linux/include/linux/mmzone.h
<<
>>
Prefs
   1#ifndef _LINUX_MMZONE_H
   2#define _LINUX_MMZONE_H
   3
   4#ifndef __ASSEMBLY__
   5#ifndef __GENERATING_BOUNDS_H
   6
   7#include <linux/spinlock.h>
   8#include <linux/list.h>
   9#include <linux/wait.h>
  10#include <linux/bitops.h>
  11#include <linux/cache.h>
  12#include <linux/threads.h>
  13#include <linux/numa.h>
  14#include <linux/init.h>
  15#include <linux/seqlock.h>
  16#include <linux/nodemask.h>
  17#include <linux/pageblock-flags.h>
  18#include <linux/page-flags-layout.h>
  19#include <linux/atomic.h>
  20#include <asm/page.h>
  21
  22/* Free memory management - zoned buddy allocator.  */
  23#ifndef CONFIG_FORCE_MAX_ZONEORDER
  24#define MAX_ORDER 11
  25#else
  26#define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
  27#endif
  28#define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
  29
  30/*
  31 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
  32 * costly to service.  That is between allocation orders which should
  33 * coalesce naturally under reasonable reclaim pressure and those which
  34 * will not.
  35 */
  36#define PAGE_ALLOC_COSTLY_ORDER 3
  37
  38enum {
  39        MIGRATE_UNMOVABLE,
  40        MIGRATE_RECLAIMABLE,
  41        MIGRATE_MOVABLE,
  42        MIGRATE_PCPTYPES,       /* the number of types on the pcp lists */
  43        MIGRATE_RESERVE = MIGRATE_PCPTYPES,
  44#ifdef CONFIG_CMA
  45        /*
  46         * MIGRATE_CMA migration type is designed to mimic the way
  47         * ZONE_MOVABLE works.  Only movable pages can be allocated
  48         * from MIGRATE_CMA pageblocks and page allocator never
  49         * implicitly change migration type of MIGRATE_CMA pageblock.
  50         *
  51         * The way to use it is to change migratetype of a range of
  52         * pageblocks to MIGRATE_CMA which can be done by
  53         * __free_pageblock_cma() function.  What is important though
  54         * is that a range of pageblocks must be aligned to
  55         * MAX_ORDER_NR_PAGES should biggest page be bigger then
  56         * a single pageblock.
  57         */
  58        MIGRATE_CMA,
  59#endif
  60#ifdef CONFIG_MEMORY_ISOLATION
  61        MIGRATE_ISOLATE,        /* can't allocate from here */
  62#endif
  63        MIGRATE_TYPES
  64};
  65
  66#ifdef CONFIG_CMA
  67#  define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
  68#else
  69#  define is_migrate_cma(migratetype) false
  70#endif
  71
  72#define for_each_migratetype_order(order, type) \
  73        for (order = 0; order < MAX_ORDER; order++) \
  74                for (type = 0; type < MIGRATE_TYPES; type++)
  75
  76extern int page_group_by_mobility_disabled;
  77
  78static inline int get_pageblock_migratetype(struct page *page)
  79{
  80        return get_pageblock_flags_group(page, PB_migrate, PB_migrate_end);
  81}
  82
  83struct free_area {
  84        struct list_head        free_list[MIGRATE_TYPES];
  85        unsigned long           nr_free;
  86};
  87
  88struct pglist_data;
  89
  90/*
  91 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
  92 * So add a wild amount of padding here to ensure that they fall into separate
  93 * cachelines.  There are very few zone structures in the machine, so space
  94 * consumption is not a concern here.
  95 */
  96#if defined(CONFIG_SMP)
  97struct zone_padding {
  98        char x[0];
  99} ____cacheline_internodealigned_in_smp;
 100#define ZONE_PADDING(name)      struct zone_padding name;
 101#else
 102#define ZONE_PADDING(name)
 103#endif
 104
 105enum zone_stat_item {
 106        /* First 128 byte cacheline (assuming 64 bit words) */
 107        NR_FREE_PAGES,
 108        NR_LRU_BASE,
 109        NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
 110        NR_ACTIVE_ANON,         /*  "     "     "   "       "         */
 111        NR_INACTIVE_FILE,       /*  "     "     "   "       "         */
 112        NR_ACTIVE_FILE,         /*  "     "     "   "       "         */
 113        NR_UNEVICTABLE,         /*  "     "     "   "       "         */
 114        NR_MLOCK,               /* mlock()ed pages found and moved off LRU */
 115        NR_ANON_PAGES,  /* Mapped anonymous pages */
 116        NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
 117                           only modified from process context */
 118        NR_FILE_PAGES,
 119        NR_FILE_DIRTY,
 120        NR_WRITEBACK,
 121        NR_SLAB_RECLAIMABLE,
 122        NR_SLAB_UNRECLAIMABLE,
 123        NR_PAGETABLE,           /* used for pagetables */
 124        NR_KERNEL_STACK,
 125        /* Second 128 byte cacheline */
 126        NR_UNSTABLE_NFS,        /* NFS unstable pages */
 127        NR_BOUNCE,
 128        NR_VMSCAN_WRITE,
 129        NR_VMSCAN_IMMEDIATE,    /* Prioritise for reclaim when writeback ends */
 130        NR_WRITEBACK_TEMP,      /* Writeback using temporary buffers */
 131        NR_ISOLATED_ANON,       /* Temporary isolated pages from anon lru */
 132        NR_ISOLATED_FILE,       /* Temporary isolated pages from file lru */
 133        NR_SHMEM,               /* shmem pages (included tmpfs/GEM pages) */
 134        NR_DIRTIED,             /* page dirtyings since bootup */
 135        NR_WRITTEN,             /* page writings since bootup */
 136#ifdef CONFIG_NUMA
 137        NUMA_HIT,               /* allocated in intended node */
 138        NUMA_MISS,              /* allocated in non intended node */
 139        NUMA_FOREIGN,           /* was intended here, hit elsewhere */
 140        NUMA_INTERLEAVE_HIT,    /* interleaver preferred this zone */
 141        NUMA_LOCAL,             /* allocation from local node */
 142        NUMA_OTHER,             /* allocation from other node */
 143#endif
 144        NR_ANON_TRANSPARENT_HUGEPAGES,
 145        NR_FREE_CMA_PAGES,
 146        NR_VM_ZONE_STAT_ITEMS };
 147
 148/*
 149 * We do arithmetic on the LRU lists in various places in the code,
 150 * so it is important to keep the active lists LRU_ACTIVE higher in
 151 * the array than the corresponding inactive lists, and to keep
 152 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
 153 *
 154 * This has to be kept in sync with the statistics in zone_stat_item
 155 * above and the descriptions in vmstat_text in mm/vmstat.c
 156 */
 157#define LRU_BASE 0
 158#define LRU_ACTIVE 1
 159#define LRU_FILE 2
 160
 161enum lru_list {
 162        LRU_INACTIVE_ANON = LRU_BASE,
 163        LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
 164        LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
 165        LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
 166        LRU_UNEVICTABLE,
 167        NR_LRU_LISTS
 168};
 169
 170#define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
 171
 172#define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
 173
 174static inline int is_file_lru(enum lru_list lru)
 175{
 176        return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
 177}
 178
 179static inline int is_active_lru(enum lru_list lru)
 180{
 181        return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
 182}
 183
 184static inline int is_unevictable_lru(enum lru_list lru)
 185{
 186        return (lru == LRU_UNEVICTABLE);
 187}
 188
 189struct zone_reclaim_stat {
 190        /*
 191         * The pageout code in vmscan.c keeps track of how many of the
 192         * mem/swap backed and file backed pages are referenced.
 193         * The higher the rotated/scanned ratio, the more valuable
 194         * that cache is.
 195         *
 196         * The anon LRU stats live in [0], file LRU stats in [1]
 197         */
 198        unsigned long           recent_rotated[2];
 199        unsigned long           recent_scanned[2];
 200};
 201
 202struct lruvec {
 203        struct list_head lists[NR_LRU_LISTS];
 204        struct zone_reclaim_stat reclaim_stat;
 205#ifdef CONFIG_MEMCG
 206        struct zone *zone;
 207#endif
 208};
 209
 210/* Mask used at gathering information at once (see memcontrol.c) */
 211#define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
 212#define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
 213#define LRU_ALL      ((1 << NR_LRU_LISTS) - 1)
 214
 215/* Isolate clean file */
 216#define ISOLATE_CLEAN           ((__force isolate_mode_t)0x1)
 217/* Isolate unmapped file */
 218#define ISOLATE_UNMAPPED        ((__force isolate_mode_t)0x2)
 219/* Isolate for asynchronous migration */
 220#define ISOLATE_ASYNC_MIGRATE   ((__force isolate_mode_t)0x4)
 221/* Isolate unevictable pages */
 222#define ISOLATE_UNEVICTABLE     ((__force isolate_mode_t)0x8)
 223
 224/* LRU Isolation modes. */
 225typedef unsigned __bitwise__ isolate_mode_t;
 226
 227enum zone_watermarks {
 228        WMARK_MIN,
 229        WMARK_LOW,
 230        WMARK_HIGH,
 231        NR_WMARK
 232};
 233
 234#define min_wmark_pages(z) (z->watermark[WMARK_MIN])
 235#define low_wmark_pages(z) (z->watermark[WMARK_LOW])
 236#define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
 237
 238struct per_cpu_pages {
 239        int count;              /* number of pages in the list */
 240        int high;               /* high watermark, emptying needed */
 241        int batch;              /* chunk size for buddy add/remove */
 242
 243        /* Lists of pages, one per migrate type stored on the pcp-lists */
 244        struct list_head lists[MIGRATE_PCPTYPES];
 245};
 246
 247struct per_cpu_pageset {
 248        struct per_cpu_pages pcp;
 249#ifdef CONFIG_NUMA
 250        s8 expire;
 251#endif
 252#ifdef CONFIG_SMP
 253        s8 stat_threshold;
 254        s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
 255#endif
 256};
 257
 258#endif /* !__GENERATING_BOUNDS.H */
 259
 260enum zone_type {
 261#ifdef CONFIG_ZONE_DMA
 262        /*
 263         * ZONE_DMA is used when there are devices that are not able
 264         * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
 265         * carve out the portion of memory that is needed for these devices.
 266         * The range is arch specific.
 267         *
 268         * Some examples
 269         *
 270         * Architecture         Limit
 271         * ---------------------------
 272         * parisc, ia64, sparc  <4G
 273         * s390                 <2G
 274         * arm                  Various
 275         * alpha                Unlimited or 0-16MB.
 276         *
 277         * i386, x86_64 and multiple other arches
 278         *                      <16M.
 279         */
 280        ZONE_DMA,
 281#endif
 282#ifdef CONFIG_ZONE_DMA32
 283        /*
 284         * x86_64 needs two ZONE_DMAs because it supports devices that are
 285         * only able to do DMA to the lower 16M but also 32 bit devices that
 286         * can only do DMA areas below 4G.
 287         */
 288        ZONE_DMA32,
 289#endif
 290        /*
 291         * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
 292         * performed on pages in ZONE_NORMAL if the DMA devices support
 293         * transfers to all addressable memory.
 294         */
 295        ZONE_NORMAL,
 296#ifdef CONFIG_HIGHMEM
 297        /*
 298         * A memory area that is only addressable by the kernel through
 299         * mapping portions into its own address space. This is for example
 300         * used by i386 to allow the kernel to address the memory beyond
 301         * 900MB. The kernel will set up special mappings (page
 302         * table entries on i386) for each page that the kernel needs to
 303         * access.
 304         */
 305        ZONE_HIGHMEM,
 306#endif
 307        ZONE_MOVABLE,
 308        __MAX_NR_ZONES
 309};
 310
 311#ifndef __GENERATING_BOUNDS_H
 312
 313struct zone {
 314        /* Fields commonly accessed by the page allocator */
 315
 316        /* zone watermarks, access with *_wmark_pages(zone) macros */
 317        unsigned long watermark[NR_WMARK];
 318
 319        /*
 320         * When free pages are below this point, additional steps are taken
 321         * when reading the number of free pages to avoid per-cpu counter
 322         * drift allowing watermarks to be breached
 323         */
 324        unsigned long percpu_drift_mark;
 325
 326        /*
 327         * We don't know if the memory that we're going to allocate will be freeable
 328         * or/and it will be released eventually, so to avoid totally wasting several
 329         * GB of ram we must reserve some of the lower zone memory (otherwise we risk
 330         * to run OOM on the lower zones despite there's tons of freeable ram
 331         * on the higher zones). This array is recalculated at runtime if the
 332         * sysctl_lowmem_reserve_ratio sysctl changes.
 333         */
 334        unsigned long           lowmem_reserve[MAX_NR_ZONES];
 335
 336        /*
 337         * This is a per-zone reserve of pages that should not be
 338         * considered dirtyable memory.
 339         */
 340        unsigned long           dirty_balance_reserve;
 341
 342#ifdef CONFIG_NUMA
 343        int node;
 344        /*
 345         * zone reclaim becomes active if more unmapped pages exist.
 346         */
 347        unsigned long           min_unmapped_pages;
 348        unsigned long           min_slab_pages;
 349#endif
 350        struct per_cpu_pageset __percpu *pageset;
 351        /*
 352         * free areas of different sizes
 353         */
 354        spinlock_t              lock;
 355        int                     all_unreclaimable; /* All pages pinned */
 356#if defined CONFIG_COMPACTION || defined CONFIG_CMA
 357        /* Set to true when the PG_migrate_skip bits should be cleared */
 358        bool                    compact_blockskip_flush;
 359
 360        /* pfns where compaction scanners should start */
 361        unsigned long           compact_cached_free_pfn;
 362        unsigned long           compact_cached_migrate_pfn;
 363#endif
 364#ifdef CONFIG_MEMORY_HOTPLUG
 365        /* see spanned/present_pages for more description */
 366        seqlock_t               span_seqlock;
 367#endif
 368        struct free_area        free_area[MAX_ORDER];
 369
 370#ifndef CONFIG_SPARSEMEM
 371        /*
 372         * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
 373         * In SPARSEMEM, this map is stored in struct mem_section
 374         */
 375        unsigned long           *pageblock_flags;
 376#endif /* CONFIG_SPARSEMEM */
 377
 378#ifdef CONFIG_COMPACTION
 379        /*
 380         * On compaction failure, 1<<compact_defer_shift compactions
 381         * are skipped before trying again. The number attempted since
 382         * last failure is tracked with compact_considered.
 383         */
 384        unsigned int            compact_considered;
 385        unsigned int            compact_defer_shift;
 386        int                     compact_order_failed;
 387#endif
 388
 389        ZONE_PADDING(_pad1_)
 390
 391        /* Fields commonly accessed by the page reclaim scanner */
 392        spinlock_t              lru_lock;
 393        struct lruvec           lruvec;
 394
 395        unsigned long           pages_scanned;     /* since last reclaim */
 396        unsigned long           flags;             /* zone flags, see below */
 397
 398        /* Zone statistics */
 399        atomic_long_t           vm_stat[NR_VM_ZONE_STAT_ITEMS];
 400
 401        /*
 402         * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on
 403         * this zone's LRU.  Maintained by the pageout code.
 404         */
 405        unsigned int inactive_ratio;
 406
 407
 408        ZONE_PADDING(_pad2_)
 409        /* Rarely used or read-mostly fields */
 410
 411        /*
 412         * wait_table           -- the array holding the hash table
 413         * wait_table_hash_nr_entries   -- the size of the hash table array
 414         * wait_table_bits      -- wait_table_size == (1 << wait_table_bits)
 415         *
 416         * The purpose of all these is to keep track of the people
 417         * waiting for a page to become available and make them
 418         * runnable again when possible. The trouble is that this
 419         * consumes a lot of space, especially when so few things
 420         * wait on pages at a given time. So instead of using
 421         * per-page waitqueues, we use a waitqueue hash table.
 422         *
 423         * The bucket discipline is to sleep on the same queue when
 424         * colliding and wake all in that wait queue when removing.
 425         * When something wakes, it must check to be sure its page is
 426         * truly available, a la thundering herd. The cost of a
 427         * collision is great, but given the expected load of the
 428         * table, they should be so rare as to be outweighed by the
 429         * benefits from the saved space.
 430         *
 431         * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
 432         * primary users of these fields, and in mm/page_alloc.c
 433         * free_area_init_core() performs the initialization of them.
 434         */
 435        wait_queue_head_t       * wait_table;
 436        unsigned long           wait_table_hash_nr_entries;
 437        unsigned long           wait_table_bits;
 438
 439        /*
 440         * Discontig memory support fields.
 441         */
 442        struct pglist_data      *zone_pgdat;
 443        /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
 444        unsigned long           zone_start_pfn;
 445
 446        /*
 447         * spanned_pages is the total pages spanned by the zone, including
 448         * holes, which is calculated as:
 449         *      spanned_pages = zone_end_pfn - zone_start_pfn;
 450         *
 451         * present_pages is physical pages existing within the zone, which
 452         * is calculated as:
 453         *      present_pages = spanned_pages - absent_pages(pags in holes);
 454         *
 455         * managed_pages is present pages managed by the buddy system, which
 456         * is calculated as (reserved_pages includes pages allocated by the
 457         * bootmem allocator):
 458         *      managed_pages = present_pages - reserved_pages;
 459         *
 460         * So present_pages may be used by memory hotplug or memory power
 461         * management logic to figure out unmanaged pages by checking
 462         * (present_pages - managed_pages). And managed_pages should be used
 463         * by page allocator and vm scanner to calculate all kinds of watermarks
 464         * and thresholds.
 465         *
 466         * Locking rules:
 467         *
 468         * zone_start_pfn and spanned_pages are protected by span_seqlock.
 469         * It is a seqlock because it has to be read outside of zone->lock,
 470         * and it is done in the main allocator path.  But, it is written
 471         * quite infrequently.
 472         *
 473         * The span_seq lock is declared along with zone->lock because it is
 474         * frequently read in proximity to zone->lock.  It's good to
 475         * give them a chance of being in the same cacheline.
 476         *
 477         * Write access to present_pages and managed_pages at runtime should
 478         * be protected by lock_memory_hotplug()/unlock_memory_hotplug().
 479         * Any reader who can't tolerant drift of present_pages and
 480         * managed_pages should hold memory hotplug lock to get a stable value.
 481         */
 482        unsigned long           spanned_pages;
 483        unsigned long           present_pages;
 484        unsigned long           managed_pages;
 485
 486        /*
 487         * rarely used fields:
 488         */
 489        const char              *name;
 490} ____cacheline_internodealigned_in_smp;
 491
 492typedef enum {
 493        ZONE_RECLAIM_LOCKED,            /* prevents concurrent reclaim */
 494        ZONE_OOM_LOCKED,                /* zone is in OOM killer zonelist */
 495        ZONE_CONGESTED,                 /* zone has many dirty pages backed by
 496                                         * a congested BDI
 497                                         */
 498} zone_flags_t;
 499
 500static inline void zone_set_flag(struct zone *zone, zone_flags_t flag)
 501{
 502        set_bit(flag, &zone->flags);
 503}
 504
 505static inline int zone_test_and_set_flag(struct zone *zone, zone_flags_t flag)
 506{
 507        return test_and_set_bit(flag, &zone->flags);
 508}
 509
 510static inline void zone_clear_flag(struct zone *zone, zone_flags_t flag)
 511{
 512        clear_bit(flag, &zone->flags);
 513}
 514
 515static inline int zone_is_reclaim_congested(const struct zone *zone)
 516{
 517        return test_bit(ZONE_CONGESTED, &zone->flags);
 518}
 519
 520static inline int zone_is_reclaim_locked(const struct zone *zone)
 521{
 522        return test_bit(ZONE_RECLAIM_LOCKED, &zone->flags);
 523}
 524
 525static inline int zone_is_oom_locked(const struct zone *zone)
 526{
 527        return test_bit(ZONE_OOM_LOCKED, &zone->flags);
 528}
 529
 530static inline unsigned long zone_end_pfn(const struct zone *zone)
 531{
 532        return zone->zone_start_pfn + zone->spanned_pages;
 533}
 534
 535static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
 536{
 537        return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
 538}
 539
 540static inline bool zone_is_initialized(struct zone *zone)
 541{
 542        return !!zone->wait_table;
 543}
 544
 545static inline bool zone_is_empty(struct zone *zone)
 546{
 547        return zone->spanned_pages == 0;
 548}
 549
 550/*
 551 * The "priority" of VM scanning is how much of the queues we will scan in one
 552 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
 553 * queues ("queue_length >> 12") during an aging round.
 554 */
 555#define DEF_PRIORITY 12
 556
 557/* Maximum number of zones on a zonelist */
 558#define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
 559
 560#ifdef CONFIG_NUMA
 561
 562/*
 563 * The NUMA zonelists are doubled because we need zonelists that restrict the
 564 * allocations to a single node for GFP_THISNODE.
 565 *
 566 * [0]  : Zonelist with fallback
 567 * [1]  : No fallback (GFP_THISNODE)
 568 */
 569#define MAX_ZONELISTS 2
 570
 571
 572/*
 573 * We cache key information from each zonelist for smaller cache
 574 * footprint when scanning for free pages in get_page_from_freelist().
 575 *
 576 * 1) The BITMAP fullzones tracks which zones in a zonelist have come
 577 *    up short of free memory since the last time (last_fullzone_zap)
 578 *    we zero'd fullzones.
 579 * 2) The array z_to_n[] maps each zone in the zonelist to its node
 580 *    id, so that we can efficiently evaluate whether that node is
 581 *    set in the current tasks mems_allowed.
 582 *
 583 * Both fullzones and z_to_n[] are one-to-one with the zonelist,
 584 * indexed by a zones offset in the zonelist zones[] array.
 585 *
 586 * The get_page_from_freelist() routine does two scans.  During the
 587 * first scan, we skip zones whose corresponding bit in 'fullzones'
 588 * is set or whose corresponding node in current->mems_allowed (which
 589 * comes from cpusets) is not set.  During the second scan, we bypass
 590 * this zonelist_cache, to ensure we look methodically at each zone.
 591 *
 592 * Once per second, we zero out (zap) fullzones, forcing us to
 593 * reconsider nodes that might have regained more free memory.
 594 * The field last_full_zap is the time we last zapped fullzones.
 595 *
 596 * This mechanism reduces the amount of time we waste repeatedly
 597 * reexaming zones for free memory when they just came up low on
 598 * memory momentarilly ago.
 599 *
 600 * The zonelist_cache struct members logically belong in struct
 601 * zonelist.  However, the mempolicy zonelists constructed for
 602 * MPOL_BIND are intentionally variable length (and usually much
 603 * shorter).  A general purpose mechanism for handling structs with
 604 * multiple variable length members is more mechanism than we want
 605 * here.  We resort to some special case hackery instead.
 606 *
 607 * The MPOL_BIND zonelists don't need this zonelist_cache (in good
 608 * part because they are shorter), so we put the fixed length stuff
 609 * at the front of the zonelist struct, ending in a variable length
 610 * zones[], as is needed by MPOL_BIND.
 611 *
 612 * Then we put the optional zonelist cache on the end of the zonelist
 613 * struct.  This optional stuff is found by a 'zlcache_ptr' pointer in
 614 * the fixed length portion at the front of the struct.  This pointer
 615 * both enables us to find the zonelist cache, and in the case of
 616 * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL)
 617 * to know that the zonelist cache is not there.
 618 *
 619 * The end result is that struct zonelists come in two flavors:
 620 *  1) The full, fixed length version, shown below, and
 621 *  2) The custom zonelists for MPOL_BIND.
 622 * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache.
 623 *
 624 * Even though there may be multiple CPU cores on a node modifying
 625 * fullzones or last_full_zap in the same zonelist_cache at the same
 626 * time, we don't lock it.  This is just hint data - if it is wrong now
 627 * and then, the allocator will still function, perhaps a bit slower.
 628 */
 629
 630
 631struct zonelist_cache {
 632        unsigned short z_to_n[MAX_ZONES_PER_ZONELIST];          /* zone->nid */
 633        DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST);      /* zone full? */
 634        unsigned long last_full_zap;            /* when last zap'd (jiffies) */
 635};
 636#else
 637#define MAX_ZONELISTS 1
 638struct zonelist_cache;
 639#endif
 640
 641/*
 642 * This struct contains information about a zone in a zonelist. It is stored
 643 * here to avoid dereferences into large structures and lookups of tables
 644 */
 645struct zoneref {
 646        struct zone *zone;      /* Pointer to actual zone */
 647        int zone_idx;           /* zone_idx(zoneref->zone) */
 648};
 649
 650/*
 651 * One allocation request operates on a zonelist. A zonelist
 652 * is a list of zones, the first one is the 'goal' of the
 653 * allocation, the other zones are fallback zones, in decreasing
 654 * priority.
 655 *
 656 * If zlcache_ptr is not NULL, then it is just the address of zlcache,
 657 * as explained above.  If zlcache_ptr is NULL, there is no zlcache.
 658 * *
 659 * To speed the reading of the zonelist, the zonerefs contain the zone index
 660 * of the entry being read. Helper functions to access information given
 661 * a struct zoneref are
 662 *
 663 * zonelist_zone()      - Return the struct zone * for an entry in _zonerefs
 664 * zonelist_zone_idx()  - Return the index of the zone for an entry
 665 * zonelist_node_idx()  - Return the index of the node for an entry
 666 */
 667struct zonelist {
 668        struct zonelist_cache *zlcache_ptr;                  // NULL or &zlcache
 669        struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
 670#ifdef CONFIG_NUMA
 671        struct zonelist_cache zlcache;                       // optional ...
 672#endif
 673};
 674
 675#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
 676struct node_active_region {
 677        unsigned long start_pfn;
 678        unsigned long end_pfn;
 679        int nid;
 680};
 681#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
 682
 683#ifndef CONFIG_DISCONTIGMEM
 684/* The array of struct pages - for discontigmem use pgdat->lmem_map */
 685extern struct page *mem_map;
 686#endif
 687
 688/*
 689 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
 690 * (mostly NUMA machines?) to denote a higher-level memory zone than the
 691 * zone denotes.
 692 *
 693 * On NUMA machines, each NUMA node would have a pg_data_t to describe
 694 * it's memory layout.
 695 *
 696 * Memory statistics and page replacement data structures are maintained on a
 697 * per-zone basis.
 698 */
 699struct bootmem_data;
 700typedef struct pglist_data {
 701        struct zone node_zones[MAX_NR_ZONES];
 702        struct zonelist node_zonelists[MAX_ZONELISTS];
 703        int nr_zones;
 704#ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
 705        struct page *node_mem_map;
 706#ifdef CONFIG_MEMCG
 707        struct page_cgroup *node_page_cgroup;
 708#endif
 709#endif
 710#ifndef CONFIG_NO_BOOTMEM
 711        struct bootmem_data *bdata;
 712#endif
 713#ifdef CONFIG_MEMORY_HOTPLUG
 714        /*
 715         * Must be held any time you expect node_start_pfn, node_present_pages
 716         * or node_spanned_pages stay constant.  Holding this will also
 717         * guarantee that any pfn_valid() stays that way.
 718         *
 719         * Nests above zone->lock and zone->size_seqlock.
 720         */
 721        spinlock_t node_size_lock;
 722#endif
 723        unsigned long node_start_pfn;
 724        unsigned long node_present_pages; /* total number of physical pages */
 725        unsigned long node_spanned_pages; /* total size of physical page
 726                                             range, including holes */
 727        int node_id;
 728        nodemask_t reclaim_nodes;       /* Nodes allowed to reclaim from */
 729        wait_queue_head_t kswapd_wait;
 730        wait_queue_head_t pfmemalloc_wait;
 731        struct task_struct *kswapd;     /* Protected by lock_memory_hotplug() */
 732        int kswapd_max_order;
 733        enum zone_type classzone_idx;
 734#ifdef CONFIG_NUMA_BALANCING
 735        /*
 736         * Lock serializing the per destination node AutoNUMA memory
 737         * migration rate limiting data.
 738         */
 739        spinlock_t numabalancing_migrate_lock;
 740
 741        /* Rate limiting time interval */
 742        unsigned long numabalancing_migrate_next_window;
 743
 744        /* Number of pages migrated during the rate limiting time interval */
 745        unsigned long numabalancing_migrate_nr_pages;
 746#endif
 747} pg_data_t;
 748
 749#define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
 750#define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
 751#ifdef CONFIG_FLAT_NODE_MEM_MAP
 752#define pgdat_page_nr(pgdat, pagenr)    ((pgdat)->node_mem_map + (pagenr))
 753#else
 754#define pgdat_page_nr(pgdat, pagenr)    pfn_to_page((pgdat)->node_start_pfn + (pagenr))
 755#endif
 756#define nid_page_nr(nid, pagenr)        pgdat_page_nr(NODE_DATA(nid),(pagenr))
 757
 758#define node_start_pfn(nid)     (NODE_DATA(nid)->node_start_pfn)
 759#define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
 760
 761static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
 762{
 763        return pgdat->node_start_pfn + pgdat->node_spanned_pages;
 764}
 765
 766static inline bool pgdat_is_empty(pg_data_t *pgdat)
 767{
 768        return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
 769}
 770
 771#include <linux/memory_hotplug.h>
 772
 773extern struct mutex zonelists_mutex;
 774void build_all_zonelists(pg_data_t *pgdat, struct zone *zone);
 775void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx);
 776bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
 777                int classzone_idx, int alloc_flags);
 778bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
 779                int classzone_idx, int alloc_flags);
 780enum memmap_context {
 781        MEMMAP_EARLY,
 782        MEMMAP_HOTPLUG,
 783};
 784extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
 785                                     unsigned long size,
 786                                     enum memmap_context context);
 787
 788extern void lruvec_init(struct lruvec *lruvec);
 789
 790static inline struct zone *lruvec_zone(struct lruvec *lruvec)
 791{
 792#ifdef CONFIG_MEMCG
 793        return lruvec->zone;
 794#else
 795        return container_of(lruvec, struct zone, lruvec);
 796#endif
 797}
 798
 799#ifdef CONFIG_HAVE_MEMORY_PRESENT
 800void memory_present(int nid, unsigned long start, unsigned long end);
 801#else
 802static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
 803#endif
 804
 805#ifdef CONFIG_HAVE_MEMORYLESS_NODES
 806int local_memory_node(int node_id);
 807#else
 808static inline int local_memory_node(int node_id) { return node_id; };
 809#endif
 810
 811#ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
 812unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
 813#endif
 814
 815/*
 816 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
 817 */
 818#define zone_idx(zone)          ((zone) - (zone)->zone_pgdat->node_zones)
 819
 820static inline int populated_zone(struct zone *zone)
 821{
 822        return (!!zone->present_pages);
 823}
 824
 825extern int movable_zone;
 826
 827static inline int zone_movable_is_highmem(void)
 828{
 829#if defined(CONFIG_HIGHMEM) && defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
 830        return movable_zone == ZONE_HIGHMEM;
 831#else
 832        return 0;
 833#endif
 834}
 835
 836static inline int is_highmem_idx(enum zone_type idx)
 837{
 838#ifdef CONFIG_HIGHMEM
 839        return (idx == ZONE_HIGHMEM ||
 840                (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
 841#else
 842        return 0;
 843#endif
 844}
 845
 846static inline int is_normal_idx(enum zone_type idx)
 847{
 848        return (idx == ZONE_NORMAL);
 849}
 850
 851/**
 852 * is_highmem - helper function to quickly check if a struct zone is a 
 853 *              highmem zone or not.  This is an attempt to keep references
 854 *              to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
 855 * @zone - pointer to struct zone variable
 856 */
 857static inline int is_highmem(struct zone *zone)
 858{
 859#ifdef CONFIG_HIGHMEM
 860        int zone_off = (char *)zone - (char *)zone->zone_pgdat->node_zones;
 861        return zone_off == ZONE_HIGHMEM * sizeof(*zone) ||
 862               (zone_off == ZONE_MOVABLE * sizeof(*zone) &&
 863                zone_movable_is_highmem());
 864#else
 865        return 0;
 866#endif
 867}
 868
 869static inline int is_normal(struct zone *zone)
 870{
 871        return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
 872}
 873
 874static inline int is_dma32(struct zone *zone)
 875{
 876#ifdef CONFIG_ZONE_DMA32
 877        return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
 878#else
 879        return 0;
 880#endif
 881}
 882
 883static inline int is_dma(struct zone *zone)
 884{
 885#ifdef CONFIG_ZONE_DMA
 886        return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
 887#else
 888        return 0;
 889#endif
 890}
 891
 892/* These two functions are used to setup the per zone pages min values */
 893struct ctl_table;
 894int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
 895                                        void __user *, size_t *, loff_t *);
 896extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
 897int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
 898                                        void __user *, size_t *, loff_t *);
 899int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
 900                                        void __user *, size_t *, loff_t *);
 901int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
 902                        void __user *, size_t *, loff_t *);
 903int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
 904                        void __user *, size_t *, loff_t *);
 905
 906extern int numa_zonelist_order_handler(struct ctl_table *, int,
 907                        void __user *, size_t *, loff_t *);
 908extern char numa_zonelist_order[];
 909#define NUMA_ZONELIST_ORDER_LEN 16      /* string buffer size */
 910
 911#ifndef CONFIG_NEED_MULTIPLE_NODES
 912
 913extern struct pglist_data contig_page_data;
 914#define NODE_DATA(nid)          (&contig_page_data)
 915#define NODE_MEM_MAP(nid)       mem_map
 916
 917#else /* CONFIG_NEED_MULTIPLE_NODES */
 918
 919#include <asm/mmzone.h>
 920
 921#endif /* !CONFIG_NEED_MULTIPLE_NODES */
 922
 923extern struct pglist_data *first_online_pgdat(void);
 924extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
 925extern struct zone *next_zone(struct zone *zone);
 926
 927/**
 928 * for_each_online_pgdat - helper macro to iterate over all online nodes
 929 * @pgdat - pointer to a pg_data_t variable
 930 */
 931#define for_each_online_pgdat(pgdat)                    \
 932        for (pgdat = first_online_pgdat();              \
 933             pgdat;                                     \
 934             pgdat = next_online_pgdat(pgdat))
 935/**
 936 * for_each_zone - helper macro to iterate over all memory zones
 937 * @zone - pointer to struct zone variable
 938 *
 939 * The user only needs to declare the zone variable, for_each_zone
 940 * fills it in.
 941 */
 942#define for_each_zone(zone)                             \
 943        for (zone = (first_online_pgdat())->node_zones; \
 944             zone;                                      \
 945             zone = next_zone(zone))
 946
 947#define for_each_populated_zone(zone)                   \
 948        for (zone = (first_online_pgdat())->node_zones; \
 949             zone;                                      \
 950             zone = next_zone(zone))                    \
 951                if (!populated_zone(zone))              \
 952                        ; /* do nothing */              \
 953                else
 954
 955static inline struct zone *zonelist_zone(struct zoneref *zoneref)
 956{
 957        return zoneref->zone;
 958}
 959
 960static inline int zonelist_zone_idx(struct zoneref *zoneref)
 961{
 962        return zoneref->zone_idx;
 963}
 964
 965static inline int zonelist_node_idx(struct zoneref *zoneref)
 966{
 967#ifdef CONFIG_NUMA
 968        /* zone_to_nid not available in this context */
 969        return zoneref->zone->node;
 970#else
 971        return 0;
 972#endif /* CONFIG_NUMA */
 973}
 974
 975/**
 976 * 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
 977 * @z - The cursor used as a starting point for the search
 978 * @highest_zoneidx - The zone index of the highest zone to return
 979 * @nodes - An optional nodemask to filter the zonelist with
 980 * @zone - The first suitable zone found is returned via this parameter
 981 *
 982 * This function returns the next zone at or below a given zone index that is
 983 * within the allowed nodemask using a cursor as the starting point for the
 984 * search. The zoneref returned is a cursor that represents the current zone
 985 * being examined. It should be advanced by one before calling
 986 * next_zones_zonelist again.
 987 */
 988struct zoneref *next_zones_zonelist(struct zoneref *z,
 989                                        enum zone_type highest_zoneidx,
 990                                        nodemask_t *nodes,
 991                                        struct zone **zone);
 992
 993/**
 994 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
 995 * @zonelist - The zonelist to search for a suitable zone
 996 * @highest_zoneidx - The zone index of the highest zone to return
 997 * @nodes - An optional nodemask to filter the zonelist with
 998 * @zone - The first suitable zone found is returned via this parameter
 999 *
1000 * This function returns the first zone at or below a given zone index that is
1001 * within the allowed nodemask. The zoneref returned is a cursor that can be
1002 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1003 * one before calling.
1004 */
1005static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1006                                        enum zone_type highest_zoneidx,
1007                                        nodemask_t *nodes,
1008                                        struct zone **zone)
1009{
1010        return next_zones_zonelist(zonelist->_zonerefs, highest_zoneidx, nodes,
1011                                                                zone);
1012}
1013
1014/**
1015 * 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
1016 * @zone - The current zone in the iterator
1017 * @z - The current pointer within zonelist->zones being iterated
1018 * @zlist - The zonelist being iterated
1019 * @highidx - The zone index of the highest zone to return
1020 * @nodemask - Nodemask allowed by the allocator
1021 *
1022 * This iterator iterates though all zones at or below a given zone index and
1023 * within a given nodemask
1024 */
1025#define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1026        for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \
1027                zone;                                                   \
1028                z = next_zones_zonelist(++z, highidx, nodemask, &zone)) \
1029
1030/**
1031 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1032 * @zone - The current zone in the iterator
1033 * @z - The current pointer within zonelist->zones being iterated
1034 * @zlist - The zonelist being iterated
1035 * @highidx - The zone index of the highest zone to return
1036 *
1037 * This iterator iterates though all zones at or below a given zone index.
1038 */
1039#define for_each_zone_zonelist(zone, z, zlist, highidx) \
1040        for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1041
1042#ifdef CONFIG_SPARSEMEM
1043#include <asm/sparsemem.h>
1044#endif
1045
1046#if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1047        !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1048static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1049{
1050        return 0;
1051}
1052#endif
1053
1054#ifdef CONFIG_FLATMEM
1055#define pfn_to_nid(pfn)         (0)
1056#endif
1057
1058#ifdef CONFIG_SPARSEMEM
1059
1060/*
1061 * SECTION_SHIFT                #bits space required to store a section #
1062 *
1063 * PA_SECTION_SHIFT             physical address to/from section number
1064 * PFN_SECTION_SHIFT            pfn to/from section number
1065 */
1066#define PA_SECTION_SHIFT        (SECTION_SIZE_BITS)
1067#define PFN_SECTION_SHIFT       (SECTION_SIZE_BITS - PAGE_SHIFT)
1068
1069#define NR_MEM_SECTIONS         (1UL << SECTIONS_SHIFT)
1070
1071#define PAGES_PER_SECTION       (1UL << PFN_SECTION_SHIFT)
1072#define PAGE_SECTION_MASK       (~(PAGES_PER_SECTION-1))
1073
1074#define SECTION_BLOCKFLAGS_BITS \
1075        ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1076
1077#if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1078#error Allocator MAX_ORDER exceeds SECTION_SIZE
1079#endif
1080
1081#define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
1082#define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
1083
1084#define SECTION_ALIGN_UP(pfn)   (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1085#define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1086
1087struct page;
1088struct page_cgroup;
1089struct mem_section {
1090        /*
1091         * This is, logically, a pointer to an array of struct
1092         * pages.  However, it is stored with some other magic.
1093         * (see sparse.c::sparse_init_one_section())
1094         *
1095         * Additionally during early boot we encode node id of
1096         * the location of the section here to guide allocation.
1097         * (see sparse.c::memory_present())
1098         *
1099         * Making it a UL at least makes someone do a cast
1100         * before using it wrong.
1101         */
1102        unsigned long section_mem_map;
1103
1104        /* See declaration of similar field in struct zone */
1105        unsigned long *pageblock_flags;
1106#ifdef CONFIG_MEMCG
1107        /*
1108         * If !SPARSEMEM, pgdat doesn't have page_cgroup pointer. We use
1109         * section. (see memcontrol.h/page_cgroup.h about this.)
1110         */
1111        struct page_cgroup *page_cgroup;
1112        unsigned long pad;
1113#endif
1114};
1115
1116#ifdef CONFIG_SPARSEMEM_EXTREME
1117#define SECTIONS_PER_ROOT       (PAGE_SIZE / sizeof (struct mem_section))
1118#else
1119#define SECTIONS_PER_ROOT       1
1120#endif
1121
1122#define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1123#define NR_SECTION_ROOTS        DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1124#define SECTION_ROOT_MASK       (SECTIONS_PER_ROOT - 1)
1125
1126#ifdef CONFIG_SPARSEMEM_EXTREME
1127extern struct mem_section *mem_section[NR_SECTION_ROOTS];
1128#else
1129extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1130#endif
1131
1132static inline struct mem_section *__nr_to_section(unsigned long nr)
1133{
1134        if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1135                return NULL;
1136        return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1137}
1138extern int __section_nr(struct mem_section* ms);
1139extern unsigned long usemap_size(void);
1140
1141/*
1142 * We use the lower bits of the mem_map pointer to store
1143 * a little bit of information.  There should be at least
1144 * 3 bits here due to 32-bit alignment.
1145 */
1146#define SECTION_MARKED_PRESENT  (1UL<<0)
1147#define SECTION_HAS_MEM_MAP     (1UL<<1)
1148#define SECTION_MAP_LAST_BIT    (1UL<<2)
1149#define SECTION_MAP_MASK        (~(SECTION_MAP_LAST_BIT-1))
1150#define SECTION_NID_SHIFT       2
1151
1152static inline struct page *__section_mem_map_addr(struct mem_section *section)
1153{
1154        unsigned long map = section->section_mem_map;
1155        map &= SECTION_MAP_MASK;
1156        return (struct page *)map;
1157}
1158
1159static inline int present_section(struct mem_section *section)
1160{
1161        return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1162}
1163
1164static inline int present_section_nr(unsigned long nr)
1165{
1166        return present_section(__nr_to_section(nr));
1167}
1168
1169static inline int valid_section(struct mem_section *section)
1170{
1171        return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1172}
1173
1174static inline int valid_section_nr(unsigned long nr)
1175{
1176        return valid_section(__nr_to_section(nr));
1177}
1178
1179static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1180{
1181        return __nr_to_section(pfn_to_section_nr(pfn));
1182}
1183
1184#ifndef CONFIG_HAVE_ARCH_PFN_VALID
1185static inline int pfn_valid(unsigned long pfn)
1186{
1187        if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1188                return 0;
1189        return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
1190}
1191#endif
1192
1193static inline int pfn_present(unsigned long pfn)
1194{
1195        if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1196                return 0;
1197        return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1198}
1199
1200/*
1201 * These are _only_ used during initialisation, therefore they
1202 * can use __initdata ...  They could have names to indicate
1203 * this restriction.
1204 */
1205#ifdef CONFIG_NUMA
1206#define pfn_to_nid(pfn)                                                 \
1207({                                                                      \
1208        unsigned long __pfn_to_nid_pfn = (pfn);                         \
1209        page_to_nid(pfn_to_page(__pfn_to_nid_pfn));                     \
1210})
1211#else
1212#define pfn_to_nid(pfn)         (0)
1213#endif
1214
1215#define early_pfn_valid(pfn)    pfn_valid(pfn)
1216void sparse_init(void);
1217#else
1218#define sparse_init()   do {} while (0)
1219#define sparse_index_init(_sec, _nid)  do {} while (0)
1220#endif /* CONFIG_SPARSEMEM */
1221
1222#ifdef CONFIG_NODES_SPAN_OTHER_NODES
1223bool early_pfn_in_nid(unsigned long pfn, int nid);
1224#else
1225#define early_pfn_in_nid(pfn, nid)      (1)
1226#endif
1227
1228#ifndef early_pfn_valid
1229#define early_pfn_valid(pfn)    (1)
1230#endif
1231
1232void memory_present(int nid, unsigned long start, unsigned long end);
1233unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
1234
1235/*
1236 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1237 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1238 * pfn_valid_within() should be used in this case; we optimise this away
1239 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1240 */
1241#ifdef CONFIG_HOLES_IN_ZONE
1242#define pfn_valid_within(pfn) pfn_valid(pfn)
1243#else
1244#define pfn_valid_within(pfn) (1)
1245#endif
1246
1247#ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1248/*
1249 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1250 * associated with it or not. In FLATMEM, it is expected that holes always
1251 * have valid memmap as long as there is valid PFNs either side of the hole.
1252 * In SPARSEMEM, it is assumed that a valid section has a memmap for the
1253 * entire section.
1254 *
1255 * However, an ARM, and maybe other embedded architectures in the future
1256 * free memmap backing holes to save memory on the assumption the memmap is
1257 * never used. The page_zone linkages are then broken even though pfn_valid()
1258 * returns true. A walker of the full memmap must then do this additional
1259 * check to ensure the memmap they are looking at is sane by making sure
1260 * the zone and PFN linkages are still valid. This is expensive, but walkers
1261 * of the full memmap are extremely rare.
1262 */
1263int memmap_valid_within(unsigned long pfn,
1264                                        struct page *page, struct zone *zone);
1265#else
1266static inline int memmap_valid_within(unsigned long pfn,
1267                                        struct page *page, struct zone *zone)
1268{
1269        return 1;
1270}
1271#endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1272
1273#endif /* !__GENERATING_BOUNDS.H */
1274#endif /* !__ASSEMBLY__ */
1275#endif /* _LINUX_MMZONE_H */
1276
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.