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 * Bits in struct cg_proto.flags
92 /* Currently active and new sockets should be assigned to cgroups */
94 /* It was ever activated; we must disarm static keys on destruction */
99 struct page_counter memory_allocated; /* Current allocated memory. */
100 struct percpu_counter sockets_allocated; /* Current number of sockets. */
105 * memcg field is used to find which memcg we belong directly
106 * Each memcg struct can hold more than one cg_proto, so container_of
109 * The elegant solution would be having an inverse function to
110 * proto_cgroup in struct proto, but that means polluting the structure
111 * for everybody, instead of just for memcg users.
113 struct mem_cgroup *memcg;
117 struct mem_cgroup_stat_cpu {
118 long count[MEM_CGROUP_STAT_NSTATS];
119 unsigned long events[MEMCG_NR_EVENTS];
120 unsigned long nr_page_events;
121 unsigned long targets[MEM_CGROUP_NTARGETS];
124 struct mem_cgroup_reclaim_iter {
125 struct mem_cgroup *position;
126 /* scan generation, increased every round-trip */
127 unsigned int generation;
131 * per-zone information in memory controller.
133 struct mem_cgroup_per_zone {
134 struct lruvec lruvec;
135 unsigned long lru_size[NR_LRU_LISTS];
137 struct mem_cgroup_reclaim_iter iter[DEF_PRIORITY + 1];
139 struct rb_node tree_node; /* RB tree node */
140 unsigned long usage_in_excess;/* Set to the value by which */
141 /* the soft limit is exceeded*/
143 struct mem_cgroup *memcg; /* Back pointer, we cannot */
144 /* use container_of */
147 struct mem_cgroup_per_node {
148 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
151 struct mem_cgroup_threshold {
152 struct eventfd_ctx *eventfd;
153 unsigned long threshold;
157 struct mem_cgroup_threshold_ary {
158 /* An array index points to threshold just below or equal to usage. */
159 int current_threshold;
160 /* Size of entries[] */
162 /* Array of thresholds */
163 struct mem_cgroup_threshold entries[0];
166 struct mem_cgroup_thresholds {
167 /* Primary thresholds array */
168 struct mem_cgroup_threshold_ary *primary;
170 * Spare threshold array.
171 * This is needed to make mem_cgroup_unregister_event() "never fail".
172 * It must be able to store at least primary->size - 1 entries.
174 struct mem_cgroup_threshold_ary *spare;
178 * The memory controller data structure. The memory controller controls both
179 * page cache and RSS per cgroup. We would eventually like to provide
180 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
181 * to help the administrator determine what knobs to tune.
184 struct cgroup_subsys_state css;
186 /* Accounted resources */
187 struct page_counter memory;
188 struct page_counter memsw;
189 struct page_counter kmem;
191 /* Normal memory consumption range */
195 unsigned long soft_limit;
197 /* vmpressure notifications */
198 struct vmpressure vmpressure;
200 /* css_online() has been completed */
204 * Should the accounting and control be hierarchical, per subtree?
208 /* protected by memcg_oom_lock */
213 /* OOM-Killer disable */
214 int oom_kill_disable;
216 /* handle for "memory.events" */
217 struct cgroup_file events_file;
219 /* protect arrays of thresholds */
220 struct mutex thresholds_lock;
222 /* thresholds for memory usage. RCU-protected */
223 struct mem_cgroup_thresholds thresholds;
225 /* thresholds for mem+swap usage. RCU-protected */
226 struct mem_cgroup_thresholds memsw_thresholds;
228 /* For oom notifier event fd */
229 struct list_head oom_notify;
232 * Should we move charges of a task when a task is moved into this
233 * mem_cgroup ? And what type of charges should we move ?
235 unsigned long move_charge_at_immigrate;
237 * set > 0 if pages under this cgroup are moving to other cgroup.
239 atomic_t moving_account;
240 /* taken only while moving_account > 0 */
241 spinlock_t move_lock;
242 struct task_struct *move_lock_task;
243 unsigned long move_lock_flags;
247 struct mem_cgroup_stat_cpu __percpu *stat;
249 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET)
250 struct cg_proto tcp_mem;
252 #if defined(CONFIG_MEMCG_KMEM)
253 /* Index in the kmem_cache->memcg_params.memcg_caches array */
255 bool kmem_acct_activated;
256 bool kmem_acct_active;
259 int last_scanned_node;
261 nodemask_t scan_nodes;
262 atomic_t numainfo_events;
263 atomic_t numainfo_updating;
266 #ifdef CONFIG_CGROUP_WRITEBACK
267 struct list_head cgwb_list;
268 struct wb_domain cgwb_domain;
271 /* List of events which userspace want to receive */
272 struct list_head event_list;
273 spinlock_t event_list_lock;
275 struct mem_cgroup_per_node *nodeinfo[0];
276 /* WARNING: nodeinfo must be the last member here */
278 extern struct cgroup_subsys_state *mem_cgroup_root_css;
281 * mem_cgroup_events - count memory events against a cgroup
282 * @memcg: the memory cgroup
283 * @idx: the event index
284 * @nr: the number of events to account for
286 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
287 enum mem_cgroup_events_index idx,
290 this_cpu_add(memcg->stat->events[idx], nr);
291 cgroup_file_notify(&memcg->events_file);
294 bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg);
296 int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
297 gfp_t gfp_mask, struct mem_cgroup **memcgp,
299 void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
300 bool lrucare, bool compound);
301 void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
303 void mem_cgroup_uncharge(struct page *page);
304 void mem_cgroup_uncharge_list(struct list_head *page_list);
306 void mem_cgroup_replace_page(struct page *oldpage, struct page *newpage);
308 struct lruvec *mem_cgroup_zone_lruvec(struct zone *, struct mem_cgroup *);
309 struct lruvec *mem_cgroup_page_lruvec(struct page *, struct zone *);
311 bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg);
312 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
313 struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg);
316 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
317 return css ? container_of(css, struct mem_cgroup, css) : NULL;
320 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
322 struct mem_cgroup_reclaim_cookie *);
323 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
325 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
326 struct mem_cgroup *root)
330 if (!root->use_hierarchy)
332 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
335 static inline bool mm_match_cgroup(struct mm_struct *mm,
336 struct mem_cgroup *memcg)
338 struct mem_cgroup *task_memcg;
342 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
344 match = mem_cgroup_is_descendant(task_memcg, memcg);
349 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
350 ino_t page_cgroup_ino(struct page *page);
352 static inline bool mem_cgroup_disabled(void)
354 return !cgroup_subsys_enabled(memory_cgrp_subsys);
358 * For memory reclaim.
360 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
362 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
365 static inline bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
367 struct mem_cgroup_per_zone *mz;
368 struct mem_cgroup *memcg;
370 if (mem_cgroup_disabled())
373 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
376 return !!(memcg->css.flags & CSS_ONLINE);
380 unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
382 struct mem_cgroup_per_zone *mz;
384 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
385 return mz->lru_size[lru];
388 static inline bool mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
390 unsigned long inactive_ratio;
391 unsigned long inactive;
392 unsigned long active;
395 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
396 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
398 gb = (inactive + active) >> (30 - PAGE_SHIFT);
400 inactive_ratio = int_sqrt(10 * gb);
404 return inactive * inactive_ratio < active;
407 void mem_cgroup_handle_over_high(void);
409 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
410 struct task_struct *p);
412 static inline void mem_cgroup_oom_enable(void)
414 WARN_ON(current->memcg_may_oom);
415 current->memcg_may_oom = 1;
418 static inline void mem_cgroup_oom_disable(void)
420 WARN_ON(!current->memcg_may_oom);
421 current->memcg_may_oom = 0;
424 static inline bool task_in_memcg_oom(struct task_struct *p)
426 return p->memcg_in_oom;
429 bool mem_cgroup_oom_synchronize(bool wait);
431 #ifdef CONFIG_MEMCG_SWAP
432 extern int do_swap_account;
435 struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page);
436 void mem_cgroup_end_page_stat(struct mem_cgroup *memcg);
439 * mem_cgroup_update_page_stat - update page state statistics
440 * @memcg: memcg to account against
441 * @idx: page state item to account
442 * @val: number of pages (positive or negative)
444 * See mem_cgroup_begin_page_stat() for locking requirements.
446 static inline void mem_cgroup_update_page_stat(struct mem_cgroup *memcg,
447 enum mem_cgroup_stat_index idx, int val)
449 VM_BUG_ON(!rcu_read_lock_held());
452 this_cpu_add(memcg->stat->count[idx], val);
455 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
456 enum mem_cgroup_stat_index idx)
458 mem_cgroup_update_page_stat(memcg, idx, 1);
461 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
462 enum mem_cgroup_stat_index idx)
464 mem_cgroup_update_page_stat(memcg, idx, -1);
467 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
469 unsigned long *total_scanned);
471 static inline void mem_cgroup_count_vm_event(struct mm_struct *mm,
472 enum vm_event_item idx)
474 struct mem_cgroup *memcg;
476 if (mem_cgroup_disabled())
480 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
481 if (unlikely(!memcg))
486 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
489 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
497 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
498 void mem_cgroup_split_huge_fixup(struct page *head);
501 #else /* CONFIG_MEMCG */
504 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
505 enum mem_cgroup_events_index idx,
510 static inline bool mem_cgroup_low(struct mem_cgroup *root,
511 struct mem_cgroup *memcg)
516 static inline int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
518 struct mem_cgroup **memcgp,
525 static inline void mem_cgroup_commit_charge(struct page *page,
526 struct mem_cgroup *memcg,
527 bool lrucare, bool compound)
531 static inline void mem_cgroup_cancel_charge(struct page *page,
532 struct mem_cgroup *memcg,
537 static inline void mem_cgroup_uncharge(struct page *page)
541 static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
545 static inline void mem_cgroup_replace_page(struct page *old, struct page *new)
549 static inline struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
550 struct mem_cgroup *memcg)
552 return &zone->lruvec;
555 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
558 return &zone->lruvec;
561 static inline bool mm_match_cgroup(struct mm_struct *mm,
562 struct mem_cgroup *memcg)
567 static inline bool task_in_mem_cgroup(struct task_struct *task,
568 const struct mem_cgroup *memcg)
573 static inline struct mem_cgroup *
574 mem_cgroup_iter(struct mem_cgroup *root,
575 struct mem_cgroup *prev,
576 struct mem_cgroup_reclaim_cookie *reclaim)
581 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
582 struct mem_cgroup *prev)
586 static inline bool mem_cgroup_disabled(void)
592 mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
597 static inline bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
602 static inline unsigned long
603 mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
609 mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
615 mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
619 static inline struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
624 static inline void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
628 static inline void mem_cgroup_handle_over_high(void)
632 static inline void mem_cgroup_oom_enable(void)
636 static inline void mem_cgroup_oom_disable(void)
640 static inline bool task_in_memcg_oom(struct task_struct *p)
645 static inline bool mem_cgroup_oom_synchronize(bool wait)
650 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
651 enum mem_cgroup_stat_index idx)
655 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
656 enum mem_cgroup_stat_index idx)
661 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
663 unsigned long *total_scanned)
668 static inline void mem_cgroup_split_huge_fixup(struct page *head)
673 void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
676 #endif /* CONFIG_MEMCG */
684 #ifdef CONFIG_CGROUP_WRITEBACK
686 struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg);
687 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
688 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
689 unsigned long *pheadroom, unsigned long *pdirty,
690 unsigned long *pwriteback);
692 #else /* CONFIG_CGROUP_WRITEBACK */
694 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
699 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
700 unsigned long *pfilepages,
701 unsigned long *pheadroom,
702 unsigned long *pdirty,
703 unsigned long *pwriteback)
707 #endif /* CONFIG_CGROUP_WRITEBACK */
710 #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
711 void sock_update_memcg(struct sock *sk);
712 void sock_release_memcg(struct sock *sk);
714 static inline void sock_update_memcg(struct sock *sk)
717 static inline void sock_release_memcg(struct sock *sk)
720 #endif /* CONFIG_INET && CONFIG_MEMCG_KMEM */
722 #ifdef CONFIG_MEMCG_KMEM
723 extern struct static_key memcg_kmem_enabled_key;
725 extern int memcg_nr_cache_ids;
726 void memcg_get_cache_ids(void);
727 void memcg_put_cache_ids(void);
730 * Helper macro to loop through all memcg-specific caches. Callers must still
731 * check if the cache is valid (it is either valid or NULL).
732 * the slab_mutex must be held when looping through those caches
734 #define for_each_memcg_cache_index(_idx) \
735 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
737 static inline bool memcg_kmem_enabled(void)
739 return static_key_false(&memcg_kmem_enabled_key);
742 static inline bool memcg_kmem_is_active(struct mem_cgroup *memcg)
744 return memcg->kmem_acct_active;
748 * In general, we'll do everything in our power to not incur in any overhead
749 * for non-memcg users for the kmem functions. Not even a function call, if we
752 * Therefore, we'll inline all those functions so that in the best case, we'll
753 * see that kmemcg is off for everybody and proceed quickly. If it is on,
754 * we'll still do most of the flag checking inline. We check a lot of
755 * conditions, but because they are pretty simple, they are expected to be
758 int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
759 struct mem_cgroup *memcg);
760 int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order);
761 void __memcg_kmem_uncharge(struct page *page, int order);
764 * helper for acessing a memcg's index. It will be used as an index in the
765 * child cache array in kmem_cache, and also to derive its name. This function
766 * will return -1 when this is not a kmem-limited memcg.
768 static inline int memcg_cache_id(struct mem_cgroup *memcg)
770 return memcg ? memcg->kmemcg_id : -1;
773 struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep);
774 void __memcg_kmem_put_cache(struct kmem_cache *cachep);
776 static inline bool __memcg_kmem_bypass(gfp_t gfp)
778 if (!memcg_kmem_enabled())
780 if (gfp & __GFP_NOACCOUNT)
782 if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
788 * memcg_kmem_charge: charge a kmem page
789 * @page: page to charge
791 * @order: allocation order
793 * Returns 0 on success, an error code on failure.
795 static __always_inline int memcg_kmem_charge(struct page *page,
796 gfp_t gfp, int order)
798 if (__memcg_kmem_bypass(gfp))
800 return __memcg_kmem_charge(page, gfp, order);
804 * memcg_kmem_uncharge: uncharge a kmem page
805 * @page: page to uncharge
806 * @order: allocation order
808 static __always_inline void memcg_kmem_uncharge(struct page *page, int order)
810 if (memcg_kmem_enabled())
811 __memcg_kmem_uncharge(page, order);
815 * memcg_kmem_get_cache: selects the correct per-memcg cache for allocation
816 * @cachep: the original global kmem cache
817 * @gfp: allocation flags.
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(gfp))
826 return __memcg_kmem_get_cache(cachep);
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_is_active(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_KMEM */
880 #endif /* _LINUX_MEMCONTROL_H */