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