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 page_counter memory_allocated; /* Current allocated memory. */
90 struct percpu_counter sockets_allocated; /* Current number of sockets. */
95 * memcg field is used to find which memcg we belong directly
96 * Each memcg struct can hold more than one cg_proto, so container_of
99 * The elegant solution would be having an inverse function to
100 * proto_cgroup in struct proto, but that means polluting the structure
101 * for everybody, instead of just for memcg users.
103 struct mem_cgroup *memcg;
107 struct mem_cgroup_stat_cpu {
108 long count[MEM_CGROUP_STAT_NSTATS];
109 unsigned long events[MEMCG_NR_EVENTS];
110 unsigned long nr_page_events;
111 unsigned long targets[MEM_CGROUP_NTARGETS];
114 struct mem_cgroup_reclaim_iter {
115 struct mem_cgroup *position;
116 /* scan generation, increased every round-trip */
117 unsigned int generation;
121 * per-zone information in memory controller.
123 struct mem_cgroup_per_zone {
124 struct lruvec lruvec;
125 unsigned long lru_size[NR_LRU_LISTS];
127 struct mem_cgroup_reclaim_iter iter[DEF_PRIORITY + 1];
129 struct rb_node tree_node; /* RB tree node */
130 unsigned long usage_in_excess;/* Set to the value by which */
131 /* the soft limit is exceeded*/
133 struct mem_cgroup *memcg; /* Back pointer, we cannot */
134 /* use container_of */
137 struct mem_cgroup_per_node {
138 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
141 struct mem_cgroup_threshold {
142 struct eventfd_ctx *eventfd;
143 unsigned long threshold;
147 struct mem_cgroup_threshold_ary {
148 /* An array index points to threshold just below or equal to usage. */
149 int current_threshold;
150 /* Size of entries[] */
152 /* Array of thresholds */
153 struct mem_cgroup_threshold entries[0];
156 struct mem_cgroup_thresholds {
157 /* Primary thresholds array */
158 struct mem_cgroup_threshold_ary *primary;
160 * Spare threshold array.
161 * This is needed to make mem_cgroup_unregister_event() "never fail".
162 * It must be able to store at least primary->size - 1 entries.
164 struct mem_cgroup_threshold_ary *spare;
168 * The memory controller data structure. The memory controller controls both
169 * page cache and RSS per cgroup. We would eventually like to provide
170 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
171 * to help the administrator determine what knobs to tune.
174 struct cgroup_subsys_state css;
176 /* Accounted resources */
177 struct page_counter memory;
178 struct page_counter memsw;
179 struct page_counter kmem;
181 /* Normal memory consumption range */
185 unsigned long soft_limit;
187 /* vmpressure notifications */
188 struct vmpressure vmpressure;
190 /* css_online() has been completed */
194 * Should the accounting and control be hierarchical, per subtree?
198 /* protected by memcg_oom_lock */
203 /* OOM-Killer disable */
204 int oom_kill_disable;
206 /* handle for "memory.events" */
207 struct cgroup_file events_file;
209 /* protect arrays of thresholds */
210 struct mutex thresholds_lock;
212 /* thresholds for memory usage. RCU-protected */
213 struct mem_cgroup_thresholds thresholds;
215 /* thresholds for mem+swap usage. RCU-protected */
216 struct mem_cgroup_thresholds memsw_thresholds;
218 /* For oom notifier event fd */
219 struct list_head oom_notify;
222 * Should we move charges of a task when a task is moved into this
223 * mem_cgroup ? And what type of charges should we move ?
225 unsigned long move_charge_at_immigrate;
227 * set > 0 if pages under this cgroup are moving to other cgroup.
229 atomic_t moving_account;
230 /* taken only while moving_account > 0 */
231 spinlock_t move_lock;
232 struct task_struct *move_lock_task;
233 unsigned long move_lock_flags;
237 struct mem_cgroup_stat_cpu __percpu *stat;
239 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET)
240 struct cg_proto tcp_mem;
242 #if defined(CONFIG_MEMCG_KMEM)
243 /* Index in the kmem_cache->memcg_params.memcg_caches array */
245 bool kmem_acct_activated;
246 bool kmem_acct_active;
249 int last_scanned_node;
251 nodemask_t scan_nodes;
252 atomic_t numainfo_events;
253 atomic_t numainfo_updating;
256 #ifdef CONFIG_CGROUP_WRITEBACK
257 struct list_head cgwb_list;
258 struct wb_domain cgwb_domain;
261 /* List of events which userspace want to receive */
262 struct list_head event_list;
263 spinlock_t event_list_lock;
265 struct mem_cgroup_per_node *nodeinfo[0];
266 /* WARNING: nodeinfo must be the last member here */
269 extern struct mem_cgroup *root_mem_cgroup;
272 * mem_cgroup_events - count memory events against a cgroup
273 * @memcg: the memory cgroup
274 * @idx: the event index
275 * @nr: the number of events to account for
277 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
278 enum mem_cgroup_events_index idx,
281 this_cpu_add(memcg->stat->events[idx], nr);
282 cgroup_file_notify(&memcg->events_file);
285 bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg);
287 int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
288 gfp_t gfp_mask, struct mem_cgroup **memcgp);
289 void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
291 void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg);
292 void mem_cgroup_uncharge(struct page *page);
293 void mem_cgroup_uncharge_list(struct list_head *page_list);
295 void mem_cgroup_replace_page(struct page *oldpage, struct page *newpage);
297 struct lruvec *mem_cgroup_zone_lruvec(struct zone *, struct mem_cgroup *);
298 struct lruvec *mem_cgroup_page_lruvec(struct page *, struct zone *);
300 bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg);
301 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
302 struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg);
305 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
306 return css ? container_of(css, struct mem_cgroup, css) : NULL;
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 *);
314 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
315 struct mem_cgroup *root)
319 if (!root->use_hierarchy)
321 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
324 static inline bool mm_match_cgroup(struct mm_struct *mm,
325 struct mem_cgroup *memcg)
327 struct mem_cgroup *task_memcg;
331 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
333 match = mem_cgroup_is_descendant(task_memcg, memcg);
338 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
339 ino_t page_cgroup_ino(struct page *page);
341 static inline bool mem_cgroup_disabled(void)
343 return !cgroup_subsys_enabled(memory_cgrp_subsys);
347 * For memory reclaim.
349 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
351 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
354 static inline bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
356 struct mem_cgroup_per_zone *mz;
357 struct mem_cgroup *memcg;
359 if (mem_cgroup_disabled())
362 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
365 return !!(memcg->css.flags & CSS_ONLINE);
369 unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
371 struct mem_cgroup_per_zone *mz;
373 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
374 return mz->lru_size[lru];
377 static inline bool mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
379 unsigned long inactive_ratio;
380 unsigned long inactive;
381 unsigned long active;
384 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
385 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
387 gb = (inactive + active) >> (30 - PAGE_SHIFT);
389 inactive_ratio = int_sqrt(10 * gb);
393 return inactive * inactive_ratio < active;
396 void mem_cgroup_handle_over_high(void);
398 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
399 struct task_struct *p);
401 static inline void mem_cgroup_oom_enable(void)
403 WARN_ON(current->memcg_may_oom);
404 current->memcg_may_oom = 1;
407 static inline void mem_cgroup_oom_disable(void)
409 WARN_ON(!current->memcg_may_oom);
410 current->memcg_may_oom = 0;
413 static inline bool task_in_memcg_oom(struct task_struct *p)
415 return p->memcg_in_oom;
418 bool mem_cgroup_oom_synchronize(bool wait);
420 #ifdef CONFIG_MEMCG_SWAP
421 extern int do_swap_account;
424 struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page);
425 void mem_cgroup_end_page_stat(struct mem_cgroup *memcg);
428 * mem_cgroup_update_page_stat - update page state statistics
429 * @memcg: memcg to account against
430 * @idx: page state item to account
431 * @val: number of pages (positive or negative)
433 * See mem_cgroup_begin_page_stat() for locking requirements.
435 static inline void mem_cgroup_update_page_stat(struct mem_cgroup *memcg,
436 enum mem_cgroup_stat_index idx, int val)
438 VM_BUG_ON(!rcu_read_lock_held());
441 this_cpu_add(memcg->stat->count[idx], val);
444 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
445 enum mem_cgroup_stat_index idx)
447 mem_cgroup_update_page_stat(memcg, idx, 1);
450 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
451 enum mem_cgroup_stat_index idx)
453 mem_cgroup_update_page_stat(memcg, idx, -1);
456 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
458 unsigned long *total_scanned);
460 static inline void mem_cgroup_count_vm_event(struct mm_struct *mm,
461 enum vm_event_item idx)
463 struct mem_cgroup *memcg;
465 if (mem_cgroup_disabled())
469 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
470 if (unlikely(!memcg))
475 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
478 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
486 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
487 void mem_cgroup_split_huge_fixup(struct page *head);
490 #else /* CONFIG_MEMCG */
493 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
494 enum mem_cgroup_events_index idx,
499 static inline bool mem_cgroup_low(struct mem_cgroup *root,
500 struct mem_cgroup *memcg)
505 static inline int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
507 struct mem_cgroup **memcgp)
513 static inline void mem_cgroup_commit_charge(struct page *page,
514 struct mem_cgroup *memcg,
519 static inline void mem_cgroup_cancel_charge(struct page *page,
520 struct mem_cgroup *memcg)
524 static inline void mem_cgroup_uncharge(struct page *page)
528 static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
532 static inline void mem_cgroup_replace_page(struct page *old, struct page *new)
536 static inline struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
537 struct mem_cgroup *memcg)
539 return &zone->lruvec;
542 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
545 return &zone->lruvec;
548 static inline bool mm_match_cgroup(struct mm_struct *mm,
549 struct mem_cgroup *memcg)
554 static inline bool task_in_mem_cgroup(struct task_struct *task,
555 const struct mem_cgroup *memcg)
560 static inline struct mem_cgroup *
561 mem_cgroup_iter(struct mem_cgroup *root,
562 struct mem_cgroup *prev,
563 struct mem_cgroup_reclaim_cookie *reclaim)
568 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
569 struct mem_cgroup *prev)
573 static inline bool mem_cgroup_disabled(void)
579 mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
584 static inline bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
589 static inline unsigned long
590 mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
596 mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
602 mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
606 static inline struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
611 static inline void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
615 static inline void mem_cgroup_handle_over_high(void)
619 static inline void mem_cgroup_oom_enable(void)
623 static inline void mem_cgroup_oom_disable(void)
627 static inline bool task_in_memcg_oom(struct task_struct *p)
632 static inline bool mem_cgroup_oom_synchronize(bool wait)
637 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
638 enum mem_cgroup_stat_index idx)
642 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
643 enum mem_cgroup_stat_index idx)
648 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
650 unsigned long *total_scanned)
655 static inline void mem_cgroup_split_huge_fixup(struct page *head)
660 void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
663 #endif /* CONFIG_MEMCG */
671 #ifdef CONFIG_CGROUP_WRITEBACK
673 struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg);
674 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
675 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
676 unsigned long *pheadroom, unsigned long *pdirty,
677 unsigned long *pwriteback);
679 #else /* CONFIG_CGROUP_WRITEBACK */
681 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
686 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
687 unsigned long *pfilepages,
688 unsigned long *pheadroom,
689 unsigned long *pdirty,
690 unsigned long *pwriteback)
694 #endif /* CONFIG_CGROUP_WRITEBACK */
697 #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
698 void sock_update_memcg(struct sock *sk);
699 void sock_release_memcg(struct sock *sk);
701 static inline void sock_update_memcg(struct sock *sk)
704 static inline void sock_release_memcg(struct sock *sk)
707 #endif /* CONFIG_INET && CONFIG_MEMCG_KMEM */
709 #ifdef CONFIG_MEMCG_KMEM
710 extern struct static_key memcg_kmem_enabled_key;
712 extern int memcg_nr_cache_ids;
713 void memcg_get_cache_ids(void);
714 void memcg_put_cache_ids(void);
717 * Helper macro to loop through all memcg-specific caches. Callers must still
718 * check if the cache is valid (it is either valid or NULL).
719 * the slab_mutex must be held when looping through those caches
721 #define for_each_memcg_cache_index(_idx) \
722 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
724 static inline bool memcg_kmem_enabled(void)
726 return static_key_false(&memcg_kmem_enabled_key);
729 static inline bool memcg_kmem_is_active(struct mem_cgroup *memcg)
731 return memcg->kmem_acct_active;
735 * In general, we'll do everything in our power to not incur in any overhead
736 * for non-memcg users for the kmem functions. Not even a function call, if we
739 * Therefore, we'll inline all those functions so that in the best case, we'll
740 * see that kmemcg is off for everybody and proceed quickly. If it is on,
741 * we'll still do most of the flag checking inline. We check a lot of
742 * conditions, but because they are pretty simple, they are expected to be
745 int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
746 struct mem_cgroup *memcg);
747 int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order);
748 void __memcg_kmem_uncharge(struct page *page, int order);
751 * helper for acessing a memcg's index. It will be used as an index in the
752 * child cache array in kmem_cache, and also to derive its name. This function
753 * will return -1 when this is not a kmem-limited memcg.
755 static inline int memcg_cache_id(struct mem_cgroup *memcg)
757 return memcg ? memcg->kmemcg_id : -1;
760 struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp);
761 void __memcg_kmem_put_cache(struct kmem_cache *cachep);
763 static inline bool __memcg_kmem_bypass(void)
765 if (!memcg_kmem_enabled())
767 if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
773 * memcg_kmem_charge: charge a kmem page
774 * @page: page to charge
776 * @order: allocation order
778 * Returns 0 on success, an error code on failure.
780 static __always_inline int memcg_kmem_charge(struct page *page,
781 gfp_t gfp, int order)
783 if (__memcg_kmem_bypass())
785 if (!(gfp & __GFP_ACCOUNT))
787 return __memcg_kmem_charge(page, gfp, order);
791 * memcg_kmem_uncharge: uncharge a kmem page
792 * @page: page to uncharge
793 * @order: allocation order
795 static __always_inline void memcg_kmem_uncharge(struct page *page, int order)
797 if (memcg_kmem_enabled())
798 __memcg_kmem_uncharge(page, order);
802 * memcg_kmem_get_cache: selects the correct per-memcg cache for allocation
803 * @cachep: the original global kmem cache
805 * All memory allocated from a per-memcg cache is charged to the owner memcg.
807 static __always_inline struct kmem_cache *
808 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
810 if (__memcg_kmem_bypass())
812 return __memcg_kmem_get_cache(cachep, gfp);
815 static __always_inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
817 if (memcg_kmem_enabled())
818 __memcg_kmem_put_cache(cachep);
821 #define for_each_memcg_cache_index(_idx) \
824 static inline bool memcg_kmem_enabled(void)
829 static inline bool memcg_kmem_is_active(struct mem_cgroup *memcg)
834 static inline int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
839 static inline void memcg_kmem_uncharge(struct page *page, int order)
843 static inline int memcg_cache_id(struct mem_cgroup *memcg)
848 static inline void memcg_get_cache_ids(void)
852 static inline void memcg_put_cache_ids(void)
856 static inline struct kmem_cache *
857 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
862 static inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
865 #endif /* CONFIG_MEMCG_KMEM */
866 #endif /* _LINUX_MEMCONTROL_H */