]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blame - include/linux/mmzone.h
Merge tag 'char-misc-4.20-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/gregk...
[mirror_ubuntu-eoan-kernel.git] / include / linux / mmzone.h
CommitLineData
b2441318 1/* SPDX-License-Identifier: GPL-2.0 */
1da177e4
LT
2#ifndef _LINUX_MMZONE_H
3#define _LINUX_MMZONE_H
4
1da177e4 5#ifndef __ASSEMBLY__
97965478 6#ifndef __GENERATING_BOUNDS_H
1da177e4 7
1da177e4
LT
8#include <linux/spinlock.h>
9#include <linux/list.h>
10#include <linux/wait.h>
e815af95 11#include <linux/bitops.h>
1da177e4
LT
12#include <linux/cache.h>
13#include <linux/threads.h>
14#include <linux/numa.h>
15#include <linux/init.h>
bdc8cb98 16#include <linux/seqlock.h>
8357f869 17#include <linux/nodemask.h>
835c134e 18#include <linux/pageblock-flags.h>
bbeae5b0 19#include <linux/page-flags-layout.h>
60063497 20#include <linux/atomic.h>
93ff66bf 21#include <asm/page.h>
1da177e4
LT
22
23/* Free memory management - zoned buddy allocator. */
24#ifndef CONFIG_FORCE_MAX_ZONEORDER
25#define MAX_ORDER 11
26#else
27#define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
28#endif
e984bb43 29#define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
1da177e4 30
5ad333eb
AW
31/*
32 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
33 * costly to service. That is between allocation orders which should
35fca53e 34 * coalesce naturally under reasonable reclaim pressure and those which
5ad333eb
AW
35 * will not.
36 */
37#define PAGE_ALLOC_COSTLY_ORDER 3
38
a6ffdc07 39enum migratetype {
47118af0 40 MIGRATE_UNMOVABLE,
47118af0 41 MIGRATE_MOVABLE,
016c13da 42 MIGRATE_RECLAIMABLE,
0aaa29a5
MG
43 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
44 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
47118af0
MN
45#ifdef CONFIG_CMA
46 /*
47 * MIGRATE_CMA migration type is designed to mimic the way
48 * ZONE_MOVABLE works. Only movable pages can be allocated
49 * from MIGRATE_CMA pageblocks and page allocator never
50 * implicitly change migration type of MIGRATE_CMA pageblock.
51 *
52 * The way to use it is to change migratetype of a range of
53 * pageblocks to MIGRATE_CMA which can be done by
54 * __free_pageblock_cma() function. What is important though
55 * is that a range of pageblocks must be aligned to
56 * MAX_ORDER_NR_PAGES should biggest page be bigger then
57 * a single pageblock.
58 */
59 MIGRATE_CMA,
60#endif
194159fb 61#ifdef CONFIG_MEMORY_ISOLATION
47118af0 62 MIGRATE_ISOLATE, /* can't allocate from here */
194159fb 63#endif
47118af0
MN
64 MIGRATE_TYPES
65};
66
60f30350
VB
67/* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
68extern char * const migratetype_names[MIGRATE_TYPES];
69
47118af0
MN
70#ifdef CONFIG_CMA
71# define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
7c15d9bb 72# define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
47118af0
MN
73#else
74# define is_migrate_cma(migratetype) false
7c15d9bb 75# define is_migrate_cma_page(_page) false
47118af0 76#endif
b2a0ac88 77
b682debd
VB
78static inline bool is_migrate_movable(int mt)
79{
80 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
81}
82
b2a0ac88
MG
83#define for_each_migratetype_order(order, type) \
84 for (order = 0; order < MAX_ORDER; order++) \
85 for (type = 0; type < MIGRATE_TYPES; type++)
86
467c996c
MG
87extern int page_group_by_mobility_disabled;
88
e58469ba
MG
89#define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1)
90#define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1)
91
dc4b0caf
MG
92#define get_pageblock_migratetype(page) \
93 get_pfnblock_flags_mask(page, page_to_pfn(page), \
94 PB_migrate_end, MIGRATETYPE_MASK)
95
1da177e4 96struct free_area {
b2a0ac88 97 struct list_head free_list[MIGRATE_TYPES];
1da177e4
LT
98 unsigned long nr_free;
99};
100
101struct pglist_data;
102
103/*
a52633d8 104 * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
1da177e4
LT
105 * So add a wild amount of padding here to ensure that they fall into separate
106 * cachelines. There are very few zone structures in the machine, so space
107 * consumption is not a concern here.
108 */
109#if defined(CONFIG_SMP)
110struct zone_padding {
111 char x[0];
22fc6ecc 112} ____cacheline_internodealigned_in_smp;
1da177e4
LT
113#define ZONE_PADDING(name) struct zone_padding name;
114#else
115#define ZONE_PADDING(name)
116#endif
117
3a321d2a
KW
118#ifdef CONFIG_NUMA
119enum numa_stat_item {
120 NUMA_HIT, /* allocated in intended node */
121 NUMA_MISS, /* allocated in non intended node */
122 NUMA_FOREIGN, /* was intended here, hit elsewhere */
123 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
124 NUMA_LOCAL, /* allocation from local node */
125 NUMA_OTHER, /* allocation from other node */
126 NR_VM_NUMA_STAT_ITEMS
127};
128#else
129#define NR_VM_NUMA_STAT_ITEMS 0
130#endif
131
2244b95a 132enum zone_stat_item {
51ed4491 133 /* First 128 byte cacheline (assuming 64 bit words) */
d23ad423 134 NR_FREE_PAGES,
71c799f4
MK
135 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
136 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
137 NR_ZONE_ACTIVE_ANON,
138 NR_ZONE_INACTIVE_FILE,
139 NR_ZONE_ACTIVE_FILE,
140 NR_ZONE_UNEVICTABLE,
5a1c84b4 141 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
5344b7e6 142 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
51ed4491 143 NR_PAGETABLE, /* used for pagetables */
d30dd8be 144 NR_KERNEL_STACK_KB, /* measured in KiB */
c6a7f572 145 /* Second 128 byte cacheline */
d2c5e30c 146 NR_BOUNCE,
91537fee
MK
147#if IS_ENABLED(CONFIG_ZSMALLOC)
148 NR_ZSPAGES, /* allocated in zsmalloc */
ca889e6c 149#endif
d1ce749a 150 NR_FREE_CMA_PAGES,
2244b95a
CL
151 NR_VM_ZONE_STAT_ITEMS };
152
75ef7184 153enum node_stat_item {
599d0c95
MG
154 NR_LRU_BASE,
155 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
156 NR_ACTIVE_ANON, /* " " " " " */
157 NR_INACTIVE_FILE, /* " " " " " */
158 NR_ACTIVE_FILE, /* " " " " " */
159 NR_UNEVICTABLE, /* " " " " " */
385386cf
JW
160 NR_SLAB_RECLAIMABLE,
161 NR_SLAB_UNRECLAIMABLE,
599d0c95
MG
162 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
163 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
68d48e6a 164 WORKINGSET_NODES,
1e6b1085
MG
165 WORKINGSET_REFAULT,
166 WORKINGSET_ACTIVATE,
1899ad18 167 WORKINGSET_RESTORE,
1e6b1085 168 WORKINGSET_NODERECLAIM,
4b9d0fab 169 NR_ANON_MAPPED, /* Mapped anonymous pages */
50658e2e
MG
170 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
171 only modified from process context */
11fb9989
MG
172 NR_FILE_PAGES,
173 NR_FILE_DIRTY,
174 NR_WRITEBACK,
175 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
176 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
177 NR_SHMEM_THPS,
178 NR_SHMEM_PMDMAPPED,
179 NR_ANON_THPS,
180 NR_UNSTABLE_NFS, /* NFS unstable pages */
c4a25635
MG
181 NR_VMSCAN_WRITE,
182 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
183 NR_DIRTIED, /* page dirtyings since bootup */
184 NR_WRITTEN, /* page writings since bootup */
b29940c1 185 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
75ef7184
MG
186 NR_VM_NODE_STAT_ITEMS
187};
188
4f98a2fe
RR
189/*
190 * We do arithmetic on the LRU lists in various places in the code,
191 * so it is important to keep the active lists LRU_ACTIVE higher in
192 * the array than the corresponding inactive lists, and to keep
193 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
194 *
195 * This has to be kept in sync with the statistics in zone_stat_item
196 * above and the descriptions in vmstat_text in mm/vmstat.c
197 */
198#define LRU_BASE 0
199#define LRU_ACTIVE 1
200#define LRU_FILE 2
201
b69408e8 202enum lru_list {
4f98a2fe
RR
203 LRU_INACTIVE_ANON = LRU_BASE,
204 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
205 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
206 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
894bc310 207 LRU_UNEVICTABLE,
894bc310
LS
208 NR_LRU_LISTS
209};
b69408e8 210
4111304d 211#define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
b69408e8 212
4111304d 213#define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
894bc310 214
4111304d 215static inline int is_file_lru(enum lru_list lru)
4f98a2fe 216{
4111304d 217 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
4f98a2fe
RR
218}
219
4111304d 220static inline int is_active_lru(enum lru_list lru)
b69408e8 221{
4111304d 222 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
b69408e8
CL
223}
224
89abfab1
HD
225struct zone_reclaim_stat {
226 /*
227 * The pageout code in vmscan.c keeps track of how many of the
59f91e5d 228 * mem/swap backed and file backed pages are referenced.
89abfab1
HD
229 * The higher the rotated/scanned ratio, the more valuable
230 * that cache is.
231 *
232 * The anon LRU stats live in [0], file LRU stats in [1]
233 */
234 unsigned long recent_rotated[2];
235 unsigned long recent_scanned[2];
236};
237
6290df54 238struct lruvec {
23047a96
JW
239 struct list_head lists[NR_LRU_LISTS];
240 struct zone_reclaim_stat reclaim_stat;
241 /* Evictions & activations on the inactive file list */
242 atomic_long_t inactive_age;
2a2e4885
JW
243 /* Refaults at the time of last reclaim cycle */
244 unsigned long refaults;
c255a458 245#ifdef CONFIG_MEMCG
599d0c95 246 struct pglist_data *pgdat;
7f5e86c2 247#endif
6290df54
JW
248};
249
bb2a0de9
KH
250/* Mask used at gathering information at once (see memcontrol.c) */
251#define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
252#define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
bb2a0de9
KH
253#define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
254
f80c0673 255/* Isolate unmapped file */
f3fd4a61 256#define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
c8244935 257/* Isolate for asynchronous migration */
f3fd4a61 258#define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
e46a2879
MK
259/* Isolate unevictable pages */
260#define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
4356f21d
MK
261
262/* LRU Isolation modes. */
9efeccac 263typedef unsigned __bitwise isolate_mode_t;
4356f21d 264
41858966
MG
265enum zone_watermarks {
266 WMARK_MIN,
267 WMARK_LOW,
268 WMARK_HIGH,
269 NR_WMARK
270};
271
272#define min_wmark_pages(z) (z->watermark[WMARK_MIN])
273#define low_wmark_pages(z) (z->watermark[WMARK_LOW])
274#define high_wmark_pages(z) (z->watermark[WMARK_HIGH])
275
1da177e4
LT
276struct per_cpu_pages {
277 int count; /* number of pages in the list */
1da177e4
LT
278 int high; /* high watermark, emptying needed */
279 int batch; /* chunk size for buddy add/remove */
5f8dcc21
MG
280
281 /* Lists of pages, one per migrate type stored on the pcp-lists */
282 struct list_head lists[MIGRATE_PCPTYPES];
1da177e4
LT
283};
284
285struct per_cpu_pageset {
3dfa5721 286 struct per_cpu_pages pcp;
4037d452
CL
287#ifdef CONFIG_NUMA
288 s8 expire;
1d90ca89 289 u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
4037d452 290#endif
2244b95a 291#ifdef CONFIG_SMP
df9ecaba 292 s8 stat_threshold;
2244b95a
CL
293 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
294#endif
99dcc3e5 295};
e7c8d5c9 296
75ef7184
MG
297struct per_cpu_nodestat {
298 s8 stat_threshold;
299 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
300};
301
97965478
CL
302#endif /* !__GENERATING_BOUNDS.H */
303
2f1b6248 304enum zone_type {
4b51d669 305#ifdef CONFIG_ZONE_DMA
2f1b6248
CL
306 /*
307 * ZONE_DMA is used when there are devices that are not able
308 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we
309 * carve out the portion of memory that is needed for these devices.
310 * The range is arch specific.
311 *
312 * Some examples
313 *
314 * Architecture Limit
315 * ---------------------------
316 * parisc, ia64, sparc <4G
317 * s390 <2G
2f1b6248
CL
318 * arm Various
319 * alpha Unlimited or 0-16MB.
320 *
321 * i386, x86_64 and multiple other arches
322 * <16M.
323 */
324 ZONE_DMA,
4b51d669 325#endif
fb0e7942 326#ifdef CONFIG_ZONE_DMA32
2f1b6248
CL
327 /*
328 * x86_64 needs two ZONE_DMAs because it supports devices that are
329 * only able to do DMA to the lower 16M but also 32 bit devices that
330 * can only do DMA areas below 4G.
331 */
332 ZONE_DMA32,
fb0e7942 333#endif
2f1b6248
CL
334 /*
335 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
336 * performed on pages in ZONE_NORMAL if the DMA devices support
337 * transfers to all addressable memory.
338 */
339 ZONE_NORMAL,
e53ef38d 340#ifdef CONFIG_HIGHMEM
2f1b6248
CL
341 /*
342 * A memory area that is only addressable by the kernel through
343 * mapping portions into its own address space. This is for example
344 * used by i386 to allow the kernel to address the memory beyond
345 * 900MB. The kernel will set up special mappings (page
346 * table entries on i386) for each page that the kernel needs to
347 * access.
348 */
349 ZONE_HIGHMEM,
e53ef38d 350#endif
2a1e274a 351 ZONE_MOVABLE,
033fbae9
DW
352#ifdef CONFIG_ZONE_DEVICE
353 ZONE_DEVICE,
354#endif
97965478 355 __MAX_NR_ZONES
033fbae9 356
2f1b6248 357};
1da177e4 358
97965478
CL
359#ifndef __GENERATING_BOUNDS_H
360
1da177e4 361struct zone {
3484b2de 362 /* Read-mostly fields */
41858966
MG
363
364 /* zone watermarks, access with *_wmark_pages(zone) macros */
365 unsigned long watermark[NR_WMARK];
366
0aaa29a5
MG
367 unsigned long nr_reserved_highatomic;
368
1da177e4 369 /*
89903327
AM
370 * We don't know if the memory that we're going to allocate will be
371 * freeable or/and it will be released eventually, so to avoid totally
372 * wasting several GB of ram we must reserve some of the lower zone
373 * memory (otherwise we risk to run OOM on the lower zones despite
374 * there being tons of freeable ram on the higher zones). This array is
375 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
376 * changes.
1da177e4 377 */
3484b2de 378 long lowmem_reserve[MAX_NR_ZONES];
ab8fabd4 379
e7c8d5c9 380#ifdef CONFIG_NUMA
d5f541ed 381 int node;
3484b2de 382#endif
3484b2de 383 struct pglist_data *zone_pgdat;
43cf38eb 384 struct per_cpu_pageset __percpu *pageset;
3484b2de 385
835c134e
MG
386#ifndef CONFIG_SPARSEMEM
387 /*
d9c23400 388 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
835c134e
MG
389 * In SPARSEMEM, this map is stored in struct mem_section
390 */
391 unsigned long *pageblock_flags;
392#endif /* CONFIG_SPARSEMEM */
393
1da177e4
LT
394 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
395 unsigned long zone_start_pfn;
396
bdc8cb98 397 /*
9feedc9d
JL
398 * spanned_pages is the total pages spanned by the zone, including
399 * holes, which is calculated as:
400 * spanned_pages = zone_end_pfn - zone_start_pfn;
bdc8cb98 401 *
9feedc9d
JL
402 * present_pages is physical pages existing within the zone, which
403 * is calculated as:
8761e31c 404 * present_pages = spanned_pages - absent_pages(pages in holes);
9feedc9d
JL
405 *
406 * managed_pages is present pages managed by the buddy system, which
407 * is calculated as (reserved_pages includes pages allocated by the
408 * bootmem allocator):
409 * managed_pages = present_pages - reserved_pages;
410 *
411 * So present_pages may be used by memory hotplug or memory power
412 * management logic to figure out unmanaged pages by checking
413 * (present_pages - managed_pages). And managed_pages should be used
414 * by page allocator and vm scanner to calculate all kinds of watermarks
415 * and thresholds.
416 *
417 * Locking rules:
418 *
419 * zone_start_pfn and spanned_pages are protected by span_seqlock.
420 * It is a seqlock because it has to be read outside of zone->lock,
421 * and it is done in the main allocator path. But, it is written
422 * quite infrequently.
423 *
424 * The span_seq lock is declared along with zone->lock because it is
bdc8cb98
DH
425 * frequently read in proximity to zone->lock. It's good to
426 * give them a chance of being in the same cacheline.
9feedc9d 427 *
c3d5f5f0 428 * Write access to present_pages at runtime should be protected by
bfc8c901
VD
429 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
430 * present_pages should get_online_mems() to get a stable value.
c3d5f5f0
JL
431 *
432 * Read access to managed_pages should be safe because it's unsigned
433 * long. Write access to zone->managed_pages and totalram_pages are
434 * protected by managed_page_count_lock at runtime. Idealy only
435 * adjust_managed_page_count() should be used instead of directly
436 * touching zone->managed_pages and totalram_pages.
bdc8cb98 437 */
3484b2de 438 unsigned long managed_pages;
9feedc9d
JL
439 unsigned long spanned_pages;
440 unsigned long present_pages;
3484b2de
MG
441
442 const char *name;
1da177e4 443
ad53f92e
JK
444#ifdef CONFIG_MEMORY_ISOLATION
445 /*
446 * Number of isolated pageblock. It is used to solve incorrect
447 * freepage counting problem due to racy retrieving migratetype
448 * of pageblock. Protected by zone->lock.
449 */
450 unsigned long nr_isolate_pageblock;
451#endif
452
3484b2de
MG
453#ifdef CONFIG_MEMORY_HOTPLUG
454 /* see spanned/present_pages for more description */
455 seqlock_t span_seqlock;
456#endif
457
9dcb8b68 458 int initialized;
3484b2de 459
0f661148 460 /* Write-intensive fields used from the page allocator */
3484b2de 461 ZONE_PADDING(_pad1_)
0f661148 462
3484b2de
MG
463 /* free areas of different sizes */
464 struct free_area free_area[MAX_ORDER];
465
466 /* zone flags, see below */
467 unsigned long flags;
468
0f661148 469 /* Primarily protects free_area */
a368ab67
MG
470 spinlock_t lock;
471
0f661148 472 /* Write-intensive fields used by compaction and vmstats. */
3484b2de
MG
473 ZONE_PADDING(_pad2_)
474
3484b2de
MG
475 /*
476 * When free pages are below this point, additional steps are taken
477 * when reading the number of free pages to avoid per-cpu counter
478 * drift allowing watermarks to be breached
479 */
480 unsigned long percpu_drift_mark;
481
482#if defined CONFIG_COMPACTION || defined CONFIG_CMA
483 /* pfn where compaction free scanner should start */
484 unsigned long compact_cached_free_pfn;
485 /* pfn where async and sync compaction migration scanner should start */
486 unsigned long compact_cached_migrate_pfn[2];
487#endif
488
489#ifdef CONFIG_COMPACTION
490 /*
491 * On compaction failure, 1<<compact_defer_shift compactions
492 * are skipped before trying again. The number attempted since
493 * last failure is tracked with compact_considered.
494 */
495 unsigned int compact_considered;
496 unsigned int compact_defer_shift;
497 int compact_order_failed;
498#endif
499
500#if defined CONFIG_COMPACTION || defined CONFIG_CMA
501 /* Set to true when the PG_migrate_skip bits should be cleared */
502 bool compact_blockskip_flush;
503#endif
504
7cf91a98
JK
505 bool contiguous;
506
3484b2de
MG
507 ZONE_PADDING(_pad3_)
508 /* Zone statistics */
509 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
3a321d2a 510 atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
22fc6ecc 511} ____cacheline_internodealigned_in_smp;
1da177e4 512
599d0c95
MG
513enum pgdat_flags {
514 PGDAT_CONGESTED, /* pgdat has many dirty pages backed by
0e093d99
MG
515 * a congested BDI
516 */
599d0c95 517 PGDAT_DIRTY, /* reclaim scanning has recently found
d43006d5
MG
518 * many dirty file pages at the tail
519 * of the LRU.
520 */
599d0c95 521 PGDAT_WRITEBACK, /* reclaim scanning has recently found
283aba9f
MG
522 * many pages under writeback
523 */
a5f5f91d 524 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
57054651 525};
e815af95 526
f9228b20 527static inline unsigned long zone_end_pfn(const struct zone *zone)
108bcc96
CS
528{
529 return zone->zone_start_pfn + zone->spanned_pages;
530}
531
532static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
533{
534 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
535}
536
2a6e3ebe
CS
537static inline bool zone_is_initialized(struct zone *zone)
538{
9dcb8b68 539 return zone->initialized;
2a6e3ebe
CS
540}
541
542static inline bool zone_is_empty(struct zone *zone)
543{
544 return zone->spanned_pages == 0;
545}
546
f1dd2cd1
MH
547/*
548 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
549 * intersection with the given zone
550 */
551static inline bool zone_intersects(struct zone *zone,
552 unsigned long start_pfn, unsigned long nr_pages)
553{
554 if (zone_is_empty(zone))
555 return false;
556 if (start_pfn >= zone_end_pfn(zone) ||
557 start_pfn + nr_pages <= zone->zone_start_pfn)
558 return false;
559
560 return true;
561}
562
1da177e4
LT
563/*
564 * The "priority" of VM scanning is how much of the queues we will scan in one
565 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
566 * queues ("queue_length >> 12") during an aging round.
567 */
568#define DEF_PRIORITY 12
569
9276b1bc
PJ
570/* Maximum number of zones on a zonelist */
571#define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
572
c00eb15a
YB
573enum {
574 ZONELIST_FALLBACK, /* zonelist with fallback */
9276b1bc 575#ifdef CONFIG_NUMA
c00eb15a
YB
576 /*
577 * The NUMA zonelists are doubled because we need zonelists that
578 * restrict the allocations to a single node for __GFP_THISNODE.
579 */
580 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
9276b1bc 581#endif
c00eb15a
YB
582 MAX_ZONELISTS
583};
9276b1bc 584
dd1a239f
MG
585/*
586 * This struct contains information about a zone in a zonelist. It is stored
587 * here to avoid dereferences into large structures and lookups of tables
588 */
589struct zoneref {
590 struct zone *zone; /* Pointer to actual zone */
591 int zone_idx; /* zone_idx(zoneref->zone) */
592};
593
1da177e4
LT
594/*
595 * One allocation request operates on a zonelist. A zonelist
596 * is a list of zones, the first one is the 'goal' of the
597 * allocation, the other zones are fallback zones, in decreasing
598 * priority.
599 *
dd1a239f
MG
600 * To speed the reading of the zonelist, the zonerefs contain the zone index
601 * of the entry being read. Helper functions to access information given
602 * a struct zoneref are
603 *
604 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
605 * zonelist_zone_idx() - Return the index of the zone for an entry
606 * zonelist_node_idx() - Return the index of the node for an entry
1da177e4
LT
607 */
608struct zonelist {
dd1a239f 609 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
1da177e4
LT
610};
611
5b99cd0e
HC
612#ifndef CONFIG_DISCONTIGMEM
613/* The array of struct pages - for discontigmem use pgdat->lmem_map */
614extern struct page *mem_map;
615#endif
616
1da177e4 617/*
1da177e4 618 * On NUMA machines, each NUMA node would have a pg_data_t to describe
618b8c20
NB
619 * it's memory layout. On UMA machines there is a single pglist_data which
620 * describes the whole memory.
1da177e4
LT
621 *
622 * Memory statistics and page replacement data structures are maintained on a
623 * per-zone basis.
624 */
625struct bootmem_data;
626typedef struct pglist_data {
627 struct zone node_zones[MAX_NR_ZONES];
523b9458 628 struct zonelist node_zonelists[MAX_ZONELISTS];
1da177e4 629 int nr_zones;
52d4b9ac 630#ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
1da177e4 631 struct page *node_mem_map;
eefa864b
JK
632#ifdef CONFIG_PAGE_EXTENSION
633 struct page_ext *node_page_ext;
634#endif
d41dee36 635#endif
3a2d7fa8 636#if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
208d54e5
DH
637 /*
638 * Must be held any time you expect node_start_pfn, node_present_pages
639 * or node_spanned_pages stay constant. Holding this will also
640 * guarantee that any pfn_valid() stays that way.
641 *
114d4b79 642 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
3a2d7fa8
PT
643 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
644 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
114d4b79 645 *
72c3b51b 646 * Nests above zone->lock and zone->span_seqlock
208d54e5
DH
647 */
648 spinlock_t node_size_lock;
649#endif
1da177e4
LT
650 unsigned long node_start_pfn;
651 unsigned long node_present_pages; /* total number of physical pages */
652 unsigned long node_spanned_pages; /* total size of physical page
653 range, including holes */
654 int node_id;
1da177e4 655 wait_queue_head_t kswapd_wait;
5515061d 656 wait_queue_head_t pfmemalloc_wait;
bfc8c901
VD
657 struct task_struct *kswapd; /* Protected by
658 mem_hotplug_begin/end() */
38087d9b
MG
659 int kswapd_order;
660 enum zone_type kswapd_classzone_idx;
661
c73322d0
JW
662 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
663
698b1b30
VB
664#ifdef CONFIG_COMPACTION
665 int kcompactd_max_order;
666 enum zone_type kcompactd_classzone_idx;
667 wait_queue_head_t kcompactd_wait;
668 struct task_struct *kcompactd;
8177a420 669#endif
281e3726
MG
670 /*
671 * This is a per-node reserve of pages that are not available
672 * to userspace allocations.
673 */
674 unsigned long totalreserve_pages;
675
a5f5f91d
MG
676#ifdef CONFIG_NUMA
677 /*
678 * zone reclaim becomes active if more unmapped pages exist.
679 */
680 unsigned long min_unmapped_pages;
681 unsigned long min_slab_pages;
682#endif /* CONFIG_NUMA */
683
a52633d8
MG
684 /* Write-intensive fields used by page reclaim */
685 ZONE_PADDING(_pad1_)
686 spinlock_t lru_lock;
3a80a7fa
MG
687
688#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
689 /*
690 * If memory initialisation on large machines is deferred then this
691 * is the first PFN that needs to be initialised.
692 */
693 unsigned long first_deferred_pfn;
d135e575
PT
694 /* Number of non-deferred pages */
695 unsigned long static_init_pgcnt;
3a80a7fa 696#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
a3d0a918
KS
697
698#ifdef CONFIG_TRANSPARENT_HUGEPAGE
699 spinlock_t split_queue_lock;
700 struct list_head split_queue;
701 unsigned long split_queue_len;
702#endif
75ef7184 703
599d0c95
MG
704 /* Fields commonly accessed by the page reclaim scanner */
705 struct lruvec lruvec;
706
599d0c95
MG
707 unsigned long flags;
708
709 ZONE_PADDING(_pad2_)
710
75ef7184
MG
711 /* Per-node vmstats */
712 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
713 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
1da177e4
LT
714} pg_data_t;
715
716#define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
717#define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
d41dee36 718#ifdef CONFIG_FLAT_NODE_MEM_MAP
408fde81 719#define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
d41dee36
AW
720#else
721#define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
722#endif
408fde81 723#define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
1da177e4 724
c6830c22 725#define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
da3649e1 726#define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
a52633d8
MG
727static inline spinlock_t *zone_lru_lock(struct zone *zone)
728{
729 return &zone->zone_pgdat->lru_lock;
730}
c6830c22 731
a9dd0a83 732static inline struct lruvec *node_lruvec(struct pglist_data *pgdat)
599d0c95 733{
a9dd0a83 734 return &pgdat->lruvec;
599d0c95
MG
735}
736
da3649e1
CS
737static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
738{
739 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
740}
741
742static inline bool pgdat_is_empty(pg_data_t *pgdat)
743{
744 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
745}
c6830c22 746
208d54e5
DH
747#include <linux/memory_hotplug.h>
748
72675e13 749void build_all_zonelists(pg_data_t *pgdat);
5ecd9d40
DR
750void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
751 enum zone_type classzone_idx);
86a294a8
MH
752bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
753 int classzone_idx, unsigned int alloc_flags,
754 long free_pages);
7aeb09f9 755bool zone_watermark_ok(struct zone *z, unsigned int order,
c603844b
MG
756 unsigned long mark, int classzone_idx,
757 unsigned int alloc_flags);
7aeb09f9 758bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
e2b19197 759 unsigned long mark, int classzone_idx);
a2f3aa02
DH
760enum memmap_context {
761 MEMMAP_EARLY,
762 MEMMAP_HOTPLUG,
763};
dc0bbf3b 764extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
b171e409 765 unsigned long size);
718127cc 766
bea8c150 767extern void lruvec_init(struct lruvec *lruvec);
7f5e86c2 768
599d0c95 769static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
7f5e86c2 770{
c255a458 771#ifdef CONFIG_MEMCG
599d0c95 772 return lruvec->pgdat;
7f5e86c2 773#else
599d0c95 774 return container_of(lruvec, struct pglist_data, lruvec);
7f5e86c2
KK
775#endif
776}
777
fd538803 778extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
23047a96 779
1da177e4
LT
780#ifdef CONFIG_HAVE_MEMORY_PRESENT
781void memory_present(int nid, unsigned long start, unsigned long end);
782#else
783static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
784#endif
785
7aac7898
LS
786#ifdef CONFIG_HAVE_MEMORYLESS_NODES
787int local_memory_node(int node_id);
788#else
789static inline int local_memory_node(int node_id) { return node_id; };
790#endif
791
1da177e4
LT
792/*
793 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
794 */
795#define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
796
89696701
OS
797#ifdef CONFIG_ZONE_DEVICE
798static inline bool is_dev_zone(const struct zone *zone)
799{
800 return zone_idx(zone) == ZONE_DEVICE;
801}
802#else
803static inline bool is_dev_zone(const struct zone *zone)
804{
805 return false;
806}
807#endif
808
6aa303de
MG
809/*
810 * Returns true if a zone has pages managed by the buddy allocator.
811 * All the reclaim decisions have to use this function rather than
812 * populated_zone(). If the whole zone is reserved then we can easily
813 * end up with populated_zone() && !managed_zone().
814 */
815static inline bool managed_zone(struct zone *zone)
816{
817 return zone->managed_pages;
818}
819
820/* Returns true if a zone has memory */
821static inline bool populated_zone(struct zone *zone)
f3fe6512 822{
6aa303de 823 return zone->present_pages;
f3fe6512
CK
824}
825
c1093b74
PT
826#ifdef CONFIG_NUMA
827static inline int zone_to_nid(struct zone *zone)
828{
829 return zone->node;
830}
831
832static inline void zone_set_nid(struct zone *zone, int nid)
833{
834 zone->node = nid;
835}
836#else
837static inline int zone_to_nid(struct zone *zone)
838{
839 return 0;
840}
841
842static inline void zone_set_nid(struct zone *zone, int nid) {}
843#endif
844
2a1e274a
MG
845extern int movable_zone;
846
d7e4a2ea 847#ifdef CONFIG_HIGHMEM
2a1e274a
MG
848static inline int zone_movable_is_highmem(void)
849{
d7e4a2ea 850#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
2a1e274a
MG
851 return movable_zone == ZONE_HIGHMEM;
852#else
d7e4a2ea 853 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
2a1e274a
MG
854#endif
855}
d7e4a2ea 856#endif
2a1e274a 857
2f1b6248 858static inline int is_highmem_idx(enum zone_type idx)
1da177e4 859{
e53ef38d 860#ifdef CONFIG_HIGHMEM
2a1e274a
MG
861 return (idx == ZONE_HIGHMEM ||
862 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
e53ef38d
CL
863#else
864 return 0;
865#endif
1da177e4
LT
866}
867
1da177e4 868/**
b4a991ec 869 * is_highmem - helper function to quickly check if a struct zone is a
1da177e4
LT
870 * highmem zone or not. This is an attempt to keep references
871 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
872 * @zone - pointer to struct zone variable
873 */
874static inline int is_highmem(struct zone *zone)
875{
e53ef38d 876#ifdef CONFIG_HIGHMEM
29f9cb53 877 return is_highmem_idx(zone_idx(zone));
e53ef38d
CL
878#else
879 return 0;
880#endif
1da177e4
LT
881}
882
1da177e4
LT
883/* These two functions are used to setup the per zone pages min values */
884struct ctl_table;
8d65af78 885int min_free_kbytes_sysctl_handler(struct ctl_table *, int,
1da177e4 886 void __user *, size_t *, loff_t *);
795ae7a0
JW
887int watermark_scale_factor_sysctl_handler(struct ctl_table *, int,
888 void __user *, size_t *, loff_t *);
d3cda233 889extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
8d65af78 890int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int,
1da177e4 891 void __user *, size_t *, loff_t *);
8d65af78 892int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
8ad4b1fb 893 void __user *, size_t *, loff_t *);
9614634f 894int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
8d65af78 895 void __user *, size_t *, loff_t *);
0ff38490 896int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
8d65af78 897 void __user *, size_t *, loff_t *);
1da177e4 898
f0c0b2b8 899extern int numa_zonelist_order_handler(struct ctl_table *, int,
8d65af78 900 void __user *, size_t *, loff_t *);
f0c0b2b8 901extern char numa_zonelist_order[];
c9bff3ee 902#define NUMA_ZONELIST_ORDER_LEN 16
f0c0b2b8 903
93b7504e 904#ifndef CONFIG_NEED_MULTIPLE_NODES
1da177e4
LT
905
906extern struct pglist_data contig_page_data;
907#define NODE_DATA(nid) (&contig_page_data)
908#define NODE_MEM_MAP(nid) mem_map
1da177e4 909
93b7504e 910#else /* CONFIG_NEED_MULTIPLE_NODES */
1da177e4
LT
911
912#include <asm/mmzone.h>
913
93b7504e 914#endif /* !CONFIG_NEED_MULTIPLE_NODES */
348f8b6c 915
95144c78
KH
916extern struct pglist_data *first_online_pgdat(void);
917extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
918extern struct zone *next_zone(struct zone *zone);
8357f869
KH
919
920/**
12d15f0d 921 * for_each_online_pgdat - helper macro to iterate over all online nodes
8357f869
KH
922 * @pgdat - pointer to a pg_data_t variable
923 */
924#define for_each_online_pgdat(pgdat) \
925 for (pgdat = first_online_pgdat(); \
926 pgdat; \
927 pgdat = next_online_pgdat(pgdat))
8357f869
KH
928/**
929 * for_each_zone - helper macro to iterate over all memory zones
930 * @zone - pointer to struct zone variable
931 *
932 * The user only needs to declare the zone variable, for_each_zone
933 * fills it in.
934 */
935#define for_each_zone(zone) \
936 for (zone = (first_online_pgdat())->node_zones; \
937 zone; \
938 zone = next_zone(zone))
939
ee99c71c
KM
940#define for_each_populated_zone(zone) \
941 for (zone = (first_online_pgdat())->node_zones; \
942 zone; \
943 zone = next_zone(zone)) \
944 if (!populated_zone(zone)) \
945 ; /* do nothing */ \
946 else
947
dd1a239f
MG
948static inline struct zone *zonelist_zone(struct zoneref *zoneref)
949{
950 return zoneref->zone;
951}
952
953static inline int zonelist_zone_idx(struct zoneref *zoneref)
954{
955 return zoneref->zone_idx;
956}
957
958static inline int zonelist_node_idx(struct zoneref *zoneref)
959{
c1093b74 960 return zone_to_nid(zoneref->zone);
dd1a239f
MG
961}
962
682a3385
MG
963struct zoneref *__next_zones_zonelist(struct zoneref *z,
964 enum zone_type highest_zoneidx,
965 nodemask_t *nodes);
966
19770b32
MG
967/**
968 * 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
969 * @z - The cursor used as a starting point for the search
970 * @highest_zoneidx - The zone index of the highest zone to return
971 * @nodes - An optional nodemask to filter the zonelist with
19770b32
MG
972 *
973 * This function returns the next zone at or below a given zone index that is
974 * within the allowed nodemask using a cursor as the starting point for the
5bead2a0
MG
975 * search. The zoneref returned is a cursor that represents the current zone
976 * being examined. It should be advanced by one before calling
977 * next_zones_zonelist again.
19770b32 978 */
682a3385 979static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
19770b32 980 enum zone_type highest_zoneidx,
682a3385
MG
981 nodemask_t *nodes)
982{
983 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
984 return z;
985 return __next_zones_zonelist(z, highest_zoneidx, nodes);
986}
dd1a239f 987
19770b32
MG
988/**
989 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
990 * @zonelist - The zonelist to search for a suitable zone
991 * @highest_zoneidx - The zone index of the highest zone to return
992 * @nodes - An optional nodemask to filter the zonelist with
ea57485a 993 * @return - Zoneref pointer for the first suitable zone found (see below)
19770b32
MG
994 *
995 * This function returns the first zone at or below a given zone index that is
996 * within the allowed nodemask. The zoneref returned is a cursor that can be
5bead2a0
MG
997 * used to iterate the zonelist with next_zones_zonelist by advancing it by
998 * one before calling.
ea57485a
VB
999 *
1000 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1001 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1002 * update due to cpuset modification.
19770b32 1003 */
dd1a239f 1004static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
19770b32 1005 enum zone_type highest_zoneidx,
c33d6c06 1006 nodemask_t *nodes)
54a6eb5c 1007{
c33d6c06 1008 return next_zones_zonelist(zonelist->_zonerefs,
05891fb0 1009 highest_zoneidx, nodes);
54a6eb5c
MG
1010}
1011
19770b32
MG
1012/**
1013 * 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
1014 * @zone - The current zone in the iterator
1015 * @z - The current pointer within zonelist->zones being iterated
1016 * @zlist - The zonelist being iterated
1017 * @highidx - The zone index of the highest zone to return
1018 * @nodemask - Nodemask allowed by the allocator
1019 *
1020 * This iterator iterates though all zones at or below a given zone index and
1021 * within a given nodemask
1022 */
1023#define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
c33d6c06 1024 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
19770b32 1025 zone; \
05891fb0 1026 z = next_zones_zonelist(++z, highidx, nodemask), \
c33d6c06
MG
1027 zone = zonelist_zone(z))
1028
1029#define for_next_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1030 for (zone = z->zone; \
1031 zone; \
1032 z = next_zones_zonelist(++z, highidx, nodemask), \
1033 zone = zonelist_zone(z))
1034
54a6eb5c
MG
1035
1036/**
1037 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1038 * @zone - The current zone in the iterator
1039 * @z - The current pointer within zonelist->zones being iterated
1040 * @zlist - The zonelist being iterated
1041 * @highidx - The zone index of the highest zone to return
1042 *
1043 * This iterator iterates though all zones at or below a given zone index.
1044 */
1045#define for_each_zone_zonelist(zone, z, zlist, highidx) \
19770b32 1046 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
54a6eb5c 1047
d41dee36
AW
1048#ifdef CONFIG_SPARSEMEM
1049#include <asm/sparsemem.h>
1050#endif
1051
c713216d 1052#if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \
0ee332c1 1053 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
b4544568
AM
1054static inline unsigned long early_pfn_to_nid(unsigned long pfn)
1055{
9d1f4b3f 1056 BUILD_BUG_ON(IS_ENABLED(CONFIG_NUMA));
b4544568
AM
1057 return 0;
1058}
b159d43f
AW
1059#endif
1060
2bdaf115
AW
1061#ifdef CONFIG_FLATMEM
1062#define pfn_to_nid(pfn) (0)
1063#endif
1064
d41dee36
AW
1065#ifdef CONFIG_SPARSEMEM
1066
1067/*
1068 * SECTION_SHIFT #bits space required to store a section #
1069 *
1070 * PA_SECTION_SHIFT physical address to/from section number
1071 * PFN_SECTION_SHIFT pfn to/from section number
1072 */
d41dee36
AW
1073#define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1074#define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1075
1076#define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1077
1078#define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1079#define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1080
835c134e 1081#define SECTION_BLOCKFLAGS_BITS \
d9c23400 1082 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
835c134e 1083
d41dee36
AW
1084#if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1085#error Allocator MAX_ORDER exceeds SECTION_SIZE
1086#endif
1087
1dd2bfc8
YI
1088static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1089{
1090 return pfn >> PFN_SECTION_SHIFT;
1091}
1092static inline unsigned long section_nr_to_pfn(unsigned long sec)
1093{
1094 return sec << PFN_SECTION_SHIFT;
1095}
e3c40f37 1096
a539f353
DK
1097#define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1098#define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1099
d41dee36 1100struct page;
eefa864b 1101struct page_ext;
d41dee36 1102struct mem_section {
29751f69
AW
1103 /*
1104 * This is, logically, a pointer to an array of struct
1105 * pages. However, it is stored with some other magic.
1106 * (see sparse.c::sparse_init_one_section())
1107 *
30c253e6
AW
1108 * Additionally during early boot we encode node id of
1109 * the location of the section here to guide allocation.
1110 * (see sparse.c::memory_present())
1111 *
29751f69
AW
1112 * Making it a UL at least makes someone do a cast
1113 * before using it wrong.
1114 */
1115 unsigned long section_mem_map;
5c0e3066
MG
1116
1117 /* See declaration of similar field in struct zone */
1118 unsigned long *pageblock_flags;
eefa864b
JK
1119#ifdef CONFIG_PAGE_EXTENSION
1120 /*
0c9ad804 1121 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
eefa864b
JK
1122 * section. (see page_ext.h about this.)
1123 */
1124 struct page_ext *page_ext;
1125 unsigned long pad;
1126#endif
55878e88
CS
1127 /*
1128 * WARNING: mem_section must be a power-of-2 in size for the
1129 * calculation and use of SECTION_ROOT_MASK to make sense.
1130 */
d41dee36
AW
1131};
1132
3e347261
BP
1133#ifdef CONFIG_SPARSEMEM_EXTREME
1134#define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1135#else
1136#define SECTIONS_PER_ROOT 1
1137#endif
802f192e 1138
3e347261 1139#define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
0faa5638 1140#define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
3e347261 1141#define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
802f192e 1142
3e347261 1143#ifdef CONFIG_SPARSEMEM_EXTREME
83e3c487 1144extern struct mem_section **mem_section;
802f192e 1145#else
3e347261
BP
1146extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1147#endif
d41dee36 1148
29751f69
AW
1149static inline struct mem_section *__nr_to_section(unsigned long nr)
1150{
83e3c487
KS
1151#ifdef CONFIG_SPARSEMEM_EXTREME
1152 if (!mem_section)
1153 return NULL;
1154#endif
3e347261
BP
1155 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1156 return NULL;
1157 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
29751f69 1158}
4ca644d9 1159extern int __section_nr(struct mem_section* ms);
04753278 1160extern unsigned long usemap_size(void);
29751f69
AW
1161
1162/*
1163 * We use the lower bits of the mem_map pointer to store
def9b71e
PT
1164 * a little bit of information. The pointer is calculated
1165 * as mem_map - section_nr_to_pfn(pnum). The result is
1166 * aligned to the minimum alignment of the two values:
1167 * 1. All mem_map arrays are page-aligned.
1168 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1169 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1170 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1171 * worst combination is powerpc with 256k pages,
1172 * which results in PFN_SECTION_SHIFT equal 6.
1173 * To sum it up, at least 6 bits are available.
29751f69
AW
1174 */
1175#define SECTION_MARKED_PRESENT (1UL<<0)
1176#define SECTION_HAS_MEM_MAP (1UL<<1)
2d070eab
MH
1177#define SECTION_IS_ONLINE (1UL<<2)
1178#define SECTION_MAP_LAST_BIT (1UL<<3)
29751f69 1179#define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
2d070eab 1180#define SECTION_NID_SHIFT 3
29751f69
AW
1181
1182static inline struct page *__section_mem_map_addr(struct mem_section *section)
1183{
1184 unsigned long map = section->section_mem_map;
1185 map &= SECTION_MAP_MASK;
1186 return (struct page *)map;
1187}
1188
540557b9 1189static inline int present_section(struct mem_section *section)
29751f69 1190{
802f192e 1191 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
29751f69
AW
1192}
1193
540557b9
AW
1194static inline int present_section_nr(unsigned long nr)
1195{
1196 return present_section(__nr_to_section(nr));
1197}
1198
1199static inline int valid_section(struct mem_section *section)
29751f69 1200{
802f192e 1201 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
29751f69
AW
1202}
1203
1204static inline int valid_section_nr(unsigned long nr)
1205{
1206 return valid_section(__nr_to_section(nr));
1207}
1208
2d070eab
MH
1209static inline int online_section(struct mem_section *section)
1210{
1211 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1212}
1213
1214static inline int online_section_nr(unsigned long nr)
1215{
1216 return online_section(__nr_to_section(nr));
1217}
1218
1219#ifdef CONFIG_MEMORY_HOTPLUG
1220void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1221#ifdef CONFIG_MEMORY_HOTREMOVE
1222void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1223#endif
1224#endif
1225
d41dee36
AW
1226static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1227{
29751f69 1228 return __nr_to_section(pfn_to_section_nr(pfn));
d41dee36
AW
1229}
1230
c4e1be9e
DH
1231extern int __highest_present_section_nr;
1232
7b7bf499 1233#ifndef CONFIG_HAVE_ARCH_PFN_VALID
d41dee36
AW
1234static inline int pfn_valid(unsigned long pfn)
1235{
1236 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1237 return 0;
29751f69 1238 return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
d41dee36 1239}
7b7bf499 1240#endif
d41dee36 1241
540557b9
AW
1242static inline int pfn_present(unsigned long pfn)
1243{
1244 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1245 return 0;
1246 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1247}
1248
d41dee36
AW
1249/*
1250 * These are _only_ used during initialisation, therefore they
1251 * can use __initdata ... They could have names to indicate
1252 * this restriction.
1253 */
1254#ifdef CONFIG_NUMA
161599ff
AW
1255#define pfn_to_nid(pfn) \
1256({ \
1257 unsigned long __pfn_to_nid_pfn = (pfn); \
1258 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1259})
2bdaf115
AW
1260#else
1261#define pfn_to_nid(pfn) (0)
d41dee36
AW
1262#endif
1263
d41dee36
AW
1264#define early_pfn_valid(pfn) pfn_valid(pfn)
1265void sparse_init(void);
1266#else
1267#define sparse_init() do {} while (0)
28ae55c9 1268#define sparse_index_init(_sec, _nid) do {} while (0)
d41dee36
AW
1269#endif /* CONFIG_SPARSEMEM */
1270
8a942fde
MG
1271/*
1272 * During memory init memblocks map pfns to nids. The search is expensive and
1273 * this caches recent lookups. The implementation of __early_pfn_to_nid
1274 * may treat start/end as pfns or sections.
1275 */
1276struct mminit_pfnnid_cache {
1277 unsigned long last_start;
1278 unsigned long last_end;
1279 int last_nid;
1280};
1281
d41dee36
AW
1282#ifndef early_pfn_valid
1283#define early_pfn_valid(pfn) (1)
1284#endif
1285
1286void memory_present(int nid, unsigned long start, unsigned long end);
d41dee36 1287
14e07298
AW
1288/*
1289 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1290 * need to check pfn validility within that MAX_ORDER_NR_PAGES block.
1291 * pfn_valid_within() should be used in this case; we optimise this away
1292 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1293 */
1294#ifdef CONFIG_HOLES_IN_ZONE
1295#define pfn_valid_within(pfn) pfn_valid(pfn)
1296#else
1297#define pfn_valid_within(pfn) (1)
1298#endif
1299
eb33575c
MG
1300#ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL
1301/*
1302 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap
2d070eab
MH
1303 * associated with it or not. This means that a struct page exists for this
1304 * pfn. The caller cannot assume the page is fully initialized in general.
1305 * Hotplugable pages might not have been onlined yet. pfn_to_online_page()
1306 * will ensure the struct page is fully online and initialized. Special pages
1307 * (e.g. ZONE_DEVICE) are never onlined and should be treated accordingly.
1308 *
1309 * In FLATMEM, it is expected that holes always have valid memmap as long as
1310 * there is valid PFNs either side of the hole. In SPARSEMEM, it is assumed
1311 * that a valid section has a memmap for the entire section.
eb33575c
MG
1312 *
1313 * However, an ARM, and maybe other embedded architectures in the future
1314 * free memmap backing holes to save memory on the assumption the memmap is
1315 * never used. The page_zone linkages are then broken even though pfn_valid()
1316 * returns true. A walker of the full memmap must then do this additional
1317 * check to ensure the memmap they are looking at is sane by making sure
1318 * the zone and PFN linkages are still valid. This is expensive, but walkers
1319 * of the full memmap are extremely rare.
1320 */
5b80287a 1321bool memmap_valid_within(unsigned long pfn,
eb33575c
MG
1322 struct page *page, struct zone *zone);
1323#else
5b80287a 1324static inline bool memmap_valid_within(unsigned long pfn,
eb33575c
MG
1325 struct page *page, struct zone *zone)
1326{
5b80287a 1327 return true;
eb33575c
MG
1328}
1329#endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */
1330
97965478 1331#endif /* !__GENERATING_BOUNDS.H */
1da177e4 1332#endif /* !__ASSEMBLY__ */
1da177e4 1333#endif /* _LINUX_MMZONE_H */