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
42 MIGRATE_PCPTYPES
, /* the number of types on the pcp lists */
43 MIGRATE_RESERVE
= MIGRATE_PCPTYPES
,
46 * MIGRATE_CMA migration type is designed to mimic the way
47 * ZONE_MOVABLE works. Only movable pages can be allocated
48 * from MIGRATE_CMA pageblocks and page allocator never
49 * implicitly change migration type of MIGRATE_CMA pageblock.
51 * The way to use it is to change migratetype of a range of
52 * pageblocks to MIGRATE_CMA which can be done by
53 * __free_pageblock_cma() function. What is important though
54 * is that a range of pageblocks must be aligned to
55 * MAX_ORDER_NR_PAGES should biggest page be bigger then
60 #ifdef CONFIG_MEMORY_ISOLATION
61 MIGRATE_ISOLATE
, /* can't allocate from here */
67 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
69 # define is_migrate_cma(migratetype) false
72 #define for_each_migratetype_order(order, type) \
73 for (order = 0; order < MAX_ORDER; order++) \
74 for (type = 0; type < MIGRATE_TYPES; type++)
76 extern int page_group_by_mobility_disabled
;
78 #define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1)
79 #define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1)
81 #define get_pageblock_migratetype(page) \
82 get_pfnblock_flags_mask(page, page_to_pfn(page), \
83 PB_migrate_end, MIGRATETYPE_MASK)
85 static inline int get_pfnblock_migratetype(struct page
*page
, unsigned long pfn
)
87 BUILD_BUG_ON(PB_migrate_end
- PB_migrate
!= 2);
88 return get_pfnblock_flags_mask(page
, pfn
, 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) */
119 NR_INACTIVE_ANON
= NR_LRU_BASE
, /* must match order of LRU_[IN]ACTIVE */
120 NR_ACTIVE_ANON
, /* " " " " " */
121 NR_INACTIVE_FILE
, /* " " " " " */
122 NR_ACTIVE_FILE
, /* " " " " " */
123 NR_UNEVICTABLE
, /* " " " " " */
124 NR_MLOCK
, /* mlock()ed pages found and moved off LRU */
125 NR_ANON_PAGES
, /* Mapped anonymous pages */
126 NR_FILE_MAPPED
, /* pagecache pages mapped into pagetables.
127 only modified from process context */
132 NR_SLAB_UNRECLAIMABLE
,
133 NR_PAGETABLE
, /* used for pagetables */
135 /* Second 128 byte cacheline */
136 NR_UNSTABLE_NFS
, /* NFS unstable pages */
139 NR_VMSCAN_IMMEDIATE
, /* Prioritise for reclaim when writeback ends */
140 NR_WRITEBACK_TEMP
, /* Writeback using temporary buffers */
141 NR_ISOLATED_ANON
, /* Temporary isolated pages from anon lru */
142 NR_ISOLATED_FILE
, /* Temporary isolated pages from file lru */
143 NR_SHMEM
, /* shmem pages (included tmpfs/GEM pages) */
144 NR_DIRTIED
, /* page dirtyings since bootup */
145 NR_WRITTEN
, /* page writings since bootup */
146 NR_PAGES_SCANNED
, /* pages scanned since last reclaim */
148 NUMA_HIT
, /* allocated in intended node */
149 NUMA_MISS
, /* allocated in non intended node */
150 NUMA_FOREIGN
, /* was intended here, hit elsewhere */
151 NUMA_INTERLEAVE_HIT
, /* interleaver preferred this zone */
152 NUMA_LOCAL
, /* allocation from local node */
153 NUMA_OTHER
, /* allocation from other node */
157 WORKINGSET_NODERECLAIM
,
158 NR_ANON_TRANSPARENT_HUGEPAGES
,
160 NR_VM_ZONE_STAT_ITEMS
};
163 * We do arithmetic on the LRU lists in various places in the code,
164 * so it is important to keep the active lists LRU_ACTIVE higher in
165 * the array than the corresponding inactive lists, and to keep
166 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
168 * This has to be kept in sync with the statistics in zone_stat_item
169 * above and the descriptions in vmstat_text in mm/vmstat.c
176 LRU_INACTIVE_ANON
= LRU_BASE
,
177 LRU_ACTIVE_ANON
= LRU_BASE
+ LRU_ACTIVE
,
178 LRU_INACTIVE_FILE
= LRU_BASE
+ LRU_FILE
,
179 LRU_ACTIVE_FILE
= LRU_BASE
+ LRU_FILE
+ LRU_ACTIVE
,
184 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
186 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
188 static inline int is_file_lru(enum lru_list lru
)
190 return (lru
== LRU_INACTIVE_FILE
|| lru
== LRU_ACTIVE_FILE
);
193 static inline int is_active_lru(enum lru_list lru
)
195 return (lru
== LRU_ACTIVE_ANON
|| lru
== LRU_ACTIVE_FILE
);
198 static inline int is_unevictable_lru(enum lru_list lru
)
200 return (lru
== LRU_UNEVICTABLE
);
203 struct zone_reclaim_stat
{
205 * The pageout code in vmscan.c keeps track of how many of the
206 * mem/swap backed and file backed pages are referenced.
207 * The higher the rotated/scanned ratio, the more valuable
210 * The anon LRU stats live in [0], file LRU stats in [1]
212 unsigned long recent_rotated
[2];
213 unsigned long recent_scanned
[2];
217 struct list_head lists
[NR_LRU_LISTS
];
218 struct zone_reclaim_stat reclaim_stat
;
224 /* Mask used at gathering information at once (see memcontrol.c) */
225 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
226 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
227 #define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
229 /* Isolate clean file */
230 #define ISOLATE_CLEAN ((__force isolate_mode_t)0x1)
231 /* Isolate unmapped file */
232 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
233 /* Isolate for asynchronous migration */
234 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
235 /* Isolate unevictable pages */
236 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
238 /* LRU Isolation modes. */
239 typedef unsigned __bitwise__ isolate_mode_t
;
241 enum zone_watermarks
{
248 #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
249 #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
250 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
252 struct per_cpu_pages
{
253 int count
; /* number of pages in the list */
254 int high
; /* high watermark, emptying needed */
255 int batch
; /* chunk size for buddy add/remove */
257 /* Lists of pages, one per migrate type stored on the pcp-lists */
258 struct list_head lists
[MIGRATE_PCPTYPES
];
261 struct per_cpu_pageset
{
262 struct per_cpu_pages pcp
;
268 s8 vm_stat_diff
[NR_VM_ZONE_STAT_ITEMS
];
272 #endif /* !__GENERATING_BOUNDS.H */
275 #ifdef CONFIG_ZONE_DMA
277 * ZONE_DMA is used when there are devices that are not able
278 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
279 * carve out the portion of memory that is needed for these devices.
280 * The range is arch specific.
285 * ---------------------------
286 * parisc, ia64, sparc <4G
289 * alpha Unlimited or 0-16MB.
291 * i386, x86_64 and multiple other arches
296 #ifdef CONFIG_ZONE_DMA32
298 * x86_64 needs two ZONE_DMAs because it supports devices that are
299 * only able to do DMA to the lower 16M but also 32 bit devices that
300 * can only do DMA areas below 4G.
305 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
306 * performed on pages in ZONE_NORMAL if the DMA devices support
307 * transfers to all addressable memory.
310 #ifdef CONFIG_HIGHMEM
312 * A memory area that is only addressable by the kernel through
313 * mapping portions into its own address space. This is for example
314 * used by i386 to allow the kernel to address the memory beyond
315 * 900MB. The kernel will set up special mappings (page
316 * table entries on i386) for each page that the kernel needs to
325 #ifndef __GENERATING_BOUNDS_H
328 /* Read-mostly fields */
330 /* zone watermarks, access with *_wmark_pages(zone) macros */
331 unsigned long watermark
[NR_WMARK
];
334 * We don't know if the memory that we're going to allocate will be freeable
335 * or/and it will be released eventually, so to avoid totally wasting several
336 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
337 * to run OOM on the lower zones despite there's tons of freeable ram
338 * on the higher zones). This array is recalculated at runtime if the
339 * sysctl_lowmem_reserve_ratio sysctl changes.
341 long lowmem_reserve
[MAX_NR_ZONES
];
348 * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on
349 * this zone's LRU. Maintained by the pageout code.
351 unsigned int inactive_ratio
;
353 struct pglist_data
*zone_pgdat
;
354 struct per_cpu_pageset __percpu
*pageset
;
357 * This is a per-zone reserve of pages that should not be
358 * considered dirtyable memory.
360 unsigned long dirty_balance_reserve
;
362 #ifndef CONFIG_SPARSEMEM
364 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
365 * In SPARSEMEM, this map is stored in struct mem_section
367 unsigned long *pageblock_flags
;
368 #endif /* CONFIG_SPARSEMEM */
372 * zone reclaim becomes active if more unmapped pages exist.
374 unsigned long min_unmapped_pages
;
375 unsigned long min_slab_pages
;
376 #endif /* CONFIG_NUMA */
378 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
379 unsigned long zone_start_pfn
;
382 * spanned_pages is the total pages spanned by the zone, including
383 * holes, which is calculated as:
384 * spanned_pages = zone_end_pfn - zone_start_pfn;
386 * present_pages is physical pages existing within the zone, which
388 * present_pages = spanned_pages - absent_pages(pages in holes);
390 * managed_pages is present pages managed by the buddy system, which
391 * is calculated as (reserved_pages includes pages allocated by the
392 * bootmem allocator):
393 * managed_pages = present_pages - reserved_pages;
395 * So present_pages may be used by memory hotplug or memory power
396 * management logic to figure out unmanaged pages by checking
397 * (present_pages - managed_pages). And managed_pages should be used
398 * by page allocator and vm scanner to calculate all kinds of watermarks
403 * zone_start_pfn and spanned_pages are protected by span_seqlock.
404 * It is a seqlock because it has to be read outside of zone->lock,
405 * and it is done in the main allocator path. But, it is written
406 * quite infrequently.
408 * The span_seq lock is declared along with zone->lock because it is
409 * frequently read in proximity to zone->lock. It's good to
410 * give them a chance of being in the same cacheline.
412 * Write access to present_pages at runtime should be protected by
413 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
414 * present_pages should get_online_mems() to get a stable value.
416 * Read access to managed_pages should be safe because it's unsigned
417 * long. Write access to zone->managed_pages and totalram_pages are
418 * protected by managed_page_count_lock at runtime. Idealy only
419 * adjust_managed_page_count() should be used instead of directly
420 * touching zone->managed_pages and totalram_pages.
422 unsigned long managed_pages
;
423 unsigned long spanned_pages
;
424 unsigned long present_pages
;
429 * Number of MIGRATE_RESERVE page block. To maintain for just
430 * optimization. Protected by zone->lock.
432 int nr_migrate_reserve_block
;
434 #ifdef CONFIG_MEMORY_ISOLATION
436 * Number of isolated pageblock. It is used to solve incorrect
437 * freepage counting problem due to racy retrieving migratetype
438 * of pageblock. Protected by zone->lock.
440 unsigned long nr_isolate_pageblock
;
443 #ifdef CONFIG_MEMORY_HOTPLUG
444 /* see spanned/present_pages for more description */
445 seqlock_t span_seqlock
;
449 * wait_table -- the array holding the hash table
450 * wait_table_hash_nr_entries -- the size of the hash table array
451 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
453 * The purpose of all these is to keep track of the people
454 * waiting for a page to become available and make them
455 * runnable again when possible. The trouble is that this
456 * consumes a lot of space, especially when so few things
457 * wait on pages at a given time. So instead of using
458 * per-page waitqueues, we use a waitqueue hash table.
460 * The bucket discipline is to sleep on the same queue when
461 * colliding and wake all in that wait queue when removing.
462 * When something wakes, it must check to be sure its page is
463 * truly available, a la thundering herd. The cost of a
464 * collision is great, but given the expected load of the
465 * table, they should be so rare as to be outweighed by the
466 * benefits from the saved space.
468 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
469 * primary users of these fields, and in mm/page_alloc.c
470 * free_area_init_core() performs the initialization of them.
472 wait_queue_head_t
*wait_table
;
473 unsigned long wait_table_hash_nr_entries
;
474 unsigned long wait_table_bits
;
477 /* free areas of different sizes */
478 struct free_area free_area
[MAX_ORDER
];
480 /* zone flags, see below */
483 /* Write-intensive fields used from the page allocator */
488 /* Write-intensive fields used by page reclaim */
490 /* Fields commonly accessed by the page reclaim scanner */
492 struct lruvec lruvec
;
494 /* Evictions & activations on the inactive file list */
495 atomic_long_t inactive_age
;
498 * When free pages are below this point, additional steps are taken
499 * when reading the number of free pages to avoid per-cpu counter
500 * drift allowing watermarks to be breached
502 unsigned long percpu_drift_mark
;
504 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
505 /* pfn where compaction free scanner should start */
506 unsigned long compact_cached_free_pfn
;
507 /* pfn where async and sync compaction migration scanner should start */
508 unsigned long compact_cached_migrate_pfn
[2];
511 #ifdef CONFIG_COMPACTION
513 * On compaction failure, 1<<compact_defer_shift compactions
514 * are skipped before trying again. The number attempted since
515 * last failure is tracked with compact_considered.
517 unsigned int compact_considered
;
518 unsigned int compact_defer_shift
;
519 int compact_order_failed
;
522 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
523 /* Set to true when the PG_migrate_skip bits should be cleared */
524 bool compact_blockskip_flush
;
528 /* Zone statistics */
529 atomic_long_t vm_stat
[NR_VM_ZONE_STAT_ITEMS
];
530 } ____cacheline_internodealigned_in_smp
;
533 ZONE_RECLAIM_LOCKED
, /* prevents concurrent reclaim */
534 ZONE_OOM_LOCKED
, /* zone is in OOM killer zonelist */
535 ZONE_CONGESTED
, /* zone has many dirty pages backed by
538 ZONE_DIRTY
, /* reclaim scanning has recently found
539 * many dirty file pages at the tail
542 ZONE_WRITEBACK
, /* reclaim scanning has recently found
543 * many pages under writeback
545 ZONE_FAIR_DEPLETED
, /* fair zone policy batch depleted */
548 static inline unsigned long zone_end_pfn(const struct zone
*zone
)
550 return zone
->zone_start_pfn
+ zone
->spanned_pages
;
553 static inline bool zone_spans_pfn(const struct zone
*zone
, unsigned long pfn
)
555 return zone
->zone_start_pfn
<= pfn
&& pfn
< zone_end_pfn(zone
);
558 static inline bool zone_is_initialized(struct zone
*zone
)
560 return !!zone
->wait_table
;
563 static inline bool zone_is_empty(struct zone
*zone
)
565 return zone
->spanned_pages
== 0;
569 * The "priority" of VM scanning is how much of the queues we will scan in one
570 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
571 * queues ("queue_length >> 12") during an aging round.
573 #define DEF_PRIORITY 12
575 /* Maximum number of zones on a zonelist */
576 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
581 * The NUMA zonelists are doubled because we need zonelists that restrict the
582 * allocations to a single node for __GFP_THISNODE.
584 * [0] : Zonelist with fallback
585 * [1] : No fallback (__GFP_THISNODE)
587 #define MAX_ZONELISTS 2
591 * We cache key information from each zonelist for smaller cache
592 * footprint when scanning for free pages in get_page_from_freelist().
594 * 1) The BITMAP fullzones tracks which zones in a zonelist have come
595 * up short of free memory since the last time (last_fullzone_zap)
596 * we zero'd fullzones.
597 * 2) The array z_to_n[] maps each zone in the zonelist to its node
598 * id, so that we can efficiently evaluate whether that node is
599 * set in the current tasks mems_allowed.
601 * Both fullzones and z_to_n[] are one-to-one with the zonelist,
602 * indexed by a zones offset in the zonelist zones[] array.
604 * The get_page_from_freelist() routine does two scans. During the
605 * first scan, we skip zones whose corresponding bit in 'fullzones'
606 * is set or whose corresponding node in current->mems_allowed (which
607 * comes from cpusets) is not set. During the second scan, we bypass
608 * this zonelist_cache, to ensure we look methodically at each zone.
610 * Once per second, we zero out (zap) fullzones, forcing us to
611 * reconsider nodes that might have regained more free memory.
612 * The field last_full_zap is the time we last zapped fullzones.
614 * This mechanism reduces the amount of time we waste repeatedly
615 * reexaming zones for free memory when they just came up low on
616 * memory momentarilly ago.
618 * The zonelist_cache struct members logically belong in struct
619 * zonelist. However, the mempolicy zonelists constructed for
620 * MPOL_BIND are intentionally variable length (and usually much
621 * shorter). A general purpose mechanism for handling structs with
622 * multiple variable length members is more mechanism than we want
623 * here. We resort to some special case hackery instead.
625 * The MPOL_BIND zonelists don't need this zonelist_cache (in good
626 * part because they are shorter), so we put the fixed length stuff
627 * at the front of the zonelist struct, ending in a variable length
628 * zones[], as is needed by MPOL_BIND.
630 * Then we put the optional zonelist cache on the end of the zonelist
631 * struct. This optional stuff is found by a 'zlcache_ptr' pointer in
632 * the fixed length portion at the front of the struct. This pointer
633 * both enables us to find the zonelist cache, and in the case of
634 * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL)
635 * to know that the zonelist cache is not there.
637 * The end result is that struct zonelists come in two flavors:
638 * 1) The full, fixed length version, shown below, and
639 * 2) The custom zonelists for MPOL_BIND.
640 * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache.
642 * Even though there may be multiple CPU cores on a node modifying
643 * fullzones or last_full_zap in the same zonelist_cache at the same
644 * time, we don't lock it. This is just hint data - if it is wrong now
645 * and then, the allocator will still function, perhaps a bit slower.
649 struct zonelist_cache
{
650 unsigned short z_to_n
[MAX_ZONES_PER_ZONELIST
]; /* zone->nid */
651 DECLARE_BITMAP(fullzones
, MAX_ZONES_PER_ZONELIST
); /* zone full? */
652 unsigned long last_full_zap
; /* when last zap'd (jiffies) */
655 #define MAX_ZONELISTS 1
656 struct zonelist_cache
;
660 * This struct contains information about a zone in a zonelist. It is stored
661 * here to avoid dereferences into large structures and lookups of tables
664 struct zone
*zone
; /* Pointer to actual zone */
665 int zone_idx
; /* zone_idx(zoneref->zone) */
669 * One allocation request operates on a zonelist. A zonelist
670 * is a list of zones, the first one is the 'goal' of the
671 * allocation, the other zones are fallback zones, in decreasing
674 * If zlcache_ptr is not NULL, then it is just the address of zlcache,
675 * as explained above. If zlcache_ptr is NULL, there is no zlcache.
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 zonelist_cache
*zlcache_ptr
; // NULL or &zlcache
687 struct zoneref _zonerefs
[MAX_ZONES_PER_ZONELIST
+ 1];
689 struct zonelist_cache zlcache
; // optional ...
693 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
694 struct node_active_region
{
695 unsigned long start_pfn
;
696 unsigned long end_pfn
;
699 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
701 #ifndef CONFIG_DISCONTIGMEM
702 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
703 extern struct page
*mem_map
;
707 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
708 * (mostly NUMA machines?) to denote a higher-level memory zone than the
711 * On NUMA machines, each NUMA node would have a pg_data_t to describe
712 * it's memory layout.
714 * Memory statistics and page replacement data structures are maintained on a
718 typedef struct pglist_data
{
719 struct zone node_zones
[MAX_NR_ZONES
];
720 struct zonelist node_zonelists
[MAX_ZONELISTS
];
722 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
723 struct page
*node_mem_map
;
724 #ifdef CONFIG_PAGE_EXTENSION
725 struct page_ext
*node_page_ext
;
728 #ifndef CONFIG_NO_BOOTMEM
729 struct bootmem_data
*bdata
;
731 #ifdef CONFIG_MEMORY_HOTPLUG
733 * Must be held any time you expect node_start_pfn, node_present_pages
734 * or node_spanned_pages stay constant. Holding this will also
735 * guarantee that any pfn_valid() stays that way.
737 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
738 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG.
740 * Nests above zone->lock and zone->span_seqlock
742 spinlock_t node_size_lock
;
744 unsigned long node_start_pfn
;
745 unsigned long node_present_pages
; /* total number of physical pages */
746 unsigned long node_spanned_pages
; /* total size of physical page
747 range, including holes */
749 wait_queue_head_t kswapd_wait
;
750 wait_queue_head_t pfmemalloc_wait
;
751 struct task_struct
*kswapd
; /* Protected by
752 mem_hotplug_begin/end() */
753 int kswapd_max_order
;
754 enum zone_type classzone_idx
;
755 #ifdef CONFIG_NUMA_BALANCING
756 /* Lock serializing the migrate rate limiting window */
757 spinlock_t numabalancing_migrate_lock
;
759 /* Rate limiting time interval */
760 unsigned long numabalancing_migrate_next_window
;
762 /* Number of pages migrated during the rate limiting time interval */
763 unsigned long numabalancing_migrate_nr_pages
;
766 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
768 * If memory initialisation on large machines is deferred then this
769 * is the first PFN that needs to be initialised.
771 unsigned long first_deferred_pfn
;
772 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
775 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
776 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
777 #ifdef CONFIG_FLAT_NODE_MEM_MAP
778 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
780 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
782 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
784 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
785 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
787 static inline unsigned long pgdat_end_pfn(pg_data_t
*pgdat
)
789 return pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
792 static inline bool pgdat_is_empty(pg_data_t
*pgdat
)
794 return !pgdat
->node_start_pfn
&& !pgdat
->node_spanned_pages
;
797 #include <linux/memory_hotplug.h>
799 extern struct mutex zonelists_mutex
;
800 void build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
);
801 void wakeup_kswapd(struct zone
*zone
, int order
, enum zone_type classzone_idx
);
802 bool zone_watermark_ok(struct zone
*z
, unsigned int order
,
803 unsigned long mark
, int classzone_idx
, int alloc_flags
);
804 bool zone_watermark_ok_safe(struct zone
*z
, unsigned int order
,
805 unsigned long mark
, int classzone_idx
, int alloc_flags
);
806 enum memmap_context
{
810 extern int init_currently_empty_zone(struct zone
*zone
, unsigned long start_pfn
,
812 enum memmap_context context
);
814 extern void lruvec_init(struct lruvec
*lruvec
);
816 static inline struct zone
*lruvec_zone(struct lruvec
*lruvec
)
821 return container_of(lruvec
, struct zone
, lruvec
);
825 #ifdef CONFIG_HAVE_MEMORY_PRESENT
826 void memory_present(int nid
, unsigned long start
, unsigned long end
);
828 static inline void memory_present(int nid
, unsigned long start
, unsigned long end
) {}
831 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
832 int local_memory_node(int node_id
);
834 static inline int local_memory_node(int node_id
) { return node_id
; };
837 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
838 unsigned long __init
node_memmap_size_bytes(int, unsigned long, unsigned long);
842 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
844 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
846 static inline int populated_zone(struct zone
*zone
)
848 return (!!zone
->present_pages
);
851 extern int movable_zone
;
853 #ifdef CONFIG_HIGHMEM
854 static inline int zone_movable_is_highmem(void)
856 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
857 return movable_zone
== ZONE_HIGHMEM
;
859 return (ZONE_MOVABLE
- 1) == ZONE_HIGHMEM
;
864 static inline int is_highmem_idx(enum zone_type idx
)
866 #ifdef CONFIG_HIGHMEM
867 return (idx
== ZONE_HIGHMEM
||
868 (idx
== ZONE_MOVABLE
&& zone_movable_is_highmem()));
875 * is_highmem - helper function to quickly check if a struct zone is a
876 * highmem zone or not. This is an attempt to keep references
877 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
878 * @zone - pointer to struct zone variable
880 static inline int is_highmem(struct zone
*zone
)
882 #ifdef CONFIG_HIGHMEM
883 int zone_off
= (char *)zone
- (char *)zone
->zone_pgdat
->node_zones
;
884 return zone_off
== ZONE_HIGHMEM
* sizeof(*zone
) ||
885 (zone_off
== ZONE_MOVABLE
* sizeof(*zone
) &&
886 zone_movable_is_highmem());
892 /* These two functions are used to setup the per zone pages min values */
894 int min_free_kbytes_sysctl_handler(struct ctl_table
*, int,
895 void __user
*, size_t *, loff_t
*);
896 extern int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1];
897 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table
*, int,
898 void __user
*, size_t *, loff_t
*);
899 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table
*, int,
900 void __user
*, size_t *, loff_t
*);
901 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table
*, int,
902 void __user
*, size_t *, loff_t
*);
903 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table
*, int,
904 void __user
*, size_t *, loff_t
*);
906 extern int numa_zonelist_order_handler(struct ctl_table
*, int,
907 void __user
*, size_t *, loff_t
*);
908 extern char numa_zonelist_order
[];
909 #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
911 #ifndef CONFIG_NEED_MULTIPLE_NODES
913 extern struct pglist_data contig_page_data
;
914 #define NODE_DATA(nid) (&contig_page_data)
915 #define NODE_MEM_MAP(nid) mem_map
917 #else /* CONFIG_NEED_MULTIPLE_NODES */
919 #include <asm/mmzone.h>
921 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
923 extern struct pglist_data
*first_online_pgdat(void);
924 extern struct pglist_data
*next_online_pgdat(struct pglist_data
*pgdat
);
925 extern struct zone
*next_zone(struct zone
*zone
);
928 * for_each_online_pgdat - helper macro to iterate over all online nodes
929 * @pgdat - pointer to a pg_data_t variable
931 #define for_each_online_pgdat(pgdat) \
932 for (pgdat = first_online_pgdat(); \
934 pgdat = next_online_pgdat(pgdat))
936 * for_each_zone - helper macro to iterate over all memory zones
937 * @zone - pointer to struct zone variable
939 * The user only needs to declare the zone variable, for_each_zone
942 #define for_each_zone(zone) \
943 for (zone = (first_online_pgdat())->node_zones; \
945 zone = next_zone(zone))
947 #define for_each_populated_zone(zone) \
948 for (zone = (first_online_pgdat())->node_zones; \
950 zone = next_zone(zone)) \
951 if (!populated_zone(zone)) \
955 static inline struct zone
*zonelist_zone(struct zoneref
*zoneref
)
957 return zoneref
->zone
;
960 static inline int zonelist_zone_idx(struct zoneref
*zoneref
)
962 return zoneref
->zone_idx
;
965 static inline int zonelist_node_idx(struct zoneref
*zoneref
)
968 /* zone_to_nid not available in this context */
969 return zoneref
->zone
->node
;
972 #endif /* CONFIG_NUMA */
976 * 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
977 * @z - The cursor used as a starting point for the search
978 * @highest_zoneidx - The zone index of the highest zone to return
979 * @nodes - An optional nodemask to filter the zonelist with
981 * This function returns the next zone at or below a given zone index that is
982 * within the allowed nodemask using a cursor as the starting point for the
983 * search. The zoneref returned is a cursor that represents the current zone
984 * being examined. It should be advanced by one before calling
985 * next_zones_zonelist again.
987 struct zoneref
*next_zones_zonelist(struct zoneref
*z
,
988 enum zone_type highest_zoneidx
,
992 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
993 * @zonelist - The zonelist to search for a suitable zone
994 * @highest_zoneidx - The zone index of the highest zone to return
995 * @nodes - An optional nodemask to filter the zonelist with
996 * @zone - The first suitable zone found is returned via this parameter
998 * This function returns the first zone at or below a given zone index that is
999 * within the allowed nodemask. The zoneref returned is a cursor that can be
1000 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1001 * one before calling.
1003 static inline struct zoneref
*first_zones_zonelist(struct zonelist
*zonelist
,
1004 enum zone_type highest_zoneidx
,
1008 struct zoneref
*z
= next_zones_zonelist(zonelist
->_zonerefs
,
1009 highest_zoneidx
, nodes
);
1010 *zone
= zonelist_zone(z
);
1015 * 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
1016 * @zone - The current zone in the iterator
1017 * @z - The current pointer within zonelist->zones being iterated
1018 * @zlist - The zonelist being iterated
1019 * @highidx - The zone index of the highest zone to return
1020 * @nodemask - Nodemask allowed by the allocator
1022 * This iterator iterates though all zones at or below a given zone index and
1023 * within a given nodemask
1025 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1026 for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \
1028 z = next_zones_zonelist(++z, highidx, nodemask), \
1029 zone = zonelist_zone(z)) \
1032 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
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
1038 * This iterator iterates though all zones at or below a given zone index.
1040 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1041 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1043 #ifdef CONFIG_SPARSEMEM
1044 #include <asm/sparsemem.h>
1047 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1048 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1049 static inline unsigned long early_pfn_to_nid(unsigned long pfn
)
1055 #ifdef CONFIG_FLATMEM
1056 #define pfn_to_nid(pfn) (0)
1059 #ifdef CONFIG_SPARSEMEM
1062 * SECTION_SHIFT #bits space required to store a section #
1064 * PA_SECTION_SHIFT physical address to/from section number
1065 * PFN_SECTION_SHIFT pfn to/from section number
1067 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1068 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1070 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1072 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1073 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1075 #define SECTION_BLOCKFLAGS_BITS \
1076 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1078 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1079 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1082 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
1083 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
1085 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1086 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1090 struct mem_section
{
1092 * This is, logically, a pointer to an array of struct
1093 * pages. However, it is stored with some other magic.
1094 * (see sparse.c::sparse_init_one_section())
1096 * Additionally during early boot we encode node id of
1097 * the location of the section here to guide allocation.
1098 * (see sparse.c::memory_present())
1100 * Making it a UL at least makes someone do a cast
1101 * before using it wrong.
1103 unsigned long section_mem_map
;
1105 /* See declaration of similar field in struct zone */
1106 unsigned long *pageblock_flags
;
1107 #ifdef CONFIG_PAGE_EXTENSION
1109 * If !SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1110 * section. (see page_ext.h about this.)
1112 struct page_ext
*page_ext
;
1116 * WARNING: mem_section must be a power-of-2 in size for the
1117 * calculation and use of SECTION_ROOT_MASK to make sense.
1121 #ifdef CONFIG_SPARSEMEM_EXTREME
1122 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1124 #define SECTIONS_PER_ROOT 1
1127 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1128 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1129 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1131 #ifdef CONFIG_SPARSEMEM_EXTREME
1132 extern struct mem_section
*mem_section
[NR_SECTION_ROOTS
];
1134 extern struct mem_section mem_section
[NR_SECTION_ROOTS
][SECTIONS_PER_ROOT
];
1137 static inline struct mem_section
*__nr_to_section(unsigned long nr
)
1139 if (!mem_section
[SECTION_NR_TO_ROOT(nr
)])
1141 return &mem_section
[SECTION_NR_TO_ROOT(nr
)][nr
& SECTION_ROOT_MASK
];
1143 extern int __section_nr(struct mem_section
* ms
);
1144 extern unsigned long usemap_size(void);
1147 * We use the lower bits of the mem_map pointer to store
1148 * a little bit of information. There should be at least
1149 * 3 bits here due to 32-bit alignment.
1151 #define SECTION_MARKED_PRESENT (1UL<<0)
1152 #define SECTION_HAS_MEM_MAP (1UL<<1)
1153 #define SECTION_MAP_LAST_BIT (1UL<<2)
1154 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1155 #define SECTION_NID_SHIFT 2
1157 static inline struct page
*__section_mem_map_addr(struct mem_section
*section
)
1159 unsigned long map
= section
->section_mem_map
;
1160 map
&= SECTION_MAP_MASK
;
1161 return (struct page
*)map
;
1164 static inline int present_section(struct mem_section
*section
)
1166 return (section
&& (section
->section_mem_map
& SECTION_MARKED_PRESENT
));
1169 static inline int present_section_nr(unsigned long nr
)
1171 return present_section(__nr_to_section(nr
));
1174 static inline int valid_section(struct mem_section
*section
)
1176 return (section
&& (section
->section_mem_map
& SECTION_HAS_MEM_MAP
));
1179 static inline int valid_section_nr(unsigned long nr
)
1181 return valid_section(__nr_to_section(nr
));
1184 static inline struct mem_section
*__pfn_to_section(unsigned long pfn
)
1186 return __nr_to_section(pfn_to_section_nr(pfn
));
1189 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1190 static inline int pfn_valid(unsigned long pfn
)
1192 if (pfn_to_section_nr(pfn
) >= NR_MEM_SECTIONS
)
1194 return valid_section(__nr_to_section(pfn_to_section_nr(pfn
)));
1198 static inline int pfn_present(unsigned long pfn
)
1200 if (pfn_to_section_nr(pfn
) >= NR_MEM_SECTIONS
)
1202 return present_section(__nr_to_section(pfn_to_section_nr(pfn
)));
1206 * These are _only_ used during initialisation, therefore they
1207 * can use __initdata ... They could have names to indicate
1211 #define pfn_to_nid(pfn) \
1213 unsigned long __pfn_to_nid_pfn = (pfn); \
1214 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1217 #define pfn_to_nid(pfn) (0)
1220 #define early_pfn_valid(pfn) pfn_valid(pfn)
1221 void sparse_init(void);
1223 #define sparse_init() do {} while (0)
1224 #define sparse_index_init(_sec, _nid) do {} while (0)
1225 #endif /* CONFIG_SPARSEMEM */
1228 * During memory init memblocks map pfns to nids. The search is expensive and
1229 * this caches recent lookups. The implementation of __early_pfn_to_nid
1230 * may treat start/end as pfns or sections.
1232 struct mminit_pfnnid_cache
{
1233 unsigned long last_start
;
1234 unsigned long last_end
;
1238 #ifndef early_pfn_valid
1239 #define early_pfn_valid(pfn) (1)
1242 void memory_present(int nid
, unsigned long start
, unsigned long end
);
1243 unsigned long __init
node_memmap_size_bytes(int, unsigned long, unsigned long);
1246 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1247 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1248 * pfn_valid_within() should be used in this case; we optimise this away
1249 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1251 #ifdef CONFIG_HOLES_IN_ZONE
1252 #define pfn_valid_within(pfn) pfn_valid(pfn)
1254 #define pfn_valid_within(pfn) (1)
1257 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1259 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1260 * associated with it or not. In FLATMEM, it is expected that holes always
1261 * have valid memmap as long as there is valid PFNs either side of the hole.
1262 * In SPARSEMEM, it is assumed that a valid section has a memmap for the
1265 * However, an ARM, and maybe other embedded architectures in the future
1266 * free memmap backing holes to save memory on the assumption the memmap is
1267 * never used. The page_zone linkages are then broken even though pfn_valid()
1268 * returns true. A walker of the full memmap must then do this additional
1269 * check to ensure the memmap they are looking at is sane by making sure
1270 * the zone and PFN linkages are still valid. This is expensive, but walkers
1271 * of the full memmap are extremely rare.
1273 int memmap_valid_within(unsigned long pfn
,
1274 struct page
*page
, struct zone
*zone
);
1276 static inline int memmap_valid_within(unsigned long pfn
,
1277 struct page
*page
, struct zone
*zone
)
1281 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1283 #endif /* !__GENERATING_BOUNDS.H */
1284 #endif /* !__ASSEMBLY__ */
1285 #endif /* _LINUX_MMZONE_H */