Re: [PATCH v6 31/33] mm: memcontrol: convert objcg to be per-memcg per-node type

[Qi Zheng <qi.zheng@linux.dev>]

On 3/7/26 7:08 PM, Usama Arif wrote:
> 
> 
> On 07/03/2026 08:51, Qi Zheng wrote:
>> Hi Usama,
>>
>> On 3/7/26 4:29 AM, Usama Arif wrote:
>>> On Thu,  5 Mar 2026 19:52:49 +0800 Qi Zheng <qi.zheng@linux.dev> wrote:
>>>
>>>> From: Qi Zheng <zhengqi.arch@bytedance.com>
>>>>
>>>> Convert objcg to be per-memcg per-node type, so that when reparent LRU
>>>> folios later, we can hold the lru lock at the node level, thus avoiding
>>>> holding too many lru locks at once.
>>>>
>>>> Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
>>>> Acked-by: Shakeel Butt <shakeel.butt@linux.dev>
>>>> ---
>>>>    include/linux/memcontrol.h | 23 +++++------
>>>>    include/linux/sched.h      |  2 +-
>>>>    mm/memcontrol.c            | 79 +++++++++++++++++++++++---------------
>>>>    3 files changed, 62 insertions(+), 42 deletions(-)
>>>>
>>>
>>> [...]
>>>
>>>> @@ -4087,7 +4100,13 @@ static int mem_cgroup_css_online(struct cgroup_subsys_state *css)
>>>>        xa_store(&mem_cgroup_private_ids, memcg->id.id, memcg, GFP_KERNEL);
>>>>          return 0;
>>>> -free_shrinker:
>>>> +free_objcg:
>>>> +    for_each_node(nid) {
>>>> +        struct mem_cgroup_per_node *pn = memcg->nodeinfo[nid];
>>>> +
>>>> +        if (pn && pn->orig_objcg)
>>>> +            obj_cgroup_put(pn->orig_objcg);
>>>
>>> Is it possible that you might call obj_cgroup_put twice on the same cgroup?
>>
>> Oh, I think you are right. Here pn->orig_objcg was not reset to NULL, so
>> obj_cgroup_put() will be called in __mem_cgroup_free() again.
>>
>>>
>>> If css_create fails, css_free_rwork_fn is queued, which ends up calling
>>> mem_cgroup_css_free which calls obj_cgroup_put again?
>>>
>>> Maybe adding pn->orig_objcg = NULL overhere after obj_cgroup_put
>>> is enough to prevent the double put from causing issues?
>>
>> Agree.
>>
>> Like this?
>>
> 
> Yes below looks good! Might be good to add a comment as well why setting
> it to NULL.

OK, will add the following comment:

/*
  * Reset pn->orig_objcg to NULL to prevent obj_cgroup_put()
  * from being called agagin in __mem_cgroup_free().
  */

> 
>> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
>> index 992a3f5caa62b..e0795aec4356b 100644
>> --- a/mm/memcontrol.c
>> +++ b/mm/memcontrol.c
>> @@ -4140,8 +4140,10 @@ static int mem_cgroup_css_online(struct cgroup_subsys_state *css)
>>          for_each_node(nid) {
>>                  struct mem_cgroup_per_node *pn = memcg->nodeinfo[nid];
>>
>> -               if (pn && pn->orig_objcg)
>> +               if (pn && pn->orig_objcg) {
>>                          obj_cgroup_put(pn->orig_objcg);
>> +                       pn->orig_objcg = NULL;
>> +               }
>>          }
>>          free_shrinker_info(memcg);
>>   offline_kmem:
>>
>> If there are no problems, I will send a fix patch later.
>>
>> Thanks,
>> Qi
>>
>>>
>>>> +    }
>>>>        free_shrinker_info(memcg);
>>>>    offline_kmem:
>>>>        memcg_offline_kmem(memcg);
>>>> -- 
>>>> 2.20.1
>>>>
>>>>
>>
>