1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /* memcontrol.h - Memory Controller
4 * Copyright IBM Corporation, 2007
7 * Copyright 2007 OpenVZ SWsoft Inc
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 long count[NR_VM_NODE_STAT_ITEMS];
112 struct batched_lruvec_stat {
113 s32 count[NR_VM_NODE_STAT_ITEMS];
117 * Bitmap and deferred work of shrinker::id corresponding to memcg-aware
118 * shrinkers, which have elements charged to this memcg.
120 struct shrinker_info {
122 atomic_long_t *nr_deferred;
127 * per-node information in memory controller.
129 struct mem_cgroup_per_node {
130 struct lruvec lruvec;
133 * Legacy local VM stats. This should be struct lruvec_stat and
134 * cannot be optimized to struct batched_lruvec_stat. Because
135 * the threshold of the lruvec_stat_cpu can be as big as
136 * MEMCG_CHARGE_BATCH * PAGE_SIZE. It can fit into s32. But this
137 * filed has no upper limit.
139 struct lruvec_stat __percpu *lruvec_stat_local;
141 /* Subtree VM stats (batched updates) */
142 struct batched_lruvec_stat __percpu *lruvec_stat_cpu;
143 atomic_long_t lruvec_stat[NR_VM_NODE_STAT_ITEMS];
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 enum memcg_kmem_state {
191 #if defined(CONFIG_SMP)
192 struct memcg_padding {
194 } ____cacheline_internodealigned_in_smp;
195 #define MEMCG_PADDING(name) struct memcg_padding name;
197 #define MEMCG_PADDING(name)
201 * Remember four most recent foreign writebacks with dirty pages in this
202 * cgroup. Inode sharing is expected to be uncommon and, even if we miss
203 * one in a given round, we're likely to catch it later if it keeps
204 * foreign-dirtying, so a fairly low count should be enough.
206 * See mem_cgroup_track_foreign_dirty_slowpath() for details.
208 #define MEMCG_CGWB_FRN_CNT 4
210 struct memcg_cgwb_frn {
211 u64 bdi_id; /* bdi->id of the foreign inode */
212 int memcg_id; /* memcg->css.id of foreign inode */
213 u64 at; /* jiffies_64 at the time of dirtying */
214 struct wb_completion done; /* tracks in-flight foreign writebacks */
218 * Bucket for arbitrarily byte-sized objects charged to a memory
219 * cgroup. The bucket can be reparented in one piece when the cgroup
220 * is destroyed, without having to round up the individual references
221 * of all live memory objects in the wild.
224 struct percpu_ref refcnt;
225 struct mem_cgroup *memcg;
226 atomic_t nr_charged_bytes;
228 struct list_head list;
234 * The memory controller data structure. The memory controller controls both
235 * page cache and RSS per cgroup. We would eventually like to provide
236 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
237 * to help the administrator determine what knobs to tune.
240 struct cgroup_subsys_state css;
242 /* Private memcg ID. Used to ID objects that outlive the cgroup */
243 struct mem_cgroup_id id;
245 /* Accounted resources */
246 struct page_counter memory; /* Both v1 & v2 */
249 struct page_counter swap; /* v2 only */
250 struct page_counter memsw; /* v1 only */
253 /* Legacy consumer-oriented counters */
254 struct page_counter kmem; /* v1 only */
255 struct page_counter tcpmem; /* v1 only */
257 /* Range enforcement for interrupt charges */
258 struct work_struct high_work;
260 unsigned long soft_limit;
262 /* vmpressure notifications */
263 struct vmpressure vmpressure;
266 * Should the OOM killer kill all belonging tasks, had it kill one?
270 /* protected by memcg_oom_lock */
275 /* OOM-Killer disable */
276 int oom_kill_disable;
278 /* memory.events and memory.events.local */
279 struct cgroup_file events_file;
280 struct cgroup_file events_local_file;
282 /* handle for "memory.swap.events" */
283 struct cgroup_file swap_events_file;
285 /* protect arrays of thresholds */
286 struct mutex thresholds_lock;
288 /* thresholds for memory usage. RCU-protected */
289 struct mem_cgroup_thresholds thresholds;
291 /* thresholds for mem+swap usage. RCU-protected */
292 struct mem_cgroup_thresholds memsw_thresholds;
294 /* For oom notifier event fd */
295 struct list_head oom_notify;
298 * Should we move charges of a task when a task is moved into this
299 * mem_cgroup ? And what type of charges should we move ?
301 unsigned long move_charge_at_immigrate;
302 /* taken only while moving_account > 0 */
303 spinlock_t move_lock;
304 unsigned long move_lock_flags;
306 MEMCG_PADDING(_pad1_);
309 struct memcg_vmstats vmstats;
312 atomic_long_t memory_events[MEMCG_NR_MEMORY_EVENTS];
313 atomic_long_t memory_events_local[MEMCG_NR_MEMORY_EVENTS];
315 unsigned long socket_pressure;
317 /* Legacy tcp memory accounting */
321 #ifdef CONFIG_MEMCG_KMEM
323 enum memcg_kmem_state kmem_state;
324 struct obj_cgroup __rcu *objcg;
325 struct list_head objcg_list; /* list of inherited objcgs */
328 MEMCG_PADDING(_pad2_);
331 * set > 0 if pages under this cgroup are moving to other cgroup.
333 atomic_t moving_account;
334 struct task_struct *move_lock_task;
336 struct memcg_vmstats_percpu __percpu *vmstats_percpu;
338 #ifdef CONFIG_CGROUP_WRITEBACK
339 struct list_head cgwb_list;
340 struct wb_domain cgwb_domain;
341 struct memcg_cgwb_frn cgwb_frn[MEMCG_CGWB_FRN_CNT];
344 /* List of events which userspace want to receive */
345 struct list_head event_list;
346 spinlock_t event_list_lock;
348 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
349 struct deferred_split deferred_split_queue;
352 struct mem_cgroup_per_node *nodeinfo[0];
353 /* WARNING: nodeinfo must be the last member here */
357 * size of first charge trial. "32" comes from vmscan.c's magic value.
358 * TODO: maybe necessary to use big numbers in big irons.
360 #define MEMCG_CHARGE_BATCH 32U
362 extern struct mem_cgroup *root_mem_cgroup;
364 enum page_memcg_data_flags {
365 /* page->memcg_data is a pointer to an objcgs vector */
366 MEMCG_DATA_OBJCGS = (1UL << 0),
367 /* page has been accounted as a non-slab kernel page */
368 MEMCG_DATA_KMEM = (1UL << 1),
369 /* the next bit after the last actual flag */
370 __NR_MEMCG_DATA_FLAGS = (1UL << 2),
373 #define MEMCG_DATA_FLAGS_MASK (__NR_MEMCG_DATA_FLAGS - 1)
375 static inline bool PageMemcgKmem(struct page *page);
378 * After the initialization objcg->memcg is always pointing at
379 * a valid memcg, but can be atomically swapped to the parent memcg.
381 * The caller must ensure that the returned memcg won't be released:
382 * e.g. acquire the rcu_read_lock or css_set_lock.
384 static inline struct mem_cgroup *obj_cgroup_memcg(struct obj_cgroup *objcg)
386 return READ_ONCE(objcg->memcg);
390 * __page_memcg - get the memory cgroup associated with a non-kmem page
391 * @page: a pointer to the page struct
393 * Returns a pointer to the memory cgroup associated with the page,
394 * or NULL. This function assumes that the page is known to have a
395 * proper memory cgroup pointer. It's not safe to call this function
396 * against some type of pages, e.g. slab pages or ex-slab pages or
399 static inline struct mem_cgroup *__page_memcg(struct page *page)
401 unsigned long memcg_data = page->memcg_data;
403 VM_BUG_ON_PAGE(PageSlab(page), page);
404 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_OBJCGS, page);
405 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, page);
407 return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
411 * __page_objcg - get the object cgroup associated with a kmem page
412 * @page: a pointer to the page struct
414 * Returns a pointer to the object cgroup associated with the page,
415 * or NULL. This function assumes that the page is known to have a
416 * proper object cgroup pointer. It's not safe to call this function
417 * against some type of pages, e.g. slab pages or ex-slab pages or
420 static inline struct obj_cgroup *__page_objcg(struct page *page)
422 unsigned long memcg_data = page->memcg_data;
424 VM_BUG_ON_PAGE(PageSlab(page), page);
425 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_OBJCGS, page);
426 VM_BUG_ON_PAGE(!(memcg_data & MEMCG_DATA_KMEM), page);
428 return (struct obj_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
432 * page_memcg - get the memory cgroup associated with a page
433 * @page: a pointer to the page struct
435 * Returns a pointer to the memory cgroup associated with the page,
436 * or NULL. This function assumes that the page is known to have a
437 * proper memory cgroup pointer. It's not safe to call this function
438 * against some type of pages, e.g. slab pages or ex-slab pages.
440 * For a non-kmem page any of the following ensures page and memcg binding
445 * - lock_page_memcg()
446 * - exclusive reference
448 * For a kmem page a caller should hold an rcu read lock to protect memcg
449 * associated with a kmem page from being released.
451 static inline struct mem_cgroup *page_memcg(struct page *page)
453 if (PageMemcgKmem(page))
454 return obj_cgroup_memcg(__page_objcg(page));
456 return __page_memcg(page);
460 * page_memcg_rcu - locklessly get the memory cgroup associated with a page
461 * @page: a pointer to the page struct
463 * Returns a pointer to the memory cgroup associated with the page,
464 * or NULL. This function assumes that the page is known to have a
465 * proper memory cgroup pointer. It's not safe to call this function
466 * against some type of pages, e.g. slab pages or ex-slab pages.
468 static inline struct mem_cgroup *page_memcg_rcu(struct page *page)
470 unsigned long memcg_data = READ_ONCE(page->memcg_data);
472 VM_BUG_ON_PAGE(PageSlab(page), page);
473 WARN_ON_ONCE(!rcu_read_lock_held());
475 if (memcg_data & MEMCG_DATA_KMEM) {
476 struct obj_cgroup *objcg;
478 objcg = (void *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
479 return obj_cgroup_memcg(objcg);
482 return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
486 * page_memcg_check - get the memory cgroup associated with a page
487 * @page: a pointer to the page struct
489 * Returns a pointer to the memory cgroup associated with the page,
490 * or NULL. This function unlike page_memcg() can take any page
491 * as an argument. It has to be used in cases when it's not known if a page
492 * has an associated memory cgroup pointer or an object cgroups vector or
495 * For a non-kmem page any of the following ensures page and memcg binding
500 * - lock_page_memcg()
501 * - exclusive reference
503 * For a kmem page a caller should hold an rcu read lock to protect memcg
504 * associated with a kmem page from being released.
506 static inline struct mem_cgroup *page_memcg_check(struct page *page)
509 * Because page->memcg_data might be changed asynchronously
510 * for slab pages, READ_ONCE() should be used here.
512 unsigned long memcg_data = READ_ONCE(page->memcg_data);
514 if (memcg_data & MEMCG_DATA_OBJCGS)
517 if (memcg_data & MEMCG_DATA_KMEM) {
518 struct obj_cgroup *objcg;
520 objcg = (void *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
521 return obj_cgroup_memcg(objcg);
524 return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
527 #ifdef CONFIG_MEMCG_KMEM
529 * PageMemcgKmem - check if the page has MemcgKmem flag set
530 * @page: a pointer to the page struct
532 * Checks if the page has MemcgKmem flag set. The caller must ensure that
533 * the page has an associated memory cgroup. It's not safe to call this function
534 * against some types of pages, e.g. slab pages.
536 static inline bool PageMemcgKmem(struct page *page)
538 VM_BUG_ON_PAGE(page->memcg_data & MEMCG_DATA_OBJCGS, page);
539 return page->memcg_data & MEMCG_DATA_KMEM;
543 * page_objcgs - get the object cgroups vector associated with a page
544 * @page: a pointer to the page struct
546 * Returns a pointer to the object cgroups vector associated with the page,
547 * or NULL. This function assumes that the page is known to have an
548 * associated object cgroups vector. It's not safe to call this function
549 * against pages, which might have an associated memory cgroup: e.g.
550 * kernel stack pages.
552 static inline struct obj_cgroup **page_objcgs(struct page *page)
554 unsigned long memcg_data = READ_ONCE(page->memcg_data);
556 VM_BUG_ON_PAGE(memcg_data && !(memcg_data & MEMCG_DATA_OBJCGS), page);
557 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, page);
559 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
563 * page_objcgs_check - get the object cgroups vector associated with a page
564 * @page: a pointer to the page struct
566 * Returns a pointer to the object cgroups vector associated with the page,
567 * or NULL. This function is safe to use if the page can be directly associated
568 * with a memory cgroup.
570 static inline struct obj_cgroup **page_objcgs_check(struct page *page)
572 unsigned long memcg_data = READ_ONCE(page->memcg_data);
574 if (!memcg_data || !(memcg_data & MEMCG_DATA_OBJCGS))
577 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, page);
579 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
583 static inline bool PageMemcgKmem(struct page *page)
588 static inline struct obj_cgroup **page_objcgs(struct page *page)
593 static inline struct obj_cgroup **page_objcgs_check(struct page *page)
599 static __always_inline bool memcg_stat_item_in_bytes(int idx)
601 if (idx == MEMCG_PERCPU_B)
603 return vmstat_item_in_bytes(idx);
606 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
608 return (memcg == root_mem_cgroup);
611 static inline bool mem_cgroup_disabled(void)
613 return !cgroup_subsys_enabled(memory_cgrp_subsys);
616 static inline unsigned long mem_cgroup_protection(struct mem_cgroup *root,
617 struct mem_cgroup *memcg,
620 if (mem_cgroup_disabled())
624 * There is no reclaim protection applied to a targeted reclaim.
625 * We are special casing this specific case here because
626 * mem_cgroup_protected calculation is not robust enough to keep
627 * the protection invariant for calculated effective values for
628 * parallel reclaimers with different reclaim target. This is
629 * especially a problem for tail memcgs (as they have pages on LRU)
630 * which would want to have effective values 0 for targeted reclaim
631 * but a different value for external reclaim.
634 * Let's have global and A's reclaim in parallel:
636 * A (low=2G, usage = 3G, max = 3G, children_low_usage = 1.5G)
638 * | C (low = 1G, usage = 2.5G)
639 * B (low = 1G, usage = 0.5G)
641 * For the global reclaim
643 * B.elow = min(B.usage, B.low) because children_low_usage <= A.elow
644 * C.elow = min(C.usage, C.low)
646 * With the effective values resetting we have A reclaim
651 * If the global reclaim races with A's reclaim then
652 * B.elow = C.elow = 0 because children_low_usage > A.elow)
653 * is possible and reclaiming B would be violating the protection.
660 return READ_ONCE(memcg->memory.emin);
662 return max(READ_ONCE(memcg->memory.emin),
663 READ_ONCE(memcg->memory.elow));
666 void mem_cgroup_calculate_protection(struct mem_cgroup *root,
667 struct mem_cgroup *memcg);
669 static inline bool mem_cgroup_supports_protection(struct mem_cgroup *memcg)
672 * The root memcg doesn't account charges, and doesn't support
675 return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg);
679 static inline bool mem_cgroup_below_low(struct mem_cgroup *memcg)
681 if (!mem_cgroup_supports_protection(memcg))
684 return READ_ONCE(memcg->memory.elow) >=
685 page_counter_read(&memcg->memory);
688 static inline bool mem_cgroup_below_min(struct mem_cgroup *memcg)
690 if (!mem_cgroup_supports_protection(memcg))
693 return READ_ONCE(memcg->memory.emin) >=
694 page_counter_read(&memcg->memory);
697 int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask);
698 int mem_cgroup_swapin_charge_page(struct page *page, struct mm_struct *mm,
699 gfp_t gfp, swp_entry_t entry);
700 void mem_cgroup_swapin_uncharge_swap(swp_entry_t entry);
702 void mem_cgroup_uncharge(struct page *page);
703 void mem_cgroup_uncharge_list(struct list_head *page_list);
705 void mem_cgroup_migrate(struct page *oldpage, struct page *newpage);
708 * mem_cgroup_lruvec - get the lru list vector for a memcg & node
709 * @memcg: memcg of the wanted lruvec
710 * @pgdat: pglist_data
712 * Returns the lru list vector holding pages for a given @memcg &
713 * @pgdat combination. This can be the node lruvec, if the memory
714 * controller is disabled.
716 static inline struct lruvec *mem_cgroup_lruvec(struct mem_cgroup *memcg,
717 struct pglist_data *pgdat)
719 struct mem_cgroup_per_node *mz;
720 struct lruvec *lruvec;
722 if (mem_cgroup_disabled()) {
723 lruvec = &pgdat->__lruvec;
728 memcg = root_mem_cgroup;
730 mz = memcg->nodeinfo[pgdat->node_id];
731 lruvec = &mz->lruvec;
734 * Since a node can be onlined after the mem_cgroup was created,
735 * we have to be prepared to initialize lruvec->pgdat here;
736 * and if offlined then reonlined, we need to reinitialize it.
738 if (unlikely(lruvec->pgdat != pgdat))
739 lruvec->pgdat = pgdat;
744 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
746 * @pgdat: pgdat of the page
748 * This function relies on page->mem_cgroup being stable.
750 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
751 struct pglist_data *pgdat)
753 struct mem_cgroup *memcg = page_memcg(page);
755 VM_WARN_ON_ONCE_PAGE(!memcg && !mem_cgroup_disabled(), page);
756 return mem_cgroup_lruvec(memcg, pgdat);
759 static inline bool lruvec_holds_page_lru_lock(struct page *page,
760 struct lruvec *lruvec)
762 pg_data_t *pgdat = page_pgdat(page);
763 const struct mem_cgroup *memcg;
764 struct mem_cgroup_per_node *mz;
766 if (mem_cgroup_disabled())
767 return lruvec == &pgdat->__lruvec;
769 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
770 memcg = page_memcg(page) ? : root_mem_cgroup;
772 return lruvec->pgdat == pgdat && mz->memcg == memcg;
775 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
777 struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm);
779 struct lruvec *lock_page_lruvec(struct page *page);
780 struct lruvec *lock_page_lruvec_irq(struct page *page);
781 struct lruvec *lock_page_lruvec_irqsave(struct page *page,
782 unsigned long *flags);
784 #ifdef CONFIG_DEBUG_VM
785 void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page);
787 static inline void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page)
793 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
794 return css ? container_of(css, struct mem_cgroup, css) : NULL;
797 static inline bool obj_cgroup_tryget(struct obj_cgroup *objcg)
799 return percpu_ref_tryget(&objcg->refcnt);
802 static inline void obj_cgroup_get(struct obj_cgroup *objcg)
804 percpu_ref_get(&objcg->refcnt);
807 static inline void obj_cgroup_get_many(struct obj_cgroup *objcg,
810 percpu_ref_get_many(&objcg->refcnt, nr);
813 static inline void obj_cgroup_put(struct obj_cgroup *objcg)
815 percpu_ref_put(&objcg->refcnt);
818 static inline void mem_cgroup_put(struct mem_cgroup *memcg)
821 css_put(&memcg->css);
824 #define mem_cgroup_from_counter(counter, member) \
825 container_of(counter, struct mem_cgroup, member)
827 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
829 struct mem_cgroup_reclaim_cookie *);
830 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
831 int mem_cgroup_scan_tasks(struct mem_cgroup *,
832 int (*)(struct task_struct *, void *), void *);
834 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
836 if (mem_cgroup_disabled())
841 struct mem_cgroup *mem_cgroup_from_id(unsigned short id);
843 static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m)
845 return mem_cgroup_from_css(seq_css(m));
848 static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec)
850 struct mem_cgroup_per_node *mz;
852 if (mem_cgroup_disabled())
855 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
860 * parent_mem_cgroup - find the accounting parent of a memcg
861 * @memcg: memcg whose parent to find
863 * Returns the parent memcg, or NULL if this is the root or the memory
864 * controller is in legacy no-hierarchy mode.
866 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
868 if (!memcg->memory.parent)
870 return mem_cgroup_from_counter(memcg->memory.parent, memory);
873 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
874 struct mem_cgroup *root)
878 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
881 static inline bool mm_match_cgroup(struct mm_struct *mm,
882 struct mem_cgroup *memcg)
884 struct mem_cgroup *task_memcg;
888 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
890 match = mem_cgroup_is_descendant(task_memcg, memcg);
895 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
896 ino_t page_cgroup_ino(struct page *page);
898 static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
900 if (mem_cgroup_disabled())
902 return !!(memcg->css.flags & CSS_ONLINE);
906 * For memory reclaim.
908 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
910 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
911 int zid, int nr_pages);
914 unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
915 enum lru_list lru, int zone_idx)
917 struct mem_cgroup_per_node *mz;
919 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
920 return READ_ONCE(mz->lru_zone_size[zone_idx][lru]);
923 void mem_cgroup_handle_over_high(void);
925 unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg);
927 unsigned long mem_cgroup_size(struct mem_cgroup *memcg);
929 void mem_cgroup_print_oom_context(struct mem_cgroup *memcg,
930 struct task_struct *p);
932 void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg);
934 static inline void mem_cgroup_enter_user_fault(void)
936 WARN_ON(current->in_user_fault);
937 current->in_user_fault = 1;
940 static inline void mem_cgroup_exit_user_fault(void)
942 WARN_ON(!current->in_user_fault);
943 current->in_user_fault = 0;
946 static inline bool task_in_memcg_oom(struct task_struct *p)
948 return p->memcg_in_oom;
951 bool mem_cgroup_oom_synchronize(bool wait);
952 struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim,
953 struct mem_cgroup *oom_domain);
954 void mem_cgroup_print_oom_group(struct mem_cgroup *memcg);
956 #ifdef CONFIG_MEMCG_SWAP
957 extern bool cgroup_memory_noswap;
960 void lock_page_memcg(struct page *page);
961 void unlock_page_memcg(struct page *page);
963 void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val);
965 /* idx can be of type enum memcg_stat_item or node_stat_item */
966 static inline void mod_memcg_state(struct mem_cgroup *memcg,
971 local_irq_save(flags);
972 __mod_memcg_state(memcg, idx, val);
973 local_irq_restore(flags);
976 static inline unsigned long lruvec_page_state(struct lruvec *lruvec,
977 enum node_stat_item idx)
979 struct mem_cgroup_per_node *pn;
982 if (mem_cgroup_disabled())
983 return node_page_state(lruvec_pgdat(lruvec), idx);
985 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
986 x = atomic_long_read(&pn->lruvec_stat[idx]);
994 static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec,
995 enum node_stat_item idx)
997 struct mem_cgroup_per_node *pn;
1001 if (mem_cgroup_disabled())
1002 return node_page_state(lruvec_pgdat(lruvec), idx);
1004 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
1005 for_each_possible_cpu(cpu)
1006 x += per_cpu(pn->lruvec_stat_local->count[idx], cpu);
1014 void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx,
1016 void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, int val);
1018 static inline void mod_lruvec_kmem_state(void *p, enum node_stat_item idx,
1021 unsigned long flags;
1023 local_irq_save(flags);
1024 __mod_lruvec_kmem_state(p, idx, val);
1025 local_irq_restore(flags);
1028 static inline void mod_memcg_lruvec_state(struct lruvec *lruvec,
1029 enum node_stat_item idx, int val)
1031 unsigned long flags;
1033 local_irq_save(flags);
1034 __mod_memcg_lruvec_state(lruvec, idx, val);
1035 local_irq_restore(flags);
1038 void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx,
1039 unsigned long count);
1041 static inline void count_memcg_events(struct mem_cgroup *memcg,
1042 enum vm_event_item idx,
1043 unsigned long count)
1045 unsigned long flags;
1047 local_irq_save(flags);
1048 __count_memcg_events(memcg, idx, count);
1049 local_irq_restore(flags);
1052 static inline void count_memcg_page_event(struct page *page,
1053 enum vm_event_item idx)
1055 struct mem_cgroup *memcg = page_memcg(page);
1058 count_memcg_events(memcg, idx, 1);
1061 static inline void count_memcg_event_mm(struct mm_struct *mm,
1062 enum vm_event_item idx)
1064 struct mem_cgroup *memcg;
1066 if (mem_cgroup_disabled())
1070 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1072 count_memcg_events(memcg, idx, 1);
1076 static inline void memcg_memory_event(struct mem_cgroup *memcg,
1077 enum memcg_memory_event event)
1079 bool swap_event = event == MEMCG_SWAP_HIGH || event == MEMCG_SWAP_MAX ||
1080 event == MEMCG_SWAP_FAIL;
1082 atomic_long_inc(&memcg->memory_events_local[event]);
1084 cgroup_file_notify(&memcg->events_local_file);
1087 atomic_long_inc(&memcg->memory_events[event]);
1089 cgroup_file_notify(&memcg->swap_events_file);
1091 cgroup_file_notify(&memcg->events_file);
1093 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
1095 if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
1097 } while ((memcg = parent_mem_cgroup(memcg)) &&
1098 !mem_cgroup_is_root(memcg));
1101 static inline void memcg_memory_event_mm(struct mm_struct *mm,
1102 enum memcg_memory_event event)
1104 struct mem_cgroup *memcg;
1106 if (mem_cgroup_disabled())
1110 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1112 memcg_memory_event(memcg, event);
1116 void split_page_memcg(struct page *head, unsigned int nr);
1118 unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
1120 unsigned long *total_scanned);
1122 #else /* CONFIG_MEMCG */
1124 #define MEM_CGROUP_ID_SHIFT 0
1125 #define MEM_CGROUP_ID_MAX 0
1127 static inline struct mem_cgroup *page_memcg(struct page *page)
1132 static inline struct mem_cgroup *page_memcg_rcu(struct page *page)
1134 WARN_ON_ONCE(!rcu_read_lock_held());
1138 static inline struct mem_cgroup *page_memcg_check(struct page *page)
1143 static inline bool PageMemcgKmem(struct page *page)
1148 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
1153 static inline bool mem_cgroup_disabled(void)
1158 static inline void memcg_memory_event(struct mem_cgroup *memcg,
1159 enum memcg_memory_event event)
1163 static inline void memcg_memory_event_mm(struct mm_struct *mm,
1164 enum memcg_memory_event event)
1168 static inline unsigned long mem_cgroup_protection(struct mem_cgroup *root,
1169 struct mem_cgroup *memcg,
1170 bool in_low_reclaim)
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 page *page, struct mm_struct *mm,
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 page *page)
1210 static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
1214 static inline void mem_cgroup_migrate(struct page *old, struct page *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 *mem_cgroup_page_lruvec(struct page *page,
1225 struct pglist_data *pgdat)
1227 return &pgdat->__lruvec;
1230 static inline bool lruvec_holds_page_lru_lock(struct page *page,
1231 struct lruvec *lruvec)
1233 pg_data_t *pgdat = page_pgdat(page);
1235 return lruvec == &pgdat->__lruvec;
1238 static inline void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page)
1242 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
1247 static inline bool mm_match_cgroup(struct mm_struct *mm,
1248 struct mem_cgroup *memcg)
1253 static inline struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
1258 static inline void mem_cgroup_put(struct mem_cgroup *memcg)
1262 static inline struct lruvec *lock_page_lruvec(struct page *page)
1264 struct pglist_data *pgdat = page_pgdat(page);
1266 spin_lock(&pgdat->__lruvec.lru_lock);
1267 return &pgdat->__lruvec;
1270 static inline struct lruvec *lock_page_lruvec_irq(struct page *page)
1272 struct pglist_data *pgdat = page_pgdat(page);
1274 spin_lock_irq(&pgdat->__lruvec.lru_lock);
1275 return &pgdat->__lruvec;
1278 static inline struct lruvec *lock_page_lruvec_irqsave(struct page *page,
1279 unsigned long *flagsp)
1281 struct pglist_data *pgdat = page_pgdat(page);
1283 spin_lock_irqsave(&pgdat->__lruvec.lru_lock, *flagsp);
1284 return &pgdat->__lruvec;
1287 static inline struct mem_cgroup *
1288 mem_cgroup_iter(struct mem_cgroup *root,
1289 struct mem_cgroup *prev,
1290 struct mem_cgroup_reclaim_cookie *reclaim)
1295 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
1296 struct mem_cgroup *prev)
1300 static inline int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
1301 int (*fn)(struct task_struct *, void *), void *arg)
1306 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
1311 static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
1314 /* XXX: This should always return root_mem_cgroup */
1318 static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m)
1323 static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec)
1328 static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
1334 unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
1335 enum lru_list lru, int zone_idx)
1340 static inline unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg)
1345 static inline unsigned long mem_cgroup_size(struct mem_cgroup *memcg)
1351 mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p)
1356 mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg)
1360 static inline void lock_page_memcg(struct page *page)
1364 static inline void unlock_page_memcg(struct page *page)
1368 static inline void mem_cgroup_handle_over_high(void)
1372 static inline void mem_cgroup_enter_user_fault(void)
1376 static inline void mem_cgroup_exit_user_fault(void)
1380 static inline bool task_in_memcg_oom(struct task_struct *p)
1385 static inline bool mem_cgroup_oom_synchronize(bool wait)
1390 static inline struct mem_cgroup *mem_cgroup_get_oom_group(
1391 struct task_struct *victim, struct mem_cgroup *oom_domain)
1396 static inline void mem_cgroup_print_oom_group(struct mem_cgroup *memcg)
1400 static inline void __mod_memcg_state(struct mem_cgroup *memcg,
1406 static inline void mod_memcg_state(struct mem_cgroup *memcg,
1412 static inline unsigned long lruvec_page_state(struct lruvec *lruvec,
1413 enum node_stat_item idx)
1415 return node_page_state(lruvec_pgdat(lruvec), idx);
1418 static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec,
1419 enum node_stat_item idx)
1421 return node_page_state(lruvec_pgdat(lruvec), idx);
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 /* Don't lock again iff page's lruvec locked */
1520 static inline struct lruvec *relock_page_lruvec_irq(struct page *page,
1521 struct lruvec *locked_lruvec)
1523 if (locked_lruvec) {
1524 if (lruvec_holds_page_lru_lock(page, locked_lruvec))
1525 return locked_lruvec;
1527 unlock_page_lruvec_irq(locked_lruvec);
1530 return lock_page_lruvec_irq(page);
1533 /* Don't lock again iff page's lruvec locked */
1534 static inline struct lruvec *relock_page_lruvec_irqsave(struct page *page,
1535 struct lruvec *locked_lruvec, unsigned long *flags)
1537 if (locked_lruvec) {
1538 if (lruvec_holds_page_lru_lock(page, locked_lruvec))
1539 return locked_lruvec;
1541 unlock_page_lruvec_irqrestore(locked_lruvec, *flags);
1544 return lock_page_lruvec_irqsave(page, flags);
1547 #ifdef CONFIG_CGROUP_WRITEBACK
1549 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
1550 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
1551 unsigned long *pheadroom, unsigned long *pdirty,
1552 unsigned long *pwriteback);
1554 void mem_cgroup_track_foreign_dirty_slowpath(struct page *page,
1555 struct bdi_writeback *wb);
1557 static inline void mem_cgroup_track_foreign_dirty(struct page *page,
1558 struct bdi_writeback *wb)
1560 if (mem_cgroup_disabled())
1563 if (unlikely(&page_memcg(page)->css != wb->memcg_css))
1564 mem_cgroup_track_foreign_dirty_slowpath(page, wb);
1567 void mem_cgroup_flush_foreign(struct bdi_writeback *wb);
1569 #else /* CONFIG_CGROUP_WRITEBACK */
1571 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
1576 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
1577 unsigned long *pfilepages,
1578 unsigned long *pheadroom,
1579 unsigned long *pdirty,
1580 unsigned long *pwriteback)
1584 static inline void mem_cgroup_track_foreign_dirty(struct page *page,
1585 struct bdi_writeback *wb)
1589 static inline void mem_cgroup_flush_foreign(struct bdi_writeback *wb)
1593 #endif /* CONFIG_CGROUP_WRITEBACK */
1596 bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
1597 void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
1599 extern struct static_key_false memcg_sockets_enabled_key;
1600 #define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_enabled_key)
1601 void mem_cgroup_sk_alloc(struct sock *sk);
1602 void mem_cgroup_sk_free(struct sock *sk);
1603 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
1605 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_pressure)
1608 if (time_before(jiffies, memcg->socket_pressure))
1610 } while ((memcg = parent_mem_cgroup(memcg)));
1614 int alloc_shrinker_info(struct mem_cgroup *memcg);
1615 void free_shrinker_info(struct mem_cgroup *memcg);
1616 void set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id);
1617 void reparent_shrinker_deferred(struct mem_cgroup *memcg);
1619 #define mem_cgroup_sockets_enabled 0
1620 static inline void mem_cgroup_sk_alloc(struct sock *sk) { };
1621 static inline void mem_cgroup_sk_free(struct sock *sk) { };
1622 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
1627 static inline void set_shrinker_bit(struct mem_cgroup *memcg,
1628 int nid, int shrinker_id)
1633 #ifdef CONFIG_MEMCG_KMEM
1634 int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order);
1635 void __memcg_kmem_uncharge_page(struct page *page, int order);
1637 struct obj_cgroup *get_obj_cgroup_from_current(void);
1639 int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size);
1640 void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size);
1642 extern struct static_key_false memcg_kmem_enabled_key;
1644 extern int memcg_nr_cache_ids;
1645 void memcg_get_cache_ids(void);
1646 void memcg_put_cache_ids(void);
1649 * Helper macro to loop through all memcg-specific caches. Callers must still
1650 * check if the cache is valid (it is either valid or NULL).
1651 * the slab_mutex must be held when looping through those caches
1653 #define for_each_memcg_cache_index(_idx) \
1654 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
1656 static inline bool memcg_kmem_enabled(void)
1658 return static_branch_likely(&memcg_kmem_enabled_key);
1661 static inline int memcg_kmem_charge_page(struct page *page, gfp_t gfp,
1664 if (memcg_kmem_enabled())
1665 return __memcg_kmem_charge_page(page, gfp, order);
1669 static inline void memcg_kmem_uncharge_page(struct page *page, int order)
1671 if (memcg_kmem_enabled())
1672 __memcg_kmem_uncharge_page(page, order);
1676 * A helper for accessing memcg's kmem_id, used for getting
1677 * corresponding LRU lists.
1679 static inline int memcg_cache_id(struct mem_cgroup *memcg)
1681 return memcg ? memcg->kmemcg_id : -1;
1684 struct mem_cgroup *mem_cgroup_from_obj(void *p);
1688 static inline int memcg_kmem_charge_page(struct page *page, gfp_t gfp,
1694 static inline void memcg_kmem_uncharge_page(struct page *page, int order)
1698 static inline int __memcg_kmem_charge_page(struct page *page, gfp_t gfp,
1704 static inline void __memcg_kmem_uncharge_page(struct page *page, int order)
1708 #define for_each_memcg_cache_index(_idx) \
1711 static inline bool memcg_kmem_enabled(void)
1716 static inline int memcg_cache_id(struct mem_cgroup *memcg)
1721 static inline void memcg_get_cache_ids(void)
1725 static inline void memcg_put_cache_ids(void)
1729 static inline struct mem_cgroup *mem_cgroup_from_obj(void *p)
1734 #endif /* CONFIG_MEMCG_KMEM */
1736 #endif /* _LINUX_MEMCONTROL_H */