1 #ifndef _LINUX_MMZONE_H
2 #define _LINUX_MMZONE_H
5 #ifndef __GENERATING_BOUNDS_H
7 #include <linux/spinlock.h>
8 #include <linux/list.h>
9 #include <linux/wait.h>
10 #include <linux/bitops.h>
11 #include <linux/cache.h>
12 #include <linux/threads.h>
13 #include <linux/numa.h>
14 #include <linux/init.h>
15 #include <linux/seqlock.h>
16 #include <linux/nodemask.h>
17 #include <linux/pageblock-flags.h>
18 #include <linux/page-flags-layout.h>
19 #include <linux/atomic.h>
22 /* Free memory management - zoned buddy allocator. */
23 #ifndef CONFIG_FORCE_MAX_ZONEORDER
26 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
28 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
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
36 #define PAGE_ALLOC_COSTLY_ORDER 3
42 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
43 MIGRATE_RESERVE = MIGRATE_PCPTYPES,
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.
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
60 #ifdef CONFIG_MEMORY_ISOLATION
61 MIGRATE_ISOLATE, /* can't allocate from here */
67 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
69 # define is_migrate_cma(migratetype) false
72 #define for_each_migratetype_order(order, type) \
73 for (order = 0; order < MAX_ORDER; order++) \
74 for (type = 0; type < MIGRATE_TYPES; type++)
76 extern int page_group_by_mobility_disabled;
78 static inline int get_pageblock_migratetype(struct page *page)
80 return get_pageblock_flags_group(page, PB_migrate, PB_migrate_end);
84 struct list_head free_list[MIGRATE_TYPES];
85 unsigned long nr_free;
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.
96 #if defined(CONFIG_SMP)
99 } ____cacheline_internodealigned_in_smp;
100 #define ZONE_PADDING(name) struct zone_padding name;
102 #define ZONE_PADDING(name)
105 enum zone_stat_item {
106 /* First 128 byte cacheline (assuming 64 bit words) */
110 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
111 NR_ACTIVE_ANON, /* " " " " " */
112 NR_INACTIVE_FILE, /* " " " " " */
113 NR_ACTIVE_FILE, /* " " " " " */
114 NR_UNEVICTABLE, /* " " " " " */
115 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
116 NR_ANON_PAGES, /* Mapped anonymous pages */
117 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
118 only modified from process context */
123 NR_SLAB_UNRECLAIMABLE,
124 NR_PAGETABLE, /* used for pagetables */
126 /* Second 128 byte cacheline */
127 NR_UNSTABLE_NFS, /* NFS unstable pages */
130 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
131 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
132 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
133 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
134 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
135 NR_DIRTIED, /* page dirtyings since bootup */
136 NR_WRITTEN, /* page writings since bootup */
138 NUMA_HIT, /* allocated in intended node */
139 NUMA_MISS, /* allocated in non intended node */
140 NUMA_FOREIGN, /* was intended here, hit elsewhere */
141 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
142 NUMA_LOCAL, /* allocation from local node */
143 NUMA_OTHER, /* allocation from other node */
147 NR_ANON_TRANSPARENT_HUGEPAGES,
149 NR_VM_ZONE_STAT_ITEMS };
152 * We do arithmetic on the LRU lists in various places in the code,
153 * so it is important to keep the active lists LRU_ACTIVE higher in
154 * the array than the corresponding inactive lists, and to keep
155 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
157 * This has to be kept in sync with the statistics in zone_stat_item
158 * above and the descriptions in vmstat_text in mm/vmstat.c
165 LRU_INACTIVE_ANON = LRU_BASE,
166 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
167 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
168 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
173 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
175 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
177 static inline int is_file_lru(enum lru_list lru)
179 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
182 static inline int is_active_lru(enum lru_list lru)
184 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
187 static inline int is_unevictable_lru(enum lru_list lru)
189 return (lru == LRU_UNEVICTABLE);
192 struct zone_reclaim_stat {
194 * The pageout code in vmscan.c keeps track of how many of the
195 * mem/swap backed and file backed pages are referenced.
196 * The higher the rotated/scanned ratio, the more valuable
199 * The anon LRU stats live in [0], file LRU stats in [1]
201 unsigned long recent_rotated[2];
202 unsigned long recent_scanned[2];
206 struct list_head lists[NR_LRU_LISTS];
207 struct zone_reclaim_stat reclaim_stat;
213 /* Mask used at gathering information at once (see memcontrol.c) */
214 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
215 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
216 #define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
218 /* Isolate clean file */
219 #define ISOLATE_CLEAN ((__force isolate_mode_t)0x1)
220 /* Isolate unmapped file */
221 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
222 /* Isolate for asynchronous migration */
223 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
224 /* Isolate unevictable pages */
225 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
227 /* LRU Isolation modes. */
228 typedef unsigned __bitwise__ isolate_mode_t;
230 enum zone_watermarks {
237 #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
238 #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
239 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
241 struct per_cpu_pages {
242 int count; /* number of pages in the list */
243 int high; /* high watermark, emptying needed */
244 int batch; /* chunk size for buddy add/remove */
246 /* Lists of pages, one per migrate type stored on the pcp-lists */
247 struct list_head lists[MIGRATE_PCPTYPES];
250 struct per_cpu_pageset {
251 struct per_cpu_pages pcp;
257 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
261 #endif /* !__GENERATING_BOUNDS.H */
264 #ifdef CONFIG_ZONE_DMA
266 * ZONE_DMA is used when there are devices that are not able
267 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
268 * carve out the portion of memory that is needed for these devices.
269 * The range is arch specific.
274 * ---------------------------
275 * parisc, ia64, sparc <4G
278 * alpha Unlimited or 0-16MB.
280 * i386, x86_64 and multiple other arches
285 #ifdef CONFIG_ZONE_DMA32
287 * x86_64 needs two ZONE_DMAs because it supports devices that are
288 * only able to do DMA to the lower 16M but also 32 bit devices that
289 * can only do DMA areas below 4G.
294 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
295 * performed on pages in ZONE_NORMAL if the DMA devices support
296 * transfers to all addressable memory.
299 #ifdef CONFIG_HIGHMEM
301 * A memory area that is only addressable by the kernel through
302 * mapping portions into its own address space. This is for example
303 * used by i386 to allow the kernel to address the memory beyond
304 * 900MB. The kernel will set up special mappings (page
305 * table entries on i386) for each page that the kernel needs to
314 #ifndef __GENERATING_BOUNDS_H
317 /* Fields commonly accessed by the page allocator */
319 /* zone watermarks, access with *_wmark_pages(zone) macros */
320 unsigned long watermark[NR_WMARK];
323 * When free pages are below this point, additional steps are taken
324 * when reading the number of free pages to avoid per-cpu counter
325 * drift allowing watermarks to be breached
327 unsigned long percpu_drift_mark;
330 * We don't know if the memory that we're going to allocate will be freeable
331 * or/and it will be released eventually, so to avoid totally wasting several
332 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
333 * to run OOM on the lower zones despite there's tons of freeable ram
334 * on the higher zones). This array is recalculated at runtime if the
335 * sysctl_lowmem_reserve_ratio sysctl changes.
337 unsigned long lowmem_reserve[MAX_NR_ZONES];
340 * This is a per-zone reserve of pages that should not be
341 * considered dirtyable memory.
343 unsigned long dirty_balance_reserve;
348 * zone reclaim becomes active if more unmapped pages exist.
350 unsigned long min_unmapped_pages;
351 unsigned long min_slab_pages;
353 struct per_cpu_pageset __percpu *pageset;
355 * free areas of different sizes
358 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
359 /* Set to true when the PG_migrate_skip bits should be cleared */
360 bool compact_blockskip_flush;
362 /* pfns where compaction scanners should start */
363 unsigned long compact_cached_free_pfn;
364 unsigned long compact_cached_migrate_pfn;
366 #ifdef CONFIG_MEMORY_HOTPLUG
367 /* see spanned/present_pages for more description */
368 seqlock_t span_seqlock;
370 struct free_area free_area[MAX_ORDER];
372 #ifndef CONFIG_SPARSEMEM
374 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
375 * In SPARSEMEM, this map is stored in struct mem_section
377 unsigned long *pageblock_flags;
378 #endif /* CONFIG_SPARSEMEM */
380 #ifdef CONFIG_COMPACTION
382 * On compaction failure, 1<<compact_defer_shift compactions
383 * are skipped before trying again. The number attempted since
384 * last failure is tracked with compact_considered.
386 unsigned int compact_considered;
387 unsigned int compact_defer_shift;
388 int compact_order_failed;
393 /* Fields commonly accessed by the page reclaim scanner */
395 struct lruvec lruvec;
397 /* Evictions & activations on the inactive file list */
398 atomic_long_t inactive_age;
400 unsigned long pages_scanned; /* since last reclaim */
401 unsigned long flags; /* zone flags, see below */
403 /* Zone statistics */
404 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
407 * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on
408 * this zone's LRU. Maintained by the pageout code.
410 unsigned int inactive_ratio;
414 /* Rarely used or read-mostly fields */
417 * wait_table -- the array holding the hash table
418 * wait_table_hash_nr_entries -- the size of the hash table array
419 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
421 * The purpose of all these is to keep track of the people
422 * waiting for a page to become available and make them
423 * runnable again when possible. The trouble is that this
424 * consumes a lot of space, especially when so few things
425 * wait on pages at a given time. So instead of using
426 * per-page waitqueues, we use a waitqueue hash table.
428 * The bucket discipline is to sleep on the same queue when
429 * colliding and wake all in that wait queue when removing.
430 * When something wakes, it must check to be sure its page is
431 * truly available, a la thundering herd. The cost of a
432 * collision is great, but given the expected load of the
433 * table, they should be so rare as to be outweighed by the
434 * benefits from the saved space.
436 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
437 * primary users of these fields, and in mm/page_alloc.c
438 * free_area_init_core() performs the initialization of them.
440 wait_queue_head_t * wait_table;
441 unsigned long wait_table_hash_nr_entries;
442 unsigned long wait_table_bits;
445 * Discontig memory support fields.
447 struct pglist_data *zone_pgdat;
448 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
449 unsigned long zone_start_pfn;
452 * spanned_pages is the total pages spanned by the zone, including
453 * holes, which is calculated as:
454 * spanned_pages = zone_end_pfn - zone_start_pfn;
456 * present_pages is physical pages existing within the zone, which
458 * present_pages = spanned_pages - absent_pages(pages in holes);
460 * managed_pages is present pages managed by the buddy system, which
461 * is calculated as (reserved_pages includes pages allocated by the
462 * bootmem allocator):
463 * managed_pages = present_pages - reserved_pages;
465 * So present_pages may be used by memory hotplug or memory power
466 * management logic to figure out unmanaged pages by checking
467 * (present_pages - managed_pages). And managed_pages should be used
468 * by page allocator and vm scanner to calculate all kinds of watermarks
473 * zone_start_pfn and spanned_pages are protected by span_seqlock.
474 * It is a seqlock because it has to be read outside of zone->lock,
475 * and it is done in the main allocator path. But, it is written
476 * quite infrequently.
478 * The span_seq lock is declared along with zone->lock because it is
479 * frequently read in proximity to zone->lock. It's good to
480 * give them a chance of being in the same cacheline.
482 * Write access to present_pages at runtime should be protected by
483 * lock_memory_hotplug()/unlock_memory_hotplug(). Any reader who can't
484 * tolerant drift of present_pages should hold memory hotplug lock to
485 * get a stable value.
487 * Read access to managed_pages should be safe because it's unsigned
488 * long. Write access to zone->managed_pages and totalram_pages are
489 * protected by managed_page_count_lock at runtime. Idealy only
490 * adjust_managed_page_count() should be used instead of directly
491 * touching zone->managed_pages and totalram_pages.
493 unsigned long spanned_pages;
494 unsigned long present_pages;
495 unsigned long managed_pages;
498 * Number of MIGRATE_RESEVE page block. To maintain for just
499 * optimization. Protected by zone->lock.
501 int nr_migrate_reserve_block;
504 * rarely used fields:
507 } ____cacheline_internodealigned_in_smp;
510 ZONE_RECLAIM_LOCKED, /* prevents concurrent reclaim */
511 ZONE_OOM_LOCKED, /* zone is in OOM killer zonelist */
512 ZONE_CONGESTED, /* zone has many dirty pages backed by
515 ZONE_TAIL_LRU_DIRTY, /* reclaim scanning has recently found
516 * many dirty file pages at the tail
519 ZONE_WRITEBACK, /* reclaim scanning has recently found
520 * many pages under writeback
524 static inline void zone_set_flag(struct zone *zone, zone_flags_t flag)
526 set_bit(flag, &zone->flags);
529 static inline int zone_test_and_set_flag(struct zone *zone, zone_flags_t flag)
531 return test_and_set_bit(flag, &zone->flags);
534 static inline void zone_clear_flag(struct zone *zone, zone_flags_t flag)
536 clear_bit(flag, &zone->flags);
539 static inline int zone_is_reclaim_congested(const struct zone *zone)
541 return test_bit(ZONE_CONGESTED, &zone->flags);
544 static inline int zone_is_reclaim_dirty(const struct zone *zone)
546 return test_bit(ZONE_TAIL_LRU_DIRTY, &zone->flags);
549 static inline int zone_is_reclaim_writeback(const struct zone *zone)
551 return test_bit(ZONE_WRITEBACK, &zone->flags);
554 static inline int zone_is_reclaim_locked(const struct zone *zone)
556 return test_bit(ZONE_RECLAIM_LOCKED, &zone->flags);
559 static inline int zone_is_oom_locked(const struct zone *zone)
561 return test_bit(ZONE_OOM_LOCKED, &zone->flags);
564 static inline unsigned long zone_end_pfn(const struct zone *zone)
566 return zone->zone_start_pfn + zone->spanned_pages;
569 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
571 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
574 static inline bool zone_is_initialized(struct zone *zone)
576 return !!zone->wait_table;
579 static inline bool zone_is_empty(struct zone *zone)
581 return zone->spanned_pages == 0;
585 * The "priority" of VM scanning is how much of the queues we will scan in one
586 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
587 * queues ("queue_length >> 12") during an aging round.
589 #define DEF_PRIORITY 12
591 /* Maximum number of zones on a zonelist */
592 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
597 * The NUMA zonelists are doubled because we need zonelists that restrict the
598 * allocations to a single node for __GFP_THISNODE.
600 * [0] : Zonelist with fallback
601 * [1] : No fallback (__GFP_THISNODE)
603 #define MAX_ZONELISTS 2
607 * We cache key information from each zonelist for smaller cache
608 * footprint when scanning for free pages in get_page_from_freelist().
610 * 1) The BITMAP fullzones tracks which zones in a zonelist have come
611 * up short of free memory since the last time (last_fullzone_zap)
612 * we zero'd fullzones.
613 * 2) The array z_to_n[] maps each zone in the zonelist to its node
614 * id, so that we can efficiently evaluate whether that node is
615 * set in the current tasks mems_allowed.
617 * Both fullzones and z_to_n[] are one-to-one with the zonelist,
618 * indexed by a zones offset in the zonelist zones[] array.
620 * The get_page_from_freelist() routine does two scans. During the
621 * first scan, we skip zones whose corresponding bit in 'fullzones'
622 * is set or whose corresponding node in current->mems_allowed (which
623 * comes from cpusets) is not set. During the second scan, we bypass
624 * this zonelist_cache, to ensure we look methodically at each zone.
626 * Once per second, we zero out (zap) fullzones, forcing us to
627 * reconsider nodes that might have regained more free memory.
628 * The field last_full_zap is the time we last zapped fullzones.
630 * This mechanism reduces the amount of time we waste repeatedly
631 * reexaming zones for free memory when they just came up low on
632 * memory momentarilly ago.
634 * The zonelist_cache struct members logically belong in struct
635 * zonelist. However, the mempolicy zonelists constructed for
636 * MPOL_BIND are intentionally variable length (and usually much
637 * shorter). A general purpose mechanism for handling structs with
638 * multiple variable length members is more mechanism than we want
639 * here. We resort to some special case hackery instead.
641 * The MPOL_BIND zonelists don't need this zonelist_cache (in good
642 * part because they are shorter), so we put the fixed length stuff
643 * at the front of the zonelist struct, ending in a variable length
644 * zones[], as is needed by MPOL_BIND.
646 * Then we put the optional zonelist cache on the end of the zonelist
647 * struct. This optional stuff is found by a 'zlcache_ptr' pointer in
648 * the fixed length portion at the front of the struct. This pointer
649 * both enables us to find the zonelist cache, and in the case of
650 * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL)
651 * to know that the zonelist cache is not there.
653 * The end result is that struct zonelists come in two flavors:
654 * 1) The full, fixed length version, shown below, and
655 * 2) The custom zonelists for MPOL_BIND.
656 * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache.
658 * Even though there may be multiple CPU cores on a node modifying
659 * fullzones or last_full_zap in the same zonelist_cache at the same
660 * time, we don't lock it. This is just hint data - if it is wrong now
661 * and then, the allocator will still function, perhaps a bit slower.
665 struct zonelist_cache {
666 unsigned short z_to_n[MAX_ZONES_PER_ZONELIST]; /* zone->nid */
667 DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST); /* zone full? */
668 unsigned long last_full_zap; /* when last zap'd (jiffies) */
671 #define MAX_ZONELISTS 1
672 struct zonelist_cache;
676 * This struct contains information about a zone in a zonelist. It is stored
677 * here to avoid dereferences into large structures and lookups of tables
680 struct zone *zone; /* Pointer to actual zone */
681 int zone_idx; /* zone_idx(zoneref->zone) */
685 * One allocation request operates on a zonelist. A zonelist
686 * is a list of zones, the first one is the 'goal' of the
687 * allocation, the other zones are fallback zones, in decreasing
690 * If zlcache_ptr is not NULL, then it is just the address of zlcache,
691 * as explained above. If zlcache_ptr is NULL, there is no zlcache.
693 * To speed the reading of the zonelist, the zonerefs contain the zone index
694 * of the entry being read. Helper functions to access information given
695 * a struct zoneref are
697 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
698 * zonelist_zone_idx() - Return the index of the zone for an entry
699 * zonelist_node_idx() - Return the index of the node for an entry
702 struct zonelist_cache *zlcache_ptr; // NULL or &zlcache
703 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
705 struct zonelist_cache zlcache; // optional ...
709 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
710 struct node_active_region {
711 unsigned long start_pfn;
712 unsigned long end_pfn;
715 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
717 #ifndef CONFIG_DISCONTIGMEM
718 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
719 extern struct page *mem_map;
723 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
724 * (mostly NUMA machines?) to denote a higher-level memory zone than the
727 * On NUMA machines, each NUMA node would have a pg_data_t to describe
728 * it's memory layout.
730 * Memory statistics and page replacement data structures are maintained on a
734 typedef struct pglist_data {
735 struct zone node_zones[MAX_NR_ZONES];
736 struct zonelist node_zonelists[MAX_ZONELISTS];
738 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
739 struct page *node_mem_map;
741 struct page_cgroup *node_page_cgroup;
744 #ifndef CONFIG_NO_BOOTMEM
745 struct bootmem_data *bdata;
747 #ifdef CONFIG_MEMORY_HOTPLUG
749 * Must be held any time you expect node_start_pfn, node_present_pages
750 * or node_spanned_pages stay constant. Holding this will also
751 * guarantee that any pfn_valid() stays that way.
753 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
754 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG.
756 * Nests above zone->lock and zone->span_seqlock
758 spinlock_t node_size_lock;
760 unsigned long node_start_pfn;
761 unsigned long node_present_pages; /* total number of physical pages */
762 unsigned long node_spanned_pages; /* total size of physical page
763 range, including holes */
765 nodemask_t reclaim_nodes; /* Nodes allowed to reclaim from */
766 wait_queue_head_t kswapd_wait;
767 wait_queue_head_t pfmemalloc_wait;
768 struct task_struct *kswapd; /* Protected by lock_memory_hotplug() */
769 int kswapd_max_order;
770 enum zone_type classzone_idx;
771 #ifdef CONFIG_NUMA_BALANCING
772 /* Lock serializing the migrate rate limiting window */
773 spinlock_t numabalancing_migrate_lock;
775 /* Rate limiting time interval */
776 unsigned long numabalancing_migrate_next_window;
778 /* Number of pages migrated during the rate limiting time interval */
779 unsigned long numabalancing_migrate_nr_pages;
783 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
784 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
785 #ifdef CONFIG_FLAT_NODE_MEM_MAP
786 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
788 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
790 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
792 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
793 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
795 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
797 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
800 static inline bool pgdat_is_empty(pg_data_t *pgdat)
802 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
805 #include <linux/memory_hotplug.h>
807 extern struct mutex zonelists_mutex;
808 void build_all_zonelists(pg_data_t *pgdat, struct zone *zone);
809 void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx);
810 bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
811 int classzone_idx, int alloc_flags);
812 bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
813 int classzone_idx, int alloc_flags);
814 enum memmap_context {
818 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
820 enum memmap_context context);
822 extern void lruvec_init(struct lruvec *lruvec);
824 static inline struct zone *lruvec_zone(struct lruvec *lruvec)
829 return container_of(lruvec, struct zone, lruvec);
833 #ifdef CONFIG_HAVE_MEMORY_PRESENT
834 void memory_present(int nid, unsigned long start, unsigned long end);
836 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
839 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
840 int local_memory_node(int node_id);
842 static inline int local_memory_node(int node_id) { return node_id; };
845 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
846 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
850 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
852 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
854 static inline int populated_zone(struct zone *zone)
856 return (!!zone->present_pages);
859 extern int movable_zone;
861 static inline int zone_movable_is_highmem(void)
863 #if defined(CONFIG_HIGHMEM) && defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
864 return movable_zone == ZONE_HIGHMEM;
870 static inline int is_highmem_idx(enum zone_type idx)
872 #ifdef CONFIG_HIGHMEM
873 return (idx == ZONE_HIGHMEM ||
874 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
881 * is_highmem - helper function to quickly check if a struct zone is a
882 * highmem zone or not. This is an attempt to keep references
883 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
884 * @zone - pointer to struct zone variable
886 static inline int is_highmem(struct zone *zone)
888 #ifdef CONFIG_HIGHMEM
889 int zone_off = (char *)zone - (char *)zone->zone_pgdat->node_zones;
890 return zone_off == ZONE_HIGHMEM * sizeof(*zone) ||
891 (zone_off == ZONE_MOVABLE * sizeof(*zone) &&
892 zone_movable_is_highmem());
898 /* These two functions are used to setup the per zone pages min values */
900 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
901 void __user *, size_t *, loff_t *);
902 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
903 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
904 void __user *, size_t *, loff_t *);
905 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
906 void __user *, size_t *, loff_t *);
907 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
908 void __user *, size_t *, loff_t *);
909 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
910 void __user *, size_t *, loff_t *);
912 extern int numa_zonelist_order_handler(struct ctl_table *, int,
913 void __user *, size_t *, loff_t *);
914 extern char numa_zonelist_order[];
915 #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
917 #ifndef CONFIG_NEED_MULTIPLE_NODES
919 extern struct pglist_data contig_page_data;
920 #define NODE_DATA(nid) (&contig_page_data)
921 #define NODE_MEM_MAP(nid) mem_map
923 #else /* CONFIG_NEED_MULTIPLE_NODES */
925 #include <asm/mmzone.h>
927 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
929 extern struct pglist_data *first_online_pgdat(void);
930 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
931 extern struct zone *next_zone(struct zone *zone);
934 * for_each_online_pgdat - helper macro to iterate over all online nodes
935 * @pgdat - pointer to a pg_data_t variable
937 #define for_each_online_pgdat(pgdat) \
938 for (pgdat = first_online_pgdat(); \
940 pgdat = next_online_pgdat(pgdat))
942 * for_each_zone - helper macro to iterate over all memory zones
943 * @zone - pointer to struct zone variable
945 * The user only needs to declare the zone variable, for_each_zone
948 #define for_each_zone(zone) \
949 for (zone = (first_online_pgdat())->node_zones; \
951 zone = next_zone(zone))
953 #define for_each_populated_zone(zone) \
954 for (zone = (first_online_pgdat())->node_zones; \
956 zone = next_zone(zone)) \
957 if (!populated_zone(zone)) \
961 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
963 return zoneref->zone;
966 static inline int zonelist_zone_idx(struct zoneref *zoneref)
968 return zoneref->zone_idx;
971 static inline int zonelist_node_idx(struct zoneref *zoneref)
974 /* zone_to_nid not available in this context */
975 return zoneref->zone->node;
978 #endif /* CONFIG_NUMA */
982 * 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
983 * @z - The cursor used as a starting point for the search
984 * @highest_zoneidx - The zone index of the highest zone to return
985 * @nodes - An optional nodemask to filter the zonelist with
986 * @zone - The first suitable zone found is returned via this parameter
988 * This function returns the next zone at or below a given zone index that is
989 * within the allowed nodemask using a cursor as the starting point for the
990 * search. The zoneref returned is a cursor that represents the current zone
991 * being examined. It should be advanced by one before calling
992 * next_zones_zonelist again.
994 struct zoneref *next_zones_zonelist(struct zoneref *z,
995 enum zone_type highest_zoneidx,
1000 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1001 * @zonelist - The zonelist to search for a suitable zone
1002 * @highest_zoneidx - The zone index of the highest zone to return
1003 * @nodes - An optional nodemask to filter the zonelist with
1004 * @zone - The first suitable zone found is returned via this parameter
1006 * This function returns the first zone at or below a given zone index that is
1007 * within the allowed nodemask. The zoneref returned is a cursor that can be
1008 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1009 * one before calling.
1011 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1012 enum zone_type highest_zoneidx,
1016 return next_zones_zonelist(zonelist->_zonerefs, highest_zoneidx, nodes,
1021 * 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
1022 * @zone - The current zone in the iterator
1023 * @z - The current pointer within zonelist->zones being iterated
1024 * @zlist - The zonelist being iterated
1025 * @highidx - The zone index of the highest zone to return
1026 * @nodemask - Nodemask allowed by the allocator
1028 * This iterator iterates though all zones at or below a given zone index and
1029 * within a given nodemask
1031 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1032 for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \
1034 z = next_zones_zonelist(++z, highidx, nodemask, &zone)) \
1037 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1038 * @zone - The current zone in the iterator
1039 * @z - The current pointer within zonelist->zones being iterated
1040 * @zlist - The zonelist being iterated
1041 * @highidx - The zone index of the highest zone to return
1043 * This iterator iterates though all zones at or below a given zone index.
1045 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1046 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1048 #ifdef CONFIG_SPARSEMEM
1049 #include <asm/sparsemem.h>
1052 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1053 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1054 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1060 #ifdef CONFIG_FLATMEM
1061 #define pfn_to_nid(pfn) (0)
1064 #ifdef CONFIG_SPARSEMEM
1067 * SECTION_SHIFT #bits space required to store a section #
1069 * PA_SECTION_SHIFT physical address to/from section number
1070 * PFN_SECTION_SHIFT pfn to/from section number
1072 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1073 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1075 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1077 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1078 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1080 #define SECTION_BLOCKFLAGS_BITS \
1081 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1083 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1084 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1087 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
1088 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
1090 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1091 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1095 struct mem_section {
1097 * This is, logically, a pointer to an array of struct
1098 * pages. However, it is stored with some other magic.
1099 * (see sparse.c::sparse_init_one_section())
1101 * Additionally during early boot we encode node id of
1102 * the location of the section here to guide allocation.
1103 * (see sparse.c::memory_present())
1105 * Making it a UL at least makes someone do a cast
1106 * before using it wrong.
1108 unsigned long section_mem_map;
1110 /* See declaration of similar field in struct zone */
1111 unsigned long *pageblock_flags;
1114 * If !SPARSEMEM, pgdat doesn't have page_cgroup pointer. We use
1115 * section. (see memcontrol.h/page_cgroup.h about this.)
1117 struct page_cgroup *page_cgroup;
1121 * WARNING: mem_section must be a power-of-2 in size for the
1122 * calculation and use of SECTION_ROOT_MASK to make sense.
1126 #ifdef CONFIG_SPARSEMEM_EXTREME
1127 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1129 #define SECTIONS_PER_ROOT 1
1132 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1133 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1134 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1136 #ifdef CONFIG_SPARSEMEM_EXTREME
1137 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
1139 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1142 static inline struct mem_section *__nr_to_section(unsigned long nr)
1144 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1146 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1148 extern int __section_nr(struct mem_section* ms);
1149 extern unsigned long usemap_size(void);
1152 * We use the lower bits of the mem_map pointer to store
1153 * a little bit of information. There should be at least
1154 * 3 bits here due to 32-bit alignment.
1156 #define SECTION_MARKED_PRESENT (1UL<<0)
1157 #define SECTION_HAS_MEM_MAP (1UL<<1)
1158 #define SECTION_MAP_LAST_BIT (1UL<<2)
1159 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1160 #define SECTION_NID_SHIFT 2
1162 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1164 unsigned long map = section->section_mem_map;
1165 map &= SECTION_MAP_MASK;
1166 return (struct page *)map;
1169 static inline int present_section(struct mem_section *section)
1171 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1174 static inline int present_section_nr(unsigned long nr)
1176 return present_section(__nr_to_section(nr));
1179 static inline int valid_section(struct mem_section *section)
1181 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1184 static inline int valid_section_nr(unsigned long nr)
1186 return valid_section(__nr_to_section(nr));
1189 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1191 return __nr_to_section(pfn_to_section_nr(pfn));
1194 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1195 static inline int pfn_valid(unsigned long pfn)
1197 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1199 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
1203 static inline int pfn_present(unsigned long pfn)
1205 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1207 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1211 * These are _only_ used during initialisation, therefore they
1212 * can use __initdata ... They could have names to indicate
1216 #define pfn_to_nid(pfn) \
1218 unsigned long __pfn_to_nid_pfn = (pfn); \
1219 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1222 #define pfn_to_nid(pfn) (0)
1225 #define early_pfn_valid(pfn) pfn_valid(pfn)
1226 void sparse_init(void);
1228 #define sparse_init() do {} while (0)
1229 #define sparse_index_init(_sec, _nid) do {} while (0)
1230 #endif /* CONFIG_SPARSEMEM */
1232 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1233 bool early_pfn_in_nid(unsigned long pfn, int nid);
1235 #define early_pfn_in_nid(pfn, nid) (1)
1238 #ifndef early_pfn_valid
1239 #define early_pfn_valid(pfn) (1)
1242 void memory_present(int nid, unsigned long start, unsigned long end);
1243 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
1246 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1247 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1248 * pfn_valid_within() should be used in this case; we optimise this away
1249 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1251 #ifdef CONFIG_HOLES_IN_ZONE
1252 #define pfn_valid_within(pfn) pfn_valid(pfn)
1254 #define pfn_valid_within(pfn) (1)
1257 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1259 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1260 * associated with it or not. In FLATMEM, it is expected that holes always
1261 * have valid memmap as long as there is valid PFNs either side of the hole.
1262 * In SPARSEMEM, it is assumed that a valid section has a memmap for the
1265 * However, an ARM, and maybe other embedded architectures in the future
1266 * free memmap backing holes to save memory on the assumption the memmap is
1267 * never used. The page_zone linkages are then broken even though pfn_valid()
1268 * returns true. A walker of the full memmap must then do this additional
1269 * check to ensure the memmap they are looking at is sane by making sure
1270 * the zone and PFN linkages are still valid. This is expensive, but walkers
1271 * of the full memmap are extremely rare.
1273 int memmap_valid_within(unsigned long pfn,
1274 struct page *page, struct zone *zone);
1276 static inline int memmap_valid_within(unsigned long pfn,
1277 struct page *page, struct zone *zone)
1281 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1283 #endif /* !__GENERATING_BOUNDS.H */
1284 #endif /* !__ASSEMBLY__ */
1285 #endif /* _LINUX_MMZONE_H */