Kernel

Kernel Git Source Tree

Root/mm/memcontrol.c

1	/* memcontrol.c - Memory Controller
2	*
3	* Copyright IBM Corporation, 2007
4	* Author Balbir Singh <balbir@linux.vnet.ibm.com>
5	*
6	* Copyright 2007 OpenVZ SWsoft Inc
7	* Author: Pavel Emelianov <xemul@openvz.org>
8	*
9	* Memory thresholds
10	* Copyright (C) 2009 Nokia Corporation
11	* Author: Kirill A. Shutemov
12	*
13	* This program is free software; you can redistribute it and/or modify
14	* it under the terms of the GNU General Public License as published by
15	* the Free Software Foundation; either version 2 of the License, or
16	* (at your option) any later version.
17	*
18	* This program is distributed in the hope that it will be useful,
19	* but WITHOUT ANY WARRANTY; without even the implied warranty of
20	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21	* GNU General Public License for more details.
22	*/
23
24	#include <linux/res_counter.h>
25	#include <linux/memcontrol.h>
26	#include <linux/cgroup.h>
27	#include <linux/mm.h>
28	#include <linux/hugetlb.h>
29	#include <linux/pagemap.h>
30	#include <linux/smp.h>
31	#include <linux/page-flags.h>
32	#include <linux/backing-dev.h>
33	#include <linux/bit_spinlock.h>
34	#include <linux/rcupdate.h>
35	#include <linux/limits.h>
36	#include <linux/export.h>
37	#include <linux/mutex.h>
38	#include <linux/rbtree.h>
39	#include <linux/slab.h>
40	#include <linux/swap.h>
41	#include <linux/swapops.h>
42	#include <linux/spinlock.h>
43	#include <linux/eventfd.h>
44	#include <linux/sort.h>
45	#include <linux/fs.h>
46	#include <linux/seq_file.h>
47	#include <linux/vmalloc.h>
48	#include <linux/mm_inline.h>
49	#include <linux/page_cgroup.h>
50	#include <linux/cpu.h>
51	#include <linux/oom.h>
52	#include "internal.h"
53	#include <net/sock.h>
54	#include <net/tcp_memcontrol.h>
55
56	#include <asm/uaccess.h>
57
58	#include <trace/events/vmscan.h>
59
60	struct cgroup_subsys mem_cgroup_subsys __read_mostly;
61	#define MEM_CGROUP_RECLAIM_RETRIES 5
62	static struct mem_cgroup *root_mem_cgroup __read_mostly;
63
64	#ifdef CONFIG_MEMCG_SWAP
65	/* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
66	int do_swap_account __read_mostly;
67
68	/* for remember boot option*/
69	#ifdef CONFIG_MEMCG_SWAP_ENABLED
70	static int really_do_swap_account __initdata = 1;
71	#else
72	static int really_do_swap_account __initdata = 0;
73	#endif
74
75	#else
76	#define do_swap_account 0
77	#endif
78
79
80	/*
81	* Statistics for memory cgroup.
82	*/
83	enum mem_cgroup_stat_index {
84	/*
85	* For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
86	*/
87	MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
88	MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
89	MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
90	MEM_CGROUP_STAT_SWAP, /* # of pages, swapped out */
91	MEM_CGROUP_STAT_NSTATS,
92	};
93
94	static const char * const mem_cgroup_stat_names[] = {
95	"cache",
96	"rss",
97	"mapped_file",
98	"swap",
99	};
100
101	enum mem_cgroup_events_index {
102	MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */
103	MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */
104	MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */
105	MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */
106	MEM_CGROUP_EVENTS_NSTATS,
107	};
108
109	static const char * const mem_cgroup_events_names[] = {
110	"pgpgin",
111	"pgpgout",
112	"pgfault",
113	"pgmajfault",
114	};
115
116	/*
117	* Per memcg event counter is incremented at every pagein/pageout. With THP,
118	* it will be incremated by the number of pages. This counter is used for
119	* for trigger some periodic events. This is straightforward and better
120	* than using jiffies etc. to handle periodic memcg event.
121	*/
122	enum mem_cgroup_events_target {
123	MEM_CGROUP_TARGET_THRESH,
124	MEM_CGROUP_TARGET_SOFTLIMIT,
125	MEM_CGROUP_TARGET_NUMAINFO,
126	MEM_CGROUP_NTARGETS,
127	};
128	#define THRESHOLDS_EVENTS_TARGET 128
129	#define SOFTLIMIT_EVENTS_TARGET 1024
130	#define NUMAINFO_EVENTS_TARGET 1024
131
132	struct mem_cgroup_stat_cpu {
133	long count[MEM_CGROUP_STAT_NSTATS];
134	unsigned long events[MEM_CGROUP_EVENTS_NSTATS];
135	unsigned long nr_page_events;
136	unsigned long targets[MEM_CGROUP_NTARGETS];
137	};
138
139	struct mem_cgroup_reclaim_iter {
140	/* css_id of the last scanned hierarchy member */
141	int position;
142	/* scan generation, increased every round-trip */
143	unsigned int generation;
144	};
145
146	/*
147	* per-zone information in memory controller.
148	*/
149	struct mem_cgroup_per_zone {
150	struct lruvec lruvec;
151	unsigned long lru_size[NR_LRU_LISTS];
152
153	struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1];
154
155	struct rb_node tree_node; /* RB tree node */
156	unsigned long long usage_in_excess;/* Set to the value by which */
157	/* the soft limit is exceeded*/
158	bool on_tree;
159	struct mem_cgroup memcg; / Back pointer, we cannot */
160	/* use container_of */
161	};
162
163	struct mem_cgroup_per_node {
164	struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
165	};
166
167	struct mem_cgroup_lru_info {
168	struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
169	};
170
171	/*
172	* Cgroups above their limits are maintained in a RB-Tree, independent of
173	* their hierarchy representation
174	*/
175
176	struct mem_cgroup_tree_per_zone {
177	struct rb_root rb_root;
178	spinlock_t lock;
179	};
180
181	struct mem_cgroup_tree_per_node {
182	struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
183	};
184
185	struct mem_cgroup_tree {
186	struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
187	};
188
189	static struct mem_cgroup_tree soft_limit_tree __read_mostly;
190
191	struct mem_cgroup_threshold {
192	struct eventfd_ctx *eventfd;
193	u64 threshold;
194	};
195
196	/* For threshold */
197	struct mem_cgroup_threshold_ary {
198	/* An array index points to threshold just below or equal to usage. */
199	int current_threshold;
200	/* Size of entries[] */
201	unsigned int size;
202	/* Array of thresholds */
203	struct mem_cgroup_threshold entries[0];
204	};
205
206	struct mem_cgroup_thresholds {
207	/* Primary thresholds array */
208	struct mem_cgroup_threshold_ary *primary;
209	/*
210	* Spare threshold array.
211	* This is needed to make mem_cgroup_unregister_event() "never fail".
212	* It must be able to store at least primary->size - 1 entries.
213	*/
214	struct mem_cgroup_threshold_ary *spare;
215	};
216
217	/* for OOM */
218	struct mem_cgroup_eventfd_list {
219	struct list_head list;
220	struct eventfd_ctx *eventfd;
221	};
222
223	static void mem_cgroup_threshold(struct mem_cgroup *memcg);
224	static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
225
226	/*
227	* The memory controller data structure. The memory controller controls both
228	* page cache and RSS per cgroup. We would eventually like to provide
229	* statistics based on the statistics developed by Rik Van Riel for clock-pro,
230	* to help the administrator determine what knobs to tune.
231	*
232	* TODO: Add a water mark for the memory controller. Reclaim will begin when
233	* we hit the water mark. May be even add a low water mark, such that
234	* no reclaim occurs from a cgroup at it's low water mark, this is
235	* a feature that will be implemented much later in the future.
236	*/
237	struct mem_cgroup {
238	struct cgroup_subsys_state css;
239	/*
240	* the counter to account for memory usage
241	*/
242	struct res_counter res;
243
244	union {
245	/*
246	* the counter to account for mem+swap usage.
247	*/
248	struct res_counter memsw;
249
250	/*
251	* rcu_freeing is used only when freeing struct mem_cgroup,
252	* so put it into a union to avoid wasting more memory.
253	* It must be disjoint from the css field. It could be
254	* in a union with the res field, but res plays a much
255	* larger part in mem_cgroup life than memsw, and might
256	* be of interest, even at time of free, when debugging.
257	* So share rcu_head with the less interesting memsw.
258	*/
259	struct rcu_head rcu_freeing;
260	/*
261	* We also need some space for a worker in deferred freeing.
262	* By the time we call it, rcu_freeing is no longer in use.
263	*/
264	struct work_struct work_freeing;
265	};
266
267	/*
268	* Per cgroup active and inactive list, similar to the
269	* per zone LRU lists.
270	*/
271	struct mem_cgroup_lru_info info;
272	int last_scanned_node;
273	#if MAX_NUMNODES > 1
274	nodemask_t scan_nodes;
275	atomic_t numainfo_events;
276	atomic_t numainfo_updating;
277	#endif
278	/*
279	* Should the accounting and control be hierarchical, per subtree?
280	*/
281	bool use_hierarchy;
282
283	bool oom_lock;
284	atomic_t under_oom;
285
286	atomic_t refcnt;
287
288	int swappiness;
289	/* OOM-Killer disable */
290	int oom_kill_disable;
291
292	/* set when res.limit == memsw.limit */
293	bool memsw_is_minimum;
294
295	/* protect arrays of thresholds */
296	struct mutex thresholds_lock;
297
298	/* thresholds for memory usage. RCU-protected */
299	struct mem_cgroup_thresholds thresholds;
300
301	/* thresholds for mem+swap usage. RCU-protected */
302	struct mem_cgroup_thresholds memsw_thresholds;
303
304	/* For oom notifier event fd */
305	struct list_head oom_notify;
306
307	/*
308	* Should we move charges of a task when a task is moved into this
309	* mem_cgroup ? And what type of charges should we move ?
310	*/
311	unsigned long move_charge_at_immigrate;
312	/*
313	* set > 0 if pages under this cgroup are moving to other cgroup.
314	*/
315	atomic_t moving_account;
316	/* taken only while moving_account > 0 */
317	spinlock_t move_lock;
318	/*
319	* percpu counter.
320	*/
321	struct mem_cgroup_stat_cpu __percpu *stat;
322	/*
323	* used when a cpu is offlined or other synchronizations
324	* See mem_cgroup_read_stat().
325	*/
326	struct mem_cgroup_stat_cpu nocpu_base;
327	spinlock_t pcp_counter_lock;
328
329	#ifdef CONFIG_INET
330	struct tcp_memcontrol tcp_mem;
331	#endif
332	};
333
334	/* Stuffs for move charges at task migration. */
335	/*
336	* Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
337	* left-shifted bitmap of these types.
338	*/
339	enum move_type {
340	MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */
341	MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */
342	NR_MOVE_TYPE,
343	};
344
345	/* "mc" and its members are protected by cgroup_mutex */
346	static struct move_charge_struct {
347	spinlock_t lock; /* for from, to */
348	struct mem_cgroup *from;
349	struct mem_cgroup *to;
350	unsigned long precharge;
351	unsigned long moved_charge;
352	unsigned long moved_swap;
353	struct task_struct moving_task; / a task moving charges */
354	wait_queue_head_t waitq; /* a waitq for other context */
355	} mc = {
356	.lock = __SPIN_LOCK_UNLOCKED(mc.lock),
357	.waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
358	};
359
360	static bool move_anon(void)
361	{
362	return test_bit(MOVE_CHARGE_TYPE_ANON,
363	&mc.to->move_charge_at_immigrate);
364	}
365
366	static bool move_file(void)
367	{
368	return test_bit(MOVE_CHARGE_TYPE_FILE,
369	&mc.to->move_charge_at_immigrate);
370	}
371
372	/*
373	* Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
374	* limit reclaim to prevent infinite loops, if they ever occur.
375	*/
376	#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
377	#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
378
379	enum charge_type {
380	MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
381	MEM_CGROUP_CHARGE_TYPE_ANON,
382	MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
383	MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
384	NR_CHARGE_TYPE,
385	};
386
387	/* for encoding cft->private value on file */
388	#define _MEM (0)
389	#define _MEMSWAP (1)
390	#define _OOM_TYPE (2)
391	#define MEMFILE_PRIVATE(x, val) ((x) << 16 \| (val))
392	#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
393	#define MEMFILE_ATTR(val) ((val) & 0xffff)
394	/* Used for OOM nofiier */
395	#define OOM_CONTROL (0)
396
397	/*
398	* Reclaim flags for mem_cgroup_hierarchical_reclaim
399	*/
400	#define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
401	#define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
402	#define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
403	#define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
404
405	static void mem_cgroup_get(struct mem_cgroup *memcg);
406	static void mem_cgroup_put(struct mem_cgroup *memcg);
407
408	static inline
409	struct mem_cgroup mem_cgroup_from_css(struct cgroup_subsys_state s)
410	{
411	return container_of(s, struct mem_cgroup, css);
412	}
413
414	/* Writing them here to avoid exposing memcg's inner layout */
415	#ifdef CONFIG_MEMCG_KMEM
416	#include <net/sock.h>
417	#include <net/ip.h>
418
419	static bool mem_cgroup_is_root(struct mem_cgroup *memcg);
420	void sock_update_memcg(struct sock *sk)
421	{
422	if (mem_cgroup_sockets_enabled) {
423	struct mem_cgroup *memcg;
424	struct cg_proto *cg_proto;
425
426	BUG_ON(!sk->sk_prot->proto_cgroup);
427
428	/* Socket cloning can throw us here with sk_cgrp already
429	* filled. It won't however, necessarily happen from
430	* process context. So the test for root memcg given
431	* the current task's memcg won't help us in this case.
432	*
433	* Respecting the original socket's memcg is a better
434	* decision in this case.
435	*/
436	if (sk->sk_cgrp) {
437	BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg));
438	mem_cgroup_get(sk->sk_cgrp->memcg);
439	return;
440	}
441
442	rcu_read_lock();
443	memcg = mem_cgroup_from_task(current);
444	cg_proto = sk->sk_prot->proto_cgroup(memcg);
445	if (!mem_cgroup_is_root(memcg) && memcg_proto_active(cg_proto)) {
446	mem_cgroup_get(memcg);
447	sk->sk_cgrp = cg_proto;
448	}
449	rcu_read_unlock();
450	}
451	}
452	EXPORT_SYMBOL(sock_update_memcg);
453
454	void sock_release_memcg(struct sock *sk)
455	{
456	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
457	struct mem_cgroup *memcg;
458	WARN_ON(!sk->sk_cgrp->memcg);
459	memcg = sk->sk_cgrp->memcg;
460	mem_cgroup_put(memcg);
461	}
462	}
463
464	#ifdef CONFIG_INET
465	struct cg_proto tcp_proto_cgroup(struct mem_cgroup memcg)
466	{
467	if (!memcg \|\| mem_cgroup_is_root(memcg))
468	return NULL;
469
470	return &memcg->tcp_mem.cg_proto;
471	}
472	EXPORT_SYMBOL(tcp_proto_cgroup);
473	#endif /* CONFIG_INET */
474	#endif /* CONFIG_MEMCG_KMEM */
475
476	#if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
477	static void disarm_sock_keys(struct mem_cgroup *memcg)
478	{
479	if (!memcg_proto_activated(&memcg->tcp_mem.cg_proto))
480	return;
481	static_key_slow_dec(&memcg_socket_limit_enabled);
482	}
483	#else
484	static void disarm_sock_keys(struct mem_cgroup *memcg)
485	{
486	}
487	#endif
488
489	static void drain_all_stock_async(struct mem_cgroup *memcg);
490
491	static struct mem_cgroup_per_zone *
492	mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid)
493	{
494	return &memcg->info.nodeinfo[nid]->zoneinfo[zid];
495	}
496
497	struct cgroup_subsys_state mem_cgroup_css(struct mem_cgroup memcg)
498	{
499	return &memcg->css;
500	}
501
502	static struct mem_cgroup_per_zone *
503	page_cgroup_zoneinfo(struct mem_cgroup memcg, struct page page)
504	{
505	int nid = page_to_nid(page);
506	int zid = page_zonenum(page);
507
508	return mem_cgroup_zoneinfo(memcg, nid, zid);
509	}
510
511	static struct mem_cgroup_tree_per_zone *
512	soft_limit_tree_node_zone(int nid, int zid)
513	{
514	return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
515	}
516
517	static struct mem_cgroup_tree_per_zone *
518	soft_limit_tree_from_page(struct page *page)
519	{
520	int nid = page_to_nid(page);
521	int zid = page_zonenum(page);
522
523	return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
524	}
525
526	static void
527	__mem_cgroup_insert_exceeded(struct mem_cgroup *memcg,
528	struct mem_cgroup_per_zone *mz,
529	struct mem_cgroup_tree_per_zone *mctz,
530	unsigned long long new_usage_in_excess)
531	{
532	struct rb_node **p = &mctz->rb_root.rb_node;
533	struct rb_node *parent = NULL;
534	struct mem_cgroup_per_zone *mz_node;
535
536	if (mz->on_tree)
537	return;
538
539	mz->usage_in_excess = new_usage_in_excess;
540	if (!mz->usage_in_excess)
541	return;
542	while (*p) {
543	parent = *p;
544	mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
545	tree_node);
546	if (mz->usage_in_excess < mz_node->usage_in_excess)
547	p = &(*p)->rb_left;
548	/*
549	* We can't avoid mem cgroups that are over their soft
550	* limit by the same amount
551	*/
552	else if (mz->usage_in_excess >= mz_node->usage_in_excess)
553	p = &(*p)->rb_right;
554	}
555	rb_link_node(&mz->tree_node, parent, p);
556	rb_insert_color(&mz->tree_node, &mctz->rb_root);
557	mz->on_tree = true;
558	}
559
560	static void
561	__mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
562	struct mem_cgroup_per_zone *mz,
563	struct mem_cgroup_tree_per_zone *mctz)
564	{
565	if (!mz->on_tree)
566	return;
567	rb_erase(&mz->tree_node, &mctz->rb_root);
568	mz->on_tree = false;
569	}
570
571	static void
572	mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
573	struct mem_cgroup_per_zone *mz,
574	struct mem_cgroup_tree_per_zone *mctz)
575	{
576	spin_lock(&mctz->lock);
577	__mem_cgroup_remove_exceeded(memcg, mz, mctz);
578	spin_unlock(&mctz->lock);
579	}
580
581
582	static void mem_cgroup_update_tree(struct mem_cgroup memcg, struct page page)
583	{
584	unsigned long long excess;
585	struct mem_cgroup_per_zone *mz;
586	struct mem_cgroup_tree_per_zone *mctz;
587	int nid = page_to_nid(page);
588	int zid = page_zonenum(page);
589	mctz = soft_limit_tree_from_page(page);
590
591	/*
592	* Necessary to update all ancestors when hierarchy is used.
593	* because their event counter is not touched.
594	*/
595	for (; memcg; memcg = parent_mem_cgroup(memcg)) {
596	mz = mem_cgroup_zoneinfo(memcg, nid, zid);
597	excess = res_counter_soft_limit_excess(&memcg->res);
598	/*
599	* We have to update the tree if mz is on RB-tree or
600	* mem is over its softlimit.
601	*/
602	if (excess \|\| mz->on_tree) {
603	spin_lock(&mctz->lock);
604	/* if on-tree, remove it */
605	if (mz->on_tree)
606	__mem_cgroup_remove_exceeded(memcg, mz, mctz);
607	/*
608	* Insert again. mz->usage_in_excess will be updated.
609	* If excess is 0, no tree ops.
610	*/
611	__mem_cgroup_insert_exceeded(memcg, mz, mctz, excess);
612	spin_unlock(&mctz->lock);
613	}
614	}
615	}
616
617	static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
618	{
619	int node, zone;
620	struct mem_cgroup_per_zone *mz;
621	struct mem_cgroup_tree_per_zone *mctz;
622
623	for_each_node(node) {
624	for (zone = 0; zone < MAX_NR_ZONES; zone++) {
625	mz = mem_cgroup_zoneinfo(memcg, node, zone);
626	mctz = soft_limit_tree_node_zone(node, zone);
627	mem_cgroup_remove_exceeded(memcg, mz, mctz);
628	}
629	}
630	}
631
632	static struct mem_cgroup_per_zone *
633	__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
634	{
635	struct rb_node *rightmost = NULL;
636	struct mem_cgroup_per_zone *mz;
637
638	retry:
639	mz = NULL;
640	rightmost = rb_last(&mctz->rb_root);
641	if (!rightmost)
642	goto done; /* Nothing to reclaim from */
643
644	mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
645	/*
646	* Remove the node now but someone else can add it back,
647	* we will to add it back at the end of reclaim to its correct
648	* position in the tree.
649	*/
650	__mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
651	if (!res_counter_soft_limit_excess(&mz->memcg->res) \|\|
652	!css_tryget(&mz->memcg->css))
653	goto retry;
654	done:
655	return mz;
656	}
657
658	static struct mem_cgroup_per_zone *
659	mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
660	{
661	struct mem_cgroup_per_zone *mz;
662
663	spin_lock(&mctz->lock);
664	mz = __mem_cgroup_largest_soft_limit_node(mctz);
665	spin_unlock(&mctz->lock);
666	return mz;
667	}
668
669	/*
670	* Implementation Note: reading percpu statistics for memcg.
671	*
672	* Both of vmstat[] and percpu_counter has threshold and do periodic
673	* synchronization to implement "quick" read. There are trade-off between
674	* reading cost and precision of value. Then, we may have a chance to implement
675	* a periodic synchronizion of counter in memcg's counter.
676	*
677	* But this _read() function is used for user interface now. The user accounts
678	* memory usage by memory cgroup and he _always_ requires exact value because
679	* he accounts memory. Even if we provide quick-and-fuzzy read, we always
680	* have to visit all online cpus and make sum. So, for now, unnecessary
681	* synchronization is not implemented. (just implemented for cpu hotplug)
682	*
683	* If there are kernel internal actions which can make use of some not-exact
684	* value, and reading all cpu value can be performance bottleneck in some
685	* common workload, threashold and synchonization as vmstat[] should be
686	* implemented.
687	*/
688	static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
689	enum mem_cgroup_stat_index idx)
690	{
691	long val = 0;
692	int cpu;
693
694	get_online_cpus();
695	for_each_online_cpu(cpu)
696	val += per_cpu(memcg->stat->count[idx], cpu);
697	#ifdef CONFIG_HOTPLUG_CPU
698	spin_lock(&memcg->pcp_counter_lock);
699	val += memcg->nocpu_base.count[idx];
700	spin_unlock(&memcg->pcp_counter_lock);
701	#endif
702	put_online_cpus();
703	return val;
704	}
705
706	static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
707	bool charge)
708	{
709	int val = (charge) ? 1 : -1;
710	this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
711	}
712
713	static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
714	enum mem_cgroup_events_index idx)
715	{
716	unsigned long val = 0;
717	int cpu;
718
719	for_each_online_cpu(cpu)
720	val += per_cpu(memcg->stat->events[idx], cpu);
721	#ifdef CONFIG_HOTPLUG_CPU
722	spin_lock(&memcg->pcp_counter_lock);
723	val += memcg->nocpu_base.events[idx];
724	spin_unlock(&memcg->pcp_counter_lock);
725	#endif
726	return val;
727	}
728
729	static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
730	bool anon, int nr_pages)
731	{
732	preempt_disable();
733
734	/*
735	* Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
736	* counted as CACHE even if it's on ANON LRU.
737	*/
738	if (anon)
739	__this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
740	nr_pages);
741	else
742	__this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
743	nr_pages);
744
745	/* pagein of a big page is an event. So, ignore page size */
746	if (nr_pages > 0)
747	__this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
748	else {
749	__this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
750	nr_pages = -nr_pages; /* for event */
751	}
752
753	__this_cpu_add(memcg->stat->nr_page_events, nr_pages);
754
755	preempt_enable();
756	}
757
758	unsigned long
759	mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
760	{
761	struct mem_cgroup_per_zone *mz;
762
763	mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
764	return mz->lru_size[lru];
765	}
766
767	static unsigned long
768	mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid,
769	unsigned int lru_mask)
770	{
771	struct mem_cgroup_per_zone *mz;
772	enum lru_list lru;
773	unsigned long ret = 0;
774
775	mz = mem_cgroup_zoneinfo(memcg, nid, zid);
776
777	for_each_lru(lru) {
778	if (BIT(lru) & lru_mask)
779	ret += mz->lru_size[lru];
780	}
781	return ret;
782	}
783
784	static unsigned long
785	mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
786	int nid, unsigned int lru_mask)
787	{
788	u64 total = 0;
789	int zid;
790
791	for (zid = 0; zid < MAX_NR_ZONES; zid++)
792	total += mem_cgroup_zone_nr_lru_pages(memcg,
793	nid, zid, lru_mask);
794
795	return total;
796	}
797
798	static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
799	unsigned int lru_mask)
800	{
801	int nid;
802	u64 total = 0;
803
804	for_each_node_state(nid, N_HIGH_MEMORY)
805	total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
806	return total;
807	}
808
809	static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
810	enum mem_cgroup_events_target target)
811	{
812	unsigned long val, next;
813
814	val = __this_cpu_read(memcg->stat->nr_page_events);
815	next = __this_cpu_read(memcg->stat->targets[target]);
816	/* from time_after() in jiffies.h */
817	if ((long)next - (long)val < 0) {
818	switch (target) {
819	case MEM_CGROUP_TARGET_THRESH:
820	next = val + THRESHOLDS_EVENTS_TARGET;
821	break;
822	case MEM_CGROUP_TARGET_SOFTLIMIT:
823	next = val + SOFTLIMIT_EVENTS_TARGET;
824	break;
825	case MEM_CGROUP_TARGET_NUMAINFO:
826	next = val + NUMAINFO_EVENTS_TARGET;
827	break;
828	default:
829	break;
830	}
831	__this_cpu_write(memcg->stat->targets[target], next);
832	return true;
833	}
834	return false;
835	}
836
837	/*
838	* Check events in order.
839	*
840	*/
841	static void memcg_check_events(struct mem_cgroup memcg, struct page page)
842	{
843	preempt_disable();
844	/* threshold event is triggered in finer grain than soft limit */
845	if (unlikely(mem_cgroup_event_ratelimit(memcg,
846	MEM_CGROUP_TARGET_THRESH))) {
847	bool do_softlimit;
848	bool do_numainfo __maybe_unused;
849
850	do_softlimit = mem_cgroup_event_ratelimit(memcg,
851	MEM_CGROUP_TARGET_SOFTLIMIT);
852	#if MAX_NUMNODES > 1
853	do_numainfo = mem_cgroup_event_ratelimit(memcg,
854	MEM_CGROUP_TARGET_NUMAINFO);
855	#endif
856	preempt_enable();
857
858	mem_cgroup_threshold(memcg);
859	if (unlikely(do_softlimit))
860	mem_cgroup_update_tree(memcg, page);
861	#if MAX_NUMNODES > 1
862	if (unlikely(do_numainfo))
863	atomic_inc(&memcg->numainfo_events);
864	#endif
865	} else
866	preempt_enable();
867	}
868
869	struct mem_cgroup mem_cgroup_from_cont(struct cgroup cont)
870	{
871	return mem_cgroup_from_css(
872	cgroup_subsys_state(cont, mem_cgroup_subsys_id));
873	}
874
875	struct mem_cgroup mem_cgroup_from_task(struct task_struct p)
876	{
877	/*
878	* mm_update_next_owner() may clear mm->owner to NULL
879	* if it races with swapoff, page migration, etc.
880	* So this can be called with p == NULL.
881	*/
882	if (unlikely(!p))
883	return NULL;
884
885	return mem_cgroup_from_css(task_subsys_state(p, mem_cgroup_subsys_id));
886	}
887
888	struct mem_cgroup try_get_mem_cgroup_from_mm(struct mm_struct mm)
889	{
890	struct mem_cgroup *memcg = NULL;
891
892	if (!mm)
893	return NULL;
894	/*
895	* Because we have no locks, mm->owner's may be being moved to other
896	* cgroup. We use css_tryget() here even if this looks
897	* pessimistic (rather than adding locks here).
898	*/
899	rcu_read_lock();
900	do {
901	memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
902	if (unlikely(!memcg))
903	break;
904	} while (!css_tryget(&memcg->css));
905	rcu_read_unlock();
906	return memcg;
907	}
908
909	/**
910	* mem_cgroup_iter - iterate over memory cgroup hierarchy
911	* @root: hierarchy root
912	* @prev: previously returned memcg, NULL on first invocation
913	* @reclaim: cookie for shared reclaim walks, NULL for full walks
914	*
915	* Returns references to children of the hierarchy below @root, or
916	* @root itself, or %NULL after a full round-trip.
917	*
918	* Caller must pass the return value in @prev on subsequent
919	* invocations for reference counting, or use mem_cgroup_iter_break()
920	* to cancel a hierarchy walk before the round-trip is complete.
921	*
922	* Reclaimers can specify a zone and a priority level in @reclaim to
923	* divide up the memcgs in the hierarchy among all concurrent
924	* reclaimers operating on the same zone and priority.
925	*/
926	struct mem_cgroup mem_cgroup_iter(struct mem_cgroup root,
927	struct mem_cgroup *prev,
928	struct mem_cgroup_reclaim_cookie *reclaim)
929	{
930	struct mem_cgroup *memcg = NULL;
931	int id = 0;
932
933	if (mem_cgroup_disabled())
934	return NULL;
935
936	if (!root)
937	root = root_mem_cgroup;
938
939	if (prev && !reclaim)
940	id = css_id(&prev->css);
941
942	if (prev && prev != root)
943	css_put(&prev->css);
944
945	if (!root->use_hierarchy && root != root_mem_cgroup) {
946	if (prev)
947	return NULL;
948	return root;
949	}
950
951	while (!memcg) {
952	struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
953	struct cgroup_subsys_state *css;
954
955	if (reclaim) {
956	int nid = zone_to_nid(reclaim->zone);
957	int zid = zone_idx(reclaim->zone);
958	struct mem_cgroup_per_zone *mz;
959
960	mz = mem_cgroup_zoneinfo(root, nid, zid);
961	iter = &mz->reclaim_iter[reclaim->priority];
962	if (prev && reclaim->generation != iter->generation)
963	return NULL;
964	id = iter->position;
965	}
966
967	rcu_read_lock();
968	css = css_get_next(&mem_cgroup_subsys, id + 1, &root->css, &id);
969	if (css) {
970	if (css == &root->css \|\| css_tryget(css))
971	memcg = mem_cgroup_from_css(css);
972	} else
973	id = 0;
974	rcu_read_unlock();
975
976	if (reclaim) {
977	iter->position = id;
978	if (!css)
979	iter->generation++;
980	else if (!prev && memcg)
981	reclaim->generation = iter->generation;
982	}
983
984	if (prev && !css)
985	return NULL;
986	}
987	return memcg;
988	}
989
990	/**
991	* mem_cgroup_iter_break - abort a hierarchy walk prematurely
992	* @root: hierarchy root
993	* @prev: last visited hierarchy member as returned by mem_cgroup_iter()
994	*/
995	void mem_cgroup_iter_break(struct mem_cgroup *root,
996	struct mem_cgroup *prev)
997	{
998	if (!root)
999	root = root_mem_cgroup;
1000	if (prev && prev != root)
1001	css_put(&prev->css);
1002	}
1003
1004	/*
1005	* Iteration constructs for visiting all cgroups (under a tree). If
1006	* loops are exited prematurely (break), mem_cgroup_iter_break() must
1007	* be used for reference counting.
1008	*/
1009	#define for_each_mem_cgroup_tree(iter, root) \
1010	for (iter = mem_cgroup_iter(root, NULL, NULL); \
1011	iter != NULL; \
1012	iter = mem_cgroup_iter(root, iter, NULL))
1013
1014	#define for_each_mem_cgroup(iter) \
1015	for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
1016	iter != NULL; \
1017	iter = mem_cgroup_iter(NULL, iter, NULL))
1018
1019	static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
1020	{
1021	return (memcg == root_mem_cgroup);
1022	}
1023
1024	void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
1025	{
1026	struct mem_cgroup *memcg;
1027
1028	if (!mm)
1029	return;
1030
1031	rcu_read_lock();
1032	memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1033	if (unlikely(!memcg))
1034	goto out;
1035
1036	switch (idx) {
1037	case PGFAULT:
1038	this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
1039	break;
1040	case PGMAJFAULT:
1041	this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
1042	break;
1043	default:
1044	BUG();
1045	}
1046	out:
1047	rcu_read_unlock();
1048	}
1049	EXPORT_SYMBOL(mem_cgroup_count_vm_event);
1050
1051	/**
1052	* mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1053	* @zone: zone of the wanted lruvec
1054	* @memcg: memcg of the wanted lruvec
1055	*
1056	* Returns the lru list vector holding pages for the given @zone and
1057	* @mem. This can be the global zone lruvec, if the memory controller
1058	* is disabled.
1059	*/
1060	struct lruvec mem_cgroup_zone_lruvec(struct zone zone,
1061	struct mem_cgroup *memcg)
1062	{
1063	struct mem_cgroup_per_zone *mz;
1064
1065	if (mem_cgroup_disabled())
1066	return &zone->lruvec;
1067
1068	mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone));
1069	return &mz->lruvec;
1070	}
1071
1072	/*
1073	* Following LRU functions are allowed to be used without PCG_LOCK.
1074	* Operations are called by routine of global LRU independently from memcg.
1075	* What we have to take care of here is validness of pc->mem_cgroup.
1076	*
1077	* Changes to pc->mem_cgroup happens when
1078	* 1. charge
1079	* 2. moving account
1080	* In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1081	* It is added to LRU before charge.
1082	* If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1083	* When moving account, the page is not on LRU. It's isolated.
1084	*/
1085
1086	/**
1087	* mem_cgroup_page_lruvec - return lruvec for adding an lru page
1088	* @page: the page
1089	* @zone: zone of the page
1090	*/
1091	struct lruvec mem_cgroup_page_lruvec(struct page page, struct zone *zone)
1092	{
1093	struct mem_cgroup_per_zone *mz;
1094	struct mem_cgroup *memcg;
1095	struct page_cgroup *pc;
1096
1097	if (mem_cgroup_disabled())
1098	return &zone->lruvec;
1099
1100	pc = lookup_page_cgroup(page);
1101	memcg = pc->mem_cgroup;
1102
1103	/*
1104	* Surreptitiously switch any uncharged offlist page to root:
1105	* an uncharged page off lru does nothing to secure
1106	* its former mem_cgroup from sudden removal.
1107	*
1108	* Our caller holds lru_lock, and PageCgroupUsed is updated
1109	* under page_cgroup lock: between them, they make all uses
1110	* of pc->mem_cgroup safe.
1111	*/
1112	if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup)
1113	pc->mem_cgroup = memcg = root_mem_cgroup;
1114
1115	mz = page_cgroup_zoneinfo(memcg, page);
1116	return &mz->lruvec;
1117	}
1118
1119	/**
1120	* mem_cgroup_update_lru_size - account for adding or removing an lru page
1121	* @lruvec: mem_cgroup per zone lru vector
1122	* @lru: index of lru list the page is sitting on
1123	* @nr_pages: positive when adding or negative when removing
1124	*
1125	* This function must be called when a page is added to or removed from an
1126	* lru list.
1127	*/
1128	void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1129	int nr_pages)
1130	{
1131	struct mem_cgroup_per_zone *mz;
1132	unsigned long *lru_size;
1133
1134	if (mem_cgroup_disabled())
1135	return;
1136
1137	mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1138	lru_size = mz->lru_size + lru;
1139	*lru_size += nr_pages;
1140	VM_BUG_ON((long)(*lru_size) < 0);
1141	}
1142
1143	/*
1144	* Checks whether given mem is same or in the root_mem_cgroup's
1145	* hierarchy subtree
1146	*/
1147	bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1148	struct mem_cgroup *memcg)
1149	{
1150	if (root_memcg == memcg)
1151	return true;
1152	if (!root_memcg->use_hierarchy \|\| !memcg)
1153	return false;
1154	return css_is_ancestor(&memcg->css, &root_memcg->css);
1155	}
1156
1157	static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1158	struct mem_cgroup *memcg)
1159	{
1160	bool ret;
1161
1162	rcu_read_lock();
1163	ret = __mem_cgroup_same_or_subtree(root_memcg, memcg);
1164	rcu_read_unlock();
1165	return ret;
1166	}
1167
1168	int task_in_mem_cgroup(struct task_struct task, const struct mem_cgroup memcg)
1169	{
1170	int ret;
1171	struct mem_cgroup *curr = NULL;
1172	struct task_struct *p;
1173
1174	p = find_lock_task_mm(task);
1175	if (p) {
1176	curr = try_get_mem_cgroup_from_mm(p->mm);
1177	task_unlock(p);
1178	} else {
1179	/*
1180	* All threads may have already detached their mm's, but the oom
1181	* killer still needs to detect if they have already been oom
1182	* killed to prevent needlessly killing additional tasks.
1183	*/
1184	task_lock(task);
1185	curr = mem_cgroup_from_task(task);
1186	if (curr)
1187	css_get(&curr->css);
1188	task_unlock(task);
1189	}
1190	if (!curr)
1191	return 0;
1192	/*
1193	* We should check use_hierarchy of "memcg" not "curr". Because checking
1194	* use_hierarchy of "curr" here make this function true if hierarchy is
1195	* enabled in "curr" and "curr" is a child of "memcg" in cgroup
1196	* hierarchy(even if use_hierarchy is disabled in "memcg").
1197	*/
1198	ret = mem_cgroup_same_or_subtree(memcg, curr);
1199	css_put(&curr->css);
1200	return ret;
1201	}
1202
1203	int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
1204	{
1205	unsigned long inactive_ratio;
1206	unsigned long inactive;
1207	unsigned long active;
1208	unsigned long gb;
1209
1210	inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
1211	active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
1212
1213	gb = (inactive + active) >> (30 - PAGE_SHIFT);
1214	if (gb)
1215	inactive_ratio = int_sqrt(10 * gb);
1216	else
1217	inactive_ratio = 1;
1218
1219	return inactive * inactive_ratio < active;
1220	}
1221
1222	int mem_cgroup_inactive_file_is_low(struct lruvec *lruvec)
1223	{
1224	unsigned long active;
1225	unsigned long inactive;
1226
1227	inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_FILE);
1228	active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_FILE);
1229
1230	return (active > inactive);
1231	}
1232
1233	#define mem_cgroup_from_res_counter(counter, member) \
1234	container_of(counter, struct mem_cgroup, member)
1235
1236	/**
1237	* mem_cgroup_margin - calculate chargeable space of a memory cgroup
1238	* @memcg: the memory cgroup
1239	*
1240	* Returns the maximum amount of memory @mem can be charged with, in
1241	* pages.
1242	*/
1243	static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
1244	{
1245	unsigned long long margin;
1246
1247	margin = res_counter_margin(&memcg->res);
1248	if (do_swap_account)
1249	margin = min(margin, res_counter_margin(&memcg->memsw));
1250	return margin >> PAGE_SHIFT;
1251	}
1252
1253	int mem_cgroup_swappiness(struct mem_cgroup *memcg)
1254	{
1255	struct cgroup *cgrp = memcg->css.cgroup;
1256
1257	/* root ? */
1258	if (cgrp->parent == NULL)
1259	return vm_swappiness;
1260
1261	return memcg->swappiness;
1262	}
1263
1264	/*
1265	* memcg->moving_account is used for checking possibility that some thread is
1266	* calling move_account(). When a thread on CPU-A starts moving pages under
1267	* a memcg, other threads should check memcg->moving_account under
1268	* rcu_read_lock(), like this:
1269	*
1270	* CPU-A CPU-B
1271	* rcu_read_lock()
1272	* memcg->moving_account+1 if (memcg->mocing_account)
1273	* take heavy locks.
1274	* synchronize_rcu() update something.
1275	* rcu_read_unlock()
1276	* start move here.
1277	*/
1278
1279	/* for quick checking without looking up memcg */
1280	atomic_t memcg_moving __read_mostly;
1281
1282	static void mem_cgroup_start_move(struct mem_cgroup *memcg)
1283	{
1284	atomic_inc(&memcg_moving);
1285	atomic_inc(&memcg->moving_account);
1286	synchronize_rcu();
1287	}
1288
1289	static void mem_cgroup_end_move(struct mem_cgroup *memcg)
1290	{
1291	/*
1292	* Now, mem_cgroup_clear_mc() may call this function with NULL.
1293	* We check NULL in callee rather than caller.
1294	*/
1295	if (memcg) {
1296	atomic_dec(&memcg_moving);
1297	atomic_dec(&memcg->moving_account);
1298	}
1299	}
1300
1301	/*
1302	* 2 routines for checking "mem" is under move_account() or not.
1303	*
1304	* mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1305	* is used for avoiding races in accounting. If true,
1306	* pc->mem_cgroup may be overwritten.
1307	*
1308	* mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1309	* under hierarchy of moving cgroups. This is for
1310	* waiting at hith-memory prressure caused by "move".
1311	*/
1312
1313	static bool mem_cgroup_stolen(struct mem_cgroup *memcg)
1314	{
1315	VM_BUG_ON(!rcu_read_lock_held());
1316	return atomic_read(&memcg->moving_account) > 0;
1317	}
1318
1319	static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
1320	{
1321	struct mem_cgroup *from;
1322	struct mem_cgroup *to;
1323	bool ret = false;
1324	/*
1325	* Unlike task_move routines, we access mc.to, mc.from not under
1326	* mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1327	*/
1328	spin_lock(&mc.lock);
1329	from = mc.from;
1330	to = mc.to;
1331	if (!from)
1332	goto unlock;
1333
1334	ret = mem_cgroup_same_or_subtree(memcg, from)
1335	\|\| mem_cgroup_same_or_subtree(memcg, to);
1336	unlock:
1337	spin_unlock(&mc.lock);
1338	return ret;
1339	}
1340
1341	static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
1342	{
1343	if (mc.moving_task && current != mc.moving_task) {
1344	if (mem_cgroup_under_move(memcg)) {
1345	DEFINE_WAIT(wait);
1346	prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1347	/* moving charge context might have finished. */
1348	if (mc.moving_task)
1349	schedule();
1350	finish_wait(&mc.waitq, &wait);
1351	return true;
1352	}
1353	}
1354	return false;
1355	}
1356
1357	/*
1358	* Take this lock when
1359	* - a code tries to modify page's memcg while it's USED.
1360	* - a code tries to modify page state accounting in a memcg.
1361	* see mem_cgroup_stolen(), too.
1362	*/
1363	static void move_lock_mem_cgroup(struct mem_cgroup *memcg,
1364	unsigned long *flags)
1365	{
1366	spin_lock_irqsave(&memcg->move_lock, *flags);
1367	}
1368
1369	static void move_unlock_mem_cgroup(struct mem_cgroup *memcg,
1370	unsigned long *flags)
1371	{
1372	spin_unlock_irqrestore(&memcg->move_lock, *flags);
1373	}
1374
1375	/**
1376	* mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1377	* @memcg: The memory cgroup that went over limit
1378	* @p: Task that is going to be killed
1379	*
1380	* NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1381	* enabled
1382	*/
1383	void mem_cgroup_print_oom_info(struct mem_cgroup memcg, struct task_struct p)
1384	{
1385	struct cgroup *task_cgrp;
1386	struct cgroup *mem_cgrp;
1387	/*
1388	* Need a buffer in BSS, can't rely on allocations. The code relies
1389	* on the assumption that OOM is serialized for memory controller.
1390	* If this assumption is broken, revisit this code.
1391	*/
1392	static char memcg_name[PATH_MAX];
1393	int ret;
1394
1395	if (!memcg \|\| !p)
1396	return;
1397
1398	rcu_read_lock();
1399
1400	mem_cgrp = memcg->css.cgroup;
1401	task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1402
1403	ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1404	if (ret < 0) {
1405	/*
1406	* Unfortunately, we are unable to convert to a useful name
1407	* But we'll still print out the usage information
1408	*/
1409	rcu_read_unlock();
1410	goto done;
1411	}
1412	rcu_read_unlock();
1413
1414	printk(KERN_INFO "Task in %s killed", memcg_name);
1415
1416	rcu_read_lock();
1417	ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1418	if (ret < 0) {
1419	rcu_read_unlock();
1420	goto done;
1421	}
1422	rcu_read_unlock();
1423
1424	/*
1425	* Continues from above, so we don't need an KERN_ level
1426	*/
1427	printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
1428	done:
1429
1430	printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
1431	res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1432	res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1433	res_counter_read_u64(&memcg->res, RES_FAILCNT));
1434	printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
1435	"failcnt %llu\n",
1436	res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1437	res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1438	res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
1439	}
1440
1441	/*
1442	* This function returns the number of memcg under hierarchy tree. Returns
1443	* 1(self count) if no children.
1444	*/
1445	static int mem_cgroup_count_children(struct mem_cgroup *memcg)
1446	{
1447	int num = 0;
1448	struct mem_cgroup *iter;
1449
1450	for_each_mem_cgroup_tree(iter, memcg)
1451	num++;
1452	return num;
1453	}
1454
1455	/*
1456	* Return the memory (and swap, if configured) limit for a memcg.
1457	*/
1458	static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
1459	{
1460	u64 limit;
1461	u64 memsw;
1462
1463	limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1464	limit += total_swap_pages << PAGE_SHIFT;
1465
1466	memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1467	/*
1468	* If memsw is finite and limits the amount of swap space available
1469	* to this memcg, return that limit.
1470	*/
1471	return min(limit, memsw);
1472	}
1473
1474	void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1475	int order)
1476	{
1477	struct mem_cgroup *iter;
1478	unsigned long chosen_points = 0;
1479	unsigned long totalpages;
1480	unsigned int points = 0;
1481	struct task_struct *chosen = NULL;
1482
1483	/*
1484	* If current has a pending SIGKILL, then automatically select it. The
1485	* goal is to allow it to allocate so that it may quickly exit and free
1486	* its memory.
1487	*/
1488	if (fatal_signal_pending(current)) {
1489	set_thread_flag(TIF_MEMDIE);
1490	return;
1491	}
1492
1493	check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL);
1494	totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1;
1495	for_each_mem_cgroup_tree(iter, memcg) {
1496	struct cgroup *cgroup = iter->css.cgroup;
1497	struct cgroup_iter it;
1498	struct task_struct *task;
1499
1500	cgroup_iter_start(cgroup, &it);
1501	while ((task = cgroup_iter_next(cgroup, &it))) {
1502	switch (oom_scan_process_thread(task, totalpages, NULL,
1503	false)) {
1504	case OOM_SCAN_SELECT:
1505	if (chosen)
1506	put_task_struct(chosen);
1507	chosen = task;
1508	chosen_points = ULONG_MAX;
1509	get_task_struct(chosen);
1510	/* fall through */
1511	case OOM_SCAN_CONTINUE:
1512	continue;
1513	case OOM_SCAN_ABORT:
1514	cgroup_iter_end(cgroup, &it);
1515	mem_cgroup_iter_break(memcg, iter);
1516	if (chosen)
1517	put_task_struct(chosen);
1518	return;
1519	case OOM_SCAN_OK:
1520	break;
1521	};
1522	points = oom_badness(task, memcg, NULL, totalpages);
1523	if (points > chosen_points) {
1524	if (chosen)
1525	put_task_struct(chosen);
1526	chosen = task;
1527	chosen_points = points;
1528	get_task_struct(chosen);
1529	}
1530	}
1531	cgroup_iter_end(cgroup, &it);
1532	}
1533
1534	if (!chosen)
1535	return;
1536	points = chosen_points * 1000 / totalpages;
1537	oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg,
1538	NULL, "Memory cgroup out of memory");
1539	}
1540
1541	static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg,
1542	gfp_t gfp_mask,
1543	unsigned long flags)
1544	{
1545	unsigned long total = 0;
1546	bool noswap = false;
1547	int loop;
1548
1549	if (flags & MEM_CGROUP_RECLAIM_NOSWAP)
1550	noswap = true;
1551	if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum)
1552	noswap = true;
1553
1554	for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) {
1555	if (loop)
1556	drain_all_stock_async(memcg);
1557	total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap);
1558	/*
1559	* Allow limit shrinkers, which are triggered directly
1560	* by userspace, to catch signals and stop reclaim
1561	* after minimal progress, regardless of the margin.
1562	*/
1563	if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK))
1564	break;
1565	if (mem_cgroup_margin(memcg))
1566	break;
1567	/*
1568	* If nothing was reclaimed after two attempts, there
1569	* may be no reclaimable pages in this hierarchy.
1570	*/
1571	if (loop && !total)
1572	break;
1573	}
1574	return total;
1575	}
1576
1577	/**
1578	* test_mem_cgroup_node_reclaimable
1579	* @memcg: the target memcg
1580	* @nid: the node ID to be checked.
1581	* @noswap : specify true here if the user wants flle only information.
1582	*
1583	* This function returns whether the specified memcg contains any
1584	* reclaimable pages on a node. Returns true if there are any reclaimable
1585	* pages in the node.
1586	*/
1587	static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
1588	int nid, bool noswap)
1589	{
1590	if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
1591	return true;
1592	if (noswap \|\| !total_swap_pages)
1593	return false;
1594	if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
1595	return true;
1596	return false;
1597
1598	}
1599	#if MAX_NUMNODES > 1
1600
1601	/*
1602	* Always updating the nodemask is not very good - even if we have an empty
1603	* list or the wrong list here, we can start from some node and traverse all
1604	* nodes based on the zonelist. So update the list loosely once per 10 secs.
1605	*
1606	*/
1607	static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
1608	{
1609	int nid;
1610	/*
1611	* numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1612	* pagein/pageout changes since the last update.
1613	*/
1614	if (!atomic_read(&memcg->numainfo_events))
1615	return;
1616	if (atomic_inc_return(&memcg->numainfo_updating) > 1)
1617	return;
1618
1619	/* make a nodemask where this memcg uses memory from */
1620	memcg->scan_nodes = node_states[N_HIGH_MEMORY];
1621
1622	for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) {
1623
1624	if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1625	node_clear(nid, memcg->scan_nodes);
1626	}
1627
1628	atomic_set(&memcg->numainfo_events, 0);
1629	atomic_set(&memcg->numainfo_updating, 0);
1630	}
1631
1632	/*
1633	* Selecting a node where we start reclaim from. Because what we need is just
1634	* reducing usage counter, start from anywhere is O,K. Considering
1635	* memory reclaim from current node, there are pros. and cons.
1636	*
1637	* Freeing memory from current node means freeing memory from a node which
1638	* we'll use or we've used. So, it may make LRU bad. And if several threads
1639	* hit limits, it will see a contention on a node. But freeing from remote
1640	* node means more costs for memory reclaim because of memory latency.
1641	*
1642	* Now, we use round-robin. Better algorithm is welcomed.
1643	*/
1644	int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
1645	{
1646	int node;
1647
1648	mem_cgroup_may_update_nodemask(memcg);
1649	node = memcg->last_scanned_node;
1650
1651	node = next_node(node, memcg->scan_nodes);
1652	if (node == MAX_NUMNODES)
1653	node = first_node(memcg->scan_nodes);
1654	/*
1655	* We call this when we hit limit, not when pages are added to LRU.
1656	* No LRU may hold pages because all pages are UNEVICTABLE or
1657	* memcg is too small and all pages are not on LRU. In that case,
1658	* we use curret node.
1659	*/
1660	if (unlikely(node == MAX_NUMNODES))
1661	node = numa_node_id();
1662
1663	memcg->last_scanned_node = node;
1664	return node;
1665	}
1666
1667	/*
1668	* Check all nodes whether it contains reclaimable pages or not.
1669	* For quick scan, we make use of scan_nodes. This will allow us to skip
1670	* unused nodes. But scan_nodes is lazily updated and may not cotain
1671	* enough new information. We need to do double check.
1672	*/
1673	static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
1674	{
1675	int nid;
1676
1677	/*
1678	* quick check...making use of scan_node.
1679	* We can skip unused nodes.
1680	*/
1681	if (!nodes_empty(memcg->scan_nodes)) {
1682	for (nid = first_node(memcg->scan_nodes);
1683	nid < MAX_NUMNODES;
1684	nid = next_node(nid, memcg->scan_nodes)) {
1685
1686	if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
1687	return true;
1688	}
1689	}
1690	/*
1691	* Check rest of nodes.
1692	*/
1693	for_each_node_state(nid, N_HIGH_MEMORY) {
1694	if (node_isset(nid, memcg->scan_nodes))
1695	continue;
1696	if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
1697	return true;
1698	}
1699	return false;
1700	}
1701
1702	#else
1703	int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
1704	{
1705	return 0;
1706	}
1707
1708	static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
1709	{
1710	return test_mem_cgroup_node_reclaimable(memcg, 0, noswap);
1711	}
1712	#endif
1713
1714	static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1715	struct zone *zone,
1716	gfp_t gfp_mask,
1717	unsigned long *total_scanned)
1718	{
1719	struct mem_cgroup *victim = NULL;
1720	int total = 0;
1721	int loop = 0;
1722	unsigned long excess;
1723	unsigned long nr_scanned;
1724	struct mem_cgroup_reclaim_cookie reclaim = {
1725	.zone = zone,
1726	.priority = 0,
1727	};
1728
1729	excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT;
1730
1731	while (1) {
1732	victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1733	if (!victim) {
1734	loop++;
1735	if (loop >= 2) {
1736	/*
1737	* If we have not been able to reclaim
1738	* anything, it might because there are
1739	* no reclaimable pages under this hierarchy
1740	*/
1741	if (!total)
1742	break;
1743	/*
1744	* We want to do more targeted reclaim.
1745	* excess >> 2 is not to excessive so as to
1746	* reclaim too much, nor too less that we keep
1747	* coming back to reclaim from this cgroup
1748	*/
1749	if (total >= (excess >> 2) \|\|
1750	(loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1751	break;
1752	}
1753	continue;
1754	}
1755	if (!mem_cgroup_reclaimable(victim, false))
1756	continue;
1757	total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1758	zone, &nr_scanned);
1759	*total_scanned += nr_scanned;
1760	if (!res_counter_soft_limit_excess(&root_memcg->res))
1761	break;
1762	}
1763	mem_cgroup_iter_break(root_memcg, victim);
1764	return total;
1765	}
1766
1767	/*
1768	* Check OOM-Killer is already running under our hierarchy.
1769	* If someone is running, return false.
1770	* Has to be called with memcg_oom_lock
1771	*/
1772	static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg)
1773	{
1774	struct mem_cgroup iter, failed = NULL;
1775
1776	for_each_mem_cgroup_tree(iter, memcg) {
1777	if (iter->oom_lock) {
1778	/*
1779	* this subtree of our hierarchy is already locked
1780	* so we cannot give a lock.
1781	*/
1782	failed = iter;
1783	mem_cgroup_iter_break(memcg, iter);
1784	break;
1785	} else
1786	iter->oom_lock = true;
1787	}
1788
1789	if (!failed)
1790	return true;
1791
1792	/*
1793	* OK, we failed to lock the whole subtree so we have to clean up
1794	* what we set up to the failing subtree
1795	*/
1796	for_each_mem_cgroup_tree(iter, memcg) {
1797	if (iter == failed) {
1798	mem_cgroup_iter_break(memcg, iter);
1799	break;
1800	}
1801	iter->oom_lock = false;
1802	}
1803	return false;
1804	}
1805
1806	/*
1807	* Has to be called with memcg_oom_lock
1808	*/
1809	static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
1810	{
1811	struct mem_cgroup *iter;
1812
1813	for_each_mem_cgroup_tree(iter, memcg)
1814	iter->oom_lock = false;
1815	return 0;
1816	}
1817
1818	static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
1819	{
1820	struct mem_cgroup *iter;
1821
1822	for_each_mem_cgroup_tree(iter, memcg)
1823	atomic_inc(&iter->under_oom);
1824	}
1825
1826	static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
1827	{
1828	struct mem_cgroup *iter;
1829
1830	/*
1831	* When a new child is created while the hierarchy is under oom,
1832	* mem_cgroup_oom_lock() may not be called. We have to use
1833	* atomic_add_unless() here.
1834	*/
1835	for_each_mem_cgroup_tree(iter, memcg)
1836	atomic_add_unless(&iter->under_oom, -1, 0);
1837	}
1838
1839	static DEFINE_SPINLOCK(memcg_oom_lock);
1840	static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1841
1842	struct oom_wait_info {
1843	struct mem_cgroup *memcg;
1844	wait_queue_t wait;
1845	};
1846
1847	static int memcg_oom_wake_function(wait_queue_t *wait,
1848	unsigned mode, int sync, void *arg)
1849	{
1850	struct mem_cgroup wake_memcg = (struct mem_cgroup )arg;
1851	struct mem_cgroup *oom_wait_memcg;
1852	struct oom_wait_info *oom_wait_info;
1853
1854	oom_wait_info = container_of(wait, struct oom_wait_info, wait);
1855	oom_wait_memcg = oom_wait_info->memcg;
1856
1857	/*
1858	* Both of oom_wait_info->memcg and wake_memcg are stable under us.
1859	* Then we can use css_is_ancestor without taking care of RCU.
1860	*/
1861	if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg)
1862	&& !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg))
1863	return 0;
1864	return autoremove_wake_function(wait, mode, sync, arg);
1865	}
1866
1867	static void memcg_wakeup_oom(struct mem_cgroup *memcg)
1868	{
1869	/* for filtering, pass "memcg" as argument. */
1870	__wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
1871	}
1872
1873	static void memcg_oom_recover(struct mem_cgroup *memcg)
1874	{
1875	if (memcg && atomic_read(&memcg->under_oom))
1876	memcg_wakeup_oom(memcg);
1877	}
1878
1879	/*
1880	* try to call OOM killer. returns false if we should exit memory-reclaim loop.
1881	*/
1882	static bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask,
1883	int order)
1884	{
1885	struct oom_wait_info owait;
1886	bool locked, need_to_kill;
1887
1888	owait.memcg = memcg;
1889	owait.wait.flags = 0;
1890	owait.wait.func = memcg_oom_wake_function;
1891	owait.wait.private = current;
1892	INIT_LIST_HEAD(&owait.wait.task_list);
1893	need_to_kill = true;
1894	mem_cgroup_mark_under_oom(memcg);
1895
1896	/* At first, try to OOM lock hierarchy under memcg.*/
1897	spin_lock(&memcg_oom_lock);
1898	locked = mem_cgroup_oom_lock(memcg);
1899	/*
1900	* Even if signal_pending(), we can't quit charge() loop without
1901	* accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1902	* under OOM is always welcomed, use TASK_KILLABLE here.
1903	*/
1904	prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
1905	if (!locked \|\| memcg->oom_kill_disable)
1906	need_to_kill = false;
1907	if (locked)
1908	mem_cgroup_oom_notify(memcg);
1909	spin_unlock(&memcg_oom_lock);
1910
1911	if (need_to_kill) {
1912	finish_wait(&memcg_oom_waitq, &owait.wait);
1913	mem_cgroup_out_of_memory(memcg, mask, order);
1914	} else {
1915	schedule();
1916	finish_wait(&memcg_oom_waitq, &owait.wait);
1917	}
1918	spin_lock(&memcg_oom_lock);
1919	if (locked)
1920	mem_cgroup_oom_unlock(memcg);
1921	memcg_wakeup_oom(memcg);
1922	spin_unlock(&memcg_oom_lock);
1923
1924	mem_cgroup_unmark_under_oom(memcg);
1925
1926	if (test_thread_flag(TIF_MEMDIE) \|\| fatal_signal_pending(current))
1927	return false;
1928	/* Give chance to dying process */
1929	schedule_timeout_uninterruptible(1);
1930	return true;
1931	}
1932
1933	/*
1934	* Currently used to update mapped file statistics, but the routine can be
1935	* generalized to update other statistics as well.
1936	*
1937	* Notes: Race condition
1938	*
1939	* We usually use page_cgroup_lock() for accessing page_cgroup member but
1940	* it tends to be costly. But considering some conditions, we doesn't need
1941	* to do so _always_.
1942	*
1943	* Considering "charge", lock_page_cgroup() is not required because all
1944	* file-stat operations happen after a page is attached to radix-tree. There
1945	* are no race with "charge".
1946	*
1947	* Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1948	* at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1949	* if there are race with "uncharge". Statistics itself is properly handled
1950	* by flags.
1951	*
1952	* Considering "move", this is an only case we see a race. To make the race
1953	* small, we check mm->moving_account and detect there are possibility of race
1954	* If there is, we take a lock.
1955	*/
1956
1957	void __mem_cgroup_begin_update_page_stat(struct page *page,
1958	bool locked, unsigned long flags)
1959	{
1960	struct mem_cgroup *memcg;
1961	struct page_cgroup *pc;
1962
1963	pc = lookup_page_cgroup(page);
1964	again:
1965	memcg = pc->mem_cgroup;
1966	if (unlikely(!memcg \|\| !PageCgroupUsed(pc)))
1967	return;
1968	/*
1969	* If this memory cgroup is not under account moving, we don't
1970	* need to take move_lock_mem_cgroup(). Because we already hold
1971	* rcu_read_lock(), any calls to move_account will be delayed until
1972	* rcu_read_unlock() if mem_cgroup_stolen() == true.
1973	*/
1974	if (!mem_cgroup_stolen(memcg))
1975	return;
1976
1977	move_lock_mem_cgroup(memcg, flags);
1978	if (memcg != pc->mem_cgroup \|\| !PageCgroupUsed(pc)) {
1979	move_unlock_mem_cgroup(memcg, flags);
1980	goto again;
1981	}
1982	*locked = true;
1983	}
1984
1985	void __mem_cgroup_end_update_page_stat(struct page page, unsigned long flags)
1986	{
1987	struct page_cgroup *pc = lookup_page_cgroup(page);
1988
1989	/*
1990	* It's guaranteed that pc->mem_cgroup never changes while
1991	* lock is held because a routine modifies pc->mem_cgroup
1992	* should take move_lock_mem_cgroup().
1993	*/
1994	move_unlock_mem_cgroup(pc->mem_cgroup, flags);
1995	}
1996
1997	void mem_cgroup_update_page_stat(struct page *page,
1998	enum mem_cgroup_page_stat_item idx, int val)
1999	{
2000	struct mem_cgroup *memcg;
2001	struct page_cgroup *pc = lookup_page_cgroup(page);
2002	unsigned long uninitialized_var(flags);
2003
2004	if (mem_cgroup_disabled())
2005	return;
2006
2007	memcg = pc->mem_cgroup;
2008	if (unlikely(!memcg \|\| !PageCgroupUsed(pc)))
2009	return;
2010
2011	switch (idx) {
2012	case MEMCG_NR_FILE_MAPPED:
2013	idx = MEM_CGROUP_STAT_FILE_MAPPED;
2014	break;
2015	default:
2016	BUG();
2017	}
2018
2019	this_cpu_add(memcg->stat->count[idx], val);
2020	}
2021
2022	/*
2023	* size of first charge trial. "32" comes from vmscan.c's magic value.
2024	* TODO: maybe necessary to use big numbers in big irons.
2025	*/
2026	#define CHARGE_BATCH 32U
2027	struct memcg_stock_pcp {
2028	struct mem_cgroup cached; / this never be root cgroup */
2029	unsigned int nr_pages;
2030	struct work_struct work;
2031	unsigned long flags;
2032	#define FLUSHING_CACHED_CHARGE 0
2033	};
2034	static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
2035	static DEFINE_MUTEX(percpu_charge_mutex);
2036
2037	/*
2038	* Try to consume stocked charge on this cpu. If success, one page is consumed
2039	* from local stock and true is returned. If the stock is 0 or charges from a
2040	* cgroup which is not current target, returns false. This stock will be
2041	* refilled.
2042	*/
2043	static bool consume_stock(struct mem_cgroup *memcg)
2044	{
2045	struct memcg_stock_pcp *stock;
2046	bool ret = true;
2047
2048	stock = &get_cpu_var(memcg_stock);
2049	if (memcg == stock->cached && stock->nr_pages)
2050	stock->nr_pages--;
2051	else /* need to call res_counter_charge */
2052	ret = false;
2053	put_cpu_var(memcg_stock);
2054	return ret;
2055	}
2056
2057	/*
2058	* Returns stocks cached in percpu to res_counter and reset cached information.
2059	*/
2060	static void drain_stock(struct memcg_stock_pcp *stock)
2061	{
2062	struct mem_cgroup *old = stock->cached;
2063
2064	if (stock->nr_pages) {
2065	unsigned long bytes = stock->nr_pages * PAGE_SIZE;
2066
2067	res_counter_uncharge(&old->res, bytes);
2068	if (do_swap_account)
2069	res_counter_uncharge(&old->memsw, bytes);
2070	stock->nr_pages = 0;
2071	}
2072	stock->cached = NULL;
2073	}
2074
2075	/*
2076	* This must be called under preempt disabled or must be called by
2077	* a thread which is pinned to local cpu.
2078	*/
2079	static void drain_local_stock(struct work_struct *dummy)
2080	{
2081	struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
2082	drain_stock(stock);
2083	clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
2084	}
2085
2086	/*
2087	* Cache charges(val) which is from res_counter, to local per_cpu area.
2088	* This will be consumed by consume_stock() function, later.
2089	*/
2090	static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
2091	{
2092	struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
2093
2094	if (stock->cached != memcg) { /* reset if necessary */
2095	drain_stock(stock);
2096	stock->cached = memcg;
2097	}
2098	stock->nr_pages += nr_pages;
2099	put_cpu_var(memcg_stock);
2100	}
2101
2102	/*
2103	* Drains all per-CPU charge caches for given root_memcg resp. subtree
2104	* of the hierarchy under it. sync flag says whether we should block
2105	* until the work is done.
2106	*/
2107	static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync)
2108	{
2109	int cpu, curcpu;
2110
2111	/* Notify other cpus that system-wide "drain" is running */
2112	get_online_cpus();
2113	curcpu = get_cpu();
2114	for_each_online_cpu(cpu) {
2115	struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
2116	struct mem_cgroup *memcg;
2117
2118	memcg = stock->cached;
2119	if (!memcg \|\| !stock->nr_pages)
2120	continue;
2121	if (!mem_cgroup_same_or_subtree(root_memcg, memcg))
2122	continue;
2123	if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2124	if (cpu == curcpu)
2125	drain_local_stock(&stock->work);
2126	else
2127	schedule_work_on(cpu, &stock->work);
2128	}
2129	}
2130	put_cpu();
2131
2132	if (!sync)
2133	goto out;
2134
2135	for_each_online_cpu(cpu) {
2136	struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
2137	if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
2138	flush_work(&stock->work);
2139	}
2140	out:
2141	put_online_cpus();
2142	}
2143
2144	/*
2145	* Tries to drain stocked charges in other cpus. This function is asynchronous
2146	* and just put a work per cpu for draining localy on each cpu. Caller can
2147	* expects some charges will be back to res_counter later but cannot wait for
2148	* it.
2149	*/
2150	static void drain_all_stock_async(struct mem_cgroup *root_memcg)
2151	{
2152	/*
2153	* If someone calls draining, avoid adding more kworker runs.
2154	*/
2155	if (!mutex_trylock(&percpu_charge_mutex))
2156	return;
2157	drain_all_stock(root_memcg, false);
2158	mutex_unlock(&percpu_charge_mutex);
2159	}
2160
2161	/* This is a synchronous drain interface. */
2162	static void drain_all_stock_sync(struct mem_cgroup *root_memcg)
2163	{
2164	/* called when force_empty is called */
2165	mutex_lock(&percpu_charge_mutex);
2166	drain_all_stock(root_memcg, true);
2167	mutex_unlock(&percpu_charge_mutex);
2168	}
2169
2170	/*
2171	* This function drains percpu counter value from DEAD cpu and
2172	* move it to local cpu. Note that this function can be preempted.
2173	*/
2174	static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
2175	{
2176	int i;
2177
2178	spin_lock(&memcg->pcp_counter_lock);
2179	for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
2180	long x = per_cpu(memcg->stat->count[i], cpu);
2181
2182	per_cpu(memcg->stat->count[i], cpu) = 0;
2183	memcg->nocpu_base.count[i] += x;
2184	}
2185	for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
2186	unsigned long x = per_cpu(memcg->stat->events[i], cpu);
2187
2188	per_cpu(memcg->stat->events[i], cpu) = 0;
2189	memcg->nocpu_base.events[i] += x;
2190	}
2191	spin_unlock(&memcg->pcp_counter_lock);
2192	}
2193
2194	static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb,
2195	unsigned long action,
2196	void *hcpu)
2197	{
2198	int cpu = (unsigned long)hcpu;
2199	struct memcg_stock_pcp *stock;
2200	struct mem_cgroup *iter;
2201
2202	if (action == CPU_ONLINE)
2203	return NOTIFY_OK;
2204
2205	if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
2206	return NOTIFY_OK;
2207
2208	for_each_mem_cgroup(iter)
2209	mem_cgroup_drain_pcp_counter(iter, cpu);
2210
2211	stock = &per_cpu(memcg_stock, cpu);
2212	drain_stock(stock);
2213	return NOTIFY_OK;
2214	}
2215
2216
2217	/* See __mem_cgroup_try_charge() for details */
2218	enum {
2219	CHARGE_OK, /* success */
2220	CHARGE_RETRY, /* need to retry but retry is not bad */
2221	CHARGE_NOMEM, /* we can't do more. return -ENOMEM */
2222	CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */
2223	CHARGE_OOM_DIE, /* the current is killed because of OOM */
2224	};
2225
2226	static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
2227	unsigned int nr_pages, bool oom_check)
2228	{
2229	unsigned long csize = nr_pages * PAGE_SIZE;
2230	struct mem_cgroup *mem_over_limit;
2231	struct res_counter *fail_res;
2232	unsigned long flags = 0;
2233	int ret;
2234
2235	ret = res_counter_charge(&memcg->res, csize, &fail_res);
2236
2237	if (likely(!ret)) {
2238	if (!do_swap_account)
2239	return CHARGE_OK;
2240	ret = res_counter_charge(&memcg->memsw, csize, &fail_res);
2241	if (likely(!ret))
2242	return CHARGE_OK;
2243
2244	res_counter_uncharge(&memcg->res, csize);
2245	mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
2246	flags \|= MEM_CGROUP_RECLAIM_NOSWAP;
2247	} else
2248	mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
2249	/*
2250	* nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2251	* of regular pages (CHARGE_BATCH), or a single regular page (1).
2252	*
2253	* Never reclaim on behalf of optional batching, retry with a
2254	* single page instead.
2255	*/
2256	if (nr_pages == CHARGE_BATCH)
2257	return CHARGE_RETRY;
2258
2259	if (!(gfp_mask & __GFP_WAIT))
2260	return CHARGE_WOULDBLOCK;
2261
2262	ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags);
2263	if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
2264	return CHARGE_RETRY;
2265	/*
2266	* Even though the limit is exceeded at this point, reclaim
2267	* may have been able to free some pages. Retry the charge
2268	* before killing the task.
2269	*
2270	* Only for regular pages, though: huge pages are rather
2271	* unlikely to succeed so close to the limit, and we fall back
2272	* to regular pages anyway in case of failure.
2273	*/
2274	if (nr_pages == 1 && ret)
2275	return CHARGE_RETRY;
2276
2277	/*
2278	* At task move, charge accounts can be doubly counted. So, it's
2279	* better to wait until the end of task_move if something is going on.
2280	*/
2281	if (mem_cgroup_wait_acct_move(mem_over_limit))
2282	return CHARGE_RETRY;
2283
2284	/* If we don't need to call oom-killer at el, return immediately */
2285	if (!oom_check)
2286	return CHARGE_NOMEM;
2287	/* check OOM */
2288	if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize)))
2289	return CHARGE_OOM_DIE;
2290
2291	return CHARGE_RETRY;
2292	}
2293
2294	/*
2295	* __mem_cgroup_try_charge() does
2296	* 1. detect memcg to be charged against from passed mm and ptr,
2297	* 2. update res_counter
2298	* 3. call memory reclaim if necessary.
2299	*
2300	* In some special case, if the task is fatal, fatal_signal_pending() or
2301	* has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2302	* to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2303	* as possible without any hazards. 2: all pages should have a valid
2304	* pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2305	* pointer, that is treated as a charge to root_mem_cgroup.
2306	*
2307	* So __mem_cgroup_try_charge() will return
2308	* 0 ... on success, filling *ptr with a valid memcg pointer.
2309	* -ENOMEM ... charge failure because of resource limits.
2310	* -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2311	*
2312	* Unlike the exported interface, an "oom" parameter is added. if oom==true,
2313	* the oom-killer can be invoked.
2314	*/
2315	static int __mem_cgroup_try_charge(struct mm_struct *mm,
2316	gfp_t gfp_mask,
2317	unsigned int nr_pages,
2318	struct mem_cgroup **ptr,
2319	bool oom)
2320	{
2321	unsigned int batch = max(CHARGE_BATCH, nr_pages);
2322	int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
2323	struct mem_cgroup *memcg = NULL;
2324	int ret;
2325
2326	/*
2327	* Unlike gloval-vm's OOM-kill, we're not in memory shortage
2328	* in system level. So, allow to go ahead dying process in addition to
2329	* MEMDIE process.
2330	*/
2331	if (unlikely(test_thread_flag(TIF_MEMDIE)
2332	\|\| fatal_signal_pending(current)))
2333	goto bypass;
2334
2335	/*
2336	* We always charge the cgroup the mm_struct belongs to.
2337	* The mm_struct's mem_cgroup changes on task migration if the
2338	* thread group leader migrates. It's possible that mm is not
2339	* set, if so charge the root memcg (happens for pagecache usage).
2340	*/
2341	if (!*ptr && !mm)
2342	*ptr = root_mem_cgroup;
2343	again:
2344	if (ptr) { / css should be a valid one */
2345	memcg = *ptr;
2346	VM_BUG_ON(css_is_removed(&memcg->css));
2347	if (mem_cgroup_is_root(memcg))
2348	goto done;
2349	if (nr_pages == 1 && consume_stock(memcg))
2350	goto done;
2351	css_get(&memcg->css);
2352	} else {
2353	struct task_struct *p;
2354
2355	rcu_read_lock();
2356	p = rcu_dereference(mm->owner);
2357	/*
2358	* Because we don't have task_lock(), "p" can exit.
2359	* In that case, "memcg" can point to root or p can be NULL with
2360	* race with swapoff. Then, we have small risk of mis-accouning.
2361	* But such kind of mis-account by race always happens because
2362	* we don't have cgroup_mutex(). It's overkill and we allo that
2363	* small race, here.
2364	* (*) swapoff at el will charge against mm-struct not against
2365	* task-struct. So, mm->owner can be NULL.
2366	*/
2367	memcg = mem_cgroup_from_task(p);
2368	if (!memcg)
2369	memcg = root_mem_cgroup;
2370	if (mem_cgroup_is_root(memcg)) {
2371	rcu_read_unlock();
2372	goto done;
2373	}
2374	if (nr_pages == 1 && consume_stock(memcg)) {
2375	/*
2376	* It seems dagerous to access memcg without css_get().
2377	* But considering how consume_stok works, it's not
2378	* necessary. If consume_stock success, some charges
2379	* from this memcg are cached on this cpu. So, we
2380	* don't need to call css_get()/css_tryget() before
2381	* calling consume_stock().
2382	*/
2383	rcu_read_unlock();
2384	goto done;
2385	}
2386	/* after here, we may be blocked. we need to get refcnt */
2387	if (!css_tryget(&memcg->css)) {
2388	rcu_read_unlock();
2389	goto again;
2390	}
2391	rcu_read_unlock();
2392	}
2393
2394	do {
2395	bool oom_check;
2396
2397	/* If killed, bypass charge */
2398	if (fatal_signal_pending(current)) {
2399	css_put(&memcg->css);
2400	goto bypass;
2401	}
2402
2403	oom_check = false;
2404	if (oom && !nr_oom_retries) {
2405	oom_check = true;
2406	nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
2407	}
2408
2409	ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, oom_check);
2410	switch (ret) {
2411	case CHARGE_OK:
2412	break;
2413	case CHARGE_RETRY: /* not in OOM situation but retry */
2414	batch = nr_pages;
2415	css_put(&memcg->css);
2416	memcg = NULL;
2417	goto again;
2418	case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
2419	css_put(&memcg->css);
2420	goto nomem;
2421	case CHARGE_NOMEM: /* OOM routine works */
2422	if (!oom) {
2423	css_put(&memcg->css);
2424	goto nomem;
2425	}
2426	/* If oom, we never return -ENOMEM */
2427	nr_oom_retries--;
2428	break;
2429	case CHARGE_OOM_DIE: /* Killed by OOM Killer */
2430	css_put(&memcg->css);
2431	goto bypass;
2432	}
2433	} while (ret != CHARGE_OK);
2434
2435	if (batch > nr_pages)
2436	refill_stock(memcg, batch - nr_pages);
2437	css_put(&memcg->css);
2438	done:
2439	*ptr = memcg;
2440	return 0;
2441	nomem:
2442	*ptr = NULL;
2443	return -ENOMEM;
2444	bypass:
2445	*ptr = root_mem_cgroup;
2446	return -EINTR;
2447	}
2448
2449	/*
2450	* Somemtimes we have to undo a charge we got by try_charge().
2451	* This function is for that and do uncharge, put css's refcnt.
2452	* gotten by try_charge().
2453	*/
2454	static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg,
2455	unsigned int nr_pages)
2456	{
2457	if (!mem_cgroup_is_root(memcg)) {
2458	unsigned long bytes = nr_pages * PAGE_SIZE;
2459
2460	res_counter_uncharge(&memcg->res, bytes);
2461	if (do_swap_account)
2462	res_counter_uncharge(&memcg->memsw, bytes);
2463	}
2464	}
2465
2466	/*
2467	* Cancel chrages in this cgroup....doesn't propagate to parent cgroup.
2468	* This is useful when moving usage to parent cgroup.
2469	*/
2470	static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg,
2471	unsigned int nr_pages)
2472	{
2473	unsigned long bytes = nr_pages * PAGE_SIZE;
2474
2475	if (mem_cgroup_is_root(memcg))
2476	return;
2477
2478	res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes);
2479	if (do_swap_account)
2480	res_counter_uncharge_until(&memcg->memsw,
2481	memcg->memsw.parent, bytes);
2482	}
2483
2484	/*
2485	* A helper function to get mem_cgroup from ID. must be called under
2486	* rcu_read_lock(). The caller must check css_is_removed() or some if
2487	* it's concern. (dropping refcnt from swap can be called against removed
2488	* memcg.)
2489	*/
2490	static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2491	{
2492	struct cgroup_subsys_state *css;
2493
2494	/* ID 0 is unused ID */
2495	if (!id)
2496	return NULL;
2497	css = css_lookup(&mem_cgroup_subsys, id);
2498	if (!css)
2499	return NULL;
2500	return mem_cgroup_from_css(css);
2501	}
2502
2503	struct mem_cgroup try_get_mem_cgroup_from_page(struct page page)
2504	{
2505	struct mem_cgroup *memcg = NULL;
2506	struct page_cgroup *pc;
2507	unsigned short id;
2508	swp_entry_t ent;
2509
2510	VM_BUG_ON(!PageLocked(page));
2511
2512	pc = lookup_page_cgroup(page);
2513	lock_page_cgroup(pc);
2514	if (PageCgroupUsed(pc)) {
2515	memcg = pc->mem_cgroup;
2516	if (memcg && !css_tryget(&memcg->css))
2517	memcg = NULL;
2518	} else if (PageSwapCache(page)) {
2519	ent.val = page_private(page);
2520	id = lookup_swap_cgroup_id(ent);
2521	rcu_read_lock();
2522	memcg = mem_cgroup_lookup(id);
2523	if (memcg && !css_tryget(&memcg->css))
2524	memcg = NULL;
2525	rcu_read_unlock();
2526	}
2527	unlock_page_cgroup(pc);
2528	return memcg;
2529	}
2530
2531	static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
2532	struct page *page,
2533	unsigned int nr_pages,
2534	enum charge_type ctype,
2535	bool lrucare)
2536	{
2537	struct page_cgroup *pc = lookup_page_cgroup(page);
2538	struct zone *uninitialized_var(zone);
2539	struct lruvec *lruvec;
2540	bool was_on_lru = false;
2541	bool anon;
2542
2543	lock_page_cgroup(pc);
2544	VM_BUG_ON(PageCgroupUsed(pc));
2545	/*
2546	* we don't need page_cgroup_lock about tail pages, becase they are not
2547	* accessed by any other context at this point.
2548	*/
2549
2550	/*
2551	* In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2552	* may already be on some other mem_cgroup's LRU. Take care of it.
2553	*/
2554	if (lrucare) {
2555	zone = page_zone(page);
2556	spin_lock_irq(&zone->lru_lock);
2557	if (PageLRU(page)) {
2558	lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
2559	ClearPageLRU(page);
2560	del_page_from_lru_list(page, lruvec, page_lru(page));
2561	was_on_lru = true;
2562	}
2563	}
2564
2565	pc->mem_cgroup = memcg;
2566	/*
2567	* We access a page_cgroup asynchronously without lock_page_cgroup().
2568	* Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2569	* is accessed after testing USED bit. To make pc->mem_cgroup visible
2570	* before USED bit, we need memory barrier here.
2571	* See mem_cgroup_add_lru_list(), etc.
2572	*/
2573	smp_wmb();
2574	SetPageCgroupUsed(pc);
2575
2576	if (lrucare) {
2577	if (was_on_lru) {
2578	lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
2579	VM_BUG_ON(PageLRU(page));
2580	SetPageLRU(page);
2581	add_page_to_lru_list(page, lruvec, page_lru(page));
2582	}
2583	spin_unlock_irq(&zone->lru_lock);
2584	}
2585
2586	if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON)
2587	anon = true;
2588	else
2589	anon = false;
2590
2591	mem_cgroup_charge_statistics(memcg, anon, nr_pages);
2592	unlock_page_cgroup(pc);
2593
2594	/*
2595	* "charge_statistics" updated event counter. Then, check it.
2596	* Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2597	* if they exceeds softlimit.
2598	*/
2599	memcg_check_events(memcg, page);
2600	}
2601
2602	#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2603
2604	#define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK \| 1 << PCG_MIGRATION)
2605	/*
2606	* Because tail pages are not marked as "used", set it. We're under
2607	* zone->lru_lock, 'splitting on pmd' and compound_lock.
2608	* charge/uncharge will be never happen and move_account() is done under
2609	* compound_lock(), so we don't have to take care of races.
2610	*/
2611	void mem_cgroup_split_huge_fixup(struct page *head)
2612	{
2613	struct page_cgroup *head_pc = lookup_page_cgroup(head);
2614	struct page_cgroup *pc;
2615	int i;
2616
2617	if (mem_cgroup_disabled())
2618	return;
2619	for (i = 1; i < HPAGE_PMD_NR; i++) {
2620	pc = head_pc + i;
2621	pc->mem_cgroup = head_pc->mem_cgroup;
2622	smp_wmb();/* see __commit_charge() */
2623	pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
2624	}
2625	}
2626	#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2627
2628	/**
2629	* mem_cgroup_move_account - move account of the page
2630	* @page: the page
2631	* @nr_pages: number of regular pages (>1 for huge pages)
2632	* @pc: page_cgroup of the page.
2633	* @from: mem_cgroup which the page is moved from.
2634	* @to: mem_cgroup which the page is moved to. @from != @to.
2635	*
2636	* The caller must confirm following.
2637	* - page is not on LRU (isolate_page() is useful.)
2638	* - compound_lock is held when nr_pages > 1
2639	*
2640	* This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
2641	* from old cgroup.
2642	*/
2643	static int mem_cgroup_move_account(struct page *page,
2644	unsigned int nr_pages,
2645	struct page_cgroup *pc,
2646	struct mem_cgroup *from,
2647	struct mem_cgroup *to)
2648	{
2649	unsigned long flags;
2650	int ret;
2651	bool anon = PageAnon(page);
2652
2653	VM_BUG_ON(from == to);
2654	VM_BUG_ON(PageLRU(page));
2655	/*
2656	* The page is isolated from LRU. So, collapse function
2657	* will not handle this page. But page splitting can happen.
2658	* Do this check under compound_page_lock(). The caller should
2659	* hold it.
2660	*/
2661	ret = -EBUSY;
2662	if (nr_pages > 1 && !PageTransHuge(page))
2663	goto out;
2664
2665	lock_page_cgroup(pc);
2666
2667	ret = -EINVAL;
2668	if (!PageCgroupUsed(pc) \|\| pc->mem_cgroup != from)
2669	goto unlock;
2670
2671	move_lock_mem_cgroup(from, &flags);
2672
2673	if (!anon && page_mapped(page)) {
2674	/* Update mapped_file data for mem_cgroup */
2675	preempt_disable();
2676	__this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2677	__this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2678	preempt_enable();
2679	}
2680	mem_cgroup_charge_statistics(from, anon, -nr_pages);
2681
2682	/* caller should have done css_get */
2683	pc->mem_cgroup = to;
2684	mem_cgroup_charge_statistics(to, anon, nr_pages);
2685	/*
2686	* We charges against "to" which may not have any tasks. Then, "to"
2687	* can be under rmdir(). But in current implementation, caller of
2688	* this function is just force_empty() and move charge, so it's
2689	* guaranteed that "to" is never removed. So, we don't check rmdir
2690	* status here.
2691	*/
2692	move_unlock_mem_cgroup(from, &flags);
2693	ret = 0;
2694	unlock:
2695	unlock_page_cgroup(pc);
2696	/*
2697	* check events
2698	*/
2699	memcg_check_events(to, page);
2700	memcg_check_events(from, page);
2701	out:
2702	return ret;
2703	}
2704
2705	/*
2706	* move charges to its parent.
2707	*/
2708
2709	static int mem_cgroup_move_parent(struct page *page,
2710	struct page_cgroup *pc,
2711	struct mem_cgroup *child)
2712	{
2713	struct mem_cgroup *parent;
2714	unsigned int nr_pages;
2715	unsigned long uninitialized_var(flags);
2716	int ret;
2717
2718	/* Is ROOT ? */
2719	if (mem_cgroup_is_root(child))
2720	return -EINVAL;
2721
2722	ret = -EBUSY;
2723	if (!get_page_unless_zero(page))
2724	goto out;
2725	if (isolate_lru_page(page))
2726	goto put;
2727
2728	nr_pages = hpage_nr_pages(page);
2729
2730	parent = parent_mem_cgroup(child);
2731	/*
2732	* If no parent, move charges to root cgroup.
2733	*/
2734	if (!parent)
2735	parent = root_mem_cgroup;
2736
2737	if (nr_pages > 1)
2738	flags = compound_lock_irqsave(page);
2739
2740	ret = mem_cgroup_move_account(page, nr_pages,
2741	pc, child, parent);
2742	if (!ret)
2743	__mem_cgroup_cancel_local_charge(child, nr_pages);
2744
2745	if (nr_pages > 1)
2746	compound_unlock_irqrestore(page, flags);
2747	putback_lru_page(page);
2748	put:
2749	put_page(page);
2750	out:
2751	return ret;
2752	}
2753
2754	/*
2755	* Charge the memory controller for page usage.
2756	* Return
2757	* 0 if the charge was successful
2758	* < 0 if the cgroup is over its limit
2759	*/
2760	static int mem_cgroup_charge_common(struct page page, struct mm_struct mm,
2761	gfp_t gfp_mask, enum charge_type ctype)
2762	{
2763	struct mem_cgroup *memcg = NULL;
2764	unsigned int nr_pages = 1;
2765	bool oom = true;
2766	int ret;
2767
2768	if (PageTransHuge(page)) {
2769	nr_pages <<= compound_order(page);
2770	VM_BUG_ON(!PageTransHuge(page));
2771	/*
2772	* Never OOM-kill a process for a huge page. The
2773	* fault handler will fall back to regular pages.
2774	*/
2775	oom = false;
2776	}
2777
2778	ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom);
2779	if (ret == -ENOMEM)
2780	return ret;
2781	__mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false);
2782	return 0;
2783	}
2784
2785	int mem_cgroup_newpage_charge(struct page *page,
2786	struct mm_struct *mm, gfp_t gfp_mask)
2787	{
2788	if (mem_cgroup_disabled())
2789	return 0;
2790	VM_BUG_ON(page_mapped(page));
2791	VM_BUG_ON(page->mapping && !PageAnon(page));
2792	VM_BUG_ON(!mm);
2793	return mem_cgroup_charge_common(page, mm, gfp_mask,
2794	MEM_CGROUP_CHARGE_TYPE_ANON);
2795	}
2796
2797	/*
2798	* While swap-in, try_charge -> commit or cancel, the page is locked.
2799	* And when try_charge() successfully returns, one refcnt to memcg without
2800	* struct page_cgroup is acquired. This refcnt will be consumed by
2801	* "commit()" or removed by "cancel()"
2802	*/
2803	static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm,
2804	struct page *page,
2805	gfp_t mask,
2806	struct mem_cgroup **memcgp)
2807	{
2808	struct mem_cgroup *memcg;
2809	struct page_cgroup *pc;
2810	int ret;
2811
2812	pc = lookup_page_cgroup(page);
2813	/*
2814	* Every swap fault against a single page tries to charge the
2815	* page, bail as early as possible. shmem_unuse() encounters
2816	* already charged pages, too. The USED bit is protected by
2817	* the page lock, which serializes swap cache removal, which
2818	* in turn serializes uncharging.
2819	*/
2820	if (PageCgroupUsed(pc))
2821	return 0;
2822	if (!do_swap_account)
2823	goto charge_cur_mm;
2824	memcg = try_get_mem_cgroup_from_page(page);
2825	if (!memcg)
2826	goto charge_cur_mm;
2827	*memcgp = memcg;
2828	ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true);
2829	css_put(&memcg->css);
2830	if (ret == -EINTR)
2831	ret = 0;
2832	return ret;
2833	charge_cur_mm:
2834	ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true);
2835	if (ret == -EINTR)
2836	ret = 0;
2837	return ret;
2838	}
2839
2840	int mem_cgroup_try_charge_swapin(struct mm_struct mm, struct page page,
2841	gfp_t gfp_mask, struct mem_cgroup **memcgp)
2842	{
2843	*memcgp = NULL;
2844	if (mem_cgroup_disabled())
2845	return 0;
2846	/*
2847	* A racing thread's fault, or swapoff, may have already
2848	* updated the pte, and even removed page from swap cache: in
2849	* those cases unuse_pte()'s pte_same() test will fail; but
2850	* there's also a KSM case which does need to charge the page.
2851	*/
2852	if (!PageSwapCache(page)) {
2853	int ret;
2854
2855	ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true);
2856	if (ret == -EINTR)
2857	ret = 0;
2858	return ret;
2859	}
2860	return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp);
2861	}
2862
2863	void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
2864	{
2865	if (mem_cgroup_disabled())
2866	return;
2867	if (!memcg)
2868	return;
2869	__mem_cgroup_cancel_charge(memcg, 1);
2870	}
2871
2872	static void
2873	__mem_cgroup_commit_charge_swapin(struct page page, struct mem_cgroup memcg,
2874	enum charge_type ctype)
2875	{
2876	if (mem_cgroup_disabled())
2877	return;
2878	if (!memcg)
2879	return;
2880	cgroup_exclude_rmdir(&memcg->css);
2881
2882	__mem_cgroup_commit_charge(memcg, page, 1, ctype, true);
2883	/*
2884	* Now swap is on-memory. This means this page may be
2885	* counted both as mem and swap....double count.
2886	* Fix it by uncharging from memsw. Basically, this SwapCache is stable
2887	* under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2888	* may call delete_from_swap_cache() before reach here.
2889	*/
2890	if (do_swap_account && PageSwapCache(page)) {
2891	swp_entry_t ent = {.val = page_private(page)};
2892	mem_cgroup_uncharge_swap(ent);
2893	}
2894	/*
2895	* At swapin, we may charge account against cgroup which has no tasks.
2896	* So, rmdir()->pre_destroy() can be called while we do this charge.
2897	* In that case, we need to call pre_destroy() again. check it here.
2898	*/
2899	cgroup_release_and_wakeup_rmdir(&memcg->css);
2900	}
2901
2902	void mem_cgroup_commit_charge_swapin(struct page *page,
2903	struct mem_cgroup *memcg)
2904	{
2905	__mem_cgroup_commit_charge_swapin(page, memcg,
2906	MEM_CGROUP_CHARGE_TYPE_ANON);
2907	}
2908
2909	int mem_cgroup_cache_charge(struct page page, struct mm_struct mm,
2910	gfp_t gfp_mask)
2911	{
2912	struct mem_cgroup *memcg = NULL;
2913	enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
2914	int ret;
2915
2916	if (mem_cgroup_disabled())
2917	return 0;
2918	if (PageCompound(page))
2919	return 0;
2920
2921	if (!PageSwapCache(page))
2922	ret = mem_cgroup_charge_common(page, mm, gfp_mask, type);
2923	else { /* page is swapcache/shmem */
2924	ret = __mem_cgroup_try_charge_swapin(mm, page,
2925	gfp_mask, &memcg);
2926	if (!ret)
2927	__mem_cgroup_commit_charge_swapin(page, memcg, type);
2928	}
2929	return ret;
2930	}
2931
2932	static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg,
2933	unsigned int nr_pages,
2934	const enum charge_type ctype)
2935	{
2936	struct memcg_batch_info *batch = NULL;
2937	bool uncharge_memsw = true;
2938
2939	/* If swapout, usage of swap doesn't decrease */
2940	if (!do_swap_account \|\| ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2941	uncharge_memsw = false;
2942
2943	batch = &current->memcg_batch;
2944	/*
2945	* In usual, we do css_get() when we remember memcg pointer.
2946	* But in this case, we keep res->usage until end of a series of
2947	* uncharges. Then, it's ok to ignore memcg's refcnt.
2948	*/
2949	if (!batch->memcg)
2950	batch->memcg = memcg;
2951	/*
2952	* do_batch > 0 when unmapping pages or inode invalidate/truncate.
2953	* In those cases, all pages freed continuously can be expected to be in
2954	* the same cgroup and we have chance to coalesce uncharges.
2955	* But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2956	* because we want to do uncharge as soon as possible.
2957	*/
2958
2959	if (!batch->do_batch \|\| test_thread_flag(TIF_MEMDIE))
2960	goto direct_uncharge;
2961
2962	if (nr_pages > 1)
2963	goto direct_uncharge;
2964
2965	/*
2966	* In typical case, batch->memcg == mem. This means we can
2967	* merge a series of uncharges to an uncharge of res_counter.
2968	* If not, we uncharge res_counter ony by one.
2969	*/
2970	if (batch->memcg != memcg)
2971	goto direct_uncharge;
2972	/* remember freed charge and uncharge it later */
2973	batch->nr_pages++;
2974	if (uncharge_memsw)
2975	batch->memsw_nr_pages++;
2976	return;
2977	direct_uncharge:
2978	res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE);
2979	if (uncharge_memsw)
2980	res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE);
2981	if (unlikely(batch->memcg != memcg))
2982	memcg_oom_recover(memcg);
2983	}
2984
2985	/*
2986	* uncharge if !page_mapped(page)
2987	*/
2988	static struct mem_cgroup *
2989	__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype,
2990	bool end_migration)
2991	{
2992	struct mem_cgroup *memcg = NULL;
2993	unsigned int nr_pages = 1;
2994	struct page_cgroup *pc;
2995	bool anon;
2996
2997	if (mem_cgroup_disabled())
2998	return NULL;
2999
3000	VM_BUG_ON(PageSwapCache(page));
3001
3002	if (PageTransHuge(page)) {
3003	nr_pages <<= compound_order(page);
3004	VM_BUG_ON(!PageTransHuge(page));
3005	}
3006	/*
3007	* Check if our page_cgroup is valid
3008	*/
3009	pc = lookup_page_cgroup(page);
3010	if (unlikely(!PageCgroupUsed(pc)))
3011	return NULL;
3012
3013	lock_page_cgroup(pc);
3014
3015	memcg = pc->mem_cgroup;
3016
3017	if (!PageCgroupUsed(pc))
3018	goto unlock_out;
3019
3020	anon = PageAnon(page);
3021
3022	switch (ctype) {
3023	case MEM_CGROUP_CHARGE_TYPE_ANON:
3024	/*
3025	* Generally PageAnon tells if it's the anon statistics to be
3026	* updated; but sometimes e.g. mem_cgroup_uncharge_page() is
3027	* used before page reached the stage of being marked PageAnon.
3028	*/
3029	anon = true;
3030	/* fallthrough */
3031	case MEM_CGROUP_CHARGE_TYPE_DROP:
3032	/* See mem_cgroup_prepare_migration() */
3033	if (page_mapped(page))
3034	goto unlock_out;
3035	/*
3036	* Pages under migration may not be uncharged. But
3037	* end_migration() /must/ be the one uncharging the
3038	* unused post-migration page and so it has to call
3039	* here with the migration bit still set. See the
3040	* res_counter handling below.
3041	*/
3042	if (!end_migration && PageCgroupMigration(pc))
3043	goto unlock_out;
3044	break;
3045	case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
3046	if (!PageAnon(page)) { /* Shared memory */
3047	if (page->mapping && !page_is_file_cache(page))
3048	goto unlock_out;
3049	} else if (page_mapped(page)) /* Anon */
3050	goto unlock_out;
3051	break;
3052	default:
3053	break;
3054	}
3055
3056	mem_cgroup_charge_statistics(memcg, anon, -nr_pages);
3057
3058	ClearPageCgroupUsed(pc);
3059	/*
3060	* pc->mem_cgroup is not cleared here. It will be accessed when it's
3061	* freed from LRU. This is safe because uncharged page is expected not
3062	* to be reused (freed soon). Exception is SwapCache, it's handled by
3063	* special functions.
3064	*/
3065
3066	unlock_page_cgroup(pc);
3067	/*
3068	* even after unlock, we have memcg->res.usage here and this memcg
3069	* will never be freed.
3070	*/
3071	memcg_check_events(memcg, page);
3072	if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
3073	mem_cgroup_swap_statistics(memcg, true);
3074	mem_cgroup_get(memcg);
3075	}
3076	/*
3077	* Migration does not charge the res_counter for the
3078	* replacement page, so leave it alone when phasing out the
3079	* page that is unused after the migration.
3080	*/
3081	if (!end_migration && !mem_cgroup_is_root(memcg))
3082	mem_cgroup_do_uncharge(memcg, nr_pages, ctype);
3083
3084	return memcg;
3085
3086	unlock_out:
3087	unlock_page_cgroup(pc);
3088	return NULL;
3089	}
3090
3091	void mem_cgroup_uncharge_page(struct page *page)
3092	{
3093	/* early check. */
3094	if (page_mapped(page))
3095	return;
3096	VM_BUG_ON(page->mapping && !PageAnon(page));
3097	if (PageSwapCache(page))
3098	return;
3099	__mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false);
3100	}
3101
3102	void mem_cgroup_uncharge_cache_page(struct page *page)
3103	{
3104	VM_BUG_ON(page_mapped(page));
3105	VM_BUG_ON(page->mapping);
3106	__mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false);
3107	}
3108
3109	/*
3110	* Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3111	* In that cases, pages are freed continuously and we can expect pages
3112	* are in the same memcg. All these calls itself limits the number of
3113	* pages freed at once, then uncharge_start/end() is called properly.
3114	* This may be called prural(2) times in a context,
3115	*/
3116
3117	void mem_cgroup_uncharge_start(void)
3118	{
3119	current->memcg_batch.do_batch++;
3120	/* We can do nest. */
3121	if (current->memcg_batch.do_batch == 1) {
3122	current->memcg_batch.memcg = NULL;
3123	current->memcg_batch.nr_pages = 0;
3124	current->memcg_batch.memsw_nr_pages = 0;
3125	}
3126	}
3127
3128	void mem_cgroup_uncharge_end(void)
3129	{
3130	struct memcg_batch_info *batch = &current->memcg_batch;
3131
3132	if (!batch->do_batch)
3133	return;
3134
3135	batch->do_batch--;
3136	if (batch->do_batch) /* If stacked, do nothing. */
3137	return;
3138
3139	if (!batch->memcg)
3140	return;
3141	/*
3142	* This "batch->memcg" is valid without any css_get/put etc...
3143	* bacause we hide charges behind us.
3144	*/
3145	if (batch->nr_pages)
3146	res_counter_uncharge(&batch->memcg->res,
3147	batch->nr_pages * PAGE_SIZE);
3148	if (batch->memsw_nr_pages)
3149	res_counter_uncharge(&batch->memcg->memsw,
3150	batch->memsw_nr_pages * PAGE_SIZE);
3151	memcg_oom_recover(batch->memcg);
3152	/* forget this pointer (for sanity check) */
3153	batch->memcg = NULL;
3154	}
3155
3156	#ifdef CONFIG_SWAP
3157	/*
3158	* called after __delete_from_swap_cache() and drop "page" account.
3159	* memcg information is recorded to swap_cgroup of "ent"
3160	*/
3161	void
3162	mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
3163	{
3164	struct mem_cgroup *memcg;
3165	int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
3166
3167	if (!swapout) /* this was a swap cache but the swap is unused ! */
3168	ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
3169
3170	memcg = __mem_cgroup_uncharge_common(page, ctype, false);
3171
3172	/*
3173	* record memcg information, if swapout && memcg != NULL,
3174	* mem_cgroup_get() was called in uncharge().
3175	*/
3176	if (do_swap_account && swapout && memcg)
3177	swap_cgroup_record(ent, css_id(&memcg->css));
3178	}
3179	#endif
3180
3181	#ifdef CONFIG_MEMCG_SWAP
3182	/*
3183	* called from swap_entry_free(). remove record in swap_cgroup and
3184	* uncharge "memsw" account.
3185	*/
3186	void mem_cgroup_uncharge_swap(swp_entry_t ent)
3187	{
3188	struct mem_cgroup *memcg;
3189	unsigned short id;
3190
3191	if (!do_swap_account)
3192	return;
3193
3194	id = swap_cgroup_record(ent, 0);
3195	rcu_read_lock();
3196	memcg = mem_cgroup_lookup(id);
3197	if (memcg) {
3198	/*
3199	* We uncharge this because swap is freed.
3200	* This memcg can be obsolete one. We avoid calling css_tryget
3201	*/
3202	if (!mem_cgroup_is_root(memcg))
3203	res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
3204	mem_cgroup_swap_statistics(memcg, false);
3205	mem_cgroup_put(memcg);
3206	}
3207	rcu_read_unlock();
3208	}
3209
3210	/**
3211	* mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3212	* @entry: swap entry to be moved
3213	* @from: mem_cgroup which the entry is moved from
3214	* @to: mem_cgroup which the entry is moved to
3215	*
3216	* It succeeds only when the swap_cgroup's record for this entry is the same
3217	* as the mem_cgroup's id of @from.
3218	*
3219	* Returns 0 on success, -EINVAL on failure.
3220	*
3221	* The caller must have charged to @to, IOW, called res_counter_charge() about
3222	* both res and memsw, and called css_get().
3223	*/
3224	static int mem_cgroup_move_swap_account(swp_entry_t entry,
3225	struct mem_cgroup from, struct mem_cgroup to)
3226	{
3227	unsigned short old_id, new_id;
3228
3229	old_id = css_id(&from->css);
3230	new_id = css_id(&to->css);
3231
3232	if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
3233	mem_cgroup_swap_statistics(from, false);
3234	mem_cgroup_swap_statistics(to, true);
3235	/*
3236	* This function is only called from task migration context now.
3237	* It postpones res_counter and refcount handling till the end
3238	* of task migration(mem_cgroup_clear_mc()) for performance
3239	* improvement. But we cannot postpone mem_cgroup_get(to)
3240	* because if the process that has been moved to @to does
3241	* swap-in, the refcount of @to might be decreased to 0.
3242	*/
3243	mem_cgroup_get(to);
3244	return 0;
3245	}
3246	return -EINVAL;
3247	}
3248	#else
3249	static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
3250	struct mem_cgroup from, struct mem_cgroup to)
3251	{
3252	return -EINVAL;
3253	}
3254	#endif
3255
3256	/*
3257	* Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3258	* page belongs to.
3259	*/
3260	void mem_cgroup_prepare_migration(struct page page, struct page newpage,
3261	struct mem_cgroup **memcgp)
3262	{
3263	struct mem_cgroup *memcg = NULL;
3264	struct page_cgroup *pc;
3265	enum charge_type ctype;
3266
3267	*memcgp = NULL;
3268
3269	VM_BUG_ON(PageTransHuge(page));
3270	if (mem_cgroup_disabled())
3271	return;
3272
3273	pc = lookup_page_cgroup(page);
3274	lock_page_cgroup(pc);
3275	if (PageCgroupUsed(pc)) {
3276	memcg = pc->mem_cgroup;
3277	css_get(&memcg->css);
3278	/*
3279	* At migrating an anonymous page, its mapcount goes down
3280	* to 0 and uncharge() will be called. But, even if it's fully
3281	* unmapped, migration may fail and this page has to be
3282	* charged again. We set MIGRATION flag here and delay uncharge
3283	* until end_migration() is called
3284	*
3285	* Corner Case Thinking
3286	* A)
3287	* When the old page was mapped as Anon and it's unmap-and-freed
3288	* while migration was ongoing.
3289	* If unmap finds the old page, uncharge() of it will be delayed
3290	* until end_migration(). If unmap finds a new page, it's
3291	* uncharged when it make mapcount to be 1->0. If unmap code
3292	* finds swap_migration_entry, the new page will not be mapped
3293	* and end_migration() will find it(mapcount==0).
3294	*
3295	* B)
3296	* When the old page was mapped but migraion fails, the kernel
3297	* remaps it. A charge for it is kept by MIGRATION flag even
3298	* if mapcount goes down to 0. We can do remap successfully
3299	* without charging it again.
3300	*
3301	* C)
3302	* The "old" page is under lock_page() until the end of
3303	* migration, so, the old page itself will not be swapped-out.
3304	* If the new page is swapped out before end_migraton, our
3305	* hook to usual swap-out path will catch the event.
3306	*/
3307	if (PageAnon(page))
3308	SetPageCgroupMigration(pc);
3309	}
3310	unlock_page_cgroup(pc);
3311	/*
3312	* If the page is not charged at this point,
3313	* we return here.
3314	*/
3315	if (!memcg)
3316	return;
3317
3318	*memcgp = memcg;
3319	/*
3320	* We charge new page before it's used/mapped. So, even if unlock_page()
3321	* is called before end_migration, we can catch all events on this new
3322	* page. In the case new page is migrated but not remapped, new page's
3323	* mapcount will be finally 0 and we call uncharge in end_migration().
3324	*/
3325	if (PageAnon(page))
3326	ctype = MEM_CGROUP_CHARGE_TYPE_ANON;
3327	else
3328	ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
3329	/*
3330	* The page is committed to the memcg, but it's not actually
3331	* charged to the res_counter since we plan on replacing the
3332	* old one and only one page is going to be left afterwards.
3333	*/
3334	__mem_cgroup_commit_charge(memcg, newpage, 1, ctype, false);
3335	}
3336
3337	/* remove redundant charge if migration failed*/
3338	void mem_cgroup_end_migration(struct mem_cgroup *memcg,
3339	struct page oldpage, struct page newpage, bool migration_ok)
3340	{
3341	struct page used, unused;
3342	struct page_cgroup *pc;
3343	bool anon;
3344
3345	if (!memcg)
3346	return;
3347	/* blocks rmdir() */
3348	cgroup_exclude_rmdir(&memcg->css);
3349	if (!migration_ok) {
3350	used = oldpage;
3351	unused = newpage;
3352	} else {
3353	used = newpage;
3354	unused = oldpage;
3355	}
3356	anon = PageAnon(used);
3357	__mem_cgroup_uncharge_common(unused,
3358	anon ? MEM_CGROUP_CHARGE_TYPE_ANON
3359	: MEM_CGROUP_CHARGE_TYPE_CACHE,
3360	true);
3361	css_put(&memcg->css);
3362	/*
3363	* We disallowed uncharge of pages under migration because mapcount
3364	* of the page goes down to zero, temporarly.
3365	* Clear the flag and check the page should be charged.
3366	*/
3367	pc = lookup_page_cgroup(oldpage);
3368	lock_page_cgroup(pc);
3369	ClearPageCgroupMigration(pc);
3370	unlock_page_cgroup(pc);
3371
3372	/*
3373	* If a page is a file cache, radix-tree replacement is very atomic
3374	* and we can skip this check. When it was an Anon page, its mapcount
3375	* goes down to 0. But because we added MIGRATION flage, it's not
3376	* uncharged yet. There are several case but page->mapcount check
3377	* and USED bit check in mem_cgroup_uncharge_page() will do enough
3378	* check. (see prepare_charge() also)
3379	*/
3380	if (anon)
3381	mem_cgroup_uncharge_page(used);
3382	/*
3383	* At migration, we may charge account against cgroup which has no
3384	* tasks.
3385	* So, rmdir()->pre_destroy() can be called while we do this charge.
3386	* In that case, we need to call pre_destroy() again. check it here.
3387	*/
3388	cgroup_release_and_wakeup_rmdir(&memcg->css);
3389	}
3390
3391	/*
3392	* At replace page cache, newpage is not under any memcg but it's on
3393	* LRU. So, this function doesn't touch res_counter but handles LRU
3394	* in correct way. Both pages are locked so we cannot race with uncharge.
3395	*/
3396	void mem_cgroup_replace_page_cache(struct page *oldpage,
3397	struct page *newpage)
3398	{
3399	struct mem_cgroup *memcg = NULL;
3400	struct page_cgroup *pc;
3401	enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
3402
3403	if (mem_cgroup_disabled())
3404	return;
3405
3406	pc = lookup_page_cgroup(oldpage);
3407	/* fix accounting on old pages */
3408	lock_page_cgroup(pc);
3409	if (PageCgroupUsed(pc)) {
3410	memcg = pc->mem_cgroup;
3411	mem_cgroup_charge_statistics(memcg, false, -1);
3412	ClearPageCgroupUsed(pc);
3413	}
3414	unlock_page_cgroup(pc);
3415
3416	/*
3417	* When called from shmem_replace_page(), in some cases the
3418	* oldpage has already been charged, and in some cases not.
3419	*/
3420	if (!memcg)
3421	return;
3422	/*
3423	* Even if newpage->mapping was NULL before starting replacement,
3424	* the newpage may be on LRU(or pagevec for LRU) already. We lock
3425	* LRU while we overwrite pc->mem_cgroup.
3426	*/
3427	__mem_cgroup_commit_charge(memcg, newpage, 1, type, true);
3428	}
3429
3430	#ifdef CONFIG_DEBUG_VM
3431	static struct page_cgroup lookup_page_cgroup_used(struct page page)
3432	{
3433	struct page_cgroup *pc;
3434
3435	pc = lookup_page_cgroup(page);
3436	/*
3437	* Can be NULL while feeding pages into the page allocator for
3438	* the first time, i.e. during boot or memory hotplug;
3439	* or when mem_cgroup_disabled().
3440	*/
3441	if (likely(pc) && PageCgroupUsed(pc))
3442	return pc;
3443	return NULL;
3444	}
3445
3446	bool mem_cgroup_bad_page_check(struct page *page)
3447	{
3448	if (mem_cgroup_disabled())
3449	return false;
3450
3451	return lookup_page_cgroup_used(page) != NULL;
3452	}
3453
3454	void mem_cgroup_print_bad_page(struct page *page)
3455	{
3456	struct page_cgroup *pc;
3457
3458	pc = lookup_page_cgroup_used(page);
3459	if (pc) {
3460	printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
3461	pc, pc->flags, pc->mem_cgroup);
3462	}
3463	}
3464	#endif
3465
3466	static DEFINE_MUTEX(set_limit_mutex);
3467
3468	static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
3469	unsigned long long val)
3470	{
3471	int retry_count;
3472	u64 memswlimit, memlimit;
3473	int ret = 0;
3474	int children = mem_cgroup_count_children(memcg);
3475	u64 curusage, oldusage;
3476	int enlarge;
3477
3478	/*
3479	* For keeping hierarchical_reclaim simple, how long we should retry
3480	* is depends on callers. We set our retry-count to be function
3481	* of # of children which we should visit in this loop.
3482	*/
3483	retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
3484
3485	oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
3486
3487	enlarge = 0;
3488	while (retry_count) {
3489	if (signal_pending(current)) {
3490	ret = -EINTR;
3491	break;
3492	}
3493	/*
3494	* Rather than hide all in some function, I do this in
3495	* open coded manner. You see what this really does.
3496	* We have to guarantee memcg->res.limit <= memcg->memsw.limit.
3497	*/
3498	mutex_lock(&set_limit_mutex);
3499	memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3500	if (memswlimit < val) {
3501	ret = -EINVAL;
3502	mutex_unlock(&set_limit_mutex);
3503	break;
3504	}
3505
3506	memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3507	if (memlimit < val)
3508	enlarge = 1;
3509
3510	ret = res_counter_set_limit(&memcg->res, val);
3511	if (!ret) {
3512	if (memswlimit == val)
3513	memcg->memsw_is_minimum = true;
3514	else
3515	memcg->memsw_is_minimum = false;
3516	}
3517	mutex_unlock(&set_limit_mutex);
3518
3519	if (!ret)
3520	break;
3521
3522	mem_cgroup_reclaim(memcg, GFP_KERNEL,
3523	MEM_CGROUP_RECLAIM_SHRINK);
3524	curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
3525	/* Usage is reduced ? */
3526	if (curusage >= oldusage)
3527	retry_count--;
3528	else
3529	oldusage = curusage;
3530	}
3531	if (!ret && enlarge)
3532	memcg_oom_recover(memcg);
3533
3534	return ret;
3535	}
3536
3537	static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
3538	unsigned long long val)
3539	{
3540	int retry_count;
3541	u64 memlimit, memswlimit, oldusage, curusage;
3542	int children = mem_cgroup_count_children(memcg);
3543	int ret = -EBUSY;
3544	int enlarge = 0;
3545
3546	/* see mem_cgroup_resize_res_limit */
3547	retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
3548	oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
3549	while (retry_count) {
3550	if (signal_pending(current)) {
3551	ret = -EINTR;
3552	break;
3553	}
3554	/*
3555	* Rather than hide all in some function, I do this in
3556	* open coded manner. You see what this really does.
3557	* We have to guarantee memcg->res.limit <= memcg->memsw.limit.
3558	*/
3559	mutex_lock(&set_limit_mutex);
3560	memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3561	if (memlimit > val) {
3562	ret = -EINVAL;
3563	mutex_unlock(&set_limit_mutex);
3564	break;
3565	}
3566	memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3567	if (memswlimit < val)
3568	enlarge = 1;
3569	ret = res_counter_set_limit(&memcg->memsw, val);
3570	if (!ret) {
3571	if (memlimit == val)
3572	memcg->memsw_is_minimum = true;
3573	else
3574	memcg->memsw_is_minimum = false;
3575	}
3576	mutex_unlock(&set_limit_mutex);
3577
3578	if (!ret)
3579	break;
3580
3581	mem_cgroup_reclaim(memcg, GFP_KERNEL,
3582	MEM_CGROUP_RECLAIM_NOSWAP \|
3583	MEM_CGROUP_RECLAIM_SHRINK);
3584	curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
3585	/* Usage is reduced ? */
3586	if (curusage >= oldusage)
3587	retry_count--;
3588	else
3589	oldusage = curusage;
3590	}
3591	if (!ret && enlarge)
3592	memcg_oom_recover(memcg);
3593	return ret;
3594	}
3595
3596	unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
3597	gfp_t gfp_mask,
3598	unsigned long *total_scanned)
3599	{
3600	unsigned long nr_reclaimed = 0;
3601	struct mem_cgroup_per_zone mz, next_mz = NULL;
3602	unsigned long reclaimed;
3603	int loop = 0;
3604	struct mem_cgroup_tree_per_zone *mctz;
3605	unsigned long long excess;
3606	unsigned long nr_scanned;
3607
3608	if (order > 0)
3609	return 0;
3610
3611	mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
3612	/*
3613	* This loop can run a while, specially if mem_cgroup's continuously
3614	* keep exceeding their soft limit and putting the system under
3615	* pressure
3616	*/
3617	do {
3618	if (next_mz)
3619	mz = next_mz;
3620	else
3621	mz = mem_cgroup_largest_soft_limit_node(mctz);
3622	if (!mz)
3623	break;
3624
3625	nr_scanned = 0;
3626	reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
3627	gfp_mask, &nr_scanned);
3628	nr_reclaimed += reclaimed;
3629	*total_scanned += nr_scanned;
3630	spin_lock(&mctz->lock);
3631
3632	/*
3633	* If we failed to reclaim anything from this memory cgroup
3634	* it is time to move on to the next cgroup
3635	*/
3636	next_mz = NULL;
3637	if (!reclaimed) {
3638	do {
3639	/*
3640	* Loop until we find yet another one.
3641	*
3642	* By the time we get the soft_limit lock
3643	* again, someone might have aded the
3644	* group back on the RB tree. Iterate to
3645	* make sure we get a different mem.
3646	* mem_cgroup_largest_soft_limit_node returns
3647	* NULL if no other cgroup is present on
3648	* the tree
3649	*/
3650	next_mz =
3651	__mem_cgroup_largest_soft_limit_node(mctz);
3652	if (next_mz == mz)
3653	css_put(&next_mz->memcg->css);
3654	else /* next_mz == NULL or other memcg */
3655	break;
3656	} while (1);
3657	}
3658	__mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
3659	excess = res_counter_soft_limit_excess(&mz->memcg->res);
3660	/*
3661	* One school of thought says that we should not add
3662	* back the node to the tree if reclaim returns 0.
3663	* But our reclaim could return 0, simply because due
3664	* to priority we are exposing a smaller subset of
3665	* memory to reclaim from. Consider this as a longer
3666	* term TODO.
3667	*/
3668	/* If excess == 0, no tree ops */
3669	__mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess);
3670	spin_unlock(&mctz->lock);
3671	css_put(&mz->memcg->css);
3672	loop++;
3673	/*
3674	* Could not reclaim anything and there are no more
3675	* mem cgroups to try or we seem to be looping without
3676	* reclaiming anything.
3677	*/
3678	if (!nr_reclaimed &&
3679	(next_mz == NULL \|\|
3680	loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
3681	break;
3682	} while (!nr_reclaimed);
3683	if (next_mz)
3684	css_put(&next_mz->memcg->css);
3685	return nr_reclaimed;
3686	}
3687
3688	/*
3689	* Traverse a specified page_cgroup list and try to drop them all. This doesn't
3690	* reclaim the pages page themselves - it just removes the page_cgroups.
3691	* Returns true if some page_cgroups were not freed, indicating that the caller
3692	* must retry this operation.
3693	*/
3694	static bool mem_cgroup_force_empty_list(struct mem_cgroup *memcg,
3695	int node, int zid, enum lru_list lru)
3696	{
3697	struct mem_cgroup_per_zone *mz;
3698	unsigned long flags, loop;
3699	struct list_head *list;
3700	struct page *busy;
3701	struct zone *zone;
3702
3703	zone = &NODE_DATA(node)->node_zones[zid];
3704	mz = mem_cgroup_zoneinfo(memcg, node, zid);
3705	list = &mz->lruvec.lists[lru];
3706
3707	loop = mz->lru_size[lru];
3708	/* give some margin against EBUSY etc...*/
3709	loop += 256;
3710	busy = NULL;
3711	while (loop--) {
3712	struct page_cgroup *pc;
3713	struct page *page;
3714
3715	spin_lock_irqsave(&zone->lru_lock, flags);
3716	if (list_empty(list)) {
3717	spin_unlock_irqrestore(&zone->lru_lock, flags);
3718	break;
3719	}
3720	page = list_entry(list->prev, struct page, lru);
3721	if (busy == page) {
3722	list_move(&page->lru, list);
3723	busy = NULL;
3724	spin_unlock_irqrestore(&zone->lru_lock, flags);
3725	continue;
3726	}
3727	spin_unlock_irqrestore(&zone->lru_lock, flags);
3728
3729	pc = lookup_page_cgroup(page);
3730
3731	if (mem_cgroup_move_parent(page, pc, memcg)) {
3732	/* found lock contention or "pc" is obsolete. */
3733	busy = page;
3734	cond_resched();
3735	} else
3736	busy = NULL;
3737	}
3738	return !list_empty(list);
3739	}
3740
3741	/*
3742	* make mem_cgroup's charge to be 0 if there is no task.
3743	* This enables deleting this mem_cgroup.
3744	*/
3745	static int mem_cgroup_force_empty(struct mem_cgroup *memcg, bool free_all)
3746	{
3747	int ret;
3748	int node, zid, shrink;
3749	int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
3750	struct cgroup *cgrp = memcg->css.cgroup;
3751
3752	css_get(&memcg->css);
3753
3754	shrink = 0;
3755	/* should free all ? */
3756	if (free_all)
3757	goto try_to_free;
3758	move_account:
3759	do {
3760	ret = -EBUSY;
3761	if (cgroup_task_count(cgrp) \|\| !list_empty(&cgrp->children))
3762	goto out;
3763	/* This is for making all used pages to be on LRU. */
3764	lru_add_drain_all();
3765	drain_all_stock_sync(memcg);
3766	ret = 0;
3767	mem_cgroup_start_move(memcg);
3768	for_each_node_state(node, N_HIGH_MEMORY) {
3769	for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
3770	enum lru_list lru;
3771	for_each_lru(lru) {
3772	ret = mem_cgroup_force_empty_list(memcg,
3773	node, zid, lru);
3774	if (ret)
3775	break;
3776	}
3777	}
3778	if (ret)
3779	break;
3780	}
3781	mem_cgroup_end_move(memcg);
3782	memcg_oom_recover(memcg);
3783	cond_resched();
3784	/* "ret" should also be checked to ensure all lists are empty. */
3785	} while (res_counter_read_u64(&memcg->res, RES_USAGE) > 0 \|\| ret);
3786	out:
3787	css_put(&memcg->css);
3788	return ret;
3789
3790	try_to_free:
3791	/* returns EBUSY if there is a task or if we come here twice. */
3792	if (cgroup_task_count(cgrp) \|\| !list_empty(&cgrp->children) \|\| shrink) {
3793	ret = -EBUSY;
3794	goto out;
3795	}
3796	/* we call try-to-free pages for make this cgroup empty */
3797	lru_add_drain_all();
3798	/* try to free all pages in this cgroup */
3799	shrink = 1;
3800	while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) {
3801	int progress;
3802
3803	if (signal_pending(current)) {
3804	ret = -EINTR;
3805	goto out;
3806	}
3807	progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL,
3808	false);
3809	if (!progress) {
3810	nr_retries--;
3811	/* maybe some writeback is necessary */
3812	congestion_wait(BLK_RW_ASYNC, HZ/10);
3813	}
3814
3815	}
3816	lru_add_drain();
3817	/* try move_account...there may be some locked pages. */
3818	goto move_account;
3819	}
3820
3821	static int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
3822	{
3823	return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
3824	}
3825
3826
3827	static u64 mem_cgroup_hierarchy_read(struct cgroup cont, struct cftype cft)
3828	{
3829	return mem_cgroup_from_cont(cont)->use_hierarchy;
3830	}
3831
3832	static int mem_cgroup_hierarchy_write(struct cgroup cont, struct cftype cft,
3833	u64 val)
3834	{
3835	int retval = 0;
3836	struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
3837	struct cgroup *parent = cont->parent;
3838	struct mem_cgroup *parent_memcg = NULL;
3839
3840	if (parent)
3841	parent_memcg = mem_cgroup_from_cont(parent);
3842
3843	cgroup_lock();
3844
3845	if (memcg->use_hierarchy == val)
3846	goto out;
3847
3848	/*
3849	* If parent's use_hierarchy is set, we can't make any modifications
3850	* in the child subtrees. If it is unset, then the change can
3851	* occur, provided the current cgroup has no children.
3852	*
3853	* For the root cgroup, parent_mem is NULL, we allow value to be
3854	* set if there are no children.
3855	*/
3856	if ((!parent_memcg \|\| !parent_memcg->use_hierarchy) &&
3857	(val == 1 \|\| val == 0)) {
3858	if (list_empty(&cont->children))
3859	memcg->use_hierarchy = val;
3860	else
3861	retval = -EBUSY;
3862	} else
3863	retval = -EINVAL;
3864
3865	out:
3866	cgroup_unlock();
3867
3868	return retval;
3869	}
3870
3871
3872	static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg,
3873	enum mem_cgroup_stat_index idx)
3874	{
3875	struct mem_cgroup *iter;
3876	long val = 0;
3877
3878	/* Per-cpu values can be negative, use a signed accumulator */
3879	for_each_mem_cgroup_tree(iter, memcg)
3880	val += mem_cgroup_read_stat(iter, idx);
3881
3882	if (val < 0) /* race ? */
3883	val = 0;
3884	return val;
3885	}
3886
3887	static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
3888	{
3889	u64 val;
3890
3891	if (!mem_cgroup_is_root(memcg)) {
3892	if (!swap)
3893	return res_counter_read_u64(&memcg->res, RES_USAGE);
3894	else
3895	return res_counter_read_u64(&memcg->memsw, RES_USAGE);
3896	}
3897
3898	val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE);
3899	val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS);
3900
3901	if (swap)
3902	val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP);
3903
3904	return val << PAGE_SHIFT;
3905	}
3906
3907	static ssize_t mem_cgroup_read(struct cgroup cont, struct cftype cft,
3908	struct file file, char __user buf,
3909	size_t nbytes, loff_t *ppos)
3910	{
3911	struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
3912	char str[64];
3913	u64 val;
3914	int type, name, len;
3915
3916	type = MEMFILE_TYPE(cft->private);
3917	name = MEMFILE_ATTR(cft->private);
3918
3919	if (!do_swap_account && type == _MEMSWAP)
3920	return -EOPNOTSUPP;
3921
3922	switch (type) {
3923	case _MEM:
3924	if (name == RES_USAGE)
3925	val = mem_cgroup_usage(memcg, false);
3926	else
3927	val = res_counter_read_u64(&memcg->res, name);
3928	break;
3929	case _MEMSWAP:
3930	if (name == RES_USAGE)
3931	val = mem_cgroup_usage(memcg, true);
3932	else
3933	val = res_counter_read_u64(&memcg->memsw, name);
3934	break;
3935	default:
3936	BUG();
3937	}
3938
3939	len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val);
3940	return simple_read_from_buffer(buf, nbytes, ppos, str, len);
3941	}
3942	/*
3943	* The user of this function is...
3944	* RES_LIMIT.
3945	*/
3946	static int mem_cgroup_write(struct cgroup cont, struct cftype cft,
3947	const char *buffer)
3948	{
3949	struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
3950	int type, name;
3951	unsigned long long val;
3952	int ret;
3953
3954	type = MEMFILE_TYPE(cft->private);
3955	name = MEMFILE_ATTR(cft->private);
3956
3957	if (!do_swap_account && type == _MEMSWAP)
3958	return -EOPNOTSUPP;
3959
3960	switch (name) {
3961	case RES_LIMIT:
3962	if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3963	ret = -EINVAL;
3964	break;
3965	}
3966	/* This function does all necessary parse...reuse it */
3967	ret = res_counter_memparse_write_strategy(buffer, &val);
3968	if (ret)
3969	break;
3970	if (type == _MEM)
3971	ret = mem_cgroup_resize_limit(memcg, val);
3972	else
3973	ret = mem_cgroup_resize_memsw_limit(memcg, val);
3974	break;
3975	case RES_SOFT_LIMIT:
3976	ret = res_counter_memparse_write_strategy(buffer, &val);
3977	if (ret)
3978	break;
3979	/*
3980	* For memsw, soft limits are hard to implement in terms
3981	* of semantics, for now, we support soft limits for
3982	* control without swap
3983	*/
3984	if (type == _MEM)
3985	ret = res_counter_set_soft_limit(&memcg->res, val);
3986	else
3987	ret = -EINVAL;
3988	break;
3989	default:
3990	ret = -EINVAL; /* should be BUG() ? */
3991	break;
3992	}
3993	return ret;
3994	}
3995
3996	static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
3997	unsigned long long mem_limit, unsigned long long memsw_limit)
3998	{
3999	struct cgroup *cgroup;
4000	unsigned long long min_limit, min_memsw_limit, tmp;
4001
4002	min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
4003	min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4004	cgroup = memcg->css.cgroup;
4005	if (!memcg->use_hierarchy)
4006	goto out;
4007
4008	while (cgroup->parent) {
4009	cgroup = cgroup->parent;
4010	memcg = mem_cgroup_from_cont(cgroup);
4011	if (!memcg->use_hierarchy)
4012	break;
4013	tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
4014	min_limit = min(min_limit, tmp);
4015	tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4016	min_memsw_limit = min(min_memsw_limit, tmp);
4017	}
4018	out:
4019	*mem_limit = min_limit;
4020	*memsw_limit = min_memsw_limit;
4021	}
4022
4023	static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
4024	{
4025	struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
4026	int type, name;
4027
4028	type = MEMFILE_TYPE(event);
4029	name = MEMFILE_ATTR(event);
4030
4031	if (!do_swap_account && type == _MEMSWAP)
4032	return -EOPNOTSUPP;
4033
4034	switch (name) {
4035	case RES_MAX_USAGE:
4036	if (type == _MEM)
4037	res_counter_reset_max(&memcg->res);
4038	else
4039	res_counter_reset_max(&memcg->memsw);
4040	break;
4041	case RES_FAILCNT:
4042	if (type == _MEM)
4043	res_counter_reset_failcnt(&memcg->res);
4044	else
4045	res_counter_reset_failcnt(&memcg->memsw);
4046	break;
4047	}
4048
4049	return 0;
4050	}
4051
4052	static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp,
4053	struct cftype *cft)
4054	{
4055	return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate;
4056	}
4057
4058	#ifdef CONFIG_MMU
4059	static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
4060	struct cftype *cft, u64 val)
4061	{
4062	struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4063
4064	if (val >= (1 << NR_MOVE_TYPE))
4065	return -EINVAL;
4066	/*
4067	* We check this value several times in both in can_attach() and
4068	* attach(), so we need cgroup lock to prevent this value from being
4069	* inconsistent.
4070	*/
4071	cgroup_lock();
4072	memcg->move_charge_at_immigrate = val;
4073	cgroup_unlock();
4074
4075	return 0;
4076	}
4077	#else
4078	static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
4079	struct cftype *cft, u64 val)
4080	{
4081	return -ENOSYS;
4082	}
4083	#endif
4084
4085	#ifdef CONFIG_NUMA
4086	static int memcg_numa_stat_show(struct cgroup cont, struct cftype cft,
4087	struct seq_file *m)
4088	{
4089	int nid;
4090	unsigned long total_nr, file_nr, anon_nr, unevictable_nr;
4091	unsigned long node_nr;
4092	struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
4093
4094	total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL);
4095	seq_printf(m, "total=%lu", total_nr);
4096	for_each_node_state(nid, N_HIGH_MEMORY) {
4097	node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL);
4098	seq_printf(m, " N%d=%lu", nid, node_nr);
4099	}
4100	seq_putc(m, '\n');
4101
4102	file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE);
4103	seq_printf(m, "file=%lu", file_nr);
4104	for_each_node_state(nid, N_HIGH_MEMORY) {
4105	node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
4106	LRU_ALL_FILE);
4107	seq_printf(m, " N%d=%lu", nid, node_nr);
4108	}
4109	seq_putc(m, '\n');
4110
4111	anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON);
4112	seq_printf(m, "anon=%lu", anon_nr);
4113	for_each_node_state(nid, N_HIGH_MEMORY) {
4114	node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
4115	LRU_ALL_ANON);
4116	seq_printf(m, " N%d=%lu", nid, node_nr);
4117	}
4118	seq_putc(m, '\n');
4119
4120	unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE));
4121	seq_printf(m, "unevictable=%lu", unevictable_nr);
4122	for_each_node_state(nid, N_HIGH_MEMORY) {
4123	node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
4124	BIT(LRU_UNEVICTABLE));
4125	seq_printf(m, " N%d=%lu", nid, node_nr);
4126	}
4127	seq_putc(m, '\n');
4128	return 0;
4129	}
4130	#endif /* CONFIG_NUMA */
4131
4132	static const char * const mem_cgroup_lru_names[] = {
4133	"inactive_anon",
4134	"active_anon",
4135	"inactive_file",
4136	"active_file",
4137	"unevictable",
4138	};
4139
4140	static inline void mem_cgroup_lru_names_not_uptodate(void)
4141	{
4142	BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
4143	}
4144
4145	static int memcg_stat_show(struct cgroup cont, struct cftype cft,
4146	struct seq_file *m)
4147	{
4148	struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
4149	struct mem_cgroup *mi;
4150	unsigned int i;
4151
4152	for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
4153	if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
4154	continue;
4155	seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i],
4156	mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
4157	}
4158
4159	for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
4160	seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
4161	mem_cgroup_read_events(memcg, i));
4162
4163	for (i = 0; i < NR_LRU_LISTS; i++)
4164	seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
4165	mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
4166
4167	/* Hierarchical information */
4168	{
4169	unsigned long long limit, memsw_limit;
4170	memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit);
4171	seq_printf(m, "hierarchical_memory_limit %llu\n", limit);
4172	if (do_swap_account)
4173	seq_printf(m, "hierarchical_memsw_limit %llu\n",
4174	memsw_limit);
4175	}
4176
4177	for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
4178	long long val = 0;
4179
4180	if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
4181	continue;
4182	for_each_mem_cgroup_tree(mi, memcg)
4183	val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
4184	seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val);
4185	}
4186
4187	for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
4188	unsigned long long val = 0;
4189
4190	for_each_mem_cgroup_tree(mi, memcg)
4191	val += mem_cgroup_read_events(mi, i);
4192	seq_printf(m, "total_%s %llu\n",
4193	mem_cgroup_events_names[i], val);
4194	}
4195
4196	for (i = 0; i < NR_LRU_LISTS; i++) {
4197	unsigned long long val = 0;
4198
4199	for_each_mem_cgroup_tree(mi, memcg)
4200	val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
4201	seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
4202	}
4203
4204	#ifdef CONFIG_DEBUG_VM
4205	{
4206	int nid, zid;
4207	struct mem_cgroup_per_zone *mz;
4208	struct zone_reclaim_stat *rstat;
4209	unsigned long recent_rotated[2] = {0, 0};
4210	unsigned long recent_scanned[2] = {0, 0};
4211
4212	for_each_online_node(nid)
4213	for (zid = 0; zid < MAX_NR_ZONES; zid++) {
4214	mz = mem_cgroup_zoneinfo(memcg, nid, zid);
4215	rstat = &mz->lruvec.reclaim_stat;
4216
4217	recent_rotated[0] += rstat->recent_rotated[0];
4218	recent_rotated[1] += rstat->recent_rotated[1];
4219	recent_scanned[0] += rstat->recent_scanned[0];
4220	recent_scanned[1] += rstat->recent_scanned[1];
4221	}
4222	seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
4223	seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
4224	seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
4225	seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
4226	}
4227	#endif
4228
4229	return 0;
4230	}
4231
4232	static u64 mem_cgroup_swappiness_read(struct cgroup cgrp, struct cftype cft)
4233	{
4234	struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4235
4236	return mem_cgroup_swappiness(memcg);
4237	}
4238
4239	static int mem_cgroup_swappiness_write(struct cgroup cgrp, struct cftype cft,
4240	u64 val)
4241	{
4242	struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4243	struct mem_cgroup *parent;
4244
4245	if (val > 100)
4246	return -EINVAL;
4247
4248	if (cgrp->parent == NULL)
4249	return -EINVAL;
4250
4251	parent = mem_cgroup_from_cont(cgrp->parent);
4252
4253	cgroup_lock();
4254
4255	/* If under hierarchy, only empty-root can set this value */
4256	if ((parent->use_hierarchy) \|\|
4257	(memcg->use_hierarchy && !list_empty(&cgrp->children))) {
4258	cgroup_unlock();
4259	return -EINVAL;
4260	}
4261
4262	memcg->swappiness = val;
4263
4264	cgroup_unlock();
4265
4266	return 0;
4267	}
4268
4269	static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
4270	{
4271	struct mem_cgroup_threshold_ary *t;
4272	u64 usage;
4273	int i;
4274
4275	rcu_read_lock();
4276	if (!swap)
4277	t = rcu_dereference(memcg->thresholds.primary);
4278	else
4279	t = rcu_dereference(memcg->memsw_thresholds.primary);
4280
4281	if (!t)
4282	goto unlock;
4283
4284	usage = mem_cgroup_usage(memcg, swap);
4285
4286	/*
4287	* current_threshold points to threshold just below or equal to usage.
4288	* If it's not true, a threshold was crossed after last
4289	* call of __mem_cgroup_threshold().
4290	*/
4291	i = t->current_threshold;
4292
4293	/*
4294	* Iterate backward over array of thresholds starting from
4295	* current_threshold and check if a threshold is crossed.
4296	* If none of thresholds below usage is crossed, we read
4297	* only one element of the array here.
4298	*/
4299	for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
4300	eventfd_signal(t->entries[i].eventfd, 1);
4301
4302	/* i = current_threshold + 1 */
4303	i++;
4304
4305	/*
4306	* Iterate forward over array of thresholds starting from
4307	* current_threshold+1 and check if a threshold is crossed.
4308	* If none of thresholds above usage is crossed, we read
4309	* only one element of the array here.
4310	*/
4311	for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
4312	eventfd_signal(t->entries[i].eventfd, 1);
4313
4314	/* Update current_threshold */
4315	t->current_threshold = i - 1;
4316	unlock:
4317	rcu_read_unlock();
4318	}
4319
4320	static void mem_cgroup_threshold(struct mem_cgroup *memcg)
4321	{
4322	while (memcg) {
4323	__mem_cgroup_threshold(memcg, false);
4324	if (do_swap_account)
4325	__mem_cgroup_threshold(memcg, true);
4326
4327	memcg = parent_mem_cgroup(memcg);
4328	}
4329	}
4330
4331	static int compare_thresholds(const void a, const void b)
4332	{
4333	const struct mem_cgroup_threshold *_a = a;
4334	const struct mem_cgroup_threshold *_b = b;
4335
4336	return _a->threshold - _b->threshold;
4337	}
4338
4339	static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
4340	{
4341	struct mem_cgroup_eventfd_list *ev;
4342
4343	list_for_each_entry(ev, &memcg->oom_notify, list)
4344	eventfd_signal(ev->eventfd, 1);
4345	return 0;
4346	}
4347
4348	static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
4349	{
4350	struct mem_cgroup *iter;
4351
4352	for_each_mem_cgroup_tree(iter, memcg)
4353	mem_cgroup_oom_notify_cb(iter);
4354	}
4355
4356	static int mem_cgroup_usage_register_event(struct cgroup *cgrp,
4357	struct cftype cft, struct eventfd_ctx eventfd, const char *args)
4358	{
4359	struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4360	struct mem_cgroup_thresholds *thresholds;
4361	struct mem_cgroup_threshold_ary *new;
4362	int type = MEMFILE_TYPE(cft->private);
4363	u64 threshold, usage;
4364	int i, size, ret;
4365
4366	ret = res_counter_memparse_write_strategy(args, &threshold);
4367	if (ret)
4368	return ret;
4369
4370	mutex_lock(&memcg->thresholds_lock);
4371
4372	if (type == _MEM)
4373	thresholds = &memcg->thresholds;
4374	else if (type == _MEMSWAP)
4375	thresholds = &memcg->memsw_thresholds;
4376	else
4377	BUG();
4378
4379	usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4380
4381	/* Check if a threshold crossed before adding a new one */
4382	if (thresholds->primary)
4383	__mem_cgroup_threshold(memcg, type == _MEMSWAP);
4384
4385	size = thresholds->primary ? thresholds->primary->size + 1 : 1;
4386
4387	/* Allocate memory for new array of thresholds */
4388	new = kmalloc(sizeof(new) + size sizeof(struct mem_cgroup_threshold),
4389	GFP_KERNEL);
4390	if (!new) {
4391	ret = -ENOMEM;
4392	goto unlock;
4393	}
4394	new->size = size;
4395
4396	/* Copy thresholds (if any) to new array */
4397	if (thresholds->primary) {
4398	memcpy(new->entries, thresholds->primary->entries, (size - 1) *
4399	sizeof(struct mem_cgroup_threshold));
4400	}
4401
4402	/* Add new threshold */
4403	new->entries[size - 1].eventfd = eventfd;
4404	new->entries[size - 1].threshold = threshold;
4405
4406	/* Sort thresholds. Registering of new threshold isn't time-critical */
4407	sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
4408	compare_thresholds, NULL);
4409
4410	/* Find current threshold */
4411	new->current_threshold = -1;
4412	for (i = 0; i < size; i++) {
4413	if (new->entries[i].threshold <= usage) {
4414	/*
4415	* new->current_threshold will not be used until
4416	* rcu_assign_pointer(), so it's safe to increment
4417	* it here.
4418	*/
4419	++new->current_threshold;
4420	} else
4421	break;
4422	}
4423
4424	/* Free old spare buffer and save old primary buffer as spare */
4425	kfree(thresholds->spare);
4426	thresholds->spare = thresholds->primary;
4427
4428	rcu_assign_pointer(thresholds->primary, new);
4429
4430	/* To be sure that nobody uses thresholds */
4431	synchronize_rcu();
4432
4433	unlock:
4434	mutex_unlock(&memcg->thresholds_lock);
4435
4436	return ret;
4437	}
4438
4439	static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp,
4440	struct cftype cft, struct eventfd_ctx eventfd)
4441	{
4442	struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4443	struct mem_cgroup_thresholds *thresholds;
4444	struct mem_cgroup_threshold_ary *new;
4445	int type = MEMFILE_TYPE(cft->private);
4446	u64 usage;
4447	int i, j, size;
4448
4449	mutex_lock(&memcg->thresholds_lock);
4450	if (type == _MEM)
4451	thresholds = &memcg->thresholds;
4452	else if (type == _MEMSWAP)
4453	thresholds = &memcg->memsw_thresholds;
4454	else
4455	BUG();
4456
4457	if (!thresholds->primary)
4458	goto unlock;
4459
4460	usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4461
4462	/* Check if a threshold crossed before removing */
4463	__mem_cgroup_threshold(memcg, type == _MEMSWAP);
4464
4465	/* Calculate new number of threshold */
4466	size = 0;
4467	for (i = 0; i < thresholds->primary->size; i++) {
4468	if (thresholds->primary->entries[i].eventfd != eventfd)
4469	size++;
4470	}
4471
4472	new = thresholds->spare;
4473
4474	/* Set thresholds array to NULL if we don't have thresholds */
4475	if (!size) {
4476	kfree(new);
4477	new = NULL;
4478	goto swap_buffers;
4479	}
4480
4481	new->size = size;
4482
4483	/* Copy thresholds and find current threshold */
4484	new->current_threshold = -1;
4485	for (i = 0, j = 0; i < thresholds->primary->size; i++) {
4486	if (thresholds->primary->entries[i].eventfd == eventfd)
4487	continue;
4488
4489	new->entries[j] = thresholds->primary->entries[i];
4490	if (new->entries[j].threshold <= usage) {
4491	/*
4492	* new->current_threshold will not be used
4493	* until rcu_assign_pointer(), so it's safe to increment
4494	* it here.
4495	*/
4496	++new->current_threshold;
4497	}
4498	j++;
4499	}
4500
4501	swap_buffers:
4502	/* Swap primary and spare array */
4503	thresholds->spare = thresholds->primary;
4504	/* If all events are unregistered, free the spare array */
4505	if (!new) {
4506	kfree(thresholds->spare);
4507	thresholds->spare = NULL;
4508	}
4509
4510	rcu_assign_pointer(thresholds->primary, new);
4511
4512	/* To be sure that nobody uses thresholds */
4513	synchronize_rcu();
4514	unlock:
4515	mutex_unlock(&memcg->thresholds_lock);
4516	}
4517
4518	static int mem_cgroup_oom_register_event(struct cgroup *cgrp,
4519	struct cftype cft, struct eventfd_ctx eventfd, const char *args)
4520	{
4521	struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4522	struct mem_cgroup_eventfd_list *event;
4523	int type = MEMFILE_TYPE(cft->private);
4524
4525	BUG_ON(type != _OOM_TYPE);
4526	event = kmalloc(sizeof(*event), GFP_KERNEL);
4527	if (!event)
4528	return -ENOMEM;
4529
4530	spin_lock(&memcg_oom_lock);
4531
4532	event->eventfd = eventfd;
4533	list_add(&event->list, &memcg->oom_notify);
4534
4535	/* already in OOM ? */
4536	if (atomic_read(&memcg->under_oom))
4537	eventfd_signal(eventfd, 1);
4538	spin_unlock(&memcg_oom_lock);
4539
4540	return 0;
4541	}
4542
4543	static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp,
4544	struct cftype cft, struct eventfd_ctx eventfd)
4545	{
4546	struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4547	struct mem_cgroup_eventfd_list ev, tmp;
4548	int type = MEMFILE_TYPE(cft->private);
4549
4550	BUG_ON(type != _OOM_TYPE);
4551
4552	spin_lock(&memcg_oom_lock);
4553
4554	list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
4555	if (ev->eventfd == eventfd) {
4556	list_del(&ev->list);
4557	kfree(ev);
4558	}
4559	}
4560
4561	spin_unlock(&memcg_oom_lock);
4562	}
4563
4564	static int mem_cgroup_oom_control_read(struct cgroup *cgrp,
4565	struct cftype cft, struct cgroup_map_cb cb)
4566	{
4567	struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4568
4569	cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable);
4570
4571	if (atomic_read(&memcg->under_oom))
4572	cb->fill(cb, "under_oom", 1);
4573	else
4574	cb->fill(cb, "under_oom", 0);
4575	return 0;
4576	}
4577
4578	static int mem_cgroup_oom_control_write(struct cgroup *cgrp,
4579	struct cftype *cft, u64 val)
4580	{
4581	struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4582	struct mem_cgroup *parent;
4583
4584	/* cannot set to root cgroup and only 0 and 1 are allowed */
4585	if (!cgrp->parent \|\| !((val == 0) \|\| (val == 1)))
4586	return -EINVAL;
4587
4588	parent = mem_cgroup_from_cont(cgrp->parent);
4589
4590	cgroup_lock();
4591	/* oom-kill-disable is a flag for subhierarchy. */
4592	if ((parent->use_hierarchy) \|\|
4593	(memcg->use_hierarchy && !list_empty(&cgrp->children))) {
4594	cgroup_unlock();
4595	return -EINVAL;
4596	}
4597	memcg->oom_kill_disable = val;
4598	if (!val)
4599	memcg_oom_recover(memcg);
4600	cgroup_unlock();
4601	return 0;
4602	}
4603
4604	#ifdef CONFIG_MEMCG_KMEM
4605	static int memcg_init_kmem(struct mem_cgroup memcg, struct cgroup_subsys ss)
4606	{
4607	return mem_cgroup_sockets_init(memcg, ss);
4608	};
4609
4610	static void kmem_cgroup_destroy(struct mem_cgroup *memcg)
4611	{
4612	mem_cgroup_sockets_destroy(memcg);
4613	}
4614	#else
4615	static int memcg_init_kmem(struct mem_cgroup memcg, struct cgroup_subsys ss)
4616	{
4617	return 0;
4618	}
4619
4620	static void kmem_cgroup_destroy(struct mem_cgroup *memcg)
4621	{
4622	}
4623	#endif
4624
4625	static struct cftype mem_cgroup_files[] = {
4626	{
4627	.name = "usage_in_bytes",
4628	.private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
4629	.read = mem_cgroup_read,
4630	.register_event = mem_cgroup_usage_register_event,
4631	.unregister_event = mem_cgroup_usage_unregister_event,
4632	},
4633	{
4634	.name = "max_usage_in_bytes",
4635	.private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
4636	.trigger = mem_cgroup_reset,
4637	.read = mem_cgroup_read,
4638	},
4639	{
4640	.name = "limit_in_bytes",
4641	.private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
4642	.write_string = mem_cgroup_write,
4643	.read = mem_cgroup_read,
4644	},
4645	{
4646	.name = "soft_limit_in_bytes",
4647	.private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
4648	.write_string = mem_cgroup_write,
4649	.read = mem_cgroup_read,
4650	},
4651	{
4652	.name = "failcnt",
4653	.private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
4654	.trigger = mem_cgroup_reset,
4655	.read = mem_cgroup_read,
4656	},
4657	{
4658	.name = "stat",
4659	.read_seq_string = memcg_stat_show,
4660	},
4661	{
4662	.name = "force_empty",
4663	.trigger = mem_cgroup_force_empty_write,
4664	},
4665	{
4666	.name = "use_hierarchy",
4667	.write_u64 = mem_cgroup_hierarchy_write,
4668	.read_u64 = mem_cgroup_hierarchy_read,
4669	},
4670	{
4671	.name = "swappiness",
4672	.read_u64 = mem_cgroup_swappiness_read,
4673	.write_u64 = mem_cgroup_swappiness_write,
4674	},
4675	{
4676	.name = "move_charge_at_immigrate",
4677	.read_u64 = mem_cgroup_move_charge_read,
4678	.write_u64 = mem_cgroup_move_charge_write,
4679	},
4680	{
4681	.name = "oom_control",
4682	.read_map = mem_cgroup_oom_control_read,
4683	.write_u64 = mem_cgroup_oom_control_write,
4684	.register_event = mem_cgroup_oom_register_event,
4685	.unregister_event = mem_cgroup_oom_unregister_event,
4686	.private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4687	},
4688	#ifdef CONFIG_NUMA
4689	{
4690	.name = "numa_stat",
4691	.read_seq_string = memcg_numa_stat_show,
4692	},
4693	#endif
4694	#ifdef CONFIG_MEMCG_SWAP
4695	{
4696	.name = "memsw.usage_in_bytes",
4697	.private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
4698	.read = mem_cgroup_read,
4699	.register_event = mem_cgroup_usage_register_event,
4700	.unregister_event = mem_cgroup_usage_unregister_event,
4701	},
4702	{
4703	.name = "memsw.max_usage_in_bytes",
4704	.private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
4705	.trigger = mem_cgroup_reset,
4706	.read = mem_cgroup_read,
4707	},
4708	{
4709	.name = "memsw.limit_in_bytes",
4710	.private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
4711	.write_string = mem_cgroup_write,
4712	.read = mem_cgroup_read,
4713	},
4714	{
4715	.name = "memsw.failcnt",
4716	.private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
4717	.trigger = mem_cgroup_reset,
4718	.read = mem_cgroup_read,
4719	},
4720	#endif
4721	{ }, /* terminate */
4722	};
4723
4724	static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
4725	{
4726	struct mem_cgroup_per_node *pn;
4727	struct mem_cgroup_per_zone *mz;
4728	int zone, tmp = node;
4729	/*
4730	* This routine is called against possible nodes.
4731	* But it's BUG to call kmalloc() against offline node.
4732	*
4733	* TODO: this routine can waste much memory for nodes which will
4734	* never be onlined. It's better to use memory hotplug callback
4735	* function.
4736	*/
4737	if (!node_state(node, N_NORMAL_MEMORY))
4738	tmp = -1;
4739	pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
4740	if (!pn)
4741	return 1;
4742
4743	for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4744	mz = &pn->zoneinfo[zone];
4745	lruvec_init(&mz->lruvec, &NODE_DATA(node)->node_zones[zone]);
4746	mz->usage_in_excess = 0;
4747	mz->on_tree = false;
4748	mz->memcg = memcg;
4749	}
4750	memcg->info.nodeinfo[node] = pn;
4751	return 0;
4752	}
4753
4754	static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
4755	{
4756	kfree(memcg->info.nodeinfo[node]);
4757	}
4758
4759	static struct mem_cgroup *mem_cgroup_alloc(void)
4760	{
4761	struct mem_cgroup *memcg;
4762	int size = sizeof(struct mem_cgroup);
4763
4764	/* Can be very big if MAX_NUMNODES is very big */
4765	if (size < PAGE_SIZE)
4766	memcg = kzalloc(size, GFP_KERNEL);
4767	else
4768	memcg = vzalloc(size);
4769
4770	if (!memcg)
4771	return NULL;
4772
4773	memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4774	if (!memcg->stat)
4775	goto out_free;
4776	spin_lock_init(&memcg->pcp_counter_lock);
4777	return memcg;
4778
4779	out_free:
4780	if (size < PAGE_SIZE)
4781	kfree(memcg);
4782	else
4783	vfree(memcg);
4784	return NULL;
4785	}
4786
4787	/*
4788	* Helpers for freeing a kmalloc()ed/vzalloc()ed mem_cgroup by RCU,
4789	* but in process context. The work_freeing structure is overlaid
4790	* on the rcu_freeing structure, which itself is overlaid on memsw.
4791	*/
4792	static void free_work(struct work_struct *work)
4793	{
4794	struct mem_cgroup *memcg;
4795	int size = sizeof(struct mem_cgroup);
4796
4797	memcg = container_of(work, struct mem_cgroup, work_freeing);
4798	/*
4799	* We need to make sure that (at least for now), the jump label
4800	* destruction code runs outside of the cgroup lock. This is because
4801	* get_online_cpus(), which is called from the static_branch update,
4802	* can't be called inside the cgroup_lock. cpusets are the ones
4803	* enforcing this dependency, so if they ever change, we might as well.
4804	*
4805	* schedule_work() will guarantee this happens. Be careful if you need
4806	* to move this code around, and make sure it is outside
4807	* the cgroup_lock.
4808	*/
4809	disarm_sock_keys(memcg);
4810	if (size < PAGE_SIZE)
4811	kfree(memcg);
4812	else
4813	vfree(memcg);
4814	}
4815
4816	static void free_rcu(struct rcu_head *rcu_head)
4817	{
4818	struct mem_cgroup *memcg;
4819
4820	memcg = container_of(rcu_head, struct mem_cgroup, rcu_freeing);
4821	INIT_WORK(&memcg->work_freeing, free_work);
4822	schedule_work(&memcg->work_freeing);
4823	}
4824
4825	/*
4826	* At destroying mem_cgroup, references from swap_cgroup can remain.
4827	* (scanning all at force_empty is too costly...)
4828	*
4829	* Instead of clearing all references at force_empty, we remember
4830	* the number of reference from swap_cgroup and free mem_cgroup when
4831	* it goes down to 0.
4832	*
4833	* Removal of cgroup itself succeeds regardless of refs from swap.
4834	*/
4835
4836	static void __mem_cgroup_free(struct mem_cgroup *memcg)
4837	{
4838	int node;
4839
4840	mem_cgroup_remove_from_trees(memcg);
4841	free_css_id(&mem_cgroup_subsys, &memcg->css);
4842
4843	for_each_node(node)
4844	free_mem_cgroup_per_zone_info(memcg, node);
4845
4846	free_percpu(memcg->stat);
4847	call_rcu(&memcg->rcu_freeing, free_rcu);
4848	}
4849
4850	static void mem_cgroup_get(struct mem_cgroup *memcg)
4851	{
4852	atomic_inc(&memcg->refcnt);
4853	}
4854
4855	static void __mem_cgroup_put(struct mem_cgroup *memcg, int count)
4856	{
4857	if (atomic_sub_and_test(count, &memcg->refcnt)) {
4858	struct mem_cgroup *parent = parent_mem_cgroup(memcg);
4859	__mem_cgroup_free(memcg);
4860	if (parent)
4861	mem_cgroup_put(parent);
4862	}
4863	}
4864
4865	static void mem_cgroup_put(struct mem_cgroup *memcg)
4866	{
4867	__mem_cgroup_put(memcg, 1);
4868	}
4869
4870	/*
4871	* Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4872	*/
4873	struct mem_cgroup parent_mem_cgroup(struct mem_cgroup memcg)
4874	{
4875	if (!memcg->res.parent)
4876	return NULL;
4877	return mem_cgroup_from_res_counter(memcg->res.parent, res);
4878	}
4879	EXPORT_SYMBOL(parent_mem_cgroup);
4880
4881	#ifdef CONFIG_MEMCG_SWAP
4882	static void __init enable_swap_cgroup(void)
4883	{
4884	if (!mem_cgroup_disabled() && really_do_swap_account)
4885	do_swap_account = 1;
4886	}
4887	#else
4888	static void __init enable_swap_cgroup(void)
4889	{
4890	}
4891	#endif
4892
4893	static int mem_cgroup_soft_limit_tree_init(void)
4894	{
4895	struct mem_cgroup_tree_per_node *rtpn;
4896	struct mem_cgroup_tree_per_zone *rtpz;
4897	int tmp, node, zone;
4898
4899	for_each_node(node) {
4900	tmp = node;
4901	if (!node_state(node, N_NORMAL_MEMORY))
4902	tmp = -1;
4903	rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
4904	if (!rtpn)
4905	goto err_cleanup;
4906
4907	soft_limit_tree.rb_tree_per_node[node] = rtpn;
4908
4909	for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4910	rtpz = &rtpn->rb_tree_per_zone[zone];
4911	rtpz->rb_root = RB_ROOT;
4912	spin_lock_init(&rtpz->lock);
4913	}
4914	}
4915	return 0;
4916
4917	err_cleanup:
4918	for_each_node(node) {
4919	if (!soft_limit_tree.rb_tree_per_node[node])
4920	break;
4921	kfree(soft_limit_tree.rb_tree_per_node[node]);
4922	soft_limit_tree.rb_tree_per_node[node] = NULL;
4923	}
4924	return 1;
4925
4926	}
4927
4928	static struct cgroup_subsys_state * __ref
4929	mem_cgroup_create(struct cgroup *cont)
4930	{
4931	struct mem_cgroup memcg, parent;
4932	long error = -ENOMEM;
4933	int node;
4934
4935	memcg = mem_cgroup_alloc();
4936	if (!memcg)
4937	return ERR_PTR(error);
4938
4939	for_each_node(node)
4940	if (alloc_mem_cgroup_per_zone_info(memcg, node))
4941	goto free_out;
4942
4943	/* root ? */
4944	if (cont->parent == NULL) {
4945	int cpu;
4946	enable_swap_cgroup();
4947	parent = NULL;
4948	if (mem_cgroup_soft_limit_tree_init())
4949	goto free_out;
4950	root_mem_cgroup = memcg;
4951	for_each_possible_cpu(cpu) {
4952	struct memcg_stock_pcp *stock =
4953	&per_cpu(memcg_stock, cpu);
4954	INIT_WORK(&stock->work, drain_local_stock);
4955	}
4956	hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
4957	} else {
4958	parent = mem_cgroup_from_cont(cont->parent);
4959	memcg->use_hierarchy = parent->use_hierarchy;
4960	memcg->oom_kill_disable = parent->oom_kill_disable;
4961	}
4962
4963	if (parent && parent->use_hierarchy) {
4964	res_counter_init(&memcg->res, &parent->res);
4965	res_counter_init(&memcg->memsw, &parent->memsw);
4966	/*
4967	* We increment refcnt of the parent to ensure that we can
4968	* safely access it on res_counter_charge/uncharge.
4969	* This refcnt will be decremented when freeing this
4970	* mem_cgroup(see mem_cgroup_put).
4971	*/
4972	mem_cgroup_get(parent);
4973	} else {
4974	res_counter_init(&memcg->res, NULL);
4975	res_counter_init(&memcg->memsw, NULL);
4976	}
4977	memcg->last_scanned_node = MAX_NUMNODES;
4978	INIT_LIST_HEAD(&memcg->oom_notify);
4979
4980	if (parent)
4981	memcg->swappiness = mem_cgroup_swappiness(parent);
4982	atomic_set(&memcg->refcnt, 1);
4983	memcg->move_charge_at_immigrate = 0;
4984	mutex_init(&memcg->thresholds_lock);
4985	spin_lock_init(&memcg->move_lock);
4986
4987	error = memcg_init_kmem(memcg, &mem_cgroup_subsys);
4988	if (error) {
4989	/*
4990	* We call put now because our (and parent's) refcnts
4991	* are already in place. mem_cgroup_put() will internally
4992	* call __mem_cgroup_free, so return directly
4993	*/
4994	mem_cgroup_put(memcg);
4995	return ERR_PTR(error);
4996	}
4997	return &memcg->css;
4998	free_out:
4999	__mem_cgroup_free(memcg);
5000	return ERR_PTR(error);
5001	}
5002
5003	static int mem_cgroup_pre_destroy(struct cgroup *cont)
5004	{
5005	struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
5006
5007	return mem_cgroup_force_empty(memcg, false);
5008	}
5009
5010	static void mem_cgroup_destroy(struct cgroup *cont)
5011	{
5012	struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
5013
5014	kmem_cgroup_destroy(memcg);
5015
5016	mem_cgroup_put(memcg);
5017	}
5018
5019	#ifdef CONFIG_MMU
5020	/* Handlers for move charge at task migration. */
5021	#define PRECHARGE_COUNT_AT_ONCE 256
5022	static int mem_cgroup_do_precharge(unsigned long count)
5023	{
5024	int ret = 0;
5025	int batch_count = PRECHARGE_COUNT_AT_ONCE;
5026	struct mem_cgroup *memcg = mc.to;
5027
5028	if (mem_cgroup_is_root(memcg)) {
5029	mc.precharge += count;
5030	/* we don't need css_get for root */
5031	return ret;
5032	}
5033	/* try to charge at once */
5034	if (count > 1) {
5035	struct res_counter *dummy;
5036	/*
5037	* "memcg" cannot be under rmdir() because we've already checked
5038	* by cgroup_lock_live_cgroup() that it is not removed and we
5039	* are still under the same cgroup_mutex. So we can postpone
5040	* css_get().
5041	*/
5042	if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy))
5043	goto one_by_one;
5044	if (do_swap_account && res_counter_charge(&memcg->memsw,
5045	PAGE_SIZE * count, &dummy)) {
5046	res_counter_uncharge(&memcg->res, PAGE_SIZE * count);
5047	goto one_by_one;
5048	}
5049	mc.precharge += count;
5050	return ret;
5051	}
5052	one_by_one:
5053	/* fall back to one by one charge */
5054	while (count--) {
5055	if (signal_pending(current)) {
5056	ret = -EINTR;
5057	break;
5058	}
5059	if (!batch_count--) {
5060	batch_count = PRECHARGE_COUNT_AT_ONCE;
5061	cond_resched();
5062	}
5063	ret = __mem_cgroup_try_charge(NULL,
5064	GFP_KERNEL, 1, &memcg, false);
5065	if (ret)
5066	/* mem_cgroup_clear_mc() will do uncharge later */
5067	return ret;
5068	mc.precharge++;
5069	}
5070	return ret;
5071	}
5072
5073	/**
5074	* get_mctgt_type - get target type of moving charge
5075	* @vma: the vma the pte to be checked belongs
5076	* @addr: the address corresponding to the pte to be checked
5077	* @ptent: the pte to be checked
5078	* @target: the pointer the target page or swap ent will be stored(can be NULL)
5079	*
5080	* Returns
5081	* 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5082	* 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5083	* move charge. if @target is not NULL, the page is stored in target->page
5084	* with extra refcnt got(Callers should handle it).
5085	* 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5086	* target for charge migration. if @target is not NULL, the entry is stored
5087	* in target->ent.
5088	*
5089	* Called with pte lock held.
5090	*/
5091	union mc_target {
5092	struct page *page;
5093	swp_entry_t ent;
5094	};
5095
5096	enum mc_target_type {
5097	MC_TARGET_NONE = 0,
5098	MC_TARGET_PAGE,
5099	MC_TARGET_SWAP,
5100	};
5101
5102	static struct page mc_handle_present_pte(struct vm_area_struct vma,
5103	unsigned long addr, pte_t ptent)
5104	{
5105	struct page *page = vm_normal_page(vma, addr, ptent);
5106
5107	if (!page \|\| !page_mapped(page))
5108	return NULL;
5109	if (PageAnon(page)) {
5110	/* we don't move shared anon */
5111	if (!move_anon())
5112	return NULL;
5113	} else if (!move_file())
5114	/* we ignore mapcount for file pages */
5115	return NULL;
5116	if (!get_page_unless_zero(page))
5117	return NULL;
5118
5119	return page;
5120	}
5121
5122	#ifdef CONFIG_SWAP
5123	static struct page mc_handle_swap_pte(struct vm_area_struct vma,
5124	unsigned long addr, pte_t ptent, swp_entry_t *entry)
5125	{
5126	struct page *page = NULL;
5127	swp_entry_t ent = pte_to_swp_entry(ptent);
5128
5129	if (!move_anon() \|\| non_swap_entry(ent))
5130	return NULL;
5131	/*
5132	* Because lookup_swap_cache() updates some statistics counter,
5133	* we call find_get_page() with swapper_space directly.
5134	*/
5135	page = find_get_page(&swapper_space, ent.val);
5136	if (do_swap_account)
5137	entry->val = ent.val;
5138
5139	return page;
5140	}
5141	#else
5142	static struct page mc_handle_swap_pte(struct vm_area_struct vma,
5143	unsigned long addr, pte_t ptent, swp_entry_t *entry)
5144	{
5145	return NULL;
5146	}
5147	#endif
5148
5149	static struct page mc_handle_file_pte(struct vm_area_struct vma,
5150	unsigned long addr, pte_t ptent, swp_entry_t *entry)
5151	{
5152	struct page *page = NULL;
5153	struct address_space *mapping;
5154	pgoff_t pgoff;
5155
5156	if (!vma->vm_file) /* anonymous vma */
5157	return NULL;
5158	if (!move_file())
5159	return NULL;
5160
5161	mapping = vma->vm_file->f_mapping;
5162	if (pte_none(ptent))
5163	pgoff = linear_page_index(vma, addr);
5164	else /* pte_file(ptent) is true */
5165	pgoff = pte_to_pgoff(ptent);
5166
5167	/* page is moved even if it's not RSS of this task(page-faulted). */
5168	page = find_get_page(mapping, pgoff);
5169
5170	#ifdef CONFIG_SWAP
5171	/* shmem/tmpfs may report page out on swap: account for that too. */
5172	if (radix_tree_exceptional_entry(page)) {
5173	swp_entry_t swap = radix_to_swp_entry(page);
5174	if (do_swap_account)
5175	*entry = swap;
5176	page = find_get_page(&swapper_space, swap.val);
5177	}
5178	#endif
5179	return page;
5180	}
5181
5182	static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
5183	unsigned long addr, pte_t ptent, union mc_target *target)
5184	{
5185	struct page *page = NULL;
5186	struct page_cgroup *pc;
5187	enum mc_target_type ret = MC_TARGET_NONE;
5188	swp_entry_t ent = { .val = 0 };
5189
5190	if (pte_present(ptent))
5191	page = mc_handle_present_pte(vma, addr, ptent);
5192	else if (is_swap_pte(ptent))
5193	page = mc_handle_swap_pte(vma, addr, ptent, &ent);
5194	else if (pte_none(ptent) \|\| pte_file(ptent))
5195	page = mc_handle_file_pte(vma, addr, ptent, &ent);
5196
5197	if (!page && !ent.val)
5198	return ret;
5199	if (page) {
5200	pc = lookup_page_cgroup(page);
5201	/*
5202	* Do only loose check w/o page_cgroup lock.
5203	* mem_cgroup_move_account() checks the pc is valid or not under
5204	* the lock.
5205	*/
5206	if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
5207	ret = MC_TARGET_PAGE;
5208	if (target)
5209	target->page = page;
5210	}
5211	if (!ret \|\| !target)
5212	put_page(page);
5213	}
5214	/* There is a swap entry and a page doesn't exist or isn't charged */
5215	if (ent.val && !ret &&
5216	css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) {
5217	ret = MC_TARGET_SWAP;
5218	if (target)
5219	target->ent = ent;
5220	}
5221	return ret;
5222	}
5223
5224	#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5225	/*
5226	* We don't consider swapping or file mapped pages because THP does not
5227	* support them for now.
5228	* Caller should make sure that pmd_trans_huge(pmd) is true.
5229	*/
5230	static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5231	unsigned long addr, pmd_t pmd, union mc_target *target)
5232	{
5233	struct page *page = NULL;
5234	struct page_cgroup *pc;
5235	enum mc_target_type ret = MC_TARGET_NONE;
5236
5237	page = pmd_page(pmd);
5238	VM_BUG_ON(!page \|\| !PageHead(page));
5239	if (!move_anon())
5240	return ret;
5241	pc = lookup_page_cgroup(page);
5242	if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
5243	ret = MC_TARGET_PAGE;
5244	if (target) {
5245	get_page(page);
5246	target->page = page;
5247	}
5248	}
5249	return ret;
5250	}
5251	#else
5252	static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5253	unsigned long addr, pmd_t pmd, union mc_target *target)
5254	{
5255	return MC_TARGET_NONE;
5256	}
5257	#endif
5258
5259	static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
5260	unsigned long addr, unsigned long end,
5261	struct mm_walk *walk)
5262	{
5263	struct vm_area_struct *vma = walk->private;
5264	pte_t *pte;
5265	spinlock_t *ptl;
5266
5267	if (pmd_trans_huge_lock(pmd, vma) == 1) {
5268	if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
5269	mc.precharge += HPAGE_PMD_NR;
5270	spin_unlock(&vma->vm_mm->page_table_lock);
5271	return 0;
5272	}
5273
5274	if (pmd_trans_unstable(pmd))
5275	return 0;
5276	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5277	for (; addr != end; pte++, addr += PAGE_SIZE)
5278	if (get_mctgt_type(vma, addr, *pte, NULL))
5279	mc.precharge++; /* increment precharge temporarily */
5280	pte_unmap_unlock(pte - 1, ptl);
5281	cond_resched();
5282
5283	return 0;
5284	}
5285
5286	static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
5287	{
5288	unsigned long precharge;
5289	struct vm_area_struct *vma;
5290
5291	down_read(&mm->mmap_sem);
5292	for (vma = mm->mmap; vma; vma = vma->vm_next) {
5293	struct mm_walk mem_cgroup_count_precharge_walk = {
5294	.pmd_entry = mem_cgroup_count_precharge_pte_range,
5295	.mm = mm,
5296	.private = vma,
5297	};
5298	if (is_vm_hugetlb_page(vma))
5299	continue;
5300	walk_page_range(vma->vm_start, vma->vm_end,
5301	&mem_cgroup_count_precharge_walk);
5302	}
5303	up_read(&mm->mmap_sem);
5304
5305	precharge = mc.precharge;
5306	mc.precharge = 0;
5307
5308	return precharge;
5309	}
5310
5311	static int mem_cgroup_precharge_mc(struct mm_struct *mm)
5312	{
5313	unsigned long precharge = mem_cgroup_count_precharge(mm);
5314
5315	VM_BUG_ON(mc.moving_task);
5316	mc.moving_task = current;
5317	return mem_cgroup_do_precharge(precharge);
5318	}
5319
5320	/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5321	static void __mem_cgroup_clear_mc(void)
5322	{
5323	struct mem_cgroup *from = mc.from;
5324	struct mem_cgroup *to = mc.to;
5325
5326	/* we must uncharge all the leftover precharges from mc.to */
5327	if (mc.precharge) {
5328	__mem_cgroup_cancel_charge(mc.to, mc.precharge);
5329	mc.precharge = 0;
5330	}
5331	/*
5332	* we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5333	* we must uncharge here.
5334	*/
5335	if (mc.moved_charge) {
5336	__mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
5337	mc.moved_charge = 0;
5338	}
5339	/* we must fixup refcnts and charges */
5340	if (mc.moved_swap) {
5341	/* uncharge swap account from the old cgroup */
5342	if (!mem_cgroup_is_root(mc.from))
5343	res_counter_uncharge(&mc.from->memsw,
5344	PAGE_SIZE * mc.moved_swap);
5345	__mem_cgroup_put(mc.from, mc.moved_swap);
5346
5347	if (!mem_cgroup_is_root(mc.to)) {
5348	/*
5349	* we charged both to->res and to->memsw, so we should
5350	* uncharge to->res.
5351	*/
5352	res_counter_uncharge(&mc.to->res,
5353	PAGE_SIZE * mc.moved_swap);
5354	}
5355	/* we've already done mem_cgroup_get(mc.to) */
5356	mc.moved_swap = 0;
5357	}
5358	memcg_oom_recover(from);
5359	memcg_oom_recover(to);
5360	wake_up_all(&mc.waitq);
5361	}
5362
5363	static void mem_cgroup_clear_mc(void)
5364	{
5365	struct mem_cgroup *from = mc.from;
5366
5367	/*
5368	* we must clear moving_task before waking up waiters at the end of
5369	* task migration.
5370	*/
5371	mc.moving_task = NULL;
5372	__mem_cgroup_clear_mc();
5373	spin_lock(&mc.lock);
5374	mc.from = NULL;
5375	mc.to = NULL;
5376	spin_unlock(&mc.lock);
5377	mem_cgroup_end_move(from);
5378	}
5379
5380	static int mem_cgroup_can_attach(struct cgroup *cgroup,
5381	struct cgroup_taskset *tset)
5382	{
5383	struct task_struct *p = cgroup_taskset_first(tset);
5384	int ret = 0;
5385	struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup);
5386
5387	if (memcg->move_charge_at_immigrate) {
5388	struct mm_struct *mm;
5389	struct mem_cgroup *from = mem_cgroup_from_task(p);
5390
5391	VM_BUG_ON(from == memcg);
5392
5393	mm = get_task_mm(p);
5394	if (!mm)
5395	return 0;
5396	/* We move charges only when we move a owner of the mm */
5397	if (mm->owner == p) {
5398	VM_BUG_ON(mc.from);
5399	VM_BUG_ON(mc.to);
5400	VM_BUG_ON(mc.precharge);
5401	VM_BUG_ON(mc.moved_charge);
5402	VM_BUG_ON(mc.moved_swap);
5403	mem_cgroup_start_move(from);
5404	spin_lock(&mc.lock);
5405	mc.from = from;
5406	mc.to = memcg;
5407	spin_unlock(&mc.lock);
5408	/* We set mc.moving_task later */
5409
5410	ret = mem_cgroup_precharge_mc(mm);
5411	if (ret)
5412	mem_cgroup_clear_mc();
5413	}
5414	mmput(mm);
5415	}
5416	return ret;
5417	}
5418
5419	static void mem_cgroup_cancel_attach(struct cgroup *cgroup,
5420	struct cgroup_taskset *tset)
5421	{
5422	mem_cgroup_clear_mc();
5423	}
5424
5425	static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
5426	unsigned long addr, unsigned long end,
5427	struct mm_walk *walk)
5428	{
5429	int ret = 0;
5430	struct vm_area_struct *vma = walk->private;
5431	pte_t *pte;
5432	spinlock_t *ptl;
5433	enum mc_target_type target_type;
5434	union mc_target target;
5435	struct page *page;
5436	struct page_cgroup *pc;
5437
5438	/*
5439	* We don't take compound_lock() here but no race with splitting thp
5440	* happens because:
5441	* - if pmd_trans_huge_lock() returns 1, the relevant thp is not
5442	* under splitting, which means there's no concurrent thp split,
5443	* - if another thread runs into split_huge_page() just after we
5444	* entered this if-block, the thread must wait for page table lock
5445	* to be unlocked in __split_huge_page_splitting(), where the main
5446	* part of thp split is not executed yet.
5447	*/
5448	if (pmd_trans_huge_lock(pmd, vma) == 1) {
5449	if (mc.precharge < HPAGE_PMD_NR) {
5450	spin_unlock(&vma->vm_mm->page_table_lock);
5451	return 0;
5452	}
5453	target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
5454	if (target_type == MC_TARGET_PAGE) {
5455	page = target.page;
5456	if (!isolate_lru_page(page)) {
5457	pc = lookup_page_cgroup(page);
5458	if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
5459	pc, mc.from, mc.to)) {
5460	mc.precharge -= HPAGE_PMD_NR;
5461	mc.moved_charge += HPAGE_PMD_NR;
5462	}
5463	putback_lru_page(page);
5464	}
5465	put_page(page);
5466	}
5467	spin_unlock(&vma->vm_mm->page_table_lock);
5468	return 0;
5469	}
5470
5471	if (pmd_trans_unstable(pmd))
5472	return 0;
5473	retry:
5474	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5475	for (; addr != end; addr += PAGE_SIZE) {
5476	pte_t ptent = *(pte++);
5477	swp_entry_t ent;
5478
5479	if (!mc.precharge)
5480	break;
5481
5482	switch (get_mctgt_type(vma, addr, ptent, &target)) {
5483	case MC_TARGET_PAGE:
5484	page = target.page;
5485	if (isolate_lru_page(page))
5486	goto put;
5487	pc = lookup_page_cgroup(page);
5488	if (!mem_cgroup_move_account(page, 1, pc,
5489	mc.from, mc.to)) {
5490	mc.precharge--;
5491	/* we uncharge from mc.from later. */
5492	mc.moved_charge++;
5493	}
5494	putback_lru_page(page);
5495	put: /* get_mctgt_type() gets the page */
5496	put_page(page);
5497	break;
5498	case MC_TARGET_SWAP:
5499	ent = target.ent;
5500	if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
5501	mc.precharge--;
5502	/* we fixup refcnts and charges later. */
5503	mc.moved_swap++;
5504	}
5505	break;
5506	default:
5507	break;
5508	}
5509	}
5510	pte_unmap_unlock(pte - 1, ptl);
5511	cond_resched();
5512
5513	if (addr != end) {
5514	/*
5515	* We have consumed all precharges we got in can_attach().
5516	* We try charge one by one, but don't do any additional
5517	* charges to mc.to if we have failed in charge once in attach()
5518	* phase.
5519	*/
5520	ret = mem_cgroup_do_precharge(1);
5521	if (!ret)
5522	goto retry;
5523	}
5524
5525	return ret;
5526	}
5527
5528	static void mem_cgroup_move_charge(struct mm_struct *mm)
5529	{
5530	struct vm_area_struct *vma;
5531
5532	lru_add_drain_all();
5533	retry:
5534	if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
5535	/*
5536	* Someone who are holding the mmap_sem might be waiting in
5537	* waitq. So we cancel all extra charges, wake up all waiters,
5538	* and retry. Because we cancel precharges, we might not be able
5539	* to move enough charges, but moving charge is a best-effort
5540	* feature anyway, so it wouldn't be a big problem.
5541	*/
5542	__mem_cgroup_clear_mc();
5543	cond_resched();
5544	goto retry;
5545	}
5546	for (vma = mm->mmap; vma; vma = vma->vm_next) {
5547	int ret;
5548	struct mm_walk mem_cgroup_move_charge_walk = {
5549	.pmd_entry = mem_cgroup_move_charge_pte_range,
5550	.mm = mm,
5551	.private = vma,
5552	};
5553	if (is_vm_hugetlb_page(vma))
5554	continue;
5555	ret = walk_page_range(vma->vm_start, vma->vm_end,
5556	&mem_cgroup_move_charge_walk);
5557	if (ret)
5558	/*
5559	* means we have consumed all precharges and failed in
5560	* doing additional charge. Just abandon here.
5561	*/
5562	break;
5563	}
5564	up_read(&mm->mmap_sem);
5565	}
5566
5567	static void mem_cgroup_move_task(struct cgroup *cont,
5568	struct cgroup_taskset *tset)
5569	{
5570	struct task_struct *p = cgroup_taskset_first(tset);
5571	struct mm_struct *mm = get_task_mm(p);
5572
5573	if (mm) {
5574	if (mc.to)
5575	mem_cgroup_move_charge(mm);
5576	mmput(mm);
5577	}
5578	if (mc.to)
5579	mem_cgroup_clear_mc();
5580	}
5581	#else /* !CONFIG_MMU */
5582	static int mem_cgroup_can_attach(struct cgroup *cgroup,
5583	struct cgroup_taskset *tset)
5584	{
5585	return 0;
5586	}
5587	static void mem_cgroup_cancel_attach(struct cgroup *cgroup,
5588	struct cgroup_taskset *tset)
5589	{
5590	}
5591	static void mem_cgroup_move_task(struct cgroup *cont,
5592	struct cgroup_taskset *tset)
5593	{
5594	}
5595	#endif
5596
5597	struct cgroup_subsys mem_cgroup_subsys = {
5598	.name = "memory",
5599	.subsys_id = mem_cgroup_subsys_id,
5600	.create = mem_cgroup_create,
5601	.pre_destroy = mem_cgroup_pre_destroy,
5602	.destroy = mem_cgroup_destroy,
5603	.can_attach = mem_cgroup_can_attach,
5604	.cancel_attach = mem_cgroup_cancel_attach,
5605	.attach = mem_cgroup_move_task,
5606	.base_cftypes = mem_cgroup_files,
5607	.early_init = 0,
5608	.use_id = 1,
5609	.__DEPRECATED_clear_css_refs = true,
5610	};
5611
5612	#ifdef CONFIG_MEMCG_SWAP
5613	static int __init enable_swap_account(char *s)
5614	{
5615	/* consider enabled if no parameter or 1 is given */
5616	if (!strcmp(s, "1"))
5617	really_do_swap_account = 1;
5618	else if (!strcmp(s, "0"))
5619	really_do_swap_account = 0;
5620	return 1;
5621	}
5622	__setup("swapaccount=", enable_swap_account);
5623
5624	#endif
5625

Download this file

Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master

Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9

Kernel

Kernel Git Source Tree

Root/mm/memcontrol.c

interactive