linux/include/linux/mmzone.h
<<
>>
Prefs
   1/* SPDX-License-Identifier: GPL-2.0 */
   2#ifndef _LINUX_MMZONE_H
   3#define _LINUX_MMZONE_H
   4
   5#ifndef __ASSEMBLY__
   6#ifndef __GENERATING_BOUNDS_H
   7
   8#include <linux/spinlock.h>
   9#include <linux/list.h>
  10#include <linux/wait.h>
  11#include <linux/bitops.h>
  12#include <linux/cache.h>
  13#include <linux/threads.h>
  14#include <linux/numa.h>
  15#include <linux/init.h>
  16#include <linux/seqlock.h>
  17#include <linux/nodemask.h>
  18#include <linux/pageblock-flags.h>
  19#include <linux/page-flags-layout.h>
  20#include <linux/atomic.h>
  21#include <linux/mm_types.h>
  22#include <linux/page-flags.h>
  23#include <asm/page.h>
  24
  25/* Free memory management - zoned buddy allocator.  */
  26#ifndef CONFIG_FORCE_MAX_ZONEORDER
  27#define MAX_ORDER 11
  28#else
  29#define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
  30#endif
  31#define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
  32
  33/*
  34 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
  35 * costly to service.  That is between allocation orders which should
  36 * coalesce naturally under reasonable reclaim pressure and those which
  37 * will not.
  38 */
  39#define PAGE_ALLOC_COSTLY_ORDER 3
  40
  41enum migratetype {
  42        MIGRATE_UNMOVABLE,
  43        MIGRATE_MOVABLE,
  44        MIGRATE_RECLAIMABLE,
  45        MIGRATE_PCPTYPES,       /* the number of types on the pcp lists */
  46        MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
  47#ifdef CONFIG_CMA
  48        /*
  49         * MIGRATE_CMA migration type is designed to mimic the way
  50         * ZONE_MOVABLE works.  Only movable pages can be allocated
  51         * from MIGRATE_CMA pageblocks and page allocator never
  52         * implicitly change migration type of MIGRATE_CMA pageblock.
  53         *
  54         * The way to use it is to change migratetype of a range of
  55         * pageblocks to MIGRATE_CMA which can be done by
  56         * __free_pageblock_cma() function.  What is important though
  57         * is that a range of pageblocks must be aligned to
  58         * MAX_ORDER_NR_PAGES should biggest page be bigger then
  59         * a single pageblock.
  60         */
  61        MIGRATE_CMA,
  62#endif
  63#ifdef CONFIG_MEMORY_ISOLATION
  64        MIGRATE_ISOLATE,        /* can't allocate from here */
  65#endif
  66        MIGRATE_TYPES
  67};
  68
  69/* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
  70extern const char * const migratetype_names[MIGRATE_TYPES];
  71
  72#ifdef CONFIG_CMA
  73#  define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
  74#  define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
  75#else
  76#  define is_migrate_cma(migratetype) false
  77#  define is_migrate_cma_page(_page) false
  78#endif
  79
  80static inline bool is_migrate_movable(int mt)
  81{
  82        return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
  83}
  84
  85#define for_each_migratetype_order(order, type) \
  86        for (order = 0; order < MAX_ORDER; order++) \
  87                for (type = 0; type < MIGRATE_TYPES; type++)
  88
  89extern int page_group_by_mobility_disabled;
  90
  91#define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1)
  92#define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1)
  93
  94#define get_pageblock_migratetype(page)                                 \
  95        get_pfnblock_flags_mask(page, page_to_pfn(page),                \
  96                        PB_migrate_end, MIGRATETYPE_MASK)
  97
  98struct free_area {
  99        struct list_head        free_list[MIGRATE_TYPES];
 100        unsigned long           nr_free;
 101};
 102
 103/* Used for pages not on another list */
 104static inline void add_to_free_area(struct page *page, struct free_area *area,
 105                             int migratetype)
 106{
 107        list_add(&page->lru, &area->free_list[migratetype]);
 108        area->nr_free++;
 109}
 110
 111/* Used for pages not on another list */
 112static inline void add_to_free_area_tail(struct page *page, struct free_area *area,
 113                                  int migratetype)
 114{
 115        list_add_tail(&page->lru, &area->free_list[migratetype]);
 116        area->nr_free++;
 117}
 118
 119#ifdef CONFIG_SHUFFLE_PAGE_ALLOCATOR
 120/* Used to preserve page allocation order entropy */
 121void add_to_free_area_random(struct page *page, struct free_area *area,
 122                int migratetype);
 123#else
 124static inline void add_to_free_area_random(struct page *page,
 125                struct free_area *area, int migratetype)
 126{
 127        add_to_free_area(page, area, migratetype);
 128}
 129#endif
 130
 131/* Used for pages which are on another list */
 132static inline void move_to_free_area(struct page *page, struct free_area *area,
 133                             int migratetype)
 134{
 135        list_move(&page->lru, &area->free_list[migratetype]);
 136}
 137
 138static inline struct page *get_page_from_free_area(struct free_area *area,
 139                                            int migratetype)
 140{
 141        return list_first_entry_or_null(&area->free_list[migratetype],
 142                                        struct page, lru);
 143}
 144
 145static inline void del_page_from_free_area(struct page *page,
 146                struct free_area *area)
 147{
 148        list_del(&page->lru);
 149        __ClearPageBuddy(page);
 150        set_page_private(page, 0);
 151        area->nr_free--;
 152}
 153
 154static inline bool free_area_empty(struct free_area *area, int migratetype)
 155{
 156        return list_empty(&area->free_list[migratetype]);
 157}
 158
 159struct pglist_data;
 160
 161/*
 162 * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
 163 * So add a wild amount of padding here to ensure that they fall into separate
 164 * cachelines.  There are very few zone structures in the machine, so space
 165 * consumption is not a concern here.
 166 */
 167#if defined(CONFIG_SMP)
 168struct zone_padding {
 169        char x[0];
 170} ____cacheline_internodealigned_in_smp;
 171#define ZONE_PADDING(name)      struct zone_padding name;
 172#else
 173#define ZONE_PADDING(name)
 174#endif
 175
 176#ifdef CONFIG_NUMA
 177enum numa_stat_item {
 178        NUMA_HIT,               /* allocated in intended node */
 179        NUMA_MISS,              /* allocated in non intended node */
 180        NUMA_FOREIGN,           /* was intended here, hit elsewhere */
 181        NUMA_INTERLEAVE_HIT,    /* interleaver preferred this zone */
 182        NUMA_LOCAL,             /* allocation from local node */
 183        NUMA_OTHER,             /* allocation from other node */
 184        NR_VM_NUMA_STAT_ITEMS
 185};
 186#else
 187#define NR_VM_NUMA_STAT_ITEMS 0
 188#endif
 189
 190enum zone_stat_item {
 191        /* First 128 byte cacheline (assuming 64 bit words) */
 192        NR_FREE_PAGES,
 193        NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
 194        NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
 195        NR_ZONE_ACTIVE_ANON,
 196        NR_ZONE_INACTIVE_FILE,
 197        NR_ZONE_ACTIVE_FILE,
 198        NR_ZONE_UNEVICTABLE,
 199        NR_ZONE_WRITE_PENDING,  /* Count of dirty, writeback and unstable pages */
 200        NR_MLOCK,               /* mlock()ed pages found and moved off LRU */
 201        NR_PAGETABLE,           /* used for pagetables */
 202        NR_KERNEL_STACK_KB,     /* measured in KiB */
 203        /* Second 128 byte cacheline */
 204        NR_BOUNCE,
 205#if IS_ENABLED(CONFIG_ZSMALLOC)
 206        NR_ZSPAGES,             /* allocated in zsmalloc */
 207#endif
 208        NR_FREE_CMA_PAGES,
 209        NR_VM_ZONE_STAT_ITEMS };
 210
 211enum node_stat_item {
 212        NR_LRU_BASE,
 213        NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
 214        NR_ACTIVE_ANON,         /*  "     "     "   "       "         */
 215        NR_INACTIVE_FILE,       /*  "     "     "   "       "         */
 216        NR_ACTIVE_FILE,         /*  "     "     "   "       "         */
 217        NR_UNEVICTABLE,         /*  "     "     "   "       "         */
 218        NR_SLAB_RECLAIMABLE,
 219        NR_SLAB_UNRECLAIMABLE,
 220        NR_ISOLATED_ANON,       /* Temporary isolated pages from anon lru */
 221        NR_ISOLATED_FILE,       /* Temporary isolated pages from file lru */
 222        WORKINGSET_NODES,
 223        WORKINGSET_REFAULT,
 224        WORKINGSET_ACTIVATE,
 225        WORKINGSET_RESTORE,
 226        WORKINGSET_NODERECLAIM,
 227        NR_ANON_MAPPED, /* Mapped anonymous pages */
 228        NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
 229                           only modified from process context */
 230        NR_FILE_PAGES,
 231        NR_FILE_DIRTY,
 232        NR_WRITEBACK,
 233        NR_WRITEBACK_TEMP,      /* Writeback using temporary buffers */
 234        NR_SHMEM,               /* shmem pages (included tmpfs/GEM pages) */
 235        NR_SHMEM_THPS,
 236        NR_SHMEM_PMDMAPPED,
 237        NR_ANON_THPS,
 238        NR_UNSTABLE_NFS,        /* NFS unstable pages */
 239        NR_VMSCAN_WRITE,
 240        NR_VMSCAN_IMMEDIATE,    /* Prioritise for reclaim when writeback ends */
 241        NR_DIRTIED,             /* page dirtyings since bootup */
 242        NR_WRITTEN,             /* page writings since bootup */
 243        NR_KERNEL_MISC_RECLAIMABLE,     /* reclaimable non-slab kernel pages */
 244        NR_VM_NODE_STAT_ITEMS
 245};
 246
 247/*
 248 * We do arithmetic on the LRU lists in various places in the code,
 249 * so it is important to keep the active lists LRU_ACTIVE higher in
 250 * the array than the corresponding inactive lists, and to keep
 251 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
 252 *
 253 * This has to be kept in sync with the statistics in zone_stat_item
 254 * above and the descriptions in vmstat_text in mm/vmstat.c
 255 */
 256#define LRU_BASE 0
 257#define LRU_ACTIVE 1
 258#define LRU_FILE 2
 259
 260enum lru_list {
 261        LRU_INACTIVE_ANON = LRU_BASE,
 262        LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
 263        LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
 264        LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
 265        LRU_UNEVICTABLE,
 266        NR_LRU_LISTS
 267};
 268
 269#define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
 270
 271#define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
 272
 273static inline int is_file_lru(enum lru_list lru)
 274{
 275        return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
 276}
 277
 278static inline int is_active_lru(enum lru_list lru)
 279{
 280        return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
 281}
 282
 283struct zone_reclaim_stat {
 284        /*
 285         * The pageout code in vmscan.c keeps track of how many of the
 286         * mem/swap backed and file backed pages are referenced.
 287         * The higher the rotated/scanned ratio, the more valuable
 288         * that cache is.
 289         *
 290         * The anon LRU stats live in [0], file LRU stats in [1]
 291         */
 292        unsigned long           recent_rotated[2];
 293        unsigned long           recent_scanned[2];
 294};
 295
 296struct lruvec {
 297        struct list_head                lists[NR_LRU_LISTS];
 298        struct zone_reclaim_stat        reclaim_stat;
 299        /* Evictions & activations on the inactive file list */
 300        atomic_long_t                   inactive_age;
 301        /* Refaults at the time of last reclaim cycle */
 302        unsigned long                   refaults;
 303#ifdef CONFIG_MEMCG
 304        struct pglist_data *pgdat;
 305#endif
 306};
 307
 308/* Isolate unmapped file */
 309#define ISOLATE_UNMAPPED        ((__force isolate_mode_t)0x2)
 310/* Isolate for asynchronous migration */
 311#define ISOLATE_ASYNC_MIGRATE   ((__force isolate_mode_t)0x4)
 312/* Isolate unevictable pages */
 313#define ISOLATE_UNEVICTABLE     ((__force isolate_mode_t)0x8)
 314
 315/* LRU Isolation modes. */
 316typedef unsigned __bitwise isolate_mode_t;
 317
 318enum zone_watermarks {
 319        WMARK_MIN,
 320        WMARK_LOW,
 321        WMARK_HIGH,
 322        NR_WMARK
 323};
 324
 325#define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
 326#define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
 327#define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
 328#define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
 329
 330struct per_cpu_pages {
 331        int count;              /* number of pages in the list */
 332        int high;               /* high watermark, emptying needed */
 333        int batch;              /* chunk size for buddy add/remove */
 334
 335        /* Lists of pages, one per migrate type stored on the pcp-lists */
 336        struct list_head lists[MIGRATE_PCPTYPES];
 337};
 338
 339struct per_cpu_pageset {
 340        struct per_cpu_pages pcp;
 341#ifdef CONFIG_NUMA
 342        s8 expire;
 343        u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
 344#endif
 345#ifdef CONFIG_SMP
 346        s8 stat_threshold;
 347        s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
 348#endif
 349};
 350
 351struct per_cpu_nodestat {
 352        s8 stat_threshold;
 353        s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
 354};
 355
 356#endif /* !__GENERATING_BOUNDS.H */
 357
 358enum zone_type {
 359#ifdef CONFIG_ZONE_DMA
 360        /*
 361         * ZONE_DMA is used when there are devices that are not able
 362         * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
 363         * carve out the portion of memory that is needed for these devices.
 364         * The range is arch specific.
 365         *
 366         * Some examples
 367         *
 368         * Architecture         Limit
 369         * ---------------------------
 370         * parisc, ia64, sparc  <4G
 371         * s390, powerpc        <2G
 372         * arm                  Various
 373         * alpha                Unlimited or 0-16MB.
 374         *
 375         * i386, x86_64 and multiple other arches
 376         *                      <16M.
 377         */
 378        ZONE_DMA,
 379#endif
 380#ifdef CONFIG_ZONE_DMA32
 381        /*
 382         * x86_64 needs two ZONE_DMAs because it supports devices that are
 383         * only able to do DMA to the lower 16M but also 32 bit devices that
 384         * can only do DMA areas below 4G.
 385         */
 386        ZONE_DMA32,
 387#endif
 388        /*
 389         * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
 390         * performed on pages in ZONE_NORMAL if the DMA devices support
 391         * transfers to all addressable memory.
 392         */
 393        ZONE_NORMAL,
 394#ifdef CONFIG_HIGHMEM
 395        /*
 396         * A memory area that is only addressable by the kernel through
 397         * mapping portions into its own address space. This is for example
 398         * used by i386 to allow the kernel to address the memory beyond
 399         * 900MB. The kernel will set up special mappings (page
 400         * table entries on i386) for each page that the kernel needs to
 401         * access.
 402         */
 403        ZONE_HIGHMEM,
 404#endif
 405        ZONE_MOVABLE,
 406#ifdef CONFIG_ZONE_DEVICE
 407        ZONE_DEVICE,
 408#endif
 409        __MAX_NR_ZONES
 410
 411};
 412
 413#ifndef __GENERATING_BOUNDS_H
 414
 415struct zone {
 416        /* Read-mostly fields */
 417
 418        /* zone watermarks, access with *_wmark_pages(zone) macros */
 419        unsigned long _watermark[NR_WMARK];
 420        unsigned long watermark_boost;
 421
 422        unsigned long nr_reserved_highatomic;
 423
 424        /*
 425         * We don't know if the memory that we're going to allocate will be
 426         * freeable or/and it will be released eventually, so to avoid totally
 427         * wasting several GB of ram we must reserve some of the lower zone
 428         * memory (otherwise we risk to run OOM on the lower zones despite
 429         * there being tons of freeable ram on the higher zones).  This array is
 430         * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
 431         * changes.
 432         */
 433        long lowmem_reserve[MAX_NR_ZONES];
 434
 435#ifdef CONFIG_NUMA
 436        int node;
 437#endif
 438        struct pglist_data      *zone_pgdat;
 439        struct per_cpu_pageset __percpu *pageset;
 440
 441#ifndef CONFIG_SPARSEMEM
 442        /*
 443         * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
 444         * In SPARSEMEM, this map is stored in struct mem_section
 445         */
 446        unsigned long           *pageblock_flags;
 447#endif /* CONFIG_SPARSEMEM */
 448
 449        /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
 450        unsigned long           zone_start_pfn;
 451
 452        /*
 453         * spanned_pages is the total pages spanned by the zone, including
 454         * holes, which is calculated as:
 455         *      spanned_pages = zone_end_pfn - zone_start_pfn;
 456         *
 457         * present_pages is physical pages existing within the zone, which
 458         * is calculated as:
 459         *      present_pages = spanned_pages - absent_pages(pages in holes);
 460         *
 461         * managed_pages is present pages managed by the buddy system, which
 462         * is calculated as (reserved_pages includes pages allocated by the
 463         * bootmem allocator):
 464         *      managed_pages = present_pages - reserved_pages;
 465         *
 466         * So present_pages may be used by memory hotplug or memory power
 467         * management logic to figure out unmanaged pages by checking
 468         * (present_pages - managed_pages). And managed_pages should be used
 469         * by page allocator and vm scanner to calculate all kinds of watermarks
 470         * and thresholds.
 471         *
 472         * Locking rules:
 473         *
 474         * zone_start_pfn and spanned_pages are protected by span_seqlock.
 475         * It is a seqlock because it has to be read outside of zone->lock,
 476         * and it is done in the main allocator path.  But, it is written
 477         * quite infrequently.
 478         *
 479         * The span_seq lock is declared along with zone->lock because it is
 480         * frequently read in proximity to zone->lock.  It's good to
 481         * give them a chance of being in the same cacheline.
 482         *
 483         * Write access to present_pages at runtime should be protected by
 484         * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
 485         * present_pages should get_online_mems() to get a stable value.
 486         */
 487        atomic_long_t           managed_pages;
 488        unsigned long           spanned_pages;
 489        unsigned long           present_pages;
 490
 491        const char              *name;
 492
 493#ifdef CONFIG_MEMORY_ISOLATION
 494        /*
 495         * Number of isolated pageblock. It is used to solve incorrect
 496         * freepage counting problem due to racy retrieving migratetype
 497         * of pageblock. Protected by zone->lock.
 498         */
 499        unsigned long           nr_isolate_pageblock;
 500#endif
 501
 502#ifdef CONFIG_MEMORY_HOTPLUG
 503        /* see spanned/present_pages for more description */
 504        seqlock_t               span_seqlock;
 505#endif
 506
 507        int initialized;
 508
 509        /* Write-intensive fields used from the page allocator */
 510        ZONE_PADDING(_pad1_)
 511
 512        /* free areas of different sizes */
 513        struct free_area        free_area[MAX_ORDER];
 514
 515        /* zone flags, see below */
 516        unsigned long           flags;
 517
 518        /* Primarily protects free_area */
 519        spinlock_t              lock;
 520
 521        /* Write-intensive fields used by compaction and vmstats. */
 522        ZONE_PADDING(_pad2_)
 523
 524        /*
 525         * When free pages are below this point, additional steps are taken
 526         * when reading the number of free pages to avoid per-cpu counter
 527         * drift allowing watermarks to be breached
 528         */
 529        unsigned long percpu_drift_mark;
 530
 531#if defined CONFIG_COMPACTION || defined CONFIG_CMA
 532        /* pfn where compaction free scanner should start */
 533        unsigned long           compact_cached_free_pfn;
 534        /* pfn where async and sync compaction migration scanner should start */
 535        unsigned long           compact_cached_migrate_pfn[2];
 536        unsigned long           compact_init_migrate_pfn;
 537        unsigned long           compact_init_free_pfn;
 538#endif
 539
 540#ifdef CONFIG_COMPACTION
 541        /*
 542         * On compaction failure, 1<<compact_defer_shift compactions
 543         * are skipped before trying again. The number attempted since
 544         * last failure is tracked with compact_considered.
 545         */
 546        unsigned int            compact_considered;
 547        unsigned int            compact_defer_shift;
 548        int                     compact_order_failed;
 549#endif
 550
 551#if defined CONFIG_COMPACTION || defined CONFIG_CMA
 552        /* Set to true when the PG_migrate_skip bits should be cleared */
 553        bool                    compact_blockskip_flush;
 554#endif
 555
 556        bool                    contiguous;
 557
 558        ZONE_PADDING(_pad3_)
 559        /* Zone statistics */
 560        atomic_long_t           vm_stat[NR_VM_ZONE_STAT_ITEMS];
 561        atomic_long_t           vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
 562} ____cacheline_internodealigned_in_smp;
 563
 564enum pgdat_flags {
 565        PGDAT_CONGESTED,                /* pgdat has many dirty pages backed by
 566                                         * a congested BDI
 567                                         */
 568        PGDAT_DIRTY,                    /* reclaim scanning has recently found
 569                                         * many dirty file pages at the tail
 570                                         * of the LRU.
 571                                         */
 572        PGDAT_WRITEBACK,                /* reclaim scanning has recently found
 573                                         * many pages under writeback
 574                                         */
 575        PGDAT_RECLAIM_LOCKED,           /* prevents concurrent reclaim */
 576};
 577
 578enum zone_flags {
 579        ZONE_BOOSTED_WATERMARK,         /* zone recently boosted watermarks.
 580                                         * Cleared when kswapd is woken.
 581                                         */
 582};
 583
 584static inline unsigned long zone_managed_pages(struct zone *zone)
 585{
 586        return (unsigned long)atomic_long_read(&zone->managed_pages);
 587}
 588
 589static inline unsigned long zone_end_pfn(const struct zone *zone)
 590{
 591        return zone->zone_start_pfn + zone->spanned_pages;
 592}
 593
 594static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
 595{
 596        return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
 597}
 598
 599static inline bool zone_is_initialized(struct zone *zone)
 600{
 601        return zone->initialized;
 602}
 603
 604static inline bool zone_is_empty(struct zone *zone)
 605{
 606        return zone->spanned_pages == 0;
 607}
 608
 609/*
 610 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
 611 * intersection with the given zone
 612 */
 613static inline bool zone_intersects(struct zone *zone,
 614                unsigned long start_pfn, unsigned long nr_pages)
 615{
 616        if (zone_is_empty(zone))
 617                return false;
 618        if (start_pfn >= zone_end_pfn(zone) ||
 619            start_pfn + nr_pages <= zone->zone_start_pfn)
 620                return false;
 621
 622        return true;
 623}
 624
 625/*
 626 * The "priority" of VM scanning is how much of the queues we will scan in one
 627 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
 628 * queues ("queue_length >> 12") during an aging round.
 629 */
 630#define DEF_PRIORITY 12
 631
 632/* Maximum number of zones on a zonelist */
 633#define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
 634
 635enum {
 636        ZONELIST_FALLBACK,      /* zonelist with fallback */
 637#ifdef CONFIG_NUMA
 638        /*
 639         * The NUMA zonelists are doubled because we need zonelists that
 640         * restrict the allocations to a single node for __GFP_THISNODE.
 641         */
 642        ZONELIST_NOFALLBACK,    /* zonelist without fallback (__GFP_THISNODE) */
 643#endif
 644        MAX_ZONELISTS
 645};
 646
 647/*
 648 * This struct contains information about a zone in a zonelist. It is stored
 649 * here to avoid dereferences into large structures and lookups of tables
 650 */
 651struct zoneref {
 652        struct zone *zone;      /* Pointer to actual zone */
 653        int zone_idx;           /* zone_idx(zoneref->zone) */
 654};
 655
 656/*
 657 * One allocation request operates on a zonelist. A zonelist
 658 * is a list of zones, the first one is the 'goal' of the
 659 * allocation, the other zones are fallback zones, in decreasing
 660 * priority.
 661 *
 662 * To speed the reading of the zonelist, the zonerefs contain the zone index
 663 * of the entry being read. Helper functions to access information given
 664 * a struct zoneref are
 665 *
 666 * zonelist_zone()      - Return the struct zone * for an entry in _zonerefs
 667 * zonelist_zone_idx()  - Return the index of the zone for an entry
 668 * zonelist_node_idx()  - Return the index of the node for an entry
 669 */
 670struct zonelist {
 671        struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
 672};
 673
 674#ifndef CONFIG_DISCONTIGMEM
 675/* The array of struct pages - for discontigmem use pgdat->lmem_map */
 676extern struct page *mem_map;
 677#endif
 678
 679/*
 680 * On NUMA machines, each NUMA node would have a pg_data_t to describe
 681 * it's memory layout. On UMA machines there is a single pglist_data which
 682 * describes the whole memory.
 683 *
 684 * Memory statistics and page replacement data structures are maintained on a
 685 * per-zone basis.
 686 */
 687struct bootmem_data;
 688typedef struct pglist_data {
 689        struct zone node_zones[MAX_NR_ZONES];
 690        struct zonelist node_zonelists[MAX_ZONELISTS];
 691        int nr_zones;
 692#ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
 693        struct page *node_mem_map;
 694#ifdef CONFIG_PAGE_EXTENSION
 695        struct page_ext *node_page_ext;
 696#endif
 697#endif
 698#if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
 699        /*
 700         * Must be held any time you expect node_start_pfn,
 701         * node_present_pages, node_spanned_pages or nr_zones to stay constant.
 702         *
 703         * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
 704         * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
 705         * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
 706         *
 707         * Nests above zone->lock and zone->span_seqlock
 708         */
 709        spinlock_t node_size_lock;
 710#endif
 711        unsigned long node_start_pfn;
 712        unsigned long node_present_pages; /* total number of physical pages */
 713        unsigned long node_spanned_pages; /* total size of physical page
 714                                             range, including holes */
 715        int node_id;
 716        wait_queue_head_t kswapd_wait;
 717        wait_queue_head_t pfmemalloc_wait;
 718        struct task_struct *kswapd;     /* Protected by
 719                                           mem_hotplug_begin/end() */
 720        int kswapd_order;
 721        enum zone_type kswapd_classzone_idx;
 722
 723        int kswapd_failures;            /* Number of 'reclaimed == 0' runs */
 724
 725#ifdef CONFIG_COMPACTION
 726        int kcompactd_max_order;
 727        enum zone_type kcompactd_classzone_idx;
 728        wait_queue_head_t kcompactd_wait;
 729        struct task_struct *kcompactd;
 730#endif
 731        /*
 732         * This is a per-node reserve of pages that are not available
 733         * to userspace allocations.
 734         */
 735        unsigned long           totalreserve_pages;
 736
 737#ifdef CONFIG_NUMA
 738        /*
 739         * zone reclaim becomes active if more unmapped pages exist.
 740         */
 741        unsigned long           min_unmapped_pages;
 742        unsigned long           min_slab_pages;
 743#endif /* CONFIG_NUMA */
 744
 745        /* Write-intensive fields used by page reclaim */
 746        ZONE_PADDING(_pad1_)
 747        spinlock_t              lru_lock;
 748
 749#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
 750        /*
 751         * If memory initialisation on large machines is deferred then this
 752         * is the first PFN that needs to be initialised.
 753         */
 754        unsigned long first_deferred_pfn;
 755#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
 756
 757#ifdef CONFIG_TRANSPARENT_HUGEPAGE
 758        spinlock_t split_queue_lock;
 759        struct list_head split_queue;
 760        unsigned long split_queue_len;
 761#endif
 762
 763        /* Fields commonly accessed by the page reclaim scanner */
 764        struct lruvec           lruvec;
 765
 766        unsigned long           flags;
 767
 768        ZONE_PADDING(_pad2_)
 769
 770        /* Per-node vmstats */
 771        struct per_cpu_nodestat __percpu *per_cpu_nodestats;
 772        atomic_long_t           vm_stat[NR_VM_NODE_STAT_ITEMS];
 773} pg_data_t;
 774
 775#define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
 776#define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
 777#ifdef CONFIG_FLAT_NODE_MEM_MAP
 778#define pgdat_page_nr(pgdat, pagenr)    ((pgdat)->node_mem_map + (pagenr))
 779#else
 780#define pgdat_page_nr(pgdat, pagenr)    pfn_to_page((pgdat)->node_start_pfn + (pagenr))
 781#endif
 782#define nid_page_nr(nid, pagenr)        pgdat_page_nr(NODE_DATA(nid),(pagenr))
 783
 784#define node_start_pfn(nid)     (NODE_DATA(nid)->node_start_pfn)
 785#define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
 786
 787static inline struct lruvec *node_lruvec(struct pglist_data *pgdat)
 788{
 789        return &pgdat->lruvec;
 790}
 791
 792static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
 793{
 794        return pgdat->node_start_pfn + pgdat->node_spanned_pages;
 795}
 796
 797static inline bool pgdat_is_empty(pg_data_t *pgdat)
 798{
 799        return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
 800}
 801
 802#include <linux/memory_hotplug.h>
 803
 804void build_all_zonelists(pg_data_t *pgdat);
 805void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
 806                   enum zone_type classzone_idx);
 807bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
 808                         int classzone_idx, unsigned int alloc_flags,
 809                         long free_pages);
 810bool zone_watermark_ok(struct zone *z, unsigned int order,
 811                unsigned long mark, int classzone_idx,
 812                unsigned int alloc_flags);
 813bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
 814                unsigned long mark, int classzone_idx);
 815enum memmap_context {
 816        MEMMAP_EARLY,
 817        MEMMAP_HOTPLUG,
 818};
 819extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
 820                                     unsigned long size);
 821
 822extern void lruvec_init(struct lruvec *lruvec);
 823
 824static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
 825{
 826#ifdef CONFIG_MEMCG
 827        return lruvec->pgdat;
 828#else
 829        return container_of(lruvec, struct pglist_data, lruvec);
 830#endif
 831}
 832
 833extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
 834
 835#ifdef CONFIG_HAVE_MEMORY_PRESENT
 836void memory_present(int nid, unsigned long start, unsigned long end);
 837#else
 838static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
 839#endif
 840
 841#if defined(CONFIG_SPARSEMEM)
 842void memblocks_present(void);
 843#else
 844static inline void memblocks_present(void) {}
 845#endif
 846
 847#ifdef CONFIG_HAVE_MEMORYLESS_NODES
 848int local_memory_node(int node_id);
 849#else
 850static inline int local_memory_node(int node_id) { return node_id; };
 851#endif
 852
 853/*
 854 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
 855 */
 856#define zone_idx(zone)          ((zone) - (zone)->zone_pgdat->node_zones)
 857
 858#ifdef CONFIG_ZONE_DEVICE
 859static inline bool is_dev_zone(const struct zone *zone)
 860{
 861        return zone_idx(zone) == ZONE_DEVICE;
 862}
 863#else
 864static inline bool is_dev_zone(const struct zone *zone)
 865{
 866        return false;
 867}
 868#endif
 869
 870/*
 871 * Returns true if a zone has pages managed by the buddy allocator.
 872 * All the reclaim decisions have to use this function rather than
 873 * populated_zone(). If the whole zone is reserved then we can easily
 874 * end up with populated_zone() && !managed_zone().
 875 */
 876static inline bool managed_zone(struct zone *zone)
 877{
 878        return zone_managed_pages(zone);
 879}
 880
 881/* Returns true if a zone has memory */
 882static inline bool populated_zone(struct zone *zone)
 883{
 884        return zone->present_pages;
 885}
 886
 887#ifdef CONFIG_NUMA
 888static inline int zone_to_nid(struct zone *zone)
 889{
 890        return zone->node;
 891}
 892
 893static inline void zone_set_nid(struct zone *zone, int nid)
 894{
 895        zone->node = nid;
 896}
 897#else
 898static inline int zone_to_nid(struct zone *zone)
 899{
 900        return 0;
 901}
 902
 903static inline void zone_set_nid(struct zone *zone, int nid) {}
 904#endif
 905
 906extern int movable_zone;
 907
 908#ifdef CONFIG_HIGHMEM
 909static inline int zone_movable_is_highmem(void)
 910{
 911#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
 912        return movable_zone == ZONE_HIGHMEM;
 913#else
 914        return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
 915#endif
 916}
 917#endif
 918
 919static inline int is_highmem_idx(enum zone_type idx)
 920{
 921#ifdef CONFIG_HIGHMEM
 922        return (idx == ZONE_HIGHMEM ||
 923                (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
 924#else
 925        return 0;
 926#endif
 927}
 928
 929/**
 930 * is_highmem - helper function to quickly check if a struct zone is a
 931 *              highmem zone or not.  This is an attempt to keep references
 932 *              to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
 933 * @zone - pointer to struct zone variable
 934 */
 935static inline int is_highmem(struct zone *zone)
 936{
 937#ifdef CONFIG_HIGHMEM
 938        return is_highmem_idx(zone_idx(zone));
 939#else
 940        return 0;
 941#endif
 942}
 943
 944/* These two functions are used to setup the per zone pages min values */
 945struct ctl_table;
 946int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
 947                                        void __user *, size_t *, loff_t *);
 948int watermark_boost_factor_sysctl_handler(struct ctl_table *, int,
 949                                        void __user *, size_t *, loff_t *);
 950int watermark_scale_factor_sysctl_handler(struct ctl_table *, int,
 951                                        void __user *, size_t *, loff_t *);
 952extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
 953int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
 954                                        void __user *, size_t *, loff_t *);
 955int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
 956                                        void __user *, size_t *, loff_t *);
 957int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
 958                        void __user *, size_t *, loff_t *);
 959int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
 960                        void __user *, size_t *, loff_t *);
 961
 962extern int numa_zonelist_order_handler(struct ctl_table *, int,
 963                        void __user *, size_t *, loff_t *);
 964extern char numa_zonelist_order[];
 965#define NUMA_ZONELIST_ORDER_LEN 16
 966
 967#ifndef CONFIG_NEED_MULTIPLE_NODES
 968
 969extern struct pglist_data contig_page_data;
 970#define NODE_DATA(nid)          (&contig_page_data)
 971#define NODE_MEM_MAP(nid)       mem_map
 972
 973#else /* CONFIG_NEED_MULTIPLE_NODES */
 974
 975#include <asm/mmzone.h>
 976
 977#endif /* !CONFIG_NEED_MULTIPLE_NODES */
 978
 979extern struct pglist_data *first_online_pgdat(void);
 980extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
 981extern struct zone *next_zone(struct zone *zone);
 982
 983/**
 984 * for_each_online_pgdat - helper macro to iterate over all online nodes
 985 * @pgdat - pointer to a pg_data_t variable
 986 */
 987#define for_each_online_pgdat(pgdat)                    \
 988        for (pgdat = first_online_pgdat();              \
 989             pgdat;                                     \
 990             pgdat = next_online_pgdat(pgdat))
 991/**
 992 * for_each_zone - helper macro to iterate over all memory zones
 993 * @zone - pointer to struct zone variable
 994 *
 995 * The user only needs to declare the zone variable, for_each_zone
 996 * fills it in.
 997 */
 998#define for_each_zone(zone)                             \
 999        for (zone = (first_online_pgdat())->node_zones; \
1000             zone;                                      \
1001             zone = next_zone(zone))
1002
1003#define for_each_populated_zone(zone)                   \
1004        for (zone = (first_online_pgdat())->node_zones; \
1005             zone;                                      \
1006             zone = next_zone(zone))                    \
1007                if (!populated_zone(zone))              \
1008                        ; /* do nothing */              \
1009                else
1010
1011static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1012{
1013        return zoneref->zone;
1014}
1015
1016static inline int zonelist_zone_idx(struct zoneref *zoneref)
1017{
1018        return zoneref->zone_idx;
1019}
1020
1021static inline int zonelist_node_idx(struct zoneref *zoneref)
1022{
1023        return zone_to_nid(zoneref->zone);
1024}
1025
1026struct zoneref *__next_zones_zonelist(struct zoneref *z,
1027                                        enum zone_type highest_zoneidx,
1028                                        nodemask_t *nodes);
1029
1030/**
1031 * next_zones_zonelist - Returns the next zone at or below highest_zoneidx within the allowed nodemask using a cursor within a zonelist as a starting point
1032 * @z - The cursor used as a starting point for the search
1033 * @highest_zoneidx - The zone index of the highest zone to return
1034 * @nodes - An optional nodemask to filter the zonelist with
1035 *
1036 * This function returns the next zone at or below a given zone index that is
1037 * within the allowed nodemask using a cursor as the starting point for the
1038 * search. The zoneref returned is a cursor that represents the current zone
1039 * being examined. It should be advanced by one before calling
1040 * next_zones_zonelist again.
1041 */
1042static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1043                                        enum zone_type highest_zoneidx,
1044                                        nodemask_t *nodes)
1045{
1046        if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1047                return z;
1048        return __next_zones_zonelist(z, highest_zoneidx, nodes);
1049}
1050
1051/**
1052 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1053 * @zonelist - The zonelist to search for a suitable zone
1054 * @highest_zoneidx - The zone index of the highest zone to return
1055 * @nodes - An optional nodemask to filter the zonelist with
1056 * @return - Zoneref pointer for the first suitable zone found (see below)
1057 *
1058 * This function returns the first zone at or below a given zone index that is
1059 * within the allowed nodemask. The zoneref returned is a cursor that can be
1060 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1061 * one before calling.
1062 *
1063 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1064 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1065 * update due to cpuset modification.
1066 */
1067static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1068                                        enum zone_type highest_zoneidx,
1069                                        nodemask_t *nodes)
1070{
1071        return next_zones_zonelist(zonelist->_zonerefs,
1072                                                        highest_zoneidx, nodes);
1073}
1074
1075/**
1076 * for_each_zone_zonelist_nodemask - helper macro to iterate over valid zones in a zonelist at or below a given zone index and within a nodemask
1077 * @zone - The current zone in the iterator
1078 * @z - The current pointer within zonelist->zones being iterated
1079 * @zlist - The zonelist being iterated
1080 * @highidx - The zone index of the highest zone to return
1081 * @nodemask - Nodemask allowed by the allocator
1082 *
1083 * This iterator iterates though all zones at or below a given zone index and
1084 * within a given nodemask
1085 */
1086#define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1087        for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z);       \
1088                zone;                                                   \
1089                z = next_zones_zonelist(++z, highidx, nodemask),        \
1090                        zone = zonelist_zone(z))
1091
1092#define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1093        for (zone = z->zone;    \
1094                zone;                                                   \
1095                z = next_zones_zonelist(++z, highidx, nodemask),        \
1096                        zone = zonelist_zone(z))
1097
1098
1099/**
1100 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1101 * @zone - The current zone in the iterator
1102 * @z - The current pointer within zonelist->zones being iterated
1103 * @zlist - The zonelist being iterated
1104 * @highidx - The zone index of the highest zone to return
1105 *
1106 * This iterator iterates though all zones at or below a given zone index.
1107 */
1108#define for_each_zone_zonelist(zone, z, zlist, highidx) \
1109        for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1110
1111#ifdef CONFIG_SPARSEMEM
1112#include <asm/sparsemem.h>
1113#endif
1114
1115#if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1116        !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1117static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1118{
1119        BUILD_BUG_ON(IS_ENABLED(CONFIG_NUMA));
1120        return 0;
1121}
1122#endif
1123
1124#ifdef CONFIG_FLATMEM
1125#define pfn_to_nid(pfn)         (0)
1126#endif
1127
1128#ifdef CONFIG_SPARSEMEM
1129
1130/*
1131 * SECTION_SHIFT                #bits space required to store a section #
1132 *
1133 * PA_SECTION_SHIFT             physical address to/from section number
1134 * PFN_SECTION_SHIFT            pfn to/from section number
1135 */
1136#define PA_SECTION_SHIFT        (SECTION_SIZE_BITS)
1137#define PFN_SECTION_SHIFT       (SECTION_SIZE_BITS - PAGE_SHIFT)
1138
1139#define NR_MEM_SECTIONS         (1UL << SECTIONS_SHIFT)
1140
1141#define PAGES_PER_SECTION       (1UL << PFN_SECTION_SHIFT)
1142#define PAGE_SECTION_MASK       (~(PAGES_PER_SECTION-1))
1143
1144#define SECTION_BLOCKFLAGS_BITS \
1145        ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1146
1147#if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1148#error Allocator MAX_ORDER exceeds SECTION_SIZE
1149#endif
1150
1151static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1152{
1153        return pfn >> PFN_SECTION_SHIFT;
1154}
1155static inline unsigned long section_nr_to_pfn(unsigned long sec)
1156{
1157        return sec << PFN_SECTION_SHIFT;
1158}
1159
1160#define SECTION_ALIGN_UP(pfn)   (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1161#define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1162
1163struct page;
1164struct page_ext;
1165struct mem_section {
1166        /*
1167         * This is, logically, a pointer to an array of struct
1168         * pages.  However, it is stored with some other magic.
1169         * (see sparse.c::sparse_init_one_section())
1170         *
1171         * Additionally during early boot we encode node id of
1172         * the location of the section here to guide allocation.
1173         * (see sparse.c::memory_present())
1174         *
1175         * Making it a UL at least makes someone do a cast
1176         * before using it wrong.
1177         */
1178        unsigned long section_mem_map;
1179
1180        /* See declaration of similar field in struct zone */
1181        unsigned long *pageblock_flags;
1182#ifdef CONFIG_PAGE_EXTENSION
1183        /*
1184         * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1185         * section. (see page_ext.h about this.)
1186         */
1187        struct page_ext *page_ext;
1188        unsigned long pad;
1189#endif
1190        /*
1191         * WARNING: mem_section must be a power-of-2 in size for the
1192         * calculation and use of SECTION_ROOT_MASK to make sense.
1193         */
1194};
1195
1196#ifdef CONFIG_SPARSEMEM_EXTREME
1197#define SECTIONS_PER_ROOT       (PAGE_SIZE / sizeof (struct mem_section))
1198#else
1199#define SECTIONS_PER_ROOT       1
1200#endif
1201
1202#define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1203#define NR_SECTION_ROOTS        DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1204#define SECTION_ROOT_MASK       (SECTIONS_PER_ROOT - 1)
1205
1206#ifdef CONFIG_SPARSEMEM_EXTREME
1207extern struct mem_section **mem_section;
1208#else
1209extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1210#endif
1211
1212static inline struct mem_section *__nr_to_section(unsigned long nr)
1213{
1214#ifdef CONFIG_SPARSEMEM_EXTREME
1215        if (!mem_section)
1216                return NULL;
1217#endif
1218        if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1219                return NULL;
1220        return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1221}
1222extern int __section_nr(struct mem_section* ms);
1223extern unsigned long usemap_size(void);
1224
1225/*
1226 * We use the lower bits of the mem_map pointer to store
1227 * a little bit of information.  The pointer is calculated
1228 * as mem_map - section_nr_to_pfn(pnum).  The result is
1229 * aligned to the minimum alignment of the two values:
1230 *   1. All mem_map arrays are page-aligned.
1231 *   2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1232 *      lowest bits.  PFN_SECTION_SHIFT is arch-specific
1233 *      (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1234 *      worst combination is powerpc with 256k pages,
1235 *      which results in PFN_SECTION_SHIFT equal 6.
1236 * To sum it up, at least 6 bits are available.
1237 */
1238#define SECTION_MARKED_PRESENT  (1UL<<0)
1239#define SECTION_HAS_MEM_MAP     (1UL<<1)
1240#define SECTION_IS_ONLINE       (1UL<<2)
1241#define SECTION_MAP_LAST_BIT    (1UL<<3)
1242#define SECTION_MAP_MASK        (~(SECTION_MAP_LAST_BIT-1))
1243#define SECTION_NID_SHIFT       3
1244
1245static inline struct page *__section_mem_map_addr(struct mem_section *section)
1246{
1247        unsigned long map = section->section_mem_map;
1248        map &= SECTION_MAP_MASK;
1249        return (struct page *)map;
1250}
1251
1252static inline int present_section(struct mem_section *section)
1253{
1254        return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1255}
1256
1257static inline int present_section_nr(unsigned long nr)
1258{
1259        return present_section(__nr_to_section(nr));
1260}
1261
1262static inline int valid_section(struct mem_section *section)
1263{
1264        return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1265}
1266
1267static inline int valid_section_nr(unsigned long nr)
1268{
1269        return valid_section(__nr_to_section(nr));
1270}
1271
1272static inline int online_section(struct mem_section *section)
1273{
1274        return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1275}
1276
1277static inline int online_section_nr(unsigned long nr)
1278{
1279        return online_section(__nr_to_section(nr));
1280}
1281
1282#ifdef CONFIG_MEMORY_HOTPLUG
1283void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1284#ifdef CONFIG_MEMORY_HOTREMOVE
1285void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1286#endif
1287#endif
1288
1289static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1290{
1291        return __nr_to_section(pfn_to_section_nr(pfn));
1292}
1293
1294extern int __highest_present_section_nr;
1295
1296#ifndef CONFIG_HAVE_ARCH_PFN_VALID
1297static inline int pfn_valid(unsigned long pfn)
1298{
1299        if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1300                return 0;
1301        return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
1302}
1303#endif
1304
1305static inline int pfn_present(unsigned long pfn)
1306{
1307        if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1308                return 0;
1309        return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1310}
1311
1312/*
1313 * These are _only_ used during initialisation, therefore they
1314 * can use __initdata ...  They could have names to indicate
1315 * this restriction.
1316 */
1317#ifdef CONFIG_NUMA
1318#define pfn_to_nid(pfn)                                                 \
1319({                                                                      \
1320        unsigned long __pfn_to_nid_pfn = (pfn);                         \
1321        page_to_nid(pfn_to_page(__pfn_to_nid_pfn));                     \
1322})
1323#else
1324#define pfn_to_nid(pfn)         (0)
1325#endif
1326
1327#define early_pfn_valid(pfn)    pfn_valid(pfn)
1328void sparse_init(void);
1329#else
1330#define sparse_init()   do {} while (0)
1331#define sparse_index_init(_sec, _nid)  do {} while (0)
1332#define pfn_present pfn_valid
1333#endif /* CONFIG_SPARSEMEM */
1334
1335/*
1336 * During memory init memblocks map pfns to nids. The search is expensive and
1337 * this caches recent lookups. The implementation of __early_pfn_to_nid
1338 * may treat start/end as pfns or sections.
1339 */
1340struct mminit_pfnnid_cache {
1341        unsigned long last_start;
1342        unsigned long last_end;
1343        int last_nid;
1344};
1345
1346#ifndef early_pfn_valid
1347#define early_pfn_valid(pfn)    (1)
1348#endif
1349
1350void memory_present(int nid, unsigned long start, unsigned long end);
1351
1352/*
1353 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1354 * need to check pfn validity within that MAX_ORDER_NR_PAGES block.
1355 * pfn_valid_within() should be used in this case; we optimise this away
1356 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1357 */
1358#ifdef CONFIG_HOLES_IN_ZONE
1359#define pfn_valid_within(pfn) pfn_valid(pfn)
1360#else
1361#define pfn_valid_within(pfn) (1)
1362#endif
1363
1364#ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1365/*
1366 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1367 * associated with it or not. This means that a struct page exists for this
1368 * pfn. The caller cannot assume the page is fully initialized in general.
1369 * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1370 * will ensure the struct page is fully online and initialized. Special pages
1371 * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1372 *
1373 * In FLATMEM, it is expected that holes always have valid memmap as long as
1374 * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1375 * that a valid section has a memmap for the entire section.
1376 *
1377 * However, an ARM, and maybe other embedded architectures in the future
1378 * free memmap backing holes to save memory on the assumption the memmap is
1379 * never used. The page_zone linkages are then broken even though pfn_valid()
1380 * returns true. A walker of the full memmap must then do this additional
1381 * check to ensure the memmap they are looking at is sane by making sure
1382 * the zone and PFN linkages are still valid. This is expensive, but walkers
1383 * of the full memmap are extremely rare.
1384 */
1385bool memmap_valid_within(unsigned long pfn,
1386                                        struct page *page, struct zone *zone);
1387#else
1388static inline bool memmap_valid_within(unsigned long pfn,
1389                                        struct page *page, struct zone *zone)
1390{
1391        return true;
1392}
1393#endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1394
1395#endif /* !__GENERATING_BOUNDS.H */
1396#endif /* !__ASSEMBLY__ */
1397#endif /* _LINUX_MMZONE_H */
1398