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 /* Used for pages not on another list */
104 static inline void add_to_free_area(struct page *page, struct free_area *area,
107 list_add(&page->lru, &area->free_list[migratetype]);
111 /* Used for pages not on another list */
112 static inline void add_to_free_area_tail(struct page *page, struct free_area *area,
115 list_add_tail(&page->lru, &area->free_list[migratetype]);
119 #ifdef CONFIG_SHUFFLE_PAGE_ALLOCATOR
120 /* Used to preserve page allocation order entropy */
121 void add_to_free_area_random(struct page *page, struct free_area *area,
124 static inline void add_to_free_area_random(struct page *page,
125 struct free_area *area, int migratetype)
127 add_to_free_area(page, area, migratetype);
131 /* Used for pages which are on another list */
132 static inline void move_to_free_area(struct page *page, struct free_area *area,
135 list_move(&page->lru, &area->free_list[migratetype]);
138 static inline struct page *get_page_from_free_area(struct free_area *area,
141 return list_first_entry_or_null(&area->free_list[migratetype],
145 static inline void del_page_from_free_area(struct page *page,
146 struct free_area *area)
148 list_del(&page->lru);
149 __ClearPageBuddy(page);
150 set_page_private(page, 0);
154 static inline bool free_area_empty(struct free_area *area, int migratetype)
156 return list_empty(&area->free_list[migratetype]);
162 * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
163 * So add a wild amount of padding here to ensure that they fall into separate
164 * cachelines. There are very few zone structures in the machine, so space
165 * consumption is not a concern here.
167 #if defined(CONFIG_SMP)
168 struct zone_padding {
170 } ____cacheline_internodealigned_in_smp;
171 #define ZONE_PADDING(name) struct zone_padding name;
173 #define ZONE_PADDING(name)
177 enum numa_stat_item {
178 NUMA_HIT, /* allocated in intended node */
179 NUMA_MISS, /* allocated in non intended node */
180 NUMA_FOREIGN, /* was intended here, hit elsewhere */
181 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
182 NUMA_LOCAL, /* allocation from local node */
183 NUMA_OTHER, /* allocation from other node */
184 NR_VM_NUMA_STAT_ITEMS
187 #define NR_VM_NUMA_STAT_ITEMS 0
190 enum zone_stat_item {
191 /* First 128 byte cacheline (assuming 64 bit words) */
193 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
194 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
196 NR_ZONE_INACTIVE_FILE,
199 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
200 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
201 NR_PAGETABLE, /* used for pagetables */
202 NR_KERNEL_STACK_KB, /* measured in KiB */
203 /* Second 128 byte cacheline */
205 #if IS_ENABLED(CONFIG_ZSMALLOC)
206 NR_ZSPAGES, /* allocated in zsmalloc */
209 NR_VM_ZONE_STAT_ITEMS };
211 enum node_stat_item {
213 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
214 NR_ACTIVE_ANON, /* " " " " " */
215 NR_INACTIVE_FILE, /* " " " " " */
216 NR_ACTIVE_FILE, /* " " " " " */
217 NR_UNEVICTABLE, /* " " " " " */
218 NR_SLAB_RECLAIMABLE, /* Please do not reorder this item */
219 NR_SLAB_UNRECLAIMABLE, /* and this one without looking at
220 * memcg_flush_percpu_vmstats() first. */
221 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
222 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
227 WORKINGSET_NODERECLAIM,
228 NR_ANON_MAPPED, /* Mapped anonymous pages */
229 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
230 only modified from process context */
234 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
235 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
241 NR_UNSTABLE_NFS, /* NFS unstable pages */
243 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
244 NR_DIRTIED, /* page dirtyings since bootup */
245 NR_WRITTEN, /* page writings since bootup */
246 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
247 NR_VM_NODE_STAT_ITEMS
251 * We do arithmetic on the LRU lists in various places in the code,
252 * so it is important to keep the active lists LRU_ACTIVE higher in
253 * the array than the corresponding inactive lists, and to keep
254 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
256 * This has to be kept in sync with the statistics in zone_stat_item
257 * above and the descriptions in vmstat_text in mm/vmstat.c
264 LRU_INACTIVE_ANON = LRU_BASE,
265 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
266 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
267 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
272 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
274 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
276 static inline bool is_file_lru(enum lru_list lru)
278 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
281 static inline bool is_active_lru(enum lru_list lru)
283 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
286 struct zone_reclaim_stat {
288 * The pageout code in vmscan.c keeps track of how many of the
289 * mem/swap backed and file backed pages are referenced.
290 * The higher the rotated/scanned ratio, the more valuable
293 * The anon LRU stats live in [0], file LRU stats in [1]
295 unsigned long recent_rotated[2];
296 unsigned long recent_scanned[2];
300 LRUVEC_CONGESTED, /* lruvec has many dirty pages
301 * backed by a congested BDI
306 struct list_head lists[NR_LRU_LISTS];
307 struct zone_reclaim_stat reclaim_stat;
308 /* Evictions & activations on the inactive file list */
309 atomic_long_t inactive_age;
310 /* Refaults at the time of last reclaim cycle */
311 unsigned long refaults;
312 /* Various lruvec state flags (enum lruvec_flags) */
315 struct pglist_data *pgdat;
319 /* Isolate unmapped pages */
320 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
321 /* Isolate for asynchronous migration */
322 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
323 /* Isolate unevictable pages */
324 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
326 /* LRU Isolation modes. */
327 typedef unsigned __bitwise isolate_mode_t;
329 enum zone_watermarks {
336 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
337 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
338 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
339 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
341 struct per_cpu_pages {
342 int count; /* number of pages in the list */
343 int high; /* high watermark, emptying needed */
344 int batch; /* chunk size for buddy add/remove */
346 /* Lists of pages, one per migrate type stored on the pcp-lists */
347 struct list_head lists[MIGRATE_PCPTYPES];
350 struct per_cpu_pageset {
351 struct per_cpu_pages pcp;
354 u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
358 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
362 struct per_cpu_nodestat {
364 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
367 #endif /* !__GENERATING_BOUNDS.H */
371 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
372 * to DMA to all of the addressable memory (ZONE_NORMAL).
373 * On architectures where this area covers the whole 32 bit address
374 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
375 * DMA addressing constraints. This distinction is important as a 32bit
376 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
377 * platforms may need both zones as they support peripherals with
378 * different DMA addressing limitations.
382 * - i386 and x86_64 have a fixed 16M ZONE_DMA and ZONE_DMA32 for the
383 * rest of the lower 4G.
385 * - arm only uses ZONE_DMA, the size, up to 4G, may vary depending on
386 * the specific device.
388 * - arm64 has a fixed 1G ZONE_DMA and ZONE_DMA32 for the rest of the
391 * - powerpc only uses ZONE_DMA, the size, up to 2G, may vary
392 * depending on the specific device.
394 * - s390 uses ZONE_DMA fixed to the lower 2G.
396 * - ia64 and riscv only use ZONE_DMA32.
398 * - parisc uses neither.
400 #ifdef CONFIG_ZONE_DMA
403 #ifdef CONFIG_ZONE_DMA32
407 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
408 * performed on pages in ZONE_NORMAL if the DMA devices support
409 * transfers to all addressable memory.
412 #ifdef CONFIG_HIGHMEM
414 * A memory area that is only addressable by the kernel through
415 * mapping portions into its own address space. This is for example
416 * used by i386 to allow the kernel to address the memory beyond
417 * 900MB. The kernel will set up special mappings (page
418 * table entries on i386) for each page that the kernel needs to
424 #ifdef CONFIG_ZONE_DEVICE
431 #ifndef __GENERATING_BOUNDS_H
434 /* Read-mostly fields */
436 /* zone watermarks, access with *_wmark_pages(zone) macros */
437 unsigned long _watermark[NR_WMARK];
438 unsigned long watermark_boost;
440 unsigned long nr_reserved_highatomic;
443 * We don't know if the memory that we're going to allocate will be
444 * freeable or/and it will be released eventually, so to avoid totally
445 * wasting several GB of ram we must reserve some of the lower zone
446 * memory (otherwise we risk to run OOM on the lower zones despite
447 * there being tons of freeable ram on the higher zones). This array is
448 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
451 long lowmem_reserve[MAX_NR_ZONES];
456 struct pglist_data *zone_pgdat;
457 struct per_cpu_pageset __percpu *pageset;
459 #ifndef CONFIG_SPARSEMEM
461 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
462 * In SPARSEMEM, this map is stored in struct mem_section
464 unsigned long *pageblock_flags;
465 #endif /* CONFIG_SPARSEMEM */
467 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
468 unsigned long zone_start_pfn;
471 * spanned_pages is the total pages spanned by the zone, including
472 * holes, which is calculated as:
473 * spanned_pages = zone_end_pfn - zone_start_pfn;
475 * present_pages is physical pages existing within the zone, which
477 * present_pages = spanned_pages - absent_pages(pages in holes);
479 * managed_pages is present pages managed by the buddy system, which
480 * is calculated as (reserved_pages includes pages allocated by the
481 * bootmem allocator):
482 * managed_pages = present_pages - reserved_pages;
484 * So present_pages may be used by memory hotplug or memory power
485 * management logic to figure out unmanaged pages by checking
486 * (present_pages - managed_pages). And managed_pages should be used
487 * by page allocator and vm scanner to calculate all kinds of watermarks
492 * zone_start_pfn and spanned_pages are protected by span_seqlock.
493 * It is a seqlock because it has to be read outside of zone->lock,
494 * and it is done in the main allocator path. But, it is written
495 * quite infrequently.
497 * The span_seq lock is declared along with zone->lock because it is
498 * frequently read in proximity to zone->lock. It's good to
499 * give them a chance of being in the same cacheline.
501 * Write access to present_pages at runtime should be protected by
502 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
503 * present_pages should get_online_mems() to get a stable value.
505 atomic_long_t managed_pages;
506 unsigned long spanned_pages;
507 unsigned long present_pages;
511 #ifdef CONFIG_MEMORY_ISOLATION
513 * Number of isolated pageblock. It is used to solve incorrect
514 * freepage counting problem due to racy retrieving migratetype
515 * of pageblock. Protected by zone->lock.
517 unsigned long nr_isolate_pageblock;
520 #ifdef CONFIG_MEMORY_HOTPLUG
521 /* see spanned/present_pages for more description */
522 seqlock_t span_seqlock;
527 /* Write-intensive fields used from the page allocator */
530 /* free areas of different sizes */
531 struct free_area free_area[MAX_ORDER];
533 /* zone flags, see below */
536 /* Primarily protects free_area */
539 /* Write-intensive fields used by compaction and vmstats. */
543 * When free pages are below this point, additional steps are taken
544 * when reading the number of free pages to avoid per-cpu counter
545 * drift allowing watermarks to be breached
547 unsigned long percpu_drift_mark;
549 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
550 /* pfn where compaction free scanner should start */
551 unsigned long compact_cached_free_pfn;
552 /* pfn where async and sync compaction migration scanner should start */
553 unsigned long compact_cached_migrate_pfn[2];
554 unsigned long compact_init_migrate_pfn;
555 unsigned long compact_init_free_pfn;
558 #ifdef CONFIG_COMPACTION
560 * On compaction failure, 1<<compact_defer_shift compactions
561 * are skipped before trying again. The number attempted since
562 * last failure is tracked with compact_considered.
564 unsigned int compact_considered;
565 unsigned int compact_defer_shift;
566 int compact_order_failed;
569 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
570 /* Set to true when the PG_migrate_skip bits should be cleared */
571 bool compact_blockskip_flush;
577 /* Zone statistics */
578 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
579 atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
580 } ____cacheline_internodealigned_in_smp;
583 PGDAT_DIRTY, /* reclaim scanning has recently found
584 * many dirty file pages at the tail
587 PGDAT_WRITEBACK, /* reclaim scanning has recently found
588 * many pages under writeback
590 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
594 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
595 * Cleared when kswapd is woken.
599 static inline unsigned long zone_managed_pages(struct zone *zone)
601 return (unsigned long)atomic_long_read(&zone->managed_pages);
604 static inline unsigned long zone_end_pfn(const struct zone *zone)
606 return zone->zone_start_pfn + zone->spanned_pages;
609 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
611 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
614 static inline bool zone_is_initialized(struct zone *zone)
616 return zone->initialized;
619 static inline bool zone_is_empty(struct zone *zone)
621 return zone->spanned_pages == 0;
625 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
626 * intersection with the given zone
628 static inline bool zone_intersects(struct zone *zone,
629 unsigned long start_pfn, unsigned long nr_pages)
631 if (zone_is_empty(zone))
633 if (start_pfn >= zone_end_pfn(zone) ||
634 start_pfn + nr_pages <= zone->zone_start_pfn)
641 * The "priority" of VM scanning is how much of the queues we will scan in one
642 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
643 * queues ("queue_length >> 12") during an aging round.
645 #define DEF_PRIORITY 12
647 /* Maximum number of zones on a zonelist */
648 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
651 ZONELIST_FALLBACK, /* zonelist with fallback */
654 * The NUMA zonelists are doubled because we need zonelists that
655 * restrict the allocations to a single node for __GFP_THISNODE.
657 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
663 * This struct contains information about a zone in a zonelist. It is stored
664 * here to avoid dereferences into large structures and lookups of tables
667 struct zone *zone; /* Pointer to actual zone */
668 int zone_idx; /* zone_idx(zoneref->zone) */
672 * One allocation request operates on a zonelist. A zonelist
673 * is a list of zones, the first one is the 'goal' of the
674 * allocation, the other zones are fallback zones, in decreasing
677 * To speed the reading of the zonelist, the zonerefs contain the zone index
678 * of the entry being read. Helper functions to access information given
679 * a struct zoneref are
681 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
682 * zonelist_zone_idx() - Return the index of the zone for an entry
683 * zonelist_node_idx() - Return the index of the node for an entry
686 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
689 #ifndef CONFIG_DISCONTIGMEM
690 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
691 extern struct page *mem_map;
694 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
695 struct deferred_split {
696 spinlock_t split_queue_lock;
697 struct list_head split_queue;
698 unsigned long split_queue_len;
703 * On NUMA machines, each NUMA node would have a pg_data_t to describe
704 * it's memory layout. On UMA machines there is a single pglist_data which
705 * describes the whole memory.
707 * Memory statistics and page replacement data structures are maintained on a
711 typedef struct pglist_data {
712 struct zone node_zones[MAX_NR_ZONES];
713 struct zonelist node_zonelists[MAX_ZONELISTS];
715 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
716 struct page *node_mem_map;
717 #ifdef CONFIG_PAGE_EXTENSION
718 struct page_ext *node_page_ext;
721 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
723 * Must be held any time you expect node_start_pfn,
724 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
726 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
727 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
728 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
730 * Nests above zone->lock and zone->span_seqlock
732 spinlock_t node_size_lock;
734 unsigned long node_start_pfn;
735 unsigned long node_present_pages; /* total number of physical pages */
736 unsigned long node_spanned_pages; /* total size of physical page
737 range, including holes */
739 wait_queue_head_t kswapd_wait;
740 wait_queue_head_t pfmemalloc_wait;
741 struct task_struct *kswapd; /* Protected by
742 mem_hotplug_begin/end() */
744 enum zone_type kswapd_classzone_idx;
746 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
748 #ifdef CONFIG_COMPACTION
749 int kcompactd_max_order;
750 enum zone_type kcompactd_classzone_idx;
751 wait_queue_head_t kcompactd_wait;
752 struct task_struct *kcompactd;
755 * This is a per-node reserve of pages that are not available
756 * to userspace allocations.
758 unsigned long totalreserve_pages;
762 * zone reclaim becomes active if more unmapped pages exist.
764 unsigned long min_unmapped_pages;
765 unsigned long min_slab_pages;
766 #endif /* CONFIG_NUMA */
768 /* Write-intensive fields used by page reclaim */
772 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
774 * If memory initialisation on large machines is deferred then this
775 * is the first PFN that needs to be initialised.
777 unsigned long first_deferred_pfn;
778 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
780 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
781 struct deferred_split deferred_split_queue;
784 /* Fields commonly accessed by the page reclaim scanner */
787 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
789 * Use mem_cgroup_lruvec() to look up lruvecs.
791 struct lruvec __lruvec;
797 /* Per-node vmstats */
798 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
799 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
802 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
803 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
804 #ifdef CONFIG_FLAT_NODE_MEM_MAP
805 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
807 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
809 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
811 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
812 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
814 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
816 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
819 static inline bool pgdat_is_empty(pg_data_t *pgdat)
821 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
824 #include <linux/memory_hotplug.h>
826 void build_all_zonelists(pg_data_t *pgdat);
827 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
828 enum zone_type classzone_idx);
829 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
830 int classzone_idx, unsigned int alloc_flags,
832 bool zone_watermark_ok(struct zone *z, unsigned int order,
833 unsigned long mark, int classzone_idx,
834 unsigned int alloc_flags);
835 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
836 unsigned long mark, int classzone_idx);
837 enum memmap_context {
841 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
844 extern void lruvec_init(struct lruvec *lruvec);
846 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
849 return lruvec->pgdat;
851 return container_of(lruvec, struct pglist_data, __lruvec);
855 extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
857 #ifdef CONFIG_HAVE_MEMORY_PRESENT
858 void memory_present(int nid, unsigned long start, unsigned long end);
860 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
863 #if defined(CONFIG_SPARSEMEM)
864 void memblocks_present(void);
866 static inline void memblocks_present(void) {}
869 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
870 int local_memory_node(int node_id);
872 static inline int local_memory_node(int node_id) { return node_id; };
876 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
878 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
881 * Returns true if a zone has pages managed by the buddy allocator.
882 * All the reclaim decisions have to use this function rather than
883 * populated_zone(). If the whole zone is reserved then we can easily
884 * end up with populated_zone() && !managed_zone().
886 static inline bool managed_zone(struct zone *zone)
888 return zone_managed_pages(zone);
891 /* Returns true if a zone has memory */
892 static inline bool populated_zone(struct zone *zone)
894 return zone->present_pages;
898 static inline int zone_to_nid(struct zone *zone)
903 static inline void zone_set_nid(struct zone *zone, int nid)
908 static inline int zone_to_nid(struct zone *zone)
913 static inline void zone_set_nid(struct zone *zone, int nid) {}
916 extern int movable_zone;
918 #ifdef CONFIG_HIGHMEM
919 static inline int zone_movable_is_highmem(void)
921 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
922 return movable_zone == ZONE_HIGHMEM;
924 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
929 static inline int is_highmem_idx(enum zone_type idx)
931 #ifdef CONFIG_HIGHMEM
932 return (idx == ZONE_HIGHMEM ||
933 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
940 * is_highmem - helper function to quickly check if a struct zone is a
941 * highmem zone or not. This is an attempt to keep references
942 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
943 * @zone - pointer to struct zone variable
945 static inline int is_highmem(struct zone *zone)
947 #ifdef CONFIG_HIGHMEM
948 return is_highmem_idx(zone_idx(zone));
954 /* These two functions are used to setup the per zone pages min values */
956 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
957 void __user *, size_t *, loff_t *);
958 int watermark_boost_factor_sysctl_handler(struct ctl_table *, int,
959 void __user *, size_t *, loff_t *);
960 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int,
961 void __user *, size_t *, loff_t *);
962 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
963 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
964 void __user *, size_t *, loff_t *);
965 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
966 void __user *, size_t *, loff_t *);
967 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
968 void __user *, size_t *, loff_t *);
969 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
970 void __user *, size_t *, loff_t *);
972 extern int numa_zonelist_order_handler(struct ctl_table *, int,
973 void __user *, size_t *, loff_t *);
974 extern char numa_zonelist_order[];
975 #define NUMA_ZONELIST_ORDER_LEN 16
977 #ifndef CONFIG_NEED_MULTIPLE_NODES
979 extern struct pglist_data contig_page_data;
980 #define NODE_DATA(nid) (&contig_page_data)
981 #define NODE_MEM_MAP(nid) mem_map
983 #else /* CONFIG_NEED_MULTIPLE_NODES */
985 #include <asm/mmzone.h>
987 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
989 extern struct pglist_data *first_online_pgdat(void);
990 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
991 extern struct zone *next_zone(struct zone *zone);
994 * for_each_online_pgdat - helper macro to iterate over all online nodes
995 * @pgdat - pointer to a pg_data_t variable
997 #define for_each_online_pgdat(pgdat) \
998 for (pgdat = first_online_pgdat(); \
1000 pgdat = next_online_pgdat(pgdat))
1002 * for_each_zone - helper macro to iterate over all memory zones
1003 * @zone - pointer to struct zone variable
1005 * The user only needs to declare the zone variable, for_each_zone
1008 #define for_each_zone(zone) \
1009 for (zone = (first_online_pgdat())->node_zones; \
1011 zone = next_zone(zone))
1013 #define for_each_populated_zone(zone) \
1014 for (zone = (first_online_pgdat())->node_zones; \
1016 zone = next_zone(zone)) \
1017 if (!populated_zone(zone)) \
1018 ; /* do nothing */ \
1021 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1023 return zoneref->zone;
1026 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1028 return zoneref->zone_idx;
1031 static inline int zonelist_node_idx(struct zoneref *zoneref)
1033 return zone_to_nid(zoneref->zone);
1036 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1037 enum zone_type highest_zoneidx,
1041 * 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
1042 * @z - The cursor used as a starting point for the search
1043 * @highest_zoneidx - The zone index of the highest zone to return
1044 * @nodes - An optional nodemask to filter the zonelist with
1046 * This function returns the next zone at or below a given zone index that is
1047 * within the allowed nodemask using a cursor as the starting point for the
1048 * search. The zoneref returned is a cursor that represents the current zone
1049 * being examined. It should be advanced by one before calling
1050 * next_zones_zonelist again.
1052 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1053 enum zone_type highest_zoneidx,
1056 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1058 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1062 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1063 * @zonelist - The zonelist to search for a suitable zone
1064 * @highest_zoneidx - The zone index of the highest zone to return
1065 * @nodes - An optional nodemask to filter the zonelist with
1066 * @return - Zoneref pointer for the first suitable zone found (see below)
1068 * This function returns the first zone at or below a given zone index that is
1069 * within the allowed nodemask. The zoneref returned is a cursor that can be
1070 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1071 * one before calling.
1073 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1074 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1075 * update due to cpuset modification.
1077 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1078 enum zone_type highest_zoneidx,
1081 return next_zones_zonelist(zonelist->_zonerefs,
1082 highest_zoneidx, nodes);
1086 * 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
1087 * @zone - The current zone in the iterator
1088 * @z - The current pointer within zonelist->_zonerefs being iterated
1089 * @zlist - The zonelist being iterated
1090 * @highidx - The zone index of the highest zone to return
1091 * @nodemask - Nodemask allowed by the allocator
1093 * This iterator iterates though all zones at or below a given zone index and
1094 * within a given nodemask
1096 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1097 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1099 z = next_zones_zonelist(++z, highidx, nodemask), \
1100 zone = zonelist_zone(z))
1102 #define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1103 for (zone = z->zone; \
1105 z = next_zones_zonelist(++z, highidx, nodemask), \
1106 zone = zonelist_zone(z))
1110 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1111 * @zone - The current zone in the iterator
1112 * @z - The current pointer within zonelist->zones being iterated
1113 * @zlist - The zonelist being iterated
1114 * @highidx - The zone index of the highest zone to return
1116 * This iterator iterates though all zones at or below a given zone index.
1118 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1119 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1121 #ifdef CONFIG_SPARSEMEM
1122 #include <asm/sparsemem.h>
1125 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1126 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1127 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1129 BUILD_BUG_ON(IS_ENABLED(CONFIG_NUMA));
1134 #ifdef CONFIG_FLATMEM
1135 #define pfn_to_nid(pfn) (0)
1138 #ifdef CONFIG_SPARSEMEM
1141 * SECTION_SHIFT #bits space required to store a section #
1143 * PA_SECTION_SHIFT physical address to/from section number
1144 * PFN_SECTION_SHIFT pfn to/from section number
1146 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1147 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1149 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1151 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1152 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1154 #define SECTION_BLOCKFLAGS_BITS \
1155 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1157 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1158 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1161 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1163 return pfn >> PFN_SECTION_SHIFT;
1165 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1167 return sec << PFN_SECTION_SHIFT;
1170 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1171 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1173 #define SUBSECTION_SHIFT 21
1175 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1176 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1177 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1179 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1180 #error Subsection size exceeds section size
1182 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1185 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1186 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1188 struct mem_section_usage {
1189 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1190 /* See declaration of similar field in struct zone */
1191 unsigned long pageblock_flags[0];
1194 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1198 struct mem_section {
1200 * This is, logically, a pointer to an array of struct
1201 * pages. However, it is stored with some other magic.
1202 * (see sparse.c::sparse_init_one_section())
1204 * Additionally during early boot we encode node id of
1205 * the location of the section here to guide allocation.
1206 * (see sparse.c::memory_present())
1208 * Making it a UL at least makes someone do a cast
1209 * before using it wrong.
1211 unsigned long section_mem_map;
1213 struct mem_section_usage *usage;
1214 #ifdef CONFIG_PAGE_EXTENSION
1216 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1217 * section. (see page_ext.h about this.)
1219 struct page_ext *page_ext;
1223 * WARNING: mem_section must be a power-of-2 in size for the
1224 * calculation and use of SECTION_ROOT_MASK to make sense.
1228 #ifdef CONFIG_SPARSEMEM_EXTREME
1229 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1231 #define SECTIONS_PER_ROOT 1
1234 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1235 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1236 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1238 #ifdef CONFIG_SPARSEMEM_EXTREME
1239 extern struct mem_section **mem_section;
1241 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1244 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1246 return ms->usage->pageblock_flags;
1249 static inline struct mem_section *__nr_to_section(unsigned long nr)
1251 #ifdef CONFIG_SPARSEMEM_EXTREME
1255 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1257 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1259 extern unsigned long __section_nr(struct mem_section *ms);
1260 extern size_t mem_section_usage_size(void);
1263 * We use the lower bits of the mem_map pointer to store
1264 * a little bit of information. The pointer is calculated
1265 * as mem_map - section_nr_to_pfn(pnum). The result is
1266 * aligned to the minimum alignment of the two values:
1267 * 1. All mem_map arrays are page-aligned.
1268 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1269 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1270 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1271 * worst combination is powerpc with 256k pages,
1272 * which results in PFN_SECTION_SHIFT equal 6.
1273 * To sum it up, at least 6 bits are available.
1275 #define SECTION_MARKED_PRESENT (1UL<<0)
1276 #define SECTION_HAS_MEM_MAP (1UL<<1)
1277 #define SECTION_IS_ONLINE (1UL<<2)
1278 #define SECTION_IS_EARLY (1UL<<3)
1279 #define SECTION_MAP_LAST_BIT (1UL<<4)
1280 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1281 #define SECTION_NID_SHIFT 3
1283 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1285 unsigned long map = section->section_mem_map;
1286 map &= SECTION_MAP_MASK;
1287 return (struct page *)map;
1290 static inline int present_section(struct mem_section *section)
1292 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1295 static inline int present_section_nr(unsigned long nr)
1297 return present_section(__nr_to_section(nr));
1300 static inline int valid_section(struct mem_section *section)
1302 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1305 static inline int early_section(struct mem_section *section)
1307 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1310 static inline int valid_section_nr(unsigned long nr)
1312 return valid_section(__nr_to_section(nr));
1315 static inline int online_section(struct mem_section *section)
1317 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1320 static inline int online_section_nr(unsigned long nr)
1322 return online_section(__nr_to_section(nr));
1325 #ifdef CONFIG_MEMORY_HOTPLUG
1326 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1327 #ifdef CONFIG_MEMORY_HOTREMOVE
1328 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1332 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1334 return __nr_to_section(pfn_to_section_nr(pfn));
1337 extern unsigned long __highest_present_section_nr;
1339 static inline int subsection_map_index(unsigned long pfn)
1341 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1344 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1345 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1347 int idx = subsection_map_index(pfn);
1349 return test_bit(idx, ms->usage->subsection_map);
1352 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1358 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1359 static inline int pfn_valid(unsigned long pfn)
1361 struct mem_section *ms;
1363 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1365 ms = __nr_to_section(pfn_to_section_nr(pfn));
1366 if (!valid_section(ms))
1369 * Traditionally early sections always returned pfn_valid() for
1370 * the entire section-sized span.
1372 return early_section(ms) || pfn_section_valid(ms, pfn);
1376 static inline int pfn_present(unsigned long pfn)
1378 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1380 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1384 * These are _only_ used during initialisation, therefore they
1385 * can use __initdata ... They could have names to indicate
1389 #define pfn_to_nid(pfn) \
1391 unsigned long __pfn_to_nid_pfn = (pfn); \
1392 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1395 #define pfn_to_nid(pfn) (0)
1398 #define early_pfn_valid(pfn) pfn_valid(pfn)
1399 void sparse_init(void);
1401 #define sparse_init() do {} while (0)
1402 #define sparse_index_init(_sec, _nid) do {} while (0)
1403 #define pfn_present pfn_valid
1404 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
1405 #endif /* CONFIG_SPARSEMEM */
1408 * During memory init memblocks map pfns to nids. The search is expensive and
1409 * this caches recent lookups. The implementation of __early_pfn_to_nid
1410 * may treat start/end as pfns or sections.
1412 struct mminit_pfnnid_cache {
1413 unsigned long last_start;
1414 unsigned long last_end;
1418 #ifndef early_pfn_valid
1419 #define early_pfn_valid(pfn) (1)
1422 void memory_present(int nid, unsigned long start, unsigned long end);
1425 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1426 * need to check pfn validity within that MAX_ORDER_NR_PAGES block.
1427 * pfn_valid_within() should be used in this case; we optimise this away
1428 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1430 #ifdef CONFIG_HOLES_IN_ZONE
1431 #define pfn_valid_within(pfn) pfn_valid(pfn)
1433 #define pfn_valid_within(pfn) (1)
1436 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1438 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1439 * associated with it or not. This means that a struct page exists for this
1440 * pfn. The caller cannot assume the page is fully initialized in general.
1441 * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1442 * will ensure the struct page is fully online and initialized. Special pages
1443 * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1445 * In FLATMEM, it is expected that holes always have valid memmap as long as
1446 * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1447 * that a valid section has a memmap for the entire section.
1449 * However, an ARM, and maybe other embedded architectures in the future
1450 * free memmap backing holes to save memory on the assumption the memmap is
1451 * never used. The page_zone linkages are then broken even though pfn_valid()
1452 * returns true. A walker of the full memmap must then do this additional
1453 * check to ensure the memmap they are looking at is sane by making sure
1454 * the zone and PFN linkages are still valid. This is expensive, but walkers
1455 * of the full memmap are extremely rare.
1457 bool memmap_valid_within(unsigned long pfn,
1458 struct page *page, struct zone *zone);
1460 static inline bool memmap_valid_within(unsigned long pfn,
1461 struct page *page, struct zone *zone)
1465 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1467 #endif /* !__GENERATING_BOUNDS.H */
1468 #endif /* !__ASSEMBLY__ */
1469 #endif /* _LINUX_MMZONE_H */