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>
34 * The corresponding mem_cgroup_stat_names is defined in mm/memcontrol.c,
35 * These two lists should keep in accord with each other.
37 enum mem_cgroup_stat_index {
39 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
41 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
42 MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
43 MEM_CGROUP_STAT_RSS_HUGE, /* # of pages charged as anon huge */
44 MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
45 MEM_CGROUP_STAT_WRITEBACK, /* # of pages under writeback */
46 MEM_CGROUP_STAT_SWAP, /* # of pages, swapped out */
47 MEM_CGROUP_STAT_NSTATS,
50 struct mem_cgroup_reclaim_cookie {
53 unsigned int generation;
56 enum mem_cgroup_filter_t {
57 VISIT, /* visit current node */
58 SKIP, /* skip the current node and continue traversal */
59 SKIP_TREE, /* skip the whole subtree and continue traversal */
63 * mem_cgroup_filter_t predicate might instruct mem_cgroup_iter_cond how to
64 * iterate through the hierarchy tree. Each tree element is checked by the
65 * predicate before it is returned by the iterator. If a filter returns
66 * SKIP or SKIP_TREE then the iterator code continues traversal (with the
67 * next node down the hierarchy or the next node that doesn't belong under the
70 typedef enum mem_cgroup_filter_t
71 (*mem_cgroup_iter_filter)(struct mem_cgroup *memcg, struct mem_cgroup *root);
75 * All "charge" functions with gfp_mask should use GFP_KERNEL or
76 * (gfp_mask & GFP_RECLAIM_MASK). In current implementatin, memcg doesn't
77 * alloc memory but reclaims memory from all available zones. So, "where I want
78 * memory from" bits of gfp_mask has no meaning. So any bits of that field is
79 * available but adding a rule is better. charge functions' gfp_mask should
80 * be set to GFP_KERNEL or gfp_mask & GFP_RECLAIM_MASK for avoiding ambiguous
82 * (Of course, if memcg does memory allocation in future, GFP_KERNEL is sane.)
85 extern int mem_cgroup_newpage_charge(struct page *page, struct mm_struct *mm,
87 /* for swap handling */
88 extern int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
89 struct page *page, gfp_t mask, struct mem_cgroup **memcgp);
90 extern void mem_cgroup_commit_charge_swapin(struct page *page,
91 struct mem_cgroup *memcg);
92 extern void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg);
94 extern int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
97 struct lruvec *mem_cgroup_zone_lruvec(struct zone *, struct mem_cgroup *);
98 struct lruvec *mem_cgroup_page_lruvec(struct page *, struct zone *);
100 /* For coalescing uncharge for reducing memcg' overhead*/
101 extern void mem_cgroup_uncharge_start(void);
102 extern void mem_cgroup_uncharge_end(void);
104 extern void mem_cgroup_uncharge_page(struct page *page);
105 extern void mem_cgroup_uncharge_cache_page(struct page *page);
107 bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
108 struct mem_cgroup *memcg);
109 bool task_in_mem_cgroup(struct task_struct *task,
110 const struct mem_cgroup *memcg);
112 extern struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page);
113 extern struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
114 extern struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm);
116 extern struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg);
117 extern struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css);
120 bool mm_match_cgroup(const struct mm_struct *mm, const struct mem_cgroup *memcg)
122 struct mem_cgroup *task_memcg;
126 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
127 match = __mem_cgroup_same_or_subtree(memcg, task_memcg);
132 extern struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg);
135 mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
136 struct mem_cgroup **memcgp);
137 extern void mem_cgroup_end_migration(struct mem_cgroup *memcg,
138 struct page *oldpage, struct page *newpage, bool migration_ok);
140 struct mem_cgroup *mem_cgroup_iter_cond(struct mem_cgroup *root,
141 struct mem_cgroup *prev,
142 struct mem_cgroup_reclaim_cookie *reclaim,
143 mem_cgroup_iter_filter cond);
145 static inline struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
146 struct mem_cgroup *prev,
147 struct mem_cgroup_reclaim_cookie *reclaim)
149 return mem_cgroup_iter_cond(root, prev, reclaim, NULL);
152 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
155 * For memory reclaim.
157 int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec);
158 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
159 unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list);
160 void mem_cgroup_update_lru_size(struct lruvec *, enum lru_list, int);
161 extern void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
162 struct task_struct *p);
163 extern void mem_cgroup_replace_page_cache(struct page *oldpage,
164 struct page *newpage);
167 * mem_cgroup_toggle_oom - toggle the memcg OOM killer for the current task
168 * @new: true to enable, false to disable
170 * Toggle whether a failed memcg charge should invoke the OOM killer
171 * or just return -ENOMEM. Returns the previous toggle state.
173 * NOTE: Any path that enables the OOM killer before charging must
174 * call mem_cgroup_oom_synchronize() afterward to finalize the
175 * OOM handling and clean up.
177 static inline bool mem_cgroup_toggle_oom(bool new)
181 old = current->memcg_oom.may_oom;
182 current->memcg_oom.may_oom = new;
187 static inline void mem_cgroup_enable_oom(void)
189 bool old = mem_cgroup_toggle_oom(true);
191 WARN_ON(old == true);
194 static inline void mem_cgroup_disable_oom(void)
196 bool old = mem_cgroup_toggle_oom(false);
198 WARN_ON(old == false);
201 static inline bool task_in_memcg_oom(struct task_struct *p)
203 return p->memcg_oom.in_memcg_oom;
206 bool mem_cgroup_oom_synchronize(void);
208 #ifdef CONFIG_MEMCG_SWAP
209 extern int do_swap_account;
212 static inline bool mem_cgroup_disabled(void)
214 if (mem_cgroup_subsys.disabled)
219 void __mem_cgroup_begin_update_page_stat(struct page *page, bool *locked,
220 unsigned long *flags);
222 extern atomic_t memcg_moving;
224 static inline void mem_cgroup_begin_update_page_stat(struct page *page,
225 bool *locked, unsigned long *flags)
227 if (mem_cgroup_disabled())
231 if (atomic_read(&memcg_moving))
232 __mem_cgroup_begin_update_page_stat(page, locked, flags);
235 void __mem_cgroup_end_update_page_stat(struct page *page,
236 unsigned long *flags);
237 static inline void mem_cgroup_end_update_page_stat(struct page *page,
238 bool *locked, unsigned long *flags)
240 if (mem_cgroup_disabled())
243 __mem_cgroup_end_update_page_stat(page, flags);
247 void mem_cgroup_update_page_stat(struct page *page,
248 enum mem_cgroup_stat_index idx,
251 static inline void mem_cgroup_inc_page_stat(struct page *page,
252 enum mem_cgroup_stat_index idx)
254 mem_cgroup_update_page_stat(page, idx, 1);
257 static inline void mem_cgroup_dec_page_stat(struct page *page,
258 enum mem_cgroup_stat_index idx)
260 mem_cgroup_update_page_stat(page, idx, -1);
263 enum mem_cgroup_filter_t
264 mem_cgroup_soft_reclaim_eligible(struct mem_cgroup *memcg,
265 struct mem_cgroup *root);
267 void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx);
268 static inline void mem_cgroup_count_vm_event(struct mm_struct *mm,
269 enum vm_event_item idx)
271 if (mem_cgroup_disabled())
273 __mem_cgroup_count_vm_event(mm, idx);
275 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
276 void mem_cgroup_split_huge_fixup(struct page *head);
279 #ifdef CONFIG_DEBUG_VM
280 bool mem_cgroup_bad_page_check(struct page *page);
281 void mem_cgroup_print_bad_page(struct page *page);
283 #else /* CONFIG_MEMCG */
286 static inline int mem_cgroup_newpage_charge(struct page *page,
287 struct mm_struct *mm, gfp_t gfp_mask)
292 static inline int mem_cgroup_cache_charge(struct page *page,
293 struct mm_struct *mm, gfp_t gfp_mask)
298 static inline int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
299 struct page *page, gfp_t gfp_mask, struct mem_cgroup **memcgp)
304 static inline void mem_cgroup_commit_charge_swapin(struct page *page,
305 struct mem_cgroup *memcg)
309 static inline void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
313 static inline void mem_cgroup_uncharge_start(void)
317 static inline void mem_cgroup_uncharge_end(void)
321 static inline void mem_cgroup_uncharge_page(struct page *page)
325 static inline void mem_cgroup_uncharge_cache_page(struct page *page)
329 static inline struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
330 struct mem_cgroup *memcg)
332 return &zone->lruvec;
335 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
338 return &zone->lruvec;
341 static inline struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
346 static inline struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
351 static inline bool mm_match_cgroup(struct mm_struct *mm,
352 struct mem_cgroup *memcg)
357 static inline bool task_in_mem_cgroup(struct task_struct *task,
358 const struct mem_cgroup *memcg)
363 static inline struct cgroup_subsys_state
364 *mem_cgroup_css(struct mem_cgroup *memcg)
370 mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
371 struct mem_cgroup **memcgp)
375 static inline void mem_cgroup_end_migration(struct mem_cgroup *memcg,
376 struct page *oldpage, struct page *newpage, bool migration_ok)
379 static inline struct mem_cgroup *
380 mem_cgroup_iter_cond(struct mem_cgroup *root,
381 struct mem_cgroup *prev,
382 struct mem_cgroup_reclaim_cookie *reclaim,
383 mem_cgroup_iter_filter cond)
385 /* first call must return non-NULL, second return NULL */
386 return (struct mem_cgroup *)(unsigned long)!prev;
389 static inline struct mem_cgroup *
390 mem_cgroup_iter(struct mem_cgroup *root,
391 struct mem_cgroup *prev,
392 struct mem_cgroup_reclaim_cookie *reclaim)
397 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
398 struct mem_cgroup *prev)
402 static inline bool mem_cgroup_disabled(void)
408 mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
413 static inline unsigned long
414 mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
420 mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
426 mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
430 static inline void mem_cgroup_begin_update_page_stat(struct page *page,
431 bool *locked, unsigned long *flags)
435 static inline void mem_cgroup_end_update_page_stat(struct page *page,
436 bool *locked, unsigned long *flags)
440 static inline bool mem_cgroup_toggle_oom(bool new)
445 static inline void mem_cgroup_enable_oom(void)
449 static inline void mem_cgroup_disable_oom(void)
453 static inline bool task_in_memcg_oom(struct task_struct *p)
458 static inline bool mem_cgroup_oom_synchronize(void)
463 static inline void mem_cgroup_inc_page_stat(struct page *page,
464 enum mem_cgroup_stat_index idx)
468 static inline void mem_cgroup_dec_page_stat(struct page *page,
469 enum mem_cgroup_stat_index idx)
474 enum mem_cgroup_filter_t
475 mem_cgroup_soft_reclaim_eligible(struct mem_cgroup *memcg,
476 struct mem_cgroup *root)
481 static inline void mem_cgroup_split_huge_fixup(struct page *head)
486 void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
489 static inline void mem_cgroup_replace_page_cache(struct page *oldpage,
490 struct page *newpage)
493 #endif /* CONFIG_MEMCG */
495 #if !defined(CONFIG_MEMCG) || !defined(CONFIG_DEBUG_VM)
497 mem_cgroup_bad_page_check(struct page *page)
503 mem_cgroup_print_bad_page(struct page *page)
515 #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
516 void sock_update_memcg(struct sock *sk);
517 void sock_release_memcg(struct sock *sk);
519 static inline void sock_update_memcg(struct sock *sk)
522 static inline void sock_release_memcg(struct sock *sk)
525 #endif /* CONFIG_INET && CONFIG_MEMCG_KMEM */
527 #ifdef CONFIG_MEMCG_KMEM
528 extern struct static_key memcg_kmem_enabled_key;
530 extern int memcg_limited_groups_array_size;
533 * Helper macro to loop through all memcg-specific caches. Callers must still
534 * check if the cache is valid (it is either valid or NULL).
535 * the slab_mutex must be held when looping through those caches
537 #define for_each_memcg_cache_index(_idx) \
538 for ((_idx) = 0; (_idx) < memcg_limited_groups_array_size; (_idx)++)
540 static inline bool memcg_kmem_enabled(void)
542 return static_key_false(&memcg_kmem_enabled_key);
546 * In general, we'll do everything in our power to not incur in any overhead
547 * for non-memcg users for the kmem functions. Not even a function call, if we
550 * Therefore, we'll inline all those functions so that in the best case, we'll
551 * see that kmemcg is off for everybody and proceed quickly. If it is on,
552 * we'll still do most of the flag checking inline. We check a lot of
553 * conditions, but because they are pretty simple, they are expected to be
556 bool __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg,
558 void __memcg_kmem_commit_charge(struct page *page,
559 struct mem_cgroup *memcg, int order);
560 void __memcg_kmem_uncharge_pages(struct page *page, int order);
562 int memcg_cache_id(struct mem_cgroup *memcg);
563 int memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s,
564 struct kmem_cache *root_cache);
565 void memcg_release_cache(struct kmem_cache *cachep);
566 void memcg_cache_list_add(struct mem_cgroup *memcg, struct kmem_cache *cachep);
568 int memcg_update_cache_size(struct kmem_cache *s, int num_groups);
569 void memcg_update_array_size(int num_groups);
572 __memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp);
574 void mem_cgroup_destroy_cache(struct kmem_cache *cachep);
575 void kmem_cache_destroy_memcg_children(struct kmem_cache *s);
578 * memcg_kmem_newpage_charge: verify if a new kmem allocation is allowed.
579 * @gfp: the gfp allocation flags.
580 * @memcg: a pointer to the memcg this was charged against.
581 * @order: allocation order.
583 * returns true if the memcg where the current task belongs can hold this
586 * We return true automatically if this allocation is not to be accounted to
590 memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg, int order)
592 if (!memcg_kmem_enabled())
596 * __GFP_NOFAIL allocations will move on even if charging is not
597 * possible. Therefore we don't even try, and have this allocation
598 * unaccounted. We could in theory charge it with
599 * res_counter_charge_nofail, but we hope those allocations are rare,
600 * and won't be worth the trouble.
602 if (!(gfp & __GFP_KMEMCG) || (gfp & __GFP_NOFAIL))
604 if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
607 /* If the test is dying, just let it go. */
608 if (unlikely(fatal_signal_pending(current)))
611 return __memcg_kmem_newpage_charge(gfp, memcg, order);
615 * memcg_kmem_uncharge_pages: uncharge pages from memcg
616 * @page: pointer to struct page being freed
617 * @order: allocation order.
619 * there is no need to specify memcg here, since it is embedded in page_cgroup
622 memcg_kmem_uncharge_pages(struct page *page, int order)
624 if (memcg_kmem_enabled())
625 __memcg_kmem_uncharge_pages(page, order);
629 * memcg_kmem_commit_charge: embeds correct memcg in a page
630 * @page: pointer to struct page recently allocated
631 * @memcg: the memcg structure we charged against
632 * @order: allocation order.
634 * Needs to be called after memcg_kmem_newpage_charge, regardless of success or
635 * failure of the allocation. if @page is NULL, this function will revert the
636 * charges. Otherwise, it will commit the memcg given by @memcg to the
637 * corresponding page_cgroup.
640 memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, int order)
642 if (memcg_kmem_enabled() && memcg)
643 __memcg_kmem_commit_charge(page, memcg, order);
647 * memcg_kmem_get_cache: selects the correct per-memcg cache for allocation
648 * @cachep: the original global kmem cache
649 * @gfp: allocation flags.
651 * This function assumes that the task allocating, which determines the memcg
652 * in the page allocator, belongs to the same cgroup throughout the whole
653 * process. Misacounting can happen if the task calls memcg_kmem_get_cache()
654 * while belonging to a cgroup, and later on changes. This is considered
655 * acceptable, and should only happen upon task migration.
657 * Before the cache is created by the memcg core, there is also a possible
658 * imbalance: the task belongs to a memcg, but the cache being allocated from
659 * is the global cache, since the child cache is not yet guaranteed to be
660 * ready. This case is also fine, since in this case the GFP_KMEMCG will not be
661 * passed and the page allocator will not attempt any cgroup accounting.
663 static __always_inline struct kmem_cache *
664 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
666 if (!memcg_kmem_enabled())
668 if (gfp & __GFP_NOFAIL)
670 if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
672 if (unlikely(fatal_signal_pending(current)))
675 return __memcg_kmem_get_cache(cachep, gfp);
678 #define for_each_memcg_cache_index(_idx) \
681 static inline bool memcg_kmem_enabled(void)
687 memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg, int order)
692 static inline void memcg_kmem_uncharge_pages(struct page *page, int order)
697 memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, int order)
701 static inline int memcg_cache_id(struct mem_cgroup *memcg)
707 memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s,
708 struct kmem_cache *root_cache)
713 static inline void memcg_release_cache(struct kmem_cache *cachep)
717 static inline void memcg_cache_list_add(struct mem_cgroup *memcg,
718 struct kmem_cache *s)
722 static inline struct kmem_cache *
723 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
728 static inline void kmem_cache_destroy_memcg_children(struct kmem_cache *s)
731 #endif /* CONFIG_MEMCG_KMEM */
732 #endif /* _LINUX_MEMCONTROL_H */