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_VM_NODE_STAT_ITEMS
187 * We do arithmetic on the LRU lists in various places in the code,
188 * so it is important to keep the active lists LRU_ACTIVE higher in
189 * the array than the corresponding inactive lists, and to keep
190 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
192 * This has to be kept in sync with the statistics in zone_stat_item
193 * above and the descriptions in vmstat_text in mm/vmstat.c
200 LRU_INACTIVE_ANON = LRU_BASE,
201 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
202 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
203 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
208 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
210 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
212 static inline int is_file_lru(enum lru_list lru)
214 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
217 static inline int is_active_lru(enum lru_list lru)
219 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
222 struct zone_reclaim_stat {
224 * The pageout code in vmscan.c keeps track of how many of the
225 * mem/swap backed and file backed pages are referenced.
226 * The higher the rotated/scanned ratio, the more valuable
229 * The anon LRU stats live in [0], file LRU stats in [1]
231 unsigned long recent_rotated[2];
232 unsigned long recent_scanned[2];
236 struct list_head lists[NR_LRU_LISTS];
237 struct zone_reclaim_stat reclaim_stat;
238 /* Evictions & activations on the inactive file list */
239 atomic_long_t inactive_age;
240 /* Refaults at the time of last reclaim cycle */
241 unsigned long refaults;
243 struct pglist_data *pgdat;
247 /* Mask used at gathering information at once (see memcontrol.c) */
248 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
249 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
250 #define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
252 /* Isolate unmapped file */
253 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
254 /* Isolate for asynchronous migration */
255 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
256 /* Isolate unevictable pages */
257 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
259 /* LRU Isolation modes. */
260 typedef unsigned __bitwise isolate_mode_t;
262 enum zone_watermarks {
269 #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
270 #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
271 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
273 struct per_cpu_pages {
274 int count; /* number of pages in the list */
275 int high; /* high watermark, emptying needed */
276 int batch; /* chunk size for buddy add/remove */
278 /* Lists of pages, one per migrate type stored on the pcp-lists */
279 struct list_head lists[MIGRATE_PCPTYPES];
282 struct per_cpu_pageset {
283 struct per_cpu_pages pcp;
286 u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
290 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
294 struct per_cpu_nodestat {
296 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
299 #endif /* !__GENERATING_BOUNDS.H */
302 #ifdef CONFIG_ZONE_DMA
304 * ZONE_DMA is used when there are devices that are not able
305 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
306 * carve out the portion of memory that is needed for these devices.
307 * The range is arch specific.
312 * ---------------------------
313 * parisc, ia64, sparc <4G
316 * alpha Unlimited or 0-16MB.
318 * i386, x86_64 and multiple other arches
323 #ifdef CONFIG_ZONE_DMA32
325 * x86_64 needs two ZONE_DMAs because it supports devices that are
326 * only able to do DMA to the lower 16M but also 32 bit devices that
327 * can only do DMA areas below 4G.
332 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
333 * performed on pages in ZONE_NORMAL if the DMA devices support
334 * transfers to all addressable memory.
337 #ifdef CONFIG_HIGHMEM
339 * A memory area that is only addressable by the kernel through
340 * mapping portions into its own address space. This is for example
341 * used by i386 to allow the kernel to address the memory beyond
342 * 900MB. The kernel will set up special mappings (page
343 * table entries on i386) for each page that the kernel needs to
349 #ifdef CONFIG_ZONE_DEVICE
356 #ifndef __GENERATING_BOUNDS_H
359 /* Read-mostly fields */
361 /* zone watermarks, access with *_wmark_pages(zone) macros */
362 unsigned long watermark[NR_WMARK];
364 unsigned long nr_reserved_highatomic;
367 * We don't know if the memory that we're going to allocate will be
368 * freeable or/and it will be released eventually, so to avoid totally
369 * wasting several GB of ram we must reserve some of the lower zone
370 * memory (otherwise we risk to run OOM on the lower zones despite
371 * there being tons of freeable ram on the higher zones). This array is
372 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
375 long lowmem_reserve[MAX_NR_ZONES];
380 struct pglist_data *zone_pgdat;
381 struct per_cpu_pageset __percpu *pageset;
383 #ifndef CONFIG_SPARSEMEM
385 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
386 * In SPARSEMEM, this map is stored in struct mem_section
388 unsigned long *pageblock_flags;
389 #endif /* CONFIG_SPARSEMEM */
391 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
392 unsigned long zone_start_pfn;
395 * spanned_pages is the total pages spanned by the zone, including
396 * holes, which is calculated as:
397 * spanned_pages = zone_end_pfn - zone_start_pfn;
399 * present_pages is physical pages existing within the zone, which
401 * present_pages = spanned_pages - absent_pages(pages in holes);
403 * managed_pages is present pages managed by the buddy system, which
404 * is calculated as (reserved_pages includes pages allocated by the
405 * bootmem allocator):
406 * managed_pages = present_pages - reserved_pages;
408 * So present_pages may be used by memory hotplug or memory power
409 * management logic to figure out unmanaged pages by checking
410 * (present_pages - managed_pages). And managed_pages should be used
411 * by page allocator and vm scanner to calculate all kinds of watermarks
416 * zone_start_pfn and spanned_pages are protected by span_seqlock.
417 * It is a seqlock because it has to be read outside of zone->lock,
418 * and it is done in the main allocator path. But, it is written
419 * quite infrequently.
421 * The span_seq lock is declared along with zone->lock because it is
422 * frequently read in proximity to zone->lock. It's good to
423 * give them a chance of being in the same cacheline.
425 * Write access to present_pages at runtime should be protected by
426 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
427 * present_pages should get_online_mems() to get a stable value.
429 * Read access to managed_pages should be safe because it's unsigned
430 * long. Write access to zone->managed_pages and totalram_pages are
431 * protected by managed_page_count_lock at runtime. Idealy only
432 * adjust_managed_page_count() should be used instead of directly
433 * touching zone->managed_pages and totalram_pages.
435 unsigned long managed_pages;
436 unsigned long spanned_pages;
437 unsigned long present_pages;
441 #ifdef CONFIG_MEMORY_ISOLATION
443 * Number of isolated pageblock. It is used to solve incorrect
444 * freepage counting problem due to racy retrieving migratetype
445 * of pageblock. Protected by zone->lock.
447 unsigned long nr_isolate_pageblock;
450 #ifdef CONFIG_MEMORY_HOTPLUG
451 /* see spanned/present_pages for more description */
452 seqlock_t span_seqlock;
457 /* Write-intensive fields used from the page allocator */
460 /* free areas of different sizes */
461 struct free_area free_area[MAX_ORDER];
463 /* zone flags, see below */
466 /* Primarily protects free_area */
469 /* Write-intensive fields used by compaction and vmstats. */
473 * When free pages are below this point, additional steps are taken
474 * when reading the number of free pages to avoid per-cpu counter
475 * drift allowing watermarks to be breached
477 unsigned long percpu_drift_mark;
479 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
480 /* pfn where compaction free scanner should start */
481 unsigned long compact_cached_free_pfn;
482 /* pfn where async and sync compaction migration scanner should start */
483 unsigned long compact_cached_migrate_pfn[2];
486 #ifdef CONFIG_COMPACTION
488 * On compaction failure, 1<<compact_defer_shift compactions
489 * are skipped before trying again. The number attempted since
490 * last failure is tracked with compact_considered.
492 unsigned int compact_considered;
493 unsigned int compact_defer_shift;
494 int compact_order_failed;
497 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
498 /* Set to true when the PG_migrate_skip bits should be cleared */
499 bool compact_blockskip_flush;
505 /* Zone statistics */
506 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
507 atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
508 } ____cacheline_internodealigned_in_smp;
511 PGDAT_CONGESTED, /* pgdat has many dirty pages backed by
514 PGDAT_DIRTY, /* reclaim scanning has recently found
515 * many dirty file pages at the tail
518 PGDAT_WRITEBACK, /* reclaim scanning has recently found
519 * many pages under writeback
521 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
524 static inline unsigned long zone_end_pfn(const struct zone *zone)
526 return zone->zone_start_pfn + zone->spanned_pages;
529 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
531 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
534 static inline bool zone_is_initialized(struct zone *zone)
536 return zone->initialized;
539 static inline bool zone_is_empty(struct zone *zone)
541 return zone->spanned_pages == 0;
545 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
546 * intersection with the given zone
548 static inline bool zone_intersects(struct zone *zone,
549 unsigned long start_pfn, unsigned long nr_pages)
551 if (zone_is_empty(zone))
553 if (start_pfn >= zone_end_pfn(zone) ||
554 start_pfn + nr_pages <= zone->zone_start_pfn)
561 * The "priority" of VM scanning is how much of the queues we will scan in one
562 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
563 * queues ("queue_length >> 12") during an aging round.
565 #define DEF_PRIORITY 12
567 /* Maximum number of zones on a zonelist */
568 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
571 ZONELIST_FALLBACK, /* zonelist with fallback */
574 * The NUMA zonelists are doubled because we need zonelists that
575 * restrict the allocations to a single node for __GFP_THISNODE.
577 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
583 * This struct contains information about a zone in a zonelist. It is stored
584 * here to avoid dereferences into large structures and lookups of tables
587 struct zone *zone; /* Pointer to actual zone */
588 int zone_idx; /* zone_idx(zoneref->zone) */
592 * One allocation request operates on a zonelist. A zonelist
593 * is a list of zones, the first one is the 'goal' of the
594 * allocation, the other zones are fallback zones, in decreasing
597 * To speed the reading of the zonelist, the zonerefs contain the zone index
598 * of the entry being read. Helper functions to access information given
599 * a struct zoneref are
601 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
602 * zonelist_zone_idx() - Return the index of the zone for an entry
603 * zonelist_node_idx() - Return the index of the node for an entry
606 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
609 #ifndef CONFIG_DISCONTIGMEM
610 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
611 extern struct page *mem_map;
615 * On NUMA machines, each NUMA node would have a pg_data_t to describe
616 * it's memory layout. On UMA machines there is a single pglist_data which
617 * describes the whole memory.
619 * Memory statistics and page replacement data structures are maintained on a
623 typedef struct pglist_data {
624 struct zone node_zones[MAX_NR_ZONES];
625 struct zonelist node_zonelists[MAX_ZONELISTS];
627 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
628 struct page *node_mem_map;
629 #ifdef CONFIG_PAGE_EXTENSION
630 struct page_ext *node_page_ext;
633 #ifndef CONFIG_NO_BOOTMEM
634 struct bootmem_data *bdata;
636 #ifdef CONFIG_MEMORY_HOTPLUG
638 * Must be held any time you expect node_start_pfn, node_present_pages
639 * or node_spanned_pages stay constant. Holding this will also
640 * guarantee that any pfn_valid() stays that way.
642 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
643 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG.
645 * Nests above zone->lock and zone->span_seqlock
647 spinlock_t node_size_lock;
649 unsigned long node_start_pfn;
650 unsigned long node_present_pages; /* total number of physical pages */
651 unsigned long node_spanned_pages; /* total size of physical page
652 range, including holes */
654 wait_queue_head_t kswapd_wait;
655 wait_queue_head_t pfmemalloc_wait;
656 struct task_struct *kswapd; /* Protected by
657 mem_hotplug_begin/end() */
659 enum zone_type kswapd_classzone_idx;
661 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
663 #ifdef CONFIG_COMPACTION
664 int kcompactd_max_order;
665 enum zone_type kcompactd_classzone_idx;
666 wait_queue_head_t kcompactd_wait;
667 struct task_struct *kcompactd;
669 #ifdef CONFIG_NUMA_BALANCING
670 /* Lock serializing the migrate rate limiting window */
671 spinlock_t numabalancing_migrate_lock;
673 /* Rate limiting time interval */
674 unsigned long numabalancing_migrate_next_window;
676 /* Number of pages migrated during the rate limiting time interval */
677 unsigned long numabalancing_migrate_nr_pages;
680 * This is a per-node reserve of pages that are not available
681 * to userspace allocations.
683 unsigned long totalreserve_pages;
687 * zone reclaim becomes active if more unmapped pages exist.
689 unsigned long min_unmapped_pages;
690 unsigned long min_slab_pages;
691 #endif /* CONFIG_NUMA */
693 /* Write-intensive fields used by page reclaim */
697 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
699 * If memory initialisation on large machines is deferred then this
700 * is the first PFN that needs to be initialised.
702 unsigned long first_deferred_pfn;
703 /* Number of non-deferred pages */
704 unsigned long static_init_pgcnt;
705 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
707 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
708 spinlock_t split_queue_lock;
709 struct list_head split_queue;
710 unsigned long split_queue_len;
713 /* Fields commonly accessed by the page reclaim scanner */
714 struct lruvec lruvec;
720 /* Per-node vmstats */
721 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
722 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
725 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
726 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
727 #ifdef CONFIG_FLAT_NODE_MEM_MAP
728 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
730 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
732 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
734 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
735 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
736 static inline spinlock_t *zone_lru_lock(struct zone *zone)
738 return &zone->zone_pgdat->lru_lock;
741 static inline struct lruvec *node_lruvec(struct pglist_data *pgdat)
743 return &pgdat->lruvec;
746 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
748 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
751 static inline bool pgdat_is_empty(pg_data_t *pgdat)
753 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
756 static inline int zone_id(const struct zone *zone)
758 struct pglist_data *pgdat = zone->zone_pgdat;
760 return zone - pgdat->node_zones;
763 #ifdef CONFIG_ZONE_DEVICE
764 static inline bool is_dev_zone(const struct zone *zone)
766 return zone_id(zone) == ZONE_DEVICE;
769 static inline bool is_dev_zone(const struct zone *zone)
775 #include <linux/memory_hotplug.h>
777 void build_all_zonelists(pg_data_t *pgdat);
778 void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx);
779 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
780 int classzone_idx, unsigned int alloc_flags,
782 bool zone_watermark_ok(struct zone *z, unsigned int order,
783 unsigned long mark, int classzone_idx,
784 unsigned int alloc_flags);
785 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
786 unsigned long mark, int classzone_idx);
787 enum memmap_context {
791 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
794 extern void lruvec_init(struct lruvec *lruvec);
796 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
799 return lruvec->pgdat;
801 return container_of(lruvec, struct pglist_data, lruvec);
805 extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
807 #ifdef CONFIG_HAVE_MEMORY_PRESENT
808 void memory_present(int nid, unsigned long start, unsigned long end);
810 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
813 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
814 int local_memory_node(int node_id);
816 static inline int local_memory_node(int node_id) { return node_id; };
819 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
820 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
824 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
826 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
829 * Returns true if a zone has pages managed by the buddy allocator.
830 * All the reclaim decisions have to use this function rather than
831 * populated_zone(). If the whole zone is reserved then we can easily
832 * end up with populated_zone() && !managed_zone().
834 static inline bool managed_zone(struct zone *zone)
836 return zone->managed_pages;
839 /* Returns true if a zone has memory */
840 static inline bool populated_zone(struct zone *zone)
842 return zone->present_pages;
845 extern int movable_zone;
847 #ifdef CONFIG_HIGHMEM
848 static inline int zone_movable_is_highmem(void)
850 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
851 return movable_zone == ZONE_HIGHMEM;
853 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
858 static inline int is_highmem_idx(enum zone_type idx)
860 #ifdef CONFIG_HIGHMEM
861 return (idx == ZONE_HIGHMEM ||
862 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
869 * is_highmem - helper function to quickly check if a struct zone is a
870 * highmem zone or not. This is an attempt to keep references
871 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
872 * @zone - pointer to struct zone variable
874 static inline int is_highmem(struct zone *zone)
876 #ifdef CONFIG_HIGHMEM
877 return is_highmem_idx(zone_idx(zone));
883 /* These two functions are used to setup the per zone pages min values */
885 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
886 void __user *, size_t *, loff_t *);
887 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int,
888 void __user *, size_t *, loff_t *);
889 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
890 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
891 void __user *, size_t *, loff_t *);
892 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
893 void __user *, size_t *, loff_t *);
894 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
895 void __user *, size_t *, loff_t *);
896 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
897 void __user *, size_t *, loff_t *);
899 extern int numa_zonelist_order_handler(struct ctl_table *, int,
900 void __user *, size_t *, loff_t *);
901 extern char numa_zonelist_order[];
902 #define NUMA_ZONELIST_ORDER_LEN 16
904 #ifndef CONFIG_NEED_MULTIPLE_NODES
906 extern struct pglist_data contig_page_data;
907 #define NODE_DATA(nid) (&contig_page_data)
908 #define NODE_MEM_MAP(nid) mem_map
910 #else /* CONFIG_NEED_MULTIPLE_NODES */
912 #include <asm/mmzone.h>
914 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
916 extern struct pglist_data *first_online_pgdat(void);
917 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
918 extern struct zone *next_zone(struct zone *zone);
921 * for_each_online_pgdat - helper macro to iterate over all online nodes
922 * @pgdat - pointer to a pg_data_t variable
924 #define for_each_online_pgdat(pgdat) \
925 for (pgdat = first_online_pgdat(); \
927 pgdat = next_online_pgdat(pgdat))
929 * for_each_zone - helper macro to iterate over all memory zones
930 * @zone - pointer to struct zone variable
932 * The user only needs to declare the zone variable, for_each_zone
935 #define for_each_zone(zone) \
936 for (zone = (first_online_pgdat())->node_zones; \
938 zone = next_zone(zone))
940 #define for_each_populated_zone(zone) \
941 for (zone = (first_online_pgdat())->node_zones; \
943 zone = next_zone(zone)) \
944 if (!populated_zone(zone)) \
948 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
950 return zoneref->zone;
953 static inline int zonelist_zone_idx(struct zoneref *zoneref)
955 return zoneref->zone_idx;
958 static inline int zonelist_node_idx(struct zoneref *zoneref)
961 /* zone_to_nid not available in this context */
962 return zoneref->zone->node;
965 #endif /* CONFIG_NUMA */
968 struct zoneref *__next_zones_zonelist(struct zoneref *z,
969 enum zone_type highest_zoneidx,
973 * 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
974 * @z - The cursor used as a starting point for the search
975 * @highest_zoneidx - The zone index of the highest zone to return
976 * @nodes - An optional nodemask to filter the zonelist with
978 * This function returns the next zone at or below a given zone index that is
979 * within the allowed nodemask using a cursor as the starting point for the
980 * search. The zoneref returned is a cursor that represents the current zone
981 * being examined. It should be advanced by one before calling
982 * next_zones_zonelist again.
984 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
985 enum zone_type highest_zoneidx,
988 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
990 return __next_zones_zonelist(z, highest_zoneidx, nodes);
994 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
995 * @zonelist - The zonelist to search for a suitable zone
996 * @highest_zoneidx - The zone index of the highest zone to return
997 * @nodes - An optional nodemask to filter the zonelist with
998 * @return - Zoneref pointer for the first suitable zone found (see below)
1000 * This function returns the first zone at or below a given zone index that is
1001 * within the allowed nodemask. The zoneref returned is a cursor that can be
1002 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1003 * one before calling.
1005 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1006 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1007 * update due to cpuset modification.
1009 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1010 enum zone_type highest_zoneidx,
1013 return next_zones_zonelist(zonelist->_zonerefs,
1014 highest_zoneidx, nodes);
1018 * 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
1019 * @zone - The current zone in the iterator
1020 * @z - The current pointer within zonelist->zones being iterated
1021 * @zlist - The zonelist being iterated
1022 * @highidx - The zone index of the highest zone to return
1023 * @nodemask - Nodemask allowed by the allocator
1025 * This iterator iterates though all zones at or below a given zone index and
1026 * within a given nodemask
1028 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1029 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1031 z = next_zones_zonelist(++z, highidx, nodemask), \
1032 zone = zonelist_zone(z))
1034 #define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1035 for (zone = z->zone; \
1037 z = next_zones_zonelist(++z, highidx, nodemask), \
1038 zone = zonelist_zone(z))
1042 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1043 * @zone - The current zone in the iterator
1044 * @z - The current pointer within zonelist->zones being iterated
1045 * @zlist - The zonelist being iterated
1046 * @highidx - The zone index of the highest zone to return
1048 * This iterator iterates though all zones at or below a given zone index.
1050 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1051 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1053 #ifdef CONFIG_SPARSEMEM
1054 #include <asm/sparsemem.h>
1057 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1058 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1059 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1061 BUILD_BUG_ON(IS_ENABLED(CONFIG_NUMA));
1066 #ifdef CONFIG_FLATMEM
1067 #define pfn_to_nid(pfn) (0)
1070 #ifdef CONFIG_SPARSEMEM
1073 * SECTION_SHIFT #bits space required to store a section #
1075 * PA_SECTION_SHIFT physical address to/from section number
1076 * PFN_SECTION_SHIFT pfn to/from section number
1078 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1079 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1081 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1083 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1084 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1086 #define SECTION_BLOCKFLAGS_BITS \
1087 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1089 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1090 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1093 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1095 return pfn >> PFN_SECTION_SHIFT;
1097 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1099 return sec << PFN_SECTION_SHIFT;
1102 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1103 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1107 struct mem_section {
1109 * This is, logically, a pointer to an array of struct
1110 * pages. However, it is stored with some other magic.
1111 * (see sparse.c::sparse_init_one_section())
1113 * Additionally during early boot we encode node id of
1114 * the location of the section here to guide allocation.
1115 * (see sparse.c::memory_present())
1117 * Making it a UL at least makes someone do a cast
1118 * before using it wrong.
1120 unsigned long section_mem_map;
1122 /* See declaration of similar field in struct zone */
1123 unsigned long *pageblock_flags;
1124 #ifdef CONFIG_PAGE_EXTENSION
1126 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1127 * section. (see page_ext.h about this.)
1129 struct page_ext *page_ext;
1133 * WARNING: mem_section must be a power-of-2 in size for the
1134 * calculation and use of SECTION_ROOT_MASK to make sense.
1138 #ifdef CONFIG_SPARSEMEM_EXTREME
1139 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1141 #define SECTIONS_PER_ROOT 1
1144 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1145 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1146 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1148 #ifdef CONFIG_SPARSEMEM_EXTREME
1149 extern struct mem_section **mem_section;
1151 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1154 static inline struct mem_section *__nr_to_section(unsigned long nr)
1156 #ifdef CONFIG_SPARSEMEM_EXTREME
1160 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1162 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1164 extern int __section_nr(struct mem_section* ms);
1165 extern unsigned long usemap_size(void);
1168 * We use the lower bits of the mem_map pointer to store
1169 * a little bit of information. There should be at least
1170 * 3 bits here due to 32-bit alignment.
1172 #define SECTION_MARKED_PRESENT (1UL<<0)
1173 #define SECTION_HAS_MEM_MAP (1UL<<1)
1174 #define SECTION_IS_ONLINE (1UL<<2)
1175 #define SECTION_MAP_LAST_BIT (1UL<<3)
1176 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1177 #define SECTION_NID_SHIFT 3
1179 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1181 unsigned long map = section->section_mem_map;
1182 map &= SECTION_MAP_MASK;
1183 return (struct page *)map;
1186 static inline int present_section(struct mem_section *section)
1188 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1191 static inline int present_section_nr(unsigned long nr)
1193 return present_section(__nr_to_section(nr));
1196 static inline int valid_section(struct mem_section *section)
1198 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1201 static inline int valid_section_nr(unsigned long nr)
1203 return valid_section(__nr_to_section(nr));
1206 static inline int online_section(struct mem_section *section)
1208 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1211 static inline int online_section_nr(unsigned long nr)
1213 return online_section(__nr_to_section(nr));
1216 #ifdef CONFIG_MEMORY_HOTPLUG
1217 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1218 #ifdef CONFIG_MEMORY_HOTREMOVE
1219 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1223 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1225 return __nr_to_section(pfn_to_section_nr(pfn));
1228 extern int __highest_present_section_nr;
1230 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1231 static inline int pfn_valid(unsigned long pfn)
1233 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1235 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
1239 static inline int pfn_present(unsigned long pfn)
1241 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1243 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1247 * These are _only_ used during initialisation, therefore they
1248 * can use __initdata ... They could have names to indicate
1252 #define pfn_to_nid(pfn) \
1254 unsigned long __pfn_to_nid_pfn = (pfn); \
1255 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1258 #define pfn_to_nid(pfn) (0)
1261 #define early_pfn_valid(pfn) pfn_valid(pfn)
1262 void sparse_init(void);
1264 #define sparse_init() do {} while (0)
1265 #define sparse_index_init(_sec, _nid) do {} while (0)
1266 #endif /* CONFIG_SPARSEMEM */
1269 * During memory init memblocks map pfns to nids. The search is expensive and
1270 * this caches recent lookups. The implementation of __early_pfn_to_nid
1271 * may treat start/end as pfns or sections.
1273 struct mminit_pfnnid_cache {
1274 unsigned long last_start;
1275 unsigned long last_end;
1279 #ifndef early_pfn_valid
1280 #define early_pfn_valid(pfn) (1)
1283 void memory_present(int nid, unsigned long start, unsigned long end);
1284 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
1287 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1288 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1289 * pfn_valid_within() should be used in this case; we optimise this away
1290 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1292 #ifdef CONFIG_HOLES_IN_ZONE
1293 #define pfn_valid_within(pfn) pfn_valid(pfn)
1295 #define pfn_valid_within(pfn) (1)
1298 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1300 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1301 * associated with it or not. This means that a struct page exists for this
1302 * pfn. The caller cannot assume the page is fully initialized in general.
1303 * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1304 * will ensure the struct page is fully online and initialized. Special pages
1305 * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1307 * In FLATMEM, it is expected that holes always have valid memmap as long as
1308 * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1309 * that a valid section has a memmap for the entire section.
1311 * However, an ARM, and maybe other embedded architectures in the future
1312 * free memmap backing holes to save memory on the assumption the memmap is
1313 * never used. The page_zone linkages are then broken even though pfn_valid()
1314 * returns true. A walker of the full memmap must then do this additional
1315 * check to ensure the memmap they are looking at is sane by making sure
1316 * the zone and PFN linkages are still valid. This is expensive, but walkers
1317 * of the full memmap are extremely rare.
1319 bool memmap_valid_within(unsigned long pfn,
1320 struct page *page, struct zone *zone);
1322 static inline bool memmap_valid_within(unsigned long pfn,
1323 struct page *page, struct zone *zone)
1327 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1329 #endif /* !__GENERATING_BOUNDS.H */
1330 #endif /* !__ASSEMBLY__ */
1331 #endif /* _LINUX_MMZONE_H */