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