]>
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> |
01fc0ac1 | 18 | #include <generated/bounds.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 | |
60 | MIGRATE_ISOLATE, /* can't allocate from here */ | |
61 | MIGRATE_TYPES | |
62 | }; | |
63 | ||
64 | #ifdef CONFIG_CMA | |
65 | # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA) | |
49f223a9 | 66 | # define cma_wmark_pages(zone) zone->min_cma_pages |
47118af0 MN |
67 | #else |
68 | # define is_migrate_cma(migratetype) false | |
49f223a9 | 69 | # define cma_wmark_pages(zone) 0 |
47118af0 | 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 | ||
78 | static inline int get_pageblock_migratetype(struct page *page) | |
79 | { | |
467c996c MG |
80 | return get_pageblock_flags_group(page, PB_migrate, PB_migrate_end); |
81 | } | |
82 | ||
1da177e4 | 83 | struct free_area { |
b2a0ac88 | 84 | struct list_head free_list[MIGRATE_TYPES]; |
1da177e4 LT |
85 | unsigned long nr_free; |
86 | }; | |
87 | ||
88 | struct pglist_data; | |
89 | ||
90 | /* | |
91 | * zone->lock and zone->lru_lock are two of the hottest locks in the kernel. | |
92 | * So add a wild amount of padding here to ensure that they fall into separate | |
93 | * cachelines. There are very few zone structures in the machine, so space | |
94 | * consumption is not a concern here. | |
95 | */ | |
96 | #if defined(CONFIG_SMP) | |
97 | struct zone_padding { | |
98 | char x[0]; | |
22fc6ecc | 99 | } ____cacheline_internodealigned_in_smp; |
1da177e4 LT |
100 | #define ZONE_PADDING(name) struct zone_padding name; |
101 | #else | |
102 | #define ZONE_PADDING(name) | |
103 | #endif | |
104 | ||
2244b95a | 105 | enum zone_stat_item { |
51ed4491 | 106 | /* First 128 byte cacheline (assuming 64 bit words) */ |
d23ad423 | 107 | NR_FREE_PAGES, |
b69408e8 | 108 | NR_LRU_BASE, |
4f98a2fe RR |
109 | NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */ |
110 | NR_ACTIVE_ANON, /* " " " " " */ | |
111 | NR_INACTIVE_FILE, /* " " " " " */ | |
112 | NR_ACTIVE_FILE, /* " " " " " */ | |
894bc310 | 113 | NR_UNEVICTABLE, /* " " " " " */ |
5344b7e6 | 114 | NR_MLOCK, /* mlock()ed pages found and moved off LRU */ |
f3dbd344 CL |
115 | NR_ANON_PAGES, /* Mapped anonymous pages */ |
116 | NR_FILE_MAPPED, /* pagecache pages mapped into pagetables. | |
65ba55f5 | 117 | only modified from process context */ |
347ce434 | 118 | NR_FILE_PAGES, |
b1e7a8fd | 119 | NR_FILE_DIRTY, |
ce866b34 | 120 | NR_WRITEBACK, |
51ed4491 CL |
121 | NR_SLAB_RECLAIMABLE, |
122 | NR_SLAB_UNRECLAIMABLE, | |
123 | NR_PAGETABLE, /* used for pagetables */ | |
c6a7f572 KM |
124 | NR_KERNEL_STACK, |
125 | /* Second 128 byte cacheline */ | |
fd39fc85 | 126 | NR_UNSTABLE_NFS, /* NFS unstable pages */ |
d2c5e30c | 127 | NR_BOUNCE, |
e129b5c2 | 128 | NR_VMSCAN_WRITE, |
49ea7eb6 | 129 | NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */ |
fc3ba692 | 130 | NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */ |
a731286d KM |
131 | NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */ |
132 | NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */ | |
4b02108a | 133 | NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */ |
ea941f0e MR |
134 | NR_DIRTIED, /* page dirtyings since bootup */ |
135 | NR_WRITTEN, /* page writings since bootup */ | |
ca889e6c CL |
136 | #ifdef CONFIG_NUMA |
137 | NUMA_HIT, /* allocated in intended node */ | |
138 | NUMA_MISS, /* allocated in non intended node */ | |
139 | NUMA_FOREIGN, /* was intended here, hit elsewhere */ | |
140 | NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */ | |
141 | NUMA_LOCAL, /* allocation from local node */ | |
142 | NUMA_OTHER, /* allocation from other node */ | |
143 | #endif | |
79134171 | 144 | NR_ANON_TRANSPARENT_HUGEPAGES, |
d1ce749a | 145 | NR_FREE_CMA_PAGES, |
2244b95a CL |
146 | NR_VM_ZONE_STAT_ITEMS }; |
147 | ||
4f98a2fe RR |
148 | /* |
149 | * We do arithmetic on the LRU lists in various places in the code, | |
150 | * so it is important to keep the active lists LRU_ACTIVE higher in | |
151 | * the array than the corresponding inactive lists, and to keep | |
152 | * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists. | |
153 | * | |
154 | * This has to be kept in sync with the statistics in zone_stat_item | |
155 | * above and the descriptions in vmstat_text in mm/vmstat.c | |
156 | */ | |
157 | #define LRU_BASE 0 | |
158 | #define LRU_ACTIVE 1 | |
159 | #define LRU_FILE 2 | |
160 | ||
b69408e8 | 161 | enum lru_list { |
4f98a2fe RR |
162 | LRU_INACTIVE_ANON = LRU_BASE, |
163 | LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE, | |
164 | LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE, | |
165 | LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE, | |
894bc310 | 166 | LRU_UNEVICTABLE, |
894bc310 LS |
167 | NR_LRU_LISTS |
168 | }; | |
b69408e8 | 169 | |
4111304d | 170 | #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++) |
b69408e8 | 171 | |
4111304d | 172 | #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++) |
894bc310 | 173 | |
4111304d | 174 | static inline int is_file_lru(enum lru_list lru) |
4f98a2fe | 175 | { |
4111304d | 176 | return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE); |
4f98a2fe RR |
177 | } |
178 | ||
4111304d | 179 | static inline int is_active_lru(enum lru_list lru) |
b69408e8 | 180 | { |
4111304d | 181 | return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE); |
b69408e8 CL |
182 | } |
183 | ||
4111304d | 184 | static inline int is_unevictable_lru(enum lru_list lru) |
894bc310 | 185 | { |
4111304d | 186 | return (lru == LRU_UNEVICTABLE); |
894bc310 LS |
187 | } |
188 | ||
89abfab1 HD |
189 | struct zone_reclaim_stat { |
190 | /* | |
191 | * The pageout code in vmscan.c keeps track of how many of the | |
59f91e5d | 192 | * mem/swap backed and file backed pages are referenced. |
89abfab1 HD |
193 | * The higher the rotated/scanned ratio, the more valuable |
194 | * that cache is. | |
195 | * | |
196 | * The anon LRU stats live in [0], file LRU stats in [1] | |
197 | */ | |
198 | unsigned long recent_rotated[2]; | |
199 | unsigned long recent_scanned[2]; | |
200 | }; | |
201 | ||
6290df54 JW |
202 | struct lruvec { |
203 | struct list_head lists[NR_LRU_LISTS]; | |
89abfab1 | 204 | struct zone_reclaim_stat reclaim_stat; |
c255a458 | 205 | #ifdef CONFIG_MEMCG |
7f5e86c2 KK |
206 | struct zone *zone; |
207 | #endif | |
6290df54 JW |
208 | }; |
209 | ||
bb2a0de9 KH |
210 | /* Mask used at gathering information at once (see memcontrol.c) */ |
211 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
212 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
bb2a0de9 KH |
213 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) |
214 | ||
39deaf85 | 215 | /* Isolate clean file */ |
f3fd4a61 | 216 | #define ISOLATE_CLEAN ((__force isolate_mode_t)0x1) |
f80c0673 | 217 | /* Isolate unmapped file */ |
f3fd4a61 | 218 | #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2) |
c8244935 | 219 | /* Isolate for asynchronous migration */ |
f3fd4a61 | 220 | #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4) |
e46a2879 MK |
221 | /* Isolate unevictable pages */ |
222 | #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8) | |
4356f21d MK |
223 | |
224 | /* LRU Isolation modes. */ | |
225 | typedef unsigned __bitwise__ isolate_mode_t; | |
226 | ||
41858966 MG |
227 | enum zone_watermarks { |
228 | WMARK_MIN, | |
229 | WMARK_LOW, | |
230 | WMARK_HIGH, | |
231 | NR_WMARK | |
232 | }; | |
233 | ||
234 | #define min_wmark_pages(z) (z->watermark[WMARK_MIN]) | |
235 | #define low_wmark_pages(z) (z->watermark[WMARK_LOW]) | |
236 | #define high_wmark_pages(z) (z->watermark[WMARK_HIGH]) | |
237 | ||
1da177e4 LT |
238 | struct per_cpu_pages { |
239 | int count; /* number of pages in the list */ | |
1da177e4 LT |
240 | int high; /* high watermark, emptying needed */ |
241 | int batch; /* chunk size for buddy add/remove */ | |
5f8dcc21 MG |
242 | |
243 | /* Lists of pages, one per migrate type stored on the pcp-lists */ | |
244 | struct list_head lists[MIGRATE_PCPTYPES]; | |
1da177e4 LT |
245 | }; |
246 | ||
247 | struct per_cpu_pageset { | |
3dfa5721 | 248 | struct per_cpu_pages pcp; |
4037d452 CL |
249 | #ifdef CONFIG_NUMA |
250 | s8 expire; | |
251 | #endif | |
2244b95a | 252 | #ifdef CONFIG_SMP |
df9ecaba | 253 | s8 stat_threshold; |
2244b95a CL |
254 | s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS]; |
255 | #endif | |
99dcc3e5 | 256 | }; |
e7c8d5c9 | 257 | |
97965478 CL |
258 | #endif /* !__GENERATING_BOUNDS.H */ |
259 | ||
2f1b6248 | 260 | enum zone_type { |
4b51d669 | 261 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 CL |
262 | /* |
263 | * ZONE_DMA is used when there are devices that are not able | |
264 | * to do DMA to all of addressable memory (ZONE_NORMAL). Then we | |
265 | * carve out the portion of memory that is needed for these devices. | |
266 | * The range is arch specific. | |
267 | * | |
268 | * Some examples | |
269 | * | |
270 | * Architecture Limit | |
271 | * --------------------------- | |
272 | * parisc, ia64, sparc <4G | |
273 | * s390 <2G | |
2f1b6248 CL |
274 | * arm Various |
275 | * alpha Unlimited or 0-16MB. | |
276 | * | |
277 | * i386, x86_64 and multiple other arches | |
278 | * <16M. | |
279 | */ | |
280 | ZONE_DMA, | |
4b51d669 | 281 | #endif |
fb0e7942 | 282 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 CL |
283 | /* |
284 | * x86_64 needs two ZONE_DMAs because it supports devices that are | |
285 | * only able to do DMA to the lower 16M but also 32 bit devices that | |
286 | * can only do DMA areas below 4G. | |
287 | */ | |
288 | ZONE_DMA32, | |
fb0e7942 | 289 | #endif |
2f1b6248 CL |
290 | /* |
291 | * Normal addressable memory is in ZONE_NORMAL. DMA operations can be | |
292 | * performed on pages in ZONE_NORMAL if the DMA devices support | |
293 | * transfers to all addressable memory. | |
294 | */ | |
295 | ZONE_NORMAL, | |
e53ef38d | 296 | #ifdef CONFIG_HIGHMEM |
2f1b6248 CL |
297 | /* |
298 | * A memory area that is only addressable by the kernel through | |
299 | * mapping portions into its own address space. This is for example | |
300 | * used by i386 to allow the kernel to address the memory beyond | |
301 | * 900MB. The kernel will set up special mappings (page | |
302 | * table entries on i386) for each page that the kernel needs to | |
303 | * access. | |
304 | */ | |
305 | ZONE_HIGHMEM, | |
e53ef38d | 306 | #endif |
2a1e274a | 307 | ZONE_MOVABLE, |
97965478 | 308 | __MAX_NR_ZONES |
2f1b6248 | 309 | }; |
1da177e4 | 310 | |
97965478 CL |
311 | #ifndef __GENERATING_BOUNDS_H |
312 | ||
1da177e4 LT |
313 | /* |
314 | * When a memory allocation must conform to specific limitations (such | |
315 | * as being suitable for DMA) the caller will pass in hints to the | |
316 | * allocator in the gfp_mask, in the zone modifier bits. These bits | |
317 | * are used to select a priority ordered list of memory zones which | |
19655d34 | 318 | * match the requested limits. See gfp_zone() in include/linux/gfp.h |
1da177e4 | 319 | */ |
fb0e7942 | 320 | |
97965478 | 321 | #if MAX_NR_ZONES < 2 |
4b51d669 | 322 | #define ZONES_SHIFT 0 |
97965478 | 323 | #elif MAX_NR_ZONES <= 2 |
19655d34 | 324 | #define ZONES_SHIFT 1 |
97965478 | 325 | #elif MAX_NR_ZONES <= 4 |
19655d34 | 326 | #define ZONES_SHIFT 2 |
4b51d669 CL |
327 | #else |
328 | #error ZONES_SHIFT -- too many zones configured adjust calculation | |
fb0e7942 | 329 | #endif |
1da177e4 | 330 | |
1da177e4 LT |
331 | struct zone { |
332 | /* Fields commonly accessed by the page allocator */ | |
41858966 MG |
333 | |
334 | /* zone watermarks, access with *_wmark_pages(zone) macros */ | |
335 | unsigned long watermark[NR_WMARK]; | |
336 | ||
aa454840 CL |
337 | /* |
338 | * When free pages are below this point, additional steps are taken | |
339 | * when reading the number of free pages to avoid per-cpu counter | |
340 | * drift allowing watermarks to be breached | |
341 | */ | |
342 | unsigned long percpu_drift_mark; | |
343 | ||
1da177e4 LT |
344 | /* |
345 | * We don't know if the memory that we're going to allocate will be freeable | |
346 | * or/and it will be released eventually, so to avoid totally wasting several | |
347 | * GB of ram we must reserve some of the lower zone memory (otherwise we risk | |
348 | * to run OOM on the lower zones despite there's tons of freeable ram | |
349 | * on the higher zones). This array is recalculated at runtime if the | |
350 | * sysctl_lowmem_reserve_ratio sysctl changes. | |
351 | */ | |
352 | unsigned long lowmem_reserve[MAX_NR_ZONES]; | |
353 | ||
ab8fabd4 JW |
354 | /* |
355 | * This is a per-zone reserve of pages that should not be | |
356 | * considered dirtyable memory. | |
357 | */ | |
358 | unsigned long dirty_balance_reserve; | |
359 | ||
e7c8d5c9 | 360 | #ifdef CONFIG_NUMA |
d5f541ed | 361 | int node; |
9614634f CL |
362 | /* |
363 | * zone reclaim becomes active if more unmapped pages exist. | |
364 | */ | |
8417bba4 | 365 | unsigned long min_unmapped_pages; |
0ff38490 | 366 | unsigned long min_slab_pages; |
e7c8d5c9 | 367 | #endif |
43cf38eb | 368 | struct per_cpu_pageset __percpu *pageset; |
1da177e4 LT |
369 | /* |
370 | * free areas of different sizes | |
371 | */ | |
372 | spinlock_t lock; | |
93e4a89a | 373 | int all_unreclaimable; /* All pages pinned */ |
bb13ffeb | 374 | #if defined CONFIG_COMPACTION || defined CONFIG_CMA |
62997027 MG |
375 | /* Set to true when the PG_migrate_skip bits should be cleared */ |
376 | bool compact_blockskip_flush; | |
c89511ab MG |
377 | |
378 | /* pfns where compaction scanners should start */ | |
379 | unsigned long compact_cached_free_pfn; | |
380 | unsigned long compact_cached_migrate_pfn; | |
bb13ffeb | 381 | #endif |
bdc8cb98 DH |
382 | #ifdef CONFIG_MEMORY_HOTPLUG |
383 | /* see spanned/present_pages for more description */ | |
384 | seqlock_t span_seqlock; | |
49f223a9 MS |
385 | #endif |
386 | #ifdef CONFIG_CMA | |
387 | /* | |
388 | * CMA needs to increase watermark levels during the allocation | |
389 | * process to make sure that the system is not starved. | |
390 | */ | |
391 | unsigned long min_cma_pages; | |
bdc8cb98 | 392 | #endif |
1da177e4 LT |
393 | struct free_area free_area[MAX_ORDER]; |
394 | ||
835c134e MG |
395 | #ifndef CONFIG_SPARSEMEM |
396 | /* | |
d9c23400 | 397 | * Flags for a pageblock_nr_pages block. See pageblock-flags.h. |
835c134e MG |
398 | * In SPARSEMEM, this map is stored in struct mem_section |
399 | */ | |
400 | unsigned long *pageblock_flags; | |
401 | #endif /* CONFIG_SPARSEMEM */ | |
402 | ||
4f92e258 MG |
403 | #ifdef CONFIG_COMPACTION |
404 | /* | |
405 | * On compaction failure, 1<<compact_defer_shift compactions | |
406 | * are skipped before trying again. The number attempted since | |
407 | * last failure is tracked with compact_considered. | |
408 | */ | |
409 | unsigned int compact_considered; | |
410 | unsigned int compact_defer_shift; | |
aff62249 | 411 | int compact_order_failed; |
4f92e258 | 412 | #endif |
1da177e4 LT |
413 | |
414 | ZONE_PADDING(_pad1_) | |
415 | ||
416 | /* Fields commonly accessed by the page reclaim scanner */ | |
6290df54 JW |
417 | spinlock_t lru_lock; |
418 | struct lruvec lruvec; | |
4f98a2fe | 419 | |
1da177e4 | 420 | unsigned long pages_scanned; /* since last reclaim */ |
e815af95 | 421 | unsigned long flags; /* zone flags, see below */ |
753ee728 | 422 | |
2244b95a CL |
423 | /* Zone statistics */ |
424 | atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS]; | |
9eeff239 | 425 | |
556adecb RR |
426 | /* |
427 | * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on | |
428 | * this zone's LRU. Maintained by the pageout code. | |
429 | */ | |
430 | unsigned int inactive_ratio; | |
431 | ||
1da177e4 LT |
432 | |
433 | ZONE_PADDING(_pad2_) | |
434 | /* Rarely used or read-mostly fields */ | |
435 | ||
436 | /* | |
437 | * wait_table -- the array holding the hash table | |
02b694de | 438 | * wait_table_hash_nr_entries -- the size of the hash table array |
1da177e4 LT |
439 | * wait_table_bits -- wait_table_size == (1 << wait_table_bits) |
440 | * | |
441 | * The purpose of all these is to keep track of the people | |
442 | * waiting for a page to become available and make them | |
443 | * runnable again when possible. The trouble is that this | |
444 | * consumes a lot of space, especially when so few things | |
445 | * wait on pages at a given time. So instead of using | |
446 | * per-page waitqueues, we use a waitqueue hash table. | |
447 | * | |
448 | * The bucket discipline is to sleep on the same queue when | |
449 | * colliding and wake all in that wait queue when removing. | |
450 | * When something wakes, it must check to be sure its page is | |
451 | * truly available, a la thundering herd. The cost of a | |
452 | * collision is great, but given the expected load of the | |
453 | * table, they should be so rare as to be outweighed by the | |
454 | * benefits from the saved space. | |
455 | * | |
456 | * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the | |
457 | * primary users of these fields, and in mm/page_alloc.c | |
458 | * free_area_init_core() performs the initialization of them. | |
459 | */ | |
460 | wait_queue_head_t * wait_table; | |
02b694de | 461 | unsigned long wait_table_hash_nr_entries; |
1da177e4 LT |
462 | unsigned long wait_table_bits; |
463 | ||
464 | /* | |
465 | * Discontig memory support fields. | |
466 | */ | |
467 | struct pglist_data *zone_pgdat; | |
1da177e4 LT |
468 | /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */ |
469 | unsigned long zone_start_pfn; | |
470 | ||
bdc8cb98 DH |
471 | /* |
472 | * zone_start_pfn, spanned_pages and present_pages are all | |
473 | * protected by span_seqlock. It is a seqlock because it has | |
474 | * to be read outside of zone->lock, and it is done in the main | |
475 | * allocator path. But, it is written quite infrequently. | |
476 | * | |
477 | * The lock is declared along with zone->lock because it is | |
478 | * frequently read in proximity to zone->lock. It's good to | |
479 | * give them a chance of being in the same cacheline. | |
480 | */ | |
1da177e4 LT |
481 | unsigned long spanned_pages; /* total size, including holes */ |
482 | unsigned long present_pages; /* amount of memory (excluding holes) */ | |
483 | ||
484 | /* | |
485 | * rarely used fields: | |
486 | */ | |
15ad7cdc | 487 | const char *name; |
702d1a6e MK |
488 | #ifdef CONFIG_MEMORY_ISOLATION |
489 | /* | |
490 | * the number of MIGRATE_ISOLATE *pageblock*. | |
491 | * We need this for free page counting. Look at zone_watermark_ok_safe. | |
492 | * It's protected by zone->lock | |
493 | */ | |
494 | int nr_pageblock_isolate; | |
495 | #endif | |
22fc6ecc | 496 | } ____cacheline_internodealigned_in_smp; |
1da177e4 | 497 | |
e815af95 | 498 | typedef enum { |
e815af95 | 499 | ZONE_RECLAIM_LOCKED, /* prevents concurrent reclaim */ |
098d7f12 | 500 | ZONE_OOM_LOCKED, /* zone is in OOM killer zonelist */ |
0e093d99 MG |
501 | ZONE_CONGESTED, /* zone has many dirty pages backed by |
502 | * a congested BDI | |
503 | */ | |
e815af95 DR |
504 | } zone_flags_t; |
505 | ||
506 | static inline void zone_set_flag(struct zone *zone, zone_flags_t flag) | |
507 | { | |
508 | set_bit(flag, &zone->flags); | |
509 | } | |
d773ed6b DR |
510 | |
511 | static inline int zone_test_and_set_flag(struct zone *zone, zone_flags_t flag) | |
512 | { | |
513 | return test_and_set_bit(flag, &zone->flags); | |
514 | } | |
515 | ||
e815af95 DR |
516 | static inline void zone_clear_flag(struct zone *zone, zone_flags_t flag) |
517 | { | |
518 | clear_bit(flag, &zone->flags); | |
519 | } | |
520 | ||
0e093d99 MG |
521 | static inline int zone_is_reclaim_congested(const struct zone *zone) |
522 | { | |
523 | return test_bit(ZONE_CONGESTED, &zone->flags); | |
524 | } | |
525 | ||
e815af95 DR |
526 | static inline int zone_is_reclaim_locked(const struct zone *zone) |
527 | { | |
528 | return test_bit(ZONE_RECLAIM_LOCKED, &zone->flags); | |
529 | } | |
d773ed6b | 530 | |
098d7f12 DR |
531 | static inline int zone_is_oom_locked(const struct zone *zone) |
532 | { | |
533 | return test_bit(ZONE_OOM_LOCKED, &zone->flags); | |
534 | } | |
e815af95 | 535 | |
1da177e4 LT |
536 | /* |
537 | * The "priority" of VM scanning is how much of the queues we will scan in one | |
538 | * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the | |
539 | * queues ("queue_length >> 12") during an aging round. | |
540 | */ | |
541 | #define DEF_PRIORITY 12 | |
542 | ||
9276b1bc PJ |
543 | /* Maximum number of zones on a zonelist */ |
544 | #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES) | |
545 | ||
546 | #ifdef CONFIG_NUMA | |
523b9458 CL |
547 | |
548 | /* | |
25a64ec1 | 549 | * The NUMA zonelists are doubled because we need zonelists that restrict the |
523b9458 CL |
550 | * allocations to a single node for GFP_THISNODE. |
551 | * | |
54a6eb5c MG |
552 | * [0] : Zonelist with fallback |
553 | * [1] : No fallback (GFP_THISNODE) | |
523b9458 | 554 | */ |
54a6eb5c | 555 | #define MAX_ZONELISTS 2 |
523b9458 CL |
556 | |
557 | ||
9276b1bc PJ |
558 | /* |
559 | * We cache key information from each zonelist for smaller cache | |
560 | * footprint when scanning for free pages in get_page_from_freelist(). | |
561 | * | |
562 | * 1) The BITMAP fullzones tracks which zones in a zonelist have come | |
563 | * up short of free memory since the last time (last_fullzone_zap) | |
564 | * we zero'd fullzones. | |
565 | * 2) The array z_to_n[] maps each zone in the zonelist to its node | |
566 | * id, so that we can efficiently evaluate whether that node is | |
567 | * set in the current tasks mems_allowed. | |
568 | * | |
569 | * Both fullzones and z_to_n[] are one-to-one with the zonelist, | |
570 | * indexed by a zones offset in the zonelist zones[] array. | |
571 | * | |
572 | * The get_page_from_freelist() routine does two scans. During the | |
573 | * first scan, we skip zones whose corresponding bit in 'fullzones' | |
574 | * is set or whose corresponding node in current->mems_allowed (which | |
575 | * comes from cpusets) is not set. During the second scan, we bypass | |
576 | * this zonelist_cache, to ensure we look methodically at each zone. | |
577 | * | |
578 | * Once per second, we zero out (zap) fullzones, forcing us to | |
579 | * reconsider nodes that might have regained more free memory. | |
580 | * The field last_full_zap is the time we last zapped fullzones. | |
581 | * | |
582 | * This mechanism reduces the amount of time we waste repeatedly | |
583 | * reexaming zones for free memory when they just came up low on | |
584 | * memory momentarilly ago. | |
585 | * | |
586 | * The zonelist_cache struct members logically belong in struct | |
587 | * zonelist. However, the mempolicy zonelists constructed for | |
588 | * MPOL_BIND are intentionally variable length (and usually much | |
589 | * shorter). A general purpose mechanism for handling structs with | |
590 | * multiple variable length members is more mechanism than we want | |
591 | * here. We resort to some special case hackery instead. | |
592 | * | |
593 | * The MPOL_BIND zonelists don't need this zonelist_cache (in good | |
594 | * part because they are shorter), so we put the fixed length stuff | |
595 | * at the front of the zonelist struct, ending in a variable length | |
596 | * zones[], as is needed by MPOL_BIND. | |
597 | * | |
598 | * Then we put the optional zonelist cache on the end of the zonelist | |
599 | * struct. This optional stuff is found by a 'zlcache_ptr' pointer in | |
600 | * the fixed length portion at the front of the struct. This pointer | |
601 | * both enables us to find the zonelist cache, and in the case of | |
602 | * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL) | |
603 | * to know that the zonelist cache is not there. | |
604 | * | |
605 | * The end result is that struct zonelists come in two flavors: | |
606 | * 1) The full, fixed length version, shown below, and | |
607 | * 2) The custom zonelists for MPOL_BIND. | |
608 | * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache. | |
609 | * | |
610 | * Even though there may be multiple CPU cores on a node modifying | |
611 | * fullzones or last_full_zap in the same zonelist_cache at the same | |
612 | * time, we don't lock it. This is just hint data - if it is wrong now | |
613 | * and then, the allocator will still function, perhaps a bit slower. | |
614 | */ | |
615 | ||
616 | ||
617 | struct zonelist_cache { | |
9276b1bc | 618 | unsigned short z_to_n[MAX_ZONES_PER_ZONELIST]; /* zone->nid */ |
7253f4ef | 619 | DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST); /* zone full? */ |
9276b1bc PJ |
620 | unsigned long last_full_zap; /* when last zap'd (jiffies) */ |
621 | }; | |
622 | #else | |
54a6eb5c | 623 | #define MAX_ZONELISTS 1 |
9276b1bc PJ |
624 | struct zonelist_cache; |
625 | #endif | |
626 | ||
dd1a239f MG |
627 | /* |
628 | * This struct contains information about a zone in a zonelist. It is stored | |
629 | * here to avoid dereferences into large structures and lookups of tables | |
630 | */ | |
631 | struct zoneref { | |
632 | struct zone *zone; /* Pointer to actual zone */ | |
633 | int zone_idx; /* zone_idx(zoneref->zone) */ | |
634 | }; | |
635 | ||
1da177e4 LT |
636 | /* |
637 | * One allocation request operates on a zonelist. A zonelist | |
638 | * is a list of zones, the first one is the 'goal' of the | |
639 | * allocation, the other zones are fallback zones, in decreasing | |
640 | * priority. | |
641 | * | |
9276b1bc PJ |
642 | * If zlcache_ptr is not NULL, then it is just the address of zlcache, |
643 | * as explained above. If zlcache_ptr is NULL, there is no zlcache. | |
dd1a239f MG |
644 | * * |
645 | * To speed the reading of the zonelist, the zonerefs contain the zone index | |
646 | * of the entry being read. Helper functions to access information given | |
647 | * a struct zoneref are | |
648 | * | |
649 | * zonelist_zone() - Return the struct zone * for an entry in _zonerefs | |
650 | * zonelist_zone_idx() - Return the index of the zone for an entry | |
651 | * zonelist_node_idx() - Return the index of the node for an entry | |
1da177e4 LT |
652 | */ |
653 | struct zonelist { | |
9276b1bc | 654 | struct zonelist_cache *zlcache_ptr; // NULL or &zlcache |
dd1a239f | 655 | struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1]; |
9276b1bc PJ |
656 | #ifdef CONFIG_NUMA |
657 | struct zonelist_cache zlcache; // optional ... | |
658 | #endif | |
1da177e4 LT |
659 | }; |
660 | ||
0ee332c1 | 661 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
c713216d MG |
662 | struct node_active_region { |
663 | unsigned long start_pfn; | |
664 | unsigned long end_pfn; | |
665 | int nid; | |
666 | }; | |
0ee332c1 | 667 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
1da177e4 | 668 | |
5b99cd0e HC |
669 | #ifndef CONFIG_DISCONTIGMEM |
670 | /* The array of struct pages - for discontigmem use pgdat->lmem_map */ | |
671 | extern struct page *mem_map; | |
672 | #endif | |
673 | ||
1da177e4 LT |
674 | /* |
675 | * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM | |
676 | * (mostly NUMA machines?) to denote a higher-level memory zone than the | |
677 | * zone denotes. | |
678 | * | |
679 | * On NUMA machines, each NUMA node would have a pg_data_t to describe | |
680 | * it's memory layout. | |
681 | * | |
682 | * Memory statistics and page replacement data structures are maintained on a | |
683 | * per-zone basis. | |
684 | */ | |
685 | struct bootmem_data; | |
686 | typedef struct pglist_data { | |
687 | struct zone node_zones[MAX_NR_ZONES]; | |
523b9458 | 688 | struct zonelist node_zonelists[MAX_ZONELISTS]; |
1da177e4 | 689 | int nr_zones; |
52d4b9ac | 690 | #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */ |
1da177e4 | 691 | struct page *node_mem_map; |
c255a458 | 692 | #ifdef CONFIG_MEMCG |
52d4b9ac KH |
693 | struct page_cgroup *node_page_cgroup; |
694 | #endif | |
d41dee36 | 695 | #endif |
08677214 | 696 | #ifndef CONFIG_NO_BOOTMEM |
1da177e4 | 697 | struct bootmem_data *bdata; |
08677214 | 698 | #endif |
208d54e5 DH |
699 | #ifdef CONFIG_MEMORY_HOTPLUG |
700 | /* | |
701 | * Must be held any time you expect node_start_pfn, node_present_pages | |
702 | * or node_spanned_pages stay constant. Holding this will also | |
703 | * guarantee that any pfn_valid() stays that way. | |
704 | * | |
705 | * Nests above zone->lock and zone->size_seqlock. | |
706 | */ | |
707 | spinlock_t node_size_lock; | |
708 | #endif | |
1da177e4 LT |
709 | unsigned long node_start_pfn; |
710 | unsigned long node_present_pages; /* total number of physical pages */ | |
711 | unsigned long node_spanned_pages; /* total size of physical page | |
712 | range, including holes */ | |
713 | int node_id; | |
957f822a | 714 | nodemask_t reclaim_nodes; /* Nodes allowed to reclaim from */ |
1da177e4 | 715 | wait_queue_head_t kswapd_wait; |
5515061d | 716 | wait_queue_head_t pfmemalloc_wait; |
d8adde17 | 717 | struct task_struct *kswapd; /* Protected by lock_memory_hotplug() */ |
1da177e4 | 718 | int kswapd_max_order; |
99504748 | 719 | enum zone_type classzone_idx; |
1da177e4 LT |
720 | } pg_data_t; |
721 | ||
722 | #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages) | |
723 | #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages) | |
d41dee36 | 724 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
408fde81 | 725 | #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr)) |
d41dee36 AW |
726 | #else |
727 | #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr)) | |
728 | #endif | |
408fde81 | 729 | #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr)) |
1da177e4 | 730 | |
c6830c22 KH |
731 | #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn) |
732 | ||
733 | #define node_end_pfn(nid) ({\ | |
734 | pg_data_t *__pgdat = NODE_DATA(nid);\ | |
735 | __pgdat->node_start_pfn + __pgdat->node_spanned_pages;\ | |
736 | }) | |
737 | ||
208d54e5 DH |
738 | #include <linux/memory_hotplug.h> |
739 | ||
4eaf3f64 | 740 | extern struct mutex zonelists_mutex; |
9adb62a5 | 741 | void build_all_zonelists(pg_data_t *pgdat, struct zone *zone); |
99504748 | 742 | void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx); |
88f5acf8 MG |
743 | bool zone_watermark_ok(struct zone *z, int order, unsigned long mark, |
744 | int classzone_idx, int alloc_flags); | |
745 | bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark, | |
7fb1d9fc | 746 | int classzone_idx, int alloc_flags); |
a2f3aa02 DH |
747 | enum memmap_context { |
748 | MEMMAP_EARLY, | |
749 | MEMMAP_HOTPLUG, | |
750 | }; | |
718127cc | 751 | extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn, |
a2f3aa02 DH |
752 | unsigned long size, |
753 | enum memmap_context context); | |
718127cc | 754 | |
bea8c150 | 755 | extern void lruvec_init(struct lruvec *lruvec); |
7f5e86c2 KK |
756 | |
757 | static inline struct zone *lruvec_zone(struct lruvec *lruvec) | |
758 | { | |
c255a458 | 759 | #ifdef CONFIG_MEMCG |
7f5e86c2 KK |
760 | return lruvec->zone; |
761 | #else | |
762 | return container_of(lruvec, struct zone, lruvec); | |
763 | #endif | |
764 | } | |
765 | ||
1da177e4 LT |
766 | #ifdef CONFIG_HAVE_MEMORY_PRESENT |
767 | void memory_present(int nid, unsigned long start, unsigned long end); | |
768 | #else | |
769 | static inline void memory_present(int nid, unsigned long start, unsigned long end) {} | |
770 | #endif | |
771 | ||
7aac7898 LS |
772 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
773 | int local_memory_node(int node_id); | |
774 | #else | |
775 | static inline int local_memory_node(int node_id) { return node_id; }; | |
776 | #endif | |
777 | ||
1da177e4 LT |
778 | #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE |
779 | unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long); | |
780 | #endif | |
781 | ||
782 | /* | |
783 | * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc. | |
784 | */ | |
785 | #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones) | |
786 | ||
f3fe6512 CK |
787 | static inline int populated_zone(struct zone *zone) |
788 | { | |
789 | return (!!zone->present_pages); | |
790 | } | |
791 | ||
2a1e274a MG |
792 | extern int movable_zone; |
793 | ||
794 | static inline int zone_movable_is_highmem(void) | |
795 | { | |
fe03025d | 796 | #if defined(CONFIG_HIGHMEM) && defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) |
2a1e274a MG |
797 | return movable_zone == ZONE_HIGHMEM; |
798 | #else | |
799 | return 0; | |
800 | #endif | |
801 | } | |
802 | ||
2f1b6248 | 803 | static inline int is_highmem_idx(enum zone_type idx) |
1da177e4 | 804 | { |
e53ef38d | 805 | #ifdef CONFIG_HIGHMEM |
2a1e274a MG |
806 | return (idx == ZONE_HIGHMEM || |
807 | (idx == ZONE_MOVABLE && zone_movable_is_highmem())); | |
e53ef38d CL |
808 | #else |
809 | return 0; | |
810 | #endif | |
1da177e4 LT |
811 | } |
812 | ||
2f1b6248 | 813 | static inline int is_normal_idx(enum zone_type idx) |
1da177e4 LT |
814 | { |
815 | return (idx == ZONE_NORMAL); | |
816 | } | |
9328b8fa | 817 | |
1da177e4 LT |
818 | /** |
819 | * is_highmem - helper function to quickly check if a struct zone is a | |
820 | * highmem zone or not. This is an attempt to keep references | |
821 | * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum. | |
822 | * @zone - pointer to struct zone variable | |
823 | */ | |
824 | static inline int is_highmem(struct zone *zone) | |
825 | { | |
e53ef38d | 826 | #ifdef CONFIG_HIGHMEM |
ddc81ed2 HH |
827 | int zone_off = (char *)zone - (char *)zone->zone_pgdat->node_zones; |
828 | return zone_off == ZONE_HIGHMEM * sizeof(*zone) || | |
829 | (zone_off == ZONE_MOVABLE * sizeof(*zone) && | |
830 | zone_movable_is_highmem()); | |
e53ef38d CL |
831 | #else |
832 | return 0; | |
833 | #endif | |
1da177e4 LT |
834 | } |
835 | ||
836 | static inline int is_normal(struct zone *zone) | |
837 | { | |
838 | return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL; | |
839 | } | |
840 | ||
9328b8fa NP |
841 | static inline int is_dma32(struct zone *zone) |
842 | { | |
fb0e7942 | 843 | #ifdef CONFIG_ZONE_DMA32 |
9328b8fa | 844 | return zone == zone->zone_pgdat->node_zones + ZONE_DMA32; |
fb0e7942 CL |
845 | #else |
846 | return 0; | |
847 | #endif | |
9328b8fa NP |
848 | } |
849 | ||
850 | static inline int is_dma(struct zone *zone) | |
851 | { | |
4b51d669 | 852 | #ifdef CONFIG_ZONE_DMA |
9328b8fa | 853 | return zone == zone->zone_pgdat->node_zones + ZONE_DMA; |
4b51d669 CL |
854 | #else |
855 | return 0; | |
856 | #endif | |
9328b8fa NP |
857 | } |
858 | ||
1da177e4 LT |
859 | /* These two functions are used to setup the per zone pages min values */ |
860 | struct ctl_table; | |
8d65af78 | 861 | int min_free_kbytes_sysctl_handler(struct ctl_table *, int, |
1da177e4 LT |
862 | void __user *, size_t *, loff_t *); |
863 | extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1]; | |
8d65af78 | 864 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, |
1da177e4 | 865 | void __user *, size_t *, loff_t *); |
8d65af78 | 866 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, |
8ad4b1fb | 867 | void __user *, size_t *, loff_t *); |
9614634f | 868 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int, |
8d65af78 | 869 | void __user *, size_t *, loff_t *); |
0ff38490 | 870 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int, |
8d65af78 | 871 | void __user *, size_t *, loff_t *); |
1da177e4 | 872 | |
f0c0b2b8 | 873 | extern int numa_zonelist_order_handler(struct ctl_table *, int, |
8d65af78 | 874 | void __user *, size_t *, loff_t *); |
f0c0b2b8 KH |
875 | extern char numa_zonelist_order[]; |
876 | #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */ | |
877 | ||
93b7504e | 878 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
879 | |
880 | extern struct pglist_data contig_page_data; | |
881 | #define NODE_DATA(nid) (&contig_page_data) | |
882 | #define NODE_MEM_MAP(nid) mem_map | |
1da177e4 | 883 | |
93b7504e | 884 | #else /* CONFIG_NEED_MULTIPLE_NODES */ |
1da177e4 LT |
885 | |
886 | #include <asm/mmzone.h> | |
887 | ||
93b7504e | 888 | #endif /* !CONFIG_NEED_MULTIPLE_NODES */ |
348f8b6c | 889 | |
95144c78 KH |
890 | extern struct pglist_data *first_online_pgdat(void); |
891 | extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat); | |
892 | extern struct zone *next_zone(struct zone *zone); | |
8357f869 KH |
893 | |
894 | /** | |
12d15f0d | 895 | * for_each_online_pgdat - helper macro to iterate over all online nodes |
8357f869 KH |
896 | * @pgdat - pointer to a pg_data_t variable |
897 | */ | |
898 | #define for_each_online_pgdat(pgdat) \ | |
899 | for (pgdat = first_online_pgdat(); \ | |
900 | pgdat; \ | |
901 | pgdat = next_online_pgdat(pgdat)) | |
8357f869 KH |
902 | /** |
903 | * for_each_zone - helper macro to iterate over all memory zones | |
904 | * @zone - pointer to struct zone variable | |
905 | * | |
906 | * The user only needs to declare the zone variable, for_each_zone | |
907 | * fills it in. | |
908 | */ | |
909 | #define for_each_zone(zone) \ | |
910 | for (zone = (first_online_pgdat())->node_zones; \ | |
911 | zone; \ | |
912 | zone = next_zone(zone)) | |
913 | ||
ee99c71c KM |
914 | #define for_each_populated_zone(zone) \ |
915 | for (zone = (first_online_pgdat())->node_zones; \ | |
916 | zone; \ | |
917 | zone = next_zone(zone)) \ | |
918 | if (!populated_zone(zone)) \ | |
919 | ; /* do nothing */ \ | |
920 | else | |
921 | ||
dd1a239f MG |
922 | static inline struct zone *zonelist_zone(struct zoneref *zoneref) |
923 | { | |
924 | return zoneref->zone; | |
925 | } | |
926 | ||
927 | static inline int zonelist_zone_idx(struct zoneref *zoneref) | |
928 | { | |
929 | return zoneref->zone_idx; | |
930 | } | |
931 | ||
932 | static inline int zonelist_node_idx(struct zoneref *zoneref) | |
933 | { | |
934 | #ifdef CONFIG_NUMA | |
935 | /* zone_to_nid not available in this context */ | |
936 | return zoneref->zone->node; | |
937 | #else | |
938 | return 0; | |
939 | #endif /* CONFIG_NUMA */ | |
940 | } | |
941 | ||
19770b32 MG |
942 | /** |
943 | * 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 | |
944 | * @z - The cursor used as a starting point for the search | |
945 | * @highest_zoneidx - The zone index of the highest zone to return | |
946 | * @nodes - An optional nodemask to filter the zonelist with | |
947 | * @zone - The first suitable zone found is returned via this parameter | |
948 | * | |
949 | * This function returns the next zone at or below a given zone index that is | |
950 | * within the allowed nodemask using a cursor as the starting point for the | |
5bead2a0 MG |
951 | * search. The zoneref returned is a cursor that represents the current zone |
952 | * being examined. It should be advanced by one before calling | |
953 | * next_zones_zonelist again. | |
19770b32 MG |
954 | */ |
955 | struct zoneref *next_zones_zonelist(struct zoneref *z, | |
956 | enum zone_type highest_zoneidx, | |
957 | nodemask_t *nodes, | |
958 | struct zone **zone); | |
dd1a239f | 959 | |
19770b32 MG |
960 | /** |
961 | * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist | |
962 | * @zonelist - The zonelist to search for a suitable zone | |
963 | * @highest_zoneidx - The zone index of the highest zone to return | |
964 | * @nodes - An optional nodemask to filter the zonelist with | |
965 | * @zone - The first suitable zone found is returned via this parameter | |
966 | * | |
967 | * This function returns the first zone at or below a given zone index that is | |
968 | * within the allowed nodemask. The zoneref returned is a cursor that can be | |
5bead2a0 MG |
969 | * used to iterate the zonelist with next_zones_zonelist by advancing it by |
970 | * one before calling. | |
19770b32 | 971 | */ |
dd1a239f | 972 | static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist, |
19770b32 MG |
973 | enum zone_type highest_zoneidx, |
974 | nodemask_t *nodes, | |
975 | struct zone **zone) | |
54a6eb5c | 976 | { |
19770b32 MG |
977 | return next_zones_zonelist(zonelist->_zonerefs, highest_zoneidx, nodes, |
978 | zone); | |
54a6eb5c MG |
979 | } |
980 | ||
19770b32 MG |
981 | /** |
982 | * 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 | |
983 | * @zone - The current zone in the iterator | |
984 | * @z - The current pointer within zonelist->zones being iterated | |
985 | * @zlist - The zonelist being iterated | |
986 | * @highidx - The zone index of the highest zone to return | |
987 | * @nodemask - Nodemask allowed by the allocator | |
988 | * | |
989 | * This iterator iterates though all zones at or below a given zone index and | |
990 | * within a given nodemask | |
991 | */ | |
992 | #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \ | |
993 | for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \ | |
994 | zone; \ | |
5bead2a0 | 995 | z = next_zones_zonelist(++z, highidx, nodemask, &zone)) \ |
54a6eb5c MG |
996 | |
997 | /** | |
998 | * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index | |
999 | * @zone - The current zone in the iterator | |
1000 | * @z - The current pointer within zonelist->zones being iterated | |
1001 | * @zlist - The zonelist being iterated | |
1002 | * @highidx - The zone index of the highest zone to return | |
1003 | * | |
1004 | * This iterator iterates though all zones at or below a given zone index. | |
1005 | */ | |
1006 | #define for_each_zone_zonelist(zone, z, zlist, highidx) \ | |
19770b32 | 1007 | for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL) |
54a6eb5c | 1008 | |
d41dee36 AW |
1009 | #ifdef CONFIG_SPARSEMEM |
1010 | #include <asm/sparsemem.h> | |
1011 | #endif | |
1012 | ||
c713216d | 1013 | #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \ |
0ee332c1 | 1014 | !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) |
b4544568 AM |
1015 | static inline unsigned long early_pfn_to_nid(unsigned long pfn) |
1016 | { | |
1017 | return 0; | |
1018 | } | |
b159d43f AW |
1019 | #endif |
1020 | ||
2bdaf115 AW |
1021 | #ifdef CONFIG_FLATMEM |
1022 | #define pfn_to_nid(pfn) (0) | |
1023 | #endif | |
1024 | ||
d41dee36 AW |
1025 | #ifdef CONFIG_SPARSEMEM |
1026 | ||
1027 | /* | |
1028 | * SECTION_SHIFT #bits space required to store a section # | |
1029 | * | |
1030 | * PA_SECTION_SHIFT physical address to/from section number | |
1031 | * PFN_SECTION_SHIFT pfn to/from section number | |
1032 | */ | |
1033 | #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS) | |
1034 | ||
1035 | #define PA_SECTION_SHIFT (SECTION_SIZE_BITS) | |
1036 | #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT) | |
1037 | ||
1038 | #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT) | |
1039 | ||
1040 | #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT) | |
1041 | #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1)) | |
1042 | ||
835c134e | 1043 | #define SECTION_BLOCKFLAGS_BITS \ |
d9c23400 | 1044 | ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS) |
835c134e | 1045 | |
d41dee36 AW |
1046 | #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS |
1047 | #error Allocator MAX_ORDER exceeds SECTION_SIZE | |
1048 | #endif | |
1049 | ||
e3c40f37 DK |
1050 | #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT) |
1051 | #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT) | |
1052 | ||
a539f353 DK |
1053 | #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK) |
1054 | #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK) | |
1055 | ||
d41dee36 | 1056 | struct page; |
52d4b9ac | 1057 | struct page_cgroup; |
d41dee36 | 1058 | struct mem_section { |
29751f69 AW |
1059 | /* |
1060 | * This is, logically, a pointer to an array of struct | |
1061 | * pages. However, it is stored with some other magic. | |
1062 | * (see sparse.c::sparse_init_one_section()) | |
1063 | * | |
30c253e6 AW |
1064 | * Additionally during early boot we encode node id of |
1065 | * the location of the section here to guide allocation. | |
1066 | * (see sparse.c::memory_present()) | |
1067 | * | |
29751f69 AW |
1068 | * Making it a UL at least makes someone do a cast |
1069 | * before using it wrong. | |
1070 | */ | |
1071 | unsigned long section_mem_map; | |
5c0e3066 MG |
1072 | |
1073 | /* See declaration of similar field in struct zone */ | |
1074 | unsigned long *pageblock_flags; | |
c255a458 | 1075 | #ifdef CONFIG_MEMCG |
52d4b9ac KH |
1076 | /* |
1077 | * If !SPARSEMEM, pgdat doesn't have page_cgroup pointer. We use | |
1078 | * section. (see memcontrol.h/page_cgroup.h about this.) | |
1079 | */ | |
1080 | struct page_cgroup *page_cgroup; | |
1081 | unsigned long pad; | |
1082 | #endif | |
d41dee36 AW |
1083 | }; |
1084 | ||
3e347261 BP |
1085 | #ifdef CONFIG_SPARSEMEM_EXTREME |
1086 | #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section)) | |
1087 | #else | |
1088 | #define SECTIONS_PER_ROOT 1 | |
1089 | #endif | |
802f192e | 1090 | |
3e347261 | 1091 | #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT) |
0faa5638 | 1092 | #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT) |
3e347261 | 1093 | #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1) |
802f192e | 1094 | |
3e347261 BP |
1095 | #ifdef CONFIG_SPARSEMEM_EXTREME |
1096 | extern struct mem_section *mem_section[NR_SECTION_ROOTS]; | |
802f192e | 1097 | #else |
3e347261 BP |
1098 | extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]; |
1099 | #endif | |
d41dee36 | 1100 | |
29751f69 AW |
1101 | static inline struct mem_section *__nr_to_section(unsigned long nr) |
1102 | { | |
3e347261 BP |
1103 | if (!mem_section[SECTION_NR_TO_ROOT(nr)]) |
1104 | return NULL; | |
1105 | return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK]; | |
29751f69 | 1106 | } |
4ca644d9 | 1107 | extern int __section_nr(struct mem_section* ms); |
04753278 | 1108 | extern unsigned long usemap_size(void); |
29751f69 AW |
1109 | |
1110 | /* | |
1111 | * We use the lower bits of the mem_map pointer to store | |
1112 | * a little bit of information. There should be at least | |
1113 | * 3 bits here due to 32-bit alignment. | |
1114 | */ | |
1115 | #define SECTION_MARKED_PRESENT (1UL<<0) | |
1116 | #define SECTION_HAS_MEM_MAP (1UL<<1) | |
1117 | #define SECTION_MAP_LAST_BIT (1UL<<2) | |
1118 | #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1)) | |
30c253e6 | 1119 | #define SECTION_NID_SHIFT 2 |
29751f69 AW |
1120 | |
1121 | static inline struct page *__section_mem_map_addr(struct mem_section *section) | |
1122 | { | |
1123 | unsigned long map = section->section_mem_map; | |
1124 | map &= SECTION_MAP_MASK; | |
1125 | return (struct page *)map; | |
1126 | } | |
1127 | ||
540557b9 | 1128 | static inline int present_section(struct mem_section *section) |
29751f69 | 1129 | { |
802f192e | 1130 | return (section && (section->section_mem_map & SECTION_MARKED_PRESENT)); |
29751f69 AW |
1131 | } |
1132 | ||
540557b9 AW |
1133 | static inline int present_section_nr(unsigned long nr) |
1134 | { | |
1135 | return present_section(__nr_to_section(nr)); | |
1136 | } | |
1137 | ||
1138 | static inline int valid_section(struct mem_section *section) | |
29751f69 | 1139 | { |
802f192e | 1140 | return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP)); |
29751f69 AW |
1141 | } |
1142 | ||
1143 | static inline int valid_section_nr(unsigned long nr) | |
1144 | { | |
1145 | return valid_section(__nr_to_section(nr)); | |
1146 | } | |
1147 | ||
d41dee36 AW |
1148 | static inline struct mem_section *__pfn_to_section(unsigned long pfn) |
1149 | { | |
29751f69 | 1150 | return __nr_to_section(pfn_to_section_nr(pfn)); |
d41dee36 AW |
1151 | } |
1152 | ||
7b7bf499 | 1153 | #ifndef CONFIG_HAVE_ARCH_PFN_VALID |
d41dee36 AW |
1154 | static inline int pfn_valid(unsigned long pfn) |
1155 | { | |
1156 | if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) | |
1157 | return 0; | |
29751f69 | 1158 | return valid_section(__nr_to_section(pfn_to_section_nr(pfn))); |
d41dee36 | 1159 | } |
7b7bf499 | 1160 | #endif |
d41dee36 | 1161 | |
540557b9 AW |
1162 | static inline int pfn_present(unsigned long pfn) |
1163 | { | |
1164 | if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) | |
1165 | return 0; | |
1166 | return present_section(__nr_to_section(pfn_to_section_nr(pfn))); | |
1167 | } | |
1168 | ||
d41dee36 AW |
1169 | /* |
1170 | * These are _only_ used during initialisation, therefore they | |
1171 | * can use __initdata ... They could have names to indicate | |
1172 | * this restriction. | |
1173 | */ | |
1174 | #ifdef CONFIG_NUMA | |
161599ff AW |
1175 | #define pfn_to_nid(pfn) \ |
1176 | ({ \ | |
1177 | unsigned long __pfn_to_nid_pfn = (pfn); \ | |
1178 | page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \ | |
1179 | }) | |
2bdaf115 AW |
1180 | #else |
1181 | #define pfn_to_nid(pfn) (0) | |
d41dee36 AW |
1182 | #endif |
1183 | ||
d41dee36 AW |
1184 | #define early_pfn_valid(pfn) pfn_valid(pfn) |
1185 | void sparse_init(void); | |
1186 | #else | |
1187 | #define sparse_init() do {} while (0) | |
28ae55c9 | 1188 | #define sparse_index_init(_sec, _nid) do {} while (0) |
d41dee36 AW |
1189 | #endif /* CONFIG_SPARSEMEM */ |
1190 | ||
75167957 | 1191 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
cc2559bc | 1192 | bool early_pfn_in_nid(unsigned long pfn, int nid); |
75167957 AW |
1193 | #else |
1194 | #define early_pfn_in_nid(pfn, nid) (1) | |
1195 | #endif | |
1196 | ||
d41dee36 AW |
1197 | #ifndef early_pfn_valid |
1198 | #define early_pfn_valid(pfn) (1) | |
1199 | #endif | |
1200 | ||
1201 | void memory_present(int nid, unsigned long start, unsigned long end); | |
1202 | unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long); | |
1203 | ||
14e07298 AW |
1204 | /* |
1205 | * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we | |
1206 | * need to check pfn validility within that MAX_ORDER_NR_PAGES block. | |
1207 | * pfn_valid_within() should be used in this case; we optimise this away | |
1208 | * when we have no holes within a MAX_ORDER_NR_PAGES block. | |
1209 | */ | |
1210 | #ifdef CONFIG_HOLES_IN_ZONE | |
1211 | #define pfn_valid_within(pfn) pfn_valid(pfn) | |
1212 | #else | |
1213 | #define pfn_valid_within(pfn) (1) | |
1214 | #endif | |
1215 | ||
eb33575c MG |
1216 | #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL |
1217 | /* | |
1218 | * pfn_valid() is meant to be able to tell if a given PFN has valid memmap | |
1219 | * associated with it or not. In FLATMEM, it is expected that holes always | |
1220 | * have valid memmap as long as there is valid PFNs either side of the hole. | |
1221 | * In SPARSEMEM, it is assumed that a valid section has a memmap for the | |
1222 | * entire section. | |
1223 | * | |
1224 | * However, an ARM, and maybe other embedded architectures in the future | |
1225 | * free memmap backing holes to save memory on the assumption the memmap is | |
1226 | * never used. The page_zone linkages are then broken even though pfn_valid() | |
1227 | * returns true. A walker of the full memmap must then do this additional | |
1228 | * check to ensure the memmap they are looking at is sane by making sure | |
1229 | * the zone and PFN linkages are still valid. This is expensive, but walkers | |
1230 | * of the full memmap are extremely rare. | |
1231 | */ | |
1232 | int memmap_valid_within(unsigned long pfn, | |
1233 | struct page *page, struct zone *zone); | |
1234 | #else | |
1235 | static inline int memmap_valid_within(unsigned long pfn, | |
1236 | struct page *page, struct zone *zone) | |
1237 | { | |
1238 | return 1; | |
1239 | } | |
1240 | #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */ | |
1241 | ||
97965478 | 1242 | #endif /* !__GENERATING_BOUNDS.H */ |
1da177e4 | 1243 | #endif /* !__ASSEMBLY__ */ |
1da177e4 | 1244 | #endif /* _LINUX_MMZONE_H */ |