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 MIGRATETYPE_MASK ((1UL << PB_migratetype_bits) - 1)
93 #define get_pageblock_migratetype(page) \
94 get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK)
97 struct list_head free_list[MIGRATE_TYPES];
98 unsigned long nr_free;
101 static inline struct page *get_page_from_free_area(struct free_area *area,
104 return list_first_entry_or_null(&area->free_list[migratetype],
108 static inline bool free_area_empty(struct free_area *area, int migratetype)
110 return list_empty(&area->free_list[migratetype]);
116 * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
117 * So add a wild amount of padding here to ensure that they fall into separate
118 * cachelines. There are very few zone structures in the machine, so space
119 * consumption is not a concern here.
121 #if defined(CONFIG_SMP)
122 struct zone_padding {
124 } ____cacheline_internodealigned_in_smp;
125 #define ZONE_PADDING(name) struct zone_padding name;
127 #define ZONE_PADDING(name)
131 enum numa_stat_item {
132 NUMA_HIT, /* allocated in intended node */
133 NUMA_MISS, /* allocated in non intended node */
134 NUMA_FOREIGN, /* was intended here, hit elsewhere */
135 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
136 NUMA_LOCAL, /* allocation from local node */
137 NUMA_OTHER, /* allocation from other node */
138 NR_VM_NUMA_STAT_ITEMS
141 #define NR_VM_NUMA_STAT_ITEMS 0
144 enum zone_stat_item {
145 /* First 128 byte cacheline (assuming 64 bit words) */
147 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
148 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
150 NR_ZONE_INACTIVE_FILE,
153 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
154 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
155 NR_PAGETABLE, /* used for pagetables */
156 /* Second 128 byte cacheline */
158 #if IS_ENABLED(CONFIG_ZSMALLOC)
159 NR_ZSPAGES, /* allocated in zsmalloc */
162 NR_VM_ZONE_STAT_ITEMS };
164 enum node_stat_item {
166 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
167 NR_ACTIVE_ANON, /* " " " " " */
168 NR_INACTIVE_FILE, /* " " " " " */
169 NR_ACTIVE_FILE, /* " " " " " */
170 NR_UNEVICTABLE, /* " " " " " */
171 NR_SLAB_RECLAIMABLE_B,
172 NR_SLAB_UNRECLAIMABLE_B,
173 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
174 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
179 WORKINGSET_NODERECLAIM,
180 NR_ANON_MAPPED, /* Mapped anonymous pages */
181 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
182 only modified from process context */
186 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
187 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
194 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
195 NR_DIRTIED, /* page dirtyings since bootup */
196 NR_WRITTEN, /* page writings since bootup */
197 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
198 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
199 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
200 NR_KERNEL_STACK_KB, /* measured in KiB */
201 #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
202 NR_KERNEL_SCS_KB, /* measured in KiB */
204 NR_VM_NODE_STAT_ITEMS
208 * Returns true if the value is measured in bytes (most vmstat values are
209 * measured in pages). This defines the API part, the internal representation
210 * might be different.
212 static __always_inline bool vmstat_item_in_bytes(int idx)
215 * Global and per-node slab counters track slab pages.
216 * It's expected that changes are multiples of PAGE_SIZE.
217 * Internally values are stored in pages.
219 * Per-memcg and per-lruvec counters track memory, consumed
220 * by individual slab objects. These counters are actually
223 return (idx == NR_SLAB_RECLAIMABLE_B ||
224 idx == NR_SLAB_UNRECLAIMABLE_B);
228 * We do arithmetic on the LRU lists in various places in the code,
229 * so it is important to keep the active lists LRU_ACTIVE higher in
230 * the array than the corresponding inactive lists, and to keep
231 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
233 * This has to be kept in sync with the statistics in zone_stat_item
234 * above and the descriptions in vmstat_text in mm/vmstat.c
241 LRU_INACTIVE_ANON = LRU_BASE,
242 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
243 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
244 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
249 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
251 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
253 static inline bool is_file_lru(enum lru_list lru)
255 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
258 static inline bool is_active_lru(enum lru_list lru)
260 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
264 LRUVEC_CONGESTED, /* lruvec has many dirty pages
265 * backed by a congested BDI
270 struct list_head lists[NR_LRU_LISTS];
272 * These track the cost of reclaiming one LRU - file or anon -
273 * over the other. As the observed cost of reclaiming one LRU
274 * increases, the reclaim scan balance tips toward the other.
276 unsigned long anon_cost;
277 unsigned long file_cost;
278 /* Non-resident age, driven by LRU movement */
279 atomic_long_t nonresident_age;
280 /* Refaults at the time of last reclaim cycle */
281 unsigned long refaults;
282 /* Various lruvec state flags (enum lruvec_flags) */
285 struct pglist_data *pgdat;
289 /* Isolate unmapped pages */
290 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
291 /* Isolate for asynchronous migration */
292 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
293 /* Isolate unevictable pages */
294 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
296 /* LRU Isolation modes. */
297 typedef unsigned __bitwise isolate_mode_t;
299 enum zone_watermarks {
306 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
307 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
308 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
309 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
311 struct per_cpu_pages {
312 int count; /* number of pages in the list */
313 int high; /* high watermark, emptying needed */
314 int batch; /* chunk size for buddy add/remove */
316 /* Lists of pages, one per migrate type stored on the pcp-lists */
317 struct list_head lists[MIGRATE_PCPTYPES];
320 struct per_cpu_pageset {
321 struct per_cpu_pages pcp;
324 u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
328 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
332 struct per_cpu_nodestat {
334 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
337 #endif /* !__GENERATING_BOUNDS.H */
341 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
342 * to DMA to all of the addressable memory (ZONE_NORMAL).
343 * On architectures where this area covers the whole 32 bit address
344 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
345 * DMA addressing constraints. This distinction is important as a 32bit
346 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
347 * platforms may need both zones as they support peripherals with
348 * different DMA addressing limitations.
352 * - i386 and x86_64 have a fixed 16M ZONE_DMA and ZONE_DMA32 for the
353 * rest of the lower 4G.
355 * - arm only uses ZONE_DMA, the size, up to 4G, may vary depending on
356 * the specific device.
358 * - arm64 has a fixed 1G ZONE_DMA and ZONE_DMA32 for the rest of the
361 * - powerpc only uses ZONE_DMA, the size, up to 2G, may vary
362 * depending on the specific device.
364 * - s390 uses ZONE_DMA fixed to the lower 2G.
366 * - ia64 and riscv only use ZONE_DMA32.
368 * - parisc uses neither.
370 #ifdef CONFIG_ZONE_DMA
373 #ifdef CONFIG_ZONE_DMA32
377 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
378 * performed on pages in ZONE_NORMAL if the DMA devices support
379 * transfers to all addressable memory.
382 #ifdef CONFIG_HIGHMEM
384 * A memory area that is only addressable by the kernel through
385 * mapping portions into its own address space. This is for example
386 * used by i386 to allow the kernel to address the memory beyond
387 * 900MB. The kernel will set up special mappings (page
388 * table entries on i386) for each page that the kernel needs to
394 #ifdef CONFIG_ZONE_DEVICE
401 #ifndef __GENERATING_BOUNDS_H
404 /* Read-mostly fields */
406 /* zone watermarks, access with *_wmark_pages(zone) macros */
407 unsigned long _watermark[NR_WMARK];
408 unsigned long watermark_boost;
410 unsigned long nr_reserved_highatomic;
413 * We don't know if the memory that we're going to allocate will be
414 * freeable or/and it will be released eventually, so to avoid totally
415 * wasting several GB of ram we must reserve some of the lower zone
416 * memory (otherwise we risk to run OOM on the lower zones despite
417 * there being tons of freeable ram on the higher zones). This array is
418 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
421 long lowmem_reserve[MAX_NR_ZONES];
426 struct pglist_data *zone_pgdat;
427 struct per_cpu_pageset __percpu *pageset;
429 #ifndef CONFIG_SPARSEMEM
431 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
432 * In SPARSEMEM, this map is stored in struct mem_section
434 unsigned long *pageblock_flags;
435 #endif /* CONFIG_SPARSEMEM */
437 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
438 unsigned long zone_start_pfn;
441 * spanned_pages is the total pages spanned by the zone, including
442 * holes, which is calculated as:
443 * spanned_pages = zone_end_pfn - zone_start_pfn;
445 * present_pages is physical pages existing within the zone, which
447 * present_pages = spanned_pages - absent_pages(pages in holes);
449 * managed_pages is present pages managed by the buddy system, which
450 * is calculated as (reserved_pages includes pages allocated by the
451 * bootmem allocator):
452 * managed_pages = present_pages - reserved_pages;
454 * So present_pages may be used by memory hotplug or memory power
455 * management logic to figure out unmanaged pages by checking
456 * (present_pages - managed_pages). And managed_pages should be used
457 * by page allocator and vm scanner to calculate all kinds of watermarks
462 * zone_start_pfn and spanned_pages are protected by span_seqlock.
463 * It is a seqlock because it has to be read outside of zone->lock,
464 * and it is done in the main allocator path. But, it is written
465 * quite infrequently.
467 * The span_seq lock is declared along with zone->lock because it is
468 * frequently read in proximity to zone->lock. It's good to
469 * give them a chance of being in the same cacheline.
471 * Write access to present_pages at runtime should be protected by
472 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
473 * present_pages should get_online_mems() to get a stable value.
475 atomic_long_t managed_pages;
476 unsigned long spanned_pages;
477 unsigned long present_pages;
481 #ifdef CONFIG_MEMORY_ISOLATION
483 * Number of isolated pageblock. It is used to solve incorrect
484 * freepage counting problem due to racy retrieving migratetype
485 * of pageblock. Protected by zone->lock.
487 unsigned long nr_isolate_pageblock;
490 #ifdef CONFIG_MEMORY_HOTPLUG
491 /* see spanned/present_pages for more description */
492 seqlock_t span_seqlock;
497 /* Write-intensive fields used from the page allocator */
500 /* free areas of different sizes */
501 struct free_area free_area[MAX_ORDER];
503 /* zone flags, see below */
506 /* Primarily protects free_area */
509 /* Write-intensive fields used by compaction and vmstats. */
513 * When free pages are below this point, additional steps are taken
514 * when reading the number of free pages to avoid per-cpu counter
515 * drift allowing watermarks to be breached
517 unsigned long percpu_drift_mark;
519 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
520 /* pfn where compaction free scanner should start */
521 unsigned long compact_cached_free_pfn;
522 /* pfn where async and sync compaction migration scanner should start */
523 unsigned long compact_cached_migrate_pfn[2];
524 unsigned long compact_init_migrate_pfn;
525 unsigned long compact_init_free_pfn;
528 #ifdef CONFIG_COMPACTION
530 * On compaction failure, 1<<compact_defer_shift compactions
531 * are skipped before trying again. The number attempted since
532 * last failure is tracked with compact_considered.
534 unsigned int compact_considered;
535 unsigned int compact_defer_shift;
536 int compact_order_failed;
539 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
540 /* Set to true when the PG_migrate_skip bits should be cleared */
541 bool compact_blockskip_flush;
547 /* Zone statistics */
548 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
549 atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
550 } ____cacheline_internodealigned_in_smp;
553 PGDAT_DIRTY, /* reclaim scanning has recently found
554 * many dirty file pages at the tail
557 PGDAT_WRITEBACK, /* reclaim scanning has recently found
558 * many pages under writeback
560 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
564 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
565 * Cleared when kswapd is woken.
569 static inline unsigned long zone_managed_pages(struct zone *zone)
571 return (unsigned long)atomic_long_read(&zone->managed_pages);
574 static inline unsigned long zone_end_pfn(const struct zone *zone)
576 return zone->zone_start_pfn + zone->spanned_pages;
579 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
581 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
584 static inline bool zone_is_initialized(struct zone *zone)
586 return zone->initialized;
589 static inline bool zone_is_empty(struct zone *zone)
591 return zone->spanned_pages == 0;
595 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
596 * intersection with the given zone
598 static inline bool zone_intersects(struct zone *zone,
599 unsigned long start_pfn, unsigned long nr_pages)
601 if (zone_is_empty(zone))
603 if (start_pfn >= zone_end_pfn(zone) ||
604 start_pfn + nr_pages <= zone->zone_start_pfn)
611 * The "priority" of VM scanning is how much of the queues we will scan in one
612 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
613 * queues ("queue_length >> 12") during an aging round.
615 #define DEF_PRIORITY 12
617 /* Maximum number of zones on a zonelist */
618 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
621 ZONELIST_FALLBACK, /* zonelist with fallback */
624 * The NUMA zonelists are doubled because we need zonelists that
625 * restrict the allocations to a single node for __GFP_THISNODE.
627 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
633 * This struct contains information about a zone in a zonelist. It is stored
634 * here to avoid dereferences into large structures and lookups of tables
637 struct zone *zone; /* Pointer to actual zone */
638 int zone_idx; /* zone_idx(zoneref->zone) */
642 * One allocation request operates on a zonelist. A zonelist
643 * is a list of zones, the first one is the 'goal' of the
644 * allocation, the other zones are fallback zones, in decreasing
647 * To speed the reading of the zonelist, the zonerefs contain the zone index
648 * of the entry being read. Helper functions to access information given
649 * a struct zoneref are
651 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
652 * zonelist_zone_idx() - Return the index of the zone for an entry
653 * zonelist_node_idx() - Return the index of the node for an entry
656 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
659 #ifndef CONFIG_DISCONTIGMEM
660 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
661 extern struct page *mem_map;
664 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
665 struct deferred_split {
666 spinlock_t split_queue_lock;
667 struct list_head split_queue;
668 unsigned long split_queue_len;
673 * On NUMA machines, each NUMA node would have a pg_data_t to describe
674 * it's memory layout. On UMA machines there is a single pglist_data which
675 * describes the whole memory.
677 * Memory statistics and page replacement data structures are maintained on a
680 typedef struct pglist_data {
682 * node_zones contains just the zones for THIS node. Not all of the
683 * zones may be populated, but it is the full list. It is referenced by
684 * this node's node_zonelists as well as other node's node_zonelists.
686 struct zone node_zones[MAX_NR_ZONES];
689 * node_zonelists contains references to all zones in all nodes.
690 * Generally the first zones will be references to this node's
693 struct zonelist node_zonelists[MAX_ZONELISTS];
695 int nr_zones; /* number of populated zones in this node */
696 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
697 struct page *node_mem_map;
698 #ifdef CONFIG_PAGE_EXTENSION
699 struct page_ext *node_page_ext;
702 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
704 * Must be held any time you expect node_start_pfn,
705 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
706 * Also synchronizes pgdat->first_deferred_pfn during deferred page
709 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
710 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
711 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
713 * Nests above zone->lock and zone->span_seqlock
715 spinlock_t node_size_lock;
717 unsigned long node_start_pfn;
718 unsigned long node_present_pages; /* total number of physical pages */
719 unsigned long node_spanned_pages; /* total size of physical page
720 range, including holes */
722 wait_queue_head_t kswapd_wait;
723 wait_queue_head_t pfmemalloc_wait;
724 struct task_struct *kswapd; /* Protected by
725 mem_hotplug_begin/end() */
727 enum zone_type kswapd_highest_zoneidx;
729 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
731 #ifdef CONFIG_COMPACTION
732 int kcompactd_max_order;
733 enum zone_type kcompactd_highest_zoneidx;
734 wait_queue_head_t kcompactd_wait;
735 struct task_struct *kcompactd;
738 * This is a per-node reserve of pages that are not available
739 * to userspace allocations.
741 unsigned long totalreserve_pages;
745 * node reclaim becomes active if more unmapped pages exist.
747 unsigned long min_unmapped_pages;
748 unsigned long min_slab_pages;
749 #endif /* CONFIG_NUMA */
751 /* Write-intensive fields used by page reclaim */
755 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
757 * If memory initialisation on large machines is deferred then this
758 * is the first PFN that needs to be initialised.
760 unsigned long first_deferred_pfn;
761 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
763 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
764 struct deferred_split deferred_split_queue;
767 /* Fields commonly accessed by the page reclaim scanner */
770 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
772 * Use mem_cgroup_lruvec() to look up lruvecs.
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 unsigned long pgdat_end_pfn(pg_data_t *pgdat)
799 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
802 static inline bool pgdat_is_empty(pg_data_t *pgdat)
804 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
807 #include <linux/memory_hotplug.h>
809 void build_all_zonelists(pg_data_t *pgdat);
810 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
811 enum zone_type highest_zoneidx);
812 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
813 int highest_zoneidx, unsigned int alloc_flags,
815 bool zone_watermark_ok(struct zone *z, unsigned int order,
816 unsigned long mark, int highest_zoneidx,
817 unsigned int alloc_flags);
818 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
819 unsigned long mark, int highest_zoneidx);
820 enum memmap_context {
824 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
827 extern void lruvec_init(struct lruvec *lruvec);
829 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
832 return lruvec->pgdat;
834 return container_of(lruvec, struct pglist_data, __lruvec);
838 extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
840 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
841 int local_memory_node(int node_id);
843 static inline int local_memory_node(int node_id) { return node_id; };
847 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
849 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
852 * Returns true if a zone has pages managed by the buddy allocator.
853 * All the reclaim decisions have to use this function rather than
854 * populated_zone(). If the whole zone is reserved then we can easily
855 * end up with populated_zone() && !managed_zone().
857 static inline bool managed_zone(struct zone *zone)
859 return zone_managed_pages(zone);
862 /* Returns true if a zone has memory */
863 static inline bool populated_zone(struct zone *zone)
865 return zone->present_pages;
869 static inline int zone_to_nid(struct zone *zone)
874 static inline void zone_set_nid(struct zone *zone, int nid)
879 static inline int zone_to_nid(struct zone *zone)
884 static inline void zone_set_nid(struct zone *zone, int nid) {}
887 extern int movable_zone;
889 #ifdef CONFIG_HIGHMEM
890 static inline int zone_movable_is_highmem(void)
892 #ifdef CONFIG_NEED_MULTIPLE_NODES
893 return movable_zone == ZONE_HIGHMEM;
895 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
900 static inline int is_highmem_idx(enum zone_type idx)
902 #ifdef CONFIG_HIGHMEM
903 return (idx == ZONE_HIGHMEM ||
904 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
911 * is_highmem - helper function to quickly check if a struct zone is a
912 * highmem zone or not. This is an attempt to keep references
913 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
914 * @zone - pointer to struct zone variable
916 static inline int is_highmem(struct zone *zone)
918 #ifdef CONFIG_HIGHMEM
919 return is_highmem_idx(zone_idx(zone));
925 /* These two functions are used to setup the per zone pages min values */
928 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, void *, size_t *,
930 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, void *,
932 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
933 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, void *,
935 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
936 void *, size_t *, loff_t *);
937 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
938 void *, size_t *, loff_t *);
939 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
940 void *, size_t *, loff_t *);
941 int numa_zonelist_order_handler(struct ctl_table *, int,
942 void *, size_t *, loff_t *);
943 extern int percpu_pagelist_fraction;
944 extern char numa_zonelist_order[];
945 #define NUMA_ZONELIST_ORDER_LEN 16
947 #ifndef CONFIG_NEED_MULTIPLE_NODES
949 extern struct pglist_data contig_page_data;
950 #define NODE_DATA(nid) (&contig_page_data)
951 #define NODE_MEM_MAP(nid) mem_map
953 #else /* CONFIG_NEED_MULTIPLE_NODES */
955 #include <asm/mmzone.h>
957 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
959 extern struct pglist_data *first_online_pgdat(void);
960 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
961 extern struct zone *next_zone(struct zone *zone);
964 * for_each_online_pgdat - helper macro to iterate over all online nodes
965 * @pgdat - pointer to a pg_data_t variable
967 #define for_each_online_pgdat(pgdat) \
968 for (pgdat = first_online_pgdat(); \
970 pgdat = next_online_pgdat(pgdat))
972 * for_each_zone - helper macro to iterate over all memory zones
973 * @zone - pointer to struct zone variable
975 * The user only needs to declare the zone variable, for_each_zone
978 #define for_each_zone(zone) \
979 for (zone = (first_online_pgdat())->node_zones; \
981 zone = next_zone(zone))
983 #define for_each_populated_zone(zone) \
984 for (zone = (first_online_pgdat())->node_zones; \
986 zone = next_zone(zone)) \
987 if (!populated_zone(zone)) \
991 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
993 return zoneref->zone;
996 static inline int zonelist_zone_idx(struct zoneref *zoneref)
998 return zoneref->zone_idx;
1001 static inline int zonelist_node_idx(struct zoneref *zoneref)
1003 return zone_to_nid(zoneref->zone);
1006 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1007 enum zone_type highest_zoneidx,
1011 * 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
1012 * @z - The cursor used as a starting point for the search
1013 * @highest_zoneidx - The zone index of the highest zone to return
1014 * @nodes - An optional nodemask to filter the zonelist with
1016 * This function returns the next zone at or below a given zone index that is
1017 * within the allowed nodemask using a cursor as the starting point for the
1018 * search. The zoneref returned is a cursor that represents the current zone
1019 * being examined. It should be advanced by one before calling
1020 * next_zones_zonelist again.
1022 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1023 enum zone_type highest_zoneidx,
1026 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1028 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1032 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1033 * @zonelist - The zonelist to search for a suitable zone
1034 * @highest_zoneidx - The zone index of the highest zone to return
1035 * @nodes - An optional nodemask to filter the zonelist with
1036 * @return - Zoneref pointer for the first suitable zone found (see below)
1038 * This function returns the first zone at or below a given zone index that is
1039 * within the allowed nodemask. The zoneref returned is a cursor that can be
1040 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1041 * one before calling.
1043 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1044 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1045 * update due to cpuset modification.
1047 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1048 enum zone_type highest_zoneidx,
1051 return next_zones_zonelist(zonelist->_zonerefs,
1052 highest_zoneidx, nodes);
1056 * 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
1057 * @zone - The current zone in the iterator
1058 * @z - The current pointer within zonelist->_zonerefs being iterated
1059 * @zlist - The zonelist being iterated
1060 * @highidx - The zone index of the highest zone to return
1061 * @nodemask - Nodemask allowed by the allocator
1063 * This iterator iterates though all zones at or below a given zone index and
1064 * within a given nodemask
1066 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1067 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1069 z = next_zones_zonelist(++z, highidx, nodemask), \
1070 zone = zonelist_zone(z))
1072 #define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1073 for (zone = z->zone; \
1075 z = next_zones_zonelist(++z, highidx, nodemask), \
1076 zone = zonelist_zone(z))
1080 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1081 * @zone - The current zone in the iterator
1082 * @z - The current pointer within zonelist->zones being iterated
1083 * @zlist - The zonelist being iterated
1084 * @highidx - The zone index of the highest zone to return
1086 * This iterator iterates though all zones at or below a given zone index.
1088 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1089 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1091 #ifdef CONFIG_SPARSEMEM
1092 #include <asm/sparsemem.h>
1095 #ifdef CONFIG_FLATMEM
1096 #define pfn_to_nid(pfn) (0)
1099 #ifdef CONFIG_SPARSEMEM
1102 * SECTION_SHIFT #bits space required to store a section #
1104 * PA_SECTION_SHIFT physical address to/from section number
1105 * PFN_SECTION_SHIFT pfn to/from section number
1107 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1108 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1110 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1112 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1113 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1115 #define SECTION_BLOCKFLAGS_BITS \
1116 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1118 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1119 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1122 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1124 return pfn >> PFN_SECTION_SHIFT;
1126 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1128 return sec << PFN_SECTION_SHIFT;
1131 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1132 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1134 #define SUBSECTION_SHIFT 21
1135 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1137 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1138 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1139 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1141 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1142 #error Subsection size exceeds section size
1144 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1147 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1148 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1150 struct mem_section_usage {
1151 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1152 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1154 /* See declaration of similar field in struct zone */
1155 unsigned long pageblock_flags[0];
1158 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1162 struct mem_section {
1164 * This is, logically, a pointer to an array of struct
1165 * pages. However, it is stored with some other magic.
1166 * (see sparse.c::sparse_init_one_section())
1168 * Additionally during early boot we encode node id of
1169 * the location of the section here to guide allocation.
1170 * (see sparse.c::memory_present())
1172 * Making it a UL at least makes someone do a cast
1173 * before using it wrong.
1175 unsigned long section_mem_map;
1177 struct mem_section_usage *usage;
1178 #ifdef CONFIG_PAGE_EXTENSION
1180 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1181 * section. (see page_ext.h about this.)
1183 struct page_ext *page_ext;
1187 * WARNING: mem_section must be a power-of-2 in size for the
1188 * calculation and use of SECTION_ROOT_MASK to make sense.
1192 #ifdef CONFIG_SPARSEMEM_EXTREME
1193 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1195 #define SECTIONS_PER_ROOT 1
1198 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1199 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1200 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1202 #ifdef CONFIG_SPARSEMEM_EXTREME
1203 extern struct mem_section **mem_section;
1205 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1208 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1210 return ms->usage->pageblock_flags;
1213 static inline struct mem_section *__nr_to_section(unsigned long nr)
1215 #ifdef CONFIG_SPARSEMEM_EXTREME
1219 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1221 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1223 extern unsigned long __section_nr(struct mem_section *ms);
1224 extern size_t mem_section_usage_size(void);
1227 * We use the lower bits of the mem_map pointer to store
1228 * a little bit of information. The pointer is calculated
1229 * as mem_map - section_nr_to_pfn(pnum). The result is
1230 * aligned to the minimum alignment of the two values:
1231 * 1. All mem_map arrays are page-aligned.
1232 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1233 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1234 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1235 * worst combination is powerpc with 256k pages,
1236 * which results in PFN_SECTION_SHIFT equal 6.
1237 * To sum it up, at least 6 bits are available.
1239 #define SECTION_MARKED_PRESENT (1UL<<0)
1240 #define SECTION_HAS_MEM_MAP (1UL<<1)
1241 #define SECTION_IS_ONLINE (1UL<<2)
1242 #define SECTION_IS_EARLY (1UL<<3)
1243 #define SECTION_MAP_LAST_BIT (1UL<<4)
1244 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1245 #define SECTION_NID_SHIFT 3
1247 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1249 unsigned long map = section->section_mem_map;
1250 map &= SECTION_MAP_MASK;
1251 return (struct page *)map;
1254 static inline int present_section(struct mem_section *section)
1256 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1259 static inline int present_section_nr(unsigned long nr)
1261 return present_section(__nr_to_section(nr));
1264 static inline int valid_section(struct mem_section *section)
1266 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1269 static inline int early_section(struct mem_section *section)
1271 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1274 static inline int valid_section_nr(unsigned long nr)
1276 return valid_section(__nr_to_section(nr));
1279 static inline int online_section(struct mem_section *section)
1281 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1284 static inline int online_section_nr(unsigned long nr)
1286 return online_section(__nr_to_section(nr));
1289 #ifdef CONFIG_MEMORY_HOTPLUG
1290 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1291 #ifdef CONFIG_MEMORY_HOTREMOVE
1292 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1296 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1298 return __nr_to_section(pfn_to_section_nr(pfn));
1301 extern unsigned long __highest_present_section_nr;
1303 static inline int subsection_map_index(unsigned long pfn)
1305 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1308 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1309 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1311 int idx = subsection_map_index(pfn);
1313 return test_bit(idx, ms->usage->subsection_map);
1316 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1322 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1323 static inline int pfn_valid(unsigned long pfn)
1325 struct mem_section *ms;
1327 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1329 ms = __nr_to_section(pfn_to_section_nr(pfn));
1330 if (!valid_section(ms))
1333 * Traditionally early sections always returned pfn_valid() for
1334 * the entire section-sized span.
1336 return early_section(ms) || pfn_section_valid(ms, pfn);
1340 static inline int pfn_in_present_section(unsigned long pfn)
1342 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1344 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1347 static inline unsigned long next_present_section_nr(unsigned long section_nr)
1349 while (++section_nr <= __highest_present_section_nr) {
1350 if (present_section_nr(section_nr))
1358 * These are _only_ used during initialisation, therefore they
1359 * can use __initdata ... They could have names to indicate
1363 #define pfn_to_nid(pfn) \
1365 unsigned long __pfn_to_nid_pfn = (pfn); \
1366 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1369 #define pfn_to_nid(pfn) (0)
1372 #define early_pfn_valid(pfn) pfn_valid(pfn)
1373 void sparse_init(void);
1375 #define sparse_init() do {} while (0)
1376 #define sparse_index_init(_sec, _nid) do {} while (0)
1377 #define pfn_in_present_section pfn_valid
1378 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
1379 #endif /* CONFIG_SPARSEMEM */
1382 * During memory init memblocks map pfns to nids. The search is expensive and
1383 * this caches recent lookups. The implementation of __early_pfn_to_nid
1384 * may treat start/end as pfns or sections.
1386 struct mminit_pfnnid_cache {
1387 unsigned long last_start;
1388 unsigned long last_end;
1392 #ifndef early_pfn_valid
1393 #define early_pfn_valid(pfn) (1)
1397 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1398 * need to check pfn validity within that MAX_ORDER_NR_PAGES block.
1399 * pfn_valid_within() should be used in this case; we optimise this away
1400 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1402 #ifdef CONFIG_HOLES_IN_ZONE
1403 #define pfn_valid_within(pfn) pfn_valid(pfn)
1405 #define pfn_valid_within(pfn) (1)
1408 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1410 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1411 * associated with it or not. This means that a struct page exists for this
1412 * pfn. The caller cannot assume the page is fully initialized in general.
1413 * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1414 * will ensure the struct page is fully online and initialized. Special pages
1415 * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1417 * In FLATMEM, it is expected that holes always have valid memmap as long as
1418 * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1419 * that a valid section has a memmap for the entire section.
1421 * However, an ARM, and maybe other embedded architectures in the future
1422 * free memmap backing holes to save memory on the assumption the memmap is
1423 * never used. The page_zone linkages are then broken even though pfn_valid()
1424 * returns true. A walker of the full memmap must then do this additional
1425 * check to ensure the memmap they are looking at is sane by making sure
1426 * the zone and PFN linkages are still valid. This is expensive, but walkers
1427 * of the full memmap are extremely rare.
1429 bool memmap_valid_within(unsigned long pfn,
1430 struct page *page, struct zone *zone);
1432 static inline bool memmap_valid_within(unsigned long pfn,
1433 struct page *page, struct zone *zone)
1437 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1439 #endif /* !__GENERATING_BOUNDS.H */
1440 #endif /* !__ASSEMBLY__ */
1441 #endif /* _LINUX_MMZONE_H */