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 atomic_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 /* protected by memcg_oom_lock */
229 bool oom_lock;
230 int under_oom;
231
232 int swappiness;
233 /* OOM-Killer disable */
234 int oom_kill_disable;
235
236 /* memory.events */
237 struct cgroup_file events_file;
238
239 /* handle for "memory.swap.events" */
240 struct cgroup_file swap_events_file;
241
242 /* protect arrays of thresholds */
243 struct mutex thresholds_lock;
244
245 /* thresholds for memory usage. RCU-protected */
246 struct mem_cgroup_thresholds thresholds;
247
248 /* thresholds for mem+swap usage. RCU-protected */
249 struct mem_cgroup_thresholds memsw_thresholds;
250
251 /* For oom notifier event fd */
252 struct list_head oom_notify;
253
254 /*
255 * Should we move charges of a task when a task is moved into this
256 * mem_cgroup ? And what type of charges should we move ?
257 */
258 unsigned long move_charge_at_immigrate;
259 /* taken only while moving_account > 0 */
260 spinlock_t move_lock;
261 unsigned long move_lock_flags;
262
263 MEMCG_PADDING(_pad1_);
264
265 /*
266 * set > 0 if pages under this cgroup are moving to other cgroup.
267 */
268 atomic_t moving_account;
269 struct task_struct *move_lock_task;
270
271 /* memory.stat */
272 struct mem_cgroup_stat_cpu __percpu *stat_cpu;
273
274 MEMCG_PADDING(_pad2_);
275
276 atomic_long_t stat[MEMCG_NR_STAT];
277 atomic_long_t events[NR_VM_EVENT_ITEMS];
278 atomic_long_t memory_events[MEMCG_NR_MEMORY_EVENTS];
279
280 unsigned long socket_pressure;
281
282 /* Legacy tcp memory accounting */
283 bool tcpmem_active;
284 int tcpmem_pressure;
285
286#ifdef CONFIG_MEMCG_KMEM
287 /* Index in the kmem_cache->memcg_params.memcg_caches array */
288 int kmemcg_id;
289 enum memcg_kmem_state kmem_state;
290 struct list_head kmem_caches;
291#endif
292
293 int last_scanned_node;
294#if MAX_NUMNODES > 1
295 nodemask_t scan_nodes;
296 atomic_t numainfo_events;
297 atomic_t numainfo_updating;
298#endif
299
300#ifdef CONFIG_CGROUP_WRITEBACK
301 struct list_head cgwb_list;
302 struct wb_domain cgwb_domain;
303#endif
304
305 /* List of events which userspace want to receive */
306 struct list_head event_list;
307 spinlock_t event_list_lock;
308
309 struct mem_cgroup_per_node *nodeinfo[0];
310 /* WARNING: nodeinfo must be the last member here */
311};
312
313/*
314 * size of first charge trial. "32" comes from vmscan.c's magic value.
315 * TODO: maybe necessary to use big numbers in big irons.
316 */
317#define MEMCG_CHARGE_BATCH 32U
318
319extern struct mem_cgroup *root_mem_cgroup;
320
321static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
322{
323 return (memcg == root_mem_cgroup);
324}
325
326static inline bool mem_cgroup_disabled(void)
327{
328 return !cgroup_subsys_enabled(memory_cgrp_subsys);
329}
330
331enum mem_cgroup_protection mem_cgroup_protected(struct mem_cgroup *root,
332 struct mem_cgroup *memcg);
333
334int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
335 gfp_t gfp_mask, struct mem_cgroup **memcgp,
336 bool compound);
337int mem_cgroup_try_charge_delay(struct page *page, struct mm_struct *mm,
338 gfp_t gfp_mask, struct mem_cgroup **memcgp,
339 bool compound);
340void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
341 bool lrucare, bool compound);
342void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
343 bool compound);
344void mem_cgroup_uncharge(struct page *page);
345void mem_cgroup_uncharge_list(struct list_head *page_list);
346
347void mem_cgroup_migrate(struct page *oldpage, struct page *newpage);
348
349static struct mem_cgroup_per_node *
350mem_cgroup_nodeinfo(struct mem_cgroup *memcg, int nid)
351{
352 return memcg->nodeinfo[nid];
353}
354
355/**
356 * mem_cgroup_lruvec - get the lru list vector for a node or a memcg zone
357 * @node: node of the wanted lruvec
358 * @memcg: memcg of the wanted lruvec
359 *
360 * Returns the lru list vector holding pages for a given @node or a given
361 * @memcg and @zone. This can be the node lruvec, if the memory controller
362 * is disabled.
363 */
364static inline struct lruvec *mem_cgroup_lruvec(struct pglist_data *pgdat,
365 struct mem_cgroup *memcg)
366{
367 struct mem_cgroup_per_node *mz;
368 struct lruvec *lruvec;
369
370 if (mem_cgroup_disabled()) {
371 lruvec = node_lruvec(pgdat);
372 goto out;
373 }
374
375 mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
376 lruvec = &mz->lruvec;
377out:
378 /*
379 * Since a node can be onlined after the mem_cgroup was created,
380 * we have to be prepared to initialize lruvec->pgdat here;
381 * and if offlined then reonlined, we need to reinitialize it.
382 */
383 if (unlikely(lruvec->pgdat != pgdat))
384 lruvec->pgdat = pgdat;
385 return lruvec;
386}
387
388struct lruvec *mem_cgroup_page_lruvec(struct page *, struct pglist_data *);
389
390bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg);
391struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
392
393struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm);
394
395struct mem_cgroup *get_mem_cgroup_from_page(struct page *page);
396
397static inline
398struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
399 return css ? container_of(css, struct mem_cgroup, css) : NULL;
400}
401
402static inline void mem_cgroup_put(struct mem_cgroup *memcg)
403{
404 if (memcg)
405 css_put(&memcg->css);
406}
407
408#define mem_cgroup_from_counter(counter, member) \
409 container_of(counter, struct mem_cgroup, member)
410
411struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
412 struct mem_cgroup *,
413 struct mem_cgroup_reclaim_cookie *);
414void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
415int mem_cgroup_scan_tasks(struct mem_cgroup *,
416 int (*)(struct task_struct *, void *), void *);
417
418static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
419{
420 if (mem_cgroup_disabled())
421 return 0;
422
423 return memcg->id.id;
424}
425struct mem_cgroup *mem_cgroup_from_id(unsigned short id);
426
427static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec)
428{
429 struct mem_cgroup_per_node *mz;
430
431 if (mem_cgroup_disabled())
432 return NULL;
433
434 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
435 return mz->memcg;
436}
437
438/**
439 * parent_mem_cgroup - find the accounting parent of a memcg
440 * @memcg: memcg whose parent to find
441 *
442 * Returns the parent memcg, or NULL if this is the root or the memory
443 * controller is in legacy no-hierarchy mode.
444 */
445static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
446{
447 if (!memcg->memory.parent)
448 return NULL;
449 return mem_cgroup_from_counter(memcg->memory.parent, memory);
450}
451
452static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
453 struct mem_cgroup *root)
454{
455 if (root == memcg)
456 return true;
457 if (!root->use_hierarchy)
458 return false;
459 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
460}
461
462static inline bool mm_match_cgroup(struct mm_struct *mm,
463 struct mem_cgroup *memcg)
464{
465 struct mem_cgroup *task_memcg;
466 bool match = false;
467
468 rcu_read_lock();
469 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
470 if (task_memcg)
471 match = mem_cgroup_is_descendant(task_memcg, memcg);
472 rcu_read_unlock();
473 return match;
474}
475
476struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
477ino_t page_cgroup_ino(struct page *page);
478
479static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
480{
481 if (mem_cgroup_disabled())
482 return true;
483 return !!(memcg->css.flags & CSS_ONLINE);
484}
485
486/*
487 * For memory reclaim.
488 */
489int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
490
491void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
492 int zid, int nr_pages);
493
494unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
495 int nid, unsigned int lru_mask);
496
497static inline
498unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
499{
500 struct mem_cgroup_per_node *mz;
501 unsigned long nr_pages = 0;
502 int zid;
503
504 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
505 for (zid = 0; zid < MAX_NR_ZONES; zid++)
506 nr_pages += mz->lru_zone_size[zid][lru];
507 return nr_pages;
508}
509
510static inline
511unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
512 enum lru_list lru, int zone_idx)
513{
514 struct mem_cgroup_per_node *mz;
515
516 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
517 return mz->lru_zone_size[zone_idx][lru];
518}
519
520void mem_cgroup_handle_over_high(void);
521
522unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg);
523
524void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
525 struct task_struct *p);
526
527static inline void mem_cgroup_enter_user_fault(void)
528{
529 WARN_ON(current->in_user_fault);
530 current->in_user_fault = 1;
531}
532
533static inline void mem_cgroup_exit_user_fault(void)
534{
535 WARN_ON(!current->in_user_fault);
536 current->in_user_fault = 0;
537}
538
539static inline bool task_in_memcg_oom(struct task_struct *p)
540{
541 return p->memcg_in_oom;
542}
543
544bool mem_cgroup_oom_synchronize(bool wait);
545
546#ifdef CONFIG_MEMCG_SWAP
547extern int do_swap_account;
548#endif
549
550struct mem_cgroup *lock_page_memcg(struct page *page);
551void __unlock_page_memcg(struct mem_cgroup *memcg);
552void unlock_page_memcg(struct page *page);
553
554/* idx can be of type enum memcg_stat_item or node_stat_item */
555static inline unsigned long memcg_page_state(struct mem_cgroup *memcg,
556 int idx)
557{
558 long x = atomic_long_read(&memcg->stat[idx]);
559#ifdef CONFIG_SMP
560 if (x < 0)
561 x = 0;
562#endif
563 return x;
564}
565
566/* idx can be of type enum memcg_stat_item or node_stat_item */
567static inline void __mod_memcg_state(struct mem_cgroup *memcg,
568 int idx, int val)
569{
570 long x;
571
572 if (mem_cgroup_disabled())
573 return;
574
575 x = val + __this_cpu_read(memcg->stat_cpu->count[idx]);
576 if (unlikely(abs(x) > MEMCG_CHARGE_BATCH)) {
577 atomic_long_add(x, &memcg->stat[idx]);
578 x = 0;
579 }
580 __this_cpu_write(memcg->stat_cpu->count[idx], x);
581}
582
583/* idx can be of type enum memcg_stat_item or node_stat_item */
584static inline void mod_memcg_state(struct mem_cgroup *memcg,
585 int idx, int val)
586{
587 unsigned long flags;
588
589 local_irq_save(flags);
590 __mod_memcg_state(memcg, idx, val);
591 local_irq_restore(flags);
592}
593
594/**
595 * mod_memcg_page_state - update page state statistics
596 * @page: the page
597 * @idx: page state item to account
598 * @val: number of pages (positive or negative)
599 *
600 * The @page must be locked or the caller must use lock_page_memcg()
601 * to prevent double accounting when the page is concurrently being
602 * moved to another memcg:
603 *
604 * lock_page(page) or lock_page_memcg(page)
605 * if (TestClearPageState(page))
606 * mod_memcg_page_state(page, state, -1);
607 * unlock_page(page) or unlock_page_memcg(page)
608 *
609 * Kernel pages are an exception to this, since they'll never move.
610 */
611static inline void __mod_memcg_page_state(struct page *page,
612 int idx, int val)
613{
614 if (page->mem_cgroup)
615 __mod_memcg_state(page->mem_cgroup, idx, val);
616}
617
618static inline void mod_memcg_page_state(struct page *page,
619 int idx, int val)
620{
621 if (page->mem_cgroup)
622 mod_memcg_state(page->mem_cgroup, idx, val);
623}
624
625static inline unsigned long lruvec_page_state(struct lruvec *lruvec,
626 enum node_stat_item idx)
627{
628 struct mem_cgroup_per_node *pn;
629 long x;
630
631 if (mem_cgroup_disabled())
632 return node_page_state(lruvec_pgdat(lruvec), idx);
633
634 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
635 x = atomic_long_read(&pn->lruvec_stat[idx]);
636#ifdef CONFIG_SMP
637 if (x < 0)
638 x = 0;
639#endif
640 return x;
641}
642
643static inline void __mod_lruvec_state(struct lruvec *lruvec,
644 enum node_stat_item idx, int val)
645{
646 struct mem_cgroup_per_node *pn;
647 long x;
648
649 /* Update node */
650 __mod_node_page_state(lruvec_pgdat(lruvec), idx, val);
651
652 if (mem_cgroup_disabled())
653 return;
654
655 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
656
657 /* Update memcg */
658 __mod_memcg_state(pn->memcg, idx, val);
659
660 /* Update lruvec */
661 x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]);
662 if (unlikely(abs(x) > MEMCG_CHARGE_BATCH)) {
663 atomic_long_add(x, &pn->lruvec_stat[idx]);
664 x = 0;
665 }
666 __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x);
667}
668
669static inline void mod_lruvec_state(struct lruvec *lruvec,
670 enum node_stat_item idx, int val)
671{
672 unsigned long flags;
673
674 local_irq_save(flags);
675 __mod_lruvec_state(lruvec, idx, val);
676 local_irq_restore(flags);
677}
678
679static inline void __mod_lruvec_page_state(struct page *page,
680 enum node_stat_item idx, int val)
681{
682 pg_data_t *pgdat = page_pgdat(page);
683 struct lruvec *lruvec;
684
685 /* Untracked pages have no memcg, no lruvec. Update only the node */
686 if (!page->mem_cgroup) {
687 __mod_node_page_state(pgdat, idx, val);
688 return;
689 }
690
691 lruvec = mem_cgroup_lruvec(pgdat, page->mem_cgroup);
692 __mod_lruvec_state(lruvec, idx, val);
693}
694
695static inline void mod_lruvec_page_state(struct page *page,
696 enum node_stat_item idx, int val)
697{
698 unsigned long flags;
699
700 local_irq_save(flags);
701 __mod_lruvec_page_state(page, idx, val);
702 local_irq_restore(flags);
703}
704
705unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
706 gfp_t gfp_mask,
707 unsigned long *total_scanned);
708
709static inline void __count_memcg_events(struct mem_cgroup *memcg,
710 enum vm_event_item idx,
711 unsigned long count)
712{
713 unsigned long x;
714
715 if (mem_cgroup_disabled())
716 return;
717
718 x = count + __this_cpu_read(memcg->stat_cpu->events[idx]);
719 if (unlikely(x > MEMCG_CHARGE_BATCH)) {
720 atomic_long_add(x, &memcg->events[idx]);
721 x = 0;
722 }
723 __this_cpu_write(memcg->stat_cpu->events[idx], x);
724}
725
726static inline void count_memcg_events(struct mem_cgroup *memcg,
727 enum vm_event_item idx,
728 unsigned long count)
729{
730 unsigned long flags;
731
732 local_irq_save(flags);
733 __count_memcg_events(memcg, idx, count);
734 local_irq_restore(flags);
735}
736
737static inline void count_memcg_page_event(struct page *page,
738 enum vm_event_item idx)
739{
740 if (page->mem_cgroup)
741 count_memcg_events(page->mem_cgroup, idx, 1);
742}
743
744static inline void count_memcg_event_mm(struct mm_struct *mm,
745 enum vm_event_item idx)
746{
747 struct mem_cgroup *memcg;
748
749 if (mem_cgroup_disabled())
750 return;
751
752 rcu_read_lock();
753 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
754 if (likely(memcg))
755 count_memcg_events(memcg, idx, 1);
756 rcu_read_unlock();
757}
758
759static inline void memcg_memory_event(struct mem_cgroup *memcg,
760 enum memcg_memory_event event)
761{
762 atomic_long_inc(&memcg->memory_events[event]);
763 cgroup_file_notify(&memcg->events_file);
764}
765
766static inline void memcg_memory_event_mm(struct mm_struct *mm,
767 enum memcg_memory_event event)
768{
769 struct mem_cgroup *memcg;
770
771 if (mem_cgroup_disabled())
772 return;
773
774 rcu_read_lock();
775 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
776 if (likely(memcg))
777 memcg_memory_event(memcg, event);
778 rcu_read_unlock();
779}
780
781#ifdef CONFIG_TRANSPARENT_HUGEPAGE
782void mem_cgroup_split_huge_fixup(struct page *head);
783#endif
784
785#else /* CONFIG_MEMCG */
786
787#define MEM_CGROUP_ID_SHIFT 0
788#define MEM_CGROUP_ID_MAX 0
789
790struct mem_cgroup;
791
792static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
793{
794 return true;
795}
796
797static inline bool mem_cgroup_disabled(void)
798{
799 return true;
800}
801
802static inline void memcg_memory_event(struct mem_cgroup *memcg,
803 enum memcg_memory_event event)
804{
805}
806
807static inline void memcg_memory_event_mm(struct mm_struct *mm,
808 enum memcg_memory_event event)
809{
810}
811
812static inline enum mem_cgroup_protection mem_cgroup_protected(
813 struct mem_cgroup *root, struct mem_cgroup *memcg)
814{
815 return MEMCG_PROT_NONE;
816}
817
818static inline int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
819 gfp_t gfp_mask,
820 struct mem_cgroup **memcgp,
821 bool compound)
822{
823 *memcgp = NULL;
824 return 0;
825}
826
827static inline int mem_cgroup_try_charge_delay(struct page *page,
828 struct mm_struct *mm,
829 gfp_t gfp_mask,
830 struct mem_cgroup **memcgp,
831 bool compound)
832{
833 *memcgp = NULL;
834 return 0;
835}
836
837static inline void mem_cgroup_commit_charge(struct page *page,
838 struct mem_cgroup *memcg,
839 bool lrucare, bool compound)
840{
841}
842
843static inline void mem_cgroup_cancel_charge(struct page *page,
844 struct mem_cgroup *memcg,
845 bool compound)
846{
847}
848
849static inline void mem_cgroup_uncharge(struct page *page)
850{
851}
852
853static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
854{
855}
856
857static inline void mem_cgroup_migrate(struct page *old, struct page *new)
858{
859}
860
861static inline struct lruvec *mem_cgroup_lruvec(struct pglist_data *pgdat,
862 struct mem_cgroup *memcg)
863{
864 return node_lruvec(pgdat);
865}
866
867static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
868 struct pglist_data *pgdat)
869{
870 return &pgdat->lruvec;
871}
872
873static inline bool mm_match_cgroup(struct mm_struct *mm,
874 struct mem_cgroup *memcg)
875{
876 return true;
877}
878
879static inline bool task_in_mem_cgroup(struct task_struct *task,
880 const struct mem_cgroup *memcg)
881{
882 return true;
883}
884
885static inline struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
886{
887 return NULL;
888}
889
890static inline struct mem_cgroup *get_mem_cgroup_from_page(struct page *page)
891{
892 return NULL;
893}
894
895static inline void mem_cgroup_put(struct mem_cgroup *memcg)
896{
897}
898
899static inline struct mem_cgroup *
900mem_cgroup_iter(struct mem_cgroup *root,
901 struct mem_cgroup *prev,
902 struct mem_cgroup_reclaim_cookie *reclaim)
903{
904 return NULL;
905}
906
907static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
908 struct mem_cgroup *prev)
909{
910}
911
912static inline int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
913 int (*fn)(struct task_struct *, void *), void *arg)
914{
915 return 0;
916}
917
918static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
919{
920 return 0;
921}
922
923static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
924{
925 WARN_ON_ONCE(id);
926 /* XXX: This should always return root_mem_cgroup */
927 return NULL;
928}
929
930static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec)
931{
932 return NULL;
933}
934
935static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
936{
937 return true;
938}
939
940static inline unsigned long
941mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
942{
943 return 0;
944}
945static inline
946unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
947 enum lru_list lru, int zone_idx)
948{
949 return 0;
950}
951
952static inline unsigned long
953mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
954 int nid, unsigned int lru_mask)
955{
956 return 0;
957}
958
959static inline unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg)
960{
961 return 0;
962}
963
964static inline void
965mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
966{
967}
968
969static inline struct mem_cgroup *lock_page_memcg(struct page *page)
970{
971 return NULL;
972}
973
974static inline void __unlock_page_memcg(struct mem_cgroup *memcg)
975{
976}
977
978static inline void unlock_page_memcg(struct page *page)
979{
980}
981
982static inline void mem_cgroup_handle_over_high(void)
983{
984}
985
986static inline void mem_cgroup_enter_user_fault(void)
987{
988}
989
990static inline void mem_cgroup_exit_user_fault(void)
991{
992}
993
994static inline bool task_in_memcg_oom(struct task_struct *p)
995{
996 return false;
997}
998
999static inline bool mem_cgroup_oom_synchronize(bool wait)
1000{
1001 return false;
1002}
1003
1004static inline unsigned long memcg_page_state(struct mem_cgroup *memcg,
1005 int idx)
1006{
1007 return 0;
1008}
1009
1010static inline void __mod_memcg_state(struct mem_cgroup *memcg,
1011 int idx,
1012 int nr)
1013{
1014}
1015
1016static inline void mod_memcg_state(struct mem_cgroup *memcg,
1017 int idx,
1018 int nr)
1019{
1020}
1021
1022static inline void __mod_memcg_page_state(struct page *page,
1023 int idx,
1024 int nr)
1025{
1026}
1027
1028static inline void mod_memcg_page_state(struct page *page,
1029 int idx,
1030 int nr)
1031{
1032}
1033
1034static inline unsigned long lruvec_page_state(struct lruvec *lruvec,
1035 enum node_stat_item idx)
1036{
1037 return node_page_state(lruvec_pgdat(lruvec), idx);
1038}
1039
1040static inline void __mod_lruvec_state(struct lruvec *lruvec,
1041 enum node_stat_item idx, int val)
1042{
1043 __mod_node_page_state(lruvec_pgdat(lruvec), idx, val);
1044}
1045
1046static inline void mod_lruvec_state(struct lruvec *lruvec,
1047 enum node_stat_item idx, int val)
1048{
1049 mod_node_page_state(lruvec_pgdat(lruvec), idx, val);
1050}
1051
1052static inline void __mod_lruvec_page_state(struct page *page,
1053 enum node_stat_item idx, int val)
1054{
1055 __mod_node_page_state(page_pgdat(page), idx, val);
1056}
1057
1058static inline void mod_lruvec_page_state(struct page *page,
1059 enum node_stat_item idx, int val)
1060{
1061 mod_node_page_state(page_pgdat(page), idx, val);
1062}
1063
1064static inline
1065unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
1066 gfp_t gfp_mask,
1067 unsigned long *total_scanned)
1068{
1069 return 0;
1070}
1071
1072static inline void mem_cgroup_split_huge_fixup(struct page *head)
1073{
1074}
1075
1076static inline void count_memcg_events(struct mem_cgroup *memcg,
1077 enum vm_event_item idx,
1078 unsigned long count)
1079{
1080}
1081
1082static inline void count_memcg_page_event(struct page *page,
1083 int idx)
1084{
1085}
1086
1087static inline
1088void count_memcg_event_mm(struct mm_struct *mm, enum vm_event_item idx)
1089{
1090}
1091#endif /* CONFIG_MEMCG */
1092
1093/* idx can be of type enum memcg_stat_item or node_stat_item */
1094static inline void __inc_memcg_state(struct mem_cgroup *memcg,
1095 int idx)
1096{
1097 __mod_memcg_state(memcg, idx, 1);
1098}
1099
1100/* idx can be of type enum memcg_stat_item or node_stat_item */
1101static inline void __dec_memcg_state(struct mem_cgroup *memcg,
1102 int idx)
1103{
1104 __mod_memcg_state(memcg, idx, -1);
1105}
1106
1107/* idx can be of type enum memcg_stat_item or node_stat_item */
1108static inline void __inc_memcg_page_state(struct page *page,
1109 int idx)
1110{
1111 __mod_memcg_page_state(page, idx, 1);
1112}
1113
1114/* idx can be of type enum memcg_stat_item or node_stat_item */
1115static inline void __dec_memcg_page_state(struct page *page,
1116 int idx)
1117{
1118 __mod_memcg_page_state(page, idx, -1);
1119}
1120
1121static inline void __inc_lruvec_state(struct lruvec *lruvec,
1122 enum node_stat_item idx)
1123{
1124 __mod_lruvec_state(lruvec, idx, 1);
1125}
1126
1127static inline void __dec_lruvec_state(struct lruvec *lruvec,
1128 enum node_stat_item idx)
1129{
1130 __mod_lruvec_state(lruvec, idx, -1);
1131}
1132
1133static inline void __inc_lruvec_page_state(struct page *page,
1134 enum node_stat_item idx)
1135{
1136 __mod_lruvec_page_state(page, idx, 1);
1137}
1138
1139static inline void __dec_lruvec_page_state(struct page *page,
1140 enum node_stat_item idx)
1141{
1142 __mod_lruvec_page_state(page, idx, -1);
1143}
1144
1145/* idx can be of type enum memcg_stat_item or node_stat_item */
1146static inline void inc_memcg_state(struct mem_cgroup *memcg,
1147 int idx)
1148{
1149 mod_memcg_state(memcg, idx, 1);
1150}
1151
1152/* idx can be of type enum memcg_stat_item or node_stat_item */
1153static inline void dec_memcg_state(struct mem_cgroup *memcg,
1154 int idx)
1155{
1156 mod_memcg_state(memcg, idx, -1);
1157}
1158
1159/* idx can be of type enum memcg_stat_item or node_stat_item */
1160static inline void inc_memcg_page_state(struct page *page,
1161 int idx)
1162{
1163 mod_memcg_page_state(page, idx, 1);
1164}
1165
1166/* idx can be of type enum memcg_stat_item or node_stat_item */
1167static inline void dec_memcg_page_state(struct page *page,
1168 int idx)
1169{
1170 mod_memcg_page_state(page, idx, -1);
1171}
1172
1173static inline void inc_lruvec_state(struct lruvec *lruvec,
1174 enum node_stat_item idx)
1175{
1176 mod_lruvec_state(lruvec, idx, 1);
1177}
1178
1179static inline void dec_lruvec_state(struct lruvec *lruvec,
1180 enum node_stat_item idx)
1181{
1182 mod_lruvec_state(lruvec, idx, -1);
1183}
1184
1185static inline void inc_lruvec_page_state(struct page *page,
1186 enum node_stat_item idx)
1187{
1188 mod_lruvec_page_state(page, idx, 1);
1189}
1190
1191static inline void dec_lruvec_page_state(struct page *page,
1192 enum node_stat_item idx)
1193{
1194 mod_lruvec_page_state(page, idx, -1);
1195}
1196
1197#ifdef CONFIG_CGROUP_WRITEBACK
1198
1199struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
1200void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
1201 unsigned long *pheadroom, unsigned long *pdirty,
1202 unsigned long *pwriteback);
1203
1204#else /* CONFIG_CGROUP_WRITEBACK */
1205
1206static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
1207{
1208 return NULL;
1209}
1210
1211static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
1212 unsigned long *pfilepages,
1213 unsigned long *pheadroom,
1214 unsigned long *pdirty,
1215 unsigned long *pwriteback)
1216{
1217}
1218
1219#endif /* CONFIG_CGROUP_WRITEBACK */
1220
1221struct sock;
1222bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
1223void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
1224#ifdef CONFIG_MEMCG
1225extern struct static_key_false memcg_sockets_enabled_key;
1226#define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_enabled_key)
1227void mem_cgroup_sk_alloc(struct sock *sk);
1228void mem_cgroup_sk_free(struct sock *sk);
1229static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
1230{
1231 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_pressure)
1232 return true;
1233 do {
1234 if (time_before(jiffies, memcg->socket_pressure))
1235 return true;
1236 } while ((memcg = parent_mem_cgroup(memcg)));
1237 return false;
1238}
1239#else
1240#define mem_cgroup_sockets_enabled 0
1241static inline void mem_cgroup_sk_alloc(struct sock *sk) { };
1242static inline void mem_cgroup_sk_free(struct sock *sk) { };
1243static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
1244{
1245 return false;
1246}
1247#endif
1248
1249struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep);
1250void memcg_kmem_put_cache(struct kmem_cache *cachep);
1251int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
1252 struct mem_cgroup *memcg);
1253int memcg_kmem_charge(struct page *page, gfp_t gfp, int order);
1254void memcg_kmem_uncharge(struct page *page, int order);
1255
1256#ifdef CONFIG_MEMCG_KMEM
1257extern struct static_key_false memcg_kmem_enabled_key;
1258extern struct workqueue_struct *memcg_kmem_cache_wq;
1259
1260extern int memcg_nr_cache_ids;
1261void memcg_get_cache_ids(void);
1262void memcg_put_cache_ids(void);
1263
1264/*
1265 * Helper macro to loop through all memcg-specific caches. Callers must still
1266 * check if the cache is valid (it is either valid or NULL).
1267 * the slab_mutex must be held when looping through those caches
1268 */
1269#define for_each_memcg_cache_index(_idx) \
1270 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
1271
1272static inline bool memcg_kmem_enabled(void)
1273{
1274 return static_branch_unlikely(&memcg_kmem_enabled_key);
1275}
1276
1277/*
1278 * helper for accessing a memcg's index. It will be used as an index in the
1279 * child cache array in kmem_cache, and also to derive its name. This function
1280 * will return -1 when this is not a kmem-limited memcg.
1281 */
1282static inline int memcg_cache_id(struct mem_cgroup *memcg)
1283{
1284 return memcg ? memcg->kmemcg_id : -1;
1285}
1286
1287extern int memcg_expand_shrinker_maps(int new_id);
1288
1289extern void memcg_set_shrinker_bit(struct mem_cgroup *memcg,
1290 int nid, int shrinker_id);
1291#else
1292#define for_each_memcg_cache_index(_idx) \
1293 for (; NULL; )
1294
1295static inline bool memcg_kmem_enabled(void)
1296{
1297 return false;
1298}
1299
1300static inline int memcg_cache_id(struct mem_cgroup *memcg)
1301{
1302 return -1;
1303}
1304
1305static inline void memcg_get_cache_ids(void)
1306{
1307}
1308
1309static inline void memcg_put_cache_ids(void)
1310{
1311}
1312
1313static inline void memcg_set_shrinker_bit(struct mem_cgroup *memcg,
1314 int nid, int shrinker_id) { }
1315#endif /* CONFIG_MEMCG_KMEM */
1316
1317#endif /* _LINUX_MEMCONTROL_H */
1318