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