1 #ifndef _LINUX_MMZONE_H
2 #define _LINUX_MMZONE_H
5 #ifndef __GENERATING_BOUNDS_H
7 #include <linux/spinlock.h>
8 #include <linux/list.h>
9 #include <linux/wait.h>
10 #include <linux/bitops.h>
11 #include <linux/cache.h>
12 #include <linux/threads.h>
13 #include <linux/numa.h>
14 #include <linux/init.h>
15 #include <linux/seqlock.h>
16 #include <linux/nodemask.h>
17 #include <linux/pageblock-flags.h>
18 #include <linux/page-flags-layout.h>
19 #include <linux/atomic.h>
22 /* Free memory management - zoned buddy allocator. */
23 #ifndef CONFIG_FORCE_MAX_ZONEORDER
26 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
28 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
31 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
32 * costly to service. That is between allocation orders which should
33 * coalesce naturally under reasonable reclaim pressure and those which
36 #define PAGE_ALLOC_COSTLY_ORDER 3
45 MIGRATE_PCPTYPES
, /* the number of types on the pcp lists */
46 MIGRATE_RESERVE
= 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 */
70 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
72 # define is_migrate_cma(migratetype) false
76 #define is_migrate_mtkpasr(mt) unlikely((mt) == MIGRATE_MTKPASR)
78 #define is_migrate_mtkpasr(mt) false
81 #define for_each_migratetype_order(order, type) \
82 for (order = 0; order < MAX_ORDER; order++) \
83 for (type = 0; type < MIGRATE_TYPES; type++)
85 extern int page_group_by_mobility_disabled
;
87 static inline int get_pageblock_migratetype(struct page
*page
)
89 return get_pageblock_flags_group(page
, PB_migrate
, PB_migrate_end
);
93 struct list_head free_list
[MIGRATE_TYPES
];
94 unsigned long nr_free
;
100 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
101 * So add a wild amount of padding here to ensure that they fall into separate
102 * cachelines. There are very few zone structures in the machine, so space
103 * consumption is not a concern here.
105 #if defined(CONFIG_SMP)
106 struct zone_padding
{
108 } ____cacheline_internodealigned_in_smp
;
109 #define ZONE_PADDING(name) struct zone_padding name;
111 #define ZONE_PADDING(name)
114 enum zone_stat_item
{
115 /* First 128 byte cacheline (assuming 64 bit words) */
118 NR_INACTIVE_ANON
= NR_LRU_BASE
, /* must match order of LRU_[IN]ACTIVE */
119 NR_ACTIVE_ANON
, /* " " " " " */
120 NR_INACTIVE_FILE
, /* " " " " " */
121 NR_ACTIVE_FILE
, /* " " " " " */
122 NR_UNEVICTABLE
, /* " " " " " */
123 NR_MLOCK
, /* mlock()ed pages found and moved off LRU */
124 NR_ANON_PAGES
, /* Mapped anonymous pages */
125 NR_FILE_MAPPED
, /* pagecache pages mapped into pagetables.
126 only modified from process context */
131 NR_SLAB_UNRECLAIMABLE
,
132 NR_PAGETABLE
, /* used for pagetables */
134 /* Second 128 byte cacheline */
135 NR_UNSTABLE_NFS
, /* NFS unstable pages */
138 NR_VMSCAN_IMMEDIATE
, /* Prioritise for reclaim when writeback ends */
139 NR_WRITEBACK_TEMP
, /* Writeback using temporary buffers */
140 NR_ISOLATED_ANON
, /* Temporary isolated pages from anon lru */
141 NR_ISOLATED_FILE
, /* Temporary isolated pages from file lru */
142 NR_SHMEM
, /* shmem pages (included tmpfs/GEM pages) */
143 NR_DIRTIED
, /* page dirtyings since bootup */
144 NR_WRITTEN
, /* page writings since bootup */
146 NUMA_HIT
, /* allocated in intended node */
147 NUMA_MISS
, /* allocated in non intended node */
148 NUMA_FOREIGN
, /* was intended here, hit elsewhere */
149 NUMA_INTERLEAVE_HIT
, /* interleaver preferred this zone */
150 NUMA_LOCAL
, /* allocation from local node */
151 NUMA_OTHER
, /* allocation from other node */
153 NR_ANON_TRANSPARENT_HUGEPAGES
,
155 NR_VM_ZONE_STAT_ITEMS
};
158 * We do arithmetic on the LRU lists in various places in the code,
159 * so it is important to keep the active lists LRU_ACTIVE higher in
160 * the array than the corresponding inactive lists, and to keep
161 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
163 * This has to be kept in sync with the statistics in zone_stat_item
164 * above and the descriptions in vmstat_text in mm/vmstat.c
171 LRU_INACTIVE_ANON
= LRU_BASE
,
172 LRU_ACTIVE_ANON
= LRU_BASE
+ LRU_ACTIVE
,
173 LRU_INACTIVE_FILE
= LRU_BASE
+ LRU_FILE
,
174 LRU_ACTIVE_FILE
= LRU_BASE
+ LRU_FILE
+ LRU_ACTIVE
,
179 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
181 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
183 static inline int is_file_lru(enum lru_list lru
)
185 return (lru
== LRU_INACTIVE_FILE
|| lru
== LRU_ACTIVE_FILE
);
188 static inline int is_active_lru(enum lru_list lru
)
190 return (lru
== LRU_ACTIVE_ANON
|| lru
== LRU_ACTIVE_FILE
);
193 static inline int is_unevictable_lru(enum lru_list lru
)
195 return (lru
== LRU_UNEVICTABLE
);
198 struct zone_reclaim_stat
{
200 * The pageout code in vmscan.c keeps track of how many of the
201 * mem/swap backed and file backed pages are referenced.
202 * The higher the rotated/scanned ratio, the more valuable
205 * The anon LRU stats live in [0], file LRU stats in [1]
207 unsigned long recent_rotated
[2];
208 unsigned long recent_scanned
[2];
212 struct list_head lists
[NR_LRU_LISTS
];
213 struct zone_reclaim_stat reclaim_stat
;
219 /* Mask used at gathering information at once (see memcontrol.c) */
220 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
221 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
222 #define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
224 /* Isolate clean file */
225 #define ISOLATE_CLEAN ((__force isolate_mode_t)0x1)
226 /* Isolate unmapped file */
227 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
228 /* Isolate for asynchronous migration */
229 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
230 /* Isolate unevictable pages */
231 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
233 /* LRU Isolation modes. */
234 typedef unsigned __bitwise__ isolate_mode_t
;
236 enum zone_watermarks
{
243 #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
244 #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
245 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
247 struct per_cpu_pages
{
248 int count
; /* number of pages in the list */
249 int high
; /* high watermark, emptying needed */
250 int batch
; /* chunk size for buddy add/remove */
252 /* Lists of pages, one per migrate type stored on the pcp-lists */
253 struct list_head lists
[MIGRATE_PCPTYPES
];
256 struct per_cpu_pageset
{
257 struct per_cpu_pages pcp
;
263 s8 vm_stat_diff
[NR_VM_ZONE_STAT_ITEMS
];
267 #endif /* !__GENERATING_BOUNDS.H */
270 #ifdef CONFIG_ZONE_DMA
272 * ZONE_DMA is used when there are devices that are not able
273 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
274 * carve out the portion of memory that is needed for these devices.
275 * The range is arch specific.
280 * ---------------------------
281 * parisc, ia64, sparc <4G
284 * alpha Unlimited or 0-16MB.
286 * i386, x86_64 and multiple other arches
291 #ifdef CONFIG_ZONE_DMA32
293 * x86_64 needs two ZONE_DMAs because it supports devices that are
294 * only able to do DMA to the lower 16M but also 32 bit devices that
295 * can only do DMA areas below 4G.
300 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
301 * performed on pages in ZONE_NORMAL if the DMA devices support
302 * transfers to all addressable memory.
305 #ifdef CONFIG_HIGHMEM
307 * A memory area that is only addressable by the kernel through
308 * mapping portions into its own address space. This is for example
309 * used by i386 to allow the kernel to address the memory beyond
310 * 900MB. The kernel will set up special mappings (page
311 * table entries on i386) for each page that the kernel needs to
320 #ifndef __GENERATING_BOUNDS_H
323 /* Fields commonly accessed by the page allocator */
325 /* zone watermarks, access with *_wmark_pages(zone) macros */
326 unsigned long watermark
[NR_WMARK
];
329 * When free pages are below this point, additional steps are taken
330 * when reading the number of free pages to avoid per-cpu counter
331 * drift allowing watermarks to be breached
333 unsigned long percpu_drift_mark
;
336 * We don't know if the memory that we're going to allocate will be freeable
337 * or/and it will be released eventually, so to avoid totally wasting several
338 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
339 * to run OOM on the lower zones despite there's tons of freeable ram
340 * on the higher zones). This array is recalculated at runtime if the
341 * sysctl_lowmem_reserve_ratio sysctl changes.
343 unsigned long lowmem_reserve
[MAX_NR_ZONES
];
346 * This is a per-zone reserve of pages that should not be
347 * considered dirtyable memory.
349 unsigned long dirty_balance_reserve
;
354 * zone reclaim becomes active if more unmapped pages exist.
356 unsigned long min_unmapped_pages
;
357 unsigned long min_slab_pages
;
359 struct per_cpu_pageset __percpu
*pageset
;
361 * free areas of different sizes
364 int all_unreclaimable
; /* All pages pinned */
365 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
366 /* Set to true when the PG_migrate_skip bits should be cleared */
367 bool compact_blockskip_flush
;
369 /* pfns where compaction scanners should start */
370 unsigned long compact_cached_free_pfn
;
371 unsigned long compact_cached_migrate_pfn
;
373 #ifdef CONFIG_MEMORY_HOTPLUG
374 /* see spanned/present_pages for more description */
375 seqlock_t span_seqlock
;
377 struct free_area free_area
[MAX_ORDER
];
379 #ifndef CONFIG_SPARSEMEM
381 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
382 * In SPARSEMEM, this map is stored in struct mem_section
384 unsigned long *pageblock_flags
;
385 #endif /* CONFIG_SPARSEMEM */
387 #ifdef CONFIG_COMPACTION
389 * On compaction failure, 1<<compact_defer_shift compactions
390 * are skipped before trying again. The number attempted since
391 * last failure is tracked with compact_considered.
393 unsigned int compact_considered
;
394 unsigned int compact_defer_shift
;
395 int compact_order_failed
;
400 /* Fields commonly accessed by the page reclaim scanner */
402 struct lruvec lruvec
;
404 unsigned long pages_scanned
; /* since last reclaim */
405 unsigned long flags
; /* zone flags, see below */
407 /* Zone statistics */
408 atomic_long_t vm_stat
[NR_VM_ZONE_STAT_ITEMS
];
411 * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on
412 * this zone's LRU. Maintained by the pageout code.
414 unsigned int inactive_ratio
;
418 /* Rarely used or read-mostly fields */
421 * wait_table -- the array holding the hash table
422 * wait_table_hash_nr_entries -- the size of the hash table array
423 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
425 * The purpose of all these is to keep track of the people
426 * waiting for a page to become available and make them
427 * runnable again when possible. The trouble is that this
428 * consumes a lot of space, especially when so few things
429 * wait on pages at a given time. So instead of using
430 * per-page waitqueues, we use a waitqueue hash table.
432 * The bucket discipline is to sleep on the same queue when
433 * colliding and wake all in that wait queue when removing.
434 * When something wakes, it must check to be sure its page is
435 * truly available, a la thundering herd. The cost of a
436 * collision is great, but given the expected load of the
437 * table, they should be so rare as to be outweighed by the
438 * benefits from the saved space.
440 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
441 * primary users of these fields, and in mm/page_alloc.c
442 * free_area_init_core() performs the initialization of them.
444 wait_queue_head_t
* wait_table
;
445 unsigned long wait_table_hash_nr_entries
;
446 unsigned long wait_table_bits
;
449 * Discontig memory support fields.
451 struct pglist_data
*zone_pgdat
;
452 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
453 unsigned long zone_start_pfn
;
456 * spanned_pages is the total pages spanned by the zone, including
457 * holes, which is calculated as:
458 * spanned_pages = zone_end_pfn - zone_start_pfn;
460 * present_pages is physical pages existing within the zone, which
462 * present_pages = spanned_pages - absent_pages(pages in holes);
464 * managed_pages is present pages managed by the buddy system, which
465 * is calculated as (reserved_pages includes pages allocated by the
466 * bootmem allocator):
467 * managed_pages = present_pages - reserved_pages;
469 * So present_pages may be used by memory hotplug or memory power
470 * management logic to figure out unmanaged pages by checking
471 * (present_pages - managed_pages). And managed_pages should be used
472 * by page allocator and vm scanner to calculate all kinds of watermarks
477 * zone_start_pfn and spanned_pages are protected by span_seqlock.
478 * It is a seqlock because it has to be read outside of zone->lock,
479 * and it is done in the main allocator path. But, it is written
480 * quite infrequently.
482 * The span_seq lock is declared along with zone->lock because it is
483 * frequently read in proximity to zone->lock. It's good to
484 * give them a chance of being in the same cacheline.
486 * Write access to present_pages and managed_pages at runtime should
487 * be protected by lock_memory_hotplug()/unlock_memory_hotplug().
488 * Any reader who can't tolerant drift of present_pages and
489 * managed_pages should hold memory hotplug lock to get a stable value.
491 unsigned long spanned_pages
;
492 unsigned long present_pages
;
493 unsigned long managed_pages
;
496 * rarely used fields:
499 } ____cacheline_internodealigned_in_smp
;
502 ZONE_RECLAIM_LOCKED
, /* prevents concurrent reclaim */
503 ZONE_OOM_LOCKED
, /* zone is in OOM killer zonelist */
504 ZONE_CONGESTED
, /* zone has many dirty pages backed by
509 static inline void zone_set_flag(struct zone
*zone
, zone_flags_t flag
)
511 set_bit(flag
, &zone
->flags
);
514 static inline int zone_test_and_set_flag(struct zone
*zone
, zone_flags_t flag
)
516 return test_and_set_bit(flag
, &zone
->flags
);
519 static inline void zone_clear_flag(struct zone
*zone
, zone_flags_t flag
)
521 clear_bit(flag
, &zone
->flags
);
524 static inline int zone_is_reclaim_congested(const struct zone
*zone
)
526 return test_bit(ZONE_CONGESTED
, &zone
->flags
);
529 static inline int zone_is_reclaim_locked(const struct zone
*zone
)
531 return test_bit(ZONE_RECLAIM_LOCKED
, &zone
->flags
);
534 static inline int zone_is_oom_locked(const struct zone
*zone
)
536 return test_bit(ZONE_OOM_LOCKED
, &zone
->flags
);
539 static inline unsigned long zone_end_pfn(const struct zone
*zone
)
541 return zone
->zone_start_pfn
+ zone
->spanned_pages
;
544 static inline bool zone_spans_pfn(const struct zone
*zone
, unsigned long pfn
)
546 return zone
->zone_start_pfn
<= pfn
&& pfn
< zone_end_pfn(zone
);
549 static inline bool zone_is_initialized(struct zone
*zone
)
551 return !!zone
->wait_table
;
554 static inline bool zone_is_empty(struct zone
*zone
)
556 return zone
->spanned_pages
== 0;
560 * The "priority" of VM scanning is how much of the queues we will scan in one
561 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
562 * queues ("queue_length >> 12") during an aging round.
564 #define DEF_PRIORITY 12
566 /* Maximum number of zones on a zonelist */
567 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
572 * The NUMA zonelists are doubled because we need zonelists that restrict the
573 * allocations to a single node for GFP_THISNODE.
575 * [0] : Zonelist with fallback
576 * [1] : No fallback (GFP_THISNODE)
578 #define MAX_ZONELISTS 2
582 * We cache key information from each zonelist for smaller cache
583 * footprint when scanning for free pages in get_page_from_freelist().
585 * 1) The BITMAP fullzones tracks which zones in a zonelist have come
586 * up short of free memory since the last time (last_fullzone_zap)
587 * we zero'd fullzones.
588 * 2) The array z_to_n[] maps each zone in the zonelist to its node
589 * id, so that we can efficiently evaluate whether that node is
590 * set in the current tasks mems_allowed.
592 * Both fullzones and z_to_n[] are one-to-one with the zonelist,
593 * indexed by a zones offset in the zonelist zones[] array.
595 * The get_page_from_freelist() routine does two scans. During the
596 * first scan, we skip zones whose corresponding bit in 'fullzones'
597 * is set or whose corresponding node in current->mems_allowed (which
598 * comes from cpusets) is not set. During the second scan, we bypass
599 * this zonelist_cache, to ensure we look methodically at each zone.
601 * Once per second, we zero out (zap) fullzones, forcing us to
602 * reconsider nodes that might have regained more free memory.
603 * The field last_full_zap is the time we last zapped fullzones.
605 * This mechanism reduces the amount of time we waste repeatedly
606 * reexaming zones for free memory when they just came up low on
607 * memory momentarilly ago.
609 * The zonelist_cache struct members logically belong in struct
610 * zonelist. However, the mempolicy zonelists constructed for
611 * MPOL_BIND are intentionally variable length (and usually much
612 * shorter). A general purpose mechanism for handling structs with
613 * multiple variable length members is more mechanism than we want
614 * here. We resort to some special case hackery instead.
616 * The MPOL_BIND zonelists don't need this zonelist_cache (in good
617 * part because they are shorter), so we put the fixed length stuff
618 * at the front of the zonelist struct, ending in a variable length
619 * zones[], as is needed by MPOL_BIND.
621 * Then we put the optional zonelist cache on the end of the zonelist
622 * struct. This optional stuff is found by a 'zlcache_ptr' pointer in
623 * the fixed length portion at the front of the struct. This pointer
624 * both enables us to find the zonelist cache, and in the case of
625 * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL)
626 * to know that the zonelist cache is not there.
628 * The end result is that struct zonelists come in two flavors:
629 * 1) The full, fixed length version, shown below, and
630 * 2) The custom zonelists for MPOL_BIND.
631 * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache.
633 * Even though there may be multiple CPU cores on a node modifying
634 * fullzones or last_full_zap in the same zonelist_cache at the same
635 * time, we don't lock it. This is just hint data - if it is wrong now
636 * and then, the allocator will still function, perhaps a bit slower.
640 struct zonelist_cache
{
641 unsigned short z_to_n
[MAX_ZONES_PER_ZONELIST
]; /* zone->nid */
642 DECLARE_BITMAP(fullzones
, MAX_ZONES_PER_ZONELIST
); /* zone full? */
643 unsigned long last_full_zap
; /* when last zap'd (jiffies) */
646 #define MAX_ZONELISTS 1
647 struct zonelist_cache
;
651 * This struct contains information about a zone in a zonelist. It is stored
652 * here to avoid dereferences into large structures and lookups of tables
655 struct zone
*zone
; /* Pointer to actual zone */
656 int zone_idx
; /* zone_idx(zoneref->zone) */
660 * One allocation request operates on a zonelist. A zonelist
661 * is a list of zones, the first one is the 'goal' of the
662 * allocation, the other zones are fallback zones, in decreasing
665 * If zlcache_ptr is not NULL, then it is just the address of zlcache,
666 * as explained above. If zlcache_ptr is NULL, there is no zlcache.
668 * To speed the reading of the zonelist, the zonerefs contain the zone index
669 * of the entry being read. Helper functions to access information given
670 * a struct zoneref are
672 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
673 * zonelist_zone_idx() - Return the index of the zone for an entry
674 * zonelist_node_idx() - Return the index of the node for an entry
677 struct zonelist_cache
*zlcache_ptr
; // NULL or &zlcache
678 struct zoneref _zonerefs
[MAX_ZONES_PER_ZONELIST
+ 1];
680 struct zonelist_cache zlcache
; // optional ...
684 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
685 struct node_active_region
{
686 unsigned long start_pfn
;
687 unsigned long end_pfn
;
690 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
692 #ifndef CONFIG_DISCONTIGMEM
693 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
694 extern struct page
*mem_map
;
698 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
699 * (mostly NUMA machines?) to denote a higher-level memory zone than the
702 * On NUMA machines, each NUMA node would have a pg_data_t to describe
703 * it's memory layout.
705 * Memory statistics and page replacement data structures are maintained on a
709 typedef struct pglist_data
{
710 struct zone node_zones
[MAX_NR_ZONES
];
711 struct zonelist node_zonelists
[MAX_ZONELISTS
];
713 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
714 struct page
*node_mem_map
;
716 struct page_cgroup
*node_page_cgroup
;
719 #ifndef CONFIG_NO_BOOTMEM
720 struct bootmem_data
*bdata
;
722 #ifdef CONFIG_MEMORY_HOTPLUG
724 * Must be held any time you expect node_start_pfn, node_present_pages
725 * or node_spanned_pages stay constant. Holding this will also
726 * guarantee that any pfn_valid() stays that way.
728 * Nests above zone->lock and zone->size_seqlock.
730 spinlock_t node_size_lock
;
732 unsigned long node_start_pfn
;
733 unsigned long node_present_pages
; /* total number of physical pages */
734 unsigned long node_spanned_pages
; /* total size of physical page
735 range, including holes */
737 nodemask_t reclaim_nodes
; /* Nodes allowed to reclaim from */
738 wait_queue_head_t kswapd_wait
;
739 wait_queue_head_t pfmemalloc_wait
;
740 struct task_struct
*kswapd
; /* Protected by lock_memory_hotplug() */
741 int kswapd_max_order
;
742 enum zone_type classzone_idx
;
743 #ifdef CONFIG_NUMA_BALANCING
745 * Lock serializing the per destination node AutoNUMA memory
746 * migration rate limiting data.
748 spinlock_t numabalancing_migrate_lock
;
750 /* Rate limiting time interval */
751 unsigned long numabalancing_migrate_next_window
;
753 /* Number of pages migrated during the rate limiting time interval */
754 unsigned long numabalancing_migrate_nr_pages
;
758 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
759 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
760 #ifdef CONFIG_FLAT_NODE_MEM_MAP
761 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
763 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
765 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
767 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
768 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
770 static inline unsigned long pgdat_end_pfn(pg_data_t
*pgdat
)
772 return pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
775 static inline bool pgdat_is_empty(pg_data_t
*pgdat
)
777 return !pgdat
->node_start_pfn
&& !pgdat
->node_spanned_pages
;
780 #include <linux/memory_hotplug.h>
782 extern struct mutex zonelists_mutex
;
783 void build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
);
784 void wakeup_kswapd(struct zone
*zone
, int order
, enum zone_type classzone_idx
);
785 bool zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
786 int classzone_idx
, int alloc_flags
);
787 bool zone_watermark_ok_safe(struct zone
*z
, int order
, unsigned long mark
,
788 int classzone_idx
, int alloc_flags
);
789 enum memmap_context
{
793 extern int init_currently_empty_zone(struct zone
*zone
, unsigned long start_pfn
,
795 enum memmap_context context
);
797 extern void lruvec_init(struct lruvec
*lruvec
);
799 static inline struct zone
*lruvec_zone(struct lruvec
*lruvec
)
804 return container_of(lruvec
, struct zone
, lruvec
);
808 #ifdef CONFIG_HAVE_MEMORY_PRESENT
809 void memory_present(int nid
, unsigned long start
, unsigned long end
);
811 static inline void memory_present(int nid
, unsigned long start
, unsigned long end
) {}
814 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
815 int local_memory_node(int node_id
);
817 static inline int local_memory_node(int node_id
) { return node_id
; };
820 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
821 unsigned long __init
node_memmap_size_bytes(int, unsigned long, unsigned long);
825 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
827 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
829 static inline int populated_zone(struct zone
*zone
)
831 return (!!zone
->present_pages
);
834 extern int movable_zone
;
836 static inline int zone_movable_is_highmem(void)
838 #if defined(CONFIG_HIGHMEM) && defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
839 return movable_zone
== ZONE_HIGHMEM
;
845 static inline int is_highmem_idx(enum zone_type idx
)
847 #ifdef CONFIG_HIGHMEM
848 return (idx
== ZONE_HIGHMEM
||
849 (idx
== ZONE_MOVABLE
&& zone_movable_is_highmem()));
855 static inline int is_normal_idx(enum zone_type idx
)
857 return (idx
== ZONE_NORMAL
);
861 * is_highmem - helper function to quickly check if a struct zone is a
862 * highmem zone or not. This is an attempt to keep references
863 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
864 * @zone - pointer to struct zone variable
866 static inline int is_highmem(struct zone
*zone
)
868 #ifdef CONFIG_HIGHMEM
869 int zone_off
= (char *)zone
- (char *)zone
->zone_pgdat
->node_zones
;
870 return zone_off
== ZONE_HIGHMEM
* sizeof(*zone
) ||
871 (zone_off
== ZONE_MOVABLE
* sizeof(*zone
) &&
872 zone_movable_is_highmem());
878 static inline int is_normal(struct zone
*zone
)
880 return zone
== zone
->zone_pgdat
->node_zones
+ ZONE_NORMAL
;
883 static inline int is_dma32(struct zone
*zone
)
885 #ifdef CONFIG_ZONE_DMA32
886 return zone
== zone
->zone_pgdat
->node_zones
+ ZONE_DMA32
;
892 static inline int is_dma(struct zone
*zone
)
894 #ifdef CONFIG_ZONE_DMA
895 return zone
== zone
->zone_pgdat
->node_zones
+ ZONE_DMA
;
901 /* These two functions are used to setup the per zone pages min values */
903 int min_free_kbytes_sysctl_handler(struct ctl_table
*, int,
904 void __user
*, size_t *, loff_t
*);
905 extern int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1];
906 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table
*, int,
907 void __user
*, size_t *, loff_t
*);
908 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table
*, int,
909 void __user
*, size_t *, loff_t
*);
910 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table
*, int,
911 void __user
*, size_t *, loff_t
*);
912 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table
*, int,
913 void __user
*, size_t *, loff_t
*);
915 extern int numa_zonelist_order_handler(struct ctl_table
*, int,
916 void __user
*, size_t *, loff_t
*);
917 extern char numa_zonelist_order
[];
918 #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
920 #ifndef CONFIG_NEED_MULTIPLE_NODES
922 extern struct pglist_data contig_page_data
;
923 #define NODE_DATA(nid) (&contig_page_data)
924 #define NODE_MEM_MAP(nid) mem_map
926 #else /* CONFIG_NEED_MULTIPLE_NODES */
928 #include <asm/mmzone.h>
930 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
932 #ifdef CONFIG_MTKPASR
933 #ifdef CONFIG_HIGHMEM
934 #define MTKPASR_ZONE (NODE_DATA(0)->node_zones + ZONE_HIGHMEM)
936 #define MTKPASR_ZONE (NODE_DATA(0)->node_zones)
940 extern struct pglist_data
*first_online_pgdat(void);
941 extern struct pglist_data
*next_online_pgdat(struct pglist_data
*pgdat
);
942 extern struct zone
*next_zone(struct zone
*zone
);
945 * for_each_online_pgdat - helper macro to iterate over all online nodes
946 * @pgdat - pointer to a pg_data_t variable
948 #define for_each_online_pgdat(pgdat) \
949 for (pgdat = first_online_pgdat(); \
951 pgdat = next_online_pgdat(pgdat))
953 * for_each_zone - helper macro to iterate over all memory zones
954 * @zone - pointer to struct zone variable
956 * The user only needs to declare the zone variable, for_each_zone
959 #define for_each_zone(zone) \
960 for (zone = (first_online_pgdat())->node_zones; \
962 zone = next_zone(zone))
964 #define for_each_populated_zone(zone) \
965 for (zone = (first_online_pgdat())->node_zones; \
967 zone = next_zone(zone)) \
968 if (!populated_zone(zone)) \
972 static inline struct zone
*zonelist_zone(struct zoneref
*zoneref
)
974 return zoneref
->zone
;
977 static inline int zonelist_zone_idx(struct zoneref
*zoneref
)
979 return zoneref
->zone_idx
;
982 static inline int zonelist_node_idx(struct zoneref
*zoneref
)
985 /* zone_to_nid not available in this context */
986 return zoneref
->zone
->node
;
989 #endif /* CONFIG_NUMA */
993 * 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
994 * @z - The cursor used as a starting point for the search
995 * @highest_zoneidx - The zone index of the highest zone to return
996 * @nodes - An optional nodemask to filter the zonelist with
997 * @zone - The first suitable zone found is returned via this parameter
999 * This function returns the next zone at or below a given zone index that is
1000 * within the allowed nodemask using a cursor as the starting point for the
1001 * search. The zoneref returned is a cursor that represents the current zone
1002 * being examined. It should be advanced by one before calling
1003 * next_zones_zonelist again.
1005 struct zoneref
*next_zones_zonelist(struct zoneref
*z
,
1006 enum zone_type highest_zoneidx
,
1008 struct zone
**zone
);
1011 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1012 * @zonelist - The zonelist to search for a suitable zone
1013 * @highest_zoneidx - The zone index of the highest zone to return
1014 * @nodes - An optional nodemask to filter the zonelist with
1015 * @zone - The first suitable zone found is returned via this parameter
1017 * This function returns the first zone at or below a given zone index that is
1018 * within the allowed nodemask. The zoneref returned is a cursor that can be
1019 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1020 * one before calling.
1022 static inline struct zoneref
*first_zones_zonelist(struct zonelist
*zonelist
,
1023 enum zone_type highest_zoneidx
,
1027 return next_zones_zonelist(zonelist
->_zonerefs
, highest_zoneidx
, nodes
,
1032 * 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
1033 * @zone - The current zone in the iterator
1034 * @z - The current pointer within zonelist->zones being iterated
1035 * @zlist - The zonelist being iterated
1036 * @highidx - The zone index of the highest zone to return
1037 * @nodemask - Nodemask allowed by the allocator
1039 * This iterator iterates though all zones at or below a given zone index and
1040 * within a given nodemask
1042 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1043 for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \
1045 z = next_zones_zonelist(++z, highidx, nodemask, &zone)) \
1048 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1049 * @zone - The current zone in the iterator
1050 * @z - The current pointer within zonelist->zones being iterated
1051 * @zlist - The zonelist being iterated
1052 * @highidx - The zone index of the highest zone to return
1054 * This iterator iterates though all zones at or below a given zone index.
1056 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1057 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1059 #ifdef CONFIG_SPARSEMEM
1060 #include <asm/sparsemem.h>
1063 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1064 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1065 static inline unsigned long early_pfn_to_nid(unsigned long pfn
)
1071 #ifdef CONFIG_FLATMEM
1072 #define pfn_to_nid(pfn) (0)
1075 #ifdef CONFIG_SPARSEMEM
1078 * SECTION_SHIFT #bits space required to store a section #
1080 * PA_SECTION_SHIFT physical address to/from section number
1081 * PFN_SECTION_SHIFT pfn to/from section number
1083 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1084 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1086 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1088 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1089 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1091 #define SECTION_BLOCKFLAGS_BITS \
1092 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1094 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1095 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1098 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
1099 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
1101 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1102 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1106 struct mem_section
{
1108 * This is, logically, a pointer to an array of struct
1109 * pages. However, it is stored with some other magic.
1110 * (see sparse.c::sparse_init_one_section())
1112 * Additionally during early boot we encode node id of
1113 * the location of the section here to guide allocation.
1114 * (see sparse.c::memory_present())
1116 * Making it a UL at least makes someone do a cast
1117 * before using it wrong.
1119 unsigned long section_mem_map
;
1121 /* See declaration of similar field in struct zone */
1122 unsigned long *pageblock_flags
;
1125 * If !SPARSEMEM, pgdat doesn't have page_cgroup pointer. We use
1126 * section. (see memcontrol.h/page_cgroup.h about this.)
1128 struct page_cgroup
*page_cgroup
;
1133 #ifdef CONFIG_SPARSEMEM_EXTREME
1134 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1136 #define SECTIONS_PER_ROOT 1
1139 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1140 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1141 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1143 #ifdef CONFIG_SPARSEMEM_EXTREME
1144 extern struct mem_section
*mem_section
[NR_SECTION_ROOTS
];
1146 extern struct mem_section mem_section
[NR_SECTION_ROOTS
][SECTIONS_PER_ROOT
];
1149 static inline struct mem_section
*__nr_to_section(unsigned long nr
)
1151 if (!mem_section
[SECTION_NR_TO_ROOT(nr
)])
1153 return &mem_section
[SECTION_NR_TO_ROOT(nr
)][nr
& SECTION_ROOT_MASK
];
1155 extern int __section_nr(struct mem_section
* ms
);
1156 extern unsigned long usemap_size(void);
1159 * We use the lower bits of the mem_map pointer to store
1160 * a little bit of information. There should be at least
1161 * 3 bits here due to 32-bit alignment.
1163 #define SECTION_MARKED_PRESENT (1UL<<0)
1164 #define SECTION_HAS_MEM_MAP (1UL<<1)
1165 #define SECTION_MAP_LAST_BIT (1UL<<2)
1166 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1167 #define SECTION_NID_SHIFT 2
1169 static inline struct page
*__section_mem_map_addr(struct mem_section
*section
)
1171 unsigned long map
= section
->section_mem_map
;
1172 map
&= SECTION_MAP_MASK
;
1173 return (struct page
*)map
;
1176 static inline int present_section(struct mem_section
*section
)
1178 return (section
&& (section
->section_mem_map
& SECTION_MARKED_PRESENT
));
1181 static inline int present_section_nr(unsigned long nr
)
1183 return present_section(__nr_to_section(nr
));
1186 static inline int valid_section(struct mem_section
*section
)
1188 return (section
&& (section
->section_mem_map
& SECTION_HAS_MEM_MAP
));
1191 static inline int valid_section_nr(unsigned long nr
)
1193 return valid_section(__nr_to_section(nr
));
1196 static inline struct mem_section
*__pfn_to_section(unsigned long pfn
)
1198 return __nr_to_section(pfn_to_section_nr(pfn
));
1201 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1202 static inline int pfn_valid(unsigned long pfn
)
1204 if (pfn_to_section_nr(pfn
) >= NR_MEM_SECTIONS
)
1206 return valid_section(__nr_to_section(pfn_to_section_nr(pfn
)));
1210 static inline int pfn_present(unsigned long pfn
)
1212 if (pfn_to_section_nr(pfn
) >= NR_MEM_SECTIONS
)
1214 return present_section(__nr_to_section(pfn_to_section_nr(pfn
)));
1218 * These are _only_ used during initialisation, therefore they
1219 * can use __initdata ... They could have names to indicate
1223 #define pfn_to_nid(pfn) \
1225 unsigned long __pfn_to_nid_pfn = (pfn); \
1226 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1229 #define pfn_to_nid(pfn) (0)
1232 #define early_pfn_valid(pfn) pfn_valid(pfn)
1233 void sparse_init(void);
1235 #define sparse_init() do {} while (0)
1236 #define sparse_index_init(_sec, _nid) do {} while (0)
1237 #endif /* CONFIG_SPARSEMEM */
1239 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1240 bool early_pfn_in_nid(unsigned long pfn
, int nid
);
1242 #define early_pfn_in_nid(pfn, nid) (1)
1245 #ifndef early_pfn_valid
1246 #define early_pfn_valid(pfn) (1)
1249 void memory_present(int nid
, unsigned long start
, unsigned long end
);
1250 unsigned long __init
node_memmap_size_bytes(int, unsigned long, unsigned long);
1253 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1254 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1255 * pfn_valid_within() should be used in this case; we optimise this away
1256 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1258 #ifdef CONFIG_HOLES_IN_ZONE
1259 #define pfn_valid_within(pfn) pfn_valid(pfn)
1261 #define pfn_valid_within(pfn) (1)
1264 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1266 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1267 * associated with it or not. In FLATMEM, it is expected that holes always
1268 * have valid memmap as long as there is valid PFNs either side of the hole.
1269 * In SPARSEMEM, it is assumed that a valid section has a memmap for the
1272 * However, an ARM, and maybe other embedded architectures in the future
1273 * free memmap backing holes to save memory on the assumption the memmap is
1274 * never used. The page_zone linkages are then broken even though pfn_valid()
1275 * returns true. A walker of the full memmap must then do this additional
1276 * check to ensure the memmap they are looking at is sane by making sure
1277 * the zone and PFN linkages are still valid. This is expensive, but walkers
1278 * of the full memmap are extremely rare.
1280 int memmap_valid_within(unsigned long pfn
,
1281 struct page
*page
, struct zone
*zone
);
1283 static inline int memmap_valid_within(unsigned long pfn
,
1284 struct page
*page
, struct zone
*zone
)
1288 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1290 #endif /* !__GENERATING_BOUNDS.H */
1291 #endif /* !__ASSEMBLY__ */
1292 #endif /* _LINUX_MMZONE_H */