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. */
93 * memcg field is used to find which memcg we belong directly
94 * Each memcg struct can hold more than one cg_proto, so container_of
97 * The elegant solution would be having an inverse function to
98 * proto_cgroup in struct proto, but that means polluting the structure
99 * for everybody, instead of just for memcg users.
101 struct mem_cgroup *memcg;
105 struct mem_cgroup_stat_cpu {
106 long count[MEM_CGROUP_STAT_NSTATS];
107 unsigned long events[MEMCG_NR_EVENTS];
108 unsigned long nr_page_events;
109 unsigned long targets[MEM_CGROUP_NTARGETS];
112 struct mem_cgroup_reclaim_iter {
113 struct mem_cgroup *position;
114 /* scan generation, increased every round-trip */
115 unsigned int generation;
119 * per-zone information in memory controller.
121 struct mem_cgroup_per_zone {
122 struct lruvec lruvec;
123 unsigned long lru_size[NR_LRU_LISTS];
125 struct mem_cgroup_reclaim_iter iter[DEF_PRIORITY + 1];
127 struct rb_node tree_node; /* RB tree node */
128 unsigned long usage_in_excess;/* Set to the value by which */
129 /* the soft limit is exceeded*/
131 struct mem_cgroup *memcg; /* Back pointer, we cannot */
132 /* use container_of */
135 struct mem_cgroup_per_node {
136 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
139 struct mem_cgroup_threshold {
140 struct eventfd_ctx *eventfd;
141 unsigned long threshold;
145 struct mem_cgroup_threshold_ary {
146 /* An array index points to threshold just below or equal to usage. */
147 int current_threshold;
148 /* Size of entries[] */
150 /* Array of thresholds */
151 struct mem_cgroup_threshold entries[0];
154 struct mem_cgroup_thresholds {
155 /* Primary thresholds array */
156 struct mem_cgroup_threshold_ary *primary;
158 * Spare threshold array.
159 * This is needed to make mem_cgroup_unregister_event() "never fail".
160 * It must be able to store at least primary->size - 1 entries.
162 struct mem_cgroup_threshold_ary *spare;
166 * The memory controller data structure. The memory controller controls both
167 * page cache and RSS per cgroup. We would eventually like to provide
168 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
169 * to help the administrator determine what knobs to tune.
172 struct cgroup_subsys_state css;
174 /* Accounted resources */
175 struct page_counter memory;
176 struct page_counter memsw;
177 struct page_counter kmem;
179 /* Normal memory consumption range */
183 unsigned long soft_limit;
185 /* vmpressure notifications */
186 struct vmpressure vmpressure;
188 /* css_online() has been completed */
192 * Should the accounting and control be hierarchical, per subtree?
196 /* protected by memcg_oom_lock */
201 /* OOM-Killer disable */
202 int oom_kill_disable;
204 /* handle for "memory.events" */
205 struct cgroup_file events_file;
207 /* protect arrays of thresholds */
208 struct mutex thresholds_lock;
210 /* thresholds for memory usage. RCU-protected */
211 struct mem_cgroup_thresholds thresholds;
213 /* thresholds for mem+swap usage. RCU-protected */
214 struct mem_cgroup_thresholds memsw_thresholds;
216 /* For oom notifier event fd */
217 struct list_head oom_notify;
220 * Should we move charges of a task when a task is moved into this
221 * mem_cgroup ? And what type of charges should we move ?
223 unsigned long move_charge_at_immigrate;
225 * set > 0 if pages under this cgroup are moving to other cgroup.
227 atomic_t moving_account;
228 /* taken only while moving_account > 0 */
229 spinlock_t move_lock;
230 struct task_struct *move_lock_task;
231 unsigned long move_lock_flags;
235 struct mem_cgroup_stat_cpu __percpu *stat;
237 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET)
238 struct cg_proto tcp_mem;
240 #if defined(CONFIG_MEMCG_KMEM)
241 /* Index in the kmem_cache->memcg_params.memcg_caches array */
243 bool kmem_acct_activated;
244 bool kmem_acct_active;
247 int last_scanned_node;
249 nodemask_t scan_nodes;
250 atomic_t numainfo_events;
251 atomic_t numainfo_updating;
254 #ifdef CONFIG_CGROUP_WRITEBACK
255 struct list_head cgwb_list;
256 struct wb_domain cgwb_domain;
259 /* List of events which userspace want to receive */
260 struct list_head event_list;
261 spinlock_t event_list_lock;
263 struct mem_cgroup_per_node *nodeinfo[0];
264 /* WARNING: nodeinfo must be the last member here */
267 extern struct mem_cgroup *root_mem_cgroup;
270 * mem_cgroup_events - count memory events against a cgroup
271 * @memcg: the memory cgroup
272 * @idx: the event index
273 * @nr: the number of events to account for
275 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
276 enum mem_cgroup_events_index idx,
279 this_cpu_add(memcg->stat->events[idx], nr);
280 cgroup_file_notify(&memcg->events_file);
283 bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg);
285 int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
286 gfp_t gfp_mask, struct mem_cgroup **memcgp);
287 void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
289 void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg);
290 void mem_cgroup_uncharge(struct page *page);
291 void mem_cgroup_uncharge_list(struct list_head *page_list);
293 void mem_cgroup_replace_page(struct page *oldpage, struct page *newpage);
295 struct lruvec *mem_cgroup_zone_lruvec(struct zone *, struct mem_cgroup *);
296 struct lruvec *mem_cgroup_page_lruvec(struct page *, struct zone *);
298 bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg);
299 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
300 struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg);
303 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
304 return css ? container_of(css, struct mem_cgroup, css) : NULL;
307 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
309 struct mem_cgroup_reclaim_cookie *);
310 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
312 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
313 struct mem_cgroup *root)
317 if (!root->use_hierarchy)
319 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
322 static inline bool mm_match_cgroup(struct mm_struct *mm,
323 struct mem_cgroup *memcg)
325 struct mem_cgroup *task_memcg;
329 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
331 match = mem_cgroup_is_descendant(task_memcg, memcg);
336 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
337 ino_t page_cgroup_ino(struct page *page);
339 static inline bool mem_cgroup_disabled(void)
341 return !cgroup_subsys_enabled(memory_cgrp_subsys);
345 * For memory reclaim.
347 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
349 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
352 static inline bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
354 struct mem_cgroup_per_zone *mz;
355 struct mem_cgroup *memcg;
357 if (mem_cgroup_disabled())
360 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
363 return !!(memcg->css.flags & CSS_ONLINE);
367 unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
369 struct mem_cgroup_per_zone *mz;
371 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
372 return mz->lru_size[lru];
375 static inline bool mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
377 unsigned long inactive_ratio;
378 unsigned long inactive;
379 unsigned long active;
382 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
383 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
385 gb = (inactive + active) >> (30 - PAGE_SHIFT);
387 inactive_ratio = int_sqrt(10 * gb);
391 return inactive * inactive_ratio < active;
394 void mem_cgroup_handle_over_high(void);
396 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
397 struct task_struct *p);
399 static inline void mem_cgroup_oom_enable(void)
401 WARN_ON(current->memcg_may_oom);
402 current->memcg_may_oom = 1;
405 static inline void mem_cgroup_oom_disable(void)
407 WARN_ON(!current->memcg_may_oom);
408 current->memcg_may_oom = 0;
411 static inline bool task_in_memcg_oom(struct task_struct *p)
413 return p->memcg_in_oom;
416 bool mem_cgroup_oom_synchronize(bool wait);
418 #ifdef CONFIG_MEMCG_SWAP
419 extern int do_swap_account;
422 struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page);
423 void mem_cgroup_end_page_stat(struct mem_cgroup *memcg);
426 * mem_cgroup_update_page_stat - update page state statistics
427 * @memcg: memcg to account against
428 * @idx: page state item to account
429 * @val: number of pages (positive or negative)
431 * See mem_cgroup_begin_page_stat() for locking requirements.
433 static inline void mem_cgroup_update_page_stat(struct mem_cgroup *memcg,
434 enum mem_cgroup_stat_index idx, int val)
436 VM_BUG_ON(!rcu_read_lock_held());
439 this_cpu_add(memcg->stat->count[idx], val);
442 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
443 enum mem_cgroup_stat_index idx)
445 mem_cgroup_update_page_stat(memcg, idx, 1);
448 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
449 enum mem_cgroup_stat_index idx)
451 mem_cgroup_update_page_stat(memcg, idx, -1);
454 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
456 unsigned long *total_scanned);
458 static inline void mem_cgroup_count_vm_event(struct mm_struct *mm,
459 enum vm_event_item idx)
461 struct mem_cgroup *memcg;
463 if (mem_cgroup_disabled())
467 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
468 if (unlikely(!memcg))
473 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
476 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
484 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
485 void mem_cgroup_split_huge_fixup(struct page *head);
488 #else /* CONFIG_MEMCG */
491 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
492 enum mem_cgroup_events_index idx,
497 static inline bool mem_cgroup_low(struct mem_cgroup *root,
498 struct mem_cgroup *memcg)
503 static inline int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
505 struct mem_cgroup **memcgp)
511 static inline void mem_cgroup_commit_charge(struct page *page,
512 struct mem_cgroup *memcg,
517 static inline void mem_cgroup_cancel_charge(struct page *page,
518 struct mem_cgroup *memcg)
522 static inline void mem_cgroup_uncharge(struct page *page)
526 static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
530 static inline void mem_cgroup_replace_page(struct page *old, struct page *new)
534 static inline struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
535 struct mem_cgroup *memcg)
537 return &zone->lruvec;
540 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
543 return &zone->lruvec;
546 static inline bool mm_match_cgroup(struct mm_struct *mm,
547 struct mem_cgroup *memcg)
552 static inline bool task_in_mem_cgroup(struct task_struct *task,
553 const struct mem_cgroup *memcg)
558 static inline struct mem_cgroup *
559 mem_cgroup_iter(struct mem_cgroup *root,
560 struct mem_cgroup *prev,
561 struct mem_cgroup_reclaim_cookie *reclaim)
566 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
567 struct mem_cgroup *prev)
571 static inline bool mem_cgroup_disabled(void)
577 mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
582 static inline bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
587 static inline unsigned long
588 mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
594 mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
600 mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
604 static inline struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
609 static inline void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
613 static inline void mem_cgroup_handle_over_high(void)
617 static inline void mem_cgroup_oom_enable(void)
621 static inline void mem_cgroup_oom_disable(void)
625 static inline bool task_in_memcg_oom(struct task_struct *p)
630 static inline bool mem_cgroup_oom_synchronize(bool wait)
635 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
636 enum mem_cgroup_stat_index idx)
640 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
641 enum mem_cgroup_stat_index idx)
646 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
648 unsigned long *total_scanned)
653 static inline void mem_cgroup_split_huge_fixup(struct page *head)
658 void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
661 #endif /* CONFIG_MEMCG */
669 #ifdef CONFIG_CGROUP_WRITEBACK
671 struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg);
672 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
673 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
674 unsigned long *pheadroom, unsigned long *pdirty,
675 unsigned long *pwriteback);
677 #else /* CONFIG_CGROUP_WRITEBACK */
679 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
684 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
685 unsigned long *pfilepages,
686 unsigned long *pheadroom,
687 unsigned long *pdirty,
688 unsigned long *pwriteback)
692 #endif /* CONFIG_CGROUP_WRITEBACK */
695 void sock_update_memcg(struct sock *sk);
696 void sock_release_memcg(struct sock *sk);
698 #ifdef CONFIG_MEMCG_KMEM
699 extern struct static_key memcg_kmem_enabled_key;
701 extern int memcg_nr_cache_ids;
702 void memcg_get_cache_ids(void);
703 void memcg_put_cache_ids(void);
706 * Helper macro to loop through all memcg-specific caches. Callers must still
707 * check if the cache is valid (it is either valid or NULL).
708 * the slab_mutex must be held when looping through those caches
710 #define for_each_memcg_cache_index(_idx) \
711 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
713 static inline bool memcg_kmem_enabled(void)
715 return static_key_false(&memcg_kmem_enabled_key);
718 static inline bool memcg_kmem_is_active(struct mem_cgroup *memcg)
720 return memcg->kmem_acct_active;
724 * In general, we'll do everything in our power to not incur in any overhead
725 * for non-memcg users for the kmem functions. Not even a function call, if we
728 * Therefore, we'll inline all those functions so that in the best case, we'll
729 * see that kmemcg is off for everybody and proceed quickly. If it is on,
730 * we'll still do most of the flag checking inline. We check a lot of
731 * conditions, but because they are pretty simple, they are expected to be
734 int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
735 struct mem_cgroup *memcg);
736 int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order);
737 void __memcg_kmem_uncharge(struct page *page, int order);
740 * helper for acessing a memcg's index. It will be used as an index in the
741 * child cache array in kmem_cache, and also to derive its name. This function
742 * will return -1 when this is not a kmem-limited memcg.
744 static inline int memcg_cache_id(struct mem_cgroup *memcg)
746 return memcg ? memcg->kmemcg_id : -1;
749 struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp);
750 void __memcg_kmem_put_cache(struct kmem_cache *cachep);
752 static inline bool __memcg_kmem_bypass(void)
754 if (!memcg_kmem_enabled())
756 if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
762 * memcg_kmem_charge: charge a kmem page
763 * @page: page to charge
765 * @order: allocation order
767 * Returns 0 on success, an error code on failure.
769 static __always_inline int memcg_kmem_charge(struct page *page,
770 gfp_t gfp, int order)
772 if (__memcg_kmem_bypass())
774 if (!(gfp & __GFP_ACCOUNT))
776 return __memcg_kmem_charge(page, gfp, order);
780 * memcg_kmem_uncharge: uncharge a kmem page
781 * @page: page to uncharge
782 * @order: allocation order
784 static __always_inline void memcg_kmem_uncharge(struct page *page, int order)
786 if (memcg_kmem_enabled())
787 __memcg_kmem_uncharge(page, order);
791 * memcg_kmem_get_cache: selects the correct per-memcg cache for allocation
792 * @cachep: the original global kmem cache
794 * All memory allocated from a per-memcg cache is charged to the owner memcg.
796 static __always_inline struct kmem_cache *
797 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
799 if (__memcg_kmem_bypass())
801 return __memcg_kmem_get_cache(cachep, gfp);
804 static __always_inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
806 if (memcg_kmem_enabled())
807 __memcg_kmem_put_cache(cachep);
810 #define for_each_memcg_cache_index(_idx) \
813 static inline bool memcg_kmem_enabled(void)
818 static inline bool memcg_kmem_is_active(struct mem_cgroup *memcg)
823 static inline int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
828 static inline void memcg_kmem_uncharge(struct page *page, int order)
832 static inline int memcg_cache_id(struct mem_cgroup *memcg)
837 static inline void memcg_get_cache_ids(void)
841 static inline void memcg_put_cache_ids(void)
845 static inline struct kmem_cache *
846 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
851 static inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
854 #endif /* CONFIG_MEMCG_KMEM */
855 #endif /* _LINUX_MEMCONTROL_H */