1 #ifndef _LINUX_MMZONE_H
2 #define _LINUX_MMZONE_H
3
4 #ifndef __ASSEMBLY__
5 #ifndef __GENERATING_BOUNDS_H
6
7 #include <linux/spinlock.h>
8 #include <linux/list.h>
9 #include <linux/wait.h>
10 #include <linux/bitops.h>
11 #include <linux/cache.h>
12 #include <linux/threads.h>
13 #include <linux/numa.h>
14 #include <linux/init.h>
15 #include <linux/seqlock.h>
16 #include <linux/nodemask.h>
17 #include <linux/pageblock-flags.h>
18 #include <linux/page-flags-layout.h>
19 #include <linux/atomic.h>
20 #include <asm/page.h>
21
22 /* Free memory management - zoned buddy allocator. */
23 #ifndef CONFIG_FORCE_MAX_ZONEORDER
24 #define MAX_ORDER 11
25 #else
26 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
27 #endif
28 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
29
30 /*
31 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
32 * costly to service. That is between allocation orders which should
33 * coalesce naturally under reasonable reclaim pressure and those which
34 * will not.
35 */
36 #define PAGE_ALLOC_COSTLY_ORDER 3
37
38 enum {
39 MIGRATE_UNMOVABLE,
40 MIGRATE_RECLAIMABLE,
41 MIGRATE_MOVABLE,
42 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
43 MIGRATE_RESERVE = MIGRATE_PCPTYPES,
44 #ifdef CONFIG_CMA
45 /*
46 * MIGRATE_CMA migration type is designed to mimic the way
47 * ZONE_MOVABLE works. Only movable pages can be allocated
48 * from MIGRATE_CMA pageblocks and page allocator never
49 * implicitly change migration type of MIGRATE_CMA pageblock.
50 *
51 * The way to use it is to change migratetype of a range of
52 * pageblocks to MIGRATE_CMA which can be done by
53 * __free_pageblock_cma() function. What is important though
54 * is that a range of pageblocks must be aligned to
55 * MAX_ORDER_NR_PAGES should biggest page be bigger then
56 * a single pageblock.
57 */
58 MIGRATE_CMA,
59 #endif
60 #ifdef CONFIG_MEMORY_ISOLATION
61 MIGRATE_ISOLATE, /* can't allocate from here */
62 #endif
63 MIGRATE_TYPES
64 };
65
66 #ifdef CONFIG_CMA
67 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
68 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
69 #else
70 # define is_migrate_cma(migratetype) false
71 # define is_migrate_cma_page(_page) false
72 #endif
73
74 #define for_each_migratetype_order(order, type) \
75 for (order = 0; order < MAX_ORDER; order++) \
76 for (type = 0; type < MIGRATE_TYPES; type++)
77
78 extern int page_group_by_mobility_disabled;
79
80 #define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1)
81 #define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1)
82
83 #define get_pageblock_migratetype(page) \
84 get_pfnblock_flags_mask(page, page_to_pfn(page), \
85 PB_migrate_end, MIGRATETYPE_MASK)
86
get_pfnblock_migratetype(struct page * page,unsigned long pfn)87 static inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn)
88 {
89 BUILD_BUG_ON(PB_migrate_end - PB_migrate != 2);
90 return get_pfnblock_flags_mask(page, pfn, PB_migrate_end,
91 MIGRATETYPE_MASK);
92 }
93
94 struct free_area {
95 struct list_head free_list[MIGRATE_TYPES];
96 unsigned long nr_free;
97 };
98
99 struct pglist_data;
100
101 /*
102 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
103 * So add a wild amount of padding here to ensure that they fall into separate
104 * cachelines. There are very few zone structures in the machine, so space
105 * consumption is not a concern here.
106 */
107 #if defined(CONFIG_SMP)
108 struct zone_padding {
109 char x[0];
110 } ____cacheline_internodealigned_in_smp;
111 #define ZONE_PADDING(name) struct zone_padding name;
112 #else
113 #define ZONE_PADDING(name)
114 #endif
115
116 enum zone_stat_item {
117 /* First 128 byte cacheline (assuming 64 bit words) */
118 NR_FREE_PAGES,
119 NR_ALLOC_BATCH,
120 NR_LRU_BASE,
121 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
122 NR_ACTIVE_ANON, /* " " " " " */
123 NR_INACTIVE_FILE, /* " " " " " */
124 NR_ACTIVE_FILE, /* " " " " " */
125 NR_UNEVICTABLE, /* " " " " " */
126 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
127 NR_ANON_PAGES, /* Mapped anonymous pages */
128 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
129 only modified from process context */
130 NR_FILE_PAGES,
131 NR_FILE_DIRTY,
132 NR_WRITEBACK,
133 NR_SLAB_RECLAIMABLE,
134 NR_SLAB_UNRECLAIMABLE,
135 NR_PAGETABLE, /* used for pagetables */
136 NR_KERNEL_STACK,
137 /* Second 128 byte cacheline */
138 NR_UNSTABLE_NFS, /* NFS unstable pages */
139 NR_BOUNCE,
140 NR_VMSCAN_WRITE,
141 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
142 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
143 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
144 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
145 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
146 NR_DIRTIED, /* page dirtyings since bootup */
147 NR_WRITTEN, /* page writings since bootup */
148 NR_PAGES_SCANNED, /* pages scanned since last reclaim */
149 #ifdef CONFIG_NUMA
150 NUMA_HIT, /* allocated in intended node */
151 NUMA_MISS, /* allocated in non intended node */
152 NUMA_FOREIGN, /* was intended here, hit elsewhere */
153 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
154 NUMA_LOCAL, /* allocation from local node */
155 NUMA_OTHER, /* allocation from other node */
156 #endif
157 WORKINGSET_REFAULT,
158 WORKINGSET_ACTIVATE,
159 WORKINGSET_NODERECLAIM,
160 NR_ANON_TRANSPARENT_HUGEPAGES,
161 NR_FREE_CMA_PAGES,
162 NR_VM_ZONE_STAT_ITEMS };
163
164 /*
165 * We do arithmetic on the LRU lists in various places in the code,
166 * so it is important to keep the active lists LRU_ACTIVE higher in
167 * the array than the corresponding inactive lists, and to keep
168 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
169 *
170 * This has to be kept in sync with the statistics in zone_stat_item
171 * above and the descriptions in vmstat_text in mm/vmstat.c
172 */
173 #define LRU_BASE 0
174 #define LRU_ACTIVE 1
175 #define LRU_FILE 2
176
177 enum lru_list {
178 LRU_INACTIVE_ANON = LRU_BASE,
179 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
180 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
181 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
182 LRU_UNEVICTABLE,
183 NR_LRU_LISTS
184 };
185
186 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
187
188 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
189
is_file_lru(enum lru_list lru)190 static inline int is_file_lru(enum lru_list lru)
191 {
192 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
193 }
194
is_active_lru(enum lru_list lru)195 static inline int is_active_lru(enum lru_list lru)
196 {
197 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
198 }
199
is_unevictable_lru(enum lru_list lru)200 static inline int is_unevictable_lru(enum lru_list lru)
201 {
202 return (lru == LRU_UNEVICTABLE);
203 }
204
205 struct zone_reclaim_stat {
206 /*
207 * The pageout code in vmscan.c keeps track of how many of the
208 * mem/swap backed and file backed pages are referenced.
209 * The higher the rotated/scanned ratio, the more valuable
210 * that cache is.
211 *
212 * The anon LRU stats live in [0], file LRU stats in [1]
213 */
214 unsigned long recent_rotated[2];
215 unsigned long recent_scanned[2];
216 };
217
218 struct lruvec {
219 struct list_head lists[NR_LRU_LISTS];
220 struct zone_reclaim_stat reclaim_stat;
221 #ifdef CONFIG_MEMCG
222 struct zone *zone;
223 #endif
224 };
225
226 /* Mask used at gathering information at once (see memcontrol.c) */
227 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
228 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
229 #define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
230
231 /* Isolate clean file */
232 #define ISOLATE_CLEAN ((__force isolate_mode_t)0x1)
233 /* Isolate unmapped file */
234 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
235 /* Isolate for asynchronous migration */
236 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
237 /* Isolate unevictable pages */
238 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
239
240 /* LRU Isolation modes. */
241 typedef unsigned __bitwise__ isolate_mode_t;
242
243 enum zone_watermarks {
244 WMARK_MIN,
245 WMARK_LOW,
246 WMARK_HIGH,
247 NR_WMARK
248 };
249
250 #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
251 #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
252 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
253
254 struct per_cpu_pages {
255 int count; /* number of pages in the list */
256 int high; /* high watermark, emptying needed */
257 int batch; /* chunk size for buddy add/remove */
258
259 /* Lists of pages, one per migrate type stored on the pcp-lists */
260 struct list_head lists[MIGRATE_PCPTYPES];
261 };
262
263 struct per_cpu_pageset {
264 struct per_cpu_pages pcp;
265 #ifdef CONFIG_NUMA
266 s8 expire;
267 #endif
268 #ifdef CONFIG_SMP
269 s8 stat_threshold;
270 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
271 #endif
272 };
273
274 #endif /* !__GENERATING_BOUNDS.H */
275
276 enum zone_type {
277 #ifdef CONFIG_ZONE_DMA
278 /*
279 * ZONE_DMA is used when there are devices that are not able
280 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
281 * carve out the portion of memory that is needed for these devices.
282 * The range is arch specific.
283 *
284 * Some examples
285 *
286 * Architecture Limit
287 * ---------------------------
288 * parisc, ia64, sparc <4G
289 * s390 <2G
290 * arm Various
291 * alpha Unlimited or 0-16MB.
292 *
293 * i386, x86_64 and multiple other arches
294 * <16M.
295 */
296 ZONE_DMA,
297 #endif
298 #ifdef CONFIG_ZONE_DMA32
299 /*
300 * x86_64 needs two ZONE_DMAs because it supports devices that are
301 * only able to do DMA to the lower 16M but also 32 bit devices that
302 * can only do DMA areas below 4G.
303 */
304 ZONE_DMA32,
305 #endif
306 /*
307 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
308 * performed on pages in ZONE_NORMAL if the DMA devices support
309 * transfers to all addressable memory.
310 */
311 ZONE_NORMAL,
312 #ifdef CONFIG_HIGHMEM
313 /*
314 * A memory area that is only addressable by the kernel through
315 * mapping portions into its own address space. This is for example
316 * used by i386 to allow the kernel to address the memory beyond
317 * 900MB. The kernel will set up special mappings (page
318 * table entries on i386) for each page that the kernel needs to
319 * access.
320 */
321 ZONE_HIGHMEM,
322 #endif
323 ZONE_MOVABLE,
324 __MAX_NR_ZONES
325 };
326
327 #ifndef __GENERATING_BOUNDS_H
328
329 struct zone {
330 /* Read-mostly fields */
331
332 /* zone watermarks, access with *_wmark_pages(zone) macros */
333 unsigned long watermark[NR_WMARK];
334
335 /*
336 * We don't know if the memory that we're going to allocate will be freeable
337 * or/and it will be released eventually, so to avoid totally wasting several
338 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
339 * to run OOM on the lower zones despite there's tons of freeable ram
340 * on the higher zones). This array is recalculated at runtime if the
341 * sysctl_lowmem_reserve_ratio sysctl changes.
342 */
343 long lowmem_reserve[MAX_NR_ZONES];
344
345 #ifdef CONFIG_NUMA
346 int node;
347 #endif
348
349 /*
350 * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on
351 * this zone's LRU. Maintained by the pageout code.
352 */
353 unsigned int inactive_ratio;
354
355 struct pglist_data *zone_pgdat;
356 struct per_cpu_pageset __percpu *pageset;
357
358 /*
359 * This is a per-zone reserve of pages that should not be
360 * considered dirtyable memory.
361 */
362 unsigned long dirty_balance_reserve;
363
364 #ifndef CONFIG_SPARSEMEM
365 /*
366 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
367 * In SPARSEMEM, this map is stored in struct mem_section
368 */
369 unsigned long *pageblock_flags;
370 #endif /* CONFIG_SPARSEMEM */
371
372 #ifdef CONFIG_NUMA
373 /*
374 * zone reclaim becomes active if more unmapped pages exist.
375 */
376 unsigned long min_unmapped_pages;
377 unsigned long min_slab_pages;
378 #endif /* CONFIG_NUMA */
379
380 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
381 unsigned long zone_start_pfn;
382
383 /*
384 * spanned_pages is the total pages spanned by the zone, including
385 * holes, which is calculated as:
386 * spanned_pages = zone_end_pfn - zone_start_pfn;
387 *
388 * present_pages is physical pages existing within the zone, which
389 * is calculated as:
390 * present_pages = spanned_pages - absent_pages(pages in holes);
391 *
392 * managed_pages is present pages managed by the buddy system, which
393 * is calculated as (reserved_pages includes pages allocated by the
394 * bootmem allocator):
395 * managed_pages = present_pages - reserved_pages;
396 *
397 * So present_pages may be used by memory hotplug or memory power
398 * management logic to figure out unmanaged pages by checking
399 * (present_pages - managed_pages). And managed_pages should be used
400 * by page allocator and vm scanner to calculate all kinds of watermarks
401 * and thresholds.
402 *
403 * Locking rules:
404 *
405 * zone_start_pfn and spanned_pages are protected by span_seqlock.
406 * It is a seqlock because it has to be read outside of zone->lock,
407 * and it is done in the main allocator path. But, it is written
408 * quite infrequently.
409 *
410 * The span_seq lock is declared along with zone->lock because it is
411 * frequently read in proximity to zone->lock. It's good to
412 * give them a chance of being in the same cacheline.
413 *
414 * Write access to present_pages at runtime should be protected by
415 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
416 * present_pages should get_online_mems() to get a stable value.
417 *
418 * Read access to managed_pages should be safe because it's unsigned
419 * long. Write access to zone->managed_pages and totalram_pages are
420 * protected by managed_page_count_lock at runtime. Idealy only
421 * adjust_managed_page_count() should be used instead of directly
422 * touching zone->managed_pages and totalram_pages.
423 */
424 unsigned long managed_pages;
425 unsigned long spanned_pages;
426 unsigned long present_pages;
427
428 const char *name;
429
430 /*
431 * Number of MIGRATE_RESEVE page block. To maintain for just
432 * optimization. Protected by zone->lock.
433 */
434 int nr_migrate_reserve_block;
435
436 #ifdef CONFIG_MEMORY_ISOLATION
437 /*
438 * Number of isolated pageblock. It is used to solve incorrect
439 * freepage counting problem due to racy retrieving migratetype
440 * of pageblock. Protected by zone->lock.
441 */
442 unsigned long nr_isolate_pageblock;
443 #endif
444
445 #ifdef CONFIG_MEMORY_HOTPLUG
446 /* see spanned/present_pages for more description */
447 seqlock_t span_seqlock;
448 #endif
449
450 /*
451 * wait_table -- the array holding the hash table
452 * wait_table_hash_nr_entries -- the size of the hash table array
453 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
454 *
455 * The purpose of all these is to keep track of the people
456 * waiting for a page to become available and make them
457 * runnable again when possible. The trouble is that this
458 * consumes a lot of space, especially when so few things
459 * wait on pages at a given time. So instead of using
460 * per-page waitqueues, we use a waitqueue hash table.
461 *
462 * The bucket discipline is to sleep on the same queue when
463 * colliding and wake all in that wait queue when removing.
464 * When something wakes, it must check to be sure its page is
465 * truly available, a la thundering herd. The cost of a
466 * collision is great, but given the expected load of the
467 * table, they should be so rare as to be outweighed by the
468 * benefits from the saved space.
469 *
470 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
471 * primary users of these fields, and in mm/page_alloc.c
472 * free_area_init_core() performs the initialization of them.
473 */
474 wait_queue_head_t *wait_table;
475 unsigned long wait_table_hash_nr_entries;
476 unsigned long wait_table_bits;
477
478 ZONE_PADDING(_pad1_)
479
480 /* Write-intensive fields used from the page allocator */
481 spinlock_t lock;
482
483 /* free areas of different sizes */
484 struct free_area free_area[MAX_ORDER];
485
486 /* zone flags, see below */
487 unsigned long flags;
488
489 ZONE_PADDING(_pad2_)
490
491 /* Write-intensive fields used by page reclaim */
492
493 /* Fields commonly accessed by the page reclaim scanner */
494 spinlock_t lru_lock;
495 struct lruvec lruvec;
496
497 /* Evictions & activations on the inactive file list */
498 atomic_long_t inactive_age;
499
500 /*
501 * When free pages are below this point, additional steps are taken
502 * when reading the number of free pages to avoid per-cpu counter
503 * drift allowing watermarks to be breached
504 */
505 unsigned long percpu_drift_mark;
506
507 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
508 /* pfn where compaction free scanner should start */
509 unsigned long compact_cached_free_pfn;
510 /* pfn where async and sync compaction migration scanner should start */
511 unsigned long compact_cached_migrate_pfn[2];
512 #endif
513
514 #ifdef CONFIG_COMPACTION
515 /*
516 * On compaction failure, 1<<compact_defer_shift compactions
517 * are skipped before trying again. The number attempted since
518 * last failure is tracked with compact_considered.
519 */
520 unsigned int compact_considered;
521 unsigned int compact_defer_shift;
522 int compact_order_failed;
523 #endif
524
525 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
526 /* Set to true when the PG_migrate_skip bits should be cleared */
527 bool compact_blockskip_flush;
528 #endif
529
530 ZONE_PADDING(_pad3_)
531 /* Zone statistics */
532 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
533 } ____cacheline_internodealigned_in_smp;
534
535 enum zone_flags {
536 ZONE_RECLAIM_LOCKED, /* prevents concurrent reclaim */
537 ZONE_OOM_LOCKED, /* zone is in OOM killer zonelist */
538 ZONE_CONGESTED, /* zone has many dirty pages backed by
539 * a congested BDI
540 */
541 ZONE_DIRTY, /* reclaim scanning has recently found
542 * many dirty file pages at the tail
543 * of the LRU.
544 */
545 ZONE_WRITEBACK, /* reclaim scanning has recently found
546 * many pages under writeback
547 */
548 ZONE_FAIR_DEPLETED, /* fair zone policy batch depleted */
549 };
550
zone_end_pfn(const struct zone * zone)551 static inline unsigned long zone_end_pfn(const struct zone *zone)
552 {
553 return zone->zone_start_pfn + zone->spanned_pages;
554 }
555
zone_spans_pfn(const struct zone * zone,unsigned long pfn)556 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
557 {
558 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
559 }
560
zone_is_initialized(struct zone * zone)561 static inline bool zone_is_initialized(struct zone *zone)
562 {
563 return !!zone->wait_table;
564 }
565
zone_is_empty(struct zone * zone)566 static inline bool zone_is_empty(struct zone *zone)
567 {
568 return zone->spanned_pages == 0;
569 }
570
571 /*
572 * The "priority" of VM scanning is how much of the queues we will scan in one
573 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
574 * queues ("queue_length >> 12") during an aging round.
575 */
576 #define DEF_PRIORITY 12
577
578 /* Maximum number of zones on a zonelist */
579 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
580
581 #ifdef CONFIG_NUMA
582
583 /*
584 * The NUMA zonelists are doubled because we need zonelists that restrict the
585 * allocations to a single node for __GFP_THISNODE.
586 *
587 * [0] : Zonelist with fallback
588 * [1] : No fallback (__GFP_THISNODE)
589 */
590 #define MAX_ZONELISTS 2
591
592
593 /*
594 * We cache key information from each zonelist for smaller cache
595 * footprint when scanning for free pages in get_page_from_freelist().
596 *
597 * 1) The BITMAP fullzones tracks which zones in a zonelist have come
598 * up short of free memory since the last time (last_fullzone_zap)
599 * we zero'd fullzones.
600 * 2) The array z_to_n[] maps each zone in the zonelist to its node
601 * id, so that we can efficiently evaluate whether that node is
602 * set in the current tasks mems_allowed.
603 *
604 * Both fullzones and z_to_n[] are one-to-one with the zonelist,
605 * indexed by a zones offset in the zonelist zones[] array.
606 *
607 * The get_page_from_freelist() routine does two scans. During the
608 * first scan, we skip zones whose corresponding bit in 'fullzones'
609 * is set or whose corresponding node in current->mems_allowed (which
610 * comes from cpusets) is not set. During the second scan, we bypass
611 * this zonelist_cache, to ensure we look methodically at each zone.
612 *
613 * Once per second, we zero out (zap) fullzones, forcing us to
614 * reconsider nodes that might have regained more free memory.
615 * The field last_full_zap is the time we last zapped fullzones.
616 *
617 * This mechanism reduces the amount of time we waste repeatedly
618 * reexaming zones for free memory when they just came up low on
619 * memory momentarilly ago.
620 *
621 * The zonelist_cache struct members logically belong in struct
622 * zonelist. However, the mempolicy zonelists constructed for
623 * MPOL_BIND are intentionally variable length (and usually much
624 * shorter). A general purpose mechanism for handling structs with
625 * multiple variable length members is more mechanism than we want
626 * here. We resort to some special case hackery instead.
627 *
628 * The MPOL_BIND zonelists don't need this zonelist_cache (in good
629 * part because they are shorter), so we put the fixed length stuff
630 * at the front of the zonelist struct, ending in a variable length
631 * zones[], as is needed by MPOL_BIND.
632 *
633 * Then we put the optional zonelist cache on the end of the zonelist
634 * struct. This optional stuff is found by a 'zlcache_ptr' pointer in
635 * the fixed length portion at the front of the struct. This pointer
636 * both enables us to find the zonelist cache, and in the case of
637 * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL)
638 * to know that the zonelist cache is not there.
639 *
640 * The end result is that struct zonelists come in two flavors:
641 * 1) The full, fixed length version, shown below, and
642 * 2) The custom zonelists for MPOL_BIND.
643 * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache.
644 *
645 * Even though there may be multiple CPU cores on a node modifying
646 * fullzones or last_full_zap in the same zonelist_cache at the same
647 * time, we don't lock it. This is just hint data - if it is wrong now
648 * and then, the allocator will still function, perhaps a bit slower.
649 */
650
651
652 struct zonelist_cache {
653 unsigned short z_to_n[MAX_ZONES_PER_ZONELIST]; /* zone->nid */
654 DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST); /* zone full? */
655 unsigned long last_full_zap; /* when last zap'd (jiffies) */
656 };
657 #else
658 #define MAX_ZONELISTS 1
659 struct zonelist_cache;
660 #endif
661
662 /*
663 * This struct contains information about a zone in a zonelist. It is stored
664 * here to avoid dereferences into large structures and lookups of tables
665 */
666 struct zoneref {
667 struct zone *zone; /* Pointer to actual zone */
668 int zone_idx; /* zone_idx(zoneref->zone) */
669 };
670
671 /*
672 * One allocation request operates on a zonelist. A zonelist
673 * is a list of zones, the first one is the 'goal' of the
674 * allocation, the other zones are fallback zones, in decreasing
675 * priority.
676 *
677 * If zlcache_ptr is not NULL, then it is just the address of zlcache,
678 * as explained above. If zlcache_ptr is NULL, there is no zlcache.
679 * *
680 * To speed the reading of the zonelist, the zonerefs contain the zone index
681 * of the entry being read. Helper functions to access information given
682 * a struct zoneref are
683 *
684 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
685 * zonelist_zone_idx() - Return the index of the zone for an entry
686 * zonelist_node_idx() - Return the index of the node for an entry
687 */
688 struct zonelist {
689 struct zonelist_cache *zlcache_ptr; // NULL or &zlcache
690 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
691 #ifdef CONFIG_NUMA
692 struct zonelist_cache zlcache; // optional ...
693 #endif
694 };
695
696 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
697 struct node_active_region {
698 unsigned long start_pfn;
699 unsigned long end_pfn;
700 int nid;
701 };
702 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
703
704 #ifndef CONFIG_DISCONTIGMEM
705 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
706 extern struct page *mem_map;
707 #endif
708
709 /*
710 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
711 * (mostly NUMA machines?) to denote a higher-level memory zone than the
712 * zone denotes.
713 *
714 * On NUMA machines, each NUMA node would have a pg_data_t to describe
715 * it's memory layout.
716 *
717 * Memory statistics and page replacement data structures are maintained on a
718 * per-zone basis.
719 */
720 struct bootmem_data;
721 typedef struct pglist_data {
722 struct zone node_zones[MAX_NR_ZONES];
723 struct zonelist node_zonelists[MAX_ZONELISTS];
724 int nr_zones;
725 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
726 struct page *node_mem_map;
727 #ifdef CONFIG_MEMCG
728 struct page_cgroup *node_page_cgroup;
729 #endif
730 #endif
731 #ifndef CONFIG_NO_BOOTMEM
732 struct bootmem_data *bdata;
733 #endif
734 #ifdef CONFIG_MEMORY_HOTPLUG
735 /*
736 * Must be held any time you expect node_start_pfn, node_present_pages
737 * or node_spanned_pages stay constant. Holding this will also
738 * guarantee that any pfn_valid() stays that way.
739 *
740 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
741 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG.
742 *
743 * Nests above zone->lock and zone->span_seqlock
744 */
745 spinlock_t node_size_lock;
746 #endif
747 unsigned long node_start_pfn;
748 unsigned long node_present_pages; /* total number of physical pages */
749 unsigned long node_spanned_pages; /* total size of physical page
750 range, including holes */
751 int node_id;
752 wait_queue_head_t kswapd_wait;
753 wait_queue_head_t pfmemalloc_wait;
754 struct task_struct *kswapd; /* Protected by
755 mem_hotplug_begin/end() */
756 int kswapd_max_order;
757 enum zone_type classzone_idx;
758 #ifdef CONFIG_NUMA_BALANCING
759 /* Lock serializing the migrate rate limiting window */
760 spinlock_t numabalancing_migrate_lock;
761
762 /* Rate limiting time interval */
763 unsigned long numabalancing_migrate_next_window;
764
765 /* Number of pages migrated during the rate limiting time interval */
766 unsigned long numabalancing_migrate_nr_pages;
767 #endif
768 } pg_data_t;
769
770 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
771 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
772 #ifdef CONFIG_FLAT_NODE_MEM_MAP
773 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
774 #else
775 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
776 #endif
777 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
778
779 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
780 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
781
pgdat_end_pfn(pg_data_t * pgdat)782 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
783 {
784 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
785 }
786
pgdat_is_empty(pg_data_t * pgdat)787 static inline bool pgdat_is_empty(pg_data_t *pgdat)
788 {
789 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
790 }
791
792 #include <linux/memory_hotplug.h>
793
794 extern struct mutex zonelists_mutex;
795 void build_all_zonelists(pg_data_t *pgdat, struct zone *zone);
796 void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx);
797 bool zone_watermark_ok(struct zone *z, unsigned int order,
798 unsigned long mark, int classzone_idx, int alloc_flags);
799 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
800 unsigned long mark, int classzone_idx, int alloc_flags);
801 enum memmap_context {
802 MEMMAP_EARLY,
803 MEMMAP_HOTPLUG,
804 };
805 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
806 unsigned long size,
807 enum memmap_context context);
808
809 extern void lruvec_init(struct lruvec *lruvec);
810
lruvec_zone(struct lruvec * lruvec)811 static inline struct zone *lruvec_zone(struct lruvec *lruvec)
812 {
813 #ifdef CONFIG_MEMCG
814 return lruvec->zone;
815 #else
816 return container_of(lruvec, struct zone, lruvec);
817 #endif
818 }
819
820 #ifdef CONFIG_HAVE_MEMORY_PRESENT
821 void memory_present(int nid, unsigned long start, unsigned long end);
822 #else
memory_present(int nid,unsigned long start,unsigned long end)823 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
824 #endif
825
826 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
827 int local_memory_node(int node_id);
828 #else
local_memory_node(int node_id)829 static inline int local_memory_node(int node_id) { return node_id; };
830 #endif
831
832 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
833 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
834 #endif
835
836 /*
837 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
838 */
839 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
840
populated_zone(struct zone * zone)841 static inline int populated_zone(struct zone *zone)
842 {
843 return (!!zone->present_pages);
844 }
845
846 extern int movable_zone;
847
zone_movable_is_highmem(void)848 static inline int zone_movable_is_highmem(void)
849 {
850 #if defined(CONFIG_HIGHMEM) && defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
851 return movable_zone == ZONE_HIGHMEM;
852 #elif defined(CONFIG_HIGHMEM)
853 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
854 #else
855 return 0;
856 #endif
857 }
858
is_highmem_idx(enum zone_type idx)859 static inline int is_highmem_idx(enum zone_type idx)
860 {
861 #ifdef CONFIG_HIGHMEM
862 return (idx == ZONE_HIGHMEM ||
863 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
864 #else
865 return 0;
866 #endif
867 }
868
869 /**
870 * is_highmem - helper function to quickly check if a struct zone is a
871 * highmem zone or not. This is an attempt to keep references
872 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
873 * @zone - pointer to struct zone variable
874 */
is_highmem(struct zone * zone)875 static inline int is_highmem(struct zone *zone)
876 {
877 #ifdef CONFIG_HIGHMEM
878 int zone_off = (char *)zone - (char *)zone->zone_pgdat->node_zones;
879 return zone_off == ZONE_HIGHMEM * sizeof(*zone) ||
880 (zone_off == ZONE_MOVABLE * sizeof(*zone) &&
881 zone_movable_is_highmem());
882 #else
883 return 0;
884 #endif
885 }
886
887 /* These two functions are used to setup the per zone pages min values */
888 struct ctl_table;
889 int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
890 void __user *, size_t *, loff_t *);
891 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
892 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
893 void __user *, size_t *, loff_t *);
894 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
895 void __user *, size_t *, loff_t *);
896 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
897 void __user *, size_t *, loff_t *);
898 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
899 void __user *, size_t *, loff_t *);
900
901 extern int numa_zonelist_order_handler(struct ctl_table *, int,
902 void __user *, size_t *, loff_t *);
903 extern char numa_zonelist_order[];
904 #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
905
906 #ifndef CONFIG_NEED_MULTIPLE_NODES
907
908 extern struct pglist_data contig_page_data;
909 #define NODE_DATA(nid) (&contig_page_data)
910 #define NODE_MEM_MAP(nid) mem_map
911
912 #else /* CONFIG_NEED_MULTIPLE_NODES */
913
914 #include <asm/mmzone.h>
915
916 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
917
918 extern struct pglist_data *first_online_pgdat(void);
919 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
920 extern struct zone *next_zone(struct zone *zone);
921
922 /**
923 * for_each_online_pgdat - helper macro to iterate over all online nodes
924 * @pgdat - pointer to a pg_data_t variable
925 */
926 #define for_each_online_pgdat(pgdat) \
927 for (pgdat = first_online_pgdat(); \
928 pgdat; \
929 pgdat = next_online_pgdat(pgdat))
930 /**
931 * for_each_zone - helper macro to iterate over all memory zones
932 * @zone - pointer to struct zone variable
933 *
934 * The user only needs to declare the zone variable, for_each_zone
935 * fills it in.
936 */
937 #define for_each_zone(zone) \
938 for (zone = (first_online_pgdat())->node_zones; \
939 zone; \
940 zone = next_zone(zone))
941
942 #define for_each_populated_zone(zone) \
943 for (zone = (first_online_pgdat())->node_zones; \
944 zone; \
945 zone = next_zone(zone)) \
946 if (!populated_zone(zone)) \
947 ; /* do nothing */ \
948 else
949
zonelist_zone(struct zoneref * zoneref)950 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
951 {
952 return zoneref->zone;
953 }
954
zonelist_zone_idx(struct zoneref * zoneref)955 static inline int zonelist_zone_idx(struct zoneref *zoneref)
956 {
957 return zoneref->zone_idx;
958 }
959
zonelist_node_idx(struct zoneref * zoneref)960 static inline int zonelist_node_idx(struct zoneref *zoneref)
961 {
962 #ifdef CONFIG_NUMA
963 /* zone_to_nid not available in this context */
964 return zoneref->zone->node;
965 #else
966 return 0;
967 #endif /* CONFIG_NUMA */
968 }
969
970 /**
971 * 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
972 * @z - The cursor used as a starting point for the search
973 * @highest_zoneidx - The zone index of the highest zone to return
974 * @nodes - An optional nodemask to filter the zonelist with
975 * @zone - The first suitable zone found is returned via this parameter
976 *
977 * This function returns the next zone at or below a given zone index that is
978 * within the allowed nodemask using a cursor as the starting point for the
979 * search. The zoneref returned is a cursor that represents the current zone
980 * being examined. It should be advanced by one before calling
981 * next_zones_zonelist again.
982 */
983 struct zoneref *next_zones_zonelist(struct zoneref *z,
984 enum zone_type highest_zoneidx,
985 nodemask_t *nodes,
986 struct zone **zone);
987
988 /**
989 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
990 * @zonelist - The zonelist to search for a suitable zone
991 * @highest_zoneidx - The zone index of the highest zone to return
992 * @nodes - An optional nodemask to filter the zonelist with
993 * @zone - The first suitable zone found is returned via this parameter
994 *
995 * This function returns the first zone at or below a given zone index that is
996 * within the allowed nodemask. The zoneref returned is a cursor that can be
997 * used to iterate the zonelist with next_zones_zonelist by advancing it by
998 * one before calling.
999 */
first_zones_zonelist(struct zonelist * zonelist,enum zone_type highest_zoneidx,nodemask_t * nodes,struct zone ** zone)1000 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1001 enum zone_type highest_zoneidx,
1002 nodemask_t *nodes,
1003 struct zone **zone)
1004 {
1005 return next_zones_zonelist(zonelist->_zonerefs, highest_zoneidx, nodes,
1006 zone);
1007 }
1008
1009 /**
1010 * 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
1011 * @zone - The current zone in the iterator
1012 * @z - The current pointer within zonelist->zones being iterated
1013 * @zlist - The zonelist being iterated
1014 * @highidx - The zone index of the highest zone to return
1015 * @nodemask - Nodemask allowed by the allocator
1016 *
1017 * This iterator iterates though all zones at or below a given zone index and
1018 * within a given nodemask
1019 */
1020 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1021 for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \
1022 zone; \
1023 z = next_zones_zonelist(++z, highidx, nodemask, &zone)) \
1024
1025 /**
1026 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1027 * @zone - The current zone in the iterator
1028 * @z - The current pointer within zonelist->zones being iterated
1029 * @zlist - The zonelist being iterated
1030 * @highidx - The zone index of the highest zone to return
1031 *
1032 * This iterator iterates though all zones at or below a given zone index.
1033 */
1034 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1035 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1036
1037 #ifdef CONFIG_SPARSEMEM
1038 #include <asm/sparsemem.h>
1039 #endif
1040
1041 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1042 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
early_pfn_to_nid(unsigned long pfn)1043 static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1044 {
1045 return 0;
1046 }
1047 #endif
1048
1049 #ifdef CONFIG_FLATMEM
1050 #define pfn_to_nid(pfn) (0)
1051 #endif
1052
1053 #ifdef CONFIG_SPARSEMEM
1054
1055 /*
1056 * SECTION_SHIFT #bits space required to store a section #
1057 *
1058 * PA_SECTION_SHIFT physical address to/from section number
1059 * PFN_SECTION_SHIFT pfn to/from section number
1060 */
1061 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1062 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1063
1064 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1065
1066 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1067 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1068
1069 #define SECTION_BLOCKFLAGS_BITS \
1070 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1071
1072 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1073 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1074 #endif
1075
1076 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
1077 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
1078
1079 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1080 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1081
1082 struct page;
1083 struct page_cgroup;
1084 struct mem_section {
1085 /*
1086 * This is, logically, a pointer to an array of struct
1087 * pages. However, it is stored with some other magic.
1088 * (see sparse.c::sparse_init_one_section())
1089 *
1090 * Additionally during early boot we encode node id of
1091 * the location of the section here to guide allocation.
1092 * (see sparse.c::memory_present())
1093 *
1094 * Making it a UL at least makes someone do a cast
1095 * before using it wrong.
1096 */
1097 unsigned long section_mem_map;
1098
1099 /* See declaration of similar field in struct zone */
1100 unsigned long *pageblock_flags;
1101 #ifdef CONFIG_MEMCG
1102 /*
1103 * If !SPARSEMEM, pgdat doesn't have page_cgroup pointer. We use
1104 * section. (see memcontrol.h/page_cgroup.h about this.)
1105 */
1106 struct page_cgroup *page_cgroup;
1107 unsigned long pad;
1108 #endif
1109 /*
1110 * WARNING: mem_section must be a power-of-2 in size for the
1111 * calculation and use of SECTION_ROOT_MASK to make sense.
1112 */
1113 };
1114
1115 #ifdef CONFIG_SPARSEMEM_EXTREME
1116 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1117 #else
1118 #define SECTIONS_PER_ROOT 1
1119 #endif
1120
1121 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1122 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1123 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1124
1125 #ifdef CONFIG_SPARSEMEM_EXTREME
1126 extern struct mem_section *mem_section[NR_SECTION_ROOTS];
1127 #else
1128 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1129 #endif
1130
__nr_to_section(unsigned long nr)1131 static inline struct mem_section *__nr_to_section(unsigned long nr)
1132 {
1133 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1134 return NULL;
1135 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1136 }
1137 extern int __section_nr(struct mem_section* ms);
1138 extern unsigned long usemap_size(void);
1139
1140 /*
1141 * We use the lower bits of the mem_map pointer to store
1142 * a little bit of information. There should be at least
1143 * 3 bits here due to 32-bit alignment.
1144 */
1145 #define SECTION_MARKED_PRESENT (1UL<<0)
1146 #define SECTION_HAS_MEM_MAP (1UL<<1)
1147 #define SECTION_MAP_LAST_BIT (1UL<<2)
1148 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1149 #define SECTION_NID_SHIFT 2
1150
__section_mem_map_addr(struct mem_section * section)1151 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1152 {
1153 unsigned long map = section->section_mem_map;
1154 map &= SECTION_MAP_MASK;
1155 return (struct page *)map;
1156 }
1157
present_section(struct mem_section * section)1158 static inline int present_section(struct mem_section *section)
1159 {
1160 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1161 }
1162
present_section_nr(unsigned long nr)1163 static inline int present_section_nr(unsigned long nr)
1164 {
1165 return present_section(__nr_to_section(nr));
1166 }
1167
valid_section(struct mem_section * section)1168 static inline int valid_section(struct mem_section *section)
1169 {
1170 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1171 }
1172
valid_section_nr(unsigned long nr)1173 static inline int valid_section_nr(unsigned long nr)
1174 {
1175 return valid_section(__nr_to_section(nr));
1176 }
1177
__pfn_to_section(unsigned long pfn)1178 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1179 {
1180 return __nr_to_section(pfn_to_section_nr(pfn));
1181 }
1182
1183 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
pfn_valid(unsigned long pfn)1184 static inline int pfn_valid(unsigned long pfn)
1185 {
1186 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1187 return 0;
1188 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
1189 }
1190 #endif
1191
pfn_present(unsigned long pfn)1192 static inline int pfn_present(unsigned long pfn)
1193 {
1194 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1195 return 0;
1196 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1197 }
1198
1199 /*
1200 * These are _only_ used during initialisation, therefore they
1201 * can use __initdata ... They could have names to indicate
1202 * this restriction.
1203 */
1204 #ifdef CONFIG_NUMA
1205 #define pfn_to_nid(pfn) \
1206 ({ \
1207 unsigned long __pfn_to_nid_pfn = (pfn); \
1208 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1209 })
1210 #else
1211 #define pfn_to_nid(pfn) (0)
1212 #endif
1213
1214 #define early_pfn_valid(pfn) pfn_valid(pfn)
1215 void sparse_init(void);
1216 #else
1217 #define sparse_init() do {} while (0)
1218 #define sparse_index_init(_sec, _nid) do {} while (0)
1219 #endif /* CONFIG_SPARSEMEM */
1220
1221 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1222 bool early_pfn_in_nid(unsigned long pfn, int nid);
1223 #else
1224 #define early_pfn_in_nid(pfn, nid) (1)
1225 #endif
1226
1227 #ifndef early_pfn_valid
1228 #define early_pfn_valid(pfn) (1)
1229 #endif
1230
1231 void memory_present(int nid, unsigned long start, unsigned long end);
1232 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
1233
1234 /*
1235 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1236 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1237 * pfn_valid_within() should be used in this case; we optimise this away
1238 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1239 */
1240 #ifdef CONFIG_HOLES_IN_ZONE
1241 #define pfn_valid_within(pfn) pfn_valid(pfn)
1242 #else
1243 #define pfn_valid_within(pfn) (1)
1244 #endif
1245
1246 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1247 /*
1248 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1249 * associated with it or not. In FLATMEM, it is expected that holes always
1250 * have valid memmap as long as there is valid PFNs either side of the hole.
1251 * In SPARSEMEM, it is assumed that a valid section has a memmap for the
1252 * entire section.
1253 *
1254 * However, an ARM, and maybe other embedded architectures in the future
1255 * free memmap backing holes to save memory on the assumption the memmap is
1256 * never used. The page_zone linkages are then broken even though pfn_valid()
1257 * returns true. A walker of the full memmap must then do this additional
1258 * check to ensure the memmap they are looking at is sane by making sure
1259 * the zone and PFN linkages are still valid. This is expensive, but walkers
1260 * of the full memmap are extremely rare.
1261 */
1262 int memmap_valid_within(unsigned long pfn,
1263 struct page *page, struct zone *zone);
1264 #else
memmap_valid_within(unsigned long pfn,struct page * page,struct zone * zone)1265 static inline int memmap_valid_within(unsigned long pfn,
1266 struct page *page, struct zone *zone)
1267 {
1268 return 1;
1269 }
1270 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1271
1272 #endif /* !__GENERATING_BOUNDS.H */
1273 #endif /* !__ASSEMBLY__ */
1274 #endif /* _LINUX_MMZONE_H */
1275