1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /* memcontrol.h - Memory Controller
4 * Copyright IBM Corporation, 2007
5 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
7 * Copyright 2007 OpenVZ SWsoft Inc
8 * Author: Pavel Emelianov <xemul@openvz.org>
11 #ifndef _LINUX_MEMCONTROL_H
12 #define _LINUX_MEMCONTROL_H
13 #include <linux/cgroup.h>
14 #include <linux/vm_event_item.h>
15 #include <linux/hardirq.h>
16 #include <linux/jump_label.h>
17 #include <linux/page_counter.h>
18 #include <linux/vmpressure.h>
19 #include <linux/eventfd.h>
21 #include <linux/vmstat.h>
22 #include <linux/writeback.h>
23 #include <linux/page-flags.h>
31 /* Cgroup-specific page state, on top of universal node page state */
32 enum memcg_stat_item {
33 MEMCG_SWAP = NR_VM_NODE_STAT_ITEMS,
39 enum memcg_memory_event {
48 MEMCG_NR_MEMORY_EVENTS,
51 struct mem_cgroup_reclaim_cookie {
53 unsigned int generation;
58 #define MEM_CGROUP_ID_SHIFT 16
59 #define MEM_CGROUP_ID_MAX USHRT_MAX
61 struct mem_cgroup_id {
67 * Per memcg event counter is incremented at every pagein/pageout. With THP,
68 * it will be incremented by the number of pages. This counter is used
69 * to trigger some periodic events. This is straightforward and better
70 * than using jiffies etc. to handle periodic memcg event.
72 enum mem_cgroup_events_target {
73 MEM_CGROUP_TARGET_THRESH,
74 MEM_CGROUP_TARGET_SOFTLIMIT,
78 struct memcg_vmstats_percpu {
79 /* Local (CPU and cgroup) page state & events */
80 long state[MEMCG_NR_STAT];
81 unsigned long events[NR_VM_EVENT_ITEMS];
83 /* Delta calculation for lockless upward propagation */
84 long state_prev[MEMCG_NR_STAT];
85 unsigned long events_prev[NR_VM_EVENT_ITEMS];
87 /* Cgroup1: threshold notifications & softlimit tree updates */
88 unsigned long nr_page_events;
89 unsigned long targets[MEM_CGROUP_NTARGETS];
92 struct memcg_vmstats {
93 /* Aggregated (CPU and subtree) page state & events */
94 long state[MEMCG_NR_STAT];
95 unsigned long events[NR_VM_EVENT_ITEMS];
97 /* Pending child counts during tree propagation */
98 long state_pending[MEMCG_NR_STAT];
99 unsigned long events_pending[NR_VM_EVENT_ITEMS];
102 struct mem_cgroup_reclaim_iter {
103 struct mem_cgroup *position;
104 /* scan generation, increased every round-trip */
105 unsigned int generation;
109 * Bitmap and deferred work of shrinker::id corresponding to memcg-aware
110 * shrinkers, which have elements charged to this memcg.
112 struct shrinker_info {
114 atomic_long_t *nr_deferred;
118 struct lruvec_stats_percpu {
119 /* Local (CPU and cgroup) state */
120 long state[NR_VM_NODE_STAT_ITEMS];
122 /* Delta calculation for lockless upward propagation */
123 long state_prev[NR_VM_NODE_STAT_ITEMS];
126 struct lruvec_stats {
127 /* Aggregated (CPU and subtree) state */
128 long state[NR_VM_NODE_STAT_ITEMS];
130 /* Pending child counts during tree propagation */
131 long state_pending[NR_VM_NODE_STAT_ITEMS];
135 * per-node information in memory controller.
137 struct mem_cgroup_per_node {
138 struct lruvec lruvec;
140 struct lruvec_stats_percpu __percpu *lruvec_stats_percpu;
141 struct lruvec_stats lruvec_stats;
143 unsigned long lru_zone_size[MAX_NR_ZONES][NR_LRU_LISTS];
145 struct mem_cgroup_reclaim_iter iter;
147 struct shrinker_info __rcu *shrinker_info;
149 struct rb_node tree_node; /* RB tree node */
150 unsigned long usage_in_excess;/* Set to the value by which */
151 /* the soft limit is exceeded*/
153 struct mem_cgroup *memcg; /* Back pointer, we cannot */
154 /* use container_of */
157 struct mem_cgroup_threshold {
158 struct eventfd_ctx *eventfd;
159 unsigned long threshold;
163 struct mem_cgroup_threshold_ary {
164 /* An array index points to threshold just below or equal to usage. */
165 int current_threshold;
166 /* Size of entries[] */
168 /* Array of thresholds */
169 struct mem_cgroup_threshold entries[];
172 struct mem_cgroup_thresholds {
173 /* Primary thresholds array */
174 struct mem_cgroup_threshold_ary *primary;
176 * Spare threshold array.
177 * This is needed to make mem_cgroup_unregister_event() "never fail".
178 * It must be able to store at least primary->size - 1 entries.
180 struct mem_cgroup_threshold_ary *spare;
183 enum memcg_kmem_state {
189 #if defined(CONFIG_SMP)
190 struct memcg_padding {
192 } ____cacheline_internodealigned_in_smp;
193 #define MEMCG_PADDING(name) struct memcg_padding name
195 #define MEMCG_PADDING(name)
199 * Remember four most recent foreign writebacks with dirty pages in this
200 * cgroup. Inode sharing is expected to be uncommon and, even if we miss
201 * one in a given round, we're likely to catch it later if it keeps
202 * foreign-dirtying, so a fairly low count should be enough.
204 * See mem_cgroup_track_foreign_dirty_slowpath() for details.
206 #define MEMCG_CGWB_FRN_CNT 4
208 struct memcg_cgwb_frn {
209 u64 bdi_id; /* bdi->id of the foreign inode */
210 int memcg_id; /* memcg->css.id of foreign inode */
211 u64 at; /* jiffies_64 at the time of dirtying */
212 struct wb_completion done; /* tracks in-flight foreign writebacks */
216 * Bucket for arbitrarily byte-sized objects charged to a memory
217 * cgroup. The bucket can be reparented in one piece when the cgroup
218 * is destroyed, without having to round up the individual references
219 * of all live memory objects in the wild.
222 struct percpu_ref refcnt;
223 struct mem_cgroup *memcg;
224 atomic_t nr_charged_bytes;
226 struct list_head list; /* protected by objcg_lock */
232 * The memory controller data structure. The memory controller controls both
233 * page cache and RSS per cgroup. We would eventually like to provide
234 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
235 * to help the administrator determine what knobs to tune.
238 struct cgroup_subsys_state css;
240 /* Private memcg ID. Used to ID objects that outlive the cgroup */
241 struct mem_cgroup_id id;
243 /* Accounted resources */
244 struct page_counter memory; /* Both v1 & v2 */
247 struct page_counter swap; /* v2 only */
248 struct page_counter memsw; /* v1 only */
251 /* Legacy consumer-oriented counters */
252 struct page_counter kmem; /* v1 only */
253 struct page_counter tcpmem; /* v1 only */
255 /* Range enforcement for interrupt charges */
256 struct work_struct high_work;
258 unsigned long soft_limit;
260 /* vmpressure notifications */
261 struct vmpressure vmpressure;
264 * Should the OOM killer kill all belonging tasks, had it kill one?
268 /* protected by memcg_oom_lock */
273 /* OOM-Killer disable */
274 int oom_kill_disable;
276 /* memory.events and memory.events.local */
277 struct cgroup_file events_file;
278 struct cgroup_file events_local_file;
280 /* handle for "memory.swap.events" */
281 struct cgroup_file swap_events_file;
283 /* protect arrays of thresholds */
284 struct mutex thresholds_lock;
286 /* thresholds for memory usage. RCU-protected */
287 struct mem_cgroup_thresholds thresholds;
289 /* thresholds for mem+swap usage. RCU-protected */
290 struct mem_cgroup_thresholds memsw_thresholds;
292 /* For oom notifier event fd */
293 struct list_head oom_notify;
296 * Should we move charges of a task when a task is moved into this
297 * mem_cgroup ? And what type of charges should we move ?
299 unsigned long move_charge_at_immigrate;
300 /* taken only while moving_account > 0 */
301 spinlock_t move_lock;
302 unsigned long move_lock_flags;
304 MEMCG_PADDING(_pad1_);
307 struct memcg_vmstats vmstats;
310 atomic_long_t memory_events[MEMCG_NR_MEMORY_EVENTS];
311 atomic_long_t memory_events_local[MEMCG_NR_MEMORY_EVENTS];
313 unsigned long socket_pressure;
315 /* Legacy tcp memory accounting */
319 #ifdef CONFIG_MEMCG_KMEM
321 enum memcg_kmem_state kmem_state;
322 struct obj_cgroup __rcu *objcg;
323 /* list of inherited objcgs, protected by objcg_lock */
324 struct list_head objcg_list;
327 MEMCG_PADDING(_pad2_);
330 * set > 0 if pages under this cgroup are moving to other cgroup.
332 atomic_t moving_account;
333 struct task_struct *move_lock_task;
335 struct memcg_vmstats_percpu __percpu *vmstats_percpu;
337 #ifdef CONFIG_CGROUP_WRITEBACK
338 struct list_head cgwb_list;
339 struct wb_domain cgwb_domain;
340 struct memcg_cgwb_frn cgwb_frn[MEMCG_CGWB_FRN_CNT];
343 /* List of events which userspace want to receive */
344 struct list_head event_list;
345 spinlock_t event_list_lock;
347 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
348 struct deferred_split deferred_split_queue;
351 struct mem_cgroup_per_node *nodeinfo[];
355 * size of first charge trial. "32" comes from vmscan.c's magic value.
356 * TODO: maybe necessary to use big numbers in big irons.
358 #define MEMCG_CHARGE_BATCH 32U
360 extern struct mem_cgroup *root_mem_cgroup;
362 enum page_memcg_data_flags {
363 /* page->memcg_data is a pointer to an objcgs vector */
364 MEMCG_DATA_OBJCGS = (1UL << 0),
365 /* page has been accounted as a non-slab kernel page */
366 MEMCG_DATA_KMEM = (1UL << 1),
367 /* the next bit after the last actual flag */
368 __NR_MEMCG_DATA_FLAGS = (1UL << 2),
371 #define MEMCG_DATA_FLAGS_MASK (__NR_MEMCG_DATA_FLAGS - 1)
373 static inline bool PageMemcgKmem(struct page *page);
376 * After the initialization objcg->memcg is always pointing at
377 * a valid memcg, but can be atomically swapped to the parent memcg.
379 * The caller must ensure that the returned memcg won't be released:
380 * e.g. acquire the rcu_read_lock or css_set_lock.
382 static inline struct mem_cgroup *obj_cgroup_memcg(struct obj_cgroup *objcg)
384 return READ_ONCE(objcg->memcg);
388 * __page_memcg - get the memory cgroup associated with a non-kmem page
389 * @page: a pointer to the page struct
391 * Returns a pointer to the memory cgroup associated with the page,
392 * or NULL. This function assumes that the page is known to have a
393 * proper memory cgroup pointer. It's not safe to call this function
394 * against some type of pages, e.g. slab pages or ex-slab pages or
397 static inline struct mem_cgroup *__page_memcg(struct page *page)
399 unsigned long memcg_data = page->memcg_data;
401 VM_BUG_ON_PAGE(PageSlab(page), page);
402 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_OBJCGS, page);
403 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, page);
405 return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
409 * __page_objcg - get the object cgroup associated with a kmem page
410 * @page: a pointer to the page struct
412 * Returns a pointer to the object cgroup associated with the page,
413 * or NULL. This function assumes that the page is known to have a
414 * proper object cgroup pointer. It's not safe to call this function
415 * against some type of pages, e.g. slab pages or ex-slab pages or
418 static inline struct obj_cgroup *__page_objcg(struct page *page)
420 unsigned long memcg_data = page->memcg_data;
422 VM_BUG_ON_PAGE(PageSlab(page), page);
423 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_OBJCGS, page);
424 VM_BUG_ON_PAGE(!(memcg_data & MEMCG_DATA_KMEM), page);
426 return (struct obj_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
430 * page_memcg - get the memory cgroup associated with a page
431 * @page: a pointer to the page struct
433 * Returns a pointer to the memory cgroup associated with the page,
434 * or NULL. This function assumes that the page is known to have a
435 * proper memory cgroup pointer. It's not safe to call this function
436 * against some type of pages, e.g. slab pages or ex-slab pages.
438 * For a non-kmem page any of the following ensures page and memcg binding
443 * - lock_page_memcg()
444 * - exclusive reference
446 * For a kmem page a caller should hold an rcu read lock to protect memcg
447 * associated with a kmem page from being released.
449 static inline struct mem_cgroup *page_memcg(struct page *page)
451 if (PageMemcgKmem(page))
452 return obj_cgroup_memcg(__page_objcg(page));
454 return __page_memcg(page);
458 * page_memcg_rcu - locklessly get the memory cgroup associated with a page
459 * @page: a pointer to the page struct
461 * Returns a pointer to the memory cgroup associated with the page,
462 * or NULL. This function assumes that the page is known to have a
463 * proper memory cgroup pointer. It's not safe to call this function
464 * against some type of pages, e.g. slab pages or ex-slab pages.
466 static inline struct mem_cgroup *page_memcg_rcu(struct page *page)
468 unsigned long memcg_data = READ_ONCE(page->memcg_data);
470 VM_BUG_ON_PAGE(PageSlab(page), page);
471 WARN_ON_ONCE(!rcu_read_lock_held());
473 if (memcg_data & MEMCG_DATA_KMEM) {
474 struct obj_cgroup *objcg;
476 objcg = (void *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
477 return obj_cgroup_memcg(objcg);
480 return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
484 * page_memcg_check - get the memory cgroup associated with a page
485 * @page: a pointer to the page struct
487 * Returns a pointer to the memory cgroup associated with the page,
488 * or NULL. This function unlike page_memcg() can take any page
489 * as an argument. It has to be used in cases when it's not known if a page
490 * has an associated memory cgroup pointer or an object cgroups vector or
493 * For a non-kmem page any of the following ensures page and memcg binding
498 * - lock_page_memcg()
499 * - exclusive reference
501 * For a kmem page a caller should hold an rcu read lock to protect memcg
502 * associated with a kmem page from being released.
504 static inline struct mem_cgroup *page_memcg_check(struct page *page)
507 * Because page->memcg_data might be changed asynchronously
508 * for slab pages, READ_ONCE() should be used here.
510 unsigned long memcg_data = READ_ONCE(page->memcg_data);
512 if (memcg_data & MEMCG_DATA_OBJCGS)
515 if (memcg_data & MEMCG_DATA_KMEM) {
516 struct obj_cgroup *objcg;
518 objcg = (void *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
519 return obj_cgroup_memcg(objcg);
522 return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
525 #ifdef CONFIG_MEMCG_KMEM
527 * PageMemcgKmem - check if the page has MemcgKmem flag set
528 * @page: a pointer to the page struct
530 * Checks if the page has MemcgKmem flag set. The caller must ensure that
531 * the page has an associated memory cgroup. It's not safe to call this function
532 * against some types of pages, e.g. slab pages.
534 static inline bool PageMemcgKmem(struct page *page)
536 VM_BUG_ON_PAGE(page->memcg_data & MEMCG_DATA_OBJCGS, page);
537 return page->memcg_data & MEMCG_DATA_KMEM;
541 * page_objcgs - get the object cgroups vector associated with a page
542 * @page: a pointer to the page struct
544 * Returns a pointer to the object cgroups vector associated with the page,
545 * or NULL. This function assumes that the page is known to have an
546 * associated object cgroups vector. It's not safe to call this function
547 * against pages, which might have an associated memory cgroup: e.g.
548 * kernel stack pages.
550 static inline struct obj_cgroup **page_objcgs(struct page *page)
552 unsigned long memcg_data = READ_ONCE(page->memcg_data);
554 VM_BUG_ON_PAGE(memcg_data && !(memcg_data & MEMCG_DATA_OBJCGS), page);
555 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, page);
557 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
561 * page_objcgs_check - get the object cgroups vector associated with a page
562 * @page: a pointer to the page struct
564 * Returns a pointer to the object cgroups vector associated with the page,
565 * or NULL. This function is safe to use if the page can be directly associated
566 * with a memory cgroup.
568 static inline struct obj_cgroup **page_objcgs_check(struct page *page)
570 unsigned long memcg_data = READ_ONCE(page->memcg_data);
572 if (!memcg_data || !(memcg_data & MEMCG_DATA_OBJCGS))
575 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, page);
577 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
581 static inline bool PageMemcgKmem(struct page *page)
586 static inline struct obj_cgroup **page_objcgs(struct page *page)
591 static inline struct obj_cgroup **page_objcgs_check(struct page *page)
597 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
599 return (memcg == root_mem_cgroup);
602 static inline bool mem_cgroup_disabled(void)
604 return !cgroup_subsys_enabled(memory_cgrp_subsys);
607 static inline void mem_cgroup_protection(struct mem_cgroup *root,
608 struct mem_cgroup *memcg,
614 if (mem_cgroup_disabled())
618 * There is no reclaim protection applied to a targeted reclaim.
619 * We are special casing this specific case here because
620 * mem_cgroup_protected calculation is not robust enough to keep
621 * the protection invariant for calculated effective values for
622 * parallel reclaimers with different reclaim target. This is
623 * especially a problem for tail memcgs (as they have pages on LRU)
624 * which would want to have effective values 0 for targeted reclaim
625 * but a different value for external reclaim.
628 * Let's have global and A's reclaim in parallel:
630 * A (low=2G, usage = 3G, max = 3G, children_low_usage = 1.5G)
632 * | C (low = 1G, usage = 2.5G)
633 * B (low = 1G, usage = 0.5G)
635 * For the global reclaim
637 * B.elow = min(B.usage, B.low) because children_low_usage <= A.elow
638 * C.elow = min(C.usage, C.low)
640 * With the effective values resetting we have A reclaim
645 * If the global reclaim races with A's reclaim then
646 * B.elow = C.elow = 0 because children_low_usage > A.elow)
647 * is possible and reclaiming B would be violating the protection.
653 *min = READ_ONCE(memcg->memory.emin);
654 *low = READ_ONCE(memcg->memory.elow);
657 void mem_cgroup_calculate_protection(struct mem_cgroup *root,
658 struct mem_cgroup *memcg);
660 static inline bool mem_cgroup_supports_protection(struct mem_cgroup *memcg)
663 * The root memcg doesn't account charges, and doesn't support
666 return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg);
670 static inline bool mem_cgroup_below_low(struct mem_cgroup *memcg)
672 if (!mem_cgroup_supports_protection(memcg))
675 return READ_ONCE(memcg->memory.elow) >=
676 page_counter_read(&memcg->memory);
679 static inline bool mem_cgroup_below_min(struct mem_cgroup *memcg)
681 if (!mem_cgroup_supports_protection(memcg))
684 return READ_ONCE(memcg->memory.emin) >=
685 page_counter_read(&memcg->memory);
688 int __mem_cgroup_charge(struct page *page, struct mm_struct *mm,
690 static inline int mem_cgroup_charge(struct page *page, struct mm_struct *mm,
693 if (mem_cgroup_disabled())
695 return __mem_cgroup_charge(page, mm, 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 static inline void mem_cgroup_uncharge(struct page *page)
705 if (mem_cgroup_disabled())
707 __mem_cgroup_uncharge(page);
710 void __mem_cgroup_uncharge_list(struct list_head *page_list);
711 static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
713 if (mem_cgroup_disabled())
715 __mem_cgroup_uncharge_list(page_list);
718 void mem_cgroup_migrate(struct page *oldpage, struct page *newpage);
721 * mem_cgroup_lruvec - get the lru list vector for a memcg & node
722 * @memcg: memcg of the wanted lruvec
723 * @pgdat: pglist_data
725 * Returns the lru list vector holding pages for a given @memcg &
726 * @pgdat combination. This can be the node lruvec, if the memory
727 * controller is disabled.
729 static inline struct lruvec *mem_cgroup_lruvec(struct mem_cgroup *memcg,
730 struct pglist_data *pgdat)
732 struct mem_cgroup_per_node *mz;
733 struct lruvec *lruvec;
735 if (mem_cgroup_disabled()) {
736 lruvec = &pgdat->__lruvec;
741 memcg = root_mem_cgroup;
743 mz = memcg->nodeinfo[pgdat->node_id];
744 lruvec = &mz->lruvec;
747 * Since a node can be onlined after the mem_cgroup was created,
748 * we have to be prepared to initialize lruvec->pgdat here;
749 * and if offlined then reonlined, we need to reinitialize it.
751 if (unlikely(lruvec->pgdat != pgdat))
752 lruvec->pgdat = pgdat;
757 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
760 * This function relies on page->mem_cgroup being stable.
762 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page)
764 pg_data_t *pgdat = page_pgdat(page);
765 struct mem_cgroup *memcg = page_memcg(page);
767 VM_WARN_ON_ONCE_PAGE(!memcg && !mem_cgroup_disabled(), page);
768 return mem_cgroup_lruvec(memcg, pgdat);
771 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
773 struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm);
775 struct lruvec *lock_page_lruvec(struct page *page);
776 struct lruvec *lock_page_lruvec_irq(struct page *page);
777 struct lruvec *lock_page_lruvec_irqsave(struct page *page,
778 unsigned long *flags);
780 #ifdef CONFIG_DEBUG_VM
781 void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page);
783 static inline void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page)
789 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
790 return css ? container_of(css, struct mem_cgroup, css) : NULL;
793 static inline bool obj_cgroup_tryget(struct obj_cgroup *objcg)
795 return percpu_ref_tryget(&objcg->refcnt);
798 static inline void obj_cgroup_get(struct obj_cgroup *objcg)
800 percpu_ref_get(&objcg->refcnt);
803 static inline void obj_cgroup_get_many(struct obj_cgroup *objcg,
806 percpu_ref_get_many(&objcg->refcnt, nr);
809 static inline void obj_cgroup_put(struct obj_cgroup *objcg)
811 percpu_ref_put(&objcg->refcnt);
814 static inline void mem_cgroup_put(struct mem_cgroup *memcg)
817 css_put(&memcg->css);
820 #define mem_cgroup_from_counter(counter, member) \
821 container_of(counter, struct mem_cgroup, member)
823 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
825 struct mem_cgroup_reclaim_cookie *);
826 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
827 int mem_cgroup_scan_tasks(struct mem_cgroup *,
828 int (*)(struct task_struct *, void *), void *);
830 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
832 if (mem_cgroup_disabled())
837 struct mem_cgroup *mem_cgroup_from_id(unsigned short id);
839 static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m)
841 return mem_cgroup_from_css(seq_css(m));
844 static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec)
846 struct mem_cgroup_per_node *mz;
848 if (mem_cgroup_disabled())
851 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
856 * parent_mem_cgroup - find the accounting parent of a memcg
857 * @memcg: memcg whose parent to find
859 * Returns the parent memcg, or NULL if this is the root or the memory
860 * controller is in legacy no-hierarchy mode.
862 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
864 if (!memcg->memory.parent)
866 return mem_cgroup_from_counter(memcg->memory.parent, memory);
869 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
870 struct mem_cgroup *root)
874 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
877 static inline bool mm_match_cgroup(struct mm_struct *mm,
878 struct mem_cgroup *memcg)
880 struct mem_cgroup *task_memcg;
884 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
886 match = mem_cgroup_is_descendant(task_memcg, memcg);
891 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
892 ino_t page_cgroup_ino(struct page *page);
894 static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
896 if (mem_cgroup_disabled())
898 return !!(memcg->css.flags & CSS_ONLINE);
901 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
902 int zid, int nr_pages);
905 unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
906 enum lru_list lru, int zone_idx)
908 struct mem_cgroup_per_node *mz;
910 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
911 return READ_ONCE(mz->lru_zone_size[zone_idx][lru]);
914 void mem_cgroup_handle_over_high(void);
916 unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg);
918 unsigned long mem_cgroup_size(struct mem_cgroup *memcg);
920 void mem_cgroup_print_oom_context(struct mem_cgroup *memcg,
921 struct task_struct *p);
923 void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg);
925 static inline void mem_cgroup_enter_user_fault(void)
927 WARN_ON(current->in_user_fault);
928 current->in_user_fault = 1;
931 static inline void mem_cgroup_exit_user_fault(void)
933 WARN_ON(!current->in_user_fault);
934 current->in_user_fault = 0;
937 static inline bool task_in_memcg_oom(struct task_struct *p)
939 return p->memcg_in_oom;
942 bool mem_cgroup_oom_synchronize(bool wait);
943 struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim,
944 struct mem_cgroup *oom_domain);
945 void mem_cgroup_print_oom_group(struct mem_cgroup *memcg);
947 #ifdef CONFIG_MEMCG_SWAP
948 extern bool cgroup_memory_noswap;
951 void lock_page_memcg(struct page *page);
952 void unlock_page_memcg(struct page *page);
954 void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val);
956 /* idx can be of type enum memcg_stat_item or node_stat_item */
957 static inline void mod_memcg_state(struct mem_cgroup *memcg,
962 local_irq_save(flags);
963 __mod_memcg_state(memcg, idx, val);
964 local_irq_restore(flags);
967 static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx)
969 return READ_ONCE(memcg->vmstats.state[idx]);
972 static inline unsigned long lruvec_page_state(struct lruvec *lruvec,
973 enum node_stat_item idx)
975 struct mem_cgroup_per_node *pn;
977 if (mem_cgroup_disabled())
978 return node_page_state(lruvec_pgdat(lruvec), idx);
980 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
981 return READ_ONCE(pn->lruvec_stats.state[idx]);
984 static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec,
985 enum node_stat_item idx)
987 struct mem_cgroup_per_node *pn;
991 if (mem_cgroup_disabled())
992 return node_page_state(lruvec_pgdat(lruvec), idx);
994 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
995 for_each_possible_cpu(cpu)
996 x += per_cpu(pn->lruvec_stats_percpu->state[idx], cpu);
1004 void mem_cgroup_flush_stats(void);
1005 void mem_cgroup_flush_stats_delayed(void);
1007 void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx,
1009 void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, int val);
1011 static inline void mod_lruvec_kmem_state(void *p, enum node_stat_item idx,
1014 unsigned long flags;
1016 local_irq_save(flags);
1017 __mod_lruvec_kmem_state(p, idx, val);
1018 local_irq_restore(flags);
1021 static inline void mod_memcg_lruvec_state(struct lruvec *lruvec,
1022 enum node_stat_item idx, int val)
1024 unsigned long flags;
1026 local_irq_save(flags);
1027 __mod_memcg_lruvec_state(lruvec, idx, val);
1028 local_irq_restore(flags);
1031 void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx,
1032 unsigned long count);
1034 static inline void count_memcg_events(struct mem_cgroup *memcg,
1035 enum vm_event_item idx,
1036 unsigned long count)
1038 unsigned long flags;
1040 local_irq_save(flags);
1041 __count_memcg_events(memcg, idx, count);
1042 local_irq_restore(flags);
1045 static inline void count_memcg_page_event(struct page *page,
1046 enum vm_event_item idx)
1048 struct mem_cgroup *memcg = page_memcg(page);
1051 count_memcg_events(memcg, idx, 1);
1054 static inline void count_memcg_event_mm(struct mm_struct *mm,
1055 enum vm_event_item idx)
1057 struct mem_cgroup *memcg;
1059 if (mem_cgroup_disabled())
1063 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1065 count_memcg_events(memcg, idx, 1);
1069 static inline void memcg_memory_event(struct mem_cgroup *memcg,
1070 enum memcg_memory_event event)
1072 bool swap_event = event == MEMCG_SWAP_HIGH || event == MEMCG_SWAP_MAX ||
1073 event == MEMCG_SWAP_FAIL;
1075 atomic_long_inc(&memcg->memory_events_local[event]);
1077 cgroup_file_notify(&memcg->events_local_file);
1080 atomic_long_inc(&memcg->memory_events[event]);
1082 cgroup_file_notify(&memcg->swap_events_file);
1084 cgroup_file_notify(&memcg->events_file);
1086 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
1088 if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
1090 } while ((memcg = parent_mem_cgroup(memcg)) &&
1091 !mem_cgroup_is_root(memcg));
1094 static inline void memcg_memory_event_mm(struct mm_struct *mm,
1095 enum memcg_memory_event event)
1097 struct mem_cgroup *memcg;
1099 if (mem_cgroup_disabled())
1103 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1105 memcg_memory_event(memcg, event);
1109 void split_page_memcg(struct page *head, unsigned int nr);
1111 unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
1113 unsigned long *total_scanned);
1115 #else /* CONFIG_MEMCG */
1117 #define MEM_CGROUP_ID_SHIFT 0
1118 #define MEM_CGROUP_ID_MAX 0
1120 static inline struct mem_cgroup *page_memcg(struct page *page)
1125 static inline struct mem_cgroup *page_memcg_rcu(struct page *page)
1127 WARN_ON_ONCE(!rcu_read_lock_held());
1131 static inline struct mem_cgroup *page_memcg_check(struct page *page)
1136 static inline bool PageMemcgKmem(struct page *page)
1141 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
1146 static inline bool mem_cgroup_disabled(void)
1151 static inline void memcg_memory_event(struct mem_cgroup *memcg,
1152 enum memcg_memory_event event)
1156 static inline void memcg_memory_event_mm(struct mm_struct *mm,
1157 enum memcg_memory_event event)
1161 static inline void mem_cgroup_protection(struct mem_cgroup *root,
1162 struct mem_cgroup *memcg,
1169 static inline void mem_cgroup_calculate_protection(struct mem_cgroup *root,
1170 struct mem_cgroup *memcg)
1174 static inline bool mem_cgroup_below_low(struct mem_cgroup *memcg)
1179 static inline bool mem_cgroup_below_min(struct mem_cgroup *memcg)
1184 static inline int mem_cgroup_charge(struct page *page, struct mm_struct *mm,
1190 static inline int mem_cgroup_swapin_charge_page(struct page *page,
1191 struct mm_struct *mm, gfp_t gfp, swp_entry_t entry)
1196 static inline void mem_cgroup_swapin_uncharge_swap(swp_entry_t entry)
1200 static inline void mem_cgroup_uncharge(struct page *page)
1204 static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
1208 static inline void mem_cgroup_migrate(struct page *old, struct page *new)
1212 static inline struct lruvec *mem_cgroup_lruvec(struct mem_cgroup *memcg,
1213 struct pglist_data *pgdat)
1215 return &pgdat->__lruvec;
1218 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page)
1220 pg_data_t *pgdat = page_pgdat(page);
1222 return &pgdat->__lruvec;
1225 static inline void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page)
1229 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
1234 static inline bool mm_match_cgroup(struct mm_struct *mm,
1235 struct mem_cgroup *memcg)
1240 static inline struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
1246 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css)
1251 static inline void mem_cgroup_put(struct mem_cgroup *memcg)
1255 static inline struct lruvec *lock_page_lruvec(struct page *page)
1257 struct pglist_data *pgdat = page_pgdat(page);
1259 spin_lock(&pgdat->__lruvec.lru_lock);
1260 return &pgdat->__lruvec;
1263 static inline struct lruvec *lock_page_lruvec_irq(struct page *page)
1265 struct pglist_data *pgdat = page_pgdat(page);
1267 spin_lock_irq(&pgdat->__lruvec.lru_lock);
1268 return &pgdat->__lruvec;
1271 static inline struct lruvec *lock_page_lruvec_irqsave(struct page *page,
1272 unsigned long *flagsp)
1274 struct pglist_data *pgdat = page_pgdat(page);
1276 spin_lock_irqsave(&pgdat->__lruvec.lru_lock, *flagsp);
1277 return &pgdat->__lruvec;
1280 static inline struct mem_cgroup *
1281 mem_cgroup_iter(struct mem_cgroup *root,
1282 struct mem_cgroup *prev,
1283 struct mem_cgroup_reclaim_cookie *reclaim)
1288 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
1289 struct mem_cgroup *prev)
1293 static inline int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
1294 int (*fn)(struct task_struct *, void *), void *arg)
1299 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
1304 static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
1307 /* XXX: This should always return root_mem_cgroup */
1311 static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m)
1316 static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec)
1321 static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
1327 unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
1328 enum lru_list lru, int zone_idx)
1333 static inline unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg)
1338 static inline unsigned long mem_cgroup_size(struct mem_cgroup *memcg)
1344 mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p)
1349 mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg)
1353 static inline void lock_page_memcg(struct page *page)
1357 static inline void unlock_page_memcg(struct page *page)
1361 static inline void mem_cgroup_handle_over_high(void)
1365 static inline void mem_cgroup_enter_user_fault(void)
1369 static inline void mem_cgroup_exit_user_fault(void)
1373 static inline bool task_in_memcg_oom(struct task_struct *p)
1378 static inline bool mem_cgroup_oom_synchronize(bool wait)
1383 static inline struct mem_cgroup *mem_cgroup_get_oom_group(
1384 struct task_struct *victim, struct mem_cgroup *oom_domain)
1389 static inline void mem_cgroup_print_oom_group(struct mem_cgroup *memcg)
1393 static inline void __mod_memcg_state(struct mem_cgroup *memcg,
1399 static inline void mod_memcg_state(struct mem_cgroup *memcg,
1405 static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx)
1410 static inline unsigned long lruvec_page_state(struct lruvec *lruvec,
1411 enum node_stat_item idx)
1413 return node_page_state(lruvec_pgdat(lruvec), idx);
1416 static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec,
1417 enum node_stat_item idx)
1419 return node_page_state(lruvec_pgdat(lruvec), idx);
1422 static inline void mem_cgroup_flush_stats(void)
1426 static inline void mem_cgroup_flush_stats_delayed(void)
1430 static inline void __mod_memcg_lruvec_state(struct lruvec *lruvec,
1431 enum node_stat_item idx, int val)
1435 static inline void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx,
1438 struct page *page = virt_to_head_page(p);
1440 __mod_node_page_state(page_pgdat(page), idx, val);
1443 static inline void mod_lruvec_kmem_state(void *p, enum node_stat_item idx,
1446 struct page *page = virt_to_head_page(p);
1448 mod_node_page_state(page_pgdat(page), idx, val);
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_events(struct mem_cgroup *memcg,
1458 enum vm_event_item idx,
1459 unsigned long count)
1463 static inline void count_memcg_page_event(struct page *page,
1469 void count_memcg_event_mm(struct mm_struct *mm, enum vm_event_item idx)
1473 static inline void split_page_memcg(struct page *head, unsigned int nr)
1478 unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
1480 unsigned long *total_scanned)
1484 #endif /* CONFIG_MEMCG */
1486 static inline void __inc_lruvec_kmem_state(void *p, enum node_stat_item idx)
1488 __mod_lruvec_kmem_state(p, idx, 1);
1491 static inline void __dec_lruvec_kmem_state(void *p, enum node_stat_item idx)
1493 __mod_lruvec_kmem_state(p, idx, -1);
1496 static inline struct lruvec *parent_lruvec(struct lruvec *lruvec)
1498 struct mem_cgroup *memcg;
1500 memcg = lruvec_memcg(lruvec);
1503 memcg = parent_mem_cgroup(memcg);
1506 return mem_cgroup_lruvec(memcg, lruvec_pgdat(lruvec));
1509 static inline void unlock_page_lruvec(struct lruvec *lruvec)
1511 spin_unlock(&lruvec->lru_lock);
1514 static inline void unlock_page_lruvec_irq(struct lruvec *lruvec)
1516 spin_unlock_irq(&lruvec->lru_lock);
1519 static inline void unlock_page_lruvec_irqrestore(struct lruvec *lruvec,
1520 unsigned long flags)
1522 spin_unlock_irqrestore(&lruvec->lru_lock, flags);
1525 /* Test requires a stable page->memcg binding, see page_memcg() */
1526 static inline bool page_matches_lruvec(struct page *page, struct lruvec *lruvec)
1528 return lruvec_pgdat(lruvec) == page_pgdat(page) &&
1529 lruvec_memcg(lruvec) == page_memcg(page);
1532 /* Don't lock again iff page's lruvec locked */
1533 static inline struct lruvec *relock_page_lruvec_irq(struct page *page,
1534 struct lruvec *locked_lruvec)
1536 if (locked_lruvec) {
1537 if (page_matches_lruvec(page, locked_lruvec))
1538 return locked_lruvec;
1540 unlock_page_lruvec_irq(locked_lruvec);
1543 return lock_page_lruvec_irq(page);
1546 /* Don't lock again iff page's lruvec locked */
1547 static inline struct lruvec *relock_page_lruvec_irqsave(struct page *page,
1548 struct lruvec *locked_lruvec, unsigned long *flags)
1550 if (locked_lruvec) {
1551 if (page_matches_lruvec(page, locked_lruvec))
1552 return locked_lruvec;
1554 unlock_page_lruvec_irqrestore(locked_lruvec, *flags);
1557 return lock_page_lruvec_irqsave(page, flags);
1560 #ifdef CONFIG_CGROUP_WRITEBACK
1562 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
1563 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
1564 unsigned long *pheadroom, unsigned long *pdirty,
1565 unsigned long *pwriteback);
1567 void mem_cgroup_track_foreign_dirty_slowpath(struct page *page,
1568 struct bdi_writeback *wb);
1570 static inline void mem_cgroup_track_foreign_dirty(struct page *page,
1571 struct bdi_writeback *wb)
1573 if (mem_cgroup_disabled())
1576 if (unlikely(&page_memcg(page)->css != wb->memcg_css))
1577 mem_cgroup_track_foreign_dirty_slowpath(page, wb);
1580 void mem_cgroup_flush_foreign(struct bdi_writeback *wb);
1582 #else /* CONFIG_CGROUP_WRITEBACK */
1584 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
1589 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
1590 unsigned long *pfilepages,
1591 unsigned long *pheadroom,
1592 unsigned long *pdirty,
1593 unsigned long *pwriteback)
1597 static inline void mem_cgroup_track_foreign_dirty(struct page *page,
1598 struct bdi_writeback *wb)
1602 static inline void mem_cgroup_flush_foreign(struct bdi_writeback *wb)
1606 #endif /* CONFIG_CGROUP_WRITEBACK */
1609 bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages,
1611 void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
1613 extern struct static_key_false memcg_sockets_enabled_key;
1614 #define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_enabled_key)
1615 void mem_cgroup_sk_alloc(struct sock *sk);
1616 void mem_cgroup_sk_free(struct sock *sk);
1617 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
1619 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_pressure)
1622 if (time_before(jiffies, memcg->socket_pressure))
1624 } while ((memcg = parent_mem_cgroup(memcg)));
1628 int alloc_shrinker_info(struct mem_cgroup *memcg);
1629 void free_shrinker_info(struct mem_cgroup *memcg);
1630 void set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id);
1631 void reparent_shrinker_deferred(struct mem_cgroup *memcg);
1633 #define mem_cgroup_sockets_enabled 0
1634 static inline void mem_cgroup_sk_alloc(struct sock *sk) { };
1635 static inline void mem_cgroup_sk_free(struct sock *sk) { };
1636 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
1641 static inline void set_shrinker_bit(struct mem_cgroup *memcg,
1642 int nid, int shrinker_id)
1647 #ifdef CONFIG_MEMCG_KMEM
1648 bool mem_cgroup_kmem_disabled(void);
1649 int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order);
1650 void __memcg_kmem_uncharge_page(struct page *page, int order);
1652 struct obj_cgroup *get_obj_cgroup_from_current(void);
1654 int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size);
1655 void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size);
1657 extern struct static_key_false memcg_kmem_enabled_key;
1659 extern int memcg_nr_cache_ids;
1660 void memcg_get_cache_ids(void);
1661 void memcg_put_cache_ids(void);
1664 * Helper macro to loop through all memcg-specific caches. Callers must still
1665 * check if the cache is valid (it is either valid or NULL).
1666 * the slab_mutex must be held when looping through those caches
1668 #define for_each_memcg_cache_index(_idx) \
1669 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
1671 static inline bool memcg_kmem_enabled(void)
1673 return static_branch_likely(&memcg_kmem_enabled_key);
1676 static inline int memcg_kmem_charge_page(struct page *page, gfp_t gfp,
1679 if (memcg_kmem_enabled())
1680 return __memcg_kmem_charge_page(page, gfp, order);
1684 static inline void memcg_kmem_uncharge_page(struct page *page, int order)
1686 if (memcg_kmem_enabled())
1687 __memcg_kmem_uncharge_page(page, order);
1691 * A helper for accessing memcg's kmem_id, used for getting
1692 * corresponding LRU lists.
1694 static inline int memcg_cache_id(struct mem_cgroup *memcg)
1696 return memcg ? memcg->kmemcg_id : -1;
1699 struct mem_cgroup *mem_cgroup_from_obj(void *p);
1702 static inline bool mem_cgroup_kmem_disabled(void)
1707 static inline int memcg_kmem_charge_page(struct page *page, gfp_t gfp,
1713 static inline void memcg_kmem_uncharge_page(struct page *page, int order)
1717 static inline int __memcg_kmem_charge_page(struct page *page, gfp_t gfp,
1723 static inline void __memcg_kmem_uncharge_page(struct page *page, int order)
1727 #define for_each_memcg_cache_index(_idx) \
1730 static inline bool memcg_kmem_enabled(void)
1735 static inline int memcg_cache_id(struct mem_cgroup *memcg)
1740 static inline void memcg_get_cache_ids(void)
1744 static inline void memcg_put_cache_ids(void)
1748 static inline struct mem_cgroup *mem_cgroup_from_obj(void *p)
1753 #endif /* CONFIG_MEMCG_KMEM */
1755 #endif /* _LINUX_MEMCONTROL_H */