Branch data Line data Source code
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 : : #else
69 : : # define is_migrate_cma(migratetype) false
70 : : #endif
71 : :
72 : : #define for_each_migratetype_order(order, type) \
73 : : for (order = 0; order < MAX_ORDER; order++) \
74 : : for (type = 0; type < MIGRATE_TYPES; type++)
75 : :
76 : : extern int page_group_by_mobility_disabled;
77 : :
78 : : static inline int get_pageblock_migratetype(struct page *page)
79 : : {
80 : 31654011 : return get_pageblock_flags_group(page, PB_migrate, PB_migrate_end);
81 : : }
82 : :
83 : : struct free_area {
84 : : struct list_head free_list[MIGRATE_TYPES];
85 : : unsigned long nr_free;
86 : : };
87 : :
88 : : struct pglist_data;
89 : :
90 : : /*
91 : : * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
92 : : * So add a wild amount of padding here to ensure that they fall into separate
93 : : * cachelines. There are very few zone structures in the machine, so space
94 : : * consumption is not a concern here.
95 : : */
96 : : #if defined(CONFIG_SMP)
97 : : struct zone_padding {
98 : : char x[0];
99 : : } ____cacheline_internodealigned_in_smp;
100 : : #define ZONE_PADDING(name) struct zone_padding name;
101 : : #else
102 : : #define ZONE_PADDING(name)
103 : : #endif
104 : :
105 : : enum zone_stat_item {
106 : : /* First 128 byte cacheline (assuming 64 bit words) */
107 : : NR_FREE_PAGES,
108 : : NR_ALLOC_BATCH,
109 : : NR_LRU_BASE,
110 : : NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
111 : : NR_ACTIVE_ANON, /* " " " " " */
112 : : NR_INACTIVE_FILE, /* " " " " " */
113 : : NR_ACTIVE_FILE, /* " " " " " */
114 : : NR_UNEVICTABLE, /* " " " " " */
115 : : NR_MLOCK, /* mlock()ed pages found and moved off LRU */
116 : : NR_ANON_PAGES, /* Mapped anonymous pages */
117 : : NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
118 : : only modified from process context */
119 : : NR_FILE_PAGES,
120 : : NR_FILE_DIRTY,
121 : : NR_WRITEBACK,
122 : : NR_SLAB_RECLAIMABLE,
123 : : NR_SLAB_UNRECLAIMABLE,
124 : : NR_PAGETABLE, /* used for pagetables */
125 : : NR_KERNEL_STACK,
126 : : /* Second 128 byte cacheline */
127 : : NR_UNSTABLE_NFS, /* NFS unstable pages */
128 : : NR_BOUNCE,
129 : : NR_VMSCAN_WRITE,
130 : : NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
131 : : NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
132 : : NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
133 : : NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
134 : : NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
135 : : NR_DIRTIED, /* page dirtyings since bootup */
136 : : NR_WRITTEN, /* page writings since bootup */
137 : : #ifdef CONFIG_NUMA
138 : : NUMA_HIT, /* allocated in intended node */
139 : : NUMA_MISS, /* allocated in non intended node */
140 : : NUMA_FOREIGN, /* was intended here, hit elsewhere */
141 : : NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
142 : : NUMA_LOCAL, /* allocation from local node */
143 : : NUMA_OTHER, /* allocation from other node */
144 : : #endif
145 : : NR_ANON_TRANSPARENT_HUGEPAGES,
146 : : NR_FREE_CMA_PAGES,
147 : : NR_VM_ZONE_STAT_ITEMS };
148 : :
149 : : /*
150 : : * We do arithmetic on the LRU lists in various places in the code,
151 : : * so it is important to keep the active lists LRU_ACTIVE higher in
152 : : * the array than the corresponding inactive lists, and to keep
153 : : * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
154 : : *
155 : : * This has to be kept in sync with the statistics in zone_stat_item
156 : : * above and the descriptions in vmstat_text in mm/vmstat.c
157 : : */
158 : : #define LRU_BASE 0
159 : : #define LRU_ACTIVE 1
160 : : #define LRU_FILE 2
161 : :
162 : : enum lru_list {
163 : : LRU_INACTIVE_ANON = LRU_BASE,
164 : : LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
165 : : LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
166 : : LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
167 : : LRU_UNEVICTABLE,
168 : : NR_LRU_LISTS
169 : : };
170 : :
171 : : #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
172 : :
173 : : #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
174 : :
175 : : static inline int is_file_lru(enum lru_list lru)
176 : : {
177 : 799599 : return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
178 : : }
179 : :
180 : : static inline int is_active_lru(enum lru_list lru)
181 : : {
182 : 123989 : return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
183 : : }
184 : :
185 : : static inline int is_unevictable_lru(enum lru_list lru)
186 : : {
187 : : return (lru == LRU_UNEVICTABLE);
188 : : }
189 : :
190 : : struct zone_reclaim_stat {
191 : : /*
192 : : * The pageout code in vmscan.c keeps track of how many of the
193 : : * mem/swap backed and file backed pages are referenced.
194 : : * The higher the rotated/scanned ratio, the more valuable
195 : : * that cache is.
196 : : *
197 : : * The anon LRU stats live in [0], file LRU stats in [1]
198 : : */
199 : : unsigned long recent_rotated[2];
200 : : unsigned long recent_scanned[2];
201 : : };
202 : :
203 : : struct lruvec {
204 : : struct list_head lists[NR_LRU_LISTS];
205 : : struct zone_reclaim_stat reclaim_stat;
206 : : #ifdef CONFIG_MEMCG
207 : : struct zone *zone;
208 : : #endif
209 : : };
210 : :
211 : : /* Mask used at gathering information at once (see memcontrol.c) */
212 : : #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
213 : : #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
214 : : #define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
215 : :
216 : : /* Isolate clean file */
217 : : #define ISOLATE_CLEAN ((__force isolate_mode_t)0x1)
218 : : /* Isolate unmapped file */
219 : : #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
220 : : /* Isolate for asynchronous migration */
221 : : #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
222 : : /* Isolate unevictable pages */
223 : : #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
224 : :
225 : : /* LRU Isolation modes. */
226 : : typedef unsigned __bitwise__ isolate_mode_t;
227 : :
228 : : enum zone_watermarks {
229 : : WMARK_MIN,
230 : : WMARK_LOW,
231 : : WMARK_HIGH,
232 : : NR_WMARK
233 : : };
234 : :
235 : : #define min_wmark_pages(z) (z->watermark[WMARK_MIN])
236 : : #define low_wmark_pages(z) (z->watermark[WMARK_LOW])
237 : : #define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
238 : :
239 : : struct per_cpu_pages {
240 : : int count; /* number of pages in the list */
241 : : int high; /* high watermark, emptying needed */
242 : : int batch; /* chunk size for buddy add/remove */
243 : :
244 : : /* Lists of pages, one per migrate type stored on the pcp-lists */
245 : : struct list_head lists[MIGRATE_PCPTYPES];
246 : : };
247 : :
248 : : struct per_cpu_pageset {
249 : : struct per_cpu_pages pcp;
250 : : #ifdef CONFIG_NUMA
251 : : s8 expire;
252 : : #endif
253 : : #ifdef CONFIG_SMP
254 : : s8 stat_threshold;
255 : : s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
256 : : #endif
257 : : };
258 : :
259 : : #endif /* !__GENERATING_BOUNDS.H */
260 : :
261 : : enum zone_type {
262 : : #ifdef CONFIG_ZONE_DMA
263 : : /*
264 : : * ZONE_DMA is used when there are devices that are not able
265 : : * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
266 : : * carve out the portion of memory that is needed for these devices.
267 : : * The range is arch specific.
268 : : *
269 : : * Some examples
270 : : *
271 : : * Architecture Limit
272 : : * ---------------------------
273 : : * parisc, ia64, sparc <4G
274 : : * s390 <2G
275 : : * arm Various
276 : : * alpha Unlimited or 0-16MB.
277 : : *
278 : : * i386, x86_64 and multiple other arches
279 : : * <16M.
280 : : */
281 : : ZONE_DMA,
282 : : #endif
283 : : #ifdef CONFIG_ZONE_DMA32
284 : : /*
285 : : * x86_64 needs two ZONE_DMAs because it supports devices that are
286 : : * only able to do DMA to the lower 16M but also 32 bit devices that
287 : : * can only do DMA areas below 4G.
288 : : */
289 : : ZONE_DMA32,
290 : : #endif
291 : : /*
292 : : * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
293 : : * performed on pages in ZONE_NORMAL if the DMA devices support
294 : : * transfers to all addressable memory.
295 : : */
296 : : ZONE_NORMAL,
297 : : #ifdef CONFIG_HIGHMEM
298 : : /*
299 : : * A memory area that is only addressable by the kernel through
300 : : * mapping portions into its own address space. This is for example
301 : : * used by i386 to allow the kernel to address the memory beyond
302 : : * 900MB. The kernel will set up special mappings (page
303 : : * table entries on i386) for each page that the kernel needs to
304 : : * access.
305 : : */
306 : : ZONE_HIGHMEM,
307 : : #endif
308 : : ZONE_MOVABLE,
309 : : __MAX_NR_ZONES
310 : : };
311 : :
312 : : #ifndef __GENERATING_BOUNDS_H
313 : :
314 : : struct zone {
315 : : /* Fields commonly accessed by the page allocator */
316 : :
317 : : /* zone watermarks, access with *_wmark_pages(zone) macros */
318 : : unsigned long watermark[NR_WMARK];
319 : :
320 : : /*
321 : : * When free pages are below this point, additional steps are taken
322 : : * when reading the number of free pages to avoid per-cpu counter
323 : : * drift allowing watermarks to be breached
324 : : */
325 : : unsigned long percpu_drift_mark;
326 : :
327 : : /*
328 : : * We don't know if the memory that we're going to allocate will be freeable
329 : : * or/and it will be released eventually, so to avoid totally wasting several
330 : : * GB of ram we must reserve some of the lower zone memory (otherwise we risk
331 : : * to run OOM on the lower zones despite there's tons of freeable ram
332 : : * on the higher zones). This array is recalculated at runtime if the
333 : : * sysctl_lowmem_reserve_ratio sysctl changes.
334 : : */
335 : : unsigned long lowmem_reserve[MAX_NR_ZONES];
336 : :
337 : : /*
338 : : * This is a per-zone reserve of pages that should not be
339 : : * considered dirtyable memory.
340 : : */
341 : : unsigned long dirty_balance_reserve;
342 : :
343 : : #ifdef CONFIG_NUMA
344 : : int node;
345 : : /*
346 : : * zone reclaim becomes active if more unmapped pages exist.
347 : : */
348 : : unsigned long min_unmapped_pages;
349 : : unsigned long min_slab_pages;
350 : : #endif
351 : : struct per_cpu_pageset __percpu *pageset;
352 : : /*
353 : : * free areas of different sizes
354 : : */
355 : : spinlock_t lock;
356 : : #if defined CONFIG_COMPACTION || defined CONFIG_CMA
357 : : /* Set to true when the PG_migrate_skip bits should be cleared */
358 : : bool compact_blockskip_flush;
359 : :
360 : : /* pfns where compaction scanners should start */
361 : : unsigned long compact_cached_free_pfn;
362 : : unsigned long compact_cached_migrate_pfn;
363 : : #endif
364 : : #ifdef CONFIG_MEMORY_HOTPLUG
365 : : /* see spanned/present_pages for more description */
366 : : seqlock_t span_seqlock;
367 : : #endif
368 : : struct free_area free_area[MAX_ORDER];
369 : :
370 : : #ifndef CONFIG_SPARSEMEM
371 : : /*
372 : : * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
373 : : * In SPARSEMEM, this map is stored in struct mem_section
374 : : */
375 : : unsigned long *pageblock_flags;
376 : : #endif /* CONFIG_SPARSEMEM */
377 : :
378 : : #ifdef CONFIG_COMPACTION
379 : : /*
380 : : * On compaction failure, 1<<compact_defer_shift compactions
381 : : * are skipped before trying again. The number attempted since
382 : : * last failure is tracked with compact_considered.
383 : : */
384 : : unsigned int compact_considered;
385 : : unsigned int compact_defer_shift;
386 : : int compact_order_failed;
387 : : #endif
388 : :
389 : : ZONE_PADDING(_pad1_)
390 : :
391 : : /* Fields commonly accessed by the page reclaim scanner */
392 : : spinlock_t lru_lock;
393 : : struct lruvec lruvec;
394 : :
395 : : unsigned long pages_scanned; /* since last reclaim */
396 : : unsigned long flags; /* zone flags, see below */
397 : :
398 : : /* Zone statistics */
399 : : atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
400 : :
401 : : /*
402 : : * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on
403 : : * this zone's LRU. Maintained by the pageout code.
404 : : */
405 : : unsigned int inactive_ratio;
406 : :
407 : :
408 : : ZONE_PADDING(_pad2_)
409 : : /* Rarely used or read-mostly fields */
410 : :
411 : : /*
412 : : * wait_table -- the array holding the hash table
413 : : * wait_table_hash_nr_entries -- the size of the hash table array
414 : : * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
415 : : *
416 : : * The purpose of all these is to keep track of the people
417 : : * waiting for a page to become available and make them
418 : : * runnable again when possible. The trouble is that this
419 : : * consumes a lot of space, especially when so few things
420 : : * wait on pages at a given time. So instead of using
421 : : * per-page waitqueues, we use a waitqueue hash table.
422 : : *
423 : : * The bucket discipline is to sleep on the same queue when
424 : : * colliding and wake all in that wait queue when removing.
425 : : * When something wakes, it must check to be sure its page is
426 : : * truly available, a la thundering herd. The cost of a
427 : : * collision is great, but given the expected load of the
428 : : * table, they should be so rare as to be outweighed by the
429 : : * benefits from the saved space.
430 : : *
431 : : * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
432 : : * primary users of these fields, and in mm/page_alloc.c
433 : : * free_area_init_core() performs the initialization of them.
434 : : */
435 : : wait_queue_head_t * wait_table;
436 : : unsigned long wait_table_hash_nr_entries;
437 : : unsigned long wait_table_bits;
438 : :
439 : : /*
440 : : * Discontig memory support fields.
441 : : */
442 : : struct pglist_data *zone_pgdat;
443 : : /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
444 : : unsigned long zone_start_pfn;
445 : :
446 : : /*
447 : : * spanned_pages is the total pages spanned by the zone, including
448 : : * holes, which is calculated as:
449 : : * spanned_pages = zone_end_pfn - zone_start_pfn;
450 : : *
451 : : * present_pages is physical pages existing within the zone, which
452 : : * is calculated as:
453 : : * present_pages = spanned_pages - absent_pages(pages in holes);
454 : : *
455 : : * managed_pages is present pages managed by the buddy system, which
456 : : * is calculated as (reserved_pages includes pages allocated by the
457 : : * bootmem allocator):
458 : : * managed_pages = present_pages - reserved_pages;
459 : : *
460 : : * So present_pages may be used by memory hotplug or memory power
461 : : * management logic to figure out unmanaged pages by checking
462 : : * (present_pages - managed_pages). And managed_pages should be used
463 : : * by page allocator and vm scanner to calculate all kinds of watermarks
464 : : * and thresholds.
465 : : *
466 : : * Locking rules:
467 : : *
468 : : * zone_start_pfn and spanned_pages are protected by span_seqlock.
469 : : * It is a seqlock because it has to be read outside of zone->lock,
470 : : * and it is done in the main allocator path. But, it is written
471 : : * quite infrequently.
472 : : *
473 : : * The span_seq lock is declared along with zone->lock because it is
474 : : * frequently read in proximity to zone->lock. It's good to
475 : : * give them a chance of being in the same cacheline.
476 : : *
477 : : * Write access to present_pages at runtime should be protected by
478 : : * lock_memory_hotplug()/unlock_memory_hotplug(). Any reader who can't
479 : : * tolerant drift of present_pages should hold memory hotplug lock to
480 : : * get a stable value.
481 : : *
482 : : * Read access to managed_pages should be safe because it's unsigned
483 : : * long. Write access to zone->managed_pages and totalram_pages are
484 : : * protected by managed_page_count_lock at runtime. Idealy only
485 : : * adjust_managed_page_count() should be used instead of directly
486 : : * touching zone->managed_pages and totalram_pages.
487 : : */
488 : : unsigned long spanned_pages;
489 : : unsigned long present_pages;
490 : : unsigned long managed_pages;
491 : :
492 : : /*
493 : : * Number of MIGRATE_RESEVE page block. To maintain for just
494 : : * optimization. Protected by zone->lock.
495 : : */
496 : : int nr_migrate_reserve_block;
497 : :
498 : : /*
499 : : * rarely used fields:
500 : : */
501 : : const char *name;
502 : : } ____cacheline_internodealigned_in_smp;
503 : :
504 : : typedef enum {
505 : : ZONE_RECLAIM_LOCKED, /* prevents concurrent reclaim */
506 : : ZONE_OOM_LOCKED, /* zone is in OOM killer zonelist */
507 : : ZONE_CONGESTED, /* zone has many dirty pages backed by
508 : : * a congested BDI
509 : : */
510 : : ZONE_TAIL_LRU_DIRTY, /* reclaim scanning has recently found
511 : : * many dirty file pages at the tail
512 : : * of the LRU.
513 : : */
514 : : ZONE_WRITEBACK, /* reclaim scanning has recently found
515 : : * many pages under writeback
516 : : */
517 : : } zone_flags_t;
518 : :
519 : : static inline void zone_set_flag(struct zone *zone, zone_flags_t flag)
520 : : {
521 : 2032 : set_bit(flag, &zone->flags);
522 : : }
523 : :
524 : : static inline int zone_test_and_set_flag(struct zone *zone, zone_flags_t flag)
525 : : {
526 : : return test_and_set_bit(flag, &zone->flags);
527 : : }
528 : :
529 : : static inline void zone_clear_flag(struct zone *zone, zone_flags_t flag)
530 : : {
531 : 51173 : clear_bit(flag, &zone->flags);
532 : : }
533 : :
534 : : static inline int zone_is_reclaim_congested(const struct zone *zone)
535 : : {
536 : : return test_bit(ZONE_CONGESTED, &zone->flags);
537 : : }
538 : :
539 : : static inline int zone_is_reclaim_dirty(const struct zone *zone)
540 : : {
541 : : return test_bit(ZONE_TAIL_LRU_DIRTY, &zone->flags);
542 : : }
543 : :
544 : : static inline int zone_is_reclaim_writeback(const struct zone *zone)
545 : : {
546 : : return test_bit(ZONE_WRITEBACK, &zone->flags);
547 : : }
548 : :
549 : : static inline int zone_is_reclaim_locked(const struct zone *zone)
550 : : {
551 : : return test_bit(ZONE_RECLAIM_LOCKED, &zone->flags);
552 : : }
553 : :
554 : : static inline int zone_is_oom_locked(const struct zone *zone)
555 : : {
556 : : return test_bit(ZONE_OOM_LOCKED, &zone->flags);
557 : : }
558 : :
559 : : static inline unsigned long zone_end_pfn(const struct zone *zone)
560 : : {
561 : 68 : return zone->zone_start_pfn + zone->spanned_pages;
562 : : }
563 : :
564 : 38 : static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
565 : : {
566 [ # # ][ # # ]: 38 : return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
[ + - ][ - + ]
[ + - ][ - + ]
567 : : }
568 : :
569 : : static inline bool zone_is_initialized(struct zone *zone)
570 : : {
571 : : return !!zone->wait_table;
572 : : }
573 : :
574 : : static inline bool zone_is_empty(struct zone *zone)
575 : : {
576 : : return zone->spanned_pages == 0;
577 : : }
578 : :
579 : : /*
580 : : * The "priority" of VM scanning is how much of the queues we will scan in one
581 : : * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
582 : : * queues ("queue_length >> 12") during an aging round.
583 : : */
584 : : #define DEF_PRIORITY 12
585 : :
586 : : /* Maximum number of zones on a zonelist */
587 : : #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
588 : :
589 : : #ifdef CONFIG_NUMA
590 : :
591 : : /*
592 : : * The NUMA zonelists are doubled because we need zonelists that restrict the
593 : : * allocations to a single node for __GFP_THISNODE.
594 : : *
595 : : * [0] : Zonelist with fallback
596 : : * [1] : No fallback (__GFP_THISNODE)
597 : : */
598 : : #define MAX_ZONELISTS 2
599 : :
600 : :
601 : : /*
602 : : * We cache key information from each zonelist for smaller cache
603 : : * footprint when scanning for free pages in get_page_from_freelist().
604 : : *
605 : : * 1) The BITMAP fullzones tracks which zones in a zonelist have come
606 : : * up short of free memory since the last time (last_fullzone_zap)
607 : : * we zero'd fullzones.
608 : : * 2) The array z_to_n[] maps each zone in the zonelist to its node
609 : : * id, so that we can efficiently evaluate whether that node is
610 : : * set in the current tasks mems_allowed.
611 : : *
612 : : * Both fullzones and z_to_n[] are one-to-one with the zonelist,
613 : : * indexed by a zones offset in the zonelist zones[] array.
614 : : *
615 : : * The get_page_from_freelist() routine does two scans. During the
616 : : * first scan, we skip zones whose corresponding bit in 'fullzones'
617 : : * is set or whose corresponding node in current->mems_allowed (which
618 : : * comes from cpusets) is not set. During the second scan, we bypass
619 : : * this zonelist_cache, to ensure we look methodically at each zone.
620 : : *
621 : : * Once per second, we zero out (zap) fullzones, forcing us to
622 : : * reconsider nodes that might have regained more free memory.
623 : : * The field last_full_zap is the time we last zapped fullzones.
624 : : *
625 : : * This mechanism reduces the amount of time we waste repeatedly
626 : : * reexaming zones for free memory when they just came up low on
627 : : * memory momentarilly ago.
628 : : *
629 : : * The zonelist_cache struct members logically belong in struct
630 : : * zonelist. However, the mempolicy zonelists constructed for
631 : : * MPOL_BIND are intentionally variable length (and usually much
632 : : * shorter). A general purpose mechanism for handling structs with
633 : : * multiple variable length members is more mechanism than we want
634 : : * here. We resort to some special case hackery instead.
635 : : *
636 : : * The MPOL_BIND zonelists don't need this zonelist_cache (in good
637 : : * part because they are shorter), so we put the fixed length stuff
638 : : * at the front of the zonelist struct, ending in a variable length
639 : : * zones[], as is needed by MPOL_BIND.
640 : : *
641 : : * Then we put the optional zonelist cache on the end of the zonelist
642 : : * struct. This optional stuff is found by a 'zlcache_ptr' pointer in
643 : : * the fixed length portion at the front of the struct. This pointer
644 : : * both enables us to find the zonelist cache, and in the case of
645 : : * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL)
646 : : * to know that the zonelist cache is not there.
647 : : *
648 : : * The end result is that struct zonelists come in two flavors:
649 : : * 1) The full, fixed length version, shown below, and
650 : : * 2) The custom zonelists for MPOL_BIND.
651 : : * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache.
652 : : *
653 : : * Even though there may be multiple CPU cores on a node modifying
654 : : * fullzones or last_full_zap in the same zonelist_cache at the same
655 : : * time, we don't lock it. This is just hint data - if it is wrong now
656 : : * and then, the allocator will still function, perhaps a bit slower.
657 : : */
658 : :
659 : :
660 : : struct zonelist_cache {
661 : : unsigned short z_to_n[MAX_ZONES_PER_ZONELIST]; /* zone->nid */
662 : : DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST); /* zone full? */
663 : : unsigned long last_full_zap; /* when last zap'd (jiffies) */
664 : : };
665 : : #else
666 : : #define MAX_ZONELISTS 1
667 : : struct zonelist_cache;
668 : : #endif
669 : :
670 : : /*
671 : : * This struct contains information about a zone in a zonelist. It is stored
672 : : * here to avoid dereferences into large structures and lookups of tables
673 : : */
674 : : struct zoneref {
675 : : struct zone *zone; /* Pointer to actual zone */
676 : : int zone_idx; /* zone_idx(zoneref->zone) */
677 : : };
678 : :
679 : : /*
680 : : * One allocation request operates on a zonelist. A zonelist
681 : : * is a list of zones, the first one is the 'goal' of the
682 : : * allocation, the other zones are fallback zones, in decreasing
683 : : * priority.
684 : : *
685 : : * If zlcache_ptr is not NULL, then it is just the address of zlcache,
686 : : * as explained above. If zlcache_ptr is NULL, there is no zlcache.
687 : : * *
688 : : * To speed the reading of the zonelist, the zonerefs contain the zone index
689 : : * of the entry being read. Helper functions to access information given
690 : : * a struct zoneref are
691 : : *
692 : : * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
693 : : * zonelist_zone_idx() - Return the index of the zone for an entry
694 : : * zonelist_node_idx() - Return the index of the node for an entry
695 : : */
696 : : struct zonelist {
697 : : struct zonelist_cache *zlcache_ptr; // NULL or &zlcache
698 : : struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
699 : : #ifdef CONFIG_NUMA
700 : : struct zonelist_cache zlcache; // optional ...
701 : : #endif
702 : : };
703 : :
704 : : #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
705 : : struct node_active_region {
706 : : unsigned long start_pfn;
707 : : unsigned long end_pfn;
708 : : int nid;
709 : : };
710 : : #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
711 : :
712 : : #ifndef CONFIG_DISCONTIGMEM
713 : : /* The array of struct pages - for discontigmem use pgdat->lmem_map */
714 : : extern struct page *mem_map;
715 : : #endif
716 : :
717 : : /*
718 : : * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
719 : : * (mostly NUMA machines?) to denote a higher-level memory zone than the
720 : : * zone denotes.
721 : : *
722 : : * On NUMA machines, each NUMA node would have a pg_data_t to describe
723 : : * it's memory layout.
724 : : *
725 : : * Memory statistics and page replacement data structures are maintained on a
726 : : * per-zone basis.
727 : : */
728 : : struct bootmem_data;
729 : : typedef struct pglist_data {
730 : : struct zone node_zones[MAX_NR_ZONES];
731 : : struct zonelist node_zonelists[MAX_ZONELISTS];
732 : : int nr_zones;
733 : : #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
734 : : struct page *node_mem_map;
735 : : #ifdef CONFIG_MEMCG
736 : : struct page_cgroup *node_page_cgroup;
737 : : #endif
738 : : #endif
739 : : #ifndef CONFIG_NO_BOOTMEM
740 : : struct bootmem_data *bdata;
741 : : #endif
742 : : #ifdef CONFIG_MEMORY_HOTPLUG
743 : : /*
744 : : * Must be held any time you expect node_start_pfn, node_present_pages
745 : : * or node_spanned_pages stay constant. Holding this will also
746 : : * guarantee that any pfn_valid() stays that way.
747 : : *
748 : : * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
749 : : * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG.
750 : : *
751 : : * Nests above zone->lock and zone->span_seqlock
752 : : */
753 : : spinlock_t node_size_lock;
754 : : #endif
755 : : unsigned long node_start_pfn;
756 : : unsigned long node_present_pages; /* total number of physical pages */
757 : : unsigned long node_spanned_pages; /* total size of physical page
758 : : range, including holes */
759 : : int node_id;
760 : : nodemask_t reclaim_nodes; /* Nodes allowed to reclaim from */
761 : : wait_queue_head_t kswapd_wait;
762 : : wait_queue_head_t pfmemalloc_wait;
763 : : struct task_struct *kswapd; /* Protected by lock_memory_hotplug() */
764 : : int kswapd_max_order;
765 : : enum zone_type classzone_idx;
766 : : #ifdef CONFIG_NUMA_BALANCING
767 : : /* Lock serializing the migrate rate limiting window */
768 : : spinlock_t numabalancing_migrate_lock;
769 : :
770 : : /* Rate limiting time interval */
771 : : unsigned long numabalancing_migrate_next_window;
772 : :
773 : : /* Number of pages migrated during the rate limiting time interval */
774 : : unsigned long numabalancing_migrate_nr_pages;
775 : : #endif
776 : : } pg_data_t;
777 : :
778 : : #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
779 : : #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
780 : : #ifdef CONFIG_FLAT_NODE_MEM_MAP
781 : : #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
782 : : #else
783 : : #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
784 : : #endif
785 : : #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
786 : :
787 : : #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
788 : : #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
789 : :
790 : : static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
791 : : {
792 : 0 : return pgdat->node_start_pfn + pgdat->node_spanned_pages;
793 : : }
794 : :
795 : : static inline bool pgdat_is_empty(pg_data_t *pgdat)
796 : : {
797 : : return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
798 : : }
799 : :
800 : : #include <linux/memory_hotplug.h>
801 : :
802 : : extern struct mutex zonelists_mutex;
803 : : void build_all_zonelists(pg_data_t *pgdat, struct zone *zone);
804 : : void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx);
805 : : bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
806 : : int classzone_idx, int alloc_flags);
807 : : bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
808 : : int classzone_idx, int alloc_flags);
809 : : enum memmap_context {
810 : : MEMMAP_EARLY,
811 : : MEMMAP_HOTPLUG,
812 : : };
813 : : extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
814 : : unsigned long size,
815 : : enum memmap_context context);
816 : :
817 : : extern void lruvec_init(struct lruvec *lruvec);
818 : :
819 : : static inline struct zone *lruvec_zone(struct lruvec *lruvec)
820 : : {
821 : : #ifdef CONFIG_MEMCG
822 : : return lruvec->zone;
823 : : #else
824 : : return container_of(lruvec, struct zone, lruvec);
825 : : #endif
826 : : }
827 : :
828 : : #ifdef CONFIG_HAVE_MEMORY_PRESENT
829 : : void memory_present(int nid, unsigned long start, unsigned long end);
830 : : #else
831 : : static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
832 : : #endif
833 : :
834 : : #ifdef CONFIG_HAVE_MEMORYLESS_NODES
835 : : int local_memory_node(int node_id);
836 : : #else
837 : : static inline int local_memory_node(int node_id) { return node_id; };
838 : : #endif
839 : :
840 : : #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
841 : : unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
842 : : #endif
843 : :
844 : : /*
845 : : * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
846 : : */
847 : : #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
848 : :
849 : : static inline int populated_zone(struct zone *zone)
850 : : {
851 : : return (!!zone->present_pages);
852 : : }
853 : :
854 : : extern int movable_zone;
855 : :
856 : : static inline int zone_movable_is_highmem(void)
857 : : {
858 : : #if defined(CONFIG_HIGHMEM) && defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
859 : : return movable_zone == ZONE_HIGHMEM;
860 : : #else
861 : : return 0;
862 : : #endif
863 : : }
864 : :
865 : : static inline int is_highmem_idx(enum zone_type idx)
866 : : {
867 : : #ifdef CONFIG_HIGHMEM
868 [ # # ][ # # ]: 0 : return (idx == ZONE_HIGHMEM ||
869 : : (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
870 : : #else
871 : : return 0;
872 : : #endif
873 : : }
874 : :
875 : : /**
876 : : * is_highmem - helper function to quickly check if a struct zone is a
877 : : * highmem zone or not. This is an attempt to keep references
878 : : * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
879 : : * @zone - pointer to struct zone variable
880 : : */
881 : : static inline int is_highmem(struct zone *zone)
882 : : {
883 : : #ifdef CONFIG_HIGHMEM
884 : 535812280 : int zone_off = (char *)zone - (char *)zone->zone_pgdat->node_zones;
885 [ + + ][ + + ]: 535812280 : return zone_off == ZONE_HIGHMEM * sizeof(*zone) ||
[ + + ][ + + ]
[ + + ][ + + ]
886 : : (zone_off == ZONE_MOVABLE * sizeof(*zone) &&
887 : : zone_movable_is_highmem());
888 : : #else
889 : : return 0;
890 : : #endif
891 : : }
892 : :
893 : : /* These two functions are used to setup the per zone pages min values */
894 : : struct ctl_table;
895 : : int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
896 : : void __user *, size_t *, loff_t *);
897 : : extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
898 : : int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
899 : : void __user *, size_t *, loff_t *);
900 : : int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
901 : : void __user *, size_t *, loff_t *);
902 : : int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
903 : : void __user *, size_t *, loff_t *);
904 : : int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
905 : : void __user *, size_t *, loff_t *);
906 : :
907 : : extern int numa_zonelist_order_handler(struct ctl_table *, int,
908 : : void __user *, size_t *, loff_t *);
909 : : extern char numa_zonelist_order[];
910 : : #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */
911 : :
912 : : #ifndef CONFIG_NEED_MULTIPLE_NODES
913 : :
914 : : extern struct pglist_data contig_page_data;
915 : : #define NODE_DATA(nid) (&contig_page_data)
916 : : #define NODE_MEM_MAP(nid) mem_map
917 : :
918 : : #else /* CONFIG_NEED_MULTIPLE_NODES */
919 : :
920 : : #include <asm/mmzone.h>
921 : :
922 : : #endif /* !CONFIG_NEED_MULTIPLE_NODES */
923 : :
924 : : extern struct pglist_data *first_online_pgdat(void);
925 : : extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
926 : : extern struct zone *next_zone(struct zone *zone);
927 : :
928 : : /**
929 : : * for_each_online_pgdat - helper macro to iterate over all online nodes
930 : : * @pgdat - pointer to a pg_data_t variable
931 : : */
932 : : #define for_each_online_pgdat(pgdat) \
933 : : for (pgdat = first_online_pgdat(); \
934 : : pgdat; \
935 : : pgdat = next_online_pgdat(pgdat))
936 : : /**
937 : : * for_each_zone - helper macro to iterate over all memory zones
938 : : * @zone - pointer to struct zone variable
939 : : *
940 : : * The user only needs to declare the zone variable, for_each_zone
941 : : * fills it in.
942 : : */
943 : : #define for_each_zone(zone) \
944 : : for (zone = (first_online_pgdat())->node_zones; \
945 : : zone; \
946 : : zone = next_zone(zone))
947 : :
948 : : #define for_each_populated_zone(zone) \
949 : : for (zone = (first_online_pgdat())->node_zones; \
950 : : zone; \
951 : : zone = next_zone(zone)) \
952 : : if (!populated_zone(zone)) \
953 : : ; /* do nothing */ \
954 : : else
955 : :
956 : : static inline struct zone *zonelist_zone(struct zoneref *zoneref)
957 : : {
958 : : return zoneref->zone;
959 : : }
960 : :
961 : : static inline int zonelist_zone_idx(struct zoneref *zoneref)
962 : : {
963 : : return zoneref->zone_idx;
964 : : }
965 : :
966 : : static inline int zonelist_node_idx(struct zoneref *zoneref)
967 : : {
968 : : #ifdef CONFIG_NUMA
969 : : /* zone_to_nid not available in this context */
970 : : return zoneref->zone->node;
971 : : #else
972 : : return 0;
973 : : #endif /* CONFIG_NUMA */
974 : : }
975 : :
976 : : /**
977 : : * 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
978 : : * @z - The cursor used as a starting point for the search
979 : : * @highest_zoneidx - The zone index of the highest zone to return
980 : : * @nodes - An optional nodemask to filter the zonelist with
981 : : * @zone - The first suitable zone found is returned via this parameter
982 : : *
983 : : * This function returns the next zone at or below a given zone index that is
984 : : * within the allowed nodemask using a cursor as the starting point for the
985 : : * search. The zoneref returned is a cursor that represents the current zone
986 : : * being examined. It should be advanced by one before calling
987 : : * next_zones_zonelist again.
988 : : */
989 : : struct zoneref *next_zones_zonelist(struct zoneref *z,
990 : : enum zone_type highest_zoneidx,
991 : : nodemask_t *nodes,
992 : : struct zone **zone);
993 : :
994 : : /**
995 : : * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
996 : : * @zonelist - The zonelist to search for a suitable zone
997 : : * @highest_zoneidx - The zone index of the highest zone to return
998 : : * @nodes - An optional nodemask to filter the zonelist with
999 : : * @zone - The first suitable zone found is returned via this parameter
1000 : : *
1001 : : * This function returns the first zone at or below a given zone index that is
1002 : : * within the allowed nodemask. The zoneref returned is a cursor that can be
1003 : : * used to iterate the zonelist with next_zones_zonelist by advancing it by
1004 : : * one before calling.
1005 : : */
1006 : : static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1007 : : enum zone_type highest_zoneidx,
1008 : : nodemask_t *nodes,
1009 : : struct zone **zone)
1010 : : {
1011 : 59217352 : return next_zones_zonelist(zonelist->_zonerefs, highest_zoneidx, nodes,
1012 : : zone);
1013 : : }
1014 : :
1015 : : /**
1016 : : * 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
1017 : : * @zone - The current zone in the iterator
1018 : : * @z - The current pointer within zonelist->zones being iterated
1019 : : * @zlist - The zonelist being iterated
1020 : : * @highidx - The zone index of the highest zone to return
1021 : : * @nodemask - Nodemask allowed by the allocator
1022 : : *
1023 : : * This iterator iterates though all zones at or below a given zone index and
1024 : : * within a given nodemask
1025 : : */
1026 : : #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1027 : : for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \
1028 : : zone; \
1029 : : z = next_zones_zonelist(++z, highidx, nodemask, &zone)) \
1030 : :
1031 : : /**
1032 : : * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1033 : : * @zone - The current zone in the iterator
1034 : : * @z - The current pointer within zonelist->zones being iterated
1035 : : * @zlist - The zonelist being iterated
1036 : : * @highidx - The zone index of the highest zone to return
1037 : : *
1038 : : * This iterator iterates though all zones at or below a given zone index.
1039 : : */
1040 : : #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1041 : : for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1042 : :
1043 : : #ifdef CONFIG_SPARSEMEM
1044 : : #include <asm/sparsemem.h>
1045 : : #endif
1046 : :
1047 : : #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
1048 : : !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1049 : : static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1050 : : {
1051 : : return 0;
1052 : : }
1053 : : #endif
1054 : :
1055 : : #ifdef CONFIG_FLATMEM
1056 : : #define pfn_to_nid(pfn) (0)
1057 : : #endif
1058 : :
1059 : : #ifdef CONFIG_SPARSEMEM
1060 : :
1061 : : /*
1062 : : * SECTION_SHIFT #bits space required to store a section #
1063 : : *
1064 : : * PA_SECTION_SHIFT physical address to/from section number
1065 : : * PFN_SECTION_SHIFT pfn to/from section number
1066 : : */
1067 : : #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1068 : : #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1069 : :
1070 : : #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1071 : :
1072 : : #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1073 : : #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1074 : :
1075 : : #define SECTION_BLOCKFLAGS_BITS \
1076 : : ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1077 : :
1078 : : #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1079 : : #error Allocator MAX_ORDER exceeds SECTION_SIZE
1080 : : #endif
1081 : :
1082 : : #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
1083 : : #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
1084 : :
1085 : : #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1086 : : #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1087 : :
1088 : : struct page;
1089 : : struct page_cgroup;
1090 : : struct mem_section {
1091 : : /*
1092 : : * This is, logically, a pointer to an array of struct
1093 : : * pages. However, it is stored with some other magic.
1094 : : * (see sparse.c::sparse_init_one_section())
1095 : : *
1096 : : * Additionally during early boot we encode node id of
1097 : : * the location of the section here to guide allocation.
1098 : : * (see sparse.c::memory_present())
1099 : : *
1100 : : * Making it a UL at least makes someone do a cast
1101 : : * before using it wrong.
1102 : : */
1103 : : unsigned long section_mem_map;
1104 : :
1105 : : /* See declaration of similar field in struct zone */
1106 : : unsigned long *pageblock_flags;
1107 : : #ifdef CONFIG_MEMCG
1108 : : /*
1109 : : * If !SPARSEMEM, pgdat doesn't have page_cgroup pointer. We use
1110 : : * section. (see memcontrol.h/page_cgroup.h about this.)
1111 : : */
1112 : : struct page_cgroup *page_cgroup;
1113 : : unsigned long pad;
1114 : : #endif
1115 : : /*
1116 : : * WARNING: mem_section must be a power-of-2 in size for the
1117 : : * calculation and use of SECTION_ROOT_MASK to make sense.
1118 : : */
1119 : : };
1120 : :
1121 : : #ifdef CONFIG_SPARSEMEM_EXTREME
1122 : : #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1123 : : #else
1124 : : #define SECTIONS_PER_ROOT 1
1125 : : #endif
1126 : :
1127 : : #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1128 : : #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1129 : : #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1130 : :
1131 : : #ifdef CONFIG_SPARSEMEM_EXTREME
1132 : : extern struct mem_section *mem_section[NR_SECTION_ROOTS];
1133 : : #else
1134 : : extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1135 : : #endif
1136 : :
1137 : : static inline struct mem_section *__nr_to_section(unsigned long nr)
1138 : : {
1139 : : if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1140 : : return NULL;
1141 : : return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1142 : : }
1143 : : extern int __section_nr(struct mem_section* ms);
1144 : : extern unsigned long usemap_size(void);
1145 : :
1146 : : /*
1147 : : * We use the lower bits of the mem_map pointer to store
1148 : : * a little bit of information. There should be at least
1149 : : * 3 bits here due to 32-bit alignment.
1150 : : */
1151 : : #define SECTION_MARKED_PRESENT (1UL<<0)
1152 : : #define SECTION_HAS_MEM_MAP (1UL<<1)
1153 : : #define SECTION_MAP_LAST_BIT (1UL<<2)
1154 : : #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1155 : : #define SECTION_NID_SHIFT 2
1156 : :
1157 : : static inline struct page *__section_mem_map_addr(struct mem_section *section)
1158 : : {
1159 : : unsigned long map = section->section_mem_map;
1160 : : map &= SECTION_MAP_MASK;
1161 : : return (struct page *)map;
1162 : : }
1163 : :
1164 : : static inline int present_section(struct mem_section *section)
1165 : : {
1166 : : return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1167 : : }
1168 : :
1169 : : static inline int present_section_nr(unsigned long nr)
1170 : : {
1171 : : return present_section(__nr_to_section(nr));
1172 : : }
1173 : :
1174 : : static inline int valid_section(struct mem_section *section)
1175 : : {
1176 : : return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1177 : : }
1178 : :
1179 : : static inline int valid_section_nr(unsigned long nr)
1180 : : {
1181 : : return valid_section(__nr_to_section(nr));
1182 : : }
1183 : :
1184 : : static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1185 : : {
1186 : : return __nr_to_section(pfn_to_section_nr(pfn));
1187 : : }
1188 : :
1189 : : #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1190 : : static inline int pfn_valid(unsigned long pfn)
1191 : : {
1192 : : if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1193 : : return 0;
1194 : : return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
1195 : : }
1196 : : #endif
1197 : :
1198 : : static inline int pfn_present(unsigned long pfn)
1199 : : {
1200 : : if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1201 : : return 0;
1202 : : return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1203 : : }
1204 : :
1205 : : /*
1206 : : * These are _only_ used during initialisation, therefore they
1207 : : * can use __initdata ... They could have names to indicate
1208 : : * this restriction.
1209 : : */
1210 : : #ifdef CONFIG_NUMA
1211 : : #define pfn_to_nid(pfn) \
1212 : : ({ \
1213 : : unsigned long __pfn_to_nid_pfn = (pfn); \
1214 : : page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1215 : : })
1216 : : #else
1217 : : #define pfn_to_nid(pfn) (0)
1218 : : #endif
1219 : :
1220 : : #define early_pfn_valid(pfn) pfn_valid(pfn)
1221 : : void sparse_init(void);
1222 : : #else
1223 : : #define sparse_init() do {} while (0)
1224 : : #define sparse_index_init(_sec, _nid) do {} while (0)
1225 : : #endif /* CONFIG_SPARSEMEM */
1226 : :
1227 : : #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1228 : : bool early_pfn_in_nid(unsigned long pfn, int nid);
1229 : : #else
1230 : : #define early_pfn_in_nid(pfn, nid) (1)
1231 : : #endif
1232 : :
1233 : : #ifndef early_pfn_valid
1234 : : #define early_pfn_valid(pfn) (1)
1235 : : #endif
1236 : :
1237 : : void memory_present(int nid, unsigned long start, unsigned long end);
1238 : : unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
1239 : :
1240 : : /*
1241 : : * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1242 : : * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1243 : : * pfn_valid_within() should be used in this case; we optimise this away
1244 : : * when we have no holes within a MAX_ORDER_NR_PAGES block.
1245 : : */
1246 : : #ifdef CONFIG_HOLES_IN_ZONE
1247 : : #define pfn_valid_within(pfn) pfn_valid(pfn)
1248 : : #else
1249 : : #define pfn_valid_within(pfn) (1)
1250 : : #endif
1251 : :
1252 : : #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1253 : : /*
1254 : : * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1255 : : * associated with it or not. In FLATMEM, it is expected that holes always
1256 : : * have valid memmap as long as there is valid PFNs either side of the hole.
1257 : : * In SPARSEMEM, it is assumed that a valid section has a memmap for the
1258 : : * entire section.
1259 : : *
1260 : : * However, an ARM, and maybe other embedded architectures in the future
1261 : : * free memmap backing holes to save memory on the assumption the memmap is
1262 : : * never used. The page_zone linkages are then broken even though pfn_valid()
1263 : : * returns true. A walker of the full memmap must then do this additional
1264 : : * check to ensure the memmap they are looking at is sane by making sure
1265 : : * the zone and PFN linkages are still valid. This is expensive, but walkers
1266 : : * of the full memmap are extremely rare.
1267 : : */
1268 : : int memmap_valid_within(unsigned long pfn,
1269 : : struct page *page, struct zone *zone);
1270 : : #else
1271 : : static inline int memmap_valid_within(unsigned long pfn,
1272 : : struct page *page, struct zone *zone)
1273 : : {
1274 : : return 1;
1275 : : }
1276 : : #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1277 : :
1278 : : #endif /* !__GENERATING_BOUNDS.H */
1279 : : #endif /* !__ASSEMBLY__ */
1280 : : #endif /* _LINUX_MMZONE_H */
|
