1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_MMZONE_H
3 #define _LINUX_MMZONE_H
4
5 #ifndef __ASSEMBLY__
6 #ifndef __GENERATING_BOUNDS_H
7
8 #include <linux/spinlock.h>
9 #include <linux/list.h>
10 #include <linux/wait.h>
11 #include <linux/bitops.h>
12 #include <linux/cache.h>
13 #include <linux/threads.h>
14 #include <linux/numa.h>
15 #include <linux/init.h>
16 #include <linux/seqlock.h>
17 #include <linux/nodemask.h>
18 #include <linux/pageblock-flags.h>
19 #include <linux/page-flags-layout.h>
20 #include <linux/atomic.h>
21 #include <linux/mm_types.h>
22 #include <linux/page-flags.h>
23 #include <asm/page.h>
24
25 /* Free memory management - zoned buddy allocator. */
26 #ifndef CONFIG_FORCE_MAX_ZONEORDER
27 #define MAX_ORDER 11
28 #else
29 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
30 #endif
31 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
32
33 /*
34 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
35 * costly to service. That is between allocation orders which should
36 * coalesce naturally under reasonable reclaim pressure and those which
37 * will not.
38 */
39 #define PAGE_ALLOC_COSTLY_ORDER 3
40
41 enum migratetype {
42 MIGRATE_UNMOVABLE,
43 MIGRATE_MOVABLE,
44 MIGRATE_RECLAIMABLE,
45 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
46 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
47 #ifdef CONFIG_CMA
48 /*
49 * MIGRATE_CMA migration type is designed to mimic the way
50 * ZONE_MOVABLE works. Only movable pages can be allocated
51 * from MIGRATE_CMA pageblocks and page allocator never
52 * implicitly change migration type of MIGRATE_CMA pageblock.
53 *
54 * The way to use it is to change migratetype of a range of
55 * pageblocks to MIGRATE_CMA which can be done by
56 * __free_pageblock_cma() function. What is important though
57 * is that a range of pageblocks must be aligned to
58 * MAX_ORDER_NR_PAGES should biggest page be bigger then
59 * a single pageblock.
60 */
61 MIGRATE_CMA,
62 #endif
63 #ifdef CONFIG_MEMORY_ISOLATION
64 MIGRATE_ISOLATE, /* can't allocate from here */
65 #endif
66 MIGRATE_TYPES
67 };
68
69 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
70 extern const char * const migratetype_names[MIGRATE_TYPES];
71
72 #ifdef CONFIG_CMA
73 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
74 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
75 #else
76 # define is_migrate_cma(migratetype) false
77 # define is_migrate_cma_page(_page) false
78 #endif
79
is_migrate_movable(int mt)80 static inline bool is_migrate_movable(int mt)
81 {
82 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
83 }
84
85 #define for_each_migratetype_order(order, type) \
86 for (order = 0; order < MAX_ORDER; order++) \
87 for (type = 0; type < MIGRATE_TYPES; type++)
88
89 extern int page_group_by_mobility_disabled;
90
91 #define MIGRATETYPE_MASK ((1UL << PB_migratetype_bits) - 1)
92
93 #define get_pageblock_migratetype(page) \
94 get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK)
95
96 struct free_area {
97 struct list_head free_list[MIGRATE_TYPES];
98 unsigned long nr_free;
99 };
100
get_page_from_free_area(struct free_area * area,int migratetype)101 static inline struct page *get_page_from_free_area(struct free_area *area,
102 int migratetype)
103 {
104 return list_first_entry_or_null(&area->free_list[migratetype],
105 struct page, lru);
106 }
107
free_area_empty(struct free_area * area,int migratetype)108 static inline bool free_area_empty(struct free_area *area, int migratetype)
109 {
110 return list_empty(&area->free_list[migratetype]);
111 }
112
113 struct pglist_data;
114
115 /*
116 * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
117 * So add a wild amount of padding here to ensure that they fall into separate
118 * cachelines. There are very few zone structures in the machine, so space
119 * consumption is not a concern here.
120 */
121 #if defined(CONFIG_SMP)
122 struct zone_padding {
123 char x[0];
124 } ____cacheline_internodealigned_in_smp;
125 #define ZONE_PADDING(name) struct zone_padding name;
126 #else
127 #define ZONE_PADDING(name)
128 #endif
129
130 #ifdef CONFIG_NUMA
131 enum numa_stat_item {
132 NUMA_HIT, /* allocated in intended node */
133 NUMA_MISS, /* allocated in non intended node */
134 NUMA_FOREIGN, /* was intended here, hit elsewhere */
135 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
136 NUMA_LOCAL, /* allocation from local node */
137 NUMA_OTHER, /* allocation from other node */
138 NR_VM_NUMA_STAT_ITEMS
139 };
140 #else
141 #define NR_VM_NUMA_STAT_ITEMS 0
142 #endif
143
144 enum zone_stat_item {
145 /* First 128 byte cacheline (assuming 64 bit words) */
146 NR_FREE_PAGES,
147 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
148 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
149 NR_ZONE_ACTIVE_ANON,
150 NR_ZONE_INACTIVE_FILE,
151 NR_ZONE_ACTIVE_FILE,
152 NR_ZONE_UNEVICTABLE,
153 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
154 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
155 NR_PAGETABLE, /* used for pagetables */
156 /* Second 128 byte cacheline */
157 NR_BOUNCE,
158 #if IS_ENABLED(CONFIG_ZSMALLOC)
159 NR_ZSPAGES, /* allocated in zsmalloc */
160 #endif
161 NR_FREE_CMA_PAGES,
162 NR_VM_ZONE_STAT_ITEMS };
163
164 enum node_stat_item {
165 NR_LRU_BASE,
166 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
167 NR_ACTIVE_ANON, /* " " " " " */
168 NR_INACTIVE_FILE, /* " " " " " */
169 NR_ACTIVE_FILE, /* " " " " " */
170 NR_UNEVICTABLE, /* " " " " " */
171 NR_SLAB_RECLAIMABLE_B,
172 NR_SLAB_UNRECLAIMABLE_B,
173 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
174 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
175 WORKINGSET_NODES,
176 WORKINGSET_REFAULT_BASE,
177 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
178 WORKINGSET_REFAULT_FILE,
179 WORKINGSET_ACTIVATE_BASE,
180 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
181 WORKINGSET_ACTIVATE_FILE,
182 WORKINGSET_RESTORE_BASE,
183 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
184 WORKINGSET_RESTORE_FILE,
185 WORKINGSET_NODERECLAIM,
186 NR_ANON_MAPPED, /* Mapped anonymous pages */
187 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
188 only modified from process context */
189 NR_FILE_PAGES,
190 NR_FILE_DIRTY,
191 NR_WRITEBACK,
192 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
193 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
194 NR_SHMEM_THPS,
195 NR_SHMEM_PMDMAPPED,
196 NR_FILE_THPS,
197 NR_FILE_PMDMAPPED,
198 NR_ANON_THPS,
199 NR_VMSCAN_WRITE,
200 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
201 NR_DIRTIED, /* page dirtyings since bootup */
202 NR_WRITTEN, /* page writings since bootup */
203 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
204 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
205 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
206 NR_KERNEL_STACK_KB, /* measured in KiB */
207 #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
208 NR_KERNEL_SCS_KB, /* measured in KiB */
209 #endif
210 NR_VM_NODE_STAT_ITEMS
211 };
212
213 /*
214 * Returns true if the value is measured in bytes (most vmstat values are
215 * measured in pages). This defines the API part, the internal representation
216 * might be different.
217 */
vmstat_item_in_bytes(int idx)218 static __always_inline bool vmstat_item_in_bytes(int idx)
219 {
220 /*
221 * Global and per-node slab counters track slab pages.
222 * It's expected that changes are multiples of PAGE_SIZE.
223 * Internally values are stored in pages.
224 *
225 * Per-memcg and per-lruvec counters track memory, consumed
226 * by individual slab objects. These counters are actually
227 * byte-precise.
228 */
229 return (idx == NR_SLAB_RECLAIMABLE_B ||
230 idx == NR_SLAB_UNRECLAIMABLE_B);
231 }
232
233 /*
234 * We do arithmetic on the LRU lists in various places in the code,
235 * so it is important to keep the active lists LRU_ACTIVE higher in
236 * the array than the corresponding inactive lists, and to keep
237 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
238 *
239 * This has to be kept in sync with the statistics in zone_stat_item
240 * above and the descriptions in vmstat_text in mm/vmstat.c
241 */
242 #define LRU_BASE 0
243 #define LRU_ACTIVE 1
244 #define LRU_FILE 2
245
246 enum lru_list {
247 LRU_INACTIVE_ANON = LRU_BASE,
248 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
249 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
250 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
251 LRU_UNEVICTABLE,
252 NR_LRU_LISTS
253 };
254
255 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
256
257 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
258
is_file_lru(enum lru_list lru)259 static inline bool is_file_lru(enum lru_list lru)
260 {
261 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
262 }
263
is_active_lru(enum lru_list lru)264 static inline bool is_active_lru(enum lru_list lru)
265 {
266 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
267 }
268
269 #define ANON_AND_FILE 2
270
271 enum lruvec_flags {
272 LRUVEC_CONGESTED, /* lruvec has many dirty pages
273 * backed by a congested BDI
274 */
275 };
276
277 struct lruvec {
278 struct list_head lists[NR_LRU_LISTS];
279 /*
280 * These track the cost of reclaiming one LRU - file or anon -
281 * over the other. As the observed cost of reclaiming one LRU
282 * increases, the reclaim scan balance tips toward the other.
283 */
284 unsigned long anon_cost;
285 unsigned long file_cost;
286 /* Non-resident age, driven by LRU movement */
287 atomic_long_t nonresident_age;
288 /* Refaults at the time of last reclaim cycle */
289 unsigned long refaults[ANON_AND_FILE];
290 /* Various lruvec state flags (enum lruvec_flags) */
291 unsigned long flags;
292 #ifdef CONFIG_MEMCG
293 struct pglist_data *pgdat;
294 #endif
295 };
296
297 /* Isolate unmapped pages */
298 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
299 /* Isolate for asynchronous migration */
300 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
301 /* Isolate unevictable pages */
302 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
303
304 /* LRU Isolation modes. */
305 typedef unsigned __bitwise isolate_mode_t;
306
307 enum zone_watermarks {
308 WMARK_MIN,
309 WMARK_LOW,
310 WMARK_HIGH,
311 NR_WMARK
312 };
313
314 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
315 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
316 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
317 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
318
319 struct per_cpu_pages {
320 int count; /* number of pages in the list */
321 int high; /* high watermark, emptying needed */
322 int batch; /* chunk size for buddy add/remove */
323
324 /* Lists of pages, one per migrate type stored on the pcp-lists */
325 struct list_head lists[MIGRATE_PCPTYPES];
326 };
327
328 struct per_cpu_pageset {
329 struct per_cpu_pages pcp;
330 #ifdef CONFIG_NUMA
331 s8 expire;
332 u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
333 #endif
334 #ifdef CONFIG_SMP
335 s8 stat_threshold;
336 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
337 #endif
338 };
339
340 struct per_cpu_nodestat {
341 s8 stat_threshold;
342 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
343 };
344
345 #endif /* !__GENERATING_BOUNDS.H */
346
347 enum zone_type {
348 /*
349 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
350 * to DMA to all of the addressable memory (ZONE_NORMAL).
351 * On architectures where this area covers the whole 32 bit address
352 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
353 * DMA addressing constraints. This distinction is important as a 32bit
354 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
355 * platforms may need both zones as they support peripherals with
356 * different DMA addressing limitations.
357 *
358 * Some examples:
359 *
360 * - i386 and x86_64 have a fixed 16M ZONE_DMA and ZONE_DMA32 for the
361 * rest of the lower 4G.
362 *
363 * - arm only uses ZONE_DMA, the size, up to 4G, may vary depending on
364 * the specific device.
365 *
366 * - arm64 has a fixed 1G ZONE_DMA and ZONE_DMA32 for the rest of the
367 * lower 4G.
368 *
369 * - powerpc only uses ZONE_DMA, the size, up to 2G, may vary
370 * depending on the specific device.
371 *
372 * - s390 uses ZONE_DMA fixed to the lower 2G.
373 *
374 * - ia64 and riscv only use ZONE_DMA32.
375 *
376 * - parisc uses neither.
377 */
378 #ifdef CONFIG_ZONE_DMA
379 ZONE_DMA,
380 #endif
381 #ifdef CONFIG_ZONE_DMA32
382 ZONE_DMA32,
383 #endif
384 /*
385 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
386 * performed on pages in ZONE_NORMAL if the DMA devices support
387 * transfers to all addressable memory.
388 */
389 ZONE_NORMAL,
390 #ifdef CONFIG_HIGHMEM
391 /*
392 * A memory area that is only addressable by the kernel through
393 * mapping portions into its own address space. This is for example
394 * used by i386 to allow the kernel to address the memory beyond
395 * 900MB. The kernel will set up special mappings (page
396 * table entries on i386) for each page that the kernel needs to
397 * access.
398 */
399 ZONE_HIGHMEM,
400 #endif
401 /*
402 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
403 * movable pages with few exceptional cases described below. Main use
404 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
405 * likely to succeed, and to locally limit unmovable allocations - e.g.,
406 * to increase the number of THP/huge pages. Notable special cases are:
407 *
408 * 1. Pinned pages: (long-term) pinning of movable pages might
409 * essentially turn such pages unmovable. Memory offlining might
410 * retry a long time.
411 * 2. memblock allocations: kernelcore/movablecore setups might create
412 * situations where ZONE_MOVABLE contains unmovable allocations
413 * after boot. Memory offlining and allocations fail early.
414 * 3. Memory holes: kernelcore/movablecore setups might create very rare
415 * situations where ZONE_MOVABLE contains memory holes after boot,
416 * for example, if we have sections that are only partially
417 * populated. Memory offlining and allocations fail early.
418 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
419 * memory offlining, such pages cannot be allocated.
420 * 5. Unmovable PG_offline pages: in paravirtualized environments,
421 * hotplugged memory blocks might only partially be managed by the
422 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
423 * parts not manged by the buddy are unmovable PG_offline pages. In
424 * some cases (virtio-mem), such pages can be skipped during
425 * memory offlining, however, cannot be moved/allocated. These
426 * techniques might use alloc_contig_range() to hide previously
427 * exposed pages from the buddy again (e.g., to implement some sort
428 * of memory unplug in virtio-mem).
429 *
430 * In general, no unmovable allocations that degrade memory offlining
431 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
432 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
433 * if has_unmovable_pages() states that there are no unmovable pages,
434 * there can be false negatives).
435 */
436 ZONE_MOVABLE,
437 #ifdef CONFIG_ZONE_DEVICE
438 ZONE_DEVICE,
439 #endif
440 __MAX_NR_ZONES
441
442 };
443
444 #ifndef __GENERATING_BOUNDS_H
445
446 #define ASYNC_AND_SYNC 2
447
448 struct zone {
449 /* Read-mostly fields */
450
451 /* zone watermarks, access with *_wmark_pages(zone) macros */
452 unsigned long _watermark[NR_WMARK];
453 unsigned long watermark_boost;
454
455 unsigned long nr_reserved_highatomic;
456
457 /*
458 * We don't know if the memory that we're going to allocate will be
459 * freeable or/and it will be released eventually, so to avoid totally
460 * wasting several GB of ram we must reserve some of the lower zone
461 * memory (otherwise we risk to run OOM on the lower zones despite
462 * there being tons of freeable ram on the higher zones). This array is
463 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
464 * changes.
465 */
466 long lowmem_reserve[MAX_NR_ZONES];
467
468 #ifdef CONFIG_NUMA
469 int node;
470 #endif
471 struct pglist_data *zone_pgdat;
472 struct per_cpu_pageset __percpu *pageset;
473
474 #ifndef CONFIG_SPARSEMEM
475 /*
476 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
477 * In SPARSEMEM, this map is stored in struct mem_section
478 */
479 unsigned long *pageblock_flags;
480 #endif /* CONFIG_SPARSEMEM */
481
482 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
483 unsigned long zone_start_pfn;
484
485 /*
486 * spanned_pages is the total pages spanned by the zone, including
487 * holes, which is calculated as:
488 * spanned_pages = zone_end_pfn - zone_start_pfn;
489 *
490 * present_pages is physical pages existing within the zone, which
491 * is calculated as:
492 * present_pages = spanned_pages - absent_pages(pages in holes);
493 *
494 * managed_pages is present pages managed by the buddy system, which
495 * is calculated as (reserved_pages includes pages allocated by the
496 * bootmem allocator):
497 * managed_pages = present_pages - reserved_pages;
498 *
499 * So present_pages may be used by memory hotplug or memory power
500 * management logic to figure out unmanaged pages by checking
501 * (present_pages - managed_pages). And managed_pages should be used
502 * by page allocator and vm scanner to calculate all kinds of watermarks
503 * and thresholds.
504 *
505 * Locking rules:
506 *
507 * zone_start_pfn and spanned_pages are protected by span_seqlock.
508 * It is a seqlock because it has to be read outside of zone->lock,
509 * and it is done in the main allocator path. But, it is written
510 * quite infrequently.
511 *
512 * The span_seq lock is declared along with zone->lock because it is
513 * frequently read in proximity to zone->lock. It's good to
514 * give them a chance of being in the same cacheline.
515 *
516 * Write access to present_pages at runtime should be protected by
517 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
518 * present_pages should get_online_mems() to get a stable value.
519 */
520 atomic_long_t managed_pages;
521 unsigned long spanned_pages;
522 unsigned long present_pages;
523
524 const char *name;
525
526 #ifdef CONFIG_MEMORY_ISOLATION
527 /*
528 * Number of isolated pageblock. It is used to solve incorrect
529 * freepage counting problem due to racy retrieving migratetype
530 * of pageblock. Protected by zone->lock.
531 */
532 unsigned long nr_isolate_pageblock;
533 #endif
534
535 #ifdef CONFIG_MEMORY_HOTPLUG
536 /* see spanned/present_pages for more description */
537 seqlock_t span_seqlock;
538 #endif
539
540 int initialized;
541
542 /* Write-intensive fields used from the page allocator */
543 ZONE_PADDING(_pad1_)
544
545 /* free areas of different sizes */
546 struct free_area free_area[MAX_ORDER];
547
548 /* zone flags, see below */
549 unsigned long flags;
550
551 /* Primarily protects free_area */
552 spinlock_t lock;
553
554 /* Write-intensive fields used by compaction and vmstats. */
555 ZONE_PADDING(_pad2_)
556
557 /*
558 * When free pages are below this point, additional steps are taken
559 * when reading the number of free pages to avoid per-cpu counter
560 * drift allowing watermarks to be breached
561 */
562 unsigned long percpu_drift_mark;
563
564 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
565 /* pfn where compaction free scanner should start */
566 unsigned long compact_cached_free_pfn;
567 /* pfn where compaction migration scanner should start */
568 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
569 unsigned long compact_init_migrate_pfn;
570 unsigned long compact_init_free_pfn;
571 #endif
572
573 #ifdef CONFIG_COMPACTION
574 /*
575 * On compaction failure, 1<<compact_defer_shift compactions
576 * are skipped before trying again. The number attempted since
577 * last failure is tracked with compact_considered.
578 * compact_order_failed is the minimum compaction failed order.
579 */
580 unsigned int compact_considered;
581 unsigned int compact_defer_shift;
582 int compact_order_failed;
583 #endif
584
585 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
586 /* Set to true when the PG_migrate_skip bits should be cleared */
587 bool compact_blockskip_flush;
588 #endif
589
590 bool contiguous;
591
592 ZONE_PADDING(_pad3_)
593 /* Zone statistics */
594 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
595 atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
596 } ____cacheline_internodealigned_in_smp;
597
598 enum pgdat_flags {
599 PGDAT_DIRTY, /* reclaim scanning has recently found
600 * many dirty file pages at the tail
601 * of the LRU.
602 */
603 PGDAT_WRITEBACK, /* reclaim scanning has recently found
604 * many pages under writeback
605 */
606 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
607 };
608
609 enum zone_flags {
610 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
611 * Cleared when kswapd is woken.
612 */
613 };
614
zone_managed_pages(struct zone * zone)615 static inline unsigned long zone_managed_pages(struct zone *zone)
616 {
617 return (unsigned long)atomic_long_read(&zone->managed_pages);
618 }
619
zone_end_pfn(const struct zone * zone)620 static inline unsigned long zone_end_pfn(const struct zone *zone)
621 {
622 return zone->zone_start_pfn + zone->spanned_pages;
623 }
624
zone_spans_pfn(const struct zone * zone,unsigned long pfn)625 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
626 {
627 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
628 }
629
zone_is_initialized(struct zone * zone)630 static inline bool zone_is_initialized(struct zone *zone)
631 {
632 return zone->initialized;
633 }
634
zone_is_empty(struct zone * zone)635 static inline bool zone_is_empty(struct zone *zone)
636 {
637 return zone->spanned_pages == 0;
638 }
639
640 /*
641 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
642 * intersection with the given zone
643 */
zone_intersects(struct zone * zone,unsigned long start_pfn,unsigned long nr_pages)644 static inline bool zone_intersects(struct zone *zone,
645 unsigned long start_pfn, unsigned long nr_pages)
646 {
647 if (zone_is_empty(zone))
648 return false;
649 if (start_pfn >= zone_end_pfn(zone) ||
650 start_pfn + nr_pages <= zone->zone_start_pfn)
651 return false;
652
653 return true;
654 }
655
656 /*
657 * The "priority" of VM scanning is how much of the queues we will scan in one
658 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
659 * queues ("queue_length >> 12") during an aging round.
660 */
661 #define DEF_PRIORITY 12
662
663 /* Maximum number of zones on a zonelist */
664 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
665
666 enum {
667 ZONELIST_FALLBACK, /* zonelist with fallback */
668 #ifdef CONFIG_NUMA
669 /*
670 * The NUMA zonelists are doubled because we need zonelists that
671 * restrict the allocations to a single node for __GFP_THISNODE.
672 */
673 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
674 #endif
675 MAX_ZONELISTS
676 };
677
678 /*
679 * This struct contains information about a zone in a zonelist. It is stored
680 * here to avoid dereferences into large structures and lookups of tables
681 */
682 struct zoneref {
683 struct zone *zone; /* Pointer to actual zone */
684 int zone_idx; /* zone_idx(zoneref->zone) */
685 };
686
687 /*
688 * One allocation request operates on a zonelist. A zonelist
689 * is a list of zones, the first one is the 'goal' of the
690 * allocation, the other zones are fallback zones, in decreasing
691 * priority.
692 *
693 * To speed the reading of the zonelist, the zonerefs contain the zone index
694 * of the entry being read. Helper functions to access information given
695 * a struct zoneref are
696 *
697 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
698 * zonelist_zone_idx() - Return the index of the zone for an entry
699 * zonelist_node_idx() - Return the index of the node for an entry
700 */
701 struct zonelist {
702 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
703 };
704
705 #ifndef CONFIG_DISCONTIGMEM
706 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
707 extern struct page *mem_map;
708 #endif
709
710 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
711 struct deferred_split {
712 spinlock_t split_queue_lock;
713 struct list_head split_queue;
714 unsigned long split_queue_len;
715 };
716 #endif
717
718 /*
719 * On NUMA machines, each NUMA node would have a pg_data_t to describe
720 * it's memory layout. On UMA machines there is a single pglist_data which
721 * describes the whole memory.
722 *
723 * Memory statistics and page replacement data structures are maintained on a
724 * per-zone basis.
725 */
726 typedef struct pglist_data {
727 /*
728 * node_zones contains just the zones for THIS node. Not all of the
729 * zones may be populated, but it is the full list. It is referenced by
730 * this node's node_zonelists as well as other node's node_zonelists.
731 */
732 struct zone node_zones[MAX_NR_ZONES];
733
734 /*
735 * node_zonelists contains references to all zones in all nodes.
736 * Generally the first zones will be references to this node's
737 * node_zones.
738 */
739 struct zonelist node_zonelists[MAX_ZONELISTS];
740
741 int nr_zones; /* number of populated zones in this node */
742 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
743 struct page *node_mem_map;
744 #ifdef CONFIG_PAGE_EXTENSION
745 struct page_ext *node_page_ext;
746 #endif
747 #endif
748 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
749 /*
750 * Must be held any time you expect node_start_pfn,
751 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
752 * Also synchronizes pgdat->first_deferred_pfn during deferred page
753 * init.
754 *
755 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
756 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
757 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
758 *
759 * Nests above zone->lock and zone->span_seqlock
760 */
761 spinlock_t node_size_lock;
762 #endif
763 unsigned long node_start_pfn;
764 unsigned long node_present_pages; /* total number of physical pages */
765 unsigned long node_spanned_pages; /* total size of physical page
766 range, including holes */
767 int node_id;
768 wait_queue_head_t kswapd_wait;
769 wait_queue_head_t pfmemalloc_wait;
770 struct task_struct *kswapd; /* Protected by
771 mem_hotplug_begin/end() */
772 int kswapd_order;
773 enum zone_type kswapd_highest_zoneidx;
774
775 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
776
777 #ifdef CONFIG_COMPACTION
778 int kcompactd_max_order;
779 enum zone_type kcompactd_highest_zoneidx;
780 wait_queue_head_t kcompactd_wait;
781 struct task_struct *kcompactd;
782 #endif
783 /*
784 * This is a per-node reserve of pages that are not available
785 * to userspace allocations.
786 */
787 unsigned long totalreserve_pages;
788
789 #ifdef CONFIG_NUMA
790 /*
791 * node reclaim becomes active if more unmapped pages exist.
792 */
793 unsigned long min_unmapped_pages;
794 unsigned long min_slab_pages;
795 #endif /* CONFIG_NUMA */
796
797 /* Write-intensive fields used by page reclaim */
798 ZONE_PADDING(_pad1_)
799 spinlock_t lru_lock;
800
801 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
802 /*
803 * If memory initialisation on large machines is deferred then this
804 * is the first PFN that needs to be initialised.
805 */
806 unsigned long first_deferred_pfn;
807 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
808
809 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
810 struct deferred_split deferred_split_queue;
811 #endif
812
813 /* Fields commonly accessed by the page reclaim scanner */
814
815 /*
816 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
817 *
818 * Use mem_cgroup_lruvec() to look up lruvecs.
819 */
820 struct lruvec __lruvec;
821
822 unsigned long flags;
823
824 ZONE_PADDING(_pad2_)
825
826 /* Per-node vmstats */
827 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
828 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
829 } pg_data_t;
830
831 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
832 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
833 #ifdef CONFIG_FLAT_NODE_MEM_MAP
834 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
835 #else
836 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
837 #endif
838 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
839
840 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
841 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
842
pgdat_end_pfn(pg_data_t * pgdat)843 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
844 {
845 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
846 }
847
pgdat_is_empty(pg_data_t * pgdat)848 static inline bool pgdat_is_empty(pg_data_t *pgdat)
849 {
850 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
851 }
852
853 #include <linux/memory_hotplug.h>
854
855 void build_all_zonelists(pg_data_t *pgdat);
856 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
857 enum zone_type highest_zoneidx);
858 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
859 int highest_zoneidx, unsigned int alloc_flags,
860 long free_pages);
861 bool zone_watermark_ok(struct zone *z, unsigned int order,
862 unsigned long mark, int highest_zoneidx,
863 unsigned int alloc_flags);
864 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
865 unsigned long mark, int highest_zoneidx);
866 /*
867 * Memory initialization context, use to differentiate memory added by
868 * the platform statically or via memory hotplug interface.
869 */
870 enum meminit_context {
871 MEMINIT_EARLY,
872 MEMINIT_HOTPLUG,
873 };
874
875 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
876 unsigned long size);
877
878 extern void lruvec_init(struct lruvec *lruvec);
879
lruvec_pgdat(struct lruvec * lruvec)880 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
881 {
882 #ifdef CONFIG_MEMCG
883 return lruvec->pgdat;
884 #else
885 return container_of(lruvec, struct pglist_data, __lruvec);
886 #endif
887 }
888
889 extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
890
891 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
892 int local_memory_node(int node_id);
893 #else
local_memory_node(int node_id)894 static inline int local_memory_node(int node_id) { return node_id; };
895 #endif
896
897 /*
898 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
899 */
900 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
901
902 /*
903 * Returns true if a zone has pages managed by the buddy allocator.
904 * All the reclaim decisions have to use this function rather than
905 * populated_zone(). If the whole zone is reserved then we can easily
906 * end up with populated_zone() && !managed_zone().
907 */
managed_zone(struct zone * zone)908 static inline bool managed_zone(struct zone *zone)
909 {
910 return zone_managed_pages(zone);
911 }
912
913 /* Returns true if a zone has memory */
populated_zone(struct zone * zone)914 static inline bool populated_zone(struct zone *zone)
915 {
916 return zone->present_pages;
917 }
918
919 #ifdef CONFIG_NUMA
zone_to_nid(struct zone * zone)920 static inline int zone_to_nid(struct zone *zone)
921 {
922 return zone->node;
923 }
924
zone_set_nid(struct zone * zone,int nid)925 static inline void zone_set_nid(struct zone *zone, int nid)
926 {
927 zone->node = nid;
928 }
929 #else
zone_to_nid(struct zone * zone)930 static inline int zone_to_nid(struct zone *zone)
931 {
932 return 0;
933 }
934
zone_set_nid(struct zone * zone,int nid)935 static inline void zone_set_nid(struct zone *zone, int nid) {}
936 #endif
937
938 extern int movable_zone;
939
940 #ifdef CONFIG_HIGHMEM
zone_movable_is_highmem(void)941 static inline int zone_movable_is_highmem(void)
942 {
943 #ifdef CONFIG_NEED_MULTIPLE_NODES
944 return movable_zone == ZONE_HIGHMEM;
945 #else
946 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
947 #endif
948 }
949 #endif
950
is_highmem_idx(enum zone_type idx)951 static inline int is_highmem_idx(enum zone_type idx)
952 {
953 #ifdef CONFIG_HIGHMEM
954 return (idx == ZONE_HIGHMEM ||
955 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
956 #else
957 return 0;
958 #endif
959 }
960
961 /**
962 * is_highmem - helper function to quickly check if a struct zone is a
963 * highmem zone or not. This is an attempt to keep references
964 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
965 * @zone - pointer to struct zone variable
966 */
is_highmem(struct zone * zone)967 static inline int is_highmem(struct zone *zone)
968 {
969 #ifdef CONFIG_HIGHMEM
970 return is_highmem_idx(zone_idx(zone));
971 #else
972 return 0;
973 #endif
974 }
975
976 /* These two functions are used to setup the per zone pages min values */
977 struct ctl_table;
978
979 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, void *, size_t *,
980 loff_t *);
981 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, void *,
982 size_t *, loff_t *);
983 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
984 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, void *,
985 size_t *, loff_t *);
986 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
987 void *, size_t *, loff_t *);
988 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
989 void *, size_t *, loff_t *);
990 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
991 void *, size_t *, loff_t *);
992 int numa_zonelist_order_handler(struct ctl_table *, int,
993 void *, size_t *, loff_t *);
994 extern int percpu_pagelist_fraction;
995 extern char numa_zonelist_order[];
996 #define NUMA_ZONELIST_ORDER_LEN 16
997
998 #ifndef CONFIG_NEED_MULTIPLE_NODES
999
1000 extern struct pglist_data contig_page_data;
1001 #define NODE_DATA(nid) (&contig_page_data)
1002 #define NODE_MEM_MAP(nid) mem_map
1003
1004 #else /* CONFIG_NEED_MULTIPLE_NODES */
1005
1006 #include <asm/mmzone.h>
1007
1008 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
1009
1010 extern struct pglist_data *first_online_pgdat(void);
1011 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1012 extern struct zone *next_zone(struct zone *zone);
1013
1014 /**
1015 * for_each_online_pgdat - helper macro to iterate over all online nodes
1016 * @pgdat - pointer to a pg_data_t variable
1017 */
1018 #define for_each_online_pgdat(pgdat) \
1019 for (pgdat = first_online_pgdat(); \
1020 pgdat; \
1021 pgdat = next_online_pgdat(pgdat))
1022 /**
1023 * for_each_zone - helper macro to iterate over all memory zones
1024 * @zone - pointer to struct zone variable
1025 *
1026 * The user only needs to declare the zone variable, for_each_zone
1027 * fills it in.
1028 */
1029 #define for_each_zone(zone) \
1030 for (zone = (first_online_pgdat())->node_zones; \
1031 zone; \
1032 zone = next_zone(zone))
1033
1034 #define for_each_populated_zone(zone) \
1035 for (zone = (first_online_pgdat())->node_zones; \
1036 zone; \
1037 zone = next_zone(zone)) \
1038 if (!populated_zone(zone)) \
1039 ; /* do nothing */ \
1040 else
1041
zonelist_zone(struct zoneref * zoneref)1042 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1043 {
1044 return zoneref->zone;
1045 }
1046
zonelist_zone_idx(struct zoneref * zoneref)1047 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1048 {
1049 return zoneref->zone_idx;
1050 }
1051
zonelist_node_idx(struct zoneref * zoneref)1052 static inline int zonelist_node_idx(struct zoneref *zoneref)
1053 {
1054 return zone_to_nid(zoneref->zone);
1055 }
1056
1057 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1058 enum zone_type highest_zoneidx,
1059 nodemask_t *nodes);
1060
1061 /**
1062 * 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
1063 * @z - The cursor used as a starting point for the search
1064 * @highest_zoneidx - The zone index of the highest zone to return
1065 * @nodes - An optional nodemask to filter the zonelist with
1066 *
1067 * This function returns the next zone at or below a given zone index that is
1068 * within the allowed nodemask using a cursor as the starting point for the
1069 * search. The zoneref returned is a cursor that represents the current zone
1070 * being examined. It should be advanced by one before calling
1071 * next_zones_zonelist again.
1072 */
next_zones_zonelist(struct zoneref * z,enum zone_type highest_zoneidx,nodemask_t * nodes)1073 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1074 enum zone_type highest_zoneidx,
1075 nodemask_t *nodes)
1076 {
1077 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1078 return z;
1079 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1080 }
1081
1082 /**
1083 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1084 * @zonelist - The zonelist to search for a suitable zone
1085 * @highest_zoneidx - The zone index of the highest zone to return
1086 * @nodes - An optional nodemask to filter the zonelist with
1087 * @return - Zoneref pointer for the first suitable zone found (see below)
1088 *
1089 * This function returns the first zone at or below a given zone index that is
1090 * within the allowed nodemask. The zoneref returned is a cursor that can be
1091 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1092 * one before calling.
1093 *
1094 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1095 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1096 * update due to cpuset modification.
1097 */
first_zones_zonelist(struct zonelist * zonelist,enum zone_type highest_zoneidx,nodemask_t * nodes)1098 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1099 enum zone_type highest_zoneidx,
1100 nodemask_t *nodes)
1101 {
1102 return next_zones_zonelist(zonelist->_zonerefs,
1103 highest_zoneidx, nodes);
1104 }
1105
1106 /**
1107 * 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
1108 * @zone - The current zone in the iterator
1109 * @z - The current pointer within zonelist->_zonerefs being iterated
1110 * @zlist - The zonelist being iterated
1111 * @highidx - The zone index of the highest zone to return
1112 * @nodemask - Nodemask allowed by the allocator
1113 *
1114 * This iterator iterates though all zones at or below a given zone index and
1115 * within a given nodemask
1116 */
1117 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1118 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1119 zone; \
1120 z = next_zones_zonelist(++z, highidx, nodemask), \
1121 zone = zonelist_zone(z))
1122
1123 #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1124 for (zone = z->zone; \
1125 zone; \
1126 z = next_zones_zonelist(++z, highidx, nodemask), \
1127 zone = zonelist_zone(z))
1128
1129
1130 /**
1131 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1132 * @zone - The current zone in the iterator
1133 * @z - The current pointer within zonelist->zones being iterated
1134 * @zlist - The zonelist being iterated
1135 * @highidx - The zone index of the highest zone to return
1136 *
1137 * This iterator iterates though all zones at or below a given zone index.
1138 */
1139 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1140 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1141
1142 #ifdef CONFIG_SPARSEMEM
1143 #include <asm/sparsemem.h>
1144 #endif
1145
1146 #ifdef CONFIG_FLATMEM
1147 #define pfn_to_nid(pfn) (0)
1148 #endif
1149
1150 #ifdef CONFIG_SPARSEMEM
1151
1152 /*
1153 * SECTION_SHIFT #bits space required to store a section #
1154 *
1155 * PA_SECTION_SHIFT physical address to/from section number
1156 * PFN_SECTION_SHIFT pfn to/from section number
1157 */
1158 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1159 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1160
1161 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1162
1163 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1164 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1165
1166 #define SECTION_BLOCKFLAGS_BITS \
1167 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1168
1169 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1170 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1171 #endif
1172
pfn_to_section_nr(unsigned long pfn)1173 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1174 {
1175 return pfn >> PFN_SECTION_SHIFT;
1176 }
section_nr_to_pfn(unsigned long sec)1177 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1178 {
1179 return sec << PFN_SECTION_SHIFT;
1180 }
1181
1182 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1183 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1184
1185 #define SUBSECTION_SHIFT 21
1186 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1187
1188 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1189 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1190 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1191
1192 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1193 #error Subsection size exceeds section size
1194 #else
1195 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1196 #endif
1197
1198 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1199 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1200
1201 struct mem_section_usage {
1202 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1203 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1204 #endif
1205 /* See declaration of similar field in struct zone */
1206 unsigned long pageblock_flags[0];
1207 };
1208
1209 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1210
1211 struct page;
1212 struct page_ext;
1213 struct mem_section {
1214 /*
1215 * This is, logically, a pointer to an array of struct
1216 * pages. However, it is stored with some other magic.
1217 * (see sparse.c::sparse_init_one_section())
1218 *
1219 * Additionally during early boot we encode node id of
1220 * the location of the section here to guide allocation.
1221 * (see sparse.c::memory_present())
1222 *
1223 * Making it a UL at least makes someone do a cast
1224 * before using it wrong.
1225 */
1226 unsigned long section_mem_map;
1227
1228 struct mem_section_usage *usage;
1229 #ifdef CONFIG_PAGE_EXTENSION
1230 /*
1231 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1232 * section. (see page_ext.h about this.)
1233 */
1234 struct page_ext *page_ext;
1235 unsigned long pad;
1236 #endif
1237 /*
1238 * WARNING: mem_section must be a power-of-2 in size for the
1239 * calculation and use of SECTION_ROOT_MASK to make sense.
1240 */
1241 };
1242
1243 #ifdef CONFIG_SPARSEMEM_EXTREME
1244 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1245 #else
1246 #define SECTIONS_PER_ROOT 1
1247 #endif
1248
1249 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1250 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1251 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1252
1253 #ifdef CONFIG_SPARSEMEM_EXTREME
1254 extern struct mem_section **mem_section;
1255 #else
1256 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1257 #endif
1258
section_to_usemap(struct mem_section * ms)1259 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1260 {
1261 return ms->usage->pageblock_flags;
1262 }
1263
__nr_to_section(unsigned long nr)1264 static inline struct mem_section *__nr_to_section(unsigned long nr)
1265 {
1266 #ifdef CONFIG_SPARSEMEM_EXTREME
1267 if (!mem_section)
1268 return NULL;
1269 #endif
1270 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1271 return NULL;
1272 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1273 }
1274 extern unsigned long __section_nr(struct mem_section *ms);
1275 extern size_t mem_section_usage_size(void);
1276
1277 /*
1278 * We use the lower bits of the mem_map pointer to store
1279 * a little bit of information. The pointer is calculated
1280 * as mem_map - section_nr_to_pfn(pnum). The result is
1281 * aligned to the minimum alignment of the two values:
1282 * 1. All mem_map arrays are page-aligned.
1283 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1284 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1285 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1286 * worst combination is powerpc with 256k pages,
1287 * which results in PFN_SECTION_SHIFT equal 6.
1288 * To sum it up, at least 6 bits are available.
1289 */
1290 #define SECTION_MARKED_PRESENT (1UL<<0)
1291 #define SECTION_HAS_MEM_MAP (1UL<<1)
1292 #define SECTION_IS_ONLINE (1UL<<2)
1293 #define SECTION_IS_EARLY (1UL<<3)
1294 #define SECTION_MAP_LAST_BIT (1UL<<4)
1295 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1296 #define SECTION_NID_SHIFT 3
1297
__section_mem_map_addr(struct mem_section * section)1298 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1299 {
1300 unsigned long map = section->section_mem_map;
1301 map &= SECTION_MAP_MASK;
1302 return (struct page *)map;
1303 }
1304
present_section(struct mem_section * section)1305 static inline int present_section(struct mem_section *section)
1306 {
1307 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1308 }
1309
present_section_nr(unsigned long nr)1310 static inline int present_section_nr(unsigned long nr)
1311 {
1312 return present_section(__nr_to_section(nr));
1313 }
1314
valid_section(struct mem_section * section)1315 static inline int valid_section(struct mem_section *section)
1316 {
1317 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1318 }
1319
early_section(struct mem_section * section)1320 static inline int early_section(struct mem_section *section)
1321 {
1322 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1323 }
1324
valid_section_nr(unsigned long nr)1325 static inline int valid_section_nr(unsigned long nr)
1326 {
1327 return valid_section(__nr_to_section(nr));
1328 }
1329
online_section(struct mem_section * section)1330 static inline int online_section(struct mem_section *section)
1331 {
1332 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1333 }
1334
online_section_nr(unsigned long nr)1335 static inline int online_section_nr(unsigned long nr)
1336 {
1337 return online_section(__nr_to_section(nr));
1338 }
1339
1340 #ifdef CONFIG_MEMORY_HOTPLUG
1341 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1342 #ifdef CONFIG_MEMORY_HOTREMOVE
1343 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1344 #endif
1345 #endif
1346
__pfn_to_section(unsigned long pfn)1347 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1348 {
1349 return __nr_to_section(pfn_to_section_nr(pfn));
1350 }
1351
1352 extern unsigned long __highest_present_section_nr;
1353
subsection_map_index(unsigned long pfn)1354 static inline int subsection_map_index(unsigned long pfn)
1355 {
1356 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1357 }
1358
1359 #ifdef CONFIG_SPARSEMEM_VMEMMAP
pfn_section_valid(struct mem_section * ms,unsigned long pfn)1360 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1361 {
1362 int idx = subsection_map_index(pfn);
1363
1364 return test_bit(idx, ms->usage->subsection_map);
1365 }
1366 #else
pfn_section_valid(struct mem_section * ms,unsigned long pfn)1367 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1368 {
1369 return 1;
1370 }
1371 #endif
1372
1373 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
pfn_valid(unsigned long pfn)1374 static inline int pfn_valid(unsigned long pfn)
1375 {
1376 struct mem_section *ms;
1377
1378 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1379 return 0;
1380 ms = __nr_to_section(pfn_to_section_nr(pfn));
1381 if (!valid_section(ms))
1382 return 0;
1383 /*
1384 * Traditionally early sections always returned pfn_valid() for
1385 * the entire section-sized span.
1386 */
1387 return early_section(ms) || pfn_section_valid(ms, pfn);
1388 }
1389 #endif
1390
pfn_in_present_section(unsigned long pfn)1391 static inline int pfn_in_present_section(unsigned long pfn)
1392 {
1393 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1394 return 0;
1395 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1396 }
1397
next_present_section_nr(unsigned long section_nr)1398 static inline unsigned long next_present_section_nr(unsigned long section_nr)
1399 {
1400 while (++section_nr <= __highest_present_section_nr) {
1401 if (present_section_nr(section_nr))
1402 return section_nr;
1403 }
1404
1405 return -1;
1406 }
1407
1408 /*
1409 * These are _only_ used during initialisation, therefore they
1410 * can use __initdata ... They could have names to indicate
1411 * this restriction.
1412 */
1413 #ifdef CONFIG_NUMA
1414 #define pfn_to_nid(pfn) \
1415 ({ \
1416 unsigned long __pfn_to_nid_pfn = (pfn); \
1417 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1418 })
1419 #else
1420 #define pfn_to_nid(pfn) (0)
1421 #endif
1422
1423 void sparse_init(void);
1424 #else
1425 #define sparse_init() do {} while (0)
1426 #define sparse_index_init(_sec, _nid) do {} while (0)
1427 #define pfn_in_present_section pfn_valid
1428 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
1429 #endif /* CONFIG_SPARSEMEM */
1430
1431 /*
1432 * During memory init memblocks map pfns to nids. The search is expensive and
1433 * this caches recent lookups. The implementation of __early_pfn_to_nid
1434 * may treat start/end as pfns or sections.
1435 */
1436 struct mminit_pfnnid_cache {
1437 unsigned long last_start;
1438 unsigned long last_end;
1439 int last_nid;
1440 };
1441
1442 /*
1443 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1444 * need to check pfn validity within that MAX_ORDER_NR_PAGES block.
1445 * pfn_valid_within() should be used in this case; we optimise this away
1446 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1447 */
1448 #ifdef CONFIG_HOLES_IN_ZONE
1449 #define pfn_valid_within(pfn) pfn_valid(pfn)
1450 #else
1451 #define pfn_valid_within(pfn) (1)
1452 #endif
1453
1454 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1455 /*
1456 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
1457 * associated with it or not. This means that a struct page exists for this
1458 * pfn. The caller cannot assume the page is fully initialized in general.
1459 * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1460 * will ensure the struct page is fully online and initialized. Special pages
1461 * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1462 *
1463 * In FLATMEM, it is expected that holes always have valid memmap as long as
1464 * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1465 * that a valid section has a memmap for the entire section.
1466 *
1467 * However, an ARM, and maybe other embedded architectures in the future
1468 * free memmap backing holes to save memory on the assumption the memmap is
1469 * never used. The page_zone linkages are then broken even though pfn_valid()
1470 * returns true. A walker of the full memmap must then do this additional
1471 * check to ensure the memmap they are looking at is sane by making sure
1472 * the zone and PFN linkages are still valid. This is expensive, but walkers
1473 * of the full memmap are extremely rare.
1474 */
1475 bool memmap_valid_within(unsigned long pfn,
1476 struct page *page, struct zone *zone);
1477 #else
memmap_valid_within(unsigned long pfn,struct page * page,struct zone * zone)1478 static inline bool memmap_valid_within(unsigned long pfn,
1479 struct page *page, struct zone *zone)
1480 {
1481 return true;
1482 }
1483 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1484
1485 #endif /* !__GENERATING_BOUNDS.H */
1486 #endif /* !__ASSEMBLY__ */
1487 #endif /* _LINUX_MMZONE_H */
1488
