Re: [PATCH] mm/memcg: bail out early when !memcg in mem_cgroup_lruvec

From: Dmitry Osipenko
Date: Mon Nov 30 2020 - 14:45:07 EST


29.11.2020 07:43, Alex Shi пишет:
>
>
> 在 2020/11/28 下午12:02, Andrew Morton 写道:
>> On Fri, 27 Nov 2020 11:08:35 +0800 Alex Shi <alex.shi@xxxxxxxxxxxxxxxxx> wrote:
>>
>>> Sometime, we use NULL memcg in mem_cgroup_lruvec(memcg, pgdat)
>>> so we could get out early in the situation to avoid useless checking.
>>>
>>> Also warning if both parameter are NULL.
>>
>> Why do you think a warning is needed here?
>
> Uh, Consider there are no problem for long time, it could be saved.
>
>>
>>> --- a/include/linux/memcontrol.h
>>> +++ b/include/linux/memcontrol.h
>>> @@ -613,14 +613,13 @@ static inline struct lruvec *mem_cgroup_lruvec(struct mem_cgroup *memcg,
>>> struct mem_cgroup_per_node *mz;
>>> struct lruvec *lruvec;
>>>
>>> - if (mem_cgroup_disabled()) {
>>> + VM_WARN_ON_ONCE(!memcg && !pgdat);
>>> +
>>> + if (mem_cgroup_disabled() || !memcg) {
>>> lruvec = &pgdat->__lruvec;
>>> goto out;
>>> }
>>>
>>> - if (!memcg)
>>> - memcg = root_mem_cgroup;
>>> -
>>
>> This change isn't obviously equivalent, is it?
>
> If !memcg, the root_mem_cgroup will still lead the lruvec to a pgdat
> same as parameter.
>
>>
>>> mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
>>> lruvec = &mz->lruvec;
>>> out:
>>
>> And the resulting code is awkward:
>>
>> if (mem_cgroup_disabled() || !memcg) {
>> lruvec = &pgdat->__lruvec;
>> goto out;
>> }
>>
>> mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
>> lruvec = &mz->lruvec;
>> out:
>>
>>
>> could be
>>
>> if (mem_cgroup_disabled() || !memcg) {
>> lruvec = &pgdat->__lruvec;
>> } else {
>> mem_cgroup_per_node mz;
>>
>> mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
>> lruvec = &mz->lruvec;
>> }
>>
>
> Right. remove 'goto' is better for understander.
>
> So, is the following patch ok?
>
> From 225f29e03b40a7cbaeb4e3bb76f8efbcd7d648a2 Mon Sep 17 00:00:00 2001
> From: Alex Shi <alex.shi@xxxxxxxxxxxxxxxxx>
> Date: Wed, 25 Nov 2020 14:06:33 +0800
> Subject: [PATCH v2] mm/memcg: bail out early when !memcg in mem_cgroup_lruvec
>
> Sometime, we use NULL memcg in mem_cgroup_lruvec(memcg, pgdat)
> so we could get out early in the situation to avoid useless checking.
>
> Polished as Andrew Morton's suggestion.
>
> Signed-off-by: Alex Shi <alex.shi@xxxxxxxxxxxxxxxxx>
> Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> Cc: Johannes Weiner <hannes@xxxxxxxxxxx>
> Cc: Shakeel Butt <shakeelb@xxxxxxxxxx>
> Cc: Roman Gushchin <guro@xxxxxx>
> Cc: Lorenzo Stoakes <lstoakes@xxxxxxxxx>
> Cc: Stephen Rothwell <sfr@xxxxxxxxxxxxxxxx>
> Cc: Alexander Duyck <alexander.h.duyck@xxxxxxxxxxxxxxx>
> Cc: Yafang Shao <laoar.shao@xxxxxxxxx>
> Cc: Wei Yang <richard.weiyang@xxxxxxxxx>
> Cc: linux-kernel@xxxxxxxxxxxxxxx
> ---
> include/linux/memcontrol.h | 15 ++++++---------
> 1 file changed, 6 insertions(+), 9 deletions(-)
>
> diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
> index 3e6a1df3bdb9..4ff2ffe2b73d 100644
> --- a/include/linux/memcontrol.h
> +++ b/include/linux/memcontrol.h
> @@ -610,20 +610,17 @@ mem_cgroup_nodeinfo(struct mem_cgroup *memcg, int nid)
> static inline struct lruvec *mem_cgroup_lruvec(struct mem_cgroup *memcg,
> struct pglist_data *pgdat)
> {
> - struct mem_cgroup_per_node *mz;
> struct lruvec *lruvec;
>
> - if (mem_cgroup_disabled()) {
> + if (mem_cgroup_disabled() || !memcg) {
> lruvec = &pgdat->__lruvec;
> - goto out;
> - }
> + } else {
> + struct mem_cgroup_per_node *mz;
>
> - if (!memcg)
> - memcg = root_mem_cgroup;
> + mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
> + lruvec = &mz->lruvec;
> + }
>
> - mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
> - lruvec = &mz->lruvec;
> -out:
> /*
> * Since a node can be onlined after the mem_cgroup was created,
> * we have to be prepared to initialize lruvec->pgdat here;
>

Hi,

This patch causes a hard lock on one of my ARM32 devices using today's
linux-next, please fix.