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/mm.h>
30#include <linux/vmstat.h>
31#include <linux/writeback.h>
32#include <linux/page-flags.h>
33
34struct mem_cgroup;
35struct page;
36struct mm_struct;
37struct kmem_cache;
38
39/* Cgroup-specific page state, on top of universal node page state */
40enum memcg_stat_item {
41 MEMCG_CACHE = NR_VM_NODE_STAT_ITEMS,
42 MEMCG_RSS,
43 MEMCG_RSS_HUGE,
44 MEMCG_SWAP,
45 MEMCG_SOCK,
46 /* XXX: why are these zone and not node counters? */
47 MEMCG_KERNEL_STACK_KB,
48 MEMCG_NR_STAT,
49};
50
51enum memcg_memory_event {
52 MEMCG_LOW,
53 MEMCG_HIGH,
54 MEMCG_MAX,
55 MEMCG_OOM,
56 MEMCG_OOM_KILL,
57 MEMCG_SWAP_MAX,
58 MEMCG_SWAP_FAIL,
59 MEMCG_NR_MEMORY_EVENTS,
60};
61
62enum mem_cgroup_protection {
63 MEMCG_PROT_NONE,
64 MEMCG_PROT_LOW,
65 MEMCG_PROT_MIN,
66};
67
68struct mem_cgroup_reclaim_cookie {
69 pg_data_t *pgdat;
70 int priority;
71 unsigned int generation;
72};
73
74#ifdef CONFIG_MEMCG
75
76#define MEM_CGROUP_ID_SHIFT 16
77#define MEM_CGROUP_ID_MAX USHRT_MAX
78
79struct mem_cgroup_id {
80 int id;
81 refcount_t ref;
82};
83
84/*
85 * Per memcg event counter is incremented at every pagein/pageout. With THP,
86 * it will be incremated by the number of pages. This counter is used for
87 * for trigger some periodic events. This is straightforward and better
88 * than using jiffies etc. to handle periodic memcg event.
89 */
90enum mem_cgroup_events_target {
91 MEM_CGROUP_TARGET_THRESH,
92 MEM_CGROUP_TARGET_SOFTLIMIT,
93 MEM_CGROUP_TARGET_NUMAINFO,
94 MEM_CGROUP_NTARGETS,
95};
96
97struct mem_cgroup_stat_cpu {
98 long count[MEMCG_NR_STAT];
99 unsigned long events[NR_VM_EVENT_ITEMS];
100 unsigned long nr_page_events;
101 unsigned long targets[MEM_CGROUP_NTARGETS];
102};
103
104struct mem_cgroup_reclaim_iter {
105 struct mem_cgroup *position;
106 /* scan generation, increased every round-trip */
107 unsigned int generation;
108};
109
110struct lruvec_stat {
111 long count[NR_VM_NODE_STAT_ITEMS];
112};
113
114/*
115 * Bitmap of shrinker::id corresponding to memcg-aware shrinkers,
116 * which have elements charged to this memcg.
117 */
118struct memcg_shrinker_map {
119 struct rcu_head rcu;
120 unsigned long map[0];
121};
122
123/*
124 * per-zone information in memory controller.
125 */
126struct mem_cgroup_per_node {
127 struct lruvec lruvec;
128
129 struct lruvec_stat __percpu *lruvec_stat_cpu;
130 atomic_long_t lruvec_stat[NR_VM_NODE_STAT_ITEMS];
131
132 unsigned long lru_zone_size[MAX_NR_ZONES][NR_LRU_LISTS];
133
134 struct mem_cgroup_reclaim_iter iter[DEF_PRIORITY + 1];
135
136#ifdef CONFIG_MEMCG_KMEM
137 struct memcg_shrinker_map __rcu *shrinker_map;
138#endif
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 bool congested; /* memcg has many dirty pages */
144 /* backed by a congested BDI */
145
146 struct mem_cgroup *memcg; /* Back pointer, we cannot */
147 /* use container_of */
148};
149
150struct mem_cgroup_threshold {
151 struct eventfd_ctx *eventfd;
152 unsigned long threshold;
153};
154
155/* For threshold */
156struct mem_cgroup_threshold_ary {
157 /* An array index points to threshold just below or equal to usage. */
158 int current_threshold;
159 /* Size of entries[] */
160 unsigned int size;
161 /* Array of thresholds */
162 struct mem_cgroup_threshold entries[0];
163};
164
165struct mem_cgroup_thresholds {
166 /* Primary thresholds array */
167 struct mem_cgroup_threshold_ary *primary;
168 /*
169 * Spare threshold array.
170 * This is needed to make mem_cgroup_unregister_event() "never fail".
171 * It must be able to store at least primary->size - 1 entries.
172 */
173 struct mem_cgroup_threshold_ary *spare;
174};
175
176enum memcg_kmem_state {
177 KMEM_NONE,
178 KMEM_ALLOCATED,
179 KMEM_ONLINE,
180};
181
182#if defined(CONFIG_SMP)
183struct memcg_padding {
184 char x[0];
185} ____cacheline_internodealigned_in_smp;
186#define MEMCG_PADDING(name) struct memcg_padding name;
187#else
188#define MEMCG_PADDING(name)
189#endif
190
191/*
192 * The memory controller data structure. The memory controller controls both
193 * page cache and RSS per cgroup. We would eventually like to provide
194 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
195 * to help the administrator determine what knobs to tune.
196 */
197struct mem_cgroup {
198 struct cgroup_subsys_state css;
199
200 /* Private memcg ID. Used to ID objects that outlive the cgroup */
201 struct mem_cgroup_id id;
202
203 /* Accounted resources */
204 struct page_counter memory;
205 struct page_counter swap;
206
207 /* Legacy consumer-oriented counters */
208 struct page_counter memsw;
209 struct page_counter kmem;
210 struct page_counter tcpmem;
211
212 /* Upper bound of normal memory consumption range */
213 unsigned long high;
214
215 /* Range enforcement for interrupt charges */
216 struct work_struct high_work;
217
218 unsigned long soft_limit;
219
220 /* vmpressure notifications */
221 struct vmpressure vmpressure;
222
223 /*
224 * Should the accounting and control be hierarchical, per subtree?
225 */
226 bool use_hierarchy;
227
228 /*
229 * Should the OOM killer kill all belonging tasks, had it kill one?
230 */
231 bool oom_group;
232
233 /* protected by memcg_oom_lock */
234 bool oom_lock;
235 int under_oom;
236
237 int swappiness;
238 /* OOM-Killer disable */
239 int oom_kill_disable;
240
241 /* memory.events */
242 struct cgroup_file events_file;
243
244 /* handle for "memory.swap.events" */
245 struct cgroup_file swap_events_file;
246
247 /* protect arrays of thresholds */
248 struct mutex thresholds_lock;
249
250 /* thresholds for memory usage. RCU-protected */
251 struct mem_cgroup_thresholds thresholds;
252
253 /* thresholds for mem+swap usage. RCU-protected */
254 struct mem_cgroup_thresholds memsw_thresholds;
255
256 /* For oom notifier event fd */
257 struct list_head oom_notify;
258
259 /*
260 * Should we move charges of a task when a task is moved into this
261 * mem_cgroup ? And what type of charges should we move ?
262 */
263 unsigned long move_charge_at_immigrate;
264 /* taken only while moving_account > 0 */
265 spinlock_t move_lock;
266 unsigned long move_lock_flags;
267
268 MEMCG_PADDING(_pad1_);
269
270 /*
271 * set > 0 if pages under this cgroup are moving to other cgroup.
272 */
273 atomic_t moving_account;
274 struct task_struct *move_lock_task;
275
276 /* memory.stat */
277 struct mem_cgroup_stat_cpu __percpu *stat_cpu;
278
279 MEMCG_PADDING(_pad2_);
280
281 atomic_long_t stat[MEMCG_NR_STAT];
282 atomic_long_t events[NR_VM_EVENT_ITEMS];
283 atomic_long_t memory_events[MEMCG_NR_MEMORY_EVENTS];
284
285 unsigned long socket_pressure;
286
287 /* Legacy tcp memory accounting */
288 bool tcpmem_active;
289 int tcpmem_pressure;
290
291#ifdef CONFIG_MEMCG_KMEM
292 /* Index in the kmem_cache->memcg_params.memcg_caches array */
293 int kmemcg_id;
294 enum memcg_kmem_state kmem_state;
295 struct list_head kmem_caches;
296#endif
297
298 int last_scanned_node;
299#if MAX_NUMNODES > 1
300 nodemask_t scan_nodes;
301 atomic_t numainfo_events;
302 atomic_t numainfo_updating;
303#endif
304
305#ifdef CONFIG_CGROUP_WRITEBACK
306 struct list_head cgwb_list;
307 struct wb_domain cgwb_domain;
308#endif
309
310 /* List of events which userspace want to receive */
311 struct list_head event_list;
312 spinlock_t event_list_lock;
313
314 struct mem_cgroup_per_node *nodeinfo[0];
315 /* WARNING: nodeinfo must be the last member here */
316};
317
318/*
319 * size of first charge trial. "32" comes from vmscan.c's magic value.
320 * TODO: maybe necessary to use big numbers in big irons.
321 */
322#define MEMCG_CHARGE_BATCH 32U
323
324extern struct mem_cgroup *root_mem_cgroup;
325
326static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
327{
328 return (memcg == root_mem_cgroup);
329}
330
331static inline bool mem_cgroup_disabled(void)
332{
333 return !cgroup_subsys_enabled(memory_cgrp_subsys);
334}
335
336enum mem_cgroup_protection mem_cgroup_protected(struct mem_cgroup *root,
337 struct mem_cgroup *memcg);
338
339int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
340 gfp_t gfp_mask, struct mem_cgroup **memcgp,
341 bool compound);
342int mem_cgroup_try_charge_delay(struct page *page, struct mm_struct *mm,
343 gfp_t gfp_mask, struct mem_cgroup **memcgp,
344 bool compound);
345void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
346 bool lrucare, bool compound);
347void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
348 bool compound);
349void mem_cgroup_uncharge(struct page *page);
350void mem_cgroup_uncharge_list(struct list_head *page_list);
351
352void mem_cgroup_migrate(struct page *oldpage, struct page *newpage);
353
354static struct mem_cgroup_per_node *
355mem_cgroup_nodeinfo(struct mem_cgroup *memcg, int nid)
356{
357 return memcg->nodeinfo[nid];
358}
359
360/**
361 * mem_cgroup_lruvec - get the lru list vector for a node or a memcg zone
362 * @node: node of the wanted lruvec
363 * @memcg: memcg of the wanted lruvec
364 *
365 * Returns the lru list vector holding pages for a given @node or a given
366 * @memcg and @zone. This can be the node lruvec, if the memory controller
367 * is disabled.
368 */
369static inline struct lruvec *mem_cgroup_lruvec(struct pglist_data *pgdat,
370 struct mem_cgroup *memcg)
371{
372 struct mem_cgroup_per_node *mz;
373 struct lruvec *lruvec;
374
375 if (mem_cgroup_disabled()) {
376 lruvec = node_lruvec(pgdat);
377 goto out;
378 }
379
380 mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
381 lruvec = &mz->lruvec;
382out:
383 /*
384 * Since a node can be onlined after the mem_cgroup was created,
385 * we have to be prepared to initialize lruvec->pgdat here;
386 * and if offlined then reonlined, we need to reinitialize it.
387 */
388 if (unlikely(lruvec->pgdat != pgdat))
389 lruvec->pgdat = pgdat;
390 return lruvec;
391}
392
393struct lruvec *mem_cgroup_page_lruvec(struct page *, struct pglist_data *);
394
395bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg);
396struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
397
398struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm);
399
400struct mem_cgroup *get_mem_cgroup_from_page(struct page *page);
401
402static inline
403struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
404 return css ? container_of(css, struct mem_cgroup, css) : NULL;
405}
406
407static inline void mem_cgroup_put(struct mem_cgroup *memcg)
408{
409 if (memcg)
410 css_put(&memcg->css);
411}
412
413#define mem_cgroup_from_counter(counter, member) \
414 container_of(counter, struct mem_cgroup, member)
415
416struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
417 struct mem_cgroup *,
418 struct mem_cgroup_reclaim_cookie *);
419void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
420int mem_cgroup_scan_tasks(struct mem_cgroup *,
421 int (*)(struct task_struct *, void *), void *);
422
423static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
424{
425 if (mem_cgroup_disabled())
426 return 0;
427
428 return memcg->id.id;
429}
430struct mem_cgroup *mem_cgroup_from_id(unsigned short id);
431
432static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m)
433{
434 return mem_cgroup_from_css(seq_css(m));
435}
436
437static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec)
438{
439 struct mem_cgroup_per_node *mz;
440
441 if (mem_cgroup_disabled())
442 return NULL;
443
444 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
445 return mz->memcg;
446}
447
448/**
449 * parent_mem_cgroup - find the accounting parent of a memcg
450 * @memcg: memcg whose parent to find
451 *
452 * Returns the parent memcg, or NULL if this is the root or the memory
453 * controller is in legacy no-hierarchy mode.
454 */
455static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
456{
457 if (!memcg->memory.parent)
458 return NULL;
459 return mem_cgroup_from_counter(memcg->memory.parent, memory);
460}
461
462static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
463 struct mem_cgroup *root)
464{
465 if (root == memcg)
466 return true;
467 if (!root->use_hierarchy)
468 return false;
469 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
470}
471
472static inline bool mm_match_cgroup(struct mm_struct *mm,
473 struct mem_cgroup *memcg)
474{
475 struct mem_cgroup *task_memcg;
476 bool match = false;
477
478 rcu_read_lock();
479 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
480 if (task_memcg)
481 match = mem_cgroup_is_descendant(task_memcg, memcg);
482 rcu_read_unlock();
483 return match;
484}
485
486struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
487ino_t page_cgroup_ino(struct page *page);
488
489static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
490{
491 if (mem_cgroup_disabled())
492 return true;
493 return !!(memcg->css.flags & CSS_ONLINE);
494}
495
496/*
497 * For memory reclaim.
498 */
499int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
500
501void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
502 int zid, int nr_pages);
503
504unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
505 int nid, unsigned int lru_mask);
506
507static inline
508unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
509{
510 struct mem_cgroup_per_node *mz;
511 unsigned long nr_pages = 0;
512 int zid;
513
514 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
515 for (zid = 0; zid < MAX_NR_ZONES; zid++)
516 nr_pages += mz->lru_zone_size[zid][lru];
517 return nr_pages;
518}
519
520static inline
521unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
522 enum lru_list lru, int zone_idx)
523{
524 struct mem_cgroup_per_node *mz;
525
526 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
527 return mz->lru_zone_size[zone_idx][lru];
528}
529
530void mem_cgroup_handle_over_high(void);
531
532unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg);
533
534void mem_cgroup_print_oom_context(struct mem_cgroup *memcg,
535 struct task_struct *p);
536
537void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg);
538
539static inline void mem_cgroup_enter_user_fault(void)
540{
541 WARN_ON(current->in_user_fault);
542 current->in_user_fault = 1;
543}
544
545static inline void mem_cgroup_exit_user_fault(void)
546{
547 WARN_ON(!current->in_user_fault);
548 current->in_user_fault = 0;
549}
550
551static inline bool task_in_memcg_oom(struct task_struct *p)
552{
553 return p->memcg_in_oom;
554}
555
556bool mem_cgroup_oom_synchronize(bool wait);
557struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim,
558 struct mem_cgroup *oom_domain);
559void mem_cgroup_print_oom_group(struct mem_cgroup *memcg);
560
561#ifdef CONFIG_MEMCG_SWAP
562extern int do_swap_account;
563#endif
564
565struct mem_cgroup *lock_page_memcg(struct page *page);
566void __unlock_page_memcg(struct mem_cgroup *memcg);
567void unlock_page_memcg(struct page *page);
568
569/* idx can be of type enum memcg_stat_item or node_stat_item */
570static inline unsigned long memcg_page_state(struct mem_cgroup *memcg,
571 int idx)
572{
573 long x = atomic_long_read(&memcg->stat[idx]);
574#ifdef CONFIG_SMP
575 if (x < 0)
576 x = 0;
577#endif
578 return x;
579}
580
581/* idx can be of type enum memcg_stat_item or node_stat_item */
582static inline void __mod_memcg_state(struct mem_cgroup *memcg,
583 int idx, int val)
584{
585 long x;
586
587 if (mem_cgroup_disabled())
588 return;
589
590 x = val + __this_cpu_read(memcg->stat_cpu->count[idx]);
591 if (unlikely(abs(x) > MEMCG_CHARGE_BATCH)) {
592 atomic_long_add(x, &memcg->stat[idx]);
593 x = 0;
594 }
595 __this_cpu_write(memcg->stat_cpu->count[idx], x);
596}
597
598/* idx can be of type enum memcg_stat_item or node_stat_item */
599static inline void mod_memcg_state(struct mem_cgroup *memcg,
600 int idx, int val)
601{
602 unsigned long flags;
603
604 local_irq_save(flags);
605 __mod_memcg_state(memcg, idx, val);
606 local_irq_restore(flags);
607}
608
609/**
610 * mod_memcg_page_state - update page state statistics
611 * @page: the page
612 * @idx: page state item to account
613 * @val: number of pages (positive or negative)
614 *
615 * The @page must be locked or the caller must use lock_page_memcg()
616 * to prevent double accounting when the page is concurrently being
617 * moved to another memcg:
618 *
619 * lock_page(page) or lock_page_memcg(page)
620 * if (TestClearPageState(page))
621 * mod_memcg_page_state(page, state, -1);
622 * unlock_page(page) or unlock_page_memcg(page)
623 *
624 * Kernel pages are an exception to this, since they'll never move.
625 */
626static inline void __mod_memcg_page_state(struct page *page,
627 int idx, int val)
628{
629 if (page->mem_cgroup)
630 __mod_memcg_state(page->mem_cgroup, idx, val);
631}
632
633static inline void mod_memcg_page_state(struct page *page,
634 int idx, int val)
635{
636 if (page->mem_cgroup)
637 mod_memcg_state(page->mem_cgroup, idx, val);
638}
639
640static inline unsigned long lruvec_page_state(struct lruvec *lruvec,
641 enum node_stat_item idx)
642{
643 struct mem_cgroup_per_node *pn;
644 long x;
645
646 if (mem_cgroup_disabled())
647 return node_page_state(lruvec_pgdat(lruvec), idx);
648
649 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
650 x = atomic_long_read(&pn->lruvec_stat[idx]);
651#ifdef CONFIG_SMP
652 if (x < 0)
653 x = 0;
654#endif
655 return x;
656}
657
658static inline void __mod_lruvec_state(struct lruvec *lruvec,
659 enum node_stat_item idx, int val)
660{
661 struct mem_cgroup_per_node *pn;
662 long x;
663
664 /* Update node */
665 __mod_node_page_state(lruvec_pgdat(lruvec), idx, val);
666
667 if (mem_cgroup_disabled())
668 return;
669
670 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
671
672 /* Update memcg */
673 __mod_memcg_state(pn->memcg, idx, val);
674
675 /* Update lruvec */
676 x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]);
677 if (unlikely(abs(x) > MEMCG_CHARGE_BATCH)) {
678 atomic_long_add(x, &pn->lruvec_stat[idx]);
679 x = 0;
680 }
681 __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x);
682}
683
684static inline void mod_lruvec_state(struct lruvec *lruvec,
685 enum node_stat_item idx, int val)
686{
687 unsigned long flags;
688
689 local_irq_save(flags);
690 __mod_lruvec_state(lruvec, idx, val);
691 local_irq_restore(flags);
692}
693
694static inline void __mod_lruvec_page_state(struct page *page,
695 enum node_stat_item idx, int val)
696{
697 pg_data_t *pgdat = page_pgdat(page);
698 struct lruvec *lruvec;
699
700 /* Untracked pages have no memcg, no lruvec. Update only the node */
701 if (!page->mem_cgroup) {
702 __mod_node_page_state(pgdat, idx, val);
703 return;
704 }
705
706 lruvec = mem_cgroup_lruvec(pgdat, page->mem_cgroup);
707 __mod_lruvec_state(lruvec, idx, val);
708}
709
710static inline void mod_lruvec_page_state(struct page *page,
711 enum node_stat_item idx, int val)
712{
713 unsigned long flags;
714
715 local_irq_save(flags);
716 __mod_lruvec_page_state(page, idx, val);
717 local_irq_restore(flags);
718}
719
720unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
721 gfp_t gfp_mask,
722 unsigned long *total_scanned);
723
724static inline void __count_memcg_events(struct mem_cgroup *memcg,
725 enum vm_event_item idx,
726 unsigned long count)
727{
728 unsigned long x;
729
730 if (mem_cgroup_disabled())
731 return;
732
733 x = count + __this_cpu_read(memcg->stat_cpu->events[idx]);
734 if (unlikely(x > MEMCG_CHARGE_BATCH)) {
735 atomic_long_add(x, &memcg->events[idx]);
736 x = 0;
737 }
738 __this_cpu_write(memcg->stat_cpu->events[idx], x);
739}
740
741static inline void count_memcg_events(struct mem_cgroup *memcg,
742 enum vm_event_item idx,
743 unsigned long count)
744{
745 unsigned long flags;
746
747 local_irq_save(flags);
748 __count_memcg_events(memcg, idx, count);
749 local_irq_restore(flags);
750}
751
752static inline void count_memcg_page_event(struct page *page,
753 enum vm_event_item idx)
754{
755 if (page->mem_cgroup)
756 count_memcg_events(page->mem_cgroup, idx, 1);
757}
758
759static inline void count_memcg_event_mm(struct mm_struct *mm,
760 enum vm_event_item idx)
761{
762 struct mem_cgroup *memcg;
763
764 if (mem_cgroup_disabled())
765 return;
766
767 rcu_read_lock();
768 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
769 if (likely(memcg))
770 count_memcg_events(memcg, idx, 1);
771 rcu_read_unlock();
772}
773
774static inline void memcg_memory_event(struct mem_cgroup *memcg,
775 enum memcg_memory_event event)
776{
777 atomic_long_inc(&memcg->memory_events[event]);
778 cgroup_file_notify(&memcg->events_file);
779}
780
781static inline void memcg_memory_event_mm(struct mm_struct *mm,
782 enum memcg_memory_event event)
783{
784 struct mem_cgroup *memcg;
785
786 if (mem_cgroup_disabled())
787 return;
788
789 rcu_read_lock();
790 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
791 if (likely(memcg))
792 memcg_memory_event(memcg, event);
793 rcu_read_unlock();
794}
795
796#ifdef CONFIG_TRANSPARENT_HUGEPAGE
797void mem_cgroup_split_huge_fixup(struct page *head);
798#endif
799
800#else /* CONFIG_MEMCG */
801
802#define MEM_CGROUP_ID_SHIFT 0
803#define MEM_CGROUP_ID_MAX 0
804
805struct mem_cgroup;
806
807static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
808{
809 return true;
810}
811
812static inline bool mem_cgroup_disabled(void)
813{
814 return true;
815}
816
817static inline void memcg_memory_event(struct mem_cgroup *memcg,
818 enum memcg_memory_event event)
819{
820}
821
822static inline void memcg_memory_event_mm(struct mm_struct *mm,
823 enum memcg_memory_event event)
824{
825}
826
827static inline enum mem_cgroup_protection mem_cgroup_protected(
828 struct mem_cgroup *root, struct mem_cgroup *memcg)
829{
830 return MEMCG_PROT_NONE;
831}
832
833static inline int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
834 gfp_t gfp_mask,
835 struct mem_cgroup **memcgp,
836 bool compound)
837{
838 *memcgp = NULL;
839 return 0;
840}
841
842static inline int mem_cgroup_try_charge_delay(struct page *page,
843 struct mm_struct *mm,
844 gfp_t gfp_mask,
845 struct mem_cgroup **memcgp,
846 bool compound)
847{
848 *memcgp = NULL;
849 return 0;
850}
851
852static inline void mem_cgroup_commit_charge(struct page *page,
853 struct mem_cgroup *memcg,
854 bool lrucare, bool compound)
855{
856}
857
858static inline void mem_cgroup_cancel_charge(struct page *page,
859 struct mem_cgroup *memcg,
860 bool compound)
861{
862}
863
864static inline void mem_cgroup_uncharge(struct page *page)
865{
866}
867
868static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
869{
870}
871
872static inline void mem_cgroup_migrate(struct page *old, struct page *new)
873{
874}
875
876static inline struct lruvec *mem_cgroup_lruvec(struct pglist_data *pgdat,
877 struct mem_cgroup *memcg)
878{
879 return node_lruvec(pgdat);
880}
881
882static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
883 struct pglist_data *pgdat)
884{
885 return &pgdat->lruvec;
886}
887
888static inline bool mm_match_cgroup(struct mm_struct *mm,
889 struct mem_cgroup *memcg)
890{
891 return true;
892}
893
894static inline bool task_in_mem_cgroup(struct task_struct *task,
895 const struct mem_cgroup *memcg)
896{
897 return true;
898}
899
900static inline struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
901{
902 return NULL;
903}
904
905static inline struct mem_cgroup *get_mem_cgroup_from_page(struct page *page)
906{
907 return NULL;
908}
909
910static inline void mem_cgroup_put(struct mem_cgroup *memcg)
911{
912}
913
914static inline struct mem_cgroup *
915mem_cgroup_iter(struct mem_cgroup *root,
916 struct mem_cgroup *prev,
917 struct mem_cgroup_reclaim_cookie *reclaim)
918{
919 return NULL;
920}
921
922static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
923 struct mem_cgroup *prev)
924{
925}
926
927static inline int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
928 int (*fn)(struct task_struct *, void *), void *arg)
929{
930 return 0;
931}
932
933static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
934{
935 return 0;
936}
937
938static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
939{
940 WARN_ON_ONCE(id);
941 /* XXX: This should always return root_mem_cgroup */
942 return NULL;
943}
944
945static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m)
946{
947 return NULL;
948}
949
950static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec)
951{
952 return NULL;
953}
954
955static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
956{
957 return true;
958}
959
960static inline unsigned long
961mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
962{
963 return 0;
964}
965static inline
966unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
967 enum lru_list lru, int zone_idx)
968{
969 return 0;
970}
971
972static inline unsigned long
973mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
974 int nid, unsigned int lru_mask)
975{
976 return 0;
977}
978
979static inline unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg)
980{
981 return 0;
982}
983
984static inline void
985mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p)
986{
987}
988
989static inline void
990mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg)
991{
992}
993
994static inline struct mem_cgroup *lock_page_memcg(struct page *page)
995{
996 return NULL;
997}
998
999static inline void __unlock_page_memcg(struct mem_cgroup *memcg)
1000{
1001}
1002
1003static inline void unlock_page_memcg(struct page *page)
1004{
1005}
1006
1007static inline void mem_cgroup_handle_over_high(void)
1008{
1009}
1010
1011static inline void mem_cgroup_enter_user_fault(void)
1012{
1013}
1014
1015static inline void mem_cgroup_exit_user_fault(void)
1016{
1017}
1018
1019static inline bool task_in_memcg_oom(struct task_struct *p)
1020{
1021 return false;
1022}
1023
1024static inline bool mem_cgroup_oom_synchronize(bool wait)
1025{
1026 return false;
1027}
1028
1029static inline struct mem_cgroup *mem_cgroup_get_oom_group(
1030 struct task_struct *victim, struct mem_cgroup *oom_domain)
1031{
1032 return NULL;
1033}
1034
1035static inline void mem_cgroup_print_oom_group(struct mem_cgroup *memcg)
1036{
1037}
1038
1039static inline unsigned long memcg_page_state(struct mem_cgroup *memcg,
1040 int idx)
1041{
1042 return 0;
1043}
1044
1045static inline void __mod_memcg_state(struct mem_cgroup *memcg,
1046 int idx,
1047 int nr)
1048{
1049}
1050
1051static inline void mod_memcg_state(struct mem_cgroup *memcg,
1052 int idx,
1053 int nr)
1054{
1055}
1056
1057static inline void __mod_memcg_page_state(struct page *page,
1058 int idx,
1059 int nr)
1060{
1061}
1062
1063static inline void mod_memcg_page_state(struct page *page,
1064 int idx,
1065 int nr)
1066{
1067}
1068
1069static inline unsigned long lruvec_page_state(struct lruvec *lruvec,
1070 enum node_stat_item idx)
1071{
1072 return node_page_state(lruvec_pgdat(lruvec), idx);
1073}
1074
1075static inline void __mod_lruvec_state(struct lruvec *lruvec,
1076 enum node_stat_item idx, int val)
1077{
1078 __mod_node_page_state(lruvec_pgdat(lruvec), idx, val);
1079}
1080
1081static inline void mod_lruvec_state(struct lruvec *lruvec,
1082 enum node_stat_item idx, int val)
1083{
1084 mod_node_page_state(lruvec_pgdat(lruvec), idx, val);
1085}
1086
1087static inline void __mod_lruvec_page_state(struct page *page,
1088 enum node_stat_item idx, int val)
1089{
1090 __mod_node_page_state(page_pgdat(page), idx, val);
1091}
1092
1093static inline void mod_lruvec_page_state(struct page *page,
1094 enum node_stat_item idx, int val)
1095{
1096 mod_node_page_state(page_pgdat(page), idx, val);
1097}
1098
1099static inline
1100unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
1101 gfp_t gfp_mask,
1102 unsigned long *total_scanned)
1103{
1104 return 0;
1105}
1106
1107static inline void mem_cgroup_split_huge_fixup(struct page *head)
1108{
1109}
1110
1111static inline void count_memcg_events(struct mem_cgroup *memcg,
1112 enum vm_event_item idx,
1113 unsigned long count)
1114{
1115}
1116
1117static inline void count_memcg_page_event(struct page *page,
1118 int idx)
1119{
1120}
1121
1122static inline
1123void count_memcg_event_mm(struct mm_struct *mm, enum vm_event_item idx)
1124{
1125}
1126#endif /* CONFIG_MEMCG */
1127
1128/* idx can be of type enum memcg_stat_item or node_stat_item */
1129static inline void __inc_memcg_state(struct mem_cgroup *memcg,
1130 int idx)
1131{
1132 __mod_memcg_state(memcg, idx, 1);
1133}
1134
1135/* idx can be of type enum memcg_stat_item or node_stat_item */
1136static inline void __dec_memcg_state(struct mem_cgroup *memcg,
1137 int idx)
1138{
1139 __mod_memcg_state(memcg, idx, -1);
1140}
1141
1142/* idx can be of type enum memcg_stat_item or node_stat_item */
1143static inline void __inc_memcg_page_state(struct page *page,
1144 int idx)
1145{
1146 __mod_memcg_page_state(page, idx, 1);
1147}
1148
1149/* idx can be of type enum memcg_stat_item or node_stat_item */
1150static inline void __dec_memcg_page_state(struct page *page,
1151 int idx)
1152{
1153 __mod_memcg_page_state(page, idx, -1);
1154}
1155
1156static inline void __inc_lruvec_state(struct lruvec *lruvec,
1157 enum node_stat_item idx)
1158{
1159 __mod_lruvec_state(lruvec, idx, 1);
1160}
1161
1162static inline void __dec_lruvec_state(struct lruvec *lruvec,
1163 enum node_stat_item idx)
1164{
1165 __mod_lruvec_state(lruvec, idx, -1);
1166}
1167
1168static inline void __inc_lruvec_page_state(struct page *page,
1169 enum node_stat_item idx)
1170{
1171 __mod_lruvec_page_state(page, idx, 1);
1172}
1173
1174static inline void __dec_lruvec_page_state(struct page *page,
1175 enum node_stat_item idx)
1176{
1177 __mod_lruvec_page_state(page, idx, -1);
1178}
1179
1180/* idx can be of type enum memcg_stat_item or node_stat_item */
1181static inline void inc_memcg_state(struct mem_cgroup *memcg,
1182 int idx)
1183{
1184 mod_memcg_state(memcg, idx, 1);
1185}
1186
1187/* idx can be of type enum memcg_stat_item or node_stat_item */
1188static inline void dec_memcg_state(struct mem_cgroup *memcg,
1189 int idx)
1190{
1191 mod_memcg_state(memcg, idx, -1);
1192}
1193
1194/* idx can be of type enum memcg_stat_item or node_stat_item */
1195static inline void inc_memcg_page_state(struct page *page,
1196 int idx)
1197{
1198 mod_memcg_page_state(page, idx, 1);
1199}
1200
1201/* idx can be of type enum memcg_stat_item or node_stat_item */
1202static inline void dec_memcg_page_state(struct page *page,
1203 int idx)
1204{
1205 mod_memcg_page_state(page, idx, -1);
1206}
1207
1208static inline void inc_lruvec_state(struct lruvec *lruvec,
1209 enum node_stat_item idx)
1210{
1211 mod_lruvec_state(lruvec, idx, 1);
1212}
1213
1214static inline void dec_lruvec_state(struct lruvec *lruvec,
1215 enum node_stat_item idx)
1216{
1217 mod_lruvec_state(lruvec, idx, -1);
1218}
1219
1220static inline void inc_lruvec_page_state(struct page *page,
1221 enum node_stat_item idx)
1222{
1223 mod_lruvec_page_state(page, idx, 1);
1224}
1225
1226static inline void dec_lruvec_page_state(struct page *page,
1227 enum node_stat_item idx)
1228{
1229 mod_lruvec_page_state(page, idx, -1);
1230}
1231
1232#ifdef CONFIG_CGROUP_WRITEBACK
1233
1234struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
1235void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
1236 unsigned long *pheadroom, unsigned long *pdirty,
1237 unsigned long *pwriteback);
1238
1239#else /* CONFIG_CGROUP_WRITEBACK */
1240
1241static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
1242{
1243 return NULL;
1244}
1245
1246static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
1247 unsigned long *pfilepages,
1248 unsigned long *pheadroom,
1249 unsigned long *pdirty,
1250 unsigned long *pwriteback)
1251{
1252}
1253
1254#endif /* CONFIG_CGROUP_WRITEBACK */
1255
1256struct sock;
1257bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
1258void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
1259#ifdef CONFIG_MEMCG
1260extern struct static_key_false memcg_sockets_enabled_key;
1261#define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_enabled_key)
1262void mem_cgroup_sk_alloc(struct sock *sk);
1263void mem_cgroup_sk_free(struct sock *sk);
1264static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
1265{
1266 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_pressure)
1267 return true;
1268 do {
1269 if (time_before(jiffies, memcg->socket_pressure))
1270 return true;
1271 } while ((memcg = parent_mem_cgroup(memcg)));
1272 return false;
1273}
1274#else
1275#define mem_cgroup_sockets_enabled 0
1276static inline void mem_cgroup_sk_alloc(struct sock *sk) { };
1277static inline void mem_cgroup_sk_free(struct sock *sk) { };
1278static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
1279{
1280 return false;
1281}
1282#endif
1283
1284struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep);
1285void memcg_kmem_put_cache(struct kmem_cache *cachep);
1286
1287#ifdef CONFIG_MEMCG_KMEM
1288int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order);
1289void __memcg_kmem_uncharge(struct page *page, int order);
1290int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
1291 struct mem_cgroup *memcg);
1292
1293extern struct static_key_false memcg_kmem_enabled_key;
1294extern struct workqueue_struct *memcg_kmem_cache_wq;
1295
1296extern int memcg_nr_cache_ids;
1297void memcg_get_cache_ids(void);
1298void memcg_put_cache_ids(void);
1299
1300/*
1301 * Helper macro to loop through all memcg-specific caches. Callers must still
1302 * check if the cache is valid (it is either valid or NULL).
1303 * the slab_mutex must be held when looping through those caches
1304 */
1305#define for_each_memcg_cache_index(_idx) \
1306 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
1307
1308static inline bool memcg_kmem_enabled(void)
1309{
1310 return static_branch_unlikely(&memcg_kmem_enabled_key);
1311}
1312
1313static inline int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
1314{
1315 if (memcg_kmem_enabled())
1316 return __memcg_kmem_charge(page, gfp, order);
1317 return 0;
1318}
1319
1320static inline void memcg_kmem_uncharge(struct page *page, int order)
1321{
1322 if (memcg_kmem_enabled())
1323 __memcg_kmem_uncharge(page, order);
1324}
1325
1326static inline int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp,
1327 int order, struct mem_cgroup *memcg)
1328{
1329 if (memcg_kmem_enabled())
1330 return __memcg_kmem_charge_memcg(page, gfp, order, memcg);
1331 return 0;
1332}
1333/*
1334 * helper for accessing a memcg's index. It will be used as an index in the
1335 * child cache array in kmem_cache, and also to derive its name. This function
1336 * will return -1 when this is not a kmem-limited memcg.
1337 */
1338static inline int memcg_cache_id(struct mem_cgroup *memcg)
1339{
1340 return memcg ? memcg->kmemcg_id : -1;
1341}
1342
1343extern int memcg_expand_shrinker_maps(int new_id);
1344
1345extern void memcg_set_shrinker_bit(struct mem_cgroup *memcg,
1346 int nid, int shrinker_id);
1347#else
1348
1349static inline int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
1350{
1351 return 0;
1352}
1353
1354static inline void memcg_kmem_uncharge(struct page *page, int order)
1355{
1356}
1357
1358static inline int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
1359{
1360 return 0;
1361}
1362
1363static inline void __memcg_kmem_uncharge(struct page *page, int order)
1364{
1365}
1366
1367#define for_each_memcg_cache_index(_idx) \
1368 for (; NULL; )
1369
1370static inline bool memcg_kmem_enabled(void)
1371{
1372 return false;
1373}
1374
1375static inline int memcg_cache_id(struct mem_cgroup *memcg)
1376{
1377 return -1;
1378}
1379
1380static inline void memcg_get_cache_ids(void)
1381{
1382}
1383
1384static inline void memcg_put_cache_ids(void)
1385{
1386}
1387
1388static inline void memcg_set_shrinker_bit(struct mem_cgroup *memcg,
1389 int nid, int shrinker_id) { }
1390#endif /* CONFIG_MEMCG_KMEM */
1391
1392#endif /* _LINUX_MEMCONTROL_H */
1393