1 /* memcontrol.h - Memory Controller
3 * Copyright IBM Corporation, 2007
6 * Copyright 2007 OpenVZ SWsoft Inc
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.
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.
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>
30 #include <linux/vmstat.h>
31 #include <linux/writeback.h>
32 #include <linux/page-flags.h>
39 /* Cgroup-specific page state, on top of universal node page state */
40 enum memcg_stat_item {
41 MEMCG_CACHE = NR_VM_NODE_STAT_ITEMS,
46 /* XXX: why are these zone and not node counters? */
47 MEMCG_KERNEL_STACK_KB,
51 enum memcg_memory_event {
59 MEMCG_NR_MEMORY_EVENTS,
62 enum mem_cgroup_protection {
68 struct mem_cgroup_reclaim_cookie {
71 unsigned int generation;
76 #define MEM_CGROUP_ID_SHIFT 16
77 #define MEM_CGROUP_ID_MAX USHRT_MAX
79 struct mem_cgroup_id {
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.
90 enum mem_cgroup_events_target {
91 MEM_CGROUP_TARGET_THRESH,
92 MEM_CGROUP_TARGET_SOFTLIMIT,
93 MEM_CGROUP_TARGET_NUMAINFO,
97 struct 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];
104 struct mem_cgroup_reclaim_iter {
105 struct mem_cgroup *position;
106 /* scan generation, increased every round-trip */
107 unsigned int generation;
111 long count[NR_VM_NODE_STAT_ITEMS];
115 * Bitmap of shrinker::id corresponding to memcg-aware shrinkers,
116 * which have elements charged to this memcg.
118 struct memcg_shrinker_map {
120 unsigned long map[0];
124 * per-zone information in memory controller.
126 struct mem_cgroup_per_node {
127 struct lruvec lruvec;
129 struct lruvec_stat __percpu *lruvec_stat_cpu;
130 atomic_long_t lruvec_stat[NR_VM_NODE_STAT_ITEMS];
132 unsigned long lru_zone_size[MAX_NR_ZONES][NR_LRU_LISTS];
134 struct mem_cgroup_reclaim_iter iter[DEF_PRIORITY + 1];
136 #ifdef CONFIG_MEMCG_KMEM
137 struct memcg_shrinker_map __rcu *shrinker_map;
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*/
143 bool congested; /* memcg has many dirty pages */
144 /* backed by a congested BDI */
146 struct mem_cgroup *memcg; /* Back pointer, we cannot */
147 /* use container_of */
150 struct mem_cgroup_threshold {
151 struct eventfd_ctx *eventfd;
152 unsigned long threshold;
156 struct 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[] */
161 /* Array of thresholds */
162 struct mem_cgroup_threshold entries[0];
165 struct mem_cgroup_thresholds {
166 /* Primary thresholds array */
167 struct mem_cgroup_threshold_ary *primary;
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.
173 struct mem_cgroup_threshold_ary *spare;
176 enum memcg_kmem_state {
182 #if defined(CONFIG_SMP)
183 struct memcg_padding {
185 } ____cacheline_internodealigned_in_smp;
186 #define MEMCG_PADDING(name) struct memcg_padding name;
188 #define MEMCG_PADDING(name)
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.
198 struct cgroup_subsys_state css;
200 /* Private memcg ID. Used to ID objects that outlive the cgroup */
201 struct mem_cgroup_id id;
203 /* Accounted resources */
204 struct page_counter memory;
205 struct page_counter swap;
207 /* Legacy consumer-oriented counters */
208 struct page_counter memsw;
209 struct page_counter kmem;
210 struct page_counter tcpmem;
212 /* Upper bound of normal memory consumption range */
215 /* Range enforcement for interrupt charges */
216 struct work_struct high_work;
218 unsigned long soft_limit;
220 /* vmpressure notifications */
221 struct vmpressure vmpressure;
224 * Should the accounting and control be hierarchical, per subtree?
229 * Should the OOM killer kill all belonging tasks, had it kill one?
233 /* protected by memcg_oom_lock */
238 /* OOM-Killer disable */
239 int oom_kill_disable;
242 struct cgroup_file events_file;
244 /* handle for "memory.swap.events" */
245 struct cgroup_file swap_events_file;
247 /* protect arrays of thresholds */
248 struct mutex thresholds_lock;
250 /* thresholds for memory usage. RCU-protected */
251 struct mem_cgroup_thresholds thresholds;
253 /* thresholds for mem+swap usage. RCU-protected */
254 struct mem_cgroup_thresholds memsw_thresholds;
256 /* For oom notifier event fd */
257 struct list_head oom_notify;
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 ?
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;
268 MEMCG_PADDING(_pad1_);
271 * set > 0 if pages under this cgroup are moving to other cgroup.
273 atomic_t moving_account;
274 struct task_struct *move_lock_task;
277 struct mem_cgroup_stat_cpu __percpu *stat_cpu;
279 MEMCG_PADDING(_pad2_);
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];
285 unsigned long socket_pressure;
287 /* Legacy tcp memory accounting */
291 #ifdef CONFIG_MEMCG_KMEM
292 /* Index in the kmem_cache->memcg_params.memcg_caches array */
294 enum memcg_kmem_state kmem_state;
295 struct list_head kmem_caches;
298 int last_scanned_node;
300 nodemask_t scan_nodes;
301 atomic_t numainfo_events;
302 atomic_t numainfo_updating;
305 #ifdef CONFIG_CGROUP_WRITEBACK
306 struct list_head cgwb_list;
307 struct wb_domain cgwb_domain;
310 /* List of events which userspace want to receive */
311 struct list_head event_list;
312 spinlock_t event_list_lock;
314 struct mem_cgroup_per_node *nodeinfo[0];
315 /* WARNING: nodeinfo must be the last member here */
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.
322 #define MEMCG_CHARGE_BATCH 32U
324 extern struct mem_cgroup *root_mem_cgroup;
326 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
328 return (memcg == root_mem_cgroup);
331 static inline bool mem_cgroup_disabled(void)
333 return !cgroup_subsys_enabled(memory_cgrp_subsys);
336 enum mem_cgroup_protection mem_cgroup_protected(struct mem_cgroup *root,
337 struct mem_cgroup *memcg);
339 int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
340 gfp_t gfp_mask, struct mem_cgroup **memcgp,
342 int mem_cgroup_try_charge_delay(struct page *page, struct mm_struct *mm,
343 gfp_t gfp_mask, struct mem_cgroup **memcgp,
345 void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
346 bool lrucare, bool compound);
347 void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
349 void mem_cgroup_uncharge(struct page *page);
350 void mem_cgroup_uncharge_list(struct list_head *page_list);
352 void mem_cgroup_migrate(struct page *oldpage, struct page *newpage);
354 static struct mem_cgroup_per_node *
355 mem_cgroup_nodeinfo(struct mem_cgroup *memcg, int nid)
357 return memcg->nodeinfo[nid];
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
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
369 static inline struct lruvec *mem_cgroup_lruvec(struct pglist_data *pgdat,
370 struct mem_cgroup *memcg)
372 struct mem_cgroup_per_node *mz;
373 struct lruvec *lruvec;
375 if (mem_cgroup_disabled()) {
376 lruvec = node_lruvec(pgdat);
380 mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
381 lruvec = &mz->lruvec;
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.
388 if (unlikely(lruvec->pgdat != pgdat))
389 lruvec->pgdat = pgdat;
393 struct lruvec *mem_cgroup_page_lruvec(struct page *, struct pglist_data *);
395 bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg);
396 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
398 struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm);
400 struct mem_cgroup *get_mem_cgroup_from_page(struct page *page);
403 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
404 return css ? container_of(css, struct mem_cgroup, css) : NULL;
407 static inline void mem_cgroup_put(struct mem_cgroup *memcg)
410 css_put(&memcg->css);
413 #define mem_cgroup_from_counter(counter, member) \
414 container_of(counter, struct mem_cgroup, member)
416 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
418 struct mem_cgroup_reclaim_cookie *);
419 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
420 int mem_cgroup_scan_tasks(struct mem_cgroup *,
421 int (*)(struct task_struct *, void *), void *);
423 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
425 if (mem_cgroup_disabled())
430 struct mem_cgroup *mem_cgroup_from_id(unsigned short id);
432 static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m)
434 return mem_cgroup_from_css(seq_css(m));
437 static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec)
439 struct mem_cgroup_per_node *mz;
441 if (mem_cgroup_disabled())
444 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
449 * parent_mem_cgroup - find the accounting parent of a memcg
450 * @memcg: memcg whose parent to find
452 * Returns the parent memcg, or NULL if this is the root or the memory
453 * controller is in legacy no-hierarchy mode.
455 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
457 if (!memcg->memory.parent)
459 return mem_cgroup_from_counter(memcg->memory.parent, memory);
462 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
463 struct mem_cgroup *root)
467 if (!root->use_hierarchy)
469 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
472 static inline bool mm_match_cgroup(struct mm_struct *mm,
473 struct mem_cgroup *memcg)
475 struct mem_cgroup *task_memcg;
479 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
481 match = mem_cgroup_is_descendant(task_memcg, memcg);
486 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
487 ino_t page_cgroup_ino(struct page *page);
489 static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
491 if (mem_cgroup_disabled())
493 return !!(memcg->css.flags & CSS_ONLINE);
497 * For memory reclaim.
499 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
501 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
502 int zid, int nr_pages);
504 unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
505 int nid, unsigned int lru_mask);
508 unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
510 struct mem_cgroup_per_node *mz;
511 unsigned long nr_pages = 0;
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];
521 unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
522 enum lru_list lru, int zone_idx)
524 struct mem_cgroup_per_node *mz;
526 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
527 return mz->lru_zone_size[zone_idx][lru];
530 void mem_cgroup_handle_over_high(void);
532 unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg);
534 void mem_cgroup_print_oom_context(struct mem_cgroup *memcg,
535 struct task_struct *p);
537 void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg);
539 static inline void mem_cgroup_enter_user_fault(void)
541 WARN_ON(current->in_user_fault);
542 current->in_user_fault = 1;
545 static inline void mem_cgroup_exit_user_fault(void)
547 WARN_ON(!current->in_user_fault);
548 current->in_user_fault = 0;
551 static inline bool task_in_memcg_oom(struct task_struct *p)
553 return p->memcg_in_oom;
556 bool mem_cgroup_oom_synchronize(bool wait);
557 struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim,
558 struct mem_cgroup *oom_domain);
559 void mem_cgroup_print_oom_group(struct mem_cgroup *memcg);
561 #ifdef CONFIG_MEMCG_SWAP
562 extern int do_swap_account;
565 struct mem_cgroup *lock_page_memcg(struct page *page);
566 void __unlock_page_memcg(struct mem_cgroup *memcg);
567 void unlock_page_memcg(struct page *page);
569 /* idx can be of type enum memcg_stat_item or node_stat_item */
570 static inline unsigned long memcg_page_state(struct mem_cgroup *memcg,
573 long x = atomic_long_read(&memcg->stat[idx]);
581 /* idx can be of type enum memcg_stat_item or node_stat_item */
582 static inline void __mod_memcg_state(struct mem_cgroup *memcg,
587 if (mem_cgroup_disabled())
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]);
595 __this_cpu_write(memcg->stat_cpu->count[idx], x);
598 /* idx can be of type enum memcg_stat_item or node_stat_item */
599 static inline void mod_memcg_state(struct mem_cgroup *memcg,
604 local_irq_save(flags);
605 __mod_memcg_state(memcg, idx, val);
606 local_irq_restore(flags);
610 * mod_memcg_page_state - update page state statistics
612 * @idx: page state item to account
613 * @val: number of pages (positive or negative)
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:
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)
624 * Kernel pages are an exception to this, since they'll never move.
626 static inline void __mod_memcg_page_state(struct page *page,
629 if (page->mem_cgroup)
630 __mod_memcg_state(page->mem_cgroup, idx, val);
633 static inline void mod_memcg_page_state(struct page *page,
636 if (page->mem_cgroup)
637 mod_memcg_state(page->mem_cgroup, idx, val);
640 static inline unsigned long lruvec_page_state(struct lruvec *lruvec,
641 enum node_stat_item idx)
643 struct mem_cgroup_per_node *pn;
646 if (mem_cgroup_disabled())
647 return node_page_state(lruvec_pgdat(lruvec), idx);
649 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
650 x = atomic_long_read(&pn->lruvec_stat[idx]);
658 static inline void __mod_lruvec_state(struct lruvec *lruvec,
659 enum node_stat_item idx, int val)
661 struct mem_cgroup_per_node *pn;
665 __mod_node_page_state(lruvec_pgdat(lruvec), idx, val);
667 if (mem_cgroup_disabled())
670 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
673 __mod_memcg_state(pn->memcg, idx, val);
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]);
681 __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x);
684 static inline void mod_lruvec_state(struct lruvec *lruvec,
685 enum node_stat_item idx, int val)
689 local_irq_save(flags);
690 __mod_lruvec_state(lruvec, idx, val);
691 local_irq_restore(flags);
694 static inline void __mod_lruvec_page_state(struct page *page,
695 enum node_stat_item idx, int val)
697 pg_data_t *pgdat = page_pgdat(page);
698 struct lruvec *lruvec;
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);
706 lruvec = mem_cgroup_lruvec(pgdat, page->mem_cgroup);
707 __mod_lruvec_state(lruvec, idx, val);
710 static inline void mod_lruvec_page_state(struct page *page,
711 enum node_stat_item idx, int val)
715 local_irq_save(flags);
716 __mod_lruvec_page_state(page, idx, val);
717 local_irq_restore(flags);
720 unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
722 unsigned long *total_scanned);
724 static inline void __count_memcg_events(struct mem_cgroup *memcg,
725 enum vm_event_item idx,
730 if (mem_cgroup_disabled())
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]);
738 __this_cpu_write(memcg->stat_cpu->events[idx], x);
741 static inline void count_memcg_events(struct mem_cgroup *memcg,
742 enum vm_event_item idx,
747 local_irq_save(flags);
748 __count_memcg_events(memcg, idx, count);
749 local_irq_restore(flags);
752 static inline void count_memcg_page_event(struct page *page,
753 enum vm_event_item idx)
755 if (page->mem_cgroup)
756 count_memcg_events(page->mem_cgroup, idx, 1);
759 static inline void count_memcg_event_mm(struct mm_struct *mm,
760 enum vm_event_item idx)
762 struct mem_cgroup *memcg;
764 if (mem_cgroup_disabled())
768 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
770 count_memcg_events(memcg, idx, 1);
774 static inline void memcg_memory_event(struct mem_cgroup *memcg,
775 enum memcg_memory_event event)
777 atomic_long_inc(&memcg->memory_events[event]);
778 cgroup_file_notify(&memcg->events_file);
781 static inline void memcg_memory_event_mm(struct mm_struct *mm,
782 enum memcg_memory_event event)
784 struct mem_cgroup *memcg;
786 if (mem_cgroup_disabled())
790 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
792 memcg_memory_event(memcg, event);
796 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
797 void mem_cgroup_split_huge_fixup(struct page *head);
800 #else /* CONFIG_MEMCG */
802 #define MEM_CGROUP_ID_SHIFT 0
803 #define MEM_CGROUP_ID_MAX 0
807 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
812 static inline bool mem_cgroup_disabled(void)
817 static inline void memcg_memory_event(struct mem_cgroup *memcg,
818 enum memcg_memory_event event)
822 static inline void memcg_memory_event_mm(struct mm_struct *mm,
823 enum memcg_memory_event event)
827 static inline enum mem_cgroup_protection mem_cgroup_protected(
828 struct mem_cgroup *root, struct mem_cgroup *memcg)
830 return MEMCG_PROT_NONE;
833 static inline int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
835 struct mem_cgroup **memcgp,
842 static inline int mem_cgroup_try_charge_delay(struct page *page,
843 struct mm_struct *mm,
845 struct mem_cgroup **memcgp,
852 static inline void mem_cgroup_commit_charge(struct page *page,
853 struct mem_cgroup *memcg,
854 bool lrucare, bool compound)
858 static inline void mem_cgroup_cancel_charge(struct page *page,
859 struct mem_cgroup *memcg,
864 static inline void mem_cgroup_uncharge(struct page *page)
868 static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
872 static inline void mem_cgroup_migrate(struct page *old, struct page *new)
876 static inline struct lruvec *mem_cgroup_lruvec(struct pglist_data *pgdat,
877 struct mem_cgroup *memcg)
879 return node_lruvec(pgdat);
882 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
883 struct pglist_data *pgdat)
885 return &pgdat->lruvec;
888 static inline bool mm_match_cgroup(struct mm_struct *mm,
889 struct mem_cgroup *memcg)
894 static inline bool task_in_mem_cgroup(struct task_struct *task,
895 const struct mem_cgroup *memcg)
900 static inline struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
905 static inline struct mem_cgroup *get_mem_cgroup_from_page(struct page *page)
910 static inline void mem_cgroup_put(struct mem_cgroup *memcg)
914 static inline struct mem_cgroup *
915 mem_cgroup_iter(struct mem_cgroup *root,
916 struct mem_cgroup *prev,
917 struct mem_cgroup_reclaim_cookie *reclaim)
922 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
923 struct mem_cgroup *prev)
927 static inline int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
928 int (*fn)(struct task_struct *, void *), void *arg)
933 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
938 static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
941 /* XXX: This should always return root_mem_cgroup */
945 static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m)
950 static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec)
955 static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
960 static inline unsigned long
961 mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
966 unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
967 enum lru_list lru, int zone_idx)
972 static inline unsigned long
973 mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
974 int nid, unsigned int lru_mask)
979 static inline unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg)
985 mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p)
990 mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg)
994 static inline struct mem_cgroup *lock_page_memcg(struct page *page)
999 static inline void __unlock_page_memcg(struct mem_cgroup *memcg)
1003 static inline void unlock_page_memcg(struct page *page)
1007 static inline void mem_cgroup_handle_over_high(void)
1011 static inline void mem_cgroup_enter_user_fault(void)
1015 static inline void mem_cgroup_exit_user_fault(void)
1019 static inline bool task_in_memcg_oom(struct task_struct *p)
1024 static inline bool mem_cgroup_oom_synchronize(bool wait)
1029 static inline struct mem_cgroup *mem_cgroup_get_oom_group(
1030 struct task_struct *victim, struct mem_cgroup *oom_domain)
1035 static inline void mem_cgroup_print_oom_group(struct mem_cgroup *memcg)
1039 static inline unsigned long memcg_page_state(struct mem_cgroup *memcg,
1045 static inline void __mod_memcg_state(struct mem_cgroup *memcg,
1051 static inline void mod_memcg_state(struct mem_cgroup *memcg,
1057 static inline void __mod_memcg_page_state(struct page *page,
1063 static inline void mod_memcg_page_state(struct page *page,
1069 static inline unsigned long lruvec_page_state(struct lruvec *lruvec,
1070 enum node_stat_item idx)
1072 return node_page_state(lruvec_pgdat(lruvec), idx);
1075 static inline void __mod_lruvec_state(struct lruvec *lruvec,
1076 enum node_stat_item idx, int val)
1078 __mod_node_page_state(lruvec_pgdat(lruvec), idx, val);
1081 static inline void mod_lruvec_state(struct lruvec *lruvec,
1082 enum node_stat_item idx, int val)
1084 mod_node_page_state(lruvec_pgdat(lruvec), idx, val);
1087 static inline void __mod_lruvec_page_state(struct page *page,
1088 enum node_stat_item idx, int val)
1090 __mod_node_page_state(page_pgdat(page), idx, val);
1093 static inline void mod_lruvec_page_state(struct page *page,
1094 enum node_stat_item idx, int val)
1096 mod_node_page_state(page_pgdat(page), idx, val);
1100 unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
1102 unsigned long *total_scanned)
1107 static inline void mem_cgroup_split_huge_fixup(struct page *head)
1111 static inline void count_memcg_events(struct mem_cgroup *memcg,
1112 enum vm_event_item idx,
1113 unsigned long count)
1117 static inline void count_memcg_page_event(struct page *page,
1123 void count_memcg_event_mm(struct mm_struct *mm, enum vm_event_item idx)
1126 #endif /* CONFIG_MEMCG */
1128 /* idx can be of type enum memcg_stat_item or node_stat_item */
1129 static inline void __inc_memcg_state(struct mem_cgroup *memcg,
1132 __mod_memcg_state(memcg, idx, 1);
1135 /* idx can be of type enum memcg_stat_item or node_stat_item */
1136 static inline void __dec_memcg_state(struct mem_cgroup *memcg,
1139 __mod_memcg_state(memcg, idx, -1);
1142 /* idx can be of type enum memcg_stat_item or node_stat_item */
1143 static inline void __inc_memcg_page_state(struct page *page,
1146 __mod_memcg_page_state(page, idx, 1);
1149 /* idx can be of type enum memcg_stat_item or node_stat_item */
1150 static inline void __dec_memcg_page_state(struct page *page,
1153 __mod_memcg_page_state(page, idx, -1);
1156 static inline void __inc_lruvec_state(struct lruvec *lruvec,
1157 enum node_stat_item idx)
1159 __mod_lruvec_state(lruvec, idx, 1);
1162 static inline void __dec_lruvec_state(struct lruvec *lruvec,
1163 enum node_stat_item idx)
1165 __mod_lruvec_state(lruvec, idx, -1);
1168 static inline void __inc_lruvec_page_state(struct page *page,
1169 enum node_stat_item idx)
1171 __mod_lruvec_page_state(page, idx, 1);
1174 static inline void __dec_lruvec_page_state(struct page *page,
1175 enum node_stat_item idx)
1177 __mod_lruvec_page_state(page, idx, -1);
1180 /* idx can be of type enum memcg_stat_item or node_stat_item */
1181 static inline void inc_memcg_state(struct mem_cgroup *memcg,
1184 mod_memcg_state(memcg, idx, 1);
1187 /* idx can be of type enum memcg_stat_item or node_stat_item */
1188 static inline void dec_memcg_state(struct mem_cgroup *memcg,
1191 mod_memcg_state(memcg, idx, -1);
1194 /* idx can be of type enum memcg_stat_item or node_stat_item */
1195 static inline void inc_memcg_page_state(struct page *page,
1198 mod_memcg_page_state(page, idx, 1);
1201 /* idx can be of type enum memcg_stat_item or node_stat_item */
1202 static inline void dec_memcg_page_state(struct page *page,
1205 mod_memcg_page_state(page, idx, -1);
1208 static inline void inc_lruvec_state(struct lruvec *lruvec,
1209 enum node_stat_item idx)
1211 mod_lruvec_state(lruvec, idx, 1);
1214 static inline void dec_lruvec_state(struct lruvec *lruvec,
1215 enum node_stat_item idx)
1217 mod_lruvec_state(lruvec, idx, -1);
1220 static inline void inc_lruvec_page_state(struct page *page,
1221 enum node_stat_item idx)
1223 mod_lruvec_page_state(page, idx, 1);
1226 static inline void dec_lruvec_page_state(struct page *page,
1227 enum node_stat_item idx)
1229 mod_lruvec_page_state(page, idx, -1);
1232 #ifdef CONFIG_CGROUP_WRITEBACK
1234 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
1235 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
1236 unsigned long *pheadroom, unsigned long *pdirty,
1237 unsigned long *pwriteback);
1239 #else /* CONFIG_CGROUP_WRITEBACK */
1241 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
1246 static 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)
1254 #endif /* CONFIG_CGROUP_WRITEBACK */
1257 bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
1258 void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
1260 extern struct static_key_false memcg_sockets_enabled_key;
1261 #define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_enabled_key)
1262 void mem_cgroup_sk_alloc(struct sock *sk);
1263 void mem_cgroup_sk_free(struct sock *sk);
1264 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
1266 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_pressure)
1269 if (time_before(jiffies, memcg->socket_pressure))
1271 } while ((memcg = parent_mem_cgroup(memcg)));
1275 #define mem_cgroup_sockets_enabled 0
1276 static inline void mem_cgroup_sk_alloc(struct sock *sk) { };
1277 static inline void mem_cgroup_sk_free(struct sock *sk) { };
1278 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
1284 struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep);
1285 void memcg_kmem_put_cache(struct kmem_cache *cachep);
1287 #ifdef CONFIG_MEMCG_KMEM
1288 int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order);
1289 void __memcg_kmem_uncharge(struct page *page, int order);
1290 int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
1291 struct mem_cgroup *memcg);
1293 extern struct static_key_false memcg_kmem_enabled_key;
1294 extern struct workqueue_struct *memcg_kmem_cache_wq;
1296 extern int memcg_nr_cache_ids;
1297 void memcg_get_cache_ids(void);
1298 void memcg_put_cache_ids(void);
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
1305 #define for_each_memcg_cache_index(_idx) \
1306 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
1308 static inline bool memcg_kmem_enabled(void)
1310 return static_branch_unlikely(&memcg_kmem_enabled_key);
1313 static inline int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
1315 if (memcg_kmem_enabled())
1316 return __memcg_kmem_charge(page, gfp, order);
1320 static inline void memcg_kmem_uncharge(struct page *page, int order)
1322 if (memcg_kmem_enabled())
1323 __memcg_kmem_uncharge(page, order);
1326 static inline int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp,
1327 int order, struct mem_cgroup *memcg)
1329 if (memcg_kmem_enabled())
1330 return __memcg_kmem_charge_memcg(page, gfp, order, memcg);
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.
1338 static inline int memcg_cache_id(struct mem_cgroup *memcg)
1340 return memcg ? memcg->kmemcg_id : -1;
1343 extern int memcg_expand_shrinker_maps(int new_id);
1345 extern void memcg_set_shrinker_bit(struct mem_cgroup *memcg,
1346 int nid, int shrinker_id);
1349 static inline int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
1354 static inline void memcg_kmem_uncharge(struct page *page, int order)
1358 static inline int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
1363 static inline void __memcg_kmem_uncharge(struct page *page, int order)
1367 #define for_each_memcg_cache_index(_idx) \
1370 static inline bool memcg_kmem_enabled(void)
1375 static inline int memcg_cache_id(struct mem_cgroup *memcg)
1380 static inline void memcg_get_cache_ids(void)
1384 static inline void memcg_put_cache_ids(void)
1388 static inline void memcg_set_shrinker_bit(struct mem_cgroup *memcg,
1389 int nid, int shrinker_id) { }
1390 #endif /* CONFIG_MEMCG_KMEM */
1392 #endif /* _LINUX_MEMCONTROL_H */