1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_MMZONE_H
3 #define _LINUX_MMZONE_H
6 #ifndef __GENERATING_BOUNDS_H
8 #include <linux/spinlock.h>
9 #include <linux/list.h>
10 #include <linux/wait.h>
11 #include <linux/bitops.h>
12 #include <linux/cache.h>
13 #include <linux/threads.h>
14 #include <linux/numa.h>
15 #include <linux/init.h>
16 #include <linux/seqlock.h>
17 #include <linux/nodemask.h>
18 #include <linux/pageblock-flags.h>
19 #include <linux/page-flags-layout.h>
20 #include <linux/atomic.h>
21 #include <linux/mm_types.h>
22 #include <linux/page-flags.h>
25 /* Free memory management - zoned buddy allocator. */
26 #ifndef CONFIG_FORCE_MAX_ZONEORDER
29 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
31 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
34 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
35 * costly to service. That is between allocation orders which should
36 * coalesce naturally under reasonable reclaim pressure and those which
39 #define PAGE_ALLOC_COSTLY_ORDER 3
45 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
46 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
49 * MIGRATE_CMA migration type is designed to mimic the way
50 * ZONE_MOVABLE works. Only movable pages can be allocated
51 * from MIGRATE_CMA pageblocks and page allocator never
52 * implicitly change migration type of MIGRATE_CMA pageblock.
54 * The way to use it is to change migratetype of a range of
55 * pageblocks to MIGRATE_CMA which can be done by
56 * __free_pageblock_cma() function. What is important though
57 * is that a range of pageblocks must be aligned to
58 * MAX_ORDER_NR_PAGES should biggest page be bigger then
63 #ifdef CONFIG_MEMORY_ISOLATION
64 MIGRATE_ISOLATE, /* can't allocate from here */
69 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
70 extern const char * const migratetype_names[MIGRATE_TYPES];
73 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
74 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
76 # define is_migrate_cma(migratetype) false
77 # define is_migrate_cma_page(_page) false
80 static inline bool is_migrate_movable(int mt)
82 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
85 #define for_each_migratetype_order(order, type) \
86 for (order = 0; order < MAX_ORDER; order++) \
87 for (type = 0; type < MIGRATE_TYPES; type++)
89 extern int page_group_by_mobility_disabled;
91 #define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1)
92 #define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1)
94 #define get_pageblock_migratetype(page) \
95 get_pfnblock_flags_mask(page, page_to_pfn(page), \
96 PB_migrate_end, MIGRATETYPE_MASK)
99 struct list_head free_list[MIGRATE_TYPES];
100 unsigned long nr_free;
103 static inline struct page *get_page_from_free_area(struct free_area *area,
106 return list_first_entry_or_null(&area->free_list[migratetype],
110 static inline bool free_area_empty(struct free_area *area, int migratetype)
112 return list_empty(&area->free_list[migratetype]);
118 * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
119 * So add a wild amount of padding here to ensure that they fall into separate
120 * cachelines. There are very few zone structures in the machine, so space
121 * consumption is not a concern here.
123 #if defined(CONFIG_SMP)
124 struct zone_padding {
126 } ____cacheline_internodealigned_in_smp;
127 #define ZONE_PADDING(name) struct zone_padding name;
129 #define ZONE_PADDING(name)
133 enum numa_stat_item {
134 NUMA_HIT, /* allocated in intended node */
135 NUMA_MISS, /* allocated in non intended node */
136 NUMA_FOREIGN, /* was intended here, hit elsewhere */
137 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
138 NUMA_LOCAL, /* allocation from local node */
139 NUMA_OTHER, /* allocation from other node */
140 NR_VM_NUMA_STAT_ITEMS
143 #define NR_VM_NUMA_STAT_ITEMS 0
146 enum zone_stat_item {
147 /* First 128 byte cacheline (assuming 64 bit words) */
149 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
150 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
152 NR_ZONE_INACTIVE_FILE,
155 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
156 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
157 NR_PAGETABLE, /* used for pagetables */
158 NR_KERNEL_STACK_KB, /* measured in KiB */
159 #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
160 NR_KERNEL_SCS_KB, /* measured in KiB */
162 /* Second 128 byte cacheline */
164 #if IS_ENABLED(CONFIG_ZSMALLOC)
165 NR_ZSPAGES, /* allocated in zsmalloc */
168 NR_VM_ZONE_STAT_ITEMS };
170 enum node_stat_item {
172 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
173 NR_ACTIVE_ANON, /* " " " " " */
174 NR_INACTIVE_FILE, /* " " " " " */
175 NR_ACTIVE_FILE, /* " " " " " */
176 NR_UNEVICTABLE, /* " " " " " */
178 NR_SLAB_UNRECLAIMABLE,
179 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
180 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
185 WORKINGSET_NODERECLAIM,
186 NR_ANON_MAPPED, /* Mapped anonymous pages */
187 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
188 only modified from process context */
192 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
193 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
200 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
201 NR_DIRTIED, /* page dirtyings since bootup */
202 NR_WRITTEN, /* page writings since bootup */
203 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
204 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
205 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
206 NR_VM_NODE_STAT_ITEMS
210 * We do arithmetic on the LRU lists in various places in the code,
211 * so it is important to keep the active lists LRU_ACTIVE higher in
212 * the array than the corresponding inactive lists, and to keep
213 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
215 * This has to be kept in sync with the statistics in zone_stat_item
216 * above and the descriptions in vmstat_text in mm/vmstat.c
223 LRU_INACTIVE_ANON = LRU_BASE,
224 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
225 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
226 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
231 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
233 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
235 static inline bool is_file_lru(enum lru_list lru)
237 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
240 static inline bool is_active_lru(enum lru_list lru)
242 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
246 LRUVEC_CONGESTED, /* lruvec has many dirty pages
247 * backed by a congested BDI
252 struct list_head lists[NR_LRU_LISTS];
254 * These track the cost of reclaiming one LRU - file or anon -
255 * over the other. As the observed cost of reclaiming one LRU
256 * increases, the reclaim scan balance tips toward the other.
258 unsigned long anon_cost;
259 unsigned long file_cost;
260 /* Non-resident age, driven by LRU movement */
261 atomic_long_t nonresident_age;
262 /* Refaults at the time of last reclaim cycle */
263 unsigned long refaults;
264 /* Various lruvec state flags (enum lruvec_flags) */
267 struct pglist_data *pgdat;
271 /* Isolate unmapped pages */
272 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
273 /* Isolate for asynchronous migration */
274 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
275 /* Isolate unevictable pages */
276 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
278 /* LRU Isolation modes. */
279 typedef unsigned __bitwise isolate_mode_t;
281 enum zone_watermarks {
288 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
289 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
290 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
291 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
293 struct per_cpu_pages {
294 int count; /* number of pages in the list */
295 int high; /* high watermark, emptying needed */
296 int batch; /* chunk size for buddy add/remove */
298 /* Lists of pages, one per migrate type stored on the pcp-lists */
299 struct list_head lists[MIGRATE_PCPTYPES];
302 struct per_cpu_pageset {
303 struct per_cpu_pages pcp;
306 u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
310 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
314 struct per_cpu_nodestat {
316 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
319 #endif /* !__GENERATING_BOUNDS.H */
323 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
324 * to DMA to all of the addressable memory (ZONE_NORMAL).
325 * On architectures where this area covers the whole 32 bit address
326 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
327 * DMA addressing constraints. This distinction is important as a 32bit
328 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
329 * platforms may need both zones as they support peripherals with
330 * different DMA addressing limitations.
334 * - i386 and x86_64 have a fixed 16M ZONE_DMA and ZONE_DMA32 for the
335 * rest of the lower 4G.
337 * - arm only uses ZONE_DMA, the size, up to 4G, may vary depending on
338 * the specific device.
340 * - arm64 has a fixed 1G ZONE_DMA and ZONE_DMA32 for the rest of the
343 * - powerpc only uses ZONE_DMA, the size, up to 2G, may vary
344 * depending on the specific device.
346 * - s390 uses ZONE_DMA fixed to the lower 2G.
348 * - ia64 and riscv only use ZONE_DMA32.
350 * - parisc uses neither.
352 #ifdef CONFIG_ZONE_DMA
355 #ifdef CONFIG_ZONE_DMA32
359 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
360 * performed on pages in ZONE_NORMAL if the DMA devices support
361 * transfers to all addressable memory.
364 #ifdef CONFIG_HIGHMEM
366 * A memory area that is only addressable by the kernel through
367 * mapping portions into its own address space. This is for example
368 * used by i386 to allow the kernel to address the memory beyond
369 * 900MB. The kernel will set up special mappings (page
370 * table entries on i386) for each page that the kernel needs to
376 #ifdef CONFIG_ZONE_DEVICE
383 #ifndef __GENERATING_BOUNDS_H
386 /* Read-mostly fields */
388 /* zone watermarks, access with *_wmark_pages(zone) macros */
389 unsigned long _watermark[NR_WMARK];
390 unsigned long watermark_boost;
392 unsigned long nr_reserved_highatomic;
395 * We don't know if the memory that we're going to allocate will be
396 * freeable or/and it will be released eventually, so to avoid totally
397 * wasting several GB of ram we must reserve some of the lower zone
398 * memory (otherwise we risk to run OOM on the lower zones despite
399 * there being tons of freeable ram on the higher zones). This array is
400 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
403 long lowmem_reserve[MAX_NR_ZONES];
408 struct pglist_data *zone_pgdat;
409 struct per_cpu_pageset __percpu *pageset;
411 #ifndef CONFIG_SPARSEMEM
413 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
414 * In SPARSEMEM, this map is stored in struct mem_section
416 unsigned long *pageblock_flags;
417 #endif /* CONFIG_SPARSEMEM */
419 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
420 unsigned long zone_start_pfn;
423 * spanned_pages is the total pages spanned by the zone, including
424 * holes, which is calculated as:
425 * spanned_pages = zone_end_pfn - zone_start_pfn;
427 * present_pages is physical pages existing within the zone, which
429 * present_pages = spanned_pages - absent_pages(pages in holes);
431 * managed_pages is present pages managed by the buddy system, which
432 * is calculated as (reserved_pages includes pages allocated by the
433 * bootmem allocator):
434 * managed_pages = present_pages - reserved_pages;
436 * So present_pages may be used by memory hotplug or memory power
437 * management logic to figure out unmanaged pages by checking
438 * (present_pages - managed_pages). And managed_pages should be used
439 * by page allocator and vm scanner to calculate all kinds of watermarks
444 * zone_start_pfn and spanned_pages are protected by span_seqlock.
445 * It is a seqlock because it has to be read outside of zone->lock,
446 * and it is done in the main allocator path. But, it is written
447 * quite infrequently.
449 * The span_seq lock is declared along with zone->lock because it is
450 * frequently read in proximity to zone->lock. It's good to
451 * give them a chance of being in the same cacheline.
453 * Write access to present_pages at runtime should be protected by
454 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
455 * present_pages should get_online_mems() to get a stable value.
457 atomic_long_t managed_pages;
458 unsigned long spanned_pages;
459 unsigned long present_pages;
463 #ifdef CONFIG_MEMORY_ISOLATION
465 * Number of isolated pageblock. It is used to solve incorrect
466 * freepage counting problem due to racy retrieving migratetype
467 * of pageblock. Protected by zone->lock.
469 unsigned long nr_isolate_pageblock;
472 #ifdef CONFIG_MEMORY_HOTPLUG
473 /* see spanned/present_pages for more description */
474 seqlock_t span_seqlock;
479 /* Write-intensive fields used from the page allocator */
482 /* free areas of different sizes */
483 struct free_area free_area[MAX_ORDER];
485 /* zone flags, see below */
488 /* Primarily protects free_area */
491 /* Write-intensive fields used by compaction and vmstats. */
495 * When free pages are below this point, additional steps are taken
496 * when reading the number of free pages to avoid per-cpu counter
497 * drift allowing watermarks to be breached
499 unsigned long percpu_drift_mark;
501 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
502 /* pfn where compaction free scanner should start */
503 unsigned long compact_cached_free_pfn;
504 /* pfn where async and sync compaction migration scanner should start */
505 unsigned long compact_cached_migrate_pfn[2];
506 unsigned long compact_init_migrate_pfn;
507 unsigned long compact_init_free_pfn;
510 #ifdef CONFIG_COMPACTION
512 * On compaction failure, 1<<compact_defer_shift compactions
513 * are skipped before trying again. The number attempted since
514 * last failure is tracked with compact_considered.
516 unsigned int compact_considered;
517 unsigned int compact_defer_shift;
518 int compact_order_failed;
521 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
522 /* Set to true when the PG_migrate_skip bits should be cleared */
523 bool compact_blockskip_flush;
529 /* Zone statistics */
530 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
531 atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
532 } ____cacheline_internodealigned_in_smp;
535 PGDAT_DIRTY, /* reclaim scanning has recently found
536 * many dirty file pages at the tail
539 PGDAT_WRITEBACK, /* reclaim scanning has recently found
540 * many pages under writeback
542 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
546 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
547 * Cleared when kswapd is woken.
551 static inline unsigned long zone_managed_pages(struct zone *zone)
553 return (unsigned long)atomic_long_read(&zone->managed_pages);
556 static inline unsigned long zone_end_pfn(const struct zone *zone)
558 return zone->zone_start_pfn + zone->spanned_pages;
561 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
563 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
566 static inline bool zone_is_initialized(struct zone *zone)
568 return zone->initialized;
571 static inline bool zone_is_empty(struct zone *zone)
573 return zone->spanned_pages == 0;
577 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
578 * intersection with the given zone
580 static inline bool zone_intersects(struct zone *zone,
581 unsigned long start_pfn, unsigned long nr_pages)
583 if (zone_is_empty(zone))
585 if (start_pfn >= zone_end_pfn(zone) ||
586 start_pfn + nr_pages <= zone->zone_start_pfn)
593 * The "priority" of VM scanning is how much of the queues we will scan in one
594 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
595 * queues ("queue_length >> 12") during an aging round.
597 #define DEF_PRIORITY 12
599 /* Maximum number of zones on a zonelist */
600 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
603 ZONELIST_FALLBACK, /* zonelist with fallback */
606 * The NUMA zonelists are doubled because we need zonelists that
607 * restrict the allocations to a single node for __GFP_THISNODE.
609 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
615 * This struct contains information about a zone in a zonelist. It is stored
616 * here to avoid dereferences into large structures and lookups of tables
619 struct zone *zone; /* Pointer to actual zone */
620 int zone_idx; /* zone_idx(zoneref->zone) */
624 * One allocation request operates on a zonelist. A zonelist
625 * is a list of zones, the first one is the 'goal' of the
626 * allocation, the other zones are fallback zones, in decreasing
629 * To speed the reading of the zonelist, the zonerefs contain the zone index
630 * of the entry being read. Helper functions to access information given
631 * a struct zoneref are
633 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
634 * zonelist_zone_idx() - Return the index of the zone for an entry
635 * zonelist_node_idx() - Return the index of the node for an entry
638 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
641 #ifndef CONFIG_DISCONTIGMEM
642 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
643 extern struct page *mem_map;
646 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
647 struct deferred_split {
648 spinlock_t split_queue_lock;
649 struct list_head split_queue;
650 unsigned long split_queue_len;
655 * On NUMA machines, each NUMA node would have a pg_data_t to describe
656 * it's memory layout. On UMA machines there is a single pglist_data which
657 * describes the whole memory.
659 * Memory statistics and page replacement data structures are maintained on a
662 typedef struct pglist_data {
664 * node_zones contains just the zones for THIS node. Not all of the
665 * zones may be populated, but it is the full list. It is referenced by
666 * this node's node_zonelists as well as other node's node_zonelists.
668 struct zone node_zones[MAX_NR_ZONES];
671 * node_zonelists contains references to all zones in all nodes.
672 * Generally the first zones will be references to this node's
675 struct zonelist node_zonelists[MAX_ZONELISTS];
677 int nr_zones; /* number of populated zones in this node */
678 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
679 struct page *node_mem_map;
680 #ifdef CONFIG_PAGE_EXTENSION
681 struct page_ext *node_page_ext;
684 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
686 * Must be held any time you expect node_start_pfn,
687 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
688 * Also synchronizes pgdat->first_deferred_pfn during deferred page
691 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
692 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
693 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
695 * Nests above zone->lock and zone->span_seqlock
697 spinlock_t node_size_lock;
699 unsigned long node_start_pfn;
700 unsigned long node_present_pages; /* total number of physical pages */
701 unsigned long node_spanned_pages; /* total size of physical page
702 range, including holes */
704 wait_queue_head_t kswapd_wait;
705 wait_queue_head_t pfmemalloc_wait;
706 struct task_struct *kswapd; /* Protected by
707 mem_hotplug_begin/end() */
709 enum zone_type kswapd_highest_zoneidx;
711 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
713 #ifdef CONFIG_COMPACTION
714 int kcompactd_max_order;
715 enum zone_type kcompactd_highest_zoneidx;
716 wait_queue_head_t kcompactd_wait;
717 struct task_struct *kcompactd;
720 * This is a per-node reserve of pages that are not available
721 * to userspace allocations.
723 unsigned long totalreserve_pages;
727 * node reclaim becomes active if more unmapped pages exist.
729 unsigned long min_unmapped_pages;
730 unsigned long min_slab_pages;
731 #endif /* CONFIG_NUMA */
733 /* Write-intensive fields used by page reclaim */
737 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
739 * If memory initialisation on large machines is deferred then this
740 * is the first PFN that needs to be initialised.
742 unsigned long first_deferred_pfn;
743 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
745 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
746 struct deferred_split deferred_split_queue;
749 /* Fields commonly accessed by the page reclaim scanner */
752 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
754 * Use mem_cgroup_lruvec() to look up lruvecs.
756 struct lruvec __lruvec;
762 /* Per-node vmstats */
763 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
764 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
767 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
768 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
769 #ifdef CONFIG_FLAT_NODE_MEM_MAP
770 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
772 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
774 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
776 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
777 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
779 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
781 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
784 static inline bool pgdat_is_empty(pg_data_t *pgdat)
786 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
789 #include <linux/memory_hotplug.h>
791 void build_all_zonelists(pg_data_t *pgdat);
792 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
793 enum zone_type highest_zoneidx);
794 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
795 int highest_zoneidx, unsigned int alloc_flags,
797 bool zone_watermark_ok(struct zone *z, unsigned int order,
798 unsigned long mark, int highest_zoneidx,
799 unsigned int alloc_flags);
800 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
801 unsigned long mark, int highest_zoneidx);
802 enum memmap_context {
806 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
809 extern void lruvec_init(struct lruvec *lruvec);
811 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
814 return lruvec->pgdat;
816 return container_of(lruvec, struct pglist_data, __lruvec);
820 extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
822 #ifdef CONFIG_HAVE_MEMORY_PRESENT
823 void memory_present(int nid, unsigned long start, unsigned long end);
825 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
828 #if defined(CONFIG_SPARSEMEM)
829 void memblocks_present(void);
831 static inline void memblocks_present(void) {}
834 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
835 int local_memory_node(int node_id);
837 static inline int local_memory_node(int node_id) { return node_id; };
841 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
843 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
846 * Returns true if a zone has pages managed by the buddy allocator.
847 * All the reclaim decisions have to use this function rather than
848 * populated_zone(). If the whole zone is reserved then we can easily
849 * end up with populated_zone() && !managed_zone().
851 static inline bool managed_zone(struct zone *zone)
853 return zone_managed_pages(zone);
856 /* Returns true if a zone has memory */
857 static inline bool populated_zone(struct zone *zone)
859 return zone->present_pages;
863 static inline int zone_to_nid(struct zone *zone)
868 static inline void zone_set_nid(struct zone *zone, int nid)
873 static inline int zone_to_nid(struct zone *zone)
878 static inline void zone_set_nid(struct zone *zone, int nid) {}
881 extern int movable_zone;
883 #ifdef CONFIG_HIGHMEM
884 static inline int zone_movable_is_highmem(void)
886 #ifdef CONFIG_NEED_MULTIPLE_NODES
887 return movable_zone == ZONE_HIGHMEM;
889 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
894 static inline int is_highmem_idx(enum zone_type idx)
896 #ifdef CONFIG_HIGHMEM
897 return (idx == ZONE_HIGHMEM ||
898 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
905 * is_highmem - helper function to quickly check if a struct zone is a
906 * highmem zone or not. This is an attempt to keep references
907 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
908 * @zone - pointer to struct zone variable
910 static inline int is_highmem(struct zone *zone)
912 #ifdef CONFIG_HIGHMEM
913 return is_highmem_idx(zone_idx(zone));
919 /* These two functions are used to setup the per zone pages min values */
922 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, void *, size_t *,
924 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, void *,
926 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
927 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, void *,
929 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
930 void *, size_t *, loff_t *);
931 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
932 void *, size_t *, loff_t *);
933 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
934 void *, size_t *, loff_t *);
935 int numa_zonelist_order_handler(struct ctl_table *, int,
936 void *, size_t *, loff_t *);
937 extern int percpu_pagelist_fraction;
938 extern char numa_zonelist_order[];
939 #define NUMA_ZONELIST_ORDER_LEN 16
941 #ifndef CONFIG_NEED_MULTIPLE_NODES
943 extern struct pglist_data contig_page_data;
944 #define NODE_DATA(nid) (&contig_page_data)
945 #define NODE_MEM_MAP(nid) mem_map
947 #else /* CONFIG_NEED_MULTIPLE_NODES */
949 #include <asm/mmzone.h>
951 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
953 extern struct pglist_data *first_online_pgdat(void);
954 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
955 extern struct zone *next_zone(struct zone *zone);
958 * for_each_online_pgdat - helper macro to iterate over all online nodes
959 * @pgdat - pointer to a pg_data_t variable
961 #define for_each_online_pgdat(pgdat) \
962 for (pgdat = first_online_pgdat(); \
964 pgdat = next_online_pgdat(pgdat))
966 * for_each_zone - helper macro to iterate over all memory zones
967 * @zone - pointer to struct zone variable
969 * The user only needs to declare the zone variable, for_each_zone
972 #define for_each_zone(zone) \
973 for (zone = (first_online_pgdat())->node_zones; \
975 zone = next_zone(zone))
977 #define for_each_populated_zone(zone) \
978 for (zone = (first_online_pgdat())->node_zones; \
980 zone = next_zone(zone)) \
981 if (!populated_zone(zone)) \
985 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
987 return zoneref->zone;
990 static inline int zonelist_zone_idx(struct zoneref *zoneref)
992 return zoneref->zone_idx;
995 static inline int zonelist_node_idx(struct zoneref *zoneref)
997 return zone_to_nid(zoneref->zone);
1000 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1001 enum zone_type highest_zoneidx,
1005 * 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
1006 * @z - The cursor used as a starting point for the search
1007 * @highest_zoneidx - The zone index of the highest zone to return
1008 * @nodes - An optional nodemask to filter the zonelist with
1010 * This function returns the next zone at or below a given zone index that is
1011 * within the allowed nodemask using a cursor as the starting point for the
1012 * search. The zoneref returned is a cursor that represents the current zone
1013 * being examined. It should be advanced by one before calling
1014 * next_zones_zonelist again.
1016 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1017 enum zone_type highest_zoneidx,
1020 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1022 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1026 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1027 * @zonelist - The zonelist to search for a suitable zone
1028 * @highest_zoneidx - The zone index of the highest zone to return
1029 * @nodes - An optional nodemask to filter the zonelist with
1030 * @return - Zoneref pointer for the first suitable zone found (see below)
1032 * This function returns the first zone at or below a given zone index that is
1033 * within the allowed nodemask. The zoneref returned is a cursor that can be
1034 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1035 * one before calling.
1037 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1038 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1039 * update due to cpuset modification.
1041 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1042 enum zone_type highest_zoneidx,
1045 return next_zones_zonelist(zonelist->_zonerefs,
1046 highest_zoneidx, nodes);
1050 * 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
1051 * @zone - The current zone in the iterator
1052 * @z - The current pointer within zonelist->_zonerefs being iterated
1053 * @zlist - The zonelist being iterated
1054 * @highidx - The zone index of the highest zone to return
1055 * @nodemask - Nodemask allowed by the allocator
1057 * This iterator iterates though all zones at or below a given zone index and
1058 * within a given nodemask
1060 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1061 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1063 z = next_zones_zonelist(++z, highidx, nodemask), \
1064 zone = zonelist_zone(z))
1066 #define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1067 for (zone = z->zone; \
1069 z = next_zones_zonelist(++z, highidx, nodemask), \
1070 zone = zonelist_zone(z))
1074 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1075 * @zone - The current zone in the iterator
1076 * @z - The current pointer within zonelist->zones being iterated
1077 * @zlist - The zonelist being iterated
1078 * @highidx - The zone index of the highest zone to return
1080 * This iterator iterates though all zones at or below a given zone index.
1082 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1083 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1085 #ifdef CONFIG_SPARSEMEM
1086 #include <asm/sparsemem.h>
1089 #ifdef CONFIG_FLATMEM
1090 #define pfn_to_nid(pfn) (0)
1093 #ifdef CONFIG_SPARSEMEM
1096 * SECTION_SHIFT #bits space required to store a section #
1098 * PA_SECTION_SHIFT physical address to/from section number
1099 * PFN_SECTION_SHIFT pfn to/from section number
1101 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1102 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1104 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1106 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1107 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1109 #define SECTION_BLOCKFLAGS_BITS \
1110 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1112 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1113 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1116 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1118 return pfn >> PFN_SECTION_SHIFT;
1120 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1122 return sec << PFN_SECTION_SHIFT;
1125 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1126 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1128 #define SUBSECTION_SHIFT 21
1129 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1131 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1132 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1133 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1135 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1136 #error Subsection size exceeds section size
1138 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1141 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1142 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1144 struct mem_section_usage {
1145 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1146 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1148 /* See declaration of similar field in struct zone */
1149 unsigned long pageblock_flags[0];
1152 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1156 struct mem_section {
1158 * This is, logically, a pointer to an array of struct
1159 * pages. However, it is stored with some other magic.
1160 * (see sparse.c::sparse_init_one_section())
1162 * Additionally during early boot we encode node id of
1163 * the location of the section here to guide allocation.
1164 * (see sparse.c::memory_present())
1166 * Making it a UL at least makes someone do a cast
1167 * before using it wrong.
1169 unsigned long section_mem_map;
1171 struct mem_section_usage *usage;
1172 #ifdef CONFIG_PAGE_EXTENSION
1174 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1175 * section. (see page_ext.h about this.)
1177 struct page_ext *page_ext;
1181 * WARNING: mem_section must be a power-of-2 in size for the
1182 * calculation and use of SECTION_ROOT_MASK to make sense.
1186 #ifdef CONFIG_SPARSEMEM_EXTREME
1187 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1189 #define SECTIONS_PER_ROOT 1
1192 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1193 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1194 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1196 #ifdef CONFIG_SPARSEMEM_EXTREME
1197 extern struct mem_section **mem_section;
1199 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1202 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1204 return ms->usage->pageblock_flags;
1207 static inline struct mem_section *__nr_to_section(unsigned long nr)
1209 #ifdef CONFIG_SPARSEMEM_EXTREME
1213 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1215 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1217 extern unsigned long __section_nr(struct mem_section *ms);
1218 extern size_t mem_section_usage_size(void);
1221 * We use the lower bits of the mem_map pointer to store
1222 * a little bit of information. The pointer is calculated
1223 * as mem_map - section_nr_to_pfn(pnum). The result is
1224 * aligned to the minimum alignment of the two values:
1225 * 1. All mem_map arrays are page-aligned.
1226 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1227 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1228 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1229 * worst combination is powerpc with 256k pages,
1230 * which results in PFN_SECTION_SHIFT equal 6.
1231 * To sum it up, at least 6 bits are available.
1233 #define SECTION_MARKED_PRESENT (1UL<<0)
1234 #define SECTION_HAS_MEM_MAP (1UL<<1)
1235 #define SECTION_IS_ONLINE (1UL<<2)
1236 #define SECTION_IS_EARLY (1UL<<3)
1237 #define SECTION_MAP_LAST_BIT (1UL<<4)
1238 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1239 #define SECTION_NID_SHIFT 3
1241 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1243 unsigned long map = section->section_mem_map;
1244 map &= SECTION_MAP_MASK;
1245 return (struct page *)map;
1248 static inline int present_section(struct mem_section *section)
1250 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1253 static inline int present_section_nr(unsigned long nr)
1255 return present_section(__nr_to_section(nr));
1258 static inline int valid_section(struct mem_section *section)
1260 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1263 static inline int early_section(struct mem_section *section)
1265 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1268 static inline int valid_section_nr(unsigned long nr)
1270 return valid_section(__nr_to_section(nr));
1273 static inline int online_section(struct mem_section *section)
1275 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1278 static inline int online_section_nr(unsigned long nr)
1280 return online_section(__nr_to_section(nr));
1283 #ifdef CONFIG_MEMORY_HOTPLUG
1284 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1285 #ifdef CONFIG_MEMORY_HOTREMOVE
1286 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1290 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1292 return __nr_to_section(pfn_to_section_nr(pfn));
1295 extern unsigned long __highest_present_section_nr;
1297 static inline int subsection_map_index(unsigned long pfn)
1299 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1302 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1303 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1305 int idx = subsection_map_index(pfn);
1307 return test_bit(idx, ms->usage->subsection_map);
1310 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1316 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1317 static inline int pfn_valid(unsigned long pfn)
1319 struct mem_section *ms;
1321 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1323 ms = __nr_to_section(pfn_to_section_nr(pfn));
1324 if (!valid_section(ms))
1327 * Traditionally early sections always returned pfn_valid() for
1328 * the entire section-sized span.
1330 return early_section(ms) || pfn_section_valid(ms, pfn);
1334 static inline int pfn_in_present_section(unsigned long pfn)
1336 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1338 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1341 static inline unsigned long next_present_section_nr(unsigned long section_nr)
1343 while (++section_nr <= __highest_present_section_nr) {
1344 if (present_section_nr(section_nr))
1352 * These are _only_ used during initialisation, therefore they
1353 * can use __initdata ... They could have names to indicate
1357 #define pfn_to_nid(pfn) \
1359 unsigned long __pfn_to_nid_pfn = (pfn); \
1360 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1363 #define pfn_to_nid(pfn) (0)
1366 #define early_pfn_valid(pfn) pfn_valid(pfn)
1367 void sparse_init(void);
1369 #define sparse_init() do {} while (0)
1370 #define sparse_index_init(_sec, _nid) do {} while (0)
1371 #define pfn_in_present_section pfn_valid
1372 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
1373 #endif /* CONFIG_SPARSEMEM */
1376 * During memory init memblocks map pfns to nids. The search is expensive and
1377 * this caches recent lookups. The implementation of __early_pfn_to_nid
1378 * may treat start/end as pfns or sections.
1380 struct mminit_pfnnid_cache {
1381 unsigned long last_start;
1382 unsigned long last_end;
1386 #ifndef early_pfn_valid
1387 #define early_pfn_valid(pfn) (1)
1390 void memory_present(int nid, unsigned long start, unsigned long end);
1393 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1394 * need to check pfn validity within that MAX_ORDER_NR_PAGES block.
1395 * pfn_valid_within() should be used in this case; we optimise this away
1396 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1398 #ifdef CONFIG_HOLES_IN_ZONE
1399 #define pfn_valid_within(pfn) pfn_valid(pfn)
1401 #define pfn_valid_within(pfn) (1)
1404 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1406 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1407 * associated with it or not. This means that a struct page exists for this
1408 * pfn. The caller cannot assume the page is fully initialized in general.
1409 * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1410 * will ensure the struct page is fully online and initialized. Special pages
1411 * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1413 * In FLATMEM, it is expected that holes always have valid memmap as long as
1414 * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1415 * that a valid section has a memmap for the entire section.
1417 * However, an ARM, and maybe other embedded architectures in the future
1418 * free memmap backing holes to save memory on the assumption the memmap is
1419 * never used. The page_zone linkages are then broken even though pfn_valid()
1420 * returns true. A walker of the full memmap must then do this additional
1421 * check to ensure the memmap they are looking at is sane by making sure
1422 * the zone and PFN linkages are still valid. This is expensive, but walkers
1423 * of the full memmap are extremely rare.
1425 bool memmap_valid_within(unsigned long pfn,
1426 struct page *page, struct zone *zone);
1428 static inline bool memmap_valid_within(unsigned long pfn,
1429 struct page *page, struct zone *zone)
1433 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1435 #endif /* !__GENERATING_BOUNDS.H */
1436 #endif /* !__ASSEMBLY__ */
1437 #endif /* _LINUX_MMZONE_H */