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. */
94 * memcg field is used to find which memcg we belong directly
95 * Each memcg struct can hold more than one cg_proto, so container_of
98 * The elegant solution would be having an inverse function to
99 * proto_cgroup in struct proto, but that means polluting the structure
100 * for everybody, instead of just for memcg users.
102 struct mem_cgroup *memcg;
106 struct mem_cgroup_stat_cpu {
107 long count[MEM_CGROUP_STAT_NSTATS];
108 unsigned long events[MEMCG_NR_EVENTS];
109 unsigned long nr_page_events;
110 unsigned long targets[MEM_CGROUP_NTARGETS];
113 struct mem_cgroup_reclaim_iter {
114 struct mem_cgroup *position;
115 /* scan generation, increased every round-trip */
116 unsigned int generation;
120 * per-zone information in memory controller.
122 struct mem_cgroup_per_zone {
123 struct lruvec lruvec;
124 unsigned long lru_size[NR_LRU_LISTS];
126 struct mem_cgroup_reclaim_iter iter[DEF_PRIORITY + 1];
128 struct rb_node tree_node; /* RB tree node */
129 unsigned long usage_in_excess;/* Set to the value by which */
130 /* the soft limit is exceeded*/
132 struct mem_cgroup *memcg; /* Back pointer, we cannot */
133 /* use container_of */
136 struct mem_cgroup_per_node {
137 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
140 struct mem_cgroup_threshold {
141 struct eventfd_ctx *eventfd;
142 unsigned long threshold;
146 struct mem_cgroup_threshold_ary {
147 /* An array index points to threshold just below or equal to usage. */
148 int current_threshold;
149 /* Size of entries[] */
151 /* Array of thresholds */
152 struct mem_cgroup_threshold entries[0];
155 struct mem_cgroup_thresholds {
156 /* Primary thresholds array */
157 struct mem_cgroup_threshold_ary *primary;
159 * Spare threshold array.
160 * This is needed to make mem_cgroup_unregister_event() "never fail".
161 * It must be able to store at least primary->size - 1 entries.
163 struct mem_cgroup_threshold_ary *spare;
167 * The memory controller data structure. The memory controller controls both
168 * page cache and RSS per cgroup. We would eventually like to provide
169 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
170 * to help the administrator determine what knobs to tune.
173 struct cgroup_subsys_state css;
175 /* Accounted resources */
176 struct page_counter memory;
177 struct page_counter memsw;
178 struct page_counter kmem;
180 /* Normal memory consumption range */
184 unsigned long soft_limit;
186 /* vmpressure notifications */
187 struct vmpressure vmpressure;
189 /* css_online() has been completed */
193 * Should the accounting and control be hierarchical, per subtree?
197 /* protected by memcg_oom_lock */
202 /* OOM-Killer disable */
203 int oom_kill_disable;
205 /* handle for "memory.events" */
206 struct cgroup_file events_file;
208 /* protect arrays of thresholds */
209 struct mutex thresholds_lock;
211 /* thresholds for memory usage. RCU-protected */
212 struct mem_cgroup_thresholds thresholds;
214 /* thresholds for mem+swap usage. RCU-protected */
215 struct mem_cgroup_thresholds memsw_thresholds;
217 /* For oom notifier event fd */
218 struct list_head oom_notify;
221 * Should we move charges of a task when a task is moved into this
222 * mem_cgroup ? And what type of charges should we move ?
224 unsigned long move_charge_at_immigrate;
226 * set > 0 if pages under this cgroup are moving to other cgroup.
228 atomic_t moving_account;
229 /* taken only while moving_account > 0 */
230 spinlock_t move_lock;
231 struct task_struct *move_lock_task;
232 unsigned long move_lock_flags;
236 struct mem_cgroup_stat_cpu __percpu *stat;
238 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET)
239 struct cg_proto tcp_mem;
241 #if defined(CONFIG_MEMCG_KMEM)
242 /* Index in the kmem_cache->memcg_params.memcg_caches array */
244 bool kmem_acct_activated;
245 bool kmem_acct_active;
248 int last_scanned_node;
250 nodemask_t scan_nodes;
251 atomic_t numainfo_events;
252 atomic_t numainfo_updating;
255 #ifdef CONFIG_CGROUP_WRITEBACK
256 struct list_head cgwb_list;
257 struct wb_domain cgwb_domain;
260 /* List of events which userspace want to receive */
261 struct list_head event_list;
262 spinlock_t event_list_lock;
264 struct mem_cgroup_per_node *nodeinfo[0];
265 /* WARNING: nodeinfo must be the last member here */
268 extern struct mem_cgroup *root_mem_cgroup;
271 * mem_cgroup_events - count memory events against a cgroup
272 * @memcg: the memory cgroup
273 * @idx: the event index
274 * @nr: the number of events to account for
276 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
277 enum mem_cgroup_events_index idx,
280 this_cpu_add(memcg->stat->events[idx], nr);
281 cgroup_file_notify(&memcg->events_file);
284 bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg);
286 int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
287 gfp_t gfp_mask, struct mem_cgroup **memcgp);
288 void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
290 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);
301 struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg);
304 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
305 return css ? container_of(css, struct mem_cgroup, css) : NULL;
308 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
310 struct mem_cgroup_reclaim_cookie *);
311 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
313 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
314 struct mem_cgroup *root)
318 if (!root->use_hierarchy)
320 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
323 static inline bool mm_match_cgroup(struct mm_struct *mm,
324 struct mem_cgroup *memcg)
326 struct mem_cgroup *task_memcg;
330 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
332 match = mem_cgroup_is_descendant(task_memcg, memcg);
337 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
338 ino_t page_cgroup_ino(struct page *page);
340 static inline bool mem_cgroup_disabled(void)
342 return !cgroup_subsys_enabled(memory_cgrp_subsys);
346 * For memory reclaim.
348 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
350 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
353 static inline bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
355 struct mem_cgroup_per_zone *mz;
356 struct mem_cgroup *memcg;
358 if (mem_cgroup_disabled())
361 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
364 return !!(memcg->css.flags & CSS_ONLINE);
368 unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
370 struct mem_cgroup_per_zone *mz;
372 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
373 return mz->lru_size[lru];
376 static inline bool mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
378 unsigned long inactive_ratio;
379 unsigned long inactive;
380 unsigned long active;
383 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
384 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
386 gb = (inactive + active) >> (30 - PAGE_SHIFT);
388 inactive_ratio = int_sqrt(10 * gb);
392 return inactive * inactive_ratio < active;
395 void mem_cgroup_handle_over_high(void);
397 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
398 struct task_struct *p);
400 static inline void mem_cgroup_oom_enable(void)
402 WARN_ON(current->memcg_may_oom);
403 current->memcg_may_oom = 1;
406 static inline void mem_cgroup_oom_disable(void)
408 WARN_ON(!current->memcg_may_oom);
409 current->memcg_may_oom = 0;
412 static inline bool task_in_memcg_oom(struct task_struct *p)
414 return p->memcg_in_oom;
417 bool mem_cgroup_oom_synchronize(bool wait);
419 #ifdef CONFIG_MEMCG_SWAP
420 extern int do_swap_account;
423 struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page);
424 void mem_cgroup_end_page_stat(struct mem_cgroup *memcg);
427 * mem_cgroup_update_page_stat - update page state statistics
428 * @memcg: memcg to account against
429 * @idx: page state item to account
430 * @val: number of pages (positive or negative)
432 * See mem_cgroup_begin_page_stat() for locking requirements.
434 static inline void mem_cgroup_update_page_stat(struct mem_cgroup *memcg,
435 enum mem_cgroup_stat_index idx, int val)
437 VM_BUG_ON(!rcu_read_lock_held());
440 this_cpu_add(memcg->stat->count[idx], val);
443 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
444 enum mem_cgroup_stat_index idx)
446 mem_cgroup_update_page_stat(memcg, idx, 1);
449 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
450 enum mem_cgroup_stat_index idx)
452 mem_cgroup_update_page_stat(memcg, idx, -1);
455 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
457 unsigned long *total_scanned);
459 static inline void mem_cgroup_count_vm_event(struct mm_struct *mm,
460 enum vm_event_item idx)
462 struct mem_cgroup *memcg;
464 if (mem_cgroup_disabled())
468 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
469 if (unlikely(!memcg))
474 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
477 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
485 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
486 void mem_cgroup_split_huge_fixup(struct page *head);
489 #else /* CONFIG_MEMCG */
492 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
493 enum mem_cgroup_events_index idx,
498 static inline bool mem_cgroup_low(struct mem_cgroup *root,
499 struct mem_cgroup *memcg)
504 static inline int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
506 struct mem_cgroup **memcgp)
512 static inline void mem_cgroup_commit_charge(struct page *page,
513 struct mem_cgroup *memcg,
518 static inline void mem_cgroup_cancel_charge(struct page *page,
519 struct mem_cgroup *memcg)
523 static inline void mem_cgroup_uncharge(struct page *page)
527 static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
531 static inline void mem_cgroup_replace_page(struct page *old, struct page *new)
535 static inline struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
536 struct mem_cgroup *memcg)
538 return &zone->lruvec;
541 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
544 return &zone->lruvec;
547 static inline bool mm_match_cgroup(struct mm_struct *mm,
548 struct mem_cgroup *memcg)
553 static inline bool task_in_mem_cgroup(struct task_struct *task,
554 const struct mem_cgroup *memcg)
559 static inline struct mem_cgroup *
560 mem_cgroup_iter(struct mem_cgroup *root,
561 struct mem_cgroup *prev,
562 struct mem_cgroup_reclaim_cookie *reclaim)
567 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
568 struct mem_cgroup *prev)
572 static inline bool mem_cgroup_disabled(void)
578 mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
583 static inline bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
588 static inline unsigned long
589 mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
595 mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
601 mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
605 static inline struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
610 static inline void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
614 static inline void mem_cgroup_handle_over_high(void)
618 static inline void mem_cgroup_oom_enable(void)
622 static inline void mem_cgroup_oom_disable(void)
626 static inline bool task_in_memcg_oom(struct task_struct *p)
631 static inline bool mem_cgroup_oom_synchronize(bool wait)
636 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
637 enum mem_cgroup_stat_index idx)
641 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
642 enum mem_cgroup_stat_index idx)
647 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
649 unsigned long *total_scanned)
654 static inline void mem_cgroup_split_huge_fixup(struct page *head)
659 void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
662 #endif /* CONFIG_MEMCG */
670 #ifdef CONFIG_CGROUP_WRITEBACK
672 struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg);
673 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
674 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
675 unsigned long *pheadroom, unsigned long *pdirty,
676 unsigned long *pwriteback);
678 #else /* CONFIG_CGROUP_WRITEBACK */
680 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
685 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
686 unsigned long *pfilepages,
687 unsigned long *pheadroom,
688 unsigned long *pdirty,
689 unsigned long *pwriteback)
693 #endif /* CONFIG_CGROUP_WRITEBACK */
696 void sock_update_memcg(struct sock *sk);
697 void sock_release_memcg(struct sock *sk);
699 #ifdef CONFIG_MEMCG_KMEM
700 extern struct static_key memcg_kmem_enabled_key;
702 extern int memcg_nr_cache_ids;
703 void memcg_get_cache_ids(void);
704 void memcg_put_cache_ids(void);
707 * Helper macro to loop through all memcg-specific caches. Callers must still
708 * check if the cache is valid (it is either valid or NULL).
709 * the slab_mutex must be held when looping through those caches
711 #define for_each_memcg_cache_index(_idx) \
712 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
714 static inline bool memcg_kmem_enabled(void)
716 return static_key_false(&memcg_kmem_enabled_key);
719 static inline bool memcg_kmem_is_active(struct mem_cgroup *memcg)
721 return memcg->kmem_acct_active;
725 * In general, we'll do everything in our power to not incur in any overhead
726 * for non-memcg users for the kmem functions. Not even a function call, if we
729 * Therefore, we'll inline all those functions so that in the best case, we'll
730 * see that kmemcg is off for everybody and proceed quickly. If it is on,
731 * we'll still do most of the flag checking inline. We check a lot of
732 * conditions, but because they are pretty simple, they are expected to be
735 int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
736 struct mem_cgroup *memcg);
737 int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order);
738 void __memcg_kmem_uncharge(struct page *page, int order);
741 * helper for acessing a memcg's index. It will be used as an index in the
742 * child cache array in kmem_cache, and also to derive its name. This function
743 * will return -1 when this is not a kmem-limited memcg.
745 static inline int memcg_cache_id(struct mem_cgroup *memcg)
747 return memcg ? memcg->kmemcg_id : -1;
750 struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp);
751 void __memcg_kmem_put_cache(struct kmem_cache *cachep);
753 static inline bool __memcg_kmem_bypass(void)
755 if (!memcg_kmem_enabled())
757 if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
763 * memcg_kmem_charge: charge a kmem page
764 * @page: page to charge
766 * @order: allocation order
768 * Returns 0 on success, an error code on failure.
770 static __always_inline int memcg_kmem_charge(struct page *page,
771 gfp_t gfp, int order)
773 if (__memcg_kmem_bypass())
775 if (!(gfp & __GFP_ACCOUNT))
777 return __memcg_kmem_charge(page, gfp, order);
781 * memcg_kmem_uncharge: uncharge a kmem page
782 * @page: page to uncharge
783 * @order: allocation order
785 static __always_inline void memcg_kmem_uncharge(struct page *page, int order)
787 if (memcg_kmem_enabled())
788 __memcg_kmem_uncharge(page, order);
792 * memcg_kmem_get_cache: selects the correct per-memcg cache for allocation
793 * @cachep: the original global kmem cache
795 * All memory allocated from a per-memcg cache is charged to the owner memcg.
797 static __always_inline struct kmem_cache *
798 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
800 if (__memcg_kmem_bypass())
802 return __memcg_kmem_get_cache(cachep, gfp);
805 static __always_inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
807 if (memcg_kmem_enabled())
808 __memcg_kmem_put_cache(cachep);
811 #define for_each_memcg_cache_index(_idx) \
814 static inline bool memcg_kmem_enabled(void)
819 static inline bool memcg_kmem_is_active(struct mem_cgroup *memcg)
824 static inline int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
829 static inline void memcg_kmem_uncharge(struct page *page, int order)
833 static inline int memcg_cache_id(struct mem_cgroup *memcg)
838 static inline void memcg_get_cache_ids(void)
842 static inline void memcg_put_cache_ids(void)
846 static inline struct kmem_cache *
847 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
852 static inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
855 #endif /* CONFIG_MEMCG_KMEM */
856 #endif /* _LINUX_MEMCONTROL_H */