1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _LINUX_MMZONE_H
3#define _LINUX_MMZONE_H
4
5#ifndef __ASSEMBLY__
6#ifndef __GENERATING_BOUNDS_H
7
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 <asm/page.h>
22
23/* Free memory management - zoned buddy allocator. */
24#ifndef CONFIG_FORCE_MAX_ZONEORDER
25#define MAX_ORDER 11
26#else
27#define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
28#endif
29#define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
30
31/*
32 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
33 * costly to service. That is between allocation orders which should
34 * coalesce naturally under reasonable reclaim pressure and those which
35 * will not.
36 */
37#define PAGE_ALLOC_COSTLY_ORDER 3
38
39enum migratetype {
40 MIGRATE_UNMOVABLE,
41 MIGRATE_MOVABLE,
42 MIGRATE_RECLAIMABLE,
43 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
44 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
45#ifdef CONFIG_CMA
46 /*
47 * MIGRATE_CMA migration type is designed to mimic the way
48 * ZONE_MOVABLE works. Only movable pages can be allocated
49 * from MIGRATE_CMA pageblocks and page allocator never
50 * implicitly change migration type of MIGRATE_CMA pageblock.
51 *
52 * The way to use it is to change migratetype of a range of
53 * pageblocks to MIGRATE_CMA which can be done by
54 * __free_pageblock_cma() function. What is important though
55 * is that a range of pageblocks must be aligned to
56 * MAX_ORDER_NR_PAGES should biggest page be bigger then
57 * a single pageblock.
58 */
59 MIGRATE_CMA,
60#endif
61#ifdef CONFIG_MEMORY_ISOLATION
62 MIGRATE_ISOLATE, /* can't allocate from here */
63#endif
64 MIGRATE_TYPES
65};
66
67/* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
68extern const char * const migratetype_names[MIGRATE_TYPES];
69
70#ifdef CONFIG_CMA
71# define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
72# define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
73#else
74# define is_migrate_cma(migratetype) false
75# define is_migrate_cma_page(_page) false
76#endif
77
78static inline bool is_migrate_movable(int mt)
79{
80 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
81}
82
83#define for_each_migratetype_order(order, type) \
84 for (order = 0; order < MAX_ORDER; order++) \
85 for (type = 0; type < MIGRATE_TYPES; type++)
86
87extern int page_group_by_mobility_disabled;
88
89#define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1)
90#define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1)
91
92#define get_pageblock_migratetype(page) \
93 get_pfnblock_flags_mask(page, page_to_pfn(page), \
94 PB_migrate_end, MIGRATETYPE_MASK)
95
96struct free_area {
97 struct list_head free_list[MIGRATE_TYPES];
98 unsigned long nr_free;
99};
100
101struct pglist_data;
102
103/*
104 * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
105 * So add a wild amount of padding here to ensure that they fall into separate
106 * cachelines. There are very few zone structures in the machine, so space
107 * consumption is not a concern here.
108 */
109#if defined(CONFIG_SMP)
110struct zone_padding {
111 char x[0];
112} ____cacheline_internodealigned_in_smp;
113#define ZONE_PADDING(name) struct zone_padding name;
114#else
115#define ZONE_PADDING(name)
116#endif
117
118#ifdef CONFIG_NUMA
119enum numa_stat_item {
120 NUMA_HIT, /* allocated in intended node */
121 NUMA_MISS, /* allocated in non intended node */
122 NUMA_FOREIGN, /* was intended here, hit elsewhere */
123 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
124 NUMA_LOCAL, /* allocation from local node */
125 NUMA_OTHER, /* allocation from other node */
126 NR_VM_NUMA_STAT_ITEMS
127};
128#else
129#define NR_VM_NUMA_STAT_ITEMS 0
130#endif
131
132enum zone_stat_item {
133 /* First 128 byte cacheline (assuming 64 bit words) */
134 NR_FREE_PAGES,
135 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
136 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
137 NR_ZONE_ACTIVE_ANON,
138 NR_ZONE_INACTIVE_FILE,
139 NR_ZONE_ACTIVE_FILE,
140 NR_ZONE_UNEVICTABLE,
141 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
142 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
143 NR_PAGETABLE, /* used for pagetables */
144 NR_KERNEL_STACK_KB, /* measured in KiB */
145 /* Second 128 byte cacheline */
146 NR_BOUNCE,
147#if IS_ENABLED(CONFIG_ZSMALLOC)
148 NR_ZSPAGES, /* allocated in zsmalloc */
149#endif
150 NR_FREE_CMA_PAGES,
151 NR_VM_ZONE_STAT_ITEMS };
152
153enum node_stat_item {
154 NR_LRU_BASE,
155 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
156 NR_ACTIVE_ANON, /* " " " " " */
157 NR_INACTIVE_FILE, /* " " " " " */
158 NR_ACTIVE_FILE, /* " " " " " */
159 NR_UNEVICTABLE, /* " " " " " */
160 NR_SLAB_RECLAIMABLE,
161 NR_SLAB_UNRECLAIMABLE,
162 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
163 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
164 WORKINGSET_NODES,
165 WORKINGSET_REFAULT,
166 WORKINGSET_ACTIVATE,
167 WORKINGSET_RESTORE,
168 WORKINGSET_NODERECLAIM,
169 NR_ANON_MAPPED, /* Mapped anonymous pages */
170 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
171 only modified from process context */
172 NR_FILE_PAGES,
173 NR_FILE_DIRTY,
174 NR_WRITEBACK,
175 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
176 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
177 NR_SHMEM_THPS,
178 NR_SHMEM_PMDMAPPED,
179 NR_ANON_THPS,
180 NR_UNSTABLE_NFS, /* NFS unstable pages */
181 NR_VMSCAN_WRITE,
182 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
183 NR_DIRTIED, /* page dirtyings since bootup */
184 NR_WRITTEN, /* page writings since bootup */
185 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
186 NR_VM_NODE_STAT_ITEMS
187};
188
189/*
190 * We do arithmetic on the LRU lists in various places in the code,
191 * so it is important to keep the active lists LRU_ACTIVE higher in
192 * the array than the corresponding inactive lists, and to keep
193 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
194 *
195 * This has to be kept in sync with the statistics in zone_stat_item
196 * above and the descriptions in vmstat_text in mm/vmstat.c
197 */
198#define LRU_BASE 0
199#define LRU_ACTIVE 1
200#define LRU_FILE 2
201
202enum lru_list {
203 LRU_INACTIVE_ANON = LRU_BASE,
204 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
205 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
206 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
207 LRU_UNEVICTABLE,
208 NR_LRU_LISTS
209};
210
211#define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
212
213#define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
214
215static inline int is_file_lru(enum lru_list lru)
216{
217 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
218}
219
220static inline int is_active_lru(enum lru_list lru)
221{
222 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
223}
224
225struct zone_reclaim_stat {
226 /*
227 * The pageout code in vmscan.c keeps track of how many of the
228 * mem/swap backed and file backed pages are referenced.
229 * The higher the rotated/scanned ratio, the more valuable
230 * that cache is.
231 *
232 * The anon LRU stats live in [0], file LRU stats in [1]
233 */
234 unsigned long recent_rotated[2];
235 unsigned long recent_scanned[2];
236};
237
238struct lruvec {
239 struct list_head lists[NR_LRU_LISTS];
240 struct zone_reclaim_stat reclaim_stat;
241 /* Evictions & activations on the inactive file list */
242 atomic_long_t inactive_age;
243 /* Refaults at the time of last reclaim cycle */
244 unsigned long refaults;
245#ifdef CONFIG_MEMCG
246 struct pglist_data *pgdat;
247#endif
248};
249
250/* Mask used at gathering information at once (see memcontrol.c) */
251#define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
252#define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
253#define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
254
255/* Isolate unmapped file */
256#define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
257/* Isolate for asynchronous migration */
258#define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
259/* Isolate unevictable pages */
260#define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
261
262/* LRU Isolation modes. */
263typedef unsigned __bitwise isolate_mode_t;
264
265enum zone_watermarks {
266 WMARK_MIN,
267 WMARK_LOW,
268 WMARK_HIGH,
269 NR_WMARK
270};
271
272#define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
273#define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
274#define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
275#define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
276
277struct per_cpu_pages {
278 int count; /* number of pages in the list */
279 int high; /* high watermark, emptying needed */
280 int batch; /* chunk size for buddy add/remove */
281
282 /* Lists of pages, one per migrate type stored on the pcp-lists */
283 struct list_head lists[MIGRATE_PCPTYPES];
284};
285
286struct per_cpu_pageset {
287 struct per_cpu_pages pcp;
288#ifdef CONFIG_NUMA
289 s8 expire;
290 u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
291#endif
292#ifdef CONFIG_SMP
293 s8 stat_threshold;
294 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
295#endif
296};
297
298struct per_cpu_nodestat {
299 s8 stat_threshold;
300 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
301};
302
303#endif /* !__GENERATING_BOUNDS.H */
304
305enum zone_type {
306#ifdef CONFIG_ZONE_DMA
307 /*
308 * ZONE_DMA is used when there are devices that are not able
309 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
310 * carve out the portion of memory that is needed for these devices.
311 * The range is arch specific.
312 *
313 * Some examples
314 *
315 * Architecture Limit
316 * ---------------------------
317 * parisc, ia64, sparc <4G
318 * s390, powerpc <2G
319 * arm Various
320 * alpha Unlimited or 0-16MB.
321 *
322 * i386, x86_64 and multiple other arches
323 * <16M.
324 */
325 ZONE_DMA,
326#endif
327#ifdef CONFIG_ZONE_DMA32
328 /*
329 * x86_64 needs two ZONE_DMAs because it supports devices that are
330 * only able to do DMA to the lower 16M but also 32 bit devices that
331 * can only do DMA areas below 4G.
332 */
333 ZONE_DMA32,
334#endif
335 /*
336 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
337 * performed on pages in ZONE_NORMAL if the DMA devices support
338 * transfers to all addressable memory.
339 */
340 ZONE_NORMAL,
341#ifdef CONFIG_HIGHMEM
342 /*
343 * A memory area that is only addressable by the kernel through
344 * mapping portions into its own address space. This is for example
345 * used by i386 to allow the kernel to address the memory beyond
346 * 900MB. The kernel will set up special mappings (page
347 * table entries on i386) for each page that the kernel needs to
348 * access.
349 */
350 ZONE_HIGHMEM,
351#endif
352 ZONE_MOVABLE,
353#ifdef CONFIG_ZONE_DEVICE
354 ZONE_DEVICE,
355#endif
356 __MAX_NR_ZONES
357
358};
359
360#ifndef __GENERATING_BOUNDS_H
361
362struct zone {
363 /* Read-mostly fields */
364
365 /* zone watermarks, access with *_wmark_pages(zone) macros */
366 unsigned long _watermark[NR_WMARK];
367 unsigned long watermark_boost;
368
369 unsigned long nr_reserved_highatomic;
370
371 /*
372 * We don't know if the memory that we're going to allocate will be
373 * freeable or/and it will be released eventually, so to avoid totally
374 * wasting several GB of ram we must reserve some of the lower zone
375 * memory (otherwise we risk to run OOM on the lower zones despite
376 * there being tons of freeable ram on the higher zones). This array is
377 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
378 * changes.
379 */
380 long lowmem_reserve[MAX_NR_ZONES];
381
382#ifdef CONFIG_NUMA
383 int node;
384#endif
385 struct pglist_data *zone_pgdat;
386 struct per_cpu_pageset __percpu *pageset;
387
388#ifndef CONFIG_SPARSEMEM
389 /*
390 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
391 * In SPARSEMEM, this map is stored in struct mem_section
392 */
393 unsigned long *pageblock_flags;
394#endif /* CONFIG_SPARSEMEM */
395
396 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
397 unsigned long zone_start_pfn;
398
399 /*
400 * spanned_pages is the total pages spanned by the zone, including
401 * holes, which is calculated as:
402 * spanned_pages = zone_end_pfn - zone_start_pfn;
403 *
404 * present_pages is physical pages existing within the zone, which
405 * is calculated as:
406 * present_pages = spanned_pages - absent_pages(pages in holes);
407 *
408 * managed_pages is present pages managed by the buddy system, which
409 * is calculated as (reserved_pages includes pages allocated by the
410 * bootmem allocator):
411 * managed_pages = present_pages - reserved_pages;
412 *
413 * So present_pages may be used by memory hotplug or memory power
414 * management logic to figure out unmanaged pages by checking
415 * (present_pages - managed_pages). And managed_pages should be used
416 * by page allocator and vm scanner to calculate all kinds of watermarks
417 * and thresholds.
418 *
419 * Locking rules:
420 *
421 * zone_start_pfn and spanned_pages are protected by span_seqlock.
422 * It is a seqlock because it has to be read outside of zone->lock,
423 * and it is done in the main allocator path. But, it is written
424 * quite infrequently.
425 *
426 * The span_seq lock is declared along with zone->lock because it is
427 * frequently read in proximity to zone->lock. It's good to
428 * give them a chance of being in the same cacheline.
429 *
430 * Write access to present_pages at runtime should be protected by
431 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
432 * present_pages should get_online_mems() to get a stable value.
433 */
434 atomic_long_t managed_pages;
435 unsigned long spanned_pages;
436 unsigned long present_pages;
437
438 const char *name;
439
440#ifdef CONFIG_MEMORY_ISOLATION
441 /*
442 * Number of isolated pageblock. It is used to solve incorrect
443 * freepage counting problem due to racy retrieving migratetype
444 * of pageblock. Protected by zone->lock.
445 */
446 unsigned long nr_isolate_pageblock;
447#endif
448
449#ifdef CONFIG_MEMORY_HOTPLUG
450 /* see spanned/present_pages for more description */
451 seqlock_t span_seqlock;
452#endif
453
454 int initialized;
455
456 /* Write-intensive fields used from the page allocator */
457 ZONE_PADDING(_pad1_)
458
459 /* free areas of different sizes */
460 struct free_area free_area[MAX_ORDER];
461
462 /* zone flags, see below */
463 unsigned long flags;
464
465 /* Primarily protects free_area */
466 spinlock_t lock;
467
468 /* Write-intensive fields used by compaction and vmstats. */
469 ZONE_PADDING(_pad2_)
470
471 /*
472 * When free pages are below this point, additional steps are taken
473 * when reading the number of free pages to avoid per-cpu counter
474 * drift allowing watermarks to be breached
475 */
476 unsigned long percpu_drift_mark;
477
478#if defined CONFIG_COMPACTION || defined CONFIG_CMA
479 /* pfn where compaction free scanner should start */
480 unsigned long compact_cached_free_pfn;
481 /* pfn where async and sync compaction migration scanner should start */
482 unsigned long compact_cached_migrate_pfn[2];
483 unsigned long compact_init_migrate_pfn;
484 unsigned long compact_init_free_pfn;
485#endif
486
487#ifdef CONFIG_COMPACTION
488 /*
489 * On compaction failure, 1<<compact_defer_shift compactions
490 * are skipped before trying again. The number attempted since
491 * last failure is tracked with compact_considered.
492 */
493 unsigned int compact_considered;
494 unsigned int compact_defer_shift;
495 int compact_order_failed;
496#endif
497
498#if defined CONFIG_COMPACTION || defined CONFIG_CMA
499 /* Set to true when the PG_migrate_skip bits should be cleared */
500 bool compact_blockskip_flush;
501#endif
502
503 bool contiguous;
504
505 ZONE_PADDING(_pad3_)
506 /* Zone statistics */
507 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
508 atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
509} ____cacheline_internodealigned_in_smp;
510
511enum pgdat_flags {
512 PGDAT_CONGESTED, /* pgdat has many dirty pages backed by
513 * a congested BDI
514 */
515 PGDAT_DIRTY, /* reclaim scanning has recently found
516 * many dirty file pages at the tail
517 * of the LRU.
518 */
519 PGDAT_WRITEBACK, /* reclaim scanning has recently found
520 * many pages under writeback
521 */
522 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
523};
524
525enum zone_flags {
526 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
527 * Cleared when kswapd is woken.
528 */
529};
530
531static inline unsigned long zone_managed_pages(struct zone *zone)
532{
533 return (unsigned long)atomic_long_read(&zone->managed_pages);
534}
535
536static inline unsigned long zone_end_pfn(const struct zone *zone)
537{
538 return zone->zone_start_pfn + zone->spanned_pages;
539}
540
541static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
542{
543 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
544}
545
546static inline bool zone_is_initialized(struct zone *zone)
547{
548 return zone->initialized;
549}
550
551static inline bool zone_is_empty(struct zone *zone)
552{
553 return zone->spanned_pages == 0;
554}
555
556/*
557 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
558 * intersection with the given zone
559 */
560static inline bool zone_intersects(struct zone *zone,
561 unsigned long start_pfn, unsigned long nr_pages)
562{
563 if (zone_is_empty(zone))
564 return false;
565 if (start_pfn >= zone_end_pfn(zone) ||
566 start_pfn + nr_pages <= zone->zone_start_pfn)
567 return false;
568
569 return true;
570}
571
572/*
573 * The "priority" of VM scanning is how much of the queues we will scan in one
574 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
575 * queues ("queue_length >> 12") during an aging round.
576 */
577#define DEF_PRIORITY 12
578
579/* Maximum number of zones on a zonelist */
580#define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
581
582enum {
583 ZONELIST_FALLBACK, /* zonelist with fallback */
584#ifdef CONFIG_NUMA
585 /*
586 * The NUMA zonelists are doubled because we need zonelists that
587 * restrict the allocations to a single node for __GFP_THISNODE.
588 */
589 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
590#endif
591 MAX_ZONELISTS
592};
593
594/*
595 * This struct contains information about a zone in a zonelist. It is stored
596 * here to avoid dereferences into large structures and lookups of tables
597 */
598struct zoneref {
599 struct zone *zone; /* Pointer to actual zone */
600 int zone_idx; /* zone_idx(zoneref->zone) */
601};
602
603/*
604 * One allocation request operates on a zonelist. A zonelist
605 * is a list of zones, the first one is the 'goal' of the
606 * allocation, the other zones are fallback zones, in decreasing
607 * priority.
608 *
609 * To speed the reading of the zonelist, the zonerefs contain the zone index
610 * of the entry being read. Helper functions to access information given
611 * a struct zoneref are
612 *
613 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
614 * zonelist_zone_idx() - Return the index of the zone for an entry
615 * zonelist_node_idx() - Return the index of the node for an entry
616 */
617struct zonelist {
618 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
619};
620
621#ifndef CONFIG_DISCONTIGMEM
622/* The array of struct pages - for discontigmem use pgdat->lmem_map */
623extern struct page *mem_map;
624#endif
625
626/*
627 * On NUMA machines, each NUMA node would have a pg_data_t to describe
628 * it's memory layout. On UMA machines there is a single pglist_data which
629 * describes the whole memory.
630 *
631 * Memory statistics and page replacement data structures are maintained on a
632 * per-zone basis.
633 */
634struct bootmem_data;
635typedef struct pglist_data {
636 struct zone node_zones[MAX_NR_ZONES];
637 struct zonelist node_zonelists[MAX_ZONELISTS];
638 int nr_zones;
639#ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
640 struct page *node_mem_map;
641#ifdef CONFIG_PAGE_EXTENSION
642 struct page_ext *node_page_ext;
643#endif
644#endif
645#if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
646 /*
647 * Must be held any time you expect node_start_pfn,
648 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
649 *
650 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
651 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
652 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
653 *
654 * Nests above zone->lock and zone->span_seqlock
655 */
656 spinlock_t node_size_lock;
657#endif
658 unsigned long node_start_pfn;
659 unsigned long node_present_pages; /* total number of physical pages */
660 unsigned long node_spanned_pages; /* total size of physical page
661 range, including holes */
662 int node_id;
663 wait_queue_head_t kswapd_wait;
664 wait_queue_head_t pfmemalloc_wait;
665 struct task_struct *kswapd; /* Protected by
666 mem_hotplug_begin/end() */
667 int kswapd_order;
668 enum zone_type kswapd_classzone_idx;
669
670 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
671
672#ifdef CONFIG_COMPACTION
673 int kcompactd_max_order;
674 enum zone_type kcompactd_classzone_idx;
675 wait_queue_head_t kcompactd_wait;
676 struct task_struct *kcompactd;
677#endif
678 /*
679 * This is a per-node reserve of pages that are not available
680 * to userspace allocations.
681 */
682 unsigned long totalreserve_pages;
683
684#ifdef CONFIG_NUMA
685 /*
686 * zone reclaim becomes active if more unmapped pages exist.
687 */
688 unsigned long min_unmapped_pages;
689 unsigned long min_slab_pages;
690#endif /* CONFIG_NUMA */
691
692 /* Write-intensive fields used by page reclaim */
693 ZONE_PADDING(_pad1_)
694 spinlock_t lru_lock;
695
696#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
697 /*
698 * If memory initialisation on large machines is deferred then this
699 * is the first PFN that needs to be initialised.
700 */
701 unsigned long first_deferred_pfn;
702#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
703
704#ifdef CONFIG_TRANSPARENT_HUGEPAGE
705 spinlock_t split_queue_lock;
706 struct list_head split_queue;
707 unsigned long split_queue_len;
708#endif
709
710 /* Fields commonly accessed by the page reclaim scanner */
711 struct lruvec lruvec;
712
713 unsigned long flags;
714
715 ZONE_PADDING(_pad2_)
716
717 /* Per-node vmstats */
718 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
719 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
720} pg_data_t;
721
722#define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
723#define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
724#ifdef CONFIG_FLAT_NODE_MEM_MAP
725#define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
726#else
727#define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
728#endif
729#define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
730
731#define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
732#define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
733
734static inline struct lruvec *node_lruvec(struct pglist_data *pgdat)
735{
736 return &pgdat->lruvec;
737}
738
739static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
740{
741 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
742}
743
744static inline bool pgdat_is_empty(pg_data_t *pgdat)
745{
746 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
747}
748
749#include <linux/memory_hotplug.h>
750
751void build_all_zonelists(pg_data_t *pgdat);
752void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
753 enum zone_type classzone_idx);
754bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
755 int classzone_idx, unsigned int alloc_flags,
756 long free_pages);
757bool zone_watermark_ok(struct zone *z, unsigned int order,
758 unsigned long mark, int classzone_idx,
759 unsigned int alloc_flags);
760bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
761 unsigned long mark, int classzone_idx);
762enum memmap_context {
763 MEMMAP_EARLY,
764 MEMMAP_HOTPLUG,
765};
766extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
767 unsigned long size);
768
769extern void lruvec_init(struct lruvec *lruvec);
770
771static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
772{
773#ifdef CONFIG_MEMCG
774 return lruvec->pgdat;
775#else
776 return container_of(lruvec, struct pglist_data, lruvec);
777#endif
778}
779
780extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
781
782#ifdef CONFIG_HAVE_MEMORY_PRESENT
783void memory_present(int nid, unsigned long start, unsigned long end);
784#else
785static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
786#endif
787
788#if defined(CONFIG_SPARSEMEM)
789void memblocks_present(void);
790#else
791static inline void memblocks_present(void) {}
792#endif
793
794#ifdef CONFIG_HAVE_MEMORYLESS_NODES
795int local_memory_node(int node_id);
796#else
797static inline int local_memory_node(int node_id) { return node_id; };
798#endif
799
800/*
801 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
802 */
803#define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
804
805#ifdef CONFIG_ZONE_DEVICE
806static inline bool is_dev_zone(const struct zone *zone)
807{
808 return zone_idx(zone) == ZONE_DEVICE;
809}
810#else
811static inline bool is_dev_zone(const struct zone *zone)
812{
813 return false;
814}
815#endif
816
817/*
818 * Returns true if a zone has pages managed by the buddy allocator.
819 * All the reclaim decisions have to use this function rather than
820 * populated_zone(). If the whole zone is reserved then we can easily
821 * end up with populated_zone() && !managed_zone().
822 */
823static inline bool managed_zone(struct zone *zone)
824{
825 return zone_managed_pages(zone);
826}
827
828/* Returns true if a zone has memory */
829static inline bool populated_zone(struct zone *zone)
830{
831 return zone->present_pages;
832}
833
834#ifdef CONFIG_NUMA
835static inline int zone_to_nid(struct zone *zone)
836{
837 return zone->node;
838}
839
840static inline void zone_set_nid(struct zone *zone, int nid)
841{
842 zone->node = nid;
843}
844#else
845static inline int zone_to_nid(struct zone *zone)
846{
847 return 0;
848}
849
850static inline void zone_set_nid(struct zone *zone, int nid) {}
851#endif
852
853extern int movable_zone;
854
855#ifdef CONFIG_HIGHMEM
856static inline int zone_movable_is_highmem(void)
857{
858#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
859 return movable_zone == ZONE_HIGHMEM;
860#else
861 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
862#endif
863}
864#endif
865
866static inline int is_highmem_idx(enum zone_type idx)
867{
868#ifdef CONFIG_HIGHMEM
869 return (idx == ZONE_HIGHMEM ||
870 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
871#else
872 return 0;
873#endif
874}
875
876/**
877 * is_highmem - helper function to quickly check if a struct zone is a
878 * highmem zone or not. This is an attempt to keep references
879 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
880 * @zone - pointer to struct zone variable
881 */
882static inline int is_highmem(struct zone *zone)
883{
884#ifdef CONFIG_HIGHMEM
885 return is_highmem_idx(zone_idx(zone));
886#else
887 return 0;
888#endif
889}
890
891/* These two functions are used to setup the per zone pages min values */
892struct ctl_table;
893int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
894 void __user *, size_t *, loff_t *);
895int watermark_boost_factor_sysctl_handler(struct ctl_table *, int,
896 void __user *, size_t *, loff_t *);
897int watermark_scale_factor_sysctl_handler(struct ctl_table *, int,
898 void __user *, size_t *, loff_t *);
899extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
900int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
901 void __user *, size_t *, loff_t *);
902int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
903 void __user *, size_t *, loff_t *);
904int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
905 void __user *, size_t *, loff_t *);
906int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
907 void __user *, size_t *, loff_t *);
908
909extern int numa_zonelist_order_handler(struct ctl_table *, int,
910 void __user *, size_t *, loff_t *);
911extern char numa_zonelist_order[];
912#define NUMA_ZONELIST_ORDER_LEN 16
913
914#ifndef CONFIG_NEED_MULTIPLE_NODES
915
916extern struct pglist_data contig_page_data;
917#define NODE_DATA(nid) (&contig_page_data)
918#define NODE_MEM_MAP(nid) mem_map
919
920#else /* CONFIG_NEED_MULTIPLE_NODES */
921
922#include <asm/mmzone.h>
923
924#endif /* !CONFIG_NEED_MULTIPLE_NODES */
925
926extern struct pglist_data *first_online_pgdat(void);
927extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
928extern struct zone *next_zone(struct zone *zone);
929
930/**
931 * for_each_online_pgdat - helper macro to iterate over all online nodes
932 * @pgdat - pointer to a pg_data_t variable
933 */
934#define for_each_online_pgdat(pgdat) \
935 for (pgdat = first_online_pgdat(); \
936 pgdat; \
937 pgdat = next_online_pgdat(pgdat))
938/**
939 * for_each_zone - helper macro to iterate over all memory zones
940 * @zone - pointer to struct zone variable
941 *
942 * The user only needs to declare the zone variable, for_each_zone
943 * fills it in.
944 */
945#define for_each_zone(zone) \
946 for (zone = (first_online_pgdat())->node_zones; \
947 zone; \
948 zone = next_zone(zone))
949
950#define for_each_populated_zone(zone) \
951 for (zone = (first_online_pgdat())->node_zones; \
952 zone; \
953 zone = next_zone(zone)) \
954 if (!populated_zone(zone)) \
955 ; /* do nothing */ \
956 else
957
958static inline struct zone *zonelist_zone(struct zoneref *zoneref)
959{
960 return zoneref->zone;
961}
962
963static inline int zonelist_zone_idx(struct zoneref *zoneref)
964{
965 return zoneref->zone_idx;
966}
967
968static inline int zonelist_node_idx(struct zoneref *zoneref)
969{
970 return zone_to_nid(zoneref->zone);
971}
972
973struct zoneref *__next_zones_zonelist(struct zoneref *z,
974 enum zone_type highest_zoneidx,
975 nodemask_t *nodes);
976
977/**
978 * 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
979 * @z - The cursor used as a starting point for the search
980 * @highest_zoneidx - The zone index of the highest zone to return
981 * @nodes - An optional nodemask to filter the zonelist with
982 *
983 * This function returns the next zone at or below a given zone index that is
984 * within the allowed nodemask using a cursor as the starting point for the
985 * search. The zoneref returned is a cursor that represents the current zone
986 * being examined. It should be advanced by one before calling
987 * next_zones_zonelist again.
988 */
989static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
990 enum zone_type highest_zoneidx,
991 nodemask_t *nodes)
992{
993 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
994 return z;
995 return __next_zones_zonelist(z, highest_zoneidx, nodes);
996}
997
998/**
999 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1000 * @zonelist - The zonelist to search for a suitable zone
1001 * @highest_zoneidx - The zone index of the highest zone to return
1002 * @nodes - An optional nodemask to filter the zonelist with
1003 * @return - Zoneref pointer for the first suitable zone found (see below)
1004 *
1005 * This function returns the first zone at or below a given zone index that is
1006 * within the allowed nodemask. The zoneref returned is a cursor that can be
1007 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1008 * one before calling.
1009 *
1010 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1011 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1012 * update due to cpuset modification.
1013 */
1014static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1015 enum zone_type highest_zoneidx,
1016 nodemask_t *nodes)
1017{
1018 return next_zones_zonelist(zonelist->_zonerefs,
1019 highest_zoneidx, nodes);
1020}
1021
1022/**
1023 * 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
1024 * @zone - The current zone in the iterator
1025 * @z - The current pointer within zonelist->zones being iterated
1026 * @zlist - The zonelist being iterated
1027 * @highidx - The zone index of the highest zone to return
1028 * @nodemask - Nodemask allowed by the allocator
1029 *
1030 * This iterator iterates though all zones at or below a given zone index and
1031 * within a given nodemask
1032 */
1033#define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1034 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1035 zone; \
1036 z = next_zones_zonelist(++z, highidx, nodemask), \
1037 zone = zonelist_zone(z))
1038
1039#define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1040 for (zone = z->zone; \
1041 zone; \
1042 z = next_zones_zonelist(++z, highidx, nodemask), \
1043 zone = zonelist_zone(z))
1044
1045
1046/**
1047 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1048 * @zone - The current zone in the iterator
1049 * @z - The current pointer within zonelist->zones being iterated
1050 * @zlist - The zonelist being iterated
1051 * @highidx - The zone index of the highest zone to return
1052 *
1053 * This iterator iterates though all zones at or below a given zone index.
1054 */
1055#define for_each_zone_zonelist(zone, z, zlist, highidx) \
1056 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1057
1058#ifdef CONFIG_SPARSEMEM
1059#include <asm/sparsemem.h>
1060#endif
1061
1062#if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1063 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1064static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1065{
1066 BUILD_BUG_ON(IS_ENABLED(CONFIG_NUMA));
1067 return 0;
1068}
1069#endif
1070
1071#ifdef CONFIG_FLATMEM
1072#define pfn_to_nid(pfn) (0)
1073#endif
1074
1075#ifdef CONFIG_SPARSEMEM
1076
1077/*
1078 * SECTION_SHIFT #bits space required to store a section #
1079 *
1080 * PA_SECTION_SHIFT physical address to/from section number
1081 * PFN_SECTION_SHIFT pfn to/from section number
1082 */
1083#define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1084#define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1085
1086#define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1087
1088#define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1089#define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1090
1091#define SECTION_BLOCKFLAGS_BITS \
1092 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1093
1094#if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1095#error Allocator MAX_ORDER exceeds SECTION_SIZE
1096#endif
1097
1098static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1099{
1100 return pfn >> PFN_SECTION_SHIFT;
1101}
1102static inline unsigned long section_nr_to_pfn(unsigned long sec)
1103{
1104 return sec << PFN_SECTION_SHIFT;
1105}
1106
1107#define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1108#define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1109
1110struct page;
1111struct page_ext;
1112struct mem_section {
1113 /*
1114 * This is, logically, a pointer to an array of struct
1115 * pages. However, it is stored with some other magic.
1116 * (see sparse.c::sparse_init_one_section())
1117 *
1118 * Additionally during early boot we encode node id of
1119 * the location of the section here to guide allocation.
1120 * (see sparse.c::memory_present())
1121 *
1122 * Making it a UL at least makes someone do a cast
1123 * before using it wrong.
1124 */
1125 unsigned long section_mem_map;
1126
1127 /* See declaration of similar field in struct zone */
1128 unsigned long *pageblock_flags;
1129#ifdef CONFIG_PAGE_EXTENSION
1130 /*
1131 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1132 * section. (see page_ext.h about this.)
1133 */
1134 struct page_ext *page_ext;
1135 unsigned long pad;
1136#endif
1137 /*
1138 * WARNING: mem_section must be a power-of-2 in size for the
1139 * calculation and use of SECTION_ROOT_MASK to make sense.
1140 */
1141};
1142
1143#ifdef CONFIG_SPARSEMEM_EXTREME
1144#define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1145#else
1146#define SECTIONS_PER_ROOT 1
1147#endif
1148
1149#define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1150#define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1151#define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1152
1153#ifdef CONFIG_SPARSEMEM_EXTREME
1154extern struct mem_section **mem_section;
1155#else
1156extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1157#endif
1158
1159static inline struct mem_section *__nr_to_section(unsigned long nr)
1160{
1161#ifdef CONFIG_SPARSEMEM_EXTREME
1162 if (!mem_section)
1163 return NULL;
1164#endif
1165 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1166 return NULL;
1167 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1168}
1169extern int __section_nr(struct mem_section* ms);
1170extern unsigned long usemap_size(void);
1171
1172/*
1173 * We use the lower bits of the mem_map pointer to store
1174 * a little bit of information. The pointer is calculated
1175 * as mem_map - section_nr_to_pfn(pnum). The result is
1176 * aligned to the minimum alignment of the two values:
1177 * 1. All mem_map arrays are page-aligned.
1178 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1179 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1180 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1181 * worst combination is powerpc with 256k pages,
1182 * which results in PFN_SECTION_SHIFT equal 6.
1183 * To sum it up, at least 6 bits are available.
1184 */
1185#define SECTION_MARKED_PRESENT (1UL<<0)
1186#define SECTION_HAS_MEM_MAP (1UL<<1)
1187#define SECTION_IS_ONLINE (1UL<<2)
1188#define SECTION_MAP_LAST_BIT (1UL<<3)
1189#define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1190#define SECTION_NID_SHIFT 3
1191
1192static inline struct page *__section_mem_map_addr(struct mem_section *section)
1193{
1194 unsigned long map = section->section_mem_map;
1195 map &= SECTION_MAP_MASK;
1196 return (struct page *)map;
1197}
1198
1199static inline int present_section(struct mem_section *section)
1200{
1201 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1202}
1203
1204static inline int present_section_nr(unsigned long nr)
1205{
1206 return present_section(__nr_to_section(nr));
1207}
1208
1209static inline int valid_section(struct mem_section *section)
1210{
1211 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1212}
1213
1214static inline int valid_section_nr(unsigned long nr)
1215{
1216 return valid_section(__nr_to_section(nr));
1217}
1218
1219static inline int online_section(struct mem_section *section)
1220{
1221 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1222}
1223
1224static inline int online_section_nr(unsigned long nr)
1225{
1226 return online_section(__nr_to_section(nr));
1227}
1228
1229#ifdef CONFIG_MEMORY_HOTPLUG
1230void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1231#ifdef CONFIG_MEMORY_HOTREMOVE
1232void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1233#endif
1234#endif
1235
1236static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1237{
1238 return __nr_to_section(pfn_to_section_nr(pfn));
1239}
1240
1241extern int __highest_present_section_nr;
1242
1243#ifndef CONFIG_HAVE_ARCH_PFN_VALID
1244static inline int pfn_valid(unsigned long pfn)
1245{
1246 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1247 return 0;
1248 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
1249}
1250#endif
1251
1252static inline int pfn_present(unsigned long pfn)
1253{
1254 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1255 return 0;
1256 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1257}
1258
1259/*
1260 * These are _only_ used during initialisation, therefore they
1261 * can use __initdata ... They could have names to indicate
1262 * this restriction.
1263 */
1264#ifdef CONFIG_NUMA
1265#define pfn_to_nid(pfn) \
1266({ \
1267 unsigned long __pfn_to_nid_pfn = (pfn); \
1268 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1269})
1270#else
1271#define pfn_to_nid(pfn) (0)
1272#endif
1273
1274#define early_pfn_valid(pfn) pfn_valid(pfn)
1275void sparse_init(void);
1276#else
1277#define sparse_init() do {} while (0)
1278#define sparse_index_init(_sec, _nid) do {} while (0)
1279#endif /* CONFIG_SPARSEMEM */
1280
1281/*
1282 * During memory init memblocks map pfns to nids. The search is expensive and
1283 * this caches recent lookups. The implementation of __early_pfn_to_nid
1284 * may treat start/end as pfns or sections.
1285 */
1286struct mminit_pfnnid_cache {
1287 unsigned long last_start;
1288 unsigned long last_end;
1289 int last_nid;
1290};
1291
1292#ifndef early_pfn_valid
1293#define early_pfn_valid(pfn) (1)
1294#endif
1295
1296void memory_present(int nid, unsigned long start, unsigned long end);
1297
1298/*
1299 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1300 * need to check pfn validity within that MAX_ORDER_NR_PAGES block.
1301 * pfn_valid_within() should be used in this case; we optimise this away
1302 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1303 */
1304#ifdef CONFIG_HOLES_IN_ZONE
1305#define pfn_valid_within(pfn) pfn_valid(pfn)
1306#else
1307#define pfn_valid_within(pfn) (1)
1308#endif
1309
1310#ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1311/*
1312 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1313 * associated with it or not. This means that a struct page exists for this
1314 * pfn. The caller cannot assume the page is fully initialized in general.
1315 * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1316 * will ensure the struct page is fully online and initialized. Special pages
1317 * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1318 *
1319 * In FLATMEM, it is expected that holes always have valid memmap as long as
1320 * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1321 * that a valid section has a memmap for the entire section.
1322 *
1323 * However, an ARM, and maybe other embedded architectures in the future
1324 * free memmap backing holes to save memory on the assumption the memmap is
1325 * never used. The page_zone linkages are then broken even though pfn_valid()
1326 * returns true. A walker of the full memmap must then do this additional
1327 * check to ensure the memmap they are looking at is sane by making sure
1328 * the zone and PFN linkages are still valid. This is expensive, but walkers
1329 * of the full memmap are extremely rare.
1330 */
1331bool memmap_valid_within(unsigned long pfn,
1332 struct page *page, struct zone *zone);
1333#else
1334static inline bool memmap_valid_within(unsigned long pfn,
1335 struct page *page, struct zone *zone)
1336{
1337 return true;
1338}
1339#endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1340
1341#endif /* !__GENERATING_BOUNDS.H */
1342#endif /* !__ASSEMBLY__ */
1343#endif /* _LINUX_MMZONE_H */
1344