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,
53 /* default hierarchy stats */
58 struct mem_cgroup_reclaim_cookie {
61 unsigned int generation;
64 enum mem_cgroup_events_index {
65 MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */
66 MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */
67 MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */
68 MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */
69 MEM_CGROUP_EVENTS_NSTATS,
70 /* default hierarchy events */
71 MEMCG_LOW = MEM_CGROUP_EVENTS_NSTATS,
79 * Per memcg event counter is incremented at every pagein/pageout. With THP,
80 * it will be incremated by the number of pages. This counter is used for
81 * for trigger some periodic events. This is straightforward and better
82 * than using jiffies etc. to handle periodic memcg event.
84 enum mem_cgroup_events_target {
85 MEM_CGROUP_TARGET_THRESH,
86 MEM_CGROUP_TARGET_SOFTLIMIT,
87 MEM_CGROUP_TARGET_NUMAINFO,
92 struct mem_cgroup_stat_cpu {
93 long count[MEMCG_NR_STAT];
94 unsigned long events[MEMCG_NR_EVENTS];
95 unsigned long nr_page_events;
96 unsigned long targets[MEM_CGROUP_NTARGETS];
99 struct mem_cgroup_reclaim_iter {
100 struct mem_cgroup *position;
101 /* scan generation, increased every round-trip */
102 unsigned int generation;
106 * per-zone information in memory controller.
108 struct mem_cgroup_per_zone {
109 struct lruvec lruvec;
110 unsigned long lru_size[NR_LRU_LISTS];
112 struct mem_cgroup_reclaim_iter iter[DEF_PRIORITY + 1];
114 struct rb_node tree_node; /* RB tree node */
115 unsigned long usage_in_excess;/* Set to the value by which */
116 /* the soft limit is exceeded*/
118 struct mem_cgroup *memcg; /* Back pointer, we cannot */
119 /* use container_of */
122 struct mem_cgroup_per_node {
123 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
126 struct mem_cgroup_threshold {
127 struct eventfd_ctx *eventfd;
128 unsigned long threshold;
132 struct mem_cgroup_threshold_ary {
133 /* An array index points to threshold just below or equal to usage. */
134 int current_threshold;
135 /* Size of entries[] */
137 /* Array of thresholds */
138 struct mem_cgroup_threshold entries[0];
141 struct mem_cgroup_thresholds {
142 /* Primary thresholds array */
143 struct mem_cgroup_threshold_ary *primary;
145 * Spare threshold array.
146 * This is needed to make mem_cgroup_unregister_event() "never fail".
147 * It must be able to store at least primary->size - 1 entries.
149 struct mem_cgroup_threshold_ary *spare;
152 enum memcg_kmem_state {
159 * The memory controller data structure. The memory controller controls both
160 * page cache and RSS per cgroup. We would eventually like to provide
161 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
162 * to help the administrator determine what knobs to tune.
165 struct cgroup_subsys_state css;
167 /* Accounted resources */
168 struct page_counter memory;
169 struct page_counter swap;
171 /* Legacy consumer-oriented counters */
172 struct page_counter memsw;
173 struct page_counter kmem;
174 struct page_counter tcpmem;
176 /* Normal memory consumption range */
180 /* Range enforcement for interrupt charges */
181 struct work_struct high_work;
183 unsigned long soft_limit;
185 /* vmpressure notifications */
186 struct vmpressure vmpressure;
189 * Should the accounting and control be hierarchical, per subtree?
193 /* protected by memcg_oom_lock */
198 /* OOM-Killer disable */
199 int oom_kill_disable;
201 /* handle for "memory.events" */
202 struct cgroup_file events_file;
204 /* protect arrays of thresholds */
205 struct mutex thresholds_lock;
207 /* thresholds for memory usage. RCU-protected */
208 struct mem_cgroup_thresholds thresholds;
210 /* thresholds for mem+swap usage. RCU-protected */
211 struct mem_cgroup_thresholds memsw_thresholds;
213 /* For oom notifier event fd */
214 struct list_head oom_notify;
217 * Should we move charges of a task when a task is moved into this
218 * mem_cgroup ? And what type of charges should we move ?
220 unsigned long move_charge_at_immigrate;
222 * set > 0 if pages under this cgroup are moving to other cgroup.
224 atomic_t moving_account;
225 /* taken only while moving_account > 0 */
226 spinlock_t move_lock;
227 struct task_struct *move_lock_task;
228 unsigned long move_lock_flags;
232 struct mem_cgroup_stat_cpu __percpu *stat;
234 unsigned long socket_pressure;
236 /* Legacy tcp memory accounting */
241 /* Index in the kmem_cache->memcg_params.memcg_caches array */
243 enum memcg_kmem_state kmem_state;
246 int last_scanned_node;
248 nodemask_t scan_nodes;
249 atomic_t numainfo_events;
250 atomic_t numainfo_updating;
253 #ifdef CONFIG_CGROUP_WRITEBACK
254 struct list_head cgwb_list;
255 struct wb_domain cgwb_domain;
258 /* List of events which userspace want to receive */
259 struct list_head event_list;
260 spinlock_t event_list_lock;
262 struct mem_cgroup_per_node *nodeinfo[0];
263 /* WARNING: nodeinfo must be the last member here */
266 extern struct mem_cgroup *root_mem_cgroup;
269 * mem_cgroup_events - count memory events against a cgroup
270 * @memcg: the memory cgroup
271 * @idx: the event index
272 * @nr: the number of events to account for
274 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
275 enum mem_cgroup_events_index idx,
278 this_cpu_add(memcg->stat->events[idx], nr);
279 cgroup_file_notify(&memcg->events_file);
282 bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg);
284 int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
285 gfp_t gfp_mask, struct mem_cgroup **memcgp,
287 void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
288 bool lrucare, bool compound);
289 void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
291 void mem_cgroup_uncharge(struct page *page);
292 void mem_cgroup_uncharge_list(struct list_head *page_list);
294 void mem_cgroup_replace_page(struct page *oldpage, struct page *newpage);
296 struct lruvec *mem_cgroup_zone_lruvec(struct zone *, struct mem_cgroup *);
297 struct lruvec *mem_cgroup_page_lruvec(struct page *, struct zone *);
299 bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg);
300 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
303 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
304 return css ? container_of(css, struct mem_cgroup, css) : NULL;
307 #define mem_cgroup_from_counter(counter, member) \
308 container_of(counter, struct mem_cgroup, member)
310 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
312 struct mem_cgroup_reclaim_cookie *);
313 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
316 * parent_mem_cgroup - find the accounting parent of a memcg
317 * @memcg: memcg whose parent to find
319 * Returns the parent memcg, or NULL if this is the root or the memory
320 * controller is in legacy no-hierarchy mode.
322 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
324 if (!memcg->memory.parent)
326 return mem_cgroup_from_counter(memcg->memory.parent, memory);
329 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
330 struct mem_cgroup *root)
334 if (!root->use_hierarchy)
336 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
339 static inline bool mm_match_cgroup(struct mm_struct *mm,
340 struct mem_cgroup *memcg)
342 struct mem_cgroup *task_memcg;
346 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
348 match = mem_cgroup_is_descendant(task_memcg, memcg);
353 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
354 ino_t page_cgroup_ino(struct page *page);
356 static inline bool mem_cgroup_disabled(void)
358 return !cgroup_subsys_enabled(memory_cgrp_subsys);
361 static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
363 if (mem_cgroup_disabled())
365 return !!(memcg->css.flags & CSS_ONLINE);
369 * For memory reclaim.
371 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
373 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
377 unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
379 struct mem_cgroup_per_zone *mz;
381 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
382 return mz->lru_size[lru];
385 static inline bool mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
387 unsigned long inactive_ratio;
388 unsigned long inactive;
389 unsigned long active;
392 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
393 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
395 gb = (inactive + active) >> (30 - PAGE_SHIFT);
397 inactive_ratio = int_sqrt(10 * gb);
401 return inactive * inactive_ratio < active;
404 void mem_cgroup_handle_over_high(void);
406 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
407 struct task_struct *p);
409 static inline void mem_cgroup_oom_enable(void)
411 WARN_ON(current->memcg_may_oom);
412 current->memcg_may_oom = 1;
415 static inline void mem_cgroup_oom_disable(void)
417 WARN_ON(!current->memcg_may_oom);
418 current->memcg_may_oom = 0;
421 static inline bool task_in_memcg_oom(struct task_struct *p)
423 return p->memcg_in_oom;
426 bool mem_cgroup_oom_synchronize(bool wait);
428 #ifdef CONFIG_MEMCG_SWAP
429 extern int do_swap_account;
432 struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page);
433 void mem_cgroup_end_page_stat(struct mem_cgroup *memcg);
436 * mem_cgroup_update_page_stat - update page state statistics
437 * @memcg: memcg to account against
438 * @idx: page state item to account
439 * @val: number of pages (positive or negative)
441 * See mem_cgroup_begin_page_stat() for locking requirements.
443 static inline void mem_cgroup_update_page_stat(struct mem_cgroup *memcg,
444 enum mem_cgroup_stat_index idx, int val)
446 VM_BUG_ON(!rcu_read_lock_held());
449 this_cpu_add(memcg->stat->count[idx], val);
452 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
453 enum mem_cgroup_stat_index idx)
455 mem_cgroup_update_page_stat(memcg, idx, 1);
458 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
459 enum mem_cgroup_stat_index idx)
461 mem_cgroup_update_page_stat(memcg, idx, -1);
464 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
466 unsigned long *total_scanned);
468 static inline void mem_cgroup_count_vm_event(struct mm_struct *mm,
469 enum vm_event_item idx)
471 struct mem_cgroup *memcg;
473 if (mem_cgroup_disabled())
477 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
478 if (unlikely(!memcg))
483 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
486 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
494 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
495 void mem_cgroup_split_huge_fixup(struct page *head);
498 #else /* CONFIG_MEMCG */
501 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
502 enum mem_cgroup_events_index idx,
507 static inline bool mem_cgroup_low(struct mem_cgroup *root,
508 struct mem_cgroup *memcg)
513 static inline int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
515 struct mem_cgroup **memcgp,
522 static inline void mem_cgroup_commit_charge(struct page *page,
523 struct mem_cgroup *memcg,
524 bool lrucare, bool compound)
528 static inline void mem_cgroup_cancel_charge(struct page *page,
529 struct mem_cgroup *memcg,
534 static inline void mem_cgroup_uncharge(struct page *page)
538 static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
542 static inline void mem_cgroup_replace_page(struct page *old, struct page *new)
546 static inline struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
547 struct mem_cgroup *memcg)
549 return &zone->lruvec;
552 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
555 return &zone->lruvec;
558 static inline bool mm_match_cgroup(struct mm_struct *mm,
559 struct mem_cgroup *memcg)
564 static inline bool task_in_mem_cgroup(struct task_struct *task,
565 const struct mem_cgroup *memcg)
570 static inline struct mem_cgroup *
571 mem_cgroup_iter(struct mem_cgroup *root,
572 struct mem_cgroup *prev,
573 struct mem_cgroup_reclaim_cookie *reclaim)
578 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
579 struct mem_cgroup *prev)
583 static inline bool mem_cgroup_disabled(void)
588 static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
594 mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
599 static inline unsigned long
600 mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
606 mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
612 mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
616 static inline struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
621 static inline void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
625 static inline void mem_cgroup_handle_over_high(void)
629 static inline void mem_cgroup_oom_enable(void)
633 static inline void mem_cgroup_oom_disable(void)
637 static inline bool task_in_memcg_oom(struct task_struct *p)
642 static inline bool mem_cgroup_oom_synchronize(bool wait)
647 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
648 enum mem_cgroup_stat_index idx)
652 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
653 enum mem_cgroup_stat_index idx)
658 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
660 unsigned long *total_scanned)
665 static inline void mem_cgroup_split_huge_fixup(struct page *head)
670 void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
673 #endif /* CONFIG_MEMCG */
675 #ifdef CONFIG_CGROUP_WRITEBACK
677 struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg);
678 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
679 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
680 unsigned long *pheadroom, unsigned long *pdirty,
681 unsigned long *pwriteback);
683 #else /* CONFIG_CGROUP_WRITEBACK */
685 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
690 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
691 unsigned long *pfilepages,
692 unsigned long *pheadroom,
693 unsigned long *pdirty,
694 unsigned long *pwriteback)
698 #endif /* CONFIG_CGROUP_WRITEBACK */
701 void sock_update_memcg(struct sock *sk);
702 void sock_release_memcg(struct sock *sk);
703 bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
704 void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
706 extern struct static_key_false memcg_sockets_enabled_key;
707 #define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_enabled_key)
708 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
710 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_pressure)
713 if (time_before(jiffies, memcg->socket_pressure))
715 } while ((memcg = parent_mem_cgroup(memcg)));
719 #define mem_cgroup_sockets_enabled 0
720 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
726 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
727 extern struct static_key_false memcg_kmem_enabled_key;
729 extern int memcg_nr_cache_ids;
730 void memcg_get_cache_ids(void);
731 void memcg_put_cache_ids(void);
734 * Helper macro to loop through all memcg-specific caches. Callers must still
735 * check if the cache is valid (it is either valid or NULL).
736 * the slab_mutex must be held when looping through those caches
738 #define for_each_memcg_cache_index(_idx) \
739 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
741 static inline bool memcg_kmem_enabled(void)
743 return static_branch_unlikely(&memcg_kmem_enabled_key);
746 static inline bool memcg_kmem_online(struct mem_cgroup *memcg)
748 return memcg->kmem_state == KMEM_ONLINE;
752 * In general, we'll do everything in our power to not incur in any overhead
753 * for non-memcg users for the kmem functions. Not even a function call, if we
756 * Therefore, we'll inline all those functions so that in the best case, we'll
757 * see that kmemcg is off for everybody and proceed quickly. If it is on,
758 * we'll still do most of the flag checking inline. We check a lot of
759 * conditions, but because they are pretty simple, they are expected to be
762 int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
763 struct mem_cgroup *memcg);
764 int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order);
765 void __memcg_kmem_uncharge(struct page *page, int order);
768 * helper for acessing a memcg's index. It will be used as an index in the
769 * child cache array in kmem_cache, and also to derive its name. This function
770 * will return -1 when this is not a kmem-limited memcg.
772 static inline int memcg_cache_id(struct mem_cgroup *memcg)
774 return memcg ? memcg->kmemcg_id : -1;
777 struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp);
778 void __memcg_kmem_put_cache(struct kmem_cache *cachep);
780 static inline bool __memcg_kmem_bypass(void)
782 if (!memcg_kmem_enabled())
784 if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
790 * memcg_kmem_charge: charge a kmem page
791 * @page: page to charge
793 * @order: allocation order
795 * Returns 0 on success, an error code on failure.
797 static __always_inline int memcg_kmem_charge(struct page *page,
798 gfp_t gfp, int order)
800 if (__memcg_kmem_bypass())
802 if (!(gfp & __GFP_ACCOUNT))
804 return __memcg_kmem_charge(page, gfp, order);
808 * memcg_kmem_uncharge: uncharge a kmem page
809 * @page: page to uncharge
810 * @order: allocation order
812 static __always_inline void memcg_kmem_uncharge(struct page *page, int order)
814 if (memcg_kmem_enabled())
815 __memcg_kmem_uncharge(page, order);
819 * memcg_kmem_get_cache: selects the correct per-memcg cache for allocation
820 * @cachep: the original global kmem cache
822 * All memory allocated from a per-memcg cache is charged to the owner memcg.
824 static __always_inline struct kmem_cache *
825 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
827 if (__memcg_kmem_bypass())
829 return __memcg_kmem_get_cache(cachep, gfp);
832 static __always_inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
834 if (memcg_kmem_enabled())
835 __memcg_kmem_put_cache(cachep);
838 #define for_each_memcg_cache_index(_idx) \
841 static inline bool memcg_kmem_enabled(void)
846 static inline bool memcg_kmem_online(struct mem_cgroup *memcg)
851 static inline int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
856 static inline void memcg_kmem_uncharge(struct page *page, int order)
860 static inline int memcg_cache_id(struct mem_cgroup *memcg)
865 static inline void memcg_get_cache_ids(void)
869 static inline void memcg_put_cache_ids(void)
873 static inline struct kmem_cache *
874 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
879 static inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
882 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
884 #endif /* _LINUX_MEMCONTROL_H */