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 long stat[MEMCG_NR_STAT];
80 unsigned long events[NR_VM_EVENT_ITEMS];
81 unsigned long nr_page_events;
82 unsigned long targets[MEM_CGROUP_NTARGETS];
85 struct mem_cgroup_reclaim_iter {
86 struct mem_cgroup *position;
87 /* scan generation, increased every round-trip */
88 unsigned int generation;
92 long count[NR_VM_NODE_STAT_ITEMS];
96 * Bitmap of shrinker::id corresponding to memcg-aware shrinkers,
97 * which have elements charged to this memcg.
99 struct memcg_shrinker_map {
105 * per-node information in memory controller.
107 struct mem_cgroup_per_node {
108 struct lruvec lruvec;
110 /* Legacy local VM stats */
111 struct lruvec_stat __percpu *lruvec_stat_local;
113 /* Subtree VM stats (batched updates) */
114 struct lruvec_stat __percpu *lruvec_stat_cpu;
115 atomic_long_t lruvec_stat[NR_VM_NODE_STAT_ITEMS];
117 unsigned long lru_zone_size[MAX_NR_ZONES][NR_LRU_LISTS];
119 struct mem_cgroup_reclaim_iter iter;
121 struct memcg_shrinker_map __rcu *shrinker_map;
123 struct rb_node tree_node; /* RB tree node */
124 unsigned long usage_in_excess;/* Set to the value by which */
125 /* the soft limit is exceeded*/
127 struct mem_cgroup *memcg; /* Back pointer, we cannot */
128 /* use container_of */
131 struct mem_cgroup_threshold {
132 struct eventfd_ctx *eventfd;
133 unsigned long threshold;
137 struct mem_cgroup_threshold_ary {
138 /* An array index points to threshold just below or equal to usage. */
139 int current_threshold;
140 /* Size of entries[] */
142 /* Array of thresholds */
143 struct mem_cgroup_threshold entries[];
146 struct mem_cgroup_thresholds {
147 /* Primary thresholds array */
148 struct mem_cgroup_threshold_ary *primary;
150 * Spare threshold array.
151 * This is needed to make mem_cgroup_unregister_event() "never fail".
152 * It must be able to store at least primary->size - 1 entries.
154 struct mem_cgroup_threshold_ary *spare;
157 enum memcg_kmem_state {
163 #if defined(CONFIG_SMP)
164 struct memcg_padding {
166 } ____cacheline_internodealigned_in_smp;
167 #define MEMCG_PADDING(name) struct memcg_padding name;
169 #define MEMCG_PADDING(name)
173 * Remember four most recent foreign writebacks with dirty pages in this
174 * cgroup. Inode sharing is expected to be uncommon and, even if we miss
175 * one in a given round, we're likely to catch it later if it keeps
176 * foreign-dirtying, so a fairly low count should be enough.
178 * See mem_cgroup_track_foreign_dirty_slowpath() for details.
180 #define MEMCG_CGWB_FRN_CNT 4
182 struct memcg_cgwb_frn {
183 u64 bdi_id; /* bdi->id of the foreign inode */
184 int memcg_id; /* memcg->css.id of foreign inode */
185 u64 at; /* jiffies_64 at the time of dirtying */
186 struct wb_completion done; /* tracks in-flight foreign writebacks */
190 * Bucket for arbitrarily byte-sized objects charged to a memory
191 * cgroup. The bucket can be reparented in one piece when the cgroup
192 * is destroyed, without having to round up the individual references
193 * of all live memory objects in the wild.
196 struct percpu_ref refcnt;
197 struct mem_cgroup *memcg;
198 atomic_t nr_charged_bytes;
200 struct list_head list;
206 * The memory controller data structure. The memory controller controls both
207 * page cache and RSS per cgroup. We would eventually like to provide
208 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
209 * to help the administrator determine what knobs to tune.
212 struct cgroup_subsys_state css;
214 /* Private memcg ID. Used to ID objects that outlive the cgroup */
215 struct mem_cgroup_id id;
217 /* Accounted resources */
218 struct page_counter memory; /* Both v1 & v2 */
221 struct page_counter swap; /* v2 only */
222 struct page_counter memsw; /* v1 only */
225 /* Legacy consumer-oriented counters */
226 struct page_counter kmem; /* v1 only */
227 struct page_counter tcpmem; /* v1 only */
229 /* Range enforcement for interrupt charges */
230 struct work_struct high_work;
232 unsigned long soft_limit;
234 /* vmpressure notifications */
235 struct vmpressure vmpressure;
238 * Should the accounting and control be hierarchical, per subtree?
243 * Should the OOM killer kill all belonging tasks, had it kill one?
247 /* protected by memcg_oom_lock */
252 /* OOM-Killer disable */
253 int oom_kill_disable;
255 /* memory.events and memory.events.local */
256 struct cgroup_file events_file;
257 struct cgroup_file events_local_file;
259 /* handle for "memory.swap.events" */
260 struct cgroup_file swap_events_file;
262 /* protect arrays of thresholds */
263 struct mutex thresholds_lock;
265 /* thresholds for memory usage. RCU-protected */
266 struct mem_cgroup_thresholds thresholds;
268 /* thresholds for mem+swap usage. RCU-protected */
269 struct mem_cgroup_thresholds memsw_thresholds;
271 /* For oom notifier event fd */
272 struct list_head oom_notify;
275 * Should we move charges of a task when a task is moved into this
276 * mem_cgroup ? And what type of charges should we move ?
278 unsigned long move_charge_at_immigrate;
279 /* taken only while moving_account > 0 */
280 spinlock_t move_lock;
281 unsigned long move_lock_flags;
283 MEMCG_PADDING(_pad1_);
286 * set > 0 if pages under this cgroup are moving to other cgroup.
288 atomic_t moving_account;
289 struct task_struct *move_lock_task;
291 /* Legacy local VM stats and events */
292 struct memcg_vmstats_percpu __percpu *vmstats_local;
294 /* Subtree VM stats and events (batched updates) */
295 struct memcg_vmstats_percpu __percpu *vmstats_percpu;
297 MEMCG_PADDING(_pad2_);
299 atomic_long_t vmstats[MEMCG_NR_STAT];
300 atomic_long_t vmevents[NR_VM_EVENT_ITEMS];
303 atomic_long_t memory_events[MEMCG_NR_MEMORY_EVENTS];
304 atomic_long_t memory_events_local[MEMCG_NR_MEMORY_EVENTS];
306 unsigned long socket_pressure;
308 /* Legacy tcp memory accounting */
312 #ifdef CONFIG_MEMCG_KMEM
313 /* Index in the kmem_cache->memcg_params.memcg_caches array */
315 enum memcg_kmem_state kmem_state;
316 struct obj_cgroup __rcu *objcg;
317 struct list_head objcg_list; /* list of inherited objcgs */
320 #ifdef CONFIG_CGROUP_WRITEBACK
321 struct list_head cgwb_list;
322 struct wb_domain cgwb_domain;
323 struct memcg_cgwb_frn cgwb_frn[MEMCG_CGWB_FRN_CNT];
326 /* List of events which userspace want to receive */
327 struct list_head event_list;
328 spinlock_t event_list_lock;
330 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
331 struct deferred_split deferred_split_queue;
334 struct mem_cgroup_per_node *nodeinfo[0];
335 /* WARNING: nodeinfo must be the last member here */
339 * size of first charge trial. "32" comes from vmscan.c's magic value.
340 * TODO: maybe necessary to use big numbers in big irons.
342 #define MEMCG_CHARGE_BATCH 32U
344 extern struct mem_cgroup *root_mem_cgroup;
346 enum page_memcg_data_flags {
347 /* page->memcg_data is a pointer to an objcgs vector */
348 MEMCG_DATA_OBJCGS = (1UL << 0),
349 /* the next bit after the last actual flag */
350 __NR_MEMCG_DATA_FLAGS = (1UL << 1),
353 #define MEMCG_DATA_FLAGS_MASK (__NR_MEMCG_DATA_FLAGS - 1)
356 * page_memcg - get the memory cgroup associated with a page
357 * @page: a pointer to the page struct
359 * Returns a pointer to the memory cgroup associated with the page,
360 * or NULL. This function assumes that the page is known to have a
361 * proper memory cgroup pointer. It's not safe to call this function
362 * against some type of pages, e.g. slab pages or ex-slab pages.
364 * Any of the following ensures page and memcg binding stability:
367 * - lock_page_memcg()
368 * - exclusive reference
370 static inline struct mem_cgroup *page_memcg(struct page *page)
372 VM_BUG_ON_PAGE(PageSlab(page), page);
373 return (struct mem_cgroup *)page->memcg_data;
377 * page_memcg_rcu - locklessly get the memory cgroup associated with a page
378 * @page: a pointer to the page struct
380 * Returns a pointer to the memory cgroup associated with the page,
381 * or NULL. This function assumes that the page is known to have a
382 * proper memory cgroup pointer. It's not safe to call this function
383 * against some type of pages, e.g. slab pages or ex-slab pages.
385 static inline struct mem_cgroup *page_memcg_rcu(struct page *page)
387 VM_BUG_ON_PAGE(PageSlab(page), page);
388 WARN_ON_ONCE(!rcu_read_lock_held());
390 return (struct mem_cgroup *)READ_ONCE(page->memcg_data);
394 * page_memcg_check - get the memory cgroup associated with a page
395 * @page: a pointer to the page struct
397 * Returns a pointer to the memory cgroup associated with the page,
398 * or NULL. This function unlike page_memcg() can take any page
399 * as an argument. It has to be used in cases when it's not known if a page
400 * has an associated memory cgroup pointer or an object cgroups vector.
402 * Any of the following ensures page and memcg binding stability:
405 * - lock_page_memcg()
406 * - exclusive reference
408 static inline struct mem_cgroup *page_memcg_check(struct page *page)
411 * Because page->memcg_data might be changed asynchronously
412 * for slab pages, READ_ONCE() should be used here.
414 unsigned long memcg_data = READ_ONCE(page->memcg_data);
416 if (memcg_data & MEMCG_DATA_OBJCGS)
419 return (struct mem_cgroup *)memcg_data;
422 #ifdef CONFIG_MEMCG_KMEM
424 * page_objcgs - get the object cgroups vector associated with a page
425 * @page: a pointer to the page struct
427 * Returns a pointer to the object cgroups vector associated with the page,
428 * or NULL. This function assumes that the page is known to have an
429 * associated object cgroups vector. It's not safe to call this function
430 * against pages, which might have an associated memory cgroup: e.g.
431 * kernel stack pages.
433 static inline struct obj_cgroup **page_objcgs(struct page *page)
435 unsigned long memcg_data = READ_ONCE(page->memcg_data);
437 VM_BUG_ON_PAGE(memcg_data && !(memcg_data & MEMCG_DATA_OBJCGS), page);
439 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
443 * page_objcgs_check - get the object cgroups vector associated with a page
444 * @page: a pointer to the page struct
446 * Returns a pointer to the object cgroups vector associated with the page,
447 * or NULL. This function is safe to use if the page can be directly associated
448 * with a memory cgroup.
450 static inline struct obj_cgroup **page_objcgs_check(struct page *page)
452 unsigned long memcg_data = READ_ONCE(page->memcg_data);
454 if (!memcg_data || !(memcg_data & MEMCG_DATA_OBJCGS))
457 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
461 * set_page_objcgs - associate a page with a object cgroups vector
462 * @page: a pointer to the page struct
463 * @objcgs: a pointer to the object cgroups vector
465 * Atomically associates a page with a vector of object cgroups.
467 static inline bool set_page_objcgs(struct page *page,
468 struct obj_cgroup **objcgs)
470 return !cmpxchg(&page->memcg_data, 0, (unsigned long)objcgs |
474 static inline struct obj_cgroup **page_objcgs(struct page *page)
479 static inline struct obj_cgroup **page_objcgs_check(struct page *page)
484 static inline bool set_page_objcgs(struct page *page,
485 struct obj_cgroup **objcgs)
491 static __always_inline bool memcg_stat_item_in_bytes(int idx)
493 if (idx == MEMCG_PERCPU_B)
495 return vmstat_item_in_bytes(idx);
498 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
500 return (memcg == root_mem_cgroup);
503 static inline bool mem_cgroup_disabled(void)
505 return !cgroup_subsys_enabled(memory_cgrp_subsys);
508 static inline unsigned long mem_cgroup_protection(struct mem_cgroup *root,
509 struct mem_cgroup *memcg,
512 if (mem_cgroup_disabled())
516 * There is no reclaim protection applied to a targeted reclaim.
517 * We are special casing this specific case here because
518 * mem_cgroup_protected calculation is not robust enough to keep
519 * the protection invariant for calculated effective values for
520 * parallel reclaimers with different reclaim target. This is
521 * especially a problem for tail memcgs (as they have pages on LRU)
522 * which would want to have effective values 0 for targeted reclaim
523 * but a different value for external reclaim.
526 * Let's have global and A's reclaim in parallel:
528 * A (low=2G, usage = 3G, max = 3G, children_low_usage = 1.5G)
530 * | C (low = 1G, usage = 2.5G)
531 * B (low = 1G, usage = 0.5G)
533 * For the global reclaim
535 * B.elow = min(B.usage, B.low) because children_low_usage <= A.elow
536 * C.elow = min(C.usage, C.low)
538 * With the effective values resetting we have A reclaim
543 * If the global reclaim races with A's reclaim then
544 * B.elow = C.elow = 0 because children_low_usage > A.elow)
545 * is possible and reclaiming B would be violating the protection.
552 return READ_ONCE(memcg->memory.emin);
554 return max(READ_ONCE(memcg->memory.emin),
555 READ_ONCE(memcg->memory.elow));
558 void mem_cgroup_calculate_protection(struct mem_cgroup *root,
559 struct mem_cgroup *memcg);
561 static inline bool mem_cgroup_supports_protection(struct mem_cgroup *memcg)
564 * The root memcg doesn't account charges, and doesn't support
567 return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg);
571 static inline bool mem_cgroup_below_low(struct mem_cgroup *memcg)
573 if (!mem_cgroup_supports_protection(memcg))
576 return READ_ONCE(memcg->memory.elow) >=
577 page_counter_read(&memcg->memory);
580 static inline bool mem_cgroup_below_min(struct mem_cgroup *memcg)
582 if (!mem_cgroup_supports_protection(memcg))
585 return READ_ONCE(memcg->memory.emin) >=
586 page_counter_read(&memcg->memory);
589 int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask);
591 void mem_cgroup_uncharge(struct page *page);
592 void mem_cgroup_uncharge_list(struct list_head *page_list);
594 void mem_cgroup_migrate(struct page *oldpage, struct page *newpage);
596 static struct mem_cgroup_per_node *
597 mem_cgroup_nodeinfo(struct mem_cgroup *memcg, int nid)
599 return memcg->nodeinfo[nid];
603 * mem_cgroup_lruvec - get the lru list vector for a memcg & node
604 * @memcg: memcg of the wanted lruvec
606 * Returns the lru list vector holding pages for a given @memcg &
607 * @node combination. This can be the node lruvec, if the memory
608 * controller is disabled.
610 static inline struct lruvec *mem_cgroup_lruvec(struct mem_cgroup *memcg,
611 struct pglist_data *pgdat)
613 struct mem_cgroup_per_node *mz;
614 struct lruvec *lruvec;
616 if (mem_cgroup_disabled()) {
617 lruvec = &pgdat->__lruvec;
622 memcg = root_mem_cgroup;
624 mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
625 lruvec = &mz->lruvec;
628 * Since a node can be onlined after the mem_cgroup was created,
629 * we have to be prepared to initialize lruvec->pgdat here;
630 * and if offlined then reonlined, we need to reinitialize it.
632 if (unlikely(lruvec->pgdat != pgdat))
633 lruvec->pgdat = pgdat;
637 struct lruvec *mem_cgroup_page_lruvec(struct page *, struct pglist_data *);
639 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
641 struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm);
643 struct mem_cgroup *get_mem_cgroup_from_page(struct page *page);
646 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
647 return css ? container_of(css, struct mem_cgroup, css) : NULL;
650 static inline bool obj_cgroup_tryget(struct obj_cgroup *objcg)
652 return percpu_ref_tryget(&objcg->refcnt);
655 static inline void obj_cgroup_get(struct obj_cgroup *objcg)
657 percpu_ref_get(&objcg->refcnt);
660 static inline void obj_cgroup_put(struct obj_cgroup *objcg)
662 percpu_ref_put(&objcg->refcnt);
666 * After the initialization objcg->memcg is always pointing at
667 * a valid memcg, but can be atomically swapped to the parent memcg.
669 * The caller must ensure that the returned memcg won't be released:
670 * e.g. acquire the rcu_read_lock or css_set_lock.
672 static inline struct mem_cgroup *obj_cgroup_memcg(struct obj_cgroup *objcg)
674 return READ_ONCE(objcg->memcg);
677 static inline void mem_cgroup_put(struct mem_cgroup *memcg)
680 css_put(&memcg->css);
683 #define mem_cgroup_from_counter(counter, member) \
684 container_of(counter, struct mem_cgroup, member)
686 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
688 struct mem_cgroup_reclaim_cookie *);
689 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
690 int mem_cgroup_scan_tasks(struct mem_cgroup *,
691 int (*)(struct task_struct *, void *), void *);
693 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
695 if (mem_cgroup_disabled())
700 struct mem_cgroup *mem_cgroup_from_id(unsigned short id);
702 static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m)
704 return mem_cgroup_from_css(seq_css(m));
707 static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec)
709 struct mem_cgroup_per_node *mz;
711 if (mem_cgroup_disabled())
714 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
719 * parent_mem_cgroup - find the accounting parent of a memcg
720 * @memcg: memcg whose parent to find
722 * Returns the parent memcg, or NULL if this is the root or the memory
723 * controller is in legacy no-hierarchy mode.
725 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
727 if (!memcg->memory.parent)
729 return mem_cgroup_from_counter(memcg->memory.parent, memory);
732 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
733 struct mem_cgroup *root)
737 if (!root->use_hierarchy)
739 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
742 static inline bool mm_match_cgroup(struct mm_struct *mm,
743 struct mem_cgroup *memcg)
745 struct mem_cgroup *task_memcg;
749 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
751 match = mem_cgroup_is_descendant(task_memcg, memcg);
756 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
757 ino_t page_cgroup_ino(struct page *page);
759 static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
761 if (mem_cgroup_disabled())
763 return !!(memcg->css.flags & CSS_ONLINE);
767 * For memory reclaim.
769 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
771 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
772 int zid, int nr_pages);
775 unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
776 enum lru_list lru, int zone_idx)
778 struct mem_cgroup_per_node *mz;
780 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
781 return READ_ONCE(mz->lru_zone_size[zone_idx][lru]);
784 void mem_cgroup_handle_over_high(void);
786 unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg);
788 unsigned long mem_cgroup_size(struct mem_cgroup *memcg);
790 void mem_cgroup_print_oom_context(struct mem_cgroup *memcg,
791 struct task_struct *p);
793 void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg);
795 static inline void mem_cgroup_enter_user_fault(void)
797 WARN_ON(current->in_user_fault);
798 current->in_user_fault = 1;
801 static inline void mem_cgroup_exit_user_fault(void)
803 WARN_ON(!current->in_user_fault);
804 current->in_user_fault = 0;
807 static inline bool task_in_memcg_oom(struct task_struct *p)
809 return p->memcg_in_oom;
812 bool mem_cgroup_oom_synchronize(bool wait);
813 struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim,
814 struct mem_cgroup *oom_domain);
815 void mem_cgroup_print_oom_group(struct mem_cgroup *memcg);
817 #ifdef CONFIG_MEMCG_SWAP
818 extern bool cgroup_memory_noswap;
821 struct mem_cgroup *lock_page_memcg(struct page *page);
822 void __unlock_page_memcg(struct mem_cgroup *memcg);
823 void unlock_page_memcg(struct page *page);
826 * idx can be of type enum memcg_stat_item or node_stat_item.
827 * Keep in sync with memcg_exact_page_state().
829 static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx)
831 long x = atomic_long_read(&memcg->vmstats[idx]);
840 * idx can be of type enum memcg_stat_item or node_stat_item.
841 * Keep in sync with memcg_exact_page_state().
843 static inline unsigned long memcg_page_state_local(struct mem_cgroup *memcg,
849 for_each_possible_cpu(cpu)
850 x += per_cpu(memcg->vmstats_local->stat[idx], cpu);
858 void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val);
860 /* idx can be of type enum memcg_stat_item or node_stat_item */
861 static inline void mod_memcg_state(struct mem_cgroup *memcg,
866 local_irq_save(flags);
867 __mod_memcg_state(memcg, idx, val);
868 local_irq_restore(flags);
872 * mod_memcg_page_state - update page state statistics
874 * @idx: page state item to account
875 * @val: number of pages (positive or negative)
877 * The @page must be locked or the caller must use lock_page_memcg()
878 * to prevent double accounting when the page is concurrently being
879 * moved to another memcg:
881 * lock_page(page) or lock_page_memcg(page)
882 * if (TestClearPageState(page))
883 * mod_memcg_page_state(page, state, -1);
884 * unlock_page(page) or unlock_page_memcg(page)
886 * Kernel pages are an exception to this, since they'll never move.
888 static inline void __mod_memcg_page_state(struct page *page,
891 struct mem_cgroup *memcg = page_memcg(page);
894 __mod_memcg_state(memcg, idx, val);
897 static inline void mod_memcg_page_state(struct page *page,
900 struct mem_cgroup *memcg = page_memcg(page);
903 mod_memcg_state(memcg, idx, val);
906 static inline unsigned long lruvec_page_state(struct lruvec *lruvec,
907 enum node_stat_item idx)
909 struct mem_cgroup_per_node *pn;
912 if (mem_cgroup_disabled())
913 return node_page_state(lruvec_pgdat(lruvec), idx);
915 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
916 x = atomic_long_read(&pn->lruvec_stat[idx]);
924 static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec,
925 enum node_stat_item idx)
927 struct mem_cgroup_per_node *pn;
931 if (mem_cgroup_disabled())
932 return node_page_state(lruvec_pgdat(lruvec), idx);
934 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
935 for_each_possible_cpu(cpu)
936 x += per_cpu(pn->lruvec_stat_local->count[idx], cpu);
944 void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx,
946 void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx,
948 void __mod_lruvec_slab_state(void *p, enum node_stat_item idx, int val);
950 void mod_memcg_obj_state(void *p, int idx, int val);
952 static inline void mod_lruvec_slab_state(void *p, enum node_stat_item idx,
957 local_irq_save(flags);
958 __mod_lruvec_slab_state(p, idx, val);
959 local_irq_restore(flags);
962 static inline void mod_memcg_lruvec_state(struct lruvec *lruvec,
963 enum node_stat_item idx, int val)
967 local_irq_save(flags);
968 __mod_memcg_lruvec_state(lruvec, idx, val);
969 local_irq_restore(flags);
972 static inline void mod_lruvec_state(struct lruvec *lruvec,
973 enum node_stat_item idx, int val)
977 local_irq_save(flags);
978 __mod_lruvec_state(lruvec, idx, val);
979 local_irq_restore(flags);
982 static inline void __mod_lruvec_page_state(struct page *page,
983 enum node_stat_item idx, int val)
985 struct page *head = compound_head(page); /* rmap on tail pages */
986 struct mem_cgroup *memcg = page_memcg(head);
987 pg_data_t *pgdat = page_pgdat(page);
988 struct lruvec *lruvec;
990 /* Untracked pages have no memcg, no lruvec. Update only the node */
992 __mod_node_page_state(pgdat, idx, val);
996 lruvec = mem_cgroup_lruvec(memcg, pgdat);
997 __mod_lruvec_state(lruvec, idx, val);
1000 static inline void mod_lruvec_page_state(struct page *page,
1001 enum node_stat_item idx, int val)
1003 unsigned long flags;
1005 local_irq_save(flags);
1006 __mod_lruvec_page_state(page, idx, val);
1007 local_irq_restore(flags);
1010 unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
1012 unsigned long *total_scanned);
1014 void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx,
1015 unsigned long count);
1017 static inline void count_memcg_events(struct mem_cgroup *memcg,
1018 enum vm_event_item idx,
1019 unsigned long count)
1021 unsigned long flags;
1023 local_irq_save(flags);
1024 __count_memcg_events(memcg, idx, count);
1025 local_irq_restore(flags);
1028 static inline void count_memcg_page_event(struct page *page,
1029 enum vm_event_item idx)
1031 struct mem_cgroup *memcg = page_memcg(page);
1034 count_memcg_events(memcg, idx, 1);
1037 static inline void count_memcg_event_mm(struct mm_struct *mm,
1038 enum vm_event_item idx)
1040 struct mem_cgroup *memcg;
1042 if (mem_cgroup_disabled())
1046 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1048 count_memcg_events(memcg, idx, 1);
1052 static inline void memcg_memory_event(struct mem_cgroup *memcg,
1053 enum memcg_memory_event event)
1055 atomic_long_inc(&memcg->memory_events_local[event]);
1056 cgroup_file_notify(&memcg->events_local_file);
1059 atomic_long_inc(&memcg->memory_events[event]);
1060 cgroup_file_notify(&memcg->events_file);
1062 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
1064 if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
1066 } while ((memcg = parent_mem_cgroup(memcg)) &&
1067 !mem_cgroup_is_root(memcg));
1070 static inline void memcg_memory_event_mm(struct mm_struct *mm,
1071 enum memcg_memory_event event)
1073 struct mem_cgroup *memcg;
1075 if (mem_cgroup_disabled())
1079 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1081 memcg_memory_event(memcg, event);
1085 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1086 void mem_cgroup_split_huge_fixup(struct page *head);
1089 #else /* CONFIG_MEMCG */
1091 #define MEM_CGROUP_ID_SHIFT 0
1092 #define MEM_CGROUP_ID_MAX 0
1096 static inline struct mem_cgroup *page_memcg(struct page *page)
1101 static inline struct mem_cgroup *page_memcg_rcu(struct page *page)
1103 WARN_ON_ONCE(!rcu_read_lock_held());
1107 static inline struct mem_cgroup *page_memcg_check(struct page *page)
1112 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
1117 static inline bool mem_cgroup_disabled(void)
1122 static inline void memcg_memory_event(struct mem_cgroup *memcg,
1123 enum memcg_memory_event event)
1127 static inline void memcg_memory_event_mm(struct mm_struct *mm,
1128 enum memcg_memory_event event)
1132 static inline unsigned long mem_cgroup_protection(struct mem_cgroup *root,
1133 struct mem_cgroup *memcg,
1134 bool in_low_reclaim)
1139 static inline void mem_cgroup_calculate_protection(struct mem_cgroup *root,
1140 struct mem_cgroup *memcg)
1144 static inline bool mem_cgroup_below_low(struct mem_cgroup *memcg)
1149 static inline bool mem_cgroup_below_min(struct mem_cgroup *memcg)
1154 static inline int mem_cgroup_charge(struct page *page, struct mm_struct *mm,
1160 static inline void mem_cgroup_uncharge(struct page *page)
1164 static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
1168 static inline void mem_cgroup_migrate(struct page *old, struct page *new)
1172 static inline struct lruvec *mem_cgroup_lruvec(struct mem_cgroup *memcg,
1173 struct pglist_data *pgdat)
1175 return &pgdat->__lruvec;
1178 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
1179 struct pglist_data *pgdat)
1181 return &pgdat->__lruvec;
1184 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
1189 static inline bool mm_match_cgroup(struct mm_struct *mm,
1190 struct mem_cgroup *memcg)
1195 static inline struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
1200 static inline struct mem_cgroup *get_mem_cgroup_from_page(struct page *page)
1205 static inline void mem_cgroup_put(struct mem_cgroup *memcg)
1209 static inline struct mem_cgroup *
1210 mem_cgroup_iter(struct mem_cgroup *root,
1211 struct mem_cgroup *prev,
1212 struct mem_cgroup_reclaim_cookie *reclaim)
1217 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
1218 struct mem_cgroup *prev)
1222 static inline int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
1223 int (*fn)(struct task_struct *, void *), void *arg)
1228 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
1233 static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
1236 /* XXX: This should always return root_mem_cgroup */
1240 static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m)
1245 static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec)
1250 static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
1256 unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
1257 enum lru_list lru, int zone_idx)
1262 static inline unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg)
1267 static inline unsigned long mem_cgroup_size(struct mem_cgroup *memcg)
1273 mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p)
1278 mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg)
1282 static inline struct mem_cgroup *lock_page_memcg(struct page *page)
1287 static inline void __unlock_page_memcg(struct mem_cgroup *memcg)
1291 static inline void unlock_page_memcg(struct page *page)
1295 static inline void mem_cgroup_handle_over_high(void)
1299 static inline void mem_cgroup_enter_user_fault(void)
1303 static inline void mem_cgroup_exit_user_fault(void)
1307 static inline bool task_in_memcg_oom(struct task_struct *p)
1312 static inline bool mem_cgroup_oom_synchronize(bool wait)
1317 static inline struct mem_cgroup *mem_cgroup_get_oom_group(
1318 struct task_struct *victim, struct mem_cgroup *oom_domain)
1323 static inline void mem_cgroup_print_oom_group(struct mem_cgroup *memcg)
1327 static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx)
1332 static inline unsigned long memcg_page_state_local(struct mem_cgroup *memcg,
1338 static inline void __mod_memcg_state(struct mem_cgroup *memcg,
1344 static inline void mod_memcg_state(struct mem_cgroup *memcg,
1350 static inline void __mod_memcg_page_state(struct page *page,
1356 static inline void mod_memcg_page_state(struct page *page,
1362 static inline unsigned long lruvec_page_state(struct lruvec *lruvec,
1363 enum node_stat_item idx)
1365 return node_page_state(lruvec_pgdat(lruvec), idx);
1368 static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec,
1369 enum node_stat_item idx)
1371 return node_page_state(lruvec_pgdat(lruvec), idx);
1374 static inline void __mod_memcg_lruvec_state(struct lruvec *lruvec,
1375 enum node_stat_item idx, int val)
1379 static inline void __mod_lruvec_state(struct lruvec *lruvec,
1380 enum node_stat_item idx, int val)
1382 __mod_node_page_state(lruvec_pgdat(lruvec), idx, val);
1385 static inline void mod_lruvec_state(struct lruvec *lruvec,
1386 enum node_stat_item idx, int val)
1388 mod_node_page_state(lruvec_pgdat(lruvec), idx, val);
1391 static inline void __mod_lruvec_page_state(struct page *page,
1392 enum node_stat_item idx, int val)
1394 __mod_node_page_state(page_pgdat(page), idx, val);
1397 static inline void mod_lruvec_page_state(struct page *page,
1398 enum node_stat_item idx, int val)
1400 mod_node_page_state(page_pgdat(page), idx, val);
1403 static inline void __mod_lruvec_slab_state(void *p, enum node_stat_item idx,
1406 struct page *page = virt_to_head_page(p);
1408 __mod_node_page_state(page_pgdat(page), idx, val);
1411 static inline void mod_lruvec_slab_state(void *p, enum node_stat_item idx,
1414 struct page *page = virt_to_head_page(p);
1416 mod_node_page_state(page_pgdat(page), idx, val);
1419 static inline void mod_memcg_obj_state(void *p, int idx, int val)
1424 unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
1426 unsigned long *total_scanned)
1431 static inline void mem_cgroup_split_huge_fixup(struct page *head)
1435 static inline void count_memcg_events(struct mem_cgroup *memcg,
1436 enum vm_event_item idx,
1437 unsigned long count)
1441 static inline void __count_memcg_events(struct mem_cgroup *memcg,
1442 enum vm_event_item idx,
1443 unsigned long count)
1447 static inline void count_memcg_page_event(struct page *page,
1453 void count_memcg_event_mm(struct mm_struct *mm, enum vm_event_item idx)
1456 #endif /* CONFIG_MEMCG */
1458 /* idx can be of type enum memcg_stat_item or node_stat_item */
1459 static inline void __inc_memcg_state(struct mem_cgroup *memcg,
1462 __mod_memcg_state(memcg, idx, 1);
1465 /* idx can be of type enum memcg_stat_item or node_stat_item */
1466 static inline void __dec_memcg_state(struct mem_cgroup *memcg,
1469 __mod_memcg_state(memcg, idx, -1);
1472 /* idx can be of type enum memcg_stat_item or node_stat_item */
1473 static inline void __inc_memcg_page_state(struct page *page,
1476 __mod_memcg_page_state(page, idx, 1);
1479 /* idx can be of type enum memcg_stat_item or node_stat_item */
1480 static inline void __dec_memcg_page_state(struct page *page,
1483 __mod_memcg_page_state(page, idx, -1);
1486 static inline void __inc_lruvec_state(struct lruvec *lruvec,
1487 enum node_stat_item idx)
1489 __mod_lruvec_state(lruvec, idx, 1);
1492 static inline void __dec_lruvec_state(struct lruvec *lruvec,
1493 enum node_stat_item idx)
1495 __mod_lruvec_state(lruvec, idx, -1);
1498 static inline void __inc_lruvec_page_state(struct page *page,
1499 enum node_stat_item idx)
1501 __mod_lruvec_page_state(page, idx, 1);
1504 static inline void __dec_lruvec_page_state(struct page *page,
1505 enum node_stat_item idx)
1507 __mod_lruvec_page_state(page, idx, -1);
1510 static inline void __inc_lruvec_slab_state(void *p, enum node_stat_item idx)
1512 __mod_lruvec_slab_state(p, idx, 1);
1515 static inline void __dec_lruvec_slab_state(void *p, enum node_stat_item idx)
1517 __mod_lruvec_slab_state(p, idx, -1);
1520 /* idx can be of type enum memcg_stat_item or node_stat_item */
1521 static inline void inc_memcg_state(struct mem_cgroup *memcg,
1524 mod_memcg_state(memcg, idx, 1);
1527 /* idx can be of type enum memcg_stat_item or node_stat_item */
1528 static inline void dec_memcg_state(struct mem_cgroup *memcg,
1531 mod_memcg_state(memcg, idx, -1);
1534 /* idx can be of type enum memcg_stat_item or node_stat_item */
1535 static inline void inc_memcg_page_state(struct page *page,
1538 mod_memcg_page_state(page, idx, 1);
1541 /* idx can be of type enum memcg_stat_item or node_stat_item */
1542 static inline void dec_memcg_page_state(struct page *page,
1545 mod_memcg_page_state(page, idx, -1);
1548 static inline void inc_lruvec_state(struct lruvec *lruvec,
1549 enum node_stat_item idx)
1551 mod_lruvec_state(lruvec, idx, 1);
1554 static inline void dec_lruvec_state(struct lruvec *lruvec,
1555 enum node_stat_item idx)
1557 mod_lruvec_state(lruvec, idx, -1);
1560 static inline void inc_lruvec_page_state(struct page *page,
1561 enum node_stat_item idx)
1563 mod_lruvec_page_state(page, idx, 1);
1566 static inline void dec_lruvec_page_state(struct page *page,
1567 enum node_stat_item idx)
1569 mod_lruvec_page_state(page, idx, -1);
1572 static inline struct lruvec *parent_lruvec(struct lruvec *lruvec)
1574 struct mem_cgroup *memcg;
1576 memcg = lruvec_memcg(lruvec);
1579 memcg = parent_mem_cgroup(memcg);
1582 return mem_cgroup_lruvec(memcg, lruvec_pgdat(lruvec));
1585 #ifdef CONFIG_CGROUP_WRITEBACK
1587 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
1588 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
1589 unsigned long *pheadroom, unsigned long *pdirty,
1590 unsigned long *pwriteback);
1592 void mem_cgroup_track_foreign_dirty_slowpath(struct page *page,
1593 struct bdi_writeback *wb);
1595 static inline void mem_cgroup_track_foreign_dirty(struct page *page,
1596 struct bdi_writeback *wb)
1598 if (mem_cgroup_disabled())
1601 if (unlikely(&page_memcg(page)->css != wb->memcg_css))
1602 mem_cgroup_track_foreign_dirty_slowpath(page, wb);
1605 void mem_cgroup_flush_foreign(struct bdi_writeback *wb);
1607 #else /* CONFIG_CGROUP_WRITEBACK */
1609 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
1614 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
1615 unsigned long *pfilepages,
1616 unsigned long *pheadroom,
1617 unsigned long *pdirty,
1618 unsigned long *pwriteback)
1622 static inline void mem_cgroup_track_foreign_dirty(struct page *page,
1623 struct bdi_writeback *wb)
1627 static inline void mem_cgroup_flush_foreign(struct bdi_writeback *wb)
1631 #endif /* CONFIG_CGROUP_WRITEBACK */
1634 bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
1635 void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
1637 extern struct static_key_false memcg_sockets_enabled_key;
1638 #define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_enabled_key)
1639 void mem_cgroup_sk_alloc(struct sock *sk);
1640 void mem_cgroup_sk_free(struct sock *sk);
1641 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
1643 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_pressure)
1646 if (time_before(jiffies, memcg->socket_pressure))
1648 } while ((memcg = parent_mem_cgroup(memcg)));
1652 extern int memcg_expand_shrinker_maps(int new_id);
1654 extern void memcg_set_shrinker_bit(struct mem_cgroup *memcg,
1655 int nid, int shrinker_id);
1657 #define mem_cgroup_sockets_enabled 0
1658 static inline void mem_cgroup_sk_alloc(struct sock *sk) { };
1659 static inline void mem_cgroup_sk_free(struct sock *sk) { };
1660 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
1665 static inline void memcg_set_shrinker_bit(struct mem_cgroup *memcg,
1666 int nid, int shrinker_id)
1671 #ifdef CONFIG_MEMCG_KMEM
1672 int __memcg_kmem_charge(struct mem_cgroup *memcg, gfp_t gfp,
1673 unsigned int nr_pages);
1674 void __memcg_kmem_uncharge(struct mem_cgroup *memcg, unsigned int nr_pages);
1675 int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order);
1676 void __memcg_kmem_uncharge_page(struct page *page, int order);
1678 struct obj_cgroup *get_obj_cgroup_from_current(void);
1680 int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size);
1681 void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size);
1683 extern struct static_key_false memcg_kmem_enabled_key;
1685 extern int memcg_nr_cache_ids;
1686 void memcg_get_cache_ids(void);
1687 void memcg_put_cache_ids(void);
1690 * Helper macro to loop through all memcg-specific caches. Callers must still
1691 * check if the cache is valid (it is either valid or NULL).
1692 * the slab_mutex must be held when looping through those caches
1694 #define for_each_memcg_cache_index(_idx) \
1695 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
1697 static inline bool memcg_kmem_enabled(void)
1699 return static_branch_likely(&memcg_kmem_enabled_key);
1702 static inline int memcg_kmem_charge_page(struct page *page, gfp_t gfp,
1705 if (memcg_kmem_enabled())
1706 return __memcg_kmem_charge_page(page, gfp, order);
1710 static inline void memcg_kmem_uncharge_page(struct page *page, int order)
1712 if (memcg_kmem_enabled())
1713 __memcg_kmem_uncharge_page(page, order);
1716 static inline int memcg_kmem_charge(struct mem_cgroup *memcg, gfp_t gfp,
1717 unsigned int nr_pages)
1719 if (memcg_kmem_enabled())
1720 return __memcg_kmem_charge(memcg, gfp, nr_pages);
1724 static inline void memcg_kmem_uncharge(struct mem_cgroup *memcg,
1725 unsigned int nr_pages)
1727 if (memcg_kmem_enabled())
1728 __memcg_kmem_uncharge(memcg, nr_pages);
1732 * helper for accessing a memcg's index. It will be used as an index in the
1733 * child cache array in kmem_cache, and also to derive its name. This function
1734 * will return -1 when this is not a kmem-limited memcg.
1736 static inline int memcg_cache_id(struct mem_cgroup *memcg)
1738 return memcg ? memcg->kmemcg_id : -1;
1741 struct mem_cgroup *mem_cgroup_from_obj(void *p);
1745 static inline int memcg_kmem_charge_page(struct page *page, gfp_t gfp,
1751 static inline void memcg_kmem_uncharge_page(struct page *page, int order)
1755 static inline int __memcg_kmem_charge_page(struct page *page, gfp_t gfp,
1761 static inline void __memcg_kmem_uncharge_page(struct page *page, int order)
1765 #define for_each_memcg_cache_index(_idx) \
1768 static inline bool memcg_kmem_enabled(void)
1773 static inline int memcg_cache_id(struct mem_cgroup *memcg)
1778 static inline void memcg_get_cache_ids(void)
1782 static inline void memcg_put_cache_ids(void)
1786 static inline struct mem_cgroup *mem_cgroup_from_obj(void *p)
1791 #endif /* CONFIG_MEMCG_KMEM */
1793 #endif /* _LINUX_MEMCONTROL_H */