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;
58 * All "charge" functions with gfp_mask should use GFP_KERNEL or
59 * (gfp_mask & GFP_RECLAIM_MASK). In current implementatin, memcg doesn't
60 * alloc memory but reclaims memory from all available zones. So, "where I want
61 * memory from" bits of gfp_mask has no meaning. So any bits of that field is
62 * available but adding a rule is better. charge functions' gfp_mask should
63 * be set to GFP_KERNEL or gfp_mask & GFP_RECLAIM_MASK for avoiding ambiguous
65 * (Of course, if memcg does memory allocation in future, GFP_KERNEL is sane.)
68 extern int mem_cgroup_newpage_charge(struct page *page, struct mm_struct *mm,
70 /* for swap handling */
71 extern int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
72 struct page *page, gfp_t mask, struct mem_cgroup **memcgp);
73 extern void mem_cgroup_commit_charge_swapin(struct page *page,
74 struct mem_cgroup *memcg);
75 extern void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg);
77 extern int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
80 struct lruvec *mem_cgroup_zone_lruvec(struct zone *, struct mem_cgroup *);
81 struct lruvec *mem_cgroup_page_lruvec(struct page *, struct zone *);
83 /* For coalescing uncharge for reducing memcg' overhead*/
84 extern void mem_cgroup_uncharge_start(void);
85 extern void mem_cgroup_uncharge_end(void);
87 extern void mem_cgroup_uncharge_page(struct page *page);
88 extern void mem_cgroup_uncharge_cache_page(struct page *page);
90 bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
91 struct mem_cgroup *memcg);
92 bool task_in_mem_cgroup(struct task_struct *task,
93 const struct mem_cgroup *memcg);
95 extern struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page);
96 extern struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
97 extern struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm);
99 extern struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg);
100 extern struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css);
103 bool mm_match_cgroup(const struct mm_struct *mm, const struct mem_cgroup *memcg)
105 struct mem_cgroup *task_memcg;
109 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
110 match = __mem_cgroup_same_or_subtree(memcg, task_memcg);
115 extern struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg);
118 mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
119 struct mem_cgroup **memcgp);
120 extern void mem_cgroup_end_migration(struct mem_cgroup *memcg,
121 struct page *oldpage, struct page *newpage, bool migration_ok);
123 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
125 struct mem_cgroup_reclaim_cookie *);
126 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
129 * For memory reclaim.
131 int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec);
132 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
133 unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list);
134 void mem_cgroup_update_lru_size(struct lruvec *, enum lru_list, int);
135 extern void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
136 struct task_struct *p);
137 extern void mem_cgroup_replace_page_cache(struct page *oldpage,
138 struct page *newpage);
141 * mem_cgroup_toggle_oom - toggle the memcg OOM killer for the current task
142 * @new: true to enable, false to disable
144 * Toggle whether a failed memcg charge should invoke the OOM killer
145 * or just return -ENOMEM. Returns the previous toggle state.
147 * NOTE: Any path that enables the OOM killer before charging must
148 * call mem_cgroup_oom_synchronize() afterward to finalize the
149 * OOM handling and clean up.
151 static inline bool mem_cgroup_toggle_oom(bool new)
155 old = current->memcg_oom.may_oom;
156 current->memcg_oom.may_oom = new;
161 static inline void mem_cgroup_enable_oom(void)
163 bool old = mem_cgroup_toggle_oom(true);
165 WARN_ON(old == true);
168 static inline void mem_cgroup_disable_oom(void)
170 bool old = mem_cgroup_toggle_oom(false);
172 WARN_ON(old == false);
175 static inline bool task_in_memcg_oom(struct task_struct *p)
177 return p->memcg_oom.in_memcg_oom;
180 bool mem_cgroup_oom_synchronize(void);
182 #ifdef CONFIG_MEMCG_SWAP
183 extern int do_swap_account;
186 static inline bool mem_cgroup_disabled(void)
188 if (mem_cgroup_subsys.disabled)
193 void __mem_cgroup_begin_update_page_stat(struct page *page, bool *locked,
194 unsigned long *flags);
196 extern atomic_t memcg_moving;
198 static inline void mem_cgroup_begin_update_page_stat(struct page *page,
199 bool *locked, unsigned long *flags)
201 if (mem_cgroup_disabled())
205 if (atomic_read(&memcg_moving))
206 __mem_cgroup_begin_update_page_stat(page, locked, flags);
209 void __mem_cgroup_end_update_page_stat(struct page *page,
210 unsigned long *flags);
211 static inline void mem_cgroup_end_update_page_stat(struct page *page,
212 bool *locked, unsigned long *flags)
214 if (mem_cgroup_disabled())
217 __mem_cgroup_end_update_page_stat(page, flags);
221 void mem_cgroup_update_page_stat(struct page *page,
222 enum mem_cgroup_stat_index idx,
225 static inline void mem_cgroup_inc_page_stat(struct page *page,
226 enum mem_cgroup_stat_index idx)
228 mem_cgroup_update_page_stat(page, idx, 1);
231 static inline void mem_cgroup_dec_page_stat(struct page *page,
232 enum mem_cgroup_stat_index idx)
234 mem_cgroup_update_page_stat(page, idx, -1);
237 bool mem_cgroup_soft_reclaim_eligible(struct mem_cgroup *memcg);
239 void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx);
240 static inline void mem_cgroup_count_vm_event(struct mm_struct *mm,
241 enum vm_event_item idx)
243 if (mem_cgroup_disabled())
245 __mem_cgroup_count_vm_event(mm, idx);
247 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
248 void mem_cgroup_split_huge_fixup(struct page *head);
251 #ifdef CONFIG_DEBUG_VM
252 bool mem_cgroup_bad_page_check(struct page *page);
253 void mem_cgroup_print_bad_page(struct page *page);
255 #else /* CONFIG_MEMCG */
258 static inline int mem_cgroup_newpage_charge(struct page *page,
259 struct mm_struct *mm, gfp_t gfp_mask)
264 static inline int mem_cgroup_cache_charge(struct page *page,
265 struct mm_struct *mm, gfp_t gfp_mask)
270 static inline int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
271 struct page *page, gfp_t gfp_mask, struct mem_cgroup **memcgp)
276 static inline void mem_cgroup_commit_charge_swapin(struct page *page,
277 struct mem_cgroup *memcg)
281 static inline void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
285 static inline void mem_cgroup_uncharge_start(void)
289 static inline void mem_cgroup_uncharge_end(void)
293 static inline void mem_cgroup_uncharge_page(struct page *page)
297 static inline void mem_cgroup_uncharge_cache_page(struct page *page)
301 static inline struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
302 struct mem_cgroup *memcg)
304 return &zone->lruvec;
307 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
310 return &zone->lruvec;
313 static inline struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
318 static inline struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
323 static inline bool mm_match_cgroup(struct mm_struct *mm,
324 struct mem_cgroup *memcg)
329 static inline bool task_in_mem_cgroup(struct task_struct *task,
330 const struct mem_cgroup *memcg)
335 static inline struct cgroup_subsys_state
336 *mem_cgroup_css(struct mem_cgroup *memcg)
342 mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
343 struct mem_cgroup **memcgp)
347 static inline void mem_cgroup_end_migration(struct mem_cgroup *memcg,
348 struct page *oldpage, struct page *newpage, bool migration_ok)
352 static inline struct mem_cgroup *
353 mem_cgroup_iter(struct mem_cgroup *root,
354 struct mem_cgroup *prev,
355 struct mem_cgroup_reclaim_cookie *reclaim)
360 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
361 struct mem_cgroup *prev)
365 static inline bool mem_cgroup_disabled(void)
371 mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
376 static inline unsigned long
377 mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
383 mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
389 mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
393 static inline void mem_cgroup_begin_update_page_stat(struct page *page,
394 bool *locked, unsigned long *flags)
398 static inline void mem_cgroup_end_update_page_stat(struct page *page,
399 bool *locked, unsigned long *flags)
403 static inline bool mem_cgroup_toggle_oom(bool new)
408 static inline void mem_cgroup_enable_oom(void)
412 static inline void mem_cgroup_disable_oom(void)
416 static inline bool task_in_memcg_oom(struct task_struct *p)
421 static inline bool mem_cgroup_oom_synchronize(void)
426 static inline void mem_cgroup_inc_page_stat(struct page *page,
427 enum mem_cgroup_stat_index idx)
431 static inline void mem_cgroup_dec_page_stat(struct page *page,
432 enum mem_cgroup_stat_index idx)
437 bool mem_cgroup_soft_reclaim_eligible(struct mem_cgroup *memcg)
442 static inline void mem_cgroup_split_huge_fixup(struct page *head)
447 void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
450 static inline void mem_cgroup_replace_page_cache(struct page *oldpage,
451 struct page *newpage)
454 #endif /* CONFIG_MEMCG */
456 #if !defined(CONFIG_MEMCG) || !defined(CONFIG_DEBUG_VM)
458 mem_cgroup_bad_page_check(struct page *page)
464 mem_cgroup_print_bad_page(struct page *page)
476 #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
477 void sock_update_memcg(struct sock *sk);
478 void sock_release_memcg(struct sock *sk);
480 static inline void sock_update_memcg(struct sock *sk)
483 static inline void sock_release_memcg(struct sock *sk)
486 #endif /* CONFIG_INET && CONFIG_MEMCG_KMEM */
488 #ifdef CONFIG_MEMCG_KMEM
489 extern struct static_key memcg_kmem_enabled_key;
491 extern int memcg_limited_groups_array_size;
494 * Helper macro to loop through all memcg-specific caches. Callers must still
495 * check if the cache is valid (it is either valid or NULL).
496 * the slab_mutex must be held when looping through those caches
498 #define for_each_memcg_cache_index(_idx) \
499 for ((_idx) = 0; (_idx) < memcg_limited_groups_array_size; (_idx)++)
501 static inline bool memcg_kmem_enabled(void)
503 return static_key_false(&memcg_kmem_enabled_key);
507 * In general, we'll do everything in our power to not incur in any overhead
508 * for non-memcg users for the kmem functions. Not even a function call, if we
511 * Therefore, we'll inline all those functions so that in the best case, we'll
512 * see that kmemcg is off for everybody and proceed quickly. If it is on,
513 * we'll still do most of the flag checking inline. We check a lot of
514 * conditions, but because they are pretty simple, they are expected to be
517 bool __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg,
519 void __memcg_kmem_commit_charge(struct page *page,
520 struct mem_cgroup *memcg, int order);
521 void __memcg_kmem_uncharge_pages(struct page *page, int order);
523 int memcg_cache_id(struct mem_cgroup *memcg);
524 int memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s,
525 struct kmem_cache *root_cache);
526 void memcg_release_cache(struct kmem_cache *cachep);
527 void memcg_cache_list_add(struct mem_cgroup *memcg, struct kmem_cache *cachep);
529 int memcg_update_cache_size(struct kmem_cache *s, int num_groups);
530 void memcg_update_array_size(int num_groups);
533 __memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp);
535 void mem_cgroup_destroy_cache(struct kmem_cache *cachep);
536 void kmem_cache_destroy_memcg_children(struct kmem_cache *s);
539 * memcg_kmem_newpage_charge: verify if a new kmem allocation is allowed.
540 * @gfp: the gfp allocation flags.
541 * @memcg: a pointer to the memcg this was charged against.
542 * @order: allocation order.
544 * returns true if the memcg where the current task belongs can hold this
547 * We return true automatically if this allocation is not to be accounted to
551 memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg, int order)
553 if (!memcg_kmem_enabled())
557 * __GFP_NOFAIL allocations will move on even if charging is not
558 * possible. Therefore we don't even try, and have this allocation
559 * unaccounted. We could in theory charge it with
560 * res_counter_charge_nofail, but we hope those allocations are rare,
561 * and won't be worth the trouble.
563 if (!(gfp & __GFP_KMEMCG) || (gfp & __GFP_NOFAIL))
565 if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
568 /* If the test is dying, just let it go. */
569 if (unlikely(fatal_signal_pending(current)))
572 return __memcg_kmem_newpage_charge(gfp, memcg, order);
576 * memcg_kmem_uncharge_pages: uncharge pages from memcg
577 * @page: pointer to struct page being freed
578 * @order: allocation order.
580 * there is no need to specify memcg here, since it is embedded in page_cgroup
583 memcg_kmem_uncharge_pages(struct page *page, int order)
585 if (memcg_kmem_enabled())
586 __memcg_kmem_uncharge_pages(page, order);
590 * memcg_kmem_commit_charge: embeds correct memcg in a page
591 * @page: pointer to struct page recently allocated
592 * @memcg: the memcg structure we charged against
593 * @order: allocation order.
595 * Needs to be called after memcg_kmem_newpage_charge, regardless of success or
596 * failure of the allocation. if @page is NULL, this function will revert the
597 * charges. Otherwise, it will commit the memcg given by @memcg to the
598 * corresponding page_cgroup.
601 memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, int order)
603 if (memcg_kmem_enabled() && memcg)
604 __memcg_kmem_commit_charge(page, memcg, order);
608 * memcg_kmem_get_cache: selects the correct per-memcg cache for allocation
609 * @cachep: the original global kmem cache
610 * @gfp: allocation flags.
612 * This function assumes that the task allocating, which determines the memcg
613 * in the page allocator, belongs to the same cgroup throughout the whole
614 * process. Misacounting can happen if the task calls memcg_kmem_get_cache()
615 * while belonging to a cgroup, and later on changes. This is considered
616 * acceptable, and should only happen upon task migration.
618 * Before the cache is created by the memcg core, there is also a possible
619 * imbalance: the task belongs to a memcg, but the cache being allocated from
620 * is the global cache, since the child cache is not yet guaranteed to be
621 * ready. This case is also fine, since in this case the GFP_KMEMCG will not be
622 * passed and the page allocator will not attempt any cgroup accounting.
624 static __always_inline struct kmem_cache *
625 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
627 if (!memcg_kmem_enabled())
629 if (gfp & __GFP_NOFAIL)
631 if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
633 if (unlikely(fatal_signal_pending(current)))
636 return __memcg_kmem_get_cache(cachep, gfp);
639 #define for_each_memcg_cache_index(_idx) \
642 static inline bool memcg_kmem_enabled(void)
648 memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg, int order)
653 static inline void memcg_kmem_uncharge_pages(struct page *page, int order)
658 memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, int order)
662 static inline int memcg_cache_id(struct mem_cgroup *memcg)
668 memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s,
669 struct kmem_cache *root_cache)
674 static inline void memcg_release_cache(struct kmem_cache *cachep)
678 static inline void memcg_cache_list_add(struct mem_cgroup *memcg,
679 struct kmem_cache *s)
683 static inline struct kmem_cache *
684 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
689 static inline void kmem_cache_destroy_memcg_children(struct kmem_cache *s)
692 #endif /* CONFIG_MEMCG_KMEM */
693 #endif /* _LINUX_MEMCONTROL_H */