1 /* memcontrol.h - Memory Controller
3 * Copyright IBM Corporation, 2007
6 * Copyright 2007 OpenVZ SWsoft Inc
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);
298 void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
300 void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg);
301 void mem_cgroup_uncharge(struct page *page);
302 void mem_cgroup_uncharge_list(struct list_head *page_list);
304 void mem_cgroup_replace_page(struct page *oldpage, struct page *newpage);
306 struct lruvec *mem_cgroup_zone_lruvec(struct zone *, struct mem_cgroup *);
307 struct lruvec *mem_cgroup_page_lruvec(struct page *, struct zone *);
309 bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg);
310 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
311 struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg);
314 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
315 return css ? container_of(css, struct mem_cgroup, css) : NULL;
318 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
320 struct mem_cgroup_reclaim_cookie *);
321 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
323 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
324 struct mem_cgroup *root)
328 if (!root->use_hierarchy)
330 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
333 static inline bool mm_match_cgroup(struct mm_struct *mm,
334 struct mem_cgroup *memcg)
336 struct mem_cgroup *task_memcg;
340 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
342 match = mem_cgroup_is_descendant(task_memcg, memcg);
347 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
348 ino_t page_cgroup_ino(struct page *page);
350 static inline bool mem_cgroup_disabled(void)
352 return !cgroup_subsys_enabled(memory_cgrp_subsys);
356 * For memory reclaim.
358 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
360 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
363 static inline bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
365 struct mem_cgroup_per_zone *mz;
366 struct mem_cgroup *memcg;
368 if (mem_cgroup_disabled())
371 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
374 return !!(memcg->css.flags & CSS_ONLINE);
378 unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
380 struct mem_cgroup_per_zone *mz;
382 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
383 return mz->lru_size[lru];
386 static inline bool mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
388 unsigned long inactive_ratio;
389 unsigned long inactive;
390 unsigned long active;
393 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
394 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
396 gb = (inactive + active) >> (30 - PAGE_SHIFT);
398 inactive_ratio = int_sqrt(10 * gb);
402 return inactive * inactive_ratio < active;
405 void mem_cgroup_handle_over_high(void);
407 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
408 struct task_struct *p);
410 static inline void mem_cgroup_oom_enable(void)
412 WARN_ON(current->memcg_may_oom);
413 current->memcg_may_oom = 1;
416 static inline void mem_cgroup_oom_disable(void)
418 WARN_ON(!current->memcg_may_oom);
419 current->memcg_may_oom = 0;
422 static inline bool task_in_memcg_oom(struct task_struct *p)
424 return p->memcg_in_oom;
427 bool mem_cgroup_oom_synchronize(bool wait);
429 #ifdef CONFIG_MEMCG_SWAP
430 extern int do_swap_account;
433 struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page);
434 void mem_cgroup_end_page_stat(struct mem_cgroup *memcg);
437 * mem_cgroup_update_page_stat - update page state statistics
438 * @memcg: memcg to account against
439 * @idx: page state item to account
440 * @val: number of pages (positive or negative)
442 * See mem_cgroup_begin_page_stat() for locking requirements.
444 static inline void mem_cgroup_update_page_stat(struct mem_cgroup *memcg,
445 enum mem_cgroup_stat_index idx, int val)
447 VM_BUG_ON(!rcu_read_lock_held());
450 this_cpu_add(memcg->stat->count[idx], val);
453 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
454 enum mem_cgroup_stat_index idx)
456 mem_cgroup_update_page_stat(memcg, idx, 1);
459 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
460 enum mem_cgroup_stat_index idx)
462 mem_cgroup_update_page_stat(memcg, idx, -1);
465 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
467 unsigned long *total_scanned);
469 static inline void mem_cgroup_count_vm_event(struct mm_struct *mm,
470 enum vm_event_item idx)
472 struct mem_cgroup *memcg;
474 if (mem_cgroup_disabled())
478 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
479 if (unlikely(!memcg))
484 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
487 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
495 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
496 void mem_cgroup_split_huge_fixup(struct page *head);
499 #else /* CONFIG_MEMCG */
502 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
503 enum mem_cgroup_events_index idx,
508 static inline bool mem_cgroup_low(struct mem_cgroup *root,
509 struct mem_cgroup *memcg)
514 static inline int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
516 struct mem_cgroup **memcgp)
522 static inline void mem_cgroup_commit_charge(struct page *page,
523 struct mem_cgroup *memcg,
528 static inline void mem_cgroup_cancel_charge(struct page *page,
529 struct mem_cgroup *memcg)
533 static inline void mem_cgroup_uncharge(struct page *page)
537 static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
541 static inline void mem_cgroup_replace_page(struct page *old, struct page *new)
545 static inline struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
546 struct mem_cgroup *memcg)
548 return &zone->lruvec;
551 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
554 return &zone->lruvec;
557 static inline bool mm_match_cgroup(struct mm_struct *mm,
558 struct mem_cgroup *memcg)
563 static inline bool task_in_mem_cgroup(struct task_struct *task,
564 const struct mem_cgroup *memcg)
569 static inline struct mem_cgroup *
570 mem_cgroup_iter(struct mem_cgroup *root,
571 struct mem_cgroup *prev,
572 struct mem_cgroup_reclaim_cookie *reclaim)
577 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
578 struct mem_cgroup *prev)
582 static inline bool mem_cgroup_disabled(void)
588 mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
593 static inline bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
598 static inline unsigned long
599 mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
605 mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
611 mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
615 static inline struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
620 static inline void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
624 static inline void mem_cgroup_handle_over_high(void)
628 static inline void mem_cgroup_oom_enable(void)
632 static inline void mem_cgroup_oom_disable(void)
636 static inline bool task_in_memcg_oom(struct task_struct *p)
641 static inline bool mem_cgroup_oom_synchronize(bool wait)
646 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
647 enum mem_cgroup_stat_index idx)
651 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
652 enum mem_cgroup_stat_index idx)
657 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
659 unsigned long *total_scanned)
664 static inline void mem_cgroup_split_huge_fixup(struct page *head)
669 void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
672 #endif /* CONFIG_MEMCG */
680 #ifdef CONFIG_CGROUP_WRITEBACK
682 struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg);
683 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
684 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
685 unsigned long *pheadroom, unsigned long *pdirty,
686 unsigned long *pwriteback);
688 #else /* CONFIG_CGROUP_WRITEBACK */
690 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
695 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
696 unsigned long *pfilepages,
697 unsigned long *pheadroom,
698 unsigned long *pdirty,
699 unsigned long *pwriteback)
703 #endif /* CONFIG_CGROUP_WRITEBACK */
706 #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
707 void sock_update_memcg(struct sock *sk);
708 void sock_release_memcg(struct sock *sk);
710 static inline void sock_update_memcg(struct sock *sk)
713 static inline void sock_release_memcg(struct sock *sk)
716 #endif /* CONFIG_INET && CONFIG_MEMCG_KMEM */
718 #ifdef CONFIG_MEMCG_KMEM
719 extern struct static_key memcg_kmem_enabled_key;
721 extern int memcg_nr_cache_ids;
722 void memcg_get_cache_ids(void);
723 void memcg_put_cache_ids(void);
726 * Helper macro to loop through all memcg-specific caches. Callers must still
727 * check if the cache is valid (it is either valid or NULL).
728 * the slab_mutex must be held when looping through those caches
730 #define for_each_memcg_cache_index(_idx) \
731 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
733 static inline bool memcg_kmem_enabled(void)
735 return static_key_false(&memcg_kmem_enabled_key);
738 static inline bool memcg_kmem_is_active(struct mem_cgroup *memcg)
740 return memcg->kmem_acct_active;
744 * In general, we'll do everything in our power to not incur in any overhead
745 * for non-memcg users for the kmem functions. Not even a function call, if we
748 * Therefore, we'll inline all those functions so that in the best case, we'll
749 * see that kmemcg is off for everybody and proceed quickly. If it is on,
750 * we'll still do most of the flag checking inline. We check a lot of
751 * conditions, but because they are pretty simple, they are expected to be
754 int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
755 struct mem_cgroup *memcg);
756 int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order);
757 void __memcg_kmem_uncharge(struct page *page, int order);
760 * helper for acessing a memcg's index. It will be used as an index in the
761 * child cache array in kmem_cache, and also to derive its name. This function
762 * will return -1 when this is not a kmem-limited memcg.
764 static inline int memcg_cache_id(struct mem_cgroup *memcg)
766 return memcg ? memcg->kmemcg_id : -1;
769 struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep);
770 void __memcg_kmem_put_cache(struct kmem_cache *cachep);
772 static inline bool __memcg_kmem_bypass(gfp_t gfp)
774 if (!memcg_kmem_enabled())
776 if (gfp & __GFP_NOACCOUNT)
778 if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
784 * memcg_kmem_charge: charge a kmem page
785 * @page: page to charge
787 * @order: allocation order
789 * Returns 0 on success, an error code on failure.
791 static __always_inline int memcg_kmem_charge(struct page *page,
792 gfp_t gfp, int order)
794 if (__memcg_kmem_bypass(gfp))
796 return __memcg_kmem_charge(page, gfp, order);
800 * memcg_kmem_uncharge: uncharge a kmem page
801 * @page: page to uncharge
802 * @order: allocation order
804 static __always_inline void memcg_kmem_uncharge(struct page *page, int order)
806 if (memcg_kmem_enabled())
807 __memcg_kmem_uncharge(page, order);
811 * memcg_kmem_get_cache: selects the correct per-memcg cache for allocation
812 * @cachep: the original global kmem cache
813 * @gfp: allocation flags.
815 * All memory allocated from a per-memcg cache is charged to the owner memcg.
817 static __always_inline struct kmem_cache *
818 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
820 if (__memcg_kmem_bypass(gfp))
822 return __memcg_kmem_get_cache(cachep);
825 static __always_inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
827 if (memcg_kmem_enabled())
828 __memcg_kmem_put_cache(cachep);
831 #define for_each_memcg_cache_index(_idx) \
834 static inline bool memcg_kmem_enabled(void)
839 static inline bool memcg_kmem_is_active(struct mem_cgroup *memcg)
844 static inline int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
849 static inline void memcg_kmem_uncharge(struct page *page, int order)
853 static inline int memcg_cache_id(struct mem_cgroup *memcg)
858 static inline void memcg_get_cache_ids(void)
862 static inline void memcg_put_cache_ids(void)
866 static inline struct kmem_cache *
867 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
872 static inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
875 #endif /* CONFIG_MEMCG_KMEM */
876 #endif /* _LINUX_MEMCONTROL_H */