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 */
176 WORKINGSET_REFAULT_BASE,
177 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
178 WORKINGSET_REFAULT_FILE,
179 WORKINGSET_ACTIVATE_BASE,
180 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
181 WORKINGSET_ACTIVATE_FILE,
182 WORKINGSET_RESTORE_BASE,
183 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
184 WORKINGSET_RESTORE_FILE,
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_KERNEL_STACK_KB, /* measured in KiB */
207 #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
208 NR_KERNEL_SCS_KB, /* measured in KiB */
210 NR_VM_NODE_STAT_ITEMS
214 * Returns true if the value is measured in bytes (most vmstat values are
215 * measured in pages). This defines the API part, the internal representation
216 * might be different.
218 static __always_inline bool vmstat_item_in_bytes(int idx)
221 * Global and per-node slab counters track slab pages.
222 * It's expected that changes are multiples of PAGE_SIZE.
223 * Internally values are stored in pages.
225 * Per-memcg and per-lruvec counters track memory, consumed
226 * by individual slab objects. These counters are actually
229 return (idx == NR_SLAB_RECLAIMABLE_B ||
230 idx == NR_SLAB_UNRECLAIMABLE_B);
234 * We do arithmetic on the LRU lists in various places in the code,
235 * so it is important to keep the active lists LRU_ACTIVE higher in
236 * the array than the corresponding inactive lists, and to keep
237 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
239 * This has to be kept in sync with the statistics in zone_stat_item
240 * above and the descriptions in vmstat_text in mm/vmstat.c
247 LRU_INACTIVE_ANON = LRU_BASE,
248 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
249 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
250 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
255 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
257 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
259 static inline bool is_file_lru(enum lru_list lru)
261 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
264 static inline bool is_active_lru(enum lru_list lru)
266 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
269 #define ANON_AND_FILE 2
272 LRUVEC_CONGESTED, /* lruvec has many dirty pages
273 * backed by a congested BDI
278 struct list_head lists[NR_LRU_LISTS];
280 * These track the cost of reclaiming one LRU - file or anon -
281 * over the other. As the observed cost of reclaiming one LRU
282 * increases, the reclaim scan balance tips toward the other.
284 unsigned long anon_cost;
285 unsigned long file_cost;
286 /* Non-resident age, driven by LRU movement */
287 atomic_long_t nonresident_age;
288 /* Refaults at the time of last reclaim cycle */
289 unsigned long refaults[ANON_AND_FILE];
290 /* Various lruvec state flags (enum lruvec_flags) */
293 struct pglist_data *pgdat;
297 /* Isolate unmapped pages */
298 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
299 /* Isolate for asynchronous migration */
300 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
301 /* Isolate unevictable pages */
302 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
304 /* LRU Isolation modes. */
305 typedef unsigned __bitwise isolate_mode_t;
307 enum zone_watermarks {
314 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
315 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
316 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
317 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
319 struct per_cpu_pages {
320 int count; /* number of pages in the list */
321 int high; /* high watermark, emptying needed */
322 int batch; /* chunk size for buddy add/remove */
324 /* Lists of pages, one per migrate type stored on the pcp-lists */
325 struct list_head lists[MIGRATE_PCPTYPES];
328 struct per_cpu_pageset {
329 struct per_cpu_pages pcp;
332 u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
336 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
340 struct per_cpu_nodestat {
342 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
345 #endif /* !__GENERATING_BOUNDS.H */
349 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
350 * to DMA to all of the addressable memory (ZONE_NORMAL).
351 * On architectures where this area covers the whole 32 bit address
352 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
353 * DMA addressing constraints. This distinction is important as a 32bit
354 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
355 * platforms may need both zones as they support peripherals with
356 * different DMA addressing limitations.
358 #ifdef CONFIG_ZONE_DMA
361 #ifdef CONFIG_ZONE_DMA32
365 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
366 * performed on pages in ZONE_NORMAL if the DMA devices support
367 * transfers to all addressable memory.
370 #ifdef CONFIG_HIGHMEM
372 * A memory area that is only addressable by the kernel through
373 * mapping portions into its own address space. This is for example
374 * used by i386 to allow the kernel to address the memory beyond
375 * 900MB. The kernel will set up special mappings (page
376 * table entries on i386) for each page that the kernel needs to
382 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
383 * movable pages with few exceptional cases described below. Main use
384 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
385 * likely to succeed, and to locally limit unmovable allocations - e.g.,
386 * to increase the number of THP/huge pages. Notable special cases are:
388 * 1. Pinned pages: (long-term) pinning of movable pages might
389 * essentially turn such pages unmovable. Memory offlining might
391 * 2. memblock allocations: kernelcore/movablecore setups might create
392 * situations where ZONE_MOVABLE contains unmovable allocations
393 * after boot. Memory offlining and allocations fail early.
394 * 3. Memory holes: kernelcore/movablecore setups might create very rare
395 * situations where ZONE_MOVABLE contains memory holes after boot,
396 * for example, if we have sections that are only partially
397 * populated. Memory offlining and allocations fail early.
398 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
399 * memory offlining, such pages cannot be allocated.
400 * 5. Unmovable PG_offline pages: in paravirtualized environments,
401 * hotplugged memory blocks might only partially be managed by the
402 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
403 * parts not manged by the buddy are unmovable PG_offline pages. In
404 * some cases (virtio-mem), such pages can be skipped during
405 * memory offlining, however, cannot be moved/allocated. These
406 * techniques might use alloc_contig_range() to hide previously
407 * exposed pages from the buddy again (e.g., to implement some sort
408 * of memory unplug in virtio-mem).
410 * In general, no unmovable allocations that degrade memory offlining
411 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
412 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
413 * if has_unmovable_pages() states that there are no unmovable pages,
414 * there can be false negatives).
417 #ifdef CONFIG_ZONE_DEVICE
424 #ifndef __GENERATING_BOUNDS_H
426 #define ASYNC_AND_SYNC 2
429 /* Read-mostly fields */
431 /* zone watermarks, access with *_wmark_pages(zone) macros */
432 unsigned long _watermark[NR_WMARK];
433 unsigned long watermark_boost;
435 unsigned long nr_reserved_highatomic;
438 * We don't know if the memory that we're going to allocate will be
439 * freeable or/and it will be released eventually, so to avoid totally
440 * wasting several GB of ram we must reserve some of the lower zone
441 * memory (otherwise we risk to run OOM on the lower zones despite
442 * there being tons of freeable ram on the higher zones). This array is
443 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
446 long lowmem_reserve[MAX_NR_ZONES];
451 struct pglist_data *zone_pgdat;
452 struct per_cpu_pageset __percpu *pageset;
454 #ifndef CONFIG_SPARSEMEM
456 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
457 * In SPARSEMEM, this map is stored in struct mem_section
459 unsigned long *pageblock_flags;
460 #endif /* CONFIG_SPARSEMEM */
462 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
463 unsigned long zone_start_pfn;
466 * spanned_pages is the total pages spanned by the zone, including
467 * holes, which is calculated as:
468 * spanned_pages = zone_end_pfn - zone_start_pfn;
470 * present_pages is physical pages existing within the zone, which
472 * present_pages = spanned_pages - absent_pages(pages in holes);
474 * managed_pages is present pages managed by the buddy system, which
475 * is calculated as (reserved_pages includes pages allocated by the
476 * bootmem allocator):
477 * managed_pages = present_pages - reserved_pages;
479 * So present_pages may be used by memory hotplug or memory power
480 * management logic to figure out unmanaged pages by checking
481 * (present_pages - managed_pages). And managed_pages should be used
482 * by page allocator and vm scanner to calculate all kinds of watermarks
487 * zone_start_pfn and spanned_pages are protected by span_seqlock.
488 * It is a seqlock because it has to be read outside of zone->lock,
489 * and it is done in the main allocator path. But, it is written
490 * quite infrequently.
492 * The span_seq lock is declared along with zone->lock because it is
493 * frequently read in proximity to zone->lock. It's good to
494 * give them a chance of being in the same cacheline.
496 * Write access to present_pages at runtime should be protected by
497 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
498 * present_pages should get_online_mems() to get a stable value.
500 atomic_long_t managed_pages;
501 unsigned long spanned_pages;
502 unsigned long present_pages;
506 #ifdef CONFIG_MEMORY_ISOLATION
508 * Number of isolated pageblock. It is used to solve incorrect
509 * freepage counting problem due to racy retrieving migratetype
510 * of pageblock. Protected by zone->lock.
512 unsigned long nr_isolate_pageblock;
515 #ifdef CONFIG_MEMORY_HOTPLUG
516 /* see spanned/present_pages for more description */
517 seqlock_t span_seqlock;
522 /* Write-intensive fields used from the page allocator */
525 /* free areas of different sizes */
526 struct free_area free_area[MAX_ORDER];
528 /* zone flags, see below */
531 /* Primarily protects free_area */
534 /* Write-intensive fields used by compaction and vmstats. */
538 * When free pages are below this point, additional steps are taken
539 * when reading the number of free pages to avoid per-cpu counter
540 * drift allowing watermarks to be breached
542 unsigned long percpu_drift_mark;
544 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
545 /* pfn where compaction free scanner should start */
546 unsigned long compact_cached_free_pfn;
547 /* pfn where compaction migration scanner should start */
548 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
549 unsigned long compact_init_migrate_pfn;
550 unsigned long compact_init_free_pfn;
553 #ifdef CONFIG_COMPACTION
555 * On compaction failure, 1<<compact_defer_shift compactions
556 * are skipped before trying again. The number attempted since
557 * last failure is tracked with compact_considered.
558 * compact_order_failed is the minimum compaction failed order.
560 unsigned int compact_considered;
561 unsigned int compact_defer_shift;
562 int compact_order_failed;
565 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
566 /* Set to true when the PG_migrate_skip bits should be cleared */
567 bool compact_blockskip_flush;
573 /* Zone statistics */
574 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
575 atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
576 } ____cacheline_internodealigned_in_smp;
579 PGDAT_DIRTY, /* reclaim scanning has recently found
580 * many dirty file pages at the tail
583 PGDAT_WRITEBACK, /* reclaim scanning has recently found
584 * many pages under writeback
586 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
590 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
591 * Cleared when kswapd is woken.
595 static inline unsigned long zone_managed_pages(struct zone *zone)
597 return (unsigned long)atomic_long_read(&zone->managed_pages);
600 static inline unsigned long zone_end_pfn(const struct zone *zone)
602 return zone->zone_start_pfn + zone->spanned_pages;
605 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
607 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
610 static inline bool zone_is_initialized(struct zone *zone)
612 return zone->initialized;
615 static inline bool zone_is_empty(struct zone *zone)
617 return zone->spanned_pages == 0;
621 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
622 * intersection with the given zone
624 static inline bool zone_intersects(struct zone *zone,
625 unsigned long start_pfn, unsigned long nr_pages)
627 if (zone_is_empty(zone))
629 if (start_pfn >= zone_end_pfn(zone) ||
630 start_pfn + nr_pages <= zone->zone_start_pfn)
637 * The "priority" of VM scanning is how much of the queues we will scan in one
638 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
639 * queues ("queue_length >> 12") during an aging round.
641 #define DEF_PRIORITY 12
643 /* Maximum number of zones on a zonelist */
644 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
647 ZONELIST_FALLBACK, /* zonelist with fallback */
650 * The NUMA zonelists are doubled because we need zonelists that
651 * restrict the allocations to a single node for __GFP_THISNODE.
653 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
659 * This struct contains information about a zone in a zonelist. It is stored
660 * here to avoid dereferences into large structures and lookups of tables
663 struct zone *zone; /* Pointer to actual zone */
664 int zone_idx; /* zone_idx(zoneref->zone) */
668 * One allocation request operates on a zonelist. A zonelist
669 * is a list of zones, the first one is the 'goal' of the
670 * allocation, the other zones are fallback zones, in decreasing
673 * To speed the reading of the zonelist, the zonerefs contain the zone index
674 * of the entry being read. Helper functions to access information given
675 * a struct zoneref are
677 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
678 * zonelist_zone_idx() - Return the index of the zone for an entry
679 * zonelist_node_idx() - Return the index of the node for an entry
682 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
685 #ifndef CONFIG_DISCONTIGMEM
686 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
687 extern struct page *mem_map;
690 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
691 struct deferred_split {
692 spinlock_t split_queue_lock;
693 struct list_head split_queue;
694 unsigned long split_queue_len;
699 * On NUMA machines, each NUMA node would have a pg_data_t to describe
700 * it's memory layout. On UMA machines there is a single pglist_data which
701 * describes the whole memory.
703 * Memory statistics and page replacement data structures are maintained on a
706 typedef struct pglist_data {
708 * node_zones contains just the zones for THIS node. Not all of the
709 * zones may be populated, but it is the full list. It is referenced by
710 * this node's node_zonelists as well as other node's node_zonelists.
712 struct zone node_zones[MAX_NR_ZONES];
715 * node_zonelists contains references to all zones in all nodes.
716 * Generally the first zones will be references to this node's
719 struct zonelist node_zonelists[MAX_ZONELISTS];
721 int nr_zones; /* number of populated zones in this node */
722 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
723 struct page *node_mem_map;
724 #ifdef CONFIG_PAGE_EXTENSION
725 struct page_ext *node_page_ext;
728 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
730 * Must be held any time you expect node_start_pfn,
731 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
732 * Also synchronizes pgdat->first_deferred_pfn during deferred page
735 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
736 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
737 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
739 * Nests above zone->lock and zone->span_seqlock
741 spinlock_t node_size_lock;
743 unsigned long node_start_pfn;
744 unsigned long node_present_pages; /* total number of physical pages */
745 unsigned long node_spanned_pages; /* total size of physical page
746 range, including holes */
748 wait_queue_head_t kswapd_wait;
749 wait_queue_head_t pfmemalloc_wait;
750 struct task_struct *kswapd; /* Protected by
751 mem_hotplug_begin/end() */
753 enum zone_type kswapd_highest_zoneidx;
755 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
757 #ifdef CONFIG_COMPACTION
758 int kcompactd_max_order;
759 enum zone_type kcompactd_highest_zoneidx;
760 wait_queue_head_t kcompactd_wait;
761 struct task_struct *kcompactd;
764 * This is a per-node reserve of pages that are not available
765 * to userspace allocations.
767 unsigned long totalreserve_pages;
771 * node reclaim becomes active if more unmapped pages exist.
773 unsigned long min_unmapped_pages;
774 unsigned long min_slab_pages;
775 #endif /* CONFIG_NUMA */
777 /* Write-intensive fields used by page reclaim */
781 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
783 * If memory initialisation on large machines is deferred then this
784 * is the first PFN that needs to be initialised.
786 unsigned long first_deferred_pfn;
787 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
789 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
790 struct deferred_split deferred_split_queue;
793 /* Fields commonly accessed by the page reclaim scanner */
796 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
798 * Use mem_cgroup_lruvec() to look up lruvecs.
800 struct lruvec __lruvec;
806 /* Per-node vmstats */
807 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
808 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
811 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
812 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
813 #ifdef CONFIG_FLAT_NODE_MEM_MAP
814 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
816 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
818 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
820 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
821 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
823 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
825 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
828 static inline bool pgdat_is_empty(pg_data_t *pgdat)
830 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
833 #include <linux/memory_hotplug.h>
835 void build_all_zonelists(pg_data_t *pgdat);
836 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
837 enum zone_type highest_zoneidx);
838 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
839 int highest_zoneidx, unsigned int alloc_flags,
841 bool zone_watermark_ok(struct zone *z, unsigned int order,
842 unsigned long mark, int highest_zoneidx,
843 unsigned int alloc_flags);
844 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
845 unsigned long mark, int highest_zoneidx);
847 * Memory initialization context, use to differentiate memory added by
848 * the platform statically or via memory hotplug interface.
850 enum meminit_context {
855 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
858 extern void lruvec_init(struct lruvec *lruvec);
860 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
863 return lruvec->pgdat;
865 return container_of(lruvec, struct pglist_data, __lruvec);
869 extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
871 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
872 int local_memory_node(int node_id);
874 static inline int local_memory_node(int node_id) { return node_id; };
878 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
880 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
883 * Returns true if a zone has pages managed by the buddy allocator.
884 * All the reclaim decisions have to use this function rather than
885 * populated_zone(). If the whole zone is reserved then we can easily
886 * end up with populated_zone() && !managed_zone().
888 static inline bool managed_zone(struct zone *zone)
890 return zone_managed_pages(zone);
893 /* Returns true if a zone has memory */
894 static inline bool populated_zone(struct zone *zone)
896 return zone->present_pages;
900 static inline int zone_to_nid(struct zone *zone)
905 static inline void zone_set_nid(struct zone *zone, int nid)
910 static inline int zone_to_nid(struct zone *zone)
915 static inline void zone_set_nid(struct zone *zone, int nid) {}
918 extern int movable_zone;
920 #ifdef CONFIG_HIGHMEM
921 static inline int zone_movable_is_highmem(void)
923 #ifdef CONFIG_NEED_MULTIPLE_NODES
924 return movable_zone == ZONE_HIGHMEM;
926 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
931 static inline int is_highmem_idx(enum zone_type idx)
933 #ifdef CONFIG_HIGHMEM
934 return (idx == ZONE_HIGHMEM ||
935 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
942 * is_highmem - helper function to quickly check if a struct zone is a
943 * highmem zone or not. This is an attempt to keep references
944 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
945 * @zone - pointer to struct zone variable
947 static inline int is_highmem(struct zone *zone)
949 #ifdef CONFIG_HIGHMEM
950 return is_highmem_idx(zone_idx(zone));
956 /* These two functions are used to setup the per zone pages min values */
959 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, void *, size_t *,
961 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, void *,
963 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
964 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, void *,
966 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
967 void *, size_t *, loff_t *);
968 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
969 void *, size_t *, loff_t *);
970 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
971 void *, size_t *, loff_t *);
972 int numa_zonelist_order_handler(struct ctl_table *, int,
973 void *, size_t *, loff_t *);
974 extern int percpu_pagelist_fraction;
975 extern char numa_zonelist_order[];
976 #define NUMA_ZONELIST_ORDER_LEN 16
978 #ifndef CONFIG_NEED_MULTIPLE_NODES
980 extern struct pglist_data contig_page_data;
981 #define NODE_DATA(nid) (&contig_page_data)
982 #define NODE_MEM_MAP(nid) mem_map
984 #else /* CONFIG_NEED_MULTIPLE_NODES */
986 #include <asm/mmzone.h>
988 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
990 extern struct pglist_data *first_online_pgdat(void);
991 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
992 extern struct zone *next_zone(struct zone *zone);
995 * for_each_online_pgdat - helper macro to iterate over all online nodes
996 * @pgdat - pointer to a pg_data_t variable
998 #define for_each_online_pgdat(pgdat) \
999 for (pgdat = first_online_pgdat(); \
1001 pgdat = next_online_pgdat(pgdat))
1003 * for_each_zone - helper macro to iterate over all memory zones
1004 * @zone - pointer to struct zone variable
1006 * The user only needs to declare the zone variable, for_each_zone
1009 #define for_each_zone(zone) \
1010 for (zone = (first_online_pgdat())->node_zones; \
1012 zone = next_zone(zone))
1014 #define for_each_populated_zone(zone) \
1015 for (zone = (first_online_pgdat())->node_zones; \
1017 zone = next_zone(zone)) \
1018 if (!populated_zone(zone)) \
1019 ; /* do nothing */ \
1022 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1024 return zoneref->zone;
1027 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1029 return zoneref->zone_idx;
1032 static inline int zonelist_node_idx(struct zoneref *zoneref)
1034 return zone_to_nid(zoneref->zone);
1037 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1038 enum zone_type highest_zoneidx,
1042 * 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
1043 * @z - The cursor used as a starting point for the search
1044 * @highest_zoneidx - The zone index of the highest zone to return
1045 * @nodes - An optional nodemask to filter the zonelist with
1047 * This function returns the next zone at or below a given zone index that is
1048 * within the allowed nodemask using a cursor as the starting point for the
1049 * search. The zoneref returned is a cursor that represents the current zone
1050 * being examined. It should be advanced by one before calling
1051 * next_zones_zonelist again.
1053 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1054 enum zone_type highest_zoneidx,
1057 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1059 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1063 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1064 * @zonelist - The zonelist to search for a suitable zone
1065 * @highest_zoneidx - The zone index of the highest zone to return
1066 * @nodes - An optional nodemask to filter the zonelist with
1067 * @return - Zoneref pointer for the first suitable zone found (see below)
1069 * This function returns the first zone at or below a given zone index that is
1070 * within the allowed nodemask. The zoneref returned is a cursor that can be
1071 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1072 * one before calling.
1074 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1075 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1076 * update due to cpuset modification.
1078 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1079 enum zone_type highest_zoneidx,
1082 return next_zones_zonelist(zonelist->_zonerefs,
1083 highest_zoneidx, nodes);
1087 * 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
1088 * @zone - The current zone in the iterator
1089 * @z - The current pointer within zonelist->_zonerefs being iterated
1090 * @zlist - The zonelist being iterated
1091 * @highidx - The zone index of the highest zone to return
1092 * @nodemask - Nodemask allowed by the allocator
1094 * This iterator iterates though all zones at or below a given zone index and
1095 * within a given nodemask
1097 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1098 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1100 z = next_zones_zonelist(++z, highidx, nodemask), \
1101 zone = zonelist_zone(z))
1103 #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1104 for (zone = z->zone; \
1106 z = next_zones_zonelist(++z, highidx, nodemask), \
1107 zone = zonelist_zone(z))
1111 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1112 * @zone - The current zone in the iterator
1113 * @z - The current pointer within zonelist->zones being iterated
1114 * @zlist - The zonelist being iterated
1115 * @highidx - The zone index of the highest zone to return
1117 * This iterator iterates though all zones at or below a given zone index.
1119 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1120 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1122 #ifdef CONFIG_SPARSEMEM
1123 #include <asm/sparsemem.h>
1126 #ifdef CONFIG_FLATMEM
1127 #define pfn_to_nid(pfn) (0)
1130 #ifdef CONFIG_SPARSEMEM
1133 * SECTION_SHIFT #bits space required to store a section #
1135 * PA_SECTION_SHIFT physical address to/from section number
1136 * PFN_SECTION_SHIFT pfn to/from section number
1138 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1139 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1141 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1143 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1144 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1146 #define SECTION_BLOCKFLAGS_BITS \
1147 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1149 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1150 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1153 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1155 return pfn >> PFN_SECTION_SHIFT;
1157 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1159 return sec << PFN_SECTION_SHIFT;
1162 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1163 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1165 #define SUBSECTION_SHIFT 21
1166 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1168 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1169 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1170 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1172 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1173 #error Subsection size exceeds section size
1175 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1178 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1179 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1181 struct mem_section_usage {
1182 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1183 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1185 /* See declaration of similar field in struct zone */
1186 unsigned long pageblock_flags[0];
1189 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1193 struct mem_section {
1195 * This is, logically, a pointer to an array of struct
1196 * pages. However, it is stored with some other magic.
1197 * (see sparse.c::sparse_init_one_section())
1199 * Additionally during early boot we encode node id of
1200 * the location of the section here to guide allocation.
1201 * (see sparse.c::memory_present())
1203 * Making it a UL at least makes someone do a cast
1204 * before using it wrong.
1206 unsigned long section_mem_map;
1208 struct mem_section_usage *usage;
1209 #ifdef CONFIG_PAGE_EXTENSION
1211 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1212 * section. (see page_ext.h about this.)
1214 struct page_ext *page_ext;
1218 * WARNING: mem_section must be a power-of-2 in size for the
1219 * calculation and use of SECTION_ROOT_MASK to make sense.
1223 #ifdef CONFIG_SPARSEMEM_EXTREME
1224 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1226 #define SECTIONS_PER_ROOT 1
1229 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1230 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1231 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1233 #ifdef CONFIG_SPARSEMEM_EXTREME
1234 extern struct mem_section **mem_section;
1236 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1239 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1241 return ms->usage->pageblock_flags;
1244 static inline struct mem_section *__nr_to_section(unsigned long nr)
1246 #ifdef CONFIG_SPARSEMEM_EXTREME
1250 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1252 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1254 extern unsigned long __section_nr(struct mem_section *ms);
1255 extern size_t mem_section_usage_size(void);
1258 * We use the lower bits of the mem_map pointer to store
1259 * a little bit of information. The pointer is calculated
1260 * as mem_map - section_nr_to_pfn(pnum). The result is
1261 * aligned to the minimum alignment of the two values:
1262 * 1. All mem_map arrays are page-aligned.
1263 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1264 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1265 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1266 * worst combination is powerpc with 256k pages,
1267 * which results in PFN_SECTION_SHIFT equal 6.
1268 * To sum it up, at least 6 bits are available.
1270 #define SECTION_MARKED_PRESENT (1UL<<0)
1271 #define SECTION_HAS_MEM_MAP (1UL<<1)
1272 #define SECTION_IS_ONLINE (1UL<<2)
1273 #define SECTION_IS_EARLY (1UL<<3)
1274 #define SECTION_MAP_LAST_BIT (1UL<<4)
1275 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1276 #define SECTION_NID_SHIFT 3
1278 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1280 unsigned long map = section->section_mem_map;
1281 map &= SECTION_MAP_MASK;
1282 return (struct page *)map;
1285 static inline int present_section(struct mem_section *section)
1287 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1290 static inline int present_section_nr(unsigned long nr)
1292 return present_section(__nr_to_section(nr));
1295 static inline int valid_section(struct mem_section *section)
1297 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1300 static inline int early_section(struct mem_section *section)
1302 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1305 static inline int valid_section_nr(unsigned long nr)
1307 return valid_section(__nr_to_section(nr));
1310 static inline int online_section(struct mem_section *section)
1312 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1315 static inline int online_section_nr(unsigned long nr)
1317 return online_section(__nr_to_section(nr));
1320 #ifdef CONFIG_MEMORY_HOTPLUG
1321 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1322 #ifdef CONFIG_MEMORY_HOTREMOVE
1323 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1327 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1329 return __nr_to_section(pfn_to_section_nr(pfn));
1332 extern unsigned long __highest_present_section_nr;
1334 static inline int subsection_map_index(unsigned long pfn)
1336 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1339 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1340 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1342 int idx = subsection_map_index(pfn);
1344 return test_bit(idx, ms->usage->subsection_map);
1347 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1353 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1354 static inline int pfn_valid(unsigned long pfn)
1356 struct mem_section *ms;
1358 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1360 ms = __nr_to_section(pfn_to_section_nr(pfn));
1361 if (!valid_section(ms))
1364 * Traditionally early sections always returned pfn_valid() for
1365 * the entire section-sized span.
1367 return early_section(ms) || pfn_section_valid(ms, pfn);
1371 static inline int pfn_in_present_section(unsigned long pfn)
1373 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1375 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1378 static inline unsigned long next_present_section_nr(unsigned long section_nr)
1380 while (++section_nr <= __highest_present_section_nr) {
1381 if (present_section_nr(section_nr))
1389 * These are _only_ used during initialisation, therefore they
1390 * can use __initdata ... They could have names to indicate
1394 #define pfn_to_nid(pfn) \
1396 unsigned long __pfn_to_nid_pfn = (pfn); \
1397 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1400 #define pfn_to_nid(pfn) (0)
1403 void sparse_init(void);
1405 #define sparse_init() do {} while (0)
1406 #define sparse_index_init(_sec, _nid) do {} while (0)
1407 #define pfn_in_present_section pfn_valid
1408 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
1409 #endif /* CONFIG_SPARSEMEM */
1412 * During memory init memblocks map pfns to nids. The search is expensive and
1413 * this caches recent lookups. The implementation of __early_pfn_to_nid
1414 * may treat start/end as pfns or sections.
1416 struct mminit_pfnnid_cache {
1417 unsigned long last_start;
1418 unsigned long last_end;
1423 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1424 * need to check pfn validity within that MAX_ORDER_NR_PAGES block.
1425 * pfn_valid_within() should be used in this case; we optimise this away
1426 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1428 #ifdef CONFIG_HOLES_IN_ZONE
1429 #define pfn_valid_within(pfn) pfn_valid(pfn)
1431 #define pfn_valid_within(pfn) (1)
1434 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1436 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1437 * associated with it or not. This means that a struct page exists for this
1438 * pfn. The caller cannot assume the page is fully initialized in general.
1439 * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1440 * will ensure the struct page is fully online and initialized. Special pages
1441 * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1443 * In FLATMEM, it is expected that holes always have valid memmap as long as
1444 * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1445 * that a valid section has a memmap for the entire section.
1447 * However, an ARM, and maybe other embedded architectures in the future
1448 * free memmap backing holes to save memory on the assumption the memmap is
1449 * never used. The page_zone linkages are then broken even though pfn_valid()
1450 * returns true. A walker of the full memmap must then do this additional
1451 * check to ensure the memmap they are looking at is sane by making sure
1452 * the zone and PFN linkages are still valid. This is expensive, but walkers
1453 * of the full memmap are extremely rare.
1455 bool memmap_valid_within(unsigned long pfn,
1456 struct page *page, struct zone *zone);
1458 static inline bool memmap_valid_within(unsigned long pfn,
1459 struct page *page, struct zone *zone)
1463 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1465 #endif /* !__GENERATING_BOUNDS.H */
1466 #endif /* !__ASSEMBLY__ */
1467 #endif /* _LINUX_MMZONE_H */