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 * Add a wild amount of padding here to ensure datas fall into separate
117 * cachelines. There are very few zone structures in the machine, so space
118 * consumption is not a concern here.
120 #if defined(CONFIG_SMP)
121 struct zone_padding {
123 } ____cacheline_internodealigned_in_smp;
124 #define ZONE_PADDING(name) struct zone_padding name;
126 #define ZONE_PADDING(name)
130 enum numa_stat_item {
131 NUMA_HIT, /* allocated in intended node */
132 NUMA_MISS, /* allocated in non intended node */
133 NUMA_FOREIGN, /* was intended here, hit elsewhere */
134 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
135 NUMA_LOCAL, /* allocation from local node */
136 NUMA_OTHER, /* allocation from other node */
137 NR_VM_NUMA_STAT_ITEMS
140 #define NR_VM_NUMA_STAT_ITEMS 0
143 enum zone_stat_item {
144 /* First 128 byte cacheline (assuming 64 bit words) */
146 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
147 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
149 NR_ZONE_INACTIVE_FILE,
152 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
153 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
154 /* Second 128 byte cacheline */
156 #if IS_ENABLED(CONFIG_ZSMALLOC)
157 NR_ZSPAGES, /* allocated in zsmalloc */
160 NR_VM_ZONE_STAT_ITEMS };
162 enum node_stat_item {
164 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
165 NR_ACTIVE_ANON, /* " " " " " */
166 NR_INACTIVE_FILE, /* " " " " " */
167 NR_ACTIVE_FILE, /* " " " " " */
168 NR_UNEVICTABLE, /* " " " " " */
169 NR_SLAB_RECLAIMABLE_B,
170 NR_SLAB_UNRECLAIMABLE_B,
171 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
172 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
174 WORKINGSET_REFAULT_BASE,
175 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
176 WORKINGSET_REFAULT_FILE,
177 WORKINGSET_ACTIVATE_BASE,
178 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
179 WORKINGSET_ACTIVATE_FILE,
180 WORKINGSET_RESTORE_BASE,
181 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
182 WORKINGSET_RESTORE_FILE,
183 WORKINGSET_NODERECLAIM,
184 NR_ANON_MAPPED, /* Mapped anonymous pages */
185 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
186 only modified from process context */
190 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
191 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
198 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
199 NR_DIRTIED, /* page dirtyings since bootup */
200 NR_WRITTEN, /* page writings since bootup */
201 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
202 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
203 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
204 NR_KERNEL_STACK_KB, /* measured in KiB */
205 #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
206 NR_KERNEL_SCS_KB, /* measured in KiB */
208 NR_PAGETABLE, /* used for pagetables */
209 NR_VM_NODE_STAT_ITEMS
213 * Returns true if the value is measured in bytes (most vmstat values are
214 * measured in pages). This defines the API part, the internal representation
215 * might be different.
217 static __always_inline bool vmstat_item_in_bytes(int idx)
220 * Global and per-node slab counters track slab pages.
221 * It's expected that changes are multiples of PAGE_SIZE.
222 * Internally values are stored in pages.
224 * Per-memcg and per-lruvec counters track memory, consumed
225 * by individual slab objects. These counters are actually
228 return (idx == NR_SLAB_RECLAIMABLE_B ||
229 idx == NR_SLAB_UNRECLAIMABLE_B);
233 * We do arithmetic on the LRU lists in various places in the code,
234 * so it is important to keep the active lists LRU_ACTIVE higher in
235 * the array than the corresponding inactive lists, and to keep
236 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
238 * This has to be kept in sync with the statistics in zone_stat_item
239 * above and the descriptions in vmstat_text in mm/vmstat.c
246 LRU_INACTIVE_ANON = LRU_BASE,
247 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
248 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
249 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
254 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
256 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
258 static inline bool is_file_lru(enum lru_list lru)
260 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
263 static inline bool is_active_lru(enum lru_list lru)
265 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
268 #define ANON_AND_FILE 2
271 LRUVEC_CONGESTED, /* lruvec has many dirty pages
272 * backed by a congested BDI
277 struct list_head lists[NR_LRU_LISTS];
278 /* per lruvec lru_lock for memcg */
281 * These track the cost of reclaiming one LRU - file or anon -
282 * over the other. As the observed cost of reclaiming one LRU
283 * increases, the reclaim scan balance tips toward the other.
285 unsigned long anon_cost;
286 unsigned long file_cost;
287 /* Non-resident age, driven by LRU movement */
288 atomic_long_t nonresident_age;
289 /* Refaults at the time of last reclaim cycle */
290 unsigned long refaults[ANON_AND_FILE];
291 /* Various lruvec state flags (enum lruvec_flags) */
294 struct pglist_data *pgdat;
298 /* Isolate unmapped pages */
299 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
300 /* Isolate for asynchronous migration */
301 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
302 /* Isolate unevictable pages */
303 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
305 /* LRU Isolation modes. */
306 typedef unsigned __bitwise isolate_mode_t;
308 enum zone_watermarks {
315 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
316 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
317 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
318 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
320 struct per_cpu_pages {
321 int count; /* number of pages in the list */
322 int high; /* high watermark, emptying needed */
323 int batch; /* chunk size for buddy add/remove */
325 /* Lists of pages, one per migrate type stored on the pcp-lists */
326 struct list_head lists[MIGRATE_PCPTYPES];
329 struct per_cpu_pageset {
330 struct per_cpu_pages pcp;
333 u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
337 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
341 struct per_cpu_nodestat {
343 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
346 #endif /* !__GENERATING_BOUNDS.H */
350 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
351 * to DMA to all of the addressable memory (ZONE_NORMAL).
352 * On architectures where this area covers the whole 32 bit address
353 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
354 * DMA addressing constraints. This distinction is important as a 32bit
355 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
356 * platforms may need both zones as they support peripherals with
357 * different DMA addressing limitations.
359 #ifdef CONFIG_ZONE_DMA
362 #ifdef CONFIG_ZONE_DMA32
366 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
367 * performed on pages in ZONE_NORMAL if the DMA devices support
368 * transfers to all addressable memory.
371 #ifdef CONFIG_HIGHMEM
373 * A memory area that is only addressable by the kernel through
374 * mapping portions into its own address space. This is for example
375 * used by i386 to allow the kernel to address the memory beyond
376 * 900MB. The kernel will set up special mappings (page
377 * table entries on i386) for each page that the kernel needs to
383 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
384 * movable pages with few exceptional cases described below. Main use
385 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
386 * likely to succeed, and to locally limit unmovable allocations - e.g.,
387 * to increase the number of THP/huge pages. Notable special cases are:
389 * 1. Pinned pages: (long-term) pinning of movable pages might
390 * essentially turn such pages unmovable. Memory offlining might
392 * 2. memblock allocations: kernelcore/movablecore setups might create
393 * situations where ZONE_MOVABLE contains unmovable allocations
394 * after boot. Memory offlining and allocations fail early.
395 * 3. Memory holes: kernelcore/movablecore setups might create very rare
396 * situations where ZONE_MOVABLE contains memory holes after boot,
397 * for example, if we have sections that are only partially
398 * populated. Memory offlining and allocations fail early.
399 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
400 * memory offlining, such pages cannot be allocated.
401 * 5. Unmovable PG_offline pages: in paravirtualized environments,
402 * hotplugged memory blocks might only partially be managed by the
403 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
404 * parts not manged by the buddy are unmovable PG_offline pages. In
405 * some cases (virtio-mem), such pages can be skipped during
406 * memory offlining, however, cannot be moved/allocated. These
407 * techniques might use alloc_contig_range() to hide previously
408 * exposed pages from the buddy again (e.g., to implement some sort
409 * of memory unplug in virtio-mem).
411 * In general, no unmovable allocations that degrade memory offlining
412 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
413 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
414 * if has_unmovable_pages() states that there are no unmovable pages,
415 * there can be false negatives).
418 #ifdef CONFIG_ZONE_DEVICE
425 #ifndef __GENERATING_BOUNDS_H
427 #define ASYNC_AND_SYNC 2
430 /* Read-mostly fields */
432 /* zone watermarks, access with *_wmark_pages(zone) macros */
433 unsigned long _watermark[NR_WMARK];
434 unsigned long watermark_boost;
436 unsigned long nr_reserved_highatomic;
439 * We don't know if the memory that we're going to allocate will be
440 * freeable or/and it will be released eventually, so to avoid totally
441 * wasting several GB of ram we must reserve some of the lower zone
442 * memory (otherwise we risk to run OOM on the lower zones despite
443 * there being tons of freeable ram on the higher zones). This array is
444 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
447 long lowmem_reserve[MAX_NR_ZONES];
452 struct pglist_data *zone_pgdat;
453 struct per_cpu_pageset __percpu *pageset;
455 * the high and batch values are copied to individual pagesets for
461 #ifndef CONFIG_SPARSEMEM
463 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
464 * In SPARSEMEM, this map is stored in struct mem_section
466 unsigned long *pageblock_flags;
467 #endif /* CONFIG_SPARSEMEM */
469 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
470 unsigned long zone_start_pfn;
473 * spanned_pages is the total pages spanned by the zone, including
474 * holes, which is calculated as:
475 * spanned_pages = zone_end_pfn - zone_start_pfn;
477 * present_pages is physical pages existing within the zone, which
479 * present_pages = spanned_pages - absent_pages(pages in holes);
481 * managed_pages is present pages managed by the buddy system, which
482 * is calculated as (reserved_pages includes pages allocated by the
483 * bootmem allocator):
484 * managed_pages = present_pages - reserved_pages;
486 * So present_pages may be used by memory hotplug or memory power
487 * management logic to figure out unmanaged pages by checking
488 * (present_pages - managed_pages). And managed_pages should be used
489 * by page allocator and vm scanner to calculate all kinds of watermarks
494 * zone_start_pfn and spanned_pages are protected by span_seqlock.
495 * It is a seqlock because it has to be read outside of zone->lock,
496 * and it is done in the main allocator path. But, it is written
497 * quite infrequently.
499 * The span_seq lock is declared along with zone->lock because it is
500 * frequently read in proximity to zone->lock. It's good to
501 * give them a chance of being in the same cacheline.
503 * Write access to present_pages at runtime should be protected by
504 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
505 * present_pages should get_online_mems() to get a stable value.
507 atomic_long_t managed_pages;
508 unsigned long spanned_pages;
509 unsigned long present_pages;
513 #ifdef CONFIG_MEMORY_ISOLATION
515 * Number of isolated pageblock. It is used to solve incorrect
516 * freepage counting problem due to racy retrieving migratetype
517 * of pageblock. Protected by zone->lock.
519 unsigned long nr_isolate_pageblock;
522 #ifdef CONFIG_MEMORY_HOTPLUG
523 /* see spanned/present_pages for more description */
524 seqlock_t span_seqlock;
529 /* Write-intensive fields used from the page allocator */
532 /* free areas of different sizes */
533 struct free_area free_area[MAX_ORDER];
535 /* zone flags, see below */
538 /* Primarily protects free_area */
541 /* Write-intensive fields used by compaction and vmstats. */
545 * When free pages are below this point, additional steps are taken
546 * when reading the number of free pages to avoid per-cpu counter
547 * drift allowing watermarks to be breached
549 unsigned long percpu_drift_mark;
551 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
552 /* pfn where compaction free scanner should start */
553 unsigned long compact_cached_free_pfn;
554 /* pfn where compaction migration scanner should start */
555 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
556 unsigned long compact_init_migrate_pfn;
557 unsigned long compact_init_free_pfn;
560 #ifdef CONFIG_COMPACTION
562 * On compaction failure, 1<<compact_defer_shift compactions
563 * are skipped before trying again. The number attempted since
564 * last failure is tracked with compact_considered.
565 * compact_order_failed is the minimum compaction failed order.
567 unsigned int compact_considered;
568 unsigned int compact_defer_shift;
569 int compact_order_failed;
572 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
573 /* Set to true when the PG_migrate_skip bits should be cleared */
574 bool compact_blockskip_flush;
580 /* Zone statistics */
581 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
582 atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
583 } ____cacheline_internodealigned_in_smp;
586 PGDAT_DIRTY, /* reclaim scanning has recently found
587 * many dirty file pages at the tail
590 PGDAT_WRITEBACK, /* reclaim scanning has recently found
591 * many pages under writeback
593 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
597 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
598 * Cleared when kswapd is woken.
602 static inline unsigned long zone_managed_pages(struct zone *zone)
604 return (unsigned long)atomic_long_read(&zone->managed_pages);
607 static inline unsigned long zone_end_pfn(const struct zone *zone)
609 return zone->zone_start_pfn + zone->spanned_pages;
612 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
614 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
617 static inline bool zone_is_initialized(struct zone *zone)
619 return zone->initialized;
622 static inline bool zone_is_empty(struct zone *zone)
624 return zone->spanned_pages == 0;
628 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
629 * intersection with the given zone
631 static inline bool zone_intersects(struct zone *zone,
632 unsigned long start_pfn, unsigned long nr_pages)
634 if (zone_is_empty(zone))
636 if (start_pfn >= zone_end_pfn(zone) ||
637 start_pfn + nr_pages <= zone->zone_start_pfn)
644 * The "priority" of VM scanning is how much of the queues we will scan in one
645 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
646 * queues ("queue_length >> 12") during an aging round.
648 #define DEF_PRIORITY 12
650 /* Maximum number of zones on a zonelist */
651 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
654 ZONELIST_FALLBACK, /* zonelist with fallback */
657 * The NUMA zonelists are doubled because we need zonelists that
658 * restrict the allocations to a single node for __GFP_THISNODE.
660 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
666 * This struct contains information about a zone in a zonelist. It is stored
667 * here to avoid dereferences into large structures and lookups of tables
670 struct zone *zone; /* Pointer to actual zone */
671 int zone_idx; /* zone_idx(zoneref->zone) */
675 * One allocation request operates on a zonelist. A zonelist
676 * is a list of zones, the first one is the 'goal' of the
677 * allocation, the other zones are fallback zones, in decreasing
680 * To speed the reading of the zonelist, the zonerefs contain the zone index
681 * of the entry being read. Helper functions to access information given
682 * a struct zoneref are
684 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
685 * zonelist_zone_idx() - Return the index of the zone for an entry
686 * zonelist_node_idx() - Return the index of the node for an entry
689 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
692 #ifndef CONFIG_DISCONTIGMEM
693 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
694 extern struct page *mem_map;
697 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
698 struct deferred_split {
699 spinlock_t split_queue_lock;
700 struct list_head split_queue;
701 unsigned long split_queue_len;
706 * On NUMA machines, each NUMA node would have a pg_data_t to describe
707 * it's memory layout. On UMA machines there is a single pglist_data which
708 * describes the whole memory.
710 * Memory statistics and page replacement data structures are maintained on a
713 typedef struct pglist_data {
715 * node_zones contains just the zones for THIS node. Not all of the
716 * zones may be populated, but it is the full list. It is referenced by
717 * this node's node_zonelists as well as other node's node_zonelists.
719 struct zone node_zones[MAX_NR_ZONES];
722 * node_zonelists contains references to all zones in all nodes.
723 * Generally the first zones will be references to this node's
726 struct zonelist node_zonelists[MAX_ZONELISTS];
728 int nr_zones; /* number of populated zones in this node */
729 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
730 struct page *node_mem_map;
731 #ifdef CONFIG_PAGE_EXTENSION
732 struct page_ext *node_page_ext;
735 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
737 * Must be held any time you expect node_start_pfn,
738 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
739 * Also synchronizes pgdat->first_deferred_pfn during deferred page
742 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
743 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
744 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
746 * Nests above zone->lock and zone->span_seqlock
748 spinlock_t node_size_lock;
750 unsigned long node_start_pfn;
751 unsigned long node_present_pages; /* total number of physical pages */
752 unsigned long node_spanned_pages; /* total size of physical page
753 range, including holes */
755 wait_queue_head_t kswapd_wait;
756 wait_queue_head_t pfmemalloc_wait;
757 struct task_struct *kswapd; /* Protected by
758 mem_hotplug_begin/end() */
760 enum zone_type kswapd_highest_zoneidx;
762 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
764 #ifdef CONFIG_COMPACTION
765 int kcompactd_max_order;
766 enum zone_type kcompactd_highest_zoneidx;
767 wait_queue_head_t kcompactd_wait;
768 struct task_struct *kcompactd;
771 * This is a per-node reserve of pages that are not available
772 * to userspace allocations.
774 unsigned long totalreserve_pages;
778 * node reclaim becomes active if more unmapped pages exist.
780 unsigned long min_unmapped_pages;
781 unsigned long min_slab_pages;
782 #endif /* CONFIG_NUMA */
784 /* Write-intensive fields used by page reclaim */
787 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
789 * If memory initialisation on large machines is deferred then this
790 * is the first PFN that needs to be initialised.
792 unsigned long first_deferred_pfn;
793 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
795 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
796 struct deferred_split deferred_split_queue;
799 /* Fields commonly accessed by the page reclaim scanner */
802 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
804 * Use mem_cgroup_lruvec() to look up lruvecs.
806 struct lruvec __lruvec;
812 /* Per-node vmstats */
813 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
814 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
817 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
818 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
819 #ifdef CONFIG_FLAT_NODE_MEM_MAP
820 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
822 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
824 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
826 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
827 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
829 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
831 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
834 static inline bool pgdat_is_empty(pg_data_t *pgdat)
836 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
839 #include <linux/memory_hotplug.h>
841 void build_all_zonelists(pg_data_t *pgdat);
842 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
843 enum zone_type highest_zoneidx);
844 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
845 int highest_zoneidx, unsigned int alloc_flags,
847 bool zone_watermark_ok(struct zone *z, unsigned int order,
848 unsigned long mark, int highest_zoneidx,
849 unsigned int alloc_flags);
850 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
851 unsigned long mark, int highest_zoneidx);
853 * Memory initialization context, use to differentiate memory added by
854 * the platform statically or via memory hotplug interface.
856 enum meminit_context {
861 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
864 extern void lruvec_init(struct lruvec *lruvec);
866 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
869 return lruvec->pgdat;
871 return container_of(lruvec, struct pglist_data, __lruvec);
875 extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
877 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
878 int local_memory_node(int node_id);
880 static inline int local_memory_node(int node_id) { return node_id; };
884 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
886 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
889 * Returns true if a zone has pages managed by the buddy allocator.
890 * All the reclaim decisions have to use this function rather than
891 * populated_zone(). If the whole zone is reserved then we can easily
892 * end up with populated_zone() && !managed_zone().
894 static inline bool managed_zone(struct zone *zone)
896 return zone_managed_pages(zone);
899 /* Returns true if a zone has memory */
900 static inline bool populated_zone(struct zone *zone)
902 return zone->present_pages;
906 static inline int zone_to_nid(struct zone *zone)
911 static inline void zone_set_nid(struct zone *zone, int nid)
916 static inline int zone_to_nid(struct zone *zone)
921 static inline void zone_set_nid(struct zone *zone, int nid) {}
924 extern int movable_zone;
926 #ifdef CONFIG_HIGHMEM
927 static inline int zone_movable_is_highmem(void)
929 #ifdef CONFIG_NEED_MULTIPLE_NODES
930 return movable_zone == ZONE_HIGHMEM;
932 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
937 static inline int is_highmem_idx(enum zone_type idx)
939 #ifdef CONFIG_HIGHMEM
940 return (idx == ZONE_HIGHMEM ||
941 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
948 * is_highmem - helper function to quickly check if a struct zone is a
949 * highmem zone or not. This is an attempt to keep references
950 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
951 * @zone - pointer to struct zone variable
953 static inline int is_highmem(struct zone *zone)
955 #ifdef CONFIG_HIGHMEM
956 return is_highmem_idx(zone_idx(zone));
962 /* These two functions are used to setup the per zone pages min values */
965 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, void *, size_t *,
967 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, void *,
969 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
970 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, void *,
972 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
973 void *, size_t *, loff_t *);
974 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
975 void *, size_t *, loff_t *);
976 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
977 void *, size_t *, loff_t *);
978 int numa_zonelist_order_handler(struct ctl_table *, int,
979 void *, size_t *, loff_t *);
980 extern int percpu_pagelist_fraction;
981 extern char numa_zonelist_order[];
982 #define NUMA_ZONELIST_ORDER_LEN 16
984 #ifndef CONFIG_NEED_MULTIPLE_NODES
986 extern struct pglist_data contig_page_data;
987 #define NODE_DATA(nid) (&contig_page_data)
988 #define NODE_MEM_MAP(nid) mem_map
990 #else /* CONFIG_NEED_MULTIPLE_NODES */
992 #include <asm/mmzone.h>
994 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
996 extern struct pglist_data *first_online_pgdat(void);
997 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
998 extern struct zone *next_zone(struct zone *zone);
1001 * for_each_online_pgdat - helper macro to iterate over all online nodes
1002 * @pgdat - pointer to a pg_data_t variable
1004 #define for_each_online_pgdat(pgdat) \
1005 for (pgdat = first_online_pgdat(); \
1007 pgdat = next_online_pgdat(pgdat))
1009 * for_each_zone - helper macro to iterate over all memory zones
1010 * @zone - pointer to struct zone variable
1012 * The user only needs to declare the zone variable, for_each_zone
1015 #define for_each_zone(zone) \
1016 for (zone = (first_online_pgdat())->node_zones; \
1018 zone = next_zone(zone))
1020 #define for_each_populated_zone(zone) \
1021 for (zone = (first_online_pgdat())->node_zones; \
1023 zone = next_zone(zone)) \
1024 if (!populated_zone(zone)) \
1025 ; /* do nothing */ \
1028 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1030 return zoneref->zone;
1033 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1035 return zoneref->zone_idx;
1038 static inline int zonelist_node_idx(struct zoneref *zoneref)
1040 return zone_to_nid(zoneref->zone);
1043 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1044 enum zone_type highest_zoneidx,
1048 * 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
1049 * @z - The cursor used as a starting point for the search
1050 * @highest_zoneidx - The zone index of the highest zone to return
1051 * @nodes - An optional nodemask to filter the zonelist with
1053 * This function returns the next zone at or below a given zone index that is
1054 * within the allowed nodemask using a cursor as the starting point for the
1055 * search. The zoneref returned is a cursor that represents the current zone
1056 * being examined. It should be advanced by one before calling
1057 * next_zones_zonelist again.
1059 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1060 enum zone_type highest_zoneidx,
1063 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1065 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1069 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1070 * @zonelist - The zonelist to search for a suitable zone
1071 * @highest_zoneidx - The zone index of the highest zone to return
1072 * @nodes - An optional nodemask to filter the zonelist with
1073 * @return - Zoneref pointer for the first suitable zone found (see below)
1075 * This function returns the first zone at or below a given zone index that is
1076 * within the allowed nodemask. The zoneref returned is a cursor that can be
1077 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1078 * one before calling.
1080 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1081 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1082 * update due to cpuset modification.
1084 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1085 enum zone_type highest_zoneidx,
1088 return next_zones_zonelist(zonelist->_zonerefs,
1089 highest_zoneidx, nodes);
1093 * 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
1094 * @zone - The current zone in the iterator
1095 * @z - The current pointer within zonelist->_zonerefs being iterated
1096 * @zlist - The zonelist being iterated
1097 * @highidx - The zone index of the highest zone to return
1098 * @nodemask - Nodemask allowed by the allocator
1100 * This iterator iterates though all zones at or below a given zone index and
1101 * within a given nodemask
1103 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1104 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1106 z = next_zones_zonelist(++z, highidx, nodemask), \
1107 zone = zonelist_zone(z))
1109 #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1110 for (zone = z->zone; \
1112 z = next_zones_zonelist(++z, highidx, nodemask), \
1113 zone = zonelist_zone(z))
1117 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1118 * @zone - The current zone in the iterator
1119 * @z - The current pointer within zonelist->zones being iterated
1120 * @zlist - The zonelist being iterated
1121 * @highidx - The zone index of the highest zone to return
1123 * This iterator iterates though all zones at or below a given zone index.
1125 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1126 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1128 #ifdef CONFIG_SPARSEMEM
1129 #include <asm/sparsemem.h>
1132 #ifdef CONFIG_FLATMEM
1133 #define pfn_to_nid(pfn) (0)
1136 #ifdef CONFIG_SPARSEMEM
1139 * SECTION_SHIFT #bits space required to store a section #
1141 * PA_SECTION_SHIFT physical address to/from section number
1142 * PFN_SECTION_SHIFT pfn to/from section number
1144 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1145 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1147 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1149 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1150 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1152 #define SECTION_BLOCKFLAGS_BITS \
1153 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1155 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1156 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1159 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1161 return pfn >> PFN_SECTION_SHIFT;
1163 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1165 return sec << PFN_SECTION_SHIFT;
1168 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1169 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1171 #define SUBSECTION_SHIFT 21
1172 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1174 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1175 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1176 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1178 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1179 #error Subsection size exceeds section size
1181 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1184 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1185 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1187 struct mem_section_usage {
1188 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1189 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1191 /* See declaration of similar field in struct zone */
1192 unsigned long pageblock_flags[0];
1195 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1199 struct mem_section {
1201 * This is, logically, a pointer to an array of struct
1202 * pages. However, it is stored with some other magic.
1203 * (see sparse.c::sparse_init_one_section())
1205 * Additionally during early boot we encode node id of
1206 * the location of the section here to guide allocation.
1207 * (see sparse.c::memory_present())
1209 * Making it a UL at least makes someone do a cast
1210 * before using it wrong.
1212 unsigned long section_mem_map;
1214 struct mem_section_usage *usage;
1215 #ifdef CONFIG_PAGE_EXTENSION
1217 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1218 * section. (see page_ext.h about this.)
1220 struct page_ext *page_ext;
1224 * WARNING: mem_section must be a power-of-2 in size for the
1225 * calculation and use of SECTION_ROOT_MASK to make sense.
1229 #ifdef CONFIG_SPARSEMEM_EXTREME
1230 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1232 #define SECTIONS_PER_ROOT 1
1235 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1236 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1237 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1239 #ifdef CONFIG_SPARSEMEM_EXTREME
1240 extern struct mem_section **mem_section;
1242 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1245 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1247 return ms->usage->pageblock_flags;
1250 static inline struct mem_section *__nr_to_section(unsigned long nr)
1252 #ifdef CONFIG_SPARSEMEM_EXTREME
1256 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1258 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1260 extern unsigned long __section_nr(struct mem_section *ms);
1261 extern size_t mem_section_usage_size(void);
1264 * We use the lower bits of the mem_map pointer to store
1265 * a little bit of information. The pointer is calculated
1266 * as mem_map - section_nr_to_pfn(pnum). The result is
1267 * aligned to the minimum alignment of the two values:
1268 * 1. All mem_map arrays are page-aligned.
1269 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1270 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1271 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1272 * worst combination is powerpc with 256k pages,
1273 * which results in PFN_SECTION_SHIFT equal 6.
1274 * To sum it up, at least 6 bits are available.
1276 #define SECTION_MARKED_PRESENT (1UL<<0)
1277 #define SECTION_HAS_MEM_MAP (1UL<<1)
1278 #define SECTION_IS_ONLINE (1UL<<2)
1279 #define SECTION_IS_EARLY (1UL<<3)
1280 #define SECTION_MAP_LAST_BIT (1UL<<4)
1281 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1282 #define SECTION_NID_SHIFT 3
1284 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1286 unsigned long map = section->section_mem_map;
1287 map &= SECTION_MAP_MASK;
1288 return (struct page *)map;
1291 static inline int present_section(struct mem_section *section)
1293 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1296 static inline int present_section_nr(unsigned long nr)
1298 return present_section(__nr_to_section(nr));
1301 static inline int valid_section(struct mem_section *section)
1303 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1306 static inline int early_section(struct mem_section *section)
1308 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1311 static inline int valid_section_nr(unsigned long nr)
1313 return valid_section(__nr_to_section(nr));
1316 static inline int online_section(struct mem_section *section)
1318 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1321 static inline int online_section_nr(unsigned long nr)
1323 return online_section(__nr_to_section(nr));
1326 #ifdef CONFIG_MEMORY_HOTPLUG
1327 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1328 #ifdef CONFIG_MEMORY_HOTREMOVE
1329 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1333 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1335 return __nr_to_section(pfn_to_section_nr(pfn));
1338 extern unsigned long __highest_present_section_nr;
1340 static inline int subsection_map_index(unsigned long pfn)
1342 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1345 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1346 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1348 int idx = subsection_map_index(pfn);
1350 return test_bit(idx, ms->usage->subsection_map);
1353 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1359 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1360 static inline int pfn_valid(unsigned long pfn)
1362 struct mem_section *ms;
1364 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1366 ms = __nr_to_section(pfn_to_section_nr(pfn));
1367 if (!valid_section(ms))
1370 * Traditionally early sections always returned pfn_valid() for
1371 * the entire section-sized span.
1373 return early_section(ms) || pfn_section_valid(ms, pfn);
1377 static inline int pfn_in_present_section(unsigned long pfn)
1379 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1381 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1384 static inline unsigned long next_present_section_nr(unsigned long section_nr)
1386 while (++section_nr <= __highest_present_section_nr) {
1387 if (present_section_nr(section_nr))
1395 * These are _only_ used during initialisation, therefore they
1396 * can use __initdata ... They could have names to indicate
1400 #define pfn_to_nid(pfn) \
1402 unsigned long __pfn_to_nid_pfn = (pfn); \
1403 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1406 #define pfn_to_nid(pfn) (0)
1409 void sparse_init(void);
1411 #define sparse_init() do {} while (0)
1412 #define sparse_index_init(_sec, _nid) do {} while (0)
1413 #define pfn_in_present_section pfn_valid
1414 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
1415 #endif /* CONFIG_SPARSEMEM */
1418 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1419 * need to check pfn validity within that MAX_ORDER_NR_PAGES block.
1420 * pfn_valid_within() should be used in this case; we optimise this away
1421 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1423 #ifdef CONFIG_HOLES_IN_ZONE
1424 #define pfn_valid_within(pfn) pfn_valid(pfn)
1426 #define pfn_valid_within(pfn) (1)
1429 #endif /* !__GENERATING_BOUNDS.H */
1430 #endif /* !__ASSEMBLY__ */
1431 #endif /* _LINUX_MMZONE_H */