1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /* memcontrol.h - Memory Controller
4 * Copyright IBM Corporation, 2007
5 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
7 * Copyright 2007 OpenVZ SWsoft Inc
8 * Author: Pavel Emelianov <xemul@openvz.org>
11 #ifndef _LINUX_MEMCONTROL_H
12 #define _LINUX_MEMCONTROL_H
13 #include <linux/cgroup.h>
14 #include <linux/vm_event_item.h>
15 #include <linux/hardirq.h>
16 #include <linux/jump_label.h>
17 #include <linux/page_counter.h>
18 #include <linux/vmpressure.h>
19 #include <linux/eventfd.h>
21 #include <linux/vmstat.h>
22 #include <linux/writeback.h>
23 #include <linux/page-flags.h>
31 /* Cgroup-specific page state, on top of universal node page state */
32 enum memcg_stat_item {
33 MEMCG_SWAP = NR_VM_NODE_STAT_ITEMS,
39 enum memcg_memory_event {
48 MEMCG_NR_MEMORY_EVENTS,
51 struct mem_cgroup_reclaim_cookie {
53 unsigned int generation;
58 #define MEM_CGROUP_ID_SHIFT 16
59 #define MEM_CGROUP_ID_MAX USHRT_MAX
61 struct mem_cgroup_id {
67 * Per memcg event counter is incremented at every pagein/pageout. With THP,
68 * it will be incremented by the number of pages. This counter is used
69 * to trigger some periodic events. This is straightforward and better
70 * than using jiffies etc. to handle periodic memcg event.
72 enum mem_cgroup_events_target {
73 MEM_CGROUP_TARGET_THRESH,
74 MEM_CGROUP_TARGET_SOFTLIMIT,
78 struct memcg_vmstats_percpu {
79 /* Local (CPU and cgroup) page state & events */
80 long state[MEMCG_NR_STAT];
81 unsigned long events[NR_VM_EVENT_ITEMS];
83 /* Delta calculation for lockless upward propagation */
84 long state_prev[MEMCG_NR_STAT];
85 unsigned long events_prev[NR_VM_EVENT_ITEMS];
87 /* Cgroup1: threshold notifications & softlimit tree updates */
88 unsigned long nr_page_events;
89 unsigned long targets[MEM_CGROUP_NTARGETS];
92 struct memcg_vmstats {
93 /* Aggregated (CPU and subtree) page state & events */
94 long state[MEMCG_NR_STAT];
95 unsigned long events[NR_VM_EVENT_ITEMS];
97 /* Pending child counts during tree propagation */
98 long state_pending[MEMCG_NR_STAT];
99 unsigned long events_pending[NR_VM_EVENT_ITEMS];
102 struct mem_cgroup_reclaim_iter {
103 struct mem_cgroup *position;
104 /* scan generation, increased every round-trip */
105 unsigned int generation;
109 * Bitmap and deferred work of shrinker::id corresponding to memcg-aware
110 * shrinkers, which have elements charged to this memcg.
112 struct shrinker_info {
114 atomic_long_t *nr_deferred;
118 struct lruvec_stats_percpu {
119 /* Local (CPU and cgroup) state */
120 long state[NR_VM_NODE_STAT_ITEMS];
122 /* Delta calculation for lockless upward propagation */
123 long state_prev[NR_VM_NODE_STAT_ITEMS];
126 struct lruvec_stats {
127 /* Aggregated (CPU and subtree) state */
128 long state[NR_VM_NODE_STAT_ITEMS];
130 /* Pending child counts during tree propagation */
131 long state_pending[NR_VM_NODE_STAT_ITEMS];
135 * per-node information in memory controller.
137 struct mem_cgroup_per_node {
138 struct lruvec lruvec;
140 struct lruvec_stats_percpu __percpu *lruvec_stats_percpu;
141 struct lruvec_stats lruvec_stats;
143 unsigned long lru_zone_size[MAX_NR_ZONES][NR_LRU_LISTS];
145 struct mem_cgroup_reclaim_iter iter;
147 struct shrinker_info __rcu *shrinker_info;
149 struct rb_node tree_node; /* RB tree node */
150 unsigned long usage_in_excess;/* Set to the value by which */
151 /* the soft limit is exceeded*/
153 struct mem_cgroup *memcg; /* Back pointer, we cannot */
154 /* use container_of */
157 struct mem_cgroup_threshold {
158 struct eventfd_ctx *eventfd;
159 unsigned long threshold;
163 struct mem_cgroup_threshold_ary {
164 /* An array index points to threshold just below or equal to usage. */
165 int current_threshold;
166 /* Size of entries[] */
168 /* Array of thresholds */
169 struct mem_cgroup_threshold entries[];
172 struct mem_cgroup_thresholds {
173 /* Primary thresholds array */
174 struct mem_cgroup_threshold_ary *primary;
176 * Spare threshold array.
177 * This is needed to make mem_cgroup_unregister_event() "never fail".
178 * It must be able to store at least primary->size - 1 entries.
180 struct mem_cgroup_threshold_ary *spare;
183 enum memcg_kmem_state {
189 #if defined(CONFIG_SMP)
190 struct memcg_padding {
192 } ____cacheline_internodealigned_in_smp;
193 #define MEMCG_PADDING(name) struct memcg_padding name
195 #define MEMCG_PADDING(name)
199 * Remember four most recent foreign writebacks with dirty pages in this
200 * cgroup. Inode sharing is expected to be uncommon and, even if we miss
201 * one in a given round, we're likely to catch it later if it keeps
202 * foreign-dirtying, so a fairly low count should be enough.
204 * See mem_cgroup_track_foreign_dirty_slowpath() for details.
206 #define MEMCG_CGWB_FRN_CNT 4
208 struct memcg_cgwb_frn {
209 u64 bdi_id; /* bdi->id of the foreign inode */
210 int memcg_id; /* memcg->css.id of foreign inode */
211 u64 at; /* jiffies_64 at the time of dirtying */
212 struct wb_completion done; /* tracks in-flight foreign writebacks */
216 * Bucket for arbitrarily byte-sized objects charged to a memory
217 * cgroup. The bucket can be reparented in one piece when the cgroup
218 * is destroyed, without having to round up the individual references
219 * of all live memory objects in the wild.
222 struct percpu_ref refcnt;
223 struct mem_cgroup *memcg;
224 atomic_t nr_charged_bytes;
226 struct list_head list;
232 * The memory controller data structure. The memory controller controls both
233 * page cache and RSS per cgroup. We would eventually like to provide
234 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
235 * to help the administrator determine what knobs to tune.
238 struct cgroup_subsys_state css;
240 /* Private memcg ID. Used to ID objects that outlive the cgroup */
241 struct mem_cgroup_id id;
243 /* Accounted resources */
244 struct page_counter memory; /* Both v1 & v2 */
247 struct page_counter swap; /* v2 only */
248 struct page_counter memsw; /* v1 only */
251 /* Legacy consumer-oriented counters */
252 struct page_counter kmem; /* v1 only */
253 struct page_counter tcpmem; /* v1 only */
255 /* Range enforcement for interrupt charges */
256 struct work_struct high_work;
258 unsigned long soft_limit;
260 /* vmpressure notifications */
261 struct vmpressure vmpressure;
264 * Should the OOM killer kill all belonging tasks, had it kill one?
268 /* protected by memcg_oom_lock */
273 /* OOM-Killer disable */
274 int oom_kill_disable;
276 /* memory.events and memory.events.local */
277 struct cgroup_file events_file;
278 struct cgroup_file events_local_file;
280 /* handle for "memory.swap.events" */
281 struct cgroup_file swap_events_file;
283 /* protect arrays of thresholds */
284 struct mutex thresholds_lock;
286 /* thresholds for memory usage. RCU-protected */
287 struct mem_cgroup_thresholds thresholds;
289 /* thresholds for mem+swap usage. RCU-protected */
290 struct mem_cgroup_thresholds memsw_thresholds;
292 /* For oom notifier event fd */
293 struct list_head oom_notify;
296 * Should we move charges of a task when a task is moved into this
297 * mem_cgroup ? And what type of charges should we move ?
299 unsigned long move_charge_at_immigrate;
300 /* taken only while moving_account > 0 */
301 spinlock_t move_lock;
302 unsigned long move_lock_flags;
304 MEMCG_PADDING(_pad1_);
307 struct memcg_vmstats vmstats;
310 atomic_long_t memory_events[MEMCG_NR_MEMORY_EVENTS];
311 atomic_long_t memory_events_local[MEMCG_NR_MEMORY_EVENTS];
313 unsigned long socket_pressure;
315 /* Legacy tcp memory accounting */
319 #ifdef CONFIG_MEMCG_KMEM
321 enum memcg_kmem_state kmem_state;
322 struct obj_cgroup __rcu *objcg;
323 struct list_head objcg_list; /* list of inherited objcgs */
326 MEMCG_PADDING(_pad2_);
329 * set > 0 if pages under this cgroup are moving to other cgroup.
331 atomic_t moving_account;
332 struct task_struct *move_lock_task;
334 struct memcg_vmstats_percpu __percpu *vmstats_percpu;
336 #ifdef CONFIG_CGROUP_WRITEBACK
337 struct list_head cgwb_list;
338 struct wb_domain cgwb_domain;
339 struct memcg_cgwb_frn cgwb_frn[MEMCG_CGWB_FRN_CNT];
342 /* List of events which userspace want to receive */
343 struct list_head event_list;
344 spinlock_t event_list_lock;
346 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
347 struct deferred_split deferred_split_queue;
350 struct mem_cgroup_per_node *nodeinfo[];
354 * size of first charge trial. "32" comes from vmscan.c's magic value.
355 * TODO: maybe necessary to use big numbers in big irons.
357 #define MEMCG_CHARGE_BATCH 32U
359 extern struct mem_cgroup *root_mem_cgroup;
361 enum page_memcg_data_flags {
362 /* page->memcg_data is a pointer to an objcgs vector */
363 MEMCG_DATA_OBJCGS = (1UL << 0),
364 /* page has been accounted as a non-slab kernel page */
365 MEMCG_DATA_KMEM = (1UL << 1),
366 /* the next bit after the last actual flag */
367 __NR_MEMCG_DATA_FLAGS = (1UL << 2),
370 #define MEMCG_DATA_FLAGS_MASK (__NR_MEMCG_DATA_FLAGS - 1)
372 static inline bool PageMemcgKmem(struct page *page);
375 * After the initialization objcg->memcg is always pointing at
376 * a valid memcg, but can be atomically swapped to the parent memcg.
378 * The caller must ensure that the returned memcg won't be released:
379 * e.g. acquire the rcu_read_lock or css_set_lock.
381 static inline struct mem_cgroup *obj_cgroup_memcg(struct obj_cgroup *objcg)
383 return READ_ONCE(objcg->memcg);
387 * __page_memcg - get the memory cgroup associated with a non-kmem page
388 * @page: a pointer to the page struct
390 * Returns a pointer to the memory cgroup associated with the page,
391 * or NULL. This function assumes that the page is known to have a
392 * proper memory cgroup pointer. It's not safe to call this function
393 * against some type of pages, e.g. slab pages or ex-slab pages or
396 static inline struct mem_cgroup *__page_memcg(struct page *page)
398 unsigned long memcg_data = page->memcg_data;
400 VM_BUG_ON_PAGE(PageSlab(page), page);
401 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_OBJCGS, page);
402 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, page);
404 return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
408 * __page_objcg - get the object cgroup associated with a kmem page
409 * @page: a pointer to the page struct
411 * Returns a pointer to the object cgroup associated with the page,
412 * or NULL. This function assumes that the page is known to have a
413 * proper object cgroup pointer. It's not safe to call this function
414 * against some type of pages, e.g. slab pages or ex-slab pages or
417 static inline struct obj_cgroup *__page_objcg(struct page *page)
419 unsigned long memcg_data = page->memcg_data;
421 VM_BUG_ON_PAGE(PageSlab(page), page);
422 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_OBJCGS, page);
423 VM_BUG_ON_PAGE(!(memcg_data & MEMCG_DATA_KMEM), page);
425 return (struct obj_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
429 * page_memcg - get the memory cgroup associated with a page
430 * @page: a pointer to the page struct
432 * Returns a pointer to the memory cgroup associated with the page,
433 * or NULL. This function assumes that the page is known to have a
434 * proper memory cgroup pointer. It's not safe to call this function
435 * against some type of pages, e.g. slab pages or ex-slab pages.
437 * For a non-kmem page any of the following ensures page and memcg binding
442 * - lock_page_memcg()
443 * - exclusive reference
445 * For a kmem page a caller should hold an rcu read lock to protect memcg
446 * associated with a kmem page from being released.
448 static inline struct mem_cgroup *page_memcg(struct page *page)
450 if (PageMemcgKmem(page))
451 return obj_cgroup_memcg(__page_objcg(page));
453 return __page_memcg(page);
457 * page_memcg_rcu - locklessly get the memory cgroup associated with a page
458 * @page: a pointer to the page struct
460 * Returns a pointer to the memory cgroup associated with the page,
461 * or NULL. This function assumes that the page is known to have a
462 * proper memory cgroup pointer. It's not safe to call this function
463 * against some type of pages, e.g. slab pages or ex-slab pages.
465 static inline struct mem_cgroup *page_memcg_rcu(struct page *page)
467 unsigned long memcg_data = READ_ONCE(page->memcg_data);
469 VM_BUG_ON_PAGE(PageSlab(page), page);
470 WARN_ON_ONCE(!rcu_read_lock_held());
472 if (memcg_data & MEMCG_DATA_KMEM) {
473 struct obj_cgroup *objcg;
475 objcg = (void *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
476 return obj_cgroup_memcg(objcg);
479 return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
483 * page_memcg_check - get the memory cgroup associated with a page
484 * @page: a pointer to the page struct
486 * Returns a pointer to the memory cgroup associated with the page,
487 * or NULL. This function unlike page_memcg() can take any page
488 * as an argument. It has to be used in cases when it's not known if a page
489 * has an associated memory cgroup pointer or an object cgroups vector or
492 * For a non-kmem page any of the following ensures page and memcg binding
497 * - lock_page_memcg()
498 * - exclusive reference
500 * For a kmem page a caller should hold an rcu read lock to protect memcg
501 * associated with a kmem page from being released.
503 static inline struct mem_cgroup *page_memcg_check(struct page *page)
506 * Because page->memcg_data might be changed asynchronously
507 * for slab pages, READ_ONCE() should be used here.
509 unsigned long memcg_data = READ_ONCE(page->memcg_data);
511 if (memcg_data & MEMCG_DATA_OBJCGS)
514 if (memcg_data & MEMCG_DATA_KMEM) {
515 struct obj_cgroup *objcg;
517 objcg = (void *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
518 return obj_cgroup_memcg(objcg);
521 return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
524 #ifdef CONFIG_MEMCG_KMEM
526 * PageMemcgKmem - check if the page has MemcgKmem flag set
527 * @page: a pointer to the page struct
529 * Checks if the page has MemcgKmem flag set. The caller must ensure that
530 * the page has an associated memory cgroup. It's not safe to call this function
531 * against some types of pages, e.g. slab pages.
533 static inline bool PageMemcgKmem(struct page *page)
535 VM_BUG_ON_PAGE(page->memcg_data & MEMCG_DATA_OBJCGS, page);
536 return page->memcg_data & MEMCG_DATA_KMEM;
540 * page_objcgs - get the object cgroups vector associated with a page
541 * @page: a pointer to the page struct
543 * Returns a pointer to the object cgroups vector associated with the page,
544 * or NULL. This function assumes that the page is known to have an
545 * associated object cgroups vector. It's not safe to call this function
546 * against pages, which might have an associated memory cgroup: e.g.
547 * kernel stack pages.
549 static inline struct obj_cgroup **page_objcgs(struct page *page)
551 unsigned long memcg_data = READ_ONCE(page->memcg_data);
553 VM_BUG_ON_PAGE(memcg_data && !(memcg_data & MEMCG_DATA_OBJCGS), page);
554 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, page);
556 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
560 * page_objcgs_check - get the object cgroups vector associated with a page
561 * @page: a pointer to the page struct
563 * Returns a pointer to the object cgroups vector associated with the page,
564 * or NULL. This function is safe to use if the page can be directly associated
565 * with a memory cgroup.
567 static inline struct obj_cgroup **page_objcgs_check(struct page *page)
569 unsigned long memcg_data = READ_ONCE(page->memcg_data);
571 if (!memcg_data || !(memcg_data & MEMCG_DATA_OBJCGS))
574 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, page);
576 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
580 static inline bool PageMemcgKmem(struct page *page)
585 static inline struct obj_cgroup **page_objcgs(struct page *page)
590 static inline struct obj_cgroup **page_objcgs_check(struct page *page)
596 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
598 return (memcg == root_mem_cgroup);
601 static inline bool mem_cgroup_disabled(void)
603 return !cgroup_subsys_enabled(memory_cgrp_subsys);
606 static inline void mem_cgroup_protection(struct mem_cgroup *root,
607 struct mem_cgroup *memcg,
613 if (mem_cgroup_disabled())
617 * There is no reclaim protection applied to a targeted reclaim.
618 * We are special casing this specific case here because
619 * mem_cgroup_protected calculation is not robust enough to keep
620 * the protection invariant for calculated effective values for
621 * parallel reclaimers with different reclaim target. This is
622 * especially a problem for tail memcgs (as they have pages on LRU)
623 * which would want to have effective values 0 for targeted reclaim
624 * but a different value for external reclaim.
627 * Let's have global and A's reclaim in parallel:
629 * A (low=2G, usage = 3G, max = 3G, children_low_usage = 1.5G)
631 * | C (low = 1G, usage = 2.5G)
632 * B (low = 1G, usage = 0.5G)
634 * For the global reclaim
636 * B.elow = min(B.usage, B.low) because children_low_usage <= A.elow
637 * C.elow = min(C.usage, C.low)
639 * With the effective values resetting we have A reclaim
644 * If the global reclaim races with A's reclaim then
645 * B.elow = C.elow = 0 because children_low_usage > A.elow)
646 * is possible and reclaiming B would be violating the protection.
652 *min = READ_ONCE(memcg->memory.emin);
653 *low = READ_ONCE(memcg->memory.elow);
656 void mem_cgroup_calculate_protection(struct mem_cgroup *root,
657 struct mem_cgroup *memcg);
659 static inline bool mem_cgroup_supports_protection(struct mem_cgroup *memcg)
662 * The root memcg doesn't account charges, and doesn't support
665 return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg);
669 static inline bool mem_cgroup_below_low(struct mem_cgroup *memcg)
671 if (!mem_cgroup_supports_protection(memcg))
674 return READ_ONCE(memcg->memory.elow) >=
675 page_counter_read(&memcg->memory);
678 static inline bool mem_cgroup_below_min(struct mem_cgroup *memcg)
680 if (!mem_cgroup_supports_protection(memcg))
683 return READ_ONCE(memcg->memory.emin) >=
684 page_counter_read(&memcg->memory);
687 int __mem_cgroup_charge(struct page *page, struct mm_struct *mm,
689 static inline int mem_cgroup_charge(struct page *page, struct mm_struct *mm,
692 if (mem_cgroup_disabled())
694 return __mem_cgroup_charge(page, mm, gfp_mask);
697 int mem_cgroup_swapin_charge_page(struct page *page, struct mm_struct *mm,
698 gfp_t gfp, swp_entry_t entry);
699 void mem_cgroup_swapin_uncharge_swap(swp_entry_t entry);
701 void __mem_cgroup_uncharge(struct page *page);
702 static inline void mem_cgroup_uncharge(struct page *page)
704 if (mem_cgroup_disabled())
706 __mem_cgroup_uncharge(page);
709 void __mem_cgroup_uncharge_list(struct list_head *page_list);
710 static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
712 if (mem_cgroup_disabled())
714 __mem_cgroup_uncharge_list(page_list);
717 void mem_cgroup_migrate(struct page *oldpage, struct page *newpage);
720 * mem_cgroup_lruvec - get the lru list vector for a memcg & node
721 * @memcg: memcg of the wanted lruvec
722 * @pgdat: pglist_data
724 * Returns the lru list vector holding pages for a given @memcg &
725 * @pgdat combination. This can be the node lruvec, if the memory
726 * controller is disabled.
728 static inline struct lruvec *mem_cgroup_lruvec(struct mem_cgroup *memcg,
729 struct pglist_data *pgdat)
731 struct mem_cgroup_per_node *mz;
732 struct lruvec *lruvec;
734 if (mem_cgroup_disabled()) {
735 lruvec = &pgdat->__lruvec;
740 memcg = root_mem_cgroup;
742 mz = memcg->nodeinfo[pgdat->node_id];
743 lruvec = &mz->lruvec;
746 * Since a node can be onlined after the mem_cgroup was created,
747 * we have to be prepared to initialize lruvec->pgdat here;
748 * and if offlined then reonlined, we need to reinitialize it.
750 if (unlikely(lruvec->pgdat != pgdat))
751 lruvec->pgdat = pgdat;
756 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
759 * This function relies on page->mem_cgroup being stable.
761 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page)
763 pg_data_t *pgdat = page_pgdat(page);
764 struct mem_cgroup *memcg = page_memcg(page);
766 VM_WARN_ON_ONCE_PAGE(!memcg && !mem_cgroup_disabled(), page);
767 return mem_cgroup_lruvec(memcg, pgdat);
770 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
772 struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm);
774 struct lruvec *lock_page_lruvec(struct page *page);
775 struct lruvec *lock_page_lruvec_irq(struct page *page);
776 struct lruvec *lock_page_lruvec_irqsave(struct page *page,
777 unsigned long *flags);
779 #ifdef CONFIG_DEBUG_VM
780 void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page);
782 static inline void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page)
788 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
789 return css ? container_of(css, struct mem_cgroup, css) : NULL;
792 static inline bool obj_cgroup_tryget(struct obj_cgroup *objcg)
794 return percpu_ref_tryget(&objcg->refcnt);
797 static inline void obj_cgroup_get(struct obj_cgroup *objcg)
799 percpu_ref_get(&objcg->refcnt);
802 static inline void obj_cgroup_get_many(struct obj_cgroup *objcg,
805 percpu_ref_get_many(&objcg->refcnt, nr);
808 static inline void obj_cgroup_put(struct obj_cgroup *objcg)
810 percpu_ref_put(&objcg->refcnt);
813 static inline void mem_cgroup_put(struct mem_cgroup *memcg)
816 css_put(&memcg->css);
819 #define mem_cgroup_from_counter(counter, member) \
820 container_of(counter, struct mem_cgroup, member)
822 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
824 struct mem_cgroup_reclaim_cookie *);
825 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
826 int mem_cgroup_scan_tasks(struct mem_cgroup *,
827 int (*)(struct task_struct *, void *), void *);
829 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
831 if (mem_cgroup_disabled())
836 struct mem_cgroup *mem_cgroup_from_id(unsigned short id);
838 static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m)
840 return mem_cgroup_from_css(seq_css(m));
843 static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec)
845 struct mem_cgroup_per_node *mz;
847 if (mem_cgroup_disabled())
850 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
855 * parent_mem_cgroup - find the accounting parent of a memcg
856 * @memcg: memcg whose parent to find
858 * Returns the parent memcg, or NULL if this is the root or the memory
859 * controller is in legacy no-hierarchy mode.
861 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
863 if (!memcg->memory.parent)
865 return mem_cgroup_from_counter(memcg->memory.parent, memory);
868 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
869 struct mem_cgroup *root)
873 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
876 static inline bool mm_match_cgroup(struct mm_struct *mm,
877 struct mem_cgroup *memcg)
879 struct mem_cgroup *task_memcg;
883 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
885 match = mem_cgroup_is_descendant(task_memcg, memcg);
890 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
891 ino_t page_cgroup_ino(struct page *page);
893 static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
895 if (mem_cgroup_disabled())
897 return !!(memcg->css.flags & CSS_ONLINE);
900 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
901 int zid, int nr_pages);
904 unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
905 enum lru_list lru, int zone_idx)
907 struct mem_cgroup_per_node *mz;
909 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
910 return READ_ONCE(mz->lru_zone_size[zone_idx][lru]);
913 void mem_cgroup_handle_over_high(void);
915 unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg);
917 unsigned long mem_cgroup_size(struct mem_cgroup *memcg);
919 void mem_cgroup_print_oom_context(struct mem_cgroup *memcg,
920 struct task_struct *p);
922 void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg);
924 static inline void mem_cgroup_enter_user_fault(void)
926 WARN_ON(current->in_user_fault);
927 current->in_user_fault = 1;
930 static inline void mem_cgroup_exit_user_fault(void)
932 WARN_ON(!current->in_user_fault);
933 current->in_user_fault = 0;
936 static inline bool task_in_memcg_oom(struct task_struct *p)
938 return p->memcg_in_oom;
941 bool mem_cgroup_oom_synchronize(bool wait);
942 struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim,
943 struct mem_cgroup *oom_domain);
944 void mem_cgroup_print_oom_group(struct mem_cgroup *memcg);
946 #ifdef CONFIG_MEMCG_SWAP
947 extern bool cgroup_memory_noswap;
950 void lock_page_memcg(struct page *page);
951 void unlock_page_memcg(struct page *page);
953 void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val);
955 /* idx can be of type enum memcg_stat_item or node_stat_item */
956 static inline void mod_memcg_state(struct mem_cgroup *memcg,
961 local_irq_save(flags);
962 __mod_memcg_state(memcg, idx, val);
963 local_irq_restore(flags);
966 static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx)
968 return READ_ONCE(memcg->vmstats.state[idx]);
971 static inline unsigned long lruvec_page_state(struct lruvec *lruvec,
972 enum node_stat_item idx)
974 struct mem_cgroup_per_node *pn;
976 if (mem_cgroup_disabled())
977 return node_page_state(lruvec_pgdat(lruvec), idx);
979 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
980 return READ_ONCE(pn->lruvec_stats.state[idx]);
983 static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec,
984 enum node_stat_item idx)
986 struct mem_cgroup_per_node *pn;
990 if (mem_cgroup_disabled())
991 return node_page_state(lruvec_pgdat(lruvec), idx);
993 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
994 for_each_possible_cpu(cpu)
995 x += per_cpu(pn->lruvec_stats_percpu->state[idx], cpu);
1003 void mem_cgroup_flush_stats(void);
1005 void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx,
1007 void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, int val);
1009 static inline void mod_lruvec_kmem_state(void *p, enum node_stat_item idx,
1012 unsigned long flags;
1014 local_irq_save(flags);
1015 __mod_lruvec_kmem_state(p, idx, val);
1016 local_irq_restore(flags);
1019 static inline void mod_memcg_lruvec_state(struct lruvec *lruvec,
1020 enum node_stat_item idx, int val)
1022 unsigned long flags;
1024 local_irq_save(flags);
1025 __mod_memcg_lruvec_state(lruvec, idx, val);
1026 local_irq_restore(flags);
1029 void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx,
1030 unsigned long count);
1032 static inline void count_memcg_events(struct mem_cgroup *memcg,
1033 enum vm_event_item idx,
1034 unsigned long count)
1036 unsigned long flags;
1038 local_irq_save(flags);
1039 __count_memcg_events(memcg, idx, count);
1040 local_irq_restore(flags);
1043 static inline void count_memcg_page_event(struct page *page,
1044 enum vm_event_item idx)
1046 struct mem_cgroup *memcg = page_memcg(page);
1049 count_memcg_events(memcg, idx, 1);
1052 static inline void count_memcg_event_mm(struct mm_struct *mm,
1053 enum vm_event_item idx)
1055 struct mem_cgroup *memcg;
1057 if (mem_cgroup_disabled())
1061 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1063 count_memcg_events(memcg, idx, 1);
1067 static inline void memcg_memory_event(struct mem_cgroup *memcg,
1068 enum memcg_memory_event event)
1070 bool swap_event = event == MEMCG_SWAP_HIGH || event == MEMCG_SWAP_MAX ||
1071 event == MEMCG_SWAP_FAIL;
1073 atomic_long_inc(&memcg->memory_events_local[event]);
1075 cgroup_file_notify(&memcg->events_local_file);
1078 atomic_long_inc(&memcg->memory_events[event]);
1080 cgroup_file_notify(&memcg->swap_events_file);
1082 cgroup_file_notify(&memcg->events_file);
1084 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
1086 if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
1088 } while ((memcg = parent_mem_cgroup(memcg)) &&
1089 !mem_cgroup_is_root(memcg));
1092 static inline void memcg_memory_event_mm(struct mm_struct *mm,
1093 enum memcg_memory_event event)
1095 struct mem_cgroup *memcg;
1097 if (mem_cgroup_disabled())
1101 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1103 memcg_memory_event(memcg, event);
1107 void split_page_memcg(struct page *head, unsigned int nr);
1109 unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
1111 unsigned long *total_scanned);
1113 #else /* CONFIG_MEMCG */
1115 #define MEM_CGROUP_ID_SHIFT 0
1116 #define MEM_CGROUP_ID_MAX 0
1118 static inline struct mem_cgroup *page_memcg(struct page *page)
1123 static inline struct mem_cgroup *page_memcg_rcu(struct page *page)
1125 WARN_ON_ONCE(!rcu_read_lock_held());
1129 static inline struct mem_cgroup *page_memcg_check(struct page *page)
1134 static inline bool PageMemcgKmem(struct page *page)
1139 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
1144 static inline bool mem_cgroup_disabled(void)
1149 static inline void memcg_memory_event(struct mem_cgroup *memcg,
1150 enum memcg_memory_event event)
1154 static inline void memcg_memory_event_mm(struct mm_struct *mm,
1155 enum memcg_memory_event event)
1159 static inline void mem_cgroup_protection(struct mem_cgroup *root,
1160 struct mem_cgroup *memcg,
1167 static inline void mem_cgroup_calculate_protection(struct mem_cgroup *root,
1168 struct mem_cgroup *memcg)
1172 static inline bool mem_cgroup_below_low(struct mem_cgroup *memcg)
1177 static inline bool mem_cgroup_below_min(struct mem_cgroup *memcg)
1182 static inline int mem_cgroup_charge(struct page *page, struct mm_struct *mm,
1188 static inline int mem_cgroup_swapin_charge_page(struct page *page,
1189 struct mm_struct *mm, gfp_t gfp, swp_entry_t entry)
1194 static inline void mem_cgroup_swapin_uncharge_swap(swp_entry_t entry)
1198 static inline void mem_cgroup_uncharge(struct page *page)
1202 static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
1206 static inline void mem_cgroup_migrate(struct page *old, struct page *new)
1210 static inline struct lruvec *mem_cgroup_lruvec(struct mem_cgroup *memcg,
1211 struct pglist_data *pgdat)
1213 return &pgdat->__lruvec;
1216 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page)
1218 pg_data_t *pgdat = page_pgdat(page);
1220 return &pgdat->__lruvec;
1223 static inline void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page)
1227 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
1232 static inline bool mm_match_cgroup(struct mm_struct *mm,
1233 struct mem_cgroup *memcg)
1238 static inline struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
1244 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css)
1249 static inline void mem_cgroup_put(struct mem_cgroup *memcg)
1253 static inline struct lruvec *lock_page_lruvec(struct page *page)
1255 struct pglist_data *pgdat = page_pgdat(page);
1257 spin_lock(&pgdat->__lruvec.lru_lock);
1258 return &pgdat->__lruvec;
1261 static inline struct lruvec *lock_page_lruvec_irq(struct page *page)
1263 struct pglist_data *pgdat = page_pgdat(page);
1265 spin_lock_irq(&pgdat->__lruvec.lru_lock);
1266 return &pgdat->__lruvec;
1269 static inline struct lruvec *lock_page_lruvec_irqsave(struct page *page,
1270 unsigned long *flagsp)
1272 struct pglist_data *pgdat = page_pgdat(page);
1274 spin_lock_irqsave(&pgdat->__lruvec.lru_lock, *flagsp);
1275 return &pgdat->__lruvec;
1278 static inline struct mem_cgroup *
1279 mem_cgroup_iter(struct mem_cgroup *root,
1280 struct mem_cgroup *prev,
1281 struct mem_cgroup_reclaim_cookie *reclaim)
1286 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
1287 struct mem_cgroup *prev)
1291 static inline int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
1292 int (*fn)(struct task_struct *, void *), void *arg)
1297 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
1302 static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
1305 /* XXX: This should always return root_mem_cgroup */
1309 static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m)
1314 static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec)
1319 static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
1325 unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
1326 enum lru_list lru, int zone_idx)
1331 static inline unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg)
1336 static inline unsigned long mem_cgroup_size(struct mem_cgroup *memcg)
1342 mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p)
1347 mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg)
1351 static inline void lock_page_memcg(struct page *page)
1355 static inline void unlock_page_memcg(struct page *page)
1359 static inline void mem_cgroup_handle_over_high(void)
1363 static inline void mem_cgroup_enter_user_fault(void)
1367 static inline void mem_cgroup_exit_user_fault(void)
1371 static inline bool task_in_memcg_oom(struct task_struct *p)
1376 static inline bool mem_cgroup_oom_synchronize(bool wait)
1381 static inline struct mem_cgroup *mem_cgroup_get_oom_group(
1382 struct task_struct *victim, struct mem_cgroup *oom_domain)
1387 static inline void mem_cgroup_print_oom_group(struct mem_cgroup *memcg)
1391 static inline void __mod_memcg_state(struct mem_cgroup *memcg,
1397 static inline void mod_memcg_state(struct mem_cgroup *memcg,
1403 static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx)
1408 static inline unsigned long lruvec_page_state(struct lruvec *lruvec,
1409 enum node_stat_item idx)
1411 return node_page_state(lruvec_pgdat(lruvec), idx);
1414 static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec,
1415 enum node_stat_item idx)
1417 return node_page_state(lruvec_pgdat(lruvec), idx);
1420 static inline void mem_cgroup_flush_stats(void)
1424 static inline void __mod_memcg_lruvec_state(struct lruvec *lruvec,
1425 enum node_stat_item idx, int val)
1429 static inline void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx,
1432 struct page *page = virt_to_head_page(p);
1434 __mod_node_page_state(page_pgdat(page), idx, val);
1437 static inline void mod_lruvec_kmem_state(void *p, enum node_stat_item idx,
1440 struct page *page = virt_to_head_page(p);
1442 mod_node_page_state(page_pgdat(page), idx, val);
1445 static inline void count_memcg_events(struct mem_cgroup *memcg,
1446 enum vm_event_item idx,
1447 unsigned long count)
1451 static inline void __count_memcg_events(struct mem_cgroup *memcg,
1452 enum vm_event_item idx,
1453 unsigned long count)
1457 static inline void count_memcg_page_event(struct page *page,
1463 void count_memcg_event_mm(struct mm_struct *mm, enum vm_event_item idx)
1467 static inline void split_page_memcg(struct page *head, unsigned int nr)
1472 unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
1474 unsigned long *total_scanned)
1478 #endif /* CONFIG_MEMCG */
1480 static inline void __inc_lruvec_kmem_state(void *p, enum node_stat_item idx)
1482 __mod_lruvec_kmem_state(p, idx, 1);
1485 static inline void __dec_lruvec_kmem_state(void *p, enum node_stat_item idx)
1487 __mod_lruvec_kmem_state(p, idx, -1);
1490 static inline struct lruvec *parent_lruvec(struct lruvec *lruvec)
1492 struct mem_cgroup *memcg;
1494 memcg = lruvec_memcg(lruvec);
1497 memcg = parent_mem_cgroup(memcg);
1500 return mem_cgroup_lruvec(memcg, lruvec_pgdat(lruvec));
1503 static inline void unlock_page_lruvec(struct lruvec *lruvec)
1505 spin_unlock(&lruvec->lru_lock);
1508 static inline void unlock_page_lruvec_irq(struct lruvec *lruvec)
1510 spin_unlock_irq(&lruvec->lru_lock);
1513 static inline void unlock_page_lruvec_irqrestore(struct lruvec *lruvec,
1514 unsigned long flags)
1516 spin_unlock_irqrestore(&lruvec->lru_lock, flags);
1519 /* Test requires a stable page->memcg binding, see page_memcg() */
1520 static inline bool page_matches_lruvec(struct page *page, struct lruvec *lruvec)
1522 return lruvec_pgdat(lruvec) == page_pgdat(page) &&
1523 lruvec_memcg(lruvec) == page_memcg(page);
1526 /* Don't lock again iff page's lruvec locked */
1527 static inline struct lruvec *relock_page_lruvec_irq(struct page *page,
1528 struct lruvec *locked_lruvec)
1530 if (locked_lruvec) {
1531 if (page_matches_lruvec(page, locked_lruvec))
1532 return locked_lruvec;
1534 unlock_page_lruvec_irq(locked_lruvec);
1537 return lock_page_lruvec_irq(page);
1540 /* Don't lock again iff page's lruvec locked */
1541 static inline struct lruvec *relock_page_lruvec_irqsave(struct page *page,
1542 struct lruvec *locked_lruvec, unsigned long *flags)
1544 if (locked_lruvec) {
1545 if (page_matches_lruvec(page, locked_lruvec))
1546 return locked_lruvec;
1548 unlock_page_lruvec_irqrestore(locked_lruvec, *flags);
1551 return lock_page_lruvec_irqsave(page, flags);
1554 #ifdef CONFIG_CGROUP_WRITEBACK
1556 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
1557 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
1558 unsigned long *pheadroom, unsigned long *pdirty,
1559 unsigned long *pwriteback);
1561 void mem_cgroup_track_foreign_dirty_slowpath(struct page *page,
1562 struct bdi_writeback *wb);
1564 static inline void mem_cgroup_track_foreign_dirty(struct page *page,
1565 struct bdi_writeback *wb)
1567 if (mem_cgroup_disabled())
1570 if (unlikely(&page_memcg(page)->css != wb->memcg_css))
1571 mem_cgroup_track_foreign_dirty_slowpath(page, wb);
1574 void mem_cgroup_flush_foreign(struct bdi_writeback *wb);
1576 #else /* CONFIG_CGROUP_WRITEBACK */
1578 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
1583 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
1584 unsigned long *pfilepages,
1585 unsigned long *pheadroom,
1586 unsigned long *pdirty,
1587 unsigned long *pwriteback)
1591 static inline void mem_cgroup_track_foreign_dirty(struct page *page,
1592 struct bdi_writeback *wb)
1596 static inline void mem_cgroup_flush_foreign(struct bdi_writeback *wb)
1600 #endif /* CONFIG_CGROUP_WRITEBACK */
1603 bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages,
1605 void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
1607 extern struct static_key_false memcg_sockets_enabled_key;
1608 #define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_enabled_key)
1609 void mem_cgroup_sk_alloc(struct sock *sk);
1610 void mem_cgroup_sk_free(struct sock *sk);
1611 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
1613 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_pressure)
1616 if (time_before(jiffies, memcg->socket_pressure))
1618 } while ((memcg = parent_mem_cgroup(memcg)));
1622 int alloc_shrinker_info(struct mem_cgroup *memcg);
1623 void free_shrinker_info(struct mem_cgroup *memcg);
1624 void set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id);
1625 void reparent_shrinker_deferred(struct mem_cgroup *memcg);
1627 #define mem_cgroup_sockets_enabled 0
1628 static inline void mem_cgroup_sk_alloc(struct sock *sk) { };
1629 static inline void mem_cgroup_sk_free(struct sock *sk) { };
1630 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
1635 static inline void set_shrinker_bit(struct mem_cgroup *memcg,
1636 int nid, int shrinker_id)
1641 #ifdef CONFIG_MEMCG_KMEM
1642 bool mem_cgroup_kmem_disabled(void);
1643 int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order);
1644 void __memcg_kmem_uncharge_page(struct page *page, int order);
1646 struct obj_cgroup *get_obj_cgroup_from_current(void);
1648 int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size);
1649 void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size);
1651 extern struct static_key_false memcg_kmem_enabled_key;
1653 extern int memcg_nr_cache_ids;
1654 void memcg_get_cache_ids(void);
1655 void memcg_put_cache_ids(void);
1658 * Helper macro to loop through all memcg-specific caches. Callers must still
1659 * check if the cache is valid (it is either valid or NULL).
1660 * the slab_mutex must be held when looping through those caches
1662 #define for_each_memcg_cache_index(_idx) \
1663 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
1665 static inline bool memcg_kmem_enabled(void)
1667 return static_branch_likely(&memcg_kmem_enabled_key);
1670 static inline int memcg_kmem_charge_page(struct page *page, gfp_t gfp,
1673 if (memcg_kmem_enabled())
1674 return __memcg_kmem_charge_page(page, gfp, order);
1678 static inline void memcg_kmem_uncharge_page(struct page *page, int order)
1680 if (memcg_kmem_enabled())
1681 __memcg_kmem_uncharge_page(page, order);
1685 * A helper for accessing memcg's kmem_id, used for getting
1686 * corresponding LRU lists.
1688 static inline int memcg_cache_id(struct mem_cgroup *memcg)
1690 return memcg ? memcg->kmemcg_id : -1;
1693 struct mem_cgroup *mem_cgroup_from_obj(void *p);
1696 static inline bool mem_cgroup_kmem_disabled(void)
1701 static inline int memcg_kmem_charge_page(struct page *page, gfp_t gfp,
1707 static inline void memcg_kmem_uncharge_page(struct page *page, int order)
1711 static inline int __memcg_kmem_charge_page(struct page *page, gfp_t gfp,
1717 static inline void __memcg_kmem_uncharge_page(struct page *page, int order)
1721 #define for_each_memcg_cache_index(_idx) \
1724 static inline bool memcg_kmem_enabled(void)
1729 static inline int memcg_cache_id(struct mem_cgroup *memcg)
1734 static inline void memcg_get_cache_ids(void)
1738 static inline void memcg_put_cache_ids(void)
1742 static inline struct mem_cgroup *mem_cgroup_from_obj(void *p)
1747 #endif /* CONFIG_MEMCG_KMEM */
1749 #endif /* _LINUX_MEMCONTROL_H */