On Sun, Apr 14, 2024 at 8:30 PM Barry Song <21cnbao@xxxxxxxxx> wrote:
>
> On Mon, Apr 15, 2024 at 3:11 PM <xu.xin16@xxxxxxxxxx> wrote:
> >
> > From: Ran Xiaokai <ran.xiaokai@xxxxxxxxxx>
> >
> > Since commit 87eaceb3faa5 ("mm: thp: make deferred split shrinker
> > memcg aware"), the THP deferred split queue is per memcg but not
> > per mem_cgroup_per_node. This has two aspects of impact:
> >
> > Impact1: for kswapd reclaim
> > =====================
> > kswapd
> > balance_pgdat
> > kswapd_shrink_node
> > shrink_node(pgdat, sc);
> > shrink_node_memcgs(pgdat, sc);
> > shrink_slab(sc->gfp_mask, pgdat->node_id, memcg...);
> > the parameter "pgdat->node_id" does not take effectct for
> > THP deferred_split_shrinker, as the deferred_split_queue of
> > specified memcg is not for a certain numa node but for all the nodes.
> > We want to makes the memcg THP deferred split shrinker aware of
> > node_id.
> >
> > Impact2: thp-deferred_split shrinker debugfs interface
> > =========================================
> > for the "count" file:
> > <cgroup inode id> <objects on node 0> <objects on node 1>
> > the output is acctually the sum of all numa nodes.
> > for the "scan" file:
> > <cgroup inode id> <numa id> <number of objects to scan>
> > Also the "numa id" input does not take effect here.
> >
> > This patch makes memcg deferred_split_queue per mem_cgroup_per_node
> > so try to conform to semantic logic.
I used to have a similar patch before,
https://lore.kernel.org/linux-mm/1569968203-64647-1-git-send-email-yang.shi@xxxxxxxxxxxxxxxxx/
But it was somehow lost in discussion.
I have no objection to this patch. However, I was thinking about using
list_lru for deferred split queue, but I didn't have time to look
deeper. Maybe we should try now?
>
> This seems to be a correct fix to me, + Yang Shi, the original author of
> commit 87eaceb3faa5.
>
> >
> > Reviewed-by: Lu Zhongjun <lu.zhongjun@xxxxxxxxxx>
> > Signed-off-by: Ran Xiaokai <ran.xiaokai@xxxxxxxxxx>
> > Cc: xu xin <xu.xin16@xxxxxxxxxx>
> > Cc: Yang Yang <yang.yang29@xxxxxxxxxx>
> > ---
> > include/linux/memcontrol.h | 7 +++----
> > mm/huge_memory.c | 6 +++---
> > mm/memcontrol.c | 11 +++++------
> > 3 files changed, 11 insertions(+), 13 deletions(-)
> >
> > diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
> > index 394fd0a887ae..7282861d5a5d 100644
> > --- a/include/linux/memcontrol.h
> > +++ b/include/linux/memcontrol.h
> > @@ -130,6 +130,9 @@ struct mem_cgroup_per_node {
> > bool on_tree;
> > struct mem_cgroup *memcg; /* Back pointer, we cannot */
> > /* use container_of */
> > +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
> > + struct deferred_split deferred_split_queue;
> > +#endif
> > };
> >
> > struct mem_cgroup_threshold {
> > @@ -327,10 +330,6 @@ struct mem_cgroup {
> > struct list_head event_list;
> > spinlock_t event_list_lock;
> >
> > -#ifdef CONFIG_TRANSPARENT_HUGEPAGE
> > - struct deferred_split deferred_split_queue;
> > -#endif
> > -
> > #ifdef CONFIG_LRU_GEN_WALKS_MMU
> > /* per-memcg mm_struct list */
> > struct lru_gen_mm_list mm_list;
> > diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> > index 9859aa4f7553..338d071070a6 100644
> > --- a/mm/huge_memory.c
> > +++ b/mm/huge_memory.c
> > @@ -774,7 +774,7 @@ struct deferred_split *get_deferred_split_queue(struct folio *folio)
> > struct pglist_data *pgdat = NODE_DATA(folio_nid(folio));
> >
> > if (memcg)
> > - return &memcg->deferred_split_queue;
> > + return &memcg->nodeinfo[pgdat->node_id]->deferred_split_queue;
> > else
> > return &pgdat->deferred_split_queue;
> > }
> > @@ -3305,7 +3305,7 @@ static unsigned long deferred_split_count(struct shrinker *shrink,
> >
> > #ifdef CONFIG_MEMCG
> > if (sc->memcg)
> > - ds_queue = &sc->memcg->deferred_split_queue;
> > + ds_queue = &sc->memcg->nodeinfo[sc->nid]->deferred_split_queue;
> > #endif
> > return READ_ONCE(ds_queue->split_queue_len);
> > }
> > @@ -3322,7 +3322,7 @@ static unsigned long deferred_split_scan(struct shrinker *shrink,
> >
> > #ifdef CONFIG_MEMCG
> > if (sc->memcg)
> > - ds_queue = &sc->memcg->deferred_split_queue;
> > + ds_queue = &sc->memcg->nodeinfo[sc->nid]->deferred_split_queue;
> > #endif
> >
> > spin_lock_irqsave(&ds_queue->split_queue_lock, flags);
> > diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> > index fabce2b50c69..cdf9f5fa3b8e 100644
> > --- a/mm/memcontrol.c
> > +++ b/mm/memcontrol.c
> > @@ -5445,7 +5445,11 @@ static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
> > kfree(pn);
> > return 1;
> > }
> > -
> > +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
> > + spin_lock_init(&pn->deferred_split_queue.split_queue_lock);
> > + INIT_LIST_HEAD(&pn->deferred_split_queue.split_queue);
> > + pn->deferred_split_queue.split_queue_len = 0;
> > +#endif
> > lruvec_init(&pn->lruvec);
> > pn->memcg = memcg;
> >
> > @@ -5545,11 +5549,6 @@ static struct mem_cgroup *mem_cgroup_alloc(struct mem_cgroup *parent)
> > for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++)
> > memcg->cgwb_frn[i].done =
> > __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq);
> > -#endif
> > -#ifdef CONFIG_TRANSPARENT_HUGEPAGE
> > - spin_lock_init(&memcg->deferred_split_queue.split_queue_lock);
> > - INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue);
> > - memcg->deferred_split_queue.split_queue_len = 0;
> > #endif
> > lru_gen_init_memcg(memcg);
> > return memcg;
> > --
> > 2.15.2
> >
>
> Thanks
> Barry
>
