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/kernel.h>
18#include <linux/page_counter.h>
19#include <linux/vmpressure.h>
20#include <linux/eventfd.h>
21#include <linux/mm.h>
22#include <linux/vmstat.h>
23#include <linux/writeback.h>
24#include <linux/page-flags.h>
25#include <linux/shrinker.h>
26
27struct mem_cgroup;
28struct obj_cgroup;
29struct page;
30struct mm_struct;
31struct kmem_cache;
32
33/* Cgroup-specific page state, on top of universal node page state */
34enum memcg_stat_item {
35 MEMCG_SWAP = NR_VM_NODE_STAT_ITEMS,
36 MEMCG_SOCK,
37 MEMCG_PERCPU_B,
38 MEMCG_VMALLOC,
39 MEMCG_KMEM,
40 MEMCG_ZSWAP_B,
41 MEMCG_ZSWAPPED,
42 MEMCG_NR_STAT,
43};
44
45enum memcg_memory_event {
46 MEMCG_LOW,
47 MEMCG_HIGH,
48 MEMCG_MAX,
49 MEMCG_OOM,
50 MEMCG_OOM_KILL,
51 MEMCG_OOM_GROUP_KILL,
52 MEMCG_SWAP_HIGH,
53 MEMCG_SWAP_MAX,
54 MEMCG_SWAP_FAIL,
55 MEMCG_NR_MEMORY_EVENTS,
56};
57
58struct mem_cgroup_reclaim_cookie {
59 pg_data_t *pgdat;
60 int generation;
61};
62
63#ifdef CONFIG_MEMCG
64
65#define MEM_CGROUP_ID_SHIFT 16
66
67struct mem_cgroup_id {
68 int id;
69 refcount_t ref;
70};
71
72struct memcg_vmstats_percpu;
73struct memcg1_events_percpu;
74struct memcg_vmstats;
75struct lruvec_stats_percpu;
76struct lruvec_stats;
77
78struct mem_cgroup_reclaim_iter {
79 struct mem_cgroup *position;
80 /* scan generation, increased every round-trip */
81 atomic_t generation;
82};
83
84/*
85 * per-node information in memory controller.
86 */
87struct mem_cgroup_per_node {
88 /* Keep the read-only fields at the start */
89 struct mem_cgroup *memcg; /* Back pointer, we cannot */
90 /* use container_of */
91
92 struct lruvec_stats_percpu __percpu *lruvec_stats_percpu;
93 struct lruvec_stats *lruvec_stats;
94 struct shrinker_info __rcu *shrinker_info;
95
96#ifdef CONFIG_MEMCG_V1
97 /*
98 * Memcg-v1 only stuff in middle as buffer between read mostly fields
99 * and update often fields to avoid false sharing. If v1 stuff is
100 * not present, an explicit padding is needed.
101 */
102
103 struct rb_node tree_node; /* RB tree node */
104 unsigned long usage_in_excess;/* Set to the value by which */
105 /* the soft limit is exceeded*/
106 bool on_tree;
107#else
108 CACHELINE_PADDING(_pad1_);
109#endif
110
111 /* Fields which get updated often at the end. */
112 struct lruvec lruvec;
113 CACHELINE_PADDING(_pad2_);
114 unsigned long lru_zone_size[MAX_NR_ZONES][NR_LRU_LISTS];
115 struct mem_cgroup_reclaim_iter iter;
116
117#ifdef CONFIG_MEMCG_NMI_SAFETY_REQUIRES_ATOMIC
118 /* slab stats for nmi context */
119 atomic_t slab_reclaimable;
120 atomic_t slab_unreclaimable;
121#endif
122};
123
124struct mem_cgroup_threshold {
125 struct eventfd_ctx *eventfd;
126 unsigned long threshold;
127};
128
129/* For threshold */
130struct mem_cgroup_threshold_ary {
131 /* An array index points to threshold just below or equal to usage. */
132 int current_threshold;
133 /* Size of entries[] */
134 unsigned int size;
135 /* Array of thresholds */
136 struct mem_cgroup_threshold entries[] __counted_by(size);
137};
138
139struct mem_cgroup_thresholds {
140 /* Primary thresholds array */
141 struct mem_cgroup_threshold_ary *primary;
142 /*
143 * Spare threshold array.
144 * This is needed to make mem_cgroup_unregister_event() "never fail".
145 * It must be able to store at least primary->size - 1 entries.
146 */
147 struct mem_cgroup_threshold_ary *spare;
148};
149
150/*
151 * Remember four most recent foreign writebacks with dirty pages in this
152 * cgroup. Inode sharing is expected to be uncommon and, even if we miss
153 * one in a given round, we're likely to catch it later if it keeps
154 * foreign-dirtying, so a fairly low count should be enough.
155 *
156 * See mem_cgroup_track_foreign_dirty_slowpath() for details.
157 */
158#define MEMCG_CGWB_FRN_CNT 4
159
160struct memcg_cgwb_frn {
161 u64 bdi_id; /* bdi->id of the foreign inode */
162 int memcg_id; /* memcg->css.id of foreign inode */
163 u64 at; /* jiffies_64 at the time of dirtying */
164 struct wb_completion done; /* tracks in-flight foreign writebacks */
165};
166
167/*
168 * Bucket for arbitrarily byte-sized objects charged to a memory
169 * cgroup. The bucket can be reparented in one piece when the cgroup
170 * is destroyed, without having to round up the individual references
171 * of all live memory objects in the wild.
172 */
173struct obj_cgroup {
174 struct percpu_ref refcnt;
175 struct mem_cgroup *memcg;
176 atomic_t nr_charged_bytes;
177 union {
178 struct list_head list; /* protected by objcg_lock */
179 struct rcu_head rcu;
180 };
181};
182
183/*
184 * The memory controller data structure. The memory controller controls both
185 * page cache and RSS per cgroup. We would eventually like to provide
186 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
187 * to help the administrator determine what knobs to tune.
188 */
189struct mem_cgroup {
190 struct cgroup_subsys_state css;
191
192 /* Private memcg ID. Used to ID objects that outlive the cgroup */
193 struct mem_cgroup_id id;
194
195 /* Accounted resources */
196 struct page_counter memory; /* Both v1 & v2 */
197
198 union {
199 struct page_counter swap; /* v2 only */
200 struct page_counter memsw; /* v1 only */
201 };
202
203 /* registered local peak watchers */
204 struct list_head memory_peaks;
205 struct list_head swap_peaks;
206 spinlock_t peaks_lock;
207
208 /* Range enforcement for interrupt charges */
209 struct work_struct high_work;
210
211#ifdef CONFIG_ZSWAP
212 unsigned long zswap_max;
213
214 /*
215 * Prevent pages from this memcg from being written back from zswap to
216 * swap, and from being swapped out on zswap store failures.
217 */
218 bool zswap_writeback;
219#endif
220
221 /* vmpressure notifications */
222 struct vmpressure vmpressure;
223
224 /*
225 * Should the OOM killer kill all belonging tasks, had it kill one?
226 */
227 bool oom_group;
228
229 int swappiness;
230
231 /* memory.events and memory.events.local */
232 struct cgroup_file events_file;
233 struct cgroup_file events_local_file;
234
235 /* handle for "memory.swap.events" */
236 struct cgroup_file swap_events_file;
237
238 /* memory.stat */
239 struct memcg_vmstats *vmstats;
240
241 /* memory.events */
242 atomic_long_t memory_events[MEMCG_NR_MEMORY_EVENTS];
243 atomic_long_t memory_events_local[MEMCG_NR_MEMORY_EVENTS];
244
245#ifdef CONFIG_MEMCG_NMI_SAFETY_REQUIRES_ATOMIC
246 /* MEMCG_KMEM for nmi context */
247 atomic_t kmem_stat;
248#endif
249 /*
250 * Hint of reclaim pressure for socket memroy management. Note
251 * that this indicator should NOT be used in legacy cgroup mode
252 * where socket memory is accounted/charged separately.
253 */
254 u64 socket_pressure;
255#if BITS_PER_LONG < 64
256 seqlock_t socket_pressure_seqlock;
257#endif
258 int kmemcg_id;
259 /*
260 * memcg->objcg is wiped out as a part of the objcg repaprenting
261 * process. memcg->orig_objcg preserves a pointer (and a reference)
262 * to the original objcg until the end of live of memcg.
263 */
264 struct obj_cgroup __rcu *objcg;
265 struct obj_cgroup *orig_objcg;
266 /* list of inherited objcgs, protected by objcg_lock */
267 struct list_head objcg_list;
268
269 struct memcg_vmstats_percpu __percpu *vmstats_percpu;
270
271#ifdef CONFIG_CGROUP_WRITEBACK
272 struct list_head cgwb_list;
273 struct wb_domain cgwb_domain;
274 struct memcg_cgwb_frn cgwb_frn[MEMCG_CGWB_FRN_CNT];
275#endif
276
277#ifdef CONFIG_TRANSPARENT_HUGEPAGE
278 struct deferred_split deferred_split_queue;
279#endif
280
281#ifdef CONFIG_LRU_GEN_WALKS_MMU
282 /* per-memcg mm_struct list */
283 struct lru_gen_mm_list mm_list;
284#endif
285
286#ifdef CONFIG_MEMCG_V1
287 /* Legacy consumer-oriented counters */
288 struct page_counter kmem; /* v1 only */
289 struct page_counter tcpmem; /* v1 only */
290
291 struct memcg1_events_percpu __percpu *events_percpu;
292
293 unsigned long soft_limit;
294
295 /* protected by memcg_oom_lock */
296 bool oom_lock;
297 int under_oom;
298
299 /* OOM-Killer disable */
300 int oom_kill_disable;
301
302 /* protect arrays of thresholds */
303 struct mutex thresholds_lock;
304
305 /* thresholds for memory usage. RCU-protected */
306 struct mem_cgroup_thresholds thresholds;
307
308 /* thresholds for mem+swap usage. RCU-protected */
309 struct mem_cgroup_thresholds memsw_thresholds;
310
311 /* For oom notifier event fd */
312 struct list_head oom_notify;
313
314 /* Legacy tcp memory accounting */
315 bool tcpmem_active;
316 int tcpmem_pressure;
317
318 /* List of events which userspace want to receive */
319 struct list_head event_list;
320 spinlock_t event_list_lock;
321#endif /* CONFIG_MEMCG_V1 */
322
323 struct mem_cgroup_per_node *nodeinfo[];
324};
325
326/*
327 * size of first charge trial.
328 * TODO: maybe necessary to use big numbers in big irons or dynamic based of the
329 * workload.
330 */
331#define MEMCG_CHARGE_BATCH 64U
332
333extern struct mem_cgroup *root_mem_cgroup;
334
335enum page_memcg_data_flags {
336 /* page->memcg_data is a pointer to an slabobj_ext vector */
337 MEMCG_DATA_OBJEXTS = (1UL << 0),
338 /* page has been accounted as a non-slab kernel page */
339 MEMCG_DATA_KMEM = (1UL << 1),
340 /* the next bit after the last actual flag */
341 __NR_MEMCG_DATA_FLAGS = (1UL << 2),
342};
343
344#define __OBJEXTS_ALLOC_FAIL MEMCG_DATA_OBJEXTS
345#define __FIRST_OBJEXT_FLAG __NR_MEMCG_DATA_FLAGS
346
347#else /* CONFIG_MEMCG */
348
349#define __OBJEXTS_ALLOC_FAIL (1UL << 0)
350#define __FIRST_OBJEXT_FLAG (1UL << 0)
351
352#endif /* CONFIG_MEMCG */
353
354enum objext_flags {
355 /*
356 * Use bit 0 with zero other bits to signal that slabobj_ext vector
357 * failed to allocate. The same bit 0 with valid upper bits means
358 * MEMCG_DATA_OBJEXTS.
359 */
360 OBJEXTS_ALLOC_FAIL = __OBJEXTS_ALLOC_FAIL,
361 /* slabobj_ext vector allocated with kmalloc_nolock() */
362 OBJEXTS_NOSPIN_ALLOC = __FIRST_OBJEXT_FLAG,
363 /* the next bit after the last actual flag */
364 __NR_OBJEXTS_FLAGS = (__FIRST_OBJEXT_FLAG << 1),
365};
366
367#define OBJEXTS_FLAGS_MASK (__NR_OBJEXTS_FLAGS - 1)
368
369#ifdef CONFIG_MEMCG
370
371static inline bool folio_memcg_kmem(struct folio *folio);
372
373/*
374 * After the initialization objcg->memcg is always pointing at
375 * a valid memcg, but can be atomically swapped to the parent memcg.
376 *
377 * The caller must ensure that the returned memcg won't be released.
378 */
379static inline struct mem_cgroup *obj_cgroup_memcg(struct obj_cgroup *objcg)
380{
381 lockdep_assert_once(rcu_read_lock_held() || lockdep_is_held(&cgroup_mutex));
382 return READ_ONCE(objcg->memcg);
383}
384
385/*
386 * __folio_memcg - Get the memory cgroup associated with a non-kmem folio
387 * @folio: Pointer to the folio.
388 *
389 * Returns a pointer to the memory cgroup associated with the folio,
390 * or NULL. This function assumes that the folio is known to have a
391 * proper memory cgroup pointer. It's not safe to call this function
392 * against some type of folios, e.g. slab folios or ex-slab folios or
393 * kmem folios.
394 */
395static inline struct mem_cgroup *__folio_memcg(struct folio *folio)
396{
397 unsigned long memcg_data = folio->memcg_data;
398
399 VM_BUG_ON_FOLIO(folio_test_slab(folio), folio);
400 VM_BUG_ON_FOLIO(memcg_data & MEMCG_DATA_OBJEXTS, folio);
401 VM_BUG_ON_FOLIO(memcg_data & MEMCG_DATA_KMEM, folio);
402
403 return (struct mem_cgroup *)(memcg_data & ~OBJEXTS_FLAGS_MASK);
404}
405
406/*
407 * __folio_objcg - get the object cgroup associated with a kmem folio.
408 * @folio: Pointer to the folio.
409 *
410 * Returns a pointer to the object cgroup associated with the folio,
411 * or NULL. This function assumes that the folio is known to have a
412 * proper object cgroup pointer. It's not safe to call this function
413 * against some type of folios, e.g. slab folios or ex-slab folios or
414 * LRU folios.
415 */
416static inline struct obj_cgroup *__folio_objcg(struct folio *folio)
417{
418 unsigned long memcg_data = folio->memcg_data;
419
420 VM_BUG_ON_FOLIO(folio_test_slab(folio), folio);
421 VM_BUG_ON_FOLIO(memcg_data & MEMCG_DATA_OBJEXTS, folio);
422 VM_BUG_ON_FOLIO(!(memcg_data & MEMCG_DATA_KMEM), folio);
423
424 return (struct obj_cgroup *)(memcg_data & ~OBJEXTS_FLAGS_MASK);
425}
426
427/*
428 * folio_memcg - Get the memory cgroup associated with a folio.
429 * @folio: Pointer to the folio.
430 *
431 * Returns a pointer to the memory cgroup associated with the folio,
432 * or NULL. This function assumes that the folio is known to have a
433 * proper memory cgroup pointer. It's not safe to call this function
434 * against some type of folios, e.g. slab folios or ex-slab folios.
435 *
436 * For a non-kmem folio any of the following ensures folio and memcg binding
437 * stability:
438 *
439 * - the folio lock
440 * - LRU isolation
441 * - exclusive reference
442 *
443 * For a kmem folio a caller should hold an rcu read lock to protect memcg
444 * associated with a kmem folio from being released.
445 */
446static inline struct mem_cgroup *folio_memcg(struct folio *folio)
447{
448 if (folio_memcg_kmem(folio))
449 return obj_cgroup_memcg(__folio_objcg(folio));
450 return __folio_memcg(folio);
451}
452
453/*
454 * folio_memcg_charged - If a folio is charged to a memory cgroup.
455 * @folio: Pointer to the folio.
456 *
457 * Returns true if folio is charged to a memory cgroup, otherwise returns false.
458 */
459static inline bool folio_memcg_charged(struct folio *folio)
460{
461 return folio->memcg_data != 0;
462}
463
464/*
465 * folio_memcg_check - Get the memory cgroup associated with a folio.
466 * @folio: Pointer to the folio.
467 *
468 * Returns a pointer to the memory cgroup associated with the folio,
469 * or NULL. This function unlike folio_memcg() can take any folio
470 * as an argument. It has to be used in cases when it's not known if a folio
471 * has an associated memory cgroup pointer or an object cgroups vector or
472 * an object cgroup.
473 *
474 * For a non-kmem folio any of the following ensures folio and memcg binding
475 * stability:
476 *
477 * - the folio lock
478 * - LRU isolation
479 * - exclusive reference
480 *
481 * For a kmem folio a caller should hold an rcu read lock to protect memcg
482 * associated with a kmem folio from being released.
483 */
484static inline struct mem_cgroup *folio_memcg_check(struct folio *folio)
485{
486 /*
487 * Because folio->memcg_data might be changed asynchronously
488 * for slabs, READ_ONCE() should be used here.
489 */
490 unsigned long memcg_data = READ_ONCE(folio->memcg_data);
491
492 if (memcg_data & MEMCG_DATA_OBJEXTS)
493 return NULL;
494
495 if (memcg_data & MEMCG_DATA_KMEM) {
496 struct obj_cgroup *objcg;
497
498 objcg = (void *)(memcg_data & ~OBJEXTS_FLAGS_MASK);
499 return obj_cgroup_memcg(objcg);
500 }
501
502 return (struct mem_cgroup *)(memcg_data & ~OBJEXTS_FLAGS_MASK);
503}
504
505static inline struct mem_cgroup *page_memcg_check(struct page *page)
506{
507 if (PageTail(page))
508 return NULL;
509 return folio_memcg_check((struct folio *)page);
510}
511
512static inline struct mem_cgroup *get_mem_cgroup_from_objcg(struct obj_cgroup *objcg)
513{
514 struct mem_cgroup *memcg;
515
516 rcu_read_lock();
517retry:
518 memcg = obj_cgroup_memcg(objcg);
519 if (unlikely(!css_tryget(&memcg->css)))
520 goto retry;
521 rcu_read_unlock();
522
523 return memcg;
524}
525
526/*
527 * folio_memcg_kmem - Check if the folio has the memcg_kmem flag set.
528 * @folio: Pointer to the folio.
529 *
530 * Checks if the folio has MemcgKmem flag set. The caller must ensure
531 * that the folio has an associated memory cgroup. It's not safe to call
532 * this function against some types of folios, e.g. slab folios.
533 */
534static inline bool folio_memcg_kmem(struct folio *folio)
535{
536 VM_BUG_ON_PGFLAGS(PageTail(&folio->page), &folio->page);
537 VM_BUG_ON_FOLIO(folio->memcg_data & MEMCG_DATA_OBJEXTS, folio);
538 return folio->memcg_data & MEMCG_DATA_KMEM;
539}
540
541static inline bool PageMemcgKmem(struct page *page)
542{
543 return folio_memcg_kmem(page_folio(page));
544}
545
546static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
547{
548 return (memcg == root_mem_cgroup);
549}
550
551static inline bool mem_cgroup_disabled(void)
552{
553 return !cgroup_subsys_enabled(memory_cgrp_subsys);
554}
555
556static inline void mem_cgroup_protection(struct mem_cgroup *root,
557 struct mem_cgroup *memcg,
558 unsigned long *min,
559 unsigned long *low)
560{
561 *min = *low = 0;
562
563 if (mem_cgroup_disabled())
564 return;
565
566 /*
567 * There is no reclaim protection applied to a targeted reclaim.
568 * We are special casing this specific case here because
569 * mem_cgroup_calculate_protection is not robust enough to keep
570 * the protection invariant for calculated effective values for
571 * parallel reclaimers with different reclaim target. This is
572 * especially a problem for tail memcgs (as they have pages on LRU)
573 * which would want to have effective values 0 for targeted reclaim
574 * but a different value for external reclaim.
575 *
576 * Example
577 * Let's have global and A's reclaim in parallel:
578 * |
579 * A (low=2G, usage = 3G, max = 3G, children_low_usage = 1.5G)
580 * |\
581 * | C (low = 1G, usage = 2.5G)
582 * B (low = 1G, usage = 0.5G)
583 *
584 * For the global reclaim
585 * A.elow = A.low
586 * B.elow = min(B.usage, B.low) because children_low_usage <= A.elow
587 * C.elow = min(C.usage, C.low)
588 *
589 * With the effective values resetting we have A reclaim
590 * A.elow = 0
591 * B.elow = B.low
592 * C.elow = C.low
593 *
594 * If the global reclaim races with A's reclaim then
595 * B.elow = C.elow = 0 because children_low_usage > A.elow)
596 * is possible and reclaiming B would be violating the protection.
597 *
598 */
599 if (root == memcg)
600 return;
601
602 *min = READ_ONCE(memcg->memory.emin);
603 *low = READ_ONCE(memcg->memory.elow);
604}
605
606void mem_cgroup_calculate_protection(struct mem_cgroup *root,
607 struct mem_cgroup *memcg);
608
609static inline bool mem_cgroup_unprotected(struct mem_cgroup *target,
610 struct mem_cgroup *memcg)
611{
612 /*
613 * The root memcg doesn't account charges, and doesn't support
614 * protection. The target memcg's protection is ignored, see
615 * mem_cgroup_calculate_protection() and mem_cgroup_protection()
616 */
617 return mem_cgroup_disabled() || mem_cgroup_is_root(memcg) ||
618 memcg == target;
619}
620
621static inline bool mem_cgroup_below_low(struct mem_cgroup *target,
622 struct mem_cgroup *memcg)
623{
624 if (mem_cgroup_unprotected(target, memcg))
625 return false;
626
627 return READ_ONCE(memcg->memory.elow) >=
628 page_counter_read(&memcg->memory);
629}
630
631static inline bool mem_cgroup_below_min(struct mem_cgroup *target,
632 struct mem_cgroup *memcg)
633{
634 if (mem_cgroup_unprotected(target, memcg))
635 return false;
636
637 return READ_ONCE(memcg->memory.emin) >=
638 page_counter_read(&memcg->memory);
639}
640
641int __mem_cgroup_charge(struct folio *folio, struct mm_struct *mm, gfp_t gfp);
642
643/**
644 * mem_cgroup_charge - Charge a newly allocated folio to a cgroup.
645 * @folio: Folio to charge.
646 * @mm: mm context of the allocating task.
647 * @gfp: Reclaim mode.
648 *
649 * Try to charge @folio to the memcg that @mm belongs to, reclaiming
650 * pages according to @gfp if necessary. If @mm is NULL, try to
651 * charge to the active memcg.
652 *
653 * Do not use this for folios allocated for swapin.
654 *
655 * Return: 0 on success. Otherwise, an error code is returned.
656 */
657static inline int mem_cgroup_charge(struct folio *folio, struct mm_struct *mm,
658 gfp_t gfp)
659{
660 if (mem_cgroup_disabled())
661 return 0;
662 return __mem_cgroup_charge(folio, mm, gfp);
663}
664
665int mem_cgroup_charge_hugetlb(struct folio* folio, gfp_t gfp);
666
667int mem_cgroup_swapin_charge_folio(struct folio *folio, struct mm_struct *mm,
668 gfp_t gfp, swp_entry_t entry);
669
670void __mem_cgroup_uncharge(struct folio *folio);
671
672/**
673 * mem_cgroup_uncharge - Uncharge a folio.
674 * @folio: Folio to uncharge.
675 *
676 * Uncharge a folio previously charged with mem_cgroup_charge().
677 */
678static inline void mem_cgroup_uncharge(struct folio *folio)
679{
680 if (mem_cgroup_disabled())
681 return;
682 __mem_cgroup_uncharge(folio);
683}
684
685void __mem_cgroup_uncharge_folios(struct folio_batch *folios);
686static inline void mem_cgroup_uncharge_folios(struct folio_batch *folios)
687{
688 if (mem_cgroup_disabled())
689 return;
690 __mem_cgroup_uncharge_folios(folios);
691}
692
693void mem_cgroup_replace_folio(struct folio *old, struct folio *new);
694void mem_cgroup_migrate(struct folio *old, struct folio *new);
695
696/**
697 * mem_cgroup_lruvec - get the lru list vector for a memcg & node
698 * @memcg: memcg of the wanted lruvec
699 * @pgdat: pglist_data
700 *
701 * Returns the lru list vector holding pages for a given @memcg &
702 * @pgdat combination. This can be the node lruvec, if the memory
703 * controller is disabled.
704 */
705static inline struct lruvec *mem_cgroup_lruvec(struct mem_cgroup *memcg,
706 struct pglist_data *pgdat)
707{
708 struct mem_cgroup_per_node *mz;
709 struct lruvec *lruvec;
710
711 if (mem_cgroup_disabled()) {
712 lruvec = &pgdat->__lruvec;
713 goto out;
714 }
715
716 if (!memcg)
717 memcg = root_mem_cgroup;
718
719 mz = memcg->nodeinfo[pgdat->node_id];
720 lruvec = &mz->lruvec;
721out:
722 /*
723 * Since a node can be onlined after the mem_cgroup was created,
724 * we have to be prepared to initialize lruvec->pgdat here;
725 * and if offlined then reonlined, we need to reinitialize it.
726 */
727 if (unlikely(lruvec->pgdat != pgdat))
728 lruvec->pgdat = pgdat;
729 return lruvec;
730}
731
732/**
733 * folio_lruvec - return lruvec for isolating/putting an LRU folio
734 * @folio: Pointer to the folio.
735 *
736 * This function relies on folio->mem_cgroup being stable.
737 */
738static inline struct lruvec *folio_lruvec(struct folio *folio)
739{
740 struct mem_cgroup *memcg = folio_memcg(folio);
741
742 VM_WARN_ON_ONCE_FOLIO(!memcg && !mem_cgroup_disabled(), folio);
743 return mem_cgroup_lruvec(memcg, folio_pgdat(folio));
744}
745
746struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
747
748struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm);
749
750struct mem_cgroup *get_mem_cgroup_from_current(void);
751
752struct mem_cgroup *get_mem_cgroup_from_folio(struct folio *folio);
753
754struct lruvec *folio_lruvec_lock(struct folio *folio);
755struct lruvec *folio_lruvec_lock_irq(struct folio *folio);
756struct lruvec *folio_lruvec_lock_irqsave(struct folio *folio,
757 unsigned long *flags);
758
759#ifdef CONFIG_DEBUG_VM
760void lruvec_memcg_debug(struct lruvec *lruvec, struct folio *folio);
761#else
762static inline
763void lruvec_memcg_debug(struct lruvec *lruvec, struct folio *folio)
764{
765}
766#endif
767
768static inline
769struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
770 return css ? container_of(css, struct mem_cgroup, css) : NULL;
771}
772
773static inline bool obj_cgroup_tryget(struct obj_cgroup *objcg)
774{
775 return percpu_ref_tryget(&objcg->refcnt);
776}
777
778static inline void obj_cgroup_get(struct obj_cgroup *objcg)
779{
780 percpu_ref_get(&objcg->refcnt);
781}
782
783static inline void obj_cgroup_get_many(struct obj_cgroup *objcg,
784 unsigned long nr)
785{
786 percpu_ref_get_many(&objcg->refcnt, nr);
787}
788
789static inline void obj_cgroup_put(struct obj_cgroup *objcg)
790{
791 if (objcg)
792 percpu_ref_put(&objcg->refcnt);
793}
794
795static inline bool mem_cgroup_tryget(struct mem_cgroup *memcg)
796{
797 return !memcg || css_tryget(&memcg->css);
798}
799
800static inline bool mem_cgroup_tryget_online(struct mem_cgroup *memcg)
801{
802 return !memcg || css_tryget_online(&memcg->css);
803}
804
805static inline void mem_cgroup_put(struct mem_cgroup *memcg)
806{
807 if (memcg)
808 css_put(&memcg->css);
809}
810
811#define mem_cgroup_from_counter(counter, member) \
812 container_of(counter, struct mem_cgroup, member)
813
814struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
815 struct mem_cgroup *,
816 struct mem_cgroup_reclaim_cookie *);
817void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
818void mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
819 int (*)(struct task_struct *, void *), void *arg);
820
821static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
822{
823 if (mem_cgroup_disabled())
824 return 0;
825
826 return memcg->id.id;
827}
828struct mem_cgroup *mem_cgroup_from_id(unsigned short id);
829
830#ifdef CONFIG_SHRINKER_DEBUG
831static inline unsigned long mem_cgroup_ino(struct mem_cgroup *memcg)
832{
833 return memcg ? cgroup_ino(memcg->css.cgroup) : 0;
834}
835
836struct mem_cgroup *mem_cgroup_get_from_ino(unsigned long ino);
837#endif
838
839static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m)
840{
841 return mem_cgroup_from_css(seq_css(m));
842}
843
844static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec)
845{
846 struct mem_cgroup_per_node *mz;
847
848 if (mem_cgroup_disabled())
849 return NULL;
850
851 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
852 return mz->memcg;
853}
854
855/**
856 * parent_mem_cgroup - find the accounting parent of a memcg
857 * @memcg: memcg whose parent to find
858 *
859 * Returns the parent memcg, or NULL if this is the root.
860 */
861static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
862{
863 return mem_cgroup_from_css(memcg->css.parent);
864}
865
866static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
867 struct mem_cgroup *root)
868{
869 if (root == memcg)
870 return true;
871 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
872}
873
874static inline bool mm_match_cgroup(struct mm_struct *mm,
875 struct mem_cgroup *memcg)
876{
877 struct mem_cgroup *task_memcg;
878 bool match = false;
879
880 rcu_read_lock();
881 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
882 if (task_memcg)
883 match = mem_cgroup_is_descendant(task_memcg, memcg);
884 rcu_read_unlock();
885 return match;
886}
887
888struct cgroup_subsys_state *mem_cgroup_css_from_folio(struct folio *folio);
889ino_t page_cgroup_ino(struct page *page);
890
891static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
892{
893 if (mem_cgroup_disabled())
894 return true;
895 return !!(memcg->css.flags & CSS_ONLINE);
896}
897
898void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
899 int zid, int nr_pages);
900
901static inline
902unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
903 enum lru_list lru, int zone_idx)
904{
905 struct mem_cgroup_per_node *mz;
906
907 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
908 return READ_ONCE(mz->lru_zone_size[zone_idx][lru]);
909}
910
911void __mem_cgroup_handle_over_high(gfp_t gfp_mask);
912
913static inline void mem_cgroup_handle_over_high(gfp_t gfp_mask)
914{
915 if (unlikely(current->memcg_nr_pages_over_high))
916 __mem_cgroup_handle_over_high(gfp_mask);
917}
918
919unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg);
920
921unsigned long mem_cgroup_size(struct mem_cgroup *memcg);
922
923void mem_cgroup_print_oom_context(struct mem_cgroup *memcg,
924 struct task_struct *p);
925
926void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg);
927
928struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim,
929 struct mem_cgroup *oom_domain);
930void mem_cgroup_print_oom_group(struct mem_cgroup *memcg);
931
932/* idx can be of type enum memcg_stat_item or node_stat_item */
933void mod_memcg_state(struct mem_cgroup *memcg,
934 enum memcg_stat_item idx, int val);
935
936static inline void mod_memcg_page_state(struct page *page,
937 enum memcg_stat_item idx, int val)
938{
939 struct mem_cgroup *memcg;
940
941 if (mem_cgroup_disabled())
942 return;
943
944 rcu_read_lock();
945 memcg = folio_memcg(page_folio(page));
946 if (memcg)
947 mod_memcg_state(memcg, idx, val);
948 rcu_read_unlock();
949}
950
951unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx);
952unsigned long lruvec_page_state(struct lruvec *lruvec, enum node_stat_item idx);
953unsigned long lruvec_page_state_local(struct lruvec *lruvec,
954 enum node_stat_item idx);
955
956void mem_cgroup_flush_stats(struct mem_cgroup *memcg);
957void mem_cgroup_flush_stats_ratelimited(struct mem_cgroup *memcg);
958
959void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, int val);
960
961static inline void mod_lruvec_kmem_state(void *p, enum node_stat_item idx,
962 int val)
963{
964 unsigned long flags;
965
966 local_irq_save(flags);
967 __mod_lruvec_kmem_state(p, idx, val);
968 local_irq_restore(flags);
969}
970
971void count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx,
972 unsigned long count);
973
974static inline void count_memcg_folio_events(struct folio *folio,
975 enum vm_event_item idx, unsigned long nr)
976{
977 struct mem_cgroup *memcg = folio_memcg(folio);
978
979 if (memcg)
980 count_memcg_events(memcg, idx, nr);
981}
982
983static inline void count_memcg_events_mm(struct mm_struct *mm,
984 enum vm_event_item idx, unsigned long count)
985{
986 struct mem_cgroup *memcg;
987
988 if (mem_cgroup_disabled())
989 return;
990
991 rcu_read_lock();
992 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
993 if (likely(memcg))
994 count_memcg_events(memcg, idx, count);
995 rcu_read_unlock();
996}
997
998static inline void count_memcg_event_mm(struct mm_struct *mm,
999 enum vm_event_item idx)
1000{
1001 count_memcg_events_mm(mm, idx, 1);
1002}
1003
1004static inline void __memcg_memory_event(struct mem_cgroup *memcg,
1005 enum memcg_memory_event event,
1006 bool allow_spinning)
1007{
1008 bool swap_event = event == MEMCG_SWAP_HIGH || event == MEMCG_SWAP_MAX ||
1009 event == MEMCG_SWAP_FAIL;
1010
1011 /* For now only MEMCG_MAX can happen with !allow_spinning context. */
1012 VM_WARN_ON_ONCE(!allow_spinning && event != MEMCG_MAX);
1013
1014 atomic_long_inc(&memcg->memory_events_local[event]);
1015 if (!swap_event && allow_spinning)
1016 cgroup_file_notify(&memcg->events_local_file);
1017
1018 do {
1019 atomic_long_inc(&memcg->memory_events[event]);
1020 if (allow_spinning) {
1021 if (swap_event)
1022 cgroup_file_notify(&memcg->swap_events_file);
1023 else
1024 cgroup_file_notify(&memcg->events_file);
1025 }
1026
1027 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
1028 break;
1029 if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
1030 break;
1031 } while ((memcg = parent_mem_cgroup(memcg)) &&
1032 !mem_cgroup_is_root(memcg));
1033}
1034
1035static inline void memcg_memory_event(struct mem_cgroup *memcg,
1036 enum memcg_memory_event event)
1037{
1038 __memcg_memory_event(memcg, event, true);
1039}
1040
1041static inline void memcg_memory_event_mm(struct mm_struct *mm,
1042 enum memcg_memory_event event)
1043{
1044 struct mem_cgroup *memcg;
1045
1046 if (mem_cgroup_disabled())
1047 return;
1048
1049 rcu_read_lock();
1050 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1051 if (likely(memcg))
1052 memcg_memory_event(memcg, event);
1053 rcu_read_unlock();
1054}
1055
1056void split_page_memcg(struct page *first, unsigned order);
1057void folio_split_memcg_refs(struct folio *folio, unsigned old_order,
1058 unsigned new_order);
1059
1060static inline u64 cgroup_id_from_mm(struct mm_struct *mm)
1061{
1062 struct mem_cgroup *memcg;
1063 u64 id;
1064
1065 if (mem_cgroup_disabled())
1066 return 0;
1067
1068 rcu_read_lock();
1069 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1070 if (!memcg)
1071 memcg = root_mem_cgroup;
1072 id = cgroup_id(memcg->css.cgroup);
1073 rcu_read_unlock();
1074 return id;
1075}
1076
1077extern int mem_cgroup_init(void);
1078#else /* CONFIG_MEMCG */
1079
1080#define MEM_CGROUP_ID_SHIFT 0
1081
1082#define root_mem_cgroup (NULL)
1083
1084static inline struct mem_cgroup *folio_memcg(struct folio *folio)
1085{
1086 return NULL;
1087}
1088
1089static inline bool folio_memcg_charged(struct folio *folio)
1090{
1091 return false;
1092}
1093
1094static inline struct mem_cgroup *folio_memcg_check(struct folio *folio)
1095{
1096 return NULL;
1097}
1098
1099static inline struct mem_cgroup *page_memcg_check(struct page *page)
1100{
1101 return NULL;
1102}
1103
1104static inline struct mem_cgroup *get_mem_cgroup_from_objcg(struct obj_cgroup *objcg)
1105{
1106 return NULL;
1107}
1108
1109static inline bool folio_memcg_kmem(struct folio *folio)
1110{
1111 return false;
1112}
1113
1114static inline bool PageMemcgKmem(struct page *page)
1115{
1116 return false;
1117}
1118
1119static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
1120{
1121 return true;
1122}
1123
1124static inline bool mem_cgroup_disabled(void)
1125{
1126 return true;
1127}
1128
1129static inline void memcg_memory_event(struct mem_cgroup *memcg,
1130 enum memcg_memory_event event)
1131{
1132}
1133
1134static inline void memcg_memory_event_mm(struct mm_struct *mm,
1135 enum memcg_memory_event event)
1136{
1137}
1138
1139static inline void mem_cgroup_protection(struct mem_cgroup *root,
1140 struct mem_cgroup *memcg,
1141 unsigned long *min,
1142 unsigned long *low)
1143{
1144 *min = *low = 0;
1145}
1146
1147static inline void mem_cgroup_calculate_protection(struct mem_cgroup *root,
1148 struct mem_cgroup *memcg)
1149{
1150}
1151
1152static inline bool mem_cgroup_unprotected(struct mem_cgroup *target,
1153 struct mem_cgroup *memcg)
1154{
1155 return true;
1156}
1157static inline bool mem_cgroup_below_low(struct mem_cgroup *target,
1158 struct mem_cgroup *memcg)
1159{
1160 return false;
1161}
1162
1163static inline bool mem_cgroup_below_min(struct mem_cgroup *target,
1164 struct mem_cgroup *memcg)
1165{
1166 return false;
1167}
1168
1169static inline int mem_cgroup_charge(struct folio *folio,
1170 struct mm_struct *mm, gfp_t gfp)
1171{
1172 return 0;
1173}
1174
1175static inline int mem_cgroup_charge_hugetlb(struct folio* folio, gfp_t gfp)
1176{
1177 return 0;
1178}
1179
1180static inline int mem_cgroup_swapin_charge_folio(struct folio *folio,
1181 struct mm_struct *mm, gfp_t gfp, swp_entry_t entry)
1182{
1183 return 0;
1184}
1185
1186static inline void mem_cgroup_uncharge(struct folio *folio)
1187{
1188}
1189
1190static inline void mem_cgroup_uncharge_folios(struct folio_batch *folios)
1191{
1192}
1193
1194static inline void mem_cgroup_replace_folio(struct folio *old,
1195 struct folio *new)
1196{
1197}
1198
1199static inline void mem_cgroup_migrate(struct folio *old, struct folio *new)
1200{
1201}
1202
1203static inline struct lruvec *mem_cgroup_lruvec(struct mem_cgroup *memcg,
1204 struct pglist_data *pgdat)
1205{
1206 return &pgdat->__lruvec;
1207}
1208
1209static inline struct lruvec *folio_lruvec(struct folio *folio)
1210{
1211 struct pglist_data *pgdat = folio_pgdat(folio);
1212 return &pgdat->__lruvec;
1213}
1214
1215static inline
1216void lruvec_memcg_debug(struct lruvec *lruvec, struct folio *folio)
1217{
1218}
1219
1220static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
1221{
1222 return NULL;
1223}
1224
1225static inline bool mm_match_cgroup(struct mm_struct *mm,
1226 struct mem_cgroup *memcg)
1227{
1228 return true;
1229}
1230
1231static inline struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
1232{
1233 return NULL;
1234}
1235
1236static inline struct mem_cgroup *get_mem_cgroup_from_current(void)
1237{
1238 return NULL;
1239}
1240
1241static inline struct mem_cgroup *get_mem_cgroup_from_folio(struct folio *folio)
1242{
1243 return NULL;
1244}
1245
1246static inline
1247struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css)
1248{
1249 return NULL;
1250}
1251
1252static inline void obj_cgroup_get(struct obj_cgroup *objcg)
1253{
1254}
1255
1256static inline void obj_cgroup_put(struct obj_cgroup *objcg)
1257{
1258}
1259
1260static inline bool mem_cgroup_tryget(struct mem_cgroup *memcg)
1261{
1262 return true;
1263}
1264
1265static inline bool mem_cgroup_tryget_online(struct mem_cgroup *memcg)
1266{
1267 return true;
1268}
1269
1270static inline void mem_cgroup_put(struct mem_cgroup *memcg)
1271{
1272}
1273
1274static inline struct lruvec *folio_lruvec_lock(struct folio *folio)
1275{
1276 struct pglist_data *pgdat = folio_pgdat(folio);
1277
1278 spin_lock(lock: &pgdat->__lruvec.lru_lock);
1279 return &pgdat->__lruvec;
1280}
1281
1282static inline struct lruvec *folio_lruvec_lock_irq(struct folio *folio)
1283{
1284 struct pglist_data *pgdat = folio_pgdat(folio);
1285
1286 spin_lock_irq(lock: &pgdat->__lruvec.lru_lock);
1287 return &pgdat->__lruvec;
1288}
1289
1290static inline struct lruvec *folio_lruvec_lock_irqsave(struct folio *folio,
1291 unsigned long *flagsp)
1292{
1293 struct pglist_data *pgdat = folio_pgdat(folio);
1294
1295 spin_lock_irqsave(&pgdat->__lruvec.lru_lock, *flagsp);
1296 return &pgdat->__lruvec;
1297}
1298
1299static inline struct mem_cgroup *
1300mem_cgroup_iter(struct mem_cgroup *root,
1301 struct mem_cgroup *prev,
1302 struct mem_cgroup_reclaim_cookie *reclaim)
1303{
1304 return NULL;
1305}
1306
1307static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
1308 struct mem_cgroup *prev)
1309{
1310}
1311
1312static inline void mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
1313 int (*fn)(struct task_struct *, void *), void *arg)
1314{
1315}
1316
1317static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
1318{
1319 return 0;
1320}
1321
1322static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
1323{
1324 WARN_ON_ONCE(id);
1325 /* XXX: This should always return root_mem_cgroup */
1326 return NULL;
1327}
1328
1329#ifdef CONFIG_SHRINKER_DEBUG
1330static inline unsigned long mem_cgroup_ino(struct mem_cgroup *memcg)
1331{
1332 return 0;
1333}
1334
1335static inline struct mem_cgroup *mem_cgroup_get_from_ino(unsigned long ino)
1336{
1337 return NULL;
1338}
1339#endif
1340
1341static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m)
1342{
1343 return NULL;
1344}
1345
1346static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec)
1347{
1348 return NULL;
1349}
1350
1351static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
1352{
1353 return true;
1354}
1355
1356static inline
1357unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
1358 enum lru_list lru, int zone_idx)
1359{
1360 return 0;
1361}
1362
1363static inline unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg)
1364{
1365 return 0;
1366}
1367
1368static inline unsigned long mem_cgroup_size(struct mem_cgroup *memcg)
1369{
1370 return 0;
1371}
1372
1373static inline void
1374mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p)
1375{
1376}
1377
1378static inline void
1379mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg)
1380{
1381}
1382
1383static inline void mem_cgroup_handle_over_high(gfp_t gfp_mask)
1384{
1385}
1386
1387static inline struct mem_cgroup *mem_cgroup_get_oom_group(
1388 struct task_struct *victim, struct mem_cgroup *oom_domain)
1389{
1390 return NULL;
1391}
1392
1393static inline void mem_cgroup_print_oom_group(struct mem_cgroup *memcg)
1394{
1395}
1396
1397static inline void mod_memcg_state(struct mem_cgroup *memcg,
1398 enum memcg_stat_item idx,
1399 int nr)
1400{
1401}
1402
1403static inline void mod_memcg_page_state(struct page *page,
1404 enum memcg_stat_item idx, int val)
1405{
1406}
1407
1408static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx)
1409{
1410 return 0;
1411}
1412
1413static inline unsigned long lruvec_page_state(struct lruvec *lruvec,
1414 enum node_stat_item idx)
1415{
1416 return node_page_state(pgdat: lruvec_pgdat(lruvec), item: idx);
1417}
1418
1419static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec,
1420 enum node_stat_item idx)
1421{
1422 return node_page_state(pgdat: lruvec_pgdat(lruvec), item: idx);
1423}
1424
1425static inline void mem_cgroup_flush_stats(struct mem_cgroup *memcg)
1426{
1427}
1428
1429static inline void mem_cgroup_flush_stats_ratelimited(struct mem_cgroup *memcg)
1430{
1431}
1432
1433static inline void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx,
1434 int val)
1435{
1436 struct page *page = virt_to_head_page(x: p);
1437
1438 __mod_node_page_state(page_pgdat(page), item: idx, val);
1439}
1440
1441static inline void mod_lruvec_kmem_state(void *p, enum node_stat_item idx,
1442 int val)
1443{
1444 struct page *page = virt_to_head_page(x: p);
1445
1446 mod_node_page_state(page_pgdat(page), idx, val);
1447}
1448
1449static inline void count_memcg_events(struct mem_cgroup *memcg,
1450 enum vm_event_item idx,
1451 unsigned long count)
1452{
1453}
1454
1455static inline void count_memcg_folio_events(struct folio *folio,
1456 enum vm_event_item idx, unsigned long nr)
1457{
1458}
1459
1460static inline void count_memcg_events_mm(struct mm_struct *mm,
1461 enum vm_event_item idx, unsigned long count)
1462{
1463}
1464
1465static inline
1466void count_memcg_event_mm(struct mm_struct *mm, enum vm_event_item idx)
1467{
1468}
1469
1470static inline void split_page_memcg(struct page *first, unsigned order)
1471{
1472}
1473
1474static inline void folio_split_memcg_refs(struct folio *folio,
1475 unsigned old_order, unsigned new_order)
1476{
1477}
1478
1479static inline u64 cgroup_id_from_mm(struct mm_struct *mm)
1480{
1481 return 0;
1482}
1483
1484static inline int mem_cgroup_init(void) { return 0; }
1485#endif /* CONFIG_MEMCG */
1486
1487/*
1488 * Extended information for slab objects stored as an array in page->memcg_data
1489 * if MEMCG_DATA_OBJEXTS is set.
1490 */
1491struct slabobj_ext {
1492#ifdef CONFIG_MEMCG
1493 struct obj_cgroup *objcg;
1494#endif
1495#ifdef CONFIG_MEM_ALLOC_PROFILING
1496 union codetag_ref ref;
1497#endif
1498} __aligned(8);
1499
1500static inline void __inc_lruvec_kmem_state(void *p, enum node_stat_item idx)
1501{
1502 __mod_lruvec_kmem_state(p, idx, val: 1);
1503}
1504
1505static inline void __dec_lruvec_kmem_state(void *p, enum node_stat_item idx)
1506{
1507 __mod_lruvec_kmem_state(p, idx, val: -1);
1508}
1509
1510static inline struct lruvec *parent_lruvec(struct lruvec *lruvec)
1511{
1512 struct mem_cgroup *memcg;
1513
1514 memcg = lruvec_memcg(lruvec);
1515 if (!memcg)
1516 return NULL;
1517 memcg = parent_mem_cgroup(memcg);
1518 if (!memcg)
1519 return NULL;
1520 return mem_cgroup_lruvec(memcg, pgdat: lruvec_pgdat(lruvec));
1521}
1522
1523static inline void unlock_page_lruvec(struct lruvec *lruvec)
1524{
1525 spin_unlock(lock: &lruvec->lru_lock);
1526}
1527
1528static inline void unlock_page_lruvec_irq(struct lruvec *lruvec)
1529{
1530 spin_unlock_irq(lock: &lruvec->lru_lock);
1531}
1532
1533static inline void unlock_page_lruvec_irqrestore(struct lruvec *lruvec,
1534 unsigned long flags)
1535{
1536 spin_unlock_irqrestore(lock: &lruvec->lru_lock, flags);
1537}
1538
1539/* Test requires a stable folio->memcg binding, see folio_memcg() */
1540static inline bool folio_matches_lruvec(struct folio *folio,
1541 struct lruvec *lruvec)
1542{
1543 return lruvec_pgdat(lruvec) == folio_pgdat(folio) &&
1544 lruvec_memcg(lruvec) == folio_memcg(folio);
1545}
1546
1547/* Don't lock again iff page's lruvec locked */
1548static inline struct lruvec *folio_lruvec_relock_irq(struct folio *folio,
1549 struct lruvec *locked_lruvec)
1550{
1551 if (locked_lruvec) {
1552 if (folio_matches_lruvec(folio, lruvec: locked_lruvec))
1553 return locked_lruvec;
1554
1555 unlock_page_lruvec_irq(lruvec: locked_lruvec);
1556 }
1557
1558 return folio_lruvec_lock_irq(folio);
1559}
1560
1561/* Don't lock again iff folio's lruvec locked */
1562static inline void folio_lruvec_relock_irqsave(struct folio *folio,
1563 struct lruvec **lruvecp, unsigned long *flags)
1564{
1565 if (*lruvecp) {
1566 if (folio_matches_lruvec(folio, lruvec: *lruvecp))
1567 return;
1568
1569 unlock_page_lruvec_irqrestore(lruvec: *lruvecp, flags: *flags);
1570 }
1571
1572 *lruvecp = folio_lruvec_lock_irqsave(folio, flagsp: flags);
1573}
1574
1575#ifdef CONFIG_CGROUP_WRITEBACK
1576
1577struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
1578void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
1579 unsigned long *pheadroom, unsigned long *pdirty,
1580 unsigned long *pwriteback);
1581
1582void mem_cgroup_track_foreign_dirty_slowpath(struct folio *folio,
1583 struct bdi_writeback *wb);
1584
1585static inline void mem_cgroup_track_foreign_dirty(struct folio *folio,
1586 struct bdi_writeback *wb)
1587{
1588 struct mem_cgroup *memcg;
1589
1590 if (mem_cgroup_disabled())
1591 return;
1592
1593 memcg = folio_memcg(folio);
1594 if (unlikely(memcg && &memcg->css != wb->memcg_css))
1595 mem_cgroup_track_foreign_dirty_slowpath(folio, wb);
1596}
1597
1598void mem_cgroup_flush_foreign(struct bdi_writeback *wb);
1599
1600#else /* CONFIG_CGROUP_WRITEBACK */
1601
1602static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
1603{
1604 return NULL;
1605}
1606
1607static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
1608 unsigned long *pfilepages,
1609 unsigned long *pheadroom,
1610 unsigned long *pdirty,
1611 unsigned long *pwriteback)
1612{
1613}
1614
1615static inline void mem_cgroup_track_foreign_dirty(struct folio *folio,
1616 struct bdi_writeback *wb)
1617{
1618}
1619
1620static inline void mem_cgroup_flush_foreign(struct bdi_writeback *wb)
1621{
1622}
1623
1624#endif /* CONFIG_CGROUP_WRITEBACK */
1625
1626struct sock;
1627#ifdef CONFIG_MEMCG
1628extern struct static_key_false memcg_sockets_enabled_key;
1629#define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_enabled_key)
1630
1631void mem_cgroup_sk_alloc(struct sock *sk);
1632void mem_cgroup_sk_free(struct sock *sk);
1633void mem_cgroup_sk_inherit(const struct sock *sk, struct sock *newsk);
1634bool mem_cgroup_sk_charge(const struct sock *sk, unsigned int nr_pages,
1635 gfp_t gfp_mask);
1636void mem_cgroup_sk_uncharge(const struct sock *sk, unsigned int nr_pages);
1637
1638#if BITS_PER_LONG < 64
1639static inline void mem_cgroup_set_socket_pressure(struct mem_cgroup *memcg)
1640{
1641 u64 val = get_jiffies_64() + HZ;
1642 unsigned long flags;
1643
1644 write_seqlock_irqsave(&memcg->socket_pressure_seqlock, flags);
1645 memcg->socket_pressure = val;
1646 write_sequnlock_irqrestore(&memcg->socket_pressure_seqlock, flags);
1647}
1648
1649static inline u64 mem_cgroup_get_socket_pressure(struct mem_cgroup *memcg)
1650{
1651 unsigned int seq;
1652 u64 val;
1653
1654 do {
1655 seq = read_seqbegin(&memcg->socket_pressure_seqlock);
1656 val = memcg->socket_pressure;
1657 } while (read_seqretry(&memcg->socket_pressure_seqlock, seq));
1658
1659 return val;
1660}
1661#else
1662static inline void mem_cgroup_set_socket_pressure(struct mem_cgroup *memcg)
1663{
1664 WRITE_ONCE(memcg->socket_pressure, jiffies + HZ);
1665}
1666
1667static inline u64 mem_cgroup_get_socket_pressure(struct mem_cgroup *memcg)
1668{
1669 return READ_ONCE(memcg->socket_pressure);
1670}
1671#endif
1672
1673int alloc_shrinker_info(struct mem_cgroup *memcg);
1674void free_shrinker_info(struct mem_cgroup *memcg);
1675void set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id);
1676void reparent_shrinker_deferred(struct mem_cgroup *memcg);
1677#else
1678#define mem_cgroup_sockets_enabled 0
1679
1680static inline void mem_cgroup_sk_alloc(struct sock *sk)
1681{
1682}
1683
1684static inline void mem_cgroup_sk_free(struct sock *sk)
1685{
1686}
1687
1688static inline void mem_cgroup_sk_inherit(const struct sock *sk, struct sock *newsk)
1689{
1690}
1691
1692static inline bool mem_cgroup_sk_charge(const struct sock *sk,
1693 unsigned int nr_pages,
1694 gfp_t gfp_mask)
1695{
1696 return false;
1697}
1698
1699static inline void mem_cgroup_sk_uncharge(const struct sock *sk,
1700 unsigned int nr_pages)
1701{
1702}
1703
1704static inline void set_shrinker_bit(struct mem_cgroup *memcg,
1705 int nid, int shrinker_id)
1706{
1707}
1708#endif
1709
1710#ifdef CONFIG_MEMCG
1711bool mem_cgroup_kmem_disabled(void);
1712int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order);
1713void __memcg_kmem_uncharge_page(struct page *page, int order);
1714
1715/*
1716 * The returned objcg pointer is safe to use without additional
1717 * protection within a scope. The scope is defined either by
1718 * the current task (similar to the "current" global variable)
1719 * or by set_active_memcg() pair.
1720 * Please, use obj_cgroup_get() to get a reference if the pointer
1721 * needs to be used outside of the local scope.
1722 */
1723struct obj_cgroup *current_obj_cgroup(void);
1724struct obj_cgroup *get_obj_cgroup_from_folio(struct folio *folio);
1725
1726static inline struct obj_cgroup *get_obj_cgroup_from_current(void)
1727{
1728 struct obj_cgroup *objcg = current_obj_cgroup();
1729
1730 if (objcg)
1731 obj_cgroup_get(objcg);
1732
1733 return objcg;
1734}
1735
1736int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size);
1737void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size);
1738
1739extern struct static_key_false memcg_bpf_enabled_key;
1740static inline bool memcg_bpf_enabled(void)
1741{
1742 return static_branch_likely(&memcg_bpf_enabled_key);
1743}
1744
1745extern struct static_key_false memcg_kmem_online_key;
1746
1747static inline bool memcg_kmem_online(void)
1748{
1749 return static_branch_likely(&memcg_kmem_online_key);
1750}
1751
1752static inline int memcg_kmem_charge_page(struct page *page, gfp_t gfp,
1753 int order)
1754{
1755 if (memcg_kmem_online())
1756 return __memcg_kmem_charge_page(page, gfp, order);
1757 return 0;
1758}
1759
1760static inline void memcg_kmem_uncharge_page(struct page *page, int order)
1761{
1762 if (memcg_kmem_online())
1763 __memcg_kmem_uncharge_page(page, order);
1764}
1765
1766/*
1767 * A helper for accessing memcg's kmem_id, used for getting
1768 * corresponding LRU lists.
1769 */
1770static inline int memcg_kmem_id(struct mem_cgroup *memcg)
1771{
1772 return memcg ? memcg->kmemcg_id : -1;
1773}
1774
1775struct mem_cgroup *mem_cgroup_from_slab_obj(void *p);
1776
1777static inline void count_objcg_events(struct obj_cgroup *objcg,
1778 enum vm_event_item idx,
1779 unsigned long count)
1780{
1781 struct mem_cgroup *memcg;
1782
1783 if (!memcg_kmem_online())
1784 return;
1785
1786 rcu_read_lock();
1787 memcg = obj_cgroup_memcg(objcg);
1788 count_memcg_events(memcg, idx, count);
1789 rcu_read_unlock();
1790}
1791
1792bool mem_cgroup_node_allowed(struct mem_cgroup *memcg, int nid);
1793
1794#else
1795static inline bool mem_cgroup_kmem_disabled(void)
1796{
1797 return true;
1798}
1799
1800static inline int memcg_kmem_charge_page(struct page *page, gfp_t gfp,
1801 int order)
1802{
1803 return 0;
1804}
1805
1806static inline void memcg_kmem_uncharge_page(struct page *page, int order)
1807{
1808}
1809
1810static inline int __memcg_kmem_charge_page(struct page *page, gfp_t gfp,
1811 int order)
1812{
1813 return 0;
1814}
1815
1816static inline void __memcg_kmem_uncharge_page(struct page *page, int order)
1817{
1818}
1819
1820static inline struct obj_cgroup *get_obj_cgroup_from_folio(struct folio *folio)
1821{
1822 return NULL;
1823}
1824
1825static inline bool memcg_bpf_enabled(void)
1826{
1827 return false;
1828}
1829
1830static inline bool memcg_kmem_online(void)
1831{
1832 return false;
1833}
1834
1835static inline int memcg_kmem_id(struct mem_cgroup *memcg)
1836{
1837 return -1;
1838}
1839
1840static inline struct mem_cgroup *mem_cgroup_from_slab_obj(void *p)
1841{
1842 return NULL;
1843}
1844
1845static inline void count_objcg_events(struct obj_cgroup *objcg,
1846 enum vm_event_item idx,
1847 unsigned long count)
1848{
1849}
1850
1851static inline ino_t page_cgroup_ino(struct page *page)
1852{
1853 return 0;
1854}
1855
1856static inline bool mem_cgroup_node_allowed(struct mem_cgroup *memcg, int nid)
1857{
1858 return true;
1859}
1860#endif /* CONFIG_MEMCG */
1861
1862#if defined(CONFIG_MEMCG) && defined(CONFIG_ZSWAP)
1863bool obj_cgroup_may_zswap(struct obj_cgroup *objcg);
1864void obj_cgroup_charge_zswap(struct obj_cgroup *objcg, size_t size);
1865void obj_cgroup_uncharge_zswap(struct obj_cgroup *objcg, size_t size);
1866bool mem_cgroup_zswap_writeback_enabled(struct mem_cgroup *memcg);
1867#else
1868static inline bool obj_cgroup_may_zswap(struct obj_cgroup *objcg)
1869{
1870 return true;
1871}
1872static inline void obj_cgroup_charge_zswap(struct obj_cgroup *objcg,
1873 size_t size)
1874{
1875}
1876static inline void obj_cgroup_uncharge_zswap(struct obj_cgroup *objcg,
1877 size_t size)
1878{
1879}
1880static inline bool mem_cgroup_zswap_writeback_enabled(struct mem_cgroup *memcg)
1881{
1882 /* if zswap is disabled, do not block pages going to the swapping device */
1883 return true;
1884}
1885#endif
1886
1887
1888/* Cgroup v1-related declarations */
1889
1890#ifdef CONFIG_MEMCG_V1
1891unsigned long memcg1_soft_limit_reclaim(pg_data_t *pgdat, int order,
1892 gfp_t gfp_mask,
1893 unsigned long *total_scanned);
1894
1895bool mem_cgroup_oom_synchronize(bool wait);
1896
1897static inline bool task_in_memcg_oom(struct task_struct *p)
1898{
1899 return p->memcg_in_oom;
1900}
1901
1902static inline void mem_cgroup_enter_user_fault(void)
1903{
1904 WARN_ON(current->in_user_fault);
1905 current->in_user_fault = 1;
1906}
1907
1908static inline void mem_cgroup_exit_user_fault(void)
1909{
1910 WARN_ON(!current->in_user_fault);
1911 current->in_user_fault = 0;
1912}
1913
1914void memcg1_swapout(struct folio *folio, swp_entry_t entry);
1915void memcg1_swapin(swp_entry_t entry, unsigned int nr_pages);
1916
1917#else /* CONFIG_MEMCG_V1 */
1918static inline
1919unsigned long memcg1_soft_limit_reclaim(pg_data_t *pgdat, int order,
1920 gfp_t gfp_mask,
1921 unsigned long *total_scanned)
1922{
1923 return 0;
1924}
1925
1926static inline bool task_in_memcg_oom(struct task_struct *p)
1927{
1928 return false;
1929}
1930
1931static inline bool mem_cgroup_oom_synchronize(bool wait)
1932{
1933 return false;
1934}
1935
1936static inline void mem_cgroup_enter_user_fault(void)
1937{
1938}
1939
1940static inline void mem_cgroup_exit_user_fault(void)
1941{
1942}
1943
1944static inline void memcg1_swapout(struct folio *folio, swp_entry_t entry)
1945{
1946}
1947
1948static inline void memcg1_swapin(swp_entry_t entry, unsigned int nr_pages)
1949{
1950}
1951
1952#endif /* CONFIG_MEMCG_V1 */
1953
1954#endif /* _LINUX_MEMCONTROL_H */
1955