memcg: sleep during flushing stats in safe contexts

Currently, all contexts that flush memcg stats do so with sleeping not
allowed.  Some of these contexts are perfectly safe to sleep in, such as
reading cgroup files from userspace or the background periodic flusher. 
Flushing is an expensive operation that scales with the number of cpus and
the number of cgroups in the system, so avoid doing it atomically where
possible.

Refactor the code to make mem_cgroup_flush_stats() non-atomic (aka
sleepable), and provide a separate atomic version.  The atomic version is
used in reclaim, refault, writeback, and in mem_cgroup_usage().  All other
code paths are left to use the non-atomic version.  This includes
callbacks for userspace reads and the periodic flusher.

Since refault is the only caller of mem_cgroup_flush_stats_ratelimited(),
change it to mem_cgroup_flush_stats_atomic_ratelimited().  Reclaim and
refault code paths are modified to do non-atomic flushing in separate
later patches -- so it will eventually be changed back to
mem_cgroup_flush_stats_ratelimited().

Link: https://lkml.kernel.org/r/20230330191801.1967435-6-yosryahmed@google.comSigned-off-by: Yosry Ahmed <yosryahmed@google.com>
Acked-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Josef Bacik <josef@toxicpanda.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Michal Koutný <mkoutny@suse.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Roman Gushchin <roman.gushchin@linux.dev>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vasily Averin <vasily.averin@linux.dev>
Cc: Zefan Li <lizefan.x@bytedance.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

include/linux/memcontrol.h

@@ -1038,7 +1038,8 @@ static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec,
 }
 
 void mem_cgroup_flush_stats(void);
-void mem_cgroup_flush_stats_ratelimited(void);
+void mem_cgroup_flush_stats_atomic(void);
+void mem_cgroup_flush_stats_atomic_ratelimited(void);
 
 void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx,
 			      int val);
@@ -1536,7 +1537,11 @@ static inline void mem_cgroup_flush_stats(void)
 {
 }
 
-static inline void mem_cgroup_flush_stats_ratelimited(void)
+static inline void mem_cgroup_flush_stats_atomic(void)
+{
+}
+
+static inline void mem_cgroup_flush_stats_atomic_ratelimited(void)
 {
 }
 

mm/memcontrol.c

@@ -635,7 +635,7 @@ static inline void memcg_rstat_updated(struct mem_cgroup *memcg, int val)
 	}
 }
 
-static void __mem_cgroup_flush_stats(void)
+static void do_flush_stats(bool atomic)
 {
 	/*
 	 * We always flush the entire tree, so concurrent flushers can just
@@ -647,26 +647,46 @@ static void __mem_cgroup_flush_stats(void)
 		return;
 
 	WRITE_ONCE(flush_next_time, jiffies_64 + 2*FLUSH_TIME);
-	cgroup_rstat_flush_atomic(root_mem_cgroup->css.cgroup);
+
+	if (atomic)
+		cgroup_rstat_flush_atomic(root_mem_cgroup->css.cgroup);
+	else
+		cgroup_rstat_flush(root_mem_cgroup->css.cgroup);
+
 	atomic_set(&stats_flush_threshold, 0);
 	atomic_set(&stats_flush_ongoing, 0);
 }
 
+static bool should_flush_stats(void)
+{
+	return atomic_read(&stats_flush_threshold) > num_online_cpus();
+}
+
 void mem_cgroup_flush_stats(void)
 {
-	if (atomic_read(&stats_flush_threshold) > num_online_cpus())
-		__mem_cgroup_flush_stats();
+	if (should_flush_stats())
+		do_flush_stats(false);
 }
 
-void mem_cgroup_flush_stats_ratelimited(void)
+void mem_cgroup_flush_stats_atomic(void)
+{
+	if (should_flush_stats())
+		do_flush_stats(true);
+}
+
+void mem_cgroup_flush_stats_atomic_ratelimited(void)
 {
 	if (time_after64(jiffies_64, READ_ONCE(flush_next_time)))
-		mem_cgroup_flush_stats();
+		mem_cgroup_flush_stats_atomic();
 }
 
 static void flush_memcg_stats_dwork(struct work_struct *w)
 {
-	__mem_cgroup_flush_stats();
+	/*
+	 * Always flush here so that flushing in latency-sensitive paths is
+	 * as cheap as possible.
+	 */
+	do_flush_stats(false);
 	queue_delayed_work(system_unbound_wq, &stats_flush_dwork, FLUSH_TIME);
 }
 
@@ -3686,9 +3706,12 @@ static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
 		 * done from irq context; use stale stats in this case.
 		 * Arguably, usage threshold events are not reliable on the root
 		 * memcg anyway since its usage is ill-defined.
+		 *
+		 * Additionally, other call paths through memcg_check_events()
+		 * disable irqs, so make sure we are flushing stats atomically.
 		 */
 		if (in_task())
-			mem_cgroup_flush_stats();
+			mem_cgroup_flush_stats_atomic();
 		val = memcg_page_state(memcg, NR_FILE_PAGES) +
 			memcg_page_state(memcg, NR_ANON_MAPPED);
 		if (swap)
@@ -4611,7 +4634,11 @@ void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
 	struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
 	struct mem_cgroup *parent;
 
-	mem_cgroup_flush_stats();
+	/*
+	 * wb_writeback() takes a spinlock and calls
+	 * wb_over_bg_thresh()->mem_cgroup_wb_stats(). Do not sleep.
+	 */
+	mem_cgroup_flush_stats_atomic();
 
 	*pdirty = memcg_page_state(memcg, NR_FILE_DIRTY);
 	*pwriteback = memcg_page_state(memcg, NR_WRITEBACK);

mm/vmscan.c

@@ -2861,7 +2861,7 @@ static void prepare_scan_count(pg_data_t *pgdat, struct scan_control *sc)
 	 * Flush the memory cgroup stats, so that we read accurate per-memcg
 	 * lruvec stats for heuristics.
 	 */
-	mem_cgroup_flush_stats();
+	mem_cgroup_flush_stats_atomic();
 
 	/*
 	 * Determine the scan balance between anon and file LRUs.

mm/workingset.c

@@ -462,7 +462,7 @@ void workingset_refault(struct folio *folio, void *shadow)
 
 	mod_lruvec_state(lruvec, WORKINGSET_REFAULT_BASE + file, nr);
 
-	mem_cgroup_flush_stats_ratelimited();
+	mem_cgroup_flush_stats_atomic_ratelimited();
 	/*
 	 * Compare the distance to the existing workingset size. We
 	 * don't activate pages that couldn't stay resident even if