Re: [PATCH] mm/memcg: Fix incorrect flushing of lruvec data in obj_stock

From: Michal Hocko <mhocko@suse.com>
To: Waiman Long <longman@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>,
	Vladimir Davydov <vdavydov.dev@gmail.com>,
	Andrew Morton <akpm@linux-foundation.org>,
	Tejun Heo <tj@kernel.org>, Christoph Lameter <cl@linux.com>,
	Pekka Enberg <penberg@kernel.org>,
	David Rientjes <rientjes@google.com>,
	Joonsoo Kim <iamjoonsoo.kim@lge.com>,
	Vlastimil Babka <vbabka@suse.cz>, Roman Gushchin <guro@fb.com>,
	linux-kernel@vger.kernel.org, cgroups@vger.kernel.org,
	linux-mm@kvack.org, Shakeel Butt <shakeelb@google.com>,
	Muchun Song <songmuchun@bytedance.com>,
	Alex Shi <alex.shi@linux.alibaba.com>,
	Chris Down <chris@chrisdown.name>,
	Yafang Shao <laoar.shao@gmail.com>,
	Wei Yang <richard.weiyang@gmail.com>,
	Masayoshi Mizuma <msys.mizuma@gmail.com>,
	Xing Zhengjun <zhengjun.xing@linux.intel.com>,
	Matthew Wilcox <willy@infradead.org>
Subject: Re: [PATCH] mm/memcg: Fix incorrect flushing of lruvec data in obj_stock
Date: Mon, 2 Aug 2021 08:28:13 +0200	[thread overview]
Message-ID: <YQeQfX3t8k+U3MIL@dhcp22.suse.cz> (raw)
In-Reply-To: <20210802022827.10192-1-longman@redhat.com>

On Sun 01-08-21 22:28:27, Waiman Long wrote:
> When mod_objcg_state() is called with a pgdat that is different from
> that in the obj_stock, the old lruvec data cached in obj_stock are
> flushed out. Unfortunately, they were flushed to the new pgdat and
> hence the wrong node, not the one cached in obj_stock.

It would be great to explicitly mention user observable problems here. I
do assume this will make slab stats skewed but the effect wouldn't be
very big, right?

> Fix that by flushing the data to the cached pgdat instead.
> 
> Fixes: 68ac5b3c8db2 ("mm/memcg: cache vmstat data in percpu memcg_stock_pcp")
> Signed-off-by: Waiman Long <longman@redhat.com>

Acked-by: Michal Hocko <mhocko@suse.com>

> ---
>  mm/memcontrol.c | 8 +++++---
>  1 file changed, 5 insertions(+), 3 deletions(-)
> 
> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> index ae1f5d0cb581..881ec4ddddcd 100644
> --- a/mm/memcontrol.c
> +++ b/mm/memcontrol.c
> @@ -3106,17 +3106,19 @@ void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat,
>  		stock->cached_pgdat = pgdat;
>  	} else if (stock->cached_pgdat != pgdat) {
>  		/* Flush the existing cached vmstat data */
> +		struct pglist_data *oldpg = stock->cached_pgdat;
> +
> +		stock->cached_pgdat = pgdat;
>  		if (stock->nr_slab_reclaimable_b) {
> -			mod_objcg_mlstate(objcg, pgdat, NR_SLAB_RECLAIMABLE_B,
> +			mod_objcg_mlstate(objcg, oldpg, NR_SLAB_RECLAIMABLE_B,
>  					  stock->nr_slab_reclaimable_b);
>  			stock->nr_slab_reclaimable_b = 0;
>  		}
>  		if (stock->nr_slab_unreclaimable_b) {
> -			mod_objcg_mlstate(objcg, pgdat, NR_SLAB_UNRECLAIMABLE_B,
> +			mod_objcg_mlstate(objcg, oldpg, NR_SLAB_UNRECLAIMABLE_B,
>  					  stock->nr_slab_unreclaimable_b);
>  			stock->nr_slab_unreclaimable_b = 0;
>  		}
> -		stock->cached_pgdat = pgdat;

Minor nit. Is there any reason to move the cached_pgdat? TBH I found the
original way better from the readability POV.

>  	}
>  
>  	bytes = (idx == NR_SLAB_RECLAIMABLE_B) ? &stock->nr_slab_reclaimable_b
> -- 
> 2.18.1

-- 
Michal Hocko
SUSE Labs