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