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 <linux/mm_types.h>
22 #include <linux/page-flags.h>
23 #include <asm/page.h>
24
25 /* Free memory management - zoned buddy allocator. */
26 #ifndef CONFIG_FORCE_MAX_ZONEORDER
27 #define MAX_ORDER 11
28 #else
29 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
30 #endif
31 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
32
33 /*
34 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
35 * costly to service. That is between allocation orders which should
36 * coalesce naturally under reasonable reclaim pressure and those which
37 * will not.
38 */
39 #define PAGE_ALLOC_COSTLY_ORDER 3
40
41 enum migratetype {
42 MIGRATE_UNMOVABLE,
43 MIGRATE_MOVABLE,
44 MIGRATE_RECLAIMABLE,
45 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
46 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
47 #ifdef CONFIG_CMA
48 /*
49 * MIGRATE_CMA migration type is designed to mimic the way
50 * ZONE_MOVABLE works. Only movable pages can be allocated
51 * from MIGRATE_CMA pageblocks and page allocator never
52 * implicitly change migration type of MIGRATE_CMA pageblock.
53 *
54 * The way to use it is to change migratetype of a range of
55 * pageblocks to MIGRATE_CMA which can be done by
56 * __free_pageblock_cma() function. What is important though
57 * is that a range of pageblocks must be aligned to
58 * MAX_ORDER_NR_PAGES should biggest page be bigger then
59 * a single pageblock.
60 */
61 MIGRATE_CMA,
62 #endif
63 #ifdef CONFIG_MEMORY_ISOLATION
64 MIGRATE_ISOLATE, /* can't allocate from here */
65 #endif
66 MIGRATE_TYPES
67 };
68
69 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
70 extern const char * const migratetype_names[MIGRATE_TYPES];
71
72 #ifdef CONFIG_CMA
73 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
74 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
75 #else
76 # define is_migrate_cma(migratetype) false
77 # define is_migrate_cma_page(_page) false
78 #endif
79
is_migrate_movable(int mt)80 static inline bool is_migrate_movable(int mt)
81 {
82 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
83 }
84
85 #define for_each_migratetype_order(order, type) \
86 for (order = 0; order < MAX_ORDER; order++) \
87 for (type = 0; type < MIGRATE_TYPES; type++)
88
89 extern int page_group_by_mobility_disabled;
90
91 #define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1)
92 #define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1)
93
94 #define get_pageblock_migratetype(page) \
95 get_pfnblock_flags_mask(page, page_to_pfn(page), \
96 PB_migrate_end, MIGRATETYPE_MASK)
97
98 struct free_area {
99 struct list_head free_list[MIGRATE_TYPES];
100 unsigned long nr_free;
101 };
102
103 /* Used for pages not on another list */
add_to_free_area(struct page * page,struct free_area * area,int migratetype)104 static inline void add_to_free_area(struct page *page, struct free_area *area,
105 int migratetype)
106 {
107 list_add(&page->lru, &area->free_list[migratetype]);
108 area->nr_free++;
109 }
110
111 /* Used for pages not on another list */
add_to_free_area_tail(struct page * page,struct free_area * area,int migratetype)112 static inline void add_to_free_area_tail(struct page *page, struct free_area *area,
113 int migratetype)
114 {
115 list_add_tail(&page->lru, &area->free_list[migratetype]);
116 area->nr_free++;
117 }
118
119 #ifdef CONFIG_SHUFFLE_PAGE_ALLOCATOR
120 /* Used to preserve page allocation order entropy */
121 void add_to_free_area_random(struct page *page, struct free_area *area,
122 int migratetype);
123 #else
add_to_free_area_random(struct page * page,struct free_area * area,int migratetype)124 static inline void add_to_free_area_random(struct page *page,
125 struct free_area *area, int migratetype)
126 {
127 add_to_free_area(page, area, migratetype);
128 }
129 #endif
130
131 /* Used for pages which are on another list */
move_to_free_area(struct page * page,struct free_area * area,int migratetype)132 static inline void move_to_free_area(struct page *page, struct free_area *area,
133 int migratetype)
134 {
135 list_move(&page->lru, &area->free_list[migratetype]);
136 }
137
get_page_from_free_area(struct free_area * area,int migratetype)138 static inline struct page *get_page_from_free_area(struct free_area *area,
139 int migratetype)
140 {
141 return list_first_entry_or_null(&area->free_list[migratetype],
142 struct page, lru);
143 }
144
del_page_from_free_area(struct page * page,struct free_area * area)145 static inline void del_page_from_free_area(struct page *page,
146 struct free_area *area)
147 {
148 list_del(&page->lru);
149 __ClearPageBuddy(page);
150 set_page_private(page, 0);
151 area->nr_free--;
152 }
153
free_area_empty(struct free_area * area,int migratetype)154 static inline bool free_area_empty(struct free_area *area, int migratetype)
155 {
156 return list_empty(&area->free_list[migratetype]);
157 }
158
159 struct pglist_data;
160
161 /*
162 * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
163 * So add a wild amount of padding here to ensure that they fall into separate
164 * cachelines. There are very few zone structures in the machine, so space
165 * consumption is not a concern here.
166 */
167 #if defined(CONFIG_SMP)
168 struct zone_padding {
169 char x[0];
170 } ____cacheline_internodealigned_in_smp;
171 #define ZONE_PADDING(name) struct zone_padding name;
172 #else
173 #define ZONE_PADDING(name)
174 #endif
175
176 #ifdef CONFIG_NUMA
177 enum numa_stat_item {
178 NUMA_HIT, /* allocated in intended node */
179 NUMA_MISS, /* allocated in non intended node */
180 NUMA_FOREIGN, /* was intended here, hit elsewhere */
181 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
182 NUMA_LOCAL, /* allocation from local node */
183 NUMA_OTHER, /* allocation from other node */
184 NR_VM_NUMA_STAT_ITEMS
185 };
186 #else
187 #define NR_VM_NUMA_STAT_ITEMS 0
188 #endif
189
190 enum zone_stat_item {
191 /* First 128 byte cacheline (assuming 64 bit words) */
192 NR_FREE_PAGES,
193 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
194 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
195 NR_ZONE_ACTIVE_ANON,
196 NR_ZONE_INACTIVE_FILE,
197 NR_ZONE_ACTIVE_FILE,
198 NR_ZONE_UNEVICTABLE,
199 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
200 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
201 NR_PAGETABLE, /* used for pagetables */
202 NR_KERNEL_STACK_KB, /* measured in KiB */
203 #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
204 NR_KERNEL_SCS_BYTES, /* measured in bytes */
205 #endif
206 /* Second 128 byte cacheline */
207 NR_BOUNCE,
208 #if IS_ENABLED(CONFIG_ZSMALLOC)
209 NR_ZSPAGES, /* allocated in zsmalloc */
210 #endif
211 NR_FREE_CMA_PAGES,
212 NR_VM_ZONE_STAT_ITEMS };
213
214 enum node_stat_item {
215 NR_LRU_BASE,
216 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
217 NR_ACTIVE_ANON, /* " " " " " */
218 NR_INACTIVE_FILE, /* " " " " " */
219 NR_ACTIVE_FILE, /* " " " " " */
220 NR_UNEVICTABLE, /* " " " " " */
221 NR_SLAB_RECLAIMABLE,
222 NR_SLAB_UNRECLAIMABLE,
223 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
224 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
225 WORKINGSET_NODES,
226 WORKINGSET_REFAULT,
227 WORKINGSET_ACTIVATE,
228 WORKINGSET_RESTORE,
229 WORKINGSET_NODERECLAIM,
230 NR_ANON_MAPPED, /* Mapped anonymous pages */
231 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
232 only modified from process context */
233 NR_FILE_PAGES,
234 NR_FILE_DIRTY,
235 NR_WRITEBACK,
236 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
237 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
238 NR_SHMEM_THPS,
239 NR_SHMEM_PMDMAPPED,
240 NR_FILE_THPS,
241 NR_FILE_PMDMAPPED,
242 NR_ANON_THPS,
243 NR_UNSTABLE_NFS, /* NFS unstable pages */
244 NR_VMSCAN_WRITE,
245 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
246 NR_DIRTIED, /* page dirtyings since bootup */
247 NR_WRITTEN, /* page writings since bootup */
248 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
249 NR_VM_NODE_STAT_ITEMS
250 };
251
252 /*
253 * We do arithmetic on the LRU lists in various places in the code,
254 * so it is important to keep the active lists LRU_ACTIVE higher in
255 * the array than the corresponding inactive lists, and to keep
256 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
257 *
258 * This has to be kept in sync with the statistics in zone_stat_item
259 * above and the descriptions in vmstat_text in mm/vmstat.c
260 */
261 #define LRU_BASE 0
262 #define LRU_ACTIVE 1
263 #define LRU_FILE 2
264
265 enum lru_list {
266 LRU_INACTIVE_ANON = LRU_BASE,
267 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
268 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
269 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
270 LRU_UNEVICTABLE,
271 NR_LRU_LISTS
272 };
273
274 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
275
276 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
277
is_file_lru(enum lru_list lru)278 static inline int is_file_lru(enum lru_list lru)
279 {
280 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
281 }
282
is_active_lru(enum lru_list lru)283 static inline int is_active_lru(enum lru_list lru)
284 {
285 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
286 }
287
288 struct zone_reclaim_stat {
289 /*
290 * The pageout code in vmscan.c keeps track of how many of the
291 * mem/swap backed and file backed pages are referenced.
292 * The higher the rotated/scanned ratio, the more valuable
293 * that cache is.
294 *
295 * The anon LRU stats live in [0], file LRU stats in [1]
296 */
297 unsigned long recent_rotated[2];
298 unsigned long recent_scanned[2];
299 };
300
301 struct lruvec {
302 struct list_head lists[NR_LRU_LISTS];
303 struct zone_reclaim_stat reclaim_stat;
304 /* Evictions & activations on the inactive file list */
305 atomic_long_t inactive_age;
306 /* Refaults at the time of last reclaim cycle */
307 unsigned long refaults;
308 #ifdef CONFIG_MEMCG
309 struct pglist_data *pgdat;
310 #endif
311 };
312
313 /* Isolate unmapped file */
314 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
315 /* Isolate for asynchronous migration */
316 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
317 /* Isolate unevictable pages */
318 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
319
320 /* LRU Isolation modes. */
321 typedef unsigned __bitwise isolate_mode_t;
322
323 enum zone_watermarks {
324 WMARK_MIN,
325 WMARK_LOW,
326 WMARK_HIGH,
327 NR_WMARK
328 };
329
330 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
331 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
332 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
333 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
334
335 struct per_cpu_pages {
336 int count; /* number of pages in the list */
337 int high; /* high watermark, emptying needed */
338 int batch; /* chunk size for buddy add/remove */
339
340 /* Lists of pages, one per migrate type stored on the pcp-lists */
341 struct list_head lists[MIGRATE_PCPTYPES];
342 };
343
344 struct per_cpu_pageset {
345 struct per_cpu_pages pcp;
346 #ifdef CONFIG_NUMA
347 s8 expire;
348 u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
349 #endif
350 #ifdef CONFIG_SMP
351 s8 stat_threshold;
352 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
353 #endif
354 };
355
356 struct per_cpu_nodestat {
357 s8 stat_threshold;
358 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
359 };
360
361 #endif /* !__GENERATING_BOUNDS.H */
362
363 enum zone_type {
364 #ifdef CONFIG_ZONE_DMA
365 /*
366 * ZONE_DMA is used when there are devices that are not able
367 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
368 * carve out the portion of memory that is needed for these devices.
369 * The range is arch specific.
370 *
371 * Some examples
372 *
373 * Architecture Limit
374 * ---------------------------
375 * parisc, ia64, sparc <4G
376 * s390, powerpc <2G
377 * arm Various
378 * alpha Unlimited or 0-16MB.
379 *
380 * i386, x86_64 and multiple other arches
381 * <16M.
382 */
383 ZONE_DMA,
384 #endif
385 #ifdef CONFIG_ZONE_DMA32
386 /*
387 * x86_64 needs two ZONE_DMAs because it supports devices that are
388 * only able to do DMA to the lower 16M but also 32 bit devices that
389 * can only do DMA areas below 4G.
390 */
391 ZONE_DMA32,
392 #endif
393 /*
394 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
395 * performed on pages in ZONE_NORMAL if the DMA devices support
396 * transfers to all addressable memory.
397 */
398 ZONE_NORMAL,
399 #ifdef CONFIG_HIGHMEM
400 /*
401 * A memory area that is only addressable by the kernel through
402 * mapping portions into its own address space. This is for example
403 * used by i386 to allow the kernel to address the memory beyond
404 * 900MB. The kernel will set up special mappings (page
405 * table entries on i386) for each page that the kernel needs to
406 * access.
407 */
408 ZONE_HIGHMEM,
409 #endif
410 ZONE_MOVABLE,
411 #ifdef CONFIG_ZONE_DEVICE
412 ZONE_DEVICE,
413 #endif
414 __MAX_NR_ZONES
415
416 };
417
418 #ifndef __GENERATING_BOUNDS_H
419
420 struct zone {
421 /* Read-mostly fields */
422
423 /* zone watermarks, access with *_wmark_pages(zone) macros */
424 unsigned long _watermark[NR_WMARK];
425 unsigned long watermark_boost;
426
427 unsigned long nr_reserved_highatomic;
428
429 /*
430 * We don't know if the memory that we're going to allocate will be
431 * freeable or/and it will be released eventually, so to avoid totally
432 * wasting several GB of ram we must reserve some of the lower zone
433 * memory (otherwise we risk to run OOM on the lower zones despite
434 * there being tons of freeable ram on the higher zones). This array is
435 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
436 * changes.
437 */
438 long lowmem_reserve[MAX_NR_ZONES];
439
440 #ifdef CONFIG_NUMA
441 int node;
442 #endif
443 struct pglist_data *zone_pgdat;
444 struct per_cpu_pageset __percpu *pageset;
445
446 #ifndef CONFIG_SPARSEMEM
447 /*
448 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
449 * In SPARSEMEM, this map is stored in struct mem_section
450 */
451 unsigned long *pageblock_flags;
452 #endif /* CONFIG_SPARSEMEM */
453
454 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
455 unsigned long zone_start_pfn;
456
457 /*
458 * spanned_pages is the total pages spanned by the zone, including
459 * holes, which is calculated as:
460 * spanned_pages = zone_end_pfn - zone_start_pfn;
461 *
462 * present_pages is physical pages existing within the zone, which
463 * is calculated as:
464 * present_pages = spanned_pages - absent_pages(pages in holes);
465 *
466 * managed_pages is present pages managed by the buddy system, which
467 * is calculated as (reserved_pages includes pages allocated by the
468 * bootmem allocator):
469 * managed_pages = present_pages - reserved_pages;
470 *
471 * So present_pages may be used by memory hotplug or memory power
472 * management logic to figure out unmanaged pages by checking
473 * (present_pages - managed_pages). And managed_pages should be used
474 * by page allocator and vm scanner to calculate all kinds of watermarks
475 * and thresholds.
476 *
477 * Locking rules:
478 *
479 * zone_start_pfn and spanned_pages are protected by span_seqlock.
480 * It is a seqlock because it has to be read outside of zone->lock,
481 * and it is done in the main allocator path. But, it is written
482 * quite infrequently.
483 *
484 * The span_seq lock is declared along with zone->lock because it is
485 * frequently read in proximity to zone->lock. It's good to
486 * give them a chance of being in the same cacheline.
487 *
488 * Write access to present_pages at runtime should be protected by
489 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
490 * present_pages should get_online_mems() to get a stable value.
491 */
492 atomic_long_t managed_pages;
493 unsigned long spanned_pages;
494 unsigned long present_pages;
495
496 const char *name;
497
498 #ifdef CONFIG_MEMORY_ISOLATION
499 /*
500 * Number of isolated pageblock. It is used to solve incorrect
501 * freepage counting problem due to racy retrieving migratetype
502 * of pageblock. Protected by zone->lock.
503 */
504 unsigned long nr_isolate_pageblock;
505 #endif
506
507 #ifdef CONFIG_MEMORY_HOTPLUG
508 /* see spanned/present_pages for more description */
509 seqlock_t span_seqlock;
510 #endif
511
512 int initialized;
513
514 /* Write-intensive fields used from the page allocator */
515 ZONE_PADDING(_pad1_)
516
517 /* free areas of different sizes */
518 struct free_area free_area[MAX_ORDER];
519
520 /* zone flags, see below */
521 unsigned long flags;
522
523 /* Primarily protects free_area */
524 spinlock_t lock;
525
526 /* Write-intensive fields used by compaction and vmstats. */
527 ZONE_PADDING(_pad2_)
528
529 /*
530 * When free pages are below this point, additional steps are taken
531 * when reading the number of free pages to avoid per-cpu counter
532 * drift allowing watermarks to be breached
533 */
534 unsigned long percpu_drift_mark;
535
536 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
537 /* pfn where compaction free scanner should start */
538 unsigned long compact_cached_free_pfn;
539 /* pfn where async and sync compaction migration scanner should start */
540 unsigned long compact_cached_migrate_pfn[2];
541 unsigned long compact_init_migrate_pfn;
542 unsigned long compact_init_free_pfn;
543 #endif
544
545 #ifdef CONFIG_COMPACTION
546 /*
547 * On compaction failure, 1<<compact_defer_shift compactions
548 * are skipped before trying again. The number attempted since
549 * last failure is tracked with compact_considered.
550 */
551 unsigned int compact_considered;
552 unsigned int compact_defer_shift;
553 int compact_order_failed;
554 #endif
555
556 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
557 /* Set to true when the PG_migrate_skip bits should be cleared */
558 bool compact_blockskip_flush;
559 #endif
560
561 bool contiguous;
562
563 ZONE_PADDING(_pad3_)
564 /* Zone statistics */
565 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
566 atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
567 } ____cacheline_internodealigned_in_smp;
568
569 enum pgdat_flags {
570 PGDAT_CONGESTED, /* pgdat has many dirty pages backed by
571 * a congested BDI
572 */
573 PGDAT_DIRTY, /* reclaim scanning has recently found
574 * many dirty file pages at the tail
575 * of the LRU.
576 */
577 PGDAT_WRITEBACK, /* reclaim scanning has recently found
578 * many pages under writeback
579 */
580 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
581 };
582
583 enum zone_flags {
584 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
585 * Cleared when kswapd is woken.
586 */
587 };
588
zone_managed_pages(struct zone * zone)589 static inline unsigned long zone_managed_pages(struct zone *zone)
590 {
591 return (unsigned long)atomic_long_read(&zone->managed_pages);
592 }
593
zone_end_pfn(const struct zone * zone)594 static inline unsigned long zone_end_pfn(const struct zone *zone)
595 {
596 return zone->zone_start_pfn + zone->spanned_pages;
597 }
598
zone_spans_pfn(const struct zone * zone,unsigned long pfn)599 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
600 {
601 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
602 }
603
zone_is_initialized(struct zone * zone)604 static inline bool zone_is_initialized(struct zone *zone)
605 {
606 return zone->initialized;
607 }
608
zone_is_empty(struct zone * zone)609 static inline bool zone_is_empty(struct zone *zone)
610 {
611 return zone->spanned_pages == 0;
612 }
613
614 /*
615 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
616 * intersection with the given zone
617 */
zone_intersects(struct zone * zone,unsigned long start_pfn,unsigned long nr_pages)618 static inline bool zone_intersects(struct zone *zone,
619 unsigned long start_pfn, unsigned long nr_pages)
620 {
621 if (zone_is_empty(zone))
622 return false;
623 if (start_pfn >= zone_end_pfn(zone) ||
624 start_pfn + nr_pages <= zone->zone_start_pfn)
625 return false;
626
627 return true;
628 }
629
630 /*
631 * The "priority" of VM scanning is how much of the queues we will scan in one
632 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
633 * queues ("queue_length >> 12") during an aging round.
634 */
635 #define DEF_PRIORITY 12
636
637 /* Maximum number of zones on a zonelist */
638 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
639
640 enum {
641 ZONELIST_FALLBACK, /* zonelist with fallback */
642 #ifdef CONFIG_NUMA
643 /*
644 * The NUMA zonelists are doubled because we need zonelists that
645 * restrict the allocations to a single node for __GFP_THISNODE.
646 */
647 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
648 #endif
649 MAX_ZONELISTS
650 };
651
652 /*
653 * This struct contains information about a zone in a zonelist. It is stored
654 * here to avoid dereferences into large structures and lookups of tables
655 */
656 struct zoneref {
657 struct zone *zone; /* Pointer to actual zone */
658 int zone_idx; /* zone_idx(zoneref->zone) */
659 };
660
661 /*
662 * One allocation request operates on a zonelist. A zonelist
663 * is a list of zones, the first one is the 'goal' of the
664 * allocation, the other zones are fallback zones, in decreasing
665 * priority.
666 *
667 * To speed the reading of the zonelist, the zonerefs contain the zone index
668 * of the entry being read. Helper functions to access information given
669 * a struct zoneref are
670 *
671 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
672 * zonelist_zone_idx() - Return the index of the zone for an entry
673 * zonelist_node_idx() - Return the index of the node for an entry
674 */
675 struct zonelist {
676 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
677 };
678
679 #ifndef CONFIG_DISCONTIGMEM
680 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
681 extern struct page *mem_map;
682 #endif
683
684 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
685 struct deferred_split {
686 spinlock_t split_queue_lock;
687 struct list_head split_queue;
688 unsigned long split_queue_len;
689 };
690 #endif
691
692 /*
693 * On NUMA machines, each NUMA node would have a pg_data_t to describe
694 * it's memory layout. On UMA machines there is a single pglist_data which
695 * describes the whole memory.
696 *
697 * Memory statistics and page replacement data structures are maintained on a
698 * per-zone basis.
699 */
700 struct bootmem_data;
701 typedef struct pglist_data {
702 struct zone node_zones[MAX_NR_ZONES];
703 struct zonelist node_zonelists[MAX_ZONELISTS];
704 int nr_zones;
705 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
706 struct page *node_mem_map;
707 #ifdef CONFIG_PAGE_EXTENSION
708 struct page_ext *node_page_ext;
709 #endif
710 #endif
711 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
712 /*
713 * Must be held any time you expect node_start_pfn,
714 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
715 *
716 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
717 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
718 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
719 *
720 * Nests above zone->lock and zone->span_seqlock
721 */
722 spinlock_t node_size_lock;
723 #endif
724 unsigned long node_start_pfn;
725 unsigned long node_present_pages; /* total number of physical pages */
726 unsigned long node_spanned_pages; /* total size of physical page
727 range, including holes */
728 int node_id;
729 wait_queue_head_t kswapd_wait;
730 wait_queue_head_t pfmemalloc_wait;
731 struct task_struct *kswapd; /* Protected by
732 mem_hotplug_begin/end() */
733 int kswapd_order;
734 enum zone_type kswapd_classzone_idx;
735
736 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
737
738 #ifdef CONFIG_COMPACTION
739 int kcompactd_max_order;
740 enum zone_type kcompactd_classzone_idx;
741 wait_queue_head_t kcompactd_wait;
742 struct task_struct *kcompactd;
743 #endif
744 /*
745 * This is a per-node reserve of pages that are not available
746 * to userspace allocations.
747 */
748 unsigned long totalreserve_pages;
749
750 #ifdef CONFIG_NUMA
751 /*
752 * zone reclaim becomes active if more unmapped pages exist.
753 */
754 unsigned long min_unmapped_pages;
755 unsigned long min_slab_pages;
756 #endif /* CONFIG_NUMA */
757
758 /* Write-intensive fields used by page reclaim */
759 ZONE_PADDING(_pad1_)
760 spinlock_t lru_lock;
761
762 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
763 /*
764 * If memory initialisation on large machines is deferred then this
765 * is the first PFN that needs to be initialised.
766 */
767 unsigned long first_deferred_pfn;
768 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
769
770 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
771 struct deferred_split deferred_split_queue;
772 #endif
773
774 /* Fields commonly accessed by the page reclaim scanner */
775 struct lruvec lruvec;
776
777 unsigned long flags;
778
779 ZONE_PADDING(_pad2_)
780
781 /* Per-node vmstats */
782 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
783 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
784 } pg_data_t;
785
786 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
787 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
788 #ifdef CONFIG_FLAT_NODE_MEM_MAP
789 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
790 #else
791 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
792 #endif
793 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
794
795 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
796 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
797
node_lruvec(struct pglist_data * pgdat)798 static inline struct lruvec *node_lruvec(struct pglist_data *pgdat)
799 {
800 return &pgdat->lruvec;
801 }
802
pgdat_end_pfn(pg_data_t * pgdat)803 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
804 {
805 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
806 }
807
pgdat_is_empty(pg_data_t * pgdat)808 static inline bool pgdat_is_empty(pg_data_t *pgdat)
809 {
810 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
811 }
812
813 #include <linux/memory_hotplug.h>
814
815 void build_all_zonelists(pg_data_t *pgdat);
816 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
817 enum zone_type classzone_idx);
818 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
819 int classzone_idx, unsigned int alloc_flags,
820 long free_pages);
821 bool zone_watermark_ok(struct zone *z, unsigned int order,
822 unsigned long mark, int classzone_idx,
823 unsigned int alloc_flags);
824 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
825 unsigned long mark, int classzone_idx);
826 enum memmap_context {
827 MEMMAP_EARLY,
828 MEMMAP_HOTPLUG,
829 };
830 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
831 unsigned long size);
832
833 extern void lruvec_init(struct lruvec *lruvec);
834
lruvec_pgdat(struct lruvec * lruvec)835 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
836 {
837 #ifdef CONFIG_MEMCG
838 return lruvec->pgdat;
839 #else
840 return container_of(lruvec, struct pglist_data, lruvec);
841 #endif
842 }
843
844 extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
845
846 #ifdef CONFIG_HAVE_MEMORY_PRESENT
847 void memory_present(int nid, unsigned long start, unsigned long end);
848 #else
memory_present(int nid,unsigned long start,unsigned long end)849 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
850 #endif
851
852 #if defined(CONFIG_SPARSEMEM)
853 void memblocks_present(void);
854 #else
memblocks_present(void)855 static inline void memblocks_present(void) {}
856 #endif
857
858 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
859 int local_memory_node(int node_id);
860 #else
local_memory_node(int node_id)861 static inline int local_memory_node(int node_id) { return node_id; };
862 #endif
863
864 /*
865 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
866 */
867 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
868
869 /*
870 * Returns true if a zone has pages managed by the buddy allocator.
871 * All the reclaim decisions have to use this function rather than
872 * populated_zone(). If the whole zone is reserved then we can easily
873 * end up with populated_zone() && !managed_zone().
874 */
managed_zone(struct zone * zone)875 static inline bool managed_zone(struct zone *zone)
876 {
877 return zone_managed_pages(zone);
878 }
879
880 /* Returns true if a zone has memory */
populated_zone(struct zone * zone)881 static inline bool populated_zone(struct zone *zone)
882 {
883 return zone->present_pages;
884 }
885
886 #ifdef CONFIG_NUMA
zone_to_nid(struct zone * zone)887 static inline int zone_to_nid(struct zone *zone)
888 {
889 return zone->node;
890 }
891
zone_set_nid(struct zone * zone,int nid)892 static inline void zone_set_nid(struct zone *zone, int nid)
893 {
894 zone->node = nid;
895 }
896 #else
zone_to_nid(struct zone * zone)897 static inline int zone_to_nid(struct zone *zone)
898 {
899 return 0;
900 }
901
zone_set_nid(struct zone * zone,int nid)902 static inline void zone_set_nid(struct zone *zone, int nid) {}
903 #endif
904
905 extern int movable_zone;
906
907 #ifdef CONFIG_HIGHMEM
zone_movable_is_highmem(void)908 static inline int zone_movable_is_highmem(void)
909 {
910 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
911 return movable_zone == ZONE_HIGHMEM;
912 #else
913 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
914 #endif
915 }
916 #endif
917
is_highmem_idx(enum zone_type idx)918 static inline int is_highmem_idx(enum zone_type idx)
919 {
920 #ifdef CONFIG_HIGHMEM
921 return (idx == ZONE_HIGHMEM ||
922 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
923 #else
924 return 0;
925 #endif
926 }
927
928 /**
929 * is_highmem - helper function to quickly check if a struct zone is a
930 * highmem zone or not. This is an attempt to keep references
931 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
932 * @zone - pointer to struct zone variable
933 */
is_highmem(struct zone * zone)934 static inline int is_highmem(struct zone *zone)
935 {
936 #ifdef CONFIG_HIGHMEM
937 return is_highmem_idx(zone_idx(zone));
938 #else
939 return 0;
940 #endif
941 }
942
943 /* These two functions are used to setup the per zone pages min values */
944 struct ctl_table;
945 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
946 void __user *, size_t *, loff_t *);
947 int watermark_boost_factor_sysctl_handler(struct ctl_table *, int,
948 void __user *, size_t *, loff_t *);
949 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int,
950 void __user *, size_t *, loff_t *);
951 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
952 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
953 void __user *, size_t *, loff_t *);
954 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
955 void __user *, size_t *, loff_t *);
956 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
957 void __user *, size_t *, loff_t *);
958 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
959 void __user *, size_t *, loff_t *);
960
961 extern int numa_zonelist_order_handler(struct ctl_table *, int,
962 void __user *, size_t *, loff_t *);
963 extern char numa_zonelist_order[];
964 #define NUMA_ZONELIST_ORDER_LEN 16
965
966 #ifndef CONFIG_NEED_MULTIPLE_NODES
967
968 extern struct pglist_data contig_page_data;
969 #define NODE_DATA(nid) (&contig_page_data)
970 #define NODE_MEM_MAP(nid) mem_map
971
972 #else /* CONFIG_NEED_MULTIPLE_NODES */
973
974 #include <asm/mmzone.h>
975
976 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
977
978 extern struct pglist_data *first_online_pgdat(void);
979 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
980 extern struct zone *next_zone(struct zone *zone);
981
982 /**
983 * for_each_online_pgdat - helper macro to iterate over all online nodes
984 * @pgdat - pointer to a pg_data_t variable
985 */
986 #define for_each_online_pgdat(pgdat) \
987 for (pgdat = first_online_pgdat(); \
988 pgdat; \
989 pgdat = next_online_pgdat(pgdat))
990 /**
991 * for_each_zone - helper macro to iterate over all memory zones
992 * @zone - pointer to struct zone variable
993 *
994 * The user only needs to declare the zone variable, for_each_zone
995 * fills it in.
996 */
997 #define for_each_zone(zone) \
998 for (zone = (first_online_pgdat())->node_zones; \
999 zone; \
1000 zone = next_zone(zone))
1001
1002 #define for_each_populated_zone(zone) \
1003 for (zone = (first_online_pgdat())->node_zones; \
1004 zone; \
1005 zone = next_zone(zone)) \
1006 if (!populated_zone(zone)) \
1007 ; /* do nothing */ \
1008 else
1009
zonelist_zone(struct zoneref * zoneref)1010 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1011 {
1012 return zoneref->zone;
1013 }
1014
zonelist_zone_idx(struct zoneref * zoneref)1015 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1016 {
1017 return zoneref->zone_idx;
1018 }
1019
zonelist_node_idx(struct zoneref * zoneref)1020 static inline int zonelist_node_idx(struct zoneref *zoneref)
1021 {
1022 return zone_to_nid(zoneref->zone);
1023 }
1024
1025 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1026 enum zone_type highest_zoneidx,
1027 nodemask_t *nodes);
1028
1029 /**
1030 * 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
1031 * @z - The cursor used as a starting point for the search
1032 * @highest_zoneidx - The zone index of the highest zone to return
1033 * @nodes - An optional nodemask to filter the zonelist with
1034 *
1035 * This function returns the next zone at or below a given zone index that is
1036 * within the allowed nodemask using a cursor as the starting point for the
1037 * search. The zoneref returned is a cursor that represents the current zone
1038 * being examined. It should be advanced by one before calling
1039 * next_zones_zonelist again.
1040 */
next_zones_zonelist(struct zoneref * z,enum zone_type highest_zoneidx,nodemask_t * nodes)1041 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1042 enum zone_type highest_zoneidx,
1043 nodemask_t *nodes)
1044 {
1045 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1046 return z;
1047 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1048 }
1049
1050 /**
1051 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1052 * @zonelist - The zonelist to search for a suitable zone
1053 * @highest_zoneidx - The zone index of the highest zone to return
1054 * @nodes - An optional nodemask to filter the zonelist with
1055 * @return - Zoneref pointer for the first suitable zone found (see below)
1056 *
1057 * This function returns the first zone at or below a given zone index that is
1058 * within the allowed nodemask. The zoneref returned is a cursor that can be
1059 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1060 * one before calling.
1061 *
1062 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1063 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1064 * update due to cpuset modification.
1065 */
first_zones_zonelist(struct zonelist * zonelist,enum zone_type highest_zoneidx,nodemask_t * nodes)1066 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1067 enum zone_type highest_zoneidx,
1068 nodemask_t *nodes)
1069 {
1070 return next_zones_zonelist(zonelist->_zonerefs,
1071 highest_zoneidx, nodes);
1072 }
1073
1074 /**
1075 * 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
1076 * @zone - The current zone in the iterator
1077 * @z - The current pointer within zonelist->zones being iterated
1078 * @zlist - The zonelist being iterated
1079 * @highidx - The zone index of the highest zone to return
1080 * @nodemask - Nodemask allowed by the allocator
1081 *
1082 * This iterator iterates though all zones at or below a given zone index and
1083 * within a given nodemask
1084 */
1085 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1086 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1087 zone; \
1088 z = next_zones_zonelist(++z, highidx, nodemask), \
1089 zone = zonelist_zone(z))
1090
1091 #define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1092 for (zone = z->zone; \
1093 zone; \
1094 z = next_zones_zonelist(++z, highidx, nodemask), \
1095 zone = zonelist_zone(z))
1096
1097
1098 /**
1099 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1100 * @zone - The current zone in the iterator
1101 * @z - The current pointer within zonelist->zones being iterated
1102 * @zlist - The zonelist being iterated
1103 * @highidx - The zone index of the highest zone to return
1104 *
1105 * This iterator iterates though all zones at or below a given zone index.
1106 */
1107 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1108 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1109
1110 #ifdef CONFIG_SPARSEMEM
1111 #include <asm/sparsemem.h>
1112 #endif
1113
1114 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1115 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
early_pfn_to_nid(unsigned long pfn)1116 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1117 {
1118 BUILD_BUG_ON(IS_ENABLED(CONFIG_NUMA));
1119 return 0;
1120 }
1121 #endif
1122
1123 #ifdef CONFIG_FLATMEM
1124 #define pfn_to_nid(pfn) (0)
1125 #endif
1126
1127 #ifdef CONFIG_SPARSEMEM
1128
1129 /*
1130 * SECTION_SHIFT #bits space required to store a section #
1131 *
1132 * PA_SECTION_SHIFT physical address to/from section number
1133 * PFN_SECTION_SHIFT pfn to/from section number
1134 */
1135 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1136 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1137
1138 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1139
1140 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1141 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1142
1143 #define SECTION_BLOCKFLAGS_BITS \
1144 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1145
1146 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1147 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1148 #endif
1149
pfn_to_section_nr(unsigned long pfn)1150 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1151 {
1152 return pfn >> PFN_SECTION_SHIFT;
1153 }
section_nr_to_pfn(unsigned long sec)1154 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1155 {
1156 return sec << PFN_SECTION_SHIFT;
1157 }
1158
1159 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1160 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1161
1162 #define SUBSECTION_SHIFT 21
1163
1164 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1165 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1166 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1167
1168 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1169 #error Subsection size exceeds section size
1170 #else
1171 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1172 #endif
1173
1174 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1175 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1176
1177 struct mem_section_usage {
1178 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1179 /* See declaration of similar field in struct zone */
1180 unsigned long pageblock_flags[0];
1181 };
1182
1183 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1184
1185 struct page;
1186 struct page_ext;
1187 struct mem_section {
1188 /*
1189 * This is, logically, a pointer to an array of struct
1190 * pages. However, it is stored with some other magic.
1191 * (see sparse.c::sparse_init_one_section())
1192 *
1193 * Additionally during early boot we encode node id of
1194 * the location of the section here to guide allocation.
1195 * (see sparse.c::memory_present())
1196 *
1197 * Making it a UL at least makes someone do a cast
1198 * before using it wrong.
1199 */
1200 unsigned long section_mem_map;
1201
1202 struct mem_section_usage *usage;
1203 #ifdef CONFIG_PAGE_EXTENSION
1204 /*
1205 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1206 * section. (see page_ext.h about this.)
1207 */
1208 struct page_ext *page_ext;
1209 unsigned long pad;
1210 #endif
1211 /*
1212 * WARNING: mem_section must be a power-of-2 in size for the
1213 * calculation and use of SECTION_ROOT_MASK to make sense.
1214 */
1215 };
1216
1217 #ifdef CONFIG_SPARSEMEM_EXTREME
1218 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1219 #else
1220 #define SECTIONS_PER_ROOT 1
1221 #endif
1222
1223 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1224 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1225 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1226
1227 #ifdef CONFIG_SPARSEMEM_EXTREME
1228 extern struct mem_section **mem_section;
1229 #else
1230 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1231 #endif
1232
section_to_usemap(struct mem_section * ms)1233 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1234 {
1235 return ms->usage->pageblock_flags;
1236 }
1237
__nr_to_section(unsigned long nr)1238 static inline struct mem_section *__nr_to_section(unsigned long nr)
1239 {
1240 #ifdef CONFIG_SPARSEMEM_EXTREME
1241 if (!mem_section)
1242 return NULL;
1243 #endif
1244 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1245 return NULL;
1246 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1247 }
1248 extern unsigned long __section_nr(struct mem_section *ms);
1249 extern size_t mem_section_usage_size(void);
1250
1251 /*
1252 * We use the lower bits of the mem_map pointer to store
1253 * a little bit of information. The pointer is calculated
1254 * as mem_map - section_nr_to_pfn(pnum). The result is
1255 * aligned to the minimum alignment of the two values:
1256 * 1. All mem_map arrays are page-aligned.
1257 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1258 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1259 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1260 * worst combination is powerpc with 256k pages,
1261 * which results in PFN_SECTION_SHIFT equal 6.
1262 * To sum it up, at least 6 bits are available.
1263 */
1264 #define SECTION_MARKED_PRESENT (1UL<<0)
1265 #define SECTION_HAS_MEM_MAP (1UL<<1)
1266 #define SECTION_IS_ONLINE (1UL<<2)
1267 #define SECTION_IS_EARLY (1UL<<3)
1268 #define SECTION_MAP_LAST_BIT (1UL<<4)
1269 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1270 #define SECTION_NID_SHIFT 3
1271
__section_mem_map_addr(struct mem_section * section)1272 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1273 {
1274 unsigned long map = section->section_mem_map;
1275 map &= SECTION_MAP_MASK;
1276 return (struct page *)map;
1277 }
1278
present_section(struct mem_section * section)1279 static inline int present_section(struct mem_section *section)
1280 {
1281 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1282 }
1283
present_section_nr(unsigned long nr)1284 static inline int present_section_nr(unsigned long nr)
1285 {
1286 return present_section(__nr_to_section(nr));
1287 }
1288
valid_section(struct mem_section * section)1289 static inline int valid_section(struct mem_section *section)
1290 {
1291 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1292 }
1293
early_section(struct mem_section * section)1294 static inline int early_section(struct mem_section *section)
1295 {
1296 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1297 }
1298
valid_section_nr(unsigned long nr)1299 static inline int valid_section_nr(unsigned long nr)
1300 {
1301 return valid_section(__nr_to_section(nr));
1302 }
1303
online_section(struct mem_section * section)1304 static inline int online_section(struct mem_section *section)
1305 {
1306 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1307 }
1308
online_section_nr(unsigned long nr)1309 static inline int online_section_nr(unsigned long nr)
1310 {
1311 return online_section(__nr_to_section(nr));
1312 }
1313
1314 #ifdef CONFIG_MEMORY_HOTPLUG
1315 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1316 #ifdef CONFIG_MEMORY_HOTREMOVE
1317 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1318 #endif
1319 #endif
1320
__pfn_to_section(unsigned long pfn)1321 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1322 {
1323 return __nr_to_section(pfn_to_section_nr(pfn));
1324 }
1325
1326 extern unsigned long __highest_present_section_nr;
1327
subsection_map_index(unsigned long pfn)1328 static inline int subsection_map_index(unsigned long pfn)
1329 {
1330 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1331 }
1332
1333 #ifdef CONFIG_SPARSEMEM_VMEMMAP
pfn_section_valid(struct mem_section * ms,unsigned long pfn)1334 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1335 {
1336 int idx = subsection_map_index(pfn);
1337
1338 return test_bit(idx, ms->usage->subsection_map);
1339 }
1340 #else
pfn_section_valid(struct mem_section * ms,unsigned long pfn)1341 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1342 {
1343 return 1;
1344 }
1345 #endif
1346
1347 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
pfn_valid(unsigned long pfn)1348 static inline int pfn_valid(unsigned long pfn)
1349 {
1350 struct mem_section *ms;
1351
1352 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1353 return 0;
1354 ms = __nr_to_section(pfn_to_section_nr(pfn));
1355 if (!valid_section(ms))
1356 return 0;
1357 /*
1358 * Traditionally early sections always returned pfn_valid() for
1359 * the entire section-sized span.
1360 */
1361 return early_section(ms) || pfn_section_valid(ms, pfn);
1362 }
1363 #endif
1364
pfn_present(unsigned long pfn)1365 static inline int pfn_present(unsigned long pfn)
1366 {
1367 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1368 return 0;
1369 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1370 }
1371
1372 /*
1373 * These are _only_ used during initialisation, therefore they
1374 * can use __initdata ... They could have names to indicate
1375 * this restriction.
1376 */
1377 #ifdef CONFIG_NUMA
1378 #define pfn_to_nid(pfn) \
1379 ({ \
1380 unsigned long __pfn_to_nid_pfn = (pfn); \
1381 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1382 })
1383 #else
1384 #define pfn_to_nid(pfn) (0)
1385 #endif
1386
1387 #define early_pfn_valid(pfn) pfn_valid(pfn)
1388 void sparse_init(void);
1389 #else
1390 #define sparse_init() do {} while (0)
1391 #define sparse_index_init(_sec, _nid) do {} while (0)
1392 #define pfn_present pfn_valid
1393 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
1394 #endif /* CONFIG_SPARSEMEM */
1395
1396 /*
1397 * During memory init memblocks map pfns to nids. The search is expensive and
1398 * this caches recent lookups. The implementation of __early_pfn_to_nid
1399 * may treat start/end as pfns or sections.
1400 */
1401 struct mminit_pfnnid_cache {
1402 unsigned long last_start;
1403 unsigned long last_end;
1404 int last_nid;
1405 };
1406
1407 #ifndef early_pfn_valid
1408 #define early_pfn_valid(pfn) (1)
1409 #endif
1410
1411 void memory_present(int nid, unsigned long start, unsigned long end);
1412
1413 /*
1414 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1415 * need to check pfn validity within that MAX_ORDER_NR_PAGES block.
1416 * pfn_valid_within() should be used in this case; we optimise this away
1417 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1418 */
1419 #ifdef CONFIG_HOLES_IN_ZONE
1420 #define pfn_valid_within(pfn) pfn_valid(pfn)
1421 #else
1422 #define pfn_valid_within(pfn) (1)
1423 #endif
1424
1425 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1426 /*
1427 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1428 * associated with it or not. This means that a struct page exists for this
1429 * pfn. The caller cannot assume the page is fully initialized in general.
1430 * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1431 * will ensure the struct page is fully online and initialized. Special pages
1432 * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1433 *
1434 * In FLATMEM, it is expected that holes always have valid memmap as long as
1435 * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1436 * that a valid section has a memmap for the entire section.
1437 *
1438 * However, an ARM, and maybe other embedded architectures in the future
1439 * free memmap backing holes to save memory on the assumption the memmap is
1440 * never used. The page_zone linkages are then broken even though pfn_valid()
1441 * returns true. A walker of the full memmap must then do this additional
1442 * check to ensure the memmap they are looking at is sane by making sure
1443 * the zone and PFN linkages are still valid. This is expensive, but walkers
1444 * of the full memmap are extremely rare.
1445 */
1446 bool memmap_valid_within(unsigned long pfn,
1447 struct page *page, struct zone *zone);
1448 #else
memmap_valid_within(unsigned long pfn,struct page * page,struct zone * zone)1449 static inline bool memmap_valid_within(unsigned long pfn,
1450 struct page *page, struct zone *zone)
1451 {
1452 return true;
1453 }
1454 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1455
1456 #endif /* !__GENERATING_BOUNDS.H */
1457 #endif /* !__ASSEMBLY__ */
1458 #endif /* _LINUX_MMZONE_H */
1459