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 #define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1)
79 #define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1)
81 static inline int get_pageblock_migratetype(struct page *page)
83 BUILD_BUG_ON(PB_migrate_end - PB_migrate != 2);
84 return get_pageblock_flags_mask(page, PB_migrate_end, MIGRATETYPE_MASK);
88 struct list_head free_list[MIGRATE_TYPES];
89 unsigned long nr_free;
95 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
96 * So add a wild amount of padding here to ensure that they fall into separate
97 * cachelines. There are very few zone structures in the machine, so space
98 * consumption is not a concern here.
100 #if defined(CONFIG_SMP)
101 struct zone_padding {
103 } ____cacheline_internodealigned_in_smp;
104 #define ZONE_PADDING(name) struct zone_padding name;
106 #define ZONE_PADDING(name)
109 enum zone_stat_item {
110 /* First 128 byte cacheline (assuming 64 bit words) */
114 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
115 NR_ACTIVE_ANON, /* " " " " " */
116 NR_INACTIVE_FILE, /* " " " " " */
117 NR_ACTIVE_FILE, /* " " " " " */
118 NR_UNEVICTABLE, /* " " " " " */
119 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
120 NR_ANON_PAGES, /* Mapped anonymous pages */
121 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
122 only modified from process context */
127 NR_SLAB_UNRECLAIMABLE,
128 NR_PAGETABLE, /* used for pagetables */
130 /* Second 128 byte cacheline */
131 NR_UNSTABLE_NFS, /* NFS unstable pages */
134 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
135 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
136 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
137 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
138 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
139 NR_DIRTIED, /* page dirtyings since bootup */
140 NR_WRITTEN, /* page writings since bootup */
142 NUMA_HIT, /* allocated in intended node */
143 NUMA_MISS, /* allocated in non intended node */
144 NUMA_FOREIGN, /* was intended here, hit elsewhere */
145 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
146 NUMA_LOCAL, /* allocation from local node */
147 NUMA_OTHER, /* allocation from other node */
149 NR_ANON_TRANSPARENT_HUGEPAGES,
151 NR_VM_ZONE_STAT_ITEMS };
154 * We do arithmetic on the LRU lists in various places in the code,
155 * so it is important to keep the active lists LRU_ACTIVE higher in
156 * the array than the corresponding inactive lists, and to keep
157 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
159 * This has to be kept in sync with the statistics in zone_stat_item
160 * above and the descriptions in vmstat_text in mm/vmstat.c
167 LRU_INACTIVE_ANON = LRU_BASE,
168 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
169 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
170 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
175 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
177 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
179 static inline int is_file_lru(enum lru_list lru)
181 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
184 static inline int is_active_lru(enum lru_list lru)
186 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
189 static inline int is_unevictable_lru(enum lru_list lru)
191 return (lru == LRU_UNEVICTABLE);
194 struct zone_reclaim_stat {
196 * The pageout code in vmscan.c keeps track of how many of the
197 * mem/swap backed and file backed pages are referenced.
198 * The higher the rotated/scanned ratio, the more valuable
201 * The anon LRU stats live in [0], file LRU stats in [1]
203 unsigned long recent_rotated[2];
204 unsigned long recent_scanned[2];
208 struct list_head lists[NR_LRU_LISTS];
209 struct zone_reclaim_stat reclaim_stat;
215 /* Mask used at gathering information at once (see memcontrol.c) */
216 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
217 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
218 #define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
220 /* Isolate clean file */
221 #define ISOLATE_CLEAN ((__force isolate_mode_t)0x1)
222 /* Isolate unmapped file */
223 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
224 /* Isolate for asynchronous migration */
225 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
226 /* Isolate unevictable pages */
227 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
229 /* LRU Isolation modes. */
230 typedef unsigned __bitwise__ isolate_mode_t;
232 enum zone_watermarks {
239 #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
240 #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
241 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
243 struct per_cpu_pages {
244 int count; /* number of pages in the list */
245 int high; /* high watermark, emptying needed */
246 int batch; /* chunk size for buddy add/remove */
248 /* Lists of pages, one per migrate type stored on the pcp-lists */
249 struct list_head lists[MIGRATE_PCPTYPES];
252 struct per_cpu_pageset {
253 struct per_cpu_pages pcp;
259 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
263 #endif /* !__GENERATING_BOUNDS.H */
266 #ifdef CONFIG_ZONE_DMA
268 * ZONE_DMA is used when there are devices that are not able
269 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
270 * carve out the portion of memory that is needed for these devices.
271 * The range is arch specific.
276 * ---------------------------
277 * parisc, ia64, sparc <4G
280 * alpha Unlimited or 0-16MB.
282 * i386, x86_64 and multiple other arches
287 #ifdef CONFIG_ZONE_DMA32
289 * x86_64 needs two ZONE_DMAs because it supports devices that are
290 * only able to do DMA to the lower 16M but also 32 bit devices that
291 * can only do DMA areas below 4G.
296 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
297 * performed on pages in ZONE_NORMAL if the DMA devices support
298 * transfers to all addressable memory.
301 #ifdef CONFIG_HIGHMEM
303 * A memory area that is only addressable by the kernel through
304 * mapping portions into its own address space. This is for example
305 * used by i386 to allow the kernel to address the memory beyond
306 * 900MB. The kernel will set up special mappings (page
307 * table entries on i386) for each page that the kernel needs to
316 #ifndef __GENERATING_BOUNDS_H
319 /* Fields commonly accessed by the page allocator */
321 /* zone watermarks, access with *_wmark_pages(zone) macros */
322 unsigned long watermark[NR_WMARK];
325 * When free pages are below this point, additional steps are taken
326 * when reading the number of free pages to avoid per-cpu counter
327 * drift allowing watermarks to be breached
329 unsigned long percpu_drift_mark;
332 * We don't know if the memory that we're going to allocate will be freeable
333 * or/and it will be released eventually, so to avoid totally wasting several
334 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
335 * to run OOM on the lower zones despite there's tons of freeable ram
336 * on the higher zones). This array is recalculated at runtime if the
337 * sysctl_lowmem_reserve_ratio sysctl changes.
339 unsigned long lowmem_reserve[MAX_NR_ZONES];
342 * This is a per-zone reserve of pages that should not be
343 * considered dirtyable memory.
345 unsigned long dirty_balance_reserve;
350 * zone reclaim becomes active if more unmapped pages exist.
352 unsigned long min_unmapped_pages;
353 unsigned long min_slab_pages;
355 struct per_cpu_pageset __percpu *pageset;
357 * free areas of different sizes
360 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
361 /* Set to true when the PG_migrate_skip bits should be cleared */
362 bool compact_blockskip_flush;
364 /* pfns where compaction scanners should start */
365 unsigned long compact_cached_free_pfn;
366 unsigned long compact_cached_migrate_pfn;
368 #ifdef CONFIG_MEMORY_HOTPLUG
369 /* see spanned/present_pages for more description */
370 seqlock_t span_seqlock;
372 struct free_area free_area[MAX_ORDER];
374 #ifndef CONFIG_SPARSEMEM
376 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
377 * In SPARSEMEM, this map is stored in struct mem_section
379 unsigned long *pageblock_flags;
380 #endif /* CONFIG_SPARSEMEM */
382 #ifdef CONFIG_COMPACTION
384 * On compaction failure, 1<<compact_defer_shift compactions
385 * are skipped before trying again. The number attempted since
386 * last failure is tracked with compact_considered.
388 unsigned int compact_considered;
389 unsigned int compact_defer_shift;
390 int compact_order_failed;
395 /* Fields commonly accessed by the page reclaim scanner */
397 struct lruvec lruvec;
399 unsigned long pages_scanned; /* since last reclaim */
400 unsigned long flags; /* zone flags, see below */
402 /* Zone statistics */
403 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
406 * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on
407 * this zone's LRU. Maintained by the pageout code.
409 unsigned int inactive_ratio;
413 /* Rarely used or read-mostly fields */
416 * wait_table -- the array holding the hash table
417 * wait_table_hash_nr_entries -- the size of the hash table array
418 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
420 * The purpose of all these is to keep track of the people
421 * waiting for a page to become available and make them
422 * runnable again when possible. The trouble is that this
423 * consumes a lot of space, especially when so few things
424 * wait on pages at a given time. So instead of using
425 * per-page waitqueues, we use a waitqueue hash table.
427 * The bucket discipline is to sleep on the same queue when
428 * colliding and wake all in that wait queue when removing.
429 * When something wakes, it must check to be sure its page is
430 * truly available, a la thundering herd. The cost of a
431 * collision is great, but given the expected load of the
432 * table, they should be so rare as to be outweighed by the
433 * benefits from the saved space.
435 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
436 * primary users of these fields, and in mm/page_alloc.c
437 * free_area_init_core() performs the initialization of them.
439 wait_queue_head_t * wait_table;
440 unsigned long wait_table_hash_nr_entries;
441 unsigned long wait_table_bits;
444 * Discontig memory support fields.
446 struct pglist_data *zone_pgdat;
447 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
448 unsigned long zone_start_pfn;
451 * spanned_pages is the total pages spanned by the zone, including
452 * holes, which is calculated as:
453 * spanned_pages = zone_end_pfn - zone_start_pfn;
455 * present_pages is physical pages existing within the zone, which
457 * present_pages = spanned_pages - absent_pages(pages in holes);
459 * managed_pages is present pages managed by the buddy system, which
460 * is calculated as (reserved_pages includes pages allocated by the
461 * bootmem allocator):
462 * managed_pages = present_pages - reserved_pages;
464 * So present_pages may be used by memory hotplug or memory power
465 * management logic to figure out unmanaged pages by checking
466 * (present_pages - managed_pages). And managed_pages should be used
467 * by page allocator and vm scanner to calculate all kinds of watermarks
472 * zone_start_pfn and spanned_pages are protected by span_seqlock.
473 * It is a seqlock because it has to be read outside of zone->lock,
474 * and it is done in the main allocator path. But, it is written
475 * quite infrequently.
477 * The span_seq 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.
481 * Write access to present_pages at runtime should be protected by
482 * lock_memory_hotplug()/unlock_memory_hotplug(). Any reader who can't
483 * tolerant drift of present_pages should hold memory hotplug lock to
484 * get a stable value.
486 * Read access to managed_pages should be safe because it's unsigned
487 * long. Write access to zone->managed_pages and totalram_pages are
488 * protected by managed_page_count_lock at runtime. Idealy only
489 * adjust_managed_page_count() should be used instead of directly
490 * touching zone->managed_pages and totalram_pages.
492 unsigned long spanned_pages;
493 unsigned long present_pages;
494 unsigned long managed_pages;
497 * Number of MIGRATE_RESEVE page block. To maintain for just
498 * optimization. Protected by zone->lock.
500 int nr_migrate_reserve_block;
503 * rarely used fields:
506 } ____cacheline_internodealigned_in_smp;
509 ZONE_RECLAIM_LOCKED, /* prevents concurrent reclaim */
510 ZONE_OOM_LOCKED, /* zone is in OOM killer zonelist */
511 ZONE_CONGESTED, /* zone has many dirty pages backed by
514 ZONE_TAIL_LRU_DIRTY, /* reclaim scanning has recently found
515 * many dirty file pages at the tail
518 ZONE_WRITEBACK, /* reclaim scanning has recently found
519 * many pages under writeback
523 static inline void zone_set_flag(struct zone *zone, zone_flags_t flag)
525 set_bit(flag, &zone->flags);
528 static inline int zone_test_and_set_flag(struct zone *zone, zone_flags_t flag)
530 return test_and_set_bit(flag, &zone->flags);
533 static inline void zone_clear_flag(struct zone *zone, zone_flags_t flag)
535 clear_bit(flag, &zone->flags);
538 static inline int zone_is_reclaim_congested(const struct zone *zone)
540 return test_bit(ZONE_CONGESTED, &zone->flags);
543 static inline int zone_is_reclaim_dirty(const struct zone *zone)
545 return test_bit(ZONE_TAIL_LRU_DIRTY, &zone->flags);
548 static inline int zone_is_reclaim_writeback(const struct zone *zone)
550 return test_bit(ZONE_WRITEBACK, &zone->flags);
553 static inline int zone_is_reclaim_locked(const struct zone *zone)
555 return test_bit(ZONE_RECLAIM_LOCKED, &zone->flags);
558 static inline int zone_is_oom_locked(const struct zone *zone)
560 return test_bit(ZONE_OOM_LOCKED, &zone->flags);
563 static inline unsigned long zone_end_pfn(const struct zone *zone)
565 return zone->zone_start_pfn + zone->spanned_pages;
568 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
570 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
573 static inline bool zone_is_initialized(struct zone *zone)
575 return !!zone->wait_table;
578 static inline bool zone_is_empty(struct zone *zone)
580 return zone->spanned_pages == 0;
584 * The "priority" of VM scanning is how much of the queues we will scan in one
585 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
586 * queues ("queue_length >> 12") during an aging round.
588 #define DEF_PRIORITY 12
590 /* Maximum number of zones on a zonelist */
591 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
596 * The NUMA zonelists are doubled because we need zonelists that restrict the
597 * allocations to a single node for __GFP_THISNODE.
599 * [0] : Zonelist with fallback
600 * [1] : No fallback (__GFP_THISNODE)
602 #define MAX_ZONELISTS 2
606 * We cache key information from each zonelist for smaller cache
607 * footprint when scanning for free pages in get_page_from_freelist().
609 * 1) The BITMAP fullzones tracks which zones in a zonelist have come
610 * up short of free memory since the last time (last_fullzone_zap)
611 * we zero'd fullzones.
612 * 2) The array z_to_n[] maps each zone in the zonelist to its node
613 * id, so that we can efficiently evaluate whether that node is
614 * set in the current tasks mems_allowed.
616 * Both fullzones and z_to_n[] are one-to-one with the zonelist,
617 * indexed by a zones offset in the zonelist zones[] array.
619 * The get_page_from_freelist() routine does two scans. During the
620 * first scan, we skip zones whose corresponding bit in 'fullzones'
621 * is set or whose corresponding node in current->mems_allowed (which
622 * comes from cpusets) is not set. During the second scan, we bypass
623 * this zonelist_cache, to ensure we look methodically at each zone.
625 * Once per second, we zero out (zap) fullzones, forcing us to
626 * reconsider nodes that might have regained more free memory.
627 * The field last_full_zap is the time we last zapped fullzones.
629 * This mechanism reduces the amount of time we waste repeatedly
630 * reexaming zones for free memory when they just came up low on
631 * memory momentarilly ago.
633 * The zonelist_cache struct members logically belong in struct
634 * zonelist. However, the mempolicy zonelists constructed for
635 * MPOL_BIND are intentionally variable length (and usually much
636 * shorter). A general purpose mechanism for handling structs with
637 * multiple variable length members is more mechanism than we want
638 * here. We resort to some special case hackery instead.
640 * The MPOL_BIND zonelists don't need this zonelist_cache (in good
641 * part because they are shorter), so we put the fixed length stuff
642 * at the front of the zonelist struct, ending in a variable length
643 * zones[], as is needed by MPOL_BIND.
645 * Then we put the optional zonelist cache on the end of the zonelist
646 * struct. This optional stuff is found by a 'zlcache_ptr' pointer in
647 * the fixed length portion at the front of the struct. This pointer
648 * both enables us to find the zonelist cache, and in the case of
649 * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL)
650 * to know that the zonelist cache is not there.
652 * The end result is that struct zonelists come in two flavors:
653 * 1) The full, fixed length version, shown below, and
654 * 2) The custom zonelists for MPOL_BIND.
655 * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache.
657 * Even though there may be multiple CPU cores on a node modifying
658 * fullzones or last_full_zap in the same zonelist_cache at the same
659 * time, we don't lock it. This is just hint data - if it is wrong now
660 * and then, the allocator will still function, perhaps a bit slower.
664 struct zonelist_cache {
665 unsigned short z_to_n[MAX_ZONES_PER_ZONELIST]; /* zone->nid */
666 DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST); /* zone full? */
667 unsigned long last_full_zap; /* when last zap'd (jiffies) */
670 #define MAX_ZONELISTS 1
671 struct zonelist_cache;
675 * This struct contains information about a zone in a zonelist. It is stored
676 * here to avoid dereferences into large structures and lookups of tables
679 struct zone *zone; /* Pointer to actual zone */
680 int zone_idx; /* zone_idx(zoneref->zone) */
684 * One allocation request operates on a zonelist. A zonelist
685 * is a list of zones, the first one is the 'goal' of the
686 * allocation, the other zones are fallback zones, in decreasing
689 * If zlcache_ptr is not NULL, then it is just the address of zlcache,
690 * as explained above. If zlcache_ptr is NULL, there is no zlcache.
692 * To speed the reading of the zonelist, the zonerefs contain the zone index
693 * of the entry being read. Helper functions to access information given
694 * a struct zoneref are
696 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
697 * zonelist_zone_idx() - Return the index of the zone for an entry
698 * zonelist_node_idx() - Return the index of the node for an entry
701 struct zonelist_cache *zlcache_ptr; // NULL or &zlcache
702 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
704 struct zonelist_cache zlcache; // optional ...
708 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
709 struct node_active_region {
710 unsigned long start_pfn;
711 unsigned long end_pfn;
714 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
716 #ifndef CONFIG_DISCONTIGMEM
717 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
718 extern struct page *mem_map;
722 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
723 * (mostly NUMA machines?) to denote a higher-level memory zone than the
726 * On NUMA machines, each NUMA node would have a pg_data_t to describe
727 * it's memory layout.
729 * Memory statistics and page replacement data structures are maintained on a
733 typedef struct pglist_data {
734 struct zone node_zones[MAX_NR_ZONES];
735 struct zonelist node_zonelists[MAX_ZONELISTS];
737 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
738 struct page *node_mem_map;
740 struct page_cgroup *node_page_cgroup;
743 #ifndef CONFIG_NO_BOOTMEM
744 struct bootmem_data *bdata;
746 #ifdef CONFIG_MEMORY_HOTPLUG
748 * Must be held any time you expect node_start_pfn, node_present_pages
749 * or node_spanned_pages stay constant. Holding this will also
750 * guarantee that any pfn_valid() stays that way.
752 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
753 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG.
755 * Nests above zone->lock and zone->span_seqlock
757 spinlock_t node_size_lock;
759 unsigned long node_start_pfn;
760 unsigned long node_present_pages; /* total number of physical pages */
761 unsigned long node_spanned_pages; /* total size of physical page
762 range, including holes */
764 nodemask_t reclaim_nodes; /* Nodes allowed to reclaim from */
765 wait_queue_head_t kswapd_wait;
766 wait_queue_head_t pfmemalloc_wait;
767 struct task_struct *kswapd; /* Protected by lock_memory_hotplug() */
768 int kswapd_max_order;
769 enum zone_type classzone_idx;
770 #ifdef CONFIG_NUMA_BALANCING
771 /* Lock serializing the migrate rate limiting window */
772 spinlock_t numabalancing_migrate_lock;
774 /* Rate limiting time interval */
775 unsigned long numabalancing_migrate_next_window;
777 /* Number of pages migrated during the rate limiting time interval */
778 unsigned long numabalancing_migrate_nr_pages;
782 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
783 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
784 #ifdef CONFIG_FLAT_NODE_MEM_MAP
785 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
787 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
789 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
791 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
792 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
794 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
796 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
799 static inline bool pgdat_is_empty(pg_data_t *pgdat)
801 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
804 #include <linux/memory_hotplug.h>
806 extern struct mutex zonelists_mutex;
807 void build_all_zonelists(pg_data_t *pgdat, struct zone *zone);
808 void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx);
809 bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
810 int classzone_idx, int alloc_flags);
811 bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
812 int classzone_idx, int alloc_flags);
813 enum memmap_context {
817 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
819 enum memmap_context context);
821 extern void lruvec_init(struct lruvec *lruvec);
823 static inline struct zone *lruvec_zone(struct lruvec *lruvec)
828 return container_of(lruvec, struct zone, lruvec);
832 #ifdef CONFIG_HAVE_MEMORY_PRESENT
833 void memory_present(int nid, unsigned long start, unsigned long end);
835 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
838 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
839 int local_memory_node(int node_id);
841 static inline int local_memory_node(int node_id) { return node_id; };
844 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
845 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
849 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
851 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
853 static inline int populated_zone(struct zone *zone)
855 return (!!zone->present_pages);
858 extern int movable_zone;
860 static inline int zone_movable_is_highmem(void)
862 #if defined(CONFIG_HIGHMEM) && defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
863 return movable_zone == ZONE_HIGHMEM;
869 static inline int is_highmem_idx(enum zone_type idx)
871 #ifdef CONFIG_HIGHMEM
872 return (idx == ZONE_HIGHMEM ||
873 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
880 * is_highmem - helper function to quickly check if a struct zone is a
881 * highmem zone or not. This is an attempt to keep references
882 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
883 * @zone - pointer to struct zone variable
885 static inline int is_highmem(struct zone *zone)
887 #ifdef CONFIG_HIGHMEM
888 int zone_off = (char *)zone - (char *)zone->zone_pgdat->node_zones;
889 return zone_off == ZONE_HIGHMEM * sizeof(*zone) ||
890 (zone_off == ZONE_MOVABLE * sizeof(*zone) &&
891 zone_movable_is_highmem());
897 /* These two functions are used to setup the per zone pages min values */
899 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
900 void __user *, size_t *, loff_t *);
901 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
902 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
903 void __user *, size_t *, loff_t *);
904 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
905 void __user *, size_t *, loff_t *);
906 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
907 void __user *, size_t *, loff_t *);
908 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
909 void __user *, size_t *, loff_t *);
911 extern int numa_zonelist_order_handler(struct ctl_table *, int,
912 void __user *, size_t *, loff_t *);
913 extern char numa_zonelist_order[];
914 #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
916 #ifndef CONFIG_NEED_MULTIPLE_NODES
918 extern struct pglist_data contig_page_data;
919 #define NODE_DATA(nid) (&contig_page_data)
920 #define NODE_MEM_MAP(nid) mem_map
922 #else /* CONFIG_NEED_MULTIPLE_NODES */
924 #include <asm/mmzone.h>
926 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
928 extern struct pglist_data *first_online_pgdat(void);
929 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
930 extern struct zone *next_zone(struct zone *zone);
933 * for_each_online_pgdat - helper macro to iterate over all online nodes
934 * @pgdat - pointer to a pg_data_t variable
936 #define for_each_online_pgdat(pgdat) \
937 for (pgdat = first_online_pgdat(); \
939 pgdat = next_online_pgdat(pgdat))
941 * for_each_zone - helper macro to iterate over all memory zones
942 * @zone - pointer to struct zone variable
944 * The user only needs to declare the zone variable, for_each_zone
947 #define for_each_zone(zone) \
948 for (zone = (first_online_pgdat())->node_zones; \
950 zone = next_zone(zone))
952 #define for_each_populated_zone(zone) \
953 for (zone = (first_online_pgdat())->node_zones; \
955 zone = next_zone(zone)) \
956 if (!populated_zone(zone)) \
960 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
962 return zoneref->zone;
965 static inline int zonelist_zone_idx(struct zoneref *zoneref)
967 return zoneref->zone_idx;
970 static inline int zonelist_node_idx(struct zoneref *zoneref)
973 /* zone_to_nid not available in this context */
974 return zoneref->zone->node;
977 #endif /* CONFIG_NUMA */
981 * 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
982 * @z - The cursor used as a starting point for the search
983 * @highest_zoneidx - The zone index of the highest zone to return
984 * @nodes - An optional nodemask to filter the zonelist with
985 * @zone - The first suitable zone found is returned via this parameter
987 * This function returns the next zone at or below a given zone index that is
988 * within the allowed nodemask using a cursor as the starting point for the
989 * search. The zoneref returned is a cursor that represents the current zone
990 * being examined. It should be advanced by one before calling
991 * next_zones_zonelist again.
993 struct zoneref *next_zones_zonelist(struct zoneref *z,
994 enum zone_type highest_zoneidx,
999 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1000 * @zonelist - The zonelist to search for a suitable zone
1001 * @highest_zoneidx - The zone index of the highest zone to return
1002 * @nodes - An optional nodemask to filter the zonelist with
1003 * @zone - The first suitable zone found is returned via this parameter
1005 * This function returns the first zone at or below a given zone index that is
1006 * within the allowed nodemask. The zoneref returned is a cursor that can be
1007 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1008 * one before calling.
1010 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1011 enum zone_type highest_zoneidx,
1015 return next_zones_zonelist(zonelist->_zonerefs, highest_zoneidx, nodes,
1020 * 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
1021 * @zone - The current zone in the iterator
1022 * @z - The current pointer within zonelist->zones being iterated
1023 * @zlist - The zonelist being iterated
1024 * @highidx - The zone index of the highest zone to return
1025 * @nodemask - Nodemask allowed by the allocator
1027 * This iterator iterates though all zones at or below a given zone index and
1028 * within a given nodemask
1030 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1031 for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \
1033 z = next_zones_zonelist(++z, highidx, nodemask, &zone)) \
1036 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1037 * @zone - The current zone in the iterator
1038 * @z - The current pointer within zonelist->zones being iterated
1039 * @zlist - The zonelist being iterated
1040 * @highidx - The zone index of the highest zone to return
1042 * This iterator iterates though all zones at or below a given zone index.
1044 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1045 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1047 #ifdef CONFIG_SPARSEMEM
1048 #include <asm/sparsemem.h>
1051 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1052 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1053 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1059 #ifdef CONFIG_FLATMEM
1060 #define pfn_to_nid(pfn) (0)
1063 #ifdef CONFIG_SPARSEMEM
1066 * SECTION_SHIFT #bits space required to store a section #
1068 * PA_SECTION_SHIFT physical address to/from section number
1069 * PFN_SECTION_SHIFT pfn to/from section number
1071 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1072 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1074 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1076 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1077 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1079 #define SECTION_BLOCKFLAGS_BITS \
1080 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1082 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1083 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1086 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
1087 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
1089 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1090 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1094 struct mem_section {
1096 * This is, logically, a pointer to an array of struct
1097 * pages. However, it is stored with some other magic.
1098 * (see sparse.c::sparse_init_one_section())
1100 * Additionally during early boot we encode node id of
1101 * the location of the section here to guide allocation.
1102 * (see sparse.c::memory_present())
1104 * Making it a UL at least makes someone do a cast
1105 * before using it wrong.
1107 unsigned long section_mem_map;
1109 /* See declaration of similar field in struct zone */
1110 unsigned long *pageblock_flags;
1113 * If !SPARSEMEM, pgdat doesn't have page_cgroup pointer. We use
1114 * section. (see memcontrol.h/page_cgroup.h about this.)
1116 struct page_cgroup *page_cgroup;
1120 * WARNING: mem_section must be a power-of-2 in size for the
1121 * calculation and use of SECTION_ROOT_MASK to make sense.
1125 #ifdef CONFIG_SPARSEMEM_EXTREME
1126 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1128 #define SECTIONS_PER_ROOT 1
1131 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1132 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1133 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1135 #ifdef CONFIG_SPARSEMEM_EXTREME
1136 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
1138 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1141 static inline struct mem_section *__nr_to_section(unsigned long nr)
1143 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1145 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1147 extern int __section_nr(struct mem_section* ms);
1148 extern unsigned long usemap_size(void);
1151 * We use the lower bits of the mem_map pointer to store
1152 * a little bit of information. There should be at least
1153 * 3 bits here due to 32-bit alignment.
1155 #define SECTION_MARKED_PRESENT (1UL<<0)
1156 #define SECTION_HAS_MEM_MAP (1UL<<1)
1157 #define SECTION_MAP_LAST_BIT (1UL<<2)
1158 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1159 #define SECTION_NID_SHIFT 2
1161 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1163 unsigned long map = section->section_mem_map;
1164 map &= SECTION_MAP_MASK;
1165 return (struct page *)map;
1168 static inline int present_section(struct mem_section *section)
1170 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1173 static inline int present_section_nr(unsigned long nr)
1175 return present_section(__nr_to_section(nr));
1178 static inline int valid_section(struct mem_section *section)
1180 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1183 static inline int valid_section_nr(unsigned long nr)
1185 return valid_section(__nr_to_section(nr));
1188 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1190 return __nr_to_section(pfn_to_section_nr(pfn));
1193 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1194 static inline int pfn_valid(unsigned long pfn)
1196 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1198 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
1202 static inline int pfn_present(unsigned long pfn)
1204 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1206 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1210 * These are _only_ used during initialisation, therefore they
1211 * can use __initdata ... They could have names to indicate
1215 #define pfn_to_nid(pfn) \
1217 unsigned long __pfn_to_nid_pfn = (pfn); \
1218 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1221 #define pfn_to_nid(pfn) (0)
1224 #define early_pfn_valid(pfn) pfn_valid(pfn)
1225 void sparse_init(void);
1227 #define sparse_init() do {} while (0)
1228 #define sparse_index_init(_sec, _nid) do {} while (0)
1229 #endif /* CONFIG_SPARSEMEM */
1231 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1232 bool early_pfn_in_nid(unsigned long pfn, int nid);
1234 #define early_pfn_in_nid(pfn, nid) (1)
1237 #ifndef early_pfn_valid
1238 #define early_pfn_valid(pfn) (1)
1241 void memory_present(int nid, unsigned long start, unsigned long end);
1242 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
1245 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1246 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1247 * pfn_valid_within() should be used in this case; we optimise this away
1248 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1250 #ifdef CONFIG_HOLES_IN_ZONE
1251 #define pfn_valid_within(pfn) pfn_valid(pfn)
1253 #define pfn_valid_within(pfn) (1)
1256 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1258 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1259 * associated with it or not. In FLATMEM, it is expected that holes always
1260 * have valid memmap as long as there is valid PFNs either side of the hole.
1261 * In SPARSEMEM, it is assumed that a valid section has a memmap for the
1264 * However, an ARM, and maybe other embedded architectures in the future
1265 * free memmap backing holes to save memory on the assumption the memmap is
1266 * never used. The page_zone linkages are then broken even though pfn_valid()
1267 * returns true. A walker of the full memmap must then do this additional
1268 * check to ensure the memmap they are looking at is sane by making sure
1269 * the zone and PFN linkages are still valid. This is expensive, but walkers
1270 * of the full memmap are extremely rare.
1272 int memmap_valid_within(unsigned long pfn,
1273 struct page *page, struct zone *zone);
1275 static inline int memmap_valid_within(unsigned long pfn,
1276 struct page *page, struct zone *zone)
1280 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1282 #endif /* !__GENERATING_BOUNDS.H */
1283 #endif /* !__ASSEMBLY__ */
1284 #endif /* _LINUX_MMZONE_H */