1/* memcontrol.h - 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 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
13 *
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17 * GNU General Public License for more details.
18 */
19
20#ifndef _LINUX_MEMCONTROL_H
21#define _LINUX_MEMCONTROL_H
22#include <linux/cgroup.h>
23#include <linux/vm_event_item.h>
24#include <linux/hardirq.h>
25#include <linux/jump_label.h>
26#include <linux/page_counter.h>
27#include <linux/vmpressure.h>
28#include <linux/eventfd.h>
29#include <linux/mmzone.h>
30#include <linux/writeback.h>
31
32struct mem_cgroup;
33struct page;
34struct mm_struct;
35struct kmem_cache;
36
37/*
38 * The corresponding mem_cgroup_stat_names is defined in mm/memcontrol.c,
39 * These two lists should keep in accord with each other.
40 */
41enum mem_cgroup_stat_index {
42	/*
43	 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
44	 */
45	MEM_CGROUP_STAT_CACHE,		/* # of pages charged as cache */
46	MEM_CGROUP_STAT_RSS,		/* # of pages charged as anon rss */
47	MEM_CGROUP_STAT_RSS_HUGE,	/* # of pages charged as anon huge */
48	MEM_CGROUP_STAT_FILE_MAPPED,	/* # of pages charged as file rss */
49	MEM_CGROUP_STAT_DIRTY,          /* # of dirty pages in page cache */
50	MEM_CGROUP_STAT_WRITEBACK,	/* # of pages under writeback */
51	MEM_CGROUP_STAT_SWAP,		/* # of pages, swapped out */
52	MEM_CGROUP_STAT_NSTATS,
53};
54
55struct mem_cgroup_reclaim_cookie {
56	struct zone *zone;
57	int priority;
58	unsigned int generation;
59};
60
61enum mem_cgroup_events_index {
62	MEM_CGROUP_EVENTS_PGPGIN,	/* # of pages paged in */
63	MEM_CGROUP_EVENTS_PGPGOUT,	/* # of pages paged out */
64	MEM_CGROUP_EVENTS_PGFAULT,	/* # of page-faults */
65	MEM_CGROUP_EVENTS_PGMAJFAULT,	/* # of major page-faults */
66	MEM_CGROUP_EVENTS_NSTATS,
67	/* default hierarchy events */
68	MEMCG_LOW = MEM_CGROUP_EVENTS_NSTATS,
69	MEMCG_HIGH,
70	MEMCG_MAX,
71	MEMCG_OOM,
72	MEMCG_NR_EVENTS,
73};
74
75/*
76 * Per memcg event counter is incremented at every pagein/pageout. With THP,
77 * it will be incremated by the number of pages. This counter is used for
78 * for trigger some periodic events. This is straightforward and better
79 * than using jiffies etc. to handle periodic memcg event.
80 */
81enum mem_cgroup_events_target {
82	MEM_CGROUP_TARGET_THRESH,
83	MEM_CGROUP_TARGET_SOFTLIMIT,
84	MEM_CGROUP_TARGET_NUMAINFO,
85	MEM_CGROUP_NTARGETS,
86};
87
88/*
89 * Bits in struct cg_proto.flags
90 */
91enum cg_proto_flags {
92	/* Currently active and new sockets should be assigned to cgroups */
93	MEMCG_SOCK_ACTIVE,
94	/* It was ever activated; we must disarm static keys on destruction */
95	MEMCG_SOCK_ACTIVATED,
96};
97
98struct cg_proto {
99	struct page_counter	memory_allocated;	/* Current allocated memory. */
100	struct percpu_counter	sockets_allocated;	/* Current number of sockets. */
101	int			memory_pressure;
102	long			sysctl_mem[3];
103	unsigned long		flags;
104	/*
105	 * memcg field is used to find which memcg we belong directly
106	 * Each memcg struct can hold more than one cg_proto, so container_of
107	 * won't really cut.
108	 *
109	 * The elegant solution would be having an inverse function to
110	 * proto_cgroup in struct proto, but that means polluting the structure
111	 * for everybody, instead of just for memcg users.
112	 */
113	struct mem_cgroup	*memcg;
114};
115
116#ifdef CONFIG_MEMCG
117struct mem_cgroup_stat_cpu {
118	long count[MEM_CGROUP_STAT_NSTATS];
119	unsigned long events[MEMCG_NR_EVENTS];
120	unsigned long nr_page_events;
121	unsigned long targets[MEM_CGROUP_NTARGETS];
122};
123
124struct mem_cgroup_reclaim_iter {
125	struct mem_cgroup *position;
126	/* scan generation, increased every round-trip */
127	unsigned int generation;
128};
129
130/*
131 * per-zone information in memory controller.
132 */
133struct mem_cgroup_per_zone {
134	struct lruvec		lruvec;
135	unsigned long		lru_size[NR_LRU_LISTS];
136
137	struct mem_cgroup_reclaim_iter	iter[DEF_PRIORITY + 1];
138
139	struct rb_node		tree_node;	/* RB tree node */
140	unsigned long		usage_in_excess;/* Set to the value by which */
141						/* the soft limit is exceeded*/
142	bool			on_tree;
143	struct mem_cgroup	*memcg;		/* Back pointer, we cannot */
144						/* use container_of	   */
145};
146
147struct mem_cgroup_per_node {
148	struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
149};
150
151struct mem_cgroup_threshold {
152	struct eventfd_ctx *eventfd;
153	unsigned long threshold;
154};
155
156/* For threshold */
157struct mem_cgroup_threshold_ary {
158	/* An array index points to threshold just below or equal to usage. */
159	int current_threshold;
160	/* Size of entries[] */
161	unsigned int size;
162	/* Array of thresholds */
163	struct mem_cgroup_threshold entries[0];
164};
165
166struct mem_cgroup_thresholds {
167	/* Primary thresholds array */
168	struct mem_cgroup_threshold_ary *primary;
169	/*
170	 * Spare threshold array.
171	 * This is needed to make mem_cgroup_unregister_event() "never fail".
172	 * It must be able to store at least primary->size - 1 entries.
173	 */
174	struct mem_cgroup_threshold_ary *spare;
175};
176
177/*
178 * The memory controller data structure. The memory controller controls both
179 * page cache and RSS per cgroup. We would eventually like to provide
180 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
181 * to help the administrator determine what knobs to tune.
182 */
183struct mem_cgroup {
184	struct cgroup_subsys_state css;
185
186	/* Accounted resources */
187	struct page_counter memory;
188	struct page_counter memsw;
189	struct page_counter kmem;
190
191	/* Normal memory consumption range */
192	unsigned long low;
193	unsigned long high;
194
195	unsigned long soft_limit;
196
197	/* vmpressure notifications */
198	struct vmpressure vmpressure;
199
200	/* css_online() has been completed */
201	int initialized;
202
203	/*
204	 * Should the accounting and control be hierarchical, per subtree?
205	 */
206	bool use_hierarchy;
207
208	/* protected by memcg_oom_lock */
209	bool		oom_lock;
210	int		under_oom;
211
212	int	swappiness;
213	/* OOM-Killer disable */
214	int		oom_kill_disable;
215
216	/* handle for "memory.events" */
217	struct cgroup_file events_file;
218
219	/* protect arrays of thresholds */
220	struct mutex thresholds_lock;
221
222	/* thresholds for memory usage. RCU-protected */
223	struct mem_cgroup_thresholds thresholds;
224
225	/* thresholds for mem+swap usage. RCU-protected */
226	struct mem_cgroup_thresholds memsw_thresholds;
227
228	/* For oom notifier event fd */
229	struct list_head oom_notify;
230
231	/*
232	 * Should we move charges of a task when a task is moved into this
233	 * mem_cgroup ? And what type of charges should we move ?
234	 */
235	unsigned long move_charge_at_immigrate;
236	/*
237	 * set > 0 if pages under this cgroup are moving to other cgroup.
238	 */
239	atomic_t		moving_account;
240	/* taken only while moving_account > 0 */
241	spinlock_t		move_lock;
242	struct task_struct	*move_lock_task;
243	unsigned long		move_lock_flags;
244	/*
245	 * percpu counter.
246	 */
247	struct mem_cgroup_stat_cpu __percpu *stat;
248
249#if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET)
250	struct cg_proto tcp_mem;
251#endif
252#if defined(CONFIG_MEMCG_KMEM)
253        /* Index in the kmem_cache->memcg_params.memcg_caches array */
254	int kmemcg_id;
255	bool kmem_acct_activated;
256	bool kmem_acct_active;
257#endif
258
259	int last_scanned_node;
260#if MAX_NUMNODES > 1
261	nodemask_t	scan_nodes;
262	atomic_t	numainfo_events;
263	atomic_t	numainfo_updating;
264#endif
265
266#ifdef CONFIG_CGROUP_WRITEBACK
267	struct list_head cgwb_list;
268	struct wb_domain cgwb_domain;
269#endif
270
271	/* List of events which userspace want to receive */
272	struct list_head event_list;
273	spinlock_t event_list_lock;
274
275	struct mem_cgroup_per_node *nodeinfo[0];
276	/* WARNING: nodeinfo must be the last member here */
277};
278extern struct cgroup_subsys_state *mem_cgroup_root_css;
279
280/**
281 * mem_cgroup_events - count memory events against a cgroup
282 * @memcg: the memory cgroup
283 * @idx: the event index
284 * @nr: the number of events to account for
285 */
286static inline void mem_cgroup_events(struct mem_cgroup *memcg,
287		       enum mem_cgroup_events_index idx,
288		       unsigned int nr)
289{
290	this_cpu_add(memcg->stat->events[idx], nr);
291	cgroup_file_notify(&memcg->events_file);
292}
293
294bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg);
295
296int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
297			  gfp_t gfp_mask, struct mem_cgroup **memcgp);
298void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
299			      bool lrucare);
300void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg);
301void mem_cgroup_uncharge(struct page *page);
302void mem_cgroup_uncharge_list(struct list_head *page_list);
303
304void mem_cgroup_replace_page(struct page *oldpage, struct page *newpage);
305
306struct lruvec *mem_cgroup_zone_lruvec(struct zone *, struct mem_cgroup *);
307struct lruvec *mem_cgroup_page_lruvec(struct page *, struct zone *);
308
309bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg);
310struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
311struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg);
312
313static inline
314struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
315	return css ? container_of(css, struct mem_cgroup, css) : NULL;
316}
317
318struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
319				   struct mem_cgroup *,
320				   struct mem_cgroup_reclaim_cookie *);
321void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
322
323static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
324			      struct mem_cgroup *root)
325{
326	if (root == memcg)
327		return true;
328	if (!root->use_hierarchy)
329		return false;
330	return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
331}
332
333static inline bool mm_match_cgroup(struct mm_struct *mm,
334				   struct mem_cgroup *memcg)
335{
336	struct mem_cgroup *task_memcg;
337	bool match = false;
338
339	rcu_read_lock();
340	task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
341	if (task_memcg)
342		match = mem_cgroup_is_descendant(task_memcg, memcg);
343	rcu_read_unlock();
344	return match;
345}
346
347struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
348ino_t page_cgroup_ino(struct page *page);
349
350static inline bool mem_cgroup_disabled(void)
351{
352	return !cgroup_subsys_enabled(memory_cgrp_subsys);
353}
354
355/*
356 * For memory reclaim.
357 */
358int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
359
360void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
361		int nr_pages);
362
363static inline bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
364{
365	struct mem_cgroup_per_zone *mz;
366	struct mem_cgroup *memcg;
367
368	if (mem_cgroup_disabled())
369		return true;
370
371	mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
372	memcg = mz->memcg;
373
374	return !!(memcg->css.flags & CSS_ONLINE);
375}
376
377static inline
378unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
379{
380	struct mem_cgroup_per_zone *mz;
381
382	mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
383	return mz->lru_size[lru];
384}
385
386static inline bool mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
387{
388	unsigned long inactive_ratio;
389	unsigned long inactive;
390	unsigned long active;
391	unsigned long gb;
392
393	inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
394	active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
395
396	gb = (inactive + active) >> (30 - PAGE_SHIFT);
397	if (gb)
398		inactive_ratio = int_sqrt(10 * gb);
399	else
400		inactive_ratio = 1;
401
402	return inactive * inactive_ratio < active;
403}
404
405void mem_cgroup_handle_over_high(void);
406
407void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
408				struct task_struct *p);
409
410static inline void mem_cgroup_oom_enable(void)
411{
412	WARN_ON(current->memcg_may_oom);
413	current->memcg_may_oom = 1;
414}
415
416static inline void mem_cgroup_oom_disable(void)
417{
418	WARN_ON(!current->memcg_may_oom);
419	current->memcg_may_oom = 0;
420}
421
422static inline bool task_in_memcg_oom(struct task_struct *p)
423{
424	return p->memcg_in_oom;
425}
426
427bool mem_cgroup_oom_synchronize(bool wait);
428
429#ifdef CONFIG_MEMCG_SWAP
430extern int do_swap_account;
431#endif
432
433struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page);
434void mem_cgroup_end_page_stat(struct mem_cgroup *memcg);
435
436/**
437 * mem_cgroup_update_page_stat - update page state statistics
438 * @memcg: memcg to account against
439 * @idx: page state item to account
440 * @val: number of pages (positive or negative)
441 *
442 * See mem_cgroup_begin_page_stat() for locking requirements.
443 */
444static inline void mem_cgroup_update_page_stat(struct mem_cgroup *memcg,
445				 enum mem_cgroup_stat_index idx, int val)
446{
447	VM_BUG_ON(!rcu_read_lock_held());
448
449	if (memcg)
450		this_cpu_add(memcg->stat->count[idx], val);
451}
452
453static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
454					    enum mem_cgroup_stat_index idx)
455{
456	mem_cgroup_update_page_stat(memcg, idx, 1);
457}
458
459static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
460					    enum mem_cgroup_stat_index idx)
461{
462	mem_cgroup_update_page_stat(memcg, idx, -1);
463}
464
465unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
466						gfp_t gfp_mask,
467						unsigned long *total_scanned);
468
469static inline void mem_cgroup_count_vm_event(struct mm_struct *mm,
470					     enum vm_event_item idx)
471{
472	struct mem_cgroup *memcg;
473
474	if (mem_cgroup_disabled())
475		return;
476
477	rcu_read_lock();
478	memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
479	if (unlikely(!memcg))
480		goto out;
481
482	switch (idx) {
483	case PGFAULT:
484		this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
485		break;
486	case PGMAJFAULT:
487		this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
488		break;
489	default:
490		BUG();
491	}
492out:
493	rcu_read_unlock();
494}
495#ifdef CONFIG_TRANSPARENT_HUGEPAGE
496void mem_cgroup_split_huge_fixup(struct page *head);
497#endif
498
499#else /* CONFIG_MEMCG */
500struct mem_cgroup;
501
502static inline void mem_cgroup_events(struct mem_cgroup *memcg,
503				     enum mem_cgroup_events_index idx,
504				     unsigned int nr)
505{
506}
507
508static inline bool mem_cgroup_low(struct mem_cgroup *root,
509				  struct mem_cgroup *memcg)
510{
511	return false;
512}
513
514static inline int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
515					gfp_t gfp_mask,
516					struct mem_cgroup **memcgp)
517{
518	*memcgp = NULL;
519	return 0;
520}
521
522static inline void mem_cgroup_commit_charge(struct page *page,
523					    struct mem_cgroup *memcg,
524					    bool lrucare)
525{
526}
527
528static inline void mem_cgroup_cancel_charge(struct page *page,
529					    struct mem_cgroup *memcg)
530{
531}
532
533static inline void mem_cgroup_uncharge(struct page *page)
534{
535}
536
537static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
538{
539}
540
541static inline void mem_cgroup_replace_page(struct page *old, struct page *new)
542{
543}
544
545static inline struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
546						    struct mem_cgroup *memcg)
547{
548	return &zone->lruvec;
549}
550
551static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
552						    struct zone *zone)
553{
554	return &zone->lruvec;
555}
556
557static inline bool mm_match_cgroup(struct mm_struct *mm,
558		struct mem_cgroup *memcg)
559{
560	return true;
561}
562
563static inline bool task_in_mem_cgroup(struct task_struct *task,
564				      const struct mem_cgroup *memcg)
565{
566	return true;
567}
568
569static inline struct mem_cgroup *
570mem_cgroup_iter(struct mem_cgroup *root,
571		struct mem_cgroup *prev,
572		struct mem_cgroup_reclaim_cookie *reclaim)
573{
574	return NULL;
575}
576
577static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
578					 struct mem_cgroup *prev)
579{
580}
581
582static inline bool mem_cgroup_disabled(void)
583{
584	return true;
585}
586
587static inline bool
588mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
589{
590	return true;
591}
592
593static inline bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
594{
595	return true;
596}
597
598static inline unsigned long
599mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
600{
601	return 0;
602}
603
604static inline void
605mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
606			      int increment)
607{
608}
609
610static inline void
611mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
612{
613}
614
615static inline struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
616{
617	return NULL;
618}
619
620static inline void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
621{
622}
623
624static inline void mem_cgroup_handle_over_high(void)
625{
626}
627
628static inline void mem_cgroup_oom_enable(void)
629{
630}
631
632static inline void mem_cgroup_oom_disable(void)
633{
634}
635
636static inline bool task_in_memcg_oom(struct task_struct *p)
637{
638	return false;
639}
640
641static inline bool mem_cgroup_oom_synchronize(bool wait)
642{
643	return false;
644}
645
646static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
647					    enum mem_cgroup_stat_index idx)
648{
649}
650
651static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
652					    enum mem_cgroup_stat_index idx)
653{
654}
655
656static inline
657unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
658					    gfp_t gfp_mask,
659					    unsigned long *total_scanned)
660{
661	return 0;
662}
663
664static inline void mem_cgroup_split_huge_fixup(struct page *head)
665{
666}
667
668static inline
669void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
670{
671}
672#endif /* CONFIG_MEMCG */
673
674enum {
675	UNDER_LIMIT,
676	SOFT_LIMIT,
677	OVER_LIMIT,
678};
679
680#ifdef CONFIG_CGROUP_WRITEBACK
681
682struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg);
683struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
684void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
685			 unsigned long *pheadroom, unsigned long *pdirty,
686			 unsigned long *pwriteback);
687
688#else	/* CONFIG_CGROUP_WRITEBACK */
689
690static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
691{
692	return NULL;
693}
694
695static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
696				       unsigned long *pfilepages,
697				       unsigned long *pheadroom,
698				       unsigned long *pdirty,
699				       unsigned long *pwriteback)
700{
701}
702
703#endif	/* CONFIG_CGROUP_WRITEBACK */
704
705struct sock;
706#if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
707void sock_update_memcg(struct sock *sk);
708void sock_release_memcg(struct sock *sk);
709#else
710static inline void sock_update_memcg(struct sock *sk)
711{
712}
713static inline void sock_release_memcg(struct sock *sk)
714{
715}
716#endif /* CONFIG_INET && CONFIG_MEMCG_KMEM */
717
718#ifdef CONFIG_MEMCG_KMEM
719extern struct static_key memcg_kmem_enabled_key;
720
721extern int memcg_nr_cache_ids;
722void memcg_get_cache_ids(void);
723void memcg_put_cache_ids(void);
724
725/*
726 * Helper macro to loop through all memcg-specific caches. Callers must still
727 * check if the cache is valid (it is either valid or NULL).
728 * the slab_mutex must be held when looping through those caches
729 */
730#define for_each_memcg_cache_index(_idx)	\
731	for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
732
733static inline bool memcg_kmem_enabled(void)
734{
735	return static_key_false(&memcg_kmem_enabled_key);
736}
737
738static inline bool memcg_kmem_is_active(struct mem_cgroup *memcg)
739{
740	return memcg->kmem_acct_active;
741}
742
743/*
744 * In general, we'll do everything in our power to not incur in any overhead
745 * for non-memcg users for the kmem functions. Not even a function call, if we
746 * can avoid it.
747 *
748 * Therefore, we'll inline all those functions so that in the best case, we'll
749 * see that kmemcg is off for everybody and proceed quickly.  If it is on,
750 * we'll still do most of the flag checking inline. We check a lot of
751 * conditions, but because they are pretty simple, they are expected to be
752 * fast.
753 */
754int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
755			      struct mem_cgroup *memcg);
756int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order);
757void __memcg_kmem_uncharge(struct page *page, int order);
758
759/*
760 * helper for acessing a memcg's index. It will be used as an index in the
761 * child cache array in kmem_cache, and also to derive its name. This function
762 * will return -1 when this is not a kmem-limited memcg.
763 */
764static inline int memcg_cache_id(struct mem_cgroup *memcg)
765{
766	return memcg ? memcg->kmemcg_id : -1;
767}
768
769struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep);
770void __memcg_kmem_put_cache(struct kmem_cache *cachep);
771
772static inline bool __memcg_kmem_bypass(gfp_t gfp)
773{
774	if (!memcg_kmem_enabled())
775		return true;
776	if (gfp & __GFP_NOACCOUNT)
777		return true;
778	if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
779		return true;
780	return false;
781}
782
783/**
784 * memcg_kmem_charge: charge a kmem page
785 * @page: page to charge
786 * @gfp: reclaim mode
787 * @order: allocation order
788 *
789 * Returns 0 on success, an error code on failure.
790 */
791static __always_inline int memcg_kmem_charge(struct page *page,
792					     gfp_t gfp, int order)
793{
794	if (__memcg_kmem_bypass(gfp))
795		return 0;
796	return __memcg_kmem_charge(page, gfp, order);
797}
798
799/**
800 * memcg_kmem_uncharge: uncharge a kmem page
801 * @page: page to uncharge
802 * @order: allocation order
803 */
804static __always_inline void memcg_kmem_uncharge(struct page *page, int order)
805{
806	if (memcg_kmem_enabled())
807		__memcg_kmem_uncharge(page, order);
808}
809
810/**
811 * memcg_kmem_get_cache: selects the correct per-memcg cache for allocation
812 * @cachep: the original global kmem cache
813 * @gfp: allocation flags.
814 *
815 * All memory allocated from a per-memcg cache is charged to the owner memcg.
816 */
817static __always_inline struct kmem_cache *
818memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
819{
820	if (__memcg_kmem_bypass(gfp))
821		return cachep;
822	return __memcg_kmem_get_cache(cachep);
823}
824
825static __always_inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
826{
827	if (memcg_kmem_enabled())
828		__memcg_kmem_put_cache(cachep);
829}
830#else
831#define for_each_memcg_cache_index(_idx)	\
832	for (; NULL; )
833
834static inline bool memcg_kmem_enabled(void)
835{
836	return false;
837}
838
839static inline bool memcg_kmem_is_active(struct mem_cgroup *memcg)
840{
841	return false;
842}
843
844static inline int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
845{
846	return 0;
847}
848
849static inline void memcg_kmem_uncharge(struct page *page, int order)
850{
851}
852
853static inline int memcg_cache_id(struct mem_cgroup *memcg)
854{
855	return -1;
856}
857
858static inline void memcg_get_cache_ids(void)
859{
860}
861
862static inline void memcg_put_cache_ids(void)
863{
864}
865
866static inline struct kmem_cache *
867memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
868{
869	return cachep;
870}
871
872static inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
873{
874}
875#endif /* CONFIG_MEMCG_KMEM */
876#endif /* _LINUX_MEMCONTROL_H */
877