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 #ifdef CONFIG_SHUFFLE_PAGE_ALLOCATOR
120 /* Used to preserve page allocation order entropy */
121 void add_to_free_area_random(struct page *page, struct free_area *area,
124 static inline void add_to_free_area_random(struct page *page,
125 struct free_area *area, int migratetype)
127 add_to_free_area(page, area, migratetype);
131 /* Used for pages which are on another list */
132 static inline void move_to_free_area(struct page *page, struct free_area *area,
135 list_move(&page->lru, &area->free_list[migratetype]);
138 static inline struct page *get_page_from_free_area(struct free_area *area,
141 return list_first_entry_or_null(&area->free_list[migratetype],
145 static inline void del_page_from_free_area(struct page *page,
146 struct free_area *area)
148 list_del(&page->lru);
149 __ClearPageBuddy(page);
150 set_page_private(page, 0);
154 static inline bool free_area_empty(struct free_area *area, int migratetype)
156 return list_empty(&area->free_list[migratetype]);
162 * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
163 * So add a wild amount of padding here to ensure that they fall into separate
164 * cachelines. There are very few zone structures in the machine, so space
165 * consumption is not a concern here.
167 #if defined(CONFIG_SMP)
168 struct zone_padding {
170 } ____cacheline_internodealigned_in_smp;
171 #define ZONE_PADDING(name) struct zone_padding name;
173 #define ZONE_PADDING(name)
177 enum numa_stat_item {
178 NUMA_HIT, /* allocated in intended node */
179 NUMA_MISS, /* allocated in non intended node */
180 NUMA_FOREIGN, /* was intended here, hit elsewhere */
181 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
182 NUMA_LOCAL, /* allocation from local node */
183 NUMA_OTHER, /* allocation from other node */
184 NR_VM_NUMA_STAT_ITEMS
187 #define NR_VM_NUMA_STAT_ITEMS 0
190 enum zone_stat_item {
191 /* First 128 byte cacheline (assuming 64 bit words) */
193 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
194 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
196 NR_ZONE_INACTIVE_FILE,
199 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
200 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
201 NR_PAGETABLE, /* used for pagetables */
202 NR_KERNEL_STACK_KB, /* measured in KiB */
203 /* Second 128 byte cacheline */
205 #if IS_ENABLED(CONFIG_ZSMALLOC)
206 NR_ZSPAGES, /* allocated in zsmalloc */
209 NR_VM_ZONE_STAT_ITEMS };
211 enum node_stat_item {
213 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
214 NR_ACTIVE_ANON, /* " " " " " */
215 NR_INACTIVE_FILE, /* " " " " " */
216 NR_ACTIVE_FILE, /* " " " " " */
217 NR_UNEVICTABLE, /* " " " " " */
219 NR_SLAB_UNRECLAIMABLE,
220 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
221 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
226 WORKINGSET_NODERECLAIM,
227 NR_ANON_MAPPED, /* Mapped anonymous pages */
228 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
229 only modified from process context */
233 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
234 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
240 NR_UNSTABLE_NFS, /* NFS unstable pages */
242 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
243 NR_DIRTIED, /* page dirtyings since bootup */
244 NR_WRITTEN, /* page writings since bootup */
245 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
246 NR_VM_NODE_STAT_ITEMS
250 * We do arithmetic on the LRU lists in various places in the code,
251 * so it is important to keep the active lists LRU_ACTIVE higher in
252 * the array than the corresponding inactive lists, and to keep
253 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
255 * This has to be kept in sync with the statistics in zone_stat_item
256 * above and the descriptions in vmstat_text in mm/vmstat.c
263 LRU_INACTIVE_ANON = LRU_BASE,
264 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
265 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
266 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
271 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
273 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
275 static inline int is_file_lru(enum lru_list lru)
277 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
280 static inline int is_active_lru(enum lru_list lru)
282 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
285 struct zone_reclaim_stat {
287 * The pageout code in vmscan.c keeps track of how many of the
288 * mem/swap backed and file backed pages are referenced.
289 * The higher the rotated/scanned ratio, the more valuable
292 * The anon LRU stats live in [0], file LRU stats in [1]
294 unsigned long recent_rotated[2];
295 unsigned long recent_scanned[2];
299 struct list_head lists[NR_LRU_LISTS];
300 struct zone_reclaim_stat reclaim_stat;
301 /* Evictions & activations on the inactive file list */
302 atomic_long_t inactive_age;
303 /* Refaults at the time of last reclaim cycle */
304 unsigned long refaults;
306 struct pglist_data *pgdat;
310 /* Isolate unmapped file */
311 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
312 /* Isolate for asynchronous migration */
313 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
314 /* Isolate unevictable pages */
315 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
317 /* LRU Isolation modes. */
318 typedef unsigned __bitwise isolate_mode_t;
320 enum zone_watermarks {
327 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
328 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
329 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
330 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
332 struct per_cpu_pages {
333 int count; /* number of pages in the list */
334 int high; /* high watermark, emptying needed */
335 int batch; /* chunk size for buddy add/remove */
337 /* Lists of pages, one per migrate type stored on the pcp-lists */
338 struct list_head lists[MIGRATE_PCPTYPES];
341 struct per_cpu_pageset {
342 struct per_cpu_pages pcp;
345 u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
349 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
353 struct per_cpu_nodestat {
355 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
358 #endif /* !__GENERATING_BOUNDS.H */
361 #ifdef CONFIG_ZONE_DMA
363 * ZONE_DMA is used when there are devices that are not able
364 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
365 * carve out the portion of memory that is needed for these devices.
366 * The range is arch specific.
371 * ---------------------------
372 * parisc, ia64, sparc <4G
375 * alpha Unlimited or 0-16MB.
377 * i386, x86_64 and multiple other arches
382 #ifdef CONFIG_ZONE_DMA32
384 * x86_64 needs two ZONE_DMAs because it supports devices that are
385 * only able to do DMA to the lower 16M but also 32 bit devices that
386 * can only do DMA areas below 4G.
391 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
392 * performed on pages in ZONE_NORMAL if the DMA devices support
393 * transfers to all addressable memory.
396 #ifdef CONFIG_HIGHMEM
398 * A memory area that is only addressable by the kernel through
399 * mapping portions into its own address space. This is for example
400 * used by i386 to allow the kernel to address the memory beyond
401 * 900MB. The kernel will set up special mappings (page
402 * table entries on i386) for each page that the kernel needs to
408 #ifdef CONFIG_ZONE_DEVICE
415 #ifndef __GENERATING_BOUNDS_H
418 /* Read-mostly fields */
420 /* zone watermarks, access with *_wmark_pages(zone) macros */
421 unsigned long _watermark[NR_WMARK];
422 unsigned long watermark_boost;
424 unsigned long nr_reserved_highatomic;
427 * We don't know if the memory that we're going to allocate will be
428 * freeable or/and it will be released eventually, so to avoid totally
429 * wasting several GB of ram we must reserve some of the lower zone
430 * memory (otherwise we risk to run OOM on the lower zones despite
431 * there being tons of freeable ram on the higher zones). This array is
432 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
435 long lowmem_reserve[MAX_NR_ZONES];
440 struct pglist_data *zone_pgdat;
441 struct per_cpu_pageset __percpu *pageset;
443 #ifndef CONFIG_SPARSEMEM
445 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
446 * In SPARSEMEM, this map is stored in struct mem_section
448 unsigned long *pageblock_flags;
449 #endif /* CONFIG_SPARSEMEM */
451 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
452 unsigned long zone_start_pfn;
455 * spanned_pages is the total pages spanned by the zone, including
456 * holes, which is calculated as:
457 * spanned_pages = zone_end_pfn - zone_start_pfn;
459 * present_pages is physical pages existing within the zone, which
461 * present_pages = spanned_pages - absent_pages(pages in holes);
463 * managed_pages is present pages managed by the buddy system, which
464 * is calculated as (reserved_pages includes pages allocated by the
465 * bootmem allocator):
466 * managed_pages = present_pages - reserved_pages;
468 * So present_pages may be used by memory hotplug or memory power
469 * management logic to figure out unmanaged pages by checking
470 * (present_pages - managed_pages). And managed_pages should be used
471 * by page allocator and vm scanner to calculate all kinds of watermarks
476 * zone_start_pfn and spanned_pages are protected by span_seqlock.
477 * It is a seqlock because it has to be read outside of zone->lock,
478 * and it is done in the main allocator path. But, it is written
479 * quite infrequently.
481 * The span_seq lock is declared along with zone->lock because it is
482 * frequently read in proximity to zone->lock. It's good to
483 * give them a chance of being in the same cacheline.
485 * Write access to present_pages at runtime should be protected by
486 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
487 * present_pages should get_online_mems() to get a stable value.
489 atomic_long_t managed_pages;
490 unsigned long spanned_pages;
491 unsigned long present_pages;
495 #ifdef CONFIG_MEMORY_ISOLATION
497 * Number of isolated pageblock. It is used to solve incorrect
498 * freepage counting problem due to racy retrieving migratetype
499 * of pageblock. Protected by zone->lock.
501 unsigned long nr_isolate_pageblock;
504 #ifdef CONFIG_MEMORY_HOTPLUG
505 /* see spanned/present_pages for more description */
506 seqlock_t span_seqlock;
511 /* Write-intensive fields used from the page allocator */
514 /* free areas of different sizes */
515 struct free_area free_area[MAX_ORDER];
517 /* zone flags, see below */
520 /* Primarily protects free_area */
523 /* Write-intensive fields used by compaction and vmstats. */
527 * When free pages are below this point, additional steps are taken
528 * when reading the number of free pages to avoid per-cpu counter
529 * drift allowing watermarks to be breached
531 unsigned long percpu_drift_mark;
533 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
534 /* pfn where compaction free scanner should start */
535 unsigned long compact_cached_free_pfn;
536 /* pfn where async and sync compaction migration scanner should start */
537 unsigned long compact_cached_migrate_pfn[2];
538 unsigned long compact_init_migrate_pfn;
539 unsigned long compact_init_free_pfn;
542 #ifdef CONFIG_COMPACTION
544 * On compaction failure, 1<<compact_defer_shift compactions
545 * are skipped before trying again. The number attempted since
546 * last failure is tracked with compact_considered.
548 unsigned int compact_considered;
549 unsigned int compact_defer_shift;
550 int compact_order_failed;
553 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
554 /* Set to true when the PG_migrate_skip bits should be cleared */
555 bool compact_blockskip_flush;
561 /* Zone statistics */
562 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
563 atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
564 } ____cacheline_internodealigned_in_smp;
567 PGDAT_CONGESTED, /* pgdat has many dirty pages backed by
570 PGDAT_DIRTY, /* reclaim scanning has recently found
571 * many dirty file pages at the tail
574 PGDAT_WRITEBACK, /* reclaim scanning has recently found
575 * many pages under writeback
577 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
581 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
582 * Cleared when kswapd is woken.
586 static inline unsigned long zone_managed_pages(struct zone *zone)
588 return (unsigned long)atomic_long_read(&zone->managed_pages);
591 static inline unsigned long zone_end_pfn(const struct zone *zone)
593 return zone->zone_start_pfn + zone->spanned_pages;
596 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
598 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
601 static inline bool zone_is_initialized(struct zone *zone)
603 return zone->initialized;
606 static inline bool zone_is_empty(struct zone *zone)
608 return zone->spanned_pages == 0;
612 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
613 * intersection with the given zone
615 static inline bool zone_intersects(struct zone *zone,
616 unsigned long start_pfn, unsigned long nr_pages)
618 if (zone_is_empty(zone))
620 if (start_pfn >= zone_end_pfn(zone) ||
621 start_pfn + nr_pages <= zone->zone_start_pfn)
628 * The "priority" of VM scanning is how much of the queues we will scan in one
629 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
630 * queues ("queue_length >> 12") during an aging round.
632 #define DEF_PRIORITY 12
634 /* Maximum number of zones on a zonelist */
635 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
638 ZONELIST_FALLBACK, /* zonelist with fallback */
641 * The NUMA zonelists are doubled because we need zonelists that
642 * restrict the allocations to a single node for __GFP_THISNODE.
644 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
650 * This struct contains information about a zone in a zonelist. It is stored
651 * here to avoid dereferences into large structures and lookups of tables
654 struct zone *zone; /* Pointer to actual zone */
655 int zone_idx; /* zone_idx(zoneref->zone) */
659 * One allocation request operates on a zonelist. A zonelist
660 * is a list of zones, the first one is the 'goal' of the
661 * allocation, the other zones are fallback zones, in decreasing
664 * To speed the reading of the zonelist, the zonerefs contain the zone index
665 * of the entry being read. Helper functions to access information given
666 * a struct zoneref are
668 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
669 * zonelist_zone_idx() - Return the index of the zone for an entry
670 * zonelist_node_idx() - Return the index of the node for an entry
673 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
676 #ifndef CONFIG_DISCONTIGMEM
677 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
678 extern struct page *mem_map;
681 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
682 struct deferred_split {
683 spinlock_t split_queue_lock;
684 struct list_head split_queue;
685 unsigned long split_queue_len;
690 * On NUMA machines, each NUMA node would have a pg_data_t to describe
691 * it's memory layout. On UMA machines there is a single pglist_data which
692 * describes the whole memory.
694 * Memory statistics and page replacement data structures are maintained on a
698 typedef struct pglist_data {
699 struct zone node_zones[MAX_NR_ZONES];
700 struct zonelist node_zonelists[MAX_ZONELISTS];
702 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
703 struct page *node_mem_map;
704 #ifdef CONFIG_PAGE_EXTENSION
705 struct page_ext *node_page_ext;
708 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
710 * Must be held any time you expect node_start_pfn,
711 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
712 * Also synchronizes pgdat->first_deferred_pfn during deferred page
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 * zone 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 */
774 struct lruvec lruvec;
780 /* Per-node vmstats */
781 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
782 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
785 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
786 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
787 #ifdef CONFIG_FLAT_NODE_MEM_MAP
788 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
790 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
792 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
794 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
795 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
797 static inline struct lruvec *node_lruvec(struct pglist_data *pgdat)
799 return &pgdat->lruvec;
802 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
804 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
807 static inline bool pgdat_is_empty(pg_data_t *pgdat)
809 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
812 #include <linux/memory_hotplug.h>
814 void build_all_zonelists(pg_data_t *pgdat);
815 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
816 enum zone_type classzone_idx);
817 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
818 int classzone_idx, unsigned int alloc_flags,
820 bool zone_watermark_ok(struct zone *z, unsigned int order,
821 unsigned long mark, int classzone_idx,
822 unsigned int alloc_flags);
823 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
824 unsigned long mark, int classzone_idx);
826 * Memory initialization context, use to differentiate memory added by
827 * the platform statically or via memory hotplug interface.
829 enum meminit_context {
834 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
837 extern void lruvec_init(struct lruvec *lruvec);
839 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
842 return lruvec->pgdat;
844 return container_of(lruvec, struct pglist_data, lruvec);
848 extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
850 #ifdef CONFIG_HAVE_MEMORY_PRESENT
851 void memory_present(int nid, unsigned long start, unsigned long end);
853 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
856 #if defined(CONFIG_SPARSEMEM)
857 void memblocks_present(void);
859 static inline void memblocks_present(void) {}
862 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
863 int local_memory_node(int node_id);
865 static inline int local_memory_node(int node_id) { return node_id; };
869 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
871 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
874 * Returns true if a zone has pages managed by the buddy allocator.
875 * All the reclaim decisions have to use this function rather than
876 * populated_zone(). If the whole zone is reserved then we can easily
877 * end up with populated_zone() && !managed_zone().
879 static inline bool managed_zone(struct zone *zone)
881 return zone_managed_pages(zone);
884 /* Returns true if a zone has memory */
885 static inline bool populated_zone(struct zone *zone)
887 return zone->present_pages;
891 static inline int zone_to_nid(struct zone *zone)
896 static inline void zone_set_nid(struct zone *zone, int nid)
901 static inline int zone_to_nid(struct zone *zone)
906 static inline void zone_set_nid(struct zone *zone, int nid) {}
909 extern int movable_zone;
911 #ifdef CONFIG_HIGHMEM
912 static inline int zone_movable_is_highmem(void)
914 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
915 return movable_zone == ZONE_HIGHMEM;
917 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
922 static inline int is_highmem_idx(enum zone_type idx)
924 #ifdef CONFIG_HIGHMEM
925 return (idx == ZONE_HIGHMEM ||
926 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
932 #ifdef CONFIG_ZONE_DMA
933 bool has_managed_dma(void);
935 static inline bool has_managed_dma(void)
942 * is_highmem - helper function to quickly check if a struct zone is a
943 * highmem zone or not. This is an attempt to keep references
944 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
945 * @zone - pointer to struct zone variable
947 static inline int is_highmem(struct zone *zone)
949 #ifdef CONFIG_HIGHMEM
950 return is_highmem_idx(zone_idx(zone));
956 /* These two functions are used to setup the per zone pages min values */
958 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
959 void __user *, size_t *, loff_t *);
960 int watermark_boost_factor_sysctl_handler(struct ctl_table *, int,
961 void __user *, size_t *, loff_t *);
962 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int,
963 void __user *, size_t *, loff_t *);
964 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
965 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
966 void __user *, size_t *, loff_t *);
967 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
968 void __user *, size_t *, loff_t *);
969 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
970 void __user *, size_t *, loff_t *);
971 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
972 void __user *, size_t *, loff_t *);
974 extern int numa_zonelist_order_handler(struct ctl_table *, int,
975 void __user *, size_t *, loff_t *);
976 extern char numa_zonelist_order[];
977 #define NUMA_ZONELIST_ORDER_LEN 16
979 #ifndef CONFIG_NEED_MULTIPLE_NODES
981 extern struct pglist_data contig_page_data;
982 #define NODE_DATA(nid) (&contig_page_data)
983 #define NODE_MEM_MAP(nid) mem_map
985 #else /* CONFIG_NEED_MULTIPLE_NODES */
987 #include <asm/mmzone.h>
989 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
991 extern struct pglist_data *first_online_pgdat(void);
992 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
993 extern struct zone *next_zone(struct zone *zone);
996 * for_each_online_pgdat - helper macro to iterate over all online nodes
997 * @pgdat - pointer to a pg_data_t variable
999 #define for_each_online_pgdat(pgdat) \
1000 for (pgdat = first_online_pgdat(); \
1002 pgdat = next_online_pgdat(pgdat))
1004 * for_each_zone - helper macro to iterate over all memory zones
1005 * @zone - pointer to struct zone variable
1007 * The user only needs to declare the zone variable, for_each_zone
1010 #define for_each_zone(zone) \
1011 for (zone = (first_online_pgdat())->node_zones; \
1013 zone = next_zone(zone))
1015 #define for_each_populated_zone(zone) \
1016 for (zone = (first_online_pgdat())->node_zones; \
1018 zone = next_zone(zone)) \
1019 if (!populated_zone(zone)) \
1020 ; /* do nothing */ \
1023 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1025 return zoneref->zone;
1028 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1030 return zoneref->zone_idx;
1033 static inline int zonelist_node_idx(struct zoneref *zoneref)
1035 return zone_to_nid(zoneref->zone);
1038 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1039 enum zone_type highest_zoneidx,
1043 * 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
1044 * @z - The cursor used as a starting point for the search
1045 * @highest_zoneidx - The zone index of the highest zone to return
1046 * @nodes - An optional nodemask to filter the zonelist with
1048 * This function returns the next zone at or below a given zone index that is
1049 * within the allowed nodemask using a cursor as the starting point for the
1050 * search. The zoneref returned is a cursor that represents the current zone
1051 * being examined. It should be advanced by one before calling
1052 * next_zones_zonelist again.
1054 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1055 enum zone_type highest_zoneidx,
1058 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1060 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1064 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1065 * @zonelist - The zonelist to search for a suitable zone
1066 * @highest_zoneidx - The zone index of the highest zone to return
1067 * @nodes - An optional nodemask to filter the zonelist with
1068 * @return - Zoneref pointer for the first suitable zone found (see below)
1070 * This function returns the first zone at or below a given zone index that is
1071 * within the allowed nodemask. The zoneref returned is a cursor that can be
1072 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1073 * one before calling.
1075 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1076 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1077 * update due to cpuset modification.
1079 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1080 enum zone_type highest_zoneidx,
1083 return next_zones_zonelist(zonelist->_zonerefs,
1084 highest_zoneidx, nodes);
1088 * 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
1089 * @zone - The current zone in the iterator
1090 * @z - The current pointer within zonelist->zones being iterated
1091 * @zlist - The zonelist being iterated
1092 * @highidx - The zone index of the highest zone to return
1093 * @nodemask - Nodemask allowed by the allocator
1095 * This iterator iterates though all zones at or below a given zone index and
1096 * within a given nodemask
1098 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1099 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1101 z = next_zones_zonelist(++z, highidx, nodemask), \
1102 zone = zonelist_zone(z))
1104 #define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1105 for (zone = z->zone; \
1107 z = next_zones_zonelist(++z, highidx, nodemask), \
1108 zone = zonelist_zone(z))
1112 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1113 * @zone - The current zone in the iterator
1114 * @z - The current pointer within zonelist->zones being iterated
1115 * @zlist - The zonelist being iterated
1116 * @highidx - The zone index of the highest zone to return
1118 * This iterator iterates though all zones at or below a given zone index.
1120 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1121 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1123 #ifdef CONFIG_SPARSEMEM
1124 #include <asm/sparsemem.h>
1127 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1128 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1129 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1131 BUILD_BUG_ON(IS_ENABLED(CONFIG_NUMA));
1136 #ifdef CONFIG_FLATMEM
1137 #define pfn_to_nid(pfn) (0)
1140 #ifdef CONFIG_SPARSEMEM
1143 * SECTION_SHIFT #bits space required to store a section #
1145 * PA_SECTION_SHIFT physical address to/from section number
1146 * PFN_SECTION_SHIFT pfn to/from section number
1148 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1149 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1151 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1153 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1154 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1156 #define SECTION_BLOCKFLAGS_BITS \
1157 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1159 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1160 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1163 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1165 return pfn >> PFN_SECTION_SHIFT;
1167 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1169 return sec << PFN_SECTION_SHIFT;
1172 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1173 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1175 #define SUBSECTION_SHIFT 21
1177 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1178 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1179 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1181 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1182 #error Subsection size exceeds section size
1184 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1187 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1188 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1190 struct mem_section_usage {
1191 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1192 /* See declaration of similar field in struct zone */
1193 unsigned long pageblock_flags[0];
1196 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1200 struct mem_section {
1202 * This is, logically, a pointer to an array of struct
1203 * pages. However, it is stored with some other magic.
1204 * (see sparse.c::sparse_init_one_section())
1206 * Additionally during early boot we encode node id of
1207 * the location of the section here to guide allocation.
1208 * (see sparse.c::memory_present())
1210 * Making it a UL at least makes someone do a cast
1211 * before using it wrong.
1213 unsigned long section_mem_map;
1215 struct mem_section_usage *usage;
1216 #ifdef CONFIG_PAGE_EXTENSION
1218 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1219 * section. (see page_ext.h about this.)
1221 struct page_ext *page_ext;
1225 * WARNING: mem_section must be a power-of-2 in size for the
1226 * calculation and use of SECTION_ROOT_MASK to make sense.
1230 #ifdef CONFIG_SPARSEMEM_EXTREME
1231 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1233 #define SECTIONS_PER_ROOT 1
1236 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1237 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1238 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1240 #ifdef CONFIG_SPARSEMEM_EXTREME
1241 extern struct mem_section **mem_section;
1243 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1246 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1248 return ms->usage->pageblock_flags;
1251 static inline struct mem_section *__nr_to_section(unsigned long nr)
1253 unsigned long root = SECTION_NR_TO_ROOT(nr);
1255 if (unlikely(root >= NR_SECTION_ROOTS))
1258 #ifdef CONFIG_SPARSEMEM_EXTREME
1259 if (!mem_section || !mem_section[root])
1262 return &mem_section[root][nr & SECTION_ROOT_MASK];
1264 extern unsigned long __section_nr(struct mem_section *ms);
1265 extern size_t mem_section_usage_size(void);
1268 * We use the lower bits of the mem_map pointer to store
1269 * a little bit of information. The pointer is calculated
1270 * as mem_map - section_nr_to_pfn(pnum). The result is
1271 * aligned to the minimum alignment of the two values:
1272 * 1. All mem_map arrays are page-aligned.
1273 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1274 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1275 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1276 * worst combination is powerpc with 256k pages,
1277 * which results in PFN_SECTION_SHIFT equal 6.
1278 * To sum it up, at least 6 bits are available.
1280 #define SECTION_MARKED_PRESENT (1UL<<0)
1281 #define SECTION_HAS_MEM_MAP (1UL<<1)
1282 #define SECTION_IS_ONLINE (1UL<<2)
1283 #define SECTION_IS_EARLY (1UL<<3)
1284 #define SECTION_MAP_LAST_BIT (1UL<<4)
1285 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1286 #define SECTION_NID_SHIFT 3
1288 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1290 unsigned long map = section->section_mem_map;
1291 map &= SECTION_MAP_MASK;
1292 return (struct page *)map;
1295 static inline int present_section(struct mem_section *section)
1297 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1300 static inline int present_section_nr(unsigned long nr)
1302 return present_section(__nr_to_section(nr));
1305 static inline int valid_section(struct mem_section *section)
1307 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1310 static inline int early_section(struct mem_section *section)
1312 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1315 static inline int valid_section_nr(unsigned long nr)
1317 return valid_section(__nr_to_section(nr));
1320 static inline int online_section(struct mem_section *section)
1322 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1325 static inline int online_section_nr(unsigned long nr)
1327 return online_section(__nr_to_section(nr));
1330 #ifdef CONFIG_MEMORY_HOTPLUG
1331 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1332 #ifdef CONFIG_MEMORY_HOTREMOVE
1333 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1337 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1339 return __nr_to_section(pfn_to_section_nr(pfn));
1342 extern unsigned long __highest_present_section_nr;
1344 static inline int subsection_map_index(unsigned long pfn)
1346 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1349 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1350 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1352 int idx = subsection_map_index(pfn);
1354 return test_bit(idx, ms->usage->subsection_map);
1357 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1363 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1364 static inline int pfn_valid(unsigned long pfn)
1366 struct mem_section *ms;
1368 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1370 ms = __nr_to_section(pfn_to_section_nr(pfn));
1371 if (!valid_section(ms))
1374 * Traditionally early sections always returned pfn_valid() for
1375 * the entire section-sized span.
1377 return early_section(ms) || pfn_section_valid(ms, pfn);
1381 static inline int pfn_present(unsigned long pfn)
1383 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1385 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1389 * These are _only_ used during initialisation, therefore they
1390 * can use __initdata ... They could have names to indicate
1394 #define pfn_to_nid(pfn) \
1396 unsigned long __pfn_to_nid_pfn = (pfn); \
1397 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1400 #define pfn_to_nid(pfn) (0)
1403 #define early_pfn_valid(pfn) pfn_valid(pfn)
1404 void sparse_init(void);
1406 #define sparse_init() do {} while (0)
1407 #define sparse_index_init(_sec, _nid) do {} while (0)
1408 #define pfn_present pfn_valid
1409 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
1410 #endif /* CONFIG_SPARSEMEM */
1413 * During memory init memblocks map pfns to nids. The search is expensive and
1414 * this caches recent lookups. The implementation of __early_pfn_to_nid
1415 * may treat start/end as pfns or sections.
1417 struct mminit_pfnnid_cache {
1418 unsigned long last_start;
1419 unsigned long last_end;
1423 #ifndef early_pfn_valid
1424 #define early_pfn_valid(pfn) (1)
1427 void memory_present(int nid, unsigned long start, unsigned long end);
1430 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1431 * need to check pfn validity within that MAX_ORDER_NR_PAGES block.
1432 * pfn_valid_within() should be used in this case; we optimise this away
1433 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1435 #ifdef CONFIG_HOLES_IN_ZONE
1436 #define pfn_valid_within(pfn) pfn_valid(pfn)
1438 #define pfn_valid_within(pfn) (1)
1441 #endif /* !__GENERATING_BOUNDS.H */
1442 #endif /* !__ASSEMBLY__ */
1443 #endif /* _LINUX_MMZONE_H */