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