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 <linux/local_lock.h>
24 #include <asm/page.h>
25
26 /* Free memory management - zoned buddy allocator. */
27 #ifndef CONFIG_FORCE_MAX_ZONEORDER
28 #define MAX_ORDER 11
29 #else
30 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
31 #endif
32 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
33
34 /*
35 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
36 * costly to service. That is between allocation orders which should
37 * coalesce naturally under reasonable reclaim pressure and those which
38 * will not.
39 */
40 #define PAGE_ALLOC_COSTLY_ORDER 3
41
42 enum migratetype {
43 MIGRATE_UNMOVABLE,
44 MIGRATE_MOVABLE,
45 MIGRATE_RECLAIMABLE,
46 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
47 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
48 #ifdef CONFIG_CMA
49 /*
50 * MIGRATE_CMA migration type is designed to mimic the way
51 * ZONE_MOVABLE works. Only movable pages can be allocated
52 * from MIGRATE_CMA pageblocks and page allocator never
53 * implicitly change migration type of MIGRATE_CMA pageblock.
54 *
55 * The way to use it is to change migratetype of a range of
56 * pageblocks to MIGRATE_CMA which can be done by
57 * __free_pageblock_cma() function. What is important though
58 * is that a range of pageblocks must be aligned to
59 * MAX_ORDER_NR_PAGES should biggest page be bigger than
60 * a single pageblock.
61 */
62 MIGRATE_CMA,
63 #endif
64 #ifdef CONFIG_MEMORY_ISOLATION
65 MIGRATE_ISOLATE, /* can't allocate from here */
66 #endif
67 MIGRATE_TYPES
68 };
69
70 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
71 extern const char * const migratetype_names[MIGRATE_TYPES];
72
73 #ifdef CONFIG_CMA
74 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
75 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
76 #else
77 # define is_migrate_cma(migratetype) false
78 # define is_migrate_cma_page(_page) false
79 #endif
80
is_migrate_movable(int mt)81 static inline bool is_migrate_movable(int mt)
82 {
83 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
84 }
85
86 #define for_each_migratetype_order(order, type) \
87 for (order = 0; order < MAX_ORDER; order++) \
88 for (type = 0; type < MIGRATE_TYPES; type++)
89
90 extern int page_group_by_mobility_disabled;
91
92 #define MIGRATETYPE_MASK ((1UL << PB_migratetype_bits) - 1)
93
94 #define get_pageblock_migratetype(page) \
95 get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK)
96
97 struct free_area {
98 struct list_head free_list[MIGRATE_TYPES];
99 unsigned long nr_free;
100 };
101
get_page_from_free_area(struct free_area * area,int migratetype)102 static inline struct page *get_page_from_free_area(struct free_area *area,
103 int migratetype)
104 {
105 return list_first_entry_or_null(&area->free_list[migratetype],
106 struct page, lru);
107 }
108
free_area_empty(struct free_area * area,int migratetype)109 static inline bool free_area_empty(struct free_area *area, int migratetype)
110 {
111 return list_empty(&area->free_list[migratetype]);
112 }
113
114 struct pglist_data;
115
116 /*
117 * Add a wild amount of padding here to ensure data 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_EVENT_ITEMS
139 };
140 #else
141 #define NR_VM_NUMA_EVENT_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 /* Second 128 byte cacheline */
156 NR_BOUNCE,
157 #if IS_ENABLED(CONFIG_ZSMALLOC)
158 NR_ZSPAGES, /* allocated in zsmalloc */
159 #endif
160 NR_FREE_CMA_PAGES,
161 NR_VM_ZONE_STAT_ITEMS };
162
163 enum node_stat_item {
164 NR_LRU_BASE,
165 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
166 NR_ACTIVE_ANON, /* " " " " " */
167 NR_INACTIVE_FILE, /* " " " " " */
168 NR_ACTIVE_FILE, /* " " " " " */
169 NR_UNEVICTABLE, /* " " " " " */
170 NR_SLAB_RECLAIMABLE_B,
171 NR_SLAB_UNRECLAIMABLE_B,
172 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
173 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
174 WORKINGSET_NODES,
175 WORKINGSET_REFAULT_BASE,
176 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
177 WORKINGSET_REFAULT_FILE,
178 WORKINGSET_ACTIVATE_BASE,
179 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
180 WORKINGSET_ACTIVATE_FILE,
181 WORKINGSET_RESTORE_BASE,
182 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
183 WORKINGSET_RESTORE_FILE,
184 WORKINGSET_NODERECLAIM,
185 NR_ANON_MAPPED, /* Mapped anonymous pages */
186 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
187 only modified from process context */
188 NR_FILE_PAGES,
189 NR_FILE_DIRTY,
190 NR_WRITEBACK,
191 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
192 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
193 NR_SHMEM_THPS,
194 NR_SHMEM_PMDMAPPED,
195 NR_FILE_THPS,
196 NR_FILE_PMDMAPPED,
197 NR_ANON_THPS,
198 NR_VMSCAN_WRITE,
199 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
200 NR_DIRTIED, /* page dirtyings since bootup */
201 NR_WRITTEN, /* page writings since bootup */
202 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
203 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
204 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
205 NR_KERNEL_STACK_KB, /* measured in KiB */
206 #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
207 NR_KERNEL_SCS_KB, /* measured in KiB */
208 #endif
209 NR_PAGETABLE, /* used for pagetables */
210 #ifdef CONFIG_SWAP
211 NR_SWAPCACHE,
212 #endif
213 NR_VM_NODE_STAT_ITEMS
214 };
215
216 /*
217 * Returns true if the item should be printed in THPs (/proc/vmstat
218 * currently prints number of anon, file and shmem THPs. But the item
219 * is charged in pages).
220 */
vmstat_item_print_in_thp(enum node_stat_item item)221 static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item)
222 {
223 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
224 return false;
225
226 return item == NR_ANON_THPS ||
227 item == NR_FILE_THPS ||
228 item == NR_SHMEM_THPS ||
229 item == NR_SHMEM_PMDMAPPED ||
230 item == NR_FILE_PMDMAPPED;
231 }
232
233 /*
234 * Returns true if the value is measured in bytes (most vmstat values are
235 * measured in pages). This defines the API part, the internal representation
236 * might be different.
237 */
vmstat_item_in_bytes(int idx)238 static __always_inline bool vmstat_item_in_bytes(int idx)
239 {
240 /*
241 * Global and per-node slab counters track slab pages.
242 * It's expected that changes are multiples of PAGE_SIZE.
243 * Internally values are stored in pages.
244 *
245 * Per-memcg and per-lruvec counters track memory, consumed
246 * by individual slab objects. These counters are actually
247 * byte-precise.
248 */
249 return (idx == NR_SLAB_RECLAIMABLE_B ||
250 idx == NR_SLAB_UNRECLAIMABLE_B);
251 }
252
253 /*
254 * We do arithmetic on the LRU lists in various places in the code,
255 * so it is important to keep the active lists LRU_ACTIVE higher in
256 * the array than the corresponding inactive lists, and to keep
257 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
258 *
259 * This has to be kept in sync with the statistics in zone_stat_item
260 * above and the descriptions in vmstat_text in mm/vmstat.c
261 */
262 #define LRU_BASE 0
263 #define LRU_ACTIVE 1
264 #define LRU_FILE 2
265
266 enum lru_list {
267 LRU_INACTIVE_ANON = LRU_BASE,
268 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
269 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
270 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
271 LRU_UNEVICTABLE,
272 NR_LRU_LISTS
273 };
274
275 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
276
277 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
278
is_file_lru(enum lru_list lru)279 static inline bool is_file_lru(enum lru_list lru)
280 {
281 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
282 }
283
is_active_lru(enum lru_list lru)284 static inline bool is_active_lru(enum lru_list lru)
285 {
286 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
287 }
288
289 #define ANON_AND_FILE 2
290
291 enum lruvec_flags {
292 LRUVEC_CONGESTED, /* lruvec has many dirty pages
293 * backed by a congested BDI
294 */
295 };
296
297 struct lruvec {
298 struct list_head lists[NR_LRU_LISTS];
299 /* per lruvec lru_lock for memcg */
300 spinlock_t lru_lock;
301 /*
302 * These track the cost of reclaiming one LRU - file or anon -
303 * over the other. As the observed cost of reclaiming one LRU
304 * increases, the reclaim scan balance tips toward the other.
305 */
306 unsigned long anon_cost;
307 unsigned long file_cost;
308 /* Non-resident age, driven by LRU movement */
309 atomic_long_t nonresident_age;
310 /* Refaults at the time of last reclaim cycle */
311 unsigned long refaults[ANON_AND_FILE];
312 /* Various lruvec state flags (enum lruvec_flags) */
313 unsigned long flags;
314 #ifdef CONFIG_MEMCG
315 struct pglist_data *pgdat;
316 #endif
317 };
318
319 /* Isolate unmapped pages */
320 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
321 /* Isolate for asynchronous migration */
322 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
323 /* Isolate unevictable pages */
324 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
325
326 /* LRU Isolation modes. */
327 typedef unsigned __bitwise isolate_mode_t;
328
329 enum zone_watermarks {
330 WMARK_MIN,
331 WMARK_LOW,
332 WMARK_HIGH,
333 NR_WMARK
334 };
335
336 /*
337 * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER plus one additional
338 * for pageblock size for THP if configured.
339 */
340 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
341 #define NR_PCP_THP 1
342 #else
343 #define NR_PCP_THP 0
344 #endif
345 #define NR_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1 + NR_PCP_THP))
346
347 /*
348 * Shift to encode migratetype and order in the same integer, with order
349 * in the least significant bits.
350 */
351 #define NR_PCP_ORDER_WIDTH 8
352 #define NR_PCP_ORDER_MASK ((1<<NR_PCP_ORDER_WIDTH) - 1)
353
354 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
355 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
356 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
357 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
358
359 /* Fields and list protected by pagesets local_lock in page_alloc.c */
360 struct per_cpu_pages {
361 int count; /* number of pages in the list */
362 int high; /* high watermark, emptying needed */
363 int batch; /* chunk size for buddy add/remove */
364 short free_factor; /* batch scaling factor during free */
365 #ifdef CONFIG_NUMA
366 short expire; /* When 0, remote pagesets are drained */
367 #endif
368
369 /* Lists of pages, one per migrate type stored on the pcp-lists */
370 struct list_head lists[NR_PCP_LISTS];
371 };
372
373 struct per_cpu_zonestat {
374 #ifdef CONFIG_SMP
375 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
376 s8 stat_threshold;
377 #endif
378 #ifdef CONFIG_NUMA
379 /*
380 * Low priority inaccurate counters that are only folded
381 * on demand. Use a large type to avoid the overhead of
382 * folding during refresh_cpu_vm_stats.
383 */
384 unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
385 #endif
386 };
387
388 struct per_cpu_nodestat {
389 s8 stat_threshold;
390 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
391 };
392
393 #endif /* !__GENERATING_BOUNDS.H */
394
395 enum zone_type {
396 /*
397 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
398 * to DMA to all of the addressable memory (ZONE_NORMAL).
399 * On architectures where this area covers the whole 32 bit address
400 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
401 * DMA addressing constraints. This distinction is important as a 32bit
402 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
403 * platforms may need both zones as they support peripherals with
404 * different DMA addressing limitations.
405 */
406 #ifdef CONFIG_ZONE_DMA
407 ZONE_DMA,
408 #endif
409 #ifdef CONFIG_ZONE_DMA32
410 ZONE_DMA32,
411 #endif
412 /*
413 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
414 * performed on pages in ZONE_NORMAL if the DMA devices support
415 * transfers to all addressable memory.
416 */
417 ZONE_NORMAL,
418 #ifdef CONFIG_HIGHMEM
419 /*
420 * A memory area that is only addressable by the kernel through
421 * mapping portions into its own address space. This is for example
422 * used by i386 to allow the kernel to address the memory beyond
423 * 900MB. The kernel will set up special mappings (page
424 * table entries on i386) for each page that the kernel needs to
425 * access.
426 */
427 ZONE_HIGHMEM,
428 #endif
429 /*
430 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
431 * movable pages with few exceptional cases described below. Main use
432 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
433 * likely to succeed, and to locally limit unmovable allocations - e.g.,
434 * to increase the number of THP/huge pages. Notable special cases are:
435 *
436 * 1. Pinned pages: (long-term) pinning of movable pages might
437 * essentially turn such pages unmovable. Therefore, we do not allow
438 * pinning long-term pages in ZONE_MOVABLE. When pages are pinned and
439 * faulted, they come from the right zone right away. However, it is
440 * still possible that address space already has pages in
441 * ZONE_MOVABLE at the time when pages are pinned (i.e. user has
442 * touches that memory before pinning). In such case we migrate them
443 * to a different zone. When migration fails - pinning fails.
444 * 2. memblock allocations: kernelcore/movablecore setups might create
445 * situations where ZONE_MOVABLE contains unmovable allocations
446 * after boot. Memory offlining and allocations fail early.
447 * 3. Memory holes: kernelcore/movablecore setups might create very rare
448 * situations where ZONE_MOVABLE contains memory holes after boot,
449 * for example, if we have sections that are only partially
450 * populated. Memory offlining and allocations fail early.
451 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
452 * memory offlining, such pages cannot be allocated.
453 * 5. Unmovable PG_offline pages: in paravirtualized environments,
454 * hotplugged memory blocks might only partially be managed by the
455 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
456 * parts not manged by the buddy are unmovable PG_offline pages. In
457 * some cases (virtio-mem), such pages can be skipped during
458 * memory offlining, however, cannot be moved/allocated. These
459 * techniques might use alloc_contig_range() to hide previously
460 * exposed pages from the buddy again (e.g., to implement some sort
461 * of memory unplug in virtio-mem).
462 * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create
463 * situations where ZERO_PAGE(0) which is allocated differently
464 * on different platforms may end up in a movable zone. ZERO_PAGE(0)
465 * cannot be migrated.
466 * 7. Memory-hotplug: when using memmap_on_memory and onlining the
467 * memory to the MOVABLE zone, the vmemmap pages are also placed in
468 * such zone. Such pages cannot be really moved around as they are
469 * self-stored in the range, but they are treated as movable when
470 * the range they describe is about to be offlined.
471 *
472 * In general, no unmovable allocations that degrade memory offlining
473 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
474 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
475 * if has_unmovable_pages() states that there are no unmovable pages,
476 * there can be false negatives).
477 */
478 ZONE_MOVABLE,
479 #ifdef CONFIG_ZONE_DEVICE
480 ZONE_DEVICE,
481 #endif
482 __MAX_NR_ZONES
483
484 };
485
486 #ifndef __GENERATING_BOUNDS_H
487
488 #define ASYNC_AND_SYNC 2
489
490 struct zone {
491 /* Read-mostly fields */
492
493 /* zone watermarks, access with *_wmark_pages(zone) macros */
494 unsigned long _watermark[NR_WMARK];
495 unsigned long watermark_boost;
496
497 unsigned long nr_reserved_highatomic;
498
499 /*
500 * We don't know if the memory that we're going to allocate will be
501 * freeable or/and it will be released eventually, so to avoid totally
502 * wasting several GB of ram we must reserve some of the lower zone
503 * memory (otherwise we risk to run OOM on the lower zones despite
504 * there being tons of freeable ram on the higher zones). This array is
505 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
506 * changes.
507 */
508 long lowmem_reserve[MAX_NR_ZONES];
509
510 #ifdef CONFIG_NUMA
511 int node;
512 #endif
513 struct pglist_data *zone_pgdat;
514 struct per_cpu_pages __percpu *per_cpu_pageset;
515 struct per_cpu_zonestat __percpu *per_cpu_zonestats;
516 /*
517 * the high and batch values are copied to individual pagesets for
518 * faster access
519 */
520 int pageset_high;
521 int pageset_batch;
522
523 #ifndef CONFIG_SPARSEMEM
524 /*
525 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
526 * In SPARSEMEM, this map is stored in struct mem_section
527 */
528 unsigned long *pageblock_flags;
529 #endif /* CONFIG_SPARSEMEM */
530
531 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
532 unsigned long zone_start_pfn;
533
534 /*
535 * spanned_pages is the total pages spanned by the zone, including
536 * holes, which is calculated as:
537 * spanned_pages = zone_end_pfn - zone_start_pfn;
538 *
539 * present_pages is physical pages existing within the zone, which
540 * is calculated as:
541 * present_pages = spanned_pages - absent_pages(pages in holes);
542 *
543 * present_early_pages is present pages existing within the zone
544 * located on memory available since early boot, excluding hotplugged
545 * memory.
546 *
547 * managed_pages is present pages managed by the buddy system, which
548 * is calculated as (reserved_pages includes pages allocated by the
549 * bootmem allocator):
550 * managed_pages = present_pages - reserved_pages;
551 *
552 * cma pages is present pages that are assigned for CMA use
553 * (MIGRATE_CMA).
554 *
555 * So present_pages may be used by memory hotplug or memory power
556 * management logic to figure out unmanaged pages by checking
557 * (present_pages - managed_pages). And managed_pages should be used
558 * by page allocator and vm scanner to calculate all kinds of watermarks
559 * and thresholds.
560 *
561 * Locking rules:
562 *
563 * zone_start_pfn and spanned_pages are protected by span_seqlock.
564 * It is a seqlock because it has to be read outside of zone->lock,
565 * and it is done in the main allocator path. But, it is written
566 * quite infrequently.
567 *
568 * The span_seq lock is declared along with zone->lock because it is
569 * frequently read in proximity to zone->lock. It's good to
570 * give them a chance of being in the same cacheline.
571 *
572 * Write access to present_pages at runtime should be protected by
573 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
574 * present_pages should get_online_mems() to get a stable value.
575 */
576 atomic_long_t managed_pages;
577 unsigned long spanned_pages;
578 unsigned long present_pages;
579 #if defined(CONFIG_MEMORY_HOTPLUG)
580 unsigned long present_early_pages;
581 #endif
582 #ifdef CONFIG_CMA
583 unsigned long cma_pages;
584 #endif
585
586 const char *name;
587
588 #ifdef CONFIG_MEMORY_ISOLATION
589 /*
590 * Number of isolated pageblock. It is used to solve incorrect
591 * freepage counting problem due to racy retrieving migratetype
592 * of pageblock. Protected by zone->lock.
593 */
594 unsigned long nr_isolate_pageblock;
595 #endif
596
597 #ifdef CONFIG_MEMORY_HOTPLUG
598 /* see spanned/present_pages for more description */
599 seqlock_t span_seqlock;
600 #endif
601
602 int initialized;
603
604 /* Write-intensive fields used from the page allocator */
605 ZONE_PADDING(_pad1_)
606
607 /* free areas of different sizes */
608 struct free_area free_area[MAX_ORDER];
609
610 /* zone flags, see below */
611 unsigned long flags;
612
613 /* Primarily protects free_area */
614 spinlock_t lock;
615
616 /* Write-intensive fields used by compaction and vmstats. */
617 ZONE_PADDING(_pad2_)
618
619 /*
620 * When free pages are below this point, additional steps are taken
621 * when reading the number of free pages to avoid per-cpu counter
622 * drift allowing watermarks to be breached
623 */
624 unsigned long percpu_drift_mark;
625
626 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
627 /* pfn where compaction free scanner should start */
628 unsigned long compact_cached_free_pfn;
629 /* pfn where compaction migration scanner should start */
630 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
631 unsigned long compact_init_migrate_pfn;
632 unsigned long compact_init_free_pfn;
633 #endif
634
635 #ifdef CONFIG_COMPACTION
636 /*
637 * On compaction failure, 1<<compact_defer_shift compactions
638 * are skipped before trying again. The number attempted since
639 * last failure is tracked with compact_considered.
640 * compact_order_failed is the minimum compaction failed order.
641 */
642 unsigned int compact_considered;
643 unsigned int compact_defer_shift;
644 int compact_order_failed;
645 #endif
646
647 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
648 /* Set to true when the PG_migrate_skip bits should be cleared */
649 bool compact_blockskip_flush;
650 #endif
651
652 bool contiguous;
653
654 ZONE_PADDING(_pad3_)
655 /* Zone statistics */
656 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
657 atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
658 } ____cacheline_internodealigned_in_smp;
659
660 enum pgdat_flags {
661 PGDAT_DIRTY, /* reclaim scanning has recently found
662 * many dirty file pages at the tail
663 * of the LRU.
664 */
665 PGDAT_WRITEBACK, /* reclaim scanning has recently found
666 * many pages under writeback
667 */
668 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
669 };
670
671 enum zone_flags {
672 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
673 * Cleared when kswapd is woken.
674 */
675 ZONE_RECLAIM_ACTIVE, /* kswapd may be scanning the zone. */
676 };
677
zone_managed_pages(struct zone * zone)678 static inline unsigned long zone_managed_pages(struct zone *zone)
679 {
680 return (unsigned long)atomic_long_read(&zone->managed_pages);
681 }
682
zone_cma_pages(struct zone * zone)683 static inline unsigned long zone_cma_pages(struct zone *zone)
684 {
685 #ifdef CONFIG_CMA
686 return zone->cma_pages;
687 #else
688 return 0;
689 #endif
690 }
691
zone_end_pfn(const struct zone * zone)692 static inline unsigned long zone_end_pfn(const struct zone *zone)
693 {
694 return zone->zone_start_pfn + zone->spanned_pages;
695 }
696
zone_spans_pfn(const struct zone * zone,unsigned long pfn)697 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
698 {
699 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
700 }
701
zone_is_initialized(struct zone * zone)702 static inline bool zone_is_initialized(struct zone *zone)
703 {
704 return zone->initialized;
705 }
706
zone_is_empty(struct zone * zone)707 static inline bool zone_is_empty(struct zone *zone)
708 {
709 return zone->spanned_pages == 0;
710 }
711
712 /*
713 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
714 * intersection with the given zone
715 */
zone_intersects(struct zone * zone,unsigned long start_pfn,unsigned long nr_pages)716 static inline bool zone_intersects(struct zone *zone,
717 unsigned long start_pfn, unsigned long nr_pages)
718 {
719 if (zone_is_empty(zone))
720 return false;
721 if (start_pfn >= zone_end_pfn(zone) ||
722 start_pfn + nr_pages <= zone->zone_start_pfn)
723 return false;
724
725 return true;
726 }
727
728 /*
729 * The "priority" of VM scanning is how much of the queues we will scan in one
730 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
731 * queues ("queue_length >> 12") during an aging round.
732 */
733 #define DEF_PRIORITY 12
734
735 /* Maximum number of zones on a zonelist */
736 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
737
738 enum {
739 ZONELIST_FALLBACK, /* zonelist with fallback */
740 #ifdef CONFIG_NUMA
741 /*
742 * The NUMA zonelists are doubled because we need zonelists that
743 * restrict the allocations to a single node for __GFP_THISNODE.
744 */
745 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
746 #endif
747 MAX_ZONELISTS
748 };
749
750 /*
751 * This struct contains information about a zone in a zonelist. It is stored
752 * here to avoid dereferences into large structures and lookups of tables
753 */
754 struct zoneref {
755 struct zone *zone; /* Pointer to actual zone */
756 int zone_idx; /* zone_idx(zoneref->zone) */
757 };
758
759 /*
760 * One allocation request operates on a zonelist. A zonelist
761 * is a list of zones, the first one is the 'goal' of the
762 * allocation, the other zones are fallback zones, in decreasing
763 * priority.
764 *
765 * To speed the reading of the zonelist, the zonerefs contain the zone index
766 * of the entry being read. Helper functions to access information given
767 * a struct zoneref are
768 *
769 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
770 * zonelist_zone_idx() - Return the index of the zone for an entry
771 * zonelist_node_idx() - Return the index of the node for an entry
772 */
773 struct zonelist {
774 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
775 };
776
777 /*
778 * The array of struct pages for flatmem.
779 * It must be declared for SPARSEMEM as well because there are configurations
780 * that rely on that.
781 */
782 extern struct page *mem_map;
783
784 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
785 struct deferred_split {
786 spinlock_t split_queue_lock;
787 struct list_head split_queue;
788 unsigned long split_queue_len;
789 };
790 #endif
791
792 /*
793 * On NUMA machines, each NUMA node would have a pg_data_t to describe
794 * it's memory layout. On UMA machines there is a single pglist_data which
795 * describes the whole memory.
796 *
797 * Memory statistics and page replacement data structures are maintained on a
798 * per-zone basis.
799 */
800 typedef struct pglist_data {
801 /*
802 * node_zones contains just the zones for THIS node. Not all of the
803 * zones may be populated, but it is the full list. It is referenced by
804 * this node's node_zonelists as well as other node's node_zonelists.
805 */
806 struct zone node_zones[MAX_NR_ZONES];
807
808 /*
809 * node_zonelists contains references to all zones in all nodes.
810 * Generally the first zones will be references to this node's
811 * node_zones.
812 */
813 struct zonelist node_zonelists[MAX_ZONELISTS];
814
815 int nr_zones; /* number of populated zones in this node */
816 #ifdef CONFIG_FLATMEM /* means !SPARSEMEM */
817 struct page *node_mem_map;
818 #ifdef CONFIG_PAGE_EXTENSION
819 struct page_ext *node_page_ext;
820 #endif
821 #endif
822 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
823 /*
824 * Must be held any time you expect node_start_pfn,
825 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
826 * Also synchronizes pgdat->first_deferred_pfn during deferred page
827 * init.
828 *
829 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
830 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
831 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
832 *
833 * Nests above zone->lock and zone->span_seqlock
834 */
835 spinlock_t node_size_lock;
836 #endif
837 unsigned long node_start_pfn;
838 unsigned long node_present_pages; /* total number of physical pages */
839 unsigned long node_spanned_pages; /* total size of physical page
840 range, including holes */
841 int node_id;
842 wait_queue_head_t kswapd_wait;
843 wait_queue_head_t pfmemalloc_wait;
844 struct task_struct *kswapd; /* Protected by
845 mem_hotplug_begin/end() */
846 int kswapd_order;
847 enum zone_type kswapd_highest_zoneidx;
848
849 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
850
851 #ifdef CONFIG_COMPACTION
852 int kcompactd_max_order;
853 enum zone_type kcompactd_highest_zoneidx;
854 wait_queue_head_t kcompactd_wait;
855 struct task_struct *kcompactd;
856 bool proactive_compact_trigger;
857 #endif
858 /*
859 * This is a per-node reserve of pages that are not available
860 * to userspace allocations.
861 */
862 unsigned long totalreserve_pages;
863
864 #ifdef CONFIG_NUMA
865 /*
866 * node reclaim becomes active if more unmapped pages exist.
867 */
868 unsigned long min_unmapped_pages;
869 unsigned long min_slab_pages;
870 #endif /* CONFIG_NUMA */
871
872 /* Write-intensive fields used by page reclaim */
873 ZONE_PADDING(_pad1_)
874
875 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
876 /*
877 * If memory initialisation on large machines is deferred then this
878 * is the first PFN that needs to be initialised.
879 */
880 unsigned long first_deferred_pfn;
881 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
882
883 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
884 struct deferred_split deferred_split_queue;
885 #endif
886
887 /* Fields commonly accessed by the page reclaim scanner */
888
889 /*
890 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
891 *
892 * Use mem_cgroup_lruvec() to look up lruvecs.
893 */
894 struct lruvec __lruvec;
895
896 unsigned long flags;
897
898 ZONE_PADDING(_pad2_)
899
900 /* Per-node vmstats */
901 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
902 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
903 } pg_data_t;
904
905 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
906 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
907 #ifdef CONFIG_FLATMEM
908 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
909 #else
910 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
911 #endif
912 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
913
914 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
915 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
916
pgdat_end_pfn(pg_data_t * pgdat)917 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
918 {
919 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
920 }
921
pgdat_is_empty(pg_data_t * pgdat)922 static inline bool pgdat_is_empty(pg_data_t *pgdat)
923 {
924 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
925 }
926
927 #include <linux/memory_hotplug.h>
928
929 void build_all_zonelists(pg_data_t *pgdat);
930 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
931 enum zone_type highest_zoneidx);
932 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
933 int highest_zoneidx, unsigned int alloc_flags,
934 long free_pages);
935 bool zone_watermark_ok(struct zone *z, unsigned int order,
936 unsigned long mark, int highest_zoneidx,
937 unsigned int alloc_flags);
938 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
939 unsigned long mark, int highest_zoneidx);
940 /*
941 * Memory initialization context, use to differentiate memory added by
942 * the platform statically or via memory hotplug interface.
943 */
944 enum meminit_context {
945 MEMINIT_EARLY,
946 MEMINIT_HOTPLUG,
947 };
948
949 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
950 unsigned long size);
951
952 extern void lruvec_init(struct lruvec *lruvec);
953
lruvec_pgdat(struct lruvec * lruvec)954 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
955 {
956 #ifdef CONFIG_MEMCG
957 return lruvec->pgdat;
958 #else
959 return container_of(lruvec, struct pglist_data, __lruvec);
960 #endif
961 }
962
963 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
964 int local_memory_node(int node_id);
965 #else
local_memory_node(int node_id)966 static inline int local_memory_node(int node_id) { return node_id; };
967 #endif
968
969 /*
970 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
971 */
972 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
973
974 #ifdef CONFIG_ZONE_DEVICE
zone_is_zone_device(struct zone * zone)975 static inline bool zone_is_zone_device(struct zone *zone)
976 {
977 return zone_idx(zone) == ZONE_DEVICE;
978 }
979 #else
zone_is_zone_device(struct zone * zone)980 static inline bool zone_is_zone_device(struct zone *zone)
981 {
982 return false;
983 }
984 #endif
985
986 /*
987 * Returns true if a zone has pages managed by the buddy allocator.
988 * All the reclaim decisions have to use this function rather than
989 * populated_zone(). If the whole zone is reserved then we can easily
990 * end up with populated_zone() && !managed_zone().
991 */
managed_zone(struct zone * zone)992 static inline bool managed_zone(struct zone *zone)
993 {
994 return zone_managed_pages(zone);
995 }
996
997 /* Returns true if a zone has memory */
populated_zone(struct zone * zone)998 static inline bool populated_zone(struct zone *zone)
999 {
1000 return zone->present_pages;
1001 }
1002
1003 #ifdef CONFIG_NUMA
zone_to_nid(struct zone * zone)1004 static inline int zone_to_nid(struct zone *zone)
1005 {
1006 return zone->node;
1007 }
1008
zone_set_nid(struct zone * zone,int nid)1009 static inline void zone_set_nid(struct zone *zone, int nid)
1010 {
1011 zone->node = nid;
1012 }
1013 #else
zone_to_nid(struct zone * zone)1014 static inline int zone_to_nid(struct zone *zone)
1015 {
1016 return 0;
1017 }
1018
zone_set_nid(struct zone * zone,int nid)1019 static inline void zone_set_nid(struct zone *zone, int nid) {}
1020 #endif
1021
1022 extern int movable_zone;
1023
is_highmem_idx(enum zone_type idx)1024 static inline int is_highmem_idx(enum zone_type idx)
1025 {
1026 #ifdef CONFIG_HIGHMEM
1027 return (idx == ZONE_HIGHMEM ||
1028 (idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM));
1029 #else
1030 return 0;
1031 #endif
1032 }
1033
1034 /**
1035 * is_highmem - helper function to quickly check if a struct zone is a
1036 * highmem zone or not. This is an attempt to keep references
1037 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1038 * @zone: pointer to struct zone variable
1039 * Return: 1 for a highmem zone, 0 otherwise
1040 */
is_highmem(struct zone * zone)1041 static inline int is_highmem(struct zone *zone)
1042 {
1043 #ifdef CONFIG_HIGHMEM
1044 return is_highmem_idx(zone_idx(zone));
1045 #else
1046 return 0;
1047 #endif
1048 }
1049
1050 /* These two functions are used to setup the per zone pages min values */
1051 struct ctl_table;
1052
1053 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, void *, size_t *,
1054 loff_t *);
1055 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, void *,
1056 size_t *, loff_t *);
1057 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
1058 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, void *,
1059 size_t *, loff_t *);
1060 int percpu_pagelist_high_fraction_sysctl_handler(struct ctl_table *, int,
1061 void *, size_t *, loff_t *);
1062 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
1063 void *, size_t *, loff_t *);
1064 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
1065 void *, size_t *, loff_t *);
1066 int numa_zonelist_order_handler(struct ctl_table *, int,
1067 void *, size_t *, loff_t *);
1068 extern int percpu_pagelist_high_fraction;
1069 extern char numa_zonelist_order[];
1070 #define NUMA_ZONELIST_ORDER_LEN 16
1071
1072 #ifndef CONFIG_NUMA
1073
1074 extern struct pglist_data contig_page_data;
NODE_DATA(int nid)1075 static inline struct pglist_data *NODE_DATA(int nid)
1076 {
1077 return &contig_page_data;
1078 }
1079 #define NODE_MEM_MAP(nid) mem_map
1080
1081 #else /* CONFIG_NUMA */
1082
1083 #include <asm/mmzone.h>
1084
1085 #endif /* !CONFIG_NUMA */
1086
1087 extern struct pglist_data *first_online_pgdat(void);
1088 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1089 extern struct zone *next_zone(struct zone *zone);
1090
1091 /**
1092 * for_each_online_pgdat - helper macro to iterate over all online nodes
1093 * @pgdat: pointer to a pg_data_t variable
1094 */
1095 #define for_each_online_pgdat(pgdat) \
1096 for (pgdat = first_online_pgdat(); \
1097 pgdat; \
1098 pgdat = next_online_pgdat(pgdat))
1099 /**
1100 * for_each_zone - helper macro to iterate over all memory zones
1101 * @zone: pointer to struct zone variable
1102 *
1103 * The user only needs to declare the zone variable, for_each_zone
1104 * fills it in.
1105 */
1106 #define for_each_zone(zone) \
1107 for (zone = (first_online_pgdat())->node_zones; \
1108 zone; \
1109 zone = next_zone(zone))
1110
1111 #define for_each_populated_zone(zone) \
1112 for (zone = (first_online_pgdat())->node_zones; \
1113 zone; \
1114 zone = next_zone(zone)) \
1115 if (!populated_zone(zone)) \
1116 ; /* do nothing */ \
1117 else
1118
zonelist_zone(struct zoneref * zoneref)1119 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1120 {
1121 return zoneref->zone;
1122 }
1123
zonelist_zone_idx(struct zoneref * zoneref)1124 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1125 {
1126 return zoneref->zone_idx;
1127 }
1128
zonelist_node_idx(struct zoneref * zoneref)1129 static inline int zonelist_node_idx(struct zoneref *zoneref)
1130 {
1131 return zone_to_nid(zoneref->zone);
1132 }
1133
1134 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1135 enum zone_type highest_zoneidx,
1136 nodemask_t *nodes);
1137
1138 /**
1139 * 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
1140 * @z: The cursor used as a starting point for the search
1141 * @highest_zoneidx: The zone index of the highest zone to return
1142 * @nodes: An optional nodemask to filter the zonelist with
1143 *
1144 * This function returns the next zone at or below a given zone index that is
1145 * within the allowed nodemask using a cursor as the starting point for the
1146 * search. The zoneref returned is a cursor that represents the current zone
1147 * being examined. It should be advanced by one before calling
1148 * next_zones_zonelist again.
1149 *
1150 * Return: the next zone at or below highest_zoneidx within the allowed
1151 * nodemask using a cursor within a zonelist as a starting point
1152 */
next_zones_zonelist(struct zoneref * z,enum zone_type highest_zoneidx,nodemask_t * nodes)1153 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1154 enum zone_type highest_zoneidx,
1155 nodemask_t *nodes)
1156 {
1157 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1158 return z;
1159 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1160 }
1161
1162 /**
1163 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1164 * @zonelist: The zonelist to search for a suitable zone
1165 * @highest_zoneidx: The zone index of the highest zone to return
1166 * @nodes: An optional nodemask to filter the zonelist with
1167 *
1168 * This function returns the first zone at or below a given zone index that is
1169 * within the allowed nodemask. The zoneref returned is a cursor that can be
1170 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1171 * one before calling.
1172 *
1173 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1174 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1175 * update due to cpuset modification.
1176 *
1177 * Return: Zoneref pointer for the first suitable zone found
1178 */
first_zones_zonelist(struct zonelist * zonelist,enum zone_type highest_zoneidx,nodemask_t * nodes)1179 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1180 enum zone_type highest_zoneidx,
1181 nodemask_t *nodes)
1182 {
1183 return next_zones_zonelist(zonelist->_zonerefs,
1184 highest_zoneidx, nodes);
1185 }
1186
1187 /**
1188 * 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
1189 * @zone: The current zone in the iterator
1190 * @z: The current pointer within zonelist->_zonerefs being iterated
1191 * @zlist: The zonelist being iterated
1192 * @highidx: The zone index of the highest zone to return
1193 * @nodemask: Nodemask allowed by the allocator
1194 *
1195 * This iterator iterates though all zones at or below a given zone index and
1196 * within a given nodemask
1197 */
1198 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1199 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1200 zone; \
1201 z = next_zones_zonelist(++z, highidx, nodemask), \
1202 zone = zonelist_zone(z))
1203
1204 #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1205 for (zone = z->zone; \
1206 zone; \
1207 z = next_zones_zonelist(++z, highidx, nodemask), \
1208 zone = zonelist_zone(z))
1209
1210
1211 /**
1212 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1213 * @zone: The current zone in the iterator
1214 * @z: The current pointer within zonelist->zones being iterated
1215 * @zlist: The zonelist being iterated
1216 * @highidx: The zone index of the highest zone to return
1217 *
1218 * This iterator iterates though all zones at or below a given zone index.
1219 */
1220 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1221 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1222
1223 #ifdef CONFIG_SPARSEMEM
1224 #include <asm/sparsemem.h>
1225 #endif
1226
1227 #ifdef CONFIG_FLATMEM
1228 #define pfn_to_nid(pfn) (0)
1229 #endif
1230
1231 #ifdef CONFIG_SPARSEMEM
1232
1233 /*
1234 * PA_SECTION_SHIFT physical address to/from section number
1235 * PFN_SECTION_SHIFT pfn to/from section number
1236 */
1237 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1238 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1239
1240 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1241
1242 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1243 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1244
1245 #define SECTION_BLOCKFLAGS_BITS \
1246 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1247
1248 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1249 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1250 #endif
1251
pfn_to_section_nr(unsigned long pfn)1252 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1253 {
1254 return pfn >> PFN_SECTION_SHIFT;
1255 }
section_nr_to_pfn(unsigned long sec)1256 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1257 {
1258 return sec << PFN_SECTION_SHIFT;
1259 }
1260
1261 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1262 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1263
1264 #define SUBSECTION_SHIFT 21
1265 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1266
1267 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1268 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1269 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1270
1271 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1272 #error Subsection size exceeds section size
1273 #else
1274 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1275 #endif
1276
1277 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1278 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1279
1280 struct mem_section_usage {
1281 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1282 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1283 #endif
1284 /* See declaration of similar field in struct zone */
1285 unsigned long pageblock_flags[0];
1286 };
1287
1288 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1289
1290 struct page;
1291 struct page_ext;
1292 struct mem_section {
1293 /*
1294 * This is, logically, a pointer to an array of struct
1295 * pages. However, it is stored with some other magic.
1296 * (see sparse.c::sparse_init_one_section())
1297 *
1298 * Additionally during early boot we encode node id of
1299 * the location of the section here to guide allocation.
1300 * (see sparse.c::memory_present())
1301 *
1302 * Making it a UL at least makes someone do a cast
1303 * before using it wrong.
1304 */
1305 unsigned long section_mem_map;
1306
1307 struct mem_section_usage *usage;
1308 #ifdef CONFIG_PAGE_EXTENSION
1309 /*
1310 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1311 * section. (see page_ext.h about this.)
1312 */
1313 struct page_ext *page_ext;
1314 unsigned long pad;
1315 #endif
1316 /*
1317 * WARNING: mem_section must be a power-of-2 in size for the
1318 * calculation and use of SECTION_ROOT_MASK to make sense.
1319 */
1320 };
1321
1322 #ifdef CONFIG_SPARSEMEM_EXTREME
1323 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1324 #else
1325 #define SECTIONS_PER_ROOT 1
1326 #endif
1327
1328 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1329 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1330 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1331
1332 #ifdef CONFIG_SPARSEMEM_EXTREME
1333 extern struct mem_section **mem_section;
1334 #else
1335 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1336 #endif
1337
section_to_usemap(struct mem_section * ms)1338 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1339 {
1340 return ms->usage->pageblock_flags;
1341 }
1342
__nr_to_section(unsigned long nr)1343 static inline struct mem_section *__nr_to_section(unsigned long nr)
1344 {
1345 #ifdef CONFIG_SPARSEMEM_EXTREME
1346 if (!mem_section)
1347 return NULL;
1348 #endif
1349 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1350 return NULL;
1351 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1352 }
1353 extern size_t mem_section_usage_size(void);
1354
1355 /*
1356 * We use the lower bits of the mem_map pointer to store
1357 * a little bit of information. The pointer is calculated
1358 * as mem_map - section_nr_to_pfn(pnum). The result is
1359 * aligned to the minimum alignment of the two values:
1360 * 1. All mem_map arrays are page-aligned.
1361 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1362 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1363 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1364 * worst combination is powerpc with 256k pages,
1365 * which results in PFN_SECTION_SHIFT equal 6.
1366 * To sum it up, at least 6 bits are available.
1367 */
1368 #define SECTION_MARKED_PRESENT (1UL<<0)
1369 #define SECTION_HAS_MEM_MAP (1UL<<1)
1370 #define SECTION_IS_ONLINE (1UL<<2)
1371 #define SECTION_IS_EARLY (1UL<<3)
1372 #define SECTION_TAINT_ZONE_DEVICE (1UL<<4)
1373 #define SECTION_MAP_LAST_BIT (1UL<<5)
1374 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1375 #define SECTION_NID_SHIFT 6
1376
__section_mem_map_addr(struct mem_section * section)1377 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1378 {
1379 unsigned long map = section->section_mem_map;
1380 map &= SECTION_MAP_MASK;
1381 return (struct page *)map;
1382 }
1383
present_section(struct mem_section * section)1384 static inline int present_section(struct mem_section *section)
1385 {
1386 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1387 }
1388
present_section_nr(unsigned long nr)1389 static inline int present_section_nr(unsigned long nr)
1390 {
1391 return present_section(__nr_to_section(nr));
1392 }
1393
valid_section(struct mem_section * section)1394 static inline int valid_section(struct mem_section *section)
1395 {
1396 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1397 }
1398
early_section(struct mem_section * section)1399 static inline int early_section(struct mem_section *section)
1400 {
1401 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1402 }
1403
valid_section_nr(unsigned long nr)1404 static inline int valid_section_nr(unsigned long nr)
1405 {
1406 return valid_section(__nr_to_section(nr));
1407 }
1408
online_section(struct mem_section * section)1409 static inline int online_section(struct mem_section *section)
1410 {
1411 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1412 }
1413
online_device_section(struct mem_section * section)1414 static inline int online_device_section(struct mem_section *section)
1415 {
1416 unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE;
1417
1418 return section && ((section->section_mem_map & flags) == flags);
1419 }
1420
online_section_nr(unsigned long nr)1421 static inline int online_section_nr(unsigned long nr)
1422 {
1423 return online_section(__nr_to_section(nr));
1424 }
1425
1426 #ifdef CONFIG_MEMORY_HOTPLUG
1427 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1428 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1429 #endif
1430
__pfn_to_section(unsigned long pfn)1431 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1432 {
1433 return __nr_to_section(pfn_to_section_nr(pfn));
1434 }
1435
1436 extern unsigned long __highest_present_section_nr;
1437
subsection_map_index(unsigned long pfn)1438 static inline int subsection_map_index(unsigned long pfn)
1439 {
1440 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1441 }
1442
1443 #ifdef CONFIG_SPARSEMEM_VMEMMAP
pfn_section_valid(struct mem_section * ms,unsigned long pfn)1444 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1445 {
1446 int idx = subsection_map_index(pfn);
1447
1448 return test_bit(idx, ms->usage->subsection_map);
1449 }
1450 #else
pfn_section_valid(struct mem_section * ms,unsigned long pfn)1451 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1452 {
1453 return 1;
1454 }
1455 #endif
1456
1457 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1458 /**
1459 * pfn_valid - check if there is a valid memory map entry for a PFN
1460 * @pfn: the page frame number to check
1461 *
1462 * Check if there is a valid memory map entry aka struct page for the @pfn.
1463 * Note, that availability of the memory map entry does not imply that
1464 * there is actual usable memory at that @pfn. The struct page may
1465 * represent a hole or an unusable page frame.
1466 *
1467 * Return: 1 for PFNs that have memory map entries and 0 otherwise
1468 */
pfn_valid(unsigned long pfn)1469 static inline int pfn_valid(unsigned long pfn)
1470 {
1471 struct mem_section *ms;
1472
1473 /*
1474 * Ensure the upper PAGE_SHIFT bits are clear in the
1475 * pfn. Else it might lead to false positives when
1476 * some of the upper bits are set, but the lower bits
1477 * match a valid pfn.
1478 */
1479 if (PHYS_PFN(PFN_PHYS(pfn)) != pfn)
1480 return 0;
1481
1482 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1483 return 0;
1484 ms = __nr_to_section(pfn_to_section_nr(pfn));
1485 if (!valid_section(ms))
1486 return 0;
1487 /*
1488 * Traditionally early sections always returned pfn_valid() for
1489 * the entire section-sized span.
1490 */
1491 return early_section(ms) || pfn_section_valid(ms, pfn);
1492 }
1493 #endif
1494
pfn_in_present_section(unsigned long pfn)1495 static inline int pfn_in_present_section(unsigned long pfn)
1496 {
1497 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1498 return 0;
1499 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1500 }
1501
next_present_section_nr(unsigned long section_nr)1502 static inline unsigned long next_present_section_nr(unsigned long section_nr)
1503 {
1504 while (++section_nr <= __highest_present_section_nr) {
1505 if (present_section_nr(section_nr))
1506 return section_nr;
1507 }
1508
1509 return -1;
1510 }
1511
1512 /*
1513 * These are _only_ used during initialisation, therefore they
1514 * can use __initdata ... They could have names to indicate
1515 * this restriction.
1516 */
1517 #ifdef CONFIG_NUMA
1518 #define pfn_to_nid(pfn) \
1519 ({ \
1520 unsigned long __pfn_to_nid_pfn = (pfn); \
1521 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1522 })
1523 #else
1524 #define pfn_to_nid(pfn) (0)
1525 #endif
1526
1527 void sparse_init(void);
1528 #else
1529 #define sparse_init() do {} while (0)
1530 #define sparse_index_init(_sec, _nid) do {} while (0)
1531 #define pfn_in_present_section pfn_valid
1532 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
1533 #endif /* CONFIG_SPARSEMEM */
1534
1535 #endif /* !__GENERATING_BOUNDS.H */
1536 #endif /* !__ASSEMBLY__ */
1537 #endif /* _LINUX_MMZONE_H */
1538
