Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | #ifndef _LINUX_MMZONE_H |
2 | #define _LINUX_MMZONE_H | |
3 | ||
1da177e4 | 4 | #ifndef __ASSEMBLY__ |
97965478 | 5 | #ifndef __GENERATING_BOUNDS_H |
1da177e4 | 6 | |
1da177e4 LT |
7 | #include <linux/spinlock.h> |
8 | #include <linux/list.h> | |
9 | #include <linux/wait.h> | |
e815af95 | 10 | #include <linux/bitops.h> |
1da177e4 LT |
11 | #include <linux/cache.h> |
12 | #include <linux/threads.h> | |
13 | #include <linux/numa.h> | |
14 | #include <linux/init.h> | |
bdc8cb98 | 15 | #include <linux/seqlock.h> |
8357f869 | 16 | #include <linux/nodemask.h> |
835c134e | 17 | #include <linux/pageblock-flags.h> |
bbeae5b0 | 18 | #include <linux/page-flags-layout.h> |
60063497 | 19 | #include <linux/atomic.h> |
93ff66bf | 20 | #include <asm/page.h> |
1da177e4 LT |
21 | |
22 | /* Free memory management - zoned buddy allocator. */ | |
23 | #ifndef CONFIG_FORCE_MAX_ZONEORDER | |
24 | #define MAX_ORDER 11 | |
25 | #else | |
26 | #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER | |
27 | #endif | |
e984bb43 | 28 | #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1)) |
1da177e4 | 29 | |
5ad333eb AW |
30 | /* |
31 | * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed | |
32 | * costly to service. That is between allocation orders which should | |
35fca53e | 33 | * coalesce naturally under reasonable reclaim pressure and those which |
5ad333eb AW |
34 | * will not. |
35 | */ | |
36 | #define PAGE_ALLOC_COSTLY_ORDER 3 | |
37 | ||
47118af0 MN |
38 | enum { |
39 | MIGRATE_UNMOVABLE, | |
40 | MIGRATE_RECLAIMABLE, | |
41 | MIGRATE_MOVABLE, | |
42 | MIGRATE_PCPTYPES, /* the number of types on the pcp lists */ | |
43 | MIGRATE_RESERVE = MIGRATE_PCPTYPES, | |
44 | #ifdef CONFIG_CMA | |
45 | /* | |
46 | * MIGRATE_CMA migration type is designed to mimic the way | |
47 | * ZONE_MOVABLE works. Only movable pages can be allocated | |
48 | * from MIGRATE_CMA pageblocks and page allocator never | |
49 | * implicitly change migration type of MIGRATE_CMA pageblock. | |
50 | * | |
51 | * The way to use it is to change migratetype of a range of | |
52 | * pageblocks to MIGRATE_CMA which can be done by | |
53 | * __free_pageblock_cma() function. What is important though | |
54 | * is that a range of pageblocks must be aligned to | |
55 | * MAX_ORDER_NR_PAGES should biggest page be bigger then | |
56 | * a single pageblock. | |
57 | */ | |
58 | MIGRATE_CMA, | |
59 | #endif | |
194159fb | 60 | #ifdef CONFIG_MEMORY_ISOLATION |
47118af0 | 61 | MIGRATE_ISOLATE, /* can't allocate from here */ |
194159fb | 62 | #endif |
47118af0 MN |
63 | MIGRATE_TYPES |
64 | }; | |
65 | ||
66 | #ifdef CONFIG_CMA | |
67 | # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA) | |
68 | #else | |
69 | # define is_migrate_cma(migratetype) false | |
70 | #endif | |
b2a0ac88 MG |
71 | |
72 | #define for_each_migratetype_order(order, type) \ | |
73 | for (order = 0; order < MAX_ORDER; order++) \ | |
74 | for (type = 0; type < MIGRATE_TYPES; type++) | |
75 | ||
467c996c MG |
76 | extern int page_group_by_mobility_disabled; |
77 | ||
e58469ba MG |
78 | #define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1) |
79 | #define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1) | |
80 | ||
dc4b0caf MG |
81 | #define get_pageblock_migratetype(page) \ |
82 | get_pfnblock_flags_mask(page, page_to_pfn(page), \ | |
83 | PB_migrate_end, MIGRATETYPE_MASK) | |
84 | ||
85 | static inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn) | |
467c996c | 86 | { |
e58469ba | 87 | BUILD_BUG_ON(PB_migrate_end - PB_migrate != 2); |
dc4b0caf MG |
88 | return get_pfnblock_flags_mask(page, pfn, PB_migrate_end, |
89 | MIGRATETYPE_MASK); | |
467c996c MG |
90 | } |
91 | ||
1da177e4 | 92 | struct free_area { |
b2a0ac88 | 93 | struct list_head free_list[MIGRATE_TYPES]; |
1da177e4 LT |
94 | unsigned long nr_free; |
95 | }; | |
96 | ||
97 | struct pglist_data; | |
98 | ||
99 | /* | |
100 | * zone->lock and zone->lru_lock are two of the hottest locks in the kernel. | |
101 | * So add a wild amount of padding here to ensure that they fall into separate | |
102 | * cachelines. There are very few zone structures in the machine, so space | |
103 | * consumption is not a concern here. | |
104 | */ | |
105 | #if defined(CONFIG_SMP) | |
106 | struct zone_padding { | |
107 | char x[0]; | |
22fc6ecc | 108 | } ____cacheline_internodealigned_in_smp; |
1da177e4 LT |
109 | #define ZONE_PADDING(name) struct zone_padding name; |
110 | #else | |
111 | #define ZONE_PADDING(name) | |
112 | #endif | |
113 | ||
2244b95a | 114 | enum zone_stat_item { |
51ed4491 | 115 | /* First 128 byte cacheline (assuming 64 bit words) */ |
d23ad423 | 116 | NR_FREE_PAGES, |
81c0a2bb | 117 | NR_ALLOC_BATCH, |
b69408e8 | 118 | NR_LRU_BASE, |
4f98a2fe RR |
119 | NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */ |
120 | NR_ACTIVE_ANON, /* " " " " " */ | |
121 | NR_INACTIVE_FILE, /* " " " " " */ | |
122 | NR_ACTIVE_FILE, /* " " " " " */ | |
894bc310 | 123 | NR_UNEVICTABLE, /* " " " " " */ |
5344b7e6 | 124 | NR_MLOCK, /* mlock()ed pages found and moved off LRU */ |
f3dbd344 CL |
125 | NR_ANON_PAGES, /* Mapped anonymous pages */ |
126 | NR_FILE_MAPPED, /* pagecache pages mapped into pagetables. | |
65ba55f5 | 127 | only modified from process context */ |
347ce434 | 128 | NR_FILE_PAGES, |
b1e7a8fd | 129 | NR_FILE_DIRTY, |
ce866b34 | 130 | NR_WRITEBACK, |
51ed4491 CL |
131 | NR_SLAB_RECLAIMABLE, |
132 | NR_SLAB_UNRECLAIMABLE, | |
133 | NR_PAGETABLE, /* used for pagetables */ | |
c6a7f572 KM |
134 | NR_KERNEL_STACK, |
135 | /* Second 128 byte cacheline */ | |
fd39fc85 | 136 | NR_UNSTABLE_NFS, /* NFS unstable pages */ |
d2c5e30c | 137 | NR_BOUNCE, |
e129b5c2 | 138 | NR_VMSCAN_WRITE, |
49ea7eb6 | 139 | NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */ |
fc3ba692 | 140 | NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */ |
a731286d KM |
141 | NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */ |
142 | NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */ | |
4b02108a | 143 | NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */ |
ea941f0e MR |
144 | NR_DIRTIED, /* page dirtyings since bootup */ |
145 | NR_WRITTEN, /* page writings since bootup */ | |
0d5d823a | 146 | NR_PAGES_SCANNED, /* pages scanned since last reclaim */ |
ca889e6c CL |
147 | #ifdef CONFIG_NUMA |
148 | NUMA_HIT, /* allocated in intended node */ | |
149 | NUMA_MISS, /* allocated in non intended node */ | |
150 | NUMA_FOREIGN, /* was intended here, hit elsewhere */ | |
151 | NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */ | |
152 | NUMA_LOCAL, /* allocation from local node */ | |
153 | NUMA_OTHER, /* allocation from other node */ | |
154 | #endif | |
a528910e JW |
155 | WORKINGSET_REFAULT, |
156 | WORKINGSET_ACTIVATE, | |
449dd698 | 157 | WORKINGSET_NODERECLAIM, |
79134171 | 158 | NR_ANON_TRANSPARENT_HUGEPAGES, |
d1ce749a | 159 | NR_FREE_CMA_PAGES, |
2244b95a CL |
160 | NR_VM_ZONE_STAT_ITEMS }; |
161 | ||
4f98a2fe RR |
162 | /* |
163 | * We do arithmetic on the LRU lists in various places in the code, | |
164 | * so it is important to keep the active lists LRU_ACTIVE higher in | |
165 | * the array than the corresponding inactive lists, and to keep | |
166 | * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists. | |
167 | * | |
168 | * This has to be kept in sync with the statistics in zone_stat_item | |
169 | * above and the descriptions in vmstat_text in mm/vmstat.c | |
170 | */ | |
171 | #define LRU_BASE 0 | |
172 | #define LRU_ACTIVE 1 | |
173 | #define LRU_FILE 2 | |
174 | ||
b69408e8 | 175 | enum lru_list { |
4f98a2fe RR |
176 | LRU_INACTIVE_ANON = LRU_BASE, |
177 | LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE, | |
178 | LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE, | |
179 | LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE, | |
894bc310 | 180 | LRU_UNEVICTABLE, |
894bc310 LS |
181 | NR_LRU_LISTS |
182 | }; | |
b69408e8 | 183 | |
4111304d | 184 | #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++) |
b69408e8 | 185 | |
4111304d | 186 | #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++) |
894bc310 | 187 | |
4111304d | 188 | static inline int is_file_lru(enum lru_list lru) |
4f98a2fe | 189 | { |
4111304d | 190 | return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE); |
4f98a2fe RR |
191 | } |
192 | ||
4111304d | 193 | static inline int is_active_lru(enum lru_list lru) |
b69408e8 | 194 | { |
4111304d | 195 | return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE); |
b69408e8 CL |
196 | } |
197 | ||
4111304d | 198 | static inline int is_unevictable_lru(enum lru_list lru) |
894bc310 | 199 | { |
4111304d | 200 | return (lru == LRU_UNEVICTABLE); |
894bc310 LS |
201 | } |
202 | ||
89abfab1 HD |
203 | struct zone_reclaim_stat { |
204 | /* | |
205 | * The pageout code in vmscan.c keeps track of how many of the | |
59f91e5d | 206 | * mem/swap backed and file backed pages are referenced. |
89abfab1 HD |
207 | * The higher the rotated/scanned ratio, the more valuable |
208 | * that cache is. | |
209 | * | |
210 | * The anon LRU stats live in [0], file LRU stats in [1] | |
211 | */ | |
212 | unsigned long recent_rotated[2]; | |
213 | unsigned long recent_scanned[2]; | |
214 | }; | |
215 | ||
6290df54 JW |
216 | struct lruvec { |
217 | struct list_head lists[NR_LRU_LISTS]; | |
89abfab1 | 218 | struct zone_reclaim_stat reclaim_stat; |
c255a458 | 219 | #ifdef CONFIG_MEMCG |
7f5e86c2 KK |
220 | struct zone *zone; |
221 | #endif | |
6290df54 JW |
222 | }; |
223 | ||
bb2a0de9 KH |
224 | /* Mask used at gathering information at once (see memcontrol.c) */ |
225 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
226 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
bb2a0de9 KH |
227 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) |
228 | ||
39deaf85 | 229 | /* Isolate clean file */ |
f3fd4a61 | 230 | #define ISOLATE_CLEAN ((__force isolate_mode_t)0x1) |
f80c0673 | 231 | /* Isolate unmapped file */ |
f3fd4a61 | 232 | #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2) |
c8244935 | 233 | /* Isolate for asynchronous migration */ |
f3fd4a61 | 234 | #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4) |
e46a2879 MK |
235 | /* Isolate unevictable pages */ |
236 | #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8) | |
4356f21d MK |
237 | |
238 | /* LRU Isolation modes. */ | |
239 | typedef unsigned __bitwise__ isolate_mode_t; | |
240 | ||
41858966 MG |
241 | enum zone_watermarks { |
242 | WMARK_MIN, | |
243 | WMARK_LOW, | |
244 | WMARK_HIGH, | |
245 | NR_WMARK | |
246 | }; | |
247 | ||
248 | #define min_wmark_pages(z) (z->watermark[WMARK_MIN]) | |
249 | #define low_wmark_pages(z) (z->watermark[WMARK_LOW]) | |
250 | #define high_wmark_pages(z) (z->watermark[WMARK_HIGH]) | |
251 | ||
1da177e4 LT |
252 | struct per_cpu_pages { |
253 | int count; /* number of pages in the list */ | |
1da177e4 LT |
254 | int high; /* high watermark, emptying needed */ |
255 | int batch; /* chunk size for buddy add/remove */ | |
5f8dcc21 MG |
256 | |
257 | /* Lists of pages, one per migrate type stored on the pcp-lists */ | |
258 | struct list_head lists[MIGRATE_PCPTYPES]; | |
1da177e4 LT |
259 | }; |
260 | ||
261 | struct per_cpu_pageset { | |
3dfa5721 | 262 | struct per_cpu_pages pcp; |
4037d452 CL |
263 | #ifdef CONFIG_NUMA |
264 | s8 expire; | |
265 | #endif | |
2244b95a | 266 | #ifdef CONFIG_SMP |
df9ecaba | 267 | s8 stat_threshold; |
2244b95a CL |
268 | s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS]; |
269 | #endif | |
99dcc3e5 | 270 | }; |
e7c8d5c9 | 271 | |
97965478 CL |
272 | #endif /* !__GENERATING_BOUNDS.H */ |
273 | ||
2f1b6248 | 274 | enum zone_type { |
4b51d669 | 275 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 CL |
276 | /* |
277 | * ZONE_DMA is used when there are devices that are not able | |
278 | * to do DMA to all of addressable memory (ZONE_NORMAL). Then we | |
279 | * carve out the portion of memory that is needed for these devices. | |
280 | * The range is arch specific. | |
281 | * | |
282 | * Some examples | |
283 | * | |
284 | * Architecture Limit | |
285 | * --------------------------- | |
286 | * parisc, ia64, sparc <4G | |
287 | * s390 <2G | |
2f1b6248 CL |
288 | * arm Various |
289 | * alpha Unlimited or 0-16MB. | |
290 | * | |
291 | * i386, x86_64 and multiple other arches | |
292 | * <16M. | |
293 | */ | |
294 | ZONE_DMA, | |
4b51d669 | 295 | #endif |
fb0e7942 | 296 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 CL |
297 | /* |
298 | * x86_64 needs two ZONE_DMAs because it supports devices that are | |
299 | * only able to do DMA to the lower 16M but also 32 bit devices that | |
300 | * can only do DMA areas below 4G. | |
301 | */ | |
302 | ZONE_DMA32, | |
fb0e7942 | 303 | #endif |
2f1b6248 CL |
304 | /* |
305 | * Normal addressable memory is in ZONE_NORMAL. DMA operations can be | |
306 | * performed on pages in ZONE_NORMAL if the DMA devices support | |
307 | * transfers to all addressable memory. | |
308 | */ | |
309 | ZONE_NORMAL, | |
e53ef38d | 310 | #ifdef CONFIG_HIGHMEM |
2f1b6248 CL |
311 | /* |
312 | * A memory area that is only addressable by the kernel through | |
313 | * mapping portions into its own address space. This is for example | |
314 | * used by i386 to allow the kernel to address the memory beyond | |
315 | * 900MB. The kernel will set up special mappings (page | |
316 | * table entries on i386) for each page that the kernel needs to | |
317 | * access. | |
318 | */ | |
319 | ZONE_HIGHMEM, | |
e53ef38d | 320 | #endif |
2a1e274a | 321 | ZONE_MOVABLE, |
97965478 | 322 | __MAX_NR_ZONES |
2f1b6248 | 323 | }; |
1da177e4 | 324 | |
97965478 CL |
325 | #ifndef __GENERATING_BOUNDS_H |
326 | ||
1da177e4 | 327 | struct zone { |
3484b2de | 328 | /* Read-mostly fields */ |
41858966 MG |
329 | |
330 | /* zone watermarks, access with *_wmark_pages(zone) macros */ | |
331 | unsigned long watermark[NR_WMARK]; | |
332 | ||
1da177e4 LT |
333 | /* |
334 | * We don't know if the memory that we're going to allocate will be freeable | |
335 | * or/and it will be released eventually, so to avoid totally wasting several | |
336 | * GB of ram we must reserve some of the lower zone memory (otherwise we risk | |
337 | * to run OOM on the lower zones despite there's tons of freeable ram | |
338 | * on the higher zones). This array is recalculated at runtime if the | |
339 | * sysctl_lowmem_reserve_ratio sysctl changes. | |
340 | */ | |
3484b2de | 341 | long lowmem_reserve[MAX_NR_ZONES]; |
ab8fabd4 | 342 | |
e7c8d5c9 | 343 | #ifdef CONFIG_NUMA |
d5f541ed | 344 | int node; |
3484b2de MG |
345 | #endif |
346 | ||
9614634f | 347 | /* |
3484b2de MG |
348 | * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on |
349 | * this zone's LRU. Maintained by the pageout code. | |
9614634f | 350 | */ |
3484b2de MG |
351 | unsigned int inactive_ratio; |
352 | ||
353 | struct pglist_data *zone_pgdat; | |
43cf38eb | 354 | struct per_cpu_pageset __percpu *pageset; |
3484b2de | 355 | |
1da177e4 | 356 | /* |
3484b2de MG |
357 | * This is a per-zone reserve of pages that should not be |
358 | * considered dirtyable memory. | |
1da177e4 | 359 | */ |
3484b2de | 360 | unsigned long dirty_balance_reserve; |
1da177e4 | 361 | |
835c134e MG |
362 | #ifndef CONFIG_SPARSEMEM |
363 | /* | |
d9c23400 | 364 | * Flags for a pageblock_nr_pages block. See pageblock-flags.h. |
835c134e MG |
365 | * In SPARSEMEM, this map is stored in struct mem_section |
366 | */ | |
367 | unsigned long *pageblock_flags; | |
368 | #endif /* CONFIG_SPARSEMEM */ | |
369 | ||
3484b2de | 370 | #ifdef CONFIG_NUMA |
1da177e4 | 371 | /* |
3484b2de | 372 | * zone reclaim becomes active if more unmapped pages exist. |
1da177e4 | 373 | */ |
3484b2de MG |
374 | unsigned long min_unmapped_pages; |
375 | unsigned long min_slab_pages; | |
376 | #endif /* CONFIG_NUMA */ | |
1da177e4 | 377 | |
1da177e4 LT |
378 | /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */ |
379 | unsigned long zone_start_pfn; | |
380 | ||
bdc8cb98 | 381 | /* |
9feedc9d JL |
382 | * spanned_pages is the total pages spanned by the zone, including |
383 | * holes, which is calculated as: | |
384 | * spanned_pages = zone_end_pfn - zone_start_pfn; | |
bdc8cb98 | 385 | * |
9feedc9d JL |
386 | * present_pages is physical pages existing within the zone, which |
387 | * is calculated as: | |
8761e31c | 388 | * present_pages = spanned_pages - absent_pages(pages in holes); |
9feedc9d JL |
389 | * |
390 | * managed_pages is present pages managed by the buddy system, which | |
391 | * is calculated as (reserved_pages includes pages allocated by the | |
392 | * bootmem allocator): | |
393 | * managed_pages = present_pages - reserved_pages; | |
394 | * | |
395 | * So present_pages may be used by memory hotplug or memory power | |
396 | * management logic to figure out unmanaged pages by checking | |
397 | * (present_pages - managed_pages). And managed_pages should be used | |
398 | * by page allocator and vm scanner to calculate all kinds of watermarks | |
399 | * and thresholds. | |
400 | * | |
401 | * Locking rules: | |
402 | * | |
403 | * zone_start_pfn and spanned_pages are protected by span_seqlock. | |
404 | * It is a seqlock because it has to be read outside of zone->lock, | |
405 | * and it is done in the main allocator path. But, it is written | |
406 | * quite infrequently. | |
407 | * | |
408 | * The span_seq lock is declared along with zone->lock because it is | |
bdc8cb98 DH |
409 | * frequently read in proximity to zone->lock. It's good to |
410 | * give them a chance of being in the same cacheline. | |
9feedc9d | 411 | * |
c3d5f5f0 | 412 | * Write access to present_pages at runtime should be protected by |
bfc8c901 VD |
413 | * mem_hotplug_begin/end(). Any reader who can't tolerant drift of |
414 | * present_pages should get_online_mems() to get a stable value. | |
c3d5f5f0 JL |
415 | * |
416 | * Read access to managed_pages should be safe because it's unsigned | |
417 | * long. Write access to zone->managed_pages and totalram_pages are | |
418 | * protected by managed_page_count_lock at runtime. Idealy only | |
419 | * adjust_managed_page_count() should be used instead of directly | |
420 | * touching zone->managed_pages and totalram_pages. | |
bdc8cb98 | 421 | */ |
3484b2de | 422 | unsigned long managed_pages; |
9feedc9d JL |
423 | unsigned long spanned_pages; |
424 | unsigned long present_pages; | |
3484b2de MG |
425 | |
426 | const char *name; | |
1da177e4 | 427 | |
943dca1a | 428 | /* |
44628d97 | 429 | * Number of MIGRATE_RESERVE page block. To maintain for just |
943dca1a YI |
430 | * optimization. Protected by zone->lock. |
431 | */ | |
432 | int nr_migrate_reserve_block; | |
433 | ||
ad53f92e JK |
434 | #ifdef CONFIG_MEMORY_ISOLATION |
435 | /* | |
436 | * Number of isolated pageblock. It is used to solve incorrect | |
437 | * freepage counting problem due to racy retrieving migratetype | |
438 | * of pageblock. Protected by zone->lock. | |
439 | */ | |
440 | unsigned long nr_isolate_pageblock; | |
441 | #endif | |
442 | ||
3484b2de MG |
443 | #ifdef CONFIG_MEMORY_HOTPLUG |
444 | /* see spanned/present_pages for more description */ | |
445 | seqlock_t span_seqlock; | |
446 | #endif | |
447 | ||
1da177e4 | 448 | /* |
3484b2de MG |
449 | * wait_table -- the array holding the hash table |
450 | * wait_table_hash_nr_entries -- the size of the hash table array | |
451 | * wait_table_bits -- wait_table_size == (1 << wait_table_bits) | |
452 | * | |
453 | * The purpose of all these is to keep track of the people | |
454 | * waiting for a page to become available and make them | |
455 | * runnable again when possible. The trouble is that this | |
456 | * consumes a lot of space, especially when so few things | |
457 | * wait on pages at a given time. So instead of using | |
458 | * per-page waitqueues, we use a waitqueue hash table. | |
459 | * | |
460 | * The bucket discipline is to sleep on the same queue when | |
461 | * colliding and wake all in that wait queue when removing. | |
462 | * When something wakes, it must check to be sure its page is | |
463 | * truly available, a la thundering herd. The cost of a | |
464 | * collision is great, but given the expected load of the | |
465 | * table, they should be so rare as to be outweighed by the | |
466 | * benefits from the saved space. | |
467 | * | |
468 | * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the | |
469 | * primary users of these fields, and in mm/page_alloc.c | |
470 | * free_area_init_core() performs the initialization of them. | |
1da177e4 | 471 | */ |
3484b2de MG |
472 | wait_queue_head_t *wait_table; |
473 | unsigned long wait_table_hash_nr_entries; | |
474 | unsigned long wait_table_bits; | |
475 | ||
476 | ZONE_PADDING(_pad1_) | |
3484b2de MG |
477 | /* free areas of different sizes */ |
478 | struct free_area free_area[MAX_ORDER]; | |
479 | ||
480 | /* zone flags, see below */ | |
481 | unsigned long flags; | |
482 | ||
a368ab67 MG |
483 | /* Write-intensive fields used from the page allocator */ |
484 | spinlock_t lock; | |
485 | ||
3484b2de MG |
486 | ZONE_PADDING(_pad2_) |
487 | ||
488 | /* Write-intensive fields used by page reclaim */ | |
489 | ||
490 | /* Fields commonly accessed by the page reclaim scanner */ | |
491 | spinlock_t lru_lock; | |
3484b2de MG |
492 | struct lruvec lruvec; |
493 | ||
494 | /* Evictions & activations on the inactive file list */ | |
495 | atomic_long_t inactive_age; | |
496 | ||
497 | /* | |
498 | * When free pages are below this point, additional steps are taken | |
499 | * when reading the number of free pages to avoid per-cpu counter | |
500 | * drift allowing watermarks to be breached | |
501 | */ | |
502 | unsigned long percpu_drift_mark; | |
503 | ||
504 | #if defined CONFIG_COMPACTION || defined CONFIG_CMA | |
505 | /* pfn where compaction free scanner should start */ | |
506 | unsigned long compact_cached_free_pfn; | |
507 | /* pfn where async and sync compaction migration scanner should start */ | |
508 | unsigned long compact_cached_migrate_pfn[2]; | |
509 | #endif | |
510 | ||
511 | #ifdef CONFIG_COMPACTION | |
512 | /* | |
513 | * On compaction failure, 1<<compact_defer_shift compactions | |
514 | * are skipped before trying again. The number attempted since | |
515 | * last failure is tracked with compact_considered. | |
516 | */ | |
517 | unsigned int compact_considered; | |
518 | unsigned int compact_defer_shift; | |
519 | int compact_order_failed; | |
520 | #endif | |
521 | ||
522 | #if defined CONFIG_COMPACTION || defined CONFIG_CMA | |
523 | /* Set to true when the PG_migrate_skip bits should be cleared */ | |
524 | bool compact_blockskip_flush; | |
525 | #endif | |
526 | ||
527 | ZONE_PADDING(_pad3_) | |
528 | /* Zone statistics */ | |
529 | atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS]; | |
22fc6ecc | 530 | } ____cacheline_internodealigned_in_smp; |
1da177e4 | 531 | |
57054651 | 532 | enum zone_flags { |
e815af95 | 533 | ZONE_RECLAIM_LOCKED, /* prevents concurrent reclaim */ |
098d7f12 | 534 | ZONE_OOM_LOCKED, /* zone is in OOM killer zonelist */ |
0e093d99 MG |
535 | ZONE_CONGESTED, /* zone has many dirty pages backed by |
536 | * a congested BDI | |
537 | */ | |
57054651 | 538 | ZONE_DIRTY, /* reclaim scanning has recently found |
d43006d5 MG |
539 | * many dirty file pages at the tail |
540 | * of the LRU. | |
541 | */ | |
283aba9f MG |
542 | ZONE_WRITEBACK, /* reclaim scanning has recently found |
543 | * many pages under writeback | |
544 | */ | |
4ffeaf35 | 545 | ZONE_FAIR_DEPLETED, /* fair zone policy batch depleted */ |
57054651 | 546 | }; |
e815af95 | 547 | |
f9228b20 | 548 | static inline unsigned long zone_end_pfn(const struct zone *zone) |
108bcc96 CS |
549 | { |
550 | return zone->zone_start_pfn + zone->spanned_pages; | |
551 | } | |
552 | ||
553 | static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn) | |
554 | { | |
555 | return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone); | |
556 | } | |
557 | ||
2a6e3ebe CS |
558 | static inline bool zone_is_initialized(struct zone *zone) |
559 | { | |
560 | return !!zone->wait_table; | |
561 | } | |
562 | ||
563 | static inline bool zone_is_empty(struct zone *zone) | |
564 | { | |
565 | return zone->spanned_pages == 0; | |
566 | } | |
567 | ||
1da177e4 LT |
568 | /* |
569 | * The "priority" of VM scanning is how much of the queues we will scan in one | |
570 | * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the | |
571 | * queues ("queue_length >> 12") during an aging round. | |
572 | */ | |
573 | #define DEF_PRIORITY 12 | |
574 | ||
9276b1bc PJ |
575 | /* Maximum number of zones on a zonelist */ |
576 | #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES) | |
577 | ||
578 | #ifdef CONFIG_NUMA | |
523b9458 CL |
579 | |
580 | /* | |
25a64ec1 | 581 | * The NUMA zonelists are doubled because we need zonelists that restrict the |
e97ca8e5 | 582 | * allocations to a single node for __GFP_THISNODE. |
523b9458 | 583 | * |
54a6eb5c | 584 | * [0] : Zonelist with fallback |
e97ca8e5 | 585 | * [1] : No fallback (__GFP_THISNODE) |
523b9458 | 586 | */ |
54a6eb5c | 587 | #define MAX_ZONELISTS 2 |
523b9458 CL |
588 | |
589 | ||
9276b1bc PJ |
590 | /* |
591 | * We cache key information from each zonelist for smaller cache | |
592 | * footprint when scanning for free pages in get_page_from_freelist(). | |
593 | * | |
594 | * 1) The BITMAP fullzones tracks which zones in a zonelist have come | |
595 | * up short of free memory since the last time (last_fullzone_zap) | |
596 | * we zero'd fullzones. | |
597 | * 2) The array z_to_n[] maps each zone in the zonelist to its node | |
598 | * id, so that we can efficiently evaluate whether that node is | |
599 | * set in the current tasks mems_allowed. | |
600 | * | |
601 | * Both fullzones and z_to_n[] are one-to-one with the zonelist, | |
602 | * indexed by a zones offset in the zonelist zones[] array. | |
603 | * | |
604 | * The get_page_from_freelist() routine does two scans. During the | |
605 | * first scan, we skip zones whose corresponding bit in 'fullzones' | |
606 | * is set or whose corresponding node in current->mems_allowed (which | |
607 | * comes from cpusets) is not set. During the second scan, we bypass | |
608 | * this zonelist_cache, to ensure we look methodically at each zone. | |
609 | * | |
610 | * Once per second, we zero out (zap) fullzones, forcing us to | |
611 | * reconsider nodes that might have regained more free memory. | |
612 | * The field last_full_zap is the time we last zapped fullzones. | |
613 | * | |
614 | * This mechanism reduces the amount of time we waste repeatedly | |
615 | * reexaming zones for free memory when they just came up low on | |
616 | * memory momentarilly ago. | |
617 | * | |
618 | * The zonelist_cache struct members logically belong in struct | |
619 | * zonelist. However, the mempolicy zonelists constructed for | |
620 | * MPOL_BIND are intentionally variable length (and usually much | |
621 | * shorter). A general purpose mechanism for handling structs with | |
622 | * multiple variable length members is more mechanism than we want | |
623 | * here. We resort to some special case hackery instead. | |
624 | * | |
625 | * The MPOL_BIND zonelists don't need this zonelist_cache (in good | |
626 | * part because they are shorter), so we put the fixed length stuff | |
627 | * at the front of the zonelist struct, ending in a variable length | |
628 | * zones[], as is needed by MPOL_BIND. | |
629 | * | |
630 | * Then we put the optional zonelist cache on the end of the zonelist | |
631 | * struct. This optional stuff is found by a 'zlcache_ptr' pointer in | |
632 | * the fixed length portion at the front of the struct. This pointer | |
633 | * both enables us to find the zonelist cache, and in the case of | |
634 | * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL) | |
635 | * to know that the zonelist cache is not there. | |
636 | * | |
637 | * The end result is that struct zonelists come in two flavors: | |
638 | * 1) The full, fixed length version, shown below, and | |
639 | * 2) The custom zonelists for MPOL_BIND. | |
640 | * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache. | |
641 | * | |
642 | * Even though there may be multiple CPU cores on a node modifying | |
643 | * fullzones or last_full_zap in the same zonelist_cache at the same | |
644 | * time, we don't lock it. This is just hint data - if it is wrong now | |
645 | * and then, the allocator will still function, perhaps a bit slower. | |
646 | */ | |
647 | ||
648 | ||
649 | struct zonelist_cache { | |
9276b1bc | 650 | unsigned short z_to_n[MAX_ZONES_PER_ZONELIST]; /* zone->nid */ |
7253f4ef | 651 | DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST); /* zone full? */ |
9276b1bc PJ |
652 | unsigned long last_full_zap; /* when last zap'd (jiffies) */ |
653 | }; | |
654 | #else | |
54a6eb5c | 655 | #define MAX_ZONELISTS 1 |
9276b1bc PJ |
656 | struct zonelist_cache; |
657 | #endif | |
658 | ||
dd1a239f MG |
659 | /* |
660 | * This struct contains information about a zone in a zonelist. It is stored | |
661 | * here to avoid dereferences into large structures and lookups of tables | |
662 | */ | |
663 | struct zoneref { | |
664 | struct zone *zone; /* Pointer to actual zone */ | |
665 | int zone_idx; /* zone_idx(zoneref->zone) */ | |
666 | }; | |
667 | ||
1da177e4 LT |
668 | /* |
669 | * One allocation request operates on a zonelist. A zonelist | |
670 | * is a list of zones, the first one is the 'goal' of the | |
671 | * allocation, the other zones are fallback zones, in decreasing | |
672 | * priority. | |
673 | * | |
9276b1bc PJ |
674 | * If zlcache_ptr is not NULL, then it is just the address of zlcache, |
675 | * as explained above. If zlcache_ptr is NULL, there is no zlcache. | |
dd1a239f MG |
676 | * * |
677 | * To speed the reading of the zonelist, the zonerefs contain the zone index | |
678 | * of the entry being read. Helper functions to access information given | |
679 | * a struct zoneref are | |
680 | * | |
681 | * zonelist_zone() - Return the struct zone * for an entry in _zonerefs | |
682 | * zonelist_zone_idx() - Return the index of the zone for an entry | |
683 | * zonelist_node_idx() - Return the index of the node for an entry | |
1da177e4 LT |
684 | */ |
685 | struct zonelist { | |
9276b1bc | 686 | struct zonelist_cache *zlcache_ptr; // NULL or &zlcache |
dd1a239f | 687 | struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1]; |
9276b1bc PJ |
688 | #ifdef CONFIG_NUMA |
689 | struct zonelist_cache zlcache; // optional ... | |
690 | #endif | |
1da177e4 LT |
691 | }; |
692 | ||
0ee332c1 | 693 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d MG |
694 | struct node_active_region { |
695 | unsigned long start_pfn; | |
696 | unsigned long end_pfn; | |
697 | int nid; | |
698 | }; | |
0ee332c1 | 699 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
1da177e4 | 700 | |
5b99cd0e HC |
701 | #ifndef CONFIG_DISCONTIGMEM |
702 | /* The array of struct pages - for discontigmem use pgdat->lmem_map */ | |
703 | extern struct page *mem_map; | |
704 | #endif | |
705 | ||
1da177e4 LT |
706 | /* |
707 | * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM | |
708 | * (mostly NUMA machines?) to denote a higher-level memory zone than the | |
709 | * zone denotes. | |
710 | * | |
711 | * On NUMA machines, each NUMA node would have a pg_data_t to describe | |
712 | * it's memory layout. | |
713 | * | |
714 | * Memory statistics and page replacement data structures are maintained on a | |
715 | * per-zone basis. | |
716 | */ | |
717 | struct bootmem_data; | |
718 | typedef struct pglist_data { | |
719 | struct zone node_zones[MAX_NR_ZONES]; | |
523b9458 | 720 | struct zonelist node_zonelists[MAX_ZONELISTS]; |
1da177e4 | 721 | int nr_zones; |
52d4b9ac | 722 | #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */ |
1da177e4 | 723 | struct page *node_mem_map; |
eefa864b JK |
724 | #ifdef CONFIG_PAGE_EXTENSION |
725 | struct page_ext *node_page_ext; | |
726 | #endif | |
d41dee36 | 727 | #endif |
08677214 | 728 | #ifndef CONFIG_NO_BOOTMEM |
1da177e4 | 729 | struct bootmem_data *bdata; |
08677214 | 730 | #endif |
208d54e5 DH |
731 | #ifdef CONFIG_MEMORY_HOTPLUG |
732 | /* | |
733 | * Must be held any time you expect node_start_pfn, node_present_pages | |
734 | * or node_spanned_pages stay constant. Holding this will also | |
735 | * guarantee that any pfn_valid() stays that way. | |
736 | * | |
114d4b79 CS |
737 | * pgdat_resize_lock() and pgdat_resize_unlock() are provided to |
738 | * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG. | |
739 | * | |
72c3b51b | 740 | * Nests above zone->lock and zone->span_seqlock |
208d54e5 DH |
741 | */ |
742 | spinlock_t node_size_lock; | |
743 | #endif | |
1da177e4 LT |
744 | unsigned long node_start_pfn; |
745 | unsigned long node_present_pages; /* total number of physical pages */ | |
746 | unsigned long node_spanned_pages; /* total size of physical page | |
747 | range, including holes */ | |
748 | int node_id; | |
1da177e4 | 749 | wait_queue_head_t kswapd_wait; |
5515061d | 750 | wait_queue_head_t pfmemalloc_wait; |
bfc8c901 VD |
751 | struct task_struct *kswapd; /* Protected by |
752 | mem_hotplug_begin/end() */ | |
1da177e4 | 753 | int kswapd_max_order; |
99504748 | 754 | enum zone_type classzone_idx; |
8177a420 | 755 | #ifdef CONFIG_NUMA_BALANCING |
1c5e9c27 | 756 | /* Lock serializing the migrate rate limiting window */ |
8177a420 AA |
757 | spinlock_t numabalancing_migrate_lock; |
758 | ||
759 | /* Rate limiting time interval */ | |
760 | unsigned long numabalancing_migrate_next_window; | |
761 | ||
762 | /* Number of pages migrated during the rate limiting time interval */ | |
763 | unsigned long numabalancing_migrate_nr_pages; | |
764 | #endif | |
1da177e4 LT |
765 | } pg_data_t; |
766 | ||
767 | #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages) | |
768 | #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages) | |
d41dee36 | 769 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
408fde81 | 770 | #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr)) |
d41dee36 AW |
771 | #else |
772 | #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr)) | |
773 | #endif | |
408fde81 | 774 | #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr)) |
1da177e4 | 775 | |
c6830c22 | 776 | #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn) |
da3649e1 | 777 | #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid)) |
c6830c22 | 778 | |
da3649e1 CS |
779 | static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat) |
780 | { | |
781 | return pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
782 | } | |
783 | ||
784 | static inline bool pgdat_is_empty(pg_data_t *pgdat) | |
785 | { | |
786 | return !pgdat->node_start_pfn && !pgdat->node_spanned_pages; | |
787 | } | |
c6830c22 | 788 | |
208d54e5 DH |
789 | #include <linux/memory_hotplug.h> |
790 | ||
4eaf3f64 | 791 | extern struct mutex zonelists_mutex; |
9adb62a5 | 792 | void build_all_zonelists(pg_data_t *pgdat, struct zone *zone); |
99504748 | 793 | void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx); |
7aeb09f9 MG |
794 | bool zone_watermark_ok(struct zone *z, unsigned int order, |
795 | unsigned long mark, int classzone_idx, int alloc_flags); | |
796 | bool zone_watermark_ok_safe(struct zone *z, unsigned int order, | |
797 | unsigned long mark, int classzone_idx, int alloc_flags); | |
a2f3aa02 DH |
798 | enum memmap_context { |
799 | MEMMAP_EARLY, | |
800 | MEMMAP_HOTPLUG, | |
801 | }; | |
718127cc | 802 | extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn, |
a2f3aa02 DH |
803 | unsigned long size, |
804 | enum memmap_context context); | |
718127cc | 805 | |
bea8c150 | 806 | extern void lruvec_init(struct lruvec *lruvec); |
7f5e86c2 KK |
807 | |
808 | static inline struct zone *lruvec_zone(struct lruvec *lruvec) | |
809 | { | |
c255a458 | 810 | #ifdef CONFIG_MEMCG |
7f5e86c2 KK |
811 | return lruvec->zone; |
812 | #else | |
813 | return container_of(lruvec, struct zone, lruvec); | |
814 | #endif | |
815 | } | |
816 | ||
1da177e4 LT |
817 | #ifdef CONFIG_HAVE_MEMORY_PRESENT |
818 | void memory_present(int nid, unsigned long start, unsigned long end); | |
819 | #else | |
820 | static inline void memory_present(int nid, unsigned long start, unsigned long end) {} | |
821 | #endif | |
822 | ||
7aac7898 LS |
823 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
824 | int local_memory_node(int node_id); | |
825 | #else | |
826 | static inline int local_memory_node(int node_id) { return node_id; }; | |
827 | #endif | |
828 | ||
1da177e4 LT |
829 | #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE |
830 | unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long); | |
831 | #endif | |
832 | ||
833 | /* | |
834 | * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc. | |
835 | */ | |
836 | #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones) | |
837 | ||
f3fe6512 CK |
838 | static inline int populated_zone(struct zone *zone) |
839 | { | |
840 | return (!!zone->present_pages); | |
841 | } | |
842 | ||
2a1e274a MG |
843 | extern int movable_zone; |
844 | ||
845 | static inline int zone_movable_is_highmem(void) | |
846 | { | |
fe03025d | 847 | #if defined(CONFIG_HIGHMEM) && defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) |
2a1e274a | 848 | return movable_zone == ZONE_HIGHMEM; |
1a4dc5bc WN |
849 | #elif defined(CONFIG_HIGHMEM) |
850 | return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM; | |
2a1e274a MG |
851 | #else |
852 | return 0; | |
853 | #endif | |
854 | } | |
855 | ||
2f1b6248 | 856 | static inline int is_highmem_idx(enum zone_type idx) |
1da177e4 | 857 | { |
e53ef38d | 858 | #ifdef CONFIG_HIGHMEM |
2a1e274a MG |
859 | return (idx == ZONE_HIGHMEM || |
860 | (idx == ZONE_MOVABLE && zone_movable_is_highmem())); | |
e53ef38d CL |
861 | #else |
862 | return 0; | |
863 | #endif | |
1da177e4 LT |
864 | } |
865 | ||
1da177e4 LT |
866 | /** |
867 | * is_highmem - helper function to quickly check if a struct zone is a | |
868 | * highmem zone or not. This is an attempt to keep references | |
869 | * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum. | |
870 | * @zone - pointer to struct zone variable | |
871 | */ | |
872 | static inline int is_highmem(struct zone *zone) | |
873 | { | |
e53ef38d | 874 | #ifdef CONFIG_HIGHMEM |
ddc81ed2 HH |
875 | int zone_off = (char *)zone - (char *)zone->zone_pgdat->node_zones; |
876 | return zone_off == ZONE_HIGHMEM * sizeof(*zone) || | |
877 | (zone_off == ZONE_MOVABLE * sizeof(*zone) && | |
878 | zone_movable_is_highmem()); | |
e53ef38d CL |
879 | #else |
880 | return 0; | |
881 | #endif | |
1da177e4 LT |
882 | } |
883 | ||
1da177e4 LT |
884 | /* These two functions are used to setup the per zone pages min values */ |
885 | struct ctl_table; | |
8d65af78 | 886 | int min_free_kbytes_sysctl_handler(struct ctl_table *, int, |
1da177e4 LT |
887 | void __user *, size_t *, loff_t *); |
888 | extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1]; | |
8d65af78 | 889 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, |
1da177e4 | 890 | void __user *, size_t *, loff_t *); |
8d65af78 | 891 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, |
8ad4b1fb | 892 | void __user *, size_t *, loff_t *); |
9614634f | 893 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int, |
8d65af78 | 894 | void __user *, size_t *, loff_t *); |
0ff38490 | 895 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int, |
8d65af78 | 896 | void __user *, size_t *, loff_t *); |
1da177e4 | 897 | |
f0c0b2b8 | 898 | extern int numa_zonelist_order_handler(struct ctl_table *, int, |
8d65af78 | 899 | void __user *, size_t *, loff_t *); |
f0c0b2b8 KH |
900 | extern char numa_zonelist_order[]; |
901 | #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */ | |
902 | ||
93b7504e | 903 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
904 | |
905 | extern struct pglist_data contig_page_data; | |
906 | #define NODE_DATA(nid) (&contig_page_data) | |
907 | #define NODE_MEM_MAP(nid) mem_map | |
1da177e4 | 908 | |
93b7504e | 909 | #else /* CONFIG_NEED_MULTIPLE_NODES */ |
1da177e4 LT |
910 | |
911 | #include <asm/mmzone.h> | |
912 | ||
93b7504e | 913 | #endif /* !CONFIG_NEED_MULTIPLE_NODES */ |
348f8b6c | 914 | |
95144c78 KH |
915 | extern struct pglist_data *first_online_pgdat(void); |
916 | extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat); | |
917 | extern struct zone *next_zone(struct zone *zone); | |
8357f869 KH |
918 | |
919 | /** | |
12d15f0d | 920 | * for_each_online_pgdat - helper macro to iterate over all online nodes |
8357f869 KH |
921 | * @pgdat - pointer to a pg_data_t variable |
922 | */ | |
923 | #define for_each_online_pgdat(pgdat) \ | |
924 | for (pgdat = first_online_pgdat(); \ | |
925 | pgdat; \ | |
926 | pgdat = next_online_pgdat(pgdat)) | |
8357f869 KH |
927 | /** |
928 | * for_each_zone - helper macro to iterate over all memory zones | |
929 | * @zone - pointer to struct zone variable | |
930 | * | |
931 | * The user only needs to declare the zone variable, for_each_zone | |
932 | * fills it in. | |
933 | */ | |
934 | #define for_each_zone(zone) \ | |
935 | for (zone = (first_online_pgdat())->node_zones; \ | |
936 | zone; \ | |
937 | zone = next_zone(zone)) | |
938 | ||
ee99c71c KM |
939 | #define for_each_populated_zone(zone) \ |
940 | for (zone = (first_online_pgdat())->node_zones; \ | |
941 | zone; \ | |
942 | zone = next_zone(zone)) \ | |
943 | if (!populated_zone(zone)) \ | |
944 | ; /* do nothing */ \ | |
945 | else | |
946 | ||
dd1a239f MG |
947 | static inline struct zone *zonelist_zone(struct zoneref *zoneref) |
948 | { | |
949 | return zoneref->zone; | |
950 | } | |
951 | ||
952 | static inline int zonelist_zone_idx(struct zoneref *zoneref) | |
953 | { | |
954 | return zoneref->zone_idx; | |
955 | } | |
956 | ||
957 | static inline int zonelist_node_idx(struct zoneref *zoneref) | |
958 | { | |
959 | #ifdef CONFIG_NUMA | |
960 | /* zone_to_nid not available in this context */ | |
961 | return zoneref->zone->node; | |
962 | #else | |
963 | return 0; | |
964 | #endif /* CONFIG_NUMA */ | |
965 | } | |
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 MG |
978 | */ |
979 | struct zoneref *next_zones_zonelist(struct zoneref *z, | |
980 | enum zone_type highest_zoneidx, | |
05891fb0 | 981 | nodemask_t *nodes); |
dd1a239f | 982 | |
19770b32 MG |
983 | /** |
984 | * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist | |
985 | * @zonelist - The zonelist to search for a suitable zone | |
986 | * @highest_zoneidx - The zone index of the highest zone to return | |
987 | * @nodes - An optional nodemask to filter the zonelist with | |
988 | * @zone - The first suitable zone found is returned via this parameter | |
989 | * | |
990 | * This function returns the first zone at or below a given zone index that is | |
991 | * within the allowed nodemask. The zoneref returned is a cursor that can be | |
5bead2a0 MG |
992 | * used to iterate the zonelist with next_zones_zonelist by advancing it by |
993 | * one before calling. | |
19770b32 | 994 | */ |
dd1a239f | 995 | static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist, |
19770b32 MG |
996 | enum zone_type highest_zoneidx, |
997 | nodemask_t *nodes, | |
998 | struct zone **zone) | |
54a6eb5c | 999 | { |
05891fb0 VB |
1000 | struct zoneref *z = next_zones_zonelist(zonelist->_zonerefs, |
1001 | highest_zoneidx, nodes); | |
1002 | *zone = zonelist_zone(z); | |
1003 | return z; | |
54a6eb5c MG |
1004 | } |
1005 | ||
19770b32 MG |
1006 | /** |
1007 | * 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 | |
1008 | * @zone - The current zone in the iterator | |
1009 | * @z - The current pointer within zonelist->zones being iterated | |
1010 | * @zlist - The zonelist being iterated | |
1011 | * @highidx - The zone index of the highest zone to return | |
1012 | * @nodemask - Nodemask allowed by the allocator | |
1013 | * | |
1014 | * This iterator iterates though all zones at or below a given zone index and | |
1015 | * within a given nodemask | |
1016 | */ | |
1017 | #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \ | |
1018 | for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \ | |
1019 | zone; \ | |
05891fb0 VB |
1020 | z = next_zones_zonelist(++z, highidx, nodemask), \ |
1021 | zone = zonelist_zone(z)) \ | |
54a6eb5c MG |
1022 | |
1023 | /** | |
1024 | * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index | |
1025 | * @zone - The current zone in the iterator | |
1026 | * @z - The current pointer within zonelist->zones being iterated | |
1027 | * @zlist - The zonelist being iterated | |
1028 | * @highidx - The zone index of the highest zone to return | |
1029 | * | |
1030 | * This iterator iterates though all zones at or below a given zone index. | |
1031 | */ | |
1032 | #define for_each_zone_zonelist(zone, z, zlist, highidx) \ | |
19770b32 | 1033 | for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL) |
54a6eb5c | 1034 | |
d41dee36 AW |
1035 | #ifdef CONFIG_SPARSEMEM |
1036 | #include <asm/sparsemem.h> | |
1037 | #endif | |
1038 | ||
c713216d | 1039 | #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \ |
0ee332c1 | 1040 | !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) |
b4544568 AM |
1041 | static inline unsigned long early_pfn_to_nid(unsigned long pfn) |
1042 | { | |
1043 | return 0; | |
1044 | } | |
b159d43f AW |
1045 | #endif |
1046 | ||
2bdaf115 AW |
1047 | #ifdef CONFIG_FLATMEM |
1048 | #define pfn_to_nid(pfn) (0) | |
1049 | #endif | |
1050 | ||
d41dee36 AW |
1051 | #ifdef CONFIG_SPARSEMEM |
1052 | ||
1053 | /* | |
1054 | * SECTION_SHIFT #bits space required to store a section # | |
1055 | * | |
1056 | * PA_SECTION_SHIFT physical address to/from section number | |
1057 | * PFN_SECTION_SHIFT pfn to/from section number | |
1058 | */ | |
d41dee36 AW |
1059 | #define PA_SECTION_SHIFT (SECTION_SIZE_BITS) |
1060 | #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT) | |
1061 | ||
1062 | #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT) | |
1063 | ||
1064 | #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT) | |
1065 | #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1)) | |
1066 | ||
835c134e | 1067 | #define SECTION_BLOCKFLAGS_BITS \ |
d9c23400 | 1068 | ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS) |
835c134e | 1069 | |
d41dee36 AW |
1070 | #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS |
1071 | #error Allocator MAX_ORDER exceeds SECTION_SIZE | |
1072 | #endif | |
1073 | ||
e3c40f37 DK |
1074 | #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT) |
1075 | #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT) | |
1076 | ||
a539f353 DK |
1077 | #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK) |
1078 | #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK) | |
1079 | ||
d41dee36 | 1080 | struct page; |
eefa864b | 1081 | struct page_ext; |
d41dee36 | 1082 | struct mem_section { |
29751f69 AW |
1083 | /* |
1084 | * This is, logically, a pointer to an array of struct | |
1085 | * pages. However, it is stored with some other magic. | |
1086 | * (see sparse.c::sparse_init_one_section()) | |
1087 | * | |
30c253e6 AW |
1088 | * Additionally during early boot we encode node id of |
1089 | * the location of the section here to guide allocation. | |
1090 | * (see sparse.c::memory_present()) | |
1091 | * | |
29751f69 AW |
1092 | * Making it a UL at least makes someone do a cast |
1093 | * before using it wrong. | |
1094 | */ | |
1095 | unsigned long section_mem_map; | |
5c0e3066 MG |
1096 | |
1097 | /* See declaration of similar field in struct zone */ | |
1098 | unsigned long *pageblock_flags; | |
eefa864b JK |
1099 | #ifdef CONFIG_PAGE_EXTENSION |
1100 | /* | |
1101 | * If !SPARSEMEM, pgdat doesn't have page_ext pointer. We use | |
1102 | * section. (see page_ext.h about this.) | |
1103 | */ | |
1104 | struct page_ext *page_ext; | |
1105 | unsigned long pad; | |
1106 | #endif | |
55878e88 CS |
1107 | /* |
1108 | * WARNING: mem_section must be a power-of-2 in size for the | |
1109 | * calculation and use of SECTION_ROOT_MASK to make sense. | |
1110 | */ | |
d41dee36 AW |
1111 | }; |
1112 | ||
3e347261 BP |
1113 | #ifdef CONFIG_SPARSEMEM_EXTREME |
1114 | #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section)) | |
1115 | #else | |
1116 | #define SECTIONS_PER_ROOT 1 | |
1117 | #endif | |
802f192e | 1118 | |
3e347261 | 1119 | #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT) |
0faa5638 | 1120 | #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT) |
3e347261 | 1121 | #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1) |
802f192e | 1122 | |
3e347261 BP |
1123 | #ifdef CONFIG_SPARSEMEM_EXTREME |
1124 | extern struct mem_section *mem_section[NR_SECTION_ROOTS]; | |
802f192e | 1125 | #else |
3e347261 BP |
1126 | extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]; |
1127 | #endif | |
d41dee36 | 1128 | |
29751f69 AW |
1129 | static inline struct mem_section *__nr_to_section(unsigned long nr) |
1130 | { | |
3e347261 BP |
1131 | if (!mem_section[SECTION_NR_TO_ROOT(nr)]) |
1132 | return NULL; | |
1133 | return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK]; | |
29751f69 | 1134 | } |
4ca644d9 | 1135 | extern int __section_nr(struct mem_section* ms); |
04753278 | 1136 | extern unsigned long usemap_size(void); |
29751f69 AW |
1137 | |
1138 | /* | |
1139 | * We use the lower bits of the mem_map pointer to store | |
1140 | * a little bit of information. There should be at least | |
1141 | * 3 bits here due to 32-bit alignment. | |
1142 | */ | |
1143 | #define SECTION_MARKED_PRESENT (1UL<<0) | |
1144 | #define SECTION_HAS_MEM_MAP (1UL<<1) | |
1145 | #define SECTION_MAP_LAST_BIT (1UL<<2) | |
1146 | #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1)) | |
30c253e6 | 1147 | #define SECTION_NID_SHIFT 2 |
29751f69 AW |
1148 | |
1149 | static inline struct page *__section_mem_map_addr(struct mem_section *section) | |
1150 | { | |
1151 | unsigned long map = section->section_mem_map; | |
1152 | map &= SECTION_MAP_MASK; | |
1153 | return (struct page *)map; | |
1154 | } | |
1155 | ||
540557b9 | 1156 | static inline int present_section(struct mem_section *section) |
29751f69 | 1157 | { |
802f192e | 1158 | return (section && (section->section_mem_map & SECTION_MARKED_PRESENT)); |
29751f69 AW |
1159 | } |
1160 | ||
540557b9 AW |
1161 | static inline int present_section_nr(unsigned long nr) |
1162 | { | |
1163 | return present_section(__nr_to_section(nr)); | |
1164 | } | |
1165 | ||
1166 | static inline int valid_section(struct mem_section *section) | |
29751f69 | 1167 | { |
802f192e | 1168 | return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP)); |
29751f69 AW |
1169 | } |
1170 | ||
1171 | static inline int valid_section_nr(unsigned long nr) | |
1172 | { | |
1173 | return valid_section(__nr_to_section(nr)); | |
1174 | } | |
1175 | ||
d41dee36 AW |
1176 | static inline struct mem_section *__pfn_to_section(unsigned long pfn) |
1177 | { | |
29751f69 | 1178 | return __nr_to_section(pfn_to_section_nr(pfn)); |
d41dee36 AW |
1179 | } |
1180 | ||
7b7bf499 | 1181 | #ifndef CONFIG_HAVE_ARCH_PFN_VALID |
d41dee36 AW |
1182 | static inline int pfn_valid(unsigned long pfn) |
1183 | { | |
1184 | if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) | |
1185 | return 0; | |
29751f69 | 1186 | return valid_section(__nr_to_section(pfn_to_section_nr(pfn))); |
d41dee36 | 1187 | } |
7b7bf499 | 1188 | #endif |
d41dee36 | 1189 | |
540557b9 AW |
1190 | static inline int pfn_present(unsigned long pfn) |
1191 | { | |
1192 | if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) | |
1193 | return 0; | |
1194 | return present_section(__nr_to_section(pfn_to_section_nr(pfn))); | |
1195 | } | |
1196 | ||
d41dee36 AW |
1197 | /* |
1198 | * These are _only_ used during initialisation, therefore they | |
1199 | * can use __initdata ... They could have names to indicate | |
1200 | * this restriction. | |
1201 | */ | |
1202 | #ifdef CONFIG_NUMA | |
161599ff AW |
1203 | #define pfn_to_nid(pfn) \ |
1204 | ({ \ | |
1205 | unsigned long __pfn_to_nid_pfn = (pfn); \ | |
1206 | page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \ | |
1207 | }) | |
2bdaf115 AW |
1208 | #else |
1209 | #define pfn_to_nid(pfn) (0) | |
d41dee36 AW |
1210 | #endif |
1211 | ||
d41dee36 AW |
1212 | #define early_pfn_valid(pfn) pfn_valid(pfn) |
1213 | void sparse_init(void); | |
1214 | #else | |
1215 | #define sparse_init() do {} while (0) | |
28ae55c9 | 1216 | #define sparse_index_init(_sec, _nid) do {} while (0) |
d41dee36 AW |
1217 | #endif /* CONFIG_SPARSEMEM */ |
1218 | ||
75167957 | 1219 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
cc2559bc | 1220 | bool early_pfn_in_nid(unsigned long pfn, int nid); |
75167957 AW |
1221 | #else |
1222 | #define early_pfn_in_nid(pfn, nid) (1) | |
1223 | #endif | |
1224 | ||
d41dee36 AW |
1225 | #ifndef early_pfn_valid |
1226 | #define early_pfn_valid(pfn) (1) | |
1227 | #endif | |
1228 | ||
1229 | void memory_present(int nid, unsigned long start, unsigned long end); | |
1230 | unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long); | |
1231 | ||
14e07298 AW |
1232 | /* |
1233 | * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we | |
1234 | * need to check pfn validility within that MAX_ORDER_NR_PAGES block. | |
1235 | * pfn_valid_within() should be used in this case; we optimise this away | |
1236 | * when we have no holes within a MAX_ORDER_NR_PAGES block. | |
1237 | */ | |
1238 | #ifdef CONFIG_HOLES_IN_ZONE | |
1239 | #define pfn_valid_within(pfn) pfn_valid(pfn) | |
1240 | #else | |
1241 | #define pfn_valid_within(pfn) (1) | |
1242 | #endif | |
1243 | ||
eb33575c MG |
1244 | #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL |
1245 | /* | |
1246 | * pfn_valid() is meant to be able to tell if a given PFN has valid memmap | |
1247 | * associated with it or not. In FLATMEM, it is expected that holes always | |
1248 | * have valid memmap as long as there is valid PFNs either side of the hole. | |
1249 | * In SPARSEMEM, it is assumed that a valid section has a memmap for the | |
1250 | * entire section. | |
1251 | * | |
1252 | * However, an ARM, and maybe other embedded architectures in the future | |
1253 | * free memmap backing holes to save memory on the assumption the memmap is | |
1254 | * never used. The page_zone linkages are then broken even though pfn_valid() | |
1255 | * returns true. A walker of the full memmap must then do this additional | |
1256 | * check to ensure the memmap they are looking at is sane by making sure | |
1257 | * the zone and PFN linkages are still valid. This is expensive, but walkers | |
1258 | * of the full memmap are extremely rare. | |
1259 | */ | |
1260 | int memmap_valid_within(unsigned long pfn, | |
1261 | struct page *page, struct zone *zone); | |
1262 | #else | |
1263 | static inline int memmap_valid_within(unsigned long pfn, | |
1264 | struct page *page, struct zone *zone) | |
1265 | { | |
1266 | return 1; | |
1267 | } | |
1268 | #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */ | |
1269 | ||
97965478 | 1270 | #endif /* !__GENERATING_BOUNDS.H */ |
1da177e4 | 1271 | #endif /* !__ASSEMBLY__ */ |
1da177e4 | 1272 | #endif /* _LINUX_MMZONE_H */ |