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 * Returns true if the value is measured in bytes (most vmstat values are
211 * measured in pages). This defines the API part, the internal representation
212 * might be different.
214 static __always_inline bool vmstat_item_in_bytes(int idx)
220 * We do arithmetic on the LRU lists in various places in the code,
221 * so it is important to keep the active lists LRU_ACTIVE higher in
222 * the array than the corresponding inactive lists, and to keep
223 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
225 * This has to be kept in sync with the statistics in zone_stat_item
226 * above and the descriptions in vmstat_text in mm/vmstat.c
233 LRU_INACTIVE_ANON = LRU_BASE,
234 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
235 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
236 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
241 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
243 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
245 static inline bool is_file_lru(enum lru_list lru)
247 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
250 static inline bool is_active_lru(enum lru_list lru)
252 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
256 LRUVEC_CONGESTED, /* lruvec has many dirty pages
257 * backed by a congested BDI
262 struct list_head lists[NR_LRU_LISTS];
264 * These track the cost of reclaiming one LRU - file or anon -
265 * over the other. As the observed cost of reclaiming one LRU
266 * increases, the reclaim scan balance tips toward the other.
268 unsigned long anon_cost;
269 unsigned long file_cost;
270 /* Non-resident age, driven by LRU movement */
271 atomic_long_t nonresident_age;
272 /* Refaults at the time of last reclaim cycle */
273 unsigned long refaults;
274 /* Various lruvec state flags (enum lruvec_flags) */
277 struct pglist_data *pgdat;
281 /* Isolate unmapped pages */
282 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
283 /* Isolate for asynchronous migration */
284 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
285 /* Isolate unevictable pages */
286 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
288 /* LRU Isolation modes. */
289 typedef unsigned __bitwise isolate_mode_t;
291 enum zone_watermarks {
298 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
299 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
300 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
301 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
303 struct per_cpu_pages {
304 int count; /* number of pages in the list */
305 int high; /* high watermark, emptying needed */
306 int batch; /* chunk size for buddy add/remove */
308 /* Lists of pages, one per migrate type stored on the pcp-lists */
309 struct list_head lists[MIGRATE_PCPTYPES];
312 struct per_cpu_pageset {
313 struct per_cpu_pages pcp;
316 u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
320 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
324 struct per_cpu_nodestat {
326 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
329 #endif /* !__GENERATING_BOUNDS.H */
333 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
334 * to DMA to all of the addressable memory (ZONE_NORMAL).
335 * On architectures where this area covers the whole 32 bit address
336 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
337 * DMA addressing constraints. This distinction is important as a 32bit
338 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
339 * platforms may need both zones as they support peripherals with
340 * different DMA addressing limitations.
344 * - i386 and x86_64 have a fixed 16M ZONE_DMA and ZONE_DMA32 for the
345 * rest of the lower 4G.
347 * - arm only uses ZONE_DMA, the size, up to 4G, may vary depending on
348 * the specific device.
350 * - arm64 has a fixed 1G ZONE_DMA and ZONE_DMA32 for the rest of the
353 * - powerpc only uses ZONE_DMA, the size, up to 2G, may vary
354 * depending on the specific device.
356 * - s390 uses ZONE_DMA fixed to the lower 2G.
358 * - ia64 and riscv only use ZONE_DMA32.
360 * - parisc uses neither.
362 #ifdef CONFIG_ZONE_DMA
365 #ifdef CONFIG_ZONE_DMA32
369 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
370 * performed on pages in ZONE_NORMAL if the DMA devices support
371 * transfers to all addressable memory.
374 #ifdef CONFIG_HIGHMEM
376 * A memory area that is only addressable by the kernel through
377 * mapping portions into its own address space. This is for example
378 * used by i386 to allow the kernel to address the memory beyond
379 * 900MB. The kernel will set up special mappings (page
380 * table entries on i386) for each page that the kernel needs to
386 #ifdef CONFIG_ZONE_DEVICE
393 #ifndef __GENERATING_BOUNDS_H
396 /* Read-mostly fields */
398 /* zone watermarks, access with *_wmark_pages(zone) macros */
399 unsigned long _watermark[NR_WMARK];
400 unsigned long watermark_boost;
402 unsigned long nr_reserved_highatomic;
405 * We don't know if the memory that we're going to allocate will be
406 * freeable or/and it will be released eventually, so to avoid totally
407 * wasting several GB of ram we must reserve some of the lower zone
408 * memory (otherwise we risk to run OOM on the lower zones despite
409 * there being tons of freeable ram on the higher zones). This array is
410 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
413 long lowmem_reserve[MAX_NR_ZONES];
418 struct pglist_data *zone_pgdat;
419 struct per_cpu_pageset __percpu *pageset;
421 #ifndef CONFIG_SPARSEMEM
423 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
424 * In SPARSEMEM, this map is stored in struct mem_section
426 unsigned long *pageblock_flags;
427 #endif /* CONFIG_SPARSEMEM */
429 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
430 unsigned long zone_start_pfn;
433 * spanned_pages is the total pages spanned by the zone, including
434 * holes, which is calculated as:
435 * spanned_pages = zone_end_pfn - zone_start_pfn;
437 * present_pages is physical pages existing within the zone, which
439 * present_pages = spanned_pages - absent_pages(pages in holes);
441 * managed_pages is present pages managed by the buddy system, which
442 * is calculated as (reserved_pages includes pages allocated by the
443 * bootmem allocator):
444 * managed_pages = present_pages - reserved_pages;
446 * So present_pages may be used by memory hotplug or memory power
447 * management logic to figure out unmanaged pages by checking
448 * (present_pages - managed_pages). And managed_pages should be used
449 * by page allocator and vm scanner to calculate all kinds of watermarks
454 * zone_start_pfn and spanned_pages are protected by span_seqlock.
455 * It is a seqlock because it has to be read outside of zone->lock,
456 * and it is done in the main allocator path. But, it is written
457 * quite infrequently.
459 * The span_seq lock is declared along with zone->lock because it is
460 * frequently read in proximity to zone->lock. It's good to
461 * give them a chance of being in the same cacheline.
463 * Write access to present_pages at runtime should be protected by
464 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
465 * present_pages should get_online_mems() to get a stable value.
467 atomic_long_t managed_pages;
468 unsigned long spanned_pages;
469 unsigned long present_pages;
473 #ifdef CONFIG_MEMORY_ISOLATION
475 * Number of isolated pageblock. It is used to solve incorrect
476 * freepage counting problem due to racy retrieving migratetype
477 * of pageblock. Protected by zone->lock.
479 unsigned long nr_isolate_pageblock;
482 #ifdef CONFIG_MEMORY_HOTPLUG
483 /* see spanned/present_pages for more description */
484 seqlock_t span_seqlock;
489 /* Write-intensive fields used from the page allocator */
492 /* free areas of different sizes */
493 struct free_area free_area[MAX_ORDER];
495 /* zone flags, see below */
498 /* Primarily protects free_area */
501 /* Write-intensive fields used by compaction and vmstats. */
505 * When free pages are below this point, additional steps are taken
506 * when reading the number of free pages to avoid per-cpu counter
507 * drift allowing watermarks to be breached
509 unsigned long percpu_drift_mark;
511 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
512 /* pfn where compaction free scanner should start */
513 unsigned long compact_cached_free_pfn;
514 /* pfn where async and sync compaction migration scanner should start */
515 unsigned long compact_cached_migrate_pfn[2];
516 unsigned long compact_init_migrate_pfn;
517 unsigned long compact_init_free_pfn;
520 #ifdef CONFIG_COMPACTION
522 * On compaction failure, 1<<compact_defer_shift compactions
523 * are skipped before trying again. The number attempted since
524 * last failure is tracked with compact_considered.
526 unsigned int compact_considered;
527 unsigned int compact_defer_shift;
528 int compact_order_failed;
531 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
532 /* Set to true when the PG_migrate_skip bits should be cleared */
533 bool compact_blockskip_flush;
539 /* Zone statistics */
540 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
541 atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
542 } ____cacheline_internodealigned_in_smp;
545 PGDAT_DIRTY, /* reclaim scanning has recently found
546 * many dirty file pages at the tail
549 PGDAT_WRITEBACK, /* reclaim scanning has recently found
550 * many pages under writeback
552 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
556 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
557 * Cleared when kswapd is woken.
561 static inline unsigned long zone_managed_pages(struct zone *zone)
563 return (unsigned long)atomic_long_read(&zone->managed_pages);
566 static inline unsigned long zone_end_pfn(const struct zone *zone)
568 return zone->zone_start_pfn + zone->spanned_pages;
571 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
573 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
576 static inline bool zone_is_initialized(struct zone *zone)
578 return zone->initialized;
581 static inline bool zone_is_empty(struct zone *zone)
583 return zone->spanned_pages == 0;
587 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
588 * intersection with the given zone
590 static inline bool zone_intersects(struct zone *zone,
591 unsigned long start_pfn, unsigned long nr_pages)
593 if (zone_is_empty(zone))
595 if (start_pfn >= zone_end_pfn(zone) ||
596 start_pfn + nr_pages <= zone->zone_start_pfn)
603 * The "priority" of VM scanning is how much of the queues we will scan in one
604 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
605 * queues ("queue_length >> 12") during an aging round.
607 #define DEF_PRIORITY 12
609 /* Maximum number of zones on a zonelist */
610 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
613 ZONELIST_FALLBACK, /* zonelist with fallback */
616 * The NUMA zonelists are doubled because we need zonelists that
617 * restrict the allocations to a single node for __GFP_THISNODE.
619 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
625 * This struct contains information about a zone in a zonelist. It is stored
626 * here to avoid dereferences into large structures and lookups of tables
629 struct zone *zone; /* Pointer to actual zone */
630 int zone_idx; /* zone_idx(zoneref->zone) */
634 * One allocation request operates on a zonelist. A zonelist
635 * is a list of zones, the first one is the 'goal' of the
636 * allocation, the other zones are fallback zones, in decreasing
639 * To speed the reading of the zonelist, the zonerefs contain the zone index
640 * of the entry being read. Helper functions to access information given
641 * a struct zoneref are
643 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
644 * zonelist_zone_idx() - Return the index of the zone for an entry
645 * zonelist_node_idx() - Return the index of the node for an entry
648 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
651 #ifndef CONFIG_DISCONTIGMEM
652 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
653 extern struct page *mem_map;
656 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
657 struct deferred_split {
658 spinlock_t split_queue_lock;
659 struct list_head split_queue;
660 unsigned long split_queue_len;
665 * On NUMA machines, each NUMA node would have a pg_data_t to describe
666 * it's memory layout. On UMA machines there is a single pglist_data which
667 * describes the whole memory.
669 * Memory statistics and page replacement data structures are maintained on a
672 typedef struct pglist_data {
674 * node_zones contains just the zones for THIS node. Not all of the
675 * zones may be populated, but it is the full list. It is referenced by
676 * this node's node_zonelists as well as other node's node_zonelists.
678 struct zone node_zones[MAX_NR_ZONES];
681 * node_zonelists contains references to all zones in all nodes.
682 * Generally the first zones will be references to this node's
685 struct zonelist node_zonelists[MAX_ZONELISTS];
687 int nr_zones; /* number of populated zones in this node */
688 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
689 struct page *node_mem_map;
690 #ifdef CONFIG_PAGE_EXTENSION
691 struct page_ext *node_page_ext;
694 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
696 * Must be held any time you expect node_start_pfn,
697 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
698 * Also synchronizes pgdat->first_deferred_pfn during deferred page
701 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
702 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
703 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
705 * Nests above zone->lock and zone->span_seqlock
707 spinlock_t node_size_lock;
709 unsigned long node_start_pfn;
710 unsigned long node_present_pages; /* total number of physical pages */
711 unsigned long node_spanned_pages; /* total size of physical page
712 range, including holes */
714 wait_queue_head_t kswapd_wait;
715 wait_queue_head_t pfmemalloc_wait;
716 struct task_struct *kswapd; /* Protected by
717 mem_hotplug_begin/end() */
719 enum zone_type kswapd_highest_zoneidx;
721 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
723 #ifdef CONFIG_COMPACTION
724 int kcompactd_max_order;
725 enum zone_type kcompactd_highest_zoneidx;
726 wait_queue_head_t kcompactd_wait;
727 struct task_struct *kcompactd;
730 * This is a per-node reserve of pages that are not available
731 * to userspace allocations.
733 unsigned long totalreserve_pages;
737 * node reclaim becomes active if more unmapped pages exist.
739 unsigned long min_unmapped_pages;
740 unsigned long min_slab_pages;
741 #endif /* CONFIG_NUMA */
743 /* Write-intensive fields used by page reclaim */
747 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
749 * If memory initialisation on large machines is deferred then this
750 * is the first PFN that needs to be initialised.
752 unsigned long first_deferred_pfn;
753 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
755 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
756 struct deferred_split deferred_split_queue;
759 /* Fields commonly accessed by the page reclaim scanner */
762 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
764 * Use mem_cgroup_lruvec() to look up lruvecs.
766 struct lruvec __lruvec;
772 /* Per-node vmstats */
773 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
774 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
777 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
778 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
779 #ifdef CONFIG_FLAT_NODE_MEM_MAP
780 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
782 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
784 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
786 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
787 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
789 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
791 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
794 static inline bool pgdat_is_empty(pg_data_t *pgdat)
796 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
799 #include <linux/memory_hotplug.h>
801 void build_all_zonelists(pg_data_t *pgdat);
802 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
803 enum zone_type highest_zoneidx);
804 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
805 int highest_zoneidx, unsigned int alloc_flags,
807 bool zone_watermark_ok(struct zone *z, unsigned int order,
808 unsigned long mark, int highest_zoneidx,
809 unsigned int alloc_flags);
810 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
811 unsigned long mark, int highest_zoneidx);
812 enum memmap_context {
816 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
819 extern void lruvec_init(struct lruvec *lruvec);
821 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
824 return lruvec->pgdat;
826 return container_of(lruvec, struct pglist_data, __lruvec);
830 extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
832 #ifdef CONFIG_HAVE_MEMORY_PRESENT
833 void memory_present(int nid, unsigned long start, unsigned long end);
835 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
838 #if defined(CONFIG_SPARSEMEM)
839 void memblocks_present(void);
841 static inline void memblocks_present(void) {}
844 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
845 int local_memory_node(int node_id);
847 static inline int local_memory_node(int node_id) { return node_id; };
851 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
853 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
856 * Returns true if a zone has pages managed by the buddy allocator.
857 * All the reclaim decisions have to use this function rather than
858 * populated_zone(). If the whole zone is reserved then we can easily
859 * end up with populated_zone() && !managed_zone().
861 static inline bool managed_zone(struct zone *zone)
863 return zone_managed_pages(zone);
866 /* Returns true if a zone has memory */
867 static inline bool populated_zone(struct zone *zone)
869 return zone->present_pages;
873 static inline int zone_to_nid(struct zone *zone)
878 static inline void zone_set_nid(struct zone *zone, int nid)
883 static inline int zone_to_nid(struct zone *zone)
888 static inline void zone_set_nid(struct zone *zone, int nid) {}
891 extern int movable_zone;
893 #ifdef CONFIG_HIGHMEM
894 static inline int zone_movable_is_highmem(void)
896 #ifdef CONFIG_NEED_MULTIPLE_NODES
897 return movable_zone == ZONE_HIGHMEM;
899 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
904 static inline int is_highmem_idx(enum zone_type idx)
906 #ifdef CONFIG_HIGHMEM
907 return (idx == ZONE_HIGHMEM ||
908 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
915 * is_highmem - helper function to quickly check if a struct zone is a
916 * highmem zone or not. This is an attempt to keep references
917 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
918 * @zone - pointer to struct zone variable
920 static inline int is_highmem(struct zone *zone)
922 #ifdef CONFIG_HIGHMEM
923 return is_highmem_idx(zone_idx(zone));
929 /* These two functions are used to setup the per zone pages min values */
932 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, void *, size_t *,
934 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, void *,
936 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
937 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, void *,
939 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
940 void *, size_t *, loff_t *);
941 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
942 void *, size_t *, loff_t *);
943 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
944 void *, size_t *, loff_t *);
945 int numa_zonelist_order_handler(struct ctl_table *, int,
946 void *, size_t *, loff_t *);
947 extern int percpu_pagelist_fraction;
948 extern char numa_zonelist_order[];
949 #define NUMA_ZONELIST_ORDER_LEN 16
951 #ifndef CONFIG_NEED_MULTIPLE_NODES
953 extern struct pglist_data contig_page_data;
954 #define NODE_DATA(nid) (&contig_page_data)
955 #define NODE_MEM_MAP(nid) mem_map
957 #else /* CONFIG_NEED_MULTIPLE_NODES */
959 #include <asm/mmzone.h>
961 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
963 extern struct pglist_data *first_online_pgdat(void);
964 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
965 extern struct zone *next_zone(struct zone *zone);
968 * for_each_online_pgdat - helper macro to iterate over all online nodes
969 * @pgdat - pointer to a pg_data_t variable
971 #define for_each_online_pgdat(pgdat) \
972 for (pgdat = first_online_pgdat(); \
974 pgdat = next_online_pgdat(pgdat))
976 * for_each_zone - helper macro to iterate over all memory zones
977 * @zone - pointer to struct zone variable
979 * The user only needs to declare the zone variable, for_each_zone
982 #define for_each_zone(zone) \
983 for (zone = (first_online_pgdat())->node_zones; \
985 zone = next_zone(zone))
987 #define for_each_populated_zone(zone) \
988 for (zone = (first_online_pgdat())->node_zones; \
990 zone = next_zone(zone)) \
991 if (!populated_zone(zone)) \
995 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
997 return zoneref->zone;
1000 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1002 return zoneref->zone_idx;
1005 static inline int zonelist_node_idx(struct zoneref *zoneref)
1007 return zone_to_nid(zoneref->zone);
1010 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1011 enum zone_type highest_zoneidx,
1015 * 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
1016 * @z - The cursor used as a starting point for the search
1017 * @highest_zoneidx - The zone index of the highest zone to return
1018 * @nodes - An optional nodemask to filter the zonelist with
1020 * This function returns the next zone at or below a given zone index that is
1021 * within the allowed nodemask using a cursor as the starting point for the
1022 * search. The zoneref returned is a cursor that represents the current zone
1023 * being examined. It should be advanced by one before calling
1024 * next_zones_zonelist again.
1026 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1027 enum zone_type highest_zoneidx,
1030 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1032 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1036 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1037 * @zonelist - The zonelist to search for a suitable zone
1038 * @highest_zoneidx - The zone index of the highest zone to return
1039 * @nodes - An optional nodemask to filter the zonelist with
1040 * @return - Zoneref pointer for the first suitable zone found (see below)
1042 * This function returns the first zone at or below a given zone index that is
1043 * within the allowed nodemask. The zoneref returned is a cursor that can be
1044 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1045 * one before calling.
1047 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1048 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1049 * update due to cpuset modification.
1051 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1052 enum zone_type highest_zoneidx,
1055 return next_zones_zonelist(zonelist->_zonerefs,
1056 highest_zoneidx, nodes);
1060 * 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
1061 * @zone - The current zone in the iterator
1062 * @z - The current pointer within zonelist->_zonerefs being iterated
1063 * @zlist - The zonelist being iterated
1064 * @highidx - The zone index of the highest zone to return
1065 * @nodemask - Nodemask allowed by the allocator
1067 * This iterator iterates though all zones at or below a given zone index and
1068 * within a given nodemask
1070 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1071 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1073 z = next_zones_zonelist(++z, highidx, nodemask), \
1074 zone = zonelist_zone(z))
1076 #define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1077 for (zone = z->zone; \
1079 z = next_zones_zonelist(++z, highidx, nodemask), \
1080 zone = zonelist_zone(z))
1084 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1085 * @zone - The current zone in the iterator
1086 * @z - The current pointer within zonelist->zones being iterated
1087 * @zlist - The zonelist being iterated
1088 * @highidx - The zone index of the highest zone to return
1090 * This iterator iterates though all zones at or below a given zone index.
1092 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1093 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1095 #ifdef CONFIG_SPARSEMEM
1096 #include <asm/sparsemem.h>
1099 #ifdef CONFIG_FLATMEM
1100 #define pfn_to_nid(pfn) (0)
1103 #ifdef CONFIG_SPARSEMEM
1106 * SECTION_SHIFT #bits space required to store a section #
1108 * PA_SECTION_SHIFT physical address to/from section number
1109 * PFN_SECTION_SHIFT pfn to/from section number
1111 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1112 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1114 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1116 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1117 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1119 #define SECTION_BLOCKFLAGS_BITS \
1120 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1122 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1123 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1126 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1128 return pfn >> PFN_SECTION_SHIFT;
1130 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1132 return sec << PFN_SECTION_SHIFT;
1135 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1136 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1138 #define SUBSECTION_SHIFT 21
1139 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1141 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1142 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1143 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1145 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1146 #error Subsection size exceeds section size
1148 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1151 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1152 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1154 struct mem_section_usage {
1155 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1156 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1158 /* See declaration of similar field in struct zone */
1159 unsigned long pageblock_flags[0];
1162 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1166 struct mem_section {
1168 * This is, logically, a pointer to an array of struct
1169 * pages. However, it is stored with some other magic.
1170 * (see sparse.c::sparse_init_one_section())
1172 * Additionally during early boot we encode node id of
1173 * the location of the section here to guide allocation.
1174 * (see sparse.c::memory_present())
1176 * Making it a UL at least makes someone do a cast
1177 * before using it wrong.
1179 unsigned long section_mem_map;
1181 struct mem_section_usage *usage;
1182 #ifdef CONFIG_PAGE_EXTENSION
1184 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1185 * section. (see page_ext.h about this.)
1187 struct page_ext *page_ext;
1191 * WARNING: mem_section must be a power-of-2 in size for the
1192 * calculation and use of SECTION_ROOT_MASK to make sense.
1196 #ifdef CONFIG_SPARSEMEM_EXTREME
1197 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1199 #define SECTIONS_PER_ROOT 1
1202 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1203 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1204 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1206 #ifdef CONFIG_SPARSEMEM_EXTREME
1207 extern struct mem_section **mem_section;
1209 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1212 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1214 return ms->usage->pageblock_flags;
1217 static inline struct mem_section *__nr_to_section(unsigned long nr)
1219 #ifdef CONFIG_SPARSEMEM_EXTREME
1223 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1225 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1227 extern unsigned long __section_nr(struct mem_section *ms);
1228 extern size_t mem_section_usage_size(void);
1231 * We use the lower bits of the mem_map pointer to store
1232 * a little bit of information. The pointer is calculated
1233 * as mem_map - section_nr_to_pfn(pnum). The result is
1234 * aligned to the minimum alignment of the two values:
1235 * 1. All mem_map arrays are page-aligned.
1236 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1237 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1238 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1239 * worst combination is powerpc with 256k pages,
1240 * which results in PFN_SECTION_SHIFT equal 6.
1241 * To sum it up, at least 6 bits are available.
1243 #define SECTION_MARKED_PRESENT (1UL<<0)
1244 #define SECTION_HAS_MEM_MAP (1UL<<1)
1245 #define SECTION_IS_ONLINE (1UL<<2)
1246 #define SECTION_IS_EARLY (1UL<<3)
1247 #define SECTION_MAP_LAST_BIT (1UL<<4)
1248 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1249 #define SECTION_NID_SHIFT 3
1251 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1253 unsigned long map = section->section_mem_map;
1254 map &= SECTION_MAP_MASK;
1255 return (struct page *)map;
1258 static inline int present_section(struct mem_section *section)
1260 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1263 static inline int present_section_nr(unsigned long nr)
1265 return present_section(__nr_to_section(nr));
1268 static inline int valid_section(struct mem_section *section)
1270 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1273 static inline int early_section(struct mem_section *section)
1275 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1278 static inline int valid_section_nr(unsigned long nr)
1280 return valid_section(__nr_to_section(nr));
1283 static inline int online_section(struct mem_section *section)
1285 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1288 static inline int online_section_nr(unsigned long nr)
1290 return online_section(__nr_to_section(nr));
1293 #ifdef CONFIG_MEMORY_HOTPLUG
1294 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1295 #ifdef CONFIG_MEMORY_HOTREMOVE
1296 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1300 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1302 return __nr_to_section(pfn_to_section_nr(pfn));
1305 extern unsigned long __highest_present_section_nr;
1307 static inline int subsection_map_index(unsigned long pfn)
1309 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1312 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1313 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1315 int idx = subsection_map_index(pfn);
1317 return test_bit(idx, ms->usage->subsection_map);
1320 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1326 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1327 static inline int pfn_valid(unsigned long pfn)
1329 struct mem_section *ms;
1331 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1333 ms = __nr_to_section(pfn_to_section_nr(pfn));
1334 if (!valid_section(ms))
1337 * Traditionally early sections always returned pfn_valid() for
1338 * the entire section-sized span.
1340 return early_section(ms) || pfn_section_valid(ms, pfn);
1344 static inline int pfn_in_present_section(unsigned long pfn)
1346 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1348 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1351 static inline unsigned long next_present_section_nr(unsigned long section_nr)
1353 while (++section_nr <= __highest_present_section_nr) {
1354 if (present_section_nr(section_nr))
1362 * These are _only_ used during initialisation, therefore they
1363 * can use __initdata ... They could have names to indicate
1367 #define pfn_to_nid(pfn) \
1369 unsigned long __pfn_to_nid_pfn = (pfn); \
1370 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1373 #define pfn_to_nid(pfn) (0)
1376 #define early_pfn_valid(pfn) pfn_valid(pfn)
1377 void sparse_init(void);
1379 #define sparse_init() do {} while (0)
1380 #define sparse_index_init(_sec, _nid) do {} while (0)
1381 #define pfn_in_present_section pfn_valid
1382 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
1383 #endif /* CONFIG_SPARSEMEM */
1386 * During memory init memblocks map pfns to nids. The search is expensive and
1387 * this caches recent lookups. The implementation of __early_pfn_to_nid
1388 * may treat start/end as pfns or sections.
1390 struct mminit_pfnnid_cache {
1391 unsigned long last_start;
1392 unsigned long last_end;
1396 #ifndef early_pfn_valid
1397 #define early_pfn_valid(pfn) (1)
1400 void memory_present(int nid, unsigned long start, unsigned long end);
1403 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1404 * need to check pfn validity within that MAX_ORDER_NR_PAGES block.
1405 * pfn_valid_within() should be used in this case; we optimise this away
1406 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1408 #ifdef CONFIG_HOLES_IN_ZONE
1409 #define pfn_valid_within(pfn) pfn_valid(pfn)
1411 #define pfn_valid_within(pfn) (1)
1414 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1416 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1417 * associated with it or not. This means that a struct page exists for this
1418 * pfn. The caller cannot assume the page is fully initialized in general.
1419 * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1420 * will ensure the struct page is fully online and initialized. Special pages
1421 * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1423 * In FLATMEM, it is expected that holes always have valid memmap as long as
1424 * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1425 * that a valid section has a memmap for the entire section.
1427 * However, an ARM, and maybe other embedded architectures in the future
1428 * free memmap backing holes to save memory on the assumption the memmap is
1429 * never used. The page_zone linkages are then broken even though pfn_valid()
1430 * returns true. A walker of the full memmap must then do this additional
1431 * check to ensure the memmap they are looking at is sane by making sure
1432 * the zone and PFN linkages are still valid. This is expensive, but walkers
1433 * of the full memmap are extremely rare.
1435 bool memmap_valid_within(unsigned long pfn,
1436 struct page *page, struct zone *zone);
1438 static inline bool memmap_valid_within(unsigned long pfn,
1439 struct page *page, struct zone *zone)
1443 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1445 #endif /* !__GENERATING_BOUNDS.H */
1446 #endif /* !__ASSEMBLY__ */
1447 #endif /* _LINUX_MMZONE_H */