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 */
145 NR_ANON_TRANSPARENT_HUGEPAGES,
147 NR_VM_ZONE_STAT_ITEMS };
150 * We do arithmetic on the LRU lists in various places in the code,
151 * so it is important to keep the active lists LRU_ACTIVE higher in
152 * the array than the corresponding inactive lists, and to keep
153 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
155 * This has to be kept in sync with the statistics in zone_stat_item
156 * above and the descriptions in vmstat_text in mm/vmstat.c
163 LRU_INACTIVE_ANON = LRU_BASE,
164 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
165 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
166 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
171 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
173 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
175 static inline int is_file_lru(enum lru_list lru)
177 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
180 static inline int is_active_lru(enum lru_list lru)
182 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
185 static inline int is_unevictable_lru(enum lru_list lru)
187 return (lru == LRU_UNEVICTABLE);
190 struct zone_reclaim_stat {
192 * The pageout code in vmscan.c keeps track of how many of the
193 * mem/swap backed and file backed pages are referenced.
194 * The higher the rotated/scanned ratio, the more valuable
197 * The anon LRU stats live in [0], file LRU stats in [1]
199 unsigned long recent_rotated[2];
200 unsigned long recent_scanned[2];
204 struct list_head lists[NR_LRU_LISTS];
205 struct zone_reclaim_stat reclaim_stat;
211 /* Mask used at gathering information at once (see memcontrol.c) */
212 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
213 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
214 #define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
216 /* Isolate clean file */
217 #define ISOLATE_CLEAN ((__force isolate_mode_t)0x1)
218 /* Isolate unmapped file */
219 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
220 /* Isolate for asynchronous migration */
221 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
222 /* Isolate unevictable pages */
223 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
225 /* LRU Isolation modes. */
226 typedef unsigned __bitwise__ isolate_mode_t;
228 enum zone_watermarks {
235 #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
236 #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
237 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
239 struct per_cpu_pages {
240 int count; /* number of pages in the list */
241 int high; /* high watermark, emptying needed */
242 int batch; /* chunk size for buddy add/remove */
244 /* Lists of pages, one per migrate type stored on the pcp-lists */
245 struct list_head lists[MIGRATE_PCPTYPES];
248 struct per_cpu_pageset {
249 struct per_cpu_pages pcp;
255 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
259 #endif /* !__GENERATING_BOUNDS.H */
262 #ifdef CONFIG_ZONE_DMA
264 * ZONE_DMA is used when there are devices that are not able
265 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
266 * carve out the portion of memory that is needed for these devices.
267 * The range is arch specific.
272 * ---------------------------
273 * parisc, ia64, sparc <4G
276 * alpha Unlimited or 0-16MB.
278 * i386, x86_64 and multiple other arches
283 #ifdef CONFIG_ZONE_DMA32
285 * x86_64 needs two ZONE_DMAs because it supports devices that are
286 * only able to do DMA to the lower 16M but also 32 bit devices that
287 * can only do DMA areas below 4G.
292 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
293 * performed on pages in ZONE_NORMAL if the DMA devices support
294 * transfers to all addressable memory.
297 #ifdef CONFIG_HIGHMEM
299 * A memory area that is only addressable by the kernel through
300 * mapping portions into its own address space. This is for example
301 * used by i386 to allow the kernel to address the memory beyond
302 * 900MB. The kernel will set up special mappings (page
303 * table entries on i386) for each page that the kernel needs to
312 #ifndef __GENERATING_BOUNDS_H
315 /* Fields commonly accessed by the page allocator */
317 /* zone watermarks, access with *_wmark_pages(zone) macros */
318 unsigned long watermark[NR_WMARK];
321 * When free pages are below this point, additional steps are taken
322 * when reading the number of free pages to avoid per-cpu counter
323 * drift allowing watermarks to be breached
325 unsigned long percpu_drift_mark;
328 * We don't know if the memory that we're going to allocate will be freeable
329 * or/and it will be released eventually, so to avoid totally wasting several
330 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
331 * to run OOM on the lower zones despite there's tons of freeable ram
332 * on the higher zones). This array is recalculated at runtime if the
333 * sysctl_lowmem_reserve_ratio sysctl changes.
335 unsigned long lowmem_reserve[MAX_NR_ZONES];
338 * This is a per-zone reserve of pages that should not be
339 * considered dirtyable memory.
341 unsigned long dirty_balance_reserve;
346 * zone reclaim becomes active if more unmapped pages exist.
348 unsigned long min_unmapped_pages;
349 unsigned long min_slab_pages;
351 struct per_cpu_pageset __percpu *pageset;
353 * free areas of different sizes
356 int all_unreclaimable; /* All pages pinned */
357 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
358 /* Set to true when the PG_migrate_skip bits should be cleared */
359 bool compact_blockskip_flush;
361 /* pfns where compaction scanners should start */
362 unsigned long compact_cached_free_pfn;
363 unsigned long compact_cached_migrate_pfn;
365 #ifdef CONFIG_MEMORY_HOTPLUG
366 /* see spanned/present_pages for more description */
367 seqlock_t span_seqlock;
369 struct free_area free_area[MAX_ORDER];
371 #ifndef CONFIG_SPARSEMEM
373 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
374 * In SPARSEMEM, this map is stored in struct mem_section
376 unsigned long *pageblock_flags;
377 #endif /* CONFIG_SPARSEMEM */
379 #ifdef CONFIG_COMPACTION
381 * On compaction failure, 1<<compact_defer_shift compactions
382 * are skipped before trying again. The number attempted since
383 * last failure is tracked with compact_considered.
385 unsigned int compact_considered;
386 unsigned int compact_defer_shift;
387 int compact_order_failed;
392 /* Fields commonly accessed by the page reclaim scanner */
394 struct lruvec lruvec;
396 unsigned long pages_scanned; /* since last reclaim */
397 unsigned long flags; /* zone flags, see below */
399 /* Zone statistics */
400 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
403 * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on
404 * this zone's LRU. Maintained by the pageout code.
406 unsigned int inactive_ratio;
410 /* Rarely used or read-mostly fields */
413 * wait_table -- the array holding the hash table
414 * wait_table_hash_nr_entries -- the size of the hash table array
415 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
417 * The purpose of all these is to keep track of the people
418 * waiting for a page to become available and make them
419 * runnable again when possible. The trouble is that this
420 * consumes a lot of space, especially when so few things
421 * wait on pages at a given time. So instead of using
422 * per-page waitqueues, we use a waitqueue hash table.
424 * The bucket discipline is to sleep on the same queue when
425 * colliding and wake all in that wait queue when removing.
426 * When something wakes, it must check to be sure its page is
427 * truly available, a la thundering herd. The cost of a
428 * collision is great, but given the expected load of the
429 * table, they should be so rare as to be outweighed by the
430 * benefits from the saved space.
432 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
433 * primary users of these fields, and in mm/page_alloc.c
434 * free_area_init_core() performs the initialization of them.
436 wait_queue_head_t * wait_table;
437 unsigned long wait_table_hash_nr_entries;
438 unsigned long wait_table_bits;
441 * Discontig memory support fields.
443 struct pglist_data *zone_pgdat;
444 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
445 unsigned long zone_start_pfn;
448 * spanned_pages is the total pages spanned by the zone, including
449 * holes, which is calculated as:
450 * spanned_pages = zone_end_pfn - zone_start_pfn;
452 * present_pages is physical pages existing within the zone, which
454 * present_pages = spanned_pages - absent_pages(pages in holes);
456 * managed_pages is present pages managed by the buddy system, which
457 * is calculated as (reserved_pages includes pages allocated by the
458 * bootmem allocator):
459 * managed_pages = present_pages - reserved_pages;
461 * So present_pages may be used by memory hotplug or memory power
462 * management logic to figure out unmanaged pages by checking
463 * (present_pages - managed_pages). And managed_pages should be used
464 * by page allocator and vm scanner to calculate all kinds of watermarks
469 * zone_start_pfn and spanned_pages are protected by span_seqlock.
470 * It is a seqlock because it has to be read outside of zone->lock,
471 * and it is done in the main allocator path. But, it is written
472 * quite infrequently.
474 * The span_seq lock is declared along with zone->lock because it is
475 * frequently read in proximity to zone->lock. It's good to
476 * give them a chance of being in the same cacheline.
478 * Write access to present_pages at runtime should be protected by
479 * lock_memory_hotplug()/unlock_memory_hotplug(). Any reader who can't
480 * tolerant drift of present_pages should hold memory hotplug lock to
481 * get a stable value.
483 * Read access to managed_pages should be safe because it's unsigned
484 * long. Write access to zone->managed_pages and totalram_pages are
485 * protected by managed_page_count_lock at runtime. Idealy only
486 * adjust_managed_page_count() should be used instead of directly
487 * touching zone->managed_pages and totalram_pages.
489 unsigned long spanned_pages;
490 unsigned long present_pages;
491 unsigned long managed_pages;
494 * rarely used fields:
497 } ____cacheline_internodealigned_in_smp;
500 ZONE_RECLAIM_LOCKED, /* prevents concurrent reclaim */
501 ZONE_OOM_LOCKED, /* zone is in OOM killer zonelist */
502 ZONE_CONGESTED, /* zone has many dirty pages backed by
505 ZONE_TAIL_LRU_DIRTY, /* reclaim scanning has recently found
506 * many dirty file pages at the tail
509 ZONE_WRITEBACK, /* reclaim scanning has recently found
510 * many pages under writeback
514 static inline void zone_set_flag(struct zone *zone, zone_flags_t flag)
516 set_bit(flag, &zone->flags);
519 static inline int zone_test_and_set_flag(struct zone *zone, zone_flags_t flag)
521 return test_and_set_bit(flag, &zone->flags);
524 static inline void zone_clear_flag(struct zone *zone, zone_flags_t flag)
526 clear_bit(flag, &zone->flags);
529 static inline int zone_is_reclaim_congested(const struct zone *zone)
531 return test_bit(ZONE_CONGESTED, &zone->flags);
534 static inline int zone_is_reclaim_dirty(const struct zone *zone)
536 return test_bit(ZONE_TAIL_LRU_DIRTY, &zone->flags);
539 static inline int zone_is_reclaim_writeback(const struct zone *zone)
541 return test_bit(ZONE_WRITEBACK, &zone->flags);
544 static inline int zone_is_reclaim_locked(const struct zone *zone)
546 return test_bit(ZONE_RECLAIM_LOCKED, &zone->flags);
549 static inline int zone_is_oom_locked(const struct zone *zone)
551 return test_bit(ZONE_OOM_LOCKED, &zone->flags);
554 static inline unsigned long zone_end_pfn(const struct zone *zone)
556 return zone->zone_start_pfn + zone->spanned_pages;
559 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
561 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
564 static inline bool zone_is_initialized(struct zone *zone)
566 return !!zone->wait_table;
569 static inline bool zone_is_empty(struct zone *zone)
571 return zone->spanned_pages == 0;
575 * The "priority" of VM scanning is how much of the queues we will scan in one
576 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
577 * queues ("queue_length >> 12") during an aging round.
579 #define DEF_PRIORITY 12
581 /* Maximum number of zones on a zonelist */
582 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
587 * The NUMA zonelists are doubled because we need zonelists that restrict the
588 * allocations to a single node for GFP_THISNODE.
590 * [0] : Zonelist with fallback
591 * [1] : No fallback (GFP_THISNODE)
593 #define MAX_ZONELISTS 2
597 * We cache key information from each zonelist for smaller cache
598 * footprint when scanning for free pages in get_page_from_freelist().
600 * 1) The BITMAP fullzones tracks which zones in a zonelist have come
601 * up short of free memory since the last time (last_fullzone_zap)
602 * we zero'd fullzones.
603 * 2) The array z_to_n[] maps each zone in the zonelist to its node
604 * id, so that we can efficiently evaluate whether that node is
605 * set in the current tasks mems_allowed.
607 * Both fullzones and z_to_n[] are one-to-one with the zonelist,
608 * indexed by a zones offset in the zonelist zones[] array.
610 * The get_page_from_freelist() routine does two scans. During the
611 * first scan, we skip zones whose corresponding bit in 'fullzones'
612 * is set or whose corresponding node in current->mems_allowed (which
613 * comes from cpusets) is not set. During the second scan, we bypass
614 * this zonelist_cache, to ensure we look methodically at each zone.
616 * Once per second, we zero out (zap) fullzones, forcing us to
617 * reconsider nodes that might have regained more free memory.
618 * The field last_full_zap is the time we last zapped fullzones.
620 * This mechanism reduces the amount of time we waste repeatedly
621 * reexaming zones for free memory when they just came up low on
622 * memory momentarilly ago.
624 * The zonelist_cache struct members logically belong in struct
625 * zonelist. However, the mempolicy zonelists constructed for
626 * MPOL_BIND are intentionally variable length (and usually much
627 * shorter). A general purpose mechanism for handling structs with
628 * multiple variable length members is more mechanism than we want
629 * here. We resort to some special case hackery instead.
631 * The MPOL_BIND zonelists don't need this zonelist_cache (in good
632 * part because they are shorter), so we put the fixed length stuff
633 * at the front of the zonelist struct, ending in a variable length
634 * zones[], as is needed by MPOL_BIND.
636 * Then we put the optional zonelist cache on the end of the zonelist
637 * struct. This optional stuff is found by a 'zlcache_ptr' pointer in
638 * the fixed length portion at the front of the struct. This pointer
639 * both enables us to find the zonelist cache, and in the case of
640 * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL)
641 * to know that the zonelist cache is not there.
643 * The end result is that struct zonelists come in two flavors:
644 * 1) The full, fixed length version, shown below, and
645 * 2) The custom zonelists for MPOL_BIND.
646 * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache.
648 * Even though there may be multiple CPU cores on a node modifying
649 * fullzones or last_full_zap in the same zonelist_cache at the same
650 * time, we don't lock it. This is just hint data - if it is wrong now
651 * and then, the allocator will still function, perhaps a bit slower.
655 struct zonelist_cache {
656 unsigned short z_to_n[MAX_ZONES_PER_ZONELIST]; /* zone->nid */
657 DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST); /* zone full? */
658 unsigned long last_full_zap; /* when last zap'd (jiffies) */
661 #define MAX_ZONELISTS 1
662 struct zonelist_cache;
666 * This struct contains information about a zone in a zonelist. It is stored
667 * here to avoid dereferences into large structures and lookups of tables
670 struct zone *zone; /* Pointer to actual zone */
671 int zone_idx; /* zone_idx(zoneref->zone) */
675 * One allocation request operates on a zonelist. A zonelist
676 * is a list of zones, the first one is the 'goal' of the
677 * allocation, the other zones are fallback zones, in decreasing
680 * If zlcache_ptr is not NULL, then it is just the address of zlcache,
681 * as explained above. If zlcache_ptr is NULL, there is no zlcache.
683 * To speed the reading of the zonelist, the zonerefs contain the zone index
684 * of the entry being read. Helper functions to access information given
685 * a struct zoneref are
687 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
688 * zonelist_zone_idx() - Return the index of the zone for an entry
689 * zonelist_node_idx() - Return the index of the node for an entry
692 struct zonelist_cache *zlcache_ptr; // NULL or &zlcache
693 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
695 struct zonelist_cache zlcache; // optional ...
699 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
700 struct node_active_region {
701 unsigned long start_pfn;
702 unsigned long end_pfn;
705 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
707 #ifndef CONFIG_DISCONTIGMEM
708 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
709 extern struct page *mem_map;
713 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
714 * (mostly NUMA machines?) to denote a higher-level memory zone than the
717 * On NUMA machines, each NUMA node would have a pg_data_t to describe
718 * it's memory layout.
720 * Memory statistics and page replacement data structures are maintained on a
724 typedef struct pglist_data {
725 struct zone node_zones[MAX_NR_ZONES];
726 struct zonelist node_zonelists[MAX_ZONELISTS];
728 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
729 struct page *node_mem_map;
731 struct page_cgroup *node_page_cgroup;
734 #ifndef CONFIG_NO_BOOTMEM
735 struct bootmem_data *bdata;
737 #ifdef CONFIG_MEMORY_HOTPLUG
739 * Must be held any time you expect node_start_pfn, node_present_pages
740 * or node_spanned_pages stay constant. Holding this will also
741 * guarantee that any pfn_valid() stays that way.
743 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
744 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG.
746 * Nests above zone->lock and zone->span_seqlock
748 spinlock_t node_size_lock;
750 unsigned long node_start_pfn;
751 unsigned long node_present_pages; /* total number of physical pages */
752 unsigned long node_spanned_pages; /* total size of physical page
753 range, including holes */
755 nodemask_t reclaim_nodes; /* Nodes allowed to reclaim from */
756 wait_queue_head_t kswapd_wait;
757 wait_queue_head_t pfmemalloc_wait;
758 struct task_struct *kswapd; /* Protected by lock_memory_hotplug() */
759 int kswapd_max_order;
760 enum zone_type classzone_idx;
761 #ifdef CONFIG_NUMA_BALANCING
763 * Lock serializing the per destination node AutoNUMA memory
764 * migration rate limiting data.
766 spinlock_t numabalancing_migrate_lock;
768 /* Rate limiting time interval */
769 unsigned long numabalancing_migrate_next_window;
771 /* Number of pages migrated during the rate limiting time interval */
772 unsigned long numabalancing_migrate_nr_pages;
776 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
777 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
778 #ifdef CONFIG_FLAT_NODE_MEM_MAP
779 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
781 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
783 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
785 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
786 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
788 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
790 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
793 static inline bool pgdat_is_empty(pg_data_t *pgdat)
795 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
798 #include <linux/memory_hotplug.h>
800 extern struct mutex zonelists_mutex;
801 void build_all_zonelists(pg_data_t *pgdat, struct zone *zone);
802 void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx);
803 bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
804 int classzone_idx, int alloc_flags);
805 bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
806 int classzone_idx, int alloc_flags);
807 enum memmap_context {
811 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
813 enum memmap_context context);
815 extern void lruvec_init(struct lruvec *lruvec);
817 static inline struct zone *lruvec_zone(struct lruvec *lruvec)
822 return container_of(lruvec, struct zone, lruvec);
826 #ifdef CONFIG_HAVE_MEMORY_PRESENT
827 void memory_present(int nid, unsigned long start, unsigned long end);
829 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
832 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
833 int local_memory_node(int node_id);
835 static inline int local_memory_node(int node_id) { return node_id; };
838 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
839 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
843 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
845 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
847 static inline int populated_zone(struct zone *zone)
849 return (!!zone->present_pages);
852 extern int movable_zone;
854 static inline int zone_movable_is_highmem(void)
856 #if defined(CONFIG_HIGHMEM) && defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
857 return movable_zone == ZONE_HIGHMEM;
863 static inline int is_highmem_idx(enum zone_type idx)
865 #ifdef CONFIG_HIGHMEM
866 return (idx == ZONE_HIGHMEM ||
867 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
874 * is_highmem - helper function to quickly check if a struct zone is a
875 * highmem zone or not. This is an attempt to keep references
876 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
877 * @zone - pointer to struct zone variable
879 static inline int is_highmem(struct zone *zone)
881 #ifdef CONFIG_HIGHMEM
882 int zone_off = (char *)zone - (char *)zone->zone_pgdat->node_zones;
883 return zone_off == ZONE_HIGHMEM * sizeof(*zone) ||
884 (zone_off == ZONE_MOVABLE * sizeof(*zone) &&
885 zone_movable_is_highmem());
891 /* These two functions are used to setup the per zone pages min values */
893 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
894 void __user *, size_t *, loff_t *);
895 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
896 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
897 void __user *, size_t *, loff_t *);
898 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
899 void __user *, size_t *, loff_t *);
900 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
901 void __user *, size_t *, loff_t *);
902 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
903 void __user *, size_t *, loff_t *);
905 extern int numa_zonelist_order_handler(struct ctl_table *, int,
906 void __user *, size_t *, loff_t *);
907 extern char numa_zonelist_order[];
908 #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
910 #ifndef CONFIG_NEED_MULTIPLE_NODES
912 extern struct pglist_data contig_page_data;
913 #define NODE_DATA(nid) (&contig_page_data)
914 #define NODE_MEM_MAP(nid) mem_map
916 #else /* CONFIG_NEED_MULTIPLE_NODES */
918 #include <asm/mmzone.h>
920 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
922 extern struct pglist_data *first_online_pgdat(void);
923 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
924 extern struct zone *next_zone(struct zone *zone);
927 * for_each_online_pgdat - helper macro to iterate over all online nodes
928 * @pgdat - pointer to a pg_data_t variable
930 #define for_each_online_pgdat(pgdat) \
931 for (pgdat = first_online_pgdat(); \
933 pgdat = next_online_pgdat(pgdat))
935 * for_each_zone - helper macro to iterate over all memory zones
936 * @zone - pointer to struct zone variable
938 * The user only needs to declare the zone variable, for_each_zone
941 #define for_each_zone(zone) \
942 for (zone = (first_online_pgdat())->node_zones; \
944 zone = next_zone(zone))
946 #define for_each_populated_zone(zone) \
947 for (zone = (first_online_pgdat())->node_zones; \
949 zone = next_zone(zone)) \
950 if (!populated_zone(zone)) \
954 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
956 return zoneref->zone;
959 static inline int zonelist_zone_idx(struct zoneref *zoneref)
961 return zoneref->zone_idx;
964 static inline int zonelist_node_idx(struct zoneref *zoneref)
967 /* zone_to_nid not available in this context */
968 return zoneref->zone->node;
971 #endif /* CONFIG_NUMA */
975 * 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
976 * @z - The cursor used as a starting point for the search
977 * @highest_zoneidx - The zone index of the highest zone to return
978 * @nodes - An optional nodemask to filter the zonelist with
979 * @zone - The first suitable zone found is returned via this parameter
981 * This function returns the next zone at or below a given zone index that is
982 * within the allowed nodemask using a cursor as the starting point for the
983 * search. The zoneref returned is a cursor that represents the current zone
984 * being examined. It should be advanced by one before calling
985 * next_zones_zonelist again.
987 struct zoneref *next_zones_zonelist(struct zoneref *z,
988 enum zone_type highest_zoneidx,
993 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
994 * @zonelist - The zonelist to search for a suitable zone
995 * @highest_zoneidx - The zone index of the highest zone to return
996 * @nodes - An optional nodemask to filter the zonelist with
997 * @zone - The first suitable zone found is returned via this parameter
999 * This function returns the first zone at or below a given zone index that is
1000 * within the allowed nodemask. The zoneref returned is a cursor that can be
1001 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1002 * one before calling.
1004 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1005 enum zone_type highest_zoneidx,
1009 return next_zones_zonelist(zonelist->_zonerefs, highest_zoneidx, nodes,
1014 * 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
1015 * @zone - The current zone in the iterator
1016 * @z - The current pointer within zonelist->zones being iterated
1017 * @zlist - The zonelist being iterated
1018 * @highidx - The zone index of the highest zone to return
1019 * @nodemask - Nodemask allowed by the allocator
1021 * This iterator iterates though all zones at or below a given zone index and
1022 * within a given nodemask
1024 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1025 for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \
1027 z = next_zones_zonelist(++z, highidx, nodemask, &zone)) \
1030 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1031 * @zone - The current zone in the iterator
1032 * @z - The current pointer within zonelist->zones being iterated
1033 * @zlist - The zonelist being iterated
1034 * @highidx - The zone index of the highest zone to return
1036 * This iterator iterates though all zones at or below a given zone index.
1038 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1039 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1041 #ifdef CONFIG_SPARSEMEM
1042 #include <asm/sparsemem.h>
1045 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1046 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1047 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1053 #ifdef CONFIG_FLATMEM
1054 #define pfn_to_nid(pfn) (0)
1057 #ifdef CONFIG_SPARSEMEM
1060 * SECTION_SHIFT #bits space required to store a section #
1062 * PA_SECTION_SHIFT physical address to/from section number
1063 * PFN_SECTION_SHIFT pfn to/from section number
1065 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1066 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1068 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1070 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1071 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1073 #define SECTION_BLOCKFLAGS_BITS \
1074 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1076 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1077 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1080 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
1081 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
1083 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1084 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1088 struct mem_section {
1090 * This is, logically, a pointer to an array of struct
1091 * pages. However, it is stored with some other magic.
1092 * (see sparse.c::sparse_init_one_section())
1094 * Additionally during early boot we encode node id of
1095 * the location of the section here to guide allocation.
1096 * (see sparse.c::memory_present())
1098 * Making it a UL at least makes someone do a cast
1099 * before using it wrong.
1101 unsigned long section_mem_map;
1103 /* See declaration of similar field in struct zone */
1104 unsigned long *pageblock_flags;
1107 * If !SPARSEMEM, pgdat doesn't have page_cgroup pointer. We use
1108 * section. (see memcontrol.h/page_cgroup.h about this.)
1110 struct page_cgroup *page_cgroup;
1114 * WARNING: mem_section must be a power-of-2 in size for the
1115 * calculation and use of SECTION_ROOT_MASK to make sense.
1119 #ifdef CONFIG_SPARSEMEM_EXTREME
1120 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1122 #define SECTIONS_PER_ROOT 1
1125 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1126 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1127 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1129 #ifdef CONFIG_SPARSEMEM_EXTREME
1130 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
1132 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1135 static inline struct mem_section *__nr_to_section(unsigned long nr)
1137 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1139 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1141 extern int __section_nr(struct mem_section* ms);
1142 extern unsigned long usemap_size(void);
1145 * We use the lower bits of the mem_map pointer to store
1146 * a little bit of information. There should be at least
1147 * 3 bits here due to 32-bit alignment.
1149 #define SECTION_MARKED_PRESENT (1UL<<0)
1150 #define SECTION_HAS_MEM_MAP (1UL<<1)
1151 #define SECTION_MAP_LAST_BIT (1UL<<2)
1152 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1153 #define SECTION_NID_SHIFT 2
1155 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1157 unsigned long map = section->section_mem_map;
1158 map &= SECTION_MAP_MASK;
1159 return (struct page *)map;
1162 static inline int present_section(struct mem_section *section)
1164 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1167 static inline int present_section_nr(unsigned long nr)
1169 return present_section(__nr_to_section(nr));
1172 static inline int valid_section(struct mem_section *section)
1174 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1177 static inline int valid_section_nr(unsigned long nr)
1179 return valid_section(__nr_to_section(nr));
1182 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1184 return __nr_to_section(pfn_to_section_nr(pfn));
1187 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1188 static inline int pfn_valid(unsigned long pfn)
1190 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1192 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
1196 static inline int pfn_present(unsigned long pfn)
1198 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1200 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1204 * These are _only_ used during initialisation, therefore they
1205 * can use __initdata ... They could have names to indicate
1209 #define pfn_to_nid(pfn) \
1211 unsigned long __pfn_to_nid_pfn = (pfn); \
1212 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1215 #define pfn_to_nid(pfn) (0)
1218 #define early_pfn_valid(pfn) pfn_valid(pfn)
1219 void sparse_init(void);
1221 #define sparse_init() do {} while (0)
1222 #define sparse_index_init(_sec, _nid) do {} while (0)
1223 #endif /* CONFIG_SPARSEMEM */
1225 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1226 bool early_pfn_in_nid(unsigned long pfn, int nid);
1228 #define early_pfn_in_nid(pfn, nid) (1)
1231 #ifndef early_pfn_valid
1232 #define early_pfn_valid(pfn) (1)
1235 void memory_present(int nid, unsigned long start, unsigned long end);
1236 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
1239 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1240 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1241 * pfn_valid_within() should be used in this case; we optimise this away
1242 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1244 #ifdef CONFIG_HOLES_IN_ZONE
1245 #define pfn_valid_within(pfn) pfn_valid(pfn)
1247 #define pfn_valid_within(pfn) (1)
1250 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1252 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1253 * associated with it or not. In FLATMEM, it is expected that holes always
1254 * have valid memmap as long as there is valid PFNs either side of the hole.
1255 * In SPARSEMEM, it is assumed that a valid section has a memmap for the
1258 * However, an ARM, and maybe other embedded architectures in the future
1259 * free memmap backing holes to save memory on the assumption the memmap is
1260 * never used. The page_zone linkages are then broken even though pfn_valid()
1261 * returns true. A walker of the full memmap must then do this additional
1262 * check to ensure the memmap they are looking at is sane by making sure
1263 * the zone and PFN linkages are still valid. This is expensive, but walkers
1264 * of the full memmap are extremely rare.
1266 int memmap_valid_within(unsigned long pfn,
1267 struct page *page, struct zone *zone);
1269 static inline int memmap_valid_within(unsigned long pfn,
1270 struct page *page, struct zone *zone)
1274 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1276 #endif /* !__GENERATING_BOUNDS.H */
1277 #endif /* !__ASSEMBLY__ */
1278 #endif /* _LINUX_MMZONE_H */