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,
40 enum memcg_memory_event {
50 MEMCG_NR_MEMORY_EVENTS,
53 struct mem_cgroup_reclaim_cookie {
55 unsigned int generation;
60 #define MEM_CGROUP_ID_SHIFT 16
61 #define MEM_CGROUP_ID_MAX USHRT_MAX
63 struct mem_cgroup_id {
69 * Per memcg event counter is incremented at every pagein/pageout. With THP,
70 * it will be incremented by the number of pages. This counter is used
71 * to trigger some periodic events. This is straightforward and better
72 * than using jiffies etc. to handle periodic memcg event.
74 enum mem_cgroup_events_target {
75 MEM_CGROUP_TARGET_THRESH,
76 MEM_CGROUP_TARGET_SOFTLIMIT,
80 struct memcg_vmstats_percpu {
81 /* Local (CPU and cgroup) page state & events */
82 long state[MEMCG_NR_STAT];
83 unsigned long events[NR_VM_EVENT_ITEMS];
85 /* Delta calculation for lockless upward propagation */
86 long state_prev[MEMCG_NR_STAT];
87 unsigned long events_prev[NR_VM_EVENT_ITEMS];
89 /* Cgroup1: threshold notifications & softlimit tree updates */
90 unsigned long nr_page_events;
91 unsigned long targets[MEM_CGROUP_NTARGETS];
94 struct memcg_vmstats {
95 /* Aggregated (CPU and subtree) page state & events */
96 long state[MEMCG_NR_STAT];
97 unsigned long events[NR_VM_EVENT_ITEMS];
99 /* Pending child counts during tree propagation */
100 long state_pending[MEMCG_NR_STAT];
101 unsigned long events_pending[NR_VM_EVENT_ITEMS];
104 struct mem_cgroup_reclaim_iter {
105 struct mem_cgroup *position;
106 /* scan generation, increased every round-trip */
107 unsigned int generation;
111 * Bitmap and deferred work of shrinker::id corresponding to memcg-aware
112 * shrinkers, which have elements charged to this memcg.
114 struct shrinker_info {
116 atomic_long_t *nr_deferred;
120 struct lruvec_stats_percpu {
121 /* Local (CPU and cgroup) state */
122 long state[NR_VM_NODE_STAT_ITEMS];
124 /* Delta calculation for lockless upward propagation */
125 long state_prev[NR_VM_NODE_STAT_ITEMS];
128 struct lruvec_stats {
129 /* Aggregated (CPU and subtree) state */
130 long state[NR_VM_NODE_STAT_ITEMS];
132 /* Pending child counts during tree propagation */
133 long state_pending[NR_VM_NODE_STAT_ITEMS];
137 * per-node information in memory controller.
139 struct mem_cgroup_per_node {
140 struct lruvec lruvec;
142 struct lruvec_stats_percpu __percpu *lruvec_stats_percpu;
143 struct lruvec_stats lruvec_stats;
145 unsigned long lru_zone_size[MAX_NR_ZONES][NR_LRU_LISTS];
147 struct mem_cgroup_reclaim_iter iter;
149 struct shrinker_info __rcu *shrinker_info;
151 struct rb_node tree_node; /* RB tree node */
152 unsigned long usage_in_excess;/* Set to the value by which */
153 /* the soft limit is exceeded*/
155 struct mem_cgroup *memcg; /* Back pointer, we cannot */
156 /* use container_of */
159 struct mem_cgroup_threshold {
160 struct eventfd_ctx *eventfd;
161 unsigned long threshold;
165 struct mem_cgroup_threshold_ary {
166 /* An array index points to threshold just below or equal to usage. */
167 int current_threshold;
168 /* Size of entries[] */
170 /* Array of thresholds */
171 struct mem_cgroup_threshold entries[];
174 struct mem_cgroup_thresholds {
175 /* Primary thresholds array */
176 struct mem_cgroup_threshold_ary *primary;
178 * Spare threshold array.
179 * This is needed to make mem_cgroup_unregister_event() "never fail".
180 * It must be able to store at least primary->size - 1 entries.
182 struct mem_cgroup_threshold_ary *spare;
185 #if defined(CONFIG_SMP)
186 struct memcg_padding {
188 } ____cacheline_internodealigned_in_smp;
189 #define MEMCG_PADDING(name) struct memcg_padding name
191 #define MEMCG_PADDING(name)
195 * Remember four most recent foreign writebacks with dirty pages in this
196 * cgroup. Inode sharing is expected to be uncommon and, even if we miss
197 * one in a given round, we're likely to catch it later if it keeps
198 * foreign-dirtying, so a fairly low count should be enough.
200 * See mem_cgroup_track_foreign_dirty_slowpath() for details.
202 #define MEMCG_CGWB_FRN_CNT 4
204 struct memcg_cgwb_frn {
205 u64 bdi_id; /* bdi->id of the foreign inode */
206 int memcg_id; /* memcg->css.id of foreign inode */
207 u64 at; /* jiffies_64 at the time of dirtying */
208 struct wb_completion done; /* tracks in-flight foreign writebacks */
212 * Bucket for arbitrarily byte-sized objects charged to a memory
213 * cgroup. The bucket can be reparented in one piece when the cgroup
214 * is destroyed, without having to round up the individual references
215 * of all live memory objects in the wild.
218 struct percpu_ref refcnt;
219 struct mem_cgroup *memcg;
220 atomic_t nr_charged_bytes;
222 struct list_head list; /* protected by objcg_lock */
228 * The memory controller data structure. The memory controller controls both
229 * page cache and RSS per cgroup. We would eventually like to provide
230 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
231 * to help the administrator determine what knobs to tune.
234 struct cgroup_subsys_state css;
236 /* Private memcg ID. Used to ID objects that outlive the cgroup */
237 struct mem_cgroup_id id;
239 /* Accounted resources */
240 struct page_counter memory; /* Both v1 & v2 */
243 struct page_counter swap; /* v2 only */
244 struct page_counter memsw; /* v1 only */
247 /* Legacy consumer-oriented counters */
248 struct page_counter kmem; /* v1 only */
249 struct page_counter tcpmem; /* v1 only */
251 /* Range enforcement for interrupt charges */
252 struct work_struct high_work;
254 unsigned long soft_limit;
256 /* vmpressure notifications */
257 struct vmpressure vmpressure;
260 * Should the OOM killer kill all belonging tasks, had it kill one?
264 /* protected by memcg_oom_lock */
269 /* OOM-Killer disable */
270 int oom_kill_disable;
272 /* memory.events and memory.events.local */
273 struct cgroup_file events_file;
274 struct cgroup_file events_local_file;
276 /* handle for "memory.swap.events" */
277 struct cgroup_file swap_events_file;
279 /* protect arrays of thresholds */
280 struct mutex thresholds_lock;
282 /* thresholds for memory usage. RCU-protected */
283 struct mem_cgroup_thresholds thresholds;
285 /* thresholds for mem+swap usage. RCU-protected */
286 struct mem_cgroup_thresholds memsw_thresholds;
288 /* For oom notifier event fd */
289 struct list_head oom_notify;
292 * Should we move charges of a task when a task is moved into this
293 * mem_cgroup ? And what type of charges should we move ?
295 unsigned long move_charge_at_immigrate;
296 /* taken only while moving_account > 0 */
297 spinlock_t move_lock;
298 unsigned long move_lock_flags;
300 MEMCG_PADDING(_pad1_);
303 struct memcg_vmstats vmstats;
306 atomic_long_t memory_events[MEMCG_NR_MEMORY_EVENTS];
307 atomic_long_t memory_events_local[MEMCG_NR_MEMORY_EVENTS];
309 unsigned long socket_pressure;
311 /* Legacy tcp memory accounting */
315 #ifdef CONFIG_MEMCG_KMEM
317 struct obj_cgroup __rcu *objcg;
318 /* list of inherited objcgs, protected by objcg_lock */
319 struct list_head objcg_list;
322 MEMCG_PADDING(_pad2_);
325 * set > 0 if pages under this cgroup are moving to other cgroup.
327 atomic_t moving_account;
328 struct task_struct *move_lock_task;
330 struct memcg_vmstats_percpu __percpu *vmstats_percpu;
332 #ifdef CONFIG_CGROUP_WRITEBACK
333 struct list_head cgwb_list;
334 struct wb_domain cgwb_domain;
335 struct memcg_cgwb_frn cgwb_frn[MEMCG_CGWB_FRN_CNT];
338 /* List of events which userspace want to receive */
339 struct list_head event_list;
340 spinlock_t event_list_lock;
342 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
343 struct deferred_split deferred_split_queue;
346 struct mem_cgroup_per_node *nodeinfo[];
350 * size of first charge trial. "32" comes from vmscan.c's magic value.
351 * TODO: maybe necessary to use big numbers in big irons.
353 #define MEMCG_CHARGE_BATCH 32U
355 extern struct mem_cgroup *root_mem_cgroup;
357 enum page_memcg_data_flags {
358 /* page->memcg_data is a pointer to an objcgs vector */
359 MEMCG_DATA_OBJCGS = (1UL << 0),
360 /* page has been accounted as a non-slab kernel page */
361 MEMCG_DATA_KMEM = (1UL << 1),
362 /* the next bit after the last actual flag */
363 __NR_MEMCG_DATA_FLAGS = (1UL << 2),
366 #define MEMCG_DATA_FLAGS_MASK (__NR_MEMCG_DATA_FLAGS - 1)
368 static inline bool folio_memcg_kmem(struct folio *folio);
371 * After the initialization objcg->memcg is always pointing at
372 * a valid memcg, but can be atomically swapped to the parent memcg.
374 * The caller must ensure that the returned memcg won't be released:
375 * e.g. acquire the rcu_read_lock or css_set_lock.
377 static inline struct mem_cgroup *obj_cgroup_memcg(struct obj_cgroup *objcg)
379 return READ_ONCE(objcg->memcg);
383 * __folio_memcg - Get the memory cgroup associated with a non-kmem folio
384 * @folio: Pointer to the folio.
386 * Returns a pointer to the memory cgroup associated with the folio,
387 * or NULL. This function assumes that the folio is known to have a
388 * proper memory cgroup pointer. It's not safe to call this function
389 * against some type of folios, e.g. slab folios or ex-slab folios or
392 static inline struct mem_cgroup *__folio_memcg(struct folio *folio)
394 unsigned long memcg_data = folio->memcg_data;
396 VM_BUG_ON_FOLIO(folio_test_slab(folio), folio);
397 VM_BUG_ON_FOLIO(memcg_data & MEMCG_DATA_OBJCGS, folio);
398 VM_BUG_ON_FOLIO(memcg_data & MEMCG_DATA_KMEM, folio);
400 return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
404 * __folio_objcg - get the object cgroup associated with a kmem folio.
405 * @folio: Pointer to the folio.
407 * Returns a pointer to the object cgroup associated with the folio,
408 * or NULL. This function assumes that the folio is known to have a
409 * proper object cgroup pointer. It's not safe to call this function
410 * against some type of folios, e.g. slab folios or ex-slab folios or
413 static inline struct obj_cgroup *__folio_objcg(struct folio *folio)
415 unsigned long memcg_data = folio->memcg_data;
417 VM_BUG_ON_FOLIO(folio_test_slab(folio), folio);
418 VM_BUG_ON_FOLIO(memcg_data & MEMCG_DATA_OBJCGS, folio);
419 VM_BUG_ON_FOLIO(!(memcg_data & MEMCG_DATA_KMEM), folio);
421 return (struct obj_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
425 * folio_memcg - Get the memory cgroup associated with a folio.
426 * @folio: Pointer to the folio.
428 * Returns a pointer to the memory cgroup associated with the folio,
429 * or NULL. This function assumes that the folio is known to have a
430 * proper memory cgroup pointer. It's not safe to call this function
431 * against some type of folios, e.g. slab folios or ex-slab folios.
433 * For a non-kmem folio any of the following ensures folio and memcg binding
438 * - lock_page_memcg()
439 * - exclusive reference
441 * For a kmem folio a caller should hold an rcu read lock to protect memcg
442 * associated with a kmem folio from being released.
444 static inline struct mem_cgroup *folio_memcg(struct folio *folio)
446 if (folio_memcg_kmem(folio))
447 return obj_cgroup_memcg(__folio_objcg(folio));
448 return __folio_memcg(folio);
451 static inline struct mem_cgroup *page_memcg(struct page *page)
453 return folio_memcg(page_folio(page));
457 * folio_memcg_rcu - Locklessly get the memory cgroup associated with a folio.
458 * @folio: Pointer to the folio.
460 * This function assumes that the folio is known to have a
461 * proper memory cgroup pointer. It's not safe to call this function
462 * against some type of folios, e.g. slab folios or ex-slab folios.
464 * Return: A pointer to the memory cgroup associated with the folio,
467 static inline struct mem_cgroup *folio_memcg_rcu(struct folio *folio)
469 unsigned long memcg_data = READ_ONCE(folio->memcg_data);
471 VM_BUG_ON_FOLIO(folio_test_slab(folio), folio);
472 WARN_ON_ONCE(!rcu_read_lock_held());
474 if (memcg_data & MEMCG_DATA_KMEM) {
475 struct obj_cgroup *objcg;
477 objcg = (void *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
478 return obj_cgroup_memcg(objcg);
481 return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
485 * page_memcg_check - get the memory cgroup associated with a page
486 * @page: a pointer to the page struct
488 * Returns a pointer to the memory cgroup associated with the page,
489 * or NULL. This function unlike page_memcg() can take any page
490 * as an argument. It has to be used in cases when it's not known if a page
491 * has an associated memory cgroup pointer or an object cgroups vector or
494 * For a non-kmem page any of the following ensures page and memcg binding
499 * - lock_page_memcg()
500 * - exclusive reference
502 * For a kmem page a caller should hold an rcu read lock to protect memcg
503 * associated with a kmem page from being released.
505 static inline struct mem_cgroup *page_memcg_check(struct page *page)
508 * Because page->memcg_data might be changed asynchronously
509 * for slab pages, READ_ONCE() should be used here.
511 unsigned long memcg_data = READ_ONCE(page->memcg_data);
513 if (memcg_data & MEMCG_DATA_OBJCGS)
516 if (memcg_data & MEMCG_DATA_KMEM) {
517 struct obj_cgroup *objcg;
519 objcg = (void *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
520 return obj_cgroup_memcg(objcg);
523 return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
526 #ifdef CONFIG_MEMCG_KMEM
528 * folio_memcg_kmem - Check if the folio has the memcg_kmem flag set.
529 * @folio: Pointer to the folio.
531 * Checks if the folio has MemcgKmem flag set. The caller must ensure
532 * that the folio has an associated memory cgroup. It's not safe to call
533 * this function against some types of folios, e.g. slab folios.
535 static inline bool folio_memcg_kmem(struct folio *folio)
537 VM_BUG_ON_PGFLAGS(PageTail(&folio->page), &folio->page);
538 VM_BUG_ON_FOLIO(folio->memcg_data & MEMCG_DATA_OBJCGS, folio);
539 return folio->memcg_data & MEMCG_DATA_KMEM;
544 static inline bool folio_memcg_kmem(struct folio *folio)
551 static inline bool PageMemcgKmem(struct page *page)
553 return folio_memcg_kmem(page_folio(page));
556 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
558 return (memcg == root_mem_cgroup);
561 static inline bool mem_cgroup_disabled(void)
563 return !cgroup_subsys_enabled(memory_cgrp_subsys);
566 static inline void mem_cgroup_protection(struct mem_cgroup *root,
567 struct mem_cgroup *memcg,
573 if (mem_cgroup_disabled())
577 * There is no reclaim protection applied to a targeted reclaim.
578 * We are special casing this specific case here because
579 * mem_cgroup_protected calculation is not robust enough to keep
580 * the protection invariant for calculated effective values for
581 * parallel reclaimers with different reclaim target. This is
582 * especially a problem for tail memcgs (as they have pages on LRU)
583 * which would want to have effective values 0 for targeted reclaim
584 * but a different value for external reclaim.
587 * Let's have global and A's reclaim in parallel:
589 * A (low=2G, usage = 3G, max = 3G, children_low_usage = 1.5G)
591 * | C (low = 1G, usage = 2.5G)
592 * B (low = 1G, usage = 0.5G)
594 * For the global reclaim
596 * B.elow = min(B.usage, B.low) because children_low_usage <= A.elow
597 * C.elow = min(C.usage, C.low)
599 * With the effective values resetting we have A reclaim
604 * If the global reclaim races with A's reclaim then
605 * B.elow = C.elow = 0 because children_low_usage > A.elow)
606 * is possible and reclaiming B would be violating the protection.
612 *min = READ_ONCE(memcg->memory.emin);
613 *low = READ_ONCE(memcg->memory.elow);
616 void mem_cgroup_calculate_protection(struct mem_cgroup *root,
617 struct mem_cgroup *memcg);
619 static inline bool mem_cgroup_supports_protection(struct mem_cgroup *memcg)
622 * The root memcg doesn't account charges, and doesn't support
625 return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg);
629 static inline bool mem_cgroup_below_low(struct mem_cgroup *memcg)
631 if (!mem_cgroup_supports_protection(memcg))
634 return READ_ONCE(memcg->memory.elow) >=
635 page_counter_read(&memcg->memory);
638 static inline bool mem_cgroup_below_min(struct mem_cgroup *memcg)
640 if (!mem_cgroup_supports_protection(memcg))
643 return READ_ONCE(memcg->memory.emin) >=
644 page_counter_read(&memcg->memory);
647 int __mem_cgroup_charge(struct folio *folio, struct mm_struct *mm, gfp_t gfp);
650 * mem_cgroup_charge - Charge a newly allocated folio to a cgroup.
651 * @folio: Folio to charge.
652 * @mm: mm context of the allocating task.
653 * @gfp: Reclaim mode.
655 * Try to charge @folio to the memcg that @mm belongs to, reclaiming
656 * pages according to @gfp if necessary. If @mm is NULL, try to
657 * charge to the active memcg.
659 * Do not use this for folios allocated for swapin.
661 * Return: 0 on success. Otherwise, an error code is returned.
663 static inline int mem_cgroup_charge(struct folio *folio, struct mm_struct *mm,
666 if (mem_cgroup_disabled())
668 return __mem_cgroup_charge(folio, mm, gfp);
671 int mem_cgroup_swapin_charge_page(struct page *page, struct mm_struct *mm,
672 gfp_t gfp, swp_entry_t entry);
673 void mem_cgroup_swapin_uncharge_swap(swp_entry_t entry);
675 void __mem_cgroup_uncharge(struct folio *folio);
678 * mem_cgroup_uncharge - Uncharge a folio.
679 * @folio: Folio to uncharge.
681 * Uncharge a folio previously charged with mem_cgroup_charge().
683 static inline void mem_cgroup_uncharge(struct folio *folio)
685 if (mem_cgroup_disabled())
687 __mem_cgroup_uncharge(folio);
690 void __mem_cgroup_uncharge_list(struct list_head *page_list);
691 static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
693 if (mem_cgroup_disabled())
695 __mem_cgroup_uncharge_list(page_list);
698 void mem_cgroup_migrate(struct folio *old, struct folio *new);
701 * mem_cgroup_lruvec - get the lru list vector for a memcg & node
702 * @memcg: memcg of the wanted lruvec
703 * @pgdat: pglist_data
705 * Returns the lru list vector holding pages for a given @memcg &
706 * @pgdat combination. This can be the node lruvec, if the memory
707 * controller is disabled.
709 static inline struct lruvec *mem_cgroup_lruvec(struct mem_cgroup *memcg,
710 struct pglist_data *pgdat)
712 struct mem_cgroup_per_node *mz;
713 struct lruvec *lruvec;
715 if (mem_cgroup_disabled()) {
716 lruvec = &pgdat->__lruvec;
721 memcg = root_mem_cgroup;
723 mz = memcg->nodeinfo[pgdat->node_id];
724 lruvec = &mz->lruvec;
727 * Since a node can be onlined after the mem_cgroup was created,
728 * we have to be prepared to initialize lruvec->pgdat here;
729 * and if offlined then reonlined, we need to reinitialize it.
731 if (unlikely(lruvec->pgdat != pgdat))
732 lruvec->pgdat = pgdat;
737 * folio_lruvec - return lruvec for isolating/putting an LRU folio
738 * @folio: Pointer to the folio.
740 * This function relies on folio->mem_cgroup being stable.
742 static inline struct lruvec *folio_lruvec(struct folio *folio)
744 struct mem_cgroup *memcg = folio_memcg(folio);
746 VM_WARN_ON_ONCE_FOLIO(!memcg && !mem_cgroup_disabled(), folio);
747 return mem_cgroup_lruvec(memcg, folio_pgdat(folio));
750 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
752 struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm);
754 struct lruvec *folio_lruvec_lock(struct folio *folio);
755 struct lruvec *folio_lruvec_lock_irq(struct folio *folio);
756 struct lruvec *folio_lruvec_lock_irqsave(struct folio *folio,
757 unsigned long *flags);
759 #ifdef CONFIG_DEBUG_VM
760 void lruvec_memcg_debug(struct lruvec *lruvec, struct folio *folio);
763 void lruvec_memcg_debug(struct lruvec *lruvec, struct folio *folio)
769 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
770 return css ? container_of(css, struct mem_cgroup, css) : NULL;
773 static inline bool obj_cgroup_tryget(struct obj_cgroup *objcg)
775 return percpu_ref_tryget(&objcg->refcnt);
778 static inline void obj_cgroup_get(struct obj_cgroup *objcg)
780 percpu_ref_get(&objcg->refcnt);
783 static inline void obj_cgroup_get_many(struct obj_cgroup *objcg,
786 percpu_ref_get_many(&objcg->refcnt, nr);
789 static inline void obj_cgroup_put(struct obj_cgroup *objcg)
791 percpu_ref_put(&objcg->refcnt);
794 static inline void mem_cgroup_put(struct mem_cgroup *memcg)
797 css_put(&memcg->css);
800 #define mem_cgroup_from_counter(counter, member) \
801 container_of(counter, struct mem_cgroup, member)
803 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
805 struct mem_cgroup_reclaim_cookie *);
806 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
807 int mem_cgroup_scan_tasks(struct mem_cgroup *,
808 int (*)(struct task_struct *, void *), void *);
810 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
812 if (mem_cgroup_disabled())
817 struct mem_cgroup *mem_cgroup_from_id(unsigned short id);
819 static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m)
821 return mem_cgroup_from_css(seq_css(m));
824 static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec)
826 struct mem_cgroup_per_node *mz;
828 if (mem_cgroup_disabled())
831 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
836 * parent_mem_cgroup - find the accounting parent of a memcg
837 * @memcg: memcg whose parent to find
839 * Returns the parent memcg, or NULL if this is the root or the memory
840 * controller is in legacy no-hierarchy mode.
842 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
844 if (!memcg->memory.parent)
846 return mem_cgroup_from_counter(memcg->memory.parent, memory);
849 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
850 struct mem_cgroup *root)
854 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
857 static inline bool mm_match_cgroup(struct mm_struct *mm,
858 struct mem_cgroup *memcg)
860 struct mem_cgroup *task_memcg;
864 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
866 match = mem_cgroup_is_descendant(task_memcg, memcg);
871 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
872 ino_t page_cgroup_ino(struct page *page);
874 static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
876 if (mem_cgroup_disabled())
878 return !!(memcg->css.flags & CSS_ONLINE);
881 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
882 int zid, int nr_pages);
885 unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
886 enum lru_list lru, int zone_idx)
888 struct mem_cgroup_per_node *mz;
890 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
891 return READ_ONCE(mz->lru_zone_size[zone_idx][lru]);
894 void mem_cgroup_handle_over_high(void);
896 unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg);
898 unsigned long mem_cgroup_size(struct mem_cgroup *memcg);
900 void mem_cgroup_print_oom_context(struct mem_cgroup *memcg,
901 struct task_struct *p);
903 void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg);
905 static inline void mem_cgroup_enter_user_fault(void)
907 WARN_ON(current->in_user_fault);
908 current->in_user_fault = 1;
911 static inline void mem_cgroup_exit_user_fault(void)
913 WARN_ON(!current->in_user_fault);
914 current->in_user_fault = 0;
917 static inline bool task_in_memcg_oom(struct task_struct *p)
919 return p->memcg_in_oom;
922 bool mem_cgroup_oom_synchronize(bool wait);
923 struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim,
924 struct mem_cgroup *oom_domain);
925 void mem_cgroup_print_oom_group(struct mem_cgroup *memcg);
927 #ifdef CONFIG_MEMCG_SWAP
928 extern bool cgroup_memory_noswap;
931 void folio_memcg_lock(struct folio *folio);
932 void folio_memcg_unlock(struct folio *folio);
933 void lock_page_memcg(struct page *page);
934 void unlock_page_memcg(struct page *page);
936 void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val);
938 /* idx can be of type enum memcg_stat_item or node_stat_item */
939 static inline void mod_memcg_state(struct mem_cgroup *memcg,
944 local_irq_save(flags);
945 __mod_memcg_state(memcg, idx, val);
946 local_irq_restore(flags);
949 static inline void mod_memcg_page_state(struct page *page,
952 struct mem_cgroup *memcg;
954 if (mem_cgroup_disabled())
958 memcg = page_memcg(page);
960 mod_memcg_state(memcg, idx, val);
964 static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx)
966 return READ_ONCE(memcg->vmstats.state[idx]);
969 static inline unsigned long lruvec_page_state(struct lruvec *lruvec,
970 enum node_stat_item idx)
972 struct mem_cgroup_per_node *pn;
974 if (mem_cgroup_disabled())
975 return node_page_state(lruvec_pgdat(lruvec), idx);
977 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
978 return READ_ONCE(pn->lruvec_stats.state[idx]);
981 static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec,
982 enum node_stat_item idx)
984 struct mem_cgroup_per_node *pn;
988 if (mem_cgroup_disabled())
989 return node_page_state(lruvec_pgdat(lruvec), idx);
991 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
992 for_each_possible_cpu(cpu)
993 x += per_cpu(pn->lruvec_stats_percpu->state[idx], cpu);
1001 void mem_cgroup_flush_stats(void);
1003 void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx,
1005 void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, int val);
1007 static inline void mod_lruvec_kmem_state(void *p, enum node_stat_item idx,
1010 unsigned long flags;
1012 local_irq_save(flags);
1013 __mod_lruvec_kmem_state(p, idx, val);
1014 local_irq_restore(flags);
1017 static inline void mod_memcg_lruvec_state(struct lruvec *lruvec,
1018 enum node_stat_item idx, int val)
1020 unsigned long flags;
1022 local_irq_save(flags);
1023 __mod_memcg_lruvec_state(lruvec, idx, val);
1024 local_irq_restore(flags);
1027 void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx,
1028 unsigned long count);
1030 static inline void count_memcg_events(struct mem_cgroup *memcg,
1031 enum vm_event_item idx,
1032 unsigned long count)
1034 unsigned long flags;
1036 local_irq_save(flags);
1037 __count_memcg_events(memcg, idx, count);
1038 local_irq_restore(flags);
1041 static inline void count_memcg_page_event(struct page *page,
1042 enum vm_event_item idx)
1044 struct mem_cgroup *memcg = page_memcg(page);
1047 count_memcg_events(memcg, idx, 1);
1050 static inline void count_memcg_event_mm(struct mm_struct *mm,
1051 enum vm_event_item idx)
1053 struct mem_cgroup *memcg;
1055 if (mem_cgroup_disabled())
1059 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1061 count_memcg_events(memcg, idx, 1);
1065 static inline void memcg_memory_event(struct mem_cgroup *memcg,
1066 enum memcg_memory_event event)
1068 bool swap_event = event == MEMCG_SWAP_HIGH || event == MEMCG_SWAP_MAX ||
1069 event == MEMCG_SWAP_FAIL;
1071 atomic_long_inc(&memcg->memory_events_local[event]);
1073 cgroup_file_notify(&memcg->events_local_file);
1076 atomic_long_inc(&memcg->memory_events[event]);
1078 cgroup_file_notify(&memcg->swap_events_file);
1080 cgroup_file_notify(&memcg->events_file);
1082 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
1084 if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
1086 } while ((memcg = parent_mem_cgroup(memcg)) &&
1087 !mem_cgroup_is_root(memcg));
1090 static inline void memcg_memory_event_mm(struct mm_struct *mm,
1091 enum memcg_memory_event event)
1093 struct mem_cgroup *memcg;
1095 if (mem_cgroup_disabled())
1099 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1101 memcg_memory_event(memcg, event);
1105 void split_page_memcg(struct page *head, unsigned int nr);
1107 unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
1109 unsigned long *total_scanned);
1111 #else /* CONFIG_MEMCG */
1113 #define MEM_CGROUP_ID_SHIFT 0
1114 #define MEM_CGROUP_ID_MAX 0
1116 static inline struct mem_cgroup *folio_memcg(struct folio *folio)
1121 static inline struct mem_cgroup *page_memcg(struct page *page)
1126 static inline struct mem_cgroup *folio_memcg_rcu(struct folio *folio)
1128 WARN_ON_ONCE(!rcu_read_lock_held());
1132 static inline struct mem_cgroup *page_memcg_check(struct page *page)
1137 static inline bool folio_memcg_kmem(struct folio *folio)
1142 static inline bool PageMemcgKmem(struct page *page)
1147 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
1152 static inline bool mem_cgroup_disabled(void)
1157 static inline void memcg_memory_event(struct mem_cgroup *memcg,
1158 enum memcg_memory_event event)
1162 static inline void memcg_memory_event_mm(struct mm_struct *mm,
1163 enum memcg_memory_event event)
1167 static inline void mem_cgroup_protection(struct mem_cgroup *root,
1168 struct mem_cgroup *memcg,
1175 static inline void mem_cgroup_calculate_protection(struct mem_cgroup *root,
1176 struct mem_cgroup *memcg)
1180 static inline bool mem_cgroup_below_low(struct mem_cgroup *memcg)
1185 static inline bool mem_cgroup_below_min(struct mem_cgroup *memcg)
1190 static inline int mem_cgroup_charge(struct folio *folio,
1191 struct mm_struct *mm, gfp_t gfp)
1196 static inline int mem_cgroup_swapin_charge_page(struct page *page,
1197 struct mm_struct *mm, gfp_t gfp, swp_entry_t entry)
1202 static inline void mem_cgroup_swapin_uncharge_swap(swp_entry_t entry)
1206 static inline void mem_cgroup_uncharge(struct folio *folio)
1210 static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
1214 static inline void mem_cgroup_migrate(struct folio *old, struct folio *new)
1218 static inline struct lruvec *mem_cgroup_lruvec(struct mem_cgroup *memcg,
1219 struct pglist_data *pgdat)
1221 return &pgdat->__lruvec;
1224 static inline struct lruvec *folio_lruvec(struct folio *folio)
1226 struct pglist_data *pgdat = folio_pgdat(folio);
1227 return &pgdat->__lruvec;
1231 void lruvec_memcg_debug(struct lruvec *lruvec, struct folio *folio)
1235 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
1240 static inline bool mm_match_cgroup(struct mm_struct *mm,
1241 struct mem_cgroup *memcg)
1246 static inline struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
1252 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css)
1257 static inline void mem_cgroup_put(struct mem_cgroup *memcg)
1261 static inline struct lruvec *folio_lruvec_lock(struct folio *folio)
1263 struct pglist_data *pgdat = folio_pgdat(folio);
1265 spin_lock(&pgdat->__lruvec.lru_lock);
1266 return &pgdat->__lruvec;
1269 static inline struct lruvec *folio_lruvec_lock_irq(struct folio *folio)
1271 struct pglist_data *pgdat = folio_pgdat(folio);
1273 spin_lock_irq(&pgdat->__lruvec.lru_lock);
1274 return &pgdat->__lruvec;
1277 static inline struct lruvec *folio_lruvec_lock_irqsave(struct folio *folio,
1278 unsigned long *flagsp)
1280 struct pglist_data *pgdat = folio_pgdat(folio);
1282 spin_lock_irqsave(&pgdat->__lruvec.lru_lock, *flagsp);
1283 return &pgdat->__lruvec;
1286 static inline struct mem_cgroup *
1287 mem_cgroup_iter(struct mem_cgroup *root,
1288 struct mem_cgroup *prev,
1289 struct mem_cgroup_reclaim_cookie *reclaim)
1294 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
1295 struct mem_cgroup *prev)
1299 static inline int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
1300 int (*fn)(struct task_struct *, void *), void *arg)
1305 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
1310 static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
1313 /* XXX: This should always return root_mem_cgroup */
1317 static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m)
1322 static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec)
1327 static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
1333 unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
1334 enum lru_list lru, int zone_idx)
1339 static inline unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg)
1344 static inline unsigned long mem_cgroup_size(struct mem_cgroup *memcg)
1350 mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p)
1355 mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg)
1359 static inline void lock_page_memcg(struct page *page)
1363 static inline void unlock_page_memcg(struct page *page)
1367 static inline void folio_memcg_lock(struct folio *folio)
1371 static inline void folio_memcg_unlock(struct folio *folio)
1375 static inline void mem_cgroup_handle_over_high(void)
1379 static inline void mem_cgroup_enter_user_fault(void)
1383 static inline void mem_cgroup_exit_user_fault(void)
1387 static inline bool task_in_memcg_oom(struct task_struct *p)
1392 static inline bool mem_cgroup_oom_synchronize(bool wait)
1397 static inline struct mem_cgroup *mem_cgroup_get_oom_group(
1398 struct task_struct *victim, struct mem_cgroup *oom_domain)
1403 static inline void mem_cgroup_print_oom_group(struct mem_cgroup *memcg)
1407 static inline void __mod_memcg_state(struct mem_cgroup *memcg,
1413 static inline void mod_memcg_state(struct mem_cgroup *memcg,
1419 static inline void mod_memcg_page_state(struct page *page,
1424 static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx)
1429 static inline unsigned long lruvec_page_state(struct lruvec *lruvec,
1430 enum node_stat_item idx)
1432 return node_page_state(lruvec_pgdat(lruvec), idx);
1435 static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec,
1436 enum node_stat_item idx)
1438 return node_page_state(lruvec_pgdat(lruvec), idx);
1441 static inline void mem_cgroup_flush_stats(void)
1445 static inline void __mod_memcg_lruvec_state(struct lruvec *lruvec,
1446 enum node_stat_item idx, int val)
1450 static inline void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx,
1453 struct page *page = virt_to_head_page(p);
1455 __mod_node_page_state(page_pgdat(page), idx, val);
1458 static inline void mod_lruvec_kmem_state(void *p, enum node_stat_item idx,
1461 struct page *page = virt_to_head_page(p);
1463 mod_node_page_state(page_pgdat(page), idx, val);
1466 static inline void count_memcg_events(struct mem_cgroup *memcg,
1467 enum vm_event_item idx,
1468 unsigned long count)
1472 static inline void __count_memcg_events(struct mem_cgroup *memcg,
1473 enum vm_event_item idx,
1474 unsigned long count)
1478 static inline void count_memcg_page_event(struct page *page,
1484 void count_memcg_event_mm(struct mm_struct *mm, enum vm_event_item idx)
1488 static inline void split_page_memcg(struct page *head, unsigned int nr)
1493 unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
1495 unsigned long *total_scanned)
1499 #endif /* CONFIG_MEMCG */
1501 static inline void __inc_lruvec_kmem_state(void *p, enum node_stat_item idx)
1503 __mod_lruvec_kmem_state(p, idx, 1);
1506 static inline void __dec_lruvec_kmem_state(void *p, enum node_stat_item idx)
1508 __mod_lruvec_kmem_state(p, idx, -1);
1511 static inline struct lruvec *parent_lruvec(struct lruvec *lruvec)
1513 struct mem_cgroup *memcg;
1515 memcg = lruvec_memcg(lruvec);
1518 memcg = parent_mem_cgroup(memcg);
1521 return mem_cgroup_lruvec(memcg, lruvec_pgdat(lruvec));
1524 static inline void unlock_page_lruvec(struct lruvec *lruvec)
1526 spin_unlock(&lruvec->lru_lock);
1529 static inline void unlock_page_lruvec_irq(struct lruvec *lruvec)
1531 spin_unlock_irq(&lruvec->lru_lock);
1534 static inline void unlock_page_lruvec_irqrestore(struct lruvec *lruvec,
1535 unsigned long flags)
1537 spin_unlock_irqrestore(&lruvec->lru_lock, flags);
1540 /* Test requires a stable page->memcg binding, see page_memcg() */
1541 static inline bool folio_matches_lruvec(struct folio *folio,
1542 struct lruvec *lruvec)
1544 return lruvec_pgdat(lruvec) == folio_pgdat(folio) &&
1545 lruvec_memcg(lruvec) == folio_memcg(folio);
1548 /* Don't lock again iff page's lruvec locked */
1549 static inline struct lruvec *folio_lruvec_relock_irq(struct folio *folio,
1550 struct lruvec *locked_lruvec)
1552 if (locked_lruvec) {
1553 if (folio_matches_lruvec(folio, locked_lruvec))
1554 return locked_lruvec;
1556 unlock_page_lruvec_irq(locked_lruvec);
1559 return folio_lruvec_lock_irq(folio);
1562 /* Don't lock again iff page's lruvec locked */
1563 static inline struct lruvec *folio_lruvec_relock_irqsave(struct folio *folio,
1564 struct lruvec *locked_lruvec, unsigned long *flags)
1566 if (locked_lruvec) {
1567 if (folio_matches_lruvec(folio, locked_lruvec))
1568 return locked_lruvec;
1570 unlock_page_lruvec_irqrestore(locked_lruvec, *flags);
1573 return folio_lruvec_lock_irqsave(folio, flags);
1576 #ifdef CONFIG_CGROUP_WRITEBACK
1578 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
1579 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
1580 unsigned long *pheadroom, unsigned long *pdirty,
1581 unsigned long *pwriteback);
1583 void mem_cgroup_track_foreign_dirty_slowpath(struct folio *folio,
1584 struct bdi_writeback *wb);
1586 static inline void mem_cgroup_track_foreign_dirty(struct folio *folio,
1587 struct bdi_writeback *wb)
1589 if (mem_cgroup_disabled())
1592 if (unlikely(&folio_memcg(folio)->css != wb->memcg_css))
1593 mem_cgroup_track_foreign_dirty_slowpath(folio, wb);
1596 void mem_cgroup_flush_foreign(struct bdi_writeback *wb);
1598 #else /* CONFIG_CGROUP_WRITEBACK */
1600 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
1605 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
1606 unsigned long *pfilepages,
1607 unsigned long *pheadroom,
1608 unsigned long *pdirty,
1609 unsigned long *pwriteback)
1613 static inline void mem_cgroup_track_foreign_dirty(struct folio *folio,
1614 struct bdi_writeback *wb)
1618 static inline void mem_cgroup_flush_foreign(struct bdi_writeback *wb)
1622 #endif /* CONFIG_CGROUP_WRITEBACK */
1625 bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages,
1627 void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
1629 extern struct static_key_false memcg_sockets_enabled_key;
1630 #define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_enabled_key)
1631 void mem_cgroup_sk_alloc(struct sock *sk);
1632 void mem_cgroup_sk_free(struct sock *sk);
1633 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
1635 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_pressure)
1638 if (time_before(jiffies, READ_ONCE(memcg->socket_pressure)))
1640 } while ((memcg = parent_mem_cgroup(memcg)));
1644 int alloc_shrinker_info(struct mem_cgroup *memcg);
1645 void free_shrinker_info(struct mem_cgroup *memcg);
1646 void set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id);
1647 void reparent_shrinker_deferred(struct mem_cgroup *memcg);
1649 #define mem_cgroup_sockets_enabled 0
1650 static inline void mem_cgroup_sk_alloc(struct sock *sk) { };
1651 static inline void mem_cgroup_sk_free(struct sock *sk) { };
1652 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
1657 static inline void set_shrinker_bit(struct mem_cgroup *memcg,
1658 int nid, int shrinker_id)
1663 #ifdef CONFIG_MEMCG_KMEM
1664 bool mem_cgroup_kmem_disabled(void);
1665 int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order);
1666 void __memcg_kmem_uncharge_page(struct page *page, int order);
1668 struct obj_cgroup *get_obj_cgroup_from_current(void);
1670 int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size);
1671 void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size);
1673 extern struct static_key_false memcg_kmem_enabled_key;
1675 extern int memcg_nr_cache_ids;
1676 void memcg_get_cache_ids(void);
1677 void memcg_put_cache_ids(void);
1680 * Helper macro to loop through all memcg-specific caches. Callers must still
1681 * check if the cache is valid (it is either valid or NULL).
1682 * the slab_mutex must be held when looping through those caches
1684 #define for_each_memcg_cache_index(_idx) \
1685 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
1687 static inline bool memcg_kmem_enabled(void)
1689 return static_branch_likely(&memcg_kmem_enabled_key);
1692 static inline int memcg_kmem_charge_page(struct page *page, gfp_t gfp,
1695 if (memcg_kmem_enabled())
1696 return __memcg_kmem_charge_page(page, gfp, order);
1700 static inline void memcg_kmem_uncharge_page(struct page *page, int order)
1702 if (memcg_kmem_enabled())
1703 __memcg_kmem_uncharge_page(page, order);
1707 * A helper for accessing memcg's kmem_id, used for getting
1708 * corresponding LRU lists.
1710 static inline int memcg_cache_id(struct mem_cgroup *memcg)
1712 return memcg ? memcg->kmemcg_id : -1;
1715 struct mem_cgroup *mem_cgroup_from_obj(void *p);
1718 static inline bool mem_cgroup_kmem_disabled(void)
1723 static inline int memcg_kmem_charge_page(struct page *page, gfp_t gfp,
1729 static inline void memcg_kmem_uncharge_page(struct page *page, int order)
1733 static inline int __memcg_kmem_charge_page(struct page *page, gfp_t gfp,
1739 static inline void __memcg_kmem_uncharge_page(struct page *page, int order)
1743 #define for_each_memcg_cache_index(_idx) \
1746 static inline bool memcg_kmem_enabled(void)
1751 static inline int memcg_cache_id(struct mem_cgroup *memcg)
1756 static inline void memcg_get_cache_ids(void)
1760 static inline void memcg_put_cache_ids(void)
1764 static inline struct mem_cgroup *mem_cgroup_from_obj(void *p)
1769 #endif /* CONFIG_MEMCG_KMEM */
1771 #endif /* _LINUX_MEMCONTROL_H */