1 /* memcontrol.h - Memory Controller
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
6 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
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>
29 #include <linux/mmzone.h>
30 #include <linux/writeback.h>
38 * The corresponding mem_cgroup_stat_names is defined in mm/memcontrol.c,
39 * These two lists should keep in accord with each other.
41 enum mem_cgroup_stat_index {
43 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
45 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
46 MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
47 MEM_CGROUP_STAT_RSS_HUGE, /* # of pages charged as anon huge */
48 MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
49 MEM_CGROUP_STAT_DIRTY, /* # of dirty pages in page cache */
50 MEM_CGROUP_STAT_WRITEBACK, /* # of pages under writeback */
51 MEM_CGROUP_STAT_SWAP, /* # of pages, swapped out */
52 MEM_CGROUP_STAT_NSTATS,
55 struct mem_cgroup_reclaim_cookie {
58 unsigned int generation;
61 enum mem_cgroup_events_index {
62 MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */
63 MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */
64 MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */
65 MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */
66 MEM_CGROUP_EVENTS_NSTATS,
67 /* default hierarchy events */
68 MEMCG_LOW = MEM_CGROUP_EVENTS_NSTATS,
76 * Per memcg event counter is incremented at every pagein/pageout. With THP,
77 * it will be incremated by the number of pages. This counter is used for
78 * for trigger some periodic events. This is straightforward and better
79 * than using jiffies etc. to handle periodic memcg event.
81 enum mem_cgroup_events_target {
82 MEM_CGROUP_TARGET_THRESH,
83 MEM_CGROUP_TARGET_SOFTLIMIT,
84 MEM_CGROUP_TARGET_NUMAINFO,
89 struct mem_cgroup_stat_cpu {
90 long count[MEM_CGROUP_STAT_NSTATS];
91 unsigned long events[MEMCG_NR_EVENTS];
92 unsigned long nr_page_events;
93 unsigned long targets[MEM_CGROUP_NTARGETS];
96 struct mem_cgroup_reclaim_iter {
97 struct mem_cgroup *position;
98 /* scan generation, increased every round-trip */
99 unsigned int generation;
103 * per-zone information in memory controller.
105 struct mem_cgroup_per_zone {
106 struct lruvec lruvec;
107 unsigned long lru_size[NR_LRU_LISTS];
109 struct mem_cgroup_reclaim_iter iter[DEF_PRIORITY + 1];
111 struct rb_node tree_node; /* RB tree node */
112 unsigned long usage_in_excess;/* Set to the value by which */
113 /* the soft limit is exceeded*/
115 struct mem_cgroup *memcg; /* Back pointer, we cannot */
116 /* use container_of */
119 struct mem_cgroup_per_node {
120 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
123 struct mem_cgroup_threshold {
124 struct eventfd_ctx *eventfd;
125 unsigned long threshold;
129 struct mem_cgroup_threshold_ary {
130 /* An array index points to threshold just below or equal to usage. */
131 int current_threshold;
132 /* Size of entries[] */
134 /* Array of thresholds */
135 struct mem_cgroup_threshold entries[0];
138 struct mem_cgroup_thresholds {
139 /* Primary thresholds array */
140 struct mem_cgroup_threshold_ary *primary;
142 * Spare threshold array.
143 * This is needed to make mem_cgroup_unregister_event() "never fail".
144 * It must be able to store at least primary->size - 1 entries.
146 struct mem_cgroup_threshold_ary *spare;
149 enum memcg_kmem_state {
156 * The memory controller data structure. The memory controller controls both
157 * page cache and RSS per cgroup. We would eventually like to provide
158 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
159 * to help the administrator determine what knobs to tune.
162 struct cgroup_subsys_state css;
164 /* Accounted resources */
165 struct page_counter memory;
167 /* Legacy consumer-oriented counters */
168 struct page_counter memsw;
169 struct page_counter kmem;
170 struct page_counter tcpmem;
172 /* Normal memory consumption range */
176 /* Range enforcement for interrupt charges */
177 struct work_struct high_work;
179 unsigned long soft_limit;
181 /* vmpressure notifications */
182 struct vmpressure vmpressure;
184 /* css_online() has been completed */
188 * Should the accounting and control be hierarchical, per subtree?
192 /* protected by memcg_oom_lock */
197 /* OOM-Killer disable */
198 int oom_kill_disable;
200 /* handle for "memory.events" */
201 struct cgroup_file events_file;
203 /* protect arrays of thresholds */
204 struct mutex thresholds_lock;
206 /* thresholds for memory usage. RCU-protected */
207 struct mem_cgroup_thresholds thresholds;
209 /* thresholds for mem+swap usage. RCU-protected */
210 struct mem_cgroup_thresholds memsw_thresholds;
212 /* For oom notifier event fd */
213 struct list_head oom_notify;
216 * Should we move charges of a task when a task is moved into this
217 * mem_cgroup ? And what type of charges should we move ?
219 unsigned long move_charge_at_immigrate;
221 * set > 0 if pages under this cgroup are moving to other cgroup.
223 atomic_t moving_account;
224 /* taken only while moving_account > 0 */
225 spinlock_t move_lock;
226 struct task_struct *move_lock_task;
227 unsigned long move_lock_flags;
231 struct mem_cgroup_stat_cpu __percpu *stat;
233 unsigned long socket_pressure;
235 /* Legacy tcp memory accounting */
240 /* Index in the kmem_cache->memcg_params.memcg_caches array */
242 enum memcg_kmem_state kmem_state;
245 int last_scanned_node;
247 nodemask_t scan_nodes;
248 atomic_t numainfo_events;
249 atomic_t numainfo_updating;
252 #ifdef CONFIG_CGROUP_WRITEBACK
253 struct list_head cgwb_list;
254 struct wb_domain cgwb_domain;
257 /* List of events which userspace want to receive */
258 struct list_head event_list;
259 spinlock_t event_list_lock;
261 struct mem_cgroup_per_node *nodeinfo[0];
262 /* WARNING: nodeinfo must be the last member here */
265 extern struct mem_cgroup *root_mem_cgroup;
268 * mem_cgroup_events - count memory events against a cgroup
269 * @memcg: the memory cgroup
270 * @idx: the event index
271 * @nr: the number of events to account for
273 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
274 enum mem_cgroup_events_index idx,
277 this_cpu_add(memcg->stat->events[idx], nr);
278 cgroup_file_notify(&memcg->events_file);
281 bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg);
283 int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
284 gfp_t gfp_mask, struct mem_cgroup **memcgp,
286 void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
287 bool lrucare, bool compound);
288 void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
290 void mem_cgroup_uncharge(struct page *page);
291 void mem_cgroup_uncharge_list(struct list_head *page_list);
293 void mem_cgroup_replace_page(struct page *oldpage, struct page *newpage);
295 struct lruvec *mem_cgroup_zone_lruvec(struct zone *, struct mem_cgroup *);
296 struct lruvec *mem_cgroup_page_lruvec(struct page *, struct zone *);
298 bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg);
299 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
302 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
303 return css ? container_of(css, struct mem_cgroup, css) : NULL;
306 #define mem_cgroup_from_counter(counter, member) \
307 container_of(counter, struct mem_cgroup, member)
309 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
311 struct mem_cgroup_reclaim_cookie *);
312 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
315 * parent_mem_cgroup - find the accounting parent of a memcg
316 * @memcg: memcg whose parent to find
318 * Returns the parent memcg, or NULL if this is the root or the memory
319 * controller is in legacy no-hierarchy mode.
321 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
323 if (!memcg->memory.parent)
325 return mem_cgroup_from_counter(memcg->memory.parent, memory);
328 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
329 struct mem_cgroup *root)
333 if (!root->use_hierarchy)
335 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
338 static inline bool mm_match_cgroup(struct mm_struct *mm,
339 struct mem_cgroup *memcg)
341 struct mem_cgroup *task_memcg;
345 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
347 match = mem_cgroup_is_descendant(task_memcg, memcg);
352 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
353 ino_t page_cgroup_ino(struct page *page);
355 static inline bool mem_cgroup_disabled(void)
357 return !cgroup_subsys_enabled(memory_cgrp_subsys);
361 * For memory reclaim.
363 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
365 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
368 static inline bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
370 struct mem_cgroup_per_zone *mz;
371 struct mem_cgroup *memcg;
373 if (mem_cgroup_disabled())
376 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
379 return !!(memcg->css.flags & CSS_ONLINE);
383 unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
385 struct mem_cgroup_per_zone *mz;
387 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
388 return mz->lru_size[lru];
391 static inline bool mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
393 unsigned long inactive_ratio;
394 unsigned long inactive;
395 unsigned long active;
398 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
399 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
401 gb = (inactive + active) >> (30 - PAGE_SHIFT);
403 inactive_ratio = int_sqrt(10 * gb);
407 return inactive * inactive_ratio < active;
410 void mem_cgroup_handle_over_high(void);
412 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
413 struct task_struct *p);
415 static inline void mem_cgroup_oom_enable(void)
417 WARN_ON(current->memcg_may_oom);
418 current->memcg_may_oom = 1;
421 static inline void mem_cgroup_oom_disable(void)
423 WARN_ON(!current->memcg_may_oom);
424 current->memcg_may_oom = 0;
427 static inline bool task_in_memcg_oom(struct task_struct *p)
429 return p->memcg_in_oom;
432 bool mem_cgroup_oom_synchronize(bool wait);
434 #ifdef CONFIG_MEMCG_SWAP
435 extern int do_swap_account;
438 struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page);
439 void mem_cgroup_end_page_stat(struct mem_cgroup *memcg);
442 * mem_cgroup_update_page_stat - update page state statistics
443 * @memcg: memcg to account against
444 * @idx: page state item to account
445 * @val: number of pages (positive or negative)
447 * See mem_cgroup_begin_page_stat() for locking requirements.
449 static inline void mem_cgroup_update_page_stat(struct mem_cgroup *memcg,
450 enum mem_cgroup_stat_index idx, int val)
452 VM_BUG_ON(!rcu_read_lock_held());
455 this_cpu_add(memcg->stat->count[idx], val);
458 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
459 enum mem_cgroup_stat_index idx)
461 mem_cgroup_update_page_stat(memcg, idx, 1);
464 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
465 enum mem_cgroup_stat_index idx)
467 mem_cgroup_update_page_stat(memcg, idx, -1);
470 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
472 unsigned long *total_scanned);
474 static inline void mem_cgroup_count_vm_event(struct mm_struct *mm,
475 enum vm_event_item idx)
477 struct mem_cgroup *memcg;
479 if (mem_cgroup_disabled())
483 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
484 if (unlikely(!memcg))
489 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
492 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
500 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
501 void mem_cgroup_split_huge_fixup(struct page *head);
504 #else /* CONFIG_MEMCG */
507 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
508 enum mem_cgroup_events_index idx,
513 static inline bool mem_cgroup_low(struct mem_cgroup *root,
514 struct mem_cgroup *memcg)
519 static inline int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
521 struct mem_cgroup **memcgp,
528 static inline void mem_cgroup_commit_charge(struct page *page,
529 struct mem_cgroup *memcg,
530 bool lrucare, bool compound)
534 static inline void mem_cgroup_cancel_charge(struct page *page,
535 struct mem_cgroup *memcg,
540 static inline void mem_cgroup_uncharge(struct page *page)
544 static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
548 static inline void mem_cgroup_replace_page(struct page *old, struct page *new)
552 static inline struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
553 struct mem_cgroup *memcg)
555 return &zone->lruvec;
558 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
561 return &zone->lruvec;
564 static inline bool mm_match_cgroup(struct mm_struct *mm,
565 struct mem_cgroup *memcg)
570 static inline bool task_in_mem_cgroup(struct task_struct *task,
571 const struct mem_cgroup *memcg)
576 static inline struct mem_cgroup *
577 mem_cgroup_iter(struct mem_cgroup *root,
578 struct mem_cgroup *prev,
579 struct mem_cgroup_reclaim_cookie *reclaim)
584 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
585 struct mem_cgroup *prev)
589 static inline bool mem_cgroup_disabled(void)
595 mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
600 static inline bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
605 static inline unsigned long
606 mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
612 mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
618 mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
622 static inline struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
627 static inline void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
631 static inline void mem_cgroup_handle_over_high(void)
635 static inline void mem_cgroup_oom_enable(void)
639 static inline void mem_cgroup_oom_disable(void)
643 static inline bool task_in_memcg_oom(struct task_struct *p)
648 static inline bool mem_cgroup_oom_synchronize(bool wait)
653 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
654 enum mem_cgroup_stat_index idx)
658 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
659 enum mem_cgroup_stat_index idx)
664 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
666 unsigned long *total_scanned)
671 static inline void mem_cgroup_split_huge_fixup(struct page *head)
676 void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
679 #endif /* CONFIG_MEMCG */
681 #ifdef CONFIG_CGROUP_WRITEBACK
683 struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg);
684 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
685 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
686 unsigned long *pheadroom, unsigned long *pdirty,
687 unsigned long *pwriteback);
689 #else /* CONFIG_CGROUP_WRITEBACK */
691 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
696 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
697 unsigned long *pfilepages,
698 unsigned long *pheadroom,
699 unsigned long *pdirty,
700 unsigned long *pwriteback)
704 #endif /* CONFIG_CGROUP_WRITEBACK */
707 void sock_update_memcg(struct sock *sk);
708 void sock_release_memcg(struct sock *sk);
709 bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
710 void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
712 extern struct static_key_false memcg_sockets_enabled_key;
713 #define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_enabled_key)
714 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
716 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_pressure)
719 if (time_before(jiffies, memcg->socket_pressure))
721 } while ((memcg = parent_mem_cgroup(memcg)));
725 #define mem_cgroup_sockets_enabled 0
726 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
732 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
733 extern struct static_key_false memcg_kmem_enabled_key;
735 extern int memcg_nr_cache_ids;
736 void memcg_get_cache_ids(void);
737 void memcg_put_cache_ids(void);
740 * Helper macro to loop through all memcg-specific caches. Callers must still
741 * check if the cache is valid (it is either valid or NULL).
742 * the slab_mutex must be held when looping through those caches
744 #define for_each_memcg_cache_index(_idx) \
745 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
747 static inline bool memcg_kmem_enabled(void)
749 return static_branch_unlikely(&memcg_kmem_enabled_key);
752 static inline bool memcg_kmem_online(struct mem_cgroup *memcg)
754 return memcg->kmem_state == KMEM_ONLINE;
758 * In general, we'll do everything in our power to not incur in any overhead
759 * for non-memcg users for the kmem functions. Not even a function call, if we
762 * Therefore, we'll inline all those functions so that in the best case, we'll
763 * see that kmemcg is off for everybody and proceed quickly. If it is on,
764 * we'll still do most of the flag checking inline. We check a lot of
765 * conditions, but because they are pretty simple, they are expected to be
768 int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
769 struct mem_cgroup *memcg);
770 int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order);
771 void __memcg_kmem_uncharge(struct page *page, int order);
774 * helper for acessing a memcg's index. It will be used as an index in the
775 * child cache array in kmem_cache, and also to derive its name. This function
776 * will return -1 when this is not a kmem-limited memcg.
778 static inline int memcg_cache_id(struct mem_cgroup *memcg)
780 return memcg ? memcg->kmemcg_id : -1;
783 struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp);
784 void __memcg_kmem_put_cache(struct kmem_cache *cachep);
786 static inline bool __memcg_kmem_bypass(void)
788 if (!memcg_kmem_enabled())
790 if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
796 * memcg_kmem_charge: charge a kmem page
797 * @page: page to charge
799 * @order: allocation order
801 * Returns 0 on success, an error code on failure.
803 static __always_inline int memcg_kmem_charge(struct page *page,
804 gfp_t gfp, int order)
806 if (__memcg_kmem_bypass())
808 if (!(gfp & __GFP_ACCOUNT))
810 return __memcg_kmem_charge(page, gfp, order);
814 * memcg_kmem_uncharge: uncharge a kmem page
815 * @page: page to uncharge
816 * @order: allocation order
818 static __always_inline void memcg_kmem_uncharge(struct page *page, int order)
820 if (memcg_kmem_enabled())
821 __memcg_kmem_uncharge(page, order);
825 * memcg_kmem_get_cache: selects the correct per-memcg cache for allocation
826 * @cachep: the original global kmem cache
828 * All memory allocated from a per-memcg cache is charged to the owner memcg.
830 static __always_inline struct kmem_cache *
831 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
833 if (__memcg_kmem_bypass())
835 return __memcg_kmem_get_cache(cachep, gfp);
838 static __always_inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
840 if (memcg_kmem_enabled())
841 __memcg_kmem_put_cache(cachep);
844 #define for_each_memcg_cache_index(_idx) \
847 static inline bool memcg_kmem_enabled(void)
852 static inline bool memcg_kmem_online(struct mem_cgroup *memcg)
857 static inline int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
862 static inline void memcg_kmem_uncharge(struct page *page, int order)
866 static inline int memcg_cache_id(struct mem_cgroup *memcg)
871 static inline void memcg_get_cache_ids(void)
875 static inline void memcg_put_cache_ids(void)
879 static inline struct kmem_cache *
880 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
885 static inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
888 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
890 #endif /* _LINUX_MEMCONTROL_H */