Re: [PATCH bpf] bpf: Make migrate_{disable,enable} always inline if in header file

[Yonghong Song <yonghong.song@linux.dev>]

On 10/30/25 3:53 AM, Peter Zijlstra wrote:
> On Wed, Oct 29, 2025 at 06:13:21PM -0700, Alexei Starovoitov wrote:
>> On Wed, Oct 29, 2025 at 11:37 AM Yonghong Song <yonghong.song@linux.dev> wrote:
>>> With latest bpf/bpf-next tree and latest pahole master, I got the following
>>> build failure:
>>>
>>>    $ make LLVM=1 -j
>>>      ...
>>>      LD      vmlinux.o
>>>      GEN     .vmlinux.objs
>>>      ...
>>>      BTF     .tmp_vmlinux1.btf.o
>>>      ...
>>>      AS      .tmp_vmlinux2.kallsyms.o
>>>      LD      vmlinux.unstripped
>>>      BTFIDS  vmlinux.unstripped
>>>    WARN: resolve_btfids: unresolved symbol migrate_enable
>>>    WARN: resolve_btfids: unresolved symbol migrate_disable
>>>    make[2]: *** [/home/yhs/work/bpf-next/scripts/Makefile.vmlinux:72: vmlinux.unstripped] Error 255
>>>    make[2]: *** Deleting file 'vmlinux.unstripped'
>>>    make[1]: *** [/home/yhs/work/bpf-next/Makefile:1242: vmlinux] Error 2
>>>    make: *** [/home/yhs/work/bpf-next/Makefile:248: __sub-make] Error 2
>>>
>>> In pahole patch [1], if two functions having identical names but different
>>> addresses, then this function name is considered ambiguous and later on
>>> this function will not be added to vmlinux/module BTF.
>>>
>>> Commit 378b7708194f ("sched: Make migrate_{en,dis}able() inline") changed
>>> original global funcitons migrate_{enable,disable} to
>>>    - in kernel/sched/core.c, migrate_{enable,disable} are global funcitons.
>>>    - in other places, migrate_{enable,disable} may survive as static functions
>>>      since they are marked as 'inline' in include/linux/sched.h and the
>>>      'inline' attribute does not garantee inlining.
>>>
>>> If I build with clang compiler (make LLVM=1 -j) (llvm21 and llvm22), I found
>>> there are four symbols for migrate_{enable,disable} respectively, three
>>> static functions and one global function. With the above pahole patch [1],
>>> migrate_{enable,disable} are not in vmlinux BTF and this will cause
>>> later resolve_btfids failure.
>>>
>>> Making migrate_{enable,disable} always inline in include/linux/sched.h
>>> can fix the problem.
>>>
>>>    [1] https://lore.kernel.org/dwarves/79a329ef-9bb3-454e-9135-731f2fd51951@oracle.com/
>>>
>>> Fixes: 378b7708194f ("sched: Make migrate_{en,dis}able() inline")
>>> Cc: Menglong Dong <menglong8.dong@gmail.com>
>>> Cc: Ihor Solodrai <ihor.solodrai@linux.dev>
>>> Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
>>> ---
>>>   include/linux/sched.h | 4 ++--
>>>   1 file changed, 2 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/include/linux/sched.h b/include/linux/sched.h
>>> index cbb7340c5866..b469878de25c 100644
>>> --- a/include/linux/sched.h
>>> +++ b/include/linux/sched.h
>>> @@ -2407,12 +2407,12 @@ static inline void __migrate_enable(void) { }
>>>    * be defined in kernel/sched/core.c.
>>>    */
>>>   #ifndef INSTANTIATE_EXPORTED_MIGRATE_DISABLE
>>> -static inline void migrate_disable(void)
>>> +static __always_inline void migrate_disable(void)
>>>   {
>>>          __migrate_disable();
>>>   }
>>>
>>> -static inline void migrate_enable(void)
>>> +static __always_inline void migrate_enable(void)
>>>   {
>>>          __migrate_enable();
>>>   }
>> Peter,
>>
>> Are you ok if we take this?
> Yes, but WTH would clang not inline this trivial function to begin with?

I checked asm codes with migrate_disable(). In the above cases, 
__migrate_disable() is inlined and the function body of 
migrate_disable() then becomes reasonably big. The caller of 
migrate_disable() are ring_buffer_resize()*/*range_tree_set()/... which 
are pretty big too.