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/page_counter.h> 18 #include <linux/vmpressure.h> 19 #include <linux/eventfd.h> 20 #include <linux/mm.h> 21 #include <linux/vmstat.h> 22 #include <linux/writeback.h> 23 #include <linux/page-flags.h> 24 25 struct mem_cgroup; 26 struct obj_cgroup; 27 struct page; 28 struct mm_struct; 29 struct kmem_cache; 30 31 /* Cgroup-specific page state, on top of universal node page state */ 32 enum memcg_stat_item { 33 MEMCG_SWAP = NR_VM_NODE_STAT_ITEMS, 34 MEMCG_SOCK, 35 MEMCG_PERCPU_B, 36 MEMCG_NR_STAT, 37 }; 38 39 enum memcg_memory_event { 40 MEMCG_LOW, 41 MEMCG_HIGH, 42 MEMCG_MAX, 43 MEMCG_OOM, 44 MEMCG_OOM_KILL, 45 MEMCG_SWAP_HIGH, 46 MEMCG_SWAP_MAX, 47 MEMCG_SWAP_FAIL, 48 MEMCG_NR_MEMORY_EVENTS, 49 }; 50 51 struct mem_cgroup_reclaim_cookie { 52 pg_data_t *pgdat; 53 unsigned int generation; 54 }; 55 56 #ifdef CONFIG_MEMCG 57 58 #define MEM_CGROUP_ID_SHIFT 16 59 #define MEM_CGROUP_ID_MAX USHRT_MAX 60 61 struct mem_cgroup_id { 62 int id; 63 refcount_t ref; 64 }; 65 66 /* 67 * Per memcg event counter is incremented at every pagein/pageout. With THP, 68 * it will be incremented by the number of pages. This counter is used 69 * to trigger some periodic events. This is straightforward and better 70 * than using jiffies etc. to handle periodic memcg event. 71 */ 72 enum mem_cgroup_events_target { 73 MEM_CGROUP_TARGET_THRESH, 74 MEM_CGROUP_TARGET_SOFTLIMIT, 75 MEM_CGROUP_NTARGETS, 76 }; 77 78 struct memcg_vmstats_percpu { 79 /* Local (CPU and cgroup) page state & events */ 80 long state[MEMCG_NR_STAT]; 81 unsigned long events[NR_VM_EVENT_ITEMS]; 82 83 /* Delta calculation for lockless upward propagation */ 84 long state_prev[MEMCG_NR_STAT]; 85 unsigned long events_prev[NR_VM_EVENT_ITEMS]; 86 87 /* Cgroup1: threshold notifications & softlimit tree updates */ 88 unsigned long nr_page_events; 89 unsigned long targets[MEM_CGROUP_NTARGETS]; 90 }; 91 92 struct memcg_vmstats { 93 /* Aggregated (CPU and subtree) page state & events */ 94 long state[MEMCG_NR_STAT]; 95 unsigned long events[NR_VM_EVENT_ITEMS]; 96 97 /* Pending child counts during tree propagation */ 98 long state_pending[MEMCG_NR_STAT]; 99 unsigned long events_pending[NR_VM_EVENT_ITEMS]; 100 }; 101 102 struct mem_cgroup_reclaim_iter { 103 struct mem_cgroup *position; 104 /* scan generation, increased every round-trip */ 105 unsigned int generation; 106 }; 107 108 struct lruvec_stat { 109 long count[NR_VM_NODE_STAT_ITEMS]; 110 }; 111 112 struct batched_lruvec_stat { 113 s32 count[NR_VM_NODE_STAT_ITEMS]; 114 }; 115 116 /* 117 * Bitmap of shrinker::id corresponding to memcg-aware shrinkers, 118 * which have elements charged to this memcg. 119 */ 120 struct memcg_shrinker_map { 121 struct rcu_head rcu; 122 unsigned long map[]; 123 }; 124 125 /* 126 * per-node information in memory controller. 127 */ 128 struct mem_cgroup_per_node { 129 struct lruvec lruvec; 130 131 /* 132 * Legacy local VM stats. This should be struct lruvec_stat and 133 * cannot be optimized to struct batched_lruvec_stat. Because 134 * the threshold of the lruvec_stat_cpu can be as big as 135 * MEMCG_CHARGE_BATCH * PAGE_SIZE. It can fit into s32. But this 136 * filed has no upper limit. 137 */ 138 struct lruvec_stat __percpu *lruvec_stat_local; 139 140 /* Subtree VM stats (batched updates) */ 141 struct batched_lruvec_stat __percpu *lruvec_stat_cpu; 142 atomic_long_t lruvec_stat[NR_VM_NODE_STAT_ITEMS]; 143 144 unsigned long lru_zone_size[MAX_NR_ZONES][NR_LRU_LISTS]; 145 146 struct mem_cgroup_reclaim_iter iter; 147 148 struct memcg_shrinker_map __rcu *shrinker_map; 149 150 struct rb_node tree_node; /* RB tree node */ 151 unsigned long usage_in_excess;/* Set to the value by which */ 152 /* the soft limit is exceeded*/ 153 bool on_tree; 154 struct mem_cgroup *memcg; /* Back pointer, we cannot */ 155 /* use container_of */ 156 }; 157 158 struct mem_cgroup_threshold { 159 struct eventfd_ctx *eventfd; 160 unsigned long threshold; 161 }; 162 163 /* For threshold */ 164 struct mem_cgroup_threshold_ary { 165 /* An array index points to threshold just below or equal to usage. */ 166 int current_threshold; 167 /* Size of entries[] */ 168 unsigned int size; 169 /* Array of thresholds */ 170 struct mem_cgroup_threshold entries[]; 171 }; 172 173 struct mem_cgroup_thresholds { 174 /* Primary thresholds array */ 175 struct mem_cgroup_threshold_ary *primary; 176 /* 177 * Spare threshold array. 178 * This is needed to make mem_cgroup_unregister_event() "never fail". 179 * It must be able to store at least primary->size - 1 entries. 180 */ 181 struct mem_cgroup_threshold_ary *spare; 182 }; 183 184 enum memcg_kmem_state { 185 KMEM_NONE, 186 KMEM_ALLOCATED, 187 KMEM_ONLINE, 188 }; 189 190 #if defined(CONFIG_SMP) 191 struct memcg_padding { 192 char x[0]; 193 } ____cacheline_internodealigned_in_smp; 194 #define MEMCG_PADDING(name) struct memcg_padding name; 195 #else 196 #define MEMCG_PADDING(name) 197 #endif 198 199 /* 200 * Remember four most recent foreign writebacks with dirty pages in this 201 * cgroup. Inode sharing is expected to be uncommon and, even if we miss 202 * one in a given round, we're likely to catch it later if it keeps 203 * foreign-dirtying, so a fairly low count should be enough. 204 * 205 * See mem_cgroup_track_foreign_dirty_slowpath() for details. 206 */ 207 #define MEMCG_CGWB_FRN_CNT 4 208 209 struct memcg_cgwb_frn { 210 u64 bdi_id; /* bdi->id of the foreign inode */ 211 int memcg_id; /* memcg->css.id of foreign inode */ 212 u64 at; /* jiffies_64 at the time of dirtying */ 213 struct wb_completion done; /* tracks in-flight foreign writebacks */ 214 }; 215 216 /* 217 * Bucket for arbitrarily byte-sized objects charged to a memory 218 * cgroup. The bucket can be reparented in one piece when the cgroup 219 * is destroyed, without having to round up the individual references 220 * of all live memory objects in the wild. 221 */ 222 struct obj_cgroup { 223 struct percpu_ref refcnt; 224 struct mem_cgroup *memcg; 225 atomic_t nr_charged_bytes; 226 union { 227 struct list_head list; 228 struct rcu_head rcu; 229 }; 230 }; 231 232 /* 233 * The memory controller data structure. The memory controller controls both 234 * page cache and RSS per cgroup. We would eventually like to provide 235 * statistics based on the statistics developed by Rik Van Riel for clock-pro, 236 * to help the administrator determine what knobs to tune. 237 */ 238 struct mem_cgroup { 239 struct cgroup_subsys_state css; 240 241 /* Private memcg ID. Used to ID objects that outlive the cgroup */ 242 struct mem_cgroup_id id; 243 244 /* Accounted resources */ 245 struct page_counter memory; /* Both v1 & v2 */ 246 247 union { 248 struct page_counter swap; /* v2 only */ 249 struct page_counter memsw; /* v1 only */ 250 }; 251 252 /* Legacy consumer-oriented counters */ 253 struct page_counter kmem; /* v1 only */ 254 struct page_counter tcpmem; /* v1 only */ 255 256 /* Range enforcement for interrupt charges */ 257 struct work_struct high_work; 258 259 unsigned long soft_limit; 260 261 /* vmpressure notifications */ 262 struct vmpressure vmpressure; 263 264 /* 265 * Should the OOM killer kill all belonging tasks, had it kill one? 266 */ 267 bool oom_group; 268 269 /* protected by memcg_oom_lock */ 270 bool oom_lock; 271 int under_oom; 272 273 int swappiness; 274 /* OOM-Killer disable */ 275 int oom_kill_disable; 276 277 /* memory.events and memory.events.local */ 278 struct cgroup_file events_file; 279 struct cgroup_file events_local_file; 280 281 /* handle for "memory.swap.events" */ 282 struct cgroup_file swap_events_file; 283 284 /* protect arrays of thresholds */ 285 struct mutex thresholds_lock; 286 287 /* thresholds for memory usage. RCU-protected */ 288 struct mem_cgroup_thresholds thresholds; 289 290 /* thresholds for mem+swap usage. RCU-protected */ 291 struct mem_cgroup_thresholds memsw_thresholds; 292 293 /* For oom notifier event fd */ 294 struct list_head oom_notify; 295 296 /* 297 * Should we move charges of a task when a task is moved into this 298 * mem_cgroup ? And what type of charges should we move ? 299 */ 300 unsigned long move_charge_at_immigrate; 301 /* taken only while moving_account > 0 */ 302 spinlock_t move_lock; 303 unsigned long move_lock_flags; 304 305 MEMCG_PADDING(_pad1_); 306 307 /* memory.stat */ 308 struct memcg_vmstats vmstats; 309 310 /* memory.events */ 311 atomic_long_t memory_events[MEMCG_NR_MEMORY_EVENTS]; 312 atomic_long_t memory_events_local[MEMCG_NR_MEMORY_EVENTS]; 313 314 unsigned long socket_pressure; 315 316 /* Legacy tcp memory accounting */ 317 bool tcpmem_active; 318 int tcpmem_pressure; 319 320 #ifdef CONFIG_MEMCG_KMEM 321 int kmemcg_id; 322 enum memcg_kmem_state kmem_state; 323 struct obj_cgroup __rcu *objcg; 324 struct list_head objcg_list; /* list of inherited objcgs */ 325 #endif 326 327 MEMCG_PADDING(_pad2_); 328 329 /* 330 * set > 0 if pages under this cgroup are moving to other cgroup. 331 */ 332 atomic_t moving_account; 333 struct task_struct *move_lock_task; 334 335 struct memcg_vmstats_percpu __percpu *vmstats_percpu; 336 337 #ifdef CONFIG_CGROUP_WRITEBACK 338 struct list_head cgwb_list; 339 struct wb_domain cgwb_domain; 340 struct memcg_cgwb_frn cgwb_frn[MEMCG_CGWB_FRN_CNT]; 341 #endif 342 343 /* List of events which userspace want to receive */ 344 struct list_head event_list; 345 spinlock_t event_list_lock; 346 347 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 348 struct deferred_split deferred_split_queue; 349 #endif 350 351 struct mem_cgroup_per_node *nodeinfo[0]; 352 /* WARNING: nodeinfo must be the last member here */ 353 }; 354 355 /* 356 * size of first charge trial. "32" comes from vmscan.c's magic value. 357 * TODO: maybe necessary to use big numbers in big irons. 358 */ 359 #define MEMCG_CHARGE_BATCH 32U 360 361 extern struct mem_cgroup *root_mem_cgroup; 362 363 enum page_memcg_data_flags { 364 /* page->memcg_data is a pointer to an objcgs vector */ 365 MEMCG_DATA_OBJCGS = (1UL << 0), 366 /* page has been accounted as a non-slab kernel page */ 367 MEMCG_DATA_KMEM = (1UL << 1), 368 /* the next bit after the last actual flag */ 369 __NR_MEMCG_DATA_FLAGS = (1UL << 2), 370 }; 371 372 #define MEMCG_DATA_FLAGS_MASK (__NR_MEMCG_DATA_FLAGS - 1) 373 374 static inline bool PageMemcgKmem(struct page *page); 375 376 /* 377 * After the initialization objcg->memcg is always pointing at 378 * a valid memcg, but can be atomically swapped to the parent memcg. 379 * 380 * The caller must ensure that the returned memcg won't be released: 381 * e.g. acquire the rcu_read_lock or css_set_lock. 382 */ 383 static inline struct mem_cgroup *obj_cgroup_memcg(struct obj_cgroup *objcg) 384 { 385 return READ_ONCE(objcg->memcg); 386 } 387 388 /* 389 * __page_memcg - get the memory cgroup associated with a non-kmem page 390 * @page: a pointer to the page struct 391 * 392 * Returns a pointer to the memory cgroup associated with the page, 393 * or NULL. This function assumes that the page is known to have a 394 * proper memory cgroup pointer. It's not safe to call this function 395 * against some type of pages, e.g. slab pages or ex-slab pages or 396 * kmem pages. 397 */ 398 static inline struct mem_cgroup *__page_memcg(struct page *page) 399 { 400 unsigned long memcg_data = page->memcg_data; 401 402 VM_BUG_ON_PAGE(PageSlab(page), page); 403 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_OBJCGS, page); 404 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, page); 405 406 return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); 407 } 408 409 /* 410 * __page_objcg - get the object cgroup associated with a kmem page 411 * @page: a pointer to the page struct 412 * 413 * Returns a pointer to the object cgroup associated with the page, 414 * or NULL. This function assumes that the page is known to have a 415 * proper object cgroup pointer. It's not safe to call this function 416 * against some type of pages, e.g. slab pages or ex-slab pages or 417 * LRU pages. 418 */ 419 static inline struct obj_cgroup *__page_objcg(struct page *page) 420 { 421 unsigned long memcg_data = page->memcg_data; 422 423 VM_BUG_ON_PAGE(PageSlab(page), page); 424 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_OBJCGS, page); 425 VM_BUG_ON_PAGE(!(memcg_data & MEMCG_DATA_KMEM), page); 426 427 return (struct obj_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); 428 } 429 430 /* 431 * page_memcg - get the memory cgroup associated with a page 432 * @page: a pointer to the page struct 433 * 434 * Returns a pointer to the memory cgroup associated with the page, 435 * or NULL. This function assumes that the page is known to have a 436 * proper memory cgroup pointer. It's not safe to call this function 437 * against some type of pages, e.g. slab pages or ex-slab pages. 438 * 439 * For a non-kmem page any of the following ensures page and memcg binding 440 * stability: 441 * 442 * - the page lock 443 * - LRU isolation 444 * - lock_page_memcg() 445 * - exclusive reference 446 * 447 * For a kmem page a caller should hold an rcu read lock to protect memcg 448 * associated with a kmem page from being released. 449 */ 450 static inline struct mem_cgroup *page_memcg(struct page *page) 451 { 452 if (PageMemcgKmem(page)) 453 return obj_cgroup_memcg(__page_objcg(page)); 454 else 455 return __page_memcg(page); 456 } 457 458 /* 459 * page_memcg_rcu - locklessly get the memory cgroup associated with a page 460 * @page: a pointer to the page struct 461 * 462 * Returns a pointer to the memory cgroup associated with the page, 463 * or NULL. This function assumes that the page is known to have a 464 * proper memory cgroup pointer. It's not safe to call this function 465 * against some type of pages, e.g. slab pages or ex-slab pages. 466 */ 467 static inline struct mem_cgroup *page_memcg_rcu(struct page *page) 468 { 469 unsigned long memcg_data = READ_ONCE(page->memcg_data); 470 471 VM_BUG_ON_PAGE(PageSlab(page), page); 472 WARN_ON_ONCE(!rcu_read_lock_held()); 473 474 if (memcg_data & MEMCG_DATA_KMEM) { 475 struct obj_cgroup *objcg; 476 477 objcg = (void *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); 478 return obj_cgroup_memcg(objcg); 479 } 480 481 return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); 482 } 483 484 /* 485 * page_memcg_check - get the memory cgroup associated with a page 486 * @page: a pointer to the page struct 487 * 488 * Returns a pointer to the memory cgroup associated with the page, 489 * or NULL. This function unlike page_memcg() can take any page 490 * as an argument. It has to be used in cases when it's not known if a page 491 * has an associated memory cgroup pointer or an object cgroups vector or 492 * an object cgroup. 493 * 494 * For a non-kmem page any of the following ensures page and memcg binding 495 * stability: 496 * 497 * - the page lock 498 * - LRU isolation 499 * - lock_page_memcg() 500 * - exclusive reference 501 * 502 * For a kmem page a caller should hold an rcu read lock to protect memcg 503 * associated with a kmem page from being released. 504 */ 505 static inline struct mem_cgroup *page_memcg_check(struct page *page) 506 { 507 /* 508 * Because page->memcg_data might be changed asynchronously 509 * for slab pages, READ_ONCE() should be used here. 510 */ 511 unsigned long memcg_data = READ_ONCE(page->memcg_data); 512 513 if (memcg_data & MEMCG_DATA_OBJCGS) 514 return NULL; 515 516 if (memcg_data & MEMCG_DATA_KMEM) { 517 struct obj_cgroup *objcg; 518 519 objcg = (void *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); 520 return obj_cgroup_memcg(objcg); 521 } 522 523 return (struct mem_cgroup *)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); 524 } 525 526 #ifdef CONFIG_MEMCG_KMEM 527 /* 528 * PageMemcgKmem - check if the page has MemcgKmem flag set 529 * @page: a pointer to the page struct 530 * 531 * Checks if the page has MemcgKmem flag set. The caller must ensure that 532 * the page has an associated memory cgroup. It's not safe to call this function 533 * against some types of pages, e.g. slab pages. 534 */ 535 static inline bool PageMemcgKmem(struct page *page) 536 { 537 VM_BUG_ON_PAGE(page->memcg_data & MEMCG_DATA_OBJCGS, page); 538 return page->memcg_data & MEMCG_DATA_KMEM; 539 } 540 541 /* 542 * page_objcgs - get the object cgroups vector associated with a page 543 * @page: a pointer to the page struct 544 * 545 * Returns a pointer to the object cgroups vector associated with the page, 546 * or NULL. This function assumes that the page is known to have an 547 * associated object cgroups vector. It's not safe to call this function 548 * against pages, which might have an associated memory cgroup: e.g. 549 * kernel stack pages. 550 */ 551 static inline struct obj_cgroup **page_objcgs(struct page *page) 552 { 553 unsigned long memcg_data = READ_ONCE(page->memcg_data); 554 555 VM_BUG_ON_PAGE(memcg_data && !(memcg_data & MEMCG_DATA_OBJCGS), page); 556 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, page); 557 558 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); 559 } 560 561 /* 562 * page_objcgs_check - get the object cgroups vector associated with a page 563 * @page: a pointer to the page struct 564 * 565 * Returns a pointer to the object cgroups vector associated with the page, 566 * or NULL. This function is safe to use if the page can be directly associated 567 * with a memory cgroup. 568 */ 569 static inline struct obj_cgroup **page_objcgs_check(struct page *page) 570 { 571 unsigned long memcg_data = READ_ONCE(page->memcg_data); 572 573 if (!memcg_data || !(memcg_data & MEMCG_DATA_OBJCGS)) 574 return NULL; 575 576 VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, page); 577 578 return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); 579 } 580 581 #else 582 static inline bool PageMemcgKmem(struct page *page) 583 { 584 return false; 585 } 586 587 static inline struct obj_cgroup **page_objcgs(struct page *page) 588 { 589 return NULL; 590 } 591 592 static inline struct obj_cgroup **page_objcgs_check(struct page *page) 593 { 594 return NULL; 595 } 596 #endif 597 598 static __always_inline bool memcg_stat_item_in_bytes(int idx) 599 { 600 if (idx == MEMCG_PERCPU_B) 601 return true; 602 return vmstat_item_in_bytes(idx); 603 } 604 605 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) 606 { 607 return (memcg == root_mem_cgroup); 608 } 609 610 static inline bool mem_cgroup_disabled(void) 611 { 612 return !cgroup_subsys_enabled(memory_cgrp_subsys); 613 } 614 615 static inline unsigned long mem_cgroup_protection(struct mem_cgroup *root, 616 struct mem_cgroup *memcg, 617 bool in_low_reclaim) 618 { 619 if (mem_cgroup_disabled()) 620 return 0; 621 622 /* 623 * There is no reclaim protection applied to a targeted reclaim. 624 * We are special casing this specific case here because 625 * mem_cgroup_protected calculation is not robust enough to keep 626 * the protection invariant for calculated effective values for 627 * parallel reclaimers with different reclaim target. This is 628 * especially a problem for tail memcgs (as they have pages on LRU) 629 * which would want to have effective values 0 for targeted reclaim 630 * but a different value for external reclaim. 631 * 632 * Example 633 * Let's have global and A's reclaim in parallel: 634 * | 635 * A (low=2G, usage = 3G, max = 3G, children_low_usage = 1.5G) 636 * |\ 637 * | C (low = 1G, usage = 2.5G) 638 * B (low = 1G, usage = 0.5G) 639 * 640 * For the global reclaim 641 * A.elow = A.low 642 * B.elow = min(B.usage, B.low) because children_low_usage <= A.elow 643 * C.elow = min(C.usage, C.low) 644 * 645 * With the effective values resetting we have A reclaim 646 * A.elow = 0 647 * B.elow = B.low 648 * C.elow = C.low 649 * 650 * If the global reclaim races with A's reclaim then 651 * B.elow = C.elow = 0 because children_low_usage > A.elow) 652 * is possible and reclaiming B would be violating the protection. 653 * 654 */ 655 if (root == memcg) 656 return 0; 657 658 if (in_low_reclaim) 659 return READ_ONCE(memcg->memory.emin); 660 661 return max(READ_ONCE(memcg->memory.emin), 662 READ_ONCE(memcg->memory.elow)); 663 } 664 665 void mem_cgroup_calculate_protection(struct mem_cgroup *root, 666 struct mem_cgroup *memcg); 667 668 static inline bool mem_cgroup_supports_protection(struct mem_cgroup *memcg) 669 { 670 /* 671 * The root memcg doesn't account charges, and doesn't support 672 * protection. 673 */ 674 return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg); 675 676 } 677 678 static inline bool mem_cgroup_below_low(struct mem_cgroup *memcg) 679 { 680 if (!mem_cgroup_supports_protection(memcg)) 681 return false; 682 683 return READ_ONCE(memcg->memory.elow) >= 684 page_counter_read(&memcg->memory); 685 } 686 687 static inline bool mem_cgroup_below_min(struct mem_cgroup *memcg) 688 { 689 if (!mem_cgroup_supports_protection(memcg)) 690 return false; 691 692 return READ_ONCE(memcg->memory.emin) >= 693 page_counter_read(&memcg->memory); 694 } 695 696 int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask); 697 int mem_cgroup_swapin_charge_page(struct page *page, struct mm_struct *mm, 698 gfp_t gfp, swp_entry_t entry); 699 void mem_cgroup_swapin_uncharge_swap(swp_entry_t entry); 700 701 void mem_cgroup_uncharge(struct page *page); 702 void mem_cgroup_uncharge_list(struct list_head *page_list); 703 704 void mem_cgroup_migrate(struct page *oldpage, struct page *newpage); 705 706 /** 707 * mem_cgroup_lruvec - get the lru list vector for a memcg & node 708 * @memcg: memcg of the wanted lruvec 709 * @pgdat: pglist_data 710 * 711 * Returns the lru list vector holding pages for a given @memcg & 712 * @pgdat combination. This can be the node lruvec, if the memory 713 * controller is disabled. 714 */ 715 static inline struct lruvec *mem_cgroup_lruvec(struct mem_cgroup *memcg, 716 struct pglist_data *pgdat) 717 { 718 struct mem_cgroup_per_node *mz; 719 struct lruvec *lruvec; 720 721 if (mem_cgroup_disabled()) { 722 lruvec = &pgdat->__lruvec; 723 goto out; 724 } 725 726 if (!memcg) 727 memcg = root_mem_cgroup; 728 729 mz = memcg->nodeinfo[pgdat->node_id]; 730 lruvec = &mz->lruvec; 731 out: 732 /* 733 * Since a node can be onlined after the mem_cgroup was created, 734 * we have to be prepared to initialize lruvec->pgdat here; 735 * and if offlined then reonlined, we need to reinitialize it. 736 */ 737 if (unlikely(lruvec->pgdat != pgdat)) 738 lruvec->pgdat = pgdat; 739 return lruvec; 740 } 741 742 /** 743 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page 744 * @page: the page 745 * @pgdat: pgdat of the page 746 * 747 * This function relies on page->mem_cgroup being stable. 748 */ 749 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page, 750 struct pglist_data *pgdat) 751 { 752 struct mem_cgroup *memcg = page_memcg(page); 753 754 VM_WARN_ON_ONCE_PAGE(!memcg && !mem_cgroup_disabled(), page); 755 return mem_cgroup_lruvec(memcg, pgdat); 756 } 757 758 static inline bool lruvec_holds_page_lru_lock(struct page *page, 759 struct lruvec *lruvec) 760 { 761 pg_data_t *pgdat = page_pgdat(page); 762 const struct mem_cgroup *memcg; 763 struct mem_cgroup_per_node *mz; 764 765 if (mem_cgroup_disabled()) 766 return lruvec == &pgdat->__lruvec; 767 768 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 769 memcg = page_memcg(page) ? : root_mem_cgroup; 770 771 return lruvec->pgdat == pgdat && mz->memcg == memcg; 772 } 773 774 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p); 775 776 struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm); 777 778 struct lruvec *lock_page_lruvec(struct page *page); 779 struct lruvec *lock_page_lruvec_irq(struct page *page); 780 struct lruvec *lock_page_lruvec_irqsave(struct page *page, 781 unsigned long *flags); 782 783 #ifdef CONFIG_DEBUG_VM 784 void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page); 785 #else 786 static inline void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page) 787 { 788 } 789 #endif 790 791 static inline 792 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){ 793 return css ? container_of(css, struct mem_cgroup, css) : NULL; 794 } 795 796 static inline bool obj_cgroup_tryget(struct obj_cgroup *objcg) 797 { 798 return percpu_ref_tryget(&objcg->refcnt); 799 } 800 801 static inline void obj_cgroup_get(struct obj_cgroup *objcg) 802 { 803 percpu_ref_get(&objcg->refcnt); 804 } 805 806 static inline void obj_cgroup_get_many(struct obj_cgroup *objcg, 807 unsigned long nr) 808 { 809 percpu_ref_get_many(&objcg->refcnt, nr); 810 } 811 812 static inline void obj_cgroup_put(struct obj_cgroup *objcg) 813 { 814 percpu_ref_put(&objcg->refcnt); 815 } 816 817 static inline void mem_cgroup_put(struct mem_cgroup *memcg) 818 { 819 if (memcg) 820 css_put(&memcg->css); 821 } 822 823 #define mem_cgroup_from_counter(counter, member) \ 824 container_of(counter, struct mem_cgroup, member) 825 826 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *, 827 struct mem_cgroup *, 828 struct mem_cgroup_reclaim_cookie *); 829 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *); 830 int mem_cgroup_scan_tasks(struct mem_cgroup *, 831 int (*)(struct task_struct *, void *), void *); 832 833 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) 834 { 835 if (mem_cgroup_disabled()) 836 return 0; 837 838 return memcg->id.id; 839 } 840 struct mem_cgroup *mem_cgroup_from_id(unsigned short id); 841 842 static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m) 843 { 844 return mem_cgroup_from_css(seq_css(m)); 845 } 846 847 static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec) 848 { 849 struct mem_cgroup_per_node *mz; 850 851 if (mem_cgroup_disabled()) 852 return NULL; 853 854 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 855 return mz->memcg; 856 } 857 858 /** 859 * parent_mem_cgroup - find the accounting parent of a memcg 860 * @memcg: memcg whose parent to find 861 * 862 * Returns the parent memcg, or NULL if this is the root or the memory 863 * controller is in legacy no-hierarchy mode. 864 */ 865 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) 866 { 867 if (!memcg->memory.parent) 868 return NULL; 869 return mem_cgroup_from_counter(memcg->memory.parent, memory); 870 } 871 872 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg, 873 struct mem_cgroup *root) 874 { 875 if (root == memcg) 876 return true; 877 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup); 878 } 879 880 static inline bool mm_match_cgroup(struct mm_struct *mm, 881 struct mem_cgroup *memcg) 882 { 883 struct mem_cgroup *task_memcg; 884 bool match = false; 885 886 rcu_read_lock(); 887 task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); 888 if (task_memcg) 889 match = mem_cgroup_is_descendant(task_memcg, memcg); 890 rcu_read_unlock(); 891 return match; 892 } 893 894 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page); 895 ino_t page_cgroup_ino(struct page *page); 896 897 static inline bool mem_cgroup_online(struct mem_cgroup *memcg) 898 { 899 if (mem_cgroup_disabled()) 900 return true; 901 return !!(memcg->css.flags & CSS_ONLINE); 902 } 903 904 /* 905 * For memory reclaim. 906 */ 907 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg); 908 909 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, 910 int zid, int nr_pages); 911 912 static inline 913 unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec, 914 enum lru_list lru, int zone_idx) 915 { 916 struct mem_cgroup_per_node *mz; 917 918 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 919 return READ_ONCE(mz->lru_zone_size[zone_idx][lru]); 920 } 921 922 void mem_cgroup_handle_over_high(void); 923 924 unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg); 925 926 unsigned long mem_cgroup_size(struct mem_cgroup *memcg); 927 928 void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, 929 struct task_struct *p); 930 931 void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg); 932 933 static inline void mem_cgroup_enter_user_fault(void) 934 { 935 WARN_ON(current->in_user_fault); 936 current->in_user_fault = 1; 937 } 938 939 static inline void mem_cgroup_exit_user_fault(void) 940 { 941 WARN_ON(!current->in_user_fault); 942 current->in_user_fault = 0; 943 } 944 945 static inline bool task_in_memcg_oom(struct task_struct *p) 946 { 947 return p->memcg_in_oom; 948 } 949 950 bool mem_cgroup_oom_synchronize(bool wait); 951 struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, 952 struct mem_cgroup *oom_domain); 953 void mem_cgroup_print_oom_group(struct mem_cgroup *memcg); 954 955 #ifdef CONFIG_MEMCG_SWAP 956 extern bool cgroup_memory_noswap; 957 #endif 958 959 void lock_page_memcg(struct page *page); 960 void unlock_page_memcg(struct page *page); 961 962 void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val); 963 964 /* idx can be of type enum memcg_stat_item or node_stat_item */ 965 static inline void mod_memcg_state(struct mem_cgroup *memcg, 966 int idx, int val) 967 { 968 unsigned long flags; 969 970 local_irq_save(flags); 971 __mod_memcg_state(memcg, idx, val); 972 local_irq_restore(flags); 973 } 974 975 static inline unsigned long lruvec_page_state(struct lruvec *lruvec, 976 enum node_stat_item idx) 977 { 978 struct mem_cgroup_per_node *pn; 979 long x; 980 981 if (mem_cgroup_disabled()) 982 return node_page_state(lruvec_pgdat(lruvec), idx); 983 984 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 985 x = atomic_long_read(&pn->lruvec_stat[idx]); 986 #ifdef CONFIG_SMP 987 if (x < 0) 988 x = 0; 989 #endif 990 return x; 991 } 992 993 static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec, 994 enum node_stat_item idx) 995 { 996 struct mem_cgroup_per_node *pn; 997 long x = 0; 998 int cpu; 999 1000 if (mem_cgroup_disabled()) 1001 return node_page_state(lruvec_pgdat(lruvec), idx); 1002 1003 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); 1004 for_each_possible_cpu(cpu) 1005 x += per_cpu(pn->lruvec_stat_local->count[idx], cpu); 1006 #ifdef CONFIG_SMP 1007 if (x < 0) 1008 x = 0; 1009 #endif 1010 return x; 1011 } 1012 1013 void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, 1014 int val); 1015 void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, int val); 1016 1017 static inline void mod_lruvec_kmem_state(void *p, enum node_stat_item idx, 1018 int val) 1019 { 1020 unsigned long flags; 1021 1022 local_irq_save(flags); 1023 __mod_lruvec_kmem_state(p, idx, val); 1024 local_irq_restore(flags); 1025 } 1026 1027 static inline void mod_memcg_lruvec_state(struct lruvec *lruvec, 1028 enum node_stat_item idx, int val) 1029 { 1030 unsigned long flags; 1031 1032 local_irq_save(flags); 1033 __mod_memcg_lruvec_state(lruvec, idx, val); 1034 local_irq_restore(flags); 1035 } 1036 1037 void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, 1038 unsigned long count); 1039 1040 static inline void count_memcg_events(struct mem_cgroup *memcg, 1041 enum vm_event_item idx, 1042 unsigned long count) 1043 { 1044 unsigned long flags; 1045 1046 local_irq_save(flags); 1047 __count_memcg_events(memcg, idx, count); 1048 local_irq_restore(flags); 1049 } 1050 1051 static inline void count_memcg_page_event(struct page *page, 1052 enum vm_event_item idx) 1053 { 1054 struct mem_cgroup *memcg = page_memcg(page); 1055 1056 if (memcg) 1057 count_memcg_events(memcg, idx, 1); 1058 } 1059 1060 static inline void count_memcg_event_mm(struct mm_struct *mm, 1061 enum vm_event_item idx) 1062 { 1063 struct mem_cgroup *memcg; 1064 1065 if (mem_cgroup_disabled()) 1066 return; 1067 1068 rcu_read_lock(); 1069 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); 1070 if (likely(memcg)) 1071 count_memcg_events(memcg, idx, 1); 1072 rcu_read_unlock(); 1073 } 1074 1075 static inline void memcg_memory_event(struct mem_cgroup *memcg, 1076 enum memcg_memory_event event) 1077 { 1078 bool swap_event = event == MEMCG_SWAP_HIGH || event == MEMCG_SWAP_MAX || 1079 event == MEMCG_SWAP_FAIL; 1080 1081 atomic_long_inc(&memcg->memory_events_local[event]); 1082 if (!swap_event) 1083 cgroup_file_notify(&memcg->events_local_file); 1084 1085 do { 1086 atomic_long_inc(&memcg->memory_events[event]); 1087 if (swap_event) 1088 cgroup_file_notify(&memcg->swap_events_file); 1089 else 1090 cgroup_file_notify(&memcg->events_file); 1091 1092 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) 1093 break; 1094 if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS) 1095 break; 1096 } while ((memcg = parent_mem_cgroup(memcg)) && 1097 !mem_cgroup_is_root(memcg)); 1098 } 1099 1100 static inline void memcg_memory_event_mm(struct mm_struct *mm, 1101 enum memcg_memory_event event) 1102 { 1103 struct mem_cgroup *memcg; 1104 1105 if (mem_cgroup_disabled()) 1106 return; 1107 1108 rcu_read_lock(); 1109 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); 1110 if (likely(memcg)) 1111 memcg_memory_event(memcg, event); 1112 rcu_read_unlock(); 1113 } 1114 1115 void split_page_memcg(struct page *head, unsigned int nr); 1116 1117 unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, 1118 gfp_t gfp_mask, 1119 unsigned long *total_scanned); 1120 1121 #else /* CONFIG_MEMCG */ 1122 1123 #define MEM_CGROUP_ID_SHIFT 0 1124 #define MEM_CGROUP_ID_MAX 0 1125 1126 static inline struct mem_cgroup *page_memcg(struct page *page) 1127 { 1128 return NULL; 1129 } 1130 1131 static inline struct mem_cgroup *page_memcg_rcu(struct page *page) 1132 { 1133 WARN_ON_ONCE(!rcu_read_lock_held()); 1134 return NULL; 1135 } 1136 1137 static inline struct mem_cgroup *page_memcg_check(struct page *page) 1138 { 1139 return NULL; 1140 } 1141 1142 static inline bool PageMemcgKmem(struct page *page) 1143 { 1144 return false; 1145 } 1146 1147 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) 1148 { 1149 return true; 1150 } 1151 1152 static inline bool mem_cgroup_disabled(void) 1153 { 1154 return true; 1155 } 1156 1157 static inline void memcg_memory_event(struct mem_cgroup *memcg, 1158 enum memcg_memory_event event) 1159 { 1160 } 1161 1162 static inline void memcg_memory_event_mm(struct mm_struct *mm, 1163 enum memcg_memory_event event) 1164 { 1165 } 1166 1167 static inline unsigned long mem_cgroup_protection(struct mem_cgroup *root, 1168 struct mem_cgroup *memcg, 1169 bool in_low_reclaim) 1170 { 1171 return 0; 1172 } 1173 1174 static inline void mem_cgroup_calculate_protection(struct mem_cgroup *root, 1175 struct mem_cgroup *memcg) 1176 { 1177 } 1178 1179 static inline bool mem_cgroup_below_low(struct mem_cgroup *memcg) 1180 { 1181 return false; 1182 } 1183 1184 static inline bool mem_cgroup_below_min(struct mem_cgroup *memcg) 1185 { 1186 return false; 1187 } 1188 1189 static inline int mem_cgroup_charge(struct page *page, struct mm_struct *mm, 1190 gfp_t gfp_mask) 1191 { 1192 return 0; 1193 } 1194 1195 static inline int mem_cgroup_swapin_charge_page(struct page *page, 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 page *page) 1206 { 1207 } 1208 1209 static inline void mem_cgroup_uncharge_list(struct list_head *page_list) 1210 { 1211 } 1212 1213 static inline void mem_cgroup_migrate(struct page *old, struct page *new) 1214 { 1215 } 1216 1217 static inline struct lruvec *mem_cgroup_lruvec(struct mem_cgroup *memcg, 1218 struct pglist_data *pgdat) 1219 { 1220 return &pgdat->__lruvec; 1221 } 1222 1223 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page, 1224 struct pglist_data *pgdat) 1225 { 1226 return &pgdat->__lruvec; 1227 } 1228 1229 static inline bool lruvec_holds_page_lru_lock(struct page *page, 1230 struct lruvec *lruvec) 1231 { 1232 pg_data_t *pgdat = page_pgdat(page); 1233 1234 return lruvec == &pgdat->__lruvec; 1235 } 1236 1237 static inline void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page) 1238 { 1239 } 1240 1241 static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) 1242 { 1243 return NULL; 1244 } 1245 1246 static inline bool mm_match_cgroup(struct mm_struct *mm, 1247 struct mem_cgroup *memcg) 1248 { 1249 return true; 1250 } 1251 1252 static inline struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) 1253 { 1254 return NULL; 1255 } 1256 1257 static inline void mem_cgroup_put(struct mem_cgroup *memcg) 1258 { 1259 } 1260 1261 static inline struct lruvec *lock_page_lruvec(struct page *page) 1262 { 1263 struct pglist_data *pgdat = page_pgdat(page); 1264 1265 spin_lock(&pgdat->__lruvec.lru_lock); 1266 return &pgdat->__lruvec; 1267 } 1268 1269 static inline struct lruvec *lock_page_lruvec_irq(struct page *page) 1270 { 1271 struct pglist_data *pgdat = page_pgdat(page); 1272 1273 spin_lock_irq(&pgdat->__lruvec.lru_lock); 1274 return &pgdat->__lruvec; 1275 } 1276 1277 static inline struct lruvec *lock_page_lruvec_irqsave(struct page *page, 1278 unsigned long *flagsp) 1279 { 1280 struct pglist_data *pgdat = page_pgdat(page); 1281 1282 spin_lock_irqsave(&pgdat->__lruvec.lru_lock, *flagsp); 1283 return &pgdat->__lruvec; 1284 } 1285 1286 static inline struct mem_cgroup * 1287 mem_cgroup_iter(struct mem_cgroup *root, 1288 struct mem_cgroup *prev, 1289 struct mem_cgroup_reclaim_cookie *reclaim) 1290 { 1291 return NULL; 1292 } 1293 1294 static inline void mem_cgroup_iter_break(struct mem_cgroup *root, 1295 struct mem_cgroup *prev) 1296 { 1297 } 1298 1299 static inline int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, 1300 int (*fn)(struct task_struct *, void *), void *arg) 1301 { 1302 return 0; 1303 } 1304 1305 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) 1306 { 1307 return 0; 1308 } 1309 1310 static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) 1311 { 1312 WARN_ON_ONCE(id); 1313 /* XXX: This should always return root_mem_cgroup */ 1314 return NULL; 1315 } 1316 1317 static inline struct mem_cgroup *mem_cgroup_from_seq(struct seq_file *m) 1318 { 1319 return NULL; 1320 } 1321 1322 static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec) 1323 { 1324 return NULL; 1325 } 1326 1327 static inline bool mem_cgroup_online(struct mem_cgroup *memcg) 1328 { 1329 return true; 1330 } 1331 1332 static inline 1333 unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec, 1334 enum lru_list lru, int zone_idx) 1335 { 1336 return 0; 1337 } 1338 1339 static inline unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) 1340 { 1341 return 0; 1342 } 1343 1344 static inline unsigned long mem_cgroup_size(struct mem_cgroup *memcg) 1345 { 1346 return 0; 1347 } 1348 1349 static inline void 1350 mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) 1351 { 1352 } 1353 1354 static inline void 1355 mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) 1356 { 1357 } 1358 1359 static inline void lock_page_memcg(struct page *page) 1360 { 1361 } 1362 1363 static inline void unlock_page_memcg(struct page *page) 1364 { 1365 } 1366 1367 static inline void mem_cgroup_handle_over_high(void) 1368 { 1369 } 1370 1371 static inline void mem_cgroup_enter_user_fault(void) 1372 { 1373 } 1374 1375 static inline void mem_cgroup_exit_user_fault(void) 1376 { 1377 } 1378 1379 static inline bool task_in_memcg_oom(struct task_struct *p) 1380 { 1381 return false; 1382 } 1383 1384 static inline bool mem_cgroup_oom_synchronize(bool wait) 1385 { 1386 return false; 1387 } 1388 1389 static inline struct mem_cgroup *mem_cgroup_get_oom_group( 1390 struct task_struct *victim, struct mem_cgroup *oom_domain) 1391 { 1392 return NULL; 1393 } 1394 1395 static inline void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) 1396 { 1397 } 1398 1399 static inline void __mod_memcg_state(struct mem_cgroup *memcg, 1400 int idx, 1401 int nr) 1402 { 1403 } 1404 1405 static inline void mod_memcg_state(struct mem_cgroup *memcg, 1406 int idx, 1407 int nr) 1408 { 1409 } 1410 1411 static inline unsigned long lruvec_page_state(struct lruvec *lruvec, 1412 enum node_stat_item idx) 1413 { 1414 return node_page_state(lruvec_pgdat(lruvec), idx); 1415 } 1416 1417 static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec, 1418 enum node_stat_item idx) 1419 { 1420 return node_page_state(lruvec_pgdat(lruvec), idx); 1421 } 1422 1423 static inline void __mod_memcg_lruvec_state(struct lruvec *lruvec, 1424 enum node_stat_item idx, int val) 1425 { 1426 } 1427 1428 static inline void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, 1429 int val) 1430 { 1431 struct page *page = virt_to_head_page(p); 1432 1433 __mod_node_page_state(page_pgdat(page), idx, val); 1434 } 1435 1436 static inline void mod_lruvec_kmem_state(void *p, enum node_stat_item idx, 1437 int val) 1438 { 1439 struct page *page = virt_to_head_page(p); 1440 1441 mod_node_page_state(page_pgdat(page), idx, val); 1442 } 1443 1444 static inline void count_memcg_events(struct mem_cgroup *memcg, 1445 enum vm_event_item idx, 1446 unsigned long count) 1447 { 1448 } 1449 1450 static inline void __count_memcg_events(struct mem_cgroup *memcg, 1451 enum vm_event_item idx, 1452 unsigned long count) 1453 { 1454 } 1455 1456 static inline void count_memcg_page_event(struct page *page, 1457 int idx) 1458 { 1459 } 1460 1461 static inline 1462 void count_memcg_event_mm(struct mm_struct *mm, enum vm_event_item idx) 1463 { 1464 } 1465 1466 static inline void split_page_memcg(struct page *head, unsigned int nr) 1467 { 1468 } 1469 1470 static inline 1471 unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, 1472 gfp_t gfp_mask, 1473 unsigned long *total_scanned) 1474 { 1475 return 0; 1476 } 1477 #endif /* CONFIG_MEMCG */ 1478 1479 static inline void __inc_lruvec_kmem_state(void *p, enum node_stat_item idx) 1480 { 1481 __mod_lruvec_kmem_state(p, idx, 1); 1482 } 1483 1484 static inline void __dec_lruvec_kmem_state(void *p, enum node_stat_item idx) 1485 { 1486 __mod_lruvec_kmem_state(p, idx, -1); 1487 } 1488 1489 static inline struct lruvec *parent_lruvec(struct lruvec *lruvec) 1490 { 1491 struct mem_cgroup *memcg; 1492 1493 memcg = lruvec_memcg(lruvec); 1494 if (!memcg) 1495 return NULL; 1496 memcg = parent_mem_cgroup(memcg); 1497 if (!memcg) 1498 return NULL; 1499 return mem_cgroup_lruvec(memcg, lruvec_pgdat(lruvec)); 1500 } 1501 1502 static inline void unlock_page_lruvec(struct lruvec *lruvec) 1503 { 1504 spin_unlock(&lruvec->lru_lock); 1505 } 1506 1507 static inline void unlock_page_lruvec_irq(struct lruvec *lruvec) 1508 { 1509 spin_unlock_irq(&lruvec->lru_lock); 1510 } 1511 1512 static inline void unlock_page_lruvec_irqrestore(struct lruvec *lruvec, 1513 unsigned long flags) 1514 { 1515 spin_unlock_irqrestore(&lruvec->lru_lock, flags); 1516 } 1517 1518 /* Don't lock again iff page's lruvec locked */ 1519 static inline struct lruvec *relock_page_lruvec_irq(struct page *page, 1520 struct lruvec *locked_lruvec) 1521 { 1522 if (locked_lruvec) { 1523 if (lruvec_holds_page_lru_lock(page, locked_lruvec)) 1524 return locked_lruvec; 1525 1526 unlock_page_lruvec_irq(locked_lruvec); 1527 } 1528 1529 return lock_page_lruvec_irq(page); 1530 } 1531 1532 /* Don't lock again iff page's lruvec locked */ 1533 static inline struct lruvec *relock_page_lruvec_irqsave(struct page *page, 1534 struct lruvec *locked_lruvec, unsigned long *flags) 1535 { 1536 if (locked_lruvec) { 1537 if (lruvec_holds_page_lru_lock(page, locked_lruvec)) 1538 return locked_lruvec; 1539 1540 unlock_page_lruvec_irqrestore(locked_lruvec, *flags); 1541 } 1542 1543 return lock_page_lruvec_irqsave(page, flags); 1544 } 1545 1546 #ifdef CONFIG_CGROUP_WRITEBACK 1547 1548 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb); 1549 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, 1550 unsigned long *pheadroom, unsigned long *pdirty, 1551 unsigned long *pwriteback); 1552 1553 void mem_cgroup_track_foreign_dirty_slowpath(struct page *page, 1554 struct bdi_writeback *wb); 1555 1556 static inline void mem_cgroup_track_foreign_dirty(struct page *page, 1557 struct bdi_writeback *wb) 1558 { 1559 if (mem_cgroup_disabled()) 1560 return; 1561 1562 if (unlikely(&page_memcg(page)->css != wb->memcg_css)) 1563 mem_cgroup_track_foreign_dirty_slowpath(page, wb); 1564 } 1565 1566 void mem_cgroup_flush_foreign(struct bdi_writeback *wb); 1567 1568 #else /* CONFIG_CGROUP_WRITEBACK */ 1569 1570 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) 1571 { 1572 return NULL; 1573 } 1574 1575 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb, 1576 unsigned long *pfilepages, 1577 unsigned long *pheadroom, 1578 unsigned long *pdirty, 1579 unsigned long *pwriteback) 1580 { 1581 } 1582 1583 static inline void mem_cgroup_track_foreign_dirty(struct page *page, 1584 struct bdi_writeback *wb) 1585 { 1586 } 1587 1588 static inline void mem_cgroup_flush_foreign(struct bdi_writeback *wb) 1589 { 1590 } 1591 1592 #endif /* CONFIG_CGROUP_WRITEBACK */ 1593 1594 struct sock; 1595 bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages); 1596 void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages); 1597 #ifdef CONFIG_MEMCG 1598 extern struct static_key_false memcg_sockets_enabled_key; 1599 #define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_enabled_key) 1600 void mem_cgroup_sk_alloc(struct sock *sk); 1601 void mem_cgroup_sk_free(struct sock *sk); 1602 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg) 1603 { 1604 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_pressure) 1605 return true; 1606 do { 1607 if (time_before(jiffies, memcg->socket_pressure)) 1608 return true; 1609 } while ((memcg = parent_mem_cgroup(memcg))); 1610 return false; 1611 } 1612 1613 extern int memcg_expand_shrinker_maps(int new_id); 1614 1615 extern void memcg_set_shrinker_bit(struct mem_cgroup *memcg, 1616 int nid, int shrinker_id); 1617 #else 1618 #define mem_cgroup_sockets_enabled 0 1619 static inline void mem_cgroup_sk_alloc(struct sock *sk) { }; 1620 static inline void mem_cgroup_sk_free(struct sock *sk) { }; 1621 static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg) 1622 { 1623 return false; 1624 } 1625 1626 static inline void memcg_set_shrinker_bit(struct mem_cgroup *memcg, 1627 int nid, int shrinker_id) 1628 { 1629 } 1630 #endif 1631 1632 #ifdef CONFIG_MEMCG_KMEM 1633 int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order); 1634 void __memcg_kmem_uncharge_page(struct page *page, int order); 1635 1636 struct obj_cgroup *get_obj_cgroup_from_current(void); 1637 1638 int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size); 1639 void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size); 1640 1641 extern struct static_key_false memcg_kmem_enabled_key; 1642 1643 extern int memcg_nr_cache_ids; 1644 void memcg_get_cache_ids(void); 1645 void memcg_put_cache_ids(void); 1646 1647 /* 1648 * Helper macro to loop through all memcg-specific caches. Callers must still 1649 * check if the cache is valid (it is either valid or NULL). 1650 * the slab_mutex must be held when looping through those caches 1651 */ 1652 #define for_each_memcg_cache_index(_idx) \ 1653 for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++) 1654 1655 static inline bool memcg_kmem_enabled(void) 1656 { 1657 return static_branch_likely(&memcg_kmem_enabled_key); 1658 } 1659 1660 static inline int memcg_kmem_charge_page(struct page *page, gfp_t gfp, 1661 int order) 1662 { 1663 if (memcg_kmem_enabled()) 1664 return __memcg_kmem_charge_page(page, gfp, order); 1665 return 0; 1666 } 1667 1668 static inline void memcg_kmem_uncharge_page(struct page *page, int order) 1669 { 1670 if (memcg_kmem_enabled()) 1671 __memcg_kmem_uncharge_page(page, order); 1672 } 1673 1674 /* 1675 * A helper for accessing memcg's kmem_id, used for getting 1676 * corresponding LRU lists. 1677 */ 1678 static inline int memcg_cache_id(struct mem_cgroup *memcg) 1679 { 1680 return memcg ? memcg->kmemcg_id : -1; 1681 } 1682 1683 struct mem_cgroup *mem_cgroup_from_obj(void *p); 1684 1685 #else 1686 1687 static inline int memcg_kmem_charge_page(struct page *page, gfp_t gfp, 1688 int order) 1689 { 1690 return 0; 1691 } 1692 1693 static inline void memcg_kmem_uncharge_page(struct page *page, int order) 1694 { 1695 } 1696 1697 static inline int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, 1698 int order) 1699 { 1700 return 0; 1701 } 1702 1703 static inline void __memcg_kmem_uncharge_page(struct page *page, int order) 1704 { 1705 } 1706 1707 #define for_each_memcg_cache_index(_idx) \ 1708 for (; NULL; ) 1709 1710 static inline bool memcg_kmem_enabled(void) 1711 { 1712 return false; 1713 } 1714 1715 static inline int memcg_cache_id(struct mem_cgroup *memcg) 1716 { 1717 return -1; 1718 } 1719 1720 static inline void memcg_get_cache_ids(void) 1721 { 1722 } 1723 1724 static inline void memcg_put_cache_ids(void) 1725 { 1726 } 1727 1728 static inline struct mem_cgroup *mem_cgroup_from_obj(void *p) 1729 { 1730 return NULL; 1731 } 1732 1733 #endif /* CONFIG_MEMCG_KMEM */ 1734 1735 #endif /* _LINUX_MEMCONTROL_H */ 1736