Re: [PATCH net] Revert "kunit: configs: Enable CONFIG_INIT_STACK_ALL_PATTERN in all_tests"

[Shuah Khan <skhan@linuxfoundation.org>]

On 6/2/25 07:34, Richard Fitzgerald wrote:
> On 31/05/2025 2:07 am, Jakub Kicinski wrote:
>> On Fri, 30 May 2025 10:59:54 -0600 Shuah Khan wrote:
>>> On 5/30/25 07:58, Jakub Kicinski wrote:
>>>> This reverts commit a571a9a1b120264e24b41eddf1ac5140131bfa84.
>>>>
>>>> The commit in question breaks kunit for older compilers:
>>>>> $ gcc --version
>>>>    gcc (GCC) 11.5.0 20240719 (Red Hat 11.5.0-5)
>>>>
>>>> $ ./tools/testing/kunit/kunit.py run  --alltests --json --arch=x86_64
>>>>    Configuring KUnit Kernel ...
>>>>    Regenerating .config ...
>>>>    Populating config with:
>>>>    $ make ARCH=x86_64 O=.kunit olddefconfig
>>>
>>>
>>>>    ERROR:root:Not all Kconfig options selected in kunitconfig were in the generated .config.
>>>>    This is probably due to unsatisfied dependencies.
>>>>    Missing: CONFIG_INIT_STACK_ALL_PATTERN=y
>>>
>>> Does adding config option work for you?
>>> ./tools/testing/kunit/kunit.py run --kconfig_add CONFIG_INIT_STACK_ALL_PATTERN
>>
>> Nope (with this patch applied):
>>
>> $ ./tools/testing/kunit/kunit.py run --kconfig_add CONFIG_INIT_STACK_ALL_PATTERN=y
>> [18:02:47] Configuring KUnit Kernel ...
>> Regenerating .config ...
>> Populating config with:
>> $ make ARCH=um O=.kunit olddefconfig
>> ERROR:root:Not all Kconfig options selected in kunitconfig were in the generated .config.
>> This is probably due to unsatisfied dependencies.
>> Missing: CONFIG_INIT_STACK_ALL_PATTERN=y
>> Note: many Kconfig options aren't available on UML. You can try running on a different architecture with something like "--arch=x86_64".
>>
>>>> Link: https://lore.kernel.org/20250529083811.778bc31b@kernel.org
>>>> Fixes: a571a9a1b120 ("kunit: configs: Enable CONFIG_INIT_STACK_ALL_PATTERN in all_tests")
>>>> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
>>>> ---
>>>> I'd like to take this in via netdev since it fixes our CI.
>>>> We'll send it to Linus next week.
>>>
>>> I am good with reverting it for now.
>>>
>>> David, Brendan,
>>> We will have to enable this at a later time. Also we saw this problem
>>> before with other configs. Anyway way to fix this for alltests case?
>>
>> FWIW Richard commented in the linked thread, IIUC this was just for
>> added coverage but not a hard requirement.
> Correct. It's not required (for me). It found a bug in my code, so it
> seemed useful to have enabled while testing. I thought this was safe to
> do, I didn't know that this only works with certain platforms and
> compilers.

Thanks for clarifying. Looks good to me.

thanks,
-- Shuah