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>
23 /* Free memory management - zoned buddy allocator. */
24 #ifndef CONFIG_FORCE_MAX_ZONEORDER
27 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
29 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
32 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
33 * costly to service. That is between allocation orders which should
34 * coalesce naturally under reasonable reclaim pressure and those which
37 #define PAGE_ALLOC_COSTLY_ORDER 3
43 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
44 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
47 * MIGRATE_CMA migration type is designed to mimic the way
48 * ZONE_MOVABLE works. Only movable pages can be allocated
49 * from MIGRATE_CMA pageblocks and page allocator never
50 * implicitly change migration type of MIGRATE_CMA pageblock.
52 * The way to use it is to change migratetype of a range of
53 * pageblocks to MIGRATE_CMA which can be done by
54 * __free_pageblock_cma() function. What is important though
55 * is that a range of pageblocks must be aligned to
56 * MAX_ORDER_NR_PAGES should biggest page be bigger then
61 #ifdef CONFIG_MEMORY_ISOLATION
62 MIGRATE_ISOLATE, /* can't allocate from here */
67 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
68 extern char * const migratetype_names[MIGRATE_TYPES];
71 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
72 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
74 # define is_migrate_cma(migratetype) false
75 # define is_migrate_cma_page(_page) false
78 static inline bool is_migrate_movable(int mt)
80 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
83 #define for_each_migratetype_order(order, type) \
84 for (order = 0; order < MAX_ORDER; order++) \
85 for (type = 0; type < MIGRATE_TYPES; type++)
87 extern int page_group_by_mobility_disabled;
89 #define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1)
90 #define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1)
92 #define get_pageblock_migratetype(page) \
93 get_pfnblock_flags_mask(page, page_to_pfn(page), \
94 PB_migrate_end, MIGRATETYPE_MASK)
97 struct list_head free_list[MIGRATE_TYPES];
98 unsigned long nr_free;
104 * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
105 * So add a wild amount of padding here to ensure that they fall into separate
106 * cachelines. There are very few zone structures in the machine, so space
107 * consumption is not a concern here.
109 #if defined(CONFIG_SMP)
110 struct zone_padding {
112 } ____cacheline_internodealigned_in_smp;
113 #define ZONE_PADDING(name) struct zone_padding name;
115 #define ZONE_PADDING(name)
119 enum numa_stat_item {
120 NUMA_HIT, /* allocated in intended node */
121 NUMA_MISS, /* allocated in non intended node */
122 NUMA_FOREIGN, /* was intended here, hit elsewhere */
123 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
124 NUMA_LOCAL, /* allocation from local node */
125 NUMA_OTHER, /* allocation from other node */
126 NR_VM_NUMA_STAT_ITEMS
129 #define NR_VM_NUMA_STAT_ITEMS 0
132 enum zone_stat_item {
133 /* First 128 byte cacheline (assuming 64 bit words) */
135 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
136 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
138 NR_ZONE_INACTIVE_FILE,
141 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
142 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
143 NR_PAGETABLE, /* used for pagetables */
144 NR_KERNEL_STACK_KB, /* measured in KiB */
145 /* Second 128 byte cacheline */
147 #if IS_ENABLED(CONFIG_ZSMALLOC)
148 NR_ZSPAGES, /* allocated in zsmalloc */
151 NR_VM_ZONE_STAT_ITEMS };
153 enum node_stat_item {
155 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
156 NR_ACTIVE_ANON, /* " " " " " */
157 NR_INACTIVE_FILE, /* " " " " " */
158 NR_ACTIVE_FILE, /* " " " " " */
159 NR_UNEVICTABLE, /* " " " " " */
161 NR_SLAB_UNRECLAIMABLE,
162 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
163 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
166 WORKINGSET_NODERECLAIM,
167 NR_ANON_MAPPED, /* Mapped anonymous pages */
168 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
169 only modified from process context */
173 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
174 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
178 NR_UNSTABLE_NFS, /* NFS unstable pages */
180 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
181 NR_DIRTIED, /* page dirtyings since bootup */
182 NR_WRITTEN, /* page writings since bootup */
183 NR_INDIRECTLY_RECLAIMABLE_BYTES, /* measured in bytes */
184 NR_VM_NODE_STAT_ITEMS
188 * We do arithmetic on the LRU lists in various places in the code,
189 * so it is important to keep the active lists LRU_ACTIVE higher in
190 * the array than the corresponding inactive lists, and to keep
191 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
193 * This has to be kept in sync with the statistics in zone_stat_item
194 * above and the descriptions in vmstat_text in mm/vmstat.c
201 LRU_INACTIVE_ANON = LRU_BASE,
202 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
203 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
204 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
209 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
211 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
213 static inline int is_file_lru(enum lru_list lru)
215 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
218 static inline int is_active_lru(enum lru_list lru)
220 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
223 struct zone_reclaim_stat {
225 * The pageout code in vmscan.c keeps track of how many of the
226 * mem/swap backed and file backed pages are referenced.
227 * The higher the rotated/scanned ratio, the more valuable
230 * The anon LRU stats live in [0], file LRU stats in [1]
232 unsigned long recent_rotated[2];
233 unsigned long recent_scanned[2];
237 struct list_head lists[NR_LRU_LISTS];
238 struct zone_reclaim_stat reclaim_stat;
239 /* Evictions & activations on the inactive file list */
240 atomic_long_t inactive_age;
241 /* Refaults at the time of last reclaim cycle */
242 unsigned long refaults;
244 struct pglist_data *pgdat;
248 /* Mask used at gathering information at once (see memcontrol.c) */
249 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
250 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
251 #define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
253 /* Isolate unmapped file */
254 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
255 /* Isolate for asynchronous migration */
256 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
257 /* Isolate unevictable pages */
258 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
260 /* LRU Isolation modes. */
261 typedef unsigned __bitwise isolate_mode_t;
263 enum zone_watermarks {
270 #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
271 #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
272 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
274 struct per_cpu_pages {
275 int count; /* number of pages in the list */
276 int high; /* high watermark, emptying needed */
277 int batch; /* chunk size for buddy add/remove */
279 /* Lists of pages, one per migrate type stored on the pcp-lists */
280 struct list_head lists[MIGRATE_PCPTYPES];
283 struct per_cpu_pageset {
284 struct per_cpu_pages pcp;
287 u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
291 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
295 struct per_cpu_nodestat {
297 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
300 #endif /* !__GENERATING_BOUNDS.H */
303 #ifdef CONFIG_ZONE_DMA
305 * ZONE_DMA is used when there are devices that are not able
306 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
307 * carve out the portion of memory that is needed for these devices.
308 * The range is arch specific.
313 * ---------------------------
314 * parisc, ia64, sparc <4G
317 * alpha Unlimited or 0-16MB.
319 * i386, x86_64 and multiple other arches
324 #ifdef CONFIG_ZONE_DMA32
326 * x86_64 needs two ZONE_DMAs because it supports devices that are
327 * only able to do DMA to the lower 16M but also 32 bit devices that
328 * can only do DMA areas below 4G.
333 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
334 * performed on pages in ZONE_NORMAL if the DMA devices support
335 * transfers to all addressable memory.
338 #ifdef CONFIG_HIGHMEM
340 * A memory area that is only addressable by the kernel through
341 * mapping portions into its own address space. This is for example
342 * used by i386 to allow the kernel to address the memory beyond
343 * 900MB. The kernel will set up special mappings (page
344 * table entries on i386) for each page that the kernel needs to
350 #ifdef CONFIG_ZONE_DEVICE
357 #ifndef __GENERATING_BOUNDS_H
360 /* Read-mostly fields */
362 /* zone watermarks, access with *_wmark_pages(zone) macros */
363 unsigned long watermark[NR_WMARK];
365 unsigned long nr_reserved_highatomic;
368 * We don't know if the memory that we're going to allocate will be
369 * freeable or/and it will be released eventually, so to avoid totally
370 * wasting several GB of ram we must reserve some of the lower zone
371 * memory (otherwise we risk to run OOM on the lower zones despite
372 * there being tons of freeable ram on the higher zones). This array is
373 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
376 long lowmem_reserve[MAX_NR_ZONES];
381 struct pglist_data *zone_pgdat;
382 struct per_cpu_pageset __percpu *pageset;
384 #ifndef CONFIG_SPARSEMEM
386 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
387 * In SPARSEMEM, this map is stored in struct mem_section
389 unsigned long *pageblock_flags;
390 #endif /* CONFIG_SPARSEMEM */
392 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
393 unsigned long zone_start_pfn;
396 * spanned_pages is the total pages spanned by the zone, including
397 * holes, which is calculated as:
398 * spanned_pages = zone_end_pfn - zone_start_pfn;
400 * present_pages is physical pages existing within the zone, which
402 * present_pages = spanned_pages - absent_pages(pages in holes);
404 * managed_pages is present pages managed by the buddy system, which
405 * is calculated as (reserved_pages includes pages allocated by the
406 * bootmem allocator):
407 * managed_pages = present_pages - reserved_pages;
409 * So present_pages may be used by memory hotplug or memory power
410 * management logic to figure out unmanaged pages by checking
411 * (present_pages - managed_pages). And managed_pages should be used
412 * by page allocator and vm scanner to calculate all kinds of watermarks
417 * zone_start_pfn and spanned_pages are protected by span_seqlock.
418 * It is a seqlock because it has to be read outside of zone->lock,
419 * and it is done in the main allocator path. But, it is written
420 * quite infrequently.
422 * The span_seq lock is declared along with zone->lock because it is
423 * frequently read in proximity to zone->lock. It's good to
424 * give them a chance of being in the same cacheline.
426 * Write access to present_pages at runtime should be protected by
427 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
428 * present_pages should get_online_mems() to get a stable value.
430 * Read access to managed_pages should be safe because it's unsigned
431 * long. Write access to zone->managed_pages and totalram_pages are
432 * protected by managed_page_count_lock at runtime. Idealy only
433 * adjust_managed_page_count() should be used instead of directly
434 * touching zone->managed_pages and totalram_pages.
436 unsigned long managed_pages;
437 unsigned long spanned_pages;
438 unsigned long present_pages;
442 #ifdef CONFIG_MEMORY_ISOLATION
444 * Number of isolated pageblock. It is used to solve incorrect
445 * freepage counting problem due to racy retrieving migratetype
446 * of pageblock. Protected by zone->lock.
448 unsigned long nr_isolate_pageblock;
451 #ifdef CONFIG_MEMORY_HOTPLUG
452 /* see spanned/present_pages for more description */
453 seqlock_t span_seqlock;
458 /* Write-intensive fields used from the page allocator */
461 /* free areas of different sizes */
462 struct free_area free_area[MAX_ORDER];
464 /* zone flags, see below */
467 /* Primarily protects free_area */
470 /* Write-intensive fields used by compaction and vmstats. */
474 * When free pages are below this point, additional steps are taken
475 * when reading the number of free pages to avoid per-cpu counter
476 * drift allowing watermarks to be breached
478 unsigned long percpu_drift_mark;
480 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
481 /* pfn where compaction free scanner should start */
482 unsigned long compact_cached_free_pfn;
483 /* pfn where async and sync compaction migration scanner should start */
484 unsigned long compact_cached_migrate_pfn[2];
487 #ifdef CONFIG_COMPACTION
489 * On compaction failure, 1<<compact_defer_shift compactions
490 * are skipped before trying again. The number attempted since
491 * last failure is tracked with compact_considered.
493 unsigned int compact_considered;
494 unsigned int compact_defer_shift;
495 int compact_order_failed;
498 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
499 /* Set to true when the PG_migrate_skip bits should be cleared */
500 bool compact_blockskip_flush;
506 /* Zone statistics */
507 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
508 atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
509 } ____cacheline_internodealigned_in_smp;
512 PGDAT_CONGESTED, /* pgdat has many dirty pages backed by
515 PGDAT_DIRTY, /* reclaim scanning has recently found
516 * many dirty file pages at the tail
519 PGDAT_WRITEBACK, /* reclaim scanning has recently found
520 * many pages under writeback
522 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
525 static inline unsigned long zone_end_pfn(const struct zone *zone)
527 return zone->zone_start_pfn + zone->spanned_pages;
530 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
532 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
535 static inline bool zone_is_initialized(struct zone *zone)
537 return zone->initialized;
540 static inline bool zone_is_empty(struct zone *zone)
542 return zone->spanned_pages == 0;
546 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
547 * intersection with the given zone
549 static inline bool zone_intersects(struct zone *zone,
550 unsigned long start_pfn, unsigned long nr_pages)
552 if (zone_is_empty(zone))
554 if (start_pfn >= zone_end_pfn(zone) ||
555 start_pfn + nr_pages <= zone->zone_start_pfn)
562 * The "priority" of VM scanning is how much of the queues we will scan in one
563 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
564 * queues ("queue_length >> 12") during an aging round.
566 #define DEF_PRIORITY 12
568 /* Maximum number of zones on a zonelist */
569 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
572 ZONELIST_FALLBACK, /* zonelist with fallback */
575 * The NUMA zonelists are doubled because we need zonelists that
576 * restrict the allocations to a single node for __GFP_THISNODE.
578 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
584 * This struct contains information about a zone in a zonelist. It is stored
585 * here to avoid dereferences into large structures and lookups of tables
588 struct zone *zone; /* Pointer to actual zone */
589 int zone_idx; /* zone_idx(zoneref->zone) */
593 * One allocation request operates on a zonelist. A zonelist
594 * is a list of zones, the first one is the 'goal' of the
595 * allocation, the other zones are fallback zones, in decreasing
598 * To speed the reading of the zonelist, the zonerefs contain the zone index
599 * of the entry being read. Helper functions to access information given
600 * a struct zoneref are
602 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
603 * zonelist_zone_idx() - Return the index of the zone for an entry
604 * zonelist_node_idx() - Return the index of the node for an entry
607 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
610 #ifndef CONFIG_DISCONTIGMEM
611 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
612 extern struct page *mem_map;
616 * On NUMA machines, each NUMA node would have a pg_data_t to describe
617 * it's memory layout. On UMA machines there is a single pglist_data which
618 * describes the whole memory.
620 * Memory statistics and page replacement data structures are maintained on a
624 typedef struct pglist_data {
625 struct zone node_zones[MAX_NR_ZONES];
626 struct zonelist node_zonelists[MAX_ZONELISTS];
628 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
629 struct page *node_mem_map;
630 #ifdef CONFIG_PAGE_EXTENSION
631 struct page_ext *node_page_ext;
634 #ifndef CONFIG_NO_BOOTMEM
635 struct bootmem_data *bdata;
637 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
639 * Must be held any time you expect node_start_pfn, node_present_pages
640 * or node_spanned_pages stay constant. Holding this will also
641 * guarantee that any pfn_valid() stays that way.
643 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
644 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
645 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
647 * Nests above zone->lock and zone->span_seqlock
649 spinlock_t node_size_lock;
651 unsigned long node_start_pfn;
652 unsigned long node_present_pages; /* total number of physical pages */
653 unsigned long node_spanned_pages; /* total size of physical page
654 range, including holes */
656 wait_queue_head_t kswapd_wait;
657 wait_queue_head_t pfmemalloc_wait;
658 struct task_struct *kswapd; /* Protected by
659 mem_hotplug_begin/end() */
661 enum zone_type kswapd_classzone_idx;
663 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
665 #ifdef CONFIG_COMPACTION
666 int kcompactd_max_order;
667 enum zone_type kcompactd_classzone_idx;
668 wait_queue_head_t kcompactd_wait;
669 struct task_struct *kcompactd;
671 #ifdef CONFIG_NUMA_BALANCING
672 /* Lock serializing the migrate rate limiting window */
673 spinlock_t numabalancing_migrate_lock;
675 /* Rate limiting time interval */
676 unsigned long numabalancing_migrate_next_window;
678 /* Number of pages migrated during the rate limiting time interval */
679 unsigned long numabalancing_migrate_nr_pages;
682 * This is a per-node reserve of pages that are not available
683 * to userspace allocations.
685 unsigned long totalreserve_pages;
689 * zone reclaim becomes active if more unmapped pages exist.
691 unsigned long min_unmapped_pages;
692 unsigned long min_slab_pages;
693 #endif /* CONFIG_NUMA */
695 /* Write-intensive fields used by page reclaim */
699 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
701 * If memory initialisation on large machines is deferred then this
702 * is the first PFN that needs to be initialised.
704 unsigned long first_deferred_pfn;
705 /* Number of non-deferred pages */
706 unsigned long static_init_pgcnt;
707 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
709 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
710 spinlock_t split_queue_lock;
711 struct list_head split_queue;
712 unsigned long split_queue_len;
715 /* Fields commonly accessed by the page reclaim scanner */
716 struct lruvec lruvec;
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 void build_all_zonelists(pg_data_t *pgdat);
780 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
781 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 void 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, int zone_idx);
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; };
823 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
825 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
828 * Returns true if a zone has pages managed by the buddy allocator.
829 * All the reclaim decisions have to use this function rather than
830 * populated_zone(). If the whole zone is reserved then we can easily
831 * end up with populated_zone() && !managed_zone().
833 static inline bool managed_zone(struct zone *zone)
835 return zone->managed_pages;
838 /* Returns true if a zone has memory */
839 static inline bool populated_zone(struct zone *zone)
841 return zone->present_pages;
844 extern int movable_zone;
846 #ifdef CONFIG_HIGHMEM
847 static inline int zone_movable_is_highmem(void)
849 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
850 return movable_zone == ZONE_HIGHMEM;
852 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
857 static inline int is_highmem_idx(enum zone_type idx)
859 #ifdef CONFIG_HIGHMEM
860 return (idx == ZONE_HIGHMEM ||
861 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
868 * is_highmem - helper function to quickly check if a struct zone is a
869 * highmem zone or not. This is an attempt to keep references
870 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
871 * @zone - pointer to struct zone variable
873 static inline int is_highmem(struct zone *zone)
875 #ifdef CONFIG_HIGHMEM
876 return is_highmem_idx(zone_idx(zone));
882 /* These two functions are used to setup the per zone pages min values */
884 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
885 void __user *, size_t *, loff_t *);
886 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int,
887 void __user *, size_t *, loff_t *);
888 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
889 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
890 void __user *, size_t *, loff_t *);
891 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
892 void __user *, size_t *, loff_t *);
893 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
894 void __user *, size_t *, loff_t *);
895 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
896 void __user *, size_t *, loff_t *);
898 extern int numa_zonelist_order_handler(struct ctl_table *, int,
899 void __user *, size_t *, loff_t *);
900 extern char numa_zonelist_order[];
901 #define NUMA_ZONELIST_ORDER_LEN 16
903 #ifndef CONFIG_NEED_MULTIPLE_NODES
905 extern struct pglist_data contig_page_data;
906 #define NODE_DATA(nid) (&contig_page_data)
907 #define NODE_MEM_MAP(nid) mem_map
909 #else /* CONFIG_NEED_MULTIPLE_NODES */
911 #include <asm/mmzone.h>
913 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
915 extern struct pglist_data *first_online_pgdat(void);
916 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
917 extern struct zone *next_zone(struct zone *zone);
920 * for_each_online_pgdat - helper macro to iterate over all online nodes
921 * @pgdat - pointer to a pg_data_t variable
923 #define for_each_online_pgdat(pgdat) \
924 for (pgdat = first_online_pgdat(); \
926 pgdat = next_online_pgdat(pgdat))
928 * for_each_zone - helper macro to iterate over all memory zones
929 * @zone - pointer to struct zone variable
931 * The user only needs to declare the zone variable, for_each_zone
934 #define for_each_zone(zone) \
935 for (zone = (first_online_pgdat())->node_zones; \
937 zone = next_zone(zone))
939 #define for_each_populated_zone(zone) \
940 for (zone = (first_online_pgdat())->node_zones; \
942 zone = next_zone(zone)) \
943 if (!populated_zone(zone)) \
947 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
949 return zoneref->zone;
952 static inline int zonelist_zone_idx(struct zoneref *zoneref)
954 return zoneref->zone_idx;
957 static inline int zonelist_node_idx(struct zoneref *zoneref)
960 /* zone_to_nid not available in this context */
961 return zoneref->zone->node;
964 #endif /* CONFIG_NUMA */
967 struct zoneref *__next_zones_zonelist(struct zoneref *z,
968 enum zone_type highest_zoneidx,
972 * 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
973 * @z - The cursor used as a starting point for the search
974 * @highest_zoneidx - The zone index of the highest zone to return
975 * @nodes - An optional nodemask to filter the zonelist with
977 * This function returns the next zone at or below a given zone index that is
978 * within the allowed nodemask using a cursor as the starting point for the
979 * search. The zoneref returned is a cursor that represents the current zone
980 * being examined. It should be advanced by one before calling
981 * next_zones_zonelist again.
983 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
984 enum zone_type highest_zoneidx,
987 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
989 return __next_zones_zonelist(z, highest_zoneidx, nodes);
993 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
994 * @zonelist - The zonelist to search for a suitable zone
995 * @highest_zoneidx - The zone index of the highest zone to return
996 * @nodes - An optional nodemask to filter the zonelist with
997 * @return - Zoneref pointer for the first suitable zone found (see below)
999 * This function returns the first zone at or below a given zone index that is
1000 * within the allowed nodemask. The zoneref returned is a cursor that can be
1001 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1002 * one before calling.
1004 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1005 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1006 * update due to cpuset modification.
1008 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1009 enum zone_type highest_zoneidx,
1012 return next_zones_zonelist(zonelist->_zonerefs,
1013 highest_zoneidx, nodes);
1017 * 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
1018 * @zone - The current zone in the iterator
1019 * @z - The current pointer within zonelist->zones being iterated
1020 * @zlist - The zonelist being iterated
1021 * @highidx - The zone index of the highest zone to return
1022 * @nodemask - Nodemask allowed by the allocator
1024 * This iterator iterates though all zones at or below a given zone index and
1025 * within a given nodemask
1027 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1028 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1030 z = next_zones_zonelist(++z, highidx, nodemask), \
1031 zone = zonelist_zone(z))
1033 #define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1034 for (zone = z->zone; \
1036 z = next_zones_zonelist(++z, highidx, nodemask), \
1037 zone = zonelist_zone(z))
1041 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1042 * @zone - The current zone in the iterator
1043 * @z - The current pointer within zonelist->zones being iterated
1044 * @zlist - The zonelist being iterated
1045 * @highidx - The zone index of the highest zone to return
1047 * This iterator iterates though all zones at or below a given zone index.
1049 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1050 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1052 #ifdef CONFIG_SPARSEMEM
1053 #include <asm/sparsemem.h>
1056 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1057 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1058 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1060 BUILD_BUG_ON(IS_ENABLED(CONFIG_NUMA));
1065 #ifdef CONFIG_FLATMEM
1066 #define pfn_to_nid(pfn) (0)
1069 #ifdef CONFIG_SPARSEMEM
1072 * SECTION_SHIFT #bits space required to store a section #
1074 * PA_SECTION_SHIFT physical address to/from section number
1075 * PFN_SECTION_SHIFT pfn to/from section number
1077 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1078 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1080 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1082 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1083 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1085 #define SECTION_BLOCKFLAGS_BITS \
1086 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1088 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1089 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1092 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1094 return pfn >> PFN_SECTION_SHIFT;
1096 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1098 return sec << PFN_SECTION_SHIFT;
1101 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1102 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1106 struct mem_section {
1108 * This is, logically, a pointer to an array of struct
1109 * pages. However, it is stored with some other magic.
1110 * (see sparse.c::sparse_init_one_section())
1112 * Additionally during early boot we encode node id of
1113 * the location of the section here to guide allocation.
1114 * (see sparse.c::memory_present())
1116 * Making it a UL at least makes someone do a cast
1117 * before using it wrong.
1119 unsigned long section_mem_map;
1121 /* See declaration of similar field in struct zone */
1122 unsigned long *pageblock_flags;
1123 #ifdef CONFIG_PAGE_EXTENSION
1125 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1126 * section. (see page_ext.h about this.)
1128 struct page_ext *page_ext;
1132 * WARNING: mem_section must be a power-of-2 in size for the
1133 * calculation and use of SECTION_ROOT_MASK to make sense.
1137 #ifdef CONFIG_SPARSEMEM_EXTREME
1138 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1140 #define SECTIONS_PER_ROOT 1
1143 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1144 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1145 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1147 #ifdef CONFIG_SPARSEMEM_EXTREME
1148 extern struct mem_section **mem_section;
1150 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1153 static inline struct mem_section *__nr_to_section(unsigned long nr)
1155 #ifdef CONFIG_SPARSEMEM_EXTREME
1159 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1161 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1163 extern int __section_nr(struct mem_section* ms);
1164 extern unsigned long usemap_size(void);
1167 * We use the lower bits of the mem_map pointer to store
1168 * a little bit of information. The pointer is calculated
1169 * as mem_map - section_nr_to_pfn(pnum). The result is
1170 * aligned to the minimum alignment of the two values:
1171 * 1. All mem_map arrays are page-aligned.
1172 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1173 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1174 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1175 * worst combination is powerpc with 256k pages,
1176 * which results in PFN_SECTION_SHIFT equal 6.
1177 * To sum it up, at least 6 bits are available.
1179 #define SECTION_MARKED_PRESENT (1UL<<0)
1180 #define SECTION_HAS_MEM_MAP (1UL<<1)
1181 #define SECTION_IS_ONLINE (1UL<<2)
1182 #define SECTION_MAP_LAST_BIT (1UL<<3)
1183 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1184 #define SECTION_NID_SHIFT 3
1186 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1188 unsigned long map = section->section_mem_map;
1189 map &= SECTION_MAP_MASK;
1190 return (struct page *)map;
1193 static inline int present_section(struct mem_section *section)
1195 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1198 static inline int present_section_nr(unsigned long nr)
1200 return present_section(__nr_to_section(nr));
1203 static inline int valid_section(struct mem_section *section)
1205 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1208 static inline int valid_section_nr(unsigned long nr)
1210 return valid_section(__nr_to_section(nr));
1213 static inline int online_section(struct mem_section *section)
1215 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1218 static inline int online_section_nr(unsigned long nr)
1220 return online_section(__nr_to_section(nr));
1223 #ifdef CONFIG_MEMORY_HOTPLUG
1224 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1225 #ifdef CONFIG_MEMORY_HOTREMOVE
1226 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1230 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1232 return __nr_to_section(pfn_to_section_nr(pfn));
1235 extern int __highest_present_section_nr;
1237 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1238 static inline int pfn_valid(unsigned long pfn)
1240 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1242 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
1246 static inline int pfn_present(unsigned long pfn)
1248 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1250 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1254 * These are _only_ used during initialisation, therefore they
1255 * can use __initdata ... They could have names to indicate
1259 #define pfn_to_nid(pfn) \
1261 unsigned long __pfn_to_nid_pfn = (pfn); \
1262 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1265 #define pfn_to_nid(pfn) (0)
1268 #define early_pfn_valid(pfn) pfn_valid(pfn)
1269 void sparse_init(void);
1271 #define sparse_init() do {} while (0)
1272 #define sparse_index_init(_sec, _nid) do {} while (0)
1273 #endif /* CONFIG_SPARSEMEM */
1276 * During memory init memblocks map pfns to nids. The search is expensive and
1277 * this caches recent lookups. The implementation of __early_pfn_to_nid
1278 * may treat start/end as pfns or sections.
1280 struct mminit_pfnnid_cache {
1281 unsigned long last_start;
1282 unsigned long last_end;
1286 #ifndef early_pfn_valid
1287 #define early_pfn_valid(pfn) (1)
1290 void memory_present(int nid, unsigned long start, unsigned long end);
1293 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1294 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1295 * pfn_valid_within() should be used in this case; we optimise this away
1296 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1298 #ifdef CONFIG_HOLES_IN_ZONE
1299 #define pfn_valid_within(pfn) pfn_valid(pfn)
1301 #define pfn_valid_within(pfn) (1)
1304 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1306 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1307 * associated with it or not. This means that a struct page exists for this
1308 * pfn. The caller cannot assume the page is fully initialized in general.
1309 * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1310 * will ensure the struct page is fully online and initialized. Special pages
1311 * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1313 * In FLATMEM, it is expected that holes always have valid memmap as long as
1314 * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1315 * that a valid section has a memmap for the entire section.
1317 * However, an ARM, and maybe other embedded architectures in the future
1318 * free memmap backing holes to save memory on the assumption the memmap is
1319 * never used. The page_zone linkages are then broken even though pfn_valid()
1320 * returns true. A walker of the full memmap must then do this additional
1321 * check to ensure the memmap they are looking at is sane by making sure
1322 * the zone and PFN linkages are still valid. This is expensive, but walkers
1323 * of the full memmap are extremely rare.
1325 bool memmap_valid_within(unsigned long pfn,
1326 struct page *page, struct zone *zone);
1328 static inline bool memmap_valid_within(unsigned long pfn,
1329 struct page *page, struct zone *zone)
1333 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1335 #endif /* !__GENERATING_BOUNDS.H */
1336 #endif /* !__ASSEMBLY__ */
1337 #endif /* _LINUX_MMZONE_H */