Re: [PATCH bpf-next] kallsyms: move module-related functions under correct configs

[Viktor Malik <vmalik@redhat.com>]

On 3/28/23 10:23, Daniel Borkmann wrote:
> On 3/27/23 9:28 PM, Luis Chamberlain wrote:
>> On Mon, Mar 27, 2023 at 08:20:56PM +0200, Viktor Malik wrote:
>>> On 3/27/23 19:47, Luis Chamberlain wrote:
>>>> On Mon, Mar 27, 2023 at 06:12:51PM +0200, Viktor Malik wrote:
>>>>> Functions for searching module kallsyms should have non-empty
>>>>> definitions only if CONFIG_MODULES=y and CONFIG_KALLSYMS=y. Until now,
>>>>> only CONFIG_MODULES check was used for many of these, which may have
>>>>> caused complilation errors on some configs.
>>>>>
>>>>> This patch moves all relevant functions under the correct configs.
>>>>>
>>>>> Signed-off-by: Viktor Malik <vmalik@redhat.com>
>>>>> Reported-by: kernel test robot <lkp@intel.com>
>>>>> Link: https://lore.kernel.org/oe-kbuild-all/202303181535.RFDCnz3E-lkp@intel.com/
>>>>
>>>> Thanks Viktor!  Does this fix something from an existing commit? If so
>>>> which one?  The commit log should mention it.
>>>
>>> Ah, right, I forgot about that. The commit log is missing:
>>>
>>> Fixes: bd5314f8dd2d ("kallsyms, bpf: Move find_kallsyms_symbol_value out of internal header")
>>>
>>> I can post v2 but I'm also fine with maintainers applying the tag.
>>
>> That patch went through the bpf tree so its fix can go throug that tree.
>> So up to Daniel if he wants a new patch.
> 
> Fixing up is fine with me. Viktor, which config combinations did you test for this
> patch and under which architectures?
> 
> I suspect kbuild bot might still complain. For example, your patch moves
> dereference_module_function_descriptor() stub definition under !CONFIG_MODULES ||
> !CONFIG_KALLSYMS. Looking at ppc's dereference_module_function_descriptor()
> implementation (!CONFIG_PPC64_ELF_ABI_V2 case), it's build under CONFIG_MODULES,
> so you'll have two different implementations of the functions, the generic one
> then w/o __weak attribute.

Right, I didn't notice this one, thanks. Looks like
dereference_module_function_descriptor will need to stay in the original
place (under CONFIG_MODULES only). I'll cross-compile for arches that
redefine this function (powerpc, ia64, parisc) with relevant config
combinations and submit v2. The rest of the functions don't have
arch-dependent implementations, so they should be fine.

> 
> Also small nit, please fix the patch up wrt indentation to align with kernel coding
> style.

Ok, will do.

Thanks,
Viktor

> 
> Thanks,
> Daniel
>