memcg: add zone_reclaim_stat

Introduce mem_cgroup_per_zone::reclaim_stat member and its statics collecting function. Now, get_scan_ratio() can calculate correct value on memcg reclaim. [hugh@veritas.com: avoid reclaim_stat oops when disabled] 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: 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: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: add zone_reclaim_stat
Introduce mem_cgroup_per_zone::reclaim_stat member and its statics collecting function. Now, get_scan_ratio() can calculate correct value on memcg reclaim. [hugh@veritas.com: avoid reclaim_stat oops when disabled] 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: 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: Hugh Dickins <hugh@veritas.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 a3d8e0549d
Showing 4 changed files with 83 additions and 23 deletions Side-by-side Diff
include/linux/memcontrol.h
mm/memcontrol.c
mm/swap.c
mm/vmscan.c
@@ -105,6 +105,10 @@
 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
 				       struct zone *zone,
 				       enum lru_list lru);
+struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
+						      struct zone *zone);
+struct zone_reclaim_stat*
+mem_cgroup_get_reclaim_stat_from_page(struct page *page);
  
 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
 extern int do_swap_account;
@@ -270,6 +274,18 @@
 	return 0;
 }
  
+
+static inline struct zone_reclaim_stat*
+mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, struct zone *zone)
+{
+	return NULL;
+}
+
+static inline struct zone_reclaim_stat*
+mem_cgroup_get_reclaim_stat_from_page(struct page *page)
+{
+	return NULL;
+}
  
 #endif /* CONFIG_CGROUP_MEM_CONT */
  
@@ -103,6 +103,8 @@
 	 */
 	struct list_head	lists[NR_LRU_LISTS];
 	unsigned long		count[NR_LRU_LISTS];
+
+	struct zone_reclaim_stat reclaim_stat;
 };
 /* Macro for accessing counter */
 #define MEM_CGROUP_ZSTAT(mz, idx)	((mz)->count[(idx)])
@@ -456,6 +458,33 @@
 	struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
  
 	return MEM_CGROUP_ZSTAT(mz, lru);
+}
+
+struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
+						      struct zone *zone)
+{
+	int nid = zone->zone_pgdat->node_id;
+	int zid = zone_idx(zone);
+	struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
+
+	return &mz->reclaim_stat;
+}
+
+struct zone_reclaim_stat *
+mem_cgroup_get_reclaim_stat_from_page(struct page *page)
+{
+	struct page_cgroup *pc;
+	struct mem_cgroup_per_zone *mz;
+
+	if (mem_cgroup_disabled())
+		return NULL;
+
+	pc = lookup_page_cgroup(page);
+	mz = page_cgroup_zoneinfo(pc);
+	if (!mz)
+		return NULL;
+
+	return &mz->reclaim_stat;
 }
  
 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
@@ -151,13 +151,32 @@
 	}
 }
  
+static void update_page_reclaim_stat(struct zone *zone, struct page *page,
+				     int file, int rotated)
+{
+	struct zone_reclaim_stat *reclaim_stat = &zone->reclaim_stat;
+	struct zone_reclaim_stat *memcg_reclaim_stat;
+
+	memcg_reclaim_stat = mem_cgroup_get_reclaim_stat_from_page(page);
+
+	reclaim_stat->recent_scanned[file]++;
+	if (rotated)
+		reclaim_stat->recent_rotated[file]++;
+
+	if (!memcg_reclaim_stat)
+		return;
+
+	memcg_reclaim_stat->recent_scanned[file]++;
+	if (rotated)
+		memcg_reclaim_stat->recent_rotated[file]++;
+}
+
 /*
  * FIXME: speed this up?
  */
 void activate_page(struct page *page)
 {
 	struct zone *zone = page_zone(page);
-	struct zone_reclaim_stat *reclaim_stat = &zone->reclaim_stat;
  
 	spin_lock_irq(&zone->lru_lock);
 	if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
@@ -170,8 +189,7 @@
 		add_page_to_lru_list(zone, page, lru);
 		__count_vm_event(PGACTIVATE);
  
-		reclaim_stat->recent_rotated[!!file]++;
-		reclaim_stat->recent_scanned[!!file]++;
+		update_page_reclaim_stat(zone, page, !!file, 1);
 	}
 	spin_unlock_irq(&zone->lru_lock);
 }
@@ -386,7 +404,6 @@
 {
 	int i;
 	struct zone *zone = NULL;
-	struct zone_reclaim_stat *reclaim_stat = NULL;
  
 	VM_BUG_ON(is_unevictable_lru(lru));
  
  
  
  
  
@@ -394,24 +411,23 @@
 		struct page *page = pvec->pages[i];
 		struct zone *pagezone = page_zone(page);
 		int file;
+		int active;
  
 		if (pagezone != zone) {
 			if (zone)
 				spin_unlock_irq(&zone->lru_lock);
 			zone = pagezone;
-			reclaim_stat = &zone->reclaim_stat;
 			spin_lock_irq(&zone->lru_lock);
 		}
 		VM_BUG_ON(PageActive(page));
 		VM_BUG_ON(PageUnevictable(page));
 		VM_BUG_ON(PageLRU(page));
 		SetPageLRU(page);
+		active = is_active_lru(lru);
 		file = is_file_lru(lru);
-		reclaim_stat->recent_scanned[file]++;
-		if (is_active_lru(lru)) {
+		if (active)
 			SetPageActive(page);
-			reclaim_stat->recent_rotated[file]++;
-		}
+		update_page_reclaim_stat(zone, page, file, active);
 		add_page_to_lru_list(zone, page, lru);
 	}
 	if (zone)
@@ -133,6 +133,9 @@
 static struct zone_reclaim_stat *get_reclaim_stat(struct zone *zone,
 						  struct scan_control *sc)
 {
+	if (!scan_global_lru(sc))
+		return mem_cgroup_get_reclaim_stat(sc->mem_cgroup, zone);
+
 	return &zone->reclaim_stat;
 }
  
  
@@ -1087,17 +1090,14 @@
 		__mod_zone_page_state(zone, NR_INACTIVE_ANON,
 						-count[LRU_INACTIVE_ANON]);
  
-		if (scan_global_lru(sc)) {
+		if (scan_global_lru(sc))
 			zone->pages_scanned += nr_scan;
-			reclaim_stat->recent_scanned[0] +=
-						      count[LRU_INACTIVE_ANON];
-			reclaim_stat->recent_scanned[0] +=
-						      count[LRU_ACTIVE_ANON];
-			reclaim_stat->recent_scanned[1] +=
-						      count[LRU_INACTIVE_FILE];
-			reclaim_stat->recent_scanned[1] +=
-						      count[LRU_ACTIVE_FILE];
-		}
+
+		reclaim_stat->recent_scanned[0] += count[LRU_INACTIVE_ANON];
+		reclaim_stat->recent_scanned[0] += count[LRU_ACTIVE_ANON];
+		reclaim_stat->recent_scanned[1] += count[LRU_INACTIVE_FILE];
+		reclaim_stat->recent_scanned[1] += count[LRU_ACTIVE_FILE];
+
 		spin_unlock_irq(&zone->lru_lock);
  
 		nr_scanned += nr_scan;
@@ -1155,7 +1155,7 @@
 			SetPageLRU(page);
 			lru = page_lru(page);
 			add_page_to_lru_list(zone, page, lru);
-			if (PageActive(page) && scan_global_lru(sc)) {
+			if (PageActive(page)) {
 				int file = !!page_is_file_cache(page);
 				reclaim_stat->recent_rotated[file]++;
 			}
  
@@ -1230,8 +1230,8 @@
 	 */
 	if (scan_global_lru(sc)) {
 		zone->pages_scanned += pgscanned;
-		reclaim_stat->recent_scanned[!!file] += pgmoved;
 	}
+	reclaim_stat->recent_scanned[!!file] += pgmoved;
  
 	if (file)
 		__mod_zone_page_state(zone, NR_ACTIVE_FILE, -pgmoved);
@@ -1272,8 +1272,7 @@
 	 * This helps balance scan pressure between file and anonymous
 	 * pages in get_scan_ratio.
 	 */
-	if (scan_global_lru(sc))
-		reclaim_stat->recent_rotated[!!file] += pgmoved;
+	reclaim_stat->recent_rotated[!!file] += pgmoved;
  
 	while (!list_empty(&l_inactive)) {
 		page = lru_to_page(&l_inactive);
...	...	@@ -105,6 +105,10 @@
105	105	unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
106	106	struct zone *zone,
107	107	enum lru_list lru);
	108	+struct zone_reclaim_stat mem_cgroup_get_reclaim_stat(struct mem_cgroup memcg,
	109	+ struct zone *zone);
	110	+struct zone_reclaim_stat*
	111	+mem_cgroup_get_reclaim_stat_from_page(struct page *page);
108	112
109	113	#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
110	114	extern int do_swap_account;
...	...	@@ -270,6 +274,18 @@
270	274	return 0;
271	275	}
272	276
	277	+
	278	+static inline struct zone_reclaim_stat*
	279	+mem_cgroup_get_reclaim_stat(struct mem_cgroup memcg, struct zone zone)
	280	+{
	281	+ return NULL;
	282	+}
	283	+
	284	+static inline struct zone_reclaim_stat*
	285	+mem_cgroup_get_reclaim_stat_from_page(struct page *page)
	286	+{
	287	+ return NULL;
	288	+}
273	289
274	290	#endif /* CONFIG_CGROUP_MEM_CONT */
275	291
...	...	@@ -103,6 +103,8 @@
103	103	*/
104	104	struct list_head lists[NR_LRU_LISTS];
105	105	unsigned long count[NR_LRU_LISTS];
	106	+
	107	+ struct zone_reclaim_stat reclaim_stat;
106	108	};
107	109	/* Macro for accessing counter */
108	110	#define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
...	...	@@ -456,6 +458,33 @@
456	458	struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
457	459
458	460	return MEM_CGROUP_ZSTAT(mz, lru);
	461	+}
	462	+
	463	+struct zone_reclaim_stat mem_cgroup_get_reclaim_stat(struct mem_cgroup memcg,
	464	+ struct zone *zone)
	465	+{
	466	+ int nid = zone->zone_pgdat->node_id;
	467	+ int zid = zone_idx(zone);
	468	+ struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
	469	+
	470	+ return &mz->reclaim_stat;
	471	+}
	472	+
	473	+struct zone_reclaim_stat *
	474	+mem_cgroup_get_reclaim_stat_from_page(struct page *page)
	475	+{
	476	+ struct page_cgroup *pc;
	477	+ struct mem_cgroup_per_zone *mz;
	478	+
	479	+ if (mem_cgroup_disabled())
	480	+ return NULL;
	481	+
	482	+ pc = lookup_page_cgroup(page);
	483	+ mz = page_cgroup_zoneinfo(pc);
	484	+ if (!mz)
	485	+ return NULL;
	486	+
	487	+ return &mz->reclaim_stat;
459	488	}
460	489
461	490	unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
...	...	@@ -151,13 +151,32 @@
151	151	}
152	152	}
153	153
	154	+static void update_page_reclaim_stat(struct zone zone, struct page page,
	155	+ int file, int rotated)
	156	+{
	157	+ struct zone_reclaim_stat *reclaim_stat = &zone->reclaim_stat;
	158	+ struct zone_reclaim_stat *memcg_reclaim_stat;
	159	+
	160	+ memcg_reclaim_stat = mem_cgroup_get_reclaim_stat_from_page(page);
	161	+
	162	+ reclaim_stat->recent_scanned[file]++;
	163	+ if (rotated)
	164	+ reclaim_stat->recent_rotated[file]++;
	165	+
	166	+ if (!memcg_reclaim_stat)
	167	+ return;
	168	+
	169	+ memcg_reclaim_stat->recent_scanned[file]++;
	170	+ if (rotated)
	171	+ memcg_reclaim_stat->recent_rotated[file]++;
	172	+}
	173	+
154	174	/*
155	175	* FIXME: speed this up?
156	176	*/
157	177	void activate_page(struct page *page)
158	178	{
159	179	struct zone *zone = page_zone(page);
160		- struct zone_reclaim_stat *reclaim_stat = &zone->reclaim_stat;
161	180
162	181	spin_lock_irq(&zone->lru_lock);
163	182	if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
...	...	@@ -170,8 +189,7 @@
170	189	add_page_to_lru_list(zone, page, lru);
171	190	__count_vm_event(PGACTIVATE);
172	191
173		- reclaim_stat->recent_rotated[!!file]++;
174		- reclaim_stat->recent_scanned[!!file]++;
	192	+ update_page_reclaim_stat(zone, page, !!file, 1);
175	193	}
176	194	spin_unlock_irq(&zone->lru_lock);
177	195	}
...	...	@@ -386,7 +404,6 @@
386	404	{
387	405	int i;
388	406	struct zone *zone = NULL;
389		- struct zone_reclaim_stat *reclaim_stat = NULL;
390	407
391	408	VM_BUG_ON(is_unevictable_lru(lru));
392	409
393	410
394	411
395	412
396	413
...	...	@@ -394,24 +411,23 @@
394	411	struct page *page = pvec->pages[i];
395	412	struct zone *pagezone = page_zone(page);
396	413	int file;
	414	+ int active;
397	415
398	416	if (pagezone != zone) {
399	417	if (zone)
400	418	spin_unlock_irq(&zone->lru_lock);
401	419	zone = pagezone;
402		- reclaim_stat = &zone->reclaim_stat;
403	420	spin_lock_irq(&zone->lru_lock);
404	421	}
405	422	VM_BUG_ON(PageActive(page));
406	423	VM_BUG_ON(PageUnevictable(page));
407	424	VM_BUG_ON(PageLRU(page));
408	425	SetPageLRU(page);
	426	+ active = is_active_lru(lru);
409	427	file = is_file_lru(lru);
410		- reclaim_stat->recent_scanned[file]++;
411		- if (is_active_lru(lru)) {
	428	+ if (active)
412	429	SetPageActive(page);
413		- reclaim_stat->recent_rotated[file]++;
414		- }
	430	+ update_page_reclaim_stat(zone, page, file, active);
415	431	add_page_to_lru_list(zone, page, lru);
416	432	}
417	433	if (zone)
...	...	@@ -133,6 +133,9 @@
133	133	static struct zone_reclaim_stat get_reclaim_stat(struct zone zone,
134	134	struct scan_control *sc)
135	135	{
	136	+ if (!scan_global_lru(sc))
	137	+ return mem_cgroup_get_reclaim_stat(sc->mem_cgroup, zone);
	138	+
136	139	return &zone->reclaim_stat;
137	140	}
138	141
139	142
...	...	@@ -1087,17 +1090,14 @@
1087	1090	__mod_zone_page_state(zone, NR_INACTIVE_ANON,
1088	1091	-count[LRU_INACTIVE_ANON]);
1089	1092
1090		- if (scan_global_lru(sc)) {
	1093	+ if (scan_global_lru(sc))
1091	1094	zone->pages_scanned += nr_scan;
1092		- reclaim_stat->recent_scanned[0] +=
1093		- count[LRU_INACTIVE_ANON];
1094		- reclaim_stat->recent_scanned[0] +=
1095		- count[LRU_ACTIVE_ANON];
1096		- reclaim_stat->recent_scanned[1] +=
1097		- count[LRU_INACTIVE_FILE];
1098		- reclaim_stat->recent_scanned[1] +=
1099		- count[LRU_ACTIVE_FILE];
1100		- }
	1095	+
	1096	+ reclaim_stat->recent_scanned[0] += count[LRU_INACTIVE_ANON];
	1097	+ reclaim_stat->recent_scanned[0] += count[LRU_ACTIVE_ANON];
	1098	+ reclaim_stat->recent_scanned[1] += count[LRU_INACTIVE_FILE];
	1099	+ reclaim_stat->recent_scanned[1] += count[LRU_ACTIVE_FILE];
	1100	+
1101	1101	spin_unlock_irq(&zone->lru_lock);
1102	1102
1103	1103	nr_scanned += nr_scan;
...	...	@@ -1155,7 +1155,7 @@
1155	1155	SetPageLRU(page);
1156	1156	lru = page_lru(page);
1157	1157	add_page_to_lru_list(zone, page, lru);
1158		- if (PageActive(page) && scan_global_lru(sc)) {
	1158	+ if (PageActive(page)) {
1159	1159	int file = !!page_is_file_cache(page);
1160	1160	reclaim_stat->recent_rotated[file]++;
1161	1161	}
1162	1162
...	...	@@ -1230,8 +1230,8 @@
1230	1230	*/
1231	1231	if (scan_global_lru(sc)) {
1232	1232	zone->pages_scanned += pgscanned;
1233		- reclaim_stat->recent_scanned[!!file] += pgmoved;
1234	1233	}
	1234	+ reclaim_stat->recent_scanned[!!file] += pgmoved;
1235	1235
1236	1236	if (file)
1237	1237	__mod_zone_page_state(zone, NR_ACTIVE_FILE, -pgmoved);
...	...	@@ -1272,8 +1272,7 @@
1272	1272	* This helps balance scan pressure between file and anonymous
1273	1273	* pages in get_scan_ratio.
1274	1274	*/
1275		- if (scan_global_lru(sc))
1276		- reclaim_stat->recent_rotated[!!file] += pgmoved;
	1275	+ reclaim_stat->recent_rotated[!!file] += pgmoved;
1277	1276
1278	1277	while (!list_empty(&l_inactive)) {
1279	1278	page = lru_to_page(&l_inactive);