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