Re: [PATCH v3] memcg: cache obj_stock by memcg, not by objcg pointer

[Harry Yoo <harry@kernel.org>]

On 5/19/26 11:02 PM, Shakeel Butt wrote:
> On Tue, May 19, 2026 at 03:46:51PM +0900, Harry Yoo wrote:
>>
>>
>> On 5/19/26 8:41 AM, Shakeel Butt wrote:
>>> On Mon, May 18, 2026 at 03:28:27PM -0700, Shakeel Butt wrote:
>>>> Commit 01b9da291c49 ("mm: memcontrol: convert objcg to be per-memcg
>>>> per-node type") split a memcg's single obj_cgroup into one per NUMA
>>>> node, but the per-CPU obj_stock_pcp still keys cached_objcg by
>>>> pointer. Cross-NUMA workloads now see a drain on every refill and a
>>>> miss on every consume that targets a sibling per-node objcg of the
>>>> same memcg, producing the 67.7% stress-ng switch-mq regression
>>>> reported by LKP.
>>>>
>>>> stock->nr_bytes are fungible across per-node objcgs of one memcg.
>>>> Treat the cache as keyed by memcg in __consume_obj_stock() and
>>>> __refill_obj_stock() so siblings share the reserve. Compare via
>>>> READ_ONCE(objcg->memcg) directly: pointer-compare only, no deref, so
>>>> the rcu_read_lock contract on obj_cgroup_memcg() does not apply.
>>>>
>>>> Sharing the reserve without re-caching means bytes funded by one
>>>> per-node objcg's slow path can be consumed/freed under a different
>>>> sibling, leaving sub-page residue on whichever sibling was cached at
>>>> drain time. The pre-existing obj_cgroup_release() path would WARN and
>>>> silently drop that residue, leaking up to nr_node_ids * (PAGE_SIZE - 1)
>>>> bytes per memcg lifecycle from the page_counter. Forward the residue
>>>> into a per-node objcg of the same (post-reparent) memcg at release time
>>>> instead, so it can be reconciled later via a refill atomic_xchg or
>>>> another release; the chain terminates at root_mem_cgroup, whose
>>>> page_counter has no enforced limit.
>>>>
>>>> Please note that this is temporary fix and will be reverted when
>>>> per-node kmem accounting is introduced.
>>
>> ... because once per-node kmem accounting is introduced,
>> "stock->nr_bytes are fungible across per-node objcgs of one memcg"
>> no longer holds?
> 
> Yes
> 
>>
>> And the follow-up plain is to revert this and address it with a multi-objcg
>> percpu stock [1], similar to a multi-memcg percpu charge cache we have now,
>> right? (regardless of per-node kmem accounting's progress)
>>
> 
> Yes

Thanks for confirming!

>> If this temporary fix imposes other potential correctness issues, would it
>> make sense to land [1] in mainline before the next LTS release and skip this
>> temporary fix?
>>
>> [1] https://lore.kernel.org/oe-lkp/agtPMpQK2jXdQAY4@linux.dev
>>
> 
> The full clean solution might take one more cycle and I think we can not just
> ignore 67% regression on 7.1.

That is valid point, unfortunately.

One more thing I have to ask... for v7.1, wouldn't it be a safer option 
to revert the per-node object change and re-introduce it once we have a 
cleaner solution?

This change was introduced in v5, but the implementation before v4 had 
been exposed in -next for a while, and I think we don't have enough 
justification to keep the per-node objcgs change, at least for v7.1, 
given that we have an unexpected last-minute regression and
correctness concerns (albeit slight).

-- 
Cheers,
Harry / Hyeonggon