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_HIGHATOMIC = 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 */
66 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
67 extern char * const migratetype_names[MIGRATE_TYPES];
70 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
71 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
73 # define is_migrate_cma(migratetype) false
74 # define is_migrate_cma_page(_page) false
77 #define for_each_migratetype_order(order, type) \
78 for (order = 0; order < MAX_ORDER; order++) \
79 for (type = 0; type < MIGRATE_TYPES; type++)
81 extern int page_group_by_mobility_disabled;
83 #define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1)
84 #define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1)
86 #define get_pageblock_migratetype(page) \
87 get_pfnblock_flags_mask(page, page_to_pfn(page), \
88 PB_migrate_end, MIGRATETYPE_MASK)
91 struct list_head free_list[MIGRATE_TYPES];
92 unsigned long nr_free;
98 * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
99 * So add a wild amount of padding here to ensure that they fall into separate
100 * cachelines. There are very few zone structures in the machine, so space
101 * consumption is not a concern here.
103 #if defined(CONFIG_SMP)
104 struct zone_padding {
106 } ____cacheline_internodealigned_in_smp;
107 #define ZONE_PADDING(name) struct zone_padding name;
109 #define ZONE_PADDING(name)
112 enum zone_stat_item {
113 /* First 128 byte cacheline (assuming 64 bit words) */
115 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
116 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
118 NR_ZONE_INACTIVE_FILE,
121 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
122 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
124 NR_SLAB_UNRECLAIMABLE,
125 NR_PAGETABLE, /* used for pagetables */
126 NR_KERNEL_STACK_KB, /* measured in KiB */
127 /* Second 128 byte cacheline */
129 #if IS_ENABLED(CONFIG_ZSMALLOC)
130 NR_ZSPAGES, /* allocated in zsmalloc */
133 NUMA_HIT, /* allocated in intended node */
134 NUMA_MISS, /* allocated in non intended node */
135 NUMA_FOREIGN, /* was intended here, hit elsewhere */
136 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
137 NUMA_LOCAL, /* allocation from local node */
138 NUMA_OTHER, /* allocation from other node */
141 NR_VM_ZONE_STAT_ITEMS };
143 enum node_stat_item {
145 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
146 NR_ACTIVE_ANON, /* " " " " " */
147 NR_INACTIVE_FILE, /* " " " " " */
148 NR_ACTIVE_FILE, /* " " " " " */
149 NR_UNEVICTABLE, /* " " " " " */
150 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
151 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
152 NR_PAGES_SCANNED, /* pages scanned since last reclaim */
155 WORKINGSET_NODERECLAIM,
156 NR_ANON_MAPPED, /* Mapped anonymous pages */
157 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
158 only modified from process context */
162 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
163 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
167 NR_UNSTABLE_NFS, /* NFS unstable pages */
169 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
170 NR_DIRTIED, /* page dirtyings since bootup */
171 NR_WRITTEN, /* page writings since bootup */
172 NR_VM_NODE_STAT_ITEMS
176 * We do arithmetic on the LRU lists in various places in the code,
177 * so it is important to keep the active lists LRU_ACTIVE higher in
178 * the array than the corresponding inactive lists, and to keep
179 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
181 * This has to be kept in sync with the statistics in zone_stat_item
182 * above and the descriptions in vmstat_text in mm/vmstat.c
189 LRU_INACTIVE_ANON = LRU_BASE,
190 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
191 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
192 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
197 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
199 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
201 static inline int is_file_lru(enum lru_list lru)
203 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
206 static inline int is_active_lru(enum lru_list lru)
208 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
211 struct zone_reclaim_stat {
213 * The pageout code in vmscan.c keeps track of how many of the
214 * mem/swap backed and file backed pages are referenced.
215 * The higher the rotated/scanned ratio, the more valuable
218 * The anon LRU stats live in [0], file LRU stats in [1]
220 unsigned long recent_rotated[2];
221 unsigned long recent_scanned[2];
225 struct list_head lists[NR_LRU_LISTS];
226 struct zone_reclaim_stat reclaim_stat;
227 /* Evictions & activations on the inactive file list */
228 atomic_long_t inactive_age;
230 struct pglist_data *pgdat;
234 /* Mask used at gathering information at once (see memcontrol.c) */
235 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
236 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
237 #define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
239 /* Isolate clean file */
240 #define ISOLATE_CLEAN ((__force isolate_mode_t)0x1)
241 /* Isolate unmapped file */
242 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
243 /* Isolate for asynchronous migration */
244 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
245 /* Isolate unevictable pages */
246 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
248 /* LRU Isolation modes. */
249 typedef unsigned __bitwise__ isolate_mode_t;
251 enum zone_watermarks {
258 #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
259 #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
260 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
262 struct per_cpu_pages {
263 int count; /* number of pages in the list */
264 int high; /* high watermark, emptying needed */
265 int batch; /* chunk size for buddy add/remove */
267 /* Lists of pages, one per migrate type stored on the pcp-lists */
268 struct list_head lists[MIGRATE_PCPTYPES];
271 struct per_cpu_pageset {
272 struct per_cpu_pages pcp;
278 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
282 struct per_cpu_nodestat {
284 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
287 #endif /* !__GENERATING_BOUNDS.H */
290 #ifdef CONFIG_ZONE_DMA
292 * ZONE_DMA is used when there are devices that are not able
293 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
294 * carve out the portion of memory that is needed for these devices.
295 * The range is arch specific.
300 * ---------------------------
301 * parisc, ia64, sparc <4G
304 * alpha Unlimited or 0-16MB.
306 * i386, x86_64 and multiple other arches
311 #ifdef CONFIG_ZONE_DMA32
313 * x86_64 needs two ZONE_DMAs because it supports devices that are
314 * only able to do DMA to the lower 16M but also 32 bit devices that
315 * can only do DMA areas below 4G.
320 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
321 * performed on pages in ZONE_NORMAL if the DMA devices support
322 * transfers to all addressable memory.
325 #ifdef CONFIG_HIGHMEM
327 * A memory area that is only addressable by the kernel through
328 * mapping portions into its own address space. This is for example
329 * used by i386 to allow the kernel to address the memory beyond
330 * 900MB. The kernel will set up special mappings (page
331 * table entries on i386) for each page that the kernel needs to
337 #ifdef CONFIG_ZONE_DEVICE
344 #ifndef __GENERATING_BOUNDS_H
347 /* Read-mostly fields */
349 /* zone watermarks, access with *_wmark_pages(zone) macros */
350 unsigned long watermark[NR_WMARK];
352 unsigned long nr_reserved_highatomic;
355 * We don't know if the memory that we're going to allocate will be
356 * freeable or/and it will be released eventually, so to avoid totally
357 * wasting several GB of ram we must reserve some of the lower zone
358 * memory (otherwise we risk to run OOM on the lower zones despite
359 * there being tons of freeable ram on the higher zones). This array is
360 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
363 long lowmem_reserve[MAX_NR_ZONES];
368 struct pglist_data *zone_pgdat;
369 struct per_cpu_pageset __percpu *pageset;
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 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
380 unsigned long zone_start_pfn;
383 * spanned_pages is the total pages spanned by the zone, including
384 * holes, which is calculated as:
385 * spanned_pages = zone_end_pfn - zone_start_pfn;
387 * present_pages is physical pages existing within the zone, which
389 * present_pages = spanned_pages - absent_pages(pages in holes);
391 * managed_pages is present pages managed by the buddy system, which
392 * is calculated as (reserved_pages includes pages allocated by the
393 * bootmem allocator):
394 * managed_pages = present_pages - reserved_pages;
396 * So present_pages may be used by memory hotplug or memory power
397 * management logic to figure out unmanaged pages by checking
398 * (present_pages - managed_pages). And managed_pages should be used
399 * by page allocator and vm scanner to calculate all kinds of watermarks
404 * zone_start_pfn and spanned_pages are protected by span_seqlock.
405 * It is a seqlock because it has to be read outside of zone->lock,
406 * and it is done in the main allocator path. But, it is written
407 * quite infrequently.
409 * The span_seq lock is declared along with zone->lock because it is
410 * frequently read in proximity to zone->lock. It's good to
411 * give them a chance of being in the same cacheline.
413 * Write access to present_pages at runtime should be protected by
414 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
415 * present_pages should get_online_mems() to get a stable value.
417 * Read access to managed_pages should be safe because it's unsigned
418 * long. Write access to zone->managed_pages and totalram_pages are
419 * protected by managed_page_count_lock at runtime. Idealy only
420 * adjust_managed_page_count() should be used instead of directly
421 * touching zone->managed_pages and totalram_pages.
423 unsigned long managed_pages;
424 unsigned long spanned_pages;
425 unsigned long present_pages;
429 #ifdef CONFIG_MEMORY_ISOLATION
431 * Number of isolated pageblock. It is used to solve incorrect
432 * freepage counting problem due to racy retrieving migratetype
433 * of pageblock. Protected by zone->lock.
435 unsigned long nr_isolate_pageblock;
438 #ifdef CONFIG_MEMORY_HOTPLUG
439 /* see spanned/present_pages for more description */
440 seqlock_t span_seqlock;
444 * wait_table -- the array holding the hash table
445 * wait_table_hash_nr_entries -- the size of the hash table array
446 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
448 * The purpose of all these is to keep track of the people
449 * waiting for a page to become available and make them
450 * runnable again when possible. The trouble is that this
451 * consumes a lot of space, especially when so few things
452 * wait on pages at a given time. So instead of using
453 * per-page waitqueues, we use a waitqueue hash table.
455 * The bucket discipline is to sleep on the same queue when
456 * colliding and wake all in that wait queue when removing.
457 * When something wakes, it must check to be sure its page is
458 * truly available, a la thundering herd. The cost of a
459 * collision is great, but given the expected load of the
460 * table, they should be so rare as to be outweighed by the
461 * benefits from the saved space.
463 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
464 * primary users of these fields, and in mm/page_alloc.c
465 * free_area_init_core() performs the initialization of them.
467 wait_queue_head_t *wait_table;
468 unsigned long wait_table_hash_nr_entries;
469 unsigned long wait_table_bits;
471 /* Write-intensive fields used from the page allocator */
474 /* free areas of different sizes */
475 struct free_area free_area[MAX_ORDER];
477 /* zone flags, see below */
480 /* Primarily protects free_area */
483 /* Write-intensive fields used by compaction and vmstats. */
487 * When free pages are below this point, additional steps are taken
488 * when reading the number of free pages to avoid per-cpu counter
489 * drift allowing watermarks to be breached
491 unsigned long percpu_drift_mark;
493 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
494 /* pfn where compaction free scanner should start */
495 unsigned long compact_cached_free_pfn;
496 /* pfn where async and sync compaction migration scanner should start */
497 unsigned long compact_cached_migrate_pfn[2];
500 #ifdef CONFIG_COMPACTION
502 * On compaction failure, 1<<compact_defer_shift compactions
503 * are skipped before trying again. The number attempted since
504 * last failure is tracked with compact_considered.
506 unsigned int compact_considered;
507 unsigned int compact_defer_shift;
508 int compact_order_failed;
511 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
512 /* Set to true when the PG_migrate_skip bits should be cleared */
513 bool compact_blockskip_flush;
519 /* Zone statistics */
520 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
521 } ____cacheline_internodealigned_in_smp;
524 PGDAT_CONGESTED, /* pgdat has many dirty pages backed by
527 PGDAT_DIRTY, /* reclaim scanning has recently found
528 * many dirty file pages at the tail
531 PGDAT_WRITEBACK, /* reclaim scanning has recently found
532 * many pages under writeback
534 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
537 static inline unsigned long zone_end_pfn(const struct zone *zone)
539 return zone->zone_start_pfn + zone->spanned_pages;
542 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
544 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
547 static inline bool zone_is_initialized(struct zone *zone)
549 return !!zone->wait_table;
552 static inline bool zone_is_empty(struct zone *zone)
554 return zone->spanned_pages == 0;
558 * The "priority" of VM scanning is how much of the queues we will scan in one
559 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
560 * queues ("queue_length >> 12") during an aging round.
562 #define DEF_PRIORITY 12
564 /* Maximum number of zones on a zonelist */
565 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
568 ZONELIST_FALLBACK, /* zonelist with fallback */
571 * The NUMA zonelists are doubled because we need zonelists that
572 * restrict the allocations to a single node for __GFP_THISNODE.
574 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
580 * This struct contains information about a zone in a zonelist. It is stored
581 * here to avoid dereferences into large structures and lookups of tables
584 struct zone *zone; /* Pointer to actual zone */
585 int zone_idx; /* zone_idx(zoneref->zone) */
589 * One allocation request operates on a zonelist. A zonelist
590 * is a list of zones, the first one is the 'goal' of the
591 * allocation, the other zones are fallback zones, in decreasing
594 * To speed the reading of the zonelist, the zonerefs contain the zone index
595 * of the entry being read. Helper functions to access information given
596 * a struct zoneref are
598 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
599 * zonelist_zone_idx() - Return the index of the zone for an entry
600 * zonelist_node_idx() - Return the index of the node for an entry
603 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
606 #ifndef CONFIG_DISCONTIGMEM
607 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
608 extern struct page *mem_map;
612 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
613 * (mostly NUMA machines?) to denote a higher-level memory zone than the
616 * On NUMA machines, each NUMA node would have a pg_data_t to describe
617 * it's memory layout.
619 * Memory statistics and page replacement data structures are maintained on a
623 typedef struct pglist_data {
624 struct zone node_zones[MAX_NR_ZONES];
625 struct zonelist node_zonelists[MAX_ZONELISTS];
627 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
628 struct page *node_mem_map;
629 #ifdef CONFIG_PAGE_EXTENSION
630 struct page_ext *node_page_ext;
633 #ifndef CONFIG_NO_BOOTMEM
634 struct bootmem_data *bdata;
636 #ifdef CONFIG_MEMORY_HOTPLUG
638 * Must be held any time you expect node_start_pfn, node_present_pages
639 * or node_spanned_pages stay constant. Holding this will also
640 * guarantee that any pfn_valid() stays that way.
642 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
643 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG.
645 * Nests above zone->lock and zone->span_seqlock
647 spinlock_t node_size_lock;
649 unsigned long node_start_pfn;
650 unsigned long node_present_pages; /* total number of physical pages */
651 unsigned long node_spanned_pages; /* total size of physical page
652 range, including holes */
654 wait_queue_head_t kswapd_wait;
655 wait_queue_head_t pfmemalloc_wait;
656 struct task_struct *kswapd; /* Protected by
657 mem_hotplug_begin/end() */
659 enum zone_type kswapd_classzone_idx;
661 #ifdef CONFIG_COMPACTION
662 int kcompactd_max_order;
663 enum zone_type kcompactd_classzone_idx;
664 wait_queue_head_t kcompactd_wait;
665 struct task_struct *kcompactd;
667 #ifdef CONFIG_NUMA_BALANCING
668 /* Lock serializing the migrate rate limiting window */
669 spinlock_t numabalancing_migrate_lock;
671 /* Rate limiting time interval */
672 unsigned long numabalancing_migrate_next_window;
674 /* Number of pages migrated during the rate limiting time interval */
675 unsigned long numabalancing_migrate_nr_pages;
678 * This is a per-node reserve of pages that are not available
679 * to userspace allocations.
681 unsigned long totalreserve_pages;
685 * zone reclaim becomes active if more unmapped pages exist.
687 unsigned long min_unmapped_pages;
688 unsigned long min_slab_pages;
689 #endif /* CONFIG_NUMA */
691 /* Write-intensive fields used by page reclaim */
695 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
697 * If memory initialisation on large machines is deferred then this
698 * is the first PFN that needs to be initialised.
700 unsigned long first_deferred_pfn;
701 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
703 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
704 spinlock_t split_queue_lock;
705 struct list_head split_queue;
706 unsigned long split_queue_len;
709 /* Fields commonly accessed by the page reclaim scanner */
710 struct lruvec lruvec;
713 * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on
714 * this node's LRU. Maintained by the pageout code.
716 unsigned int inactive_ratio;
722 /* Per-node vmstats */
723 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
724 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
727 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
728 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
729 #ifdef CONFIG_FLAT_NODE_MEM_MAP
730 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
732 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
734 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
736 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
737 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
738 static inline spinlock_t *zone_lru_lock(struct zone *zone)
740 return &zone->zone_pgdat->lru_lock;
743 static inline struct lruvec *node_lruvec(struct pglist_data *pgdat)
745 return &pgdat->lruvec;
748 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
750 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
753 static inline bool pgdat_is_empty(pg_data_t *pgdat)
755 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
758 static inline int zone_id(const struct zone *zone)
760 struct pglist_data *pgdat = zone->zone_pgdat;
762 return zone - pgdat->node_zones;
765 #ifdef CONFIG_ZONE_DEVICE
766 static inline bool is_dev_zone(const struct zone *zone)
768 return zone_id(zone) == ZONE_DEVICE;
771 static inline bool is_dev_zone(const struct zone *zone)
777 #include <linux/memory_hotplug.h>
779 extern struct mutex zonelists_mutex;
780 void build_all_zonelists(pg_data_t *pgdat, struct zone *zone);
781 void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx);
782 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
783 int classzone_idx, unsigned int alloc_flags,
785 bool zone_watermark_ok(struct zone *z, unsigned int order,
786 unsigned long mark, int classzone_idx,
787 unsigned int alloc_flags);
788 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
789 unsigned long mark, int classzone_idx);
790 enum memmap_context {
794 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
797 extern void lruvec_init(struct lruvec *lruvec);
799 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
802 return lruvec->pgdat;
804 return container_of(lruvec, struct pglist_data, lruvec);
808 extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru);
810 #ifdef CONFIG_HAVE_MEMORY_PRESENT
811 void memory_present(int nid, unsigned long start, unsigned long end);
813 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
816 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
817 int local_memory_node(int node_id);
819 static inline int local_memory_node(int node_id) { return node_id; };
822 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
823 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
827 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
829 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
831 static inline int populated_zone(struct zone *zone)
833 return (!!zone->present_pages);
836 extern int movable_zone;
838 #ifdef CONFIG_HIGHMEM
839 static inline int zone_movable_is_highmem(void)
841 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
842 return movable_zone == ZONE_HIGHMEM;
844 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
849 static inline int is_highmem_idx(enum zone_type idx)
851 #ifdef CONFIG_HIGHMEM
852 return (idx == ZONE_HIGHMEM ||
853 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
860 * is_highmem - helper function to quickly check if a struct zone is a
861 * highmem zone or not. This is an attempt to keep references
862 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
863 * @zone - pointer to struct zone variable
865 static inline int is_highmem(struct zone *zone)
867 #ifdef CONFIG_HIGHMEM
868 return is_highmem_idx(zone_idx(zone));
874 /* These two functions are used to setup the per zone pages min values */
876 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
877 void __user *, size_t *, loff_t *);
878 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int,
879 void __user *, size_t *, loff_t *);
880 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
881 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
882 void __user *, size_t *, loff_t *);
883 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
884 void __user *, size_t *, loff_t *);
885 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
886 void __user *, size_t *, loff_t *);
887 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
888 void __user *, size_t *, loff_t *);
890 extern int numa_zonelist_order_handler(struct ctl_table *, int,
891 void __user *, size_t *, loff_t *);
892 extern char numa_zonelist_order[];
893 #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
895 #ifndef CONFIG_NEED_MULTIPLE_NODES
897 extern struct pglist_data contig_page_data;
898 #define NODE_DATA(nid) (&contig_page_data)
899 #define NODE_MEM_MAP(nid) mem_map
901 #else /* CONFIG_NEED_MULTIPLE_NODES */
903 #include <asm/mmzone.h>
905 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
907 extern struct pglist_data *first_online_pgdat(void);
908 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
909 extern struct zone *next_zone(struct zone *zone);
912 * for_each_online_pgdat - helper macro to iterate over all online nodes
913 * @pgdat - pointer to a pg_data_t variable
915 #define for_each_online_pgdat(pgdat) \
916 for (pgdat = first_online_pgdat(); \
918 pgdat = next_online_pgdat(pgdat))
920 * for_each_zone - helper macro to iterate over all memory zones
921 * @zone - pointer to struct zone variable
923 * The user only needs to declare the zone variable, for_each_zone
926 #define for_each_zone(zone) \
927 for (zone = (first_online_pgdat())->node_zones; \
929 zone = next_zone(zone))
931 #define for_each_populated_zone(zone) \
932 for (zone = (first_online_pgdat())->node_zones; \
934 zone = next_zone(zone)) \
935 if (!populated_zone(zone)) \
939 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
941 return zoneref->zone;
944 static inline int zonelist_zone_idx(struct zoneref *zoneref)
946 return zoneref->zone_idx;
949 static inline int zonelist_node_idx(struct zoneref *zoneref)
952 /* zone_to_nid not available in this context */
953 return zoneref->zone->node;
956 #endif /* CONFIG_NUMA */
959 struct zoneref *__next_zones_zonelist(struct zoneref *z,
960 enum zone_type highest_zoneidx,
964 * 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
965 * @z - The cursor used as a starting point for the search
966 * @highest_zoneidx - The zone index of the highest zone to return
967 * @nodes - An optional nodemask to filter the zonelist with
969 * This function returns the next zone at or below a given zone index that is
970 * within the allowed nodemask using a cursor as the starting point for the
971 * search. The zoneref returned is a cursor that represents the current zone
972 * being examined. It should be advanced by one before calling
973 * next_zones_zonelist again.
975 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
976 enum zone_type highest_zoneidx,
979 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
981 return __next_zones_zonelist(z, highest_zoneidx, nodes);
985 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
986 * @zonelist - The zonelist to search for a suitable zone
987 * @highest_zoneidx - The zone index of the highest zone to return
988 * @nodes - An optional nodemask to filter the zonelist with
989 * @zone - The first suitable zone found is returned via this parameter
991 * This function returns the first zone at or below a given zone index that is
992 * within the allowed nodemask. The zoneref returned is a cursor that can be
993 * used to iterate the zonelist with next_zones_zonelist by advancing it by
994 * one before calling.
996 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
997 enum zone_type highest_zoneidx,
1000 return next_zones_zonelist(zonelist->_zonerefs,
1001 highest_zoneidx, nodes);
1005 * 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
1006 * @zone - The current zone in the iterator
1007 * @z - The current pointer within zonelist->zones being iterated
1008 * @zlist - The zonelist being iterated
1009 * @highidx - The zone index of the highest zone to return
1010 * @nodemask - Nodemask allowed by the allocator
1012 * This iterator iterates though all zones at or below a given zone index and
1013 * within a given nodemask
1015 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1016 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1018 z = next_zones_zonelist(++z, highidx, nodemask), \
1019 zone = zonelist_zone(z))
1021 #define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1022 for (zone = z->zone; \
1024 z = next_zones_zonelist(++z, highidx, nodemask), \
1025 zone = zonelist_zone(z))
1029 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1030 * @zone - The current zone in the iterator
1031 * @z - The current pointer within zonelist->zones being iterated
1032 * @zlist - The zonelist being iterated
1033 * @highidx - The zone index of the highest zone to return
1035 * This iterator iterates though all zones at or below a given zone index.
1037 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1038 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1040 #ifdef CONFIG_SPARSEMEM
1041 #include <asm/sparsemem.h>
1044 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1045 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1046 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1052 #ifdef CONFIG_FLATMEM
1053 #define pfn_to_nid(pfn) (0)
1056 #ifdef CONFIG_SPARSEMEM
1059 * SECTION_SHIFT #bits space required to store a section #
1061 * PA_SECTION_SHIFT physical address to/from section number
1062 * PFN_SECTION_SHIFT pfn to/from section number
1064 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1065 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1067 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1069 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1070 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1072 #define SECTION_BLOCKFLAGS_BITS \
1073 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1075 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1076 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1079 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
1080 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
1082 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1083 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1087 struct mem_section {
1089 * This is, logically, a pointer to an array of struct
1090 * pages. However, it is stored with some other magic.
1091 * (see sparse.c::sparse_init_one_section())
1093 * Additionally during early boot we encode node id of
1094 * the location of the section here to guide allocation.
1095 * (see sparse.c::memory_present())
1097 * Making it a UL at least makes someone do a cast
1098 * before using it wrong.
1100 unsigned long section_mem_map;
1102 /* See declaration of similar field in struct zone */
1103 unsigned long *pageblock_flags;
1104 #ifdef CONFIG_PAGE_EXTENSION
1106 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1107 * section. (see page_ext.h about this.)
1109 struct page_ext *page_ext;
1113 * WARNING: mem_section must be a power-of-2 in size for the
1114 * calculation and use of SECTION_ROOT_MASK to make sense.
1118 #ifdef CONFIG_SPARSEMEM_EXTREME
1119 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1121 #define SECTIONS_PER_ROOT 1
1124 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1125 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1126 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1128 #ifdef CONFIG_SPARSEMEM_EXTREME
1129 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
1131 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1134 static inline struct mem_section *__nr_to_section(unsigned long nr)
1136 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1138 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1140 extern int __section_nr(struct mem_section* ms);
1141 extern unsigned long usemap_size(void);
1144 * We use the lower bits of the mem_map pointer to store
1145 * a little bit of information. There should be at least
1146 * 3 bits here due to 32-bit alignment.
1148 #define SECTION_MARKED_PRESENT (1UL<<0)
1149 #define SECTION_HAS_MEM_MAP (1UL<<1)
1150 #define SECTION_MAP_LAST_BIT (1UL<<2)
1151 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1152 #define SECTION_NID_SHIFT 2
1154 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1156 unsigned long map = section->section_mem_map;
1157 map &= SECTION_MAP_MASK;
1158 return (struct page *)map;
1161 static inline int present_section(struct mem_section *section)
1163 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1166 static inline int present_section_nr(unsigned long nr)
1168 return present_section(__nr_to_section(nr));
1171 static inline int valid_section(struct mem_section *section)
1173 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1176 static inline int valid_section_nr(unsigned long nr)
1178 return valid_section(__nr_to_section(nr));
1181 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1183 return __nr_to_section(pfn_to_section_nr(pfn));
1186 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1187 static inline int pfn_valid(unsigned long pfn)
1189 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1191 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
1195 static inline int pfn_present(unsigned long pfn)
1197 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1199 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1203 * These are _only_ used during initialisation, therefore they
1204 * can use __initdata ... They could have names to indicate
1208 #define pfn_to_nid(pfn) \
1210 unsigned long __pfn_to_nid_pfn = (pfn); \
1211 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1214 #define pfn_to_nid(pfn) (0)
1217 #define early_pfn_valid(pfn) pfn_valid(pfn)
1218 void sparse_init(void);
1220 #define sparse_init() do {} while (0)
1221 #define sparse_index_init(_sec, _nid) do {} while (0)
1222 #endif /* CONFIG_SPARSEMEM */
1225 * During memory init memblocks map pfns to nids. The search is expensive and
1226 * this caches recent lookups. The implementation of __early_pfn_to_nid
1227 * may treat start/end as pfns or sections.
1229 struct mminit_pfnnid_cache {
1230 unsigned long last_start;
1231 unsigned long last_end;
1235 #ifndef early_pfn_valid
1236 #define early_pfn_valid(pfn) (1)
1239 void memory_present(int nid, unsigned long start, unsigned long end);
1240 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
1243 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1244 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1245 * pfn_valid_within() should be used in this case; we optimise this away
1246 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1248 #ifdef CONFIG_HOLES_IN_ZONE
1249 #define pfn_valid_within(pfn) pfn_valid(pfn)
1251 #define pfn_valid_within(pfn) (1)
1254 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1256 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1257 * associated with it or not. In FLATMEM, it is expected that holes always
1258 * have valid memmap as long as there is valid PFNs either side of the hole.
1259 * In SPARSEMEM, it is assumed that a valid section has a memmap for the
1262 * However, an ARM, and maybe other embedded architectures in the future
1263 * free memmap backing holes to save memory on the assumption the memmap is
1264 * never used. The page_zone linkages are then broken even though pfn_valid()
1265 * returns true. A walker of the full memmap must then do this additional
1266 * check to ensure the memmap they are looking at is sane by making sure
1267 * the zone and PFN linkages are still valid. This is expensive, but walkers
1268 * of the full memmap are extremely rare.
1270 bool memmap_valid_within(unsigned long pfn,
1271 struct page *page, struct zone *zone);
1273 static inline bool memmap_valid_within(unsigned long pfn,
1274 struct page *page, struct zone *zone)
1278 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1280 #endif /* !__GENERATING_BOUNDS.H */
1281 #endif /* !__ASSEMBLY__ */
1282 #endif /* _LINUX_MMZONE_H */