Re: [PATCH] mm, slab: clean up slab->obj_exts always

[Zhenhua Huang <quic_zhenhuah@quicinc.com>]

Thanks Suren.

On 2025/4/19 7:09, Suren Baghdasaryan wrote:
> On Fri, Apr 18, 2025 at 8:17 AM Harry Yoo <harry.yoo@oracle.com> wrote:
>>
>> On Fri, Apr 18, 2025 at 02:14:59PM +0800, Zhenhua Huang wrote:
>>> When memory allocation profiling is disabled at runtime or due to an
>>> error, shutdown_mem_profiling() is called: slab->obj_exts which
>>> previously allocated remains.
>>> It won't be cleared by unaccount_slab() because of
>>> mem_alloc_profiling_enabled() not true. It's incorrect, slab->obj_exts
>>> should always be cleaned up in unaccount_slab() to avoid following error:
>>>
>>> [...]BUG: Bad page state in process...
>>> ..
>>> [...]page dumped because: page still charged to cgroup
>>>
>>> Fixes: 21c690a349baa ("mm: introduce slabobj_ext to support slab object extensions")
>>> Signed-off-by: Zhenhua Huang <quic_zhenhuah@quicinc.com>
> 
> Thanks for reporting and fixing the issue!
> 
>>> ---
>>
>> Acked-by: Harry Yoo <harry.yoo@oracle.com>
>>
>> I reproduced the issue locally and confirmed that this patch fixes
>> the issue.
>>
>> Tested-by: Harry Yoo <harry.yoo@oracle.com>
>>
>> By the way, I think this should probably be backported to -stable?
>>
>> --
>> Cheers,
>> Harry / Hyeonggon
>>
>>>   mm/slub.c | 8 ++++++--
>>>   1 file changed, 6 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/mm/slub.c b/mm/slub.c
>>> index dac149df1be1..b42ce3a88806 100644
>>> --- a/mm/slub.c
>>> +++ b/mm/slub.c
>>> @@ -2023,7 +2023,7 @@ int alloc_slab_obj_exts(struct slab *slab, struct kmem_cache *s,
>>>        return 0;
>>>   }
>>>
>>> -/* Should be called only if mem_alloc_profiling_enabled() */
>>> +/* Should be called if slab_obj_exts(slab) */
>>>   static noinline void free_slab_obj_exts(struct slab *slab)
>>>   {
>>>        struct slabobj_ext *obj_exts;
>>> @@ -2592,7 +2592,11 @@ static __always_inline void account_slab(struct slab *slab, int order,
>>>   static __always_inline void unaccount_slab(struct slab *slab, int order,
>>>                                           struct kmem_cache *s)
>>>   {
>>> -     if (memcg_kmem_online() || need_slab_obj_ext())
>>> +     /*
>>> +      * The slab object extensions should now be freed regardless of
>>> +      * whether mem_alloc_profiling_enabled() or not now.
> 
> This comment does not explain why. I amended it in my suggestion below.

I will follow your suggestion.

> 
>>> +      */
>>> +     if (memcg_kmem_online() || slab_obj_exts(slab))
>>>                free_slab_obj_exts(slab);
> 
> free_slab_obj_exts() will be checking again that for
> slab_obj_exts(slab) != NULL. Since this change effectively removes the
> static key check (mem_alloc_profiling_enabled() call inside
> need_slab_obj_ext()), I think we can simply make free_slab_obj_exts()
> inline function and remove the above condition completely. IOW:
> 

Got it. I'll send another version that includes these changes.

> static inline void free_slab_obj_exts(struct slab *slab)
> {
>          struct slabobj_ext *obj_exts;
> 
>          obj_exts = slab_obj_exts(slab);
>          if (!obj_exts)
>                  return;
>          ...
>          slab->obj_exts = 0;
> }
> 
> static __always_inline void unaccount_slab(...)
> {
>       /*
>        * The slab object extensions should be freed regardless of
>        * whether mem_alloc_profiling_enabled() or not because profiling
>        * might have been disabled after slab->obj_exts got allocated.
>        */
>        free_slab_obj_exts(slab);
>        ...
> }
> 
>>>
>>>        mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s),
>>> --
>>> 2.25.1
>>>
>>>