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;
166 struct page_counter swap;
168 /* Legacy consumer-oriented counters */
169 struct page_counter memsw;
170 struct page_counter kmem;
171 struct page_counter tcpmem;
173 /* Normal memory consumption range */
177 /* Range enforcement for interrupt charges */
178 struct work_struct high_work;
180 unsigned long soft_limit;
182 /* vmpressure notifications */
183 struct vmpressure vmpressure;
186 * Should the accounting and control be hierarchical, per subtree?
190 /* protected by memcg_oom_lock */
195 /* OOM-Killer disable */
196 int oom_kill_disable;
198 /* handle for "memory.events" */
199 struct cgroup_file events_file;
201 /* protect arrays of thresholds */
202 struct mutex thresholds_lock;
204 /* thresholds for memory usage. RCU-protected */
205 struct mem_cgroup_thresholds thresholds;
207 /* thresholds for mem+swap usage. RCU-protected */
208 struct mem_cgroup_thresholds memsw_thresholds;
210 /* For oom notifier event fd */
211 struct list_head oom_notify;
214 * Should we move charges of a task when a task is moved into this
215 * mem_cgroup ? And what type of charges should we move ?
217 unsigned long move_charge_at_immigrate;
219 * set > 0 if pages under this cgroup are moving to other cgroup.
221 atomic_t moving_account;
222 /* taken only while moving_account > 0 */
223 spinlock_t move_lock;
224 struct task_struct *move_lock_task;
225 unsigned long move_lock_flags;
229 struct mem_cgroup_stat_cpu __percpu *stat;
231 unsigned long socket_pressure;
233 /* Legacy tcp memory accounting */
238 /* Index in the kmem_cache->memcg_params.memcg_caches array */
240 enum memcg_kmem_state kmem_state;
243 int last_scanned_node;
245 nodemask_t scan_nodes;
246 atomic_t numainfo_events;
247 atomic_t numainfo_updating;
250 #ifdef CONFIG_CGROUP_WRITEBACK
251 struct list_head cgwb_list;
252 struct wb_domain cgwb_domain;
255 /* List of events which userspace want to receive */
256 struct list_head event_list;
257 spinlock_t event_list_lock;
259 struct mem_cgroup_per_node *nodeinfo[0];
260 /* WARNING: nodeinfo must be the last member here */
263 extern struct mem_cgroup *root_mem_cgroup;
266 * mem_cgroup_events - count memory events against a cgroup
267 * @memcg: the memory cgroup
268 * @idx: the event index
269 * @nr: the number of events to account for
271 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
272 enum mem_cgroup_events_index idx,
275 this_cpu_add(memcg->stat->events[idx], nr);
276 cgroup_file_notify(&memcg->events_file);
279 bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg);
281 int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
282 gfp_t gfp_mask, struct mem_cgroup **memcgp,
284 void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
285 bool lrucare, bool compound);
286 void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
288 void mem_cgroup_uncharge(struct page *page);
289 void mem_cgroup_uncharge_list(struct list_head *page_list);
291 void mem_cgroup_replace_page(struct page *oldpage, struct page *newpage);
293 struct lruvec *mem_cgroup_zone_lruvec(struct zone *, struct mem_cgroup *);
294 struct lruvec *mem_cgroup_page_lruvec(struct page *, struct zone *);
296 bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg);
297 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
300 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
301 return css ? container_of(css, struct mem_cgroup, css) : NULL;
304 #define mem_cgroup_from_counter(counter, member) \
305 container_of(counter, struct mem_cgroup, member)
307 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
309 struct mem_cgroup_reclaim_cookie *);
310 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
313 * parent_mem_cgroup - find the accounting parent of a memcg
314 * @memcg: memcg whose parent to find
316 * Returns the parent memcg, or NULL if this is the root or the memory
317 * controller is in legacy no-hierarchy mode.
319 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
321 if (!memcg->memory.parent)
323 return mem_cgroup_from_counter(memcg->memory.parent, memory);
326 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
327 struct mem_cgroup *root)
331 if (!root->use_hierarchy)
333 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
336 static inline bool mm_match_cgroup(struct mm_struct *mm,
337 struct mem_cgroup *memcg)
339 struct mem_cgroup *task_memcg;
343 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
345 match = mem_cgroup_is_descendant(task_memcg, memcg);
350 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
351 ino_t page_cgroup_ino(struct page *page);
353 static inline bool mem_cgroup_disabled(void)
355 return !cgroup_subsys_enabled(memory_cgrp_subsys);
358 static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
360 if (mem_cgroup_disabled())
362 return !!(memcg->css.flags & CSS_ONLINE);
366 * For memory reclaim.
368 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
370 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
374 unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
376 struct mem_cgroup_per_zone *mz;
378 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
379 return mz->lru_size[lru];
382 static inline bool mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
384 unsigned long inactive_ratio;
385 unsigned long inactive;
386 unsigned long active;
389 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
390 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
392 gb = (inactive + active) >> (30 - PAGE_SHIFT);
394 inactive_ratio = int_sqrt(10 * gb);
398 return inactive * inactive_ratio < active;
401 void mem_cgroup_handle_over_high(void);
403 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
404 struct task_struct *p);
406 static inline void mem_cgroup_oom_enable(void)
408 WARN_ON(current->memcg_may_oom);
409 current->memcg_may_oom = 1;
412 static inline void mem_cgroup_oom_disable(void)
414 WARN_ON(!current->memcg_may_oom);
415 current->memcg_may_oom = 0;
418 static inline bool task_in_memcg_oom(struct task_struct *p)
420 return p->memcg_in_oom;
423 bool mem_cgroup_oom_synchronize(bool wait);
425 #ifdef CONFIG_MEMCG_SWAP
426 extern int do_swap_account;
429 struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page);
430 void mem_cgroup_end_page_stat(struct mem_cgroup *memcg);
433 * mem_cgroup_update_page_stat - update page state statistics
434 * @memcg: memcg to account against
435 * @idx: page state item to account
436 * @val: number of pages (positive or negative)
438 * See mem_cgroup_begin_page_stat() for locking requirements.
440 static inline void mem_cgroup_update_page_stat(struct mem_cgroup *memcg,
441 enum mem_cgroup_stat_index idx, int val)
443 VM_BUG_ON(!rcu_read_lock_held());
446 this_cpu_add(memcg->stat->count[idx], val);
449 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
450 enum mem_cgroup_stat_index idx)
452 mem_cgroup_update_page_stat(memcg, idx, 1);
455 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
456 enum mem_cgroup_stat_index idx)
458 mem_cgroup_update_page_stat(memcg, idx, -1);
461 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
463 unsigned long *total_scanned);
465 static inline void mem_cgroup_count_vm_event(struct mm_struct *mm,
466 enum vm_event_item idx)
468 struct mem_cgroup *memcg;
470 if (mem_cgroup_disabled())
474 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
475 if (unlikely(!memcg))
480 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
483 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
491 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
492 void mem_cgroup_split_huge_fixup(struct page *head);
495 #else /* CONFIG_MEMCG */
498 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
499 enum mem_cgroup_events_index idx,
504 static inline bool mem_cgroup_low(struct mem_cgroup *root,
505 struct mem_cgroup *memcg)
510 static inline int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
512 struct mem_cgroup **memcgp,
519 static inline void mem_cgroup_commit_charge(struct page *page,
520 struct mem_cgroup *memcg,
521 bool lrucare, bool compound)
525 static inline void mem_cgroup_cancel_charge(struct page *page,
526 struct mem_cgroup *memcg,
531 static inline void mem_cgroup_uncharge(struct page *page)
535 static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
539 static inline void mem_cgroup_replace_page(struct page *old, struct page *new)
543 static inline struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
544 struct mem_cgroup *memcg)
546 return &zone->lruvec;
549 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
552 return &zone->lruvec;
555 static inline bool mm_match_cgroup(struct mm_struct *mm,
556 struct mem_cgroup *memcg)
561 static inline bool task_in_mem_cgroup(struct task_struct *task,
562 const struct mem_cgroup *memcg)
567 static inline struct mem_cgroup *
568 mem_cgroup_iter(struct mem_cgroup *root,
569 struct mem_cgroup *prev,
570 struct mem_cgroup_reclaim_cookie *reclaim)
575 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
576 struct mem_cgroup *prev)
580 static inline bool mem_cgroup_disabled(void)
585 static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
591 mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
596 static inline unsigned long
597 mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
603 mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
609 mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
613 static inline struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
618 static inline void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
622 static inline void mem_cgroup_handle_over_high(void)
626 static inline void mem_cgroup_oom_enable(void)
630 static inline void mem_cgroup_oom_disable(void)
634 static inline bool task_in_memcg_oom(struct task_struct *p)
639 static inline bool mem_cgroup_oom_synchronize(bool wait)
644 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
645 enum mem_cgroup_stat_index idx)
649 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
650 enum mem_cgroup_stat_index idx)
655 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
657 unsigned long *total_scanned)
662 static inline void mem_cgroup_split_huge_fixup(struct page *head)
667 void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
670 #endif /* CONFIG_MEMCG */
672 #ifdef CONFIG_CGROUP_WRITEBACK
674 struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg);
675 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
676 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
677 unsigned long *pheadroom, unsigned long *pdirty,
678 unsigned long *pwriteback);
680 #else /* CONFIG_CGROUP_WRITEBACK */
682 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
687 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
688 unsigned long *pfilepages,
689 unsigned long *pheadroom,
690 unsigned long *pdirty,
691 unsigned long *pwriteback)
695 #endif /* CONFIG_CGROUP_WRITEBACK */
698 void sock_update_memcg(struct sock *sk);
699 void sock_release_memcg(struct sock *sk);
700 bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
701 void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
703 extern struct static_key_false memcg_sockets_enabled_key;
704 #define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_enabled_key)
705 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
707 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_pressure)
710 if (time_before(jiffies, memcg->socket_pressure))
712 } while ((memcg = parent_mem_cgroup(memcg)));
716 #define mem_cgroup_sockets_enabled 0
717 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
723 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
724 extern struct static_key_false memcg_kmem_enabled_key;
726 extern int memcg_nr_cache_ids;
727 void memcg_get_cache_ids(void);
728 void memcg_put_cache_ids(void);
731 * Helper macro to loop through all memcg-specific caches. Callers must still
732 * check if the cache is valid (it is either valid or NULL).
733 * the slab_mutex must be held when looping through those caches
735 #define for_each_memcg_cache_index(_idx) \
736 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
738 static inline bool memcg_kmem_enabled(void)
740 return static_branch_unlikely(&memcg_kmem_enabled_key);
743 static inline bool memcg_kmem_online(struct mem_cgroup *memcg)
745 return memcg->kmem_state == KMEM_ONLINE;
749 * In general, we'll do everything in our power to not incur in any overhead
750 * for non-memcg users for the kmem functions. Not even a function call, if we
753 * Therefore, we'll inline all those functions so that in the best case, we'll
754 * see that kmemcg is off for everybody and proceed quickly. If it is on,
755 * we'll still do most of the flag checking inline. We check a lot of
756 * conditions, but because they are pretty simple, they are expected to be
759 int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
760 struct mem_cgroup *memcg);
761 int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order);
762 void __memcg_kmem_uncharge(struct page *page, int order);
765 * helper for acessing a memcg's index. It will be used as an index in the
766 * child cache array in kmem_cache, and also to derive its name. This function
767 * will return -1 when this is not a kmem-limited memcg.
769 static inline int memcg_cache_id(struct mem_cgroup *memcg)
771 return memcg ? memcg->kmemcg_id : -1;
774 struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp);
775 void __memcg_kmem_put_cache(struct kmem_cache *cachep);
777 static inline bool __memcg_kmem_bypass(void)
779 if (!memcg_kmem_enabled())
781 if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
787 * memcg_kmem_charge: charge a kmem page
788 * @page: page to charge
790 * @order: allocation order
792 * Returns 0 on success, an error code on failure.
794 static __always_inline int memcg_kmem_charge(struct page *page,
795 gfp_t gfp, int order)
797 if (__memcg_kmem_bypass())
799 if (!(gfp & __GFP_ACCOUNT))
801 return __memcg_kmem_charge(page, gfp, order);
805 * memcg_kmem_uncharge: uncharge a kmem page
806 * @page: page to uncharge
807 * @order: allocation order
809 static __always_inline void memcg_kmem_uncharge(struct page *page, int order)
811 if (memcg_kmem_enabled())
812 __memcg_kmem_uncharge(page, order);
816 * memcg_kmem_get_cache: selects the correct per-memcg cache for allocation
817 * @cachep: the original global kmem cache
819 * All memory allocated from a per-memcg cache is charged to the owner memcg.
821 static __always_inline struct kmem_cache *
822 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
824 if (__memcg_kmem_bypass())
826 return __memcg_kmem_get_cache(cachep, gfp);
829 static __always_inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
831 if (memcg_kmem_enabled())
832 __memcg_kmem_put_cache(cachep);
835 #define for_each_memcg_cache_index(_idx) \
838 static inline bool memcg_kmem_enabled(void)
843 static inline bool memcg_kmem_online(struct mem_cgroup *memcg)
848 static inline int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
853 static inline void memcg_kmem_uncharge(struct page *page, int order)
857 static inline int memcg_cache_id(struct mem_cgroup *memcg)
862 static inline void memcg_get_cache_ids(void)
866 static inline void memcg_put_cache_ids(void)
870 static inline struct kmem_cache *
871 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
876 static inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
879 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
881 #endif /* _LINUX_MEMCONTROL_H */