Loading...
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#include <linux/page-flags.h>
32
33struct mem_cgroup;
34struct page;
35struct mm_struct;
36struct kmem_cache;
37
38/*
39 * The corresponding mem_cgroup_stat_names is defined in mm/memcontrol.c,
40 * These two lists should keep in accord with each other.
41 */
42enum mem_cgroup_stat_index {
43 /*
44 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
45 */
46 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
47 MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
48 MEM_CGROUP_STAT_RSS_HUGE, /* # of pages charged as anon huge */
49 MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
50 MEM_CGROUP_STAT_DIRTY, /* # of dirty pages in page cache */
51 MEM_CGROUP_STAT_WRITEBACK, /* # of pages under writeback */
52 MEM_CGROUP_STAT_SWAP, /* # of pages, swapped out */
53 MEM_CGROUP_STAT_NSTATS,
54 /* default hierarchy stats */
55 MEMCG_KERNEL_STACK_KB = MEM_CGROUP_STAT_NSTATS,
56 MEMCG_SLAB_RECLAIMABLE,
57 MEMCG_SLAB_UNRECLAIMABLE,
58 MEMCG_SOCK,
59 MEMCG_NR_STAT,
60};
61
62struct mem_cgroup_reclaim_cookie {
63 pg_data_t *pgdat;
64 int priority;
65 unsigned int generation;
66};
67
68enum mem_cgroup_events_index {
69 MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */
70 MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */
71 MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */
72 MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */
73 MEM_CGROUP_EVENTS_NSTATS,
74 /* default hierarchy events */
75 MEMCG_LOW = MEM_CGROUP_EVENTS_NSTATS,
76 MEMCG_HIGH,
77 MEMCG_MAX,
78 MEMCG_OOM,
79 MEMCG_NR_EVENTS,
80};
81
82/*
83 * Per memcg event counter is incremented at every pagein/pageout. With THP,
84 * it will be incremated by the number of pages. This counter is used for
85 * for trigger some periodic events. This is straightforward and better
86 * than using jiffies etc. to handle periodic memcg event.
87 */
88enum mem_cgroup_events_target {
89 MEM_CGROUP_TARGET_THRESH,
90 MEM_CGROUP_TARGET_SOFTLIMIT,
91 MEM_CGROUP_TARGET_NUMAINFO,
92 MEM_CGROUP_NTARGETS,
93};
94
95#ifdef CONFIG_MEMCG
96
97#define MEM_CGROUP_ID_SHIFT 16
98#define MEM_CGROUP_ID_MAX USHRT_MAX
99
100struct mem_cgroup_id {
101 int id;
102 atomic_t ref;
103};
104
105struct mem_cgroup_stat_cpu {
106 long count[MEMCG_NR_STAT];
107 unsigned long events[MEMCG_NR_EVENTS];
108 unsigned long nr_page_events;
109 unsigned long targets[MEM_CGROUP_NTARGETS];
110};
111
112struct mem_cgroup_reclaim_iter {
113 struct mem_cgroup *position;
114 /* scan generation, increased every round-trip */
115 unsigned int generation;
116};
117
118/*
119 * per-zone information in memory controller.
120 */
121struct mem_cgroup_per_node {
122 struct lruvec lruvec;
123 unsigned long lru_zone_size[MAX_NR_ZONES][NR_LRU_LISTS];
124
125 struct mem_cgroup_reclaim_iter iter[DEF_PRIORITY + 1];
126
127 struct rb_node tree_node; /* RB tree node */
128 unsigned long usage_in_excess;/* Set to the value by which */
129 /* the soft limit is exceeded*/
130 bool on_tree;
131 struct mem_cgroup *memcg; /* Back pointer, we cannot */
132 /* use container_of */
133};
134
135struct mem_cgroup_threshold {
136 struct eventfd_ctx *eventfd;
137 unsigned long threshold;
138};
139
140/* For threshold */
141struct mem_cgroup_threshold_ary {
142 /* An array index points to threshold just below or equal to usage. */
143 int current_threshold;
144 /* Size of entries[] */
145 unsigned int size;
146 /* Array of thresholds */
147 struct mem_cgroup_threshold entries[0];
148};
149
150struct mem_cgroup_thresholds {
151 /* Primary thresholds array */
152 struct mem_cgroup_threshold_ary *primary;
153 /*
154 * Spare threshold array.
155 * This is needed to make mem_cgroup_unregister_event() "never fail".
156 * It must be able to store at least primary->size - 1 entries.
157 */
158 struct mem_cgroup_threshold_ary *spare;
159};
160
161enum memcg_kmem_state {
162 KMEM_NONE,
163 KMEM_ALLOCATED,
164 KMEM_ONLINE,
165};
166
167/*
168 * The memory controller data structure. The memory controller controls both
169 * page cache and RSS per cgroup. We would eventually like to provide
170 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
171 * to help the administrator determine what knobs to tune.
172 */
173struct mem_cgroup {
174 struct cgroup_subsys_state css;
175
176 /* Private memcg ID. Used to ID objects that outlive the cgroup */
177 struct mem_cgroup_id id;
178
179 /* Accounted resources */
180 struct page_counter memory;
181 struct page_counter swap;
182
183 /* Legacy consumer-oriented counters */
184 struct page_counter memsw;
185 struct page_counter kmem;
186 struct page_counter tcpmem;
187
188 /* Normal memory consumption range */
189 unsigned long low;
190 unsigned long high;
191
192 /* Range enforcement for interrupt charges */
193 struct work_struct high_work;
194
195 unsigned long soft_limit;
196
197 /* vmpressure notifications */
198 struct vmpressure vmpressure;
199
200 /*
201 * Should the accounting and control be hierarchical, per subtree?
202 */
203 bool use_hierarchy;
204
205 /* protected by memcg_oom_lock */
206 bool oom_lock;
207 int under_oom;
208
209 int swappiness;
210 /* OOM-Killer disable */
211 int oom_kill_disable;
212
213 /* handle for "memory.events" */
214 struct cgroup_file events_file;
215
216 /* protect arrays of thresholds */
217 struct mutex thresholds_lock;
218
219 /* thresholds for memory usage. RCU-protected */
220 struct mem_cgroup_thresholds thresholds;
221
222 /* thresholds for mem+swap usage. RCU-protected */
223 struct mem_cgroup_thresholds memsw_thresholds;
224
225 /* For oom notifier event fd */
226 struct list_head oom_notify;
227
228 /*
229 * Should we move charges of a task when a task is moved into this
230 * mem_cgroup ? And what type of charges should we move ?
231 */
232 unsigned long move_charge_at_immigrate;
233 /*
234 * set > 0 if pages under this cgroup are moving to other cgroup.
235 */
236 atomic_t moving_account;
237 /* taken only while moving_account > 0 */
238 spinlock_t move_lock;
239 struct task_struct *move_lock_task;
240 unsigned long move_lock_flags;
241 /*
242 * percpu counter.
243 */
244 struct mem_cgroup_stat_cpu __percpu *stat;
245
246 unsigned long socket_pressure;
247
248 /* Legacy tcp memory accounting */
249 bool tcpmem_active;
250 int tcpmem_pressure;
251
252#ifndef CONFIG_SLOB
253 /* Index in the kmem_cache->memcg_params.memcg_caches array */
254 int kmemcg_id;
255 enum memcg_kmem_state kmem_state;
256#endif
257
258 int last_scanned_node;
259#if MAX_NUMNODES > 1
260 nodemask_t scan_nodes;
261 atomic_t numainfo_events;
262 atomic_t numainfo_updating;
263#endif
264
265#ifdef CONFIG_CGROUP_WRITEBACK
266 struct list_head cgwb_list;
267 struct wb_domain cgwb_domain;
268#endif
269
270 /* List of events which userspace want to receive */
271 struct list_head event_list;
272 spinlock_t event_list_lock;
273
274 struct mem_cgroup_per_node *nodeinfo[0];
275 /* WARNING: nodeinfo must be the last member here */
276};
277
278extern struct mem_cgroup *root_mem_cgroup;
279
280static inline bool mem_cgroup_disabled(void)
281{
282 return !cgroup_subsys_enabled(memory_cgrp_subsys);
283}
284
285/**
286 * mem_cgroup_events - count memory events against a cgroup
287 * @memcg: the memory cgroup
288 * @idx: the event index
289 * @nr: the number of events to account for
290 */
291static inline void mem_cgroup_events(struct mem_cgroup *memcg,
292 enum mem_cgroup_events_index idx,
293 unsigned int nr)
294{
295 this_cpu_add(memcg->stat->events[idx], nr);
296 cgroup_file_notify(&memcg->events_file);
297}
298
299bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg);
300
301int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
302 gfp_t gfp_mask, struct mem_cgroup **memcgp,
303 bool compound);
304void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
305 bool lrucare, bool compound);
306void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
307 bool compound);
308void mem_cgroup_uncharge(struct page *page);
309void mem_cgroup_uncharge_list(struct list_head *page_list);
310
311void mem_cgroup_migrate(struct page *oldpage, struct page *newpage);
312
313static struct mem_cgroup_per_node *
314mem_cgroup_nodeinfo(struct mem_cgroup *memcg, int nid)
315{
316 return memcg->nodeinfo[nid];
317}
318
319/**
320 * mem_cgroup_lruvec - get the lru list vector for a node or a memcg zone
321 * @node: node of the wanted lruvec
322 * @memcg: memcg of the wanted lruvec
323 *
324 * Returns the lru list vector holding pages for a given @node or a given
325 * @memcg and @zone. This can be the node lruvec, if the memory controller
326 * is disabled.
327 */
328static inline struct lruvec *mem_cgroup_lruvec(struct pglist_data *pgdat,
329 struct mem_cgroup *memcg)
330{
331 struct mem_cgroup_per_node *mz;
332 struct lruvec *lruvec;
333
334 if (mem_cgroup_disabled()) {
335 lruvec = node_lruvec(pgdat);
336 goto out;
337 }
338
339 mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
340 lruvec = &mz->lruvec;
341out:
342 /*
343 * Since a node can be onlined after the mem_cgroup was created,
344 * we have to be prepared to initialize lruvec->pgdat here;
345 * and if offlined then reonlined, we need to reinitialize it.
346 */
347 if (unlikely(lruvec->pgdat != pgdat))
348 lruvec->pgdat = pgdat;
349 return lruvec;
350}
351
352struct lruvec *mem_cgroup_page_lruvec(struct page *, struct pglist_data *);
353
354bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg);
355struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
356
357static inline
358struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
359 return css ? container_of(css, struct mem_cgroup, css) : NULL;
360}
361
362#define mem_cgroup_from_counter(counter, member) \
363 container_of(counter, struct mem_cgroup, member)
364
365struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
366 struct mem_cgroup *,
367 struct mem_cgroup_reclaim_cookie *);
368void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
369int mem_cgroup_scan_tasks(struct mem_cgroup *,
370 int (*)(struct task_struct *, void *), void *);
371
372static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
373{
374 if (mem_cgroup_disabled())
375 return 0;
376
377 return memcg->id.id;
378}
379struct mem_cgroup *mem_cgroup_from_id(unsigned short id);
380
381/**
382 * parent_mem_cgroup - find the accounting parent of a memcg
383 * @memcg: memcg whose parent to find
384 *
385 * Returns the parent memcg, or NULL if this is the root or the memory
386 * controller is in legacy no-hierarchy mode.
387 */
388static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
389{
390 if (!memcg->memory.parent)
391 return NULL;
392 return mem_cgroup_from_counter(memcg->memory.parent, memory);
393}
394
395static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
396 struct mem_cgroup *root)
397{
398 if (root == memcg)
399 return true;
400 if (!root->use_hierarchy)
401 return false;
402 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
403}
404
405static inline bool mm_match_cgroup(struct mm_struct *mm,
406 struct mem_cgroup *memcg)
407{
408 struct mem_cgroup *task_memcg;
409 bool match = false;
410
411 rcu_read_lock();
412 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
413 if (task_memcg)
414 match = mem_cgroup_is_descendant(task_memcg, memcg);
415 rcu_read_unlock();
416 return match;
417}
418
419struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
420ino_t page_cgroup_ino(struct page *page);
421
422static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
423{
424 if (mem_cgroup_disabled())
425 return true;
426 return !!(memcg->css.flags & CSS_ONLINE);
427}
428
429/*
430 * For memory reclaim.
431 */
432int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
433
434void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
435 int zid, int nr_pages);
436
437unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
438 int nid, unsigned int lru_mask);
439
440static inline
441unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
442{
443 struct mem_cgroup_per_node *mz;
444 unsigned long nr_pages = 0;
445 int zid;
446
447 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
448 for (zid = 0; zid < MAX_NR_ZONES; zid++)
449 nr_pages += mz->lru_zone_size[zid][lru];
450 return nr_pages;
451}
452
453static inline
454unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
455 enum lru_list lru, int zone_idx)
456{
457 struct mem_cgroup_per_node *mz;
458
459 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
460 return mz->lru_zone_size[zone_idx][lru];
461}
462
463void mem_cgroup_handle_over_high(void);
464
465unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg);
466
467void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
468 struct task_struct *p);
469
470static inline void mem_cgroup_oom_enable(void)
471{
472 WARN_ON(current->memcg_may_oom);
473 current->memcg_may_oom = 1;
474}
475
476static inline void mem_cgroup_oom_disable(void)
477{
478 WARN_ON(!current->memcg_may_oom);
479 current->memcg_may_oom = 0;
480}
481
482static inline bool task_in_memcg_oom(struct task_struct *p)
483{
484 return p->memcg_in_oom;
485}
486
487bool mem_cgroup_oom_synchronize(bool wait);
488
489#ifdef CONFIG_MEMCG_SWAP
490extern int do_swap_account;
491#endif
492
493void lock_page_memcg(struct page *page);
494void unlock_page_memcg(struct page *page);
495
496/**
497 * mem_cgroup_update_page_stat - update page state statistics
498 * @page: the page
499 * @idx: page state item to account
500 * @val: number of pages (positive or negative)
501 *
502 * The @page must be locked or the caller must use lock_page_memcg()
503 * to prevent double accounting when the page is concurrently being
504 * moved to another memcg:
505 *
506 * lock_page(page) or lock_page_memcg(page)
507 * if (TestClearPageState(page))
508 * mem_cgroup_update_page_stat(page, state, -1);
509 * unlock_page(page) or unlock_page_memcg(page)
510 */
511static inline void mem_cgroup_update_page_stat(struct page *page,
512 enum mem_cgroup_stat_index idx, int val)
513{
514 VM_BUG_ON(!(rcu_read_lock_held() || PageLocked(page)));
515
516 if (page->mem_cgroup)
517 this_cpu_add(page->mem_cgroup->stat->count[idx], val);
518}
519
520static inline void mem_cgroup_inc_page_stat(struct page *page,
521 enum mem_cgroup_stat_index idx)
522{
523 mem_cgroup_update_page_stat(page, idx, 1);
524}
525
526static inline void mem_cgroup_dec_page_stat(struct page *page,
527 enum mem_cgroup_stat_index idx)
528{
529 mem_cgroup_update_page_stat(page, idx, -1);
530}
531
532unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
533 gfp_t gfp_mask,
534 unsigned long *total_scanned);
535
536static inline void mem_cgroup_count_vm_event(struct mm_struct *mm,
537 enum vm_event_item idx)
538{
539 struct mem_cgroup *memcg;
540
541 if (mem_cgroup_disabled())
542 return;
543
544 rcu_read_lock();
545 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
546 if (unlikely(!memcg))
547 goto out;
548
549 switch (idx) {
550 case PGFAULT:
551 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
552 break;
553 case PGMAJFAULT:
554 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
555 break;
556 default:
557 BUG();
558 }
559out:
560 rcu_read_unlock();
561}
562#ifdef CONFIG_TRANSPARENT_HUGEPAGE
563void mem_cgroup_split_huge_fixup(struct page *head);
564#endif
565
566#else /* CONFIG_MEMCG */
567
568#define MEM_CGROUP_ID_SHIFT 0
569#define MEM_CGROUP_ID_MAX 0
570
571struct mem_cgroup;
572
573static inline bool mem_cgroup_disabled(void)
574{
575 return true;
576}
577
578static inline void mem_cgroup_events(struct mem_cgroup *memcg,
579 enum mem_cgroup_events_index idx,
580 unsigned int nr)
581{
582}
583
584static inline bool mem_cgroup_low(struct mem_cgroup *root,
585 struct mem_cgroup *memcg)
586{
587 return false;
588}
589
590static inline int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
591 gfp_t gfp_mask,
592 struct mem_cgroup **memcgp,
593 bool compound)
594{
595 *memcgp = NULL;
596 return 0;
597}
598
599static inline void mem_cgroup_commit_charge(struct page *page,
600 struct mem_cgroup *memcg,
601 bool lrucare, bool compound)
602{
603}
604
605static inline void mem_cgroup_cancel_charge(struct page *page,
606 struct mem_cgroup *memcg,
607 bool compound)
608{
609}
610
611static inline void mem_cgroup_uncharge(struct page *page)
612{
613}
614
615static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
616{
617}
618
619static inline void mem_cgroup_migrate(struct page *old, struct page *new)
620{
621}
622
623static inline struct lruvec *mem_cgroup_lruvec(struct pglist_data *pgdat,
624 struct mem_cgroup *memcg)
625{
626 return node_lruvec(pgdat);
627}
628
629static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
630 struct pglist_data *pgdat)
631{
632 return &pgdat->lruvec;
633}
634
635static inline bool mm_match_cgroup(struct mm_struct *mm,
636 struct mem_cgroup *memcg)
637{
638 return true;
639}
640
641static inline bool task_in_mem_cgroup(struct task_struct *task,
642 const struct mem_cgroup *memcg)
643{
644 return true;
645}
646
647static inline struct mem_cgroup *
648mem_cgroup_iter(struct mem_cgroup *root,
649 struct mem_cgroup *prev,
650 struct mem_cgroup_reclaim_cookie *reclaim)
651{
652 return NULL;
653}
654
655static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
656 struct mem_cgroup *prev)
657{
658}
659
660static inline int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
661 int (*fn)(struct task_struct *, void *), void *arg)
662{
663 return 0;
664}
665
666static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
667{
668 return 0;
669}
670
671static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
672{
673 WARN_ON_ONCE(id);
674 /* XXX: This should always return root_mem_cgroup */
675 return NULL;
676}
677
678static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
679{
680 return true;
681}
682
683static inline unsigned long
684mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
685{
686 return 0;
687}
688static inline
689unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
690 enum lru_list lru, int zone_idx)
691{
692 return 0;
693}
694
695static inline unsigned long
696mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
697 int nid, unsigned int lru_mask)
698{
699 return 0;
700}
701
702static inline unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg)
703{
704 return 0;
705}
706
707static inline void
708mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
709{
710}
711
712static inline void lock_page_memcg(struct page *page)
713{
714}
715
716static inline void unlock_page_memcg(struct page *page)
717{
718}
719
720static inline void mem_cgroup_handle_over_high(void)
721{
722}
723
724static inline void mem_cgroup_oom_enable(void)
725{
726}
727
728static inline void mem_cgroup_oom_disable(void)
729{
730}
731
732static inline bool task_in_memcg_oom(struct task_struct *p)
733{
734 return false;
735}
736
737static inline bool mem_cgroup_oom_synchronize(bool wait)
738{
739 return false;
740}
741
742static inline void mem_cgroup_update_page_stat(struct page *page,
743 enum mem_cgroup_stat_index idx,
744 int nr)
745{
746}
747
748static inline void mem_cgroup_inc_page_stat(struct page *page,
749 enum mem_cgroup_stat_index idx)
750{
751}
752
753static inline void mem_cgroup_dec_page_stat(struct page *page,
754 enum mem_cgroup_stat_index idx)
755{
756}
757
758static inline
759unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
760 gfp_t gfp_mask,
761 unsigned long *total_scanned)
762{
763 return 0;
764}
765
766static inline void mem_cgroup_split_huge_fixup(struct page *head)
767{
768}
769
770static inline
771void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
772{
773}
774#endif /* CONFIG_MEMCG */
775
776#ifdef CONFIG_CGROUP_WRITEBACK
777
778struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg);
779struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
780void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
781 unsigned long *pheadroom, unsigned long *pdirty,
782 unsigned long *pwriteback);
783
784#else /* CONFIG_CGROUP_WRITEBACK */
785
786static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
787{
788 return NULL;
789}
790
791static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
792 unsigned long *pfilepages,
793 unsigned long *pheadroom,
794 unsigned long *pdirty,
795 unsigned long *pwriteback)
796{
797}
798
799#endif /* CONFIG_CGROUP_WRITEBACK */
800
801struct sock;
802bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
803void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
804#ifdef CONFIG_MEMCG
805extern struct static_key_false memcg_sockets_enabled_key;
806#define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_enabled_key)
807void mem_cgroup_sk_alloc(struct sock *sk);
808void mem_cgroup_sk_free(struct sock *sk);
809static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
810{
811 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_pressure)
812 return true;
813 do {
814 if (time_before(jiffies, memcg->socket_pressure))
815 return true;
816 } while ((memcg = parent_mem_cgroup(memcg)));
817 return false;
818}
819#else
820#define mem_cgroup_sockets_enabled 0
821static inline void mem_cgroup_sk_alloc(struct sock *sk) { };
822static inline void mem_cgroup_sk_free(struct sock *sk) { };
823static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
824{
825 return false;
826}
827#endif
828
829struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep);
830void memcg_kmem_put_cache(struct kmem_cache *cachep);
831int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
832 struct mem_cgroup *memcg);
833int memcg_kmem_charge(struct page *page, gfp_t gfp, int order);
834void memcg_kmem_uncharge(struct page *page, int order);
835
836#if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
837extern struct static_key_false memcg_kmem_enabled_key;
838
839extern int memcg_nr_cache_ids;
840void memcg_get_cache_ids(void);
841void memcg_put_cache_ids(void);
842
843/*
844 * Helper macro to loop through all memcg-specific caches. Callers must still
845 * check if the cache is valid (it is either valid or NULL).
846 * the slab_mutex must be held when looping through those caches
847 */
848#define for_each_memcg_cache_index(_idx) \
849 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
850
851static inline bool memcg_kmem_enabled(void)
852{
853 return static_branch_unlikely(&memcg_kmem_enabled_key);
854}
855
856/*
857 * helper for accessing a memcg's index. It will be used as an index in the
858 * child cache array in kmem_cache, and also to derive its name. This function
859 * will return -1 when this is not a kmem-limited memcg.
860 */
861static inline int memcg_cache_id(struct mem_cgroup *memcg)
862{
863 return memcg ? memcg->kmemcg_id : -1;
864}
865
866/**
867 * memcg_kmem_update_page_stat - update kmem page state statistics
868 * @page: the page
869 * @idx: page state item to account
870 * @val: number of pages (positive or negative)
871 */
872static inline void memcg_kmem_update_page_stat(struct page *page,
873 enum mem_cgroup_stat_index idx, int val)
874{
875 if (memcg_kmem_enabled() && page->mem_cgroup)
876 this_cpu_add(page->mem_cgroup->stat->count[idx], val);
877}
878
879#else
880#define for_each_memcg_cache_index(_idx) \
881 for (; NULL; )
882
883static inline bool memcg_kmem_enabled(void)
884{
885 return false;
886}
887
888static inline int memcg_cache_id(struct mem_cgroup *memcg)
889{
890 return -1;
891}
892
893static inline void memcg_get_cache_ids(void)
894{
895}
896
897static inline void memcg_put_cache_ids(void)
898{
899}
900
901static inline void memcg_kmem_update_page_stat(struct page *page,
902 enum mem_cgroup_stat_index idx, int val)
903{
904}
905#endif /* CONFIG_MEMCG && !CONFIG_SLOB */
906
907#endif /* _LINUX_MEMCONTROL_H */
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
27struct mem_cgroup;
28struct page_cgroup;
29struct page;
30struct mm_struct;
31struct kmem_cache;
32
33/*
34 * The corresponding mem_cgroup_stat_names is defined in mm/memcontrol.c,
35 * These two lists should keep in accord with each other.
36 */
37enum mem_cgroup_stat_index {
38 /*
39 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
40 */
41 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
42 MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
43 MEM_CGROUP_STAT_RSS_HUGE, /* # of pages charged as anon huge */
44 MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
45 MEM_CGROUP_STAT_WRITEBACK, /* # of pages under writeback */
46 MEM_CGROUP_STAT_SWAP, /* # of pages, swapped out */
47 MEM_CGROUP_STAT_NSTATS,
48};
49
50struct mem_cgroup_reclaim_cookie {
51 struct zone *zone;
52 int priority;
53 unsigned int generation;
54};
55
56#ifdef CONFIG_MEMCG
57/*
58 * All "charge" functions with gfp_mask should use GFP_KERNEL or
59 * (gfp_mask & GFP_RECLAIM_MASK). In current implementatin, memcg doesn't
60 * alloc memory but reclaims memory from all available zones. So, "where I want
61 * memory from" bits of gfp_mask has no meaning. So any bits of that field is
62 * available but adding a rule is better. charge functions' gfp_mask should
63 * be set to GFP_KERNEL or gfp_mask & GFP_RECLAIM_MASK for avoiding ambiguous
64 * codes.
65 * (Of course, if memcg does memory allocation in future, GFP_KERNEL is sane.)
66 */
67
68extern int mem_cgroup_charge_anon(struct page *page, struct mm_struct *mm,
69 gfp_t gfp_mask);
70/* for swap handling */
71extern int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
72 struct page *page, gfp_t mask, struct mem_cgroup **memcgp);
73extern void mem_cgroup_commit_charge_swapin(struct page *page,
74 struct mem_cgroup *memcg);
75extern void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg);
76
77extern int mem_cgroup_charge_file(struct page *page, struct mm_struct *mm,
78 gfp_t gfp_mask);
79
80struct lruvec *mem_cgroup_zone_lruvec(struct zone *, struct mem_cgroup *);
81struct lruvec *mem_cgroup_page_lruvec(struct page *, struct zone *);
82
83/* For coalescing uncharge for reducing memcg' overhead*/
84extern void mem_cgroup_uncharge_start(void);
85extern void mem_cgroup_uncharge_end(void);
86
87extern void mem_cgroup_uncharge_page(struct page *page);
88extern void mem_cgroup_uncharge_cache_page(struct page *page);
89
90bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
91 struct mem_cgroup *memcg);
92bool task_in_mem_cgroup(struct task_struct *task,
93 const struct mem_cgroup *memcg);
94
95extern struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page);
96extern struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
97
98extern struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg);
99extern struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css);
100
101static inline
102bool mm_match_cgroup(const struct mm_struct *mm, const struct mem_cgroup *memcg)
103{
104 struct mem_cgroup *task_memcg;
105 bool match;
106
107 rcu_read_lock();
108 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
109 match = __mem_cgroup_same_or_subtree(memcg, task_memcg);
110 rcu_read_unlock();
111 return match;
112}
113
114extern struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg);
115
116extern void
117mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
118 struct mem_cgroup **memcgp);
119extern void mem_cgroup_end_migration(struct mem_cgroup *memcg,
120 struct page *oldpage, struct page *newpage, bool migration_ok);
121
122struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
123 struct mem_cgroup *,
124 struct mem_cgroup_reclaim_cookie *);
125void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
126
127/*
128 * For memory reclaim.
129 */
130int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec);
131int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
132unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list);
133void mem_cgroup_update_lru_size(struct lruvec *, enum lru_list, int);
134extern void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
135 struct task_struct *p);
136extern void mem_cgroup_replace_page_cache(struct page *oldpage,
137 struct page *newpage);
138
139static inline void mem_cgroup_oom_enable(void)
140{
141 WARN_ON(current->memcg_oom.may_oom);
142 current->memcg_oom.may_oom = 1;
143}
144
145static inline void mem_cgroup_oom_disable(void)
146{
147 WARN_ON(!current->memcg_oom.may_oom);
148 current->memcg_oom.may_oom = 0;
149}
150
151static inline bool task_in_memcg_oom(struct task_struct *p)
152{
153 return p->memcg_oom.memcg;
154}
155
156bool mem_cgroup_oom_synchronize(bool wait);
157
158#ifdef CONFIG_MEMCG_SWAP
159extern int do_swap_account;
160#endif
161
162static inline bool mem_cgroup_disabled(void)
163{
164 if (memory_cgrp_subsys.disabled)
165 return true;
166 return false;
167}
168
169void __mem_cgroup_begin_update_page_stat(struct page *page, bool *locked,
170 unsigned long *flags);
171
172extern atomic_t memcg_moving;
173
174static inline void mem_cgroup_begin_update_page_stat(struct page *page,
175 bool *locked, unsigned long *flags)
176{
177 if (mem_cgroup_disabled())
178 return;
179 rcu_read_lock();
180 *locked = false;
181 if (atomic_read(&memcg_moving))
182 __mem_cgroup_begin_update_page_stat(page, locked, flags);
183}
184
185void __mem_cgroup_end_update_page_stat(struct page *page,
186 unsigned long *flags);
187static inline void mem_cgroup_end_update_page_stat(struct page *page,
188 bool *locked, unsigned long *flags)
189{
190 if (mem_cgroup_disabled())
191 return;
192 if (*locked)
193 __mem_cgroup_end_update_page_stat(page, flags);
194 rcu_read_unlock();
195}
196
197void mem_cgroup_update_page_stat(struct page *page,
198 enum mem_cgroup_stat_index idx,
199 int val);
200
201static inline void mem_cgroup_inc_page_stat(struct page *page,
202 enum mem_cgroup_stat_index idx)
203{
204 mem_cgroup_update_page_stat(page, idx, 1);
205}
206
207static inline void mem_cgroup_dec_page_stat(struct page *page,
208 enum mem_cgroup_stat_index idx)
209{
210 mem_cgroup_update_page_stat(page, idx, -1);
211}
212
213unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
214 gfp_t gfp_mask,
215 unsigned long *total_scanned);
216
217void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx);
218static inline void mem_cgroup_count_vm_event(struct mm_struct *mm,
219 enum vm_event_item idx)
220{
221 if (mem_cgroup_disabled())
222 return;
223 __mem_cgroup_count_vm_event(mm, idx);
224}
225#ifdef CONFIG_TRANSPARENT_HUGEPAGE
226void mem_cgroup_split_huge_fixup(struct page *head);
227#endif
228
229#ifdef CONFIG_DEBUG_VM
230bool mem_cgroup_bad_page_check(struct page *page);
231void mem_cgroup_print_bad_page(struct page *page);
232#endif
233#else /* CONFIG_MEMCG */
234struct mem_cgroup;
235
236static inline int mem_cgroup_charge_anon(struct page *page,
237 struct mm_struct *mm, gfp_t gfp_mask)
238{
239 return 0;
240}
241
242static inline int mem_cgroup_charge_file(struct page *page,
243 struct mm_struct *mm, gfp_t gfp_mask)
244{
245 return 0;
246}
247
248static inline int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
249 struct page *page, gfp_t gfp_mask, struct mem_cgroup **memcgp)
250{
251 return 0;
252}
253
254static inline void mem_cgroup_commit_charge_swapin(struct page *page,
255 struct mem_cgroup *memcg)
256{
257}
258
259static inline void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
260{
261}
262
263static inline void mem_cgroup_uncharge_start(void)
264{
265}
266
267static inline void mem_cgroup_uncharge_end(void)
268{
269}
270
271static inline void mem_cgroup_uncharge_page(struct page *page)
272{
273}
274
275static inline void mem_cgroup_uncharge_cache_page(struct page *page)
276{
277}
278
279static inline struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
280 struct mem_cgroup *memcg)
281{
282 return &zone->lruvec;
283}
284
285static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
286 struct zone *zone)
287{
288 return &zone->lruvec;
289}
290
291static inline struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
292{
293 return NULL;
294}
295
296static inline bool mm_match_cgroup(struct mm_struct *mm,
297 struct mem_cgroup *memcg)
298{
299 return true;
300}
301
302static inline bool task_in_mem_cgroup(struct task_struct *task,
303 const struct mem_cgroup *memcg)
304{
305 return true;
306}
307
308static inline struct cgroup_subsys_state
309 *mem_cgroup_css(struct mem_cgroup *memcg)
310{
311 return NULL;
312}
313
314static inline void
315mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
316 struct mem_cgroup **memcgp)
317{
318}
319
320static inline void mem_cgroup_end_migration(struct mem_cgroup *memcg,
321 struct page *oldpage, struct page *newpage, bool migration_ok)
322{
323}
324
325static inline struct mem_cgroup *
326mem_cgroup_iter(struct mem_cgroup *root,
327 struct mem_cgroup *prev,
328 struct mem_cgroup_reclaim_cookie *reclaim)
329{
330 return NULL;
331}
332
333static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
334 struct mem_cgroup *prev)
335{
336}
337
338static inline bool mem_cgroup_disabled(void)
339{
340 return true;
341}
342
343static inline int
344mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
345{
346 return 1;
347}
348
349static inline unsigned long
350mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
351{
352 return 0;
353}
354
355static inline void
356mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
357 int increment)
358{
359}
360
361static inline void
362mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
363{
364}
365
366static inline void mem_cgroup_begin_update_page_stat(struct page *page,
367 bool *locked, unsigned long *flags)
368{
369}
370
371static inline void mem_cgroup_end_update_page_stat(struct page *page,
372 bool *locked, unsigned long *flags)
373{
374}
375
376static inline void mem_cgroup_oom_enable(void)
377{
378}
379
380static inline void mem_cgroup_oom_disable(void)
381{
382}
383
384static inline bool task_in_memcg_oom(struct task_struct *p)
385{
386 return false;
387}
388
389static inline bool mem_cgroup_oom_synchronize(bool wait)
390{
391 return false;
392}
393
394static inline void mem_cgroup_inc_page_stat(struct page *page,
395 enum mem_cgroup_stat_index idx)
396{
397}
398
399static inline void mem_cgroup_dec_page_stat(struct page *page,
400 enum mem_cgroup_stat_index idx)
401{
402}
403
404static inline
405unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
406 gfp_t gfp_mask,
407 unsigned long *total_scanned)
408{
409 return 0;
410}
411
412static inline void mem_cgroup_split_huge_fixup(struct page *head)
413{
414}
415
416static inline
417void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
418{
419}
420static inline void mem_cgroup_replace_page_cache(struct page *oldpage,
421 struct page *newpage)
422{
423}
424#endif /* CONFIG_MEMCG */
425
426#if !defined(CONFIG_MEMCG) || !defined(CONFIG_DEBUG_VM)
427static inline bool
428mem_cgroup_bad_page_check(struct page *page)
429{
430 return false;
431}
432
433static inline void
434mem_cgroup_print_bad_page(struct page *page)
435{
436}
437#endif
438
439enum {
440 UNDER_LIMIT,
441 SOFT_LIMIT,
442 OVER_LIMIT,
443};
444
445struct sock;
446#if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
447void sock_update_memcg(struct sock *sk);
448void sock_release_memcg(struct sock *sk);
449#else
450static inline void sock_update_memcg(struct sock *sk)
451{
452}
453static inline void sock_release_memcg(struct sock *sk)
454{
455}
456#endif /* CONFIG_INET && CONFIG_MEMCG_KMEM */
457
458#ifdef CONFIG_MEMCG_KMEM
459extern struct static_key memcg_kmem_enabled_key;
460
461extern int memcg_limited_groups_array_size;
462
463/*
464 * Helper macro to loop through all memcg-specific caches. Callers must still
465 * check if the cache is valid (it is either valid or NULL).
466 * the slab_mutex must be held when looping through those caches
467 */
468#define for_each_memcg_cache_index(_idx) \
469 for ((_idx) = 0; (_idx) < memcg_limited_groups_array_size; (_idx)++)
470
471static inline bool memcg_kmem_enabled(void)
472{
473 return static_key_false(&memcg_kmem_enabled_key);
474}
475
476/*
477 * In general, we'll do everything in our power to not incur in any overhead
478 * for non-memcg users for the kmem functions. Not even a function call, if we
479 * can avoid it.
480 *
481 * Therefore, we'll inline all those functions so that in the best case, we'll
482 * see that kmemcg is off for everybody and proceed quickly. If it is on,
483 * we'll still do most of the flag checking inline. We check a lot of
484 * conditions, but because they are pretty simple, they are expected to be
485 * fast.
486 */
487bool __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg,
488 int order);
489void __memcg_kmem_commit_charge(struct page *page,
490 struct mem_cgroup *memcg, int order);
491void __memcg_kmem_uncharge_pages(struct page *page, int order);
492
493int memcg_cache_id(struct mem_cgroup *memcg);
494
495char *memcg_create_cache_name(struct mem_cgroup *memcg,
496 struct kmem_cache *root_cache);
497int memcg_alloc_cache_params(struct mem_cgroup *memcg, struct kmem_cache *s,
498 struct kmem_cache *root_cache);
499void memcg_free_cache_params(struct kmem_cache *s);
500void memcg_register_cache(struct kmem_cache *s);
501void memcg_unregister_cache(struct kmem_cache *s);
502
503int memcg_update_cache_size(struct kmem_cache *s, int num_groups);
504void memcg_update_array_size(int num_groups);
505
506struct kmem_cache *
507__memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp);
508
509void mem_cgroup_destroy_cache(struct kmem_cache *cachep);
510int __kmem_cache_destroy_memcg_children(struct kmem_cache *s);
511
512/**
513 * memcg_kmem_newpage_charge: verify if a new kmem allocation is allowed.
514 * @gfp: the gfp allocation flags.
515 * @memcg: a pointer to the memcg this was charged against.
516 * @order: allocation order.
517 *
518 * returns true if the memcg where the current task belongs can hold this
519 * allocation.
520 *
521 * We return true automatically if this allocation is not to be accounted to
522 * any memcg.
523 */
524static inline bool
525memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg, int order)
526{
527 if (!memcg_kmem_enabled())
528 return true;
529
530 /*
531 * __GFP_NOFAIL allocations will move on even if charging is not
532 * possible. Therefore we don't even try, and have this allocation
533 * unaccounted. We could in theory charge it with
534 * res_counter_charge_nofail, but we hope those allocations are rare,
535 * and won't be worth the trouble.
536 */
537 if (!(gfp & __GFP_KMEMCG) || (gfp & __GFP_NOFAIL))
538 return true;
539 if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
540 return true;
541
542 /* If the test is dying, just let it go. */
543 if (unlikely(fatal_signal_pending(current)))
544 return true;
545
546 return __memcg_kmem_newpage_charge(gfp, memcg, order);
547}
548
549/**
550 * memcg_kmem_uncharge_pages: uncharge pages from memcg
551 * @page: pointer to struct page being freed
552 * @order: allocation order.
553 *
554 * there is no need to specify memcg here, since it is embedded in page_cgroup
555 */
556static inline void
557memcg_kmem_uncharge_pages(struct page *page, int order)
558{
559 if (memcg_kmem_enabled())
560 __memcg_kmem_uncharge_pages(page, order);
561}
562
563/**
564 * memcg_kmem_commit_charge: embeds correct memcg in a page
565 * @page: pointer to struct page recently allocated
566 * @memcg: the memcg structure we charged against
567 * @order: allocation order.
568 *
569 * Needs to be called after memcg_kmem_newpage_charge, regardless of success or
570 * failure of the allocation. if @page is NULL, this function will revert the
571 * charges. Otherwise, it will commit the memcg given by @memcg to the
572 * corresponding page_cgroup.
573 */
574static inline void
575memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, int order)
576{
577 if (memcg_kmem_enabled() && memcg)
578 __memcg_kmem_commit_charge(page, memcg, order);
579}
580
581/**
582 * memcg_kmem_get_cache: selects the correct per-memcg cache for allocation
583 * @cachep: the original global kmem cache
584 * @gfp: allocation flags.
585 *
586 * This function assumes that the task allocating, which determines the memcg
587 * in the page allocator, belongs to the same cgroup throughout the whole
588 * process. Misacounting can happen if the task calls memcg_kmem_get_cache()
589 * while belonging to a cgroup, and later on changes. This is considered
590 * acceptable, and should only happen upon task migration.
591 *
592 * Before the cache is created by the memcg core, there is also a possible
593 * imbalance: the task belongs to a memcg, but the cache being allocated from
594 * is the global cache, since the child cache is not yet guaranteed to be
595 * ready. This case is also fine, since in this case the GFP_KMEMCG will not be
596 * passed and the page allocator will not attempt any cgroup accounting.
597 */
598static __always_inline struct kmem_cache *
599memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
600{
601 if (!memcg_kmem_enabled())
602 return cachep;
603 if (gfp & __GFP_NOFAIL)
604 return cachep;
605 if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
606 return cachep;
607 if (unlikely(fatal_signal_pending(current)))
608 return cachep;
609
610 return __memcg_kmem_get_cache(cachep, gfp);
611}
612#else
613#define for_each_memcg_cache_index(_idx) \
614 for (; NULL; )
615
616static inline bool memcg_kmem_enabled(void)
617{
618 return false;
619}
620
621static inline bool
622memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg, int order)
623{
624 return true;
625}
626
627static inline void memcg_kmem_uncharge_pages(struct page *page, int order)
628{
629}
630
631static inline void
632memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, int order)
633{
634}
635
636static inline int memcg_cache_id(struct mem_cgroup *memcg)
637{
638 return -1;
639}
640
641static inline int memcg_alloc_cache_params(struct mem_cgroup *memcg,
642 struct kmem_cache *s, struct kmem_cache *root_cache)
643{
644 return 0;
645}
646
647static inline void memcg_free_cache_params(struct kmem_cache *s)
648{
649}
650
651static inline void memcg_register_cache(struct kmem_cache *s)
652{
653}
654
655static inline void memcg_unregister_cache(struct kmem_cache *s)
656{
657}
658
659static inline struct kmem_cache *
660memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
661{
662 return cachep;
663}
664#endif /* CONFIG_MEMCG_KMEM */
665#endif /* _LINUX_MEMCONTROL_H */
666