Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
mmzone.h
Go to the documentation of this file.
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 <linux/atomic.h>
20 #include <asm/page.h>
21 
22 /* Free memory management - zoned buddy allocator. */
23 #ifndef CONFIG_FORCE_MAX_ZONEORDER
24 #define MAX_ORDER 11
25 #else
26 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
27 #endif
28 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
29 
30 /*
31  * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
32  * costly to service. That is between allocation orders which should
33  * coalesce naturally under reasonable reclaim pressure and those which
34  * will not.
35  */
36 #define PAGE_ALLOC_COSTLY_ORDER 3
37 
38 enum {
42  MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
44 #ifdef CONFIG_CMA
45  /*
46  * MIGRATE_CMA migration type is designed to mimic the way
47  * ZONE_MOVABLE works. Only movable pages can be allocated
48  * from MIGRATE_CMA pageblocks and page allocator never
49  * implicitly change migration type of MIGRATE_CMA pageblock.
50  *
51  * The way to use it is to change migratetype of a range of
52  * pageblocks to MIGRATE_CMA which can be done by
53  * __free_pageblock_cma() function. What is important though
54  * is that a range of pageblocks must be aligned to
55  * MAX_ORDER_NR_PAGES should biggest page be bigger then
56  * a single pageblock.
57  */
58  MIGRATE_CMA,
59 #endif
60  MIGRATE_ISOLATE, /* can't allocate from here */
62 };
63 
64 #ifdef CONFIG_CMA
65 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
66 # define cma_wmark_pages(zone) zone->min_cma_pages
67 #else
68 # define is_migrate_cma(migratetype) false
69 # define cma_wmark_pages(zone) 0
70 #endif
71 
72 #define for_each_migratetype_order(order, type) \
73  for (order = 0; order < MAX_ORDER; order++) \
74  for (type = 0; type < MIGRATE_TYPES; type++)
75 
77 
78 static inline int get_pageblock_migratetype(struct page *page)
79 {
81 }
82 
83 struct free_area {
85  unsigned long nr_free;
86 };
87 
88 struct pglist_data;
89 
90 /*
91  * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
92  * So add a wild amount of padding here to ensure that they fall into separate
93  * cachelines. There are very few zone structures in the machine, so space
94  * consumption is not a concern here.
95  */
96 #if defined(CONFIG_SMP)
97 struct zone_padding {
98  char x[0];
100 #define ZONE_PADDING(name) struct zone_padding name;
101 #else
102 #define ZONE_PADDING(name)
103 #endif
104 
106  /* First 128 byte cacheline (assuming 64 bit words) */
109  NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
110  NR_ACTIVE_ANON, /* " " " " " */
111  NR_INACTIVE_FILE, /* " " " " " */
112  NR_ACTIVE_FILE, /* " " " " " */
113  NR_UNEVICTABLE, /* " " " " " */
114  NR_MLOCK, /* mlock()ed pages found and moved off LRU */
115  NR_ANON_PAGES, /* Mapped anonymous pages */
116  NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
117  only modified from process context */
123  NR_PAGETABLE, /* used for pagetables */
125  /* Second 128 byte cacheline */
126  NR_UNSTABLE_NFS, /* NFS unstable pages */
129  NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
130  NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
131  NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
132  NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
133  NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
134  NR_DIRTIED, /* page dirtyings since bootup */
135  NR_WRITTEN, /* page writings since bootup */
136 #ifdef CONFIG_NUMA
137  NUMA_HIT, /* allocated in intended node */
138  NUMA_MISS, /* allocated in non intended node */
139  NUMA_FOREIGN, /* was intended here, hit elsewhere */
140  NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
141  NUMA_LOCAL, /* allocation from local node */
142  NUMA_OTHER, /* allocation from other node */
143 #endif
147 
148 /*
149  * We do arithmetic on the LRU lists in various places in the code,
150  * so it is important to keep the active lists LRU_ACTIVE higher in
151  * the array than the corresponding inactive lists, and to keep
152  * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
153  *
154  * This has to be kept in sync with the statistics in zone_stat_item
155  * above and the descriptions in vmstat_text in mm/vmstat.c
156  */
157 #define LRU_BASE 0
158 #define LRU_ACTIVE 1
159 #define LRU_FILE 2
160 
161 enum lru_list {
168 };
169 
170 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
171 
172 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
173 
174 static inline int is_file_lru(enum lru_list lru)
175 {
176  return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
177 }
178 
179 static inline int is_active_lru(enum lru_list lru)
180 {
181  return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
182 }
183 
184 static inline int is_unevictable_lru(enum lru_list lru)
185 {
186  return (lru == LRU_UNEVICTABLE);
187 }
188 
190  /*
191  * The pageout code in vmscan.c keeps track of how many of the
192  * mem/swap backed and file backed pages are referenced.
193  * The higher the rotated/scanned ratio, the more valuable
194  * that cache is.
195  *
196  * The anon LRU stats live in [0], file LRU stats in [1]
197  */
198  unsigned long recent_rotated[2];
199  unsigned long recent_scanned[2];
200 };
201 
202 struct lruvec {
205 #ifdef CONFIG_MEMCG
206  struct zone *zone;
207 #endif
208 };
209 
210 /* Mask used at gathering information at once (see memcontrol.c) */
211 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
212 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
213 #define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
214 
215 /* Isolate clean file */
216 #define ISOLATE_CLEAN ((__force isolate_mode_t)0x1)
217 /* Isolate unmapped file */
218 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
219 /* Isolate for asynchronous migration */
220 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
221 /* Isolate unevictable pages */
222 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
223 
224 /* LRU Isolation modes. */
225 typedef unsigned __bitwise__ isolate_mode_t;
226 
232 };
233 
234 #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
235 #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
236 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
237 
239  int count; /* number of pages in the list */
240  int high; /* high watermark, emptying needed */
241  int batch; /* chunk size for buddy add/remove */
242 
243  /* Lists of pages, one per migrate type stored on the pcp-lists */
245 };
246 
249 #ifdef CONFIG_NUMA
250  s8 expire;
251 #endif
252 #ifdef CONFIG_SMP
253  s8 stat_threshold;
254  s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
255 #endif
256 };
257 
258 #endif /* !__GENERATING_BOUNDS.H */
259 
260 enum zone_type {
261 #ifdef CONFIG_ZONE_DMA
262  /*
263  * ZONE_DMA is used when there are devices that are not able
264  * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
265  * carve out the portion of memory that is needed for these devices.
266  * The range is arch specific.
267  *
268  * Some examples
269  *
270  * Architecture Limit
271  * ---------------------------
272  * parisc, ia64, sparc <4G
273  * s390 <2G
274  * arm Various
275  * alpha Unlimited or 0-16MB.
276  *
277  * i386, x86_64 and multiple other arches
278  * <16M.
279  */
280  ZONE_DMA,
281 #endif
282 #ifdef CONFIG_ZONE_DMA32
283  /*
284  * x86_64 needs two ZONE_DMAs because it supports devices that are
285  * only able to do DMA to the lower 16M but also 32 bit devices that
286  * can only do DMA areas below 4G.
287  */
288  ZONE_DMA32,
289 #endif
290  /*
291  * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
292  * performed on pages in ZONE_NORMAL if the DMA devices support
293  * transfers to all addressable memory.
294  */
296 #ifdef CONFIG_HIGHMEM
297  /*
298  * A memory area that is only addressable by the kernel through
299  * mapping portions into its own address space. This is for example
300  * used by i386 to allow the kernel to address the memory beyond
301  * 900MB. The kernel will set up special mappings (page
302  * table entries on i386) for each page that the kernel needs to
303  * access.
304  */
305  ZONE_HIGHMEM,
306 #endif
309 };
310 
311 #ifndef __GENERATING_BOUNDS_H
312 
313 /*
314  * When a memory allocation must conform to specific limitations (such
315  * as being suitable for DMA) the caller will pass in hints to the
316  * allocator in the gfp_mask, in the zone modifier bits. These bits
317  * are used to select a priority ordered list of memory zones which
318  * match the requested limits. See gfp_zone() in include/linux/gfp.h
319  */
320 
321 #if MAX_NR_ZONES < 2
322 #define ZONES_SHIFT 0
323 #elif MAX_NR_ZONES <= 2
324 #define ZONES_SHIFT 1
325 #elif MAX_NR_ZONES <= 4
326 #define ZONES_SHIFT 2
327 #else
328 #error ZONES_SHIFT -- too many zones configured adjust calculation
329 #endif
330 
331 struct zone {
332  /* Fields commonly accessed by the page allocator */
333 
334  /* zone watermarks, access with *_wmark_pages(zone) macros */
335  unsigned long watermark[NR_WMARK];
336 
337  /*
338  * When free pages are below this point, additional steps are taken
339  * when reading the number of free pages to avoid per-cpu counter
340  * drift allowing watermarks to be breached
341  */
342  unsigned long percpu_drift_mark;
343 
344  /*
345  * We don't know if the memory that we're going to allocate will be freeable
346  * or/and it will be released eventually, so to avoid totally wasting several
347  * GB of ram we must reserve some of the lower zone memory (otherwise we risk
348  * to run OOM on the lower zones despite there's tons of freeable ram
349  * on the higher zones). This array is recalculated at runtime if the
350  * sysctl_lowmem_reserve_ratio sysctl changes.
351  */
352  unsigned long lowmem_reserve[MAX_NR_ZONES];
353 
354  /*
355  * This is a per-zone reserve of pages that should not be
356  * considered dirtyable memory.
357  */
358  unsigned long dirty_balance_reserve;
359 
360 #ifdef CONFIG_NUMA
361  int node;
362  /*
363  * zone reclaim becomes active if more unmapped pages exist.
364  */
365  unsigned long min_unmapped_pages;
366  unsigned long min_slab_pages;
367 #endif
369  /*
370  * free areas of different sizes
371  */
373  int all_unreclaimable; /* All pages pinned */
374 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
375  /* Set to true when the PG_migrate_skip bits should be cleared */
376  bool compact_blockskip_flush;
377 
378  /* pfns where compaction scanners should start */
379  unsigned long compact_cached_free_pfn;
380  unsigned long compact_cached_migrate_pfn;
381 #endif
382 #ifdef CONFIG_MEMORY_HOTPLUG
383  /* see spanned/present_pages for more description */
384  seqlock_t span_seqlock;
385 #endif
386 #ifdef CONFIG_CMA
387  /*
388  * CMA needs to increase watermark levels during the allocation
389  * process to make sure that the system is not starved.
390  */
391  unsigned long min_cma_pages;
392 #endif
394 
395 #ifndef CONFIG_SPARSEMEM
396  /*
397  * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
398  * In SPARSEMEM, this map is stored in struct mem_section
399  */
400  unsigned long *pageblock_flags;
401 #endif /* CONFIG_SPARSEMEM */
402 
403 #ifdef CONFIG_COMPACTION
404  /*
405  * On compaction failure, 1<<compact_defer_shift compactions
406  * are skipped before trying again. The number attempted since
407  * last failure is tracked with compact_considered.
408  */
409  unsigned int compact_considered;
410  unsigned int compact_defer_shift;
411  int compact_order_failed;
412 #endif
413 
414  ZONE_PADDING(_pad1_)
415 
416  /* Fields commonly accessed by the page reclaim scanner */
419 
420  unsigned long pages_scanned; /* since last reclaim */
421  unsigned long flags; /* zone flags, see below */
422 
423  /* Zone statistics */
425 
426  /*
427  * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on
428  * this zone's LRU. Maintained by the pageout code.
429  */
430  unsigned int inactive_ratio;
431 
432 
433  ZONE_PADDING(_pad2_)
434  /* Rarely used or read-mostly fields */
435 
436  /*
437  * wait_table -- the array holding the hash table
438  * wait_table_hash_nr_entries -- the size of the hash table array
439  * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
440  *
441  * The purpose of all these is to keep track of the people
442  * waiting for a page to become available and make them
443  * runnable again when possible. The trouble is that this
444  * consumes a lot of space, especially when so few things
445  * wait on pages at a given time. So instead of using
446  * per-page waitqueues, we use a waitqueue hash table.
447  *
448  * The bucket discipline is to sleep on the same queue when
449  * colliding and wake all in that wait queue when removing.
450  * When something wakes, it must check to be sure its page is
451  * truly available, a la thundering herd. The cost of a
452  * collision is great, but given the expected load of the
453  * table, they should be so rare as to be outweighed by the
454  * benefits from the saved space.
455  *
456  * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
457  * primary users of these fields, and in mm/page_alloc.c
458  * free_area_init_core() performs the initialization of them.
459  */
462  unsigned long wait_table_bits;
463 
464  /*
465  * Discontig memory support fields.
466  */
468  /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
469  unsigned long zone_start_pfn;
470 
471  /*
472  * zone_start_pfn, spanned_pages and present_pages are all
473  * protected by span_seqlock. It is a seqlock because it has
474  * to be read outside of zone->lock, and it is done in the main
475  * allocator path. But, it is written quite infrequently.
476  *
477  * The lock is declared along with zone->lock because it is
478  * frequently read in proximity to zone->lock. It's good to
479  * give them a chance of being in the same cacheline.
480  */
481  unsigned long spanned_pages; /* total size, including holes */
482  unsigned long present_pages; /* amount of memory (excluding holes) */
483 
484  /*
485  * rarely used fields:
486  */
487  const char *name;
488 #ifdef CONFIG_MEMORY_ISOLATION
489  /*
490  * the number of MIGRATE_ISOLATE *pageblock*.
491  * We need this for free page counting. Look at zone_watermark_ok_safe.
492  * It's protected by zone->lock
493  */
494  int nr_pageblock_isolate;
495 #endif
497 
498 typedef enum {
499  ZONE_RECLAIM_LOCKED, /* prevents concurrent reclaim */
500  ZONE_OOM_LOCKED, /* zone is in OOM killer zonelist */
501  ZONE_CONGESTED, /* zone has many dirty pages backed by
502  * a congested BDI
503  */
504 } zone_flags_t;
505 
506 static inline void zone_set_flag(struct zone *zone, zone_flags_t flag)
507 {
508  set_bit(flag, &zone->flags);
509 }
510 
511 static inline int zone_test_and_set_flag(struct zone *zone, zone_flags_t flag)
512 {
513  return test_and_set_bit(flag, &zone->flags);
514 }
515 
516 static inline void zone_clear_flag(struct zone *zone, zone_flags_t flag)
517 {
518  clear_bit(flag, &zone->flags);
519 }
520 
521 static inline int zone_is_reclaim_congested(const struct zone *zone)
522 {
523  return test_bit(ZONE_CONGESTED, &zone->flags);
524 }
525 
526 static inline int zone_is_reclaim_locked(const struct zone *zone)
527 {
528  return test_bit(ZONE_RECLAIM_LOCKED, &zone->flags);
529 }
530 
531 static inline int zone_is_oom_locked(const struct zone *zone)
532 {
533  return test_bit(ZONE_OOM_LOCKED, &zone->flags);
534 }
535 
536 /*
537  * The "priority" of VM scanning is how much of the queues we will scan in one
538  * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
539  * queues ("queue_length >> 12") during an aging round.
540  */
541 #define DEF_PRIORITY 12
542 
543 /* Maximum number of zones on a zonelist */
544 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
545 
546 #ifdef CONFIG_NUMA
547 
548 /*
549  * The NUMA zonelists are doubled because we need zonelists that restrict the
550  * allocations to a single node for GFP_THISNODE.
551  *
552  * [0] : Zonelist with fallback
553  * [1] : No fallback (GFP_THISNODE)
554  */
555 #define MAX_ZONELISTS 2
556 
557 
558 /*
559  * We cache key information from each zonelist for smaller cache
560  * footprint when scanning for free pages in get_page_from_freelist().
561  *
562  * 1) The BITMAP fullzones tracks which zones in a zonelist have come
563  * up short of free memory since the last time (last_fullzone_zap)
564  * we zero'd fullzones.
565  * 2) The array z_to_n[] maps each zone in the zonelist to its node
566  * id, so that we can efficiently evaluate whether that node is
567  * set in the current tasks mems_allowed.
568  *
569  * Both fullzones and z_to_n[] are one-to-one with the zonelist,
570  * indexed by a zones offset in the zonelist zones[] array.
571  *
572  * The get_page_from_freelist() routine does two scans. During the
573  * first scan, we skip zones whose corresponding bit in 'fullzones'
574  * is set or whose corresponding node in current->mems_allowed (which
575  * comes from cpusets) is not set. During the second scan, we bypass
576  * this zonelist_cache, to ensure we look methodically at each zone.
577  *
578  * Once per second, we zero out (zap) fullzones, forcing us to
579  * reconsider nodes that might have regained more free memory.
580  * The field last_full_zap is the time we last zapped fullzones.
581  *
582  * This mechanism reduces the amount of time we waste repeatedly
583  * reexaming zones for free memory when they just came up low on
584  * memory momentarilly ago.
585  *
586  * The zonelist_cache struct members logically belong in struct
587  * zonelist. However, the mempolicy zonelists constructed for
588  * MPOL_BIND are intentionally variable length (and usually much
589  * shorter). A general purpose mechanism for handling structs with
590  * multiple variable length members is more mechanism than we want
591  * here. We resort to some special case hackery instead.
592  *
593  * The MPOL_BIND zonelists don't need this zonelist_cache (in good
594  * part because they are shorter), so we put the fixed length stuff
595  * at the front of the zonelist struct, ending in a variable length
596  * zones[], as is needed by MPOL_BIND.
597  *
598  * Then we put the optional zonelist cache on the end of the zonelist
599  * struct. This optional stuff is found by a 'zlcache_ptr' pointer in
600  * the fixed length portion at the front of the struct. This pointer
601  * both enables us to find the zonelist cache, and in the case of
602  * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL)
603  * to know that the zonelist cache is not there.
604  *
605  * The end result is that struct zonelists come in two flavors:
606  * 1) The full, fixed length version, shown below, and
607  * 2) The custom zonelists for MPOL_BIND.
608  * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache.
609  *
610  * Even though there may be multiple CPU cores on a node modifying
611  * fullzones or last_full_zap in the same zonelist_cache at the same
612  * time, we don't lock it. This is just hint data - if it is wrong now
613  * and then, the allocator will still function, perhaps a bit slower.
614  */
615 
616 
617 struct zonelist_cache {
618  unsigned short z_to_n[MAX_ZONES_PER_ZONELIST]; /* zone->nid */
619  DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST); /* zone full? */
620  unsigned long last_full_zap; /* when last zap'd (jiffies) */
621 };
622 #else
623 #define MAX_ZONELISTS 1
624 struct zonelist_cache;
625 #endif
626 
627 /*
628  * This struct contains information about a zone in a zonelist. It is stored
629  * here to avoid dereferences into large structures and lookups of tables
630  */
631 struct zoneref {
632  struct zone *zone; /* Pointer to actual zone */
633  int zone_idx; /* zone_idx(zoneref->zone) */
634 };
635 
636 /*
637  * One allocation request operates on a zonelist. A zonelist
638  * is a list of zones, the first one is the 'goal' of the
639  * allocation, the other zones are fallback zones, in decreasing
640  * priority.
641  *
642  * If zlcache_ptr is not NULL, then it is just the address of zlcache,
643  * as explained above. If zlcache_ptr is NULL, there is no zlcache.
644  * *
645  * To speed the reading of the zonelist, the zonerefs contain the zone index
646  * of the entry being read. Helper functions to access information given
647  * a struct zoneref are
648  *
649  * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
650  * zonelist_zone_idx() - Return the index of the zone for an entry
651  * zonelist_node_idx() - Return the index of the node for an entry
652  */
653 struct zonelist {
654  struct zonelist_cache *zlcache_ptr; // NULL or &zlcache
656 #ifdef CONFIG_NUMA
657  struct zonelist_cache zlcache; // optional ...
658 #endif
659 };
660 
661 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
662 struct node_active_region {
663  unsigned long start_pfn;
664  unsigned long end_pfn;
665  int nid;
666 };
667 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
668 
669 #ifndef CONFIG_DISCONTIGMEM
670 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
671 extern struct page *mem_map;
672 #endif
673 
674 /*
675  * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
676  * (mostly NUMA machines?) to denote a higher-level memory zone than the
677  * zone denotes.
678  *
679  * On NUMA machines, each NUMA node would have a pg_data_t to describe
680  * it's memory layout.
681  *
682  * Memory statistics and page replacement data structures are maintained on a
683  * per-zone basis.
684  */
685 struct bootmem_data;
686 typedef struct pglist_data {
687  struct zone node_zones[MAX_NR_ZONES];
689  int nr_zones;
690 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
691  struct page *node_mem_map;
692 #ifdef CONFIG_MEMCG
693  struct page_cgroup *node_page_cgroup;
694 #endif
695 #endif
696 #ifndef CONFIG_NO_BOOTMEM
698 #endif
699 #ifdef CONFIG_MEMORY_HOTPLUG
700  /*
701  * Must be held any time you expect node_start_pfn, node_present_pages
702  * or node_spanned_pages stay constant. Holding this will also
703  * guarantee that any pfn_valid() stays that way.
704  *
705  * Nests above zone->lock and zone->size_seqlock.
706  */
707  spinlock_t node_size_lock;
708 #endif
709  unsigned long node_start_pfn;
710  unsigned long node_present_pages; /* total number of physical pages */
711  unsigned long node_spanned_pages; /* total size of physical page
712  range, including holes */
713  int node_id;
714  nodemask_t reclaim_nodes; /* Nodes allowed to reclaim from */
717  struct task_struct *kswapd; /* Protected by lock_memory_hotplug() */
720 } pg_data_t;
721 
722 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
723 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
724 #ifdef CONFIG_FLAT_NODE_MEM_MAP
725 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
726 #else
727 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
728 #endif
729 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
730 
731 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
732 
733 #define node_end_pfn(nid) ({\
734  pg_data_t *__pgdat = NODE_DATA(nid);\
735  __pgdat->node_start_pfn + __pgdat->node_spanned_pages;\
736 })
737 
738 #include <linux/memory_hotplug.h>
739 
740 extern struct mutex zonelists_mutex;
741 void build_all_zonelists(pg_data_t *pgdat, struct zone *zone);
742 void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx);
743 bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
744  int classzone_idx, int alloc_flags);
745 bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
746  int classzone_idx, int alloc_flags);
750 };
751 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
752  unsigned long size,
753  enum memmap_context context);
754 
755 extern void lruvec_init(struct lruvec *lruvec);
756 
757 static inline struct zone *lruvec_zone(struct lruvec *lruvec)
758 {
759 #ifdef CONFIG_MEMCG
760  return lruvec->zone;
761 #else
762  return container_of(lruvec, struct zone, lruvec);
763 #endif
764 }
765 
766 #ifdef CONFIG_HAVE_MEMORY_PRESENT
767 void memory_present(int nid, unsigned long start, unsigned long end);
768 #else
769 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
770 #endif
771 
772 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
773 int local_memory_node(int node_id);
774 #else
775 static inline int local_memory_node(int node_id) { return node_id; };
776 #endif
777 
778 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
779 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
780 #endif
781 
782 /*
783  * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
784  */
785 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
786 
787 static inline int populated_zone(struct zone *zone)
788 {
789  return (!!zone->present_pages);
790 }
791 
792 extern int movable_zone;
793 
794 static inline int zone_movable_is_highmem(void)
795 {
796 #if defined(CONFIG_HIGHMEM) && defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
797  return movable_zone == ZONE_HIGHMEM;
798 #else
799  return 0;
800 #endif
801 }
802 
803 static inline int is_highmem_idx(enum zone_type idx)
804 {
805 #ifdef CONFIG_HIGHMEM
806  return (idx == ZONE_HIGHMEM ||
807  (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
808 #else
809  return 0;
810 #endif
811 }
812 
813 static inline int is_normal_idx(enum zone_type idx)
814 {
815  return (idx == ZONE_NORMAL);
816 }
817 
824 static inline int is_highmem(struct zone *zone)
825 {
826 #ifdef CONFIG_HIGHMEM
827  int zone_off = (char *)zone - (char *)zone->zone_pgdat->node_zones;
828  return zone_off == ZONE_HIGHMEM * sizeof(*zone) ||
829  (zone_off == ZONE_MOVABLE * sizeof(*zone) &&
830  zone_movable_is_highmem());
831 #else
832  return 0;
833 #endif
834 }
835 
836 static inline int is_normal(struct zone *zone)
837 {
838  return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
839 }
840 
841 static inline int is_dma32(struct zone *zone)
842 {
843 #ifdef CONFIG_ZONE_DMA32
844  return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
845 #else
846  return 0;
847 #endif
848 }
849 
850 static inline int is_dma(struct zone *zone)
851 {
852 #ifdef CONFIG_ZONE_DMA
853  return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
854 #else
855  return 0;
856 #endif
857 }
858 
859 /* These two functions are used to setup the per zone pages min values */
860 struct ctl_table;
861 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
862  void __user *, size_t *, loff_t *);
863 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
865  void __user *, size_t *, loff_t *);
867  void __user *, size_t *, loff_t *);
869  void __user *, size_t *, loff_t *);
871  void __user *, size_t *, loff_t *);
872 
873 extern int numa_zonelist_order_handler(struct ctl_table *, int,
874  void __user *, size_t *, loff_t *);
875 extern char numa_zonelist_order[];
876 #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
877 
878 #ifndef CONFIG_NEED_MULTIPLE_NODES
879 
880 extern struct pglist_data contig_page_data;
881 #define NODE_DATA(nid) (&contig_page_data)
882 #define NODE_MEM_MAP(nid) mem_map
883 
884 #else /* CONFIG_NEED_MULTIPLE_NODES */
885 
886 #include <asm/mmzone.h>
887 
888 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
889 
890 extern struct pglist_data *first_online_pgdat(void);
891 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
892 extern struct zone *next_zone(struct zone *zone);
893 
898 #define for_each_online_pgdat(pgdat) \
899  for (pgdat = first_online_pgdat(); \
900  pgdat; \
901  pgdat = next_online_pgdat(pgdat))
902 
909 #define for_each_zone(zone) \
910  for (zone = (first_online_pgdat())->node_zones; \
911  zone; \
912  zone = next_zone(zone))
913 
914 #define for_each_populated_zone(zone) \
915  for (zone = (first_online_pgdat())->node_zones; \
916  zone; \
917  zone = next_zone(zone)) \
918  if (!populated_zone(zone)) \
919  ; /* do nothing */ \
920  else
921 
922 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
923 {
924  return zoneref->zone;
925 }
926 
927 static inline int zonelist_zone_idx(struct zoneref *zoneref)
928 {
929  return zoneref->zone_idx;
930 }
931 
932 static inline int zonelist_node_idx(struct zoneref *zoneref)
933 {
934 #ifdef CONFIG_NUMA
935  /* zone_to_nid not available in this context */
936  return zoneref->zone->node;
937 #else
938  return 0;
939 #endif /* CONFIG_NUMA */
940 }
941 
955 struct zoneref *next_zones_zonelist(struct zoneref *z,
956  enum zone_type highest_zoneidx,
957  nodemask_t *nodes,
958  struct zone **zone);
959 
972 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
973  enum zone_type highest_zoneidx,
974  nodemask_t *nodes,
975  struct zone **zone)
976 {
977  return next_zones_zonelist(zonelist->_zonerefs, highest_zoneidx, nodes,
978  zone);
979 }
980 
992 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
993  for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \
994  zone; \
995  z = next_zones_zonelist(++z, highidx, nodemask, &zone)) \
996 
997 
1006 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1007  for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1008 
1009 #ifdef CONFIG_SPARSEMEM
1010 #include <asm/sparsemem.h>
1011 #endif
1012 
1013 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1014  !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1015 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1016 {
1017  return 0;
1018 }
1019 #endif
1020 
1021 #ifdef CONFIG_FLATMEM
1022 #define pfn_to_nid(pfn) (0)
1023 #endif
1024 
1025 #ifdef CONFIG_SPARSEMEM
1026 
1027 /*
1028  * SECTION_SHIFT #bits space required to store a section #
1029  *
1030  * PA_SECTION_SHIFT physical address to/from section number
1031  * PFN_SECTION_SHIFT pfn to/from section number
1032  */
1033 #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
1034 
1035 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1036 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1037 
1038 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1039 
1040 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1041 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1042 
1043 #define SECTION_BLOCKFLAGS_BITS \
1044  ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1045 
1046 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1047 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1048 #endif
1049 
1050 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
1051 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
1052 
1053 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1054 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1055 
1056 struct page;
1057 struct page_cgroup;
1058 struct mem_section {
1059  /*
1060  * This is, logically, a pointer to an array of struct
1061  * pages. However, it is stored with some other magic.
1062  * (see sparse.c::sparse_init_one_section())
1063  *
1064  * Additionally during early boot we encode node id of
1065  * the location of the section here to guide allocation.
1066  * (see sparse.c::memory_present())
1067  *
1068  * Making it a UL at least makes someone do a cast
1069  * before using it wrong.
1070  */
1071  unsigned long section_mem_map;
1072 
1073  /* See declaration of similar field in struct zone */
1074  unsigned long *pageblock_flags;
1075 #ifdef CONFIG_MEMCG
1076  /*
1077  * If !SPARSEMEM, pgdat doesn't have page_cgroup pointer. We use
1078  * section. (see memcontrol.h/page_cgroup.h about this.)
1079  */
1080  struct page_cgroup *page_cgroup;
1081  unsigned long pad;
1082 #endif
1083 };
1084 
1085 #ifdef CONFIG_SPARSEMEM_EXTREME
1086 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1087 #else
1088 #define SECTIONS_PER_ROOT 1
1089 #endif
1090 
1091 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1092 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1093 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1094 
1095 #ifdef CONFIG_SPARSEMEM_EXTREME
1096 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
1097 #else
1098 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1099 #endif
1100 
1101 static inline struct mem_section *__nr_to_section(unsigned long nr)
1102 {
1103  if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1104  return NULL;
1105  return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1106 }
1107 extern int __section_nr(struct mem_section* ms);
1108 extern unsigned long usemap_size(void);
1109 
1110 /*
1111  * We use the lower bits of the mem_map pointer to store
1112  * a little bit of information. There should be at least
1113  * 3 bits here due to 32-bit alignment.
1114  */
1115 #define SECTION_MARKED_PRESENT (1UL<<0)
1116 #define SECTION_HAS_MEM_MAP (1UL<<1)
1117 #define SECTION_MAP_LAST_BIT (1UL<<2)
1118 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1119 #define SECTION_NID_SHIFT 2
1120 
1121 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1122 {
1123  unsigned long map = section->section_mem_map;
1124  map &= SECTION_MAP_MASK;
1125  return (struct page *)map;
1126 }
1127 
1128 static inline int present_section(struct mem_section *section)
1129 {
1130  return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1131 }
1132 
1133 static inline int present_section_nr(unsigned long nr)
1134 {
1135  return present_section(__nr_to_section(nr));
1136 }
1137 
1138 static inline int valid_section(struct mem_section *section)
1139 {
1140  return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1141 }
1142 
1143 static inline int valid_section_nr(unsigned long nr)
1144 {
1145  return valid_section(__nr_to_section(nr));
1146 }
1147 
1148 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1149 {
1150  return __nr_to_section(pfn_to_section_nr(pfn));
1151 }
1152 
1153 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1154 static inline int pfn_valid(unsigned long pfn)
1155 {
1156  if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1157  return 0;
1158  return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
1159 }
1160 #endif
1161 
1162 static inline int pfn_present(unsigned long pfn)
1163 {
1164  if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1165  return 0;
1166  return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1167 }
1168 
1169 /*
1170  * These are _only_ used during initialisation, therefore they
1171  * can use __initdata ... They could have names to indicate
1172  * this restriction.
1173  */
1174 #ifdef CONFIG_NUMA
1175 #define pfn_to_nid(pfn) \
1176 ({ \
1177  unsigned long __pfn_to_nid_pfn = (pfn); \
1178  page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1179 })
1180 #else
1181 #define pfn_to_nid(pfn) (0)
1182 #endif
1183 
1184 #define early_pfn_valid(pfn) pfn_valid(pfn)
1185 void sparse_init(void);
1186 #else
1187 #define sparse_init() do {} while (0)
1188 #define sparse_index_init(_sec, _nid) do {} while (0)
1189 #endif /* CONFIG_SPARSEMEM */
1190 
1191 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1192 bool early_pfn_in_nid(unsigned long pfn, int nid);
1193 #else
1194 #define early_pfn_in_nid(pfn, nid) (1)
1195 #endif
1196 
1197 #ifndef early_pfn_valid
1198 #define early_pfn_valid(pfn) (1)
1199 #endif
1200 
1201 void memory_present(int nid, unsigned long start, unsigned long end);
1202 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
1203 
1204 /*
1205  * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1206  * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1207  * pfn_valid_within() should be used in this case; we optimise this away
1208  * when we have no holes within a MAX_ORDER_NR_PAGES block.
1209  */
1210 #ifdef CONFIG_HOLES_IN_ZONE
1211 #define pfn_valid_within(pfn) pfn_valid(pfn)
1212 #else
1213 #define pfn_valid_within(pfn) (1)
1214 #endif
1215 
1216 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1217 /*
1218  * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1219  * associated with it or not. In FLATMEM, it is expected that holes always
1220  * have valid memmap as long as there is valid PFNs either side of the hole.
1221  * In SPARSEMEM, it is assumed that a valid section has a memmap for the
1222  * entire section.
1223  *
1224  * However, an ARM, and maybe other embedded architectures in the future
1225  * free memmap backing holes to save memory on the assumption the memmap is
1226  * never used. The page_zone linkages are then broken even though pfn_valid()
1227  * returns true. A walker of the full memmap must then do this additional
1228  * check to ensure the memmap they are looking at is sane by making sure
1229  * the zone and PFN linkages are still valid. This is expensive, but walkers
1230  * of the full memmap are extremely rare.
1231  */
1232 int memmap_valid_within(unsigned long pfn,
1233  struct page *page, struct zone *zone);
1234 #else
1235 static inline int memmap_valid_within(unsigned long pfn,
1236  struct page *page, struct zone *zone)
1237 {
1238  return 1;
1239 }
1240 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1241 
1242 #endif /* !__GENERATING_BOUNDS.H */
1243 #endif /* !__ASSEMBLY__ */
1244 #endif /* _LINUX_MMZONE_H */