memcg: oom notifier

Considering containers or other resource management softwares in userland, event notification of OOM in memcg should be implemented. Now, memcg has "threshold" notifier which uses eventfd, we can make use of it for oom notification. This patch adds oom notification eventfd callback for memcg. The usage is very similar to threshold notifier, but control file is memory.oom_control and no arguments other than eventfd is required. % cgroup_event_notifier /cgroup/A/memory.oom_control dummy (About cgroup_event_notifier, see Documentation/cgroup/) Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Cc: Balbir Singh <balbir@in.ibm.com> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Cc: David Rientjes <rientjes@google.com> Cc: Davide Libenzi <davidel@xmailserver.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: oom notifier
Considering containers or other resource management softwares in userland, event notification of OOM in memcg should be implemented. Now, memcg has "threshold" notifier which uses eventfd, we can make use of it for oom notification. This patch adds oom notification eventfd callback for memcg. The usage is very similar to threshold notifier, but control file is memory.oom_control and no arguments other than eventfd is required. % cgroup_event_notifier /cgroup/A/memory.oom_control dummy (About cgroup_event_notifier, see Documentation/cgroup/) Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Cc: Balbir Singh <balbir@in.ibm.com> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Cc: David Rientjes <rientjes@google.com> Cc: Davide Libenzi <davidel@xmailserver.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
KAMEZAWA Hiroyuki · Linus Torvalds
1 parent dc98df5a1b
Showing 2 changed files with 111 additions and 9 deletions Side-by-side Diff
Documentation/cgroups/memory.txt
mm/memcontrol.c
@@ -184,6 +184,9 @@
  
 Note2: When panic_on_oom is set to "2", the whole system will panic.
  
+When oom event notifier is registered, event will be delivered.
+(See oom_control section)
+
 2. Locking
  
 The memory controller uses the following hierarchy
@@ -488,7 +491,22 @@
  
 It's applicable for root and non-root cgroup.
  
-10. TODO
+10. OOM Control
+
+Memory controler implements oom notifier using cgroup notification
+API (See cgroups.txt). It allows to register multiple oom notification
+delivery and gets notification when oom happens.
+
+To register a notifier, application need:
+ - create an eventfd using eventfd(2)
+ - open memory.oom_control file
+ - write string like "<event_fd> <memory.oom_control>" to cgroup.event_control
+
+Application will be notifier through eventfd when oom happens.
+OOM notification doesn't work for root cgroup.
+
+
+11. TODO
  
 1. Add support for accounting huge pages (as a separate controller)
 2. Make per-cgroup scanner reclaim not-shared pages first
@@ -149,6 +149,7 @@
 	u64 threshold;
 };
  
+/* For threshold */
 struct mem_cgroup_threshold_ary {
 	/* An array index points to threshold just below usage. */
 	atomic_t current_threshold;
  
@@ -157,8 +158,14 @@
 	/* Array of thresholds */
 	struct mem_cgroup_threshold entries[0];
 };
+/* for OOM */
+struct mem_cgroup_eventfd_list {
+	struct list_head list;
+	struct eventfd_ctx *eventfd;
+};
  
 static void mem_cgroup_threshold(struct mem_cgroup *mem);
+static void mem_cgroup_oom_notify(struct mem_cgroup *mem);
  
 /*
  * The memory controller data structure. The memory controller controls both
@@ -220,6 +227,9 @@
 	/* thresholds for mem+swap usage. RCU-protected */
 	struct mem_cgroup_threshold_ary *memsw_thresholds;
  
+	/* For oom notifier event fd */
+	struct list_head oom_notify;
+
 	/*
 	 * Should we move charges of a task when a task is moved into this
 	 * mem_cgroup ? And what type of charges should we move ?
  
@@ -282,9 +292,12 @@
 /* for encoding cft->private value on file */
 #define _MEM			(0)
 #define _MEMSWAP		(1)
+#define _OOM_TYPE		(2)
 #define MEMFILE_PRIVATE(x, val)	(((x) << 16) | (val))
 #define MEMFILE_TYPE(val)	(((val) >> 16) & 0xffff)
 #define MEMFILE_ATTR(val)	((val) & 0xffff)
+/* Used for OOM nofiier */
+#define OOM_CONTROL		(0)
  
 /*
  * Reclaim flags for mem_cgroup_hierarchical_reclaim
@@ -1353,6 +1366,8 @@
 	 */
 	if (!locked)
 		prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
+	else
+		mem_cgroup_oom_notify(mem);
 	mutex_unlock(&memcg_oom_mutex);
  
 	if (locked)
  
@@ -3398,9 +3413,23 @@
 	return _a->threshold - _b->threshold;
 }
  
-static int mem_cgroup_register_event(struct cgroup *cgrp, struct cftype *cft,
-		struct eventfd_ctx *eventfd, const char *args)
+static int mem_cgroup_oom_notify_cb(struct mem_cgroup *mem, void *data)
 {
+	struct mem_cgroup_eventfd_list *ev;
+
+	list_for_each_entry(ev, &mem->oom_notify, list)
+		eventfd_signal(ev->eventfd, 1);
+	return 0;
+}
+
+static void mem_cgroup_oom_notify(struct mem_cgroup *mem)
+{
+	mem_cgroup_walk_tree(mem, NULL, mem_cgroup_oom_notify_cb);
+}
+
+static int mem_cgroup_usage_register_event(struct cgroup *cgrp,
+	struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
+{
 	struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
 	struct mem_cgroup_threshold_ary *thresholds, *thresholds_new;
 	int type = MEMFILE_TYPE(cft->private);
@@ -3483,8 +3512,8 @@
 	return ret;
 }
  
-static int mem_cgroup_unregister_event(struct cgroup *cgrp, struct cftype *cft,
-		struct eventfd_ctx *eventfd)
+static int mem_cgroup_usage_unregister_event(struct cgroup *cgrp,
+	struct cftype *cft, struct eventfd_ctx *eventfd)
 {
 	struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
 	struct mem_cgroup_threshold_ary *thresholds, *thresholds_new;
  
@@ -3568,13 +3597,61 @@
 	return ret;
 }
  
+static int mem_cgroup_oom_register_event(struct cgroup *cgrp,
+	struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
+{
+	struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
+	struct mem_cgroup_eventfd_list *event;
+	int type = MEMFILE_TYPE(cft->private);
+
+	BUG_ON(type != _OOM_TYPE);
+	event = kmalloc(sizeof(*event),	GFP_KERNEL);
+	if (!event)
+		return -ENOMEM;
+
+	mutex_lock(&memcg_oom_mutex);
+
+	event->eventfd = eventfd;
+	list_add(&event->list, &memcg->oom_notify);
+
+	/* already in OOM ? */
+	if (atomic_read(&memcg->oom_lock))
+		eventfd_signal(eventfd, 1);
+	mutex_unlock(&memcg_oom_mutex);
+
+	return 0;
+}
+
+static int mem_cgroup_oom_unregister_event(struct cgroup *cgrp,
+	struct cftype *cft, struct eventfd_ctx *eventfd)
+{
+	struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
+	struct mem_cgroup_eventfd_list *ev, *tmp;
+	int type = MEMFILE_TYPE(cft->private);
+
+	BUG_ON(type != _OOM_TYPE);
+
+	mutex_lock(&memcg_oom_mutex);
+
+	list_for_each_entry_safe(ev, tmp, &mem->oom_notify, list) {
+		if (ev->eventfd == eventfd) {
+			list_del(&ev->list);
+			kfree(ev);
+		}
+	}
+
+	mutex_unlock(&memcg_oom_mutex);
+
+	return 0;
+}
+
 static struct cftype mem_cgroup_files[] = {
 	{
 		.name = "usage_in_bytes",
 		.private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
 		.read_u64 = mem_cgroup_read,
-		.register_event = mem_cgroup_register_event,
-		.unregister_event = mem_cgroup_unregister_event,
+		.register_event = mem_cgroup_usage_register_event,
+		.unregister_event = mem_cgroup_usage_unregister_event,
 	},
 	{
 		.name = "max_usage_in_bytes",
@@ -3623,6 +3700,12 @@
 		.read_u64 = mem_cgroup_move_charge_read,
 		.write_u64 = mem_cgroup_move_charge_write,
 	},
+	{
+		.name = "oom_control",
+		.register_event = mem_cgroup_oom_register_event,
+		.unregister_event = mem_cgroup_oom_unregister_event,
+		.private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
+	},
 };
  
 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
@@ -3631,8 +3714,8 @@
 		.name = "memsw.usage_in_bytes",
 		.private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
 		.read_u64 = mem_cgroup_read,
-		.register_event = mem_cgroup_register_event,
-		.unregister_event = mem_cgroup_unregister_event,
+		.register_event = mem_cgroup_usage_register_event,
+		.unregister_event = mem_cgroup_usage_unregister_event,
 	},
 	{
 		.name = "memsw.max_usage_in_bytes",
@@ -3878,6 +3961,7 @@
 	}
 	mem->last_scanned_child = 0;
 	spin_lock_init(&mem->reclaim_param_lock);
+	INIT_LIST_HEAD(&mem->oom_notify);
  
 	if (parent)
 		mem->swappiness = get_swappiness(parent);
...	...	@@ -184,6 +184,9 @@
184	184
185	185	Note2: When panic_on_oom is set to "2", the whole system will panic.
186	186
	187	+When oom event notifier is registered, event will be delivered.
	188	+(See oom_control section)
	189	+
187	190	2. Locking
188	191
189	192	The memory controller uses the following hierarchy
...	...	@@ -488,7 +491,22 @@
488	491
489	492	It's applicable for root and non-root cgroup.
490	493
491		-10. TODO
	494	+10. OOM Control
	495	+
	496	+Memory controler implements oom notifier using cgroup notification
	497	+API (See cgroups.txt). It allows to register multiple oom notification
	498	+delivery and gets notification when oom happens.
	499	+
	500	+To register a notifier, application need:
	501	+ - create an eventfd using eventfd(2)
	502	+ - open memory.oom_control file
	503	+ - write string like "<event_fd> <memory.oom_control>" to cgroup.event_control
	504	+
	505	+Application will be notifier through eventfd when oom happens.
	506	+OOM notification doesn't work for root cgroup.
	507	+
	508	+
	509	+11. TODO
492	510
493	511	1. Add support for accounting huge pages (as a separate controller)
494	512	2. Make per-cgroup scanner reclaim not-shared pages first
...	...	@@ -149,6 +149,7 @@
149	149	u64 threshold;
150	150	};
151	151
	152	+/* For threshold */
152	153	struct mem_cgroup_threshold_ary {
153	154	/* An array index points to threshold just below usage. */
154	155	atomic_t current_threshold;
155	156
...	...	@@ -157,8 +158,14 @@
157	158	/* Array of thresholds */
158	159	struct mem_cgroup_threshold entries[0];
159	160	};
	161	+/* for OOM */
	162	+struct mem_cgroup_eventfd_list {
	163	+ struct list_head list;
	164	+ struct eventfd_ctx *eventfd;
	165	+};
160	166
161	167	static void mem_cgroup_threshold(struct mem_cgroup *mem);
	168	+static void mem_cgroup_oom_notify(struct mem_cgroup *mem);
162	169
163	170	/*
164	171	* The memory controller data structure. The memory controller controls both
...	...	@@ -220,6 +227,9 @@
220	227	/* thresholds for mem+swap usage. RCU-protected */
221	228	struct mem_cgroup_threshold_ary *memsw_thresholds;
222	229
	230	+ /* For oom notifier event fd */
	231	+ struct list_head oom_notify;
	232	+
223	233	/*
224	234	* Should we move charges of a task when a task is moved into this
225	235	* mem_cgroup ? And what type of charges should we move ?
226	236
...	...	@@ -282,9 +292,12 @@
282	292	/* for encoding cft->private value on file */
283	293	#define _MEM (0)
284	294	#define _MEMSWAP (1)
	295	+#define _OOM_TYPE (2)
285	296	#define MEMFILE_PRIVATE(x, val) (((x) << 16) \| (val))
286	297	#define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
287	298	#define MEMFILE_ATTR(val) ((val) & 0xffff)
	299	+/* Used for OOM nofiier */
	300	+#define OOM_CONTROL (0)
288	301
289	302	/*
290	303	* Reclaim flags for mem_cgroup_hierarchical_reclaim
...	...	@@ -1353,6 +1366,8 @@
1353	1366	*/
1354	1367	if (!locked)
1355	1368	prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
	1369	+ else
	1370	+ mem_cgroup_oom_notify(mem);
1356	1371	mutex_unlock(&memcg_oom_mutex);
1357	1372
1358	1373	if (locked)
1359	1374
...	...	@@ -3398,9 +3413,23 @@
3398	3413	return _a->threshold - _b->threshold;
3399	3414	}
3400	3415
3401		-static int mem_cgroup_register_event(struct cgroup cgrp, struct cftype cft,
3402		- struct eventfd_ctx eventfd, const char args)
	3416	+static int mem_cgroup_oom_notify_cb(struct mem_cgroup mem, void data)
3403	3417	{
	3418	+ struct mem_cgroup_eventfd_list *ev;
	3419	+
	3420	+ list_for_each_entry(ev, &mem->oom_notify, list)
	3421	+ eventfd_signal(ev->eventfd, 1);
	3422	+ return 0;
	3423	+}
	3424	+
	3425	+static void mem_cgroup_oom_notify(struct mem_cgroup *mem)
	3426	+{
	3427	+ mem_cgroup_walk_tree(mem, NULL, mem_cgroup_oom_notify_cb);
	3428	+}
	3429	+
	3430	+static int mem_cgroup_usage_register_event(struct cgroup *cgrp,
	3431	+ struct cftype cft, struct eventfd_ctx eventfd, const char *args)
	3432	+{
3404	3433	struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3405	3434	struct mem_cgroup_threshold_ary thresholds, thresholds_new;
3406	3435	int type = MEMFILE_TYPE(cft->private);
...	...	@@ -3483,8 +3512,8 @@
3483	3512	return ret;
3484	3513	}
3485	3514
3486		-static int mem_cgroup_unregister_event(struct cgroup cgrp, struct cftype cft,
3487		- struct eventfd_ctx *eventfd)
	3515	+static int mem_cgroup_usage_unregister_event(struct cgroup *cgrp,
	3516	+ struct cftype cft, struct eventfd_ctx eventfd)
3488	3517	{
3489	3518	struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3490	3519	struct mem_cgroup_threshold_ary thresholds, thresholds_new;
3491	3520
...	...	@@ -3568,13 +3597,61 @@
3568	3597	return ret;
3569	3598	}
3570	3599
	3600	+static int mem_cgroup_oom_register_event(struct cgroup *cgrp,
	3601	+ struct cftype cft, struct eventfd_ctx eventfd, const char *args)
	3602	+{
	3603	+ struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
	3604	+ struct mem_cgroup_eventfd_list *event;
	3605	+ int type = MEMFILE_TYPE(cft->private);
	3606	+
	3607	+ BUG_ON(type != _OOM_TYPE);
	3608	+ event = kmalloc(sizeof(*event), GFP_KERNEL);
	3609	+ if (!event)
	3610	+ return -ENOMEM;
	3611	+
	3612	+ mutex_lock(&memcg_oom_mutex);
	3613	+
	3614	+ event->eventfd = eventfd;
	3615	+ list_add(&event->list, &memcg->oom_notify);
	3616	+
	3617	+ /* already in OOM ? */
	3618	+ if (atomic_read(&memcg->oom_lock))
	3619	+ eventfd_signal(eventfd, 1);
	3620	+ mutex_unlock(&memcg_oom_mutex);
	3621	+
	3622	+ return 0;
	3623	+}
	3624	+
	3625	+static int mem_cgroup_oom_unregister_event(struct cgroup *cgrp,
	3626	+ struct cftype cft, struct eventfd_ctx eventfd)
	3627	+{
	3628	+ struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
	3629	+ struct mem_cgroup_eventfd_list ev, tmp;
	3630	+ int type = MEMFILE_TYPE(cft->private);
	3631	+
	3632	+ BUG_ON(type != _OOM_TYPE);
	3633	+
	3634	+ mutex_lock(&memcg_oom_mutex);
	3635	+
	3636	+ list_for_each_entry_safe(ev, tmp, &mem->oom_notify, list) {
	3637	+ if (ev->eventfd == eventfd) {
	3638	+ list_del(&ev->list);
	3639	+ kfree(ev);
	3640	+ }
	3641	+ }
	3642	+
	3643	+ mutex_unlock(&memcg_oom_mutex);
	3644	+
	3645	+ return 0;
	3646	+}
	3647	+
3571	3648	static struct cftype mem_cgroup_files[] = {
3572	3649	{
3573	3650	.name = "usage_in_bytes",
3574	3651	.private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
3575	3652	.read_u64 = mem_cgroup_read,
3576		- .register_event = mem_cgroup_register_event,
3577		- .unregister_event = mem_cgroup_unregister_event,
	3653	+ .register_event = mem_cgroup_usage_register_event,
	3654	+ .unregister_event = mem_cgroup_usage_unregister_event,
3578	3655	},
3579	3656	{
3580	3657	.name = "max_usage_in_bytes",
...	...	@@ -3623,6 +3700,12 @@
3623	3700	.read_u64 = mem_cgroup_move_charge_read,
3624	3701	.write_u64 = mem_cgroup_move_charge_write,
3625	3702	},
	3703	+ {
	3704	+ .name = "oom_control",
	3705	+ .register_event = mem_cgroup_oom_register_event,
	3706	+ .unregister_event = mem_cgroup_oom_unregister_event,
	3707	+ .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
	3708	+ },
3626	3709	};
3627	3710
3628	3711	#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
...	...	@@ -3631,8 +3714,8 @@
3631	3714	.name = "memsw.usage_in_bytes",
3632	3715	.private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
3633	3716	.read_u64 = mem_cgroup_read,
3634		- .register_event = mem_cgroup_register_event,
3635		- .unregister_event = mem_cgroup_unregister_event,
	3717	+ .register_event = mem_cgroup_usage_register_event,
	3718	+ .unregister_event = mem_cgroup_usage_unregister_event,
3636	3719	},
3637	3720	{
3638	3721	.name = "memsw.max_usage_in_bytes",
...	...	@@ -3878,6 +3961,7 @@
3878	3961	}
3879	3962	mem->last_scanned_child = 0;
3880	3963	spin_lock_init(&mem->reclaim_param_lock);
	3964	+ INIT_LIST_HEAD(&mem->oom_notify);
3881	3965
3882	3966	if (parent)
3883	3967	mem->swappiness = get_swappiness(parent);