Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/* memcontrol.h - Memory Controller
3 *
4 * Copyright IBM Corporation, 2007
5 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
6 *
7 * Copyright 2007 OpenVZ SWsoft Inc
8 * Author: Pavel Emelianov <xemul@openvz.org>
9 */
10
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>
20#include <linux/mm.h>
21#include <linux/vmstat.h>
22#include <linux/writeback.h>
23#include <linux/page-flags.h>
24
25struct mem_cgroup;
26struct obj_cgroup;
27struct page;
28struct mm_struct;
29struct kmem_cache;
30
31/* Cgroup-specific page state, on top of universal node page state */
32enum memcg_stat_item {
33 MEMCG_SWAP = NR_VM_NODE_STAT_ITEMS,
34 MEMCG_SOCK,
35 MEMCG_PERCPU_B,
36 MEMCG_NR_STAT,
37};
38
39enum memcg_memory_event {
40 MEMCG_LOW,
41 MEMCG_HIGH,
42 MEMCG_MAX,
43 MEMCG_OOM,
44 MEMCG_OOM_KILL,
45 MEMCG_SWAP_HIGH,
46 MEMCG_SWAP_MAX,
47 MEMCG_SWAP_FAIL,
48 MEMCG_NR_MEMORY_EVENTS,
49};
50
51struct mem_cgroup_reclaim_cookie {
52 pg_data_t *pgdat;
53 unsigned int generation;
54};
55
56#ifdef CONFIG_MEMCG
57
58#define MEM_CGROUP_ID_SHIFT 16
59#define MEM_CGROUP_ID_MAX USHRT_MAX
60
61struct mem_cgroup_id {
62 int id;
63 refcount_t ref;
64};
65
66/*
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.
71 */
72enum mem_cgroup_events_target {
73 MEM_CGROUP_TARGET_THRESH,
74 MEM_CGROUP_TARGET_SOFTLIMIT,
75 MEM_CGROUP_NTARGETS,
76};
77
78struct 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];
83};
84
85struct mem_cgroup_reclaim_iter {
86 struct mem_cgroup *position;
87 /* scan generation, increased every round-trip */
88 unsigned int generation;
89};
90
91struct lruvec_stat {
92 long count[NR_VM_NODE_STAT_ITEMS];
93};
94
95struct batched_lruvec_stat {
96 s32 count[NR_VM_NODE_STAT_ITEMS];
97};
98
99/*
100 * Bitmap of shrinker::id corresponding to memcg-aware shrinkers,
101 * which have elements charged to this memcg.
102 */
103struct memcg_shrinker_map {
104 struct rcu_head rcu;
105 unsigned long map[];
106};
107
108/*
109 * per-node information in memory controller.
110 */
111struct mem_cgroup_per_node {
112 struct lruvec lruvec;
113
114 /*
115 * Legacy local VM stats. This should be struct lruvec_stat and
116 * cannot be optimized to struct batched_lruvec_stat. Because
117 * the threshold of the lruvec_stat_cpu can be as big as
118 * MEMCG_CHARGE_BATCH * PAGE_SIZE. It can fit into s32. But this
119 * filed has no upper limit.
120 */
121 struct lruvec_stat __percpu *lruvec_stat_local;
122
123 /* Subtree VM stats (batched updates) */
124 struct batched_lruvec_stat __percpu *lruvec_stat_cpu;
125 atomic_long_t lruvec_stat[NR_VM_NODE_STAT_ITEMS];
126
127 unsigned long lru_zone_size[MAX_NR_ZONES][NR_LRU_LISTS];
128
129 struct mem_cgroup_reclaim_iter iter;
130
131 struct memcg_shrinker_map __rcu *shrinker_map;
132
133 struct rb_node tree_node; /* RB tree node */
134 unsigned long usage_in_excess;/* Set to the value by which */
135 /* the soft limit is exceeded*/
136 bool on_tree;
137 struct mem_cgroup *memcg; /* Back pointer, we cannot */
138 /* use container_of */
139};
140
141struct mem_cgroup_threshold {
142 struct eventfd_ctx *eventfd;
143 unsigned long threshold;
144};
145
146/* For threshold */
147struct mem_cgroup_threshold_ary {
148 /* An array index points to threshold just below or equal to usage. */
149 int current_threshold;
150 /* Size of entries[] */
151 unsigned int size;
152 /* Array of thresholds */
153 struct mem_cgroup_threshold entries[];
154};
155
156struct mem_cgroup_thresholds {
157 /* Primary thresholds array */
158 struct mem_cgroup_threshold_ary *primary;
159 /*
160 * Spare threshold array.
161 * This is needed to make mem_cgroup_unregister_event() "never fail".
162 * It must be able to store at least primary->size - 1 entries.
163 */
164 struct mem_cgroup_threshold_ary *spare;
165};
166
167enum memcg_kmem_state {
168 KMEM_NONE,
169 KMEM_ALLOCATED,
170 KMEM_ONLINE,
171};
172
173#if defined(CONFIG_SMP)
174struct memcg_padding {
175 char x[0];
176} ____cacheline_internodealigned_in_smp;
177#define MEMCG_PADDING(name) struct memcg_padding name;
178#else
179#define MEMCG_PADDING(name)
180#endif
181
182/*
183 * Remember four most recent foreign writebacks with dirty pages in this
184 * cgroup. Inode sharing is expected to be uncommon and, even if we miss
185 * one in a given round, we're likely to catch it later if it keeps
186 * foreign-dirtying, so a fairly low count should be enough.
187 *
188 * See mem_cgroup_track_foreign_dirty_slowpath() for details.
189 */
190#define MEMCG_CGWB_FRN_CNT 4
191
192struct memcg_cgwb_frn {
193 u64 bdi_id; /* bdi->id of the foreign inode */
194 int memcg_id; /* memcg->css.id of foreign inode */
195 u64 at; /* jiffies_64 at the time of dirtying */
196 struct wb_completion done; /* tracks in-flight foreign writebacks */
197};
198
199/*
200 * Bucket for arbitrarily byte-sized objects charged to a memory
201 * cgroup. The bucket can be reparented in one piece when the cgroup
202 * is destroyed, without having to round up the individual references
203 * of all live memory objects in the wild.
204 */
205struct obj_cgroup {
206 struct percpu_ref refcnt;
207 struct mem_cgroup *memcg;
208 atomic_t nr_charged_bytes;
209 union {
210 struct list_head list;
211 struct rcu_head rcu;
212 };
213};
214
215/*
216 * The memory controller data structure. The memory controller controls both
217 * page cache and RSS per cgroup. We would eventually like to provide
218 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
219 * to help the administrator determine what knobs to tune.
220 */
221struct mem_cgroup {
222 struct cgroup_subsys_state css;
223
224 /* Private memcg ID. Used to ID objects that outlive the cgroup */
225 struct mem_cgroup_id id;
226
227 /* Accounted resources */
228 struct page_counter memory; /* Both v1 & v2 */
229
230 union {
231 struct page_counter swap; /* v2 only */
232 struct page_counter memsw; /* v1 only */
233 };
234
235 /* Legacy consumer-oriented counters */
236 struct page_counter kmem; /* v1 only */
237 struct page_counter tcpmem; /* v1 only */
238
239 /* Range enforcement for interrupt charges */
240 struct work_struct high_work;
241
242 unsigned long soft_limit;
243
244 /* vmpressure notifications */
245 struct vmpressure vmpressure;
246
247 /*
248 * Should the OOM killer kill all belonging tasks, had it kill one?
249 */
250 bool oom_group;
251
252 /* protected by memcg_oom_lock */
253 bool oom_lock;
254 int under_oom;
255
256 int swappiness;
257 /* OOM-Killer disable */
258 int oom_kill_disable;
259
260 /* memory.events and memory.events.local */
261 struct cgroup_file events_file;
262 struct cgroup_file events_local_file;
263
264 /* handle for "memory.swap.events" */
265 struct cgroup_file swap_events_file;
266
267 /* protect arrays of thresholds */
268 struct mutex thresholds_lock;
269
270 /* thresholds for memory usage. RCU-protected */
271 struct mem_cgroup_thresholds thresholds;
272
273 /* thresholds for mem+swap usage. RCU-protected */
274 struct mem_cgroup_thresholds memsw_thresholds;
275
276 /* For oom notifier event fd */
277 struct list_head oom_notify;
278
279 /*
280 * Should we move charges of a task when a task is moved into this
281 * mem_cgroup ? And what type of charges should we move ?
282 */
283 unsigned long move_charge_at_immigrate;
284 /* taken only while moving_account > 0 */
285 spinlock_t move_lock;
286 unsigned long move_lock_flags;
287
288 MEMCG_PADDING(_pad1_);
289
290 atomic_long_t vmstats[MEMCG_NR_STAT];
291 atomic_long_t vmevents[NR_VM_EVENT_ITEMS];
292
293 /* memory.events */
294 atomic_long_t memory_events[MEMCG_NR_MEMORY_EVENTS];
295 atomic_long_t memory_events_local[MEMCG_NR_MEMORY_EVENTS];
296
297 unsigned long socket_pressure;
298
299 /* Legacy tcp memory accounting */
300 bool tcpmem_active;
301 int tcpmem_pressure;
302
303#ifdef CONFIG_MEMCG_KMEM
304 int kmemcg_id;
305 enum memcg_kmem_state kmem_state;
306 struct obj_cgroup __rcu *objcg;
307 struct list_head objcg_list; /* list of inherited objcgs */
308#endif
309
310 MEMCG_PADDING(_pad2_);
311
312 /*
313 * set > 0 if pages under this cgroup are moving to other cgroup.
314 */
315 atomic_t moving_account;
316 struct task_struct *move_lock_task;
317
318 /* Legacy local VM stats and events */
319 struct memcg_vmstats_percpu __percpu *vmstats_local;
320
321 /* Subtree VM stats and events (batched updates) */
322 struct memcg_vmstats_percpu __percpu *vmstats_percpu;
323
324#ifdef CONFIG_CGROUP_WRITEBACK
325 struct list_head cgwb_list;
326 struct wb_domain cgwb_domain;
327 struct memcg_cgwb_frn cgwb_frn[MEMCG_CGWB_FRN_CNT];
328#endif
329
330 /* List of events which userspace want to receive */
331 struct list_head event_list;
332 spinlock_t event_list_lock;
333
334#ifdef CONFIG_TRANSPARENT_HUGEPAGE
335 struct deferred_split deferred_split_queue;
336#endif
337
338 struct mem_cgroup_per_node *nodeinfo[0];
339 /* WARNING: nodeinfo must be the last member here */
340};
341
342/*
343 * size of first charge trial. "32" comes from vmscan.c's magic value.
344 * TODO: maybe necessary to use big numbers in big irons.
345 */
346#define MEMCG_CHARGE_BATCH 32U
347
348extern struct mem_cgroup *root_mem_cgroup;
349
350enum page_memcg_data_flags {
351 /* page->memcg_data is a pointer to an objcgs vector */
352 MEMCG_DATA_OBJCGS = (1UL << 0),
353 /* page has been accounted as a non-slab kernel page */
354 MEMCG_DATA_KMEM = (1UL << 1),
355 /* the next bit after the last actual flag */
356 __NR_MEMCG_DATA_FLAGS = (1UL << 2),
357};
358
359#define MEMCG_DATA_FLAGS_MASK (__NR_MEMCG_DATA_FLAGS - 1)
360
361/*
362 * page_memcg - get the memory cgroup associated with a page
363 * @page: a pointer to the page struct
364 *
365 * Returns a pointer to the memory cgroup associated with the page,
366 * or NULL. This function assumes that the page is known to have a
367 * proper memory cgroup pointer. It's not safe to call this function
368 * against some type of pages, e.g. slab pages or ex-slab pages.
369 *
370 * Any of the following ensures page and memcg binding stability:
371 * - the page lock
372 * - LRU isolation
373 * - lock_page_memcg()
374 * - exclusive reference
375 */
376static inline struct mem_cgroup *page_memcg(struct page *page)
377{
378 unsigned long memcg_data = page->memcg_data;
379
380 VM_BUG_ON_PAGE(PageSlab(page), page);
381 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_OBJCGS, page);
382
383 return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
384}
385
386/*
387 * page_memcg_rcu - locklessly get the memory cgroup associated with a page
388 * @page: a pointer to the page struct
389 *
390 * Returns a pointer to the memory cgroup associated with the page,
391 * or NULL. This function assumes that the page is known to have a
392 * proper memory cgroup pointer. It's not safe to call this function
393 * against some type of pages, e.g. slab pages or ex-slab pages.
394 */
395static inline struct mem_cgroup *page_memcg_rcu(struct page *page)
396{
397 VM_BUG_ON_PAGE(PageSlab(page), page);
398 WARN_ON_ONCE(!rcu_read_lock_held());
399
400 return (struct mem_cgroup *)(READ_ONCE(page->memcg_data) &
401 ~MEMCG_DATA_FLAGS_MASK);
402}
403
404/*
405 * page_memcg_check - get the memory cgroup associated with a page
406 * @page: a pointer to the page struct
407 *
408 * Returns a pointer to the memory cgroup associated with the page,
409 * or NULL. This function unlike page_memcg() can take any page
410 * as an argument. It has to be used in cases when it's not known if a page
411 * has an associated memory cgroup pointer or an object cgroups vector.
412 *
413 * Any of the following ensures page and memcg binding stability:
414 * - the page lock
415 * - LRU isolation
416 * - lock_page_memcg()
417 * - exclusive reference
418 */
419static inline struct mem_cgroup *page_memcg_check(struct page *page)
420{
421 /*
422 * Because page->memcg_data might be changed asynchronously
423 * for slab pages, READ_ONCE() should be used here.
424 */
425 unsigned long memcg_data = READ_ONCE(page->memcg_data);
426
427 if (memcg_data & MEMCG_DATA_OBJCGS)
428 return NULL;
429
430 return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
431}
432
433/*
434 * PageMemcgKmem - check if the page has MemcgKmem flag set
435 * @page: a pointer to the page struct
436 *
437 * Checks if the page has MemcgKmem flag set. The caller must ensure that
438 * the page has an associated memory cgroup. It's not safe to call this function
439 * against some types of pages, e.g. slab pages.
440 */
441static inline bool PageMemcgKmem(struct page *page)
442{
443 VM_BUG_ON_PAGE(page->memcg_data & MEMCG_DATA_OBJCGS, page);
444 return page->memcg_data & MEMCG_DATA_KMEM;
445}
446
447#ifdef CONFIG_MEMCG_KMEM
448/*
449 * page_objcgs - get the object cgroups vector associated with a page
450 * @page: a pointer to the page struct
451 *
452 * Returns a pointer to the object cgroups vector associated with the page,
453 * or NULL. This function assumes that the page is known to have an
454 * associated object cgroups vector. It's not safe to call this function
455 * against pages, which might have an associated memory cgroup: e.g.
456 * kernel stack pages.
457 */
458static inline struct obj_cgroup **page_objcgs(struct page *page)
459{
460 unsigned long memcg_data = READ_ONCE(page->memcg_data);
461
462 VM_BUG_ON_PAGE(memcg_data && !(memcg_data & MEMCG_DATA_OBJCGS), page);
463 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, page);
464
465 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
466}
467
468/*
469 * page_objcgs_check - get the object cgroups vector associated with a page
470 * @page: a pointer to the page struct
471 *
472 * Returns a pointer to the object cgroups vector associated with the page,
473 * or NULL. This function is safe to use if the page can be directly associated
474 * with a memory cgroup.
475 */
476static inline struct obj_cgroup **page_objcgs_check(struct page *page)
477{
478 unsigned long memcg_data = READ_ONCE(page->memcg_data);
479
480 if (!memcg_data || !(memcg_data & MEMCG_DATA_OBJCGS))
481 return NULL;
482
483 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, page);
484
485 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
486}
487
488#else
489static inline struct obj_cgroup **page_objcgs(struct page *page)
490{
491 return NULL;
492}
493
494static inline struct obj_cgroup **page_objcgs_check(struct page *page)
495{
496 return NULL;
497}
498#endif
499
500static __always_inline bool memcg_stat_item_in_bytes(int idx)
501{
502 if (idx == MEMCG_PERCPU_B)
503 return true;
504 return vmstat_item_in_bytes(idx);
505}
506
507static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
508{
509 return (memcg == root_mem_cgroup);
510}
511
512static inline bool mem_cgroup_disabled(void)
513{
514 return !cgroup_subsys_enabled(memory_cgrp_subsys);
515}
516
517static inline unsigned long mem_cgroup_protection(struct mem_cgroup *root,
518 struct mem_cgroup *memcg,
519 bool in_low_reclaim)
520{
521 if (mem_cgroup_disabled())
522 return 0;
523
524 /*
525 * There is no reclaim protection applied to a targeted reclaim.
526 * We are special casing this specific case here because
527 * mem_cgroup_protected calculation is not robust enough to keep
528 * the protection invariant for calculated effective values for
529 * parallel reclaimers with different reclaim target. This is
530 * especially a problem for tail memcgs (as they have pages on LRU)
531 * which would want to have effective values 0 for targeted reclaim
532 * but a different value for external reclaim.
533 *
534 * Example
535 * Let's have global and A's reclaim in parallel:
536 * |
537 * A (low=2G, usage = 3G, max = 3G, children_low_usage = 1.5G)
538 * |\
539 * | C (low = 1G, usage = 2.5G)
540 * B (low = 1G, usage = 0.5G)
541 *
542 * For the global reclaim
543 * A.elow = A.low
544 * B.elow = min(B.usage, B.low) because children_low_usage <= A.elow
545 * C.elow = min(C.usage, C.low)
546 *
547 * With the effective values resetting we have A reclaim
548 * A.elow = 0
549 * B.elow = B.low
550 * C.elow = C.low
551 *
552 * If the global reclaim races with A's reclaim then
553 * B.elow = C.elow = 0 because children_low_usage > A.elow)
554 * is possible and reclaiming B would be violating the protection.
555 *
556 */
557 if (root == memcg)
558 return 0;
559
560 if (in_low_reclaim)
561 return READ_ONCE(memcg->memory.emin);
562
563 return max(READ_ONCE(memcg->memory.emin),
564 READ_ONCE(memcg->memory.elow));
565}
566
567void mem_cgroup_calculate_protection(struct mem_cgroup *root,
568 struct mem_cgroup *memcg);
569
570static inline bool mem_cgroup_supports_protection(struct mem_cgroup *memcg)
571{
572 /*
573 * The root memcg doesn't account charges, and doesn't support
574 * protection.
575 */
576 return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg);
577
578}
579
580static inline bool mem_cgroup_below_low(struct mem_cgroup *memcg)
581{
582 if (!mem_cgroup_supports_protection(memcg))
583 return false;
584
585 return READ_ONCE(memcg->memory.elow) >=
586 page_counter_read(&memcg->memory);
587}
588
589static inline bool mem_cgroup_below_min(struct mem_cgroup *memcg)
590{
591 if (!mem_cgroup_supports_protection(memcg))
592 return false;
593
594 return READ_ONCE(memcg->memory.emin) >=
595 page_counter_read(&memcg->memory);
596}
597
598int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask);
599
600void mem_cgroup_uncharge(struct page *page);
601void mem_cgroup_uncharge_list(struct list_head *page_list);
602
603void mem_cgroup_migrate(struct page *oldpage, struct page *newpage);
604
605static struct mem_cgroup_per_node *
606mem_cgroup_nodeinfo(struct mem_cgroup *memcg, int nid)
607{
608 return memcg->nodeinfo[nid];
609}
610
611/**
612 * mem_cgroup_lruvec - get the lru list vector for a memcg & node
613 * @memcg: memcg of the wanted lruvec
614 * @pgdat: pglist_data
615 *
616 * Returns the lru list vector holding pages for a given @memcg &
617 * @pgdat combination. This can be the node lruvec, if the memory
618 * controller is disabled.
619 */
620static inline struct lruvec *mem_cgroup_lruvec(struct mem_cgroup *memcg,
621 struct pglist_data *pgdat)
622{
623 struct mem_cgroup_per_node *mz;
624 struct lruvec *lruvec;
625
626 if (mem_cgroup_disabled()) {
627 lruvec = &pgdat->__lruvec;
628 goto out;
629 }
630
631 if (!memcg)
632 memcg = root_mem_cgroup;
633
634 mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
635 lruvec = &mz->lruvec;
636out:
637 /*
638 * Since a node can be onlined after the mem_cgroup was created,
639 * we have to be prepared to initialize lruvec->pgdat here;
640 * and if offlined then reonlined, we need to reinitialize it.
641 */
642 if (unlikely(lruvec->pgdat != pgdat))
643 lruvec->pgdat = pgdat;
644 return lruvec;
645}
646
647/**
648 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
649 * @page: the page
650 * @pgdat: pgdat of the page
651 *
652 * This function relies on page->mem_cgroup being stable.
653 */
654static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
655 struct pglist_data *pgdat)
656{
657 struct mem_cgroup *memcg = page_memcg(page);
658
659 VM_WARN_ON_ONCE_PAGE(!memcg && !mem_cgroup_disabled(), page);
660 return mem_cgroup_lruvec(memcg, pgdat);
661}
662
663static inline bool lruvec_holds_page_lru_lock(struct page *page,
664 struct lruvec *lruvec)
665{
666 pg_data_t *pgdat = page_pgdat(page);
667 const struct mem_cgroup *memcg;
668 struct mem_cgroup_per_node *mz;
669
670 if (mem_cgroup_disabled())
671 return lruvec == &pgdat->__lruvec;
672
673 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
674 memcg = page_memcg(page) ? : root_mem_cgroup;
675
676 return lruvec->pgdat == pgdat && mz->memcg == memcg;
677}
678
679struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
680
681struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm);
682
683struct lruvec *lock_page_lruvec(struct page *page);
684struct lruvec *lock_page_lruvec_irq(struct page *page);
685struct lruvec *lock_page_lruvec_irqsave(struct page *page,
686 unsigned long *flags);
687
688#ifdef CONFIG_DEBUG_VM
689void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page);
690#else
691static inline void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page)
692{
693}
694#endif
695
696static inline
697struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
698 return css ? container_of(css, struct mem_cgroup, css) : NULL;
699}
700
701static inline bool obj_cgroup_tryget(struct obj_cgroup *objcg)
702{
703 return percpu_ref_tryget(&objcg->refcnt);
704}
705
706static inline void obj_cgroup_get(struct obj_cgroup *objcg)
707{
708 percpu_ref_get(&objcg->refcnt);
709}
710
711static inline void obj_cgroup_put(struct obj_cgroup *objcg)
712{
713 percpu_ref_put(&objcg->refcnt);
714}
715
716/*
717 * After the initialization objcg->memcg is always pointing at
718 * a valid memcg, but can be atomically swapped to the parent memcg.
719 *
720 * The caller must ensure that the returned memcg won't be released:
721 * e.g. acquire the rcu_read_lock or css_set_lock.
722 */
723static inline struct mem_cgroup *obj_cgroup_memcg(struct obj_cgroup *objcg)
724{
725 return READ_ONCE(objcg->memcg);
726}
727
728static inline void mem_cgroup_put(struct mem_cgroup *memcg)
729{
730 if (memcg)
731 css_put(&memcg->css);
732}
733
734#define mem_cgroup_from_counter(counter, member) \
735 container_of(counter, struct mem_cgroup, member)
736
737struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
738 struct mem_cgroup *,
739 struct mem_cgroup_reclaim_cookie *);
740void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
741int mem_cgroup_scan_tasks(struct mem_cgroup *,
742 int (*)(struct task_struct *, void *), void *);
743
744static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
745{
746 if (mem_cgroup_disabled())
747 return 0;
748
749 return memcg->id.id;
750}
751struct mem_cgroup *mem_cgroup_from_id(unsigned short id);
752
753static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m)
754{
755 return mem_cgroup_from_css(seq_css(m));
756}
757
758static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec)
759{
760 struct mem_cgroup_per_node *mz;
761
762 if (mem_cgroup_disabled())
763 return NULL;
764
765 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
766 return mz->memcg;
767}
768
769/**
770 * parent_mem_cgroup - find the accounting parent of a memcg
771 * @memcg: memcg whose parent to find
772 *
773 * Returns the parent memcg, or NULL if this is the root or the memory
774 * controller is in legacy no-hierarchy mode.
775 */
776static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
777{
778 if (!memcg->memory.parent)
779 return NULL;
780 return mem_cgroup_from_counter(memcg->memory.parent, memory);
781}
782
783static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
784 struct mem_cgroup *root)
785{
786 if (root == memcg)
787 return true;
788 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
789}
790
791static inline bool mm_match_cgroup(struct mm_struct *mm,
792 struct mem_cgroup *memcg)
793{
794 struct mem_cgroup *task_memcg;
795 bool match = false;
796
797 rcu_read_lock();
798 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
799 if (task_memcg)
800 match = mem_cgroup_is_descendant(task_memcg, memcg);
801 rcu_read_unlock();
802 return match;
803}
804
805struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
806ino_t page_cgroup_ino(struct page *page);
807
808static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
809{
810 if (mem_cgroup_disabled())
811 return true;
812 return !!(memcg->css.flags & CSS_ONLINE);
813}
814
815/*
816 * For memory reclaim.
817 */
818int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
819
820void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
821 int zid, int nr_pages);
822
823static inline
824unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
825 enum lru_list lru, int zone_idx)
826{
827 struct mem_cgroup_per_node *mz;
828
829 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
830 return READ_ONCE(mz->lru_zone_size[zone_idx][lru]);
831}
832
833void mem_cgroup_handle_over_high(void);
834
835unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg);
836
837unsigned long mem_cgroup_size(struct mem_cgroup *memcg);
838
839void mem_cgroup_print_oom_context(struct mem_cgroup *memcg,
840 struct task_struct *p);
841
842void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg);
843
844static inline void mem_cgroup_enter_user_fault(void)
845{
846 WARN_ON(current->in_user_fault);
847 current->in_user_fault = 1;
848}
849
850static inline void mem_cgroup_exit_user_fault(void)
851{
852 WARN_ON(!current->in_user_fault);
853 current->in_user_fault = 0;
854}
855
856static inline bool task_in_memcg_oom(struct task_struct *p)
857{
858 return p->memcg_in_oom;
859}
860
861bool mem_cgroup_oom_synchronize(bool wait);
862struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim,
863 struct mem_cgroup *oom_domain);
864void mem_cgroup_print_oom_group(struct mem_cgroup *memcg);
865
866#ifdef CONFIG_MEMCG_SWAP
867extern bool cgroup_memory_noswap;
868#endif
869
870struct mem_cgroup *lock_page_memcg(struct page *page);
871void __unlock_page_memcg(struct mem_cgroup *memcg);
872void unlock_page_memcg(struct page *page);
873
874/*
875 * idx can be of type enum memcg_stat_item or node_stat_item.
876 * Keep in sync with memcg_exact_page_state().
877 */
878static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx)
879{
880 long x = atomic_long_read(&memcg->vmstats[idx]);
881#ifdef CONFIG_SMP
882 if (x < 0)
883 x = 0;
884#endif
885 return x;
886}
887
888/*
889 * idx can be of type enum memcg_stat_item or node_stat_item.
890 * Keep in sync with memcg_exact_page_state().
891 */
892static inline unsigned long memcg_page_state_local(struct mem_cgroup *memcg,
893 int idx)
894{
895 long x = 0;
896 int cpu;
897
898 for_each_possible_cpu(cpu)
899 x += per_cpu(memcg->vmstats_local->stat[idx], cpu);
900#ifdef CONFIG_SMP
901 if (x < 0)
902 x = 0;
903#endif
904 return x;
905}
906
907void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val);
908
909/* idx can be of type enum memcg_stat_item or node_stat_item */
910static inline void mod_memcg_state(struct mem_cgroup *memcg,
911 int idx, int val)
912{
913 unsigned long flags;
914
915 local_irq_save(flags);
916 __mod_memcg_state(memcg, idx, val);
917 local_irq_restore(flags);
918}
919
920static inline unsigned long lruvec_page_state(struct lruvec *lruvec,
921 enum node_stat_item idx)
922{
923 struct mem_cgroup_per_node *pn;
924 long x;
925
926 if (mem_cgroup_disabled())
927 return node_page_state(lruvec_pgdat(lruvec), idx);
928
929 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
930 x = atomic_long_read(&pn->lruvec_stat[idx]);
931#ifdef CONFIG_SMP
932 if (x < 0)
933 x = 0;
934#endif
935 return x;
936}
937
938static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec,
939 enum node_stat_item idx)
940{
941 struct mem_cgroup_per_node *pn;
942 long x = 0;
943 int cpu;
944
945 if (mem_cgroup_disabled())
946 return node_page_state(lruvec_pgdat(lruvec), idx);
947
948 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
949 for_each_possible_cpu(cpu)
950 x += per_cpu(pn->lruvec_stat_local->count[idx], cpu);
951#ifdef CONFIG_SMP
952 if (x < 0)
953 x = 0;
954#endif
955 return x;
956}
957
958void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx,
959 int val);
960void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, int val);
961
962static inline void mod_lruvec_kmem_state(void *p, enum node_stat_item idx,
963 int val)
964{
965 unsigned long flags;
966
967 local_irq_save(flags);
968 __mod_lruvec_kmem_state(p, idx, val);
969 local_irq_restore(flags);
970}
971
972static inline void mod_memcg_lruvec_state(struct lruvec *lruvec,
973 enum node_stat_item idx, int val)
974{
975 unsigned long flags;
976
977 local_irq_save(flags);
978 __mod_memcg_lruvec_state(lruvec, idx, val);
979 local_irq_restore(flags);
980}
981
982unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
983 gfp_t gfp_mask,
984 unsigned long *total_scanned);
985
986void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx,
987 unsigned long count);
988
989static inline void count_memcg_events(struct mem_cgroup *memcg,
990 enum vm_event_item idx,
991 unsigned long count)
992{
993 unsigned long flags;
994
995 local_irq_save(flags);
996 __count_memcg_events(memcg, idx, count);
997 local_irq_restore(flags);
998}
999
1000static inline void count_memcg_page_event(struct page *page,
1001 enum vm_event_item idx)
1002{
1003 struct mem_cgroup *memcg = page_memcg(page);
1004
1005 if (memcg)
1006 count_memcg_events(memcg, idx, 1);
1007}
1008
1009static inline void count_memcg_event_mm(struct mm_struct *mm,
1010 enum vm_event_item idx)
1011{
1012 struct mem_cgroup *memcg;
1013
1014 if (mem_cgroup_disabled())
1015 return;
1016
1017 rcu_read_lock();
1018 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1019 if (likely(memcg))
1020 count_memcg_events(memcg, idx, 1);
1021 rcu_read_unlock();
1022}
1023
1024static inline void memcg_memory_event(struct mem_cgroup *memcg,
1025 enum memcg_memory_event event)
1026{
1027 bool swap_event = event == MEMCG_SWAP_HIGH || event == MEMCG_SWAP_MAX ||
1028 event == MEMCG_SWAP_FAIL;
1029
1030 atomic_long_inc(&memcg->memory_events_local[event]);
1031 if (!swap_event)
1032 cgroup_file_notify(&memcg->events_local_file);
1033
1034 do {
1035 atomic_long_inc(&memcg->memory_events[event]);
1036 if (swap_event)
1037 cgroup_file_notify(&memcg->swap_events_file);
1038 else
1039 cgroup_file_notify(&memcg->events_file);
1040
1041 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
1042 break;
1043 if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
1044 break;
1045 } while ((memcg = parent_mem_cgroup(memcg)) &&
1046 !mem_cgroup_is_root(memcg));
1047}
1048
1049static inline void memcg_memory_event_mm(struct mm_struct *mm,
1050 enum memcg_memory_event event)
1051{
1052 struct mem_cgroup *memcg;
1053
1054 if (mem_cgroup_disabled())
1055 return;
1056
1057 rcu_read_lock();
1058 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1059 if (likely(memcg))
1060 memcg_memory_event(memcg, event);
1061 rcu_read_unlock();
1062}
1063
1064#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1065void mem_cgroup_split_huge_fixup(struct page *head);
1066#endif
1067
1068#else /* CONFIG_MEMCG */
1069
1070#define MEM_CGROUP_ID_SHIFT 0
1071#define MEM_CGROUP_ID_MAX 0
1072
1073struct mem_cgroup;
1074
1075static inline struct mem_cgroup *page_memcg(struct page *page)
1076{
1077 return NULL;
1078}
1079
1080static inline struct mem_cgroup *page_memcg_rcu(struct page *page)
1081{
1082 WARN_ON_ONCE(!rcu_read_lock_held());
1083 return NULL;
1084}
1085
1086static inline struct mem_cgroup *page_memcg_check(struct page *page)
1087{
1088 return NULL;
1089}
1090
1091static inline bool PageMemcgKmem(struct page *page)
1092{
1093 return false;
1094}
1095
1096static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
1097{
1098 return true;
1099}
1100
1101static inline bool mem_cgroup_disabled(void)
1102{
1103 return true;
1104}
1105
1106static inline void memcg_memory_event(struct mem_cgroup *memcg,
1107 enum memcg_memory_event event)
1108{
1109}
1110
1111static inline void memcg_memory_event_mm(struct mm_struct *mm,
1112 enum memcg_memory_event event)
1113{
1114}
1115
1116static inline unsigned long mem_cgroup_protection(struct mem_cgroup *root,
1117 struct mem_cgroup *memcg,
1118 bool in_low_reclaim)
1119{
1120 return 0;
1121}
1122
1123static inline void mem_cgroup_calculate_protection(struct mem_cgroup *root,
1124 struct mem_cgroup *memcg)
1125{
1126}
1127
1128static inline bool mem_cgroup_below_low(struct mem_cgroup *memcg)
1129{
1130 return false;
1131}
1132
1133static inline bool mem_cgroup_below_min(struct mem_cgroup *memcg)
1134{
1135 return false;
1136}
1137
1138static inline int mem_cgroup_charge(struct page *page, struct mm_struct *mm,
1139 gfp_t gfp_mask)
1140{
1141 return 0;
1142}
1143
1144static inline void mem_cgroup_uncharge(struct page *page)
1145{
1146}
1147
1148static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
1149{
1150}
1151
1152static inline void mem_cgroup_migrate(struct page *old, struct page *new)
1153{
1154}
1155
1156static inline struct lruvec *mem_cgroup_lruvec(struct mem_cgroup *memcg,
1157 struct pglist_data *pgdat)
1158{
1159 return &pgdat->__lruvec;
1160}
1161
1162static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
1163 struct pglist_data *pgdat)
1164{
1165 return &pgdat->__lruvec;
1166}
1167
1168static inline bool lruvec_holds_page_lru_lock(struct page *page,
1169 struct lruvec *lruvec)
1170{
1171 pg_data_t *pgdat = page_pgdat(page);
1172
1173 return lruvec == &pgdat->__lruvec;
1174}
1175
1176static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
1177{
1178 return NULL;
1179}
1180
1181static inline bool mm_match_cgroup(struct mm_struct *mm,
1182 struct mem_cgroup *memcg)
1183{
1184 return true;
1185}
1186
1187static inline struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
1188{
1189 return NULL;
1190}
1191
1192static inline void mem_cgroup_put(struct mem_cgroup *memcg)
1193{
1194}
1195
1196static inline struct lruvec *lock_page_lruvec(struct page *page)
1197{
1198 struct pglist_data *pgdat = page_pgdat(page);
1199
1200 spin_lock(&pgdat->__lruvec.lru_lock);
1201 return &pgdat->__lruvec;
1202}
1203
1204static inline struct lruvec *lock_page_lruvec_irq(struct page *page)
1205{
1206 struct pglist_data *pgdat = page_pgdat(page);
1207
1208 spin_lock_irq(&pgdat->__lruvec.lru_lock);
1209 return &pgdat->__lruvec;
1210}
1211
1212static inline struct lruvec *lock_page_lruvec_irqsave(struct page *page,
1213 unsigned long *flagsp)
1214{
1215 struct pglist_data *pgdat = page_pgdat(page);
1216
1217 spin_lock_irqsave(&pgdat->__lruvec.lru_lock, *flagsp);
1218 return &pgdat->__lruvec;
1219}
1220
1221static inline struct mem_cgroup *
1222mem_cgroup_iter(struct mem_cgroup *root,
1223 struct mem_cgroup *prev,
1224 struct mem_cgroup_reclaim_cookie *reclaim)
1225{
1226 return NULL;
1227}
1228
1229static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
1230 struct mem_cgroup *prev)
1231{
1232}
1233
1234static inline int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
1235 int (*fn)(struct task_struct *, void *), void *arg)
1236{
1237 return 0;
1238}
1239
1240static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
1241{
1242 return 0;
1243}
1244
1245static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
1246{
1247 WARN_ON_ONCE(id);
1248 /* XXX: This should always return root_mem_cgroup */
1249 return NULL;
1250}
1251
1252static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m)
1253{
1254 return NULL;
1255}
1256
1257static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec)
1258{
1259 return NULL;
1260}
1261
1262static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
1263{
1264 return true;
1265}
1266
1267static inline
1268unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
1269 enum lru_list lru, int zone_idx)
1270{
1271 return 0;
1272}
1273
1274static inline unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg)
1275{
1276 return 0;
1277}
1278
1279static inline unsigned long mem_cgroup_size(struct mem_cgroup *memcg)
1280{
1281 return 0;
1282}
1283
1284static inline void
1285mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p)
1286{
1287}
1288
1289static inline void
1290mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg)
1291{
1292}
1293
1294static inline struct mem_cgroup *lock_page_memcg(struct page *page)
1295{
1296 return NULL;
1297}
1298
1299static inline void __unlock_page_memcg(struct mem_cgroup *memcg)
1300{
1301}
1302
1303static inline void unlock_page_memcg(struct page *page)
1304{
1305}
1306
1307static inline void mem_cgroup_handle_over_high(void)
1308{
1309}
1310
1311static inline void mem_cgroup_enter_user_fault(void)
1312{
1313}
1314
1315static inline void mem_cgroup_exit_user_fault(void)
1316{
1317}
1318
1319static inline bool task_in_memcg_oom(struct task_struct *p)
1320{
1321 return false;
1322}
1323
1324static inline bool mem_cgroup_oom_synchronize(bool wait)
1325{
1326 return false;
1327}
1328
1329static inline struct mem_cgroup *mem_cgroup_get_oom_group(
1330 struct task_struct *victim, struct mem_cgroup *oom_domain)
1331{
1332 return NULL;
1333}
1334
1335static inline void mem_cgroup_print_oom_group(struct mem_cgroup *memcg)
1336{
1337}
1338
1339static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx)
1340{
1341 return 0;
1342}
1343
1344static inline unsigned long memcg_page_state_local(struct mem_cgroup *memcg,
1345 int idx)
1346{
1347 return 0;
1348}
1349
1350static inline void __mod_memcg_state(struct mem_cgroup *memcg,
1351 int idx,
1352 int nr)
1353{
1354}
1355
1356static inline void mod_memcg_state(struct mem_cgroup *memcg,
1357 int idx,
1358 int nr)
1359{
1360}
1361
1362static inline unsigned long lruvec_page_state(struct lruvec *lruvec,
1363 enum node_stat_item idx)
1364{
1365 return node_page_state(lruvec_pgdat(lruvec), idx);
1366}
1367
1368static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec,
1369 enum node_stat_item idx)
1370{
1371 return node_page_state(lruvec_pgdat(lruvec), idx);
1372}
1373
1374static inline void __mod_memcg_lruvec_state(struct lruvec *lruvec,
1375 enum node_stat_item idx, int val)
1376{
1377}
1378
1379static inline void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx,
1380 int val)
1381{
1382 struct page *page = virt_to_head_page(p);
1383
1384 __mod_node_page_state(page_pgdat(page), idx, val);
1385}
1386
1387static inline void mod_lruvec_kmem_state(void *p, enum node_stat_item idx,
1388 int val)
1389{
1390 struct page *page = virt_to_head_page(p);
1391
1392 mod_node_page_state(page_pgdat(page), idx, val);
1393}
1394
1395static inline
1396unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
1397 gfp_t gfp_mask,
1398 unsigned long *total_scanned)
1399{
1400 return 0;
1401}
1402
1403static inline void mem_cgroup_split_huge_fixup(struct page *head)
1404{
1405}
1406
1407static inline void count_memcg_events(struct mem_cgroup *memcg,
1408 enum vm_event_item idx,
1409 unsigned long count)
1410{
1411}
1412
1413static inline void __count_memcg_events(struct mem_cgroup *memcg,
1414 enum vm_event_item idx,
1415 unsigned long count)
1416{
1417}
1418
1419static inline void count_memcg_page_event(struct page *page,
1420 int idx)
1421{
1422}
1423
1424static inline
1425void count_memcg_event_mm(struct mm_struct *mm, enum vm_event_item idx)
1426{
1427}
1428
1429static inline void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page)
1430{
1431}
1432#endif /* CONFIG_MEMCG */
1433
1434static inline void __inc_lruvec_kmem_state(void *p, enum node_stat_item idx)
1435{
1436 __mod_lruvec_kmem_state(p, idx, 1);
1437}
1438
1439static inline void __dec_lruvec_kmem_state(void *p, enum node_stat_item idx)
1440{
1441 __mod_lruvec_kmem_state(p, idx, -1);
1442}
1443
1444static inline struct lruvec *parent_lruvec(struct lruvec *lruvec)
1445{
1446 struct mem_cgroup *memcg;
1447
1448 memcg = lruvec_memcg(lruvec);
1449 if (!memcg)
1450 return NULL;
1451 memcg = parent_mem_cgroup(memcg);
1452 if (!memcg)
1453 return NULL;
1454 return mem_cgroup_lruvec(memcg, lruvec_pgdat(lruvec));
1455}
1456
1457static inline void unlock_page_lruvec(struct lruvec *lruvec)
1458{
1459 spin_unlock(&lruvec->lru_lock);
1460}
1461
1462static inline void unlock_page_lruvec_irq(struct lruvec *lruvec)
1463{
1464 spin_unlock_irq(&lruvec->lru_lock);
1465}
1466
1467static inline void unlock_page_lruvec_irqrestore(struct lruvec *lruvec,
1468 unsigned long flags)
1469{
1470 spin_unlock_irqrestore(&lruvec->lru_lock, flags);
1471}
1472
1473/* Don't lock again iff page's lruvec locked */
1474static inline struct lruvec *relock_page_lruvec_irq(struct page *page,
1475 struct lruvec *locked_lruvec)
1476{
1477 if (locked_lruvec) {
1478 if (lruvec_holds_page_lru_lock(page, locked_lruvec))
1479 return locked_lruvec;
1480
1481 unlock_page_lruvec_irq(locked_lruvec);
1482 }
1483
1484 return lock_page_lruvec_irq(page);
1485}
1486
1487/* Don't lock again iff page's lruvec locked */
1488static inline struct lruvec *relock_page_lruvec_irqsave(struct page *page,
1489 struct lruvec *locked_lruvec, unsigned long *flags)
1490{
1491 if (locked_lruvec) {
1492 if (lruvec_holds_page_lru_lock(page, locked_lruvec))
1493 return locked_lruvec;
1494
1495 unlock_page_lruvec_irqrestore(locked_lruvec, *flags);
1496 }
1497
1498 return lock_page_lruvec_irqsave(page, flags);
1499}
1500
1501#ifdef CONFIG_CGROUP_WRITEBACK
1502
1503struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
1504void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
1505 unsigned long *pheadroom, unsigned long *pdirty,
1506 unsigned long *pwriteback);
1507
1508void mem_cgroup_track_foreign_dirty_slowpath(struct page *page,
1509 struct bdi_writeback *wb);
1510
1511static inline void mem_cgroup_track_foreign_dirty(struct page *page,
1512 struct bdi_writeback *wb)
1513{
1514 if (mem_cgroup_disabled())
1515 return;
1516
1517 if (unlikely(&page_memcg(page)->css != wb->memcg_css))
1518 mem_cgroup_track_foreign_dirty_slowpath(page, wb);
1519}
1520
1521void mem_cgroup_flush_foreign(struct bdi_writeback *wb);
1522
1523#else /* CONFIG_CGROUP_WRITEBACK */
1524
1525static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
1526{
1527 return NULL;
1528}
1529
1530static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
1531 unsigned long *pfilepages,
1532 unsigned long *pheadroom,
1533 unsigned long *pdirty,
1534 unsigned long *pwriteback)
1535{
1536}
1537
1538static inline void mem_cgroup_track_foreign_dirty(struct page *page,
1539 struct bdi_writeback *wb)
1540{
1541}
1542
1543static inline void mem_cgroup_flush_foreign(struct bdi_writeback *wb)
1544{
1545}
1546
1547#endif /* CONFIG_CGROUP_WRITEBACK */
1548
1549struct sock;
1550bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
1551void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
1552#ifdef CONFIG_MEMCG
1553extern struct static_key_false memcg_sockets_enabled_key;
1554#define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_enabled_key)
1555void mem_cgroup_sk_alloc(struct sock *sk);
1556void mem_cgroup_sk_free(struct sock *sk);
1557static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
1558{
1559 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_pressure)
1560 return true;
1561 do {
1562 if (time_before(jiffies, memcg->socket_pressure))
1563 return true;
1564 } while ((memcg = parent_mem_cgroup(memcg)));
1565 return false;
1566}
1567
1568extern int memcg_expand_shrinker_maps(int new_id);
1569
1570extern void memcg_set_shrinker_bit(struct mem_cgroup *memcg,
1571 int nid, int shrinker_id);
1572#else
1573#define mem_cgroup_sockets_enabled 0
1574static inline void mem_cgroup_sk_alloc(struct sock *sk) { };
1575static inline void mem_cgroup_sk_free(struct sock *sk) { };
1576static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
1577{
1578 return false;
1579}
1580
1581static inline void memcg_set_shrinker_bit(struct mem_cgroup *memcg,
1582 int nid, int shrinker_id)
1583{
1584}
1585#endif
1586
1587#ifdef CONFIG_MEMCG_KMEM
1588int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order);
1589void __memcg_kmem_uncharge_page(struct page *page, int order);
1590
1591struct obj_cgroup *get_obj_cgroup_from_current(void);
1592
1593int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size);
1594void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size);
1595
1596extern struct static_key_false memcg_kmem_enabled_key;
1597
1598extern int memcg_nr_cache_ids;
1599void memcg_get_cache_ids(void);
1600void memcg_put_cache_ids(void);
1601
1602/*
1603 * Helper macro to loop through all memcg-specific caches. Callers must still
1604 * check if the cache is valid (it is either valid or NULL).
1605 * the slab_mutex must be held when looping through those caches
1606 */
1607#define for_each_memcg_cache_index(_idx) \
1608 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
1609
1610static inline bool memcg_kmem_enabled(void)
1611{
1612 return static_branch_likely(&memcg_kmem_enabled_key);
1613}
1614
1615static inline int memcg_kmem_charge_page(struct page *page, gfp_t gfp,
1616 int order)
1617{
1618 if (memcg_kmem_enabled())
1619 return __memcg_kmem_charge_page(page, gfp, order);
1620 return 0;
1621}
1622
1623static inline void memcg_kmem_uncharge_page(struct page *page, int order)
1624{
1625 if (memcg_kmem_enabled())
1626 __memcg_kmem_uncharge_page(page, order);
1627}
1628
1629/*
1630 * A helper for accessing memcg's kmem_id, used for getting
1631 * corresponding LRU lists.
1632 */
1633static inline int memcg_cache_id(struct mem_cgroup *memcg)
1634{
1635 return memcg ? memcg->kmemcg_id : -1;
1636}
1637
1638struct mem_cgroup *mem_cgroup_from_obj(void *p);
1639
1640#else
1641
1642static inline int memcg_kmem_charge_page(struct page *page, gfp_t gfp,
1643 int order)
1644{
1645 return 0;
1646}
1647
1648static inline void memcg_kmem_uncharge_page(struct page *page, int order)
1649{
1650}
1651
1652static inline int __memcg_kmem_charge_page(struct page *page, gfp_t gfp,
1653 int order)
1654{
1655 return 0;
1656}
1657
1658static inline void __memcg_kmem_uncharge_page(struct page *page, int order)
1659{
1660}
1661
1662#define for_each_memcg_cache_index(_idx) \
1663 for (; NULL; )
1664
1665static inline bool memcg_kmem_enabled(void)
1666{
1667 return false;
1668}
1669
1670static inline int memcg_cache_id(struct mem_cgroup *memcg)
1671{
1672 return -1;
1673}
1674
1675static inline void memcg_get_cache_ids(void)
1676{
1677}
1678
1679static inline void memcg_put_cache_ids(void)
1680{
1681}
1682
1683static inline struct mem_cgroup *mem_cgroup_from_obj(void *p)
1684{
1685 return NULL;
1686}
1687
1688#endif /* CONFIG_MEMCG_KMEM */
1689
1690#endif /* _LINUX_MEMCONTROL_H */