include/linux/mmzone.h

   1 #ifndef _LINUX_MMZONE_H
   2 #define _LINUX_MMZONE_H
   3
   4 #ifdef __KERNEL__
   5 #ifndef __ASSEMBLY__
   6
   7 #include <linux/spinlock.h>
   8 #include <linux/list.h>
   9 #include <linux/wait.h>
  10 #include <linux/cache.h>
  11 #include <linux/threads.h>
  12 #include <linux/numa.h>
  13 #include <linux/init.h>
  14 #include <linux/seqlock.h>
  15 #include <linux/nodemask.h>
  16 #include <asm/atomic.h>
  17 #include <asm/page.h>
  18
  19 /* Free memory management - zoned buddy allocator.  */
  20 #ifndef CONFIG_FORCE_MAX_ZONEORDER
  21 #define MAX_ORDER 11
  22 #else
  23 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
  24 #endif
  25 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
  26
  27 struct free_area {
  28         struct list_head        free_list;
  29         unsigned long           nr_free;
  30 };
  31
  32 struct pglist_data;
  33
  34 /*
  35  * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
  36  * So add a wild amount of padding here to ensure that they fall into separate
  37  * cachelines.  There are very few zone structures in the machine, so space
  38  * consumption is not a concern here.
  39  */
  40 #if defined(CONFIG_SMP)
  41 struct zone_padding {
  42         char x[0];
  43 } ____cacheline_internodealigned_in_smp;
  44 #define ZONE_PADDING(name)      struct zone_padding name;
  45 #else
  46 #define ZONE_PADDING(name)
  47 #endif
  48
  49 enum zone_stat_item {
  50         NR_ANON_PAGES,  /* Mapped anonymous pages */
  51         NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
  52                            only modified from process context */
  53         NR_FILE_PAGES,
  54         NR_SLAB_RECLAIMABLE,
  55         NR_SLAB_UNRECLAIMABLE,
  56         NR_PAGETABLE,   /* used for pagetables */
  57         NR_FILE_DIRTY,
  58         NR_WRITEBACK,
  59         NR_UNSTABLE_NFS,        /* NFS unstable pages */
  60         NR_BOUNCE,
  61         NR_VMSCAN_WRITE,
  62 #ifdef CONFIG_NUMA
  63         NUMA_HIT,               /* allocated in intended node */
  64         NUMA_MISS,              /* allocated in non intended node */
  65         NUMA_FOREIGN,           /* was intended here, hit elsewhere */
  66         NUMA_INTERLEAVE_HIT,    /* interleaver preferred this zone */
  67         NUMA_LOCAL,             /* allocation from local node */
  68         NUMA_OTHER,             /* allocation from other node */
  69 #endif
  70         NR_VM_ZONE_STAT_ITEMS };
  71
  72 struct per_cpu_pages {
  73         int count;              /* number of pages in the list */
  74         int high;               /* high watermark, emptying needed */
  75         int batch;              /* chunk size for buddy add/remove */
  76         struct list_head list;  /* the list of pages */
  77 };
  78
  79 struct per_cpu_pageset {
  80         struct per_cpu_pages pcp[2];    /* 0: hot.  1: cold */
  81 #ifdef CONFIG_SMP
  82         s8 stat_threshold;
  83         s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
  84 #endif
  85 } ____cacheline_aligned_in_smp;
  86
  87 #ifdef CONFIG_NUMA
  88 #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)])
  89 #else
  90 #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)])
  91 #endif
  92
  93 enum zone_type {
  94         /*
  95          * ZONE_DMA is used when there are devices that are not able
  96          * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
  97          * carve out the portion of memory that is needed for these devices.
  98          * The range is arch specific.
  99          *
 100          * Some examples
 101          *
 102          * Architecture         Limit
 103          * ---------------------------
 104          * parisc, ia64, sparc  <4G
 105          * s390                 <2G
 106          * arm26                <48M
 107          * arm                  Various
 108          * alpha                Unlimited or 0-16MB.
 109          *
 110          * i386, x86_64 and multiple other arches
 111          *                      <16M.
 112          */
 113         ZONE_DMA,
 114 #ifdef CONFIG_ZONE_DMA32
 115         /*
 116          * x86_64 needs two ZONE_DMAs because it supports devices that are
 117          * only able to do DMA to the lower 16M but also 32 bit devices that
 118          * can only do DMA areas below 4G.
 119          */
 120         ZONE_DMA32,
 121 #endif
 122         /*
 123          * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
 124          * performed on pages in ZONE_NORMAL if the DMA devices support
 125          * transfers to all addressable memory.
 126          */
 127         ZONE_NORMAL,
 128 #ifdef CONFIG_HIGHMEM
 129         /*
 130          * A memory area that is only addressable by the kernel through
 131          * mapping portions into its own address space. This is for example
 132          * used by i386 to allow the kernel to address the memory beyond
 133          * 900MB. The kernel will set up special mappings (page
 134          * table entries on i386) for each page that the kernel needs to
 135          * access.
 136          */
 137         ZONE_HIGHMEM,
 138 #endif
 139         MAX_NR_ZONES
 140 };
 141
 142 /*
 143  * When a memory allocation must conform to specific limitations (such
 144  * as being suitable for DMA) the caller will pass in hints to the
 145  * allocator in the gfp_mask, in the zone modifier bits.  These bits
 146  * are used to select a priority ordered list of memory zones which
 147  * match the requested limits. See gfp_zone() in include/linux/gfp.h
 148  */
 149
 150 #if !defined(CONFIG_ZONE_DMA32) && !defined(CONFIG_HIGHMEM)
 151 #define ZONES_SHIFT 1
 152 #else
 153 #define ZONES_SHIFT 2
 154 #endif
 155
 156 struct zone {
 157         /* Fields commonly accessed by the page allocator */
 158         unsigned long           free_pages;
 159         unsigned long           pages_min, pages_low, pages_high;
 160         /*
 161          * We don't know if the memory that we're going to allocate will be freeable
 162          * or/and it will be released eventually, so to avoid totally wasting several
 163          * GB of ram we must reserve some of the lower zone memory (otherwise we risk
 164          * to run OOM on the lower zones despite there's tons of freeable ram
 165          * on the higher zones). This array is recalculated at runtime if the
 166          * sysctl_lowmem_reserve_ratio sysctl changes.
 167          */
 168         unsigned long           lowmem_reserve[MAX_NR_ZONES];
 169
 170 #ifdef CONFIG_NUMA
 171         int node;
 172         /*
 173          * zone reclaim becomes active if more unmapped pages exist.
 174          */
 175         unsigned long           min_unmapped_pages;
 176         unsigned long           min_slab_pages;
 177         struct per_cpu_pageset  *pageset[NR_CPUS];
 178 #else
 179         struct per_cpu_pageset  pageset[NR_CPUS];
 180 #endif
 181         /*
 182          * free areas of different sizes
 183          */
 184         spinlock_t              lock;
 185 #ifdef CONFIG_MEMORY_HOTPLUG
 186         /* see spanned/present_pages for more description */
 187         seqlock_t               span_seqlock;
 188 #endif
 189         struct free_area        free_area[MAX_ORDER];
 190
 191
 192         ZONE_PADDING(_pad1_)
 193
 194         /* Fields commonly accessed by the page reclaim scanner */
 195         spinlock_t              lru_lock;
 196         struct list_head        active_list;
 197         struct list_head        inactive_list;
 198         unsigned long           nr_scan_active;
 199         unsigned long           nr_scan_inactive;
 200         unsigned long           nr_active;
 201         unsigned long           nr_inactive;
 202         unsigned long           pages_scanned;     /* since last reclaim */
 203         int                     all_unreclaimable; /* All pages pinned */
 204
 205         /* A count of how many reclaimers are scanning this zone */
 206         atomic_t                reclaim_in_progress;
 207
 208         /* Zone statistics */
 209         atomic_long_t           vm_stat[NR_VM_ZONE_STAT_ITEMS];
 210
 211         /*
 212          * prev_priority holds the scanning priority for this zone.  It is
 213          * defined as the scanning priority at which we achieved our reclaim
 214          * target at the previous try_to_free_pages() or balance_pgdat()
 215          * invokation.
 216          *
 217          * We use prev_priority as a measure of how much stress page reclaim is
 218          * under - it drives the swappiness decision: whether to unmap mapped
 219          * pages.
 220          *
 221          * Access to both this field is quite racy even on uniprocessor.  But
 222          * it is expected to average out OK.
 223          */
 224         int prev_priority;
 225
 226
 227         ZONE_PADDING(_pad2_)
 228         /* Rarely used or read-mostly fields */
 229
 230         /*
 231          * wait_table           -- the array holding the hash table
 232          * wait_table_hash_nr_entries   -- the size of the hash table array
 233          * wait_table_bits      -- wait_table_size == (1 << wait_table_bits)
 234          *
 235          * The purpose of all these is to keep track of the people
 236          * waiting for a page to become available and make them
 237          * runnable again when possible. The trouble is that this
 238          * consumes a lot of space, especially when so few things
 239          * wait on pages at a given time. So instead of using
 240          * per-page waitqueues, we use a waitqueue hash table.
 241          *
 242          * The bucket discipline is to sleep on the same queue when
 243          * colliding and wake all in that wait queue when removing.
 244          * When something wakes, it must check to be sure its page is
 245          * truly available, a la thundering herd. The cost of a
 246          * collision is great, but given the expected load of the
 247          * table, they should be so rare as to be outweighed by the
 248          * benefits from the saved space.
 249          *
 250          * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
 251          * primary users of these fields, and in mm/page_alloc.c
 252          * free_area_init_core() performs the initialization of them.
 253          */
 254         wait_queue_head_t       * wait_table;
 255         unsigned long           wait_table_hash_nr_entries;
 256         unsigned long           wait_table_bits;
 257
 258         /*
 259          * Discontig memory support fields.
 260          */
 261         struct pglist_data      *zone_pgdat;
 262         /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
 263         unsigned long           zone_start_pfn;
 264
 265         /*
 266          * zone_start_pfn, spanned_pages and present_pages are all
 267          * protected by span_seqlock.  It is a seqlock because it has
 268          * to be read outside of zone->lock, and it is done in the main
 269          * allocator path.  But, it is written quite infrequently.
 270          *
 271          * The lock is declared along with zone->lock because it is
 272          * frequently read in proximity to zone->lock.  It's good to
 273          * give them a chance of being in the same cacheline.
 274          */
 275         unsigned long           spanned_pages;  /* total size, including holes */
 276         unsigned long           present_pages;  /* amount of memory (excluding holes) */
 277
 278         /*
 279          * rarely used fields:
 280          */
 281         const char              *name;
 282 } ____cacheline_internodealigned_in_smp;
 283
 284 /*
 285  * The "priority" of VM scanning is how much of the queues we will scan in one
 286  * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
 287  * queues ("queue_length >> 12") during an aging round.
 288  */
 289 #define DEF_PRIORITY 12
 290
 291 /* Maximum number of zones on a zonelist */
 292 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
 293
 294 #ifdef CONFIG_NUMA
 295 /*
 296  * We cache key information from each zonelist for smaller cache
 297  * footprint when scanning for free pages in get_page_from_freelist().
 298  *
 299  * 1) The BITMAP fullzones tracks which zones in a zonelist have come
 300  *    up short of free memory since the last time (last_fullzone_zap)
 301  *    we zero'd fullzones.
 302  * 2) The array z_to_n[] maps each zone in the zonelist to its node
 303  *    id, so that we can efficiently evaluate whether that node is
 304  *    set in the current tasks mems_allowed.
 305  *
 306  * Both fullzones and z_to_n[] are one-to-one with the zonelist,
 307  * indexed by a zones offset in the zonelist zones[] array.
 308  *
 309  * The get_page_from_freelist() routine does two scans.  During the
 310  * first scan, we skip zones whose corresponding bit in 'fullzones'
 311  * is set or whose corresponding node in current->mems_allowed (which
 312  * comes from cpusets) is not set.  During the second scan, we bypass
 313  * this zonelist_cache, to ensure we look methodically at each zone.
 314  *
 315  * Once per second, we zero out (zap) fullzones, forcing us to
 316  * reconsider nodes that might have regained more free memory.
 317  * The field last_full_zap is the time we last zapped fullzones.
 318  *
 319  * This mechanism reduces the amount of time we waste repeatedly
 320  * reexaming zones for free memory when they just came up low on
 321  * memory momentarilly ago.
 322  *
 323  * The zonelist_cache struct members logically belong in struct
 324  * zonelist.  However, the mempolicy zonelists constructed for
 325  * MPOL_BIND are intentionally variable length (and usually much
 326  * shorter).  A general purpose mechanism for handling structs with
 327  * multiple variable length members is more mechanism than we want
 328  * here.  We resort to some special case hackery instead.
 329  *
 330  * The MPOL_BIND zonelists don't need this zonelist_cache (in good
 331  * part because they are shorter), so we put the fixed length stuff
 332  * at the front of the zonelist struct, ending in a variable length
 333  * zones[], as is needed by MPOL_BIND.
 334  *
 335  * Then we put the optional zonelist cache on the end of the zonelist
 336  * struct.  This optional stuff is found by a 'zlcache_ptr' pointer in
 337  * the fixed length portion at the front of the struct.  This pointer
 338  * both enables us to find the zonelist cache, and in the case of
 339  * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL)
 340  * to know that the zonelist cache is not there.
 341  *
 342  * The end result is that struct zonelists come in two flavors:
 343  *  1) The full, fixed length version, shown below, and
 344  *  2) The custom zonelists for MPOL_BIND.
 345  * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache.
 346  *
 347  * Even though there may be multiple CPU cores on a node modifying
 348  * fullzones or last_full_zap in the same zonelist_cache at the same
 349  * time, we don't lock it.  This is just hint data - if it is wrong now
 350  * and then, the allocator will still function, perhaps a bit slower.
 351  */
 352
 353
 354 struct zonelist_cache {
 355         unsigned short z_to_n[MAX_ZONES_PER_ZONELIST];          /* zone->nid */
 356         DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST);      /* zone full? */
 357         unsigned long last_full_zap;            /* when last zap'd (jiffies) */
 358 };
 359 #else
 360 struct zonelist_cache;
 361 #endif
 362
 363 /*
 364  * One allocation request operates on a zonelist. A zonelist
 365  * is a list of zones, the first one is the 'goal' of the
 366  * allocation, the other zones are fallback zones, in decreasing
 367  * priority.
 368  *
 369  * If zlcache_ptr is not NULL, then it is just the address of zlcache,
 370  * as explained above.  If zlcache_ptr is NULL, there is no zlcache.
 371  */
 372
 373 struct zonelist {
 374         struct zonelist_cache *zlcache_ptr;                  // NULL or &zlcache
 375         struct zone *zones[MAX_ZONES_PER_ZONELIST + 1];      // NULL delimited
 376 #ifdef CONFIG_NUMA
 377         struct zonelist_cache zlcache;                       // optional ...
 378 #endif
 379 };
 380
 381 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
 382 struct node_active_region {
 383         unsigned long start_pfn;
 384         unsigned long end_pfn;
 385         int nid;
 386 };
 387 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
 388
 389 #ifndef CONFIG_DISCONTIGMEM
 390 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
 391 extern struct page *mem_map;
 392 #endif
 393
 394 /*
 395  * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
 396  * (mostly NUMA machines?) to denote a higher-level memory zone than the
 397  * zone denotes.
 398  *
 399  * On NUMA machines, each NUMA node would have a pg_data_t to describe
 400  * it's memory layout.
 401  *
 402  * Memory statistics and page replacement data structures are maintained on a
 403  * per-zone basis.
 404  */
 405 struct bootmem_data;
 406 typedef struct pglist_data {
 407         struct zone node_zones[MAX_NR_ZONES];
 408         struct zonelist node_zonelists[MAX_NR_ZONES];
 409         int nr_zones;
 410 #ifdef CONFIG_FLAT_NODE_MEM_MAP
 411         struct page *node_mem_map;
 412 #endif
 413         struct bootmem_data *bdata;
 414 #ifdef CONFIG_MEMORY_HOTPLUG
 415         /*
 416          * Must be held any time you expect node_start_pfn, node_present_pages
 417          * or node_spanned_pages stay constant.  Holding this will also
 418          * guarantee that any pfn_valid() stays that way.
 419          *
 420          * Nests above zone->lock and zone->size_seqlock.
 421          */
 422         spinlock_t node_size_lock;
 423 #endif
 424         unsigned long node_start_pfn;
 425         unsigned long node_present_pages; /* total number of physical pages */
 426         unsigned long node_spanned_pages; /* total size of physical page
 427                                              range, including holes */
 428         int node_id;
 429         wait_queue_head_t kswapd_wait;
 430         struct task_struct *kswapd;
 431         int kswapd_max_order;
 432 } pg_data_t;
 433
 434 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
 435 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
 436 #ifdef CONFIG_FLAT_NODE_MEM_MAP
 437 #define pgdat_page_nr(pgdat, pagenr)    ((pgdat)->node_mem_map + (pagenr))
 438 #else
 439 #define pgdat_page_nr(pgdat, pagenr)    pfn_to_page((pgdat)->node_start_pfn + (pagenr))
 440 #endif
 441 #define nid_page_nr(nid, pagenr)        pgdat_page_nr(NODE_DATA(nid),(pagenr))
 442
 443 #include <linux/memory_hotplug.h>
 444
 445 void __get_zone_counts(unsigned long *active, unsigned long *inactive,
 446                         unsigned long *free, struct pglist_data *pgdat);
 447 void get_zone_counts(unsigned long *active, unsigned long *inactive,
 448                         unsigned long *free);
 449 void build_all_zonelists(void);
 450 void wakeup_kswapd(struct zone *zone, int order);
 451 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
 452                 int classzone_idx, int alloc_flags);
 453
 454 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
 455                                      unsigned long size);
 456
 457 #ifdef CONFIG_HAVE_MEMORY_PRESENT
 458 void memory_present(int nid, unsigned long start, unsigned long end);
 459 #else
 460 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
 461 #endif
 462
 463 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
 464 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
 465 #endif
 466
 467 /*
 468  * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
 469  */
 470 #define zone_idx(zone)          ((zone) - (zone)->zone_pgdat->node_zones)
 471
 472 static inline int populated_zone(struct zone *zone)
 473 {
 474         return (!!zone->present_pages);
 475 }
 476
 477 static inline int is_highmem_idx(enum zone_type idx)
 478 {
 479 #ifdef CONFIG_HIGHMEM
 480         return (idx == ZONE_HIGHMEM);
 481 #else
 482         return 0;
 483 #endif
 484 }
 485
 486 static inline int is_normal_idx(enum zone_type idx)
 487 {
 488         return (idx == ZONE_NORMAL);
 489 }
 490
 491 /**
 492  * is_highmem - helper function to quickly check if a struct zone is a
 493  *              highmem zone or not.  This is an attempt to keep references
 494  *              to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
 495  * @zone - pointer to struct zone variable
 496  */
 497 static inline int is_highmem(struct zone *zone)
 498 {
 499 #ifdef CONFIG_HIGHMEM
 500         return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM;
 501 #else
 502         return 0;
 503 #endif
 504 }
 505
 506 static inline int is_normal(struct zone *zone)
 507 {
 508         return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
 509 }
 510
 511 static inline int is_dma32(struct zone *zone)
 512 {
 513 #ifdef CONFIG_ZONE_DMA32
 514         return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
 515 #else
 516         return 0;
 517 #endif
 518 }
 519
 520 static inline int is_dma(struct zone *zone)
 521 {
 522         return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
 523 }
 524
 525 /* These two functions are used to setup the per zone pages min values */
 526 struct ctl_table;
 527 struct file;
 528 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *,
 529                                         void __user *, size_t *, loff_t *);
 530 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
 531 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *,
 532                                         void __user *, size_t *, loff_t *);
 533 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *,
 534                                         void __user *, size_t *, loff_t *);
 535 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
 536                         struct file *, void __user *, size_t *, loff_t *);
 537 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
 538                         struct file *, void __user *, size_t *, loff_t *);
 539
 540 #include <linux/topology.h>
 541 /* Returns the number of the current Node. */
 542 #ifndef numa_node_id
 543 #define numa_node_id()          (cpu_to_node(raw_smp_processor_id()))
 544 #endif
 545
 546 #ifndef CONFIG_NEED_MULTIPLE_NODES
 547
 548 extern struct pglist_data contig_page_data;
 549 #define NODE_DATA(nid)          (&contig_page_data)
 550 #define NODE_MEM_MAP(nid)       mem_map
 551 #define MAX_NODES_SHIFT         1
 552
 553 #else /* CONFIG_NEED_MULTIPLE_NODES */
 554
 555 #include <asm/mmzone.h>
 556
 557 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 558
 559 extern struct pglist_data *first_online_pgdat(void);
 560 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
 561 extern struct zone *next_zone(struct zone *zone);
 562
 563 /**
 564  * for_each_pgdat - helper macro to iterate over all nodes
 565  * @pgdat - pointer to a pg_data_t variable
 566  */
 567 #define for_each_online_pgdat(pgdat)                    \
 568         for (pgdat = first_online_pgdat();              \
 569              pgdat;                                     \
 570              pgdat = next_online_pgdat(pgdat))
 571 /**
 572  * for_each_zone - helper macro to iterate over all memory zones
 573  * @zone - pointer to struct zone variable
 574  *
 575  * The user only needs to declare the zone variable, for_each_zone
 576  * fills it in.
 577  */
 578 #define for_each_zone(zone)                             \
 579         for (zone = (first_online_pgdat())->node_zones; \
 580              zone;                                      \
 581              zone = next_zone(zone))
 582
 583 #ifdef CONFIG_SPARSEMEM
 584 #include <asm/sparsemem.h>
 585 #endif
 586
 587 #if BITS_PER_LONG == 32
 588 /*
 589  * with 32 bit page->flags field, we reserve 9 bits for node/zone info.
 590  * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes.
 591  */
 592 #define FLAGS_RESERVED          9
 593
 594 #elif BITS_PER_LONG == 64
 595 /*
 596  * with 64 bit flags field, there's plenty of room.
 597  */
 598 #define FLAGS_RESERVED          32
 599
 600 #else
 601
 602 #error BITS_PER_LONG not defined
 603
 604 #endif
 605
 606 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
 607         !defined(CONFIG_ARCH_POPULATES_NODE_MAP)
 608 #define early_pfn_to_nid(nid)  (0UL)
 609 #endif
 610
 611 #ifdef CONFIG_FLATMEM
 612 #define pfn_to_nid(pfn)         (0)
 613 #endif
 614
 615 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
 616 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
 617
 618 #ifdef CONFIG_SPARSEMEM
 619
 620 /*
 621  * SECTION_SHIFT                #bits space required to store a section #
 622  *
 623  * PA_SECTION_SHIFT             physical address to/from section number
 624  * PFN_SECTION_SHIFT            pfn to/from section number
 625  */
 626 #define SECTIONS_SHIFT          (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
 627
 628 #define PA_SECTION_SHIFT        (SECTION_SIZE_BITS)
 629 #define PFN_SECTION_SHIFT       (SECTION_SIZE_BITS - PAGE_SHIFT)
 630
 631 #define NR_MEM_SECTIONS         (1UL << SECTIONS_SHIFT)
 632
 633 #define PAGES_PER_SECTION       (1UL << PFN_SECTION_SHIFT)
 634 #define PAGE_SECTION_MASK       (~(PAGES_PER_SECTION-1))
 635
 636 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
 637 #error Allocator MAX_ORDER exceeds SECTION_SIZE
 638 #endif
 639
 640 struct page;
 641 struct mem_section {
 642         /*
 643          * This is, logically, a pointer to an array of struct
 644          * pages.  However, it is stored with some other magic.
 645          * (see sparse.c::sparse_init_one_section())
 646          *
 647          * Additionally during early boot we encode node id of
 648          * the location of the section here to guide allocation.
 649          * (see sparse.c::memory_present())
 650          *
 651          * Making it a UL at least makes someone do a cast
 652          * before using it wrong.
 653          */
 654         unsigned long section_mem_map;
 655 };
 656
 657 #ifdef CONFIG_SPARSEMEM_EXTREME
 658 #define SECTIONS_PER_ROOT       (PAGE_SIZE / sizeof (struct mem_section))
 659 #else
 660 #define SECTIONS_PER_ROOT       1
 661 #endif
 662
 663 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
 664 #define NR_SECTION_ROOTS        (NR_MEM_SECTIONS / SECTIONS_PER_ROOT)
 665 #define SECTION_ROOT_MASK       (SECTIONS_PER_ROOT - 1)
 666
 667 #ifdef CONFIG_SPARSEMEM_EXTREME
 668 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
 669 #else
 670 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
 671 #endif
 672
 673 static inline struct mem_section *__nr_to_section(unsigned long nr)
 674 {
 675         if (!mem_section[SECTION_NR_TO_ROOT(nr)])
 676                 return NULL;
 677         return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
 678 }
 679 extern int __section_nr(struct mem_section* ms);
 680
 681 /*
 682  * We use the lower bits of the mem_map pointer to store
 683  * a little bit of information.  There should be at least
 684  * 3 bits here due to 32-bit alignment.
 685  */
 686 #define SECTION_MARKED_PRESENT  (1UL<<0)
 687 #define SECTION_HAS_MEM_MAP     (1UL<<1)
 688 #define SECTION_MAP_LAST_BIT    (1UL<<2)
 689 #define SECTION_MAP_MASK        (~(SECTION_MAP_LAST_BIT-1))
 690 #define SECTION_NID_SHIFT       2
 691
 692 static inline struct page *__section_mem_map_addr(struct mem_section *section)
 693 {
 694         unsigned long map = section->section_mem_map;
 695         map &= SECTION_MAP_MASK;
 696         return (struct page *)map;
 697 }
 698
 699 static inline int valid_section(struct mem_section *section)
 700 {
 701         return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
 702 }
 703
 704 static inline int section_has_mem_map(struct mem_section *section)
 705 {
 706         return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
 707 }
 708
 709 static inline int valid_section_nr(unsigned long nr)
 710 {
 711         return valid_section(__nr_to_section(nr));
 712 }
 713
 714 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
 715 {
 716         return __nr_to_section(pfn_to_section_nr(pfn));
 717 }
 718
 719 static inline int pfn_valid(unsigned long pfn)
 720 {
 721         if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
 722                 return 0;
 723         return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
 724 }
 725
 726 /*
 727  * These are _only_ used during initialisation, therefore they
 728  * can use __initdata ...  They could have names to indicate
 729  * this restriction.
 730  */
 731 #ifdef CONFIG_NUMA
 732 #define pfn_to_nid(pfn)                                                 \
 733 ({                                                                      \
 734         unsigned long __pfn_to_nid_pfn = (pfn);                         \
 735         page_to_nid(pfn_to_page(__pfn_to_nid_pfn));                     \
 736 })
 737 #else
 738 #define pfn_to_nid(pfn)         (0)
 739 #endif
 740
 741 #define early_pfn_valid(pfn)    pfn_valid(pfn)
 742 void sparse_init(void);
 743 #else
 744 #define sparse_init()   do {} while (0)
 745 #define sparse_index_init(_sec, _nid)  do {} while (0)
 746 #endif /* CONFIG_SPARSEMEM */
 747
 748 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
 749 #define early_pfn_in_nid(pfn, nid)      (early_pfn_to_nid(pfn) == (nid))
 750 #else
 751 #define early_pfn_in_nid(pfn, nid)      (1)
 752 #endif
 753
 754 #ifndef early_pfn_valid
 755 #define early_pfn_valid(pfn)    (1)
 756 #endif
 757
 758 void memory_present(int nid, unsigned long start, unsigned long end);
 759 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
 760
 761 #endif /* !__ASSEMBLY__ */
 762 #endif /* __KERNEL__ */
 763 #endif /* _LINUX_MMZONE_H */