1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* memcontrol.h - Memory Controller 3 * 4 * Copyright IBM Corporation, 2007 5 * Author Balbir Singh <[email protected]> 6 * 7 * Copyright 2007 OpenVZ SWsoft Inc 8 * Author: Pavel Emelianov <[email protected]> 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 27 struct mem_cgroup; 28 struct obj_cgroup; 29 struct page; 30 struct mm_struct; 31 struct kmem_cache; 32 33 /* Cgroup-specific page state, on top of universal node page state */ 34 enum 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 45 enum 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 58 struct mem_cgroup_reclaim_cookie { 59 pg_data_t *pgdat; 60 unsigned int generation; 61 }; 62 63 #ifdef CONFIG_MEMCG 64 65 #define MEM_CGROUP_ID_SHIFT 16 66 67 struct mem_cgroup_id { 68 int id; 69 refcount_t ref; 70 }; 71 72 struct memcg_vmstats_percpu; 73 struct memcg1_events_percpu; 74 struct memcg_vmstats; 75 struct lruvec_stats_percpu; 76 struct lruvec_stats; 77 78 struct mem_cgroup_reclaim_iter { 79 struct mem_cgroup *position; 80 /* scan generation, increased every round-trip */ 81 unsigned int generation; 82 }; 83 84 /* 85 * per-node information in memory controller. 86 */ 87 struct 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 }; 117 118 struct mem_cgroup_threshold { 119 struct eventfd_ctx *eventfd; 120 unsigned long threshold; 121 }; 122 123 /* For threshold */ 124 struct 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 }; 132 133 struct 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 4 153 154 struct 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 */ 167 struct 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 */ 183 struct 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 struct memcg1_events_percpu __percpu *events_percpu; 259 260 #ifdef CONFIG_CGROUP_WRITEBACK 261 struct list_head cgwb_list; 262 struct wb_domain cgwb_domain; 263 struct memcg_cgwb_frn cgwb_frn[MEMCG_CGWB_FRN_CNT]; 264 #endif 265 266 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 267 struct deferred_split deferred_split_queue; 268 #endif 269 270 #ifdef CONFIG_LRU_GEN_WALKS_MMU 271 /* per-memcg mm_struct list */ 272 struct lru_gen_mm_list mm_list; 273 #endif 274 275 #ifdef CONFIG_MEMCG_V1 276 /* Legacy consumer-oriented counters */ 277 struct page_counter kmem; /* v1 only */ 278 struct page_counter tcpmem; /* v1 only */ 279 280 unsigned long soft_limit; 281 282 /* protected by memcg_oom_lock */ 283 bool oom_lock; 284 int under_oom; 285 286 /* OOM-Killer disable */ 287 int oom_kill_disable; 288 289 /* protect arrays of thresholds */ 290 struct mutex thresholds_lock; 291 292 /* thresholds for memory usage. RCU-protected */ 293 struct mem_cgroup_thresholds thresholds; 294 295 /* thresholds for mem+swap usage. RCU-protected */ 296 struct mem_cgroup_thresholds memsw_thresholds; 297 298 /* For oom notifier event fd */ 299 struct list_head oom_notify; 300 301 /* 302 * Should we move charges of a task when a task is moved into this 303 * mem_cgroup ? And what type of charges should we move ? 304 */ 305 unsigned long move_charge_at_immigrate; 306 /* taken only while moving_account > 0 */ 307 spinlock_t move_lock; 308 unsigned long move_lock_flags; 309 310 /* Legacy tcp memory accounting */ 311 bool tcpmem_active; 312 int tcpmem_pressure; 313 314 /* 315 * set > 0 if pages under this cgroup are moving to other cgroup. 316 */ 317 atomic_t moving_account; 318 struct task_struct *move_lock_task; 319 320 /* List of events which userspace want to receive */ 321 struct list_head event_list; 322 spinlock_t event_list_lock; 323 #endif /* CONFIG_MEMCG_V1 */ 324 325 struct mem_cgroup_per_node *nodeinfo[]; 326 }; 327 328 /* 329 * size of first charge trial. 330 * TODO: maybe necessary to use big numbers in big irons or dynamic based of the 331 * workload. 332 */ 333 #define MEMCG_CHARGE_BATCH 64U 334 335 extern struct mem_cgroup *root_mem_cgroup; 336 337 enum page_memcg_data_flags { 338 /* page->memcg_data is a pointer to an slabobj_ext vector */ 339 MEMCG_DATA_OBJEXTS = (1UL << 0), 340 /* page has been accounted as a non-slab kernel page */ 341 MEMCG_DATA_KMEM = (1UL << 1), 342 /* the next bit after the last actual flag */ 343 __NR_MEMCG_DATA_FLAGS = (1UL << 2), 344 }; 345 346 #define __FIRST_OBJEXT_FLAG __NR_MEMCG_DATA_FLAGS 347 348 #else /* CONFIG_MEMCG */ 349 350 #define __FIRST_OBJEXT_FLAG (1UL << 0) 351 352 #endif /* CONFIG_MEMCG */ 353 354 enum objext_flags { 355 /* slabobj_ext vector failed to allocate */ 356 OBJEXTS_ALLOC_FAIL = __FIRST_OBJEXT_FLAG, 357 /* the next bit after the last actual flag */ 358 __NR_OBJEXTS_FLAGS = (__FIRST_OBJEXT_FLAG << 1), 359 }; 360 361 #define OBJEXTS_FLAGS_MASK (__NR_OBJEXTS_FLAGS - 1) 362 363 #ifdef CONFIG_MEMCG 364 365 static inline bool folio_memcg_kmem(struct folio *folio); 366 367 /* 368 * After the initialization objcg->memcg is always pointing at 369 * a valid memcg, but can be atomically swapped to the parent memcg. 370 * 371 * The caller must ensure that the returned memcg won't be released. 372 */ 373 static inline struct mem_cgroup *obj_cgroup_memcg(struct obj_cgroup *objcg) 374 { 375 lockdep_assert_once(rcu_read_lock_held() || lockdep_is_held(&cgroup_mutex)); 376 return READ_ONCE(objcg->memcg); 377 } 378 379 /* 380 * __folio_memcg - Get the memory cgroup associated with a non-kmem folio 381 * @folio: Pointer to the folio. 382 * 383 * Returns a pointer to the memory cgroup associated with the folio, 384 * or NULL. This function assumes that the folio is known to have a 385 * proper memory cgroup pointer. It's not safe to call this function 386 * against some type of folios, e.g. slab folios or ex-slab folios or 387 * kmem folios. 388 */ 389 static inline struct mem_cgroup *__folio_memcg(struct folio *folio) 390 { 391 unsigned long memcg_data = folio->memcg_data; 392 393 VM_BUG_ON_FOLIO(folio_test_slab(folio), folio); 394 VM_BUG_ON_FOLIO(memcg_data & MEMCG_DATA_OBJEXTS, folio); 395 VM_BUG_ON_FOLIO(memcg_data & MEMCG_DATA_KMEM, folio); 396 397 return (struct mem_cgroup *)(memcg_data & ~OBJEXTS_FLAGS_MASK); 398 } 399 400 /* 401 * __folio_objcg - get the object cgroup associated with a kmem folio. 402 * @folio: Pointer to the folio. 403 * 404 * Returns a pointer to the object cgroup associated with the folio, 405 * or NULL. This function assumes that the folio is known to have a 406 * proper object cgroup pointer. It's not safe to call this function 407 * against some type of folios, e.g. slab folios or ex-slab folios or 408 * LRU folios. 409 */ 410 static inline struct obj_cgroup *__folio_objcg(struct folio *folio) 411 { 412 unsigned long memcg_data = folio->memcg_data; 413 414 VM_BUG_ON_FOLIO(folio_test_slab(folio), folio); 415 VM_BUG_ON_FOLIO(memcg_data & MEMCG_DATA_OBJEXTS, folio); 416 VM_BUG_ON_FOLIO(!(memcg_data & MEMCG_DATA_KMEM), folio); 417 418 return (struct obj_cgroup *)(memcg_data & ~OBJEXTS_FLAGS_MASK); 419 } 420 421 /* 422 * folio_memcg - Get the memory cgroup associated with a folio. 423 * @folio: Pointer to the folio. 424 * 425 * Returns a pointer to the memory cgroup associated with the folio, 426 * or NULL. This function assumes that the folio is known to have a 427 * proper memory cgroup pointer. It's not safe to call this function 428 * against some type of folios, e.g. slab folios or ex-slab folios. 429 * 430 * For a non-kmem folio any of the following ensures folio and memcg binding 431 * stability: 432 * 433 * - the folio lock 434 * - LRU isolation 435 * - folio_memcg_lock() 436 * - exclusive reference 437 * - mem_cgroup_trylock_pages() 438 * 439 * For a kmem folio a caller should hold an rcu read lock to protect memcg 440 * associated with a kmem folio from being released. 441 */ 442 static inline struct mem_cgroup *folio_memcg(struct folio *folio) 443 { 444 if (folio_memcg_kmem(folio)) 445 return obj_cgroup_memcg(__folio_objcg(folio)); 446 return __folio_memcg(folio); 447 } 448 449 /* 450 * folio_memcg_charged - If a folio is charged to a memory cgroup. 451 * @folio: Pointer to the folio. 452 * 453 * Returns true if folio is charged to a memory cgroup, otherwise returns false. 454 */ 455 static inline bool folio_memcg_charged(struct folio *folio) 456 { 457 if (folio_memcg_kmem(folio)) 458 return __folio_objcg(folio) != NULL; 459 return __folio_memcg(folio) != NULL; 460 } 461 462 /** 463 * folio_memcg_rcu - Locklessly get the memory cgroup associated with a folio. 464 * @folio: Pointer to the folio. 465 * 466 * This function assumes that the folio is known to have a 467 * proper memory cgroup pointer. It's not safe to call this function 468 * against some type of folios, e.g. slab folios or ex-slab folios. 469 * 470 * Return: A pointer to the memory cgroup associated with the folio, 471 * or NULL. 472 */ 473 static inline struct mem_cgroup *folio_memcg_rcu(struct folio *folio) 474 { 475 unsigned long memcg_data = READ_ONCE(folio->memcg_data); 476 477 VM_BUG_ON_FOLIO(folio_test_slab(folio), folio); 478 479 if (memcg_data & MEMCG_DATA_KMEM) { 480 struct obj_cgroup *objcg; 481 482 objcg = (void *)(memcg_data & ~OBJEXTS_FLAGS_MASK); 483 return obj_cgroup_memcg(objcg); 484 } 485 486 WARN_ON_ONCE(!rcu_read_lock_held()); 487 488 return (struct mem_cgroup *)(memcg_data & ~OBJEXTS_FLAGS_MASK); 489 } 490 491 /* 492 * folio_memcg_check - Get the memory cgroup associated with a folio. 493 * @folio: Pointer to the folio. 494 * 495 * Returns a pointer to the memory cgroup associated with the folio, 496 * or NULL. This function unlike folio_memcg() can take any folio 497 * as an argument. It has to be used in cases when it's not known if a folio 498 * has an associated memory cgroup pointer or an object cgroups vector or 499 * an object cgroup. 500 * 501 * For a non-kmem folio any of the following ensures folio and memcg binding 502 * stability: 503 * 504 * - the folio lock 505 * - LRU isolation 506 * - lock_folio_memcg() 507 * - exclusive reference 508 * - mem_cgroup_trylock_pages() 509 * 510 * For a kmem folio a caller should hold an rcu read lock to protect memcg 511 * associated with a kmem folio from being released. 512 */ 513 static inline struct mem_cgroup *folio_memcg_check(struct folio *folio) 514 { 515 /* 516 * Because folio->memcg_data might be changed asynchronously 517 * for slabs, READ_ONCE() should be used here. 518 */ 519 unsigned long memcg_data = READ_ONCE(folio->memcg_data); 520 521 if (memcg_data & MEMCG_DATA_OBJEXTS) 522 return NULL; 523 524 if (memcg_data & MEMCG_DATA_KMEM) { 525 struct obj_cgroup *objcg; 526 527 objcg = (void *)(memcg_data & ~OBJEXTS_FLAGS_MASK); 528 return obj_cgroup_memcg(objcg); 529 } 530 531 return (struct mem_cgroup *)(memcg_data & ~OBJEXTS_FLAGS_MASK); 532 } 533 534 static inline struct mem_cgroup *page_memcg_check(struct page *page) 535 { 536 if (PageTail(page)) 537 return NULL; 538 return folio_memcg_check((struct folio *)page); 539 } 540 541 static inline struct mem_cgroup *get_mem_cgroup_from_objcg(struct obj_cgroup *objcg) 542 { 543 struct mem_cgroup *memcg; 544 545 rcu_read_lock(); 546 retry: 547 memcg = obj_cgroup_memcg(objcg); 548 if (unlikely(!css_tryget(&memcg->css))) 549 goto retry; 550 rcu_read_unlock(); 551 552 return memcg; 553 } 554 555 /* 556 * folio_memcg_kmem - Check if the folio has the memcg_kmem flag set. 557 * @folio: Pointer to the folio. 558 * 559 * Checks if the folio has MemcgKmem flag set. The caller must ensure 560 * that the folio has an associated memory cgroup. It's not safe to call 561 * this function against some types of folios, e.g. slab folios. 562 */ 563 static inline bool folio_memcg_kmem(struct folio *folio) 564 { 565 VM_BUG_ON_PGFLAGS(PageTail(&folio->page), &folio->page); 566 VM_BUG_ON_FOLIO(folio->memcg_data & MEMCG_DATA_OBJEXTS, folio); 567 return folio->memcg_data & MEMCG_DATA_KMEM; 568 } 569 570 static inline bool PageMemcgKmem(struct page *page) 571 { 572 return folio_memcg_kmem(page_folio(page)); 573 } 574 575 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) 576 { 577 return (memcg == root_mem_cgroup); 578 } 579 580 static inline bool mem_cgroup_disabled(void) 581 { 582 return !cgroup_subsys_enabled(memory_cgrp_subsys); 583 } 584 585 static inline void mem_cgroup_protection(struct mem_cgroup *root, 586 struct mem_cgroup *memcg, 587 unsigned long *min, 588 unsigned long *low) 589 { 590 *min = *low = 0; 591 592 if (mem_cgroup_disabled()) 593 return; 594 595 /* 596 * There is no reclaim protection applied to a targeted reclaim. 597 * We are special casing this specific case here because 598 * mem_cgroup_calculate_protection is not robust enough to keep 599 * the protection invariant for calculated effective values for 600 * parallel reclaimers with different reclaim target. This is 601 * especially a problem for tail memcgs (as they have pages on LRU) 602 * which would want to have effective values 0 for targeted reclaim 603 * but a different value for external reclaim. 604 * 605 * Example 606 * Let's have global and A's reclaim in parallel: 607 * | 608 * A (low=2G, usage = 3G, max = 3G, children_low_usage = 1.5G) 609 * |\ 610 * | C (low = 1G, usage = 2.5G) 611 * B (low = 1G, usage = 0.5G) 612 * 613 * For the global reclaim 614 * A.elow = A.low 615 * B.elow = min(B.usage, B.low) because children_low_usage <= A.elow 616 * C.elow = min(C.usage, C.low) 617 * 618 * With the effective values resetting we have A reclaim 619 * A.elow = 0 620 * B.elow = B.low 621 * C.elow = C.low 622 * 623 * If the global reclaim races with A's reclaim then 624 * B.elow = C.elow = 0 because children_low_usage > A.elow) 625 * is possible and reclaiming B would be violating the protection. 626 * 627 */ 628 if (root == memcg) 629 return; 630 631 *min = READ_ONCE(memcg->memory.emin); 632 *low = READ_ONCE(memcg->memory.elow); 633 } 634 635 void mem_cgroup_calculate_protection(struct mem_cgroup *root, 636 struct mem_cgroup *memcg); 637 638 static inline bool mem_cgroup_unprotected(struct mem_cgroup *target, 639 struct mem_cgroup *memcg) 640 { 641 /* 642 * The root memcg doesn't account charges, and doesn't support 643 * protection. The target memcg's protection is ignored, see 644 * mem_cgroup_calculate_protection() and mem_cgroup_protection() 645 */ 646 return mem_cgroup_disabled() || mem_cgroup_is_root(memcg) || 647 memcg == target; 648 } 649 650 static inline bool mem_cgroup_below_low(struct mem_cgroup *target, 651 struct mem_cgroup *memcg) 652 { 653 if (mem_cgroup_unprotected(target, memcg)) 654 return false; 655 656 return READ_ONCE(memcg->memory.elow) >= 657 page_counter_read(&memcg->memory); 658 } 659 660 static inline bool mem_cgroup_below_min(struct mem_cgroup *target, 661 struct mem_cgroup *memcg) 662 { 663 if (mem_cgroup_unprotected(target, memcg)) 664 return false; 665 666 return READ_ONCE(memcg->memory.emin) >= 667 page_counter_read(&memcg->memory); 668 } 669 670 void mem_cgroup_commit_charge(struct folio *folio, struct mem_cgroup *memcg); 671 672 int __mem_cgroup_charge(struct folio *folio, struct mm_struct *mm, gfp_t gfp); 673 674 /** 675 * mem_cgroup_charge - Charge a newly allocated folio to a cgroup. 676 * @folio: Folio to charge. 677 * @mm: mm context of the allocating task. 678 * @gfp: Reclaim mode. 679 * 680 * Try to charge @folio to the memcg that @mm belongs to, reclaiming 681 * pages according to @gfp if necessary. If @mm is NULL, try to 682 * charge to the active memcg. 683 * 684 * Do not use this for folios allocated for swapin. 685 * 686 * Return: 0 on success. Otherwise, an error code is returned. 687 */ 688 static inline int mem_cgroup_charge(struct folio *folio, struct mm_struct *mm, 689 gfp_t gfp) 690 { 691 if (mem_cgroup_disabled()) 692 return 0; 693 return __mem_cgroup_charge(folio, mm, gfp); 694 } 695 696 int mem_cgroup_hugetlb_try_charge(struct mem_cgroup *memcg, gfp_t gfp, 697 long nr_pages); 698 699 int mem_cgroup_swapin_charge_folio(struct folio *folio, struct mm_struct *mm, 700 gfp_t gfp, swp_entry_t entry); 701 void mem_cgroup_swapin_uncharge_swap(swp_entry_t entry); 702 703 void __mem_cgroup_uncharge(struct folio *folio); 704 705 /** 706 * mem_cgroup_uncharge - Uncharge a folio. 707 * @folio: Folio to uncharge. 708 * 709 * Uncharge a folio previously charged with mem_cgroup_charge(). 710 */ 711 static inline void mem_cgroup_uncharge(struct folio *folio) 712 { 713 if (mem_cgroup_disabled()) 714 return; 715 __mem_cgroup_uncharge(folio); 716 } 717 718 void __mem_cgroup_uncharge_folios(struct folio_batch *folios); 719 static inline void mem_cgroup_uncharge_folios(struct folio_batch *folios) 720 { 721 if (mem_cgroup_disabled()) 722 return; 723 __mem_cgroup_uncharge_folios(folios); 724 } 725 726 void mem_cgroup_cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages); 727 void mem_cgroup_replace_folio(struct folio *old, struct folio *new); 728 void mem_cgroup_migrate(struct folio *old, struct folio *new); 729 730 /** 731 * mem_cgroup_lruvec - get the lru list vector for a memcg & node 732 * @memcg: memcg of the wanted lruvec 733 * @pgdat: pglist_data 734 * 735 * Returns the lru list vector holding pages for a given @memcg & 736 * @pgdat combination. This can be the node lruvec, if the memory 737 * controller is disabled. 738 */ 739 static inline struct lruvec *mem_cgroup_lruvec(struct mem_cgroup *memcg, 740 struct pglist_data *pgdat) 741 { 742 struct mem_cgroup_per_node *mz; 743 struct lruvec *lruvec; 744 745 if (mem_cgroup_disabled()) { 746 lruvec = &pgdat->__lruvec; 747 goto out; 748 } 749 750 if (!memcg) 751 memcg = root_mem_cgroup; 752 753 mz = memcg->nodeinfo[pgdat->node_id]; 754 lruvec = &mz->lruvec; 755 out: 756 /* 757 * Since a node can be onlined after the mem_cgroup was created, 758 * we have to be prepared to initialize lruvec->pgdat here; 759 * and if offlined then reonlined, we need to reinitialize it. 760 */ 761 if (unlikely(lruvec->pgdat != pgdat)) 762 lruvec->pgdat = pgdat; 763 return lruvec; 764 } 765 766 /** 767 * folio_lruvec - return lruvec for isolating/putting an LRU folio 768 * @folio: Pointer to the folio. 769 * 770 * This function relies on folio->mem_cgroup being stable. 771 */ 772 static inline struct lruvec *folio_lruvec(struct folio *folio) 773 { 774 struct mem_cgroup *memcg = folio_memcg(folio); 775 776 VM_WARN_ON_ONCE_FOLIO(!memcg && !mem_cgroup_disabled(), folio); 777 return mem_cgroup_lruvec(memcg, folio_pgdat(folio)); 778 } 779 780 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p); 781 782 struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm); 783 784 struct mem_cgroup *get_mem_cgroup_from_current(void); 785 786 struct lruvec *folio_lruvec_lock(struct folio *folio); 787 struct lruvec *folio_lruvec_lock_irq(struct folio *folio); 788 struct lruvec *folio_lruvec_lock_irqsave(struct folio *folio, 789 unsigned long *flags); 790 791 #ifdef CONFIG_DEBUG_VM 792 void lruvec_memcg_debug(struct lruvec *lruvec, struct folio *folio); 793 #else 794 static inline 795 void lruvec_memcg_debug(struct lruvec *lruvec, struct folio *folio) 796 { 797 } 798 #endif 799 800 static inline 801 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){ 802 return css ? container_of(css, struct mem_cgroup, css) : NULL; 803 } 804 805 static inline bool obj_cgroup_tryget(struct obj_cgroup *objcg) 806 { 807 return percpu_ref_tryget(&objcg->refcnt); 808 } 809 810 static inline void obj_cgroup_get(struct obj_cgroup *objcg) 811 { 812 percpu_ref_get(&objcg->refcnt); 813 } 814 815 static inline void obj_cgroup_get_many(struct obj_cgroup *objcg, 816 unsigned long nr) 817 { 818 percpu_ref_get_many(&objcg->refcnt, nr); 819 } 820 821 static inline void obj_cgroup_put(struct obj_cgroup *objcg) 822 { 823 if (objcg) 824 percpu_ref_put(&objcg->refcnt); 825 } 826 827 static inline bool mem_cgroup_tryget(struct mem_cgroup *memcg) 828 { 829 return !memcg || css_tryget(&memcg->css); 830 } 831 832 static inline bool mem_cgroup_tryget_online(struct mem_cgroup *memcg) 833 { 834 return !memcg || css_tryget_online(&memcg->css); 835 } 836 837 static inline void mem_cgroup_put(struct mem_cgroup *memcg) 838 { 839 if (memcg) 840 css_put(&memcg->css); 841 } 842 843 #define mem_cgroup_from_counter(counter, member) \ 844 container_of(counter, struct mem_cgroup, member) 845 846 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *, 847 struct mem_cgroup *, 848 struct mem_cgroup_reclaim_cookie *); 849 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *); 850 void mem_cgroup_scan_tasks(struct mem_cgroup *memcg, 851 int (*)(struct task_struct *, void *), void *arg); 852 853 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) 854 { 855 if (mem_cgroup_disabled()) 856 return 0; 857 858 return memcg->id.id; 859 } 860 struct mem_cgroup *mem_cgroup_from_id(unsigned short id); 861 862 #ifdef CONFIG_SHRINKER_DEBUG 863 static inline unsigned long mem_cgroup_ino(struct mem_cgroup *memcg) 864 { 865 return memcg ? cgroup_ino(memcg->css.cgroup) : 0; 866 } 867 868 struct mem_cgroup *mem_cgroup_get_from_ino(unsigned long ino); 869 #endif 870 871 static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m) 872 { 873 return mem_cgroup_from_css(seq_css(m)); 874 } 875 876 static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec) 877 { 878 struct mem_cgroup_per_node *mz; 879 880 if (mem_cgroup_disabled()) 881 return NULL; 882 883 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 884 return mz->memcg; 885 } 886 887 /** 888 * parent_mem_cgroup - find the accounting parent of a memcg 889 * @memcg: memcg whose parent to find 890 * 891 * Returns the parent memcg, or NULL if this is the root. 892 */ 893 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) 894 { 895 return mem_cgroup_from_css(memcg->css.parent); 896 } 897 898 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg, 899 struct mem_cgroup *root) 900 { 901 if (root == memcg) 902 return true; 903 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup); 904 } 905 906 static inline bool mm_match_cgroup(struct mm_struct *mm, 907 struct mem_cgroup *memcg) 908 { 909 struct mem_cgroup *task_memcg; 910 bool match = false; 911 912 rcu_read_lock(); 913 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); 914 if (task_memcg) 915 match = mem_cgroup_is_descendant(task_memcg, memcg); 916 rcu_read_unlock(); 917 return match; 918 } 919 920 struct cgroup_subsys_state *mem_cgroup_css_from_folio(struct folio *folio); 921 ino_t page_cgroup_ino(struct page *page); 922 923 static inline bool mem_cgroup_online(struct mem_cgroup *memcg) 924 { 925 if (mem_cgroup_disabled()) 926 return true; 927 return !!(memcg->css.flags & CSS_ONLINE); 928 } 929 930 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, 931 int zid, int nr_pages); 932 933 static inline 934 unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec, 935 enum lru_list lru, int zone_idx) 936 { 937 struct mem_cgroup_per_node *mz; 938 939 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 940 return READ_ONCE(mz->lru_zone_size[zone_idx][lru]); 941 } 942 943 void mem_cgroup_handle_over_high(gfp_t gfp_mask); 944 945 unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg); 946 947 unsigned long mem_cgroup_size(struct mem_cgroup *memcg); 948 949 void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, 950 struct task_struct *p); 951 952 void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg); 953 954 struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, 955 struct mem_cgroup *oom_domain); 956 void mem_cgroup_print_oom_group(struct mem_cgroup *memcg); 957 958 void __mod_memcg_state(struct mem_cgroup *memcg, enum memcg_stat_item idx, 959 int val); 960 961 /* idx can be of type enum memcg_stat_item or node_stat_item */ 962 static inline void mod_memcg_state(struct mem_cgroup *memcg, 963 enum memcg_stat_item idx, int val) 964 { 965 unsigned long flags; 966 967 local_irq_save(flags); 968 __mod_memcg_state(memcg, idx, val); 969 local_irq_restore(flags); 970 } 971 972 static inline void mod_memcg_page_state(struct page *page, 973 enum memcg_stat_item idx, int val) 974 { 975 struct mem_cgroup *memcg; 976 977 if (mem_cgroup_disabled()) 978 return; 979 980 rcu_read_lock(); 981 memcg = folio_memcg(page_folio(page)); 982 if (memcg) 983 mod_memcg_state(memcg, idx, val); 984 rcu_read_unlock(); 985 } 986 987 unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx); 988 unsigned long lruvec_page_state(struct lruvec *lruvec, enum node_stat_item idx); 989 unsigned long lruvec_page_state_local(struct lruvec *lruvec, 990 enum node_stat_item idx); 991 992 void mem_cgroup_flush_stats(struct mem_cgroup *memcg); 993 void mem_cgroup_flush_stats_ratelimited(struct mem_cgroup *memcg); 994 995 void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, int val); 996 997 static inline void mod_lruvec_kmem_state(void *p, enum node_stat_item idx, 998 int val) 999 { 1000 unsigned long flags; 1001 1002 local_irq_save(flags); 1003 __mod_lruvec_kmem_state(p, idx, val); 1004 local_irq_restore(flags); 1005 } 1006 1007 void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, 1008 unsigned long count); 1009 1010 static inline void count_memcg_events(struct mem_cgroup *memcg, 1011 enum vm_event_item idx, 1012 unsigned long count) 1013 { 1014 unsigned long flags; 1015 1016 local_irq_save(flags); 1017 __count_memcg_events(memcg, idx, count); 1018 local_irq_restore(flags); 1019 } 1020 1021 static inline void count_memcg_folio_events(struct folio *folio, 1022 enum vm_event_item idx, unsigned long nr) 1023 { 1024 struct mem_cgroup *memcg = folio_memcg(folio); 1025 1026 if (memcg) 1027 count_memcg_events(memcg, idx, nr); 1028 } 1029 1030 static inline void count_memcg_event_mm(struct mm_struct *mm, 1031 enum vm_event_item idx) 1032 { 1033 struct mem_cgroup *memcg; 1034 1035 if (mem_cgroup_disabled()) 1036 return; 1037 1038 rcu_read_lock(); 1039 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); 1040 if (likely(memcg)) 1041 count_memcg_events(memcg, idx, 1); 1042 rcu_read_unlock(); 1043 } 1044 1045 static inline void memcg_memory_event(struct mem_cgroup *memcg, 1046 enum memcg_memory_event event) 1047 { 1048 bool swap_event = event == MEMCG_SWAP_HIGH || event == MEMCG_SWAP_MAX || 1049 event == MEMCG_SWAP_FAIL; 1050 1051 atomic_long_inc(&memcg->memory_events_local[event]); 1052 if (!swap_event) 1053 cgroup_file_notify(&memcg->events_local_file); 1054 1055 do { 1056 atomic_long_inc(&memcg->memory_events[event]); 1057 if (swap_event) 1058 cgroup_file_notify(&memcg->swap_events_file); 1059 else 1060 cgroup_file_notify(&memcg->events_file); 1061 1062 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) 1063 break; 1064 if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS) 1065 break; 1066 } while ((memcg = parent_mem_cgroup(memcg)) && 1067 !mem_cgroup_is_root(memcg)); 1068 } 1069 1070 static inline void memcg_memory_event_mm(struct mm_struct *mm, 1071 enum memcg_memory_event event) 1072 { 1073 struct mem_cgroup *memcg; 1074 1075 if (mem_cgroup_disabled()) 1076 return; 1077 1078 rcu_read_lock(); 1079 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); 1080 if (likely(memcg)) 1081 memcg_memory_event(memcg, event); 1082 rcu_read_unlock(); 1083 } 1084 1085 void split_page_memcg(struct page *head, int old_order, int new_order); 1086 1087 #else /* CONFIG_MEMCG */ 1088 1089 #define MEM_CGROUP_ID_SHIFT 0 1090 1091 static inline struct mem_cgroup *folio_memcg(struct folio *folio) 1092 { 1093 return NULL; 1094 } 1095 1096 static inline struct mem_cgroup *folio_memcg_rcu(struct folio *folio) 1097 { 1098 WARN_ON_ONCE(!rcu_read_lock_held()); 1099 return NULL; 1100 } 1101 1102 static inline struct mem_cgroup *folio_memcg_check(struct folio *folio) 1103 { 1104 return NULL; 1105 } 1106 1107 static inline struct mem_cgroup *page_memcg_check(struct page *page) 1108 { 1109 return NULL; 1110 } 1111 1112 static inline struct mem_cgroup *get_mem_cgroup_from_objcg(struct obj_cgroup *objcg) 1113 { 1114 return NULL; 1115 } 1116 1117 static inline bool folio_memcg_kmem(struct folio *folio) 1118 { 1119 return false; 1120 } 1121 1122 static inline bool PageMemcgKmem(struct page *page) 1123 { 1124 return false; 1125 } 1126 1127 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) 1128 { 1129 return true; 1130 } 1131 1132 static inline bool mem_cgroup_disabled(void) 1133 { 1134 return true; 1135 } 1136 1137 static inline void memcg_memory_event(struct mem_cgroup *memcg, 1138 enum memcg_memory_event event) 1139 { 1140 } 1141 1142 static inline void memcg_memory_event_mm(struct mm_struct *mm, 1143 enum memcg_memory_event event) 1144 { 1145 } 1146 1147 static inline void mem_cgroup_protection(struct mem_cgroup *root, 1148 struct mem_cgroup *memcg, 1149 unsigned long *min, 1150 unsigned long *low) 1151 { 1152 *min = *low = 0; 1153 } 1154 1155 static inline void mem_cgroup_calculate_protection(struct mem_cgroup *root, 1156 struct mem_cgroup *memcg) 1157 { 1158 } 1159 1160 static inline bool mem_cgroup_unprotected(struct mem_cgroup *target, 1161 struct mem_cgroup *memcg) 1162 { 1163 return true; 1164 } 1165 static inline bool mem_cgroup_below_low(struct mem_cgroup *target, 1166 struct mem_cgroup *memcg) 1167 { 1168 return false; 1169 } 1170 1171 static inline bool mem_cgroup_below_min(struct mem_cgroup *target, 1172 struct mem_cgroup *memcg) 1173 { 1174 return false; 1175 } 1176 1177 static inline void mem_cgroup_commit_charge(struct folio *folio, 1178 struct mem_cgroup *memcg) 1179 { 1180 } 1181 1182 static inline int mem_cgroup_charge(struct folio *folio, 1183 struct mm_struct *mm, gfp_t gfp) 1184 { 1185 return 0; 1186 } 1187 1188 static inline int mem_cgroup_hugetlb_try_charge(struct mem_cgroup *memcg, 1189 gfp_t gfp, long nr_pages) 1190 { 1191 return 0; 1192 } 1193 1194 static inline int mem_cgroup_swapin_charge_folio(struct folio *folio, 1195 struct mm_struct *mm, gfp_t gfp, swp_entry_t entry) 1196 { 1197 return 0; 1198 } 1199 1200 static inline void mem_cgroup_swapin_uncharge_swap(swp_entry_t entry) 1201 { 1202 } 1203 1204 static inline void mem_cgroup_uncharge(struct folio *folio) 1205 { 1206 } 1207 1208 static inline void mem_cgroup_uncharge_folios(struct folio_batch *folios) 1209 { 1210 } 1211 1212 static inline void mem_cgroup_cancel_charge(struct mem_cgroup *memcg, 1213 unsigned int nr_pages) 1214 { 1215 } 1216 1217 static inline void mem_cgroup_replace_folio(struct folio *old, 1218 struct folio *new) 1219 { 1220 } 1221 1222 static inline void mem_cgroup_migrate(struct folio *old, struct folio *new) 1223 { 1224 } 1225 1226 static inline struct lruvec *mem_cgroup_lruvec(struct mem_cgroup *memcg, 1227 struct pglist_data *pgdat) 1228 { 1229 return &pgdat->__lruvec; 1230 } 1231 1232 static inline struct lruvec *folio_lruvec(struct folio *folio) 1233 { 1234 struct pglist_data *pgdat = folio_pgdat(folio); 1235 return &pgdat->__lruvec; 1236 } 1237 1238 static inline 1239 void lruvec_memcg_debug(struct lruvec *lruvec, struct folio *folio) 1240 { 1241 } 1242 1243 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) 1244 { 1245 return NULL; 1246 } 1247 1248 static inline bool mm_match_cgroup(struct mm_struct *mm, 1249 struct mem_cgroup *memcg) 1250 { 1251 return true; 1252 } 1253 1254 static inline struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) 1255 { 1256 return NULL; 1257 } 1258 1259 static inline struct mem_cgroup *get_mem_cgroup_from_current(void) 1260 { 1261 return NULL; 1262 } 1263 1264 static inline 1265 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css) 1266 { 1267 return NULL; 1268 } 1269 1270 static inline void obj_cgroup_put(struct obj_cgroup *objcg) 1271 { 1272 } 1273 1274 static inline bool mem_cgroup_tryget(struct mem_cgroup *memcg) 1275 { 1276 return true; 1277 } 1278 1279 static inline bool mem_cgroup_tryget_online(struct mem_cgroup *memcg) 1280 { 1281 return true; 1282 } 1283 1284 static inline void mem_cgroup_put(struct mem_cgroup *memcg) 1285 { 1286 } 1287 1288 static inline struct lruvec *folio_lruvec_lock(struct folio *folio) 1289 { 1290 struct pglist_data *pgdat = folio_pgdat(folio); 1291 1292 spin_lock(&pgdat->__lruvec.lru_lock); 1293 return &pgdat->__lruvec; 1294 } 1295 1296 static inline struct lruvec *folio_lruvec_lock_irq(struct folio *folio) 1297 { 1298 struct pglist_data *pgdat = folio_pgdat(folio); 1299 1300 spin_lock_irq(&pgdat->__lruvec.lru_lock); 1301 return &pgdat->__lruvec; 1302 } 1303 1304 static inline struct lruvec *folio_lruvec_lock_irqsave(struct folio *folio, 1305 unsigned long *flagsp) 1306 { 1307 struct pglist_data *pgdat = folio_pgdat(folio); 1308 1309 spin_lock_irqsave(&pgdat->__lruvec.lru_lock, *flagsp); 1310 return &pgdat->__lruvec; 1311 } 1312 1313 static inline struct mem_cgroup * 1314 mem_cgroup_iter(struct mem_cgroup *root, 1315 struct mem_cgroup *prev, 1316 struct mem_cgroup_reclaim_cookie *reclaim) 1317 { 1318 return NULL; 1319 } 1320 1321 static inline void mem_cgroup_iter_break(struct mem_cgroup *root, 1322 struct mem_cgroup *prev) 1323 { 1324 } 1325 1326 static inline void mem_cgroup_scan_tasks(struct mem_cgroup *memcg, 1327 int (*fn)(struct task_struct *, void *), void *arg) 1328 { 1329 } 1330 1331 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) 1332 { 1333 return 0; 1334 } 1335 1336 static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) 1337 { 1338 WARN_ON_ONCE(id); 1339 /* XXX: This should always return root_mem_cgroup */ 1340 return NULL; 1341 } 1342 1343 #ifdef CONFIG_SHRINKER_DEBUG 1344 static inline unsigned long mem_cgroup_ino(struct mem_cgroup *memcg) 1345 { 1346 return 0; 1347 } 1348 1349 static inline struct mem_cgroup *mem_cgroup_get_from_ino(unsigned long ino) 1350 { 1351 return NULL; 1352 } 1353 #endif 1354 1355 static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m) 1356 { 1357 return NULL; 1358 } 1359 1360 static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec) 1361 { 1362 return NULL; 1363 } 1364 1365 static inline bool mem_cgroup_online(struct mem_cgroup *memcg) 1366 { 1367 return true; 1368 } 1369 1370 static inline 1371 unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec, 1372 enum lru_list lru, int zone_idx) 1373 { 1374 return 0; 1375 } 1376 1377 static inline unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) 1378 { 1379 return 0; 1380 } 1381 1382 static inline unsigned long mem_cgroup_size(struct mem_cgroup *memcg) 1383 { 1384 return 0; 1385 } 1386 1387 static inline void 1388 mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) 1389 { 1390 } 1391 1392 static inline void 1393 mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) 1394 { 1395 } 1396 1397 static inline void mem_cgroup_handle_over_high(gfp_t gfp_mask) 1398 { 1399 } 1400 1401 static inline struct mem_cgroup *mem_cgroup_get_oom_group( 1402 struct task_struct *victim, struct mem_cgroup *oom_domain) 1403 { 1404 return NULL; 1405 } 1406 1407 static inline void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) 1408 { 1409 } 1410 1411 static inline void __mod_memcg_state(struct mem_cgroup *memcg, 1412 enum memcg_stat_item idx, 1413 int nr) 1414 { 1415 } 1416 1417 static inline void mod_memcg_state(struct mem_cgroup *memcg, 1418 enum memcg_stat_item idx, 1419 int nr) 1420 { 1421 } 1422 1423 static inline void mod_memcg_page_state(struct page *page, 1424 enum memcg_stat_item idx, int val) 1425 { 1426 } 1427 1428 static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx) 1429 { 1430 return 0; 1431 } 1432 1433 static inline unsigned long lruvec_page_state(struct lruvec *lruvec, 1434 enum node_stat_item idx) 1435 { 1436 return node_page_state(lruvec_pgdat(lruvec), idx); 1437 } 1438 1439 static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec, 1440 enum node_stat_item idx) 1441 { 1442 return node_page_state(lruvec_pgdat(lruvec), idx); 1443 } 1444 1445 static inline void mem_cgroup_flush_stats(struct mem_cgroup *memcg) 1446 { 1447 } 1448 1449 static inline void mem_cgroup_flush_stats_ratelimited(struct mem_cgroup *memcg) 1450 { 1451 } 1452 1453 static inline void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, 1454 int val) 1455 { 1456 struct page *page = virt_to_head_page(p); 1457 1458 __mod_node_page_state(page_pgdat(page), idx, val); 1459 } 1460 1461 static inline void mod_lruvec_kmem_state(void *p, enum node_stat_item idx, 1462 int val) 1463 { 1464 struct page *page = virt_to_head_page(p); 1465 1466 mod_node_page_state(page_pgdat(page), idx, val); 1467 } 1468 1469 static inline void count_memcg_events(struct mem_cgroup *memcg, 1470 enum vm_event_item idx, 1471 unsigned long count) 1472 { 1473 } 1474 1475 static inline void __count_memcg_events(struct mem_cgroup *memcg, 1476 enum vm_event_item idx, 1477 unsigned long count) 1478 { 1479 } 1480 1481 static inline void count_memcg_folio_events(struct folio *folio, 1482 enum vm_event_item idx, unsigned long nr) 1483 { 1484 } 1485 1486 static inline 1487 void count_memcg_event_mm(struct mm_struct *mm, enum vm_event_item idx) 1488 { 1489 } 1490 1491 static inline void split_page_memcg(struct page *head, int old_order, int new_order) 1492 { 1493 } 1494 #endif /* CONFIG_MEMCG */ 1495 1496 /* 1497 * Extended information for slab objects stored as an array in page->memcg_data 1498 * if MEMCG_DATA_OBJEXTS is set. 1499 */ 1500 struct slabobj_ext { 1501 #ifdef CONFIG_MEMCG 1502 struct obj_cgroup *objcg; 1503 #endif 1504 #ifdef CONFIG_MEM_ALLOC_PROFILING 1505 union codetag_ref ref; 1506 #endif 1507 } __aligned(8); 1508 1509 static inline void __inc_lruvec_kmem_state(void *p, enum node_stat_item idx) 1510 { 1511 __mod_lruvec_kmem_state(p, idx, 1); 1512 } 1513 1514 static inline void __dec_lruvec_kmem_state(void *p, enum node_stat_item idx) 1515 { 1516 __mod_lruvec_kmem_state(p, idx, -1); 1517 } 1518 1519 static inline struct lruvec *parent_lruvec(struct lruvec *lruvec) 1520 { 1521 struct mem_cgroup *memcg; 1522 1523 memcg = lruvec_memcg(lruvec); 1524 if (!memcg) 1525 return NULL; 1526 memcg = parent_mem_cgroup(memcg); 1527 if (!memcg) 1528 return NULL; 1529 return mem_cgroup_lruvec(memcg, lruvec_pgdat(lruvec)); 1530 } 1531 1532 static inline void unlock_page_lruvec(struct lruvec *lruvec) 1533 { 1534 spin_unlock(&lruvec->lru_lock); 1535 } 1536 1537 static inline void unlock_page_lruvec_irq(struct lruvec *lruvec) 1538 { 1539 spin_unlock_irq(&lruvec->lru_lock); 1540 } 1541 1542 static inline void unlock_page_lruvec_irqrestore(struct lruvec *lruvec, 1543 unsigned long flags) 1544 { 1545 spin_unlock_irqrestore(&lruvec->lru_lock, flags); 1546 } 1547 1548 /* Test requires a stable folio->memcg binding, see folio_memcg() */ 1549 static inline bool folio_matches_lruvec(struct folio *folio, 1550 struct lruvec *lruvec) 1551 { 1552 return lruvec_pgdat(lruvec) == folio_pgdat(folio) && 1553 lruvec_memcg(lruvec) == folio_memcg(folio); 1554 } 1555 1556 /* Don't lock again iff page's lruvec locked */ 1557 static inline struct lruvec *folio_lruvec_relock_irq(struct folio *folio, 1558 struct lruvec *locked_lruvec) 1559 { 1560 if (locked_lruvec) { 1561 if (folio_matches_lruvec(folio, locked_lruvec)) 1562 return locked_lruvec; 1563 1564 unlock_page_lruvec_irq(locked_lruvec); 1565 } 1566 1567 return folio_lruvec_lock_irq(folio); 1568 } 1569 1570 /* Don't lock again iff folio's lruvec locked */ 1571 static inline void folio_lruvec_relock_irqsave(struct folio *folio, 1572 struct lruvec **lruvecp, unsigned long *flags) 1573 { 1574 if (*lruvecp) { 1575 if (folio_matches_lruvec(folio, *lruvecp)) 1576 return; 1577 1578 unlock_page_lruvec_irqrestore(*lruvecp, *flags); 1579 } 1580 1581 *lruvecp = folio_lruvec_lock_irqsave(folio, flags); 1582 } 1583 1584 #ifdef CONFIG_CGROUP_WRITEBACK 1585 1586 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb); 1587 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, 1588 unsigned long *pheadroom, unsigned long *pdirty, 1589 unsigned long *pwriteback); 1590 1591 void mem_cgroup_track_foreign_dirty_slowpath(struct folio *folio, 1592 struct bdi_writeback *wb); 1593 1594 static inline void mem_cgroup_track_foreign_dirty(struct folio *folio, 1595 struct bdi_writeback *wb) 1596 { 1597 struct mem_cgroup *memcg; 1598 1599 if (mem_cgroup_disabled()) 1600 return; 1601 1602 memcg = folio_memcg(folio); 1603 if (unlikely(memcg && &memcg->css != wb->memcg_css)) 1604 mem_cgroup_track_foreign_dirty_slowpath(folio, wb); 1605 } 1606 1607 void mem_cgroup_flush_foreign(struct bdi_writeback *wb); 1608 1609 #else /* CONFIG_CGROUP_WRITEBACK */ 1610 1611 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) 1612 { 1613 return NULL; 1614 } 1615 1616 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb, 1617 unsigned long *pfilepages, 1618 unsigned long *pheadroom, 1619 unsigned long *pdirty, 1620 unsigned long *pwriteback) 1621 { 1622 } 1623 1624 static inline void mem_cgroup_track_foreign_dirty(struct folio *folio, 1625 struct bdi_writeback *wb) 1626 { 1627 } 1628 1629 static inline void mem_cgroup_flush_foreign(struct bdi_writeback *wb) 1630 { 1631 } 1632 1633 #endif /* CONFIG_CGROUP_WRITEBACK */ 1634 1635 struct sock; 1636 bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages, 1637 gfp_t gfp_mask); 1638 void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages); 1639 #ifdef CONFIG_MEMCG 1640 extern struct static_key_false memcg_sockets_enabled_key; 1641 #define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_enabled_key) 1642 void mem_cgroup_sk_alloc(struct sock *sk); 1643 void mem_cgroup_sk_free(struct sock *sk); 1644 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg) 1645 { 1646 #ifdef CONFIG_MEMCG_V1 1647 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) 1648 return !!memcg->tcpmem_pressure; 1649 #endif /* CONFIG_MEMCG_V1 */ 1650 do { 1651 if (time_before(jiffies, READ_ONCE(memcg->socket_pressure))) 1652 return true; 1653 } while ((memcg = parent_mem_cgroup(memcg))); 1654 return false; 1655 } 1656 1657 int alloc_shrinker_info(struct mem_cgroup *memcg); 1658 void free_shrinker_info(struct mem_cgroup *memcg); 1659 void set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id); 1660 void reparent_shrinker_deferred(struct mem_cgroup *memcg); 1661 #else 1662 #define mem_cgroup_sockets_enabled 0 1663 static inline void mem_cgroup_sk_alloc(struct sock *sk) { }; 1664 static inline void mem_cgroup_sk_free(struct sock *sk) { }; 1665 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg) 1666 { 1667 return false; 1668 } 1669 1670 static inline void set_shrinker_bit(struct mem_cgroup *memcg, 1671 int nid, int shrinker_id) 1672 { 1673 } 1674 #endif 1675 1676 #ifdef CONFIG_MEMCG 1677 bool mem_cgroup_kmem_disabled(void); 1678 int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order); 1679 void __memcg_kmem_uncharge_page(struct page *page, int order); 1680 1681 /* 1682 * The returned objcg pointer is safe to use without additional 1683 * protection within a scope. The scope is defined either by 1684 * the current task (similar to the "current" global variable) 1685 * or by set_active_memcg() pair. 1686 * Please, use obj_cgroup_get() to get a reference if the pointer 1687 * needs to be used outside of the local scope. 1688 */ 1689 struct obj_cgroup *current_obj_cgroup(void); 1690 struct obj_cgroup *get_obj_cgroup_from_folio(struct folio *folio); 1691 1692 static inline struct obj_cgroup *get_obj_cgroup_from_current(void) 1693 { 1694 struct obj_cgroup *objcg = current_obj_cgroup(); 1695 1696 if (objcg) 1697 obj_cgroup_get(objcg); 1698 1699 return objcg; 1700 } 1701 1702 int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size); 1703 void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size); 1704 1705 extern struct static_key_false memcg_bpf_enabled_key; 1706 static inline bool memcg_bpf_enabled(void) 1707 { 1708 return static_branch_likely(&memcg_bpf_enabled_key); 1709 } 1710 1711 extern struct static_key_false memcg_kmem_online_key; 1712 1713 static inline bool memcg_kmem_online(void) 1714 { 1715 return static_branch_likely(&memcg_kmem_online_key); 1716 } 1717 1718 static inline int memcg_kmem_charge_page(struct page *page, gfp_t gfp, 1719 int order) 1720 { 1721 if (memcg_kmem_online()) 1722 return __memcg_kmem_charge_page(page, gfp, order); 1723 return 0; 1724 } 1725 1726 static inline void memcg_kmem_uncharge_page(struct page *page, int order) 1727 { 1728 if (memcg_kmem_online()) 1729 __memcg_kmem_uncharge_page(page, order); 1730 } 1731 1732 /* 1733 * A helper for accessing memcg's kmem_id, used for getting 1734 * corresponding LRU lists. 1735 */ 1736 static inline int memcg_kmem_id(struct mem_cgroup *memcg) 1737 { 1738 return memcg ? memcg->kmemcg_id : -1; 1739 } 1740 1741 struct mem_cgroup *mem_cgroup_from_slab_obj(void *p); 1742 1743 static inline void count_objcg_event(struct obj_cgroup *objcg, 1744 enum vm_event_item idx) 1745 { 1746 struct mem_cgroup *memcg; 1747 1748 if (!memcg_kmem_online()) 1749 return; 1750 1751 rcu_read_lock(); 1752 memcg = obj_cgroup_memcg(objcg); 1753 count_memcg_events(memcg, idx, 1); 1754 rcu_read_unlock(); 1755 } 1756 1757 #else 1758 static inline bool mem_cgroup_kmem_disabled(void) 1759 { 1760 return true; 1761 } 1762 1763 static inline int memcg_kmem_charge_page(struct page *page, gfp_t gfp, 1764 int order) 1765 { 1766 return 0; 1767 } 1768 1769 static inline void memcg_kmem_uncharge_page(struct page *page, int order) 1770 { 1771 } 1772 1773 static inline int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, 1774 int order) 1775 { 1776 return 0; 1777 } 1778 1779 static inline void __memcg_kmem_uncharge_page(struct page *page, int order) 1780 { 1781 } 1782 1783 static inline struct obj_cgroup *get_obj_cgroup_from_folio(struct folio *folio) 1784 { 1785 return NULL; 1786 } 1787 1788 static inline bool memcg_bpf_enabled(void) 1789 { 1790 return false; 1791 } 1792 1793 static inline bool memcg_kmem_online(void) 1794 { 1795 return false; 1796 } 1797 1798 static inline int memcg_kmem_id(struct mem_cgroup *memcg) 1799 { 1800 return -1; 1801 } 1802 1803 static inline struct mem_cgroup *mem_cgroup_from_slab_obj(void *p) 1804 { 1805 return NULL; 1806 } 1807 1808 static inline void count_objcg_event(struct obj_cgroup *objcg, 1809 enum vm_event_item idx) 1810 { 1811 } 1812 1813 #endif /* CONFIG_MEMCG */ 1814 1815 #if defined(CONFIG_MEMCG) && defined(CONFIG_ZSWAP) 1816 bool obj_cgroup_may_zswap(struct obj_cgroup *objcg); 1817 void obj_cgroup_charge_zswap(struct obj_cgroup *objcg, size_t size); 1818 void obj_cgroup_uncharge_zswap(struct obj_cgroup *objcg, size_t size); 1819 bool mem_cgroup_zswap_writeback_enabled(struct mem_cgroup *memcg); 1820 #else 1821 static inline bool obj_cgroup_may_zswap(struct obj_cgroup *objcg) 1822 { 1823 return true; 1824 } 1825 static inline void obj_cgroup_charge_zswap(struct obj_cgroup *objcg, 1826 size_t size) 1827 { 1828 } 1829 static inline void obj_cgroup_uncharge_zswap(struct obj_cgroup *objcg, 1830 size_t size) 1831 { 1832 } 1833 static inline bool mem_cgroup_zswap_writeback_enabled(struct mem_cgroup *memcg) 1834 { 1835 /* if zswap is disabled, do not block pages going to the swapping device */ 1836 return true; 1837 } 1838 #endif 1839 1840 1841 /* Cgroup v1-related declarations */ 1842 1843 #ifdef CONFIG_MEMCG_V1 1844 unsigned long memcg1_soft_limit_reclaim(pg_data_t *pgdat, int order, 1845 gfp_t gfp_mask, 1846 unsigned long *total_scanned); 1847 1848 bool mem_cgroup_oom_synchronize(bool wait); 1849 1850 static inline bool task_in_memcg_oom(struct task_struct *p) 1851 { 1852 return p->memcg_in_oom; 1853 } 1854 1855 void folio_memcg_lock(struct folio *folio); 1856 void folio_memcg_unlock(struct folio *folio); 1857 1858 /* try to stablize folio_memcg() for all the pages in a memcg */ 1859 static inline bool mem_cgroup_trylock_pages(struct mem_cgroup *memcg) 1860 { 1861 rcu_read_lock(); 1862 1863 if (mem_cgroup_disabled() || !atomic_read(&memcg->moving_account)) 1864 return true; 1865 1866 rcu_read_unlock(); 1867 return false; 1868 } 1869 1870 static inline void mem_cgroup_unlock_pages(void) 1871 { 1872 rcu_read_unlock(); 1873 } 1874 1875 static inline void mem_cgroup_enter_user_fault(void) 1876 { 1877 WARN_ON(current->in_user_fault); 1878 current->in_user_fault = 1; 1879 } 1880 1881 static inline void mem_cgroup_exit_user_fault(void) 1882 { 1883 WARN_ON(!current->in_user_fault); 1884 current->in_user_fault = 0; 1885 } 1886 1887 #else /* CONFIG_MEMCG_V1 */ 1888 static inline 1889 unsigned long memcg1_soft_limit_reclaim(pg_data_t *pgdat, int order, 1890 gfp_t gfp_mask, 1891 unsigned long *total_scanned) 1892 { 1893 return 0; 1894 } 1895 1896 static inline void folio_memcg_lock(struct folio *folio) 1897 { 1898 } 1899 1900 static inline void folio_memcg_unlock(struct folio *folio) 1901 { 1902 } 1903 1904 static inline bool mem_cgroup_trylock_pages(struct mem_cgroup *memcg) 1905 { 1906 /* to match folio_memcg_rcu() */ 1907 rcu_read_lock(); 1908 return true; 1909 } 1910 1911 static inline void mem_cgroup_unlock_pages(void) 1912 { 1913 rcu_read_unlock(); 1914 } 1915 1916 static inline bool task_in_memcg_oom(struct task_struct *p) 1917 { 1918 return false; 1919 } 1920 1921 static inline bool mem_cgroup_oom_synchronize(bool wait) 1922 { 1923 return false; 1924 } 1925 1926 static inline void mem_cgroup_enter_user_fault(void) 1927 { 1928 } 1929 1930 static inline void mem_cgroup_exit_user_fault(void) 1931 { 1932 } 1933 1934 #endif /* CONFIG_MEMCG_V1 */ 1935 1936 #endif /* _LINUX_MEMCONTROL_H */ 1937