Re: [PATCH] mm/bpf_memcontrol: mark BPF memcg kfuncs static

[Alexei Starovoitov]

On Wed Jun 17, 2026 at 3:31 PM PDT, Roman Gushchin wrote:
> Alexei Starovoitov <alexei.starovoitov@xxxxxxxxx> writes:
>
>> On Wed, Jun 17, 2026 at 1:22 PM JP Kobryn <jp.kobryn@xxxxxxxxx> wrote:
>>>
>>> The kfuncs in bpf_memcontrol.c are not called by in-kernel C code. They are
>>> only referenced by BPF programs and resolved through BTF.
>>>
>>> Since no external linkage is needed, mark these functions static.
>>>
>>> Fixes: 5904db9891f8 ("mm: introduce BPF kfuncs to deal with memcg pointers")
>>> Fixes: 5c7db3239c9f ("mm: introduce bpf_get_root_mem_cgroup() BPF kfunc")
>>> Fixes: 99430ab8b804 ("mm: introduce BPF kfuncs to access memcg statistics and events")
>>> Reported-by: kernel test robot <lkp@xxxxxxxxx>
>>> Closes: https://lore.kernel.org/oe-kbuild-all/202606150332.Xy4Egd9s-lkp@xxxxxxxxx/
>>> Signed-off-by: JP Kobryn <jp.kobryn@xxxxxxxxx>
>>> ---
>>>  mm/bpf_memcontrol.c | 21 +++++++++++----------
>>>  1 file changed, 11 insertions(+), 10 deletions(-)
>>>
>>> diff --git a/mm/bpf_memcontrol.c b/mm/bpf_memcontrol.c
>>> index 716df49d7647..eff79bb2c758 100644
>>> --- a/mm/bpf_memcontrol.c
>>> +++ b/mm/bpf_memcontrol.c
>>> @@ -20,7 +20,7 @@ __bpf_kfunc_start_defs();
>>>   *
>>>   * Return: A pointer to the root memory cgroup.
>>>   */
>>> -__bpf_kfunc struct mem_cgroup *bpf_get_root_mem_cgroup(void)
>>> +__bpf_kfunc static struct mem_cgroup *bpf_get_root_mem_cgroup(void)
>>
>> sashiko disappoints. It couldn't notice that kfuncs should not be static.
>> We have few "__bpf_kfunc static" in net/ipv4/
>> that work by sort-of "luck", since their addresses are taken.
>>
>> In general kfuncs cannot be declared static.
>
> It seems like it thought hard on it but failed based on the existing
> code and a bit vague documentation in Documentation/bpf/kfuncs.rst:
>
>   kfunc definitions should also always be annotated with the ``__bpf_kfunc``
>   macro. This prevents issues such as the compiler inlining the kfunc if it's a
>   static kernel function, or the function ...
>
> So it reads like bpf kfuncs can be static.
>
> This is the reasoning from the logs (a dismissed concern, to be precise):
> {
>       "type": "Dead Code / Linker Error",
>
>       "description": "Marking kfuncs as static could cause the compiler
>       to optimize them away, breaking BPF verifier BTF resolution.",
>
>       "reasoning": "The patch adds `static` to multiple `__bpf_kfunc`
>       annotated functions that are not called anywhere in C
>       code. Normally, a compiler would remove unused static functions or
>       fail to expose them in object files, which could prevent the BTF
>       ID generation logic (`resolve_btfids`) from finding them. However,
>       the `__bpf_kfunc` macro expands to include `__used` and
>       `__retain`, which strictly instructs the compiler and linker to
>       preserve the functions in the object file regardless of
>       usage. Additionally, `resolve_btfids` and `BTF_ID_FLAGS`
>       successfully parse and resolve `STB_LOCAL` static symbols, making
>       this change completely safe and correct.",

Almost, except it's not clear what takes priority for the compiler.
__bpf_kfunc used to have '__used' and it was enough until clang
got too aggressive in LTO mode. So we added '__retain' and so far
it seems to be holding, but sashiko is missing that both of these
attributes don't say anything about _name_ of the function.
'static' keyword means that compiler is free to rename it and
compilers do take advantage of that from time to time,
while __used and __retain don't add 'do-not-rename' restriction
to the compiler.

Also look at it from the other angle.. why anyone would add
'static' when the function is clearly not static to this particular
compilation unit. It will be used outside of it by bpf progs
that are obviuosly not part of this .c file.

Documentation/bpf/kfuncs.rst needs to be updated.

JP,
pls send a patch to update the docs and include few words
to ignore sparse.