linux/include/linux/memcontrol.h
<<
>>
Prefs
   1/* memcontrol.h - Memory Controller
   2 *
   3 * Copyright IBM Corporation, 2007
   4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
   5 *
   6 * Copyright 2007 OpenVZ SWsoft Inc
   7 * Author: Pavel Emelianov <xemul@openvz.org>
   8 *
   9 * This program is free software; you can redistribute it and/or modify
  10 * it under the terms of the GNU General Public License as published by
  11 * the Free Software Foundation; either version 2 of the License, or
  12 * (at your option) any later version.
  13 *
  14 * This program is distributed in the hope that it will be useful,
  15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17 * GNU General Public License for more details.
  18 */
  19
  20#ifndef _LINUX_MEMCONTROL_H
  21#define _LINUX_MEMCONTROL_H
  22#include <linux/cgroup.h>
  23#include <linux/vm_event_item.h>
  24#include <linux/hardirq.h>
  25#include <linux/jump_label.h>
  26#include <linux/page_counter.h>
  27#include <linux/vmpressure.h>
  28#include <linux/eventfd.h>
  29#include <linux/mm.h>
  30#include <linux/vmstat.h>
  31#include <linux/writeback.h>
  32#include <linux/page-flags.h>
  33
  34struct mem_cgroup;
  35struct page;
  36struct mm_struct;
  37struct kmem_cache;
  38
  39/* Cgroup-specific page state, on top of universal node page state */
  40enum memcg_stat_item {
  41        MEMCG_CACHE = NR_VM_NODE_STAT_ITEMS,
  42        MEMCG_RSS,
  43        MEMCG_RSS_HUGE,
  44        MEMCG_SWAP,
  45        MEMCG_SOCK,
  46        /* XXX: why are these zone and not node counters? */
  47        MEMCG_KERNEL_STACK_KB,
  48        MEMCG_NR_STAT,
  49};
  50
  51enum memcg_memory_event {
  52        MEMCG_LOW,
  53        MEMCG_HIGH,
  54        MEMCG_MAX,
  55        MEMCG_OOM,
  56        MEMCG_OOM_KILL,
  57        MEMCG_SWAP_MAX,
  58        MEMCG_SWAP_FAIL,
  59        MEMCG_NR_MEMORY_EVENTS,
  60};
  61
  62enum mem_cgroup_protection {
  63        MEMCG_PROT_NONE,
  64        MEMCG_PROT_LOW,
  65        MEMCG_PROT_MIN,
  66};
  67
  68struct mem_cgroup_reclaim_cookie {
  69        pg_data_t *pgdat;
  70        int priority;
  71        unsigned int generation;
  72};
  73
  74#ifdef CONFIG_MEMCG
  75
  76#define MEM_CGROUP_ID_SHIFT     16
  77#define MEM_CGROUP_ID_MAX       USHRT_MAX
  78
  79struct mem_cgroup_id {
  80        int id;
  81        atomic_t ref;
  82};
  83
  84/*
  85 * Per memcg event counter is incremented at every pagein/pageout. With THP,
  86 * it will be incremated by the number of pages. This counter is used for
  87 * for trigger some periodic events. This is straightforward and better
  88 * than using jiffies etc. to handle periodic memcg event.
  89 */
  90enum mem_cgroup_events_target {
  91        MEM_CGROUP_TARGET_THRESH,
  92        MEM_CGROUP_TARGET_SOFTLIMIT,
  93        MEM_CGROUP_TARGET_NUMAINFO,
  94        MEM_CGROUP_NTARGETS,
  95};
  96
  97struct mem_cgroup_stat_cpu {
  98        long count[MEMCG_NR_STAT];
  99        unsigned long events[NR_VM_EVENT_ITEMS];
 100        unsigned long nr_page_events;
 101        unsigned long targets[MEM_CGROUP_NTARGETS];
 102};
 103
 104struct mem_cgroup_reclaim_iter {
 105        struct mem_cgroup *position;
 106        /* scan generation, increased every round-trip */
 107        unsigned int generation;
 108};
 109
 110struct lruvec_stat {
 111        long count[NR_VM_NODE_STAT_ITEMS];
 112};
 113
 114/*
 115 * per-zone information in memory controller.
 116 */
 117struct mem_cgroup_per_node {
 118        struct lruvec           lruvec;
 119
 120        struct lruvec_stat __percpu *lruvec_stat_cpu;
 121        atomic_long_t           lruvec_stat[NR_VM_NODE_STAT_ITEMS];
 122
 123        unsigned long           lru_zone_size[MAX_NR_ZONES][NR_LRU_LISTS];
 124
 125        struct mem_cgroup_reclaim_iter  iter[DEF_PRIORITY + 1];
 126
 127        struct rb_node          tree_node;      /* RB tree node */
 128        unsigned long           usage_in_excess;/* Set to the value by which */
 129                                                /* the soft limit is exceeded*/
 130        bool                    on_tree;
 131        bool                    congested;      /* memcg has many dirty pages */
 132                                                /* backed by a congested BDI */
 133
 134        struct mem_cgroup       *memcg;         /* Back pointer, we cannot */
 135                                                /* use container_of        */
 136};
 137
 138struct mem_cgroup_threshold {
 139        struct eventfd_ctx *eventfd;
 140        unsigned long threshold;
 141};
 142
 143/* For threshold */
 144struct mem_cgroup_threshold_ary {
 145        /* An array index points to threshold just below or equal to usage. */
 146        int current_threshold;
 147        /* Size of entries[] */
 148        unsigned int size;
 149        /* Array of thresholds */
 150        struct mem_cgroup_threshold entries[0];
 151};
 152
 153struct mem_cgroup_thresholds {
 154        /* Primary thresholds array */
 155        struct mem_cgroup_threshold_ary *primary;
 156        /*
 157         * Spare threshold array.
 158         * This is needed to make mem_cgroup_unregister_event() "never fail".
 159         * It must be able to store at least primary->size - 1 entries.
 160         */
 161        struct mem_cgroup_threshold_ary *spare;
 162};
 163
 164enum memcg_kmem_state {
 165        KMEM_NONE,
 166        KMEM_ALLOCATED,
 167        KMEM_ONLINE,
 168};
 169
 170#if defined(CONFIG_SMP)
 171struct memcg_padding {
 172        char x[0];
 173} ____cacheline_internodealigned_in_smp;
 174#define MEMCG_PADDING(name)      struct memcg_padding name;
 175#else
 176#define MEMCG_PADDING(name)
 177#endif
 178
 179/*
 180 * The memory controller data structure. The memory controller controls both
 181 * page cache and RSS per cgroup. We would eventually like to provide
 182 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
 183 * to help the administrator determine what knobs to tune.
 184 */
 185struct mem_cgroup {
 186        struct cgroup_subsys_state css;
 187
 188        /* Private memcg ID. Used to ID objects that outlive the cgroup */
 189        struct mem_cgroup_id id;
 190
 191        /* Accounted resources */
 192        struct page_counter memory;
 193        struct page_counter swap;
 194
 195        /* Legacy consumer-oriented counters */
 196        struct page_counter memsw;
 197        struct page_counter kmem;
 198        struct page_counter tcpmem;
 199
 200        /* Upper bound of normal memory consumption range */
 201        unsigned long high;
 202
 203        /* Range enforcement for interrupt charges */
 204        struct work_struct high_work;
 205
 206        unsigned long soft_limit;
 207
 208        /* vmpressure notifications */
 209        struct vmpressure vmpressure;
 210
 211        /*
 212         * Should the accounting and control be hierarchical, per subtree?
 213         */
 214        bool use_hierarchy;
 215
 216        /* protected by memcg_oom_lock */
 217        bool            oom_lock;
 218        int             under_oom;
 219
 220        int     swappiness;
 221        /* OOM-Killer disable */
 222        int             oom_kill_disable;
 223
 224        /* memory.events */
 225        struct cgroup_file events_file;
 226
 227        /* handle for "memory.swap.events" */
 228        struct cgroup_file swap_events_file;
 229
 230        /* protect arrays of thresholds */
 231        struct mutex thresholds_lock;
 232
 233        /* thresholds for memory usage. RCU-protected */
 234        struct mem_cgroup_thresholds thresholds;
 235
 236        /* thresholds for mem+swap usage. RCU-protected */
 237        struct mem_cgroup_thresholds memsw_thresholds;
 238
 239        /* For oom notifier event fd */
 240        struct list_head oom_notify;
 241
 242        /*
 243         * Should we move charges of a task when a task is moved into this
 244         * mem_cgroup ? And what type of charges should we move ?
 245         */
 246        unsigned long move_charge_at_immigrate;
 247        /* taken only while moving_account > 0 */
 248        spinlock_t              move_lock;
 249        unsigned long           move_lock_flags;
 250
 251        MEMCG_PADDING(_pad1_);
 252
 253        /*
 254         * set > 0 if pages under this cgroup are moving to other cgroup.
 255         */
 256        atomic_t                moving_account;
 257        struct task_struct      *move_lock_task;
 258
 259        /* memory.stat */
 260        struct mem_cgroup_stat_cpu __percpu *stat_cpu;
 261
 262        MEMCG_PADDING(_pad2_);
 263
 264        atomic_long_t           stat[MEMCG_NR_STAT];
 265        atomic_long_t           events[NR_VM_EVENT_ITEMS];
 266        atomic_long_t memory_events[MEMCG_NR_MEMORY_EVENTS];
 267
 268        unsigned long           socket_pressure;
 269
 270        /* Legacy tcp memory accounting */
 271        bool                    tcpmem_active;
 272        int                     tcpmem_pressure;
 273
 274#ifndef CONFIG_SLOB
 275        /* Index in the kmem_cache->memcg_params.memcg_caches array */
 276        int kmemcg_id;
 277        enum memcg_kmem_state kmem_state;
 278        struct list_head kmem_caches;
 279#endif
 280
 281        int last_scanned_node;
 282#if MAX_NUMNODES > 1
 283        nodemask_t      scan_nodes;
 284        atomic_t        numainfo_events;
 285        atomic_t        numainfo_updating;
 286#endif
 287
 288#ifdef CONFIG_CGROUP_WRITEBACK
 289        struct list_head cgwb_list;
 290        struct wb_domain cgwb_domain;
 291#endif
 292
 293        /* List of events which userspace want to receive */
 294        struct list_head event_list;
 295        spinlock_t event_list_lock;
 296
 297        struct mem_cgroup_per_node *nodeinfo[0];
 298        /* WARNING: nodeinfo must be the last member here */
 299};
 300
 301/*
 302 * size of first charge trial. "32" comes from vmscan.c's magic value.
 303 * TODO: maybe necessary to use big numbers in big irons.
 304 */
 305#define MEMCG_CHARGE_BATCH 32U
 306
 307extern struct mem_cgroup *root_mem_cgroup;
 308
 309static inline bool mem_cgroup_disabled(void)
 310{
 311        return !cgroup_subsys_enabled(memory_cgrp_subsys);
 312}
 313
 314enum mem_cgroup_protection mem_cgroup_protected(struct mem_cgroup *root,
 315                                                struct mem_cgroup *memcg);
 316
 317int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
 318                          gfp_t gfp_mask, struct mem_cgroup **memcgp,
 319                          bool compound);
 320void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
 321                              bool lrucare, bool compound);
 322void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
 323                bool compound);
 324void mem_cgroup_uncharge(struct page *page);
 325void mem_cgroup_uncharge_list(struct list_head *page_list);
 326
 327void mem_cgroup_migrate(struct page *oldpage, struct page *newpage);
 328
 329static struct mem_cgroup_per_node *
 330mem_cgroup_nodeinfo(struct mem_cgroup *memcg, int nid)
 331{
 332        return memcg->nodeinfo[nid];
 333}
 334
 335/**
 336 * mem_cgroup_lruvec - get the lru list vector for a node or a memcg zone
 337 * @node: node of the wanted lruvec
 338 * @memcg: memcg of the wanted lruvec
 339 *
 340 * Returns the lru list vector holding pages for a given @node or a given
 341 * @memcg and @zone. This can be the node lruvec, if the memory controller
 342 * is disabled.
 343 */
 344static inline struct lruvec *mem_cgroup_lruvec(struct pglist_data *pgdat,
 345                                struct mem_cgroup *memcg)
 346{
 347        struct mem_cgroup_per_node *mz;
 348        struct lruvec *lruvec;
 349
 350        if (mem_cgroup_disabled()) {
 351                lruvec = node_lruvec(pgdat);
 352                goto out;
 353        }
 354
 355        mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
 356        lruvec = &mz->lruvec;
 357out:
 358        /*
 359         * Since a node can be onlined after the mem_cgroup was created,
 360         * we have to be prepared to initialize lruvec->pgdat here;
 361         * and if offlined then reonlined, we need to reinitialize it.
 362         */
 363        if (unlikely(lruvec->pgdat != pgdat))
 364                lruvec->pgdat = pgdat;
 365        return lruvec;
 366}
 367
 368struct lruvec *mem_cgroup_page_lruvec(struct page *, struct pglist_data *);
 369
 370bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg);
 371struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
 372
 373static inline
 374struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
 375        return css ? container_of(css, struct mem_cgroup, css) : NULL;
 376}
 377
 378#define mem_cgroup_from_counter(counter, member)        \
 379        container_of(counter, struct mem_cgroup, member)
 380
 381struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
 382                                   struct mem_cgroup *,
 383                                   struct mem_cgroup_reclaim_cookie *);
 384void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
 385int mem_cgroup_scan_tasks(struct mem_cgroup *,
 386                          int (*)(struct task_struct *, void *), void *);
 387
 388static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
 389{
 390        if (mem_cgroup_disabled())
 391                return 0;
 392
 393        return memcg->id.id;
 394}
 395struct mem_cgroup *mem_cgroup_from_id(unsigned short id);
 396
 397static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec)
 398{
 399        struct mem_cgroup_per_node *mz;
 400
 401        if (mem_cgroup_disabled())
 402                return NULL;
 403
 404        mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
 405        return mz->memcg;
 406}
 407
 408/**
 409 * parent_mem_cgroup - find the accounting parent of a memcg
 410 * @memcg: memcg whose parent to find
 411 *
 412 * Returns the parent memcg, or NULL if this is the root or the memory
 413 * controller is in legacy no-hierarchy mode.
 414 */
 415static inline struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
 416{
 417        if (!memcg->memory.parent)
 418                return NULL;
 419        return mem_cgroup_from_counter(memcg->memory.parent, memory);
 420}
 421
 422static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
 423                              struct mem_cgroup *root)
 424{
 425        if (root == memcg)
 426                return true;
 427        if (!root->use_hierarchy)
 428                return false;
 429        return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
 430}
 431
 432static inline bool mm_match_cgroup(struct mm_struct *mm,
 433                                   struct mem_cgroup *memcg)
 434{
 435        struct mem_cgroup *task_memcg;
 436        bool match = false;
 437
 438        rcu_read_lock();
 439        task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
 440        if (task_memcg)
 441                match = mem_cgroup_is_descendant(task_memcg, memcg);
 442        rcu_read_unlock();
 443        return match;
 444}
 445
 446struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
 447ino_t page_cgroup_ino(struct page *page);
 448
 449static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
 450{
 451        if (mem_cgroup_disabled())
 452                return true;
 453        return !!(memcg->css.flags & CSS_ONLINE);
 454}
 455
 456/*
 457 * For memory reclaim.
 458 */
 459int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
 460
 461void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
 462                int zid, int nr_pages);
 463
 464unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
 465                                           int nid, unsigned int lru_mask);
 466
 467static inline
 468unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
 469{
 470        struct mem_cgroup_per_node *mz;
 471        unsigned long nr_pages = 0;
 472        int zid;
 473
 474        mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
 475        for (zid = 0; zid < MAX_NR_ZONES; zid++)
 476                nr_pages += mz->lru_zone_size[zid][lru];
 477        return nr_pages;
 478}
 479
 480static inline
 481unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
 482                enum lru_list lru, int zone_idx)
 483{
 484        struct mem_cgroup_per_node *mz;
 485
 486        mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
 487        return mz->lru_zone_size[zone_idx][lru];
 488}
 489
 490void mem_cgroup_handle_over_high(void);
 491
 492unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg);
 493
 494void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
 495                                struct task_struct *p);
 496
 497static inline void mem_cgroup_oom_enable(void)
 498{
 499        WARN_ON(current->memcg_may_oom);
 500        current->memcg_may_oom = 1;
 501}
 502
 503static inline void mem_cgroup_oom_disable(void)
 504{
 505        WARN_ON(!current->memcg_may_oom);
 506        current->memcg_may_oom = 0;
 507}
 508
 509static inline bool task_in_memcg_oom(struct task_struct *p)
 510{
 511        return p->memcg_in_oom;
 512}
 513
 514bool mem_cgroup_oom_synchronize(bool wait);
 515
 516#ifdef CONFIG_MEMCG_SWAP
 517extern int do_swap_account;
 518#endif
 519
 520struct mem_cgroup *lock_page_memcg(struct page *page);
 521void __unlock_page_memcg(struct mem_cgroup *memcg);
 522void unlock_page_memcg(struct page *page);
 523
 524/* idx can be of type enum memcg_stat_item or node_stat_item */
 525static inline unsigned long memcg_page_state(struct mem_cgroup *memcg,
 526                                             int idx)
 527{
 528        long x = atomic_long_read(&memcg->stat[idx]);
 529#ifdef CONFIG_SMP
 530        if (x < 0)
 531                x = 0;
 532#endif
 533        return x;
 534}
 535
 536/* idx can be of type enum memcg_stat_item or node_stat_item */
 537static inline void __mod_memcg_state(struct mem_cgroup *memcg,
 538                                     int idx, int val)
 539{
 540        long x;
 541
 542        if (mem_cgroup_disabled())
 543                return;
 544
 545        x = val + __this_cpu_read(memcg->stat_cpu->count[idx]);
 546        if (unlikely(abs(x) > MEMCG_CHARGE_BATCH)) {
 547                atomic_long_add(x, &memcg->stat[idx]);
 548                x = 0;
 549        }
 550        __this_cpu_write(memcg->stat_cpu->count[idx], x);
 551}
 552
 553/* idx can be of type enum memcg_stat_item or node_stat_item */
 554static inline void mod_memcg_state(struct mem_cgroup *memcg,
 555                                   int idx, int val)
 556{
 557        unsigned long flags;
 558
 559        local_irq_save(flags);
 560        __mod_memcg_state(memcg, idx, val);
 561        local_irq_restore(flags);
 562}
 563
 564/**
 565 * mod_memcg_page_state - update page state statistics
 566 * @page: the page
 567 * @idx: page state item to account
 568 * @val: number of pages (positive or negative)
 569 *
 570 * The @page must be locked or the caller must use lock_page_memcg()
 571 * to prevent double accounting when the page is concurrently being
 572 * moved to another memcg:
 573 *
 574 *   lock_page(page) or lock_page_memcg(page)
 575 *   if (TestClearPageState(page))
 576 *     mod_memcg_page_state(page, state, -1);
 577 *   unlock_page(page) or unlock_page_memcg(page)
 578 *
 579 * Kernel pages are an exception to this, since they'll never move.
 580 */
 581static inline void __mod_memcg_page_state(struct page *page,
 582                                          int idx, int val)
 583{
 584        if (page->mem_cgroup)
 585                __mod_memcg_state(page->mem_cgroup, idx, val);
 586}
 587
 588static inline void mod_memcg_page_state(struct page *page,
 589                                        int idx, int val)
 590{
 591        if (page->mem_cgroup)
 592                mod_memcg_state(page->mem_cgroup, idx, val);
 593}
 594
 595static inline unsigned long lruvec_page_state(struct lruvec *lruvec,
 596                                              enum node_stat_item idx)
 597{
 598        struct mem_cgroup_per_node *pn;
 599        long x;
 600
 601        if (mem_cgroup_disabled())
 602                return node_page_state(lruvec_pgdat(lruvec), idx);
 603
 604        pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
 605        x = atomic_long_read(&pn->lruvec_stat[idx]);
 606#ifdef CONFIG_SMP
 607        if (x < 0)
 608                x = 0;
 609#endif
 610        return x;
 611}
 612
 613static inline void __mod_lruvec_state(struct lruvec *lruvec,
 614                                      enum node_stat_item idx, int val)
 615{
 616        struct mem_cgroup_per_node *pn;
 617        long x;
 618
 619        /* Update node */
 620        __mod_node_page_state(lruvec_pgdat(lruvec), idx, val);
 621
 622        if (mem_cgroup_disabled())
 623                return;
 624
 625        pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
 626
 627        /* Update memcg */
 628        __mod_memcg_state(pn->memcg, idx, val);
 629
 630        /* Update lruvec */
 631        x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]);
 632        if (unlikely(abs(x) > MEMCG_CHARGE_BATCH)) {
 633                atomic_long_add(x, &pn->lruvec_stat[idx]);
 634                x = 0;
 635        }
 636        __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x);
 637}
 638
 639static inline void mod_lruvec_state(struct lruvec *lruvec,
 640                                    enum node_stat_item idx, int val)
 641{
 642        unsigned long flags;
 643
 644        local_irq_save(flags);
 645        __mod_lruvec_state(lruvec, idx, val);
 646        local_irq_restore(flags);
 647}
 648
 649static inline void __mod_lruvec_page_state(struct page *page,
 650                                           enum node_stat_item idx, int val)
 651{
 652        pg_data_t *pgdat = page_pgdat(page);
 653        struct lruvec *lruvec;
 654
 655        /* Untracked pages have no memcg, no lruvec. Update only the node */
 656        if (!page->mem_cgroup) {
 657                __mod_node_page_state(pgdat, idx, val);
 658                return;
 659        }
 660
 661        lruvec = mem_cgroup_lruvec(pgdat, page->mem_cgroup);
 662        __mod_lruvec_state(lruvec, idx, val);
 663}
 664
 665static inline void mod_lruvec_page_state(struct page *page,
 666                                         enum node_stat_item idx, int val)
 667{
 668        unsigned long flags;
 669
 670        local_irq_save(flags);
 671        __mod_lruvec_page_state(page, idx, val);
 672        local_irq_restore(flags);
 673}
 674
 675unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
 676                                                gfp_t gfp_mask,
 677                                                unsigned long *total_scanned);
 678
 679static inline void __count_memcg_events(struct mem_cgroup *memcg,
 680                                        enum vm_event_item idx,
 681                                        unsigned long count)
 682{
 683        unsigned long x;
 684
 685        if (mem_cgroup_disabled())
 686                return;
 687
 688        x = count + __this_cpu_read(memcg->stat_cpu->events[idx]);
 689        if (unlikely(x > MEMCG_CHARGE_BATCH)) {
 690                atomic_long_add(x, &memcg->events[idx]);
 691                x = 0;
 692        }
 693        __this_cpu_write(memcg->stat_cpu->events[idx], x);
 694}
 695
 696static inline void count_memcg_events(struct mem_cgroup *memcg,
 697                                      enum vm_event_item idx,
 698                                      unsigned long count)
 699{
 700        unsigned long flags;
 701
 702        local_irq_save(flags);
 703        __count_memcg_events(memcg, idx, count);
 704        local_irq_restore(flags);
 705}
 706
 707static inline void count_memcg_page_event(struct page *page,
 708                                          enum vm_event_item idx)
 709{
 710        if (page->mem_cgroup)
 711                count_memcg_events(page->mem_cgroup, idx, 1);
 712}
 713
 714static inline void count_memcg_event_mm(struct mm_struct *mm,
 715                                        enum vm_event_item idx)
 716{
 717        struct mem_cgroup *memcg;
 718
 719        if (mem_cgroup_disabled())
 720                return;
 721
 722        rcu_read_lock();
 723        memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
 724        if (likely(memcg))
 725                count_memcg_events(memcg, idx, 1);
 726        rcu_read_unlock();
 727}
 728
 729static inline void memcg_memory_event(struct mem_cgroup *memcg,
 730                                      enum memcg_memory_event event)
 731{
 732        atomic_long_inc(&memcg->memory_events[event]);
 733        cgroup_file_notify(&memcg->events_file);
 734}
 735
 736static inline void memcg_memory_event_mm(struct mm_struct *mm,
 737                                         enum memcg_memory_event event)
 738{
 739        struct mem_cgroup *memcg;
 740
 741        if (mem_cgroup_disabled())
 742                return;
 743
 744        rcu_read_lock();
 745        memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
 746        if (likely(memcg))
 747                memcg_memory_event(memcg, event);
 748        rcu_read_unlock();
 749}
 750
 751#ifdef CONFIG_TRANSPARENT_HUGEPAGE
 752void mem_cgroup_split_huge_fixup(struct page *head);
 753#endif
 754
 755#else /* CONFIG_MEMCG */
 756
 757#define MEM_CGROUP_ID_SHIFT     0
 758#define MEM_CGROUP_ID_MAX       0
 759
 760struct mem_cgroup;
 761
 762static inline bool mem_cgroup_disabled(void)
 763{
 764        return true;
 765}
 766
 767static inline void memcg_memory_event(struct mem_cgroup *memcg,
 768                                      enum memcg_memory_event event)
 769{
 770}
 771
 772static inline void memcg_memory_event_mm(struct mm_struct *mm,
 773                                         enum memcg_memory_event event)
 774{
 775}
 776
 777static inline enum mem_cgroup_protection mem_cgroup_protected(
 778        struct mem_cgroup *root, struct mem_cgroup *memcg)
 779{
 780        return MEMCG_PROT_NONE;
 781}
 782
 783static inline int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
 784                                        gfp_t gfp_mask,
 785                                        struct mem_cgroup **memcgp,
 786                                        bool compound)
 787{
 788        *memcgp = NULL;
 789        return 0;
 790}
 791
 792static inline void mem_cgroup_commit_charge(struct page *page,
 793                                            struct mem_cgroup *memcg,
 794                                            bool lrucare, bool compound)
 795{
 796}
 797
 798static inline void mem_cgroup_cancel_charge(struct page *page,
 799                                            struct mem_cgroup *memcg,
 800                                            bool compound)
 801{
 802}
 803
 804static inline void mem_cgroup_uncharge(struct page *page)
 805{
 806}
 807
 808static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
 809{
 810}
 811
 812static inline void mem_cgroup_migrate(struct page *old, struct page *new)
 813{
 814}
 815
 816static inline struct lruvec *mem_cgroup_lruvec(struct pglist_data *pgdat,
 817                                struct mem_cgroup *memcg)
 818{
 819        return node_lruvec(pgdat);
 820}
 821
 822static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
 823                                                    struct pglist_data *pgdat)
 824{
 825        return &pgdat->lruvec;
 826}
 827
 828static inline bool mm_match_cgroup(struct mm_struct *mm,
 829                struct mem_cgroup *memcg)
 830{
 831        return true;
 832}
 833
 834static inline bool task_in_mem_cgroup(struct task_struct *task,
 835                                      const struct mem_cgroup *memcg)
 836{
 837        return true;
 838}
 839
 840static inline struct mem_cgroup *
 841mem_cgroup_iter(struct mem_cgroup *root,
 842                struct mem_cgroup *prev,
 843                struct mem_cgroup_reclaim_cookie *reclaim)
 844{
 845        return NULL;
 846}
 847
 848static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
 849                                         struct mem_cgroup *prev)
 850{
 851}
 852
 853static inline int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
 854                int (*fn)(struct task_struct *, void *), void *arg)
 855{
 856        return 0;
 857}
 858
 859static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
 860{
 861        return 0;
 862}
 863
 864static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
 865{
 866        WARN_ON_ONCE(id);
 867        /* XXX: This should always return root_mem_cgroup */
 868        return NULL;
 869}
 870
 871static inline struct mem_cgroup *lruvec_memcg(struct lruvec *lruvec)
 872{
 873        return NULL;
 874}
 875
 876static inline bool mem_cgroup_online(struct mem_cgroup *memcg)
 877{
 878        return true;
 879}
 880
 881static inline unsigned long
 882mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
 883{
 884        return 0;
 885}
 886static inline
 887unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
 888                enum lru_list lru, int zone_idx)
 889{
 890        return 0;
 891}
 892
 893static inline unsigned long
 894mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
 895                             int nid, unsigned int lru_mask)
 896{
 897        return 0;
 898}
 899
 900static inline unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg)
 901{
 902        return 0;
 903}
 904
 905static inline void
 906mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
 907{
 908}
 909
 910static inline struct mem_cgroup *lock_page_memcg(struct page *page)
 911{
 912        return NULL;
 913}
 914
 915static inline void __unlock_page_memcg(struct mem_cgroup *memcg)
 916{
 917}
 918
 919static inline void unlock_page_memcg(struct page *page)
 920{
 921}
 922
 923static inline void mem_cgroup_handle_over_high(void)
 924{
 925}
 926
 927static inline void mem_cgroup_oom_enable(void)
 928{
 929}
 930
 931static inline void mem_cgroup_oom_disable(void)
 932{
 933}
 934
 935static inline bool task_in_memcg_oom(struct task_struct *p)
 936{
 937        return false;
 938}
 939
 940static inline bool mem_cgroup_oom_synchronize(bool wait)
 941{
 942        return false;
 943}
 944
 945static inline unsigned long memcg_page_state(struct mem_cgroup *memcg,
 946                                             int idx)
 947{
 948        return 0;
 949}
 950
 951static inline void __mod_memcg_state(struct mem_cgroup *memcg,
 952                                     int idx,
 953                                     int nr)
 954{
 955}
 956
 957static inline void mod_memcg_state(struct mem_cgroup *memcg,
 958                                   int idx,
 959                                   int nr)
 960{
 961}
 962
 963static inline void __mod_memcg_page_state(struct page *page,
 964                                          int idx,
 965                                          int nr)
 966{
 967}
 968
 969static inline void mod_memcg_page_state(struct page *page,
 970                                        int idx,
 971                                        int nr)
 972{
 973}
 974
 975static inline unsigned long lruvec_page_state(struct lruvec *lruvec,
 976                                              enum node_stat_item idx)
 977{
 978        return node_page_state(lruvec_pgdat(lruvec), idx);
 979}
 980
 981static inline void __mod_lruvec_state(struct lruvec *lruvec,
 982                                      enum node_stat_item idx, int val)
 983{
 984        __mod_node_page_state(lruvec_pgdat(lruvec), idx, val);
 985}
 986
 987static inline void mod_lruvec_state(struct lruvec *lruvec,
 988                                    enum node_stat_item idx, int val)
 989{
 990        mod_node_page_state(lruvec_pgdat(lruvec), idx, val);
 991}
 992
 993static inline void __mod_lruvec_page_state(struct page *page,
 994                                           enum node_stat_item idx, int val)
 995{
 996        __mod_node_page_state(page_pgdat(page), idx, val);
 997}
 998
 999static inline void mod_lruvec_page_state(struct page *page,
1000                                         enum node_stat_item idx, int val)
1001{
1002        mod_node_page_state(page_pgdat(page), idx, val);
1003}
1004
1005static inline
1006unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
1007                                            gfp_t gfp_mask,
1008                                            unsigned long *total_scanned)
1009{
1010        return 0;
1011}
1012
1013static inline void mem_cgroup_split_huge_fixup(struct page *head)
1014{
1015}
1016
1017static inline void count_memcg_events(struct mem_cgroup *memcg,
1018                                      enum vm_event_item idx,
1019                                      unsigned long count)
1020{
1021}
1022
1023static inline void count_memcg_page_event(struct page *page,
1024                                          int idx)
1025{
1026}
1027
1028static inline
1029void count_memcg_event_mm(struct mm_struct *mm, enum vm_event_item idx)
1030{
1031}
1032#endif /* CONFIG_MEMCG */
1033
1034/* idx can be of type enum memcg_stat_item or node_stat_item */
1035static inline void __inc_memcg_state(struct mem_cgroup *memcg,
1036                                     int idx)
1037{
1038        __mod_memcg_state(memcg, idx, 1);
1039}
1040
1041/* idx can be of type enum memcg_stat_item or node_stat_item */
1042static inline void __dec_memcg_state(struct mem_cgroup *memcg,
1043                                     int idx)
1044{
1045        __mod_memcg_state(memcg, idx, -1);
1046}
1047
1048/* idx can be of type enum memcg_stat_item or node_stat_item */
1049static inline void __inc_memcg_page_state(struct page *page,
1050                                          int idx)
1051{
1052        __mod_memcg_page_state(page, idx, 1);
1053}
1054
1055/* idx can be of type enum memcg_stat_item or node_stat_item */
1056static inline void __dec_memcg_page_state(struct page *page,
1057                                          int idx)
1058{
1059        __mod_memcg_page_state(page, idx, -1);
1060}
1061
1062static inline void __inc_lruvec_state(struct lruvec *lruvec,
1063                                      enum node_stat_item idx)
1064{
1065        __mod_lruvec_state(lruvec, idx, 1);
1066}
1067
1068static inline void __dec_lruvec_state(struct lruvec *lruvec,
1069                                      enum node_stat_item idx)
1070{
1071        __mod_lruvec_state(lruvec, idx, -1);
1072}
1073
1074static inline void __inc_lruvec_page_state(struct page *page,
1075                                           enum node_stat_item idx)
1076{
1077        __mod_lruvec_page_state(page, idx, 1);
1078}
1079
1080static inline void __dec_lruvec_page_state(struct page *page,
1081                                           enum node_stat_item idx)
1082{
1083        __mod_lruvec_page_state(page, idx, -1);
1084}
1085
1086/* idx can be of type enum memcg_stat_item or node_stat_item */
1087static inline void inc_memcg_state(struct mem_cgroup *memcg,
1088                                   int idx)
1089{
1090        mod_memcg_state(memcg, idx, 1);
1091}
1092
1093/* idx can be of type enum memcg_stat_item or node_stat_item */
1094static inline void dec_memcg_state(struct mem_cgroup *memcg,
1095                                   int idx)
1096{
1097        mod_memcg_state(memcg, idx, -1);
1098}
1099
1100/* idx can be of type enum memcg_stat_item or node_stat_item */
1101static inline void inc_memcg_page_state(struct page *page,
1102                                        int idx)
1103{
1104        mod_memcg_page_state(page, idx, 1);
1105}
1106
1107/* idx can be of type enum memcg_stat_item or node_stat_item */
1108static inline void dec_memcg_page_state(struct page *page,
1109                                        int idx)
1110{
1111        mod_memcg_page_state(page, idx, -1);
1112}
1113
1114static inline void inc_lruvec_state(struct lruvec *lruvec,
1115                                    enum node_stat_item idx)
1116{
1117        mod_lruvec_state(lruvec, idx, 1);
1118}
1119
1120static inline void dec_lruvec_state(struct lruvec *lruvec,
1121                                    enum node_stat_item idx)
1122{
1123        mod_lruvec_state(lruvec, idx, -1);
1124}
1125
1126static inline void inc_lruvec_page_state(struct page *page,
1127                                         enum node_stat_item idx)
1128{
1129        mod_lruvec_page_state(page, idx, 1);
1130}
1131
1132static inline void dec_lruvec_page_state(struct page *page,
1133                                         enum node_stat_item idx)
1134{
1135        mod_lruvec_page_state(page, idx, -1);
1136}
1137
1138#ifdef CONFIG_CGROUP_WRITEBACK
1139
1140struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
1141void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
1142                         unsigned long *pheadroom, unsigned long *pdirty,
1143                         unsigned long *pwriteback);
1144
1145#else   /* CONFIG_CGROUP_WRITEBACK */
1146
1147static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
1148{
1149        return NULL;
1150}
1151
1152static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
1153                                       unsigned long *pfilepages,
1154                                       unsigned long *pheadroom,
1155                                       unsigned long *pdirty,
1156                                       unsigned long *pwriteback)
1157{
1158}
1159
1160#endif  /* CONFIG_CGROUP_WRITEBACK */
1161
1162struct sock;
1163bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
1164void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages);
1165#ifdef CONFIG_MEMCG
1166extern struct static_key_false memcg_sockets_enabled_key;
1167#define mem_cgroup_sockets_enabled static_branch_unlikely(&memcg_sockets_enabled_key)
1168void mem_cgroup_sk_alloc(struct sock *sk);
1169void mem_cgroup_sk_free(struct sock *sk);
1170static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
1171{
1172        if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_pressure)
1173                return true;
1174        do {
1175                if (time_before(jiffies, memcg->socket_pressure))
1176                        return true;
1177        } while ((memcg = parent_mem_cgroup(memcg)));
1178        return false;
1179}
1180#else
1181#define mem_cgroup_sockets_enabled 0
1182static inline void mem_cgroup_sk_alloc(struct sock *sk) { };
1183static inline void mem_cgroup_sk_free(struct sock *sk) { };
1184static inline bool mem_cgroup_under_socket_pressure(struct mem_cgroup *memcg)
1185{
1186        return false;
1187}
1188#endif
1189
1190struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep);
1191void memcg_kmem_put_cache(struct kmem_cache *cachep);
1192int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
1193                            struct mem_cgroup *memcg);
1194int memcg_kmem_charge(struct page *page, gfp_t gfp, int order);
1195void memcg_kmem_uncharge(struct page *page, int order);
1196
1197#if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
1198extern struct static_key_false memcg_kmem_enabled_key;
1199extern struct workqueue_struct *memcg_kmem_cache_wq;
1200
1201extern int memcg_nr_cache_ids;
1202void memcg_get_cache_ids(void);
1203void memcg_put_cache_ids(void);
1204
1205/*
1206 * Helper macro to loop through all memcg-specific caches. Callers must still
1207 * check if the cache is valid (it is either valid or NULL).
1208 * the slab_mutex must be held when looping through those caches
1209 */
1210#define for_each_memcg_cache_index(_idx)        \
1211        for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
1212
1213static inline bool memcg_kmem_enabled(void)
1214{
1215        return static_branch_unlikely(&memcg_kmem_enabled_key);
1216}
1217
1218/*
1219 * helper for accessing a memcg's index. It will be used as an index in the
1220 * child cache array in kmem_cache, and also to derive its name. This function
1221 * will return -1 when this is not a kmem-limited memcg.
1222 */
1223static inline int memcg_cache_id(struct mem_cgroup *memcg)
1224{
1225        return memcg ? memcg->kmemcg_id : -1;
1226}
1227
1228#else
1229#define for_each_memcg_cache_index(_idx)        \
1230        for (; NULL; )
1231
1232static inline bool memcg_kmem_enabled(void)
1233{
1234        return false;
1235}
1236
1237static inline int memcg_cache_id(struct mem_cgroup *memcg)
1238{
1239        return -1;
1240}
1241
1242static inline void memcg_get_cache_ids(void)
1243{
1244}
1245
1246static inline void memcg_put_cache_ids(void)
1247{
1248}
1249
1250#endif /* CONFIG_MEMCG && !CONFIG_SLOB */
1251
1252#endif /* _LINUX_MEMCONTROL_H */
1253