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