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>
31 #include <linux/page-flags.h>
39 * The corresponding mem_cgroup_stat_names is defined in mm/memcontrol.c,
40 * These two lists should keep in accord with each other.
42 enum mem_cgroup_stat_index {
44 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
46 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
47 MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
48 MEM_CGROUP_STAT_RSS_HUGE, /* # of pages charged as anon huge */
49 MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
50 MEM_CGROUP_STAT_DIRTY, /* # of dirty pages in page cache */
51 MEM_CGROUP_STAT_WRITEBACK, /* # of pages under writeback */
52 MEM_CGROUP_STAT_SWAP, /* # of pages, swapped out */
53 MEM_CGROUP_STAT_NSTATS,
54 /* default hierarchy stats */
55 MEMCG_KERNEL_STACK = MEM_CGROUP_STAT_NSTATS,
56 MEMCG_SLAB_RECLAIMABLE,
57 MEMCG_SLAB_UNRECLAIMABLE,
62 struct mem_cgroup_reclaim_cookie {
65 unsigned int generation;
68 enum mem_cgroup_events_index {
69 MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */
70 MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */
71 MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */
72 MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */
73 MEM_CGROUP_EVENTS_NSTATS,
74 /* default hierarchy events */
75 MEMCG_LOW = MEM_CGROUP_EVENTS_NSTATS,
83 * Per memcg event counter is incremented at every pagein/pageout. With THP,
84 * it will be incremated by the number of pages. This counter is used for
85 * for trigger some periodic events. This is straightforward and better
86 * than using jiffies etc. to handle periodic memcg event.
88 enum mem_cgroup_events_target {
89 MEM_CGROUP_TARGET_THRESH,
90 MEM_CGROUP_TARGET_SOFTLIMIT,
91 MEM_CGROUP_TARGET_NUMAINFO,
97 #define MEM_CGROUP_ID_SHIFT 16
98 #define MEM_CGROUP_ID_MAX USHRT_MAX
100 struct mem_cgroup_id {
105 struct mem_cgroup_stat_cpu {
106 long count[MEMCG_NR_STAT];
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;
165 enum memcg_kmem_state {
172 * The memory controller data structure. The memory controller controls both
173 * page cache and RSS per cgroup. We would eventually like to provide
174 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
175 * to help the administrator determine what knobs to tune.
178 struct cgroup_subsys_state css;
180 /* Private memcg ID. Used to ID objects that outlive the cgroup */
181 struct mem_cgroup_id id;
183 /* Accounted resources */
184 struct page_counter memory;
185 struct page_counter swap;
187 /* Legacy consumer-oriented counters */
188 struct page_counter memsw;
189 struct page_counter kmem;
190 struct page_counter tcpmem;
192 /* Normal memory consumption range */
196 /* Range enforcement for interrupt charges */
197 struct work_struct high_work;
199 unsigned long soft_limit;
201 /* vmpressure notifications */
202 struct vmpressure vmpressure;
205 * Should the accounting and control be hierarchical, per subtree?
209 /* protected by memcg_oom_lock */
214 /* OOM-Killer disable */
215 int oom_kill_disable;
217 /* handle for "memory.events" */
218 struct cgroup_file events_file;
220 /* protect arrays of thresholds */
221 struct mutex thresholds_lock;
223 /* thresholds for memory usage. RCU-protected */
224 struct mem_cgroup_thresholds thresholds;
226 /* thresholds for mem+swap usage. RCU-protected */
227 struct mem_cgroup_thresholds memsw_thresholds;
229 /* For oom notifier event fd */
230 struct list_head oom_notify;
233 * Should we move charges of a task when a task is moved into this
234 * mem_cgroup ? And what type of charges should we move ?
236 unsigned long move_charge_at_immigrate;
238 * set > 0 if pages under this cgroup are moving to other cgroup.
240 atomic_t moving_account;
241 /* taken only while moving_account > 0 */
242 spinlock_t move_lock;
243 struct task_struct *move_lock_task;
244 unsigned long move_lock_flags;
248 struct mem_cgroup_stat_cpu __percpu *stat;
250 unsigned long socket_pressure;
252 /* Legacy tcp memory accounting */
257 /* Index in the kmem_cache->memcg_params.memcg_caches array */
259 enum memcg_kmem_state kmem_state;
262 int last_scanned_node;
264 nodemask_t scan_nodes;
265 atomic_t numainfo_events;
266 atomic_t numainfo_updating;
269 #ifdef CONFIG_CGROUP_WRITEBACK
270 struct list_head cgwb_list;
271 struct wb_domain cgwb_domain;
274 /* List of events which userspace want to receive */
275 struct list_head event_list;
276 spinlock_t event_list_lock;
278 struct mem_cgroup_per_node *nodeinfo[0];
279 /* WARNING: nodeinfo must be the last member here */
282 extern struct mem_cgroup *root_mem_cgroup;
284 static inline bool mem_cgroup_disabled(void)
286 return !cgroup_subsys_enabled(memory_cgrp_subsys);
290 * mem_cgroup_events - count memory events against a cgroup
291 * @memcg: the memory cgroup
292 * @idx: the event index
293 * @nr: the number of events to account for
295 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
296 enum mem_cgroup_events_index idx,
299 this_cpu_add(memcg->stat->events[idx], nr);
300 cgroup_file_notify(&memcg->events_file);
303 bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg);
305 int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
306 gfp_t gfp_mask, struct mem_cgroup **memcgp,
308 void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
309 bool lrucare, bool compound);
310 void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
312 void mem_cgroup_uncharge(struct page *page);
313 void mem_cgroup_uncharge_list(struct list_head *page_list);
315 void mem_cgroup_migrate(struct page *oldpage, struct page *newpage);
317 struct lruvec *mem_cgroup_zone_lruvec(struct zone *, struct mem_cgroup *);
318 struct lruvec *mem_cgroup_page_lruvec(struct page *, struct zone *);
320 bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg);
321 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
324 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
325 return css ? container_of(css, struct mem_cgroup, css) : NULL;
328 #define mem_cgroup_from_counter(counter, member) \
329 container_of(counter, struct mem_cgroup, member)
331 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
333 struct mem_cgroup_reclaim_cookie *);
334 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
336 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
338 if (mem_cgroup_disabled())
343 struct mem_cgroup *mem_cgroup_from_id(unsigned short id);
346 * parent_mem_cgroup - find the accounting parent of a memcg
347 * @memcg: memcg whose parent to find
349 * Returns the parent memcg, or NULL if this is the root or the memory
350 * controller is in legacy no-hierarchy mode.
352 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
354 if (!memcg->memory.parent)
356 return mem_cgroup_from_counter(memcg->memory.parent, memory);
359 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
360 struct mem_cgroup *root)
364 if (!root->use_hierarchy)
366 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
369 static inline bool mm_match_cgroup(struct mm_struct *mm,
370 struct mem_cgroup *memcg)
372 struct mem_cgroup *task_memcg;
376 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
378 match = mem_cgroup_is_descendant(task_memcg, memcg);
383 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
384 ino_t page_cgroup_ino(struct page *page);
386 static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
388 if (mem_cgroup_disabled())
390 return !!(memcg->css.flags & CSS_ONLINE);
394 * For memory reclaim.
396 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
398 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
401 unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
402 int nid, unsigned int lru_mask);
405 unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
407 struct mem_cgroup_per_zone *mz;
409 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
410 return mz->lru_size[lru];
413 void mem_cgroup_handle_over_high(void);
415 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
416 struct task_struct *p);
418 static inline void mem_cgroup_oom_enable(void)
420 WARN_ON(current->memcg_may_oom);
421 current->memcg_may_oom = 1;
424 static inline void mem_cgroup_oom_disable(void)
426 WARN_ON(!current->memcg_may_oom);
427 current->memcg_may_oom = 0;
430 static inline bool task_in_memcg_oom(struct task_struct *p)
432 return p->memcg_in_oom;
435 bool mem_cgroup_oom_synchronize(bool wait);
437 #ifdef CONFIG_MEMCG_SWAP
438 extern int do_swap_account;
441 void lock_page_memcg(struct page *page);
442 void unlock_page_memcg(struct page *page);
445 * mem_cgroup_update_page_stat - update page state statistics
447 * @idx: page state item to account
448 * @val: number of pages (positive or negative)
450 * The @page must be locked or the caller must use lock_page_memcg()
451 * to prevent double accounting when the page is concurrently being
452 * moved to another memcg:
454 * lock_page(page) or lock_page_memcg(page)
455 * if (TestClearPageState(page))
456 * mem_cgroup_update_page_stat(page, state, -1);
457 * unlock_page(page) or unlock_page_memcg(page)
459 static inline void mem_cgroup_update_page_stat(struct page *page,
460 enum mem_cgroup_stat_index idx, int val)
462 VM_BUG_ON(!(rcu_read_lock_held() || PageLocked(page)));
464 if (page->mem_cgroup)
465 this_cpu_add(page->mem_cgroup->stat->count[idx], val);
468 static inline void mem_cgroup_inc_page_stat(struct page *page,
469 enum mem_cgroup_stat_index idx)
471 mem_cgroup_update_page_stat(page, idx, 1);
474 static inline void mem_cgroup_dec_page_stat(struct page *page,
475 enum mem_cgroup_stat_index idx)
477 mem_cgroup_update_page_stat(page, idx, -1);
480 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
482 unsigned long *total_scanned);
484 static inline void mem_cgroup_count_vm_event(struct mm_struct *mm,
485 enum vm_event_item idx)
487 struct mem_cgroup *memcg;
489 if (mem_cgroup_disabled())
493 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
494 if (unlikely(!memcg))
499 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
502 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
510 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
511 void mem_cgroup_split_huge_fixup(struct page *head);
514 #else /* CONFIG_MEMCG */
516 #define MEM_CGROUP_ID_SHIFT 0
517 #define MEM_CGROUP_ID_MAX 0
521 static inline bool mem_cgroup_disabled(void)
526 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
527 enum mem_cgroup_events_index idx,
532 static inline bool mem_cgroup_low(struct mem_cgroup *root,
533 struct mem_cgroup *memcg)
538 static inline int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
540 struct mem_cgroup **memcgp,
547 static inline void mem_cgroup_commit_charge(struct page *page,
548 struct mem_cgroup *memcg,
549 bool lrucare, bool compound)
553 static inline void mem_cgroup_cancel_charge(struct page *page,
554 struct mem_cgroup *memcg,
559 static inline void mem_cgroup_uncharge(struct page *page)
563 static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
567 static inline void mem_cgroup_migrate(struct page *old, struct page *new)
571 static inline struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
572 struct mem_cgroup *memcg)
574 return &zone->lruvec;
577 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
580 return &zone->lruvec;
583 static inline bool mm_match_cgroup(struct mm_struct *mm,
584 struct mem_cgroup *memcg)
589 static inline bool task_in_mem_cgroup(struct task_struct *task,
590 const struct mem_cgroup *memcg)
595 static inline struct mem_cgroup *
596 mem_cgroup_iter(struct mem_cgroup *root,
597 struct mem_cgroup *prev,
598 struct mem_cgroup_reclaim_cookie *reclaim)
603 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
604 struct mem_cgroup *prev)
608 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
613 static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
616 /* XXX: This should always return root_mem_cgroup */
620 static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
625 static inline unsigned long
626 mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
631 static inline unsigned long
632 mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
633 int nid, unsigned int lru_mask)
639 mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
643 static inline void lock_page_memcg(struct page *page)
647 static inline void unlock_page_memcg(struct page *page)
651 static inline void mem_cgroup_handle_over_high(void)
655 static inline void mem_cgroup_oom_enable(void)
659 static inline void mem_cgroup_oom_disable(void)
663 static inline bool task_in_memcg_oom(struct task_struct *p)
668 static inline bool mem_cgroup_oom_synchronize(bool wait)
673 static inline void mem_cgroup_inc_page_stat(struct page *page,
674 enum mem_cgroup_stat_index idx)
678 static inline void mem_cgroup_dec_page_stat(struct page *page,
679 enum mem_cgroup_stat_index idx)
684 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
686 unsigned long *total_scanned)
691 static inline void mem_cgroup_split_huge_fixup(struct page *head)
696 void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
699 #endif /* CONFIG_MEMCG */
701 #ifdef CONFIG_CGROUP_WRITEBACK
703 struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg);
704 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
705 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
706 unsigned long *pheadroom, unsigned long *pdirty,
707 unsigned long *pwriteback);
709 #else /* CONFIG_CGROUP_WRITEBACK */
711 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
716 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
717 unsigned long *pfilepages,
718 unsigned long *pheadroom,
719 unsigned long *pdirty,
720 unsigned long *pwriteback)
724 #endif /* CONFIG_CGROUP_WRITEBACK */
727 void sock_update_memcg(struct sock *sk);
728 void sock_release_memcg(struct sock *sk);
729 bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
730 void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
732 extern struct static_key_false memcg_sockets_enabled_key;
733 #define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_enabled_key)
734 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
736 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_pressure)
739 if (time_before(jiffies, memcg->socket_pressure))
741 } while ((memcg = parent_mem_cgroup(memcg)));
745 #define mem_cgroup_sockets_enabled 0
746 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
752 struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep);
753 void memcg_kmem_put_cache(struct kmem_cache *cachep);
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);
759 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
760 extern struct static_key_false memcg_kmem_enabled_key;
762 extern int memcg_nr_cache_ids;
763 void memcg_get_cache_ids(void);
764 void memcg_put_cache_ids(void);
767 * Helper macro to loop through all memcg-specific caches. Callers must still
768 * check if the cache is valid (it is either valid or NULL).
769 * the slab_mutex must be held when looping through those caches
771 #define for_each_memcg_cache_index(_idx) \
772 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
774 static inline bool memcg_kmem_enabled(void)
776 return static_branch_unlikely(&memcg_kmem_enabled_key);
780 * helper for accessing 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;
790 * memcg_kmem_update_page_stat - update kmem page state statistics
792 * @idx: page state item to account
793 * @val: number of pages (positive or negative)
795 static inline void memcg_kmem_update_page_stat(struct page *page,
796 enum mem_cgroup_stat_index idx, int val)
798 if (memcg_kmem_enabled() && page->mem_cgroup)
799 this_cpu_add(page->mem_cgroup->stat->count[idx], val);
803 #define for_each_memcg_cache_index(_idx) \
806 static inline bool memcg_kmem_enabled(void)
811 static inline int memcg_cache_id(struct mem_cgroup *memcg)
816 static inline void memcg_get_cache_ids(void)
820 static inline void memcg_put_cache_ids(void)
824 static inline void memcg_kmem_update_page_stat(struct page *page,
825 enum mem_cgroup_stat_index idx, int val)
828 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
830 #endif /* _LINUX_MEMCONTROL_H */