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. */
95 struct mem_cgroup_stat_cpu {
96 long count[MEM_CGROUP_STAT_NSTATS];
97 unsigned long events[MEMCG_NR_EVENTS];
98 unsigned long nr_page_events;
99 unsigned long targets[MEM_CGROUP_NTARGETS];
102 struct mem_cgroup_reclaim_iter {
103 struct mem_cgroup *position;
104 /* scan generation, increased every round-trip */
105 unsigned int generation;
109 * per-zone information in memory controller.
111 struct mem_cgroup_per_zone {
112 struct lruvec lruvec;
113 unsigned long lru_size[NR_LRU_LISTS];
115 struct mem_cgroup_reclaim_iter iter[DEF_PRIORITY + 1];
117 struct rb_node tree_node; /* RB tree node */
118 unsigned long usage_in_excess;/* Set to the value by which */
119 /* the soft limit is exceeded*/
121 struct mem_cgroup *memcg; /* Back pointer, we cannot */
122 /* use container_of */
125 struct mem_cgroup_per_node {
126 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
129 struct mem_cgroup_threshold {
130 struct eventfd_ctx *eventfd;
131 unsigned long threshold;
135 struct mem_cgroup_threshold_ary {
136 /* An array index points to threshold just below or equal to usage. */
137 int current_threshold;
138 /* Size of entries[] */
140 /* Array of thresholds */
141 struct mem_cgroup_threshold entries[0];
144 struct mem_cgroup_thresholds {
145 /* Primary thresholds array */
146 struct mem_cgroup_threshold_ary *primary;
148 * Spare threshold array.
149 * This is needed to make mem_cgroup_unregister_event() "never fail".
150 * It must be able to store at least primary->size - 1 entries.
152 struct mem_cgroup_threshold_ary *spare;
155 enum memcg_kmem_state {
162 * The memory controller data structure. The memory controller controls both
163 * page cache and RSS per cgroup. We would eventually like to provide
164 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
165 * to help the administrator determine what knobs to tune.
168 struct cgroup_subsys_state css;
170 /* Accounted resources */
171 struct page_counter memory;
172 struct page_counter memsw;
173 struct page_counter kmem;
175 /* Normal memory consumption range */
179 /* Range enforcement for interrupt charges */
180 struct work_struct high_work;
182 unsigned long soft_limit;
184 /* vmpressure notifications */
185 struct vmpressure vmpressure;
187 /* css_online() has been completed */
191 * Should the accounting and control be hierarchical, per subtree?
195 /* protected by memcg_oom_lock */
200 /* OOM-Killer disable */
201 int oom_kill_disable;
203 /* handle for "memory.events" */
204 struct cgroup_file events_file;
206 /* protect arrays of thresholds */
207 struct mutex thresholds_lock;
209 /* thresholds for memory usage. RCU-protected */
210 struct mem_cgroup_thresholds thresholds;
212 /* thresholds for mem+swap usage. RCU-protected */
213 struct mem_cgroup_thresholds memsw_thresholds;
215 /* For oom notifier event fd */
216 struct list_head oom_notify;
219 * Should we move charges of a task when a task is moved into this
220 * mem_cgroup ? And what type of charges should we move ?
222 unsigned long move_charge_at_immigrate;
224 * set > 0 if pages under this cgroup are moving to other cgroup.
226 atomic_t moving_account;
227 /* taken only while moving_account > 0 */
228 spinlock_t move_lock;
229 struct task_struct *move_lock_task;
230 unsigned long move_lock_flags;
234 struct mem_cgroup_stat_cpu __percpu *stat;
236 #if defined(CONFIG_MEMCG_LEGACY_KMEM) && defined(CONFIG_INET)
237 struct cg_proto tcp_mem;
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;
258 unsigned long socket_pressure;
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,
290 void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
291 bool lrucare, bool compound);
292 void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
294 void mem_cgroup_uncharge(struct page *page);
295 void mem_cgroup_uncharge_list(struct list_head *page_list);
297 void mem_cgroup_replace_page(struct page *oldpage, struct page *newpage);
299 struct lruvec *mem_cgroup_zone_lruvec(struct zone *, struct mem_cgroup *);
300 struct lruvec *mem_cgroup_page_lruvec(struct page *, struct zone *);
302 bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg);
303 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
306 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
307 return css ? container_of(css, struct mem_cgroup, css) : NULL;
310 #define mem_cgroup_from_counter(counter, member) \
311 container_of(counter, struct mem_cgroup, member)
313 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
315 struct mem_cgroup_reclaim_cookie *);
316 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
319 * parent_mem_cgroup - find the accounting parent of a memcg
320 * @memcg: memcg whose parent to find
322 * Returns the parent memcg, or NULL if this is the root or the memory
323 * controller is in legacy no-hierarchy mode.
325 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
327 if (!memcg->memory.parent)
329 return mem_cgroup_from_counter(memcg->memory.parent, memory);
332 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
333 struct mem_cgroup *root)
337 if (!root->use_hierarchy)
339 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
342 static inline bool mm_match_cgroup(struct mm_struct *mm,
343 struct mem_cgroup *memcg)
345 struct mem_cgroup *task_memcg;
349 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
351 match = mem_cgroup_is_descendant(task_memcg, memcg);
356 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
357 ino_t page_cgroup_ino(struct page *page);
359 static inline bool mem_cgroup_disabled(void)
361 return !cgroup_subsys_enabled(memory_cgrp_subsys);
365 * For memory reclaim.
367 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
369 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
372 static inline bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
374 struct mem_cgroup_per_zone *mz;
375 struct mem_cgroup *memcg;
377 if (mem_cgroup_disabled())
380 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
383 return !!(memcg->css.flags & CSS_ONLINE);
387 unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
389 struct mem_cgroup_per_zone *mz;
391 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
392 return mz->lru_size[lru];
395 static inline bool mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
397 unsigned long inactive_ratio;
398 unsigned long inactive;
399 unsigned long active;
402 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
403 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
405 gb = (inactive + active) >> (30 - PAGE_SHIFT);
407 inactive_ratio = int_sqrt(10 * gb);
411 return inactive * inactive_ratio < active;
414 void mem_cgroup_handle_over_high(void);
416 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
417 struct task_struct *p);
419 static inline void mem_cgroup_oom_enable(void)
421 WARN_ON(current->memcg_may_oom);
422 current->memcg_may_oom = 1;
425 static inline void mem_cgroup_oom_disable(void)
427 WARN_ON(!current->memcg_may_oom);
428 current->memcg_may_oom = 0;
431 static inline bool task_in_memcg_oom(struct task_struct *p)
433 return p->memcg_in_oom;
436 bool mem_cgroup_oom_synchronize(bool wait);
438 #ifdef CONFIG_MEMCG_SWAP
439 extern int do_swap_account;
442 struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page);
443 void mem_cgroup_end_page_stat(struct mem_cgroup *memcg);
446 * mem_cgroup_update_page_stat - update page state statistics
447 * @memcg: memcg to account against
448 * @idx: page state item to account
449 * @val: number of pages (positive or negative)
451 * See mem_cgroup_begin_page_stat() for locking requirements.
453 static inline void mem_cgroup_update_page_stat(struct mem_cgroup *memcg,
454 enum mem_cgroup_stat_index idx, int val)
456 VM_BUG_ON(!rcu_read_lock_held());
459 this_cpu_add(memcg->stat->count[idx], val);
462 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
463 enum mem_cgroup_stat_index idx)
465 mem_cgroup_update_page_stat(memcg, idx, 1);
468 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
469 enum mem_cgroup_stat_index idx)
471 mem_cgroup_update_page_stat(memcg, idx, -1);
474 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
476 unsigned long *total_scanned);
478 static inline void mem_cgroup_count_vm_event(struct mm_struct *mm,
479 enum vm_event_item idx)
481 struct mem_cgroup *memcg;
483 if (mem_cgroup_disabled())
487 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
488 if (unlikely(!memcg))
493 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
496 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
504 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
505 void mem_cgroup_split_huge_fixup(struct page *head);
508 #else /* CONFIG_MEMCG */
511 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
512 enum mem_cgroup_events_index idx,
517 static inline bool mem_cgroup_low(struct mem_cgroup *root,
518 struct mem_cgroup *memcg)
523 static inline int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
525 struct mem_cgroup **memcgp,
532 static inline void mem_cgroup_commit_charge(struct page *page,
533 struct mem_cgroup *memcg,
534 bool lrucare, bool compound)
538 static inline void mem_cgroup_cancel_charge(struct page *page,
539 struct mem_cgroup *memcg,
544 static inline void mem_cgroup_uncharge(struct page *page)
548 static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
552 static inline void mem_cgroup_replace_page(struct page *old, struct page *new)
556 static inline struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
557 struct mem_cgroup *memcg)
559 return &zone->lruvec;
562 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
565 return &zone->lruvec;
568 static inline bool mm_match_cgroup(struct mm_struct *mm,
569 struct mem_cgroup *memcg)
574 static inline bool task_in_mem_cgroup(struct task_struct *task,
575 const struct mem_cgroup *memcg)
580 static inline struct mem_cgroup *
581 mem_cgroup_iter(struct mem_cgroup *root,
582 struct mem_cgroup *prev,
583 struct mem_cgroup_reclaim_cookie *reclaim)
588 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
589 struct mem_cgroup *prev)
593 static inline bool mem_cgroup_disabled(void)
599 mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
604 static inline bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
609 static inline unsigned long
610 mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
616 mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
622 mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
626 static inline struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
631 static inline void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
635 static inline void mem_cgroup_handle_over_high(void)
639 static inline void mem_cgroup_oom_enable(void)
643 static inline void mem_cgroup_oom_disable(void)
647 static inline bool task_in_memcg_oom(struct task_struct *p)
652 static inline bool mem_cgroup_oom_synchronize(bool wait)
657 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
658 enum mem_cgroup_stat_index idx)
662 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
663 enum mem_cgroup_stat_index idx)
668 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
670 unsigned long *total_scanned)
675 static inline void mem_cgroup_split_huge_fixup(struct page *head)
680 void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
683 #endif /* CONFIG_MEMCG */
685 #ifdef CONFIG_CGROUP_WRITEBACK
687 struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg);
688 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
689 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
690 unsigned long *pheadroom, unsigned long *pdirty,
691 unsigned long *pwriteback);
693 #else /* CONFIG_CGROUP_WRITEBACK */
695 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
700 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
701 unsigned long *pfilepages,
702 unsigned long *pheadroom,
703 unsigned long *pdirty,
704 unsigned long *pwriteback)
708 #endif /* CONFIG_CGROUP_WRITEBACK */
711 void sock_update_memcg(struct sock *sk);
712 void sock_release_memcg(struct sock *sk);
713 bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
714 void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
715 #if defined(CONFIG_MEMCG) && defined(CONFIG_INET)
716 extern struct static_key_false memcg_sockets_enabled_key;
717 #define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_enabled_key)
718 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
720 #ifdef CONFIG_MEMCG_LEGACY_KMEM
721 if (memcg->tcp_mem.memory_pressure)
725 if (time_before(jiffies, memcg->socket_pressure))
727 } while ((memcg = parent_mem_cgroup(memcg)));
731 #define mem_cgroup_sockets_enabled 0
732 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
738 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
739 extern struct static_key_false memcg_kmem_enabled_key;
741 extern int memcg_nr_cache_ids;
742 void memcg_get_cache_ids(void);
743 void memcg_put_cache_ids(void);
746 * Helper macro to loop through all memcg-specific caches. Callers must still
747 * check if the cache is valid (it is either valid or NULL).
748 * the slab_mutex must be held when looping through those caches
750 #define for_each_memcg_cache_index(_idx) \
751 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
753 static inline bool memcg_kmem_enabled(void)
755 return static_branch_unlikely(&memcg_kmem_enabled_key);
758 static inline bool memcg_kmem_online(struct mem_cgroup *memcg)
760 return memcg->kmem_state == KMEM_ONLINE;
764 * In general, we'll do everything in our power to not incur in any overhead
765 * for non-memcg users for the kmem functions. Not even a function call, if we
768 * Therefore, we'll inline all those functions so that in the best case, we'll
769 * see that kmemcg is off for everybody and proceed quickly. If it is on,
770 * we'll still do most of the flag checking inline. We check a lot of
771 * conditions, but because they are pretty simple, they are expected to be
774 int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
775 struct mem_cgroup *memcg);
776 int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order);
777 void __memcg_kmem_uncharge(struct page *page, int order);
780 * helper for acessing a memcg's index. It will be used as an index in the
781 * child cache array in kmem_cache, and also to derive its name. This function
782 * will return -1 when this is not a kmem-limited memcg.
784 static inline int memcg_cache_id(struct mem_cgroup *memcg)
786 return memcg ? memcg->kmemcg_id : -1;
789 struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp);
790 void __memcg_kmem_put_cache(struct kmem_cache *cachep);
792 static inline bool __memcg_kmem_bypass(void)
794 if (!memcg_kmem_enabled())
796 if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
802 * memcg_kmem_charge: charge a kmem page
803 * @page: page to charge
805 * @order: allocation order
807 * Returns 0 on success, an error code on failure.
809 static __always_inline int memcg_kmem_charge(struct page *page,
810 gfp_t gfp, int order)
812 if (__memcg_kmem_bypass())
814 if (!(gfp & __GFP_ACCOUNT))
816 return __memcg_kmem_charge(page, gfp, order);
820 * memcg_kmem_uncharge: uncharge a kmem page
821 * @page: page to uncharge
822 * @order: allocation order
824 static __always_inline void memcg_kmem_uncharge(struct page *page, int order)
826 if (memcg_kmem_enabled())
827 __memcg_kmem_uncharge(page, order);
831 * memcg_kmem_get_cache: selects the correct per-memcg cache for allocation
832 * @cachep: the original global kmem cache
834 * All memory allocated from a per-memcg cache is charged to the owner memcg.
836 static __always_inline struct kmem_cache *
837 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
839 if (__memcg_kmem_bypass())
841 return __memcg_kmem_get_cache(cachep, gfp);
844 static __always_inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
846 if (memcg_kmem_enabled())
847 __memcg_kmem_put_cache(cachep);
850 #define for_each_memcg_cache_index(_idx) \
853 static inline bool memcg_kmem_enabled(void)
858 static inline bool memcg_kmem_online(struct mem_cgroup *memcg)
863 static inline int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
868 static inline void memcg_kmem_uncharge(struct page *page, int order)
872 static inline int memcg_cache_id(struct mem_cgroup *memcg)
877 static inline void memcg_get_cache_ids(void)
881 static inline void memcg_put_cache_ids(void)
885 static inline struct kmem_cache *
886 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
891 static inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
894 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
896 #endif /* _LINUX_MEMCONTROL_H */