah Khan]

On 8/19/25 01:44, David Gow wrote:
> On Tue, 19 Aug 2025 at 00:32, Shuah Khan <sk...@linuxfoundation.org> wrote:
>>
>> On 8/17/25 20:08, Stephen Rothwell wrote:
>>> Hi all,
>>>
>>> After merging the kunit-next tree, today's linux-next build (x86_64
>>> allmodconfig) failed like this:
>>
>> Thank you Stephen. I did a allmodconfig build on 6.17-rc1 base - didn't
>> see the error.
>>
>> Marie, David, can you take a look this. Looks like conflict with drm
>> in next?
>>
>
> Thanks, Shuah. I've managed to reproduce this with:
> ./tools/testing/kunit/kunit.py run --arch x86_64 --kunitconfig
> drivers/gpu/drm/xe
>
> These patches fix it (and a corresponding drm/xe test failure):
> https://lore.kernel.org/linux-next/20250819073434.1...@google.com/T/#t
>
> Ideally, they'll be squashed into the corresponding patches, as
> otherwise there'd be some temporary breakage during bisections. I can
> squash these into the original series and re-send it out if that works
> best for you.
>

David,

Please squash them and resend - also I see a kernel test robot
error in patch 1/2.

I was going to squash them, but I saw the kernel test robot error patch.

thanks,
-- Shuah

[David Gow]

Thanks, Shuah.

A v2 of the fix series, with the kernel test robot error fixed, is
here: https://lore.kernel.org/linux-kselftest/20250821135447.1...@google.com/

I've also squashed the fixes into a v4 of the original series here:
https://lore.kernel.org/linux-kselftest/20250826091341.1...@google.com/

Sorry for the mess!
-- David

[Shuah Khan]

I applied these to kunit next and ran test:

./tools/testing/kunit/kunit.py run --arch x86_64 --kunitconfig drivers/gpu/drm/xe

Looks good. Hopefully next is happy now.

thanks,
-- Shuah

[David Gow]

Thanks very much -- this is still passing all of my KUnit test runs
(save for a couple of preexisting genirq-related test failures on some
architectures, for which there are already patches en-route to the irq
folks).

-- David