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>
33 /* Stats that can be updated by kernel. */
34 enum mem_cgroup_page_stat_item {
35 MEMCG_NR_FILE_MAPPED, /* # of pages charged as file rss */
38 struct mem_cgroup_reclaim_cookie {
41 unsigned int generation;
44 enum mem_cgroup_filter_t {
45 VISIT, /* visit current node */
46 SKIP, /* skip the current node and continue traversal */
47 SKIP_TREE, /* skip the whole subtree and continue traversal */
51 * mem_cgroup_filter_t predicate might instruct mem_cgroup_iter_cond how to
52 * iterate through the hierarchy tree. Each tree element is checked by the
53 * predicate before it is returned by the iterator. If a filter returns
54 * SKIP or SKIP_TREE then the iterator code continues traversal (with the
55 * next node down the hierarchy or the next node that doesn't belong under the
58 typedef enum mem_cgroup_filter_t
59 (*mem_cgroup_iter_filter)(struct mem_cgroup *memcg, struct mem_cgroup *root);
63 * All "charge" functions with gfp_mask should use GFP_KERNEL or
64 * (gfp_mask & GFP_RECLAIM_MASK). In current implementatin, memcg doesn't
65 * alloc memory but reclaims memory from all available zones. So, "where I want
66 * memory from" bits of gfp_mask has no meaning. So any bits of that field is
67 * available but adding a rule is better. charge functions' gfp_mask should
68 * be set to GFP_KERNEL or gfp_mask & GFP_RECLAIM_MASK for avoiding ambiguous
70 * (Of course, if memcg does memory allocation in future, GFP_KERNEL is sane.)
73 extern int mem_cgroup_newpage_charge(struct page *page, struct mm_struct *mm,
75 /* for swap handling */
76 extern int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
77 struct page *page, gfp_t mask, struct mem_cgroup **memcgp);
78 extern void mem_cgroup_commit_charge_swapin(struct page *page,
79 struct mem_cgroup *memcg);
80 extern void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg);
82 extern int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
85 struct lruvec *mem_cgroup_zone_lruvec(struct zone *, struct mem_cgroup *);
86 struct lruvec *mem_cgroup_page_lruvec(struct page *, struct zone *);
88 /* For coalescing uncharge for reducing memcg' overhead*/
89 extern void mem_cgroup_uncharge_start(void);
90 extern void mem_cgroup_uncharge_end(void);
92 extern void mem_cgroup_uncharge_page(struct page *page);
93 extern void mem_cgroup_uncharge_cache_page(struct page *page);
95 bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
96 struct mem_cgroup *memcg);
97 bool task_in_mem_cgroup(struct task_struct *task,
98 const struct mem_cgroup *memcg);
100 extern struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page);
101 extern struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
102 extern struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm);
104 extern struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg);
105 extern struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css);
108 bool mm_match_cgroup(const struct mm_struct *mm, const struct mem_cgroup *memcg)
110 struct mem_cgroup *task_memcg;
114 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
115 match = __mem_cgroup_same_or_subtree(memcg, task_memcg);
120 extern struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg);
123 mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
124 struct mem_cgroup **memcgp);
125 extern void mem_cgroup_end_migration(struct mem_cgroup *memcg,
126 struct page *oldpage, struct page *newpage, bool migration_ok);
128 struct mem_cgroup *mem_cgroup_iter_cond(struct mem_cgroup *root,
129 struct mem_cgroup *prev,
130 struct mem_cgroup_reclaim_cookie *reclaim,
131 mem_cgroup_iter_filter cond);
133 static inline struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
134 struct mem_cgroup *prev,
135 struct mem_cgroup_reclaim_cookie *reclaim)
137 return mem_cgroup_iter_cond(root, prev, reclaim, NULL);
140 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
143 * For memory reclaim.
145 int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec);
146 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
147 unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list);
148 void mem_cgroup_update_lru_size(struct lruvec *, enum lru_list, int);
149 extern void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
150 struct task_struct *p);
151 extern void mem_cgroup_replace_page_cache(struct page *oldpage,
152 struct page *newpage);
155 * mem_cgroup_toggle_oom - toggle the memcg OOM killer for the current task
156 * @new: true to enable, false to disable
158 * Toggle whether a failed memcg charge should invoke the OOM killer
159 * or just return -ENOMEM. Returns the previous toggle state.
161 static inline bool mem_cgroup_toggle_oom(bool new)
165 old = current->memcg_oom.may_oom;
166 current->memcg_oom.may_oom = new;
171 static inline void mem_cgroup_enable_oom(void)
173 bool old = mem_cgroup_toggle_oom(true);
175 WARN_ON(old == true);
178 static inline void mem_cgroup_disable_oom(void)
180 bool old = mem_cgroup_toggle_oom(false);
182 WARN_ON(old == false);
185 #ifdef CONFIG_MEMCG_SWAP
186 extern int do_swap_account;
189 static inline bool mem_cgroup_disabled(void)
191 if (mem_cgroup_subsys.disabled)
196 void __mem_cgroup_begin_update_page_stat(struct page *page, bool *locked,
197 unsigned long *flags);
199 extern atomic_t memcg_moving;
201 static inline void mem_cgroup_begin_update_page_stat(struct page *page,
202 bool *locked, unsigned long *flags)
204 if (mem_cgroup_disabled())
208 if (atomic_read(&memcg_moving))
209 __mem_cgroup_begin_update_page_stat(page, locked, flags);
212 void __mem_cgroup_end_update_page_stat(struct page *page,
213 unsigned long *flags);
214 static inline void mem_cgroup_end_update_page_stat(struct page *page,
215 bool *locked, unsigned long *flags)
217 if (mem_cgroup_disabled())
220 __mem_cgroup_end_update_page_stat(page, flags);
224 void mem_cgroup_update_page_stat(struct page *page,
225 enum mem_cgroup_page_stat_item idx,
228 static inline void mem_cgroup_inc_page_stat(struct page *page,
229 enum mem_cgroup_page_stat_item idx)
231 mem_cgroup_update_page_stat(page, idx, 1);
234 static inline void mem_cgroup_dec_page_stat(struct page *page,
235 enum mem_cgroup_page_stat_item idx)
237 mem_cgroup_update_page_stat(page, idx, -1);
240 enum mem_cgroup_filter_t
241 mem_cgroup_soft_reclaim_eligible(struct mem_cgroup *memcg,
242 struct mem_cgroup *root);
244 void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx);
245 static inline void mem_cgroup_count_vm_event(struct mm_struct *mm,
246 enum vm_event_item idx)
248 if (mem_cgroup_disabled())
250 __mem_cgroup_count_vm_event(mm, idx);
252 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
253 void mem_cgroup_split_huge_fixup(struct page *head);
256 #ifdef CONFIG_DEBUG_VM
257 bool mem_cgroup_bad_page_check(struct page *page);
258 void mem_cgroup_print_bad_page(struct page *page);
260 #else /* CONFIG_MEMCG */
263 static inline int mem_cgroup_newpage_charge(struct page *page,
264 struct mm_struct *mm, gfp_t gfp_mask)
269 static inline int mem_cgroup_cache_charge(struct page *page,
270 struct mm_struct *mm, gfp_t gfp_mask)
275 static inline int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
276 struct page *page, gfp_t gfp_mask, struct mem_cgroup **memcgp)
281 static inline void mem_cgroup_commit_charge_swapin(struct page *page,
282 struct mem_cgroup *memcg)
286 static inline void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
290 static inline void mem_cgroup_uncharge_start(void)
294 static inline void mem_cgroup_uncharge_end(void)
298 static inline void mem_cgroup_uncharge_page(struct page *page)
302 static inline void mem_cgroup_uncharge_cache_page(struct page *page)
306 static inline struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
307 struct mem_cgroup *memcg)
309 return &zone->lruvec;
312 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
315 return &zone->lruvec;
318 static inline struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
323 static inline struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
328 static inline bool mm_match_cgroup(struct mm_struct *mm,
329 struct mem_cgroup *memcg)
334 static inline bool task_in_mem_cgroup(struct task_struct *task,
335 const struct mem_cgroup *memcg)
340 static inline struct cgroup_subsys_state
341 *mem_cgroup_css(struct mem_cgroup *memcg)
347 mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
348 struct mem_cgroup **memcgp)
352 static inline void mem_cgroup_end_migration(struct mem_cgroup *memcg,
353 struct page *oldpage, struct page *newpage, bool migration_ok)
356 static inline struct mem_cgroup *
357 mem_cgroup_iter_cond(struct mem_cgroup *root,
358 struct mem_cgroup *prev,
359 struct mem_cgroup_reclaim_cookie *reclaim,
360 mem_cgroup_iter_filter cond)
362 /* first call must return non-NULL, second return NULL */
363 return (struct mem_cgroup *)(unsigned long)!prev;
366 static inline struct mem_cgroup *
367 mem_cgroup_iter(struct mem_cgroup *root,
368 struct mem_cgroup *prev,
369 struct mem_cgroup_reclaim_cookie *reclaim)
374 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
375 struct mem_cgroup *prev)
379 static inline bool mem_cgroup_disabled(void)
385 mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
390 static inline unsigned long
391 mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
397 mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
403 mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
407 static inline void mem_cgroup_begin_update_page_stat(struct page *page,
408 bool *locked, unsigned long *flags)
412 static inline void mem_cgroup_end_update_page_stat(struct page *page,
413 bool *locked, unsigned long *flags)
417 static inline bool mem_cgroup_toggle_oom(bool new)
422 static inline void mem_cgroup_enable_oom(void)
426 static inline void mem_cgroup_disable_oom(void)
430 static inline void mem_cgroup_inc_page_stat(struct page *page,
431 enum mem_cgroup_page_stat_item idx)
435 static inline void mem_cgroup_dec_page_stat(struct page *page,
436 enum mem_cgroup_page_stat_item idx)
441 enum mem_cgroup_filter_t
442 mem_cgroup_soft_reclaim_eligible(struct mem_cgroup *memcg,
443 struct mem_cgroup *root)
448 static inline void mem_cgroup_split_huge_fixup(struct page *head)
453 void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
456 static inline void mem_cgroup_replace_page_cache(struct page *oldpage,
457 struct page *newpage)
460 #endif /* CONFIG_MEMCG */
462 #if !defined(CONFIG_MEMCG) || !defined(CONFIG_DEBUG_VM)
464 mem_cgroup_bad_page_check(struct page *page)
470 mem_cgroup_print_bad_page(struct page *page)
482 #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
483 void sock_update_memcg(struct sock *sk);
484 void sock_release_memcg(struct sock *sk);
486 static inline void sock_update_memcg(struct sock *sk)
489 static inline void sock_release_memcg(struct sock *sk)
492 #endif /* CONFIG_INET && CONFIG_MEMCG_KMEM */
494 #ifdef CONFIG_MEMCG_KMEM
495 extern struct static_key memcg_kmem_enabled_key;
497 extern int memcg_limited_groups_array_size;
500 * Helper macro to loop through all memcg-specific caches. Callers must still
501 * check if the cache is valid (it is either valid or NULL).
502 * the slab_mutex must be held when looping through those caches
504 #define for_each_memcg_cache_index(_idx) \
505 for ((_idx) = 0; (_idx) < memcg_limited_groups_array_size; (_idx)++)
507 static inline bool memcg_kmem_enabled(void)
509 return static_key_false(&memcg_kmem_enabled_key);
513 * In general, we'll do everything in our power to not incur in any overhead
514 * for non-memcg users for the kmem functions. Not even a function call, if we
517 * Therefore, we'll inline all those functions so that in the best case, we'll
518 * see that kmemcg is off for everybody and proceed quickly. If it is on,
519 * we'll still do most of the flag checking inline. We check a lot of
520 * conditions, but because they are pretty simple, they are expected to be
523 bool __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg,
525 void __memcg_kmem_commit_charge(struct page *page,
526 struct mem_cgroup *memcg, int order);
527 void __memcg_kmem_uncharge_pages(struct page *page, int order);
529 int memcg_cache_id(struct mem_cgroup *memcg);
530 int memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s,
531 struct kmem_cache *root_cache);
532 void memcg_release_cache(struct kmem_cache *cachep);
533 void memcg_cache_list_add(struct mem_cgroup *memcg, struct kmem_cache *cachep);
535 int memcg_update_cache_size(struct kmem_cache *s, int num_groups);
536 void memcg_update_array_size(int num_groups);
539 __memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp);
541 void mem_cgroup_destroy_cache(struct kmem_cache *cachep);
542 void kmem_cache_destroy_memcg_children(struct kmem_cache *s);
545 * memcg_kmem_newpage_charge: verify if a new kmem allocation is allowed.
546 * @gfp: the gfp allocation flags.
547 * @memcg: a pointer to the memcg this was charged against.
548 * @order: allocation order.
550 * returns true if the memcg where the current task belongs can hold this
553 * We return true automatically if this allocation is not to be accounted to
557 memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg, int order)
559 if (!memcg_kmem_enabled())
563 * __GFP_NOFAIL allocations will move on even if charging is not
564 * possible. Therefore we don't even try, and have this allocation
565 * unaccounted. We could in theory charge it with
566 * res_counter_charge_nofail, but we hope those allocations are rare,
567 * and won't be worth the trouble.
569 if (!(gfp & __GFP_KMEMCG) || (gfp & __GFP_NOFAIL))
571 if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
574 /* If the test is dying, just let it go. */
575 if (unlikely(fatal_signal_pending(current)))
578 return __memcg_kmem_newpage_charge(gfp, memcg, order);
582 * memcg_kmem_uncharge_pages: uncharge pages from memcg
583 * @page: pointer to struct page being freed
584 * @order: allocation order.
586 * there is no need to specify memcg here, since it is embedded in page_cgroup
589 memcg_kmem_uncharge_pages(struct page *page, int order)
591 if (memcg_kmem_enabled())
592 __memcg_kmem_uncharge_pages(page, order);
596 * memcg_kmem_commit_charge: embeds correct memcg in a page
597 * @page: pointer to struct page recently allocated
598 * @memcg: the memcg structure we charged against
599 * @order: allocation order.
601 * Needs to be called after memcg_kmem_newpage_charge, regardless of success or
602 * failure of the allocation. if @page is NULL, this function will revert the
603 * charges. Otherwise, it will commit the memcg given by @memcg to the
604 * corresponding page_cgroup.
607 memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, int order)
609 if (memcg_kmem_enabled() && memcg)
610 __memcg_kmem_commit_charge(page, memcg, order);
614 * memcg_kmem_get_cache: selects the correct per-memcg cache for allocation
615 * @cachep: the original global kmem cache
616 * @gfp: allocation flags.
618 * This function assumes that the task allocating, which determines the memcg
619 * in the page allocator, belongs to the same cgroup throughout the whole
620 * process. Misacounting can happen if the task calls memcg_kmem_get_cache()
621 * while belonging to a cgroup, and later on changes. This is considered
622 * acceptable, and should only happen upon task migration.
624 * Before the cache is created by the memcg core, there is also a possible
625 * imbalance: the task belongs to a memcg, but the cache being allocated from
626 * is the global cache, since the child cache is not yet guaranteed to be
627 * ready. This case is also fine, since in this case the GFP_KMEMCG will not be
628 * passed and the page allocator will not attempt any cgroup accounting.
630 static __always_inline struct kmem_cache *
631 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
633 if (!memcg_kmem_enabled())
635 if (gfp & __GFP_NOFAIL)
637 if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
639 if (unlikely(fatal_signal_pending(current)))
642 return __memcg_kmem_get_cache(cachep, gfp);
645 #define for_each_memcg_cache_index(_idx) \
648 static inline bool memcg_kmem_enabled(void)
654 memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg, int order)
659 static inline void memcg_kmem_uncharge_pages(struct page *page, int order)
664 memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, int order)
668 static inline int memcg_cache_id(struct mem_cgroup *memcg)
674 memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s,
675 struct kmem_cache *root_cache)
680 static inline void memcg_release_cache(struct kmem_cache *cachep)
684 static inline void memcg_cache_list_add(struct mem_cgroup *memcg,
685 struct kmem_cache *s)
689 static inline struct kmem_cache *
690 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
695 static inline void kmem_cache_destroy_memcg_children(struct kmem_cache *s)
698 #endif /* CONFIG_MEMCG_KMEM */
699 #endif /* _LINUX_MEMCONTROL_H */