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
44 * MIGRATE_CMA migration type is designed to mimic the way
45 * ZONE_MOVABLE works. Only movable pages can be allocated
46 * from MIGRATE_CMA pageblocks and page allocator never
47 * implicitly change migration type of MIGRATE_CMA pageblock.
49 * The way to use it is to change migratetype of a range of
50 * pageblocks to MIGRATE_CMA which can be done by
51 * __free_pageblock_cma() function. What is important though
52 * is that a range of pageblocks must be aligned to
53 * MAX_ORDER_NR_PAGES should biggest page be bigger then
58 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
59 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
60 #ifdef CONFIG_MEMORY_ISOLATION
61 MIGRATE_ISOLATE, /* can't allocate from here */
67 * Returns a list which contains the migrate types on to which
68 * an allocation falls back when the free list for the migrate
69 * type mtype is depleted.
70 * The end of the list is delimited by the type MIGRATE_TYPES.
72 extern int *get_migratetype_fallbacks(int mtype);
74 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
75 extern char * const migratetype_names[MIGRATE_TYPES];
78 bool is_cma_pageblock(struct page *page);
79 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
80 # define get_cma_migrate_type() MIGRATE_CMA
81 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
83 # define is_cma_pageblock(page) false
84 # define is_migrate_cma(migratetype) false
85 # define get_cma_migrate_type() MIGRATE_MOVABLE
86 # define is_migrate_cma_page(_page) false
89 #define for_each_migratetype_order(order, type) \
90 for (order = 0; order < MAX_ORDER; order++) \
91 for (type = 0; type < MIGRATE_TYPES; type++)
93 extern int page_group_by_mobility_disabled;
95 #define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1)
96 #define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1)
98 #define get_pageblock_migratetype(page) \
99 get_pfnblock_flags_mask(page, page_to_pfn(page), \
100 PB_migrate_end, MIGRATETYPE_MASK)
102 static inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn)
104 BUILD_BUG_ON(PB_migrate_end - PB_migrate != 2);
105 return get_pfnblock_flags_mask(page, pfn, PB_migrate_end,
110 struct list_head free_list[MIGRATE_TYPES];
111 unsigned long nr_free;
117 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
118 * So add a wild amount of padding here to ensure that they fall into separate
119 * cachelines. There are very few zone structures in the machine, so space
120 * consumption is not a concern here.
122 #if defined(CONFIG_SMP)
123 struct zone_padding {
125 } ____cacheline_internodealigned_in_smp;
126 #define ZONE_PADDING(name) struct zone_padding name;
128 #define ZONE_PADDING(name)
131 enum zone_stat_item {
132 /* First 128 byte cacheline (assuming 64 bit words) */
136 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
137 NR_ACTIVE_ANON, /* " " " " " */
138 NR_INACTIVE_FILE, /* " " " " " */
139 NR_ACTIVE_FILE, /* " " " " " */
140 NR_UNEVICTABLE, /* " " " " " */
141 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
142 NR_ANON_PAGES, /* Mapped anonymous pages */
143 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
144 only modified from process context */
149 NR_SLAB_UNRECLAIMABLE,
150 NR_PAGETABLE, /* used for pagetables */
151 /* Second 128 byte cacheline */
154 NR_UNSTABLE_NFS, /* NFS unstable pages */
157 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
158 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
159 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
160 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
161 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
162 NR_DIRTIED, /* page dirtyings since bootup */
163 NR_WRITTEN, /* page writings since bootup */
164 NR_PAGES_SCANNED, /* pages scanned since last reclaim */
166 NUMA_HIT, /* allocated in intended node */
167 NUMA_MISS, /* allocated in non intended node */
168 NUMA_FOREIGN, /* was intended here, hit elsewhere */
169 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
170 NUMA_LOCAL, /* allocation from local node */
171 NUMA_OTHER, /* allocation from other node */
175 WORKINGSET_NODERECLAIM,
176 NR_ANON_TRANSPARENT_HUGEPAGES,
179 NR_INDIRECTLY_RECLAIMABLE_BYTES, /* measured in bytes */
180 NR_VM_ZONE_STAT_ITEMS };
183 * We do arithmetic on the LRU lists in various places in the code,
184 * so it is important to keep the active lists LRU_ACTIVE higher in
185 * the array than the corresponding inactive lists, and to keep
186 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
188 * This has to be kept in sync with the statistics in zone_stat_item
189 * above and the descriptions in vmstat_text in mm/vmstat.c
196 LRU_INACTIVE_ANON = LRU_BASE,
197 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
198 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
199 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
204 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
206 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
208 static inline int is_file_lru(enum lru_list lru)
210 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
213 static inline int is_active_lru(enum lru_list lru)
215 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
218 static inline int is_unevictable_lru(enum lru_list lru)
220 return (lru == LRU_UNEVICTABLE);
223 struct zone_reclaim_stat {
225 * The pageout code in vmscan.c keeps track of how many of the
226 * mem/swap backed and file backed pages are referenced.
227 * The higher the rotated/scanned ratio, the more valuable
230 * The anon LRU stats live in [0], file LRU stats in [1]
232 unsigned long recent_rotated[2];
233 unsigned long recent_scanned[2];
237 struct list_head lists[NR_LRU_LISTS];
238 struct zone_reclaim_stat reclaim_stat;
244 /* Mask used at gathering information at once (see memcontrol.c) */
245 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
246 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
247 #define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
249 /* Isolate clean file */
250 #define ISOLATE_CLEAN ((__force isolate_mode_t)0x1)
251 /* Isolate unmapped file */
252 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
253 /* Isolate for asynchronous migration */
254 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
255 /* Isolate unevictable pages */
256 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
258 /* LRU Isolation modes. */
259 typedef unsigned __bitwise__ isolate_mode_t;
261 enum zone_watermarks {
268 #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
269 #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
270 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
272 struct per_cpu_pages {
273 int count; /* number of pages in the list */
274 int high; /* high watermark, emptying needed */
275 int batch; /* chunk size for buddy add/remove */
277 /* Lists of pages, one per migrate type stored on the pcp-lists */
278 struct list_head lists[MIGRATE_PCPTYPES];
281 struct per_cpu_pageset {
282 struct per_cpu_pages pcp;
288 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
292 #endif /* !__GENERATING_BOUNDS.H */
295 #ifdef CONFIG_ZONE_DMA
297 * ZONE_DMA is used when there are devices that are not able
298 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
299 * carve out the portion of memory that is needed for these devices.
300 * The range is arch specific.
305 * ---------------------------
306 * parisc, ia64, sparc <4G
309 * alpha Unlimited or 0-16MB.
311 * i386, x86_64 and multiple other arches
316 #ifdef CONFIG_ZONE_DMA32
318 * x86_64 needs two ZONE_DMAs because it supports devices that are
319 * only able to do DMA to the lower 16M but also 32 bit devices that
320 * can only do DMA areas below 4G.
325 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
326 * performed on pages in ZONE_NORMAL if the DMA devices support
327 * transfers to all addressable memory.
330 #ifdef CONFIG_HIGHMEM
332 * A memory area that is only addressable by the kernel through
333 * mapping portions into its own address space. This is for example
334 * used by i386 to allow the kernel to address the memory beyond
335 * 900MB. The kernel will set up special mappings (page
336 * table entries on i386) for each page that the kernel needs to
342 #ifdef CONFIG_ZONE_DEVICE
349 #ifndef __GENERATING_BOUNDS_H
352 /* Read-mostly fields */
354 /* zone watermarks, access with *_wmark_pages(zone) macros */
355 unsigned long watermark[NR_WMARK];
357 unsigned long nr_reserved_highatomic;
360 * We don't know if the memory that we're going to allocate will be
361 * freeable or/and it will be released eventually, so to avoid totally
362 * wasting several GB of ram we must reserve some of the lower zone
363 * memory (otherwise we risk to run OOM on the lower zones despite
364 * there being tons of freeable ram on the higher zones). This array is
365 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
368 long lowmem_reserve[MAX_NR_ZONES];
375 * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on
376 * this zone's LRU. Maintained by the pageout code.
378 unsigned int inactive_ratio;
380 struct pglist_data *zone_pgdat;
381 struct per_cpu_pageset __percpu *pageset;
384 * This is a per-zone reserve of pages that are not available
385 * to userspace allocations.
387 unsigned long totalreserve_pages;
392 #ifndef CONFIG_SPARSEMEM
394 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
395 * In SPARSEMEM, this map is stored in struct mem_section
397 unsigned long *pageblock_flags;
398 #endif /* CONFIG_SPARSEMEM */
402 * zone reclaim becomes active if more unmapped pages exist.
404 unsigned long min_unmapped_pages;
405 unsigned long min_slab_pages;
406 #endif /* CONFIG_NUMA */
408 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
409 unsigned long zone_start_pfn;
412 * spanned_pages is the total pages spanned by the zone, including
413 * holes, which is calculated as:
414 * spanned_pages = zone_end_pfn - zone_start_pfn;
416 * present_pages is physical pages existing within the zone, which
418 * present_pages = spanned_pages - absent_pages(pages in holes);
420 * managed_pages is present pages managed by the buddy system, which
421 * is calculated as (reserved_pages includes pages allocated by the
422 * bootmem allocator):
423 * managed_pages = present_pages - reserved_pages;
425 * So present_pages may be used by memory hotplug or memory power
426 * management logic to figure out unmanaged pages by checking
427 * (present_pages - managed_pages). And managed_pages should be used
428 * by page allocator and vm scanner to calculate all kinds of watermarks
433 * zone_start_pfn and spanned_pages are protected by span_seqlock.
434 * It is a seqlock because it has to be read outside of zone->lock,
435 * and it is done in the main allocator path. But, it is written
436 * quite infrequently.
438 * The span_seq lock is declared along with zone->lock because it is
439 * frequently read in proximity to zone->lock. It's good to
440 * give them a chance of being in the same cacheline.
442 * Write access to present_pages at runtime should be protected by
443 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
444 * present_pages should get_online_mems() to get a stable value.
446 * Read access to managed_pages should be safe because it's unsigned
447 * long. Write access to zone->managed_pages and totalram_pages are
448 * protected by managed_page_count_lock at runtime. Idealy only
449 * adjust_managed_page_count() should be used instead of directly
450 * touching zone->managed_pages and totalram_pages.
452 unsigned long managed_pages;
453 unsigned long spanned_pages;
454 unsigned long present_pages;
458 #ifdef CONFIG_MEMORY_ISOLATION
460 * Number of isolated pageblock. It is used to solve incorrect
461 * freepage counting problem due to racy retrieving migratetype
462 * of pageblock. Protected by zone->lock.
464 unsigned long nr_isolate_pageblock;
467 #ifdef CONFIG_MEMORY_HOTPLUG
468 /* see spanned/present_pages for more description */
469 seqlock_t span_seqlock;
473 * wait_table -- the array holding the hash table
474 * wait_table_hash_nr_entries -- the size of the hash table array
475 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
477 * The purpose of all these is to keep track of the people
478 * waiting for a page to become available and make them
479 * runnable again when possible. The trouble is that this
480 * consumes a lot of space, especially when so few things
481 * wait on pages at a given time. So instead of using
482 * per-page waitqueues, we use a waitqueue hash table.
484 * The bucket discipline is to sleep on the same queue when
485 * colliding and wake all in that wait queue when removing.
486 * When something wakes, it must check to be sure its page is
487 * truly available, a la thundering herd. The cost of a
488 * collision is great, but given the expected load of the
489 * table, they should be so rare as to be outweighed by the
490 * benefits from the saved space.
492 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
493 * primary users of these fields, and in mm/page_alloc.c
494 * free_area_init_core() performs the initialization of them.
496 wait_queue_head_t *wait_table;
497 unsigned long wait_table_hash_nr_entries;
498 unsigned long wait_table_bits;
501 /* free areas of different sizes */
502 struct free_area free_area[MAX_ORDER];
504 /* zone flags, see below */
507 /* Write-intensive fields used from the page allocator */
512 /* Write-intensive fields used by page reclaim */
514 /* Fields commonly accessed by the page reclaim scanner */
516 struct lruvec lruvec;
518 /* Evictions & activations on the inactive file list */
519 atomic_long_t inactive_age;
522 * When free pages are below this point, additional steps are taken
523 * when reading the number of free pages to avoid per-cpu counter
524 * drift allowing watermarks to be breached
526 unsigned long percpu_drift_mark;
528 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
529 /* pfn where compaction free scanner should start */
530 unsigned long compact_cached_free_pfn;
531 /* pfn where async and sync compaction migration scanner should start */
532 unsigned long compact_cached_migrate_pfn[2];
535 #ifdef CONFIG_COMPACTION
537 * On compaction failure, 1<<compact_defer_shift compactions
538 * are skipped before trying again. The number attempted since
539 * last failure is tracked with compact_considered.
541 unsigned int compact_considered;
542 unsigned int compact_defer_shift;
543 int compact_order_failed;
546 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
547 /* Set to true when the PG_migrate_skip bits should be cleared */
548 bool compact_blockskip_flush;
552 /* Zone statistics */
553 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
554 } ____cacheline_internodealigned_in_smp;
557 ZONE_RECLAIM_LOCKED, /* prevents concurrent reclaim */
558 ZONE_OOM_LOCKED, /* zone is in OOM killer zonelist */
559 ZONE_CONGESTED, /* zone has many dirty pages backed by
562 ZONE_DIRTY, /* reclaim scanning has recently found
563 * many dirty file pages at the tail
566 ZONE_WRITEBACK, /* reclaim scanning has recently found
567 * many pages under writeback
569 ZONE_FAIR_DEPLETED, /* fair zone policy batch depleted */
572 static inline unsigned long zone_end_pfn(const struct zone *zone)
574 return zone->zone_start_pfn + zone->spanned_pages;
577 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
579 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
582 static inline bool zone_is_initialized(struct zone *zone)
584 return !!zone->wait_table;
587 static inline bool zone_is_empty(struct zone *zone)
589 return zone->spanned_pages == 0;
593 * The "priority" of VM scanning is how much of the queues we will scan in one
594 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
595 * queues ("queue_length >> 12") during an aging round.
597 #define DEF_PRIORITY 12
599 /* Maximum number of zones on a zonelist */
600 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
605 * The NUMA zonelists are doubled because we need zonelists that restrict the
606 * allocations to a single node for __GFP_THISNODE.
608 * [0] : Zonelist with fallback
609 * [1] : No fallback (__GFP_THISNODE)
611 #define MAX_ZONELISTS 2
613 #define MAX_ZONELISTS 1
617 * This struct contains information about a zone in a zonelist. It is stored
618 * here to avoid dereferences into large structures and lookups of tables
621 struct zone *zone; /* Pointer to actual zone */
622 int zone_idx; /* zone_idx(zoneref->zone) */
626 * One allocation request operates on a zonelist. A zonelist
627 * is a list of zones, the first one is the 'goal' of the
628 * allocation, the other zones are fallback zones, in decreasing
631 * To speed the reading of the zonelist, the zonerefs contain the zone index
632 * of the entry being read. Helper functions to access information given
633 * a struct zoneref are
635 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
636 * zonelist_zone_idx() - Return the index of the zone for an entry
637 * zonelist_node_idx() - Return the index of the node for an entry
640 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
643 #ifndef CONFIG_DISCONTIGMEM
644 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
645 extern struct page *mem_map;
649 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
650 * (mostly NUMA machines?) to denote a higher-level memory zone than the
653 * On NUMA machines, each NUMA node would have a pg_data_t to describe
654 * it's memory layout.
656 * Memory statistics and page replacement data structures are maintained on a
660 typedef struct pglist_data {
661 struct zone node_zones[MAX_NR_ZONES];
662 struct zonelist node_zonelists[MAX_ZONELISTS];
664 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
665 struct page *node_mem_map;
666 #ifdef CONFIG_PAGE_EXTENSION
667 struct page_ext *node_page_ext;
670 #ifndef CONFIG_NO_BOOTMEM
671 struct bootmem_data *bdata;
673 #ifdef CONFIG_MEMORY_HOTPLUG
675 * Must be held any time you expect node_start_pfn, node_present_pages
676 * or node_spanned_pages stay constant. Holding this will also
677 * guarantee that any pfn_valid() stays that way.
679 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
680 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG.
682 * Nests above zone->lock and zone->span_seqlock
684 spinlock_t node_size_lock;
686 unsigned long node_start_pfn;
687 unsigned long node_present_pages; /* total number of physical pages */
688 unsigned long node_spanned_pages; /* total size of physical page
689 range, including holes */
691 wait_queue_head_t kswapd_wait;
692 wait_queue_head_t pfmemalloc_wait;
693 struct task_struct *kswapd; /* Protected by
694 mem_hotplug_begin/end() */
695 int kswapd_max_order;
696 enum zone_type classzone_idx;
697 #ifdef CONFIG_COMPACTION
698 int kcompactd_max_order;
699 enum zone_type kcompactd_classzone_idx;
700 wait_queue_head_t kcompactd_wait;
701 struct task_struct *kcompactd;
703 #ifdef CONFIG_NUMA_BALANCING
704 /* Lock serializing the migrate rate limiting window */
705 spinlock_t numabalancing_migrate_lock;
707 /* Rate limiting time interval */
708 unsigned long numabalancing_migrate_next_window;
710 /* Number of pages migrated during the rate limiting time interval */
711 unsigned long numabalancing_migrate_nr_pages;
714 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
716 * If memory initialisation on large machines is deferred then this
717 * is the first PFN that needs to be initialised.
719 unsigned long first_deferred_pfn;
720 /* Number of non-deferred pages */
721 unsigned long static_init_pgcnt;
722 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
725 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
726 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
727 #ifdef CONFIG_FLAT_NODE_MEM_MAP
728 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
730 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
732 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
734 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
735 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
737 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
739 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
742 static inline bool pgdat_is_empty(pg_data_t *pgdat)
744 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
747 static inline int zone_id(const struct zone *zone)
749 struct pglist_data *pgdat = zone->zone_pgdat;
751 return zone - pgdat->node_zones;
754 #ifdef CONFIG_ZONE_DEVICE
755 static inline bool is_dev_zone(const struct zone *zone)
757 return zone_id(zone) == ZONE_DEVICE;
760 static inline bool is_dev_zone(const struct zone *zone)
766 #include <linux/memory_hotplug.h>
768 extern struct mutex zonelists_mutex;
769 void build_all_zonelists(pg_data_t *pgdat, struct zone *zone);
770 void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx);
771 bool zone_watermark_ok(struct zone *z, unsigned int order,
772 unsigned long mark, int classzone_idx, int alloc_flags);
773 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
774 unsigned long mark, int classzone_idx);
775 enum memmap_context {
779 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
782 extern void lruvec_init(struct lruvec *lruvec);
784 static inline struct zone *lruvec_zone(struct lruvec *lruvec)
789 return container_of(lruvec, struct zone, lruvec);
793 #ifdef CONFIG_HAVE_MEMORY_PRESENT
794 void memory_present(int nid, unsigned long start, unsigned long end);
796 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
799 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
800 int local_memory_node(int node_id);
802 static inline int local_memory_node(int node_id) { return node_id; };
805 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
806 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
810 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
812 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
814 static inline int populated_zone(struct zone *zone)
816 return (!!zone->present_pages);
819 extern int movable_zone;
821 #ifdef CONFIG_HIGHMEM
822 static inline int zone_movable_is_highmem(void)
824 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
825 return movable_zone == ZONE_HIGHMEM;
827 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
832 static inline int is_highmem_idx(enum zone_type idx)
834 #ifdef CONFIG_HIGHMEM
835 return (idx == ZONE_HIGHMEM ||
836 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
843 * is_highmem - helper function to quickly check if a struct zone is a
844 * highmem zone or not. This is an attempt to keep references
845 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
846 * @zone - pointer to struct zone variable
848 static inline int is_highmem(struct zone *zone)
850 #ifdef CONFIG_HIGHMEM
851 int zone_off = (char *)zone - (char *)zone->zone_pgdat->node_zones;
852 return zone_off == ZONE_HIGHMEM * sizeof(*zone) ||
853 (zone_off == ZONE_MOVABLE * sizeof(*zone) &&
854 zone_movable_is_highmem());
860 /* These two functions are used to setup the per zone pages min values */
862 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
863 void __user *, size_t *, loff_t *);
864 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
865 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
866 void __user *, size_t *, loff_t *);
867 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
868 void __user *, size_t *, loff_t *);
869 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
870 void __user *, size_t *, loff_t *);
871 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
872 void __user *, size_t *, loff_t *);
874 extern int numa_zonelist_order_handler(struct ctl_table *, int,
875 void __user *, size_t *, loff_t *);
876 extern char numa_zonelist_order[];
877 #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
879 #ifndef CONFIG_NEED_MULTIPLE_NODES
881 extern struct pglist_data contig_page_data;
882 #define NODE_DATA(nid) (&contig_page_data)
883 #define NODE_MEM_MAP(nid) mem_map
885 #else /* CONFIG_NEED_MULTIPLE_NODES */
887 #include <asm/mmzone.h>
889 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
891 extern struct pglist_data *first_online_pgdat(void);
892 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
893 extern struct zone *next_zone(struct zone *zone);
896 * for_each_online_pgdat - helper macro to iterate over all online nodes
897 * @pgdat - pointer to a pg_data_t variable
899 #define for_each_online_pgdat(pgdat) \
900 for (pgdat = first_online_pgdat(); \
902 pgdat = next_online_pgdat(pgdat))
904 * for_each_zone - helper macro to iterate over all memory zones
905 * @zone - pointer to struct zone variable
907 * The user only needs to declare the zone variable, for_each_zone
910 #define for_each_zone(zone) \
911 for (zone = (first_online_pgdat())->node_zones; \
913 zone = next_zone(zone))
915 #define for_each_populated_zone(zone) \
916 for (zone = (first_online_pgdat())->node_zones; \
918 zone = next_zone(zone)) \
919 if (!populated_zone(zone)) \
923 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
925 return zoneref->zone;
928 static inline int zonelist_zone_idx(struct zoneref *zoneref)
930 return zoneref->zone_idx;
933 static inline int zonelist_node_idx(struct zoneref *zoneref)
936 /* zone_to_nid not available in this context */
937 return zoneref->zone->node;
940 #endif /* CONFIG_NUMA */
944 * 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
945 * @z - The cursor used as a starting point for the search
946 * @highest_zoneidx - The zone index of the highest zone to return
947 * @nodes - An optional nodemask to filter the zonelist with
949 * This function returns the next zone at or below a given zone index that is
950 * within the allowed nodemask using a cursor as the starting point for the
951 * search. The zoneref returned is a cursor that represents the current zone
952 * being examined. It should be advanced by one before calling
953 * next_zones_zonelist again.
955 struct zoneref *next_zones_zonelist(struct zoneref *z,
956 enum zone_type highest_zoneidx,
960 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
961 * @zonelist - The zonelist to search for a suitable zone
962 * @highest_zoneidx - The zone index of the highest zone to return
963 * @nodes - An optional nodemask to filter the zonelist with
964 * @zone - The first suitable zone found is returned via this parameter
966 * This function returns the first zone at or below a given zone index that is
967 * within the allowed nodemask. The zoneref returned is a cursor that can be
968 * used to iterate the zonelist with next_zones_zonelist by advancing it by
969 * one before calling.
971 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
972 enum zone_type highest_zoneidx,
976 struct zoneref *z = next_zones_zonelist(zonelist->_zonerefs,
977 highest_zoneidx, nodes);
978 *zone = zonelist_zone(z);
983 * 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
984 * @zone - The current zone in the iterator
985 * @z - The current pointer within zonelist->zones being iterated
986 * @zlist - The zonelist being iterated
987 * @highidx - The zone index of the highest zone to return
988 * @nodemask - Nodemask allowed by the allocator
990 * This iterator iterates though all zones at or below a given zone index and
991 * within a given nodemask
993 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
994 for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \
996 z = next_zones_zonelist(++z, highidx, nodemask), \
997 zone = zonelist_zone(z)) \
1000 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1001 * @zone - The current zone in the iterator
1002 * @z - The current pointer within zonelist->zones being iterated
1003 * @zlist - The zonelist being iterated
1004 * @highidx - The zone index of the highest zone to return
1006 * This iterator iterates though all zones at or below a given zone index.
1008 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1009 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1011 #ifdef CONFIG_SPARSEMEM
1012 #include <asm/sparsemem.h>
1015 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1016 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1017 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1023 #ifdef CONFIG_FLATMEM
1024 #define pfn_to_nid(pfn) (0)
1027 #ifdef CONFIG_SPARSEMEM
1030 * SECTION_SHIFT #bits space required to store a section #
1032 * PA_SECTION_SHIFT physical address to/from section number
1033 * PFN_SECTION_SHIFT pfn to/from section number
1035 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1036 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1038 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1040 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1041 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1043 #define SECTION_BLOCKFLAGS_BITS \
1044 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1046 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1047 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1050 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
1051 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
1053 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1054 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1058 struct mem_section {
1060 * This is, logically, a pointer to an array of struct
1061 * pages. However, it is stored with some other magic.
1062 * (see sparse.c::sparse_init_one_section())
1064 * Additionally during early boot we encode node id of
1065 * the location of the section here to guide allocation.
1066 * (see sparse.c::memory_present())
1068 * Making it a UL at least makes someone do a cast
1069 * before using it wrong.
1071 unsigned long section_mem_map;
1073 /* See declaration of similar field in struct zone */
1074 unsigned long *pageblock_flags;
1075 #ifdef CONFIG_PAGE_EXTENSION
1077 * If !SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1078 * section. (see page_ext.h about this.)
1080 struct page_ext *page_ext;
1084 * WARNING: mem_section must be a power-of-2 in size for the
1085 * calculation and use of SECTION_ROOT_MASK to make sense.
1089 #ifdef CONFIG_SPARSEMEM_EXTREME
1090 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1092 #define SECTIONS_PER_ROOT 1
1095 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1096 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1097 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1099 #ifdef CONFIG_SPARSEMEM_EXTREME
1100 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
1102 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1105 static inline struct mem_section *__nr_to_section(unsigned long nr)
1107 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1109 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1111 extern int __section_nr(struct mem_section* ms);
1112 extern unsigned long usemap_size(void);
1115 * We use the lower bits of the mem_map pointer to store
1116 * a little bit of information. There should be at least
1117 * 3 bits here due to 32-bit alignment.
1119 #define SECTION_MARKED_PRESENT (1UL<<0)
1120 #define SECTION_HAS_MEM_MAP (1UL<<1)
1121 #define SECTION_MAP_LAST_BIT (1UL<<2)
1122 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1123 #define SECTION_NID_SHIFT 2
1125 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1127 unsigned long map = section->section_mem_map;
1128 map &= SECTION_MAP_MASK;
1129 return (struct page *)map;
1132 static inline int present_section(struct mem_section *section)
1134 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1137 static inline int present_section_nr(unsigned long nr)
1139 return present_section(__nr_to_section(nr));
1142 static inline int valid_section(struct mem_section *section)
1144 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1147 static inline int valid_section_nr(unsigned long nr)
1149 return valid_section(__nr_to_section(nr));
1152 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1154 return __nr_to_section(pfn_to_section_nr(pfn));
1157 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1158 static inline int pfn_valid(unsigned long pfn)
1160 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1162 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
1166 static inline int pfn_present(unsigned long pfn)
1168 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1170 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1174 * These are _only_ used during initialisation, therefore they
1175 * can use __initdata ... They could have names to indicate
1179 #define pfn_to_nid(pfn) \
1181 unsigned long __pfn_to_nid_pfn = (pfn); \
1182 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1185 #define pfn_to_nid(pfn) (0)
1188 #define early_pfn_valid(pfn) pfn_valid(pfn)
1189 void sparse_init(void);
1191 #define sparse_init() do {} while (0)
1192 #define sparse_index_init(_sec, _nid) do {} while (0)
1193 #endif /* CONFIG_SPARSEMEM */
1196 * During memory init memblocks map pfns to nids. The search is expensive and
1197 * this caches recent lookups. The implementation of __early_pfn_to_nid
1198 * may treat start/end as pfns or sections.
1200 struct mminit_pfnnid_cache {
1201 unsigned long last_start;
1202 unsigned long last_end;
1206 #ifndef early_pfn_valid
1207 #define early_pfn_valid(pfn) (1)
1210 void memory_present(int nid, unsigned long start, unsigned long end);
1211 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
1214 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1215 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1216 * pfn_valid_within() should be used in this case; we optimise this away
1217 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1219 #ifdef CONFIG_HOLES_IN_ZONE
1220 #define pfn_valid_within(pfn) pfn_valid(pfn)
1222 #define pfn_valid_within(pfn) (1)
1225 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1227 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1228 * associated with it or not. In FLATMEM, it is expected that holes always
1229 * have valid memmap as long as there is valid PFNs either side of the hole.
1230 * In SPARSEMEM, it is assumed that a valid section has a memmap for the
1233 * However, an ARM, and maybe other embedded architectures in the future
1234 * free memmap backing holes to save memory on the assumption the memmap is
1235 * never used. The page_zone linkages are then broken even though pfn_valid()
1236 * returns true. A walker of the full memmap must then do this additional
1237 * check to ensure the memmap they are looking at is sane by making sure
1238 * the zone and PFN linkages are still valid. This is expensive, but walkers
1239 * of the full memmap are extremely rare.
1241 int memmap_valid_within(unsigned long pfn,
1242 struct page *page, struct zone *zone);
1244 static inline int memmap_valid_within(unsigned long pfn,
1245 struct page *page, struct zone *zone)
1249 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1251 #endif /* !__GENERATING_BOUNDS.H */
1252 #endif /* !__ASSEMBLY__ */
1253 #endif /* _LINUX_MMZONE_H */