memcg: add inactive_anon_is_low()

The inactive_anon_is_low() is key component of active/inactive anon balancing on reclaim. However current inactive_anon_is_low() function only consider global reclaim. Therefore, we need following ugly scan_global_lru() condition. if (lru == LRU_ACTIVE_ANON && (!scan_global_lru(sc) || inactive_anon_is_low(zone))) { shrink_active_list(nr_to_scan, zone, sc, priority, file); return 0; it cause that memcg reclaim always deactivate pages when shrink_list() is called. To make mem_cgroup_inactive_anon_is_low() improve active/inactive anon balancing of memcgroup. Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Acked-by: Rik van Riel <riel@redhat.com> Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: "Pekka Enberg" <penberg@cs.helsinki.fi> Cc: Balbir Singh <balbir@in.ibm.com> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Cc: Hugh Dickins <hugh@veritas.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: add inactive_anon_is_low()
The inactive_anon_is_low() is key component of active/inactive anon balancing on reclaim. However current inactive_anon_is_low() function only consider global reclaim. Therefore, we need following ugly scan_global_lru() condition. if (lru == LRU_ACTIVE_ANON && (!scan_global_lru(sc) || inactive_anon_is_low(zone))) { shrink_active_list(nr_to_scan, zone, sc, priority, file); return 0; it cause that memcg reclaim always deactivate pages when shrink_list() is called. To make mem_cgroup_inactive_anon_is_low() improve active/inactive anon balancing of memcgroup. Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Acked-by: Rik van Riel <riel@redhat.com> Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Cc: "Pekka Enberg" <penberg@cs.helsinki.fi> Cc: Balbir Singh <balbir@in.ibm.com> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Cc: Hugh Dickins <hugh@veritas.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
KOSAKI Motohiro · Linus Torvalds
1 parent 549927620b
Showing 3 changed files with 77 additions and 15 deletions Side-by-side Diff
include/linux/memcontrol.h
mm/memcontrol.c
mm/vmscan.c
@@ -100,6 +100,8 @@
  
 extern long mem_cgroup_calc_reclaim(struct mem_cgroup *mem, struct zone *zone,
 					int priority, enum lru_list lru);
+int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg,
+				    struct zone *zone);
  
 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
 extern int do_swap_account;
@@ -251,6 +253,13 @@
 {
 	return false;
 }
+
+static inline int
+mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg, struct zone *zone)
+{
+	return 1;
+}
+
 #endif /* CONFIG_CGROUP_MEM_CONT */
  
 #endif /* _LINUX_MEMCONTROL_H */
@@ -156,6 +156,9 @@
 	unsigned long	last_oom_jiffies;
 	int		obsolete;
 	atomic_t	refcnt;
+
+	unsigned int inactive_ratio;
+
 	/*
 	 * statistics. This must be placed at the end of memcg.
 	 */
@@ -431,6 +434,20 @@
 	return (nr_pages >> priority);
 }
  
+int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg, struct zone *zone)
+{
+	unsigned long active;
+	unsigned long inactive;
+
+	inactive = mem_cgroup_get_all_zonestat(memcg, LRU_INACTIVE_ANON);
+	active = mem_cgroup_get_all_zonestat(memcg, LRU_ACTIVE_ANON);
+
+	if (inactive * memcg->inactive_ratio < active)
+		return 1;
+
+	return 0;
+}
+
 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
 					struct list_head *dst,
 					unsigned long *scanned, int order,
@@ -1360,6 +1377,29 @@
 	return 0;
 }
  
+/*
+ * The inactive anon list should be small enough that the VM never has to
+ * do too much work, but large enough that each inactive page has a chance
+ * to be referenced again before it is swapped out.
+ *
+ * this calculation is straightforward porting from
+ * page_alloc.c::setup_per_zone_inactive_ratio().
+ * it describe more detail.
+ */
+static void mem_cgroup_set_inactive_ratio(struct mem_cgroup *memcg)
+{
+	unsigned int gb, ratio;
+
+	gb = res_counter_read_u64(&memcg->res, RES_LIMIT) >> 30;
+	if (gb)
+		ratio = int_sqrt(10 * gb);
+	else
+		ratio = 1;
+
+	memcg->inactive_ratio = ratio;
+
+}
+
 static DEFINE_MUTEX(set_limit_mutex);
  
 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
@@ -1398,6 +1438,10 @@
 				GFP_KERNEL, false);
   		if (!progress)			retry_count--;
 	}
+
+	if (!ret)
+		mem_cgroup_set_inactive_ratio(memcg);
+
 	return ret;
 }
  
@@ -1982,7 +2026,7 @@
 		res_counter_init(&mem->res, NULL);
 		res_counter_init(&mem->memsw, NULL);
 	}
-
+	mem_cgroup_set_inactive_ratio(mem);
 	mem->last_scanned_child = NULL;
  
 	return &mem->css;
@@ -1310,14 +1310,7 @@
 	pagevec_release(&pvec);
 }
  
-/**
- * inactive_anon_is_low - check if anonymous pages need to be deactivated
- * @zone: zone to check
- *
- * Returns true if the zone does not have enough inactive anon pages,
- * meaning some active anon pages need to be deactivated.
- */
-static int inactive_anon_is_low(struct zone *zone)
+static int inactive_anon_is_low_global(struct zone *zone)
 {
 	unsigned long active, inactive;
  
@@ -1330,6 +1323,25 @@
 	return 0;
 }
  
+/**
+ * inactive_anon_is_low - check if anonymous pages need to be deactivated
+ * @zone: zone to check
+ * @sc:   scan control of this context
+ *
+ * Returns true if the zone does not have enough inactive anon pages,
+ * meaning some active anon pages need to be deactivated.
+ */
+static int inactive_anon_is_low(struct zone *zone, struct scan_control *sc)
+{
+	int low;
+
+	if (scan_global_lru(sc))
+		low = inactive_anon_is_low_global(zone);
+	else
+		low = mem_cgroup_inactive_anon_is_low(sc->mem_cgroup, zone);
+	return low;
+}
+
 static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan,
 	struct zone *zone, struct scan_control *sc, int priority)
 {
@@ -1340,8 +1352,7 @@
 		return 0;
 	}
  
-	if (lru == LRU_ACTIVE_ANON &&
-	    (!scan_global_lru(sc) || inactive_anon_is_low(zone))) {
+	if (lru == LRU_ACTIVE_ANON && inactive_anon_is_low(zone, sc)) {
 		shrink_active_list(nr_to_scan, zone, sc, priority, file);
 		return 0;
 	}
  
@@ -1509,10 +1520,8 @@
 	 * Even if we did not try to evict anon pages at all, we want to
 	 * rebalance the anon lru active/inactive ratio.
 	 */
-	if (!scan_global_lru(sc) || inactive_anon_is_low(zone))
+	if (inactive_anon_is_low(zone, sc))
 		shrink_active_list(SWAP_CLUSTER_MAX, zone, sc, priority, 0);
-	else if (!scan_global_lru(sc))
-		shrink_active_list(SWAP_CLUSTER_MAX, zone, sc, priority, 0);
  
 	throttle_vm_writeout(sc->gfp_mask);
 }
@@ -1807,7 +1816,7 @@
 			 * Do some background aging of the anon list, to give
 			 * pages a chance to be referenced before reclaiming.
 			 */
-			if (inactive_anon_is_low(zone))
+			if (inactive_anon_is_low(zone, &sc))
 				shrink_active_list(SWAP_CLUSTER_MAX, zone,
 							&sc, priority, 0);
...	...	@@ -100,6 +100,8 @@
100	100
101	101	extern long mem_cgroup_calc_reclaim(struct mem_cgroup mem, struct zone zone,
102	102	int priority, enum lru_list lru);
	103	+int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg,
	104	+ struct zone *zone);
103	105
104	106	#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
105	107	extern int do_swap_account;
...	...	@@ -251,6 +253,13 @@
251	253	{
252	254	return false;
253	255	}
	256	+
	257	+static inline int
	258	+mem_cgroup_inactive_anon_is_low(struct mem_cgroup memcg, struct zone zone)
	259	+{
	260	+ return 1;
	261	+}
	262	+
254	263	#endif /* CONFIG_CGROUP_MEM_CONT */
255	264
256	265	#endif /* _LINUX_MEMCONTROL_H */
...	...	@@ -156,6 +156,9 @@
156	156	unsigned long last_oom_jiffies;
157	157	int obsolete;
158	158	atomic_t refcnt;
	159	+
	160	+ unsigned int inactive_ratio;
	161	+
159	162	/*
160	163	* statistics. This must be placed at the end of memcg.
161	164	*/
...	...	@@ -431,6 +434,20 @@
431	434	return (nr_pages >> priority);
432	435	}
433	436
	437	+int mem_cgroup_inactive_anon_is_low(struct mem_cgroup memcg, struct zone zone)
	438	+{
	439	+ unsigned long active;
	440	+ unsigned long inactive;
	441	+
	442	+ inactive = mem_cgroup_get_all_zonestat(memcg, LRU_INACTIVE_ANON);
	443	+ active = mem_cgroup_get_all_zonestat(memcg, LRU_ACTIVE_ANON);
	444	+
	445	+ if (inactive * memcg->inactive_ratio < active)
	446	+ return 1;
	447	+
	448	+ return 0;
	449	+}
	450	+
434	451	unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
435	452	struct list_head *dst,
436	453	unsigned long *scanned, int order,
...	...	@@ -1360,6 +1377,29 @@
1360	1377	return 0;
1361	1378	}
1362	1379
	1380	+/*
	1381	+ * The inactive anon list should be small enough that the VM never has to
	1382	+ * do too much work, but large enough that each inactive page has a chance
	1383	+ * to be referenced again before it is swapped out.
	1384	+ *
	1385	+ * this calculation is straightforward porting from
	1386	+ * page_alloc.c::setup_per_zone_inactive_ratio().
	1387	+ * it describe more detail.
	1388	+ */
	1389	+static void mem_cgroup_set_inactive_ratio(struct mem_cgroup *memcg)
	1390	+{
	1391	+ unsigned int gb, ratio;
	1392	+
	1393	+ gb = res_counter_read_u64(&memcg->res, RES_LIMIT) >> 30;
	1394	+ if (gb)
	1395	+ ratio = int_sqrt(10 * gb);
	1396	+ else
	1397	+ ratio = 1;
	1398	+
	1399	+ memcg->inactive_ratio = ratio;
	1400	+
	1401	+}
	1402	+
1363	1403	static DEFINE_MUTEX(set_limit_mutex);
1364	1404
1365	1405	static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
...	...	@@ -1398,6 +1438,10 @@
1398	1438	GFP_KERNEL, false);
1399	1439	if (!progress) retry_count--;
1400	1440	}
	1441	+
	1442	+ if (!ret)
	1443	+ mem_cgroup_set_inactive_ratio(memcg);
	1444	+
1401	1445	return ret;
1402	1446	}
1403	1447
...	...	@@ -1982,7 +2026,7 @@
1982	2026	res_counter_init(&mem->res, NULL);
1983	2027	res_counter_init(&mem->memsw, NULL);
1984	2028	}
1985		-
	2029	+ mem_cgroup_set_inactive_ratio(mem);
1986	2030	mem->last_scanned_child = NULL;
1987	2031
1988	2032	return &mem->css;
...	...	@@ -1310,14 +1310,7 @@
1310	1310	pagevec_release(&pvec);
1311	1311	}
1312	1312
1313		-/**
1314		- * inactive_anon_is_low - check if anonymous pages need to be deactivated
1315		- * @zone: zone to check
1316		- *
1317		- * Returns true if the zone does not have enough inactive anon pages,
1318		- * meaning some active anon pages need to be deactivated.
1319		- */
1320		-static int inactive_anon_is_low(struct zone *zone)
	1313	+static int inactive_anon_is_low_global(struct zone *zone)
1321	1314	{
1322	1315	unsigned long active, inactive;
1323	1316
...	...	@@ -1330,6 +1323,25 @@
1330	1323	return 0;
1331	1324	}
1332	1325
	1326	+/**
	1327	+ * inactive_anon_is_low - check if anonymous pages need to be deactivated
	1328	+ * @zone: zone to check
	1329	+ * @sc: scan control of this context
	1330	+ *
	1331	+ * Returns true if the zone does not have enough inactive anon pages,
	1332	+ * meaning some active anon pages need to be deactivated.
	1333	+ */
	1334	+static int inactive_anon_is_low(struct zone zone, struct scan_control sc)
	1335	+{
	1336	+ int low;
	1337	+
	1338	+ if (scan_global_lru(sc))
	1339	+ low = inactive_anon_is_low_global(zone);
	1340	+ else
	1341	+ low = mem_cgroup_inactive_anon_is_low(sc->mem_cgroup, zone);
	1342	+ return low;
	1343	+}
	1344	+
1333	1345	static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan,
1334	1346	struct zone zone, struct scan_control sc, int priority)
1335	1347	{
...	...	@@ -1340,8 +1352,7 @@
1340	1352	return 0;
1341	1353	}
1342	1354
1343		- if (lru == LRU_ACTIVE_ANON &&
1344		- (!scan_global_lru(sc) \|\| inactive_anon_is_low(zone))) {
	1355	+ if (lru == LRU_ACTIVE_ANON && inactive_anon_is_low(zone, sc)) {
1345	1356	shrink_active_list(nr_to_scan, zone, sc, priority, file);
1346	1357	return 0;
1347	1358	}
1348	1359
...	...	@@ -1509,10 +1520,8 @@
1509	1520	* Even if we did not try to evict anon pages at all, we want to
1510	1521	* rebalance the anon lru active/inactive ratio.
1511	1522	*/
1512		- if (!scan_global_lru(sc) \|\| inactive_anon_is_low(zone))
	1523	+ if (inactive_anon_is_low(zone, sc))
1513	1524	shrink_active_list(SWAP_CLUSTER_MAX, zone, sc, priority, 0);
1514		- else if (!scan_global_lru(sc))
1515		- shrink_active_list(SWAP_CLUSTER_MAX, zone, sc, priority, 0);
1516	1525
1517	1526	throttle_vm_writeout(sc->gfp_mask);
1518	1527	}
...	...	@@ -1807,7 +1816,7 @@
1807	1816	* Do some background aging of the anon list, to give
1808	1817	* pages a chance to be referenced before reclaiming.
1809	1818	*/
1810		- if (inactive_anon_is_low(zone))
	1819	+ if (inactive_anon_is_low(zone, &sc))
1811	1820	shrink_active_list(SWAP_CLUSTER_MAX, zone,
1812	1821	&sc, priority, 0);
1813	1822