1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_MMZONE_H
3 #define _LINUX_MMZONE_H
6 #ifndef __GENERATING_BOUNDS_H
8 #include <linux/spinlock.h>
9 #include <linux/list.h>
10 #include <linux/wait.h>
11 #include <linux/bitops.h>
12 #include <linux/cache.h>
13 #include <linux/threads.h>
14 #include <linux/numa.h>
15 #include <linux/init.h>
16 #include <linux/seqlock.h>
17 #include <linux/nodemask.h>
18 #include <linux/pageblock-flags.h>
19 #include <linux/page-flags-layout.h>
20 #include <linux/atomic.h>
21 #include <linux/mm_types.h>
22 #include <linux/page-flags.h>
25 /* Free memory management - zoned buddy allocator. */
26 #ifndef CONFIG_FORCE_MAX_ZONEORDER
29 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
31 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
34 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
35 * costly to service. That is between allocation orders which should
36 * coalesce naturally under reasonable reclaim pressure and those which
39 #define PAGE_ALLOC_COSTLY_ORDER 3
45 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
46 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
49 * MIGRATE_CMA migration type is designed to mimic the way
50 * ZONE_MOVABLE works. Only movable pages can be allocated
51 * from MIGRATE_CMA pageblocks and page allocator never
52 * implicitly change migration type of MIGRATE_CMA pageblock.
54 * The way to use it is to change migratetype of a range of
55 * pageblocks to MIGRATE_CMA which can be done by
56 * __free_pageblock_cma() function. What is important though
57 * is that a range of pageblocks must be aligned to
58 * MAX_ORDER_NR_PAGES should biggest page be bigger then
63 #ifdef CONFIG_MEMORY_ISOLATION
64 MIGRATE_ISOLATE, /* can't allocate from here */
69 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
70 extern const char * const migratetype_names[MIGRATE_TYPES];
73 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
74 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
76 # define is_migrate_cma(migratetype) false
77 # define is_migrate_cma_page(_page) false
80 static inline bool is_migrate_movable(int mt)
82 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
85 #define for_each_migratetype_order(order, type) \
86 for (order = 0; order < MAX_ORDER; order++) \
87 for (type = 0; type < MIGRATE_TYPES; type++)
89 extern int page_group_by_mobility_disabled;
91 #define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1)
92 #define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1)
94 #define get_pageblock_migratetype(page) \
95 get_pfnblock_flags_mask(page, page_to_pfn(page), \
96 PB_migrate_end, MIGRATETYPE_MASK)
99 struct list_head free_list[MIGRATE_TYPES];
100 unsigned long nr_free;
103 /* Used for pages not on another list */
104 static inline void add_to_free_area(struct page *page, struct free_area *area,
107 list_add(&page->lru, &area->free_list[migratetype]);
111 /* Used for pages not on another list */
112 static inline void add_to_free_area_tail(struct page *page, struct free_area *area,
115 list_add_tail(&page->lru, &area->free_list[migratetype]);
119 /* Used for pages which are on another list */
120 static inline void move_to_free_area(struct page *page, struct free_area *area,
123 list_move(&page->lru, &area->free_list[migratetype]);
126 static inline struct page *get_page_from_free_area(struct free_area *area,
129 return list_first_entry_or_null(&area->free_list[migratetype],
133 static inline void del_page_from_free_area(struct page *page,
134 struct free_area *area)
136 list_del(&page->lru);
137 __ClearPageBuddy(page);
138 set_page_private(page, 0);
142 static inline bool free_area_empty(struct free_area *area, int migratetype)
144 return list_empty(&area->free_list[migratetype]);
150 * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
151 * So add a wild amount of padding here to ensure that they fall into separate
152 * cachelines. There are very few zone structures in the machine, so space
153 * consumption is not a concern here.
155 #if defined(CONFIG_SMP)
156 struct zone_padding {
158 } ____cacheline_internodealigned_in_smp;
159 #define ZONE_PADDING(name) struct zone_padding name;
161 #define ZONE_PADDING(name)
165 enum numa_stat_item {
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 */
172 NR_VM_NUMA_STAT_ITEMS
175 #define NR_VM_NUMA_STAT_ITEMS 0
178 enum zone_stat_item {
179 /* First 128 byte cacheline (assuming 64 bit words) */
181 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
182 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
184 NR_ZONE_INACTIVE_FILE,
187 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
188 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
189 NR_PAGETABLE, /* used for pagetables */
190 NR_KERNEL_STACK_KB, /* measured in KiB */
191 /* Second 128 byte cacheline */
193 #if IS_ENABLED(CONFIG_ZSMALLOC)
194 NR_ZSPAGES, /* allocated in zsmalloc */
197 NR_VM_ZONE_STAT_ITEMS };
199 enum node_stat_item {
201 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
202 NR_ACTIVE_ANON, /* " " " " " */
203 NR_INACTIVE_FILE, /* " " " " " */
204 NR_ACTIVE_FILE, /* " " " " " */
205 NR_UNEVICTABLE, /* " " " " " */
207 NR_SLAB_UNRECLAIMABLE,
208 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
209 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
214 WORKINGSET_NODERECLAIM,
215 NR_ANON_MAPPED, /* Mapped anonymous pages */
216 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
217 only modified from process context */
221 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
222 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
228 NR_UNSTABLE_NFS, /* NFS unstable pages */
230 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
231 NR_DIRTIED, /* page dirtyings since bootup */
232 NR_WRITTEN, /* page writings since bootup */
233 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
234 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
235 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
236 NR_VM_NODE_STAT_ITEMS
240 * We do arithmetic on the LRU lists in various places in the code,
241 * so it is important to keep the active lists LRU_ACTIVE higher in
242 * the array than the corresponding inactive lists, and to keep
243 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
245 * This has to be kept in sync with the statistics in zone_stat_item
246 * above and the descriptions in vmstat_text in mm/vmstat.c
253 LRU_INACTIVE_ANON = LRU_BASE,
254 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
255 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
256 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
261 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
263 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
265 static inline bool is_file_lru(enum lru_list lru)
267 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
270 static inline bool is_active_lru(enum lru_list lru)
272 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
275 struct zone_reclaim_stat {
277 * The pageout code in vmscan.c keeps track of how many of the
278 * mem/swap backed and file backed pages are referenced.
279 * The higher the rotated/scanned ratio, the more valuable
282 * The anon LRU stats live in [0], file LRU stats in [1]
284 unsigned long recent_rotated[2];
285 unsigned long recent_scanned[2];
289 LRUVEC_CONGESTED, /* lruvec has many dirty pages
290 * backed by a congested BDI
295 struct list_head lists[NR_LRU_LISTS];
296 struct zone_reclaim_stat reclaim_stat;
297 /* Evictions & activations on the inactive file list */
298 atomic_long_t inactive_age;
299 /* Refaults at the time of last reclaim cycle */
300 unsigned long refaults;
301 /* Various lruvec state flags (enum lruvec_flags) */
304 struct pglist_data *pgdat;
308 /* Isolate unmapped pages */
309 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
310 /* Isolate for asynchronous migration */
311 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
312 /* Isolate unevictable pages */
313 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
315 /* LRU Isolation modes. */
316 typedef unsigned __bitwise isolate_mode_t;
318 enum zone_watermarks {
325 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
326 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
327 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
328 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
330 struct per_cpu_pages {
331 int count; /* number of pages in the list */
332 int high; /* high watermark, emptying needed */
333 int batch; /* chunk size for buddy add/remove */
335 /* Lists of pages, one per migrate type stored on the pcp-lists */
336 struct list_head lists[MIGRATE_PCPTYPES];
339 struct per_cpu_pageset {
340 struct per_cpu_pages pcp;
343 u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
347 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
351 struct per_cpu_nodestat {
353 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
356 #endif /* !__GENERATING_BOUNDS.H */
360 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
361 * to DMA to all of the addressable memory (ZONE_NORMAL).
362 * On architectures where this area covers the whole 32 bit address
363 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
364 * DMA addressing constraints. This distinction is important as a 32bit
365 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
366 * platforms may need both zones as they support peripherals with
367 * different DMA addressing limitations.
371 * - i386 and x86_64 have a fixed 16M ZONE_DMA and ZONE_DMA32 for the
372 * rest of the lower 4G.
374 * - arm only uses ZONE_DMA, the size, up to 4G, may vary depending on
375 * the specific device.
377 * - arm64 has a fixed 1G ZONE_DMA and ZONE_DMA32 for the rest of the
380 * - powerpc only uses ZONE_DMA, the size, up to 2G, may vary
381 * depending on the specific device.
383 * - s390 uses ZONE_DMA fixed to the lower 2G.
385 * - ia64 and riscv only use ZONE_DMA32.
387 * - parisc uses neither.
389 #ifdef CONFIG_ZONE_DMA
392 #ifdef CONFIG_ZONE_DMA32
396 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
397 * performed on pages in ZONE_NORMAL if the DMA devices support
398 * transfers to all addressable memory.
401 #ifdef CONFIG_HIGHMEM
403 * A memory area that is only addressable by the kernel through
404 * mapping portions into its own address space. This is for example
405 * used by i386 to allow the kernel to address the memory beyond
406 * 900MB. The kernel will set up special mappings (page
407 * table entries on i386) for each page that the kernel needs to
413 #ifdef CONFIG_ZONE_DEVICE
420 #ifndef __GENERATING_BOUNDS_H
423 /* Read-mostly fields */
425 /* zone watermarks, access with *_wmark_pages(zone) macros */
426 unsigned long _watermark[NR_WMARK];
427 unsigned long watermark_boost;
429 unsigned long nr_reserved_highatomic;
432 * We don't know if the memory that we're going to allocate will be
433 * freeable or/and it will be released eventually, so to avoid totally
434 * wasting several GB of ram we must reserve some of the lower zone
435 * memory (otherwise we risk to run OOM on the lower zones despite
436 * there being tons of freeable ram on the higher zones). This array is
437 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
440 long lowmem_reserve[MAX_NR_ZONES];
445 struct pglist_data *zone_pgdat;
446 struct per_cpu_pageset __percpu *pageset;
448 #ifndef CONFIG_SPARSEMEM
450 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
451 * In SPARSEMEM, this map is stored in struct mem_section
453 unsigned long *pageblock_flags;
454 #endif /* CONFIG_SPARSEMEM */
456 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
457 unsigned long zone_start_pfn;
460 * spanned_pages is the total pages spanned by the zone, including
461 * holes, which is calculated as:
462 * spanned_pages = zone_end_pfn - zone_start_pfn;
464 * present_pages is physical pages existing within the zone, which
466 * present_pages = spanned_pages - absent_pages(pages in holes);
468 * managed_pages is present pages managed by the buddy system, which
469 * is calculated as (reserved_pages includes pages allocated by the
470 * bootmem allocator):
471 * managed_pages = present_pages - reserved_pages;
473 * So present_pages may be used by memory hotplug or memory power
474 * management logic to figure out unmanaged pages by checking
475 * (present_pages - managed_pages). And managed_pages should be used
476 * by page allocator and vm scanner to calculate all kinds of watermarks
481 * zone_start_pfn and spanned_pages are protected by span_seqlock.
482 * It is a seqlock because it has to be read outside of zone->lock,
483 * and it is done in the main allocator path. But, it is written
484 * quite infrequently.
486 * The span_seq lock is declared along with zone->lock because it is
487 * frequently read in proximity to zone->lock. It's good to
488 * give them a chance of being in the same cacheline.
490 * Write access to present_pages at runtime should be protected by
491 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
492 * present_pages should get_online_mems() to get a stable value.
494 atomic_long_t managed_pages;
495 unsigned long spanned_pages;
496 unsigned long present_pages;
500 #ifdef CONFIG_MEMORY_ISOLATION
502 * Number of isolated pageblock. It is used to solve incorrect
503 * freepage counting problem due to racy retrieving migratetype
504 * of pageblock. Protected by zone->lock.
506 unsigned long nr_isolate_pageblock;
509 #ifdef CONFIG_MEMORY_HOTPLUG
510 /* see spanned/present_pages for more description */
511 seqlock_t span_seqlock;
516 /* Write-intensive fields used from the page allocator */
519 /* free areas of different sizes */
520 struct free_area free_area[MAX_ORDER];
522 /* zone flags, see below */
525 /* Primarily protects free_area */
528 /* Write-intensive fields used by compaction and vmstats. */
532 * When free pages are below this point, additional steps are taken
533 * when reading the number of free pages to avoid per-cpu counter
534 * drift allowing watermarks to be breached
536 unsigned long percpu_drift_mark;
538 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
539 /* pfn where compaction free scanner should start */
540 unsigned long compact_cached_free_pfn;
541 /* pfn where async and sync compaction migration scanner should start */
542 unsigned long compact_cached_migrate_pfn[2];
543 unsigned long compact_init_migrate_pfn;
544 unsigned long compact_init_free_pfn;
547 #ifdef CONFIG_COMPACTION
549 * On compaction failure, 1<<compact_defer_shift compactions
550 * are skipped before trying again. The number attempted since
551 * last failure is tracked with compact_considered.
553 unsigned int compact_considered;
554 unsigned int compact_defer_shift;
555 int compact_order_failed;
558 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
559 /* Set to true when the PG_migrate_skip bits should be cleared */
560 bool compact_blockskip_flush;
566 /* Zone statistics */
567 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
568 atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
569 } ____cacheline_internodealigned_in_smp;
572 PGDAT_DIRTY, /* reclaim scanning has recently found
573 * many dirty file pages at the tail
576 PGDAT_WRITEBACK, /* reclaim scanning has recently found
577 * many pages under writeback
579 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
583 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
584 * Cleared when kswapd is woken.
588 static inline unsigned long zone_managed_pages(struct zone *zone)
590 return (unsigned long)atomic_long_read(&zone->managed_pages);
593 static inline unsigned long zone_end_pfn(const struct zone *zone)
595 return zone->zone_start_pfn + zone->spanned_pages;
598 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
600 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
603 static inline bool zone_is_initialized(struct zone *zone)
605 return zone->initialized;
608 static inline bool zone_is_empty(struct zone *zone)
610 return zone->spanned_pages == 0;
614 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
615 * intersection with the given zone
617 static inline bool zone_intersects(struct zone *zone,
618 unsigned long start_pfn, unsigned long nr_pages)
620 if (zone_is_empty(zone))
622 if (start_pfn >= zone_end_pfn(zone) ||
623 start_pfn + nr_pages <= zone->zone_start_pfn)
630 * The "priority" of VM scanning is how much of the queues we will scan in one
631 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
632 * queues ("queue_length >> 12") during an aging round.
634 #define DEF_PRIORITY 12
636 /* Maximum number of zones on a zonelist */
637 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
640 ZONELIST_FALLBACK, /* zonelist with fallback */
643 * The NUMA zonelists are doubled because we need zonelists that
644 * restrict the allocations to a single node for __GFP_THISNODE.
646 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
652 * This struct contains information about a zone in a zonelist. It is stored
653 * here to avoid dereferences into large structures and lookups of tables
656 struct zone *zone; /* Pointer to actual zone */
657 int zone_idx; /* zone_idx(zoneref->zone) */
661 * One allocation request operates on a zonelist. A zonelist
662 * is a list of zones, the first one is the 'goal' of the
663 * allocation, the other zones are fallback zones, in decreasing
666 * To speed the reading of the zonelist, the zonerefs contain the zone index
667 * of the entry being read. Helper functions to access information given
668 * a struct zoneref are
670 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
671 * zonelist_zone_idx() - Return the index of the zone for an entry
672 * zonelist_node_idx() - Return the index of the node for an entry
675 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
678 #ifndef CONFIG_DISCONTIGMEM
679 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
680 extern struct page *mem_map;
683 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
684 struct deferred_split {
685 spinlock_t split_queue_lock;
686 struct list_head split_queue;
687 unsigned long split_queue_len;
692 * On NUMA machines, each NUMA node would have a pg_data_t to describe
693 * it's memory layout. On UMA machines there is a single pglist_data which
694 * describes the whole memory.
696 * Memory statistics and page replacement data structures are maintained on a
700 typedef struct pglist_data {
701 struct zone node_zones[MAX_NR_ZONES];
702 struct zonelist node_zonelists[MAX_ZONELISTS];
704 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
705 struct page *node_mem_map;
706 #ifdef CONFIG_PAGE_EXTENSION
707 struct page_ext *node_page_ext;
710 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
712 * Must be held any time you expect node_start_pfn,
713 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
715 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
716 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
717 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
719 * Nests above zone->lock and zone->span_seqlock
721 spinlock_t node_size_lock;
723 unsigned long node_start_pfn;
724 unsigned long node_present_pages; /* total number of physical pages */
725 unsigned long node_spanned_pages; /* total size of physical page
726 range, including holes */
728 wait_queue_head_t kswapd_wait;
729 wait_queue_head_t pfmemalloc_wait;
730 struct task_struct *kswapd; /* Protected by
731 mem_hotplug_begin/end() */
733 enum zone_type kswapd_classzone_idx;
735 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
737 #ifdef CONFIG_COMPACTION
738 int kcompactd_max_order;
739 enum zone_type kcompactd_classzone_idx;
740 wait_queue_head_t kcompactd_wait;
741 struct task_struct *kcompactd;
744 * This is a per-node reserve of pages that are not available
745 * to userspace allocations.
747 unsigned long totalreserve_pages;
751 * node reclaim becomes active if more unmapped pages exist.
753 unsigned long min_unmapped_pages;
754 unsigned long min_slab_pages;
755 #endif /* CONFIG_NUMA */
757 /* Write-intensive fields used by page reclaim */
761 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
763 * If memory initialisation on large machines is deferred then this
764 * is the first PFN that needs to be initialised.
766 unsigned long first_deferred_pfn;
767 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
769 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
770 struct deferred_split deferred_split_queue;
773 /* Fields commonly accessed by the page reclaim scanner */
776 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
778 * Use mem_cgroup_lruvec() to look up lruvecs.
780 struct lruvec __lruvec;
786 /* Per-node vmstats */
787 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
788 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
791 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
792 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
793 #ifdef CONFIG_FLAT_NODE_MEM_MAP
794 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
796 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
798 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
800 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
801 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
803 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
805 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
808 static inline bool pgdat_is_empty(pg_data_t *pgdat)
810 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
813 #include <linux/memory_hotplug.h>
815 void build_all_zonelists(pg_data_t *pgdat);
816 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
817 enum zone_type classzone_idx);
818 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
819 int classzone_idx, unsigned int alloc_flags,
821 bool zone_watermark_ok(struct zone *z, unsigned int order,
822 unsigned long mark, int classzone_idx,
823 unsigned int alloc_flags);
824 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
825 unsigned long mark, int classzone_idx);
826 enum memmap_context {
830 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
833 extern void lruvec_init(struct lruvec *lruvec);
835 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
838 return lruvec->pgdat;
840 return container_of(lruvec, struct pglist_data, __lruvec);
844 extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
846 #ifdef CONFIG_HAVE_MEMORY_PRESENT
847 void memory_present(int nid, unsigned long start, unsigned long end);
849 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
852 #if defined(CONFIG_SPARSEMEM)
853 void memblocks_present(void);
855 static inline void memblocks_present(void) {}
858 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
859 int local_memory_node(int node_id);
861 static inline int local_memory_node(int node_id) { return node_id; };
865 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
867 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
870 * Returns true if a zone has pages managed by the buddy allocator.
871 * All the reclaim decisions have to use this function rather than
872 * populated_zone(). If the whole zone is reserved then we can easily
873 * end up with populated_zone() && !managed_zone().
875 static inline bool managed_zone(struct zone *zone)
877 return zone_managed_pages(zone);
880 /* Returns true if a zone has memory */
881 static inline bool populated_zone(struct zone *zone)
883 return zone->present_pages;
887 static inline int zone_to_nid(struct zone *zone)
892 static inline void zone_set_nid(struct zone *zone, int nid)
897 static inline int zone_to_nid(struct zone *zone)
902 static inline void zone_set_nid(struct zone *zone, int nid) {}
905 extern int movable_zone;
907 #ifdef CONFIG_HIGHMEM
908 static inline int zone_movable_is_highmem(void)
910 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
911 return movable_zone == ZONE_HIGHMEM;
913 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
918 static inline int is_highmem_idx(enum zone_type idx)
920 #ifdef CONFIG_HIGHMEM
921 return (idx == ZONE_HIGHMEM ||
922 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
929 * is_highmem - helper function to quickly check if a struct zone is a
930 * highmem zone or not. This is an attempt to keep references
931 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
932 * @zone - pointer to struct zone variable
934 static inline int is_highmem(struct zone *zone)
936 #ifdef CONFIG_HIGHMEM
937 return is_highmem_idx(zone_idx(zone));
943 /* These two functions are used to setup the per zone pages min values */
945 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
946 void __user *, size_t *, loff_t *);
947 int watermark_boost_factor_sysctl_handler(struct ctl_table *, int,
948 void __user *, size_t *, loff_t *);
949 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int,
950 void __user *, size_t *, loff_t *);
951 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
952 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
953 void __user *, size_t *, loff_t *);
954 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
955 void __user *, size_t *, loff_t *);
956 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
957 void __user *, size_t *, loff_t *);
958 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
959 void __user *, size_t *, loff_t *);
961 extern int numa_zonelist_order_handler(struct ctl_table *, int,
962 void __user *, size_t *, loff_t *);
963 extern char numa_zonelist_order[];
964 #define NUMA_ZONELIST_ORDER_LEN 16
966 #ifndef CONFIG_NEED_MULTIPLE_NODES
968 extern struct pglist_data contig_page_data;
969 #define NODE_DATA(nid) (&contig_page_data)
970 #define NODE_MEM_MAP(nid) mem_map
972 #else /* CONFIG_NEED_MULTIPLE_NODES */
974 #include <asm/mmzone.h>
976 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
978 extern struct pglist_data *first_online_pgdat(void);
979 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
980 extern struct zone *next_zone(struct zone *zone);
983 * for_each_online_pgdat - helper macro to iterate over all online nodes
984 * @pgdat - pointer to a pg_data_t variable
986 #define for_each_online_pgdat(pgdat) \
987 for (pgdat = first_online_pgdat(); \
989 pgdat = next_online_pgdat(pgdat))
991 * for_each_zone - helper macro to iterate over all memory zones
992 * @zone - pointer to struct zone variable
994 * The user only needs to declare the zone variable, for_each_zone
997 #define for_each_zone(zone) \
998 for (zone = (first_online_pgdat())->node_zones; \
1000 zone = next_zone(zone))
1002 #define for_each_populated_zone(zone) \
1003 for (zone = (first_online_pgdat())->node_zones; \
1005 zone = next_zone(zone)) \
1006 if (!populated_zone(zone)) \
1007 ; /* do nothing */ \
1010 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1012 return zoneref->zone;
1015 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1017 return zoneref->zone_idx;
1020 static inline int zonelist_node_idx(struct zoneref *zoneref)
1022 return zone_to_nid(zoneref->zone);
1025 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1026 enum zone_type highest_zoneidx,
1030 * 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
1031 * @z - The cursor used as a starting point for the search
1032 * @highest_zoneidx - The zone index of the highest zone to return
1033 * @nodes - An optional nodemask to filter the zonelist with
1035 * This function returns the next zone at or below a given zone index that is
1036 * within the allowed nodemask using a cursor as the starting point for the
1037 * search. The zoneref returned is a cursor that represents the current zone
1038 * being examined. It should be advanced by one before calling
1039 * next_zones_zonelist again.
1041 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1042 enum zone_type highest_zoneidx,
1045 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1047 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1051 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1052 * @zonelist - The zonelist to search for a suitable zone
1053 * @highest_zoneidx - The zone index of the highest zone to return
1054 * @nodes - An optional nodemask to filter the zonelist with
1055 * @return - Zoneref pointer for the first suitable zone found (see below)
1057 * This function returns the first zone at or below a given zone index that is
1058 * within the allowed nodemask. The zoneref returned is a cursor that can be
1059 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1060 * one before calling.
1062 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1063 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1064 * update due to cpuset modification.
1066 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1067 enum zone_type highest_zoneidx,
1070 return next_zones_zonelist(zonelist->_zonerefs,
1071 highest_zoneidx, nodes);
1075 * 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
1076 * @zone - The current zone in the iterator
1077 * @z - The current pointer within zonelist->_zonerefs being iterated
1078 * @zlist - The zonelist being iterated
1079 * @highidx - The zone index of the highest zone to return
1080 * @nodemask - Nodemask allowed by the allocator
1082 * This iterator iterates though all zones at or below a given zone index and
1083 * within a given nodemask
1085 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1086 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1088 z = next_zones_zonelist(++z, highidx, nodemask), \
1089 zone = zonelist_zone(z))
1091 #define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1092 for (zone = z->zone; \
1094 z = next_zones_zonelist(++z, highidx, nodemask), \
1095 zone = zonelist_zone(z))
1099 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1100 * @zone - The current zone in the iterator
1101 * @z - The current pointer within zonelist->zones being iterated
1102 * @zlist - The zonelist being iterated
1103 * @highidx - The zone index of the highest zone to return
1105 * This iterator iterates though all zones at or below a given zone index.
1107 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1108 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1110 #ifdef CONFIG_SPARSEMEM
1111 #include <asm/sparsemem.h>
1114 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1115 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1116 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1118 BUILD_BUG_ON(IS_ENABLED(CONFIG_NUMA));
1123 #ifdef CONFIG_FLATMEM
1124 #define pfn_to_nid(pfn) (0)
1127 #ifdef CONFIG_SPARSEMEM
1130 * SECTION_SHIFT #bits space required to store a section #
1132 * PA_SECTION_SHIFT physical address to/from section number
1133 * PFN_SECTION_SHIFT pfn to/from section number
1135 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1136 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1138 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1140 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1141 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1143 #define SECTION_BLOCKFLAGS_BITS \
1144 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1146 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1147 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1150 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1152 return pfn >> PFN_SECTION_SHIFT;
1154 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1156 return sec << PFN_SECTION_SHIFT;
1159 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1160 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1162 #define SUBSECTION_SHIFT 21
1164 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1165 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1166 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1168 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1169 #error Subsection size exceeds section size
1171 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1174 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1175 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1177 struct mem_section_usage {
1178 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1179 /* See declaration of similar field in struct zone */
1180 unsigned long pageblock_flags[0];
1183 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1187 struct mem_section {
1189 * This is, logically, a pointer to an array of struct
1190 * pages. However, it is stored with some other magic.
1191 * (see sparse.c::sparse_init_one_section())
1193 * Additionally during early boot we encode node id of
1194 * the location of the section here to guide allocation.
1195 * (see sparse.c::memory_present())
1197 * Making it a UL at least makes someone do a cast
1198 * before using it wrong.
1200 unsigned long section_mem_map;
1202 struct mem_section_usage *usage;
1203 #ifdef CONFIG_PAGE_EXTENSION
1205 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1206 * section. (see page_ext.h about this.)
1208 struct page_ext *page_ext;
1212 * WARNING: mem_section must be a power-of-2 in size for the
1213 * calculation and use of SECTION_ROOT_MASK to make sense.
1217 #ifdef CONFIG_SPARSEMEM_EXTREME
1218 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1220 #define SECTIONS_PER_ROOT 1
1223 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1224 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1225 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1227 #ifdef CONFIG_SPARSEMEM_EXTREME
1228 extern struct mem_section **mem_section;
1230 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1233 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1235 return ms->usage->pageblock_flags;
1238 static inline struct mem_section *__nr_to_section(unsigned long nr)
1240 #ifdef CONFIG_SPARSEMEM_EXTREME
1244 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1246 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1248 extern unsigned long __section_nr(struct mem_section *ms);
1249 extern size_t mem_section_usage_size(void);
1252 * We use the lower bits of the mem_map pointer to store
1253 * a little bit of information. The pointer is calculated
1254 * as mem_map - section_nr_to_pfn(pnum). The result is
1255 * aligned to the minimum alignment of the two values:
1256 * 1. All mem_map arrays are page-aligned.
1257 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1258 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1259 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1260 * worst combination is powerpc with 256k pages,
1261 * which results in PFN_SECTION_SHIFT equal 6.
1262 * To sum it up, at least 6 bits are available.
1264 #define SECTION_MARKED_PRESENT (1UL<<0)
1265 #define SECTION_HAS_MEM_MAP (1UL<<1)
1266 #define SECTION_IS_ONLINE (1UL<<2)
1267 #define SECTION_IS_EARLY (1UL<<3)
1268 #define SECTION_MAP_LAST_BIT (1UL<<4)
1269 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1270 #define SECTION_NID_SHIFT 3
1272 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1274 unsigned long map = section->section_mem_map;
1275 map &= SECTION_MAP_MASK;
1276 return (struct page *)map;
1279 static inline int present_section(struct mem_section *section)
1281 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1284 static inline int present_section_nr(unsigned long nr)
1286 return present_section(__nr_to_section(nr));
1289 static inline int valid_section(struct mem_section *section)
1291 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1294 static inline int early_section(struct mem_section *section)
1296 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1299 static inline int valid_section_nr(unsigned long nr)
1301 return valid_section(__nr_to_section(nr));
1304 static inline int online_section(struct mem_section *section)
1306 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1309 static inline int online_section_nr(unsigned long nr)
1311 return online_section(__nr_to_section(nr));
1314 #ifdef CONFIG_MEMORY_HOTPLUG
1315 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1316 #ifdef CONFIG_MEMORY_HOTREMOVE
1317 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1321 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1323 return __nr_to_section(pfn_to_section_nr(pfn));
1326 extern unsigned long __highest_present_section_nr;
1328 static inline int subsection_map_index(unsigned long pfn)
1330 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1333 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1334 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1336 int idx = subsection_map_index(pfn);
1338 return test_bit(idx, ms->usage->subsection_map);
1341 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1347 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1348 static inline int pfn_valid(unsigned long pfn)
1350 struct mem_section *ms;
1352 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1354 ms = __nr_to_section(pfn_to_section_nr(pfn));
1355 if (!valid_section(ms))
1358 * Traditionally early sections always returned pfn_valid() for
1359 * the entire section-sized span.
1361 return early_section(ms) || pfn_section_valid(ms, pfn);
1365 static inline int pfn_in_present_section(unsigned long pfn)
1367 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1369 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1372 static inline unsigned long next_present_section_nr(unsigned long section_nr)
1374 while (++section_nr <= __highest_present_section_nr) {
1375 if (present_section_nr(section_nr))
1383 * These are _only_ used during initialisation, therefore they
1384 * can use __initdata ... They could have names to indicate
1388 #define pfn_to_nid(pfn) \
1390 unsigned long __pfn_to_nid_pfn = (pfn); \
1391 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1394 #define pfn_to_nid(pfn) (0)
1397 #define early_pfn_valid(pfn) pfn_valid(pfn)
1398 void sparse_init(void);
1400 #define sparse_init() do {} while (0)
1401 #define sparse_index_init(_sec, _nid) do {} while (0)
1402 #define pfn_in_present_section pfn_valid
1403 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
1404 #endif /* CONFIG_SPARSEMEM */
1407 * During memory init memblocks map pfns to nids. The search is expensive and
1408 * this caches recent lookups. The implementation of __early_pfn_to_nid
1409 * may treat start/end as pfns or sections.
1411 struct mminit_pfnnid_cache {
1412 unsigned long last_start;
1413 unsigned long last_end;
1417 #ifndef early_pfn_valid
1418 #define early_pfn_valid(pfn) (1)
1421 void memory_present(int nid, unsigned long start, unsigned long end);
1424 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1425 * need to check pfn validity within that MAX_ORDER_NR_PAGES block.
1426 * pfn_valid_within() should be used in this case; we optimise this away
1427 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1429 #ifdef CONFIG_HOLES_IN_ZONE
1430 #define pfn_valid_within(pfn) pfn_valid(pfn)
1432 #define pfn_valid_within(pfn) (1)
1435 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1437 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1438 * associated with it or not. This means that a struct page exists for this
1439 * pfn. The caller cannot assume the page is fully initialized in general.
1440 * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1441 * will ensure the struct page is fully online and initialized. Special pages
1442 * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1444 * In FLATMEM, it is expected that holes always have valid memmap as long as
1445 * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1446 * that a valid section has a memmap for the entire section.
1448 * However, an ARM, and maybe other embedded architectures in the future
1449 * free memmap backing holes to save memory on the assumption the memmap is
1450 * never used. The page_zone linkages are then broken even though pfn_valid()
1451 * returns true. A walker of the full memmap must then do this additional
1452 * check to ensure the memmap they are looking at is sane by making sure
1453 * the zone and PFN linkages are still valid. This is expensive, but walkers
1454 * of the full memmap are extremely rare.
1456 bool memmap_valid_within(unsigned long pfn,
1457 struct page *page, struct zone *zone);
1459 static inline bool memmap_valid_within(unsigned long pfn,
1460 struct page *page, struct zone *zone)
1464 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1466 #endif /* !__GENERATING_BOUNDS.H */
1467 #endif /* !__ASSEMBLY__ */
1468 #endif /* _LINUX_MMZONE_H */